Vous êtes sur la page 1sur 347

llr

Universit du Qubec
cole de technologie suprieure
DPARTEMENT DE GNIE DE LA PRODUCTION AUTOMATISE
GES-802
Analyse de faisabilit
NOTES DE COURS
Par
Ali Gharbi
Professeur
Rvis : Aot 2009
CHAPITRE 1:
CHAPITRE 2:
CHAPITRE 3:
CHAPITRE 4:
CHAPITRE 5:
CHAPITRE 6:
CHAPITRE 7:
CHAPITRE 8:
CHAPITRE 9:
ANNEXE 1:
Table des matires
ESTIMATION DES MARCHS, REVENUS ET COTS
INTRODUCTION LA NOTION DE RISQUE ET D'INCERTITUDE
ANALYSE CONOMIQUE TRADITIONNELLE ET AVANCE SOUS
RISQUE ET INCERTITUDE
TECHNIQUES DE DCISION STATISTIQUE AVEC INFORMATIONS
ADDITIONNELLES
L'ARBRE DE DCISION
LMENTS DE BASE DE LA SIMULATION DES SYSTMES
VALUATION DES NOUVELLES TECHNOLOGIES : UNE APPROCHE
MULTICRITRE
DCIDER FACE LA COMPLEXIT: ANALYSE HIRARCHIQUE DES
PROCDS (AHP)
LMENTS DE CONCEPTION D'UN PROJET
MANUEL D'UTILISATION DU LOGICIEL EXPERT CHOICE
estimating markets,
revenues,
and costs
chapter 1
5.1 INTRODUCTION
In any planning analysis for strategie manufacturing investments, there is
great need for reasonably ace urate estimates of market conditions (e.g.,
market size and share) : pricing structure, and cost patterns. Severa! tech-
niques for forecasting the size of the market for new or m_odified product s
are presented in .. Quantitative forecasting l)let_hods covered are
regression techn_i_qyes in-
elude the Once an estimate of
market demand for a new or modified product has been preparea, the priing
policy and revenues associated with that product become a matter of con-
to ... .. toplc iii the .. subject of SectiQJJ $.3. Numerous
for capital, material, and .. overhead costs are
dis,c.ussed in Section 5.4. Finally, the impact of inflation on. the cost estimat-
add_r:essed in Section 5.5.
-
Estimating Markets, Revenues, and Costs Chap. 5
5.2 FORECASTING THE MARKET
A forecast is useful if it reduces the uncertainty surrounding an event, and by
doing so results in a decision that has increased value in excess of the cost
required to produce the forecast. Forecasts are also useful to givc manage-
ment an image of the firm it guides. Companies are often what they want
themselves to be. An open-minded, optimistic vision of the future can lead
to growth and profit; a narrow. pessimistic view can lead to stagnation or
worse.
5.2.1 Forecasting Procedure
A conscientious forecaster will follow a logical step-by-step procedure
in developing and revising forecasts . The first step is to decide what re-
sponse, or quantity, to forecast. Th en pertinent numerical data are gathered
and summarized in graphical form, whenever possible. Often the data repre-
sent sorne response as a function of time, such as sales per quarter. Such
data are referred to as comprising a rime series. The time series should be
examined for underlying patterns, and the forecaster should attempt to ex-
plain these patterns . For example, a large sudden increase in sales m ~ y ~ e
explained by increased government spending. A long-lerm steady nse m
sales may be a result of graduai public acceptance of a new product.
Typical forecasts based on lime-series data merely ex tend historical
patterns into the future and do not result in predicting abrupt changes (tum-
ing points) in activity levels . ln this respect, the forecaster should search for
causal relationships between the response variable (e.g., sales) and other
variables . For example, a rise in interest rates is usually tollowed by a
decline in new construction. This is a causal relationship thal is useful in
forecasting major upswings and downturns in the construction industry.
The next step of the forecasting process is to apply expert judgment to
forecasts obtained by utilizing causal relationships and/or time-series analy-
sis . The result of this evaluation, whether it takes the form of a Del phi study
or simply a group's or an individual's opinion, should be documented and
included in the forecast. Thus forecasting involves a coordinated approach
to collecting data, identifying and mathematically testing causal relation-
ships , and/or selecting a procedure for projecting time-series data, and
choosing a structured approach for eliciting exper1 opinion concerning the
problem under consideration.
One of the most effective forecasting strategies is first to use mathe-
matical techniques based on past data and then t introduce judgment in
tt mpting 1 d cid h w th futu r pr bobly wi ll di ff r fr m th p st. A
pr irn udva nr n of this opproa h is th at 1t r lu th l\111 11 r f f11 t M ~ 1
whi h h11111 tn
1
ri n 11111 1 1 ' 1ppl 11 Il wl ly ' ll nll 11 11 1111lh 11111
Sec. 5.2 , Forecasting the Market
83
cal techniques and informedjudgment, both methods can serve as checks on
each other and tend to minimize gross errors.
To evaluate any quantitative forecasting technique, the procedure first
should be defined and theo applied to representative data. For example, if
one's goal is to forecast sales one quarter in advance, one could start by
applying the proposed forecasting technique to, say. the last three years of
quarterly data. Begin with the first quarter and apply the procedure to take
each new quarter's actual sales into account when developing the next quar-
ter's forecast. Theo for the historical data one can measure the difference
between the forecast and actual sales. The difference is the forecasting er-
ror, and severa! techniques should be evaluated in an effort to minimize this
err or.
5.2.2 Forecastlng New Products
Because CIM usually requires rethinking the way a firm does business,
modified existing products and new products form the basis of forecasting
and eventually justifying the enabling technology. When a company wishes
to forecast with reference to a particular product, it must consider the stage
of the product's li fe cycle for which it is making the forecast. The availabil-
ity of data and the possibility of establishing relationships between the fac-
tors depend directly on the maturity of a product, and hence the !ife-cycle
stage is a prime determinant of the forecasting method to be used (e.g .. see
Ref. [1]).
At each phase in the li fe of a product, from conception to maturity, the
decisions that management must make are characteristically quite different,
and they require different kinds of information as a base. ln the earl y stages
of product development, the manager wants answers to questions such as
these:
What are the alternative growth opportunities of pursuing product X?
How have established products similar to product X fared?
Should we enter this business, and if so, in what segments?
How should we allocate R&D efforts and funds?
How successful will different product concepts be?
How will product X fit into the markets 5 or W years from now?
'
!
1
J
1
i.
1
1
' 1
: ~
EstimatlnJ.! Markets, Reuenues , and osts Chap. 5
product that is still "only a gleam in the eye," information about its likely
performance can be gathered in a number of ways, provided that the market
in which it is to be sold is a known entity. For example, one can compare a
proposed product with competitors' present and planned products, ranking
J it on quantitative scales for different factors. If this approach is to be suc-
cessful, it is essential th at the persons (hopefully. experts) who pro vide the
basic data come from different disciplines-marketing, R&D, manufactur-
ing, legal, and so on-and that their opinions be unbiased.
Frequently, however, the market for a new product is weakly defined,
the product concept is still fluid, and history seems irrelevant. This has been
the typical case for gas turbines, modular housing, pollution measurement
deviees, and computer terminais. Before a product can enter its (hopefully)
rapid penetration stage, the market potential must be and the .product
must be introduced. Then market testing may be advtsable. At thts stage,
management needs answers to such questions as:
What shall our marketing plan be-which markets should we enter,
and with what production quantities and priees?
How much manufacturing capacity will the early production stages
require?
As demand grows, where should we build this capacity?
How shall we allocate our R&D resources over time?
Significant profits depend on finding the right answers, and it is there-
fore economically feasible to expend relatively large amounts of effort and
money on obtaining good forecasts (short, medium, and long range) .
A sales forecast at this stage should provide three points of informa-
tion: the date when rapid sales will begin, the rate of market penetration
du ring the growth phase, and the ultimate levei of penetration, or sales rate,
during the maturity phase.
5.2.3 Quantitative Forecasting Techniques
Two relatively simple, yet extremely useful, techniques for obtaining
initial time series forecasts of demand for a new or modified product are
described in this section: ( 1) regression analysis and (2) exponential smooth-
ing. Results of quantitative forecasting are coupled with subjective forecasts
to develop the final projection of market conditions.
5.2.3.1 Linear Regression Analysis
Regression is a statistical method of fitting a Ii ne through data to mini-
mize squared error. It is exact; however, graphing might be used to.provide
a satisfactory approximation. With linear regression, statistics can be used
to obtain a forecast and a confidence interval for the forecast . For example,
Sec. 5.2
Forecasting tite Market
85
a person using regression might forecast sales for a particular product for the
next month to be, say, $425,000 and also can state that she is, for example,
95% confident that actual sales will be between $390,000 and $460,000. This
assumes that basi.c market conditions and trends will remain unchanged .
Confidence mtervals are useful in identifying a change in the trend of a
set of data. If two or three consecutive observations fall outside these timits,
the can be fairly sure that the basic relationship has changed since
the htstoncal were gathered. This information may prove financially
valuable by enabhng one to re act quickly to a change in trend. If the trend
changes, however, the forecaster is advised to discard or substantiafly dis-
count data regarding the previous trend.
In linear regression involving one independent and one dependent vari-
able, the relationship that is used to fit n data points ( 1 :s; ; :s; n) is
y= a+ bx
(5-1)
A mathematical statement of expressions used to estimate a and b in
the simple linear regression equation 5-1 is as follows:
i X;Yai - Xi Yai
b = i = l i = l
i XJ- xi X;
(5-2)
; ,. , i = l
a= Ya- bi
(5-3)
Here x and J? are averages of the independent variable and dependent vari-
able, respecttvely, for the n data points .
Example 5-l
A durable goods manufacturer has found persona! disposable income in its market
region in a given quarter to be strongly related to sales in the following quarter.
data are listed and summarized in Table 5-1. Because a plot of these data
md1cates an approximately linear relationship between the dependent variable (on
. the Y axis) and the independent variable (on the x axis), linear regression is used to fit
an equation to the data. The data and calculations summarized in Table 5- J are
utilized below to determine the linear regression equation:
n n
L X;Yai - XL Yai
b = i l i:l _ 2,626,817 - 250.6(9,788)
- 1,416,926 - 250.6(5,01 2)
2: xl - X 2: x;
1=1 1= 1
173,944.2
160,918.8 = 1.
081
a = Ya - bi= 489.4 - 1.081(250.6) = 218.5
y= 218.5 + 1.081x

'

Table 5-l
CALCULATIONS FOR S IMPLE LINEAR REGRESSION
Data Point, i
(period) Y ai
X; X;Yai x!
360 121 43,560 14,641
2 260 118 30,680 13,924
3 440 271 119,240 73,441
4 400 190 76,000 36, 100
5 360 75 27,000 5,625
6 500 263 131,500 69, 169
7 580 334 193,720 111,556
8 560 368 206,080 135,424
9 505 305 154,025 93,025
10 480 210 100,800 44, 100
t l 602 387 232,974 149,769
12 540 270 145,800 72,900
13 415 2 18 90,470 47,524
14 590 342 201,780 116,964
15 492 173 85, 11 6 29,929
16 660 370 244,200 136,900
17 360 170 61,200 28,900
18 410 205 84,050 42,025
19 680 339 230,520 114,921
20 594 283 168, 102 80,089
Totals 9,788 5,012 2,626,81 7 1,416,926
" "
LX; "' 5,0 12 2: y., = 9,788
i ::. l i l
"
LX;
"
< = !.:..!__ = 5,012 = 250 6 Lx!= 1,416,926
. " 20 .
i l
"
2: y.,
"
y.= i = l, =



= 489.4 2: x,y.1 = 2,626,817
il

y., - actual quarterl y sales ($ 10
1
) for period i
x
1
di sposnble lncome in precedl ng period (S IO')
!li
Sec. 5.2 Forecasting the Marker
87
Figure 5- l shows the plotted data and the calculated regression tine (as weil as
confidence li mits, to be discussed later) . As an ex ample of how the regression equa-
tion is used, suppose that disposable income for the previous quarter is $3 10 x 10 6.
Then our forecast or estima te of sales for the current quarter, y, is
J = 218.5 + 1. 081(310) = 553 .6 or $553.6 x JOl
To calcula te a confi dence interval for our forecast, we must obtai n the standard
deviation of a single esti mated value, s(y). This quantity is a measure of the tight-
ness of the individual data points about the regression tine,
[
" ] l/2 [ - 2 ] 112 2: (y.;_ y,)2
1
+ .!_ + (xo - x)
,,. 1 n "
s( y ) = n _ 2 L (x; - iV
i l
(5-4)
Here x
0
is a value of the independent variable being used to obtai n t he quarter! y sales
forecast, y, and it lies in the range of observed x
1
(i .e., Xmin :5 x
0
:5 xm. , ) :
y = a + bxo

.....
700 1-
..




., ...
.!! 600
,,<f:'


0

-8
500
oP
c;
,.,_, s
0
vo ::
'
>
,/. . .

:x
>-
,
i
/
t:
..
" 0
200
llii!HIJII ''"'ln (lf'lfoUI"u l llorl '< luil ll httiO II I rtollrtl
11'"" l 'htl l lll lllliltl l - ll1111 f111 1' \"' "1'111 \ 1,
1
1
xx EstimatinK Markets, Revenues, and Costs Chap . 5
The 1 00( 1 - a) percent confidence interval for the forecast of actual quarterly sales is
defined to be between
Y + (la/l,n- l)s(y) (upper limit) (5-5)
and
Y - (tan,n- l)s( y) (lower limit) (5-6)
The term tan,n-l is an appropriate value of the t statistic. A table of t values and
instructions for using this table are given in Appendix C of Ref. [14] as weil as in most
statistics texts.
For a 95% confidence interval (a = 0.05, a/2 = 0.025) and for n data points,
there are n - 2 degrees of freedom. Thi s means that with repeated forecasts involv-
ing n data points, approximately 95% of the confidence intervals calculated with the
equation above will con tain the true value of the response variable if the underlying
causal rel ationship does not change .
The information necessary to determine the confidence interval for this fore-
cast is summarized in Table 5-2 and is used with Eq. 5-4 to perform the following
calculations.
- [52,482.8] l/l [ 1 (31 0 - 250.6)
2
]
112
s(y)- --18-
1
+ 20 + 160,919.6 .
= 55.88
For a 95% confidence interval we have
to.o2S.I8 = 2. 1009
and by utilizing Eqs. 5-5 and 5-6, these limits are obtained:
Y+ Uan,n- 2)s(y) = 553.6 + 2.1009(55.88) = 671
.v - Uan,n-2)s(y) = 553.6- 2. 1009(55.88) = 436
(upper limit)
(lower limit)
To summarize, our forecast of next quarter's sales is $553,600, and we are 95%
confident that the true value (outcome) of quarterly sales will occur between
$436,000 and $671,000.
The correlation coefficient is a measure of the strength of the relation-
ship between two variables only if the variables are linearly related. In
Example 5-1 the correlation coefficient, r, can be determined as follows :
r =
where - 1 < r < 1.
n
2: (Yai - Y;)
2
1.0- :_i::_l ---
2: (Yai - Ya)
2
i-l
If there is no relationship at ali (a shotgun etfect) between the depen-
dent and independent variables , r will be zero, or nearly zero. A negative
Sec. 5 .2
Foreca.ftIII( the Market
Table 52
CALCULATIONS TO DETERMINE CONFIDENCE INTERVAL
Data Point , i
(period)
y., x, y, (y.,- y,)' (x1 - i)
1
(y., - .Y.J'
360 121 349.3
114.5 16,796.2
16,744
2 260 118 346. 1 7, 413 ,2
17,582.8 52,624
3 440 271 511.4
5,098.0
416.2 2,440
4
400 190 423.9
571 .2 3,672.4
7,992
5 360 75 299.6 3,648.2
30,835.4 16,744
6 500 263 502.8
7.8 153.8
11 2
7 580 334 580.0
0.0 6,955.6
8,208
8 560 368 616.3 3,169.7
13,782.8
4,984
9 505 305 541! .2 1,866.2
2,959.4
243
10
480 210 445 .5 1, 190.2
1,648.4
88
Il 602 387 636.8 1,211.0
18,605.0
12,679
12 540 270 510.4
1!76.2
376.4 2,560
13 4 15 218 454.2 1,536.6
1,1162.8
5,535
14
590 342 588.2
3.2 8,354.0 10, 120
15 492 173 405.5 7.482.2
6,021.8
7
16
660 370 618.5 1,722.2 14,256.4
29,104
17 360 170 402.3 1,789.3
6,496.4 16,744
18 410 205 440.1
906.0 2,079.4
6,304
19 680 339 585.0 9,025.0 7,814.6
36,328
20 594 283 524.4 4,844.2
1,049.8 10,941
Totals
52,482.8
160,919.6 240,501
Key:
y. 1 = actual quarter! y sales ($ 101) for period i
y1 = estimated quarterly sales ($101) for period i
x1 = disposable income in preceding period ($10)
From calculations associa led with Table 5-1 :
y = 218.5 + 1.081x; and x = 250.6 and .Y. = 489.4
value of r indicates that one variable decreases as the other increases, whil e
the variables both increase (or decrease) at the same time when r is posi-
tive. The closer ris to -1 or +1, the more "perfect" is the correl ation .
Using Eq, 5-7, the correlation coefficient for data in Table 5-2 is
r=
n
2: (Ya;- Y;Jl
1.0 - .:-. ':..:.. 1 ----
2: (Yai - Ya)
2
i=l
= V0.78I8 = 0.88
j
52,483
1.0 - 240,501
This value of r indicates a fairly good relationship between the independent
and dependent variables.
1!9
-
Vi
.1:
' 1
1
1
.
90 Estimating Markets, Revenues, and Costs Chap. 5
5.2.3.2 Exponential Smoothing
A major ad van tage of the exponential smoothing method compared to
simple linear regression is that it permits the forecaster to place relatively
more weight on current data rather than treating ali data points with equal
importance. Forecasting equations can quickly be revised with a relatively
small number of calculations as each new data point is collected. Also,
exponential smoothing does not assume linearity.
The main disadvantage of exponential smoothing concerns the basic
assomption that trends and patterns of the past will continue into the fu-
ture. However, it is more sensitive to changes than is linear regression.
Because time-series analysis cannot predict turning points in the future,
expert judgment and/or analysis of suspected causal factors should be used
in interpreting results of forecasting methods discussed in this chapter.
The basic exponential smoothing model that we shall discuss and illus-
trate is shown as follows:
S, = ot'x, + (1 - a')S
1
-
1
(5-8)
or
new = a' ( new ) + (1 _ a') (previous)
estimate datum estimate
This term, a', the smoothing constant, merely provides a relative weighting
for the new datum point compared to previous estimates. In general, a'
should lie between 0.01 and 0.30, but the analyst should not hesitate to use a
value outside this range if it gives better results with representative historical
data.
Single exponential smoothing is Hlustrated with data listed in Table 5-3
that are graphed in Fig. 5-2. Based on Eq. 5-8, the following are example
calculations forS,, which is termed a "smoothed statistic" for period t.
S, = a'x, + (1 - a')S,_
1
s 1 = 0.3(50) + o. 7(50) = 50
s2 = 0.3(52) + o. 7(50) = 50.6
s] = o.3(47) + o.7(50;6) = 49.52
s4 = o.J(5I) + 0.7(49.52) = 49.96
ln determining S
1
above, a value of S
0
must be estimated. Ifthere is no trend
in the initi al data, an estimate based on the average of the flrst few data
roi nts is adequotc. Here S0 wos cstimutcd lO be 50.
To bctt cr undetstnnd the mc:n nlog of Nlll oOthi n!(, the fol
o u how d!111 11 11 ll l lh'tll \ 1 , 1 ,
1
, 11 1, '" 11 lnd111l cd
-\.
80
- 70
:J
'
"
0

ill 60
5
2
-g
j
1
Table 5-3
EXPONENTI AL SMOOTHING EXAMPLE
Period
Number, 1
0
l
2
3
4
5
6
7
8
9
10
Il
12
13
14
15
16
17
18
19
20
Demand. x,
(1,000 units)
-
50
52
47
51
49
48
51
40
48
52
51
59
57
64
68
67
69
76
75
80
S,
(a' = 0.30)
50.00
50.00
50.60
49.52
49.96
49.67
49. 17
49.72
46.80
47. 16
48.61
49.33
52. 23
53.66
56.76
60. 13
62. 19
64.23
67.76
69.93
72.95
lt!
,.,
,.,
/
.....
,...,
..

/
.....
.1''
' l'/
-
--. Demand. x,
l'ori;tl l lllllltlit
--- JJ. Exponont lally
lffiOOthod
esti mete, $,
1 1
1
14 1 10 1tl

,1 A, 1 'J tl lllll li lttW


-
' 1
92 Estimating Markets , Revenues, and Costs Clwp. 5
S, = a'x, + (1 - a ' )S,_
1
(5-9)
where
Sr - I = a'Xr-l + (1 - a')S,-2 (5-1 0)
and then
S, = a'x, + a'(l - a')x, _
1
+ (1 - a')
2
Sr- 2
lt is possible to continue substituting smoothed values in the same fashion
until we get the following:
S, = a'x, + a ' (l - a')x, _
1
+ a'(l - a')
2
X1- z
(5-12)
+ a ' (l - a')
3
x, _3 + + (1 - a')'So
As one can see, every previous value of x is included in S, . The i
values are weighted so that the values most distant in time have the smallest
weighting factors . A large a' will place very little weight on remote data.
The following calculations illustrate the weighting of the different data points
included in S
4
above as described by Eq. 5-12.
s. = 0.3(51) + 0.3(0.7)47 + 0.3(0.7)
2
52 + 0.3(0.7)
3
50 + (0.7)
4
50
= 0.30(51) + 0.21(47) + 0. 147(52) + 0. 1029(50) + 0.2401(50)
= 49.96
This agrees with our calculation of S4 earlier.
Table 5-3 shows a continuation of the example above for hypothetical
demand data over 20 periods. It should be noted that the higher the value of
a '. the cl oser the new estimate will be to the most recent datum point.
5.2.4 Subjective Forecasting Techniques
The quantitative forecasting methods discussed above are underpinned
by the premise that the future is an extension of the past. However, the
future will contain events that today are poorly understood or completely
unanticipated . Hence the aim of this section is to explain and illustrate two
techniques for developing subjective (i.e., qualitative) information for pur-
poses of making a forecast : (1) the Delphi method and (2) technology fore-
casting.
5.2.4.1 The Delphi Method
Most decision makers draw on the ad vice of experts as they form their
judgments. Often the decision situation is highly complex and poorly under-
stood, so that no single person can be expected to make an informed deci-
sion. The traditional approach to decision making in such cases is to obtain
Sec. 5.2 Forecasring the Market
93
expert opinion through open discussions and to attempt to determine a con-
sensus among the experts. However, results of panel discussions are sorne-
times unsatisfactory because group opinion is highly influenced by dominant
individuals and/or because a majority opinion may be used to create the
"band wagon effect. '
The Delphi method attempts to overcome these difficulties by forcing
persons involved in the forecasting exercise to voice their opinions anony-
mously and through an intermediary. The intermediary acts as a control
center in analyzing responses to each round of opinion gathering and in
feeding back opinion to participants in subsequent rounds . By following
such a procedure, it is hoped that the responses will converge on a consen-
sus forecast that turns out to be a good estimator of the true outcome.
Two premises underlie the Delphi method. The first is that persons
who are highly knowledgeable in a particular field make the most plausible
forecasts . Second, it is believed that the combined knowledge of severa!
persons is at least as good as that of one person .
Typically, the technique is initiated by writing an unambiguous de-
scription of the forecasting problem and sending this, along with relevant
background information, to each participant in the study. Often the partici-
pants are invited to list major areas of concern in their particular specialty as
they may relate to the problem being addressed by the study. The first
questionnaire sent out might request the opinion of each expert regarding
likely dates for the occurrence of an event identified in the problem state-
ment. Because responses to this type of question will normally reveal a
spread of opinions, interquartile ranges are customarily computed and pre-
sented to the experts at the beginning of the second round. Interquartile
ranges identify upper and Iower quartile values in the continuum of re-
sponses such that 50% of the responses fall within that range .
In the second round of the Del phi technique, the participants are asked
to review their response in the first round relative to interquartile ranges
from that round. They then have the opportunity to revise their estimates in
tight of the group response. At this point, participants can request that
additional information relevant to the forecasting problem be gathered and
sent to them.
If an estima te departs appreciably from the group median, the res pon-
dent who furnished it is asked to give reasons for his or her position.
Frequent) y, ali panelists are urged to conceive statements that challenge or
support estimates falling outside the interquartile or sorne other range of
responses. These reasons, along with routine second-round estimates for
the entire group, are again analyzed and statistically summarized (usually as
interquartile ranges, although other measures capable of showing group con-
vergence or divergence could be used) .
In those cases where a third-round questionnaire is fe ft necessary,
participants receive a su rn mary of second-round responses plus a request to
1
"ii
1 ~
l
"
j
;
1.
1
1
,,
1
Ji.:
;: ..
:ii
1
. 1
:
i:
~
. \ ~
:L
~ !:
.:
' !
: ij
;i
,.
. ! ~
94 Estimating Markets, Revenues, and Costs Chap. 5
reconsider and/or explain their estimate in view of group responses in the
second round. They are again asked to reassess their earlier responses and
possibly to explain why their estimates do not conform to the majority of
group opinion.
An example of quantitative result s of the Delphi technique is summa-
rized in Fig. 5-3. The problem of concern to a manufacturer of large earth-
moving equipment was to develop a forecast of total company sales during
fi scal year 19XX. Six marketing and sales experts (A through F) were asked
to consider historical company and industry data and anonymously prepare
a forecast of bookings for a particular product. First-round questionnaire
result s are shown in the top part of Fig. 5-3. The median response in round 1
was 229 and the interquartile range of the responses was 85. Results of
round 1 were fed back to each participant along with additional information
pertaining to the forecast that each person requested. The second and third
rounds were completed in a similar manner. Notice how forecasts in the
three iterations of questioning tend to converge, with a final median forecast
value of 260 unit s and an interquartile range of 47.
5.2.4.2 Technological Forecasting
Technological forecasting is a name given to a myriad of specialized
forecasting techniques. It provides procedures for data collection and anal
ysis to predict future technological developments and the impacts such de-
velopments will have on the environment and lifestyles of people. These
techniques seek to make potential technological developments explicit, but
more important , they force deci sion makers to try to anticipate future devel-
opments.
Technology forecasting is a method that can be used to estimate the
growth and direction of a technology. A typical question that technology
forecasting attempts to answer is: What will be the machining tolerances of
numerically controlled machine tools in the future? In dealing with operat-
ing parameters, the forecast is limited to specifie technical units and dimen
SIOn S.
Trend extrapolation is often used to make technological forecast s.
This technique is based on a historie time series for a selected technological
parame ter. It often is assumed that the factors influencing historical data are
likely to re main constant rather than to change in the future . Usually, a
single-function parameter such as speed, horsepower, or weight is extrapo-
lated. A good trend extrapolati on depends on selection and predi ction of
key pararneters of performance. The trend undcr stud y shoul d be cnpnble of
qunnti fi cnti on in ordcr that it cnn be portroycd nurncricnll y, 011d nn ndcqunt
(lut o hnsc cxist on which to hnsc 11 roli11blo trend lloo. An oxurnpl
n f)ltliiOilt cd ln I'IH. Nntko tli t1t tll 11 y 1

en -
': 8
::: "'
>-

:;::

5
.,
N
..
'

"
i
o.

g.
:;;



,;:
w
..
. 2 :J
l ]

.. ..
a.t:
a. ..
::> ::>
CT
!ij
1i
:::;:
..
f

CT
"' 1':::3
Il IL

Il'"
::;;a:
io
'
' 1

/
9
N


'

/



(.,
"'
"
u
/
/
/
I.L

'6
1?
i!
/
/
/
/
..
N,._
Il LI

Il'"
::;;CC

0
w
/
/'
K/
1
<(
. .,
/
/
/
/
/
/
I.L
N

"
0
l6
N ,.._
" ...

Il ..
::;;a:
' -.... -.
.L.
,--.
/
/
/
-

/
/
/

__.j
M
-g
il
a:
1? ....
.. -g li
- 8 ::> ..
.=: a
8. . ti .
-;:
Ill - - u
.c. 8. 1
ILl
0
6
'B t:<i
il - .. IQ
3: E i}!l
.B .... t=
u c ;:;lll a. _8
a::
! !: "
we<(
! '8.1 ::-: ,;,
m "'-;::n;g ::: -c:1

e- 8. .....

-c; .... y
Q
Gl O Il ::J -
.2 - & l '
'"';( -c E c - .,
::JIll 41 0
3
- fi
ii .
- .... 0
z "'

li:
.. 2

l! 8.
c
0
0

a.
Ill o.


., ::> 3:
t
c .,
::>
1?
::> c > -c
...
e ::> .. c
> ::>
e
== 10
0
-g t::
c

-e
.. ..
..
E
0 ::> a.
'ii
'*
L: o c.
: f- 2 ::<
...

- <
1 ......
9
'D

1ii


a.
.,.
>-
....
"'
c
30
20
.rrlmatlrr g Markets, Revenues, arr d Cosrs
Ethylene
Production "' plant capacity
Selling priee
lrap. 5
8. 4
c

a;
c
.2
tl
"
"' 0
ct
0.4
0.3
0 .2
Date of points 1955- 1969
(continuous from loft to right)

0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8
Average copacity of l ive largest plants
Figure 5-4. An example of linear trend extrapolation.
An advantage of trend extrapolation is that historical data are often
readily obtainable. A straight-line or fitted-curve projection of the future is
easily understood and used. A drawback to extrapolation stems from the
assumption that factors that shaped the past will continue to hold basically
unchanged in the future . Trend extrapolation techniques cannot predict un-
Sec. 5.2 oreca.rrlng the Marker
foreseen technology interactions such as unprecedented changes or inexpli-
cable discoveries.
The substitution curue is based on the belief that a product or technol-
ogy that exhibits a relative increase in performance over an older (i.e.,
established or conventional) product or technology will eventually substitute
for the one having lesser performance. The relative increase in performance
is the important factor in the substitution of one technology for another. A
basic assumption with this method is that once the substitution of one tech-
nology for another has begun it will irreversibly continue to completion.
Listed below are sorne common examples of the substitution effect:
Old Technology
Petroleum lamps
Horse-drawn carriages
Steam locomotive
Cotton
Leather
Soap
Reciprocating engines
Hardwood floor
New Technology
Electric lamps
Automobiles
Diesel locomotive
Synthetic
Vin yi
Detergent
Turbojet engines
Plastic flooring
The forecast starts with the observation that a new technology is start-
ing to displace an older technology. A measurement term that best defines
the fraction of total usage of each technology must be selected, and lime-
series data are gathered for both technologies . These data are used to estab-
lish the initial takeover rate and to predict the year in which takeover will
reach 50%. A typical substitution effect for two technologies is shown in
Fig. 5-5.
Forecasting by analysis of precursor events uses the correlation of
performance trends between two innovative technologies. Because techno-
logical advance usually follows a pattern of continuous increase, situations
frequently occur in which an indicator of technical progress lags another by a
given period of time. lt is thus possible to utilize the leading technology to
predict the status of the lagging technology over a ti me period equal to the
lag ti me. The frequent! y cited example of precursor events shown in Fig. 5-6
concems the historical relationship between maximum speed of military
aircraft to the maximum speed of commercial aircraft .
In this example, it was found that the speed of commercial aircraft
followed the speed of military aircraft by six years in the 1920s and eleven
years in the 1950s [9): As a result, it was predicted that commercial aircraft
with speeds of Mach 2 may be expected no later than 1970, or if such aircraft
were not introduced at this time, aircraft with speeds of Mach 3 will be
introduced near 1976. The forecast in this situation says, in effect, that there
is a logical time for the introduction of Mach 2 aircraft but if other forces

' 1.-9
\':
1:
!:
.i
.,,
i 1
. ;
i
1.
i i
' i"l 1
1
1.
1 j
: 1
1
j
l
1!
' ' , ...
::

Il
:K


[1
1
[1 l
i;
fr

r
1;
'

j:
;, j

>
:
1
1


1
1
98
2

'6
=



0..
g

'6
.<:
6
c:


:;;
0..
Estimating Markets, Revenues, and Costs
- ...
.... ,
\
\
\
\
' Ti me-..
New
technology
Point at which the new
tochnology has a 50%
takeover rate
Old
Typical technology lita cycle
Decline in
rate of growth
-....

'
' \
\
Ti me----..
'
'
'
'
......
Figure 5-5. An example of substitution curves.
Chap. 5
(e.g. , economie, social, and political) cause this time to be passed over, this
performance range will also be passed over.
5.3 PRICES AS A RETURN ON INVESTMENT*
Attention in Section 5.2 was directed at various techniques to forecast mar-
ket demand (sales dollars and/or units of production) for existing and new
manufactured products . Once the annual sales in dollars, S, or production
volume in unit s, q, have been estimated for a particular product, it is
normall y ncccssary to develop revenue cstimates so that bencfit s of product
chnngcs, includi ng new or modifl cd mnnufacturi ng proccsscs/mutcrials, cnn
Ah lli CII rrtllll liN he il 1
1
lltll)tl!lll
"Il ''"""'"'' 1, llli'VIWhl 0 1 lM il y l'uhll -hlt!N l ' til pillnllilol , N.,w Y111 k. hy jll'i
111 lili l' 1111111111 - h1111il
Sec. 5.3
3000
2000
1500

:;:
!
l
E
"
400
E

3001-
Priees as a Return on lnvestment
)'

B I'T .,_
1930 1940
Year
B bomber
F fighter
T transport
Mach 3
{at 35,000 f1
altitudal
19BO
Figure 5-6. Comparative speed trends of combat and transport aircraft.
99
be assessed. In this section the objective is to demonstrate how priees can
be set through policy adopted by a firm and then used to estimate annual
revenues for selected product lines. Es timation of costs is considered in
Section 5.4.
h
----'""No .,....., ,.,, ... ----.-.-,....;--r-
1
'\
.. , ... , "'":e:as :::cac: aazau!l nz.awa.:Sl.t aew ;p:4 w .. es .......
; i
\: 1
'.:;
1
1
: ..
,.
iJ
1 i
\J
:. 1
; :
!
1
; j
il
p
1
"

. : i
! 1
':q
:l ,i
1' 1

: iii
'!
ji;
t.l
;!


'
1
t
100 Estimating Markets, Revenues, and Costs Chap. 5
the circumstances. Maximization of profit, return on investment, market
share, and/or total sales are typical objectives. For example, market share
can be the dominant pricing objective for standard products, while custom-
made products may focus on return on investment.
Cost is a significant factor in ali pricing decisions-il is needed to
calculate the minimum or "floor priee'' that must be met or exceeded. A
significant portion of total product cost is shifting to equipment-related costs
as facilities become more automated. Clearly, capacity utilization needs to
be recognized in setting priees, and in the long term priees will need to
recover full costs plus provide an adequate return on the capital invested.
In procedures of this type, the costs of production should be based on
the long-run average to allow for abnormal operations which will occur from
time to time. These considerations can usually be estimated from an eco-
nomie break-even chart (discussed in Chapter 2) at sorne reasonable percent-
age of plant capacity.
The acceptable rate of return on investment varies with ali the eco-
nomie factors applicable to a given situation, product, company, or indus-
try. Historical values are commonly-used guides in setting acceptable earn-
ing rates . They average from a low of perhaps 5% for the manufacture of
durable goods up to perhaps 50% for a new pharmaceutical product. A
general average may be 12 to 15%; this range can serve as a criterion in
setting priees.
To visualize the mathematical calculations, the following relations are
useful, where the fiscal period is 1 year, 1 is the effective income tax rate,
1 is the investment, and S is the sales dollars :
return on investment, i* = net profit /i nvestment = YI/
return on sales, i, = net profit/sales dollars = YI S
gross profit, Z = sales dollars minus total costs = (S - Cr)
net after-tax profit. Y= gross profit minus income taxes= (S- Cr)(l - 1)
capital ratio, R = investment/sales dollars, IlS
capital turnover, T =sales dollars/i nvestment = SI!
total costs, Cr= fixed costs + variable costs
. sales dollars net profit
return on mvestment = investment x sales dollars , so
. ., = (S- Cr)(! - 1) = (S- Cr)(! - t)
1
1 s 1
(5- 13)
Si neeS- Cris the markup, Eq. (5-13) may be solved for the markup in
terms of total dollars for a given amount of sales, S, at cost, Cr, excluding
financial costs, in various forms:
Sec. 5.3 Priees as a Return on lnuestment 1\JI
i"'l y
S- Cr=--=--
1 - 1 1 - 1
(5-14)
and the fractional markup or margin based on total costs would be
S-Cr i*I
---c;- = Cr(l - 1)
(5-1 5)
Alternative! y, the average markup in doi;ars per unit where q is the units
sold is:
S - Cr = i(l /q)
q 1 - 1
(5-16)
Thus, where the costs, Cr, are known and the acceptable rate of re-
tum, ;, can be established , the markup and final priee can be determined for
the desired management policy. Note that markup is synonymous with
gross profit, Z. Ca re must be exercised that the wording of markup is clearly
defined. For example, if the fractional net profit based on sales is to be an i"
fraction of the capital ratio, R, it wou id mean th at
(l - Cr)= i* (f)
(5-17)
However, if the wording is such that the fractional net profit on net sales is a
certain amount, i, , then the fractional markup based on net sales is defined
different! y:
(1 - I)(S -Cr).
s = '
(5-18)
Any of these equations may be solved for any one of the variables
when ali the others are known or specified. The values for priees and costs
may be for total output or per unit of output. It should be observed further
that the markup (S - Cr) can be based on any component or combinations
of components desired by management, but diligence should be applied that
no costs are omitted. Furthermore, the basis should be a logical one, utiliz-
ing modern costing principles, so that erroneous priee levels are avoided,
which may impair the number of sales (if set too high) or which may produce
!osses in the long run (if priced too low).
To demonstrate certain pricing procedures, a typical example is pre-
sented.
Example S-2
In manufacturing products M and Nin amounts of 20,000 and 16,000 units annually,
a company analysis shows the following data. It also hasan effective income tax rate
of 52%.
.l
ij
i

i : .
:
:,
1

:
i

!
:
?
.,
ll
[!l''
1)
! '
H
j
H
\:-
;: u

: r
:.
d
;
;,,
1 1
lh
''
''! :1:
: 1
. l
!02 Estimating Markets, Revenues, and Costs Chap.5
lnvestment, 1 (dollars)
Production, q (annual units)
Cost (dollars)
Materials, C..,
Direct , C0
Others, fixed
Total, CT
Unit costs, CT/q
Product
M
200,000
20,000
60,000
120,000
40,000
220,000
Il
N
140,000
16,000
20,000
60,000
80,000
160,000
JO
Il is desired to compute the priees for each producl for each of the following policies:
(a) 20% return on investment.
(b) 30% markup based on total cost .
(c) 20% profit on net sales priee.
(d) 40% markup based on incrementa! costs.
(e) 50% markup based on value added by manufacture.
Solution
(a) A 20% return on investment, i*, requires calculations of the required sales
dollars, S. using Eq. 5-14. This equation is rearranged:
and
i*I
s = "''=t + CT
s
p=-q
0.20(200,000)
For product M: S =
1
_

+ 220,000 = $303,500
= $303,500 = $15 18
p 20,000 .
0.20( 140,000)
For product N : S = 0.4
8
+ 160,000 = $218,400
= $2 18,400 = $13 60
p 16,000 .
(h) A 30% mnrkup on t ot nl cos t s wou!() menn thnt the <Jifforen
nn<J tOi ni CQ,'I I S OHI NI Ct jll (\1 30% or the !CHili 0
,\
1
( ' , (l \ (',, 111 ,\' 1 1 c, , hi11Hill l tlllillll
bctween snlos
Sec. 5.3 Priees as a Return on Investment 103
Then the unit priee p would be
s - 1.3 x 220,000 = $14.30
For product M: P = 20,000 - 20,000
s - 1.3 x 160,000 = $13.00
For product N: P = 16,000 - 16,000
(c) A 20% profit on sales would mean that the net profit as a fraction of the sales
priee must be 0.20, that is, YIS = i, = 0.20, or from Eq. 5-18,
(1 - I)(S - CT) .
S = ls
S = (l - t)CT
(1 - 1) - i,
For product M: p = _S_ = (1 - 0.52)220,000
20,000 20,000(0.48 - 0.20) = $16.80
For product N: p = _ S_ = (1 - 0.52)160,000
16,000 or nnMn n n = $17. 15
(d) A 40% markup on incrementai costs means that
S - CT = 0.4(C.., + Co)
For product M: The sales dollars are
S = 1.4(C.., + C
0
) + F
= _s_ = J.4(6o,ooo + 12o,ooo> + 40,ooo = s
14 60
p 20,000 20,000 .
For product N:
= _ s_ = 1.4(2o,ooo + 60,000) + 8o,ooo = s
12 00
p 16,000 16,000 .
(e) For a 50% markup based on value added, the basic relation is
Thus:
For Product M:
s
s - CT= 0.5(CT- C..,)
u c, - o.
s
IJ ,IHIH
j(l ,()()())
s 11.<10
l

r
1
1
1
1
, .
1

1::
:
'il
Li
r,
;-j
f

(
'1
1; 1

::
''

i l

l'

1
l'i
:i

1 (j
15

. ;
. '
:. r
1
'
t
l'
1
1,1

r,
1
104
Estimating Markets , Revenues, otid Costs
For Product N: S _

= $14.38
p = 16,000-
A summary of the unit priees is:
(a) Based on investment
(b) Based on total costs
(c) Based on net sales
(d) Based on incrementai costs
(e) Based on value added
Product
M
$15. 18
14.30
16.80
14.60
15.00
Product
N
$13.60
13.00
17.15
12.00
14.38
Chap . 5
Since there are numerous variations in the pricing procedures illustrated, any
policy requires clear statement to avoid misunderstanding of exactly how a final
priee was established. The results in parts (a) and (b) are perhaps the most mean-
ingful.
5.4 COST ESTIMATING: PERSPECTIVE AND APPROACHES
In Chapter 2 we described various cost concepts that are important to eco-
nomie analyses. A key point was that the costs which are important to
economie comparisons are the marginal (variable, or incrementai) costs for
the future, and that these costs may weil include opportunities foregone as
weil as actual cash flows .
The basic difficulty in estimating costs for purposes of making eco-
nomie analyses is that most prospective projects are unique; that is, substan-
tially similar projects have not been undertaken in the past under conditions
that are the same as expected for the future . Hence, data that can be used in
estimating costs directly and without modification often do not exist. lt may
be possible, however, to gather data on certain past outcomes th at are re-
lated to the outcomes being estimated and then to make forecasts regarding
anticipated future conditions .
Whenever an economie analysis is for a major new product or process,
estimating for that analysis should be an integral part of a comprehensive
planning procedure. Such comprehensive planning would require the active
participation of at )east the marketing, design engineering, manufacturing,
finance, and top-management functions. It would generally include the foi-
Iowing ingredients :
1. A realistic master plan for product development, testing, phase into
production, and operation
2. Provision for working capital and facilities requirements
Sec. 5.4 Cost Estimating: Perspective and Approaches
10.5
3. Integration with other company plans
4. Evaluation against company objectives for market position, sales vol-
ume, profit, and investment
5. Provision of operating controls, assuming that the project is adopted
Obviously, such comprehensive planning is costly in time and effort ,
but wh en a new product or process has major implications for the future of a
finn, it is generally a sound rule to devote a greater rather than a fesser
amou nt of effort to complete planning, including estimates for the economie
analyses that are a part of the planning. The application of this rule, of
course, is bounded by constraints of limited time and talent. Following the
rule will tend to minimize the chance of poor decisions or Jack of prepared-
ness to implement projects once the decision to invest has been made.
As technology continues to accelerate, quick and reliable methods are
needed for determining the costs associated with adding new and/or improv-
ing existing manufacturing processes. The methods discussed below are
excellent for estimating costs oflabor, equipment, materials , operations, and
engineering.
5.4.1 Sources of Data
The variety of sources from which information can be obtained is too
great for complete enumeration. The following four major sources, which
are ordered roughly according to decreasing importance, are described in
subsequent sections: (1) accounting records, (2) other sources within the
finn, (3) sources outside the firm, and (4) research and development.
5.4.1.1 Accounting Records
It should be emphasized that although data available from the records
of the accounting function are a prime source of information for economie
analyses, these data are very often not sui table for direct, unadjusted use. In
its most basic sense, accounting consists of a series of procedures for keep-
. ing a detailed record of monetary transactions between established catego-
ries of accounts, each of which has an accepted interpretation useful for its
own purposes . The data generated by the accounting function are often
inherently misleading .for economie analyses, not only because they are
based on past results, but also because of the following limitations.
First, the accounting system is rigidly categorized. These categories
for a given firm may be perfectly appropriate for operating decisions and
financial summaries, but rarely are they fully appropriate to the needs of
economie analyses for longer-term decisions .
Another limitation of accounting data for obtaining estimates is the
misstatements embedded by convention into accounting practice. These are
\
\..)...>
1
l
:!
i

Il
1
r.l

'[
ll!o
106 Estimating Markets , Revenues, and Costs Chap. 5
based philosophically on the idea that management should avoid overstating
the value of its assets and should therefore assess them very conserva-
tively. This leads to such practices as (1) not changing the stated value of
one's resources as they appreciate due to rising market priees, and (2) depre-
ciating assets over a much shorter fife than actually expected. As a result of
such accounting customs, the analyst should always be careful about treat-
ing such resources as cheaply (or, sometimes, as expensively!) as they might
be represented.
The final limitations of accounting data are its illusory precision and
implied authoritativeness. Even though it is usual to present data to the
nearest dollar or cent, the records are not nearly that accurate in general.
Further, the results are often affected by arbitrary allocations and optional
charges and/or credits.
In summary, accounting records are a good source of historical data,
but they have severe limitations when used in making estimates for eco-
nomie analyses. Further, accounting records seldom directly contain the
marginal costs, especially marginal opportunity costs, appropriate for eco-
nomie analyses.
5.4.1.2 Other Sources within the Finn
The usual firm has a large number of people and records which may be
excellent sources of estimates or information from which estimates can be
made. Colleagues, supervisors, and workers can pro vide insights or suggest
sources that can readily be obtained.
Examples of records that exist in most firms are sales, production,
inventory, quality, purchasing, industrial engineering, and personnel. Table
5-4 provides a list of the types of data that might be needed for cost esti-
mating purposes, together with typical sources (mostly intrafirm) for the
data.
5.4.1.3 Sources Outside the Finn
There are innumerable sources outside the firm that can provide infor-
mation helpful for estimating. The main problem is to determine which
sources potentially are most fruitful for particular needs.
Published information such as technical directories, trade joumals,
U .S. government publi cations, and comprehensive reference books offer a
wealth of informati on to the knowl edgeable or persistent searcher.
T'crsonal contacts are excell ent potenti al sources . Vendors, snlespco-
plc, profcss ional ncquaintnnccs, customcrs, bankll, govcn, mcnt fi KCncics ,
of commci'Cc, nnct vvcn compctil ors nrc ofl lln wlll lnA to
dt1d lui<H 11 11 ' 1ion 1r til t\ ICiiHi.lNt 111 t<crlnoN 11 11 d lul tf\11,
Sec. 5.4 Cost Estimating: Perspective and Approaches 107
Description of Data
General design specifications
Quantity and rate of production
Assembly or Jayout drawings
General tooling plans and list of pro-
posed subassemblies of product
Detail drawings and bill of material
Test and inspection procedures and
equipment
Machine tool and equipment require-
ments
Packaging and/or transportation require-
ments
Manufacturing routings and operation
sheets
Detail tool, gage, machine, and equip-
ment requirements
Operation analysis and workplace studies
Standard time data
Material release data
Subcontractor cost and de li very data
Area and building requirements
Historical records of previous cost es ti -
mates (for comparison purposes, etc.)
Current costs of items presently in pro-
duction
Table 5-4
TYPES AND SOURCES OF COST ESTIMA TING DATA
Sources
Product engineering and/or sales depart-
ment
Request for estimate or sales department
Product engineering or sales department
or customer' s contact man
Product engineering or manufaciuring
engineering
Product engineering or sales department
Quality control or product engineering or
sales department
Manufacturing engineering or vendors of
materials
Sales department or shipping department
or product engineering (government
specifications)
Manufacturing engineering or methods
engineering
Manufacturing engineering or material
vendors
Methods engineering
Special charts , tables, time studies, and
technical books and magazines
Manufacturing engineering and/or pur-
chasing department or materials ven-
dors
Manufacturing engineering and/or pur-
chasing department or customer
Manufacturing engineering or plant
layout or plant engineer
Manufacturing engineering or cost de-
partment or sales department
Cost department or treasurer or comp-
troller
Source: Reproduced by permission of ASTME, Manufacturing Planning and Esrimaring
Handbook, McGraw-Hill Book Company, New York, 1%3, pp. 3-20.
5.4.1.4 Research and Development
.._
t
,1
i
-!
"i
10!1 EstimatIII( Markets , Reue11ues, and Costs Chap. 5
are very important estimates to be made and when the sources mentioned
above are known to be inadequate.
5.4.2 Estimating Methods
Severa! methods of estimating exist. However, we limit our discussion
to the (1) conference, (2) comparison, (3) unit, and (4) detailed anal y sis
methods. The first three methods are preliminary techniques for projects
that have not been designed yet or that are in the design phase. The detailed
analysis method is used for preparing final estimates .
5.4.2.1 Conference Method
In the conference method, representatives from different functions in a
firm work together to estimate the various manufacturing costs (and reve-
nues) that are needed. Primary advantages of this method are (1) its speed
and (2) the fact that it causes people with a diversity of backgrounds to focus
on the estimate(s). It is also useful when a company does not have-a formai
estimating depart ment. The main disadvantage of the conference method is
th at it may lack accuracy.
5.4.2.2 Comparison Method
In the comparison method an accumulation of historical data and expe-
rience is utilized. According to a publication by the Society of Manufactur-
ing Engineers (SME): "The estimator applies up-to-date costs derived from
si milar parts to the project, adjusting these costs to suit material, labor, and
processing variations. Care should be taken when using data from larger or
smaller manufacturing quantities. For instance, smaller production lot sizes
will usually increase setup costs per item in the lot. Converse! y, Iabor costs
normally decline with increased efficiency developed over longer production
runs" [15] . The comparison method , like the conference method, is used for
quick estimates.
5.4.2.3 Unit Method
The most popular of the preliminary estimating methods is the unit
method, which iovolves using an assumed or estimated per unit factor.
Sorne examples are:
Capital cost of plant per kilowatt of capacity
Fuel cost per kilowatt-hour generated
Capital cost per installed phone
Revenue per long-distance cali
ost Estimating: Perspective and Approaches 109
Operating cost per mile
Maintenance cost per day of use
These factors may be multiplied by the appropriate unit to provide the
total estimate. The following examples may be used for breaking quantities
to be estimated into units that can be estimated readily:
1. In different units (Example: dollars per week, to convert to dollars per
year)
2. A proportion instead of a number ( Example: percent defective , to
convert to number of defects)
3. A number instead of a proportion (Examp/e : number defective and
number produced, to convert to percent defective)
4. A rate instead of a number (Example. : miles per gallon, to convert to
gallons consumed)
S. A number instead of a rate (Example : miles and hours traveled, to
convert to average speed)
6. Using an adjustment factor to increase or decrease a known or esti-
mated number (Example: defectives reported, to convert to total de-
fectives)
Although the unit method is useful for preliminary estimating, the val-
ues can be very misleading because there is no consideration of the principle
of economies of sc ale or economies of scope .
5.4.2.4 Factor Technique
The factor technique is an extension of the unit method in which one
sums the product of one or more quantities or components involving unit
factors and adds these to any components estimated directly. That is ,
c = 2: cd + 2: Ji x V;
whre C = value (cost, priee, etc.) being estimated
Cd = cost of selected components estimated directly
fi = cost per unit of component i
U; = number of units of component i
(5-19)
Suppose that we need a slightly refined estimate of the cost of a hou se consisting of
1,500 ft
2
, two porches, and a garage. Using unit factors of$40/ft
2
, $2,000/porch, and
$3,000/garage, we can calculate the estimate as
$40 x 1,500 + $2,000 x 2 + $3,000 = $67,000
1 1
1
' V'
110 Estimating Markets, Revenues, and Costs Chap.5
5.4.2.5 Detailed Analysis Method
The d.etailed analysis method is the most reliable form of estimating
and is normally performed by one estimator (with input from other knowl-
edgeable people as needed) . The following steps should be followed in pre-
paring a de tai led analysis estimate [ 15]:
1. Calculate raw material usage, including scrap allowances and salvage
material (direct material).
2. Process each individual component (write the operation sheet).
3. Compute the production time (direct labor) for each operation.
4. Determine the equipment required (new, rework, or on hand) .
5. Determine the required tools, gages, and special fixtures .
6. Determine any additional equipment needed for inspection and tes ting.
Even though the conference and comparison methods are quicker,
most manufacturers use a detailed analysis for a preliminary estimate. It
provides the most accurate results and may be used again as the final esti-
mate aftei- revisions are made.
5.4.3 Labor Analysis
Labor is one of the most important expenses to be considered in cost
estimating. To provide an estimate of tabor costs, the amount of lime in-
volved must first be measured. Four methods of measuring labor time in-
elude standard time studies, Iabor-hour reports, work sampling, and prede-
termined time systems.
5.4.3.1 Standard Time Studies
' 'A standard ti me may be defined as the time required by an average
worker to complete a given task while working at a normal pace" [10) .
When a job is considered to be repetitive, the stopwatch procedure may be
used to obtain this standard time . The following seven steps are involved:
(1) divide the task into elements; (2) observe and record time to complete
each element each time it is performed; (3) determine the pace (rating) at
which the work was performed ; (4) determine allowances for personal
needs, fatigue, and unavoidable delays; (5) deJete observed times that are
not performed under normal working situat ions or conditi ons; (6) determine
the normal ti me; and (7) calcul ate the standard time by adding all owances to
the nornu11 timc. Thus
Tl'\ NT, 1 (N'l' ' docl11 ml) (5-20)
Sec. 5.4 Cos/ Estimating: Perspective and Approaches
where TS
1
= standard ti me for work element j, j = 1, 2, ... , rn
NT; = normal time for work element
rn = number of elements
Ill
The standard times for ali elements are then summed to provide the
total standard time, TS, per unit. lfTS is in minutes, picccs pcr hour may be
calculated as 60/TS.
5.4.3.2 Labor-Hour Reports
Labor-hour reports may be used as a basis to determine tabor hours for
nonrepetitive work or when time studies are not available. These are merely
summaries of hours worked laken from job tickets, with whatever adjust-
ments are appropriate for estimating the times desired.
5.4.3.3 Work Sampling
Work sampling is another method of obtaining tabor hours for nonre-
petitive work. This method does not require a stopwatch or historical re-
ports and is convenient and inexpensivc . According to Ostwald:
A work sampling study consists of a number of observations taken pertaining
to the specifie activities of the person(s) or machine(s) at random intervals.
These observations are classitied into predetined categories directly related to
the work situation. Du ring the course of the work sampling study, tally marks
are made by the technician, such as "work," "idle," or "absent." The key to
the accuracy is the number of observations, which may vary according to the
requirements. [Il, p. 42]
Work sampling is a statistical technique ; therefore, probability theory
must be utilized to plan for or describe accuracy of results. Work sampling
statistics generally are based on the binomial distribution in which the mean
is equal to Np; and the variance is equal to Np;(l - p;), where N = number
of observations and p; = probability thal event i will occur. Since N is
usually large, this bino:nial distribution can be approximated by a normal
distribution.
To begin the process, N ' observations must be made so that p; may be
estimated:
Ni
Pi = N'
(5-21)
where pi = obscrved proportion of occurrence of an event i expressed as a
decimal
N; initinl 11111111
N' luitl nl 11111111
b.s orvutionR of event 1
l llfldlllil llhNCI VIl {
1 -
'\
::-
...
,,

:.
'
:r.
.f
.,
:
. '
k

: j
il
l '
\!
1;
11
L
'
11
r
1
1
" j
11 2 Estimating Markets, Revenues, and Costs . Chap. 5
Once p; has be en calculated, the estimator must decide on the maximum
desirable interval, /, as a measure of the accuracy of the estimate of p
1

Suppose that p' = 50%; if a maximum interval of 2% is desired, the estimate
of p ranges from p'- 1/2 top'+ 112 (49 to 51%). The estimator must also
decide on the confidence leve!. A confidence levet of 90% is common for
work-sampling studies. From a normal distribution table (see Appendix D),
for a confidence leve! of 90%, Z = 1.645. Other values of Z (number of
standard deviations for a standard normal distribution) for different confi-
dence levels are listed in Table 5-5 .
The number of observations of event i necessary to achieve the desired
confidence lev el and interval may be found from the following equation [ 11]:
N = 4Z2p;(J .,...- p;)
1 [2
(5-22)
Once these observations have been taken, the labor hours may be calculated
as
H = (N;IN)HR(l + PF&D)
' Np
where H, = standard labor hours per job element i
N
1
= total number of event i observations
N = total number of observations
H = total tabor hours worked during study
R = rating factor
PF&D = allowance decimal
(5-23)
NP = work units accomplished during period of observing this event
Example 5-4
Suppose that it is desired to know H, for machining a part using work sampling. A
confidence of 90% and an interval of 4% have been chosen, and 125 times out of 150
Table 5-5
Z VALUES FROM STANDARD NORMAL DISTRIBUTION TABLE
Area Between Number of
Limits Standard
(Confidence Level) Deviations Area Outside
(%) -z to +Z Limits (%)
68 :!:1.000 32
90 :!: 1.645 10
95 :!: 1.960 5
99 :!:2.576
Source: Phillip F. Ostwald, Cos/ Estimating, 2nd ed.,
Prentice-Hall, lnc., Englewood Cliffs, N.J., 1984, pp. 43-
44. Reproduced by permission of the publisher.
... ,,..,. - .... - .. ,_.
Cost Estimating: Perspective and Approaches
113
initial observations the operator has been busy. From these values we canuse Eqs .
5-21 and 5-22 to find the number of observations necessary to achieve the desired
confidence levet and interval :
p; = = 0.8333
N = 4(1.645)2(0.83333)(0. 1667) =
940
1 (0.04)2
It turned out th at a total nu rn ber of observations, N = 1 ,262 were needed to
observe 940 in which the operator was busy. The opera tor worked at a rate of about
115%; therefore, R = 1.15. The PF&D allowance factor is 0. 15. The study was done
over a period of H = 6 hr, in which 20 parts were machined. From Eq. 5-23, the
standard time in hours per part is
H = (940/ 1 ,262)(6 hr)( 1.15)( 1 + 0. 15)
s 20
= 0.295 hr per part
5.4.3.4 Predetennined Time Systems
A predetermined time system is defined to be "an organized body of
information, procedures, and techniques employed in the study and evalua-
tion of work elements ... " [10, p. 34] . Such systems are often preferred to
time studies because they are considered more accurate and because they
are generally more acceptable to workers . Numerous systems exist, but
MTM (methods time measurement) appears to be the most widely used.
Malstrom describes MTM as follows:
It assigns times for the performance of elemental motions that a re completed
by the hands, arms, and other parts of the human body. Times required to
complete different types of motions are assigned specifie values. Ti me values
are expressed intime measurement units (tmu). One tmu is assigned the value
of hr. or 0.036 sec. [10, pp. 34-35]
lt should be noted that extensive training is required to use MTM with
competence.
5.4.3.5 Leaming Curves
For manufacturing in which a large amount of direct tabor is involved,
the average tabor time required to produce a unit is typically found to de-
crease over time . As a general rule, each time the cumulative production
doubles, the total time required per unit is reduced by 4%. In this case when
unit time is plotted against cumulative production, the plot is known as a
96% learning curve. An 80% learning curve would show a 20% reduction
' 1
- ,"{j
1
114
Estimating Markets, Revenues, and Costs
Chap. 5
(i.e . . rate of improvement) in cumulative average labor ti me wh en produc-
tion doubles. This simple mode! is an exponential function [8]:
Y;= y,;-h
where Y1 = cumulative average direct labor hours or cost through the ith
production unit (thus Y10 would represent the cumulative aver-
age ti me required for the first ten units, while Y
1
would represent
that for the first unit produced)
i = cumulative production count
b = the learning curve exponent
To solve for b, suppose that as cumulative production doubles from 20 to 40,
there is a 20% reduction in cumulative average Jabor hours per unit . Then
0.80 = Y4o = Y,(40) - h = 2-h
Y2o Y,(20) - h
Thus. -h log 2 = log(0.8), or -h = log(0.8)/log 2. This can be solved to find
th at h = 0.3219. Fin ally, the equation Y; = Y
1
;-
0

3219
represents an 80%
learning curve.
In general, the learning curve exponent, b, can be easily computed:
b = log of percent learning/log 2. For example, b = 0. 1520 for 90% learning
and b = 0.2345 for 85% learning. The following example shows the use of
the learning curve.
Example 5-5
One thousand units of a new product are to be produced. Direct labor has been
cstimated to average 8 hours per unit. Six months after production has begun, man-
agement has requested the following information:
1. Average labor hours per unit, to date
2. Labor hours per unit, for the latest month' s units
3. The % learning curve thal production has followed . The cumulative labor
hours and units produced are listed in Table 5-6. Dividing the cumulative labor
hours by the cumulative production shows that the average labor hours per
unit to date is equal to 15.75 labor hours . For the latest month's production, an
average of 410/42 = 9.76 labor hours per unit was reached.
In examining the data, it can be seen that the cumulative production at the end
of mon th 6 ( 132 units) is double the cumulative production at the end of mon th 4 (66
units) . The learning curve can be approximated by dividing the cumul ative average
labor hours pcr unit during month 6 by the cumulati ve average labor hours pcr
unit during month 4. The lcarning curvc in effect is approxi motely 75%, as shown
low.
1' .. 1 t hli )\)1' li t HII
ll h,hn, 1111111

............... ...
-a <
' f-
.,.., <
"Cl
:0 z
"' 0
....
:>
0

0..
" ...
. _g '
j ;,.;.-
E ..;, 5 Il
::l > 0
u..::::r:
... -
0 -
.J::J c
..c;ro::l
;:..J-;;,
0 . ...
.... Ol) ::l
""' > 0
" >
<::r:


"' u
"' 1-o V'l
- 0 ...
c .0 ::l
0 "' 0
::E..J::r:;
" c
,g
'(;


c
"' 0
..Ov-.\Or-"''f'V"'
,..... r-- ,..... 0\ \1"'1 r---

NNNN--
-.o-c-.oor-\GI ,......,..... __ oor-

N t""l---

--- N
-or-..oVo
ONOV)OO-
rl"''f'rlf"">N"''f'

=s r; 1 - r- ..,a f"'l '1:7" N
c: ::::s --- t'l ,...,....,.
0'1j
::E 0

..c:
- 1 - N ..0
::E
, 1
i
:i
,.,

i
.. !
'
J
h(
j i
' i:
:(;
r;
1!
!
,,
,,
i1
ji
lj
1
.j
1
i
j ,
i'
1
1:
1
:
j:
1
1
l

i:
t:
1
11 6 Estimatin11 Markets , Revenues, and C:os ts Clwp. 5
To determine the equation for this leaming curve, the parameter b must be
calculated. This can be accomplished by applying equation 5-25 and solving for b, as
shown below.
Reducing and solving, we find :
Y,( 132) - h
Y
1
(66) - b = 0.75
2- b = 0.75
-b log 2 = log 0.75
b = 0.415
Th us the equation is Y; = Y, ; -o.l5
5.4.4 Capital Cost Analysis
Frequently, the larges! cost of a project is attributable to new equip-
ment, buildings, and in sorne cases, entire production plants. Severa! meth-
ods for estimating capital costs are discussed below.
5.4.4.1 Analogy Costing
Normally, the information necessary for a detailed estima te of capital
costs is not readily available du ring the initial stages of a project. Therefore,
the first estimate is usually fairly crude and uses analogy costing. Analogy
costing is based on similarity among production plants with known capital
costs. Equipment for which a cost estimate is desired is located at an exist-
ing installation, and by analogy an estimate is prepared for the new equip-
ment. ln this situation differences in cost from the existing to the new equip-
ment would typically arise because of cost inflation and/or productivity
improvements . However, these two influences on cost have a tendency to
counteract each other in such a way that two projects, whose installations
are separated by only a few years, may have estirrated costs that are approx-
imately the same. given that the existing plant was not grossly over- or
underdesigned. Other factors must be considered when the same plant is to
be built in a different location, such as different delivery charges, different
site preparations, and different costs of supplementary equipment and facili-
ties to support operations [4] .
5.4.4.2 Exponential Costing
Exponential cos ting may be used when the proposed plant has a differ-
ent production capacity than the existing plant. According to de la Mare,
" the principle of exponential cos ting states that for many real-world produc-
tion systems proportionate increases in production capacity can be achieved
by Jess than proportionate increases in capital cost. This principle is a spe-
.>tC. 5.4 Cost Estimati1111 : Perspective a11 d Approaclr es 11 ,
cial manifestation of the law of increasing returns to scale, and is known as
the law of increasing technical returns to scale" [4, p. 151].
The following general equation represents most types of equipment:
c. = (Q)Il (5-26)
cb Qb
where Ca = capital cost of the proposed facility
Cb = known capital cost of an existing facility
Q. = production capacity of the proposed facility
Qb = production capacity of the existing plant
f3 = cost exponent factor, which can range from 0.4 to greater than
1.00, but is usually in the range 0.5 to 0.8
Table 5-7 provides typical cost exponent factors for selected types of indus-
trial equipment. Equation 5-27 permits one to obtain an estimate of the
capital cost of a proposed project by including factors to adjust for the
effects of inflation and productivity differences as follows :
(
Qu)(J
Ca = Cb Qb 1,1,.
(5-27)
where 1, = index of cost inflation
lp = index of productivity improvement
The accuracy of the exponential costing method depends largely on the
similarity between the two projects, and the accuracy of the cost exponent
factor {3. Generally, error ranges from 10 to 30% of the actual final cost .
A certain steam-generating boiler in the utility plant of a manufacturing complex
produces 50,000 lb/hr of saturated steam. This boiler was purchased for $250,000
eight years ago. If the priee index for this type of boil er has increased at an average
rate of 12% per year for the past 8 years, and productivity improvements have
averaged 4% per year du ring this lime, how much would a 150,000-lb/hr boiler cost
now? The cost exponent factor for this boiler is 0.50.
Utilizing Eq. 5-27,
(
Q.)Il
c. = cb Qb 1,1p
with /
1
= (1 + 0. 12)
8
and lp = (1 - 0.04)
8
, the cost of the proposed boiter can be
estimated as follows:
c. == $250,000 [ ( ~ ; ~ , ; : ) 0.
5
( 1. 12)
8
(0. 96)8]
= $250,000[(3.00)
0

1
(2.476)(0.7214)]
= $773,441
~ . . . . . ! : . : )
~ ~
. ~
~
t
~
4
$
IIH
r--cxt

"' "'

1- <
t.L.
....

z

w
!;;
0
u
...!
<
u
0::
>-
1-

"' E
Q.
.;:;
cr
w
c;;

::>
""' .5

u
c
"' Q


Q.
:;
3"
..
= :;;;
c
"'
:J;=
c;;
c
"'
'"
:::2

"' E
Q.
:;
cr
w
Ol
'
::>
""' .5
"'
"'
"'
2
0..
"' Q()
cxtl ": "'"' "f ""!"' ""! "f ooo..,oo o,.....
E

CXl.
E

0 d
- V)
0 u ...
Vl 0 V) t'
-gtl ot-
ct.E:S
Eu -u..ocouV)
Vl 0
u c t":S:!:! .s
.... = e c "' u :; Cil 1-
<<uCwciia.l
00 (7-.
oo,....Oo-.-.:t"O
1 1
00\C)OO,...._
0 d
t;
c
:.2 _g
_.,:.
E "'
00
0UU'-"'

gg i: i; t 2:
t":S000-0
a.lUUUW:I:
V)r<"''r-0\0\0\
o-=ooooV)

000000
E


:;
"' 01)
01) c
"' c "' (1) .... ....
u 0 ..= u


.-:: c "' "' - E c

N"

c
0
""'
c
j
ci
1-
....l
Ol
c
.s
'"
u
::>
-o
w
::::1

"' u


"'
"

.,
!; ..:
" "' - .=
., "'
"'"" "'>J:J
""::>
"' Q.
.s
"'-

::s c
"' 0 t:l v;

. E
"' ...
... "'
"' Q.
:::2"'
.D

"' "' u
""' ::>

fr

.
""' '-"'
::s-
" .
"" Q.
Sec. 5.4 Cost Estimating: Perspective and Approaches 119
5.4.5 Materials Analysis
Depending on the product, materials can be a very large, and therefore
important, part of a cost estimate. ln many industries it is common for
materials costs to comprise 50% or more of total product cost. Ma teri al cost
analysis involves making and using a bill of materials, deciding whether to
make or purchase parts, measuring the amou nt of raw material that goes into
each unit , and finding the costs of parts and raw materials.
5.4.5.1 Bill of Materials
The bill of mate rials is a complete listing of ali parts (mate rials, su bas
semblies, purchased items, etc.) going into the final prodpct. lt also lists the
part number, the quantity required to produce one unit, the material specifi-
cations , whether the part is purchased or made, and perhaps other products
in which the parts are used as weil as a cross-reference to similar parts. This
information is vital to the cost estimator so that an accurate estimate may be
made.
. 5.4.5.2 Decision to Purchase or Make
In most instances the decision to purchase or make parts is determined
by considering the incrementai costs of each option. That is, parts will be
purchased rather than made when the purchase priee is Jess than the incre-
mentalcost to manufacture. There are situations, though, where it is desir-
able to select the more expensive option . Suppose that a part can be pur-
chased for Jess than the cost of in-house manufacture, but the vendor cannot
guarantee delivery when the parts are needed. ln this case the part probably
should be made in-house to avoid the possibility of missing the due date.
Furthermore, sometimes a part may not be readily available for purchase
without risking the Joss of competitive information .
5.4.5.3 Estlmates of Raw Materials and Purchased
Parts Costs
Once the estimator has determined which parts are to be purchased
and which are to be manufactured, the next step is to obtain cost estimates
for both raw materials and purchased parts . According to Malstrom:
In estimating material costs, severa! factors about the part to be purchased
should interest the cost esti mator. Exampl es of factors to be considered in-
elude the quantit y to be purchased, whether or not the part has been purchascd
beforc, and the lcnsth of tin1C requircd to obt11ln 11n occumtc priee quOllt lon.
(10, p. SS I
l lll) (l fC IH ll y 11111 111( II,U,' 'IIIL
11 n oll on ktpltlll ll l OIIIJ'IIIt' l ill (l litiiiiiii' Y
'J , .
:;
i
1
120
Dale
3/84
9/85
8/86
12/87
Estimatinf( Markets. Revenues, and Costs
Table 5-8
HISTORICAL DATA
Order Quantity Unit Cost
250 1.65
1.000 1.25
160 1.!!7
730 1.72
are easily accessible. Table 5-8 provides a list of historical data. First, the
estimator picks two order quantities that bracket the order quantity de-
si red . Suppose that the desired order quantity is 424; he or she must then
choose the quantities of 250 and 730. If today's date is 6/88 and the average
inflation for this material has been 1% per mon th du ring the last 4 years, the
unit cost today of an order size of 250 would be*
V25o = $1.65(1 + o.ol)
11
= $2.74
For a lot size of 730, the unit cost today would be
u730 = $1.72(1 + o.Ol)
6
= $1.83
Interpolation can then be used to obtain a unit cost for an order quantity of
424 unit s. If we assume thal the relationship is linear, it can be solved as
(
730 - 424)
u424 = $1.83 +
730
_
250
($2.74- $1.83) = $2.41
There are severa! problems with this method. For example, inflation
rates do not usually remain constant. As a result, this method is recom-
mended only for parts with a low unit cost.
When the unit cost of a part is high, it is normal to obtain at least one
priee quotation even if the part has been purchased before. A priee quote
from a manufacturer's catalog is acceptable if it is up to date. "Substitution
is acceptable as long as the priee of the part may be obtained from a catalog
in a period of time equal to or Jess than that required to obtain a quote" [10,
p. 56]. If the dollar value of the order is Jess than the cost to ob tain the
estimate, the estimator may make an educated guess, use a manufacturer's
catalog, or use historical data to obtain a reasonable estimate. In instances
where a priee is not readily available, a part number match may be used as a
basis for estimating. The estimator should also be aware that priee breaks
may exist in the purchase of parts. If the company is willing to purchase
sufficient quantity of a part at one ti me (perhaps by combining requirements
See Section 5.5, Eq. 5-37: (1 + 0.01)" = (F/ P, 1%, 51), which merely adjusts "RS"
into "A$" by compounding inflation at 1% per mon th for the 51 months from 3/84 to 6/88.
.'-: '..., ----- - - ---- __ _ .._
Sec. 5.4 Cost Estimatinf( : Perspective and Approaclzes 121
for numerous products), there could be a significant difference in the per unit
cost.
5.4.6 Other Costa
The estimator should recognize that costs for working capital, setups,
run cycles, tooling, and factory burden need to be considered when relevant.
5.4.6.1 Working Capital
Working capital is the moncy required to transform a new capital
project into an operating process, excluding fixed assets (such as equipment,
machinery, or buildings). A common accounting definition is:
working capital = current assets - current liabilities (5-28)
Table 5-9 shows an example of current as sets and liabilities for a company.
Using equation 5-28, working capital= $4.35 x 10
6
- $1 .2 x 10
6
= $3 . 15 x
W. Although working capital is constantly being liquidated and regen-
erated, it is not available for other uses, so it must be regarded as an in-
vestment.
5.4.6.2 Setup Costs
Setup is the work required to prepare the machine , process, or bench
for operation. Normally, the setup ti me can be fou nd by using standard
data, but a study of the elements necessary to perform the setup may be
required. These elements may include operator punching in or out, obtain-
Table 5-9
A COMPANY's ASSETS AND LIABILITIES
Fixed assets
Current assets
Cas h
ln ven tory
Accounts receivable
Current liabilities
Accounts payable
Short-lerm debt
Working capi tal
Total assets employed
$ 50,000
3,300,000
1,000,000
$4,350,000
$ 700,000
500,000
$1,200,000
$ 8,000, 000
$ 3,150,000
$11 ' 150,000
Source: R. F. de la Mare , Manufacturing Sy.rtem.r Eco-
nomie.<, Cassell Educational LTD.. London. 1982,
p. 167. Reproduced by permission of the publisher.
1

!(
1;
1
''
,1
:f
t
\
1
;
,.
! ' .
1'
122 Estimating Markets, Revenues, and Costs
ing tools, doing paperwork, making machine adjustments, and positioning
necessary materials. Also included in the setup are tasks performed after a
run cycle is completed-such as teardown, returning tooling, and cleaning
workstation. Normally, the setup cost is handled as an overhead charge for
mass production [Il].
' 5.4.6.3 Cycle Time Costs
Cycle time is the time required to produce 1 unit of a product after
setup has been completed. Depending on the process, there are many dif-
ferent methods of calculating the cycle time of a machine. As an example,
the steps in a metal cutting operation are ( 1) load work, (2) ad vance tool, (3)
retract tool, and (4) unload work.
The time to load and unload the workpiece is called the handling lime
and may include the time to ad vance and retract the tool from the piece. The
handling cos/ may be found by
handling cost = Coth
where C
0
= direct labor cost, dollars/min
th = time for handling, min
See Fig. 5-7a as an example of handling cost plotted against cutting speed.
To find the time to machine 1 unit, the following equation is used [10,
p. 270) :
L LrrD
tm = JN = 12 Vf
where tm = machining time, min
L = length of eut for metal cutting, in
f = feed rate, in/rev
N
. d 12 v 1 .
= rotary cuttmg spee = rrD , rev mm
D = diameter, in
V = cutting speed, ft/min
machining cost = Colm
The unit cost decreases as cutting velocity increases, as shown in Fig.
5-7b. Note that the diameter, D, may be for either the tool or the workpiece,
depending on the type of machine. The diameter of the tool is used for
milling and drilling; in lathe turning, the diameter of the workpi ece is used.
5. 4.6.4 Costs
' utti l1fl, 11111y hilC(lll l uull, 1\ nd thc r c f () lt' 11\IINI h ll 0
l i' PhH' tl d ll \i' t ol of' Jlplll l.' l ll l-lll l n 111 11(1 luiJHHiillll l11 Il
tlt' l ll' ld l nll 11 11 lll t'll ll llll llll il l llthlll 1111d l l ti ii' I I'!JII IInl Atillld
Sec. 5.4 Cost Estimating: Perspective and Approaches
c:
::1
..
o.

Tool changing
Cutt ing speed
le)
:
::1
8.
il
u
Cutting speed
(d)
Figure 57. Graphie costs for four parts of machine turning economies . Source:
Ostwald, Phillip F. , Cosr Esrimaring, Second Edition . Eng)ewood Cliffs. NJ :
Prentice-Hall , lnc ., 1984, p. 271. Reproduced by permission of the publisher.
123
ing to Ostwald [Il], the aue rage /ife of a cutting tool may be fou nd by the
T = K""
v
where T = average tool !ife, minutes per cutting edge
(5-32)
n, K = empirical constants resulting from regression analysis and field
studies 0 s n ::5 1, K ;:::: 0
When V = 200ft/min and the tool -life equation is VT
0

16
= 400, then 76
min of machining may be done before the tool must be indexed to a new
The tool-changing cost per operation may be found by
tool-changing cost = Co te
(5 -33)
.
Il
1
,.

i
,\
124 Esrimaring Markets , Revenues , and Cosrs Chap. 5
Figure 5-7c shows the relationship of tool-changing cost to cutting speed.
The cast of cutting tools must also be calculated:
tool cost per operation = C
1
(t;) (5-34)
where C, denotes the tool cost in dollars and T is the average tool !ife in
minutes . Ostwald observes that "for insert tooling, tool cost is a function of
the insert priee, and the number of cutting edges per insert. For regrindable
tooling, the tool cost is a function of original priee, total number of cutting
edges in the li fe of the tool , and the cost to grind per edge. " As shown in
Figure 5-7d, the tool cos! increases as cutting speed increases [Il , p. 272].
5.4.6.5 Factory Burden
Ali costs that cannot, conveniently and economically, be allocated to a
particular product are called factory burden or ouerhead. Burden is typi-
cally divided into two categories: fixed and variable. Fixed burden consists
of costs that do not change with production volume, such as salaries , mort-
gage payments, rent, and insurance. Variable burden increases or decreases
as production volume rises or falls . Examples include electricity to operate
equipment. indirect materials and labor , and certain tooling. Often, compa-
nies allocate burden to different cost centers and/or products in proportion
to sorne activit y basis such as direct labor hours or prime (tabor + materials)
cost.
5.4.7 Engineering Costa
When the product to be produced is similar to an existing product,
engineering costs can be estimated by using historical records. The estimate
must include costs of basic engineering services such as product design,
specifications, and bill of materi als. as weil as maintenance engineering for
the li fe of the product. The costs are made up of wages, housing, and utili-
ties. In addition, costs for projects that did not materialize into final prod-
uct s must be considered in the wage costs [Il) . Spreadsheets su ch as th at
shown in Fig. 5-8 may be used in figuring the cost of tabor for engineers.
When representing engineering rate per unit , C,, the following equation may
be used:
C = total engineering cost
projected production quantity
(5-35)
where the " production quantity" can be in any relevant units , which are
often " hours worked ."
Sometimes factors are used to multiply by expected salaries to provide
an estimate. They can range from 1.8 to 3.0 and depend on the complexity,
Sec. 5.4 Cosr Esrimaring: Perspective and Apprnaches 125
Design engineering estimate for Speed 1t educe1t producn
Description
6olt te6.t w eed
Customer
No.t
lnquiry or quota no. Date
Based on quantity of
480 u>t.U&
Ouri ng period of
Type of labor Hours
Rate Extended
per hour la bor
Suei'I..I>.t ,
de.g n 40 $33 . 00 $1. 32 0
Engwee.lt , 800 $30 . 00 $24,000
e.lec..tM.ca.t
a.i.M 160 $IL 75 $3. 000
Tech. Wlt.i..telt
1
80 $1 7 . 00 $1,360

Mode.l 6hop
To.ta.t labolt 1080 $29 ,680
- --- - - ----
Figure S-8. Estimating for engineering cos ts. Source: Ostwald , Phill ip F., Cmt
Esrimating, Second Editi on. Englewood Cliffs. NJ : Prentice-Hall , lnc. , 1984,
p. 318. Reproduced by permission of the publi sher .
1
novelty, and sec recy of the work. Other factors, su ch as drawing size, are
a!so being used to estimate engineering design costs .
5.4.8 Comprehensive Cost Estimating Example
The XYZ company has decided to btlild a new plant that is very similar
to an existing plant. The plant will be used to produce 5,000 units of the part
shown in Fig. 5-9. The loss due to scrap is expected to be 2.5%. The com-
pany that sells the raw materials to the XYZ company will buy back any
scrap or waste for $0.95 per cubic inch . The recovery cos! is $0.20 per cubic
inch. Approximately 15% of the waste is not recovered for resale. Histori-
l.
1
0:
126
Estimating Markets, Revenues, and Costs
0.25-in. dia.
T
"" I
0.66-in. dia. - +-.,--------,------'------.
0.25-in.
Figure 5-9. Sample part .
cal data show that on January 1, 1987, the cost of this material was $1.45 per
cu bic inch, in a quantity of 8,000 inches
3
On March 1, 1988, the unit cost of
3,000 inches
3
was $1.89 per cubic inch . For the past 4 years we will assume
that inflation rates have remained stable at an annual rate of 12% and that the
date wh en the material is to be purchased is February 1, 1990.
The approximate capital cost of the existing facility which was pur-
chased recently was $12,000, and its production capacity is 20 units per
day. It is desirable for the future facility to have a production capacity of 30
units per day. Capital cost can be estimated using Eq. 5-27 with /, = 1.12,
lp = 1. 10, and (3 = 2/3. Studies were do ne at the existing facility showing thal
approximately 0.35 hr is necessary to produce each unit. This time was
achieved over a production of 3,500 units and production has followed an
85% Iearning curve. Labor costs $15 per hour, including benefits. The parts
are produced in lot quantities of 25 and the machine setup time is 1 hr. Part
of the setup involves changing the cutting tool. The cost of the tool is $20.
Factory burden is 40% of direct material. Estimate the manufacturing cost
from the information given .
Solution
The actual volume of material required per part is calculated to be
1.5375 inches-
1
. The total material necessary to produce 5,000 units is there-
fore (1.5375 inches-
1
)(5,000) = 7,688 inches
3
. The amount of m'aterial avail-
able to sell back consists of both scrap and waste. The amount of scrap per
parti s ( 1.5375 inches
3
)(0.025) = 0.0384 inches
3
The amount ofwaste (from
dri ll ing holes) is
( 1 -

in thick) = o. l 663 inl


Sec. 5.4 Cost Estimating: Perspective and Approaches .127
From the historical data we can find the cost of the raw material if it is
purchased on February 1, 1990. *
(
0.12)
23
U3,000 = $1.89 1 + 12"" = $2.38
(
0; 12)
37
Us,ooo = $1.45 1 + 12"" = $2.10
By interpolation, the cost of a unit of raw material is $2. 12 per cu bic inch.
Therefore, the total cost of materials is
($2. 12/in
3
)(7 ,688 in
3
) - ($0.1535)(5,000) = $15,531
The capital cost of the plant may be approximated by using Eq. 5-27 as:
(
30)0.6667
CA= $12,000 20 (1.12)(1.10) = $19,373
The average labor hours over the production of ali 5,000 units may be
found by use of the Iearning curve and Equation 5-24:
b = log(0.85)/log 2 = 0.2345
So
Y; = y
1
; - o.2345
For Y1 = 2.03, we find Y;= (2.03)(5,000)-
0 2345
= 0.275
The labor cost of producing 5,000 units is
(0.275 hr/unit)(5,000 units)($15/hr) = $20,625
The setup cost may be found by
(5,000/25)(1 hr)($15/hr) = $3,000
The cutting tool cost is
(5 ,000/25)($20) = $4,000
Factory burden is 40% of direct material , and thus is
(0.40)($2.12)(7 ,688) = $6,519
From the above costs, the total manufacturing cost, Cr, may be calculated
as:
Mat criai
$1
Producti on
la bor
1 $20,
Setup
$1,000
Cutting
tools
Factory
burdcn = '-T
$4,000 + $6,5 19 - $69.048
\ , , , tlq i \/ tnt UIIillllll olllloiM 1111111111
-

1
l'
( .
\
i ;
1
,, .
... ... '
128
Estimating Markets, Revenues, and Costs
Chap. 5
5.5 COST ESTIMATING RELATIONSHIPS AND INFLATION
Except for Sections 5.4.5.3 and 5.4.8 we have assumed that prices.up to this
point will remain relatively unchanged over si.Jbstantial periods of time, or
that the effect of such changes is the same on any alternatives considered.
U nfortunatly. the se are not realistic assumptions in general. Inflation can,
indeed, affect the economie comparison of alternatives. Hence the nature of
inflation and methods of taking inflation into account will be considered
below.
5.5. 1 Actual Dollars versus Real Dollars
Inflation describes the situation in which priees of fixed amounts of
goods and services are increasing. As priees ri se, the value of mo ney, that
is, its purchasing power (in real dollars , as defined below), decreases corre-
spondingly.
Let us defi ne two distinct kinds of dollars (or other monetary units such
as pesos or ru bles) with which we can work in economie analyses, if done
properly:
1. Ac tuai dollars: the actual nu rn ber of dollars as of the point in ti me they
occur and the usual kind of dollar terms in which people think. Sorne-
times called then-current dollars , or current dollars , or even infiated
dollars, they will be denoted as "A$" whenever a distinction needs to
be made in this book.
2. Real dollars: dollars of purchasing power as of sorne base point in
time, regardless of the point in time the actual dollars occur. Sorne-
times called constant worth dollars, or constant dollars, or even unin-
fiated dollars, they will be denoted as "R$" whenever a distinction
needs to be made in this book. If the base point in time, k, needs to be
specified (it is usually the time of the study or the initial investment),
that can be shown with a superscript [i .e., R$Ck>].
Actual dollars at any time, n, can be converted into real dollars at time
n, of purchasing power as of any base time k , by
(
1 )"-k
= A$n l + f, = A$n(PIF,f%, n - k)"' (5-36)
Similarly,
A$n = R$fkl( J + f)n -k =

n - k)
(5-37)
where fis the average inflation rate per period over the n periods.
By ''inflation rate" is meant the average rate of inflation for the partic-
ular items over the period in question . Sorne of the most valuable sources of
lnterest factors and symbols are explained in chapter 6. Factors are tabled in
Appendix A.
Sec. 5.5 Cmt Estimating Relationships and Inflat ion 129
data for estima ting inflation rates (either general, such as cost of living, or for
specifie commodities and services) are cost indexes. Cost indexes are di-
mensionless nu rn bers th at provide a means for converting past costs to costs
at any later time, or vice versa.
There are many cost indexes and they cover almost every area of
interest. Sorne are based on national averages; others are very specialized
by type of item and perhaps locality. Indicative values for se veral frequent! y
used indexes are shown in Table 5-10. Probably the most commonly-used
measure of the general inflation rate is the Consumer Priee Index of the U .S.
Department of Labor . Note that "1-year change" (i.e. , percent increase or
decrease for each year) in Table 5-10 is also shown for two of the indexes .
Additionally, the compounded an nuai rate of inflation for the 8-year period is
shown at the bottom of the column for each index.
If we want to fi nd the cost of something at ti me n and know its cost as
of sorne (base) time k, then the relationship that can be used is
Cn ln
ck = "h
where Cn and Ck are costs and ln and h are the appropriate index values at
times n and k, respective! y. Solving gives
c. = ck (i)
(5-38)
Recognizing that Cn could be A$n, and Ck cou id be we can substitute
into Eq. 5-38 to find
A$n = i
Combining Eqs . 5-37 and 5-39, we find that
a;, ln
(F/P,f7o, n - k) = h
Example 5-7
(5-39)
(5-40)
Use the Consumer Priee Index in Table 5-10 to estimate the 1983 cost of products
that cost $1 million in 1980. Then find the annual compounded rate of inflation.
Solution
Using Eq. 5-38 yields
(
/83)
CsJ = Cso lw
$
. . (298.4) $ 209 . .
= ( 1 mtlhon)
246
.
8
= 1. mtlhon
\
'(--..>

l '
.
.
1
'
i
iJ
; r
v
'
i
,,
i
1
li
q


' ' : i
!
;
1
i
!
i
'
il
1;
/;:
1
0"'
- w
' )(
.,., w
..!! 0
.Cl z
.. -
1-f-
"' 0
u
w
..J
...

<
)(
[JJ


0
'"
0::
8
"


..
.
0..
z
0
><
"
" .::
"
<.)
c
0..

" E

c
0
u
"
00
c
"'



>;-
8
- "'
Il
N C
,_-
a-

c
"'


"'
"
>:'
8
- ><
Il



u 0.. "'
E u
ro:O-r-c
"'"'--r--
ou r.n -Cl\
> 0 0-
<u
"' u
c .. :
v;
0 ::J
-6 8
0 a..--"'

..... :::J "
0 .. - r- c
V)- r--
r.l'l .... :l 0\

00
>o 0 "'

z
5
>-
..........,..

O..:v-\v\r-o:oOO..:O..:..Or-\
0\
v\NOOMov\c---:v\

- - - - - - - C'l N
-ooV"lr-f"f""Vl<ot:t-N
a\v-l...or..:....:t'f"\o..or"")
NV""'V)"'=tVOO"'=t-"':f' r-
....:c:::)_:v-)r--:...0No\O oci
-.cr-ooQ\-""':fr-ooa-..
----NNNNN
00\0"'=t'V'lr--V""'<ot:t...,.
.
---NNNN
o-8ooNOI""'V"'\O o
o\


g ti
'.:J-


"' y


"''?>J:
?.
"'
1:!
"
.5
"
c
:::J
&.
E
0
<.)
E
e
(J.
....;
0
....;
0
c:i
00

0.0


c(
s
"' .; :::1
- 0
0
.9 ]
<
Vi (,) 0
- - (J.
.. 5 . E
E g .-:.
"'0 8
;: < ;: UJ





o>.o""
=' 0.0 ::l L.o
1 .... 1 0

:::J [JJ :::J
Q_Q<
>o
..... OIJ '- 0
0 ,_ 0 -
.Cl " .c> E
cu c "" 0
..J [JJ ..J c
'-'-'-0
0 0 0 <.)
- - - [JJ
c c c '-
4.) u u 0
E E E :::
la la la g
a. o. o. ::J
0 u u 0
0 0 Cl u
vivi vivi
::i ::i ::i ::i

oo.o
....
.... "
';j'g.

.,-,u
=
I'O
"'..,
00 .Cl
a- 't:


c(<

- .,
0 c
,. K.
V'\ <+
$<
Cost Estimating Relationships and Inflation 131
Using Eq. 5-39, we obtain
298.4
(FIP,f%, 83 - 80) =
246
_
8
= 1.209
From compound iriterest tables, (FIP, 8%, 3) = 1.2.597 and (FIP, 6%, 3) = 1.1910 .
lnterpolating, the inflation rate for that 3-year periO<I is approximately 6 . .5% com-
pounded annually. It might be noted that the 6.5% is close to, but is not the same as,
the average of the 1-year changes shown in Table 5-10 .
5.5.2 Real lnterest Rate, Combined lnterest Rate,
and Inflation Rate
Let us define severa! types of rates and show how they are used:
1. Real interest rate: Increase in real purchasing power expressed as a
percent per period, or the interest rate at which R$ outflow is equiva-
lent to R$ inflow. lt is sometimes known as real monetary rate or
uninfiated rate and denoted as i, when it needs to be distinguished from
ic (below) .
2. Combined interest rate: Increase in dollar amount to cover real inter-
est and inflation expressed as a percent per period, and is the interest
rate at which A$ outflow is equivalent to A$ inflow. It is sometimes
known as actual rate or infiated rate and is denoted as ic whenever it
needs to be distinguished from i, (above).
3. Inflation rate: As defined previously, the increase in priee of given
goods or services as a percent per period. It is denoted asf.
Because the real interest rate and the inflation rate have a multiplica-
tive or compounding efTect,
ic = (1 + i,)(l + f) - 1
ic = i, + J + (i, X f)
. ic- f
1
' = 1 + f
Where fis not large relative to the accuracy desired, th en
ic !! i, + f
nd
(5-41)
(5-42)
(5-43)
(5-44)
1
,,
!
,
1

132 Estimating Markets , Revenues, and Costs
5.5.3 What lnterest Rate to Use in
Economy Studies
In general, the interest rate that is appropriate for time-value calcula-
lions in economy studies depends on the type of cash flow estimates as
follows :
Method
A
B
If Cash Flows Are
Estimated in Terms of:
Actual $, AS
Real$, RS
Then the lnterest Rate to Use ls:
Combined interest rate, i,
Real interest rate, i,
The above is made intuitively consistent if one thinks in terms of
method A as working with inflated (actual) dollars and interest , and method
8 being applicable to uninflated (real) dollars and interest. Method A is the
most natural to use because we usually think in terms of A$. Since interest
paid or earned is based on A$, it is a combined interest rate, ic.
method 8 is sometimes easier to use.
5.5.4 Summary of Formulas for Relating Single
Sum A$ and R$ over Time
A$n N S . .
(= A$) ote: needed only to clarify
base lime.
k Time scale n
Base time for ex pressing R$
Type of Dollars
Method or Conversion
A A$ (inflated $)
B R$ (uninflated $)
Moving Forward in Time
A$. = AS,(F/P, i " n - k)
= RS\"(F/ P, i,, n - k)
lnflating (at given time)
Moving Backward in Time
A$,= A$.(PIF, i " n - k)
RSI" =

i., n - k)
Detlating (at given time)
From R$ to A$, or AS. = n - k) = AS.(PIF.J, n - k)
from AS to R$ ,
at a given time
Example 5-8
A certain expense at the end of 1991 is estimated to be $10,000. The end of 1991 is
the base point for considering inflation. (Thus $10,000 = =

Find its
equivalent worth in 1996 for the following circumstances (paralleling the formulas
above) . and using the following rates .
Co.ft Estimating Relationship.f and Inflation
real rate, i, = 4%
inflation rate, f = 6%
combined rate i, = (1 + 0.04)(1 + 0.06) -
= 0.1024 = 10.24%
"" JO%
A$96 = ic%, 96 - 91)
= $10,000(FIP, 10. 24%, 5) = $16,282
(Note: If 10% approximation were used, answer would be $16,105. )


=

i,%, 96 - 91)
= $10,000(FIP, 4%, 5) = $12,167
In 1996 dollars, beginning with 1996 equivalent in 1991 dollars :
A$96 = 96- 91)
= $12,167(F/P, 6%, 5) = $16,282
5.5.5 Manipulating Series Which Are Uniform
in R$
133
If cash flows expressed in R$ are uniform each year (and th us the A$' s
increase each year at the average rate of inflation, J), they can be conve-
niently converted to equivalent worth(s) at other point(s) in time using uni-
form series formulas at the firm's i, = real MARR.
Thus, if there is a uniform series in which each end-of-period pa y ment,
A, for n periods, is expressed in R$<kl (and th us is denoted A <k
1
), then
Phk> = (A<ki)(PIA. i" n) (5-46)
= (A<>)(FIA, i" n) (5-47)
The superscripts make clear the base point in time at which the dollars (like
present or future equivalents) are expressed. Normal! y k = 0, but estimates
can be converted to any base point in ti me at the inflation rate, f, using Eq.
5-36 or 5-37.
An equivalent formula for finding the present worth of a uniform series
that is escalating at the inflation rate f for n years is
(


1 ,
,
j
i
'
'
\!
1
i
1
'
1
1 '
1
Il\
134 Estimating Markets, Revenues, and Costs Chap. 5
p <o> _ A<O>(FIP,f, 1)[1 -(PIF, ic%, n)(FIP,f%, n)]
0
- ic - J
Note that A<
0
>( FIP,f, 1) = A(l). Thus
p<O> _ Am[l -(PIF, ic%, n)(FIP,f%, n)]
0
- ic - J
For the special case in which ic = f, so thal the real monetary rate = 0, Eqs.
5-48 and 5-49 become


= A<Ol(n)
and
P&
0
> = Am(PIF,f, l)(n)
Example 59
A person who is earning $21,600 salary for {assumed end of) year 1 expects that
saJary to inflate (escalate) at 10%/year, which is the same as the general rate of
inflation. If her real monetary rate, i,. is 5%, then i, = 0.05 + 0. 10 + 0.05{0. 10)"'
0. 155 , or 15.5%. Find the present worth, P&
01
for 3 years of salary.
(a) Use the approach of Eq. 5-46 to find P&
11
and then convert to Pb
0
>.
(b) Use Eq. 5-49.
Solution
(a) P&
11
= Alll(PIA, i,. 3)
= $21,600(PIA, 5%, 3) = $58,821
Pb
01
= Fb
1
J(PIF,J, 1)
= $58,821 (Pl F, 10%, 1) = $53,474
(b) p<o _ $21,600[1 -(PIF, ic, 3)(FIP,f, 3)]
0
- i,- f
$21,600[1 -(PIF, 15.5%, 3)(FIP, 10%, 3)]
0. 155- 0.10
= $53,496 [same as for part {a) except for round-off error)
5.5.6 Manipulating Series That lnflate (Escalate)
at Rate Different from General Inflation
Whcn a cas h flow series expressed in A$ infl ates or escalates at a rate
different from general infl ati on. it wi ll not be a uniform seri es when cx-
nrcsllcd in R$. The use of whot might be cnll cd "differenti ai oscolation
ITI ICII" Cll n he Il hnmly C(l!)lPIII OI () tllli COII VCfl ic nCC in ll ti Ch Il Cll llC, hui WC wil l
nol ll ln Ntt ltll' llt t nl h
Sec. 5.6 Summary and Prologue 135
eral ways to convert a series of cash ftows subject to whatever rate of
inflation (which might vary from year to year) into either R$ or A$, and then
to manipulate them (in this case, into present worths) at the general inflation
rate.
Given the same individual salary situation as in Example 5-9, the only difference
being that the salary will inflate (escalate) at 8%1year, which differs from the general
inflation rate of 10%. To repeat, the end-of-tirst-year salary A$
1
= R$\n = $21,600,
i, = 5%./ = 10%, and ic = 15.5%.
(a) Show her salary for 3 years expressed in R$ and in A$.
(b) Show the present worth (as of beginning of the tirst year) of both ways
of expressing this salary.
ln A$. :
Year n
2
3
ln


Year 11
1
2
3
( 1)
A$. = A$1(FI P, 8%, 11 - Il
S21,600(FIP, 8%, 0) = 21,600
21,600(FI P, 8%, 1) = 23,328
21,600(FIP, 8%, 2) = 25,194


= AS.(PI F, 10%, 11)
S21,600(PIF, 10%, IJ = 19,637
23 ,328(PIF, 10%, 2) = 19,278
25 ,194(PIF, 10%, 3) = 18,928
(2)
(PI F, 15.5%, n )
0.8658
0.7495
0.6489
(PIF, 5%, 11)
0.9524
0.9070
0.8638
(3) = (1) x (2)

$18,701
17,485
16,348
$52,534

$18,702
17,485
16,350
$52,537
Note in the lower column {1) that even though her A$ salary is going up (at 8%
per year) , the R$ (purchasing power) of that salary is going down {approximately
10%- 8% = 2% per year). The present worths of both ways of ex pressing the sa lary
are the same, except for minor round-oiT error.
5.6 SUMMARY AND PROLOGUE
ln 1hls chnpt cr w
dMIIJWCCiil'lllr y f'or
usscd toni es el u l eu lo ocvelonmcnt of
lhc lllld C(lSIS \l fii (IV!IO{' I: d OHIIHtfhl' lU
1

d
'.
'




l

l
136 Estimating Markets , Revenues, and Costs Chap. 5
ing systems . Specifically, we addressed forecasting the market for new/
modified products; estimating revenues and costs, including consideration of
learning curves; and dealing with inflation in the estimating process .
Once monetary benefits and costs of investing in new and innovative
technology are quantified , they need to be combined into economie
measure(s) of merit. This is the important subject of Chapter 6 (before-tax
analyses) and Chapter 7 (after-tax analyses, including replacement
studies) .
REFERENCES
1. Chambers. J . C., S. K. Mullick, and D. D. Smith, "How to Choose the Right
Forecasting Technique," Harvard Business Review, July-August 1971, pp.
45-74.
2. Corr, A. V., "The Role of Cost in Pricing," Management Accounting, Novem-
ber 1974.
3. DeGarmo, E. P. , W. G. Sullivan, and J. R. Canada, Engineering Economy,
7 ed., Macmillan Publishing Company, New York, 1984.
4. de la Mare, R. F., Manufacturing Systems Economies, Holt, Rinehart and Win-
ston, New York, 1982.
5. Draper. N. R .. and H. Smith, App/ied Regression Analysis, John Wiley & Sons,
!ne .. New York, 1978.
6. Gilchrist, W., Statistical Forecasting, John Wiley & Sons, lnc., New York,
1976.
7. Jones, B. W. , Inflation in Engineering Economie Analysis, John Wiley & Sons,
!ne. , New York, 1982.
8. Kleinfeld, l. H., Engineering and Managerial Economies , Holt, Rinehart and
Winston, New York, 1986.
9. R. C. Lenz, Jr., Technologica/ Forecasting, ASD-TDR-63-414, Aeronautical
Systems Division, Air Force Systems Command, June 1962.
10. Malstrom, E. M. , What Every Engineer Shou/d Know about Manufacturing Cos/
Estimating, Marcel Dekker, !ne., New York, 1981.
Il. Ostwald, P. F., Cost Estimating, 2nd ed. , Prentice-Hall, Inc., Englewood Cliffs,
N.J., 1984.
12. Porter, M. E. , Competitive Strategy, Techniques for Analyzing Industries and
Competitors, The Free Press, New York, 1980.
13. Schweyer, H. E., Analytical Models for Managerial and Engineering Eco-
nomies, Reinhold Publishing Corporation, New York, 1964.
14. Sullivan, W. G . and W. W. Claycombe, Fundamentals of Forecasting, Reston
Publishing Co. , !ne., Reston, Va. , 1977.
15. Vernon, 1. R. (ed.), Realistic Cost Estimating for Manufacturing; Society of
Manufacturing Engineers. Dearborn, Mich., 1968.
Chap. 5 Exercises 137
16. Wheelwright, S. C., and S. Makridakis, Forecastinf( Methodsfor Management ,
2nd ed., John Wiley & Sons , lnc .. New York, 1977.
EXERCISES
5-1. List as many considerations as you can th at might affect the choice of forecast -
ing techniques by a local company familiar to you.
5-2. If a company manufactures a "faddish" li ne of sporting goods equipment, how
would its forecasting strategy differ from that of another company producing
aluminum beverage containers?
5-3. Explain how .the different stages of a product's !ife cycle will influence the
choice of a forecasting strategy.
S-4. (Section 5.2.3) Suppose that you own a small company that manufactures
metal castings for severa! large automotive companies. Over the past severa!
years you have found that quarter! y new-car sales tend to lag the prime interest
rate by 3 months . Y ou would like to make a forecast ofnext quarter' s car sales
so that the size of your work force can be anticipated. There is a direct rela-
tionship between car sales and demand for castings that your company pro-
duces. The following data are gathered:
lnterest Next
Rate Quarter
Year Quarter (%) Sales ($M)
1 8.00 $23
2 8.25 17
3 8.50 18
4 8.25 20
2 1 7.75 21
2 7.25 25
3 7.70 24
4 7.25 29
3 1 7.50 24
2 7.75 23
3 7.25 26
4 7.00 30
ln te rest
Rate
Year Quarter (%)
4 1 7.00
2 7.50
3 7.50
4 8.25
5 1 8.75
2 8.50
3 7.50
4 7.00
6 7.50
Ne xl
Quarter
Sales ($M)
$25
26
17
20
15
18
22
23
(a) Calculate a linear regression equation for these data, assuming that the
interest rate is the independent variable.
(b) Calculate the correlation coefficient.
(c) Make a forecast of sales for the next quarter based on this quarter's prime
interest rate of 7.50%.
5-S. (Section 5. 1.3) Consider the following ti me-series data of demand for a certain
company's product.
\
~
:.:
\'
L
1
;.j
:tl :t
j
1 \Jj
ft
11
~ ' '"
138 Estimating Markets, Revenues, and Costs Chap. 5
Mon th Demand (Booked Orders) Mon th Demand (Booked Orders)
4
5
6
7
8
9
10
3,1109
2,641
2.934
3.239
3,490
2,569
3,205
2,561
3,047
2,607
Il
12
13
14
15
16
17
18
19
20
(a) Plot these data on a piece of graph paper.
3,387
3,138
2,908
3,512
3,291
2,804
3,096
3,106
3,195
3,605
(b) Apply single exponential smoothing to the data when a'= 0.20 and make a
forecast for T = 1 month into the future .
(c) Repeat part (b) when a' = 0.05 and T = 3 months.
5-6. (Section 5.2.4) Discuss the principal advantages and disadvantages of the
Delphi method of forecasting.
5-7. (Section 5.2.4) In your class, attempt torun a Delphi study to determine the
priee of a compact disk player 3 years from now. Was group consensus af.
fected by conducting two or three rounds of the procedure?
5-8. (Section 5.2.4) Try to think of sorne products that are presently in the carly
stages of a substitution curve effect. List them and try to estimate when the
newer product will take more than half the market.
5-9. (Section 5.2.4) How could trend extrapolation be used to forecast future inno-
vations in the aerospace industry? What performance characteristics do you
believe are important here?
5-10. (Section 5.3)
(a) Determine the priees for two products A and B, basing priee on a return on
investment of 15%.
(b) At the same priee as in part (a) , what is the markup based on (1) total
costs. (2) incrementai costs, and (3) value added in production?
Available data for a year's operation. with an effective income tax rate of 52%
and 10,000 annual units of each product , are as follows:
Initi al investment . 1

Mnterinls,
ircct, Cn
ther tl xed
1\!HII ,
Product
A
s 90,000
20, 000
30.000
50.000
1()(),00()
B
$120,000
40,000
I.S.OOO
15.000
10,000
Chap . 5 Exercises
139
5-11. (Section 5.4.2) Use the factor technique to estimate the cost of installing a local
area network in a factory environment having the following characteristics.
One large building on a single leve! will require a total of 3,000 ft of coaxial
(broadband) cable to network its six departments . Six network interface units
(NIUs) will be required and a total of 50 taps will have to be made to connect
ali the anticipated workstations and programmable deviees . Two modems are
needed in addition to one network manager/analyzer that costs $30,000. The
information necessary to make the estimate may be obtained from the work-
sheet shown. How accurate do you think such an estimate would be?
Component Cost Estimating Relationship
1. Interbuilding connections $100- 150 per foot x
2. lntrabuilding connections $20-50 per foot x
3. Cable installation $20 per foot x
4. Equipment
a. Broadband
CA TV amplifier $500- 1500 x
Taps $17- 20 each x
Splitters $5- 15
x __ = __
NI Us $500-$1 .000 per port
x __ = _ _
Modems $1,000 each
x __ = __
b. Baseband
NI Us $600 per port x
Repeaters SI ,200-1 ,500 each x
Taps/transceivers $200- 300 each
x __ = __
c. Network manager $10,000-30,000
Network analyzer $30,000
5-U. (Section 5.4.2) A residential builder just finished constructing a 3,000-ft 2 home
for $96,000. This cost did not include the lot or utility access fees . A detailed
breakout of costs for this job is as follows :
Item Fraction of Finished Cost
Lumber and carpentry
Electrical wiring
Plumbing
Concrete and masonry
Wall board
Flooring
Foundation preparation
and npplinnces
Heu ti ng and air contlilionlng
RoonnR
l'ul nt ln
M nllllriOUII N
0.20
0. 10
0. 14
0.09
0.04
0.06
0.02
0.0
O.llY
() , 1()
1) ,()/
tl 01
1 IMI
i 1
''
' (
1
'
1
j :
,.
\.
-
\
140 Estimating Markets, Revenues, and Costs
(a) What is the unit cost for the just-finished home?
(b) If a 4,000-ft
2
home is to be built, es ti mate the total cost from the answer to
part (a) and compare it with the total of estimated item costs based on the
breakout given above.
5-13. (Section 5.4.3) Develop the standard time to perform a manual assembly task
having the following elements and compute the standard number of assemblies
that can be completed each hour.
Normal Time Allowance
Element (min) (%)
1 4.2 15
2 1.7 15
3 2.9 15
5-14. (Section 5.4.3) For a particular activity that is being work sampled, determine
the number of observations necessary to ens ure a confidence interval of 95%
and a maximum interval of 5%. The observed proportion of occurrence of this
activity is 0.4 (i.e. , 40% of the random observations have identified this activity
in progress) . Next, repeat this exercise when the confidence interval is low-
ered to 90%.
5-15. (Section 5.4.3) In a production run of R-130jet engines, the cumulative time to
produce units 1 through 100 is 80,000 hr.
(a) Calculate the unit time for the fiftieth engine with a 71 % leaming curve in
effect.
(b) Calculate the cumulative time to produce units 1 through 40 with a 71%
leaming curve.
5-16. (Section 5.4.4) A 100-kW diesel generator cost $140,000 seven years ago when
a certain equipment cost index was arbitrarily set at 100. A similarly designed
generator rated at 150 kW is now being proposed and the cost index is 140.
The cos! exponent factor. {3. is O. 7 for this type of equipment.
(a) Determine the estimated cost of the proposed generator by using the ap-
proptiate cost estimating relationship.
(b) Repeat part (a) when {3 = 0.4.
5-17. (Section 5.4.4) The Neptune Manufactuting Company is consideting abandon-
ing their old plant , built 23 years ago, and constructing a new plant that has
50% more square footage. The original cost of the old plant was $300,000 and
its capacity, in terms of standardized production units, was 250,000 units per
year. Capacity of the proposed plant is to be 500,000 units per year. During
the past 23 years, costs of plant construction have risen by an average of 5%
per year. If the cost exponent factor is 0.8, what is the estimated cost of the
new plant ?
5-18. (Section 5 .5) Answer the following questions based on the Consumer Priee
Index in Table 5-10.
(a) What is the annual compounded rate of inflation in living costs from 1978
through 1981 ?
Exercises 141
(b) If a person had take-home pa y of $20,000 in 1978, how much should it have
been in 1981 in order for that person to be living comparably to 1978?
(c) If, during 1978-1981 a person was earning 8% through investments (such
as moncy-market accounts}, what would have been his real rate of return?
S-19. (Section 5.4)
(a) It is desired to estimate the 1990 construction cost of a new plant for which
1985 component costs, and applicable rates of inflation, are as follows :
La bor
Building materials
Equipment
Total
19!!5 Cost
(Millions)
$1
5
3
$9
Projected Annual Inflation Rate
( 1985 through 1990)
(%)
10
0
15
(b) Using the answer to part (a) and using a (combined) MARR of 25%, find
the equivalent worth of the 1990 construction cost as of 1985.
S-20. (Section 5.5) Estimate the salary you hope to be making 5 years from now,
stated in today's purchasing power (i.e. , RS\
0
l) . If the average inflation rate for
goods and services you expect to bu y is 10% compounded annually, wh at will
your hoped-for salary need to be in actual dollars (i .e., A$
5
)?
S-21. (Section 5.5)
(a) Y ou hope to prepare for someone's college education costs by depositing
enough in a bank savings account today to provide, as of 15 years from
now, an amou nt equivalent to $20,000 of today's purchasing power. If the
bank will pa y 8% and the estimated inflation rate is 5% (both compounded
annually), what lump sum should you deposit today?
(b) Repeat part (a) except that the estimated inflation rate is 10% per year .
S-22. (Section 5.5) Labor costs for alternatives X and Y are estimated on different
bases as follows:
End of Year n Ait. X (AS.) Ait. B ( R S ~ " J
1 S200M S190M
2 210M 190M
3 220M 190M
4 230M 190M
If the combined (real and inflation) MARR = 25% and the real MARR
15%/yr, show which alternative has the lower equivalent :
(a) Present worth of costs at year O.
(b) Future worth of costs at year 4.
l
vv
-
i
L \.
\, t
l
1
(
,,
~
f
j
1
~ .
!
Chapitre 2
INTRODUCTION LA NOTION DE RISQUE ET D'INCERTITUDE
2.1 INTRODUCTION:
Dans le monde de l'industrie et du commerce les dcideurs sont conscients qu'ils n'auront pas
toujours raison quand il s'agit d'valuer un projet ou de choisir entre plusieurs possibilits. En
effet, le profil d'un Cash flow est influenc par les vnements futurs (international , national ,
local .. . ) qui sont incertains et qui ne peuvent tre contrls par le dcideur.
Il n'est pas possible de prvoir l'avance l'arrive de tels vnements et la rentabilit d'un
certain projet ne peut tre value avec certitude, qu' la fin de celui-ci. Donc, la rentabilit
d'un projet donn dpend d'un facteur risque qui peut entraner qu'elle sera diffrente de celle
prvue au moment de la prise de dcision.
Nous avons considr, jusqu' maintenant dans toutes les tudes conomiques effectues,
connatre tous les lments avec certitude; nous pouvions donc dfinir la valeur d'un paramtre
conomique par un seul nombre. Mais en ralit dans une situation concrte ces paramtres:
vie utile, valeur de rcupration, revenus et cots ne peuvent tre dtermins par une valeur
unique. Pour plusieurs tudes conomiques, il devient ncessaire ou prfrable de considrer
l'aspect reli aux risques et l'incertitude introduit par la variation des diffrents paramtres.
Dans les chapitres 3,4 et 5, nous allons voir diffrentes techniques pour tenir compte du risque
et de l'incertitude dans une analyse conomique.
2.2 QU'EST-CE-QUE LE RISQUE ET L'INCERTITUDE
Avant de passer aux diffrentes techniques d'analyse conomique avec risque et incertitude,
nous tenterons d'claircir cette notion dans les sections suivantes:
2. 2.1 Terminologie
Les situations de dcision pouvent tre classes en trois catgories: certitude, risque et
incertitude.
Toutes les tudes conomiques qui ont t conduites dans les chapitres prcdents ont
t ralises dans des conditions de certitude, c'est--dire que la probabilit de succs
a t considre gale 1.0 pour chaque projet valu (fig. 2.1 a) . Cependant en
ralit, les exemples o les paramtres sont estims avec certitude sont rares. On parle
alors de risque ou d'incertitude.
2-2
Les analystes prsentent une distinction classique entre le risque et l'incertitude pour un
paramtre ou une analyse conomique; par le risque, l'analyste connat les probabilits
que les tats de la nature, les revenus et les cots se produisent (figure. 2.1 b) alors
qu'en prsence d' incertitude il ne possde pas la connaissance de la frquence de leur
distribution (figure. 2. l.c). La figure 2.1 montre graphiquement ces diffrences.
Jrnl\.
ROCl.E
intitu:E
j_
a=3
m
?
~ ;
\Ae\Jile \Ae Uile
\Ae\Jile
7friS V" 7friS
(a) (b) (c)
figure 2.1
D'autres auteurs diffrencient le risque de l'incertitude en affirmant que pour le risque ils
peuvent utiliser la distribution de la probabilit d'un paramtre alors que pour l'incertitude ils
n'ont .pas confiance que l' estim de cette probabilit soit correcte.
2.2.2. Causes du risque et de l'incertitude
Nous allons maintenant considrer quelques unes des principales causes du risquee d
l'incertitude:
1. Un nombre insuffisant d'investissement similaire
Chaque firme a un faible nombre d'investissement d'un type particulier. L' e -
des revenus dfavorables peut difficilement tre statistiquement compens par
les revenus favorables; ce phnomne devient critique si une compagnie in es .
un montant considrable par rapport ses ressources totales.
2-3
2. Erreurs subjectives dans les donnes
L'analyse conomique dpend de l'optimisme ou du pessimisme de l'analyse,
ou des facteurs intangibles qui ne devraient pas influencer l'analyse conomi-
que.
3. Changement dans l'environnement conomique invalidant l'exprience
prcdente
Il arrive souvent que l'analyste fait ses prdictions futures partir des donnes
accumules prcdemment mais sans ajustement pour les conditions futures
prvues. Ceci implique qu'il assume implicitement que le futur se droulera
comme le pass; parfois des circonstances peuvent invalider cette hypothse, par
exemple la crise d'nergie.
4. Mauvaise interprtation des donnes
Diffrents facteurs influenant les donnes peuvent conduire une mauvaise
interprtation.
5. Erreurs d'analyse
L'analyse technique et conomique peut comporter des erreurs.
6. Disponibilit d'administrateurs comptents
La limitation des ressources humaines et physiques l'intrieur d'une
organisation peut handicaper certains projets relativement d'autres.
7. Rcupration de l'investissement
Une industrie qui va se procurer de l'quipement trs spcialis doit se rendre
compte qu'il pourra tre reveridu avec beaucoup plus de risque que de
l'quipement d'application gnrale.
8. Dsuet
De nouveaux procds peuvent rendre l'quipement actuel dsuet; ce dernier
facteur constitue aussi une cause plus ou moins grande de risque.
De Garmo et Canada C) dans leur livre, regroupent en quatre parties les cau
risque et de l'incertitude:
a) L'imprcision possible des valeurs utilises dans l'tude conom1
l'estimation des revenus entrane gnralement le plus d'erreur.
b) Le genre d'industrie considr; par exemple une entreprise minire co
gnralement plus de risque qu'une picerie.
c) Le type d'usine et d'quipement; de l'quipement spcialis se re end
difficilement que de l'quipement rgulier.
2
d) Le temps qui doit s'couler avant que toutes les conditions de l'tude .so =-:
ralises; un temps plus long diminue la probabilit que tous les a<
demeurent tels qu'estims, donc le risque augmente puisqu'on ne
prophtiser exactement ce que l 'avenir apportera. La mthode de la pri
recouvrement fournit une indication de ce niveau de risque.
2.2.3 Faiblesse du traitement probabiliste
Les probabilits utilises dans une analyse du risque constituent gnralem
lments non vrifiables et trs subjectifs. De plus, les diffrentes probabilits u '
peuvent dans une mme tude conomique diffrer en qualit et en quantit entree es.
L'anal y ste se doit de raliser que l'utilisation de probabilits ne fait pas dispara
risque et l'incertitude mais fournit des informations supplmentaires qui aidero
prise de dcision. L'utilisation des limites de confiance des probabilits aidera
gestionnaire prendre une dcision. Pour les problmes industriels les ex
recommandent d'utiliser une limite de confiance de 95%; c'est--dire d'ajoute
valeur moyenne plus ou moins deux fois la dviation standard (2a).
2.2.4 Moyens de changer le degr d'incertitude
Une firme peut prendre diffrents moyens pour changer son degr d' ince
mentionnons:
1) Obtenir des informations techniques et conomiques supplmentaires.
2) Augmenter le nombre d'activits pour avoir suffisamment de pro
d'investissement pour que les valeurs moyennes utilises soient statistique
valables.
3) Diversifier les oprations; choisir des produits que les cycles conomiq
influencent diffremment.
2-5
2.2.5 Choix entre diffrents niveaux de risques et retours sur l'investissement
Les conomistes acceptent gnralement qu'un projet plus risqu doit rapporter un
retour sur l'investissement plus lev. La connaissance d'une relation entre le risque
et le retour permettrait de comparer tous les projets sur une mme base. Malheureu-
sement cette relation ne peut tre dtermine d'une faon gnrale et prcise; la
politique d'une firme par rapport au risque doit donc tre connue avant qu'une dcision
d'investir puisse tre prise. Cette politique dpend grandement des prfrences du
responsable dcisionnel et du risque total auquel la firme doit dj faire face.
2.2.6 Guide pour considrer le risque et l'incertitude
Le risque et l'incertitude existent sous une varit de forme et de degr dans toutes les
analyses conomiques. Limportance de considrer le risque et l'incertitude pour un
investissement augmente, avec l'augmentation de la complexit du projet et de son
importance conomique relativement la valeur totale de la firme. Une analyse
conomique supplmentaire tenant compte du risque et de l'incertitude se justifie si elle
cote moins que les conomies potentielles qui en rsulteraient; en pratique leurs
estimations deviennent difficiles.
En gnral, lorsqu'une firme dsire effectuer une tude conomique, soit pour un
investissement ou un remplacement, elle considrera les diffrentes tapes telles que le
montre la figure 2.2. Ces tapes s'appliquent aussi bien des projets mutuellement
exclusifs ou non. Elle fait voir l'existence de quatre points o les projets peuvent tre
accepts ou rejets.
Le point 1: comprend l'ensemble de l'argent que la compagnie devra payer si elle
accepte le projet. Elle effectuera un choix immdiatement si les montants pour chaque
projet demeurent faibles compars ceux pour une tude supplmentaire. Ce critre
subjectif dpend de l'tat financier de l'entreprise, de la grosseur des projets
gnralement considrs et de la disponibilit des ressources pour une analyse plus
pousse.
Le point 2: considre le rapprochement des choix; si l'un se dtache nettement la
compagnie arrtera sa dcision.
Le point 3: exan1ine les analyses dj effectues en fonction de l'importance relative du
projet pour la compagnie, elle peut alors dcider de faire une tude plus prcise de
certains lments.
Le point 4; la compagnie rpte les considrations du point 3.
2-6
2.3 DCISION SOUS INCERTITUDE COMPLTE
Les diffrentes techniques d'analyse sont utilises selon qu'on soit dans le cas de risque ou celui
d'incertitude. Dans le prsent chapitre nous allons dvelopper les techniques d'anal se
conomique dans le cas d'un investissement sous incertitude COMPLTE c'est--dire dans une
situation ou l'analyste ne peut pas assumer des probabilits aux diffrents tats de la nature.
Supposons qu'une dcision doit tre prise entre plusieurs possibilits A
1
, A
2
, ~ ; l e dcideu
dsire choisir la meilleure. n y a rn conditions futures, sous appeles tat de la nature SI s1
... Sm. Ces tats peuvent survenir indpendamment du contrle de celui qui doit prendre la
dcision.
Prenons P;i = Valeur pour celui qui prend la dcision s'il choisit la possibilit A; et qu ' en
mme temps, l'tat de la nature Si se produit. Supposons que les valeurs positives indiquent
des situations favorables alors que celles ngatives en montrent des non-favorables. Ces
valeurs peuvent tre des revenus, des valeurs prsentes, des utilits ou tout autre indice
caractrisant un investissement. Les diffrents choix et tats de la nature peuvent tre prsents
sous la forme d'une matrice, tableau 2.1 illustre une matrice de dcision.
Possibilits
(contrl par
le dcideur)
A
c
T
I
0
N
s
A\S.
A
A?
A ~
1
A,
1
AN
sl
p
p?l
p ~ l
1
Pn
1
PN1
Tableau 2.1: matrice de dcisions
TATS DE LA NATURE
'aucun contrle du dcideur
s?
---
SI
---
SM
Pn
---
p.
--- P_rn_
Pn
---
1
---
1
p ~ ? ---
1
---
1
1 1 1 1 1
Pn --- p..l.l_
l J
1 1 1 1 1
Pm
--- --- ---
PNM

_ __JI
QEis llmrtaui

FI>ft3 ....... l1bnct <Dnrn tt
...... crwdt
IWillictl
A
2-7
R:irt4
rrn
OlrSrer le mrq.Je
c:on:re aralyse sertMt,
10.2: ders reca m OOes p:u u-e amy1yoo
2-8
(:\

2-9
Maintenant considrons l'exemple prsent dans le tableau 2.2
Tableau 2.2 Exemple d'une matrice de dcision
TATS DE LA NATURE
A\S.
1 J
rcession prosprit grande
normale prosprit
Possibilits (actions) Sl S2 S3
Aucune construction Al +25 -15 -50
Construire une petite usine A2 -50 +40 +50
Construire une grosse usine A3 -150 +10 +200
Une analyse conomique sous certitude comporte un seul tat de la nature c'est--dire une seule
colonne. La meilleure possibilit consiste prendre la valeur la plus leve dans la colonne
unique (ex: en cas de rcession (S
1
) aucune construction (A
1
)). Mais, maintenant dans une
analyse sous incertitude avec plusieurs tats de la nature donc plusieurs colonnes, quelles
possibilits choisir?
En premier, liminons toutes les possibilits qui mdpendamment du principe de choix que le
dcideur utilisera montrent des infriorits vidntes, et ensuite, on procdera au choix entre
les autres possibilits en tenant compte de l'incertitude.
Exemple 2.1
Considrons la matrice du profit du tableau 2.3 qui reprsente la valeur prsente des profits de
projets (A
1
, A
2
, A
3
, A
4
et A
5
) selon quatre tats de la nature (S
1
, S
2
, S
3
et S
4
).
2- 10
Tableau 2.3: matrice de profits
TATS DE LA NATURE
Possibilits s s? S, s ~ ~ minimum maximum
A 4 4 0 2 0 4
A? 2 2 2 2 2 2
A, 0 8 0 0 0 8
A4 2 6 0 0 0 6
As 0 1 1 0 0 1
L'analyste se rend immdiatement compte par inspection de la matrice qu'indpendammen e
l'tat de la nature qui surviendra, il aura toujours avantage choisir la possibilit A
2
au li
de As. Il peut donc liminer immdiatement la possibilit As de toutes considrations futures.
On se rfre cette situation en mentionnant que la possibilit A
2
domine la possibilit A
5

D'une faon gnrale s'il y a deux possibilits Ag et Ah tel P gj ~ p h i pour toutes valeurs de
alors la possibilit Ah domine la possibilit Ag qui peut-tre limin de cette faon. ll restera.
plusieurs possibilits et l'analyste aura besoin de procdures supplmentaires pour faire
choix.
2.3.1 tude de quelques principes sous incertitude>>
2.3.1.1
Nous allons maintenant considrer diffrents principes de choix que les experts no
suggrent. Nous discuterons aussi des points forts et des faiblesses qu 'on
principes.
Principes de l'opportunit gale
Les analystes se rfrent souvent ce principe sous les noms de <<principe de
insuffisante ou principe de Laplace. En absence d'vidence contraire l 'anal
assume que tous les tats de nature Si ont une probabilit gale de se produire.
choisit donc l'action qui permettra d'optimiser la valeur espre.
Appliquons ce principe l'exemple 2.1
pi: probabilit que l'tat de la nature si se produise
PI =p2 =p3 =p4 = !,4
2.3.1.2
2.3.1.3
2-11
La possibilit A
1
a donc une valeur espre gale :
E(A
1
)= IAx4+
1
Ax4+ IAxO+
1
Ax2=2.5
De la mme faon les autres possibilits ont comme v ~ e u r s espres:
E(A0=E(A
3
)=E(A
4
)=2.0
Le choix se porte donc sur la possibilit A
1
qui permet d'avoir un profit espr maximal
de 2.5.
Principe maximin ou minimax
Ce principe consiste choisir la possibilit qui maximise la valeur minimale qu'on
pourrait recevoir ou minimiser la valeur maximale qu'on pourrait dbourser. En
d'autres mots, l'analyste choisit la possibilit Ai de la faon suivante dans le cas d'une
matrice de profits.
maximum minimum Pij
J
Ce principe reflte une attitude trs pessimisme et conservatrice; en effet, il suppose
qu'indpendamment de la possibilit choisie, l'tat de la nature le plus dfavorable va
se produire. Dans le cas d'une matrice contenant des cots au lieu des profits, le
principe consistera choisir l'action qui minimisera le cot maximal. Le principe
s'appellera alors minimax:
minimum maximum p ..
IJ
j
Dans l'exemple 2.1, le principe maximin choisit la possibilit A
2
laquelle correspond
le maximum (2) des minimums de (0,2,0 et 0) respectivement pour les possibilits (A
1
,
A
2
, A
3
et A
4
), (tableau 2.3).
Principe maximax ou minimin
Ce principe consiste choisir la possibilit Ai, qui maximise la valeur maximum des
profits, c'est--dire l'analyste choisit la possibilit Ai associe avec le maximum Pij dans
le cas d'une matrice de profits. Dans le cas d'une matrice de cots, ce principe permet
de slectionner la possibilit qui minimise le cot minimum.
minimum minimum Pij
i j
Cette rgle optimisme et aventureuse suppose que, indpendamment de la possibilit
choisie, le meilleur tat de la nature possible associ cette possibilit va arriver.
2.3.1.4
2- 12
Pour l'exemple 2. 1, le principe maximax choisit la possibilit A
3
laquelle correspond
le maximum (8) des maximums de (2,8,6,et 1) respectivement pour les possibilits (A
1
,
A
2
, A
3
et A
4
) , (Tableau 2.3).
Principe d'Hurwicz
Hurwicz a gnralis les deux principes ci-dessus pour tous les niveaux d' optimisme
entre le pessimisme extrme de MAXIMIN et l 'optimisme exagr de MAXIMAX.
L'analyste choisit une indice d'optimisme o< a< 1. Alors pour chaque possibilit
il calcule la relation:
Hi = a max Pii +(1-a) min Pii
1 1
Dans le cas de matrice de profits, on choisit la possibilit qui maximise Hi et dans le
cas de matrice de cots, on choisit la possibilit qui minimise Hi, cette relation se
dfinie comme le principe d 'Hurwicz. Aux deux limites lorsque le degr d' op ti mis o
gale zro et un, on retrouve respectivement les principes maximin et maximax.
a=O=>Hi= min Pii
j
= max min P .. =>maximin
IJ
i j
a=1 =>Hi= max Pii
j
Max Hi => max max Pi(>maximax
i j
La figure 2.3 nous fait voir le principe d'Hurwicz appliqu l'exemple 2. 1.
2.3.1.5
Valeur du
critre
d'Hurwicz
8 ~ A 3
6 A4
4 A1
~ ~ A 2
0.25
Indice d'optimiste: a
Supposons un degr d'optimisme a=0.25 dans l'exemple 2.1
H
1
pour A
1
= 0.25 x 4 + O. 75 x 0 = 1
H
2
pour A
2
= 0.25 x 2 + 0.75 x 2 = 2 = > choisir A
2
ou A
3
H
3
pour A
3
= 0.25 x 8 + O. 75 x 0 = 2
H
4
pour A
4
= 0.25 x 6 + O. 75 x 0 = 1.5
2-13
Dans l'exemple 2.1, le choix se portera sur la possibilit A
2
pour des degrs
d'optimisme (a) infrieur 0.25. Alors qu'il choisira A
3
pour des degrs suprieurs
0.25. Pour un degr d'optimisme gale la valeur 0.25, le principe n'indique aucune
prfrence entre les possibilits A
2
et A
3

Dans la relation d'Hurwicz, le degr d'optimisme a influencera les profits maximaux
et les cots minimaux moins de l'utilisation d'une autre convention.
Principe du regret minimax (Savage)
Considrons de nouveau l'exemple 2.1, si aprs qu'une possibilit ait t choisie et les
mesures prises pour son application, le dcideur constate que l'tat de la nature S
2
se
produit, il aimerait certainement avoir choisi la possibilit A
3
S'il avait dans les faits
choisi une autre possibilit, disons A
2
il regretterait son choix actuel. Pour un tat de
la nature donn la mesure du regret qu'il peut avoir; s'appelle REGRET. Pour une
matrice content ~ possibilit et Sm tats de la nature les regrets pourront tre calculs
1
. max {Pii}-Pi
1
=Ri
1
..
par la re atwn '
L'application de la dfinition du regret une possibilit A
2
et l'tat de la nature S
2
de
l'exemple 2.1, donne un regret de R
22
=8-2=6. Considrant d'une faon similaire les
2-14
15 autres combinaisons possibles (Ai, Sj) de la matrice de cet exemple l'analyste pourra
construire la matrice du regret, (tableau 2.3).
Tableau 2.4: matrice de regrets
TAT DE LA NATURE
Possibilits
sl s2 s3 s4
Regret maximum pour
pour la possibilits A
A1 0 4 2 0 4
A? 2 6 0 0 6
A3 4 0 2 2 4
A4 2 2 2 2 2
Dans cet exemple, le choix de la possibilit A
4
minimisera le regret maximal .
. 2.3.2 Problme de choix de principe
2.3.2.1
En rsum les diffrents principes considrs jusqu' maintenant choisissent:
L'opportunit gale - > A
1
Hurwicz pour a< 0.25- > A
2
Hurwicz pour a> 0.25- > A
3
Regret minimax - > A
4
Nous avons donc converti le problme de choisir une possibilit en celui de choisir
principe. Pour faciliter ce choix, considrons maintenant quelques unes des proprits
de ces principes.
Problmes dans l'utilisation du principe d'opportunit gale
Si l'on utilise le principe d'opportunit gale, le choix d'une possibilit est sensibl 2
la faon dont les diffrents tats futurs sont prsents.
Considrons l'exemple suivant:
Exemple 2.2
On dsire choisir entre deux pices d'quipement; l'une tant d'application
gnrale, l'autre ayant un emploi bien spcifique. Or, on peut obtenir trois co
possibles pour utiliser la machine. Il est aussi possible que l'on ait aucun contra
2-15
ce cas la machine d'application gnrale pourra tre utilise pour une autre application,
mais la machine d'application spcialise demeurera inutilise.
Donc la liste des tats futurs possibles:
S
1
aucun contrat
S
2
un contrat
A
1
on achte la machine spciale
A
2
on achte la machine gnrale
La matrice, prsente dans le tableau 2.5, montre les profits (pertes) correspondant
chaque possibilit et tat de la nature; avec le principe d'opportunit gale p
1
=p
2
=lf2
Tableau 2.5: matrice de profits
(2 tats de la nature)
ETAT DE LA NATURE
Possibilits s s ?
A -1 6
A? 1 5
Le profit espr rsultant du choix de chacune des actions gales: (p
1
=p
2
=
1
/z)
E{A
1
) =
1
/z(-1)+
1
/z{6) =2.5
E(A
2
)=
1
/z ( 1)+
1
/z(5) =3.0
Ce principe permet de slectionner la possibilit A
2
, car elle a un profit espr (3.0), suprieur
celui de la possibilit A
1
(2.5) .
Maintenant, supposons que 1 'analyste veut solutionner le mme problme en prsentant les tats
de la nature (obtention d'un contrat) de la faon suivante:
S
1
aucun contrat
S
2
contrat A
S
3
contrat B
s4 contrat c
La matrice des profits devient celle prsente au tableau 2.6
2.3.2.2
2-16
Tableau 2.6: matrice de profits
( 4 tats de la nature)
TAT DELA NATURE
Possibilits
SL
S, s3_ sd.
A1 -1 6 6 6
A, 1 5 5 5
Le principe d'opportunit gale nous donne donc maintenant:
Les profits esprs des deux possibilits deviennent:
E(AI) =
1
A(-1) +
1
A (6) +
1
A (6) +
1
A (6) = 4.25
E(A2) = lA( 1) +
1
A (5) +
1
A (5) + 1,4 (5) = 4.00
Maintenant avec le principe d'opportunit gale l'analyte choisi la possibilit A
1
, car
elle gnre un profit espr de 4.25 suprieur celui de la possibilit A
2

Pour viter cet inconvnient, l'analyste devrait tablir la liste des tats de la na
de faon que chacun ait la mme probabilit d'arriver. Ceci implique qu'il conruu
d'avance les probabilits, alors que sous incertitude complte>> il assume ne pas les
connatre.
Problme dans l'utilisation du principe d'Hurwicz
Considrons maintenant des exemples pour illustrer deux problmes que le principe
d'Hurwicz engendre le tableau 2. 7 illustre une matrice des profits.
Exemple 2.3
Soit la matrice des profits (tableau 2. 7).
Tableau 2.7: matrice de profits
TATS DELA NATURE
Possibilits
SI
A1 1
A? 3
H
1
= a(4) +(l-a) 1= 3 a+l
H
2
= a(3) + (l-a) 3= 3
s2 max Pu
.i
4 4
3 3
2-17
min p ..
IJ
j
1
3
Le choix de la possibilit A
1
donne selon le critre d 'Hurwicz les profits de 3a + 1,
alors que pour celui de la possibilit A
2
, il donne un profit constant de 3, qui ne dpend
pas du degr d'optimisme a.
galant ces deux quations, 3a+ 1 =3l'analyste trouve une valeur de 2/3 pour le degr
d'optimisme a au point d'intersection. Le choix se porte donc sur la possibilit A
2
pour
un degr d'optimisme infrieur 2/3, car ces profits galent 3, alors que ceux de la
possibilit A
1
demeurent infrieurs, car il galent 3a + 1.
Maintenant, ajoutons une constante 5 chaque lment de la premire colonne de la
matrice du tableau 2.7; le tableau 2.8 fait voir la nouvelle matrice.
1
Tableau 2.8: matrice de _Qrofits
TATS DE LA NATURE
Possibillts
SI s2
max min Pu
p ..
IJ
j
.i
A 6 4 6 4
A? 8 3 8 3
Le critre d 'Hurwicz permet de calculer les profits pour
la possibilit A
1
: H
1
de a(6) + (l-a) 4= 2 a+4
la possibilit A
2
: H
2
de a(8) + (l-a) 3= 5 a+3
1
2-18
Le point d'intersection devient:
5a+3 = 2a+4
La solution de cette quation donne une degr d'optimisme a, gale 113, donc po
toute valeur de a infrieur 113, le choix se dplace de la possibilit A
2
la possibilit
A
1
par rapport la solution prcdente, car 2a+4 demeure suprieur 5a+3.
Les experts rfrent ce problme comme un manque de linarit de la colonne- .
Si le futur S
1
se produit, l'avnement de l'tat de la nature S
1
procurera cinq units
supplmentaires de profit indpendamment de la possibilit choisie. Le changemen de
la possibilit A
2
A
1
dpend donc, seulement de l'tat de la nature et non pas de la
possibilit choisie; idalement la dcision prise ne devrait pas tre influence.
Le principe du regret minimum et de 1 'opportunit gale possdent la linarit d la
colonne.
Exemple 2.4
Considrons la matrice des profits prsente dans le tableau 2.9, pour illustrer
deuxime problme associ au principe d'Hurwicz:
Tableau 2.9: matrice de profits
TATS DE LA NATURE
Possibilits sl s? s ~ s<l s ~ S,., s7 SR
SQ slO
A1
A?
1 0 0 0
0 .
0 0 0 0 0
0 1 1 1 1 1 1 1 1 1
L'analyste qui utilise le critre d'Hurwicz considrera les possibilits A
1
et A
2
co e
quivalentes, car pour:
H
1
=a(1) +(1-a) O=a
H
2
=a(1) +(1-a) O=a
et pour
Mais la plupart des gens auront une prfrence intuitive pour la possibilit _
impliquant une notion prconue au sujet des probabilits, ce qui ne peut tre le cas
lorsqu'on traite d'un problme sous incertitude complte. Pour cet exemple l'anal
qui utilise le principe d'opportunit gale choisira la possibilit A
2

2.3.2.3
2-19
Problmes associs au principe du regret minimax
Savage a tabli ce principe pour contourner le trs grand conservatisme du principe
MAXIMIN (ou MINIMAX). Le tableau 2.10, illustre une matrice de profit o
l'utilisation du principe de Savage a un avantage sur celui du principe maximin.
Tableau 2.10: matrice de profits .
TATS DE LA NATURE
Possibilits
SI s2
min p ..
IJ
j
A, 0 1000000 0
A, 1 1 1
L'analyste qui utilise le principe maximin choisit la possibilit A
2
, mais intuitivement
la possibilit A
1
semble plus dsirable, mais comme on peut le voir partir de la
matrice du regret montre au tableau 2.11, le principe du regret minimax choisit la
possibilit A
1

Tableau 2.11: matrice de regrets
TATS DE LA NATURE
Possibilits s_j
s2
Regret maximum
A, 1 0 1
A"
0 999999 999999
Par ailleurs, le principe du regret minimax a aussi des problmes propres;
Exemple 2.6
Les tableaux 2.12 et 2.13 illustrent respectivement une matrice des profits etla matrice
de regrets correspondants.
2-20
Tableau 2.12: matrice de profits
TATS DE LA NATURE
Possibilits s,
s2 s3
A, 1 6 4
A? 5 3 6
Tableau 2.13: matrice de regrets
TATS DE LA NATURE
Possibilits s,
s2 s3
regret
maximum
A, 4 0 2 4
A? 0 3 0 3
L'utilisation du principe du regret minimax permet de choisir la possibilit ~ .
Maintenant, supposons qu'une 3e possibilit A
3
devient possible. La matrice des pro :.s
est prsente dans le tableau 2.14 et celle des regrets par le tableau 2.15.
Tableau 2.14: matrice de profits
TATS DELA NATURE
Possibilits S, s? S,
A., 1 6 4
A? 5 3 6
A, 4 8 1
2-21
Tableau 2.15: matrice de regrets
TATS DE LA NATURE
Possibilits
SI s2 s3
regrets
maximum
A, 4 2 2 4
A? 0 5 0 5
A_3 1 0 5 5
L'ajout d'une troisime possibilit A
3
, entrane que l'analyste choisit maintenant en
utilisant le principe de regret minimax la possibilit A
1
Ce rsultat dsappointe, car
le fait de considrer une nouvelle possibilit ne devrait pas changer le choix fait parmi
les possibilits prcdemment considres.
L'exemple suivant illustre ce point. Une personne dsire acheter une nouvelle auto
aprs avoir considr les modles Ford et Chevrolet, elle choisit un modle Ford.
Avant d'avoir complt la transaction, elle apprend qu'un nouveau dtaillant des
modles Dodge vient de s'installer prs de chez elle. Elle va donc voir ces autres
modles pour dterminer s'ils prsentent des avantages par rapport au modle Ford
choisi. Mais voil qu' aprs avoir vu les modles Dodge, cette personne se ravise et
dcide maintenant de choisir une auto de modle Chevrolet.
L'utilisation du principe du regret minimax dpend donc de l'addition d'une nouvelle
possibilit qui peut-tre ou non pertinente. Les principes d'opportunit gale et
d 'Hurwicz ne possdent pas ce problme.
2.3.3 Choix d'un principe
Plusieurs auteurs ont list les proprits dsirables qu'un principe devrait avoir. L'une
des plus connue a t faite par Milnor (2). Il liste les dix conditions auxquelles on
aimerait qu'une bonne rgle de dcision respecte:
1. Classification: Classifier les diverses actions.
2. Symtrie: La classification devrait tre indpendante de l'ordre des colonnes et
des ranges de la matrice.
2-22
3. Forte domination: L'analyste prfre la possibilit A; Ak, s'il prfre la valeur
P;i Pki pour tous lesj.
4. Continuit: Dans une matrice convergente, la classification ne doit pas tre
inverse la limite.
5. Linarit: La classification ne devrait pas changer si l'on remplace les valeurs
de la matrice (P;i) par a Pij + ~ ou a> 0
6. Addition d'une range: L'addition de nouvelles possibilits, i.e. de nou e es
ranges la matrice, ne devrait pas changer la classification des possibilits.
7. Linarit de la colonne: L'addition d'une constante tous les lments d'
colonne ne devrait pas changer la classification des possibilits.
8. La duplication d'une colonne: L'addition la matrice d'une nouvelle colonn
identique une existante ne devrait changer la classification des possibilits.
9. Convexit: Si les possibilits A; et Ak s'quivaillent aucune n'est prfre.
2. Addition d'une range: L"addition d'une nouvelle range o aucun lmen de
cette range n'est prfr aux lments correspondants de toutes ranges d ,.
existantes ne devrait pas changer la classification des possibilits.
Milnor (2) a prouv qu'aucun des principes discuts prcdemment ne satisfait les dix
conditions. On peut voir facilement que la rgle de Laplace viole la condition
Minimax la condition 7, Hurwicz la condition 7 et 9 et Savage la condition 6.
Milnor a aussi prouv mathmatiquement que le principe:
de Laplace satisfait les conditions 1, 2, 3, 6 et 7
Minimax satisfait les conditions 1, 2, 3, 4, 6, 8 et 9
d'Hurwicz satisfait les conditions 1, 2, 3, 4, 5, 6 et 8
de Savage satisfait les conditions 1, 2, 3, 4, 7, 8, 9 et 10
Pour choisir un principe, l'analyste doit donc d ~ i d e r des proprits les plus importan
pour un cas bien particulier et choisir celui qui en satisfait le plus. La liste d
proprits ne peut tre trs longue.
2-23
Problme no 2.1
tablir un principe personnel qui vous permet de prendre une dcision sous incertitude
complte. Monter un exemple comment ce principe peut-tre utilis. Quels sont les
avantages et inconvnients de votre principe?
Problme no 2.2
Appliquer les diffrents principes de choix que nous avons vu aux cours la matrice
de cots suivante:
TATS DE LA NATURE
Possibilits S, s? s1
s"
A1 18 18 10 14
A? 14 14 14 14
A1 5 26 10 10
A"
14 22 10 10
A_5_ 10 12 12 10
Problme no 2.3
Former une nouvelle matrice en prenant chaque cot de la matrice du problme 2.2 et
en lui additionnant 2 et en multipliant le rsultat par 3. Utiliser les mmes principes
de choix qu'au problme 2.2. Qu'est-ce-que ces rsultats vous suggrent sur les
proprits de diffrents principes?
Problme no 2.4
Dans la matrice de profits ci-dessous, quelle possibilit choisiriez-vous en utilisant le
principe du regret minimal?
2-2
TAT DE LA NATURE
Possibilits s,
s2 s3
A 8 7 4
A? 10 0 4
A, 1 9 5
At
5 6 7
1 . De Garmo et Canada, Engineering Economy
Chapitre 3
ANALYSE CONOMIQUE TRADITIONNELLE ET A V ANCE
SOUS RISQUE ET INCERTITUDE
3.1 INTRODUCTION:
Dans ce chapitre nous allons dvelopper d'autres techniques d'analyse conomique avec risque
et incertitude qu'on classera en deux grandes catgories soit, celles d'analyse traditionnelle et
celles d'analyse avance. L'accent sera mis sur l'analyse conomique avec risque, c'est--dire
le cas o l'analyste connat les probabilits des divers paramtres conomiques, notamment,
que les tats de la nature, les revenus et les cots se produisent.
3.2 BREF RAPPEL DES NOTIONS DE BASE DE PROBABILITS
La probabilit peut-tre dfinie comme le rapport du nombre de cas favorables l'vnement
sur le nombre de cas possibles.
Si un vnement A peut survenir S>> fois partir d'un nombre total de n>> cas possibles et
galement identiques, la probabilit que l'vnement survienne sera de:
s
P(A) =
n
o P(A) = probabilit que l'vnement survienne
s = nombre de cas o l'vnement peut survenir
n = nombre total de cas possibles.
Par exemple, en jouant pile ou face, la probabilit d'obtenir pile est de
1
/2 et la probabilit
d'obtenir face est de
1
h. Quand un d est lanc sur la table, la probabilit d'obtenir 1 est de
1/6, la probabilit d'obtenir 2 est de 1/6, ... , la probabilit d'obtenir un 6 est de 116. Aux
cartes, la probabilit d'obtenir du carreau est de 13/52. ll en est de mme pour le pique, le
coeur et le trfle.
La dfinition prcdente est utile dans les cas de situations finies o S>> et n>> sont connus.
Pour une situation infinie, o n = "", la dfinition conduira toujours une probabilit de O.
Ainsi, dans un cas de situation infinie, la probabilit qu'un vnement survienne est
proportionnelle la distribution de l'univers.
3.2.1 Thormes de probabilit
Thorme 1. La probabilit est exprime par un nombre variant entre 1.00 et 0, o la
valeur 1.00 reprsente la certitude que l'vnement arrivera et 0, la certitude que
l'vnement n'arrivera pas.
3-2
soit un vnement E; 0:5; P(E;) :5; 1 (quation 3.1)
Thorme 2. Si P(A) est la probabilit que l'vnement A survienne, la probabilit que
A n'arrive pas est donne par P(A)=l.OO- P(A)
P(A) + P(A) = 1 (quation 3.2)
Exemple: Si la probabilit d'obtenir une pice d'quipement lectrique dfectueuse es
0.04, la probabilit d'obtenir une bonne pice d'quipement est:
P(B) = 1.00 - P(A)
= 1.00- 0.04
= 0.96
Ainsi la probabilit d'obtenir une pice d'quipement acceptable est de 0.96.
Thorme 3. Si A et B sont deux vnements mutuellement exclusifs, alor la
probabilit que soit l'vnement A ou soit l'vnement B survienne est la somme de
leurs probabilits respectives.
P(A ou B) = P(A) + P (B) (quation .3)
L'expression mutuellement exclusive signifie que le fait qu'un vnement se produise
rend l'autre vnement impossible. Ainsi, si en jetant un d, on obtient un
(vnement A) , il n'est pas possible d'obtenir un 5.
Thorme 4. Si les vnements A et B sont indpendants, la probabilit q e
l'vnement A ou l'vnement B ou que A et B se produisent est donne par
P(A ou B ou les deux) = P(A) + P(B) - P(A et B) (quation 3. )
Un vnement indpendant est celui dont la ralisation ne prsente pas d'influences
la probabilit de ralisation de l'autre ou des autres vnements.
Exemple: Dterminer la probabilit d'obtenir un valet ou un carreau en tirant u
carte d'un paquet.
P(A ou B ou les deux) = P(A) + P(B) - P(A et B)
4 13 1
P(valet ou carreau) =- +---
52 52 52
16
=-=0.308
52
3-3
Thorme 5. La somme des probabilits des vnements relatifs une situation est
gale 1.
P(A) + P(B) + ... P(N) = 1.00 (quation 3.5)
Exemple: Un inspecteur examine 3 produits d'un sous-groupe pour dterminer s'ils
sont dfectueux. L'exprience nous dit que la probabilit de ne trouver aucun produit
dfectueux d'un chantillon de 3 est 0.89, que la probabilit de trouver un produit
dfectueux d'un chantillon de 3 est 0.06, et la probabilit de trouver deux produits
dfectueux d'un chantillon de 3 est 0.03. Quelle est la probabilit de trouver 3
produits dfectueux d'un chantillon de 3?
Cette situation ne comprends que 4 vnements: 0, 1, 2 et 3 produits dfectueux.
P(O) + P(l) + P(2) + P(3) = 1.00
0.89 + 0.06 + 0.03 + P(3) = 1.00
P(3) = 0.02
Thorme 6. Si A et B sont deux vnements indpendants, la probabilit que les deux
vnements arrivent est le produit de leurs probabilits respectives.
P(A et B) = P(A) x P(B) (quation 3.6)
Exemple: La probabilit que le projet A soit rentable est de 0.30 et la probabilit
que le projet B soit rentable est de 0.10. Quelle est la probabilit que les deux projets
soient rentables.
P(A et B) = P(A) x P(B)
= (0. 30) (0.10)
= 0.03
Thorme 7. Si A et B sont deux vnements dpendants, la probabilit que les deux
vnements se produisent est le produit de la probabilit que A arrive et que si A arrive,
B arrivera aussi.
P(A et B) = P(A) P(B/ A)
P(B/A)
= P(A et B) si P(A) *O
P(A)
(quation 3. 7)
(quation 3.8)
Le symbole P(B/ A) est dfini comme tant la probabilit que 1 'vnement B se
produise, si 1 'vnement A est arriv. Un vnement dpendant est celui dont la
ralisation influence la probabilit de l'autre ou des autres vnements.
3
Exemple: Une bote contient 50 engrenages et 3 d'entre eux sont dfectueux. i
un chantillon de 2 est enlev, quelle est la probabilit que les deux soient dfectueux?
La probabilit que le premier soit dfectueux (vnement A) est P(A) = 3/50. La
probabilit que le second soit dfectueux est P(B/A) = 2/49, tant donn que la
premire pice dfectueuse a dj t enleve.
P(A et B) = P(A) x P(B/ A)
P(A et B) = (3/50) x (2/49) = 0.002
3.2.2 Valeur espre et variance
Soit X
1
, X
2
.. . X, des vnements mutuels exclusifs qui reprsentent des re en
possibles et P
1
, P
2
, ... Pn, les probabilits de ces vnements tel que:
La valeur espre des revenus (moyenne) est de:
m
E(x) = P.. = L xi Pi
i ~ I
(quation . 10)
La mesure de dispersion de la distribution de probabilit appel variante V(x) ou ci es
donne que d'expression.
m
o
2
= V(n) = L ((xi-E(x))
2
Pi (quation . 10)
i ~ I
o
2
= V(x) = E(x
2
)- [E(x)f (quation . 11
Exemple 3.1: Une compagnie considre l'investissement dans un projet don 1
revenus probabilistes sont donns au tableau 3 .1.
Tableau 3.1:
donne pour l'exemple 3.1
Revenu Probabilit des revenus
X.$ P(xi)
-llO 000 0.8
530 000 0.2
Dterminer les quantits suivantes:
a) La valeur espre du projet
b) La variante du projet
Solution:
a) de l'quation 3.9
E(X) = X1 P(Xl) + X2 P(X2)
= -110 000 x (0.8) + 530 000 x (0.2) = 18 000$
b) de l'quation 3.10
a
2
= V(X) = (X
1
- E(X
1
))
2
x P(X
1
) + (X
2
- E(X
2
))
2
x P(X
2
)
= (-110 000- 18 000)
2
x (0.8) + (530 000-18 000)
2
x (0.2)
a
2
= V(X) = 6,553.10
10
$
3.3 ANALYSES TRADITIONNELLES
3-5
Nous allons en premier voir quelques unes des analyses traditionnelles le plus importantes
utilises pour considrer le risque.
1. Jugement intuitif
2. Ajustement conservateur
3. Optimiste-pessimiste
4. Sensibilit
5. Point mort
6. Escompte du risque
3.3.1 Jugement intuitif
Cette mthode consiste effectuer l'analyse conomique en supposant que l'analyste
connat tous les paramtres conomiques avec certitude et applique un jugement
subjectif sur le risque de chaque projet; il peut renverser la dcision base sur les
rsultats quantitatifs s'il considre le risque suffisamment grand.
3.3.2 Ajustement conservateur
Cette procdure consiste changer les estims d'un ou plusieurs paramtres d'une
analyse dans une direction conservatrice, de faon diminuer le risque qu'une situation
survienne qui ne soit pas aussi favorable que celle prdite par l'analyse conomique.
L'analyste peut faire l'tude diffrents degrs de conservatisme. Le choix de valeurs
conservatrices pour les paramtres conomiques augmente la certitude d'un rsultat
favorable.
3-6
Exemple 3. 2:
Considrons l'achat d'une nouvelle machine pour imprimer les circuits lectriques e
remplacer celle existante, ceci devrait permettre une compagnie de raliser une
conomie. L' achat et l'installation cotera 10 000$, avec certitude. Elle esti me
produire 5 000 circuits de plus par anne durant les six prochaines annes et obtenir une
conomie annuelle de 5 200$. La compagnie estime que la nouvelle machine aura une
vie utile de six ans, possdera alors une valeur de rcupration de zro et cotera
000$ par an de plus pour 1 'opration et la maintenance. La banque financera cet achat
un taux annuel effectif de 5%.
Considrant la valeur des paramtres estims comme exacte, on peut calculer le profi
annuel ralis durant les six prochaines annes.
Rcupration du capital avec un retour 10 000 (A/P, 5, 6)
Cot d'opration et de maintenance supplmentaire
Cot total annuel
Revenu total annuel
Profit annuel
(3 9 )
5 2
1 2
a) Pour tenir compte de la variation possible des valeurs des estims, la compagnie
appliquera un jugement conservateur en utilisant un taux d'intrt de 25 %. Les
cots et les revenus annuels deviennent:
Rcupration du capital avec une retour 10 000 (AfP, 25, 6)
Cot d'opration et de maintenance supplmentaire
Cot total annuel
Revenu total annuel
Profit (perte) annuel
(3 390 )
(2 )
(5 39
52
-(190 )
Un taux d'intrt de 25% entranera une perte de 190$ par an et la compagnie pourra
dcider de ne pas effectuer le remplacement.
b) D'une faon similaire la compagnie pourra conserver le taux d' intrt de 5 o
appliquer un jugement conservateur en rduisant la priode de rcupration de
6 3 ans. Les cots et les revenus annuels deviennent:
Rcupration du capital avec une retour 10 000 (AfP, 5, 3)
Cot d'opration et de maintenance supplmentaire
Cot total annuel
Revenu total annuel
Profit (perte) annuel
De nouveau le jugement conservateur rend le remplacement non favorable.
3. 3. 3 Optimiste-pessimiste
3-7
(3 670$)
(2 000$)
(5 670$)
5 200$
-(470$)
Cette procdure consiste changer la valeur estime pour un ou plusieurs paramtres
d'une faon favorable (optimiste) et non-favorable (pessimiste), pour dterminer l'effet
de ces variations sur les rsultats de l'analyse conomique. L'utilisation de mthodes
statistiques permet de mesurer le degr d'optimisme. La valeur pessimiste correspond
l'estim des rsultats lorsque l'analyste attribue chaque paramtre conomique la
valeur la moins favorable qui a une chance raisonnable de se produire. Cette valeur ne
correspond pas la plus mauvaise valeur qui peut se produire. L'estim doit tre fait
pour chaque paramtre individuel.
La valeur optimiste correspond l'estim des rsultats lorsque l'analyste attribue
chaque paramtre la valeur la plus favorable qui a une chance raisonnable de se
produire. Canada dans son livre dfmit un estim optimiste comme la valeur d'un
paramtre dont l'analyste prvoit qu'elle sera atteinte pas plus que 5% du temps, alors
qu'un estim pessimiste correspond la valeur d'un paramtre dont il prvoit qu'elle
sera plus favorable que la valeur finale pas plus que 5% du temps.
La figure 3.1 illustre une reprsentation possible des valeurs pessimistes et optimistes
des revenus.
valeur prvue
estim pessimist; .. optimiste
5% 2 a 2 a 5%
distribution des revenus
Fig. 3.1
o: dviation standard
Cette mthode optimisme-pessimisme apporte des informations additionnelles dans une
tude conomique; elle montre les consquences rsultantes des dviations de la valeur
prvue. Les valeurs pessimistes utilises dans cette analyse peuvent correspondre aux
valeurs conservatrices de l'anal y se prcdente.
Exemple 3.3
Appliquons cette analyse la situation dcrite dans le tableau 3. 1
Tableau 3.1: donne de l'exemple 3. 3
Optimiste Prvue Pessimiste
(estim)
Investissement $ 10 000 10 000 10 000
Vie utile, ans 7 5 4
Valeur de rcupration $ 2 000 2 000 2 000
Profit annuel, $ 6 000 5 000 4 500
Cot annuel $ 2 200 2 200 2 400
Taux d'intrts % 8 8 8
3-9
Calculons les profits (cots) quivalents annuels en considrant successivement les
valeurs pessimiste, prvue et optimiste des diffrents paramtres.
AE (pessimiste) = -[(10 000-2 000) (A/P, 8%, 4) +2 000 (.08)]
- 2 400 + 4 500 = -(470$)
AE (prvue) = -[(10 000- 2 000) (A/P, 8%, 5) +2 000 (.08)]
- 2 200 + 5 000 = 630$
AE (optimiste) = -[(10 000- 2 000) (A/P, 8%, 7) +2 000 (.08)]
- 2 200 + 6 000 = 2 100$
Les informations additionnelles dans des conditions d'optimiste et de pessimiste aideront
l'investisseur prendre une dcision en considrant les profts et les pertes possibles
dans ces conditions extrmes.
Note: Dans cet exemple, l'analyste a considr que tous les lments pessimistes ou
optimistes se produisent en mme temps, mais il pourrait considrer diffrentes
combinaisons de ces valeurs.
3. 3. 4 Analyse de sensibilit
Une analyse de sensibilit apporte des informations supplmentaires une simple
analyse sous certitude. Elle permet de dterminer le degr de sensibilit des rsultats
d'une analyse conomique aux variations des diffrents paramtres utiliss. Les
experts, parfois, dfiniront le concept de sensibilit comme le changement relatif de
grandeur d'un ou plusieurs paramtres, dans une analyse conomique, qui va renverser
le choix des actions possibles: cette analyse s'effectue d'une faon similaire celle
optimiste-pessimiste.
L'une des faons les plus intressantes pour tudier la sensibilit consiste mettre en
graphique les rsultats de tous les paramtres indpendants considrs; l'analyste
montre la variation de chacun sur une abscisse commune, comme pourcentage de la
valeur estime; l'ordonne reprsentant le critre conomique utilis pour valuer le
projet.
La figure 3.2 montre un exemple de ce type:
Q)
::::J
c:
c:
m
laux d'int,
ert
-100 -50 0 50 100
Dviation de la valeur estime, %
Fig.3.2 Analyse de sensibilit
-10
Dans ce cas particulier la variation des cots annuels et de la vie utile ont une influence beaucoup pl
importante sur la variation des profits annuels, que celle du taux d'intrt; l'analyste devrait leu
accorder une attention particulire.
Exemple 3.4
Une socit 0 dsirant construire un nouvel atelier de fabrication de pte doi
dterminer le rendement en pte le plus conomique. Une analyse prliminaire a
permis d'identifier les cots du bois et du raffinage comme les paramtres les plus
importants sur le cot de la fabrication de la pte. La socit a conduit une tude d
sensibilit sur ces deux paramtres. La figure 3.3 montre l'influence des variations de
60% respectivement dans les cots du bois et de l'nergie sur le cot de fabrication
de la pte, pour trois niveaux de rendement.
titre d'illustration cette analyse permet de constater que si les cots d'ne e
augmentaient par 50%, il n'y aurait pratiquement pas d'avantage conomique d
produire une pte 90% de rendement au lieu de 70%. De mme, si une compagni
pouvait rduire ses cots en bois d'environ 30%, elle pourrait utiliser aussi cono
quement des ptes bas (50%) et haut rendements (90%). On peut facilem
comprendre que ce genre d'information a grandement aider la compagnie prendre
dcision concernant son investissement.
8
6
du cot d'n
-60/o -40 -20
Variation du cot individuel
Rendement en pte
50
7
cot dubois
0 +40 +600!o'
Fig.3.3: cot de production de pte diffrents niveaux de rendement
3-11
3.3.4.1
8
6
4
2
0
3- 1
Autre type d'analyse de sensibilit
Les experts suggrent aussi une autre faon d'tudier la sensibilit; elle consiste
mettre en graphique deux des paramtres importants, par exemple le taux d'intrt
et la vie utile, et de tracer une ou des iso-courbes, pour le critre conomique
choisi, tels les profits annuels. Une ligne d'indiffrence limitera deux zones; l 'une
o l'analyste suggre l'investissement et l'autre o ille dconseille. La figure 3.
illustre cette mthode.
Investir
pas investir
0 2 4 6 8 1 0 12 14 16 18 20
Taux d'intrt, %
Fig . 3.4 : sensibilit d'un projet deux paramtres importants
Exemple 3.5
Appliquons cette analyse de sensibilit aux valeurs prvues de l'exemple
considr pour illustrer l'analyse optimiste-pessimiste. L'analyste dsire conna
l'effet d'une diminution de la vie utile de 40% et d'une augmentation du t2.
d'intrt de 50%, car il considre ces deux lments comme critiques. La m od
des cots annuels quivalents donne:
CA= + 5 000-2 200 -[(10 000-2 000) (A/P, 12%, 3) + 2 000 (. 12)]
=- 773$
Prcdemment pour un taux d'intrt de 8% et une dure de vie utile de 5 ans.
cette mthode indiquait un profit annuel de 630$. Dans le mme exemp
l'analyste dsire trouver pour quelle valeur du taux d'intrt la dcision d'in
changera. Il suffit donc de trouver la valeur du taux <;l'intrt pour que les co
annuels galent zro.
CA= 5 000-2 200 -[(10 000-2 000) (A/P, i%, 5) + 2 000 (i %)] =0
pour i = 12%
i = 20%
CA= +330
CA= -300
3- 13
Donc par interpolation, ou graphiquement l'on trouve que:
1=12% . (20%-12%)=16.2%
. ( 330 l
330+300
3.3.5 Analyse du point mort
L' analyste utilise cette analyse dans des tudes conomiques o il existe une
incertitude particulire, concernant un des paramtres. Il dfinit le point mort
comme la valeur du paramtre auquel la rentabilit du projet devient marginale.
Dans le cas de plusieurs choix, il dfinit le point mort comme la valeur du
paramtre auquel les choix deviennent galement dsirables.
Exemple 3.6
Supposons que dans l'exemple 3.3, l'estimation de la vie utile soit trs incertaine,
l'analyste dsire dterminer la vie utile minimal permettant de justifier le projet.
galant les cots annuels quivalents zro l'anal y ste dtermine la dure de vie
utile (N) minimal.
5 000-2 200-((10 000-2 000) (A/P,8%,N)-2 000(.08)]=0
(AIP, 8%,N)=-
2640
=0.330 Dans les tables on trouve un
8000
temps de 3. 7 ans qui correspond la valeur de 0.330.
Exemple 3.7
Une compagnie considre la fabrication d'un nouveau produit, elle dispose des
donnes suivantes:
Prix de vente:$/unit
Cot d'quipement:$
Frais fixe par anne:$
Cot d'opration et de maintenance:$/heure
Temps de production /1000 unit, heure
Dure de vie du projet, ans
TRAM,%
12,50
200 000
50 000
25
100
5
15
La compagnie assume une valeur de rcupration des quipements nulle aprs 5 ans
et prend X= volume de vente /anne. La valeur annuelle de l'investissement est:
VA = -200 000 (A/P, 15, 5)- 50 000- 0.1 * (25) X + 12.5.X
VA = -109 660 + 10.X
Au point mort V A = 0 = > X = 10 966 units/ anne
3-1
La figure 3.5 montre une reprsentation graphique de ce problme.
1 1 1 1 1 1 1 1 1 1









J
1
Cots
annuel
1

l
1
20 00
1 1 1 1 1 1 1 1
1
1
0
5000
11000 150 000
X (Ventes annuelles)
Fig. 3.5 : graphique du point mort
3.3.6 L'escompte du risque
Cette analyse utilise un taux d'intrt proportionnel au degr de risque; lorsque ce
dernier crot le retour sur l'investissement doit tre plus lev. L'analyste trou e
gnralement difficile et subjectif d'tablir un taux d'intrt associ un degr de
risque. Il doit connatre les revenus et les cots pour chaque proje .
L'interprtation de cette analyse doit tre change lorsqu'un projet comporte u
revenu net ngatif, car un taux d'intrt plus lev pour un risque supri
entranera une valeur prsente du cot plus faible pour le projet. Cette analyse ne
s'applique pas aux projets ayant une dure de vie utile trs faible, car dans ce cas
l'analyste n'escompte pas les revenus et les cots.
Pour viter d'avoir attribuer un taux d'intrt diffrent chaque projet selon le
risque qu'il comporte, l'analyste peut calculer les valeurs prsentes pour cha e
projet correspondant divers taux d'intrt. Mettant en graphique la val
prsente en fonction du taux d'intrt, pour les projets les plus risqus il dtermin
les taux d'intrt ou de retour correspondant une mme valeur prsente.
3-15
Taux d'intrt,
0
/o
Fig. 3.5 : exemple d'analyse avec escompte de risque
Les projets A et B ont une mme valeur prsente de 10 lorsque l'analyste escompte
les revenus du projet B par un taux de 5% et celui du projet A par 20%. On peut
voir aussi, que s'il escompte les revenus du projet A un taux suffisamment lev,
si le projet comporte un haut niveau de risque, la valeur prsente du projet B peut
devenir suprieure celle du projet A. Cette analyse fournira un lment de plus
pour la prise de dcision.
3.4 ANALYSE AVANCE
Aprs avoir considr quelques analyses traditionnelles, nous allons maintenant voir quatre
(4) mthodes d'analyse conomique plus avances:
1. Montaire probabilistique
2. Estim - variance
3. Escompte des variables
4. Simulation de Monte-Carlo
3.4.1 Analyse montaire probabilistisque
L'anal y se montaire probabilistique comprend l'estimation de certains paramtres
dont l'analyste considre la variation comme importante en terme de probabilit.
Ces valeurs estimes peuvent tre modifies pour obtenir les caractristiques
3.4.1.1
3.4.1.1.1
3-16
dsires de la mesure du mrite (ex. la valeur prsente). Les rsultats d'une telle
tude peuvent se prsenter sous l'une des formes suivantes:
a) L'estim des bnfices:
i) valeur prvue
) tat futur le plus probable
i) niveau d'aspiration
b) L'estim des bnfices et de leurs variations possibles.
L'estim des bnfices
Considrons en premier l'analyse relie l'estim des bnfices.
Posons: Pi = probabilit que l'tat futur possible Si se produise
Pi ~ 0 pour tous les j et
j = rn
:E
P.
J
1
j = 1
Dans une situation sous risque l'analyste connat ces probabilits ou peut les
estimer alors que dans une situations sous incertitude complte, il n'a pas confiance
que l'estim des probabilits soit correct.
V ale urs prvues
La rgle de la valeur prvue sert souvent pour faire un choix dans une situatio
comportant des risques. L'analyste choisit la possibilit qui maximise le gain pr
ou minimise les pertes prvues.
Pour la possibilit A;, on calcule le gain ou la perte prvue partir, de la
sommation du produit de chacun des gains ou pertes par la probabilit qu'il se
produise:
j = rn
L
P .. P.
IJ J
j = 1
La dtermination des probabilits constitue l'un des principaux problmes lors
l'analyste dsire utiliser la mthode de la valeur prvue. Elles peuvent tre bases
sur des donnes historiques ou sur un traitement statistique rigoureux, mais dans
la plupart des cas l'analyste doit se servir de son intuition et de son j ugemen .
ne possde aucune assurance que le futur se droulera d'une faon identique a
pass mme si les probabilits peuvent avoir pour base des donnes historiques.
3-17
En revanche, l'obligation de penser et d'estimer la valeur des diffrents facteurs
dans le futur procurera des rsultats suprieurs ceux provenant d'aucune
rflexion. Certains analystes trouvent utile d'utiliser des distributions statistiques
(ex: Poisson, Normale) pour dterminer les valeurs prvues.
Parmi les distributions de probabilit thoriques , qui sont le plus utilises dans
l'analyse conomique avec risque, on cite les distributions Normale et Beta. La
figure 3.6 montre un exemple de distribution Normale. Les probabilits qu'une
variable alatoire X, normalement distribue ait une valeur dans un intervalle de
a, 2a ou 3a de chaque ct de la moyenne sont respectivement 68.26, 95.45 et
99.73%.
99.73%
Fig. 3.6 : exemple d'une distribution normale
Dans certains cas, on ne peut reprsenter la distribution de probabilit par une
distribution thorique connue. Une approche qui peut-tre utilise pour estimer
une distribution de probabilit subjective consiste dterminer un estim
pessimiste, optimiste et la valeur la plus probable d'une variable alatoire.
La rgle de la valeur prvue constitue une mthode trs rpandue; pourquoi est-elle
bonne? Considrons une squence de dcisions ayant des rsultats indpendants:
j = rn
E
Pj ou j = 1 ,2,3, .. . m
j = 1
z'r.. gale la somme des valeurs des rsultats des N premires dcisions lorsque
l'analyste applique le principe de la valeur prvue. Ici , par hypothse pour chaque
dcision l'analyste choisit la possibilit qui maximise VP; zN symbolise cette
somme partielle.
Alors que: ~ gale la somme des valeurs des rsultats des N premires dcisions
lorsque l'analyste applique tout autre principe de choix. Lorsque le nombre de
3.4.1.1.2
3.4.1.1.3
3-1
dcision N croit et tend vers l'infini la probabilit que zN soit plus grand q u e ~
tend vers un. Ceci constitue la Loi du succs long terme. Plusieurs volumes
de statistiques prsentent la preuve de cette loi.
L'tat futur le plus probable
Cette rgle consiste trouver l'tat de la nature possible, Sj, qui possde la
probabilit maximal de se produire; pour cet tat de la nature choisir la possibilit
~ qui maximisera Pij Cette rgle considre que l'tat de la nature le plus probable
va se produire avec certitude. En pratique, beaucoup d'investisseurs prennent des
dcisions de cette faon.
Niveau d'aspiration
Pour un iveau d'aspiration K, slectionnez la possibilit Ai qui maximisera la
probabilit que la valeur du rsultat soit plus grande ou gale K; K tant une
mesure d'un niveau de rentabilit que le propritaire d' une firme dsire obten
pour un projet. Cette rgle reflte le fait que la plupart des humains essaient de
dcouvrir et de slectionner des possibilits satisfaisantes et non pas ncessairement
optimales.
Exemple 3.7
Pour mieux comprendre l'application de ces rgles, considrons la matrice de
profits (tableau 3.2); elle montre aussi les probabilits qu'ont chaque tat de la
nature de se produire.
Tableau 3.2: matrice de profits
TATS DE LA NATURE
Possibilits
SI
S2 S3
actions
p_l = .2 p2 = .7 P
1
= . 1
Al 50 120 2 000
A, 130 130 130
AJ 500 110 100
Calculons en premier la valeur prvue pour chaque action:
VA(A
1
) = 50* .2 + 120 * .7 + 2 000 * .1 = 294
VA(A
2
) = 130 * .2 + 130 * .7 + 130 * . 1 = 130
VA(A
3
) = 500 * .2 + 110 * .7 + 100 * .1 = 187
3-19
Donc ici la rgle de la valeur prvue>> va permettre de choisir la possibilit A
1
, car
294> 130 et 187.
L'tat de la nature S
2
possde la probabilit la plus leve de se produire donc la
rgle de l'tat futur le plus probable va slectionner pour cet tat de la nature
l'action qui rapportera le profit maximal; ici, la possibilit A
2
qui donnera un profit
de 130 compar 120 et 110 respectivement pour A
1
et A
3

Maintenant, supposons que l'investisseur dsire un profit suprieur ou gal 300
(K = 300), il considrera pour chaque possibilit la ou les probabilits d'obtenir un
tel profit.
Pour la possibilit A
1
seul l'tat de la nature S
3
qui a une probabilit de 0.1 de se
produire pourra lui procurer un profit de 2 000. Pour la possibilit A
2
aucun tat
de la nature ne pourra lui procurer un tel profit. Pour la possibilit A
3
, l'tat de
nature S
1
qui a une probabilit de 0.2 de se produire lui procurera un profit de 500.
Donc, la rgle de niveau d'aspiration, pour un niveau de profit de 300, slectionne-
ra la possibilit A
3
qui a une probabilit de 0.2 de lui procurer ce profit alors que
les possibilit A
1
et A
2
ont respectivement des probabilits de 0.1 et 0.0.
Plusieurs autres principes de choix peuvent tre prsents, mais lequel devons-nous
utiliser? Beaucoup d'tudes utilisent le principe de la valeur prvue principale-
ment cause de la loi du succs long terme. En revanche, cette loi s'applique
une longue squence de dcisions. Or, la plupart des investisseurs ont effectuer
un seul investissement ou un nombre restreint ce qui n'assure pas la validit de la
loi du succs long terme.
Exemple 3.8
Considrons un exemple familier chacun d'entre nous, soit celui d'acheter ou non
une assurance automobile. Les statistiques publies montrent qu'un individu a une
chance sur mille d'avoir un accident (P
1
= 0.001); donc, la probabilit de ne pas
avoir d'accident gale 0.999. Si une police d'assurance cot 600$ et que le cot
moyen d'un accident gale 40 000$, l'individu doit-il ou non acheter une
assurance? La matrice du tableau 3.3 montre les diffrents cots:
Les rgles de la valeur prvue (VP(A
1
) =-600 et E(A
2
) =-40) de l'tat futur le plus
probable ( S ~ et du niveau d'aspiration pour K suprieur -600 slectionnent toutes
la possibilit A
2
c'est--dire de ne pas acheter d'assurance.
Mais beaucoup de personnes dans ce cas vont prfrer payer 600$ pour liminer
mme une chance trs faible de perdre une somme considrable de 40 000$.
1
3.4.1.2
3-20
Tableau 3.3: donne de l'exemQle 3.8
1
TATS DELA NATURE
Avoir un accident ne pas avoir
Possibilits
SI
d'accident
s2
P=O.OOl P=0.999
A
1
: acheter de l'assurance
-600 -600
A
2
: ne pas acheter d'assurance -40 000 0
Estim des bnfices et de leurs variations possibles
La rgle de la valeur prvue peut-tre considrablement amliore lorsque
1' analyste ajoute aux valeurs prvues une mesure de la variation possible des
bnfices et si possible une description complte de variation de la probabilit des
bnfices.
Pour mieux comprendre cette notion, considrons l'exemple suivant:
Exemple 3.9
Un projet A a un bnfice annuel prvu de 10 000$ avec une dviation standard de
25 000$. Un projet Ba un bnfice annuel prvu de 8 000$ avec une dviatio
standard de 4 000$. Un investisseur qui dsire prendre le minimum de risque
pourra choisir le projet B qui a un bnfice annuel prvu infrieur au projet A.
Supposons que les bnfices prvus possdent une distribution normale; la forme
de la courbe de distribution des bnfices procurera l'investisseur des informa-
tions supplmentaires et lui permettra de mieux j u g ~ r des avantages respectifs des
diffrents choix.
La figure 3.7 montre les bnfices de l'exemple distribus normalement.
3.4.1.3
3.4.1.3.1
probabilit de perte
pour le projet A
-20 -10
3-21
projet B
~ = 8 000 0 = 4 000
projet A ~ = 1 0 000
~ = 25 000
0
10 20 30
40
Fig. 3.7 : distribution de bnfices
On peut voir, que le projet A a une probabilit plus leve d'encourir une perte que
le projet B; le projet B pourrait tre alors choisi. Une distribution statistique des
bnfices autre que normale pourrait changer la dcision.
Les sries de Taylor peuvent servir pour approximer les variations simples des
diffrents paramtres conomiques, mais 1 'analyste doit recourir aux techniques de
simulation pour l'tude des variations plus complexes.
Exemple sur l'analyse avance
Exemple sur la mthode montaire probabilistique
Le dpartement de gnie industriel a fourni aux cadres d'une usine les informations
suivantes concernant un projet d'investissement.
Investissement,$
dure de vie, ans
3
Valeur de rcupration, $
Revenu annuel, $
Cot annuel , $
5
7
Financement (taux d'intrt annuel effectif), %
10 000
Probabilit
0.3
0.4
0.3
2 000
5000
2 200
8
3.4.1.3.2
3-22
Les cadres peuvent donc calculer l'aide de la mthode des revenus (cots) annuels
quivalents les bnfices prvus et leur dviation standard.
Pour une vie de 3 ans
VA=5 000-2 200 -[(10 000-2 000) (A/P, 8%, 3) + 2 000 (8%)] =- 460$
Pour une vie de 5 ans
VA=5 000 - 2 200 -[(10 000-2 000) (A/P, 8%, 5) + 2 000 (8%)] = 630$
Pour une vie de 7 ans
VA=5 000-2 200 -[(10 000-2 000) (A/P, 8%, 7) + 2 000 (8%)] = 1 110
L'investissement aura donc un revenu annuel prvu de:
VAprvu= -460$ (0.3) + 630$ (0.4) +1110 (0.3) = 446$
Les bnfices auront comme variance:
2
Variance(a =L [(AE)
2
* probabilit(AE)] - (AEprvu)
2
AE
= ( -460? * 0.3 + (630? * 0.4 + (1110)
2
* 0.3- (446)
2
= 401000
a= dviation standard= V401000 = 631$
Pour une limite de confiance de 95% les bnfices pourront varier entre - 81 6S e
1 708$ car;
AE = JL 2a
AE = 446$ + 2 * 631$
AE = - 816$ 1 708$
Les cadres savent maintenant que l'investissement a un bnfice prvu de 44
mais qu'il pourra encourir aussi une perte de 816$ et un bnfice maximum de
1 708$ pour les limites de confiance fixes.
Exemple sur la valeur et l'utilit prvues
Un investisseur qui a la courbe utilitaire ci-dessous dsire tudier la possibili
d'investir dans un projet qui a des revenus prvus en fonction de diffren
probabilits.
30
20
10
1 / 1 ~ - - - - - - - - - - - - - - - - - - - - - - - - - - -
/
-2 000 i
1
1
1
Revenu
$
5 000
20 000 000
10 000 000
0
-2 000 000
10 000 15 000 20 000
Dollars
Probabilit
0.05
0.15
0.30
0.50
3-23
L'investisseur dsire effectuer l'tude en utilisant les mthodes de l'utilit et de la
valeur prvues. Le tableau 3.4 ci-dessous montre ces diffrents calculs.
Tableau 3.4
Revenu Probabilit A *B Utilit B*D
(1 000 000$) du revenu (1 000 000$) du revenu
(A) (B) (C) (D)
20 0.05 1.0 26.3 1.31
10 0. 15 1.5 20.0 3.00
0 0.30 0 1.0 0.30
-2 0.50 -1.0 -10.0 -5.00
Revenu prvu 1. 5 utilit _Qrvue -0.39
3-24
La mthode de l'utilit prvue indique que le projet n'est pas favorable car il a une
utilit prvue ngative alors que l'utilit d'un revenu de 0 gale un. Par contre,
. la mthode de la valeur prvue indique un projet favorable, car elle montre un
revenu prvu suprieur zro.
3.4.2 Estim- variance:
Cette mthode d'analyse conomique, parfois appele mthode quivalente la
certitude, consiste rduire les considrations conomiques en une mesure simple
des bnfices prvus ainsi que de leur variation. Mathmatique, elles 'exprime par
la relation:
V= f.1.- Aa
o V= estim - variance
f.1. = bnfices prvus
a= dviation standard des bnfices
A= coefficient d'aversion du risque
Les experts suggrent de prendre
u"
A= ___ f.l. ou u"
2
correspond
la drive seconde de la fonction utilit versus bnfices. Cette valeur de A
s'applique lorsque l'utilit marginale de l'argent diminue; ce cas se rencontre
souvent en pratique.
3.4.3 L'escompte des variables
Une faon de considrer une augmentation du risque avec des temps de plus en
plus loigns consiste escompter les diffrents revenus possibles par des taux
diffrents; des taux plus levs accorderont une pondration plus faible aux
bnfices les plus loigns. Les experts suggrent des modles mathmatiques pour
tablir les diffrents taux d'intrt, mais ici nous nous limiterons au principe.
3.4.4 Simulation de Monte-Carlo
Mthode dveloppe durant les annes 40 par Von Neuman, Vlan et Fermi pour
rsoudre certains problmes de design d'crans anti-rayonnement. Elle es
coteuse exprimentalement et difficile rsoudre analytiquement.
Elle consiste reprsenter un problme dterministe par un processus stochastique
dont les distributions de probabilit satisfont les relations mathmatiques du
problme dterministe complexe. La premire tape consiste construire un
modle analytique qui reprsente l'investissement considr (exemple une quation
pour la valeur prsente) .
3-25
Dans la deuxime tape, l'analyste dveloppe une distribution de probabilit pour
chaque facteur qui est sujet incertitude dans le modle. Il peut utiliser des
donnes historiques ou une approche subjective.
partir des distributions de probabilits pour chaque facteur dans le modle, Il
gnre au hasard une rponse tentative. La rptition de ce mode d'chantillon-
nage un nombre important de fois, permet de gnrer une distribution de frquence
pour la rponse (ex. Valeur Prsente). On peut utiliser n'importe laquelle des
mthodes de mesure de rentabilit. La distribution de frquence rsultant peut tre
utilise pour obtenir des donnes probabilistiques sur le problme original.
La simulation Monte-Carlo est une technique efficace pour l'analyse conomique
sous risque . Elle permet de varier les facteurs simultanment, en tenant compte,
aussi bien de leur tendance centrale que de leur dispersion, et cela, quelque soit la
distribution de probabilit qui les reprsente. Le chapitre 6 permet d' expliquer plus
en dtail cette technique.
3-26
Problme no 3.1
Des amateurs de tennis dsirent construire deux cours intrieurs au cot de 125
000$. Ce tennis intrieur aura une dure de vie utile de 10 ans et une valeur de
rcupration de 20 000$. Les cots d'entretien sont estims 23 000$ par anne.
S'ils peuvent louer au taux de 7$ par heure par cours, combien d'heures devront-ils
louer chaque anne pour oprer au point mort? Le taux d'intrt bancaire est
actuellement de 10%:
Problme no 3.2
Une ligne de production sera requise encore pour trois ans. Elle a actuellement des
cots annuels de 310 000$; l'on prvoit qu'ils vont demeurer constants. Un
ingnieur industriel a soumis un plan pour une nouvelle conception de la ligne de
production. Cette nouvelle conception coterait 150 000$ installer et il y a 50%
des chances qu'elle permettrait de rduire les cots annuels de fonctionnement
210 000$. Cependant, il y a une probabilit de 25% que les cots annuels de la
nouvelle ligne augmentent de 20 000$ dans chacune des deux dernires annes.
Il y a aussi 25% que l'augmentation soit de 75 000$ dans chacune des deux
dernires annes.
Doit-on conserver la ligne actuelle ou installer la nouvelle?
Problme no 3.3
Considrant la matrice de profits annuels en milliers de dollars ci-dessous:
TATS DE LA NATURE
Actions
SI s2 s3 s4 s5 s6
P=.l . 1 .4 .2 .1 .1
Al 12 5 -8 -3 6 9
A? 7 0 1 5 20 7
Al 3 3 7 9 -5 5
A4 0 12 15 2 8 -200
A ~ -10 22 9 0 4 12
a) Quelle(s) action(s) devriez-vous choisir pour minimiser votre probabilit de
faire des pertes?
3-27
b) Quelle(s) action(s) choisiriez-vous si vous aviez un niveau d'aspiration d' au
moins 9 000$?
Problme no 3.4
Deux (2) projets A et B ont les profits prvus selon les probabilits montres dans
la matrice ci-dessous. Lequel choisiriez-vous et pourquoi?
A Probabilit
B Probabilit
Profits
0.2
0
-11
0.2
0.4
-3
Problme 3.5
0.2
0.3
5
0.2
0.2
13
0.2
-0.1
21
Il existe plusieurs mthodes pour dtecter des soudures imparfaites. Une
compagnie considre actuellement deux mthodes possibles. La mthode 1 cote
0.50$ par inspection et dtecte les dfauts 80% par temps. La mthode 2 cote
2.00$ par test et dtecte toujours une mauvaise soudure. Lorsqu'une soudure
dfectueuse est non dtecte, la compagnie estime 30.00$ les pertes qu'elle doit
encourir pour le remplacement et la mauvaise publicit qui en rsulte. La
probabilit d'avoir une soudure dfectueuse est de 0.05. Utilisant la mthode de
la "valeur prvue", dtermine quelle mthode devrait tre utilise ou est-il
prfrable pour la compagnie de n'en utiliser aucune.
Problme 3. 6
Appliquer les diffrents principes de choix (valeur prvue, l'tat futur le plus
favorable>>, niveau d'aspiration) la matrice suivante; les valeurs de cette matrice
sont des profits.
TATS DE LA NATURE
P
1
=0.1 P
2
=0.7 P
3
=0.2
Actions SI s 2 s3
At
200 10 -60
A2
150 30 0
A3
35 35 35
4.1 INTRODUCTION:
Chapitre 4
TECHNIQUES DE DCISION STATISTIQUE
AVEC INFORMA TI ONS ADDITIONNELLES
Plusieurs techniques statistiques ont t dveloppes pour aider l'investisseur prendre une
dcision. Mme en prsence d'incertitude des informations additionnelles peuvent tre obtenues
par exprience. Dans ce chapitre on verra l'importance d'informations additionnelles sur la prise
de dcision dans le cas de risque.
4.2 Statistique Bayesian
Pour tenir compte des informations additionnelles dans le cas de calcul des probabilits, on fait
appel aux statistiques Bayesian. Les statistiques Bayesian sont caractrises par l'ajustement des
probabilits connues priori pour un paramtre inconnu a des probabilits posteriori plus
certaines. Ces probabilits peuvent tre obtenues de simples vidences ou d'tudes supplmentaires.
4.2.1 Thorme de Bayes
La probabilit que 2 vnements se produisent est donne par la thorie des probabilits.
P(A et B) = P(A/B)*P(B) [quation 4. 1]
ou P(A et B) = probabilit que deux vnements A et B se produisent ensemble
P(A/B) = probabilit que A se produise lorsque B s'est dj produit
P(B) = probabilit que B se produise
On a aussi que: P(A et B) = P(B et A)
Donc P(B et A) = P(B/A)*P(A) et il rsulte que:
P(A/B)*P(B) = P(B/A)*P(A)
Le thorme de Bayes peut tre dduit si l'on assume que P(B) et P(A) * 0
P(AIB)
= P(BIA) P(A)
P(B)
[quation 4.2]
[quation 4.3]
P(BIA)
Exemple 4.1
= P (Al B) P(B)
P(A)
[quation . ]
La probabilit d'obtenir l'as de pique d'un jeu de 52 cartes = 1/52 ou peut tre calcul:
P(as et pique)= P(as/pique) P(pique)
= (1 as/13 piques) (13 piques/52 cartes)
= (1/13) (13/52) = 1152
P(as/pique) = P(pique/as
P(piqu
= (1 pique/4 as) (4 as/52 cartes)
(13 piques/52 cartes)
= (114 (1/13) = 1/13
1/4
Donc, en gnral s'il yan tats de la nature S
1
, S
2
, S
3
, Sn et le rsultat d'une tud
additionnelle tel que chantillonner, est X tel que X soit discrte et P(X) y; 0 e 1
probabilits priori P(S;) onft tablies, alors le thorme de Bayes peut s'crire:
P(XIS;) P(S)
P(SJX) = P(X)
[quation .
La probabilit postrieure P(S/X) est la probabilit d'avoir Si si le rsultat d'une d
additionnelle donne X. La probabilit de X et Si de se produire P(X/S;) P(S;) est la
probabilit conjointe de X et Si . La somme de toutes les probabilits conjointes gale la
probabilit de X. Donc l'quation [4.5] peut s'crire:
P(S/X) =
L P(X/S;) P(S;) [quation .6]
4-3
Le tableau 4.1 reprsente un modle pour 1 'application du thorme de Bayes.
Tableau 4.1: application du thorme de Bayes
1 2 3 4 5
tat de Probabilit Probabilit Probabilit Probabilit
la priori d'obtenir conjointe postrieur
nature 1' chantillon X P(S./X)
SI
s2
S;
sn
P(S
1
) P(X/S
1
) P(X/S
1
) P(S
1
) P(X/S
1
lX(S.
1
}
P(X)
P(S
2
) P(X/S
2
) P(X/S
2
) P(S
2
)
P(X)
P(S;) P(X/S;) P(X/S;) P(S;) P(X/SJ__ffil
P(X)
P(SJ P(X/SJ P(X/Sj P(SJ

P_{X)
n
n n
P(S) =1.
L P(XIS)P(S)=P(X L P(SJX)=
1
i =l i =l
Colonne:
1 S; : les tats de la nature possibles
2 P(S;): la probabilit priori S; estime (la somme gale 1)
3 P(X/S;): la probabilit conditionnelle d'obtenir un chantillon ou des rsultats d'une tude
supplmentaire X connaissant l'tat de natureS;
4 P(X/S;) P(S;): la probabilit commune d'obtenir S; et X; la somme de cette colonne est P(X), la
probabilit que 1 'chantillon ou les rsultats d'une tude supplmentaire soit X.
5 P(S/X): la robabilit postrieure deS;, tant donn qu'une tude additionnelle a produit le
rsultat X; la iime valeur gale la iime entre dans la colonne 4 divise par la somme de la colonne
4.
s.
1
SI
s?
Exemple 4.2
Une production requiert que le procd soit arrang pour fabriquer 200 units. Si les ajustements
sont bien faits, des dfauts dans les units produites se produiront avec une probabilit de o.o-.
Un mauvais ajustement avec une probabilit de 0.2 alors le taux des dfauts dans les units
produites est de 0.25. Posant S
1
pour un bon ajustement, S
2
pour un mauvais et X l'vnemen
qu'un chantillon contienne un dfaut. On dsire calculer la probabilit postrieure que les
ajustements soit bien fait ou mal fait tant donn que X s'est produit.
Solution:
Dans ce cas P(S
1
) = O. 8 et P(S
2
) = 0.2
Le tableau 4.2 rsume les calculs:
P(S) P(XIS) P(X/S)P(S) P(S/X)
0.80 0.05 0.04 4/9 = 0.44
0.20 0.25 0.05 5/9 = 0.56
L
P(Si) = 1.0 L =P(X)= 0.09
L
P(S/X) = 1.00
i i i
Tableau 4.2: rsultat de l'exemple 4.8
Donc la probabilit a priori de 0.8 que l'ajustement soit bien fait est rvise, par le fait qu 'on a
trouv une pice dfectueuse, en une probabilit postrieure de 0.44.
Exemple 4.3
Considrons un investissement ayant un retour (valeur prsente nette) de 6 000$ si l'vnemen
1
se produit et 4 000$ si c'est l'vnement S
2
qui survient. Les probabilits priori sont de 0.4 pour
S
1
et 0.6 pour S
2
Donc le retour prvu E(R):
E(R) =0.4 (6 000$) + 0.6 (- 4 000$) = 0
Supposant que la possibilit de ne pas investir un retour prvu nul; donc il y aura alors ni perte
ni gain.
4-5
Un tude additionnelle permettra d'obtenir soit X
1
montrant un retour de 6 000$ ou X
2
pour une
perte de 4 000$. Si S
1
se produit, X
1
sera indiqu avec une probabilit de 0.8; de mme pour S
2
avec une probabilit 0.6.
Le problme se rsume ainsi:
E(S
1
) = 6 000$
E(S
2
) = - 4 000$ P ( X / S ~ = 1.0 - P ( X
2
/ S ~ = 0.4
Les probabilits postrieures rsultantes de l'tude additionnelle peuvent tre calculs et la dcision
d'investir ou non peut maintenant tre dtermine en fonction du rsultat X de l'tude additionnelle.
Probabilits postrieures tant donnes que X
1
se produit.
s.
1
P(SJ P(X/ SJ P(X/SJP(SJ P(S/X
1
)
SI 0.4 0.8 0.32 0.32/0.56=0.57 _,
s2
0.6 \ 0.4 0.24 0.24/0.56=0.43
I:=
1.0 L =P(X) = 0.56
I:
1.00
Quand xl se produit la probabilit d'avoir SI change de 0.4 (priori) 0.57 (posteriori). Le retour
prvu devient alors:
E(R/X
1
)=0.57 (6,000$) + 0.43 (-4,000$) = 1 714$
Donc si X
1
se produit, le projet devra tre entrepris pour obtenir un retour de 1 714$.
Probabilit postrieures tant donn que X
2
se produit:
si P(SJ P(XiSJ
s, 0.4 0.2
s2 0.6 0.6
L: = Lo
P(X
2
1SJP(SJ
0.08
0.36
L: = o.44
P(S/X
2
)
0.18
0.82
L: Loo
Il rsulte une probabilit de 0.18 d'obtenir S
1
du fait que X
2
se produit; le retour prvu de vien
alors;
E(R/X
2
)=0.18 (6 000$) + 0.82 (-4 000$) = -2 182$
Comme E(R/X
2
) est ngatif on prfrera ne pas investir si X
2
se produit prfrant un retour de zro.
Donc on aura:
E(R)=
[
0 si X=X
2
1 714$ si X=X
1
Comme P(X
1
) =0.56 et P(X
2
) = 0.44l'tude additionnelle donnera le retour suivant.
E(R) = E(R/X
1
) P(X
1
) + E(R/X
2
) P(X
2
)
= (1 714$) (0.56) + (0) 0.44 = 960$
Donc en gnral le .retour prvu suite une tude additionnelle (ou chantillonnage) Xj est:
E ( R 1 s 1) = ~ max [E (actions) 1 xi] P(Xj)
J
ou E(R/SI) est le retour prvu tant donn les informations de l'chantillon.
Le changement de la valeur prvu de 0 960$ s'appelle souvent la valeur prvue provenant de
l'chantillon (VPPE) qui se reprsente:
VPPE = E(R/SI) - E(R)
4-7
Exemple 4.4
Un lot de cinq articles doit tre accept (probabilit a
1
) ou rejet L'tat de la nature
est caractrise par le nombre de probabilits. M d'articles dfectueux parmi les 5; donc M=O,
1, 2, 3, 4 ou 5. Pour simplifier, supposons que les pertes galent deux fois le nombre d'articles
dfectueux (2M) dans un lot qui est vendu; et gal le nombre de bons articles dans un lot qui est
rejet (5-M). Dterminez la possibilit Bayes si les probabilits priori sont P(O) = 0.6, P(1) =
0.2, P(2) = P(3) = 0.1, P(4) = P(5) =O.
Solution
Le tableau 4.3, ci-dessous, rsume les donnes
Tableau 4.3
Donnes pour l'exemple 4.4
Nombre d'article
dfectueux S; = 0 1 2 3 4 5
Action: a
1
(accept)= 0 2 4 6 8 10
Action: (rejet)= 5 4 3 2 1 0
P(S) ( priori)= .6 .2 .1 .1 0 0
Calculons les pertes si l'on accepte et rejete respectivement les lots.
B(a
1
) = 0 * 0.6 + 2 * 0.2+ 0.1 * 4 + 0.1 * 6 + 0 * 8
=1.4
B(a
2
) - 5 * 0.6 + 4 * 0.2 + 3 * 0.1 + 2 * 0.1 + 1 * 0 + 0 * 0
= 4.3
Donc, accepter les lots entranera une perte moindre
Maintenant, partir des donnes prcdentes et des probabilits conditionnelles pour les donnes
X tant donnes et l'tat de la nature M ci-dessous
P(X/M) =
M
5
si X = 1 (mauvais)
1 - M
5
si X = 0 (bon)
Calculons successivement les probabilits posteriori et la nouvelle action Bayes.
Sachant que P(X et M) = P(X/M) P(M)
X\M 0 1 2
(bon) 0 1 (.6) = .6 4/5 (.2) = .16 3/5 (.1) = .06
!(mauvais) 1 0 (.6) = 0 1/5 (.2) = .04 2/5 (.1) = .04
X\M 4 5
(bon) 0 1/5 (0) = 0 0 (0) = 0
!(mauvais) 1 l4-!5 (0) = 0 1 (0) = 0
Donc, P(X = 0) = 0.6 + 0.16 + 0.04 + 0 = 0.86
P(X = 1) = 0.04 + 0.04 + 0.06 + 0 = 0.14
P(X et M)
Calculant les probabilits posteriori : P(M/X)
P(X)
X\M 0 1 2 3 4
3
2/5 (.1) = .04
3/5 (.1) = .06
5
0 0.6 30 0.16 8 0.06 3 0.04 2 0 0
--- --=- ---- ---- -=0 -=0
.86 43 0.86 43 0.86 43 0.86 43 0.86 0.86
1
0 0.04 2 0.04 2 0.06 3 0 0
-=0 ---- --- - - --- -=0 -=0
0.14 0.14 7 0.14 7 0.14 7 0.14 0.14
4-8
On dsire avoir le minimum de perte; les pertes peuvent tre calcules pour X = 0 c'est--dire
lorsque l'chantillon test est bon. Donc, les pertes totales galent la somme des pertes pour les
diffrents tats de la nature.
30 . 8 3 2 40
=- (0)+- (2)+- (4)+- (6)+ 0 (8)=-
43 43 43 43 43
Donc, si on prend la possibilit a
2
c'est--dire si on rejette le lot lorsque l'chantillon test est bo
(X =0) les pertes totales seront:
30 8 3
=5 * - + 4 *- + 3 * - + 2
2 195
*- --
43 43 43 43 43
40 195
Donc comme - <-
43 43
pour X = 0, la possibilit Bayes est a
1
En d'autres mots, lors ue
l'chantillon test est bon, on accepte le lot. De la mme faon si l'chantillon test est mau ais.
on peut calculer les pertes pour les 2 possibilits a
1
et a
2
Pertes pour action a
1
:
4-9
2 2
+ 6 *
3 30
=0 * 0 + 2 * - + 4
*
- -
7 7 7 7
Pertes pour action a
2
:
2
+ 3 *
2
+ 2 *
3 20
=5 * (0) + 4 * -
--
7 7 7 7
20 30
Donc, comme - <-
7 7
, la possibilit Bayes est az pour X = 1, on rejette le lot si
l'chantillon test est mauvais de faon minimiser les pertes.
4.2.2 Mthode de Bayes applique un chantillonnage
Nous avons vu prcdemment qu'une compagnie pouvait raliser une conomie, en testant
un chantillon dans un lot de cinq. Maintenant le management dsire savoir s'il peut
raliser une conomie supplmentaire en testant un deuxime chantillon.
Dans ce cas les probabilits posteriori trouves prcdemment aprs avoir test un
chantillon deviendront les probabilits priori avant de tester le deuxime chantillon. Le
tableau ci-dessous montre ces probabilits.
X\M 0 1 2 3 4 5
0 30 8 3 2
0 0 - - - -
43 43 43 43
1
2 2 3
0 0
0
- - -
7 7 7
Considrant maintenant qu'un chantillon a dj t test et qu'il reste seulement quatre
chantillons dans le lot. Les relations ci-dessous caractriseront 1 'tat de la nature lorsque
le testeur aura trouv un premier chantillon bon.
P(X/M)=
M
4
si x2 = 1 (mauvais)
M
P(X/M)= 1 - - si X
2
= 0 (bon)
4
Par contre lorsqu'il aura trouv un premier chantillon mauvais les relations suivantes
caractriseront 1' tat de la nature.
-10
P(X/M)=
M- 1
si X
2
= 1 (mauvais)
4
P(X/M)= 1
M-1
si X
2
= 0 (bon)
---
4
M a t dfinit comme le nombre de mauvais chantillon dans le lot initial.
partir de ces relations calculons en premier les probabilits conjointes en utilisant la
relation P (X et M) = P(X/M) P (M) lorsque le testeur a trouv un premier chantillon bon.
X\M 0 1 2 3 4
0 (bon)
1
30 3 8 1 3 1 2 0 * 0
*-
-
* - - *-
-
*-
43 4 43 2 43 4 43
0.70 0.14 0.035 0.011
0
1 (mauvais)
0
30 1 8 1 3 3 2
0 * 1
*- - *- - *- - *-
43 4 43 2 43 4 43
0 0.046 0.035 0.035
0
tant donn que le testeur a trouv un premier chantillon bon:
La probabilit que le deuxime chantillon soit bon gale:
P(X
2
=0) = 0.70 + 0.14 + 0.035 + 0.011 = 0.885
De mme, la probabilit que le deuxime chantillon soit mauvais gale:
P(X
2
= 1) = 0.046 + 0.035 + 0.035 = 0.115
' 1
ons1 erons mamtenant e cas ou
, h 11
e testeur a trouve un premier ec antl on mauv ais.
X\M 0 1 2 3 4
0 (bon) 0 * 0
1
2 3 2 1 3 1
* 0 *- - *- - *
- -
0
7 4 7 2 7 4
0.286 0.214 0.214
0
1 (mauvais) 0 * 0 2 1 2 1 3 3
* 0 0* - -
*-
-
* -
-
0
7 4 7 2 7 4
0 0.071 0.214
0
4-11
tant donn que le testeur a trouv un premier chantillon mauvais:
La probabilit que le deuxime chantillon soit bon gale:
P(X
2
=0) = 0.286 + 0.214 + 0.214 = 0.714
De mme, la probabilit que le deuxime chantillon soit mauvais gale:
P(X
2
= 1) = 0.071 + 0.214 = 0.286
On peut maintenant calculer les probabilits posteriori que le deuxime chantillon soit
bon ou mauvais tant donn que le testeur a trouv un premier chantillon bon.
X\M 0 1 2 3 4
0 (bon) .70 . 14 .035 .011 0
-- - . -
-- --
.885 .885 .885 .885
0.79 0.158 0.0395 0.0124
1 (mauvais) 0 .046 .035 .035 0
-- -- --
.115 .115 . 115
0
0.40 0.30 0.30
0
Calculons de la mme faon pour le cas o le testeur a trouv un premier chantillon
mauvais.
X\M 0 1 2 3 4
0 (bon) 0 .286 .214 .214 0
-- -- --
.714 .714 .714
0.400 0.300 0.300
1 (mauvais) 0 0 .071 .214 0
- - - -
.286 .286
0 0
0.249 0.748
Considrons maintenant la logique des rsultats. Lorsque le testeur a test un seul
chantillon et qu'ill'a trouv bon, il rsultait une probabilit posteriori de 3/43 (0.0698)
d'avoir deux chantillons de mauvais dans le lot. Si maintenant il teste un deuxime
chantillon et qu'ille trouve encore bon, la probabilit posteriori deviendra (.0395). Il
est donc normal que la probabilit de trouver deux chantillons mauvais dans le lot diminue
du fait qu'on en a trouv un deuxime de bon. On pourrait dmontrer un raisonnement
similaire pour les autres probabilits.
-1
Connaissant les probabilits posteriori, calculons les pertes lorsque le testeur trou e u
premier chantillon bon et un deuxime:
a) bon
accepter le lot: 0.79 * 0 + 0.158 * 2 + 0.0395 * 4 + 0.0124 * 6
: 0 + 0.316 + 0.158 + 0.0744 = 0.548
rejeter le lot: 0.79 * 5 + 0.158 * 4 + 0.0395 * 3 + 0.0124 * 2 +0
: 3.95 + 0.632 + 0.118 + 0.024 = 4.72
Le management minimise ses pertes en acceptant le lot.
b) mauvais
accepter le lot: 0.40 * 2 + 0.30 * 4 + 0.30 * 6
: 0.80 + 1.20 + 1.80 = 3.80
rejeter le lot : 0.40 * 4 + 0.30 * 3 + 0.30 * 2
: 1.60 + 0.90 + 0.60 = 3.10
Le management minimise ses pertes en rejetant le lot.
Maintenant calculons les pertes lorsque le testeur trouve un premier chantillon mau ais et
un deuxime:
a) bon
accepter le lot: 0.40 * 2 + 0.30 * 4 + 0.30 * 6
: 0.80 + 1.20 + 1.80 = 3.80
rejeter le lot : 0.40 * 4 + 0.30 * 3 + 0.30 * 2
: 1.60 + 0.90 + 0.60 = 3.10
Le management minimise ses pertes en rejetant le lot.
b) mauvais
On peut immdiatement prvoir que le management rejettera le lot puisqu'ille fai
lorsque le testeur trouve un deuxime chantillon bon.
accepter le lot: 0.249 * 4 + O. 748 * 6
: 0.996 + 4.48 = 5.48
rejeter le lot : 0.249 * 3 + O. 748 ~ 2
4-13
: 0.747 + 1.49 = 2.24
Le management minimisera ses pertes en rejetant le lot.
Donc, pour minimiser l'ensemble de ses pertes, le management adoptera comme
stratgie de contrle:
Rejeter le lot lorsque le testeur trouve un premier chantillon mauvais et ne pas en
tester un deuxime. S'il trouve un premier chantillon bon il en teste un deuxime;
un mauvais chantillon entranera, le rejet du lot alors qu'un bon chantillon son
acceptation.
Prcdemment l'acceptation du lot sans aucun test entranait des pertes de 1.4. Le
test d'un chantillon permettait de rduire les pertes 1.2. Le test d'un deuxime
chantillon et l'application de la stratgie de contrle ci-dessus entranera des pertes
de:
0.86 (0.885 * 0.548 + 0.115 * 3.10) + 0.14 * 2.85
0.86 (0.485 + 0.359) + 0.40 = 1.13
Donc le test d'un deuxime chantillon permettra de rduire les pertes de 1.20- 1. 13
= 0.07
Dpendamment des cots pour effectuer ces tests le management dcidera
d'effectuer, aucun test un ou deux chantillons par lot.
4.2.3 L'utilisation de donnes dans la prise de dcision
Par donnes, on entend les rsultats d'une ou plusieurs observations que l'on crot
intimement relis l'tat de la nature. La disponibilit de telles donnes va fournir une
certaine illumination sur l'tat de la nature facilitant ainsi le choix d'une action. Une
utilisation intelligente de ces donnes va permettre d'obtenir des pertes infrieures celles
qu'on auraient eues sans leur utilisation. Pour obtenir les donnes, on doit effectuer une
exprience (tude) dont les rsultats doivent tre diffrents selon les divers tats de la nature.
Exemple 4.5
Supposons qu'une personne entrant dans un difice peut voir les lumires sur les planchers
infrieurs et suprieurs d'un ascenseur. Lorsque quelqu'un demande l'ascenseur la lumire,
s'allume et s'teint ds que la porte de l'ascenseur s'ouvre. Les donnes peuvent tre
exprimes ainsi: x = nombre de lumires allumes.
x= 0 :
x= 1:
x= 2:
personne n'a demand l'ascenseur.
l'ascenseur a t demand sur un des planchers et peut ou non tre e
mouvement.
l'ascenseur a t demand sur les deux tages (sous-sol et premier tage)
peut alors penser que l'ascenseur ne fonctionne pas puisqu'elle a t demand
depuis un certain temps.
Supposons que l'on connaisse les probabilits pour x = 0, 1, 2 pour chaque tat de la na
et le tableau des pertes. Dans ce cas particulier, on pourrait assimiler les pertes l 'ner e
ncessaire pour monter un escalier. Plus on devra monter de niveau plus on perdra de
l'nergie.
Actions
Tableau des pertes
tats de la nature
Fonctionne Ne fonctionne
pas
A
1
(descendre au sous-sol)
A
2
(monter au premier)
0 6
1 5
Probabilit pour chaque tat de la nature
SI s2
x Fonctionne Ne fonctionne pas
0 .6 .1
1 .3 .4
2 .1 .5
On peut se demander d'o peuvent provenir de telles informations?
Il y a deux sources possibles:
. a) par observation sur une longue priode de temps.
b) de considrations thoriques; exemple; un modle mathmatique qui tiendrait compte
du dbit de personnes entrant dans l'difice et celles utilisant l'ascenseur. On note
qu'on doit aussi faire des dterminations pour tablir les paramtres du modle.
4-15
Les probabilits postrieures utilisant la relation de Bayes, sachant par 1 'exprience passe
que l'ascenseur fonctionne 7 fois sur 10, c'est--dire P(S
1
) = 0.7, P(S:z) = 0.3
P(SJX)= P(X/S) P(Si)
P(X)
tat de la nature
x=O
fonctionne S
1
:
(.6)(.7) 14
--
(.6)(. 7) + (.1)(.3) 15
ne fonctionne pas S
2
:
(;1)(.3) 1
--
(.6)(.7) + (.1)(.3) 15
x = 1 x=2
(.3)(.7) 7 (.1) (. 7) 7
-- --
(.3)(.7) + (.4)(.3) 11 (.1)(.7) + (.5)(.3) 22
(.4)(.3) 4 (.5) (.3) 15
-- -
(.3)(.7) + (.4)(.3) 11 (.1)(.7) + (.5)(.3) 22
Il nous faut maintenant une rgle pour prendre une dcision. Si la personne entrant dans
1' difice trouve deux lumires allumes
X=2, il sait alors que P(S/2) = 7/22 et P(S
2
/2) = 15/22.
Se servant de ces valeurs les pertes postrieurs peuvent alors tre calcules:
La possibilit a
1
: descendre =
7 15 90
0*-+6*-=-
22 22 22
7 15 82
1*-+5*-=-
22 22 22
La monter =
Donc, pour X = 2 les pertes sont infrieures en prenant la possibilit a
2
c'est--dire de
monter.
De mme pour X = 1
La possibilit a
1
: descendre =
7 4
0*-+6*-
11 11
24
11
La possibilit a
2
: monter =
7 4
1*-+5*
11 11
27
11
Donc, pour minimiser les retards on choisit a
1
, descendre
Pour X= 0
ai : 0 * 14 + 6
15
1
*-
15
6
15
14 19
~ : 1 * - + 5 *
15 15 15
1
Donc la rgle sera de descendre au sous-sol s'il n'y a pas de lumire allume ou une seule
lumire et de monter au premier tage si les deux lumires sont allumes.
Donc en utilisant les probabilits priori que 1' ascenseur fonction c'est--dire O. 7, on aura
comme perte prvue:
Si on descend: E = 0.7 * 0 + 0.3 * 6 = 1.8
Si on monte: E = 0.7 * 1 + 0.3 * 5 = 2.2
Donc, si on descend toujours, on aura une perte de 1. 8 qui est une possibilit Bayes qui
minimise les pertes.
Maintenant, si on utilise les rsultats de 1 'tude, quelle sera alors la perte?
Comme P(X) = E P(XIS) P(S)
i
P(X = 0) = 0.6 * 0.7 + 0.1 * 0.3 = 0.45
P(X = 1) = 0.3 * 0.7 + 0.4 * 0.3 = 0.33
P(X = 2) = 0.1 * 0.7 + 0.5 * 0.3 = 0.22
Total = 1.0
- (.45) +
24
(.33) +
82
(.22) = 1
11 22
Donc, dans ce cas, le prix que rapporte l'chantillonnage est de:
1.80- 1. 72 = 0.08
4-17
On peut noter que ce problme pourrait tre solutionner en considrant l'ensemble des rgles
possibles. Dans ce problme, il y a deux possibilits: a
1
: descendre et a
2
monter. De plus,
il y a trois valeurs possibles observes: X = 0, 1, 2; donc, il y a 2
3
c'est--dire 8
combinaisons ou rgles possibles (d) de dcisions.
x
0
1
2
En calculant les pertes pour chaque rgle possible et en choisissant celle qui donne la plus
faible perte, on obtiendrait le mme rsultat que prcdemment. Mais, on peut immdiate-
ment voir que la rgle d
3
signifie: que si on observe 0 ou 2 lumires allumes on choisit la
possibilit a
2
, c'est--dire monter et la possibilit a
1
descendre si une seule lumire est
allume. Les rgles d
1
et d
8
ignorent les donnes. On remarque aussi que les rgles d
4
et
d
5
sont opposes.
4-17
On peut noter que ce problme pourrait tre solutionner en considrant 1' ensemble des rgles
possibles. Dans ce problme, il y a deux possibilits: a
1
: descendre et a
2
monter. De plus,
il y a trois valeurs possibles observes: X = 0, 1, 2; donc, il y a 2
3
c'est--dire 8
combinaisons ou rgles possibles (d) de dcisions.
x
0
1
2
En calculant les pertes pour chaque rgle possible et en choisissant celle qui donne la plus
faible perte, on obtiendrait le mme rsultat que prcdemment. Mais, on peut immdiate-
ment voir que la rgle d
3
signifie: que si on observe 0 ou 2 lumires allumes on choisit la
possibilit a
2
, c'est--dire monter et la possibilit a
1
descendre si une seule lumire est
allume. Les rgles d
1
et d
8
ignorent les donnes. On remarque aussi que les rgles d
4
et
d
5
sont opposes.
4-18
Problme 4.1
La compagnie Papier dsire augmenter sa production; elle considre actuellement les trois choix
suivants:
cl: faire un grand agrandissement
C
2
: faire un petit agrandissement
C
3
: louer les facilits additionnelles ncessaires
La compagnie demeure incertaine sur son choix, car la demande pour ses produits pourra tre:
leve (S
1
), bonne (SJ, normale (S
3
) ou faible (S
4
). Son dpartement de marketing value les
revenus et probabilits relis chaque niveau de demande de la faon suivante:
Revenus et probabilits de la demande
(Revenus en million de dollars)
Choix leve(S
1
) bonne(SJ normale (S
3
) faible (S
4
)
P(Sl) 0.2 P(S
1
) 0.45 P(S
1
) 0.25 P(SJ_ 0.1
Grand agrandissement (C
1
) 18 12 6 -12
Petit agrandissement (C
2
) 12 12 9 -6
Location (C
1
) 13 10 6 -1
Considrant sa difficult arrter son choix, la compagnie dcide de faire appel un consultant.
Ce dernier s'engage prdire trois niveaux .de la demande avec certaines probabilits; le tableau
suivant montre ses prdictions.
Les probabilits de la prdiction du consultant
Demande future leve-bonne bonne-normale normale-faible
xl x2 x3
~ l e v e S
1
0.6 0.3 0.1
bonne s2 0.3 0.5 0.2
normale s3 0.1 0.3 0.6
faible S
4
0.1 0.1 0.8
titre d'exemple si le consultant prdit une demande leve-bonne dans 60% des cas cette
demande se rvlera dans les faits leve P (X
1
1 S
1
) = 0.6
a) Dterminer les revenus prvus pour chacun des choix avant de faire appel au consultant?
4-19
b) Dterminer les probabilits postrieures l'tude du consultant?
c) Dterminer les revenus prvus aprs l'tude du consultant?
d) Quel est le montant maximal que la compagnie Papier voudra payer le consultant?
Problme no 4.2
Un tudiant possde trois (3) ds; un vert, un rouge et un blanc. Le d vert a cinq (5) faces
marques H et une face marque T, de mme, le d rouge a deux (2) faces marques H et quatre
(4) faces marques Tet le blanc trois (3) faces marques H et trois (3) faces marques T.
L'tudiant tire les trois (3) ds sans que vous puissiez les voir. Il en cache deux (2) et vous informe
que celui non cach montre la face H. Dmontrez l'aide du thorme de Bayes le choix de la
couleur du d que vous feriez pour maximiser vos chances.
Problme no 4.3
Lorsqu'une machine est ajuste correctement 9 sur 10, des articles proluits sont acceptables. Lors
d'un mauvais ajustement, il y a 0.4 chance de produire des articles acceptables. La probabilit que
la machine soit bien ajuste est de 0.95.
a) Si le premier article test aprs un ajustement est non acceptable, quelle est la probabilit
que la machine soit bien ajuste?
b) Si les deux premiers articles tests sont acceptables, quelle est la probabilit que la machine
soit bien ajuste?
Problme 4.4
Nous avons vu, aux cours, l'application de la mthode de Bayes l'chantillonnage (voir notes).
Avant de prendre une dcision finale sur le nombre d'chantillons qu'il faudra tester, votre patron
dsire connatre quelques informations supplmentaires; veuillez lui fournir ces informations.
tant donn que le testeur a trouv un premier chantillon bon et un deuxime mauvais:
a) Quelles seront les relations qui caractriseront 1 'tat de la nature?
b) Quelles seront les probabilits a priori avant de tester le troisime chantillon?
c) Quelle est la probabilit que le troisime chantillon soit bon?
4-20
d) Dterminer les probabilits a posteriori que le troisime chantillon soit bon.
e) Sans rien calculer, si les cots pour tester ce troisime chantillon sont nuls, quelle est
l'conomie maximale qu'on pourra raliser en le testant? (maximum 5 lignes)
Problme 4.5
La socit lectronique doit dcider si elle va produire un nouveau capteur; elle devra alors investir
5 $ millions. La demande pour ce capteur n'est pas connue. Si la demande tait leve, la socit
ferait un profit de 2 $ millions par anne durant cinq (5) ans. Si la demande est modre, le profit
serait de 1,6$ million par anne durant quatre (4) ans. Pour une demande faible, le profit ne serait
que de 0,8 $ million par anne durant quatre (4) ans.
On estime 10% les chances d'une faible demande et 50%, celles pour une demande leve. Le
taux d'intrt est actuellement de 10%.
a) Utilisant la mthode de la valeur prsente, la socit doit-elle faire l'investissement?
b) L'un des cadres dsire que l'on fasse une tude de march; quel est le montant maximal que
l'on pourrait tre prt dpenser pour cette tude?
c) Un consultant propose une tude au cot de 75 000$ qui a les caractristiques suivantes?
Si la demande est
leve
modre
faible
L'tude indiquera favorable ou non-favorable
avec les probabilits
favorable '&.
0.9
0.5
0
non-favorable
0.1
0.5
1.0
Est-ce que la socit doit demander l'tude?
d) Si l'tude indique non-favorable, la socit doit-elle produire le capteur? Quel est le profit
prvu dans ce cas?
4-21
Problme 4.6
Un diteur d'une maison d'dition doit dterminer s'il va accepter ou non un manuscrit pour
publication. Il a dj dpens 1 000$ pour le dveloppement de ce manuscrit. Il a le choix entre:
- A 1: rejeter le manuscrit et perdre 1 000$
-A 2: accepter le manuscrit sans consulter un expert extrieur
- A 3: obtenir l'avis d'un expert extrieur au cot de 800$
Les donnes sont rsumes dans le tableau ci-dessous:
Dcision et rsultats sivaluation de l'diteur valuation de l'expert Retour si
2ublis {10 c a s ~ ~ 2 0 c a s ~ _Qubli
Mauvais Acceptable Trs bon $
MO : pas publier 4 9 0 0 0
Ml : demande faible 3 3 2 0 -10 000
M2 : bonne vente 2 1 1 1 50 000
M3 : grand succs 1 0 2 1 200 000
Il y a trois niveaux de succs possibles si le livre est publi: Ml, M2 et M3. La dernire colonne
donne la valeur prsente des revenus moins les cots de publication pour chaque condition de
march. Le tableau indique aussi les dcisions antrieures prises aprs l'avis de l'expert. L' diteur
a accept 6 des 10 derniers manuscrits soumis; un a eu un trs grand succs. L'expert en a class
20 comme mauvais, acceptable et trs bon et charge 800$ par tude.
a) Quel est le profit prvu si l'diteur dcide sans consulter l'expert?
b) Quel est le profit prvu aprs consultation de l'expert?
c) Quel est le montant maximal que l'diteur sera prt payer pour l'expert?
Problme 4.7
Un inventeur a dvelopp un instrument pour mesurer le taux de bchettes dans le papier. Une
compagnie considre actuellement la possibilit d'acheter les brevets relis cet instrument pour
un montant de 40 000$. Elle pense qu'il y actuellement une chance sur cinq que l'instrument puisse
tre commercialis avec succs; dans un tel cas, elle obtiendrait des revenus nets de 150 000$ par
anne durant les cinq prochaines annes. Dans les autres cas l'instrument ne rapportera aucun
revenu. Cette compagnie utilise un taux d'intrt annuel effectif de 15%.
4-22
a) La compagnie doit-elle l'acheter?
b) Si un dpartement de marketing peut fournir une information avec certitude du march
futur, combien la compagnie pourrait tre prte payer pour ce service?
c) Ce mme dpartement de marketing peut aussi faire une tude de march avec une
probabilit de 0.7 que si le rsultat de l'tude s'avre positif la commercialisation sera un
succs et une probabilit de 0.9 que si l'tude montre un rsultat non-favorable l'instrument
ne pourra tre commercialis. Dterminer combien la compagnie voudra payer pour cette
dernire tude?
Chapitre 5
V ARBRE DE DCISION
Dans les problmes dcisionnels l'analyste doit reconnatre que des dcisions prises immdiatement
influenceront directement celles qui devront tre prises dans le futur. Trs souvent les industriels
prennent des dcisions sans considrer les effets long terme; une dcision qui semble bonne
actuellement pourrait placer la compagnie dans une position dfavorable et parfois non comptitive par
rapport d'autres prendre dans l'avenir. L'utilisation des arbres de dcision s'avre un moyen trs
efficace pour reprsenter l'interaction entre une squence de dcision. Ce diagramme des vnements
futurs qui pourraient influer sur la dcision prsente permet d'clairer le dcideur sur les choix faire
en prsence de risques pour atteindre les objectifs fixs.
L'arbre de dcision explicite graphiquement la squence des dcisions prendre et les divers
vnements qui peuvent arriver. Il montre les mmes informations qu'un tableau de cots mais d'une
faon beaucoup plus clairement, particulirement dans le cas de situations complexes.
Pour Belzile et al (1) la procdure consiste :
bien se situer dans le temps;
dfinir les vnements possibles qui peuvent arriver (tenir compte de l'chelle de temps);
dterminer les actions qui peuvent tre entreprises;
dterminer la valeur en dollars ou en utilit de chaque action combine avec l'vnement;
associer chaque vnement une probabilit d'occurrence;
trouver la valeur espre de chaque solution possible;
choisir la possibilit qui offre le rsultat espr le plus lev (en dollars ou en utilit).
Graphiquement l'arbre se compose d'une srie de noeuds et de branches. Chaque branche reprsente
une action possible. Le noeud localis l'extrmit gauche de chaque branche reprsente une chance
que l'vnement ou l'tat de la nature se produise. Toute autre branche la droite du noeud reprsente
d'autres actions possibles associes aussi un tat de la nature reprsent par le noeud. Un cot, un
profit ou une utilit correspond chacune des actions reprsentes; il est indiqu l'extrmit droite
de chacune des brariches terminales.
Dans le prsent texte les carrs reprsentent les points de dcision et les cercles les points de chance.
Un arbre de dcision comprendra donc toujours une combinaison d'actions possible entirement sous
le contrle du dcideur relies par des points de chance, lesquels sont soumis aux des lois des
probabilits dont la dcision ne possde aucun contrle sur leur occurrence.
L'arbre de dcision contiendra seulement les actions et les tats de la nature qui ont une importance
pour l'investisseur et dont les consquences dsirent tre compares. De plus, cette approche assume
comme hypothse qu'aucun changement important n'influencera les diffrentes situations dans le
temps.
5-2
Un arbre de dcision ne donne pas au dcideur la rponse un problme dcisionnel mais l ' aide plut
dterminer quelle action en un point particulier du temps va produire l'utilit ou le profit le plus
lev. Il aide ainsi visualiser l'information pour prendre une dcision. Le dcideur devra toujours
considrer les gains prvus en fonction des risques encourir. La nature de ces risques, dpendant
de la faon qu'illes percevra, influencera non seulement les hypothses qu'il a faites mais aussi la
stratgie suivre pour en tenir compte.
La mthode de construction d'un arbre de dcision peut ~ e rsume ainsi :
1. identifier les points de dcision et les actions prsentes chacun des points;
2. identifier les points de risque et la valeur des revenus ou utilits qu'engendreront les actions
chaque point;
3. estimer les valeurs ncessaires pour faire l'analyse; plus spcifiquement dterminer les
probabilits des diffrents tats de la nature, les cots, les gains ou les utilits associs
chacune des actions;
4. analyser les valeurs des choix pour prendre une dcision.
Solution d'un arbre de dcision
En pratique, l'analyste utilise une procdure contre courant pour analyser le problme reprsent par
un arbre de dcision. Il effectue l'analyse en partant aux bouts des branches et en se dplaant ers
le tronc en appliquant les deux rgles suivantes :
1. s'il rencontre un point de chance, c'est--dire associ un tat de nature, il calcule la valeur
prvue du cot, du gain ou de l'utilit en se basant sur les valeurs places la droite du point;
2. s'il rencontre un point de dcision, c'est--dire une action prendre, il choisit le profit ou
l'utilit maximale ou les cots minimaux en considrant les branches de droite adjacentes au
point.
En procdant de cette faon, il peut liminer certaines actions de considration future; ceci est
particulirement important dans le cas de situation complexe qui exige la construction d'arbre trs
grand et l'utilisation d'ordinateurs pour leur solution.
Rfrence
1. Belzile R., Mercier G., Rassif F., <<Analyse et gestion financire Presse universit du Qubec,
Qubec 1989.
5-3
Arbre d dcision
Exemple
La socit PAPIER VERT dsire fabriquer un nouveau type de papier sanitaire, la fois
plus conomique et plus cologique, partir d'un nouveau procd de fabrication de pte
mcanique. Malgr les tudes techniques et l'analyse de march effectues, la socit
n'est pas certaine si la qualit de son produit permettra un volume de vente important.
La socit considre actuellement deux possibilits; la premire consiste construire un
atelier de fabrication de pte mcanique de 500 t/d et de tenter de prendre une part
importante du march. La deuxime possibilit serait de construire une usine de
seulement 250 t/d et de faire une nouvelle tude aprs deux ans. Si la demande est
suffisamment leve, la socit construirait une autre unit de 250 t/d pour minimiser
l'influence de ses cots indirects de fabrication et maximiser ses profits.
Par contre, si la demande pour ce nouveau produit tait faible, la socit se retirerait du
march et rorienterait la fabrication de ses 250 t/d vers la fabrication de papiers ayant
une valeur ajoute infrieure.
Les rsultats de l'analyse de march font voir qu'il y a 60 % de chance que la demande
initiale soit leve et demeure leve durant vingt ans et 10 % de chance qu'elle soit
initialement leve et qu'elle devienne faible aprs deux ans; donc, il y a 70 %de chance
que la demande initiale soit leve pour ce type de papier et 30 % qu' elle soit faible dans
deux ans. Il y a 86 % (60/70) des chances qu'une demande leve resterait leve et
seulement 14% ( 1 0/70) qu'elle devienne faible. Si, avec toute la publicit qui entourerait
le lancement de ce nouveau produit, la demande initiale tait faible, elle demeurerait
certainement faible.
Pour sa part, le dpartement de gnie industriel de la socit estime que les cots de
construction seraient respectivement de trente et soixante millions de dollars pour un
atelier de 500 et 250 t/d. L'addition de 250 t/d, aprs deux ans, coterait vingt millions
de dollars.
Le tableau 1 fait voir l'estim des profits selon que l'on aurait une capacit de production
de 250 et 500 t/D et que la demande soit leve ou faible.
La figure 1 illustre, sous la forme d'un arbre de dcision, l'ensemble des informations.
Cette reprsentation n'apporte par de solution particulire au problme de la socit
PAPIER VERT, mais elle aide ses dirigeants visualiser la suite des dcisions prendre.
Cette socit exige un taux de retour minimum de 10 5/6 sur cet investissement qui
permettrait de consolider sa position sur les marchs internationaux.
5-4
L'analyse utilise la procdure contre-courant pour solutionner cet arbre, en commenant
au point de dcision 2 et en attribuant une valeur montaire qui lui permettra de prendre
une dcision au point 1. Le tableau 2 illustre l'analyse de la dcision au point 2 et fait
voir les valeurs prsentes prvues.
Donc, si la socit dcide maintenant, c'est--dire au point 1, de construire l'atelier de
250 t/d, elle tiendra compte que l'addition de 250 t/d aprs deux ans lui procurerait un
profit net de quarante-quatre millions de dollars.
La figure 2 illustre l'arbre de dcision simplifi et ne contenant plus que le point de
dcision 1. Multipliant les valeurs prsentes des revenues prvues par les probabilits.
L'analyste obtient les profits pour la construction d'une usine de 500 t/d et d'une usine
de 250 t/d et cela, aprs que la dcision au point 2 a t prise de procder une addition
de 250 t/d dans les deux ans. Le tableau 3 illustre les calculs des valeurs prsentes
prvues au point de dcision 1.
La construction immdiate d'un atelier de 500 t/d entranerait:
0.6 x 85.1 + 0.10 x (17.3 + 6.78) + 0.30 x 8.51-30 = 26.1
Celle d'un atelier de 250 t/d:
o. 7 x {7.81 + 36.4) + 0.3 x 29.8- 16 = 23.9
Donc, la maximisation du revenu proviendrait de la construction d'un atelier de 500 t/d
immdiatement. Avec ces informations et d'autres pertinentes, le management de cette
socit pourra prendre un dcision finale.
Tableau 5.1
Estimation des profits annuels en fonction de la
grosseur de l'atelier et du volume de la demande
Atelier tat de la nature Revenu 1 an
capacit, t/d de la demande (millions $)
leve 10
500 faible 1
leve 4.5
250 faible 3.5
leve 9
250 + 250 faible 0.5
5-5
Si la demande reste faible, la socit rorientera la production et le profit annuel
ne sera que de 3.5 millions de dollars.
5-6
Tableau 5.2
Les valeurs prsentes prvues au point de dcision 2
Revenu
tat de la an Valeur Valeur
Choix nature de Probabilit (millions $) prsente prvue
la demande durant (millions $) escompte
18 ans (millions$}
ajouter 250 Ud leve 0.86 9 73.8 63.5
ajouter 250 Ud faible 0.14 0.5 4.10 0.57
Total 64.1
- l'investis.
.2.L
NETTE 44.1
pas de leve 0.86 4.5 36.9 31.7
changement
pas de faible 0.14 3.5 28.7 4.02
changement
Total 35.7
- l'investis. 0
NETTE 35.7
Tableau 5.3
Les valeurs prsentes prvues au point de dcision 1
tat de Revenu Revenu Valeur
Choix la nature Probabilit an Nombre escompt prvue
de la (millions $) d'annes (millions $) escompte
demande (millions $)
leve 0.60 10 20 85.1 51 .1
initiale 0.10 10 2 17.3
leve et 2.41
aprs faible 1 18 6.78
faible 0.30 1 20 8.51 2.55
Total 56.1
- l' investis.
~
'
NETTE 26.1
leve 0.70 4.5 2 7.81
et
valeur de la
dcision au
point 2 31 .0
44.1 la 36.4
fin de la 2
9
anne
faible 0.30 3.5 20 29.8 8.94
Total 39.9
- l'investis. -16
NETTE 23.9
-------- - -- --
5-8
Demande Probabilit Revenu/an Nombre
d'annes
leve
0.6 10
20
leve
0.1
10 2

Faible 1
18
Faible 0.3 1

)
20



c? 0"'
t':l"''lle
0.86 9
18

<:!
'-.__.../
0.14 0.5
18
"
)


19

0.86 4.5

i:\e'lle

18
'7.
6'
'- / i:I/Ofe
/:'l/6
0.14 3.5
18
v&o
.a
"-
Faible 0.3 3.5
20
0 2 20
ans
Figure 5.1 L'arbre de dcision pour la construction d'un atelier de fabrication de pte mcanique
Demande Probabilit Revenu/an Nombre d'annes
0.6 10 20

/
leve initiale 0.1 10
2
)
Faible aprs
1 18

-:-.<..0


0
"6;$
cP
0.3 1 20

Provenant de

l'addition de
.t.:
250 tld aprs 2 ans
'?k
(\)d'
(9
0.7 2 \$'(9. +4.5

., ....
'7..
6'
0.3 3.5 20
Figure 5.2 L'arbre de dcision simplifi: aprs que la dcision a t prise de procder l'addition de 250 t/d aprs 2 ans
5-10
Problme 5. 1
Un jeune dbrouillard a dcid de tirer avantage de ses connaissances en statistiques
pour faire quelques dollars durant ses vacances. Il s'est achet deux chapeaux
identiques; dans le fond intrieur il a crit X dans l'un et Y dans l'autre. Dans le chapeau
marqu d'un X, il dpose 5 balles de golf; 3 blanches et 2 jaunes, alors que dans celui
identifi par un Y, il y dpose aussi 5 balles, mais cette fois, 4 jaunes et 1 blanche.
Il propose des amis les paris suivants:
Payer 10 $ et identifier le chapeau marqu d'un X. S'il l'identifie correctement, il
gagnera 8 $,s'il se trompe, il perdra son 10 $;
Payer 2 $ et tirer une balle de l'un des chapeaux, aprs qu'il ait vu si elle est
blanche ou jaune il peut ou non payer 10 $ et tenter d'identifier le chapeau marqu
d'un X. Il pourrait dcider de tirer 2 ou 3 balles et payer 2 $ pour chacune d'elles
avant de payer 10 $ pour identifier le chapeau . L'ami a toujours la possibilit de
refuser le pari.
a) Tracer un arbre de dcision pour montrer toutes les possibilits du jeux.
b) Solutionner l'arbre et dterminer la possibilit la plus intressante pour l'ami.
Problme 5.2
La compagnie Boisbriand dsire investir 100 millions de dollars dans la construction d'un
nouveau complexe rsidentiel. Elle prvoit des revenus de la location de ces rsidences
qui varieront avec la situation conomique de la province. Le tableau 1 dcrit ces
prdictions.
tat de la nature
de l'conomie
Prospre 0 0. A
Normal . .... . ... ..... B
Rcession 0 c
Tableau 1
Variation des revenus nets
Valeur actuelle des revenus
nets - Mill ions $
140
120
50
Probabilit
d'occurrence
0.2
0.5
0.3
5- 11
Comme la compagnie n'a pas encore pris la dcision finale d'investir, elle cons idre la
possibilit d'engager un consultant . Ce dernier exige un montant de 100 000 $ et
s'engage prdire l'tat de la nature avec certaines prcisions. Ces dernires
proviennent des rsultats de ses tudes prcdentes. Le tableau 2 rsume ces
prdictions.
Prdictions
du consultant
An
Bn
en
Tableau 2
Les probabilits que le consultant
a dj obtenues d'tudes similaires
Les probabilits d'obt enir
les tats de la nature
A 8
0.8 0.1
0 . 1 0.9
0 . 1 0.0
c
0.1
0.2
0.7
a) Dmontrer, l'aide d'un arbre de dcision, si a compagnie doit faire
l'investissement en se basant sur ses propres prdictions .
b) Est-il avantageux pour la compagnie d'engager le consultant?
c) Quel est le montant maximal que la compagni 'e pourrait payer un consultant?
d) Si une information parfaite pouvait tre obtenue, combien la compagnie pourrai t-
elle payer?
---- ---------
13.3 SEQUENTIAL DECISION MAKING: USE
OF DECISION TREES
Decision trees, also commonly called decision flow networks and decision
diagrams, are powerful means for depicting and facilitating the analysis of
important problems, especially those that involve sequential decisions and
variable outcomes over time. Decision trees have great usefulness in prac-
tice because they make it possible to look at a large complicated problem in
terms of a series of smaller simple problems and they enable objective anal y-
sis and decision ma king which includes explicit consideration of the risk and
effect of the future.
The name decision tree is descriptive of the appearance of a graphical
portrayal, for it shows branches for each possible alternative for a given
decision and branches for each possible outcome (event) which can result
from each alternative. Such networks reduce abstract thinking to a logical
visual pattern of cause and effect. When costs and benefits are associ ated
with each branch and probabilities are estimated for each possible outcome,
then analysis of the How network can clarify choices and ri sks.
13.3.1 Deterministic Exemple
The most basic form of decision tree occurs when each alternative can
be assumed to result in a single outcome-that is, when certainty is as-
sumed. The replacement problem in Fig. 13-4 illustrates this. The problem
as shown retlects that the decision on whether to replace the old machine
with the new machine is not just a one-lime decision, but rather one which
recurs periodically. That is, if the decision is made to keep the old machine
at decision point 1, then la ter, at decision point 2, a choice again has to be
made. Similarly, if the old machine is chosen at decision point 2, then a
choice again has to be made at decision point 3. For each alternative, the
cash inflow and duration of the project is shown above the arrow and the
cash investment opportunity cost is shown below the arrow.
For this problem, one is concerned initially with which alternative to
choose at decision point 1. But an intelligent choice at decision point 1
should take into account the later alternatives and decisions which stem
from it. Hence, the correct procedure in analyzing this type of problem is to
start at the most distant decision point, determine the best alternative and
quantitative result of that alternative, and then " roll back" to each sucees-
Vl
N
342
1-.....
x
".s:

.9J-,..
Risk and Uncertainty
% ...


.s:
'.r
': 6'
J-,..
Old + $3M/yr, 3 yr
3yr
1< -$2M



J--:
Figure 13-4. Deterministic replacement example.
Chap. 13
sive decision point, repeating the procedure until finally the choice at the
initial or present decision point is determined. By this procedure, one can
make a present decision which directly takes into account the alternatives
and expected decisions of the future.
For simplicity in this example, timing of the monetary outcomes will
first be neglected. which means that a dollar has the same value regardless of
the year in which it occurs. Table 13-5 shows the necessary computations
and decisions . Note that the monetary outcome of the best alternative at
decision point 3 ($7.0M for the "old") becomes part of the outcome for the
"old" alternative at decision point 2. Similarly, the best alternative at deci-
sion point 2 ($22.0M for the "new") becomes part of the outcome for the
"old " alternative at decision point 1.
Table 13-5
MONETARY UTCOMES AND D ECISIONS AT EAOI POI NT-DETERMINISTIC
REPLACEMENT EXAMPLE OF FIG. 13-4'
Decision
Point Alternative Monetary Out come Choice
3
{
Old S3M(3) - $2M = S 7.0M Old
New S6.5M(3)- SISM = S 1.5M
{
Old $7M + $3.5M(3) - SIM = Sl6.5M
2
New $6, 5M(6) - Sl7M = S22.0M New
{
Old S22.0M + S4M(3) - SO.SM = $33.2M Old
New S5M(9)- $15M = SJO.OM
lnterest = 0%; that is. ignoring timing.
Sec. 13.3
Sequential Decision Making: Use of Decision Trees
343
By following the computations in Table 13-5, one can see that the
answer is to keep the "old" now and plan to replace it with the "new" at the
end of 3 years (at decision point 2). But this does not mean that the old
machine should necessarily be kept for a full 3 years and then a new machine
bought without question at the end of 3 years. Conditions may change at any
time, thus necessitating a fresh analysis-probably a decision tree analy-
sis-based on estimates which are reasonable in light of conditions at that
later time.
13.3.1.1 Dctenninistic Example Considcring Timing
For decision tree analyses, which involve working from the most dis-
tant decision point to the nearest decision point, the easiest way to take into
account the timing of moncy is to use the present worth approach and th us
discount ali monetary outcomes to the decision points in question. To dem-
onstrate, Table 13-6 shows computations for the same repl acement problem
of Fig. 13-4 using an interest rate of 25% per year.
Note from Table 13-6 that when taking into account the effect of timing
by calculating present worths at each decision point, the indicated choice is
not only to keep the "old" at decision point 1, but also to keep the " old'' at
decision points 2 and 3 as weil. This result is not surprising sincc the high
interest rate tends to favor the alternatives with lower initial investments,
and it also tends to place Jess weight on long-term returns .
3.3.2 Typical Example with Random Outcomes
The following is a brief description of a typical automation problem for
which the analytical capabilities of decision trees are very useful . A decision
maker must decide whether to automate or not to automate a given pro-
cess. He is uncertain whether the results will turn out to be poor. fair, or
excellent-depending on the technological success of the automation proj-
ect. The net payoffs for possible outcomes (expressed in net present
worths) are $-90M, $40M, and $300M, respectively. The initially estimatcd
probabilities thal each outcome will occur are 0.5, 0.3. and 0.2. respec-
tively. Figure 13-5 is a decision tree depicting this simple situation. Table
13-7 shows calculations to determine that the best choice for the firm is to
Figure 1:35. Automation problem di a-
gram before consideration of techno(ogy
study. ($ are in net PW)
Poor S - 90M (0.5)
Fair S40M 10.3)
Exc. SJOOM (0.2)
Not automate
10
Y'
-......
.......
Decision
Point
3
{
{
{
Risk and Uncertainty Chap. 13
Table 13-6
DECISIONS I'IT EI'ICH POINT WITH INTEREST = 25% PER YEAR FOR
DETERMINISTIC REPLACEMENT EXAMPLE OF fiG. 13-4
Alternative PW of Monetary Outcome
Old $3M(P/A,3) - S2M
S3M( 1. 95) - $2M = S3.85M
--
New $6.5M(PIA.3)- $18M
$6.5M(I.95)- SI8M = -$5.33M
Old $3.85(PIF.3) + S3.5M(P/A.3)- SIM
$3.85(0.512) + SIM = $7.79M
New $6.5M(PIA,6)- $17M
S6.5M(2.95) - SI7M = $2.18M
Old S7.79M(P/F.3) + $4M(P/A.3)- $0.8M
$7.79M(0.512) + $4M(I.95) - $0.8M SI0.99M
New $15M
$5.0M(3.46) - SI5M = S2.30M
Table 13-7
EXPECTED NET PW CALCULATION FOR THE AUTOMATION
PROBLEM BEFORE CONSIDERATION OF TECHNOLOGY
FEASIBILITY STUDY
Automate: S-90M(0.5) + $40M(0.3) + $300M(0.2) = $27M
Not automate : = $0
Choice
Old
Old
Old
automate based on an expected* net PW of $27M versus $0 if it does not
automate. Nevertheless. this may not be a clear-cut decision because of the
risk of a $90M Joss and because the decision maker might reduce the risk by
obtaining further information.
Suppose that it is possible for the decision maker to make a further
technology study at the PW cost of $10M. The study will disclose that the
enabling technology is either "shaky," "promising," or "solid."
lnstead of using Bayesian statistics for revision of probabilities at this
point (this topic will be covered in Section 13.3.5 and Appendix 13-A), let us
assume that the probabilities of the various possible automation outcomes,
given the technology study outcomes, are as shown in Fig. 13-6, which is a
decision tree diagram for the entire problem.
Expected (long-run average. or mean) monetary value will be used as a choice criterion
for problems involving probabilistic monetary outcomes al most throughout this book. [Excep-
tion: Where certain monetary equivalents are used, as in Section 13.4].
Sec. 13.3 Sequential Decision Making: Use of Decision Trees
345
Poor $-90M (0.6)
Poor 1-00M (0.73)
Fair $40M 10.3) Fair $40M 10.22)

Exc. $300M 10.05)
$0 Not automate
"\

Poor $-90M (0.43)
.l'o-f, "o/. Fair $40M (0.34)
Q,
*
.
")- "l q"' .J\tt'-e Exc. $300M (0.23)
o, "o:
"), 'J.
""'
'J. '_,
---
Not automate
Poor $ - 90M 10.21)
Fair $40M (0.37)
Exc. $300M (0.42)
Figure 13-6. Automation example with technology study laken in consideration.
($are in net PW be fore consideration of any cost for "Technology Study").
10
10
10
The tree diagram shows expected future events (outcomes), along with
their respective cash flows and probabilities of occurrence. The rectangular
blocks represent (decision) points intime at which the decision maker must
elect to take one and only one of the paths (alternatives) considered. These
decisions are normally based on a quantifiable measure, su ch as money,
which has been determined to be the predominant "cost" or "reward" for
comparing alternatives. The general approach is to find the action or alter-
native that will maximize the expected net PW equivalent of future cash
flows at each decision point, starting with the furthermost decision point(s)
and then "rolling back" until the initial decision pointis reached .
The circular (chance) nodes represent points at which uncertain events
(outcomes) occur. At a chance node the expected value of ali paths leading
(from the right) into the node can be calculated as the sum of the anticipated
value of each path multiplied by its respective probability. (The probabilities
of ali paths leading into a node must sum to 1.) As the project progresses
through time, the chance nodes are automatically reduced to a single out-
come based on the "state of nature" which occurs at that time.
The solution to the problem in Fig. 13-6 is shown in Table 13-8. It can
be noted that the alternative "technology study" is shown to be best with an
expected net PW of $34.62M.
346

Poinl
2A
28
2C
Risk and Uncertainty Chap. 13
Table 13-8
ExrF.\THl NFT PW CALCUl.ATIONs FOR THE AuTOMATION PROBLEM WITH
CoNSIDERATION oF TECHNOLOGY STUDY
Alternai ive
{Automate
Not automate
{Automate
Not automate
{Automate
Not automate
{ '"'".'" Not automate
Technology
study
Expecled Net PW
$-90M(0.73) + $40M(0. 22)
$ + JOOM(0.05) = $-4 1.9M
$0
S- 90M(0.43) + S40M(0.34)
$+ 300M(0. 2Jl = $43.9M
so
$- 90M(0.21) + $40M(0.37)
$ + JOOM(0.42) = $12l.9M
$0
From labie 1 J-7 $27M
From table 1 3-7 $0
$0(0.41) + $43. 9M(0.35)
$+ 121.9M(0.24) - $10M = $34.62M
Choice
Not automate
Automate
Automate
Technology
study
13.3.3 Decision Tree Steps
Now thal decision trees (diagrams) have been introduced and the me-
chanics of using the diagrams to arrive at an initial decision have been
illustrated, the steps involved can be summarized as follows :
1. Identify the points of decision and alternatives available at each point.
2. Identify the points of uncertainty and the type or range of possible
outcomes at each point (layout of decision flow network) .
3. Estimate the values needed to make the analysis, especially the proba-
bilities of different outcomes and the costs/returns for various out-
cornes and alternative actions.
4. Analyze the alternatives, starting with the most distant decision
point(s) and working back, to choose the best initial decision.
The example above used the expected net PW as the decision crite-
rion. However. if outcomes can be expressed in terms of utility units, the
decision ma ker can use the expected utility as a decision criterion. Alterna-
tive) y, the decision maker may be willing to express his or her certain mone-
tary equivalent for each chance outcome node and use thal as his or her
decision criterion as explained in Sections 13.4 and 13.5.
Because a decision diagram can quickly become discouragingly, if not
unmanageably, large, it is generally best to start out by structuring a problem
Sec. 13.3 Sequential Decision Making: Use of De_cision 347
simply by considering only major alternatives and outcomes in order to gel
an initial understanding or "feel" for the problem. Then one can develop
more information on alternatives and outcomes which seem sufficiently im-
portant to affect the final decision until one is satisfied thal the study is
sufficiently complete in view of the nature and importance of the problem
and the lime and stndy resources available .
13.3.4 General Principles of Diagramming
The proper diagramming of a decision problem is, in itself, general! y
very useful to the understanding of the problem, and it is essential to correct
subsequent analysis.
The placement of decision points and chance outcome nodes from the
initial decision point to the base of any later decision point should give a
correct representation of the information thal will and will not be available
when the decision maker actually has to make the choice represented by the
decision point in question. The decision tree diagram should show the fol-
lowing:
1. Ail initial or immediate alternatives among which the decision maker
wishes to choose
2. Ali uncertain outcomes and future alternatives thal the decision maker
wishes to consider because they may directly affect the consequences
of initial alternatives
3. Ali uncertain outcomes thal the decision maker wishes to consider
because they may provide information thal can affect his or her future
choices among alternatives and hence indirectly affect the conse-
quences of initial alternatives
It should also be noted !hat the alternatives at any decision point and
the outcomes at any outcome node must be:
1. Mutually excl usive; thal is, no more than one can possibly be chosen .
2. Collectively exhaustive; that is, sorne one must be chosen or some-
thing must occur if the decision point or outcome node is reached.
Figure 13-6 retlects these points. For example, decision points 2A, 28 ,
and 2C are each reached only after one of the mutually exclusive results of
the technology study are known; and each decision point retlects ali alterna-
tives to be considered at thal point. Further, ali possible outcomes to be
considered are shown as evidenced by the fact thal the probabilities sum to
1.0 for each chance node.
'-h
1
..._
'-\
-48
Risk and Uncertainty
13.3.5 Use of Statistics to Evaluate the Worth
of Further lnvestigative Study
Chap. 13
As was included in the automation example above, one alternative that
frequcntly exists in an investment decision problem is further research or
investigation before deciding on the investment. This means making an in-
tensive objective study. hopefully by a fresh group of people. It may involve
su ch aspects as undertaking additional research and development study,
making a new analysis of market demand, or possibly studying anew future
operating costs for particular alternatives.
The concepts of Bayesian statistics pro vide a means for utilizing subse-
quent information to modify estimates of probabilities and also a means for
cstimating the economie value of further investigation study.
To illustrate, consider the one-stage decision situation shown in Fig.
13-7 in which each alternative has two possible chance outcomes: "high" or
"low" demand. lt is estimated that each outcome is equally likely to occur,
and the monetary result expressed as PW is shown above the arrow for each
outcome. Again. the amount of investment for each alternative is shown
below the respective !ines. Based on these amounts, the calculation of the
expected monetary values (in net PW) is shown in Table 13-9, which indi-
cates that the "new FMS" should be selected.
To demonstrate the use of Bayesian statistics, suppose that one is
considering the advisability of undertaking a fresh intensive investigation
before deciding upon the "old system" versus the "new FMS." Suppose
also that this further study would cost $2.0M. To use the Bayesian ap-
proach. it is necessary to assess the conditional probabilities that the investi-
gation or "technology study" will yield certain results. These probabilities
reflect explicit measures of management's confidence in .the ability of the
investigation to predict the outcome. Sample assessments are shown in Ta-
ble 13-1 O. As an explanation, P(ltill) means the probability that the pre-
dicted demand is "high" (Ir), given that the actual demand will turn out to be
"high" (Il) .
PW of returns
H $45M (0.5)
$27.5M (0.5)
H $SOM (0.5)
$48M (0.5) Figure 13-7. One-stage FMS replace
ment problem.
Sec. /3.3 Sequential Decision Making: Use of Decision Tree.t 349
Appendix 13-A contains a formai statement of Bayes' theorem as weil
as a tabular format for ease of calculations in the discrete outcome case.
Tables 13-1 1 and 13-12 use this format for revision of probabilities based on
the data in Table 13-10 and the prior probabilitics of0.5 that the dcmand will
be high and 0.5 that the demand will be low.
Table 139
EXPECTF.D NF.T PW FOR PRORLHI IN FIG. 13 7
Old system: S45M(0.5) + $27.5M(0.5) - SJOM S21i.25M
New FMS: S80M(0.5) + $48M(0.5) - $35M S29.0M
(1)
State
(Actual
Demand)
H
L
( 1)
State
(Actual
Demand)
H
L
Table 13-10
MANAGEMF.NT's AssESSMENT OF CoNFIDENCE
IN INVESTIGATION RESULTS'
P(hj HJ O. 70
P(/1jL) = 0.20
= 0.30
= 0.80
lnvestigation-predictcd dcmand: h , high;
/, low. Actual demand: Il, high; L, Jow.
Table 1311
CoMPUTATION OF PoSTERIOR PRORABIUTIF.S THAT
(NVESTIGATION-PREDICTED DEMAND ls HIGII Ch)
(2) (3) (4) (5)
= (2)(3) = (4)(4)
Prior Confidence Posterior
Probability, Assessment. Joint Probability,
P (State) P (lrjState) Probability P (Statejlr)
0.5 0.70 0.35 0.7R
0.5 0.20 0. 10 0.22
:E = 0.45
Table 13-12
CoMPUTATION oF PosTERIOR PROBABILITIES GIVEN THAT
INVESTIGATION-PREDICTED DEMAND (S LOW (/)
(2) 0) (4) (5)
(2)(3) = (4)/L(4)
Prior Confidence Posterior
Probability, Assessment, Joint Probability,
P(State) P (ljState) Probability P (Statej/)
0.30 0.15 0.27
0.5/ 0.80 0.40 0.73
:E = 0.55
V\
\

350 Risk and Uncertainty Chap. /3
The probabilities calculated in Tables 13-11 and 13-12 can now be used
to assess the "technology study" alternative . Figure 13-8 shows a decision
tree diagram for this alternative as weil as the two original alternatives .
Note the demand probabilities are entered on the branches according to
whether the investigation indicates "high" or "low" demand.
PW of benefits
H
$45M (0.5)
$27 .5M (0.5)
H $45M (0.78)
L $27.5M (0.22)
H $80M (0.78)
L $48M (0.22)
T eehnology study

H $45M (0.27)
$27.6M (0.73)
H $80M (0.27)
$48M (0.73)
H $80M (0.5)
$48M (0.5)
figure 13-8. Replacement problem with alternative of further investigation.
Sec. 13.3
Sequential Decision Making: Use of Decision
351
The expected outcome for the "technology study" alternative can now
be calculated. This is done by the standard decision tree "rollback" princi -
ple. This is shown in Table 13-13. His worthy of note thal the 0.45 and 0.55
probabilities that investigation-predicted demand will be "high .. and "low:
respectively, are obtained from the totals in column (4) of the Bayesian
revision calculations shown in Tables 13-11 and 13-12.
Thus, from Table 13-13, it can be seen that the "new FMS .. alternative
with an expected net PW of $29.0M is the best initial course of action by a
slight margin. Wh ile the figures used here do not reftect any ad van tage to
this technology study, the ad v an tage potentially can be great.
A 3.3.6 Expected Value of Perfect Information
The expected value of perfect information (EVPI) is the maximum
expected loss due to imperfect information or foreknowledge as to wh at will
be the outcome(s) in a situation involving probabilities associated with the
possible returns or costs. It is described more complete( y in Appendix 13-B.
Calculations of the EVPI for the cxample replacement problem of Fig-
ure 13-7 are shown in Table 13-14. It can be seen in the table thal the
expected net PW with perfect foreknowledge, $31.25M, is greater than the
expected net PW of the preferred alternative before the further investigation.
$29.0M (from Table 13-9), by $2.25M, which is a measure of the maximum
expected value of further investigation.
For the same example problem except with the different probabi lities
resulting from "technology study" as given in Fig. 13-8 and Table 13-13. the
technology study results in a decrease in the expected net PW of $1.70M.
(i.e., $29.0M for the "new FMS" minus $27.3M for "technology study .. ). If
the "technology study" had cost $0 instead of $2.0M assumed in the prob-
Decision
Point Alternative
2A
{
Old system
New FMS
2B {
Old system
New FMS
1
{
Technology
study
Old system
New FMS
Table 13-13
EXPECTED NET PW FOR REPLACEMENT PRORLEM OF FIG. 13-8
Expected Net PW Choice
$45M(0.78) + $27.5M(0.22)- SIOM = S31.15M
$80M(0.78) + $48M(0.22) - $35M = S37.%M New FMS
$45M(0.27) + $27.5M(0.73)- SIOM = $22.23M Old system
$80M(0.27) + S48M(0.73) - $35M = S21.64M
$37.96M(0.45) + $22.23M(0.55)- S2M = S27.31M
(from Table 13-9)
(from Table 13-9)
S26.25M
S29.0M New FMS
0
-- '\-)
.,
3.54 Risk and Uncertainty
13.5 ALTERNATIVE METHOD OF DECISION TREE
ANALYSIS; NOTATION/ANALYSIS CONVENTIONS
Chap. 13
Sorne analysts and decision makers prefer to show the criterion values at the
end of each possible pa th through the tree. (Note: If the monies along the
path are at significantly different points in time, the criterion values should
be expressed in terms of equivalent worths, such as PW or FW.) Then the
"rollback" technique can be used to determine the optimal choice at each
decision point and obtain the same initial decision as when using the pre-
vious method.
Suppose, for the replacement problem in Fig. 13-8, that it is desired to
evaluate the alternatives using the certainty equivalent (CE) criterion while
formally considering th at the decision maker's initial asset position is, say,
$10M. Figure 13-10 depicts this as weil as a hypothetical solution to the
problem. Note !hat the initial asse! position is shown in a dashed oval to the
left of the initial decision point. and that the solid ovals at the extreme right
indicate the outcomc's criterion values, (including the initial asset position)
for each path of branches. The conventions which produce the results in
Fig. 13-10 are as follows:
1. At each chance node (circle), the CE is shown in a rectangular box and
represents the value of being at that node, and th us takes into account
everything subsequent to (i.e., to the right of) the node.
2. At each decision point node (square) the CE of the best alternative is
shown in a rectangular box. Further, ali alternatives except the best at
each decision node are marked with double slashes across their respec-
tive paths.
To complete the solution, the decision maker must specify his CE for
ali chance outcome nodes. Note that the uppermost node is the same as in
Fig. 13-9a. As in thal example, let's assume thal the decision maker decides
that a certain $35.5M is just as desirable as 0.5 chance at $45M and 0.5
chance at $27.5M. This is shown in the small block above the node. Subjec-
tively determined CEs are shown at ali other nodes. The "rollback" analy-
sis procedure is applied but this time taking into account only the probabili-
ties . and not the cash flow s. along each path (sin ce the effect of the cash
flows along each path has already been laken into account by the equivalent
worth outcomes (in ovals) at the end of each path).
The final "rollback" to decision point 1 shows thal the alternative of
"technology study" (with a CE of $35.7M) is better than either "old sys-
tem (with a CE of $35.5M) .or "new FMS" (with a CE of $34M). This
differs from the expected monetary (net PW) results in Table 13-13 as typi
cally would be anticipated.
Sec. 13.5

Initial asset
position
Alternative Method of Decision Tree Analysi.t
H
355

127.5M 10.51
H $SOM
$48M


'""""!
Outcome1
incl udlng
ini tia18'SS!et

Figure 1310. Replacement problem with alternative of further investigation
(showing outcome criterion values) at the end of each branch and subsequent
analysis using CE criterion and including initial asse! position.
Example for lowest branch path: SIOM - SJ5M + S48M = SZJ M.
V\
'


If
demand

Risk and Uncerrainty Chap. 13
Table 13-14
EXPECTED VALUE OF PF.RFECT INFORMATION FOR NESTAGE REPLACEMENT
PROBLEM OF FIG. 13-7 (ALL $ARE IN PW)
Oulcomes
for alternatives
would he:
Then the And the The prior Thus the
preferred monetary probability expected
alternative is: oulcome is: estimate is: value is:
High Old: - $10M = $J5M New FMS $45M 0.5 S22.50M
New: - 35M = 45M
Low Old: 27.5M - IOM = 17.5M Old system 17.5M 0.5
New: 48M - 35M =
Expected net PW with perfect foreknowledge:
Expected net PW (from Table 13-9) = S29.0M
:. EVPI = SJ1.25M - S29.0M = $2.25M (in net PW)
8.75M
r = S3J.25M
lem. it would have resulted in an increase in the expected net PW of $-1.7M
+ $2.0M = $+0.3M. Thal $0.3M represents the "expected value of sample
information:: i.e. , how much out of the possible maximum EVPI of $2.25M
can be achieved by this particular "technology study."
13.4 USE OF CERTAIN MONETARY EQUIVALENTS
Any problem involving variable outcomes, such as the replacement problem
in Fig. 13-8. could just as weil be analyzed using the so-called certain mone-
tary equivalent (CME) or certainty equivalent (CE) criterion rather than an
expccted monetary value criterion.
A certainty equivalent (CE) for any situation involving variable out-
cornes is a monetary amou nt (lump sum) which the decision maker considers
to be equally as valuable (to pay or to receive) as the variable possible
outcomes for thal situation. So-called "risk-averse" (risk-avoiding) deci-
sion. makers typically have (or will choose) a CE which is Jess than the
expected monetary value (EMV) for a given situation. The amount by which
the EMV is greater than the CE, called a "risk premium" (RP), is a measure
of how much the decision maker is willing to give up, on the average, to
eliminate the variability (riskiness) in a decision problem. Thus
CE = EMV - RP and RP = EMV - CE
While RP is positive for a risk-averse decision maker (i .e., he or she would
opt to "bu y out" of a ris ky situation), it is negative in the case of a so-called
"risk-seeking" (risk-taking) decision maker, and it is 0 in the case of a "risk-
neutral" decision maker.
j.
Sec. 13.4
Use of Certain Monetary Equivalents
53
Example 13-l(a)
Suppose thal it is desired for a decision maker to specify his CE for the situation in
the left part of Fig. 13-9. The question to be answered can be phrased as: What
amount for certain (to be received or paid) would be just as good Cor bad) as a 50%
chance of $45M and a 50% chance of $27.5M? Another way of phrasing this is :
"How much would you be wiUing to pay (or get paid) in place of [sorne probabilistic
payoff situation]?" The decision maker might consider th at situation not very ris ky
and answer, say $35.5M. The EMV for the situation is $36.25M. which mcans that
RP = $36.25M - $35.5M = $0.75M.
Example 13-l(b)
Suppose thal the outcomes in Example 13-l(a) are made more risky (highly variable)
by increasing the $45M by $40M to $85M and decreasing the $27.5M by $40M to
$-12.5M. The situation is shown in the righi part of Fig. 13-9. The EMV is stiJl
$36.25M, but the CE answer by a given decision maker might vary considerably
according to his perception of the impact of the new possible outcomes . If the
decision maker is very slightly risk averse for the amounts involved, he might stiJl
have a CE close to the $35.5M in Example 13-l(a) . However, if he is significantly
concerned about the impact of a possible low outcome su ch as the - $12.5M. his CE
might be significantly lower , such as , say, $28.0M. If so. his RP = $3t>.25M -
$28.0M = $8.25M.
A decision maker's certainty equivalent for a set of uncertain cash flow
outcomes should depend on the complete context in which the cash flow
occurs . Thal is, the decision maker should be cognizant of the firm's finan-
cial position when deciding on a CE. One way this could be taken into
account is to specify the decision maker' s financial position before the
initial decision is to be made. For example, perhaps net liquid assets (such
as net assets thal can be converted into cash within a month or within a
year) might be laken as the most appropriate mcasurc of the firm ' s initial
asse! position.
(0.5) S45.0M
10.5) S85M
(0.51 S27.5M
10.5) S- 12.5M
Choice 2 CE a 7 +- (to equal choie 11
(a)
(b)
Figure 13-9.
determined.
Example of outcomes for which certainty equivalents are lo be
3Sb Risk and Uncertainty Chap. 13
13.6 STOCHASTIC DECISION TREES
lt is generally true that the number of possible outcomes that could result
from the choice of an alternative is greater than the few outcomes typically
assumed in decision tree problems so that the number of branches is kept to
manageable size. Hence, information (and sometimes, objectivity) is sacri-
ficed by the assumed reduction of branches. It is corn mon that the possible
out cornes emana ting from an alternative could be described most adequately
as a continuous distribution. These are sometimes called stochastic decision
trees .
As an example of a problem for which sorne kind of stochastic decision
tree analysis is probably applicable, consider the automation problem in Fig.
13-6. Only thrce possible outcomes were considered in case the process
were automated. However. the problem would have been more adequately
mode led if, say, ten or more possible monetary outcomes were considered if
the process were automated, taking into account the variable costs and
benefits of automation. lndeed, the problem probably can be represented
best by continuous probability distributions. Monte Carlo simulation can be
used to analyze such decision situations efficiently to arrive at probability
distributions of the decision criterion for ali initial alternatives. Section 13.7
will explain Monte Carlo simulation in general.
Often, variable outcomes are thought to have a continuous distribution
. but the decision maker or analyst feels that it is not.feasible or reasonable to
specify the distribution (i.e., the decision problem fits the classical definition
of uncertainty). A useful way to diagram an uncertain outcomefor a deci-
sion tree problem is shown in Fig. 13-11 . Note that this outcome "fan" can
show maximum and minimum values if these can be estimated.
As a further example, suppose the above automation problem involves
corltinuous outcomes for which probabilities cannot be estimated but for
which the minimum and maximum extremes for each outcome can be esti-
mated. In this case, a diagram representing the problem (with assumed max-


Figure 13-11. Representation of a con-
tinuous outcome for which probabilities
are not known.
Sec. 13.6 Stochastic Decision Trees

' /
Not autom. $0
Figure 131Z. Automation problem assuming continuous outcomes for which
probabilities may not be known, but extremes can be estimated.
j 57
imums and minimums) could be as shown in Fig. 13-12. Note that even the
outcomes of the technology study are represented to have continuous, but
unknown, probability distributions. White information in this form does not
le nd itself to mathematical calculation of decision criteria such as EMV, the
mere diagramming of the problem can be very helpful in reaching a deci-
sion. Indeed, a decision maker can be asked to determine his CE at any
outcome node even though probabilities of the various outcomes are not
known. Any such CE would be a gross guesstimate (certainly, it would be
more subjective than if probabilities of various possible outcomes were
known), but at !east it represents the decision maker's intuitive judgment
under the circumstances and can be used for comparing alternatives .


r
, ..
J74
Risk and Uncertainty Chap. 13
/)rcinn Trers (Srctions 13.3 tllrou!!ll 13.6) :
B-9. Given : There is a 60% chance !hat Dynamics General Corporation will require
a conversion from haselmnd Ethernet cabling in a certain manufacturing plant
to hroadhand Ethernet cabling at a rate of 1000 feet per year (there is a 40%
chance that the conversion will occur at a rate of 500 feet per year). The
recabling will be required if a new government contract is negotiated with them
now. which is 10 years prior to expiration of the present con tract.
The VP of Manufacturingfeels that if he wait.r for the present contract to
expire in 10 years. there is a 25% chance that the new contract will require 5000
feet per year of conversion from baseband to broadband Ethernet cable and a
75% chance of conversion at the rate of 1000 feet per year. Each 1000 feet of
Ethernet conversion (baseband to broadband) costs $100,000.
IJ-10.
Pwpert y ta x for either type of cable is the sa me and should not be
C<lnsidcred . Ail new cabling will be broadband Ethernet; and at the MARR of
21l<::f. pcr year. it can be considered lo have perpetuai life. The terms of any
new contract will cali for a continuing program of converting existing
bascband cable to broadband cable until the total anticipated cabling require-
ment of 25.000 feet has been completed.
Required: Draw a decision tree for this problem, including the appropri
ate cash nows and probabilities. State any assumptions you make. Carefully
label ali alternatives.
A purchasing manager is faced with deciding whether or not to stock a large
supply of metal. The uncertain variable is the future priee of the metal. The
following are present worths of consequences and prior probabilities for the
various perceived outcomes:
Future Priee P(Furure Priee) PW If Stock
PW If Don't Stock
High
O.J $ 100,000 $0
Medium 0.5 - 10,000
$0
Low
0.2 -100,000
$0
For $6.000 it is possible to hire a consulting firm that would be able to make a
fairly accurate forecast in terms of whether the priee will go up or down as
follows :
If the future
is going to be:
High
Medium
Low
Then the probabilities
r he consultant will
predict the priee will
go up or down are as
follows:
Up
0.9
0.4
0.8
Down
0.1
0.6
0.2
Chap. 13 Exercise.r 375
(a) Diagram the problem in the form of a decision trec .
(b) Determine what would be the best alternative using the EMY criterion.
(c) Determine the maximum expected value of the consulting firm' s
if the firm could perfectly predict the future.
(d) Determine the best alternative if you were the decision ma ker using your
own unique CE values.
1311. The Norva Company has already spent $80,000 developing a new electronic
gage and is now considering whether or not to market it. Tooling for produc-
tion wou id cost $50,000. If the gage is produced and marketed, the company
estimates that there is only one chance in four thal the gage would be success
fui. If successful, the net cash infiows would be $100.000 per year for 8
years . If not successful, the net cash outnows wou id be $30,000 per year for 2
years, after which lime the venture would be terminated. The minimum at
tractive rate of return on money is 20% per year.
(a) Draw a decision tree and determine the best alternative using the ex-
pected net PW criterion .
(b) If there is a market research group that can provide perfect information
about the success of this product, what would be the most the company
should be willing to pay for the group's service?
(c) Suppose the market research group can make a market survey thal with
probability 0.8 will predict a successif the gage will actually tu rn out to be
a success and with probability 0.9 will predict failure if the gage will tu rn
out to be unsuccessful. Should the survey be undertaken first? What is
the expected value of the survey to the company?
13-U. Given the following two-stage decision situation shown in Fig. E 13-12. deter-
mine which is the best initial decision. Notice thal sorne annual costs are
negatively signed, which makes them positive cash flows. Use the expected
PW of costs method and a MARR of 12%. To give the problem a physical
context, the following letter symbols have been employed for each alterna-
tive:
BSW: Build small warehouse.
RLW: Rent large warehouse.
BA: Build addition.
NC: No change.
13-13. A firm must decide between purchasing an automatic machine which costs
$50,000 and will last 10 years and have 0 salvage value, or purchasing a
manual machine which costs $20,000 and will last 5 years and have 0 salvage
value. If the manual machine is purchased initially, a ft er 5 years a decision
will have to be between a manual machine having the same characteristics
affecting cost as the first manual machine and a semiautomatic machine cost
ing $40,000 which would have a $20,000 salvage value after 5 years of life.
The annual operating tosts for each of the machines are as follows : auto-
matie, $10,000; manual, $14,000; semiautomatic, $11.000.
(a) Graphically construct a decision tree to represent this situation.

N
-
376 Risk and Uncertainty Chap. /3
First cost and resale
value of alternatives
Annual cost (or savings, if negative) Probabillties
-<-.
"J.

$-2,000/yr (10 yr) 0.15
11 ,000/ yr (3 yr) and 110,000/ yr (7 yr) 0.15
1- 8,000/ yr (7 yr) 0.2
12.000/ yr (7 yr) 0.8
$1 ,000/ yr (7 yr) 0. 2
118,000/yr (7 yr) 0.8
118,000/ yr l10yr)
118,000/ yr (3yr)and $21,000/yr (7 yr) 0.15
$23,000/ yr (3 yr) and $21 ,000/ yr (7 yr) 0.56
123,000/yr (10 yr) 0.14
Figure E13-12. Two-stage decision situation for Exercise 13-12.
(b) Determine which decision would be made at each point using the PW of
cost s method and an interest rate of 10%.
(c) At what interest rate would the decision between the manual and semiau-
tomatic machine be reversed?
13-14. Suppose one is faced with the same alternatives and dollar outcome conse-
quences as in the replacement problem depicted in Fig. 13-8. However, the
initi al estimates of probability of demand are: high, 0.6; low, 0.4. Further-
more. management's assessment of confidence in further investigation
results, using the notation in Table 13- 10 are
P(h !Hl = 0.80
P(hiL) = 0.40
P(liH) = 0.20
P(I!Ll = 0.60
Calculate the choice at each decision poini to determine the best initial dcci- .
sion. How close is the initial decision with these revised probabilities to the
initial decision for the original problem depicted in Fig. 13-8?
13-15. Figure E13- 15 is a decision trec portrayal of a building lease versus buy
problem with input data supplied. lnvestment requirements are shown as
negative numbers : probabilities associated with each outcome are shown in
Chap. 13 Exercises Jll
$20,000/ yr, 25 yr
1 15,000/yr, 15 yr
Keep 11 0,000/ yr, 1 0 yr, 10 salvoqe
Abandon $0/yr, 10 yr
Salvage 120,000
$0/yr, 15 vr. salvage $25,000
$20,000/yr, 15 yr
Keep $15,000/yr, 15 yr, 10
Abandon 10/yr, 15 yr
Salvage $30,000
$20,000/ yr, 25 yr
110,000/ yr, 15 yr
Keep $5,000/ yr, 10 yr
Abandon $0/yr , 10 yr
Abandon 10/yr, 15 yr
Figure E13-15. Decision lree for Exercise 13- 15.
parentheses. The annual cash savings and duration of those savings are
shown together at each relevant outcome. Salvage values in the cases of
abandonments are assumed to occur at the end of the 25-year study period .
Determine the best decision using the expected net PW method with a mini-
mum required RR of 0%.
13-16. Set up a decision tree to reHect the persona! automobile alternatives which
you expect over the next severa! years. Carry the trec far enough in ti me to
show severa! decision points at which a decision must be made between
keeping an old car and buying a new car (from a possible choice of several).
Show your roughly estimated assumed certain investment costs and salvage
values and an nuai operating costs for each alternative and determine the best
initial decision using the PW method and a 10% minimum persona! opportu-
nity cost of moncy.
13-17. Suppose, for the automation problem for which prior probabilities are shown
in Fig. 13-5, the probabilities of the possible outcomes for the added study
(technology study) are as follows :
V\
\

('0
371!
Givcn lhnl lhc
mllomnlion will
IUrn OUI 10 be:
Pnor
Fair
Excellcnl
Ri.<k and Uncertainty
Thcn lhc prohahilily lhal
lhc lcchnnlogy
will indicnle - --- is:
Shaky Sol id
0.6 0.3 0. 1
O.J 0.4 0.3
0. 1 0.4 0.5
Clzap. 13
lise Rayes" thcorcm to cnlculnlc the posterior prohnhililies regarding whal
the automation results will lurn oui ln he. givcn whal lhe lcchnology sludy
in<l ic ntcs . Chct"k ynur rcsulls n!!ainsl lhe pmhahilitics shown in Fig. 13-6.
J\
1


6-1
CHAPITRE6
LMENTS DE BASE DE LA SIMULATION DES SYSTMES
6.1 Technique de Monte Carlo :
Mthode dveloppe durant les annes 40 par Von Neuman, Vlan et Fermi pour rsoudre
certains problmes de design d'crans anti-rayonnement. Elle est coteuse exprimentalement et
difficile rsoudre analytiquement.
Elle consiste reprsenter un problme dterministe par un processus stochastique dont les
distributions de probabilit satisfont les relations mathmatiques du problme dterministe complexe.
La premire tape consiste construire un modle analytique qui reprsente l'investissement considr
(exemple: une quation pour la valeur prsente).
Dans la deuxime tape, l'analyste dveloppe une distribution de probabilit pour chaque
facteur qui est sujet l'incertitude dans le modle. Il peut utiliser des donnes historiques ou une
approche subjective.
partir des distributions de probabilits pour chaque facteur dans le modle. Il gnre au
hasard une rponse tentative. La rptition de ce mode d'chantillonnage un nombre important de fois,
permet de gnrer une distribution de frquence pour la rponse. On peut utiliser n'importe laquelle
des mthodes de mesure de rentabilit. La distribution de frquence rsultant peut tre utilise pour
obtenir des donnes probabilistes sur le problme original.
Exemple 6.1 : Calcul de 1t
Si on met alatoirement un certain nombre de points (n) dans le carr.
Un certain nombre de points appartiendront au cercle et d'autres non
- Quand n est grand, la surface du carr sera couverte.
n => oo ~ = >
#cercle
# total
nb. pt. _ds c ercle
nb. pt. ds carr
surface cercle
surface du carr
(2a)
2
n--
4
nb. ds cercle
n = 4 x ------
nb. ds. carr
a = 1 M = [ : : ]
r
1
r
2
des nombres alatoires entre -1 et 1.
On a gnr 30 pts, 24 e cercle et 6 e cercles.
n = 4 * ~ = 3 .2
30
Sin-oo, 1t-3.14
Exemple 6.2 : Le pari
6-2
Supposons que l'on vous propose de jouer au jeu suivant: pour une mise de 20 dollars, on vous
permet de jouer pile ou face avec une pice bien quilibre et de gagner 2\ x tant le nombre de lancs
ncessaires pour avoir pile. Accepterez-vous de faire la mise pour jouer? Quel est le montant maximum
(suprieur ou infrieur 20 $)que vous accepterez de risquer pour jouer? Votre rponse dpendra de
votre got pour ce type de risque. Je n'ai encore jamais rencontr quelqu'un qui dit accepter de miser
jusqu' 100 $. Pourtant. ..
Une approche analytique de ce problme de pari suggrera le recours l'esprance mathmatique.
Dans ce cas, on arrivera la conclusion qu'il faut toujours jouer ce jeu quel que soit le montant de la
mise demande, car l'esprance mathmatique du gain est infinie!
C'est ce qui ressort du calcul ci-aprs.
6-3
E ( gai n) = " ( 2n 2n- 1 2 )
LJ 2 . 2
E (gain ) =L 1
= !><)
Une approche de simulation est facile concevoir et mettre en oeuvre: essayez sans frais, avec
une pice de monnaie, de jouer pile ou face et dterminez ce qu'auraient t vos revenus si vous aviez
rellement jou.
Vous pourriez obtenir les rsultats suivants :
1
2
3
4
5
ftfp
fp
ffifp
p
p
Chaque essai est bien termin avec une sortie de pile et les montants gagns sont de 2 une
puissance gale au nombre de fois qu'il a fallu lancer la pice pour avoir pile.
la vue de ces rsultats on va tre frapp par le gros gain comme 32 ou le petit 2 selon qu'on a
le temprament de joueur ou non. En faisant la moyenne des gains on trouve 56/5, soit environ 11 . Mme
si ces rsultats ne sont pas trs significatifs (chantillon trop petit et probablement biais par la
manipulation), ils livrent nanmoins un message intressant: au del de 30 la mise est trop risque.
La simulation ne nous donne pas LA solution; elle nous donne des indications nous permettant de
prendre une dcision selon notre temprament. Cette situation est confortable pour le dcideur car on ne
lui impose pas une solution fut-elle optimale.
L'exemple du pari va nous permettre d'illustrer aussi un aspect essentiel de la simulation Monte
Carlo : la simulation de la ralisation d'un vnement partir de la probabilit attache la ralisation de
cet vnement. Une pice de monnaie permet normalement de bien simuler le comportement d'une autre
pice de monnaie quivalente.
Cependant les rsultats obtenus par le jet d'une pice de monnaie ne sont pas tout fait alatoires:
les faons de jeter la pice, de la saisir sont propres l'oprateur et non au hasard. Une autre limitation
de la technique de simulation par le jet de la pice de monnaie apparat lorsque l'on envisage de simuler
la ralisation non pas de deux vnements quiprobables (pile -face) mais de plusieurs (pile 1 fois ou 2 fois
ou 3 fois ou 4 fois en 4 lancs).
6-4
La simulation Monte Carlo permet d'arriver au mme rsultat que la simulation manuelle du jet
de la pice de monnaie. En faisant la correspondance suivante :
pile= {0, 1, 2, 3, 4}
face= {5, 6, 7, 8, 9}
et en supposant que lors d'un tirage au sort parmi ces nombres quiprobables la sortie d'un chiffre traduit
le fait que la pice de monnaie a montr le ct (pile ou face ) correspondant. Par exemple, si le nombre
tir est 2, c'est pile; 8 c'est face, etc. Il ne restera plus alors qu' trouver un systme de gnration de
nombres alatoires pour simuler le jet de la pice de monnaie.
Ce procd a l'avantage d'tre plus neutre. En plus, il permet de faire face au cas o il y a plus de
deux vnements. Si nous avons trois vnements par exemple, il suffit de connatre la probabilit attache
chacun de ces vnements et de faire la correspondance des chiffres consquemment.
Exemple 6.3 : Estim des ventes annuelles
On a estim que les ventes annuelles d'un produit d'entretien mnager ont 60% de chance
d'atteindre 200,000 dollars, 30% de chance d'atteindre 300, 000 dollars et 10% de chance d'atteindre
400,000 dollars.
Pour simuler le comportement des ventes annuelles, on fera les correspondances suivantes sur la
base de 10 :
200 000$
300 000$
400 000$
60% {0, 1, 2, 3, 4, 5}
30% {6, 7, 8}
10% {9}
Il ne restera plus qu' tirer les nombres alatoires pour simuler le comportement des ventes
annuelles. Si les pourcentages n'taient pas arrondis, il aurait fallu faire la correspondance sur la base de
100 nombres. Essayez pour l'exemple prcdent : 61%, 29% et 10% respectivement. Essayez ensuite
61.5%, 28.5% et 10%.
Exemple 6.4 : Comparaison de rsultats analytiques et de simulation
[!] Une clinique mdicale dispose de 5 centres.
[!] Une ( 1) infirmire est affecte chaque centre.
[!] Il y a 3 infirmires rservistes qui peuvent tre appeles pour remplacer celles qui
s'absenteraient.
Les infirmires peuvent s' absenter en cas de maladie. La distribution de probabilit des infirmi res
6-5
malades est reprsente dans le tableau suivant :
Nb. de malades 0 1 . 2 3 4 5
Probabilit .2 .25 .2 .15 .1 .1
On dsire simuler 25 jours de fonctionnement de la clinique l'aide de la technique MONTE
CARLO pour obtenir des estims de :
[!] le taux d'utilisation des infirmires rservistes;
[!] la probabilit qu'au moins un centre mdical soit ferm par manque d'infirmires.
Feuilles de statistiques pour la simulation du cas des infirmires.
NIG
NRU
NJCF
NIG
NRU
NJCF
Analytiquement
Nombre moyen d'infirmires malades=
Nombre moyen d'infirmires rservistes utilises=
La probabilit qu'au moins un centre mdical soit ferm par manque d'infirmires =
Exprience de simulation
Nombre moyen d'infirmires malades =
Nombre moyen d'infirmires rservistes utilises =
La probabilit qu'au moins un centre mdical soit ferm par manque d'infirmires =
Disque d'chantillonnage pour le cas des infirmires :
D
1
s
Q
u
E
NIG U) = 0
NIG U) = 1
NIG (j) =nombre d'infinnires malades au jour j
6-6
Logigramme du modle de simulation
INITIALISA noN
NRU = NJFC = 0
NIG=O
UIR = NRU/75
PCF = NJCF /25
NJCF = NJCF + 1
NRU= NRU+ 3
NIG = NIG + NIG UJ
_j 1 Ill lill 1 1 1 1 Il 1 1 1 1 Il 1 Il 1 1 1 1 1 1 1 1 1 1 1 1 Il 1 1 1 1 1 1 L
GLOSSAIRE
NJCF = #jour oa au moins 1 centre ferm
NRU = # rserviste -jours utiliss
NIG (j) = 11 infinmire en grve au jour j
UIR = utilisation des infinmires de rserve
PCF = probabilrt qu'au moins un centre sort fenm
1 11111111 Il l Ill 111 11 11 Ill 1 I l Ill lill 1 I l Il l I l 1 1
6-7
6-8
6.2 Considrations dans la sim.ulation de Monte-Carlo
Une question importante connatre dans la simulation de Monte-Carlo est le nombre d'essais (ou
d'excution) ncessaire pour avoir une rponse satisfaisante. La rponse est que le nombre d'essais doit tre
assez lev pour atteindre le rgime permanent. On peut dfinir les conditions de rgime permanent comme
une situation dans laquelle le rsultat d'essais conscutifs ne varie pas significativement.
Dans un simulateur, on utilise habituellement une priode de rchauffement avant de collecter les
statistiques, l'utilisation des simulations de Monte-Carlo prsente des avantages pour:
Les situations dans lesquelles des mthodes de rsolution analytique n'ont pas encore t trouves;
Des cas o les modles analytiques sont trs complexes rsoudre;
Ne pas perturber le fonctionnement de la compagnie;
Des situations pour lesquelles il est trs difficile ou dangereux de crer les mmes conditions
d'opration pour chaque exprience;
Faciliter l'exprimentation de plusieurs possibilits qui ne peuvent tre ralise en vraie vie;
Des situations qui peuvent coter chres et/ou prennent beaucoup de temps pour gnrer le rsultat
voulu.
Les modles analytiques et la simulation de Monte-Carlo ont les mmes problmes, savoir, la
validit du modle. Cependant, dans le cas de la simulation de Monte-Carlo, la taille de l'chantillon doit tre
assez grande pour dcrotre la variation d'chantillonnage un niveau acceptable. Les probabilits d'es
vnements doivent tre bases sur les jugements de personnes impliques dans le projet, par la suite,
l'analyse doit tre dose sur ces estims bien qu'ils soient subjectifs tant donn que chaque modle est aussi
bien que les estims de ses paramtres d'entre, la simulation de Monte-Carlo doit tre utilise avec le
jugement du dcideur.
6.3 GNRATION DE NOMBRES ALATOIRES
Les systmes habituellement simuls comportent des processus stochastiques;
ex. : demande pour un produit;
temps d'opration sur une machine;
arriv de clients une station de service;
temps de ralisation d'un projet.
etc.
Pour simuler adquatement la ralisation des vnements dont on connat les probabilits de
ralisation, il ne suffit pas de faire les correspondances que nous venons de voir. Il faut avoir aussi un
systme de gnration de nombres quiprobables (ou alatoires). Malgr l'abondance des sries de chiffre
dans notre entourage: bottin de tlphone, numro de rue, etc., on ne compte que deux sources fiables de
nombres alatoires: les tables de nombres alatoires publies (on en trouve dans les annexes de la plupart des
manuels de statistiques et de simulation) et les gnrateurs de nombres alatoires des ordinateurs.
6-9
Les tables se prtent l'utilisation manuelle. Les nombres contenus dans ces tables ont t soumis
divers tests statistiques pour s'assurer de leur caractre alatoire dans toutes les directions: horizontale,
verticale et oblique. Tous les chiffres sont quiprobables; il en est de mme des nombres de deux chiffres,
de trois chiffres. Lorsque l'on veut utiliser ces tables, il est conseill de fixer la direction de lecture de
nombres et la position de dpart et de ne pas revenir sur ses pas. Voir la table de nombres alatoires.
Les ordinateurs possdent des gnrateurs de nombres alatoires. Il suffit gnralement de les
voquer. On peut aussi programmer la gnration des nombres alatoires ou les calculer soi-mme.
On a donc besoin de gnrer des variables alatoires (dpendantes ou indpendantes) ayant des
distributions spcifies. C'est--dire un chantillonnage d'une certaine population.
L'ide est : gnrer des variables alatoires uniformment distribues;
-t gnrer des variables d'autres distributions.
6.3.1 Distributions uniformes
Ses caractristiques sont :
1) Chacun des nombres de la population a la mme probabilit d'tre dans la srie :
2) Chacun des nombre est compltement indpendant de tous les autres dj gnrs.
....
Variables alatoires indpendantes uniformment distribues .
6.3.1.1 Mthode manuelle
Pile ou face, ds, cartes, roulette, etc.
Avantages : Trs simples, honntes.
Dsavantages : Lentes, peu pratiques, impossible reproduire.
6.3.1.2 Tables
Pris dans n'importe quelle direction prdtermine, partir de n'importe quel point.
A vanta ge : reproductibilit
Dsavantages : Lenteur de l'emploi, utilisation continuelle des mmes squences (tables limites).
6-10
Table de nombres alatoires
Table A.t. RANOOM DIGITS
94737 08225 35614 24826 88319 05595 58701 57365 i4759
87259 85982 13296 89326 74863 99986 68558 06391 50248
63856 14016 18527 11634 96908 52146 53496 51730 03500
66612 54714 46783 61934 30258 61674 07471 67566 31635
30712 58582 05704 23172 86689 94834 99057 55832 21012
69607 24145 43886 86477 05317 30445 33456 34029 09603
37792 27282 94107 41967 21425 04743 42822 28111 09757
01488 56680 73847 64930 11108 44834 45390 86043 23973
66248 97697 38244 50918 55441 51217 54786 04940 50807
51453 03462 61157 65366 61130 26204 15016 85665 97714
92168 82530 19271 86999 96499 12765 20926 25282 39119
36463 07331 54590 00546 03337 41583 46439 40173 46455
47097 78780 04210 87084 44484 75377 57753 41415 09890
80400 45972 +till 99708 45935 03694. 81421 60170 58457
94554 13863 88239 91624 00022 40471 78462 96265 55360
31567 53597 08490 73544 72573 30961 12282 97033 13676
07821 24759 47266 21747 72496 77755 50391 59554 31177
09056 10709 69314 11449 40531 02917 95878 74587 60906
19922 37025 80731 26179 16039 01518 82697 73227 13160
29923 02570 80164 36108 73689 26342 35712 49137 13482
29602 29464 99219 20308 82109 03898 82072 85199 13103
94135 94661 8724 88187 62191 70607 63099 40494 49069
87926 34092 34334 55064 43152 01610 03126 47312 59578
85039 19212 59160 83537 54414 19856 90527 21756 64783
66070 38480 74636 45095 86576 79337 39578 40851 53503
78166 82521 9261 12570 10930 47564 77869 16480 43972
94672 07912 26153 10531 12715 63142 88937 94466 31388
56406 70023 27734 22254 27685 67518 63966 33203 70803
67726 57805 94264 77009 08682 18784 47554 59869 66320
07516 45979 16i35 46509 17696 67177 92600 55512 17245
43070 22671 00152 81326 89428 16368 S659 79424 57604
36917 60370 80812 87225 02850 47118 23790 55043 75117
03919 82922 02312 31106 05573 17470 25900 91080
46724 22558 64303 78804 05762 70650 56117 06707 90035
16108 61281 86823 20286 14025 24909 38391 12183 89393
74541 75808 89669 87680 72758 60851 55292 95663 883:!6
82919 31285 01850 72550 42986 57518 01159 01786 98145
31388 26809 77'158 99360 92362 21979 41319 75739
17190 75522 15687 07161 99745 48767 03121 20046 28013
00466 88068 68631 98745 97810 35886 90230 69264
6-11
6.3.1.3 Utilisation de formules rcursives
Utilisation d'quations mathmatiques de rcursion (ex. : partir d'un nombre, un autre nombre de
la srie est gnr). Puis qu'on peut reproduire ces nombres, ils sont appels pseudo-alatoire. Cette
reproductivit est leur avantage majeur.
6.3.1.3.1 Mthode du centre du carr
Chaque nombre alatoire compos den chiffres est form des n chiffres du centre du carr du nombre
prcdent.
Exemple : nombre alatoire de 2 chiffres (0 ;!; Xj ;!; 99).
Nombre
Xo= 12
X
1
=44
X
2
=93
X
3
=64
Source= 12
Xo
2
= 144
x
2
= 1936
1
X/= 8649
(NB. On pourrait aussi alterner)
Dsavantage : Dgnrescence rapide (cycle court) notons qu'aussitt qu'un nombre apparat pour
la 2ime fois, il y a cration d'un cycle.
6.3.1.3.2 Mthode du centre du produit
Chaque nombre alatoire Xj compos de n chiffre est form des n chiffres du centre de

ou
de

o K est une constante :
Exemple : Nombre de 2 chiffres
Nombre source X
1
=8; X
2
= 86.
X
1
=8
X
2
= 82
X
3
=56
X
4
=59
X
1
X
2
= 656
X
2
X
3
= 4592
6.3.1.3.3 Mthodes congruentielles
Additives : "n+I =(a "n + C) (modulo rn)
Multiplicatives : "n+I = ( a ~ (modulo rn)
{I(modulo rn)= reste de la division deI par rn}
Exemple:
Exemple :
13 (mod 5) = 3.
rn, c, a, sont des entiers positifs.
a= 2; rn= 11; X
0
= 4; c =O.
x
1
= (2*4) mod 11 = 8
x
2
= (2*8) mod 11 = 5
x
3
= (2*5) mod 11 = 10 ...
Notons que les nombres gnrs sont entre 0 et 10. 0 :S "n :S m-1.
TOUTES CES MTHODES POSSDENT 1 OU PLUSJEURS DES CARACTRISTIQUES:
les nombres peuvent tre reproduits;
possibilit de cyclage;
possibilit de dgnrescence (toujours mme nombre);
les n.a. sont uniformes.
6-12
Il existe des tests pour vrifier l'uniformit et l'auto-corrlation des sries (pas dans le cadre du
cours) .
La mthode Congruentielle est la plus utilise en pratique :
Poser rn = 2b (b entier positif)
a = 8t 3 (t entier positif)
a= 2 bn
ko = entier impair.
Tous les nombres seront :S 2b- 1.
Il y aura au maximum 2b-
2
rcursions avant de former un cycle.
Exemple : Si b = 21 ; 2
19
.
Pour U(O, 1 ), diviser les nombres par 2b - 1.
6-13
Exemple 6.5 : Prvision de la demande
La compagnie SOW Inc. cherche se faire une ide plus prcise de ce que sera la demande pour son
nouveau modle d'ordinateur. Ce type de produit informatique tant vite dsuet, la compagnie ne veut pas
se retrouver avec des invendus la fin de la priode de planification de trois mois.
Utilisons des donnes historiques pour simuler la demande trimestrielle.
.09
3 .27
2 4 .37
3 2 . 18
4 1 .09
Total= 11
Correspondances : 0 = {00, 01, 02, 03, 04, 05, 06, 07, 08} (9)
1 = {09, 10, 11, 12, 13, 14, ... '35} (27)
2 = {36, 37, 38, 39, 40, 41, ... ' 72} (37)
3 = {73, 74, 75, 76, 77, 78, 000 ' 90} (18)
4 = {91, 92, 93, 94, 95, 96, 97, 98, 99} (9)
6-14
Tirage de nombres alatoires:
Dans la table des nombres alatoires, prenons la colonne forme par les chiffres l'extrme droite
et associons chaque nombre trouv avec sa demande correspondante. Simulons 20 semaines.
- -=--, ... -. f -
"' .... -Je_atgire,..,,..... _. _-
95 4
31 1
13 1
86 3
02 0
79 3
85 3
87 3
87 3
64 2
54 2
59 2
39 2
03 0
52 2
53 2
22 1
34 1
74 2
38 2
Total=> 39
Ce qui donne une demande de 1.95 en moyenne, arrondie 2.
On pourrait penser utiliser l'esprance mathmatique pour faire l'estim de la demande partir des
probabilits attaches la ralisation de chaque volume de demande hebdomadaire. Il faut remarquer
cependant que l'esprance mathmatique est fige. Elle indique ce vers quoi tend la moyenne des demandes -
sur le long terme sans permettre dame nature d'tre capricieuse. La simulation permet ce caprice de la
nature: par exemple, mme si la demande de 1 unit ne survient normalement que 9 fois sur 100, la
simulation peut gnrer plus ou moins que cette proportion, donnant ce volume de demande un poids plus
grand ou plus petit dans la moyenne calcule. L'esprance mathmatique figera ce poids 9 pour cent.
6-15
Exemple 6.6 : Gestion des stocks
Modle analytique de base
Le problme de la gestion des stocks est bien classique: minimiser le cot total de cette gestion en
faisant le trade-off entre le cot occasionn par les lancements de commandes d'approvisionnement et le cot
de conserver des 1veaux levs d'inventaire. Un modle trs simple a t mis au point par l'approche
analytique sur la base de quelques hypothses simplificatrices.
Soient :
D la demande annuelle
A le cot de lancement d'une commande
H le cot unitaire de maintient des inventaires
L le dlais de livraison
R le niveau de stocks partir duquel on doit commander
Q la quantit commander chaque fois qu'il faut le faire
La quantit optimale commander note Q* pour minimiser le cot total de la gestion des stocks est
donne par la formule suivante :
Q* = /(2AD/H)
et
R =dL (d tant la demande durant le dlais de livraison R)
Exemple :
D = 1 000 units par an; A = 20 $ par commande; H = 4 $ par unit;
L = 2 semaines; alors :
Q* = v'(2*20* 1000/4)
= 100 units
R = (1 000 units/50 semaines) * 2 semaines = 40 units
On le voit : la formule de la quantit optimale commander est lgante et facile utiliser. Cependant
on comprend aussi qu'elle n'est approprie que pour des cas d'une simplicit rarement rencontre dans la vie
relle: demande connue, fixe et rgulire; dlais de livraison connus et constants, etc. La simulation permet
de faire face aux situations plus ralistes.
Modle de simulation
Ici on va supposer que la demande varie la semaine, que les dlais de livraison sont susceptibles de
changer; que des ruptures de stocks peuvent survenir et que cela comporte un cot estim 30 dollars par
unit manquante. Le lancement d'une commande cote comme prcdemment 20 $; on suppose que l'on
commence avec 40 units en inventaire; le cot de maintient des inventaires est de 4 $ par unit par semaine.
Par ailleurs on a obtenu les donnes suivantes sur les demandes hebdomadaires et les dlais de
livraison observs par le pass.
10
20
30
40
10
20
10
10
1
2
3
4
8
4
6-16
La simulation nous permettra de comparer plusieurs politiques alternatives dans la gestion des stocks,
chacune des politiques consistant dans une rponse aux 2 questions :
Combien commander (Q = ?)
Quand (R= ?)
Dans cet exemple manuel, on se contentera de faire la simulation pour la politique qui consiste
prendre Q = 40, R = 1 O.
Les semaines successives traduisant le droulement du temps; les nombres alatoires pour simuler
la demande de chaque semaine; les nombres alatoires pour simuler les dlais de livraison lorsqu'une
commande est lance; les demandes simules; les dlais simuls; les commandes lances; les commandes
reues; les units en stock; les cots de maintien; les cots de commandes et les cots de rupture.
6-17
!

.. ' "'(k;'' <



--.- "'


, . 1>
.. t
1\.. .Do'. )... ' ' <.J._

)'IJ.fi.
t D d D c ar B H A ru Tot
0 40
1 69 45 30 40 10 40 20 60
2 48 52 20 40 (10) 20 300 320
3 84 11 40 40 40 20 20
4 46 19 20 80 60 240 240
5 66 _76 30 30 120 120
6 27 38 20 40 10 40 20 60
7 49 41 20 40
(1 0) 20 300 320
8 69 74 30 40 40 10 40 20 60
9 68 95 30
_40 20 80 80
10 33 46 20 40 40 160 160
11 20 52 20 20 80 80
12 09 77 10 40 10 40 20 60
13 05 10 JO 40 20 20
14 72 52 30 40 40 10 40 20 60
15 80 98 40 40 40 160 160
16 24 59 20 40 60 240 240
17 08 27 10 50 200 200
18 50 42 20 30 120 120
19 08 31 10 20 80 80
?0
f)Q
_50 30_ 40 (10) 0 ?0 100 1?0
6- 18
Exemples 6.8: Dure de vie d'un quipement
Considrons la distribution des probabilits de la dure de vie utile d'une pice d'quipement :
An Probabilit
3 .20
5 .40 LP=l.O
7 .25
10 .15
Pour simuler la vie utile on assigne un nombre pris au hasard chaque valeur de faon ce qu'il soit
proportionnel aux probabilits respectives. On prend le mme nombre de chiffres significatifs que pour les
probabilits.
A Nombres
3 00-19
5 20-59
7 60-84
10 85-99
On simule une dure de vie utile en prenant un nombre au hasard dans une table cet effet. Si on
obtient un nombre entre 00 et 19 la vie utile serait de 3 ans.
Exemple 6.9: Paramtre suivant une loi normale
Considrons qu'un cash flow net annuel a une valeur moyenne de 50 000$, une dviation standard de 10
000$ et il suit une distribution normale. la valeur simule ou cash flow pour une priode de 5 ans peut
tre calcule
AN Dviation standard prise Cash Flow net
au hasard (DSH) [50 000 + DSH (10 000$)]
1 0.090 50 900
2 0.240 52 400
3 -0.448 45 520
4 0.295 52 950
5 -0.292 47 080
:E = 248 850 1 5 = 49 770$
Cette valeur moyenne se compare la valeur connue de 50 000$; erreur de 0.46%
6-19
Exemple 6.10: Paramtre suivant une loi uniforme
La distribution de probabilit qui dcrit un vnement au hasard est uniforme et continue avec une
valeur minimale A et une valeur maximale B. La rponse simule s'obtient:
OuNH:
(NH)M:
, N H
Reponse = A+ (B-A)
(N H) M
Nombre au choisi hasard
Nombre au hasard maximal possible. (9 pour un chiffre significatif, 99 pour deux
... )
Par exemple, la valeur de rcupration l'anne d'une pice d'quipement suit une distribution
uniforme et continue entre 8 000$ et 12 000$. La rponse simule pour un nombre au hasard de
74 se calcule:
Rponse = 8 000+_2i (12 000 - 8 000) =10 990$
99
L'utilisation adquate de ces procdures avec un modle prcis permet d'obtenir des
approximations valables. Un nombre suffisant d'essais doit tre fait pour que la valeur moyenne
cumulative demeure relativement constante.
Exemple 6.11: Simulation d'un cash flow complet
Considrons un investissement normalement distribu ayant une moyenne de
- 50 000$ et une dviation standard de 1 000$. Une vie utile uniformment distribue
minimale de 10 ans et maximale de 14 ans.
Les revenus annuels prvus $
35 000
40 000
45 000
probabilits
0.4
0.5
0.1
Les dpenses annuelles sont normalement distribues avec une moyenne de
- 30 000$ et une dviation standard de 2 000$. Le gestionnaire dsire connatre la probabilit
que l'investissement soit profitable, il utilise un taux d'intrt de 10% et la mthode de la
valeur prsente.
Le tableau 6.4 illustre l'utilisation de la simulation de Monte Carlo pour 5 essais. Pour
obtenir des rsultats plus prcis, il faudrait utiliser un ordinateur et des centaines d'essais.
On doit se rappeler que la prcision de la rponse dpend directement de la prcision du
modle et des probabilits utilises.
La valeur prsente moyenne des 5 essais gale 19 004/5 = 3 801$. La solution de ce mme
6-20
problme en utilisant 3 160 essais sur un ordinateur a permis d'obtenir une moyenne de 7
759$.
...
....
l:l
10
71
71
7.
72
70
...
...
....

10
...
...
lW
...
JO
....
....
....
2.
oiO
31
31
:M
32
La figure illustre les rsultats provenant de l'ordinateur. La valeur mdiane gale 6 700$ et
59.5% de toutes les simulations donne une valeur prsente de 0 et plus. Donc, le
management aura environ 4 chances sur 10 de ne pas atteindre le 10% de retour sur son
investissement.
. ................. .
. ............... .
. .............. ..' .... .
.........................
.........................
.........................
.........................
...........................................
30.
21
2t ~
24'
22.
10.
tt'
,. ' ....................................................... .
14'
,. ' ................................................................ .
10'
1.
4 .................................. ...... .. . ............................ i .... .
: 1 . : : ; : : : : ; : : : : ; : : : : ; : : : : ; : : :: ; : : :: ; :: : : :: : : : : : : ; : : : : ; ::: :; : :: : ; : : : : ; :: : : ; : : : : ; : : : : ; : :: : ; : : : : ; : : : 1 . . . ;
C:.... 8 I'G 15 20 2$ 30 35 oiO 41 JO 56 JO Ill 70 71 JO 85 90 115 100 106 1 10
t
. 211;:
l'-< E ,.,.._
Essai
1
2
3
4
5
Essai
1
2
3
4
5
Dviation
Standard
Hasard (DSH)
-1.003
-0.358
1. 294
-0.019
0.147
Nombre au
hasard
2
0
4
9
8
6-21
Tableau 6.4 Simulation de Monte Carlo
Investissement ( I) Nombre au Vie utile (N) Dure de
50 000$+DSH(1 000) hasard (NH)
1 0 + ( NH ) ( 14 - 1 0 )
vie utile (N)
999 arrondie
48 997 807 13.23 13
49 642 657 12.63 13
51 294 488 11.95 12
49 981 282 11.13 11
50 147 504 12.02 12
Tableau 6.4 (suite)
Revenus annuel (R) Dviation Dpense annuelle VP=-I+ (R-E)
35 000$ pour 0-3 standard au 30 000 +DSH (2 000) (P/A, 10%, N)
40 000$ pour 4-8 hasard DSH
45 000$ pour 9
35 000 - 0.036 29 928 -12 970
35 obo 0.605 31 210 -22 724
40 000 1. 4 70 32 940 - 3 189
45 000 1.864 33 728 +23 233
40 000 -1.223 27 554 +34 654
+19 004
<J'\
1
Problme faire:
r)
0'
J"
6-22
Une compagnie considre l'installation d'un nouvel quipement qui rduira les cots de
fonctionnement annuels de 5 000$. La valeur de rcupration de 1' quipement est zro la fm
de sa vie utile. Les cots d'installation d ~ quipements sont normalement distribu avec une
moyenne de 25 000$ et une cart type de 2 500$. La vie utile de l'quipement a une
distribution de frquence montre au tableau suivant:
dure de 8 10 12 14 16
vie
(anne)
Probabilit 0.1 0.25 0.35 0.2 0.1
La gnration de 20 nombres alatoires est montre au tableau suivant:
Exp. Nombre Dviation
alatoire nombre alatoire
1 48 0.951
2 86 1.065
3 25 0.742
4 89 0.579
5 40 0.844
6 16 2.323
7 20 -0.800
8 15 0.485
9 34 0.396
10 33 1.925
Utiliser la simulation de Monte-Carlo pour l'chantillonnage pour dterminer les quantits
suivantes:
1) Le taux interne de rendement de chaque exprience;
2) La distribution de frquence des taux internes de rendement et les probabilits associs;
3) La moyenne et la variance des taux internes de rendement.
6-23
6.3.2 Autres distributions de probabilit
Dans un modle de simulation on doit utiliser la bonne distribution de probabilit. Dans l'annexe
1, on montre l' importance d' un tel principe. Une foi , la distribution de probabilit choisie, le
simulateur est capable de la reproduire un nombre infinie de fois . La mthode utilise le plus
frquemment est la mthode de la transformation inverse.
Supposons qu'une variable alatoire X a une distribution dfinie par :
x
P[ X ~ x ) = F = Jf (x) d x
x x
La fonction cumulative de distribution est dfinie de 0 1 inclusivement .
Puisqu'on veut obtenir des valeurs successives de x qui correspondent fx(x), il suffit de gnrer
un nombre alatoire uniforme entre 0 et 1, U(O, 1) et de l'galer Fx(x). La transformation inverse nous
donnera x selon la distribution fx(x).
Mathmatiquement: gnrer r - U(0, 1)
fixer r =Fx(x)
ainsi x =Fx"
1
(r)
La Figure 6.2 prsente une illustration graphique du procd.
fx{x)
aires gales
noter ci-dessous:
1
Fx, (x)
1
1
X'b Xc
1
1
1
1 1 1 1
1 1 1 .
re =Fx (Xc) --t -- - - - - - - -t - - - j- - - 1
1 1 1 .
1
1 1
1
rb =f=x (x - - : - - - - - - - - - - - ~ - - - - ~
0 - . -
1
- -.-- --- --- --- --!.. --
1
(re-r b)=(r!Jr
8
)
1 -
1
Figure.2 : Gnration des n.a. x. et xb partir der. et rb "'N(O, 1)
x
o x=((rJ
x
6.3.2.1 Gnration d'une distribution exponentielle
X suit une expon.: Exp (li.)
insi f x (x) = ll.e - ,\x
0 < x < "' ; 0 autrement
X
et Fx (x) = J f x ( t) dt = 1 - e - ,\x
0
fixons r = 1 - e - ,\x ou r - U ( 0 , 1)
1 - r = e - ,\x
ln (1 - r) = ll.x
- ln(l- r)
,\
donc
= x
x =
- ln(l-r)
,\
= F-
1
(r)
X
6-24
Ainsi pour obtenir un chantillon d'une distribution exponentielle ayant une moyenne gale 1/.,
il suffit de gnrer 1) et d'effectuer la transformation inverse.
6.3.2.2 Gnration d'une distribution normale x -N(J.t, <J
2
)
La gnration de nombres alatoires normalement distribus n'est pas aussi simple que pour la
distribution exponentielle, car il est plus difficile d'obtenir la fonction inverse de Fx(x).
Pour cette raison, on utilise la proprit suivante :
Si r
1
et r
2
sont 2 nombres U(O, 1) indpendants, on a que
xl= (-2ln rlY'" cos 2nr2
x2 = (-2ln rl?' sin 2nr2
sont deux variables alatoires normalement distribues avec =0 et cr
2
= 1
Pour gnrer des. nombres d'une distribution N()l,cr
2
) on utilisera l'une ou l'autre des identits pour obtenir
un X N(O, 1) et on le transformera comme suit :
X= (-2 ln r
1
Y'" cos 21tf
2
Y=
Pour obtenir Y

Notons que cette transformation est base sur l'identit


Y-u 1 ), expression connue pour la loi normale.
cr
6-25
6.3.2.3 Gnration de variables stochastiques de distributions empiriques.
Il est habituellement utile de pouvoir gnrer des nombres entiers selon une distribution empirique
(donnes exprimentales par exemple) non-exprime sous forme analytique.
Les donnes peuvent tre regroupes sous la forme suivante :
i
Valeur b P(x = b) = Pi fx (b) = L pj
j = 1
Ou sous forme graphique:
Fx (bi)
1- Pn
--- -- -------- -- --- - -.----l
P1 + P2 + P3
P1 + P2
P1
bn bi
6.4 RGIME TRANSITOIRE ET PERMANENT
On s'intresse au systme, via le modle de simulation dans son rgime permanent : (c'est--dire
dans ses conditions normales d'opration) . Mais, les 1r moments d'opration ne reprsentent que trs
peu les conditions normales d'opration car :
1- Les conditions initiales du modle sont gnralement vides (inoccup, zro) . Ces conditions
peuvent tre assez diffrentes des conditions normales de fonctionnement de systme rel:
- atelier rarement sans ouvrage;
- file d'attente rarement vide (dpend du taux d'arrive)
- hpital rarement sans malade.
2- Simuler consiste gnrer des N.A. et les faire interagir : ce processus ncessite un certain nombre
de valeur avant d'obtenir des tats significatifs.
Dans un simulateur on utilise habituellement une priode de rchauffement avant de collecter les
statistiques. Notons qu'il y a des systmes qui n'entrent jamais en rgime permanent ex.: conomie
nationale et industrielle, un rgime permanent peut consister en une oscillation.
Exemple : Arrive de clients la caisse.
#
1 10
2 4
3 10
4 4
5 10
6 4
7 4
8 10
9 10
10 4
11 10
12 10
13 10
14 4
15 4
16 4
17 10
18 4
19 10
20 10
21 4
22 4
23 10
24 10
25 10
26 10
27 10
28 4
29 10
30 10
31 4
32 4
33 4
34 10
35 10
36 4
37 4
38 4
39 4
,,..
Intervalle de 10 min. avec une probabilit de .5
Intervalle de 4 min. avec une probabilit de .5
Simule l'aide d'une pice de monnaie.
Si pile prochaine arrive dans 10 min.
Si face prochaine arrive dans 4 min.
a eau T bi d' une expenence
- -
1 Il # 1 Il #
10 10. 41 10 290 81 4
14 7.0 42 4 294 82 10
24 8.0 43 4 298 83 10
28 7.0 44 4 302 6.9 84 4
38 7.6 45 4 306 85 10
42 7.0 46 4 310 86 4
46 6.6 47 10 320 87 10
56 7.0 48 10 330 88 10
66 49 4 334 89 4
70 50 4 338 90 4
80 51 4 342 91 10
90 52 4 346 92 4
100 7.7 53 4 350 6.6 93 4
104 54 10 360 94 10
108 55 4 364 95 4
110 56 4 368 96 4
122 57 10 378 97 10
126 58 10 388 98 4
136 59 4 392 99 4
146 60 4 396 100 4
150 7.1 61 4 400 6.6 101 10
154 62 10 410 102 4
164 63 10 420 103 4
174 64 10 430 104 10
184 65 10 440 105 4
194 66 10 450 6.8 106 4
204 7.6 67 4 454 4
208 68 10 464 108 10
218 69 10 474 109 4
228 70 4 478 110 10
232 71 10 488 111 4
236 72 4 492 112 10
240 73 4 496 6.8 113 4
250 7.4 74 10 506 114 4
260 75 4 510 115 4
264 76 10 520 116 4
268 77 10 530 117 10
272 78 4 534 118 4
276 79 4 538 119 10
"'Of\
,..,..
~ ~
, ~ , . .
6-26
- -
-
1 Il # Il 1 Il
546 6.7 121 4 814
556 122 4 818
566 123 10 828
570 124 10 888
580 125 4 842
584 126 10 852 6.8
594 127 10 862
604 6.9 128 10 872
608 129 4 876
612 130 4 880
622 131 4 884
626 132 4 888
630 133 4 892
640 134 4 896 6.7
644 135 10 906
648 6.8 136 10 916
658 137 4 920
662 138 4 924
672 139 10 934
682 140 10 944
692 141 4 948 6.7
696 6.8 142 4 952
706 143 10 962
716 144 10 972
720 145 4 976
724 146 10 986
728 147 10 996 6.8
738 148 10 1006
742 149 4 1010
752 6.8 150 10 1020
756 151 4 1024
766 152 10 1034
770 153 10 1044
774 154 4 1048 6.8
778 155 4 1052
782 156 4 1056
792 157 10 1066
796 6.8 158 10 1076
806 159 10 1086
nn
,,,.. ,,..
11'\C\L
(min)
i
1
7.5 --.---
1
1
1
7 -.---
1
6.5 -.---
6
5.5
Transitoire Permanent
200 400 600 800 1 000
Stabilit croissante partir de 400 500 min x "" 6.8 min. On sait qu' la longue a tend vers
7 min. = (0.5 x 4 + 0.5 x 10)
6-27
Le rgime transitoire n'est pas trs reprsentatif de la ralit - minimiser l'effet de l'tat
transitoire.
1. Utiliser des expriences trs longues; pour ngliger l'effet des rsultats transitoires sur le total.
Ceci peut tre trs coteux en temps d'ordinateur.
2. Rejeter les rsultats des premiers moments, ex. : ne pas compiler les premiers 50 jours, on
compte lorsque 1 000 commande sont traites .. .
3. Utiliser des conditions initiales trs proches de celles attendues en rgime permenent. Ca peut
biaiser si le rgime permanent n'est pas atteint.
6-28
A quel moment le rgime transitoire se termine-t-il?
Il n' y a pas de rgle satisfaisante pour dterminer avec certitude et prcision l'entre en rgime
permanent.
1. Effectuer des expriences pilotes
2. Observer le comportement du systme
3. Dcider quoi rejeter
Le nombre source est d'une grande influence sur la priode d'instabilit.
Les mthodes suivantes ont t utilises avec un certain succs :
1. Rejeter une suite de mesure. La premire de la suite n' est ni maximum ni minimum des donnes
suivantes (Conway) .
2. Observer une squence de rsultats : si les nombres d'obs o le rsultat est > J..l est = le
nombre o le rsultat est <J..l alors le R.P. est probablement atteint (Emshojj, Sisson) .
3. Calculer la priode K la plus longue (durant une exprience pilote) pour laquelle les rsultats
ont une auto-corrlation significative; rejeter une priode initiale de mme longueur pour les expriences
suivantes (Fishman).
4. Le plus long cycle du modle doit avoir t excut au moins 3 4 fois avant que les effets
transitoires soient disparus (Tacher) .
5. Calculer en moyenne mobile des rsultats et supposer que le rgime permanent est atteint
lorsque cette moyenne ne change pas de faon significative avec le temps.
6. Utiliser diffrents nombres sources pour constituer un chantillon de mesure des dures de
simulation * et effectuer des tests sur des moyennes.
13-24. Use lhe Monte Carlo simulation technique to obtain five pairs of present
worth values for the alternatives described in Exercise 13-23. What percent-
age of the present worth combinat ions fa vors alternative A 1
,\(rl'fft' Cnrl,
1.\-1!\ . The of a is n random variable depemlcnt upon
the lifc ,,f the ncn>rding ln the f'>llowing tnhlc:
Prnhnhility ,,r VRille
l .ifc
s_,_ooo SIO,OOO Sl5,000 S20,000 $25,000
0.20 0.50 0.30
4 - 0.20 0.50 0.20 0. 10
6 O.JO O.JO O.JO 0. 10
R O.JO 0.50 0.20
Il is thought thal each of the asse! lives is equally likely to occur. If the
in vcstmcnt in the asse! is $50.000 and the interest rate is 15%, show how one
cnn ohtain a distribution of the cap.ital recovery cost for the asse! by sctting up
a table and generating fi vc trial outcomcs . Then calculatc cstimates of the
meon and variance.
1.'-19. The cstimatcd clement \>Utcomcs for a certain flexible manufacturing system
arc as follows :
ln vestmenl
Li fe
Net flow
vnl uc
Normally distributed with a mean of
$1,000.000 and a variance of 16,000,000
.'i years with probability 0. 2
7 years with prohability 0.7
9 years with probability 0. 1
Uniformly distrihutcd bctwcen $120,000
nnd $340,000 per year
0
Ali clement 0lllcomcs arc indcpendent of cach other. Demonstrate how to
0bt ain n distribution of the net A W by generating fivc outcomes . From this
1
Chap. 13 Exercise.1 )79
obtain estimates of the mean and variance of the net A W. Assume a mini
mum attractive rate of return of 10%.
13-20. A dicseltractor is expectcd to require an investmenl of $100.000 and lo have a
!ife which ean be best describcd by a uniform distribution with a minimum of S
ycars and a maximum of 15 years . The salvage value is cxpectcd to he
$40,000 if the lifc is less than 8 ycars. $20,000 if the lifc is R to 12 ycars. and
$15,000 if the life is 13 or more years . Show how lo build a distribution of the
capi tal recovery cos! for this project by gcnerating tivc outcomcs using the
Monte Carlo technique . Use an intcrest rate of 10% a nd round the proj ect
li ves to the nearest wholc ycar. From your rcsults cstimale the e xpcct ed
value and variance of the distribution of capital recovery cost.
13-21. Projecl X is expected to require an inveslmenl of $40,()()() and to ha ve a li fe
which is normally distribulcd with a mean of 5 years and a slandanJ deviation
of 1 year (rounded to the nearcsl intcger). The salvagc value is e xpec ted 10
vnry nccording lo the relationship $!!,000 - $1 ,000 x Clife in yearq .
13-22.
Project Y is cxpccted to requirc an invcslmcnt of a nd to
lifc which is uniformly distrihuted hctween 5 and 15 yc:1rs fr ounded tn the
nearest in leger). The salvage value is expectcd to be nil regMdl ess of li fe .
Show how to build a distribution of the difTerence in capit:ol recovcry costs for
the two projects (X minus Y) by gcncrating three out comcs the Mon te
Carlo technique. Interest is 15%. The project lives vary independently .
Records for a certain invcntory item type indicate thal the numbcr of da ys
lead lime required for replenishment can be expected to have the following
probability distribution:
Lead Time (days)
1
2
3
Probabilil y
0.25
0.50
0.25
Demand du ring any day of le ad ti me is a normally distributed random variable
independent of the number of da ys lead ti mc and with a mean of 4 units and a
standard deviation of 1 unit of the item. lnventory holding cosl.s average $100
per unit per yr and the opportunity cost of a stock-out (failing to have a unit
when demanded) is estimated at $70 pcr unit. Show how to use the Monte
Carlo technique to determine (eventu:o ll yl) the most econo mical si7.e of safety
stock for the item, wherc safety stock is detined as the difTerence between the
number of units in inventory whcn the item is reordered and the expec ted
demand during the Jead lime period . lllustrate by simulating tive reo rder
periods with safety stocks of 0 units and thcn 4 units .
13-23. Two invcstment alternatives, A and 13 , arc under consideration : one must be
seleeted. Alternative A rcquircs an initi al invcstmcnt nf $15.000 in equir
ment; annual operating and maintenance costs arc anticip:oted to be normally
distributed, with a mean of $6,000 and a stand:ord deviation of $1i00: the
terminal _ salvage value at the end of the 10-year study pcriod is anti cipated to
be normally distributed with a mean of $2,000 and a standard deviation of
$500. Alternative B req!Jires an investment of $5,000 and end-of-year annual
expenditures which are normally distributed with a mean of $9,000 and a
standard deviation of $900. Use a MARR of 10%. Generating five trials,
estimate the orobability !hat alternative A is the oreferred alternative.
(}'\
!:
' 1
j
'
0:
,. ,
, :0
- . ~ ~
40 Economie Concepts and Accounting: Financial Considerations Chap. 2
Determine which method is most economical if actual overhead costs for the
two methods are thought to
(a) Vary according to the allocation formula
(b) Be the same for each method
(c) Be $5,000 more for the new method than for the old method
2-14. (Section 2.4) Describe why the return on net worth calculation in Fig. 2-6 (for
example) is a short-term and relatively narrow assessment of the benefits
associated with an advanced manufacturing system investment.
tl
computer -integrated
manufaduring:
the new technologies
chapter7
3.1 INTRODUCTION AND HISTORICAL CONTEXT
The gross national product (GNP) of the U.S. economy is a barometer of the
nation' s real wealth generation. As of the tnid-1980s it was comprised of
service industries (63%), manufacturing industries (24%), and the extractive
and construction industries (13%) . [7) Extractive industries include agricul-
ture and mining, while service industries encompass banking, insurance,
health care, entertainment, and so on. Because activities of a nation 's serv-
ice industries by themselves can be said not to increase the financial , long-
lerm well-being of the populace, one can surmise from the percentages
above that manufacturing industries are responsible for about two-thirds [24/
(24 + 13)] of this country' s ability to crea te real wealth. Any tech nol ogy that
can significantly improve productivity in the manufacturing sector of the
U.S. economy will have a multiplied effect on our GNP.
The purpose of this chapter is to explore the various components of
computer-integrated manufacturing (CIM) and to better understand why the
justification of CIM systems offers a special challenge to engineering and
management personnel. Manufacturing is the conversion of raw materials
41
"'-.!
l-"
il '
~ :1;
':!il;
t.q
42 Computer-lntegrated Manufacturing: the New Technologies Chap. 3
into end products. It includes ali activities from the perception of a need for
a product through the conception, design, and development of the product;
to the production, marketing, and ultimately supporting the product in use.
Ali these activities are closely related and interactive; therefore, they are
best treated as parts of a single system. [4]
Modem manufacturing began in 1798 when Eli Whitney accepted a
govemment contract to deliver 10,000 muskets in two years. Up to that time
arms had been produced by individual craftsmen who made custom-ordered
pieces in small quantities. This method lacked both consistency and quality,
and Whitney decided to di vide the production process into small jobs which
required that ali component parts of the muskets be interchangeable. Whit-
ney proceeded to design hand tools. to produce these parts accurately, uni-
formly, and quickly, and he trained his workers to use them. Eli Whitney's
technique was the first dmonstration of interchangeable parts and mass
production.
ln 1857, both priee and quality determined who would win government
contracts in the competitive arms market when Samuel Colt began making
his legendary revolver. Colt realized the need for accurate and efficient
production and began concentrating on methods that would eliminate man-
ua! fashioning of parts. His strategy was to assemble a staff of machinists
who could build machine tools for the rapid production of precise forgings,
jigs, and stamped parts. The success ofthese new machines led to new firms
producing tools for newly emerging manufacturers of sewing machines and
typewriters.
In 1870, Isaac Singer combined new manufacturing techniques with
aggressive marketing to dominate the emerging market for sewing ma-
chines . Singer realized that mass market sales required acceptance by both
lower- and middle-class households, and could be attained only by produc-
ing a variety of models consistent in quality and reliability. To accomplish
these objectives, he constructed a factory that ushered in a new era of
manufacturing. With the employment of gauging, forging, and milling ma-
chines, Singer soon produced interchangeable parts accurate to the ten-
thousandth of an inch, and then used these parts as standard components
placed inside a wide range of cabinets .
In 1908, Henry Ford revolutionized the manufacturing assembly proc-
ess. Assembly was organized according to the necessary flow ofwork rather
than the nature of the operation involved. This technology rapidly advanced
to the point where it reduced production time from 13 hours to 1 hour per
automobile chassis .
In 1946, ENIAC (the first digital computer) was produced. It weighed
30 tons, and used 18,000 vacuum tubes, 70,000 resistors, and 10,000 capaci-
tors. Two decades later, much smaller and more powerful computers be-
came the foundation for pioneering work in computer-aided design (CAD) as
researchers emp)oyed interactive computer graphies to display and manipu-
Sec. 3.2 Automation of Manufacturing Operations
43
late !ines and shapes instead of numbers. This was expanded to computer-
aided manufacturing-including robotics, flexible manufacturing systems,
and other automated systems.
Effective implementation of advanced manufacturing technologies
through a CIM system is the cornerstone of factory modernization. As a
company goes through a modemization effort, it typically identifies numer-
ous individual improvement projects. However, CIM must be viewed as a
total system which provides an automatic link between product design, man-
ufacturing engineering, and the factory floor . The methodology selected for
screening and prioritizing the se individual projects, with consideration of the
synergistic effect, is critical-it must provide a framework that ensures
consistency with the company's strategie objectives.
3.2 AUTOMATION OF MANUFACTURING OPERATIONS
Automation has been defined as "the technology concerned with the appli-
cation of complex mechanical, electronic, and computer-based systems in
the operation and control of production" [3]. This includes automatic ma-
chine tools, automated material handling systems, automatic assembly ma-
chines, robotics, computer-aided design, computer-aided manufacturing
systems, and other computerized systems. Tying ali these systems together
with an integrated data base architecture of computers is the basic goal of a
CIM system.
Automated systems are usually one of two types: fixed or programma-
ble. Fixed systems feature high production rates for high volume demand,
but they are inflexible, making changes in the process costly and difficult.
Fixed systems often are used in automatic assembly lines, transfer !ines, oil
refineries, and chemical processes. Programmable automation allows for
flexibility by making the systems suitable for low-quantity production of
different products. Figure 3-1 illustrates the integration of manufacturing
functions through a common data base (in the center), as is perceived by the
General Electric Company for the "factory of the future ."
Benefits achieved through automation are numerous and expanding. A
National Research Council report cited in Table 3-1 provides the findings of
interviews in five major companies having automation projects that ali re-
sulted in reduced work in process, excellence in design, lead-time reduction,
and improved productivity and quality. The values shown are representa-
tive of intermediate benefits of those companies' 10- to 20-year efforts.
Further benefits are expected as full integration is approached. Examples of
automation that typify the associated benefits of new technology within the
Generai.Electric Company are described in Fig. 3-2. Descriptions of various
types of manufacturing and the components of CIM systems are provided in
the remainder of this chapter.
'-J
N
:-,

44
"@

.,
0
"
""

g


:::>
:;
""' u
-5
....
0
t'
8

.,
-5
=
0
0

a
.,
.5
c
.g
"'

]
ci
.... u
,), u
'5

ti: ill
Sec. 3.2 Automation of Manufacturing Operations 45
Examples of Automation within GE
GE S<beacctady Proe.
CAD/CAM Boo1t1
Producdvlty.
The Thrblnt
Gentntor eq,aumenrln Schtn<"C
udy, NY, 11 one of thr l.aratu uen
o f CAOJCAM tqulpmtnt ln (;E.
They lntullcd Ont Catm:a
lmenctlve graphie ln
19'78. That tytrtm, :md tub.cqutnl
cnh:mcetncnlt, havi!' Ince- trlpkd
detlan and dnfrlng producrlvlty.
The rruon: lr ' s one o f the lr.cy
componenu ln rhe erne lon o f an
lncegrartd cnlllnccrl ngf
manuhccurlna dau1 bue.
C:ma CAO/CAM Sy1tem
Hel pa Save J 750.000
Annu:ly.
GE Schcnc-ct:ady 1 ;an excdknt
cumpiC" of :a tystcm
c:an llnlr. CAO CAM ln thc-
OC"IIIng.IDNC O:amc-cuttlnj\
operatio n ln 1hc bbrtc:aclon thop.
Utlng the paru d:au codcd
ttOrC"d ln the ho u computer, the
m:anuf:accurtna tpccl:allst c:al h up o n
the KJC<n :ail the- p:atii iO be CUl .lOd
the l i 1C" Of the' t tttl plate thC"y ' ll lw:
rue fro m.
1bc: PIallu thr"n 2rnnAn the
p:arta ln u doM: :a p:attem u
poulble, for the NC cool
p:uh. The- NC procHI data la
downl oadcd 10 the burnlng
muhlnct vi :a the' GE DNC llnk .
Orlgln:ally thlt Gf butlnnt
proj("cccd lncrc:ax ln
matcrl:al utlllnclon . They lncn:aacd
h
Thil will mule ln :a sa...lngs of
thr quaucn o f :a milli o n dotl an
pcr yC"Jr, an 18month
p:ayb;aclr: .
Gl!. Eric PrO'f'U P:actory
wldc Bcncnt of
Automatcd Machlnlna.
At Genenl F.lecul c' locomntlvlt'
m:anufacrurlna plana ln P. rie, PA , a
computcr contrnlkd Onlhle
machlnlnM ty"cm (FMS) w:a1
lmulled 10 muhlne a famlly o f tc n
mot or fu mn f or thelr li ne nf
l oc:nm(Hive. fra mn ln

ln C"xcn of 2, ltlt . ThcrC" :arc
urf:ac n 10 lw m:achlnC"d, more
th:an .,.lth tla,htt oln:ancc, .
Ni ne medium :and hu.,. y m:ac hlnc
10011 .,. orklnlliO!lt:lhn, cach
cqulppcd .,.hh Gcncnl EIC'Ctrlc OC
adjuuahk Jpccd splndlc drlvn
moton from o40
- 100 horKJ)().,.cr, :arc :an tmpor
une put o f ttw PMS. The m:achlnc
tool' conuln common drl.,.e
h:ard ... :ue chroua,hout , ue
cqulpptd torque prOfi:r:tm
mina :and Ill o n orient fnr cool
c h:analna. :and SIOrc o vcr 500
C'UUIR[tiOOII ,
A 2-4 , 1 Reduction ln Cycle
Tlme.
Bd ore the FMS .,.ent on llnc,
m:acblnlnA the motor ft11mct
29 machine toolt ,
:and the manuhcturlna c ycle- toolr.
16 d :a ya, the- nlnc m:achlnc-
tool sytC"m complete :a fume ln
juu 16houn.
The :auto mat\o n lnvc-ttm('('lt :alao
labor produc tiYity 20%,
lmpro.,.c-d part qual h y :and conslt-
t('('ICJ', ;and lncrnscd np;aclty } A%,
ustna 2S% noor tp:acc .
GB l!end.aJc ProTes NC Errecthen<
GE Louin Ille PtOYet the
V.alucotPCs.
ln 1979 , uur Olh,..nhtf'
PrrKJuc u h<E'aan pl:annln
tn :aut o mate lu m:anuf:ac turln"
WhC"n :actu:al produc tion
h<E'KJ.n ln the
projC"ct ln l.nulllvlllc , KY, h:ad mn
or exce-t"ckd :all i tt dntan :and
mJnuf:ac turtna ao:alt
ThlrtC"cn perce-nt reductio n ln choc:
toul cott of I2Mr. mJtttl:al , fu tory
ovcrhc:ad, trantporutton :and
78 % p:aru Jnd
:and :an Jnt iclp:accd
reduction ln Krvlcc
c:alls .
The m:anuhccurlnl yuan th :at
ddlvtn thnc lmprovcmcnu 11
cntln=l y sovemed by a dl u rtbut rd
contro l made up of H
Serin Six PCs. Thnc PC
concro llcd by :a computcrtud
F:accory M:an:aacmcoc System chat
coordln:atn the ('('ltl rc fac to ry
opentlnn, from o rckr nury 10 nruJ
and tculnft .
A 10% Reduction l.o the Total
Cot or Maout:a.cturlng.
Thlt computc-rlu:d tyttnn
prtorttl:r.n .,.orlr: to mttt produc
tl on tchrduln .1 minimum of
ln.,. cntory. Il upd:atet the' ;UtUI Of
ln progrnt baKd on d..:au
from ScriC'I Slx PCt . Il on CVCT'I
Knd JC"J)Jir lnfomullon co :a CRT ln
the malntmancc u for (:ut
rnponK.
Jl.cllablc, nulble PCt , plu
lntc-v:atlon of ali m:anuf:acturlnJ
compoMnt Yi:a IOphlulc:atrd
communlc:atlotU tc-ch nol ogy,
pl:aycd key roln ln m"tlnR :and
CXCdlnJ m:anub cturlnR JO"aJI fo r
rcduccd :and lm pro nd
pf'Oductlvlty and qu:allty.
Th(' General f:kctrk Ct1mp:any ' t Alrcrafc F.natnc- M:anuf:act ur.'
lng OIYblon ln I!.Yrnd:ak, O H, :autom:atlna lu ftct ory hy
:applytnaa f>NC to 111 n:ltlnll numC"rlnlly c nntrulled
machine cool buc. Toda y the h:u 101 NC :and CNC
pcrhap1 IURCII diltrlhutcd numcrlc:al co ntrol
ln che
c-d o n), CUl nY('UJI c-nn ua)'l 10 ll "e :an erfec tiW('
cap:achy inc rcuc cqulv:alcru to ddlna more machine' tflotlil .
T'hnoe lOt NC mJchlnct run on p:art provam
from an lntcgntc:d syucm ofmlnkomputcn. The: b:ulc dhtrlb
uted numC"tic.l control tystem, wlth che :additio n of :a nuchl nc
IOOittt ut mo nltOrlnJ IJICem te ill :li 1 sJ:anCC
nuchlnes :arC" ln K"rvlcc- , maintenance, o r boelnR
Productlvlty JmprOTem<nll or 10%. And StiJl
Cllroblna.
Wlth tht ncrallc co.t o f :a machine toolln t he ncla,hbof"hood
o f I'SO,OOO, QC'ttlnK thltlr:Jnd of c:ap:aclty throuRh NC tt<:hnol
Oii:Y :a IUblt:antlaluvlnp. And ln :additio n to th('K'
uvlnJI..' , the-re' :a;c :al.o bi& produc tiYity lmprovcmcnu: 10% to
date :and tt ill cllmbln1.
Figure 3-2. Examples of automation (courtes y of General Electric Co.) .
'-1
w
.,
l '
if; !
t\ :
46 Computer-Integrated Mamifacruring : the New Technologies Chap. 3
Table 3-1
BENEFITS ACHIEVED WITH DATA-DRIVEN AUTOMATION (BASED ON INTERVI EWS OF
fiVE MAJOR COMPANIES)
Reduction in engineering design cos!
Reduction in overall lead time
Increased product quality as measured by yield of accept-
able product
Increased capability of engineers as measured by extent
and depth of analysis in same or Jess time !han bef ore
lncreased productivity of production operation (complete
assemblies)
lncreased productivity (operating time)
Reduction of work in process
Reduction of personnel costs
15-30%
30-60%
2-5 times previous levels
35 times
40-70%
2-3 times
30-60%
5-20%
Source: National Research Cou neil, Computer Integration of Enginuring Design and Pro-
duction, National Academy Press, Washington, D.C., 1984, p. 17. Reproduced with permis-
sion of the publisher.
3.3 BASIC TYPES OF MANUFACTURING FIRMS AND
LEVELS OF AUTOMATION
There are three basic types of manufacturing firms as classified by their
associated production volumes:
1. Job shop production
2. Batch production
3. Mass production
Job shop production is characterized by low-volume production with
smalllot size (sometimes one of a kind) and it is often used to meet specifie
customer orders . This normal! y requires that the plant perform a great vari-
ety of work; therefore, the production equipment must be general purpose.
Aircraft industry contracts, as weil as special machines and deviees required
by the electronics industry, are in this category.
Batch production involves the manufacture of medium-sized lots of the
same item or product. The lots may be produced only once, or at regular
intervals. The purpose of batch production is often to satisfy periodic eus-
tomer demand for an item. The manufacturing equipment is usually general
purpose, but designed for higher rates of production. Machine tools are
often combined with special-madejigs and fixtures to accommodate diversi-
fied processes and increase production.
Sec. 3.3
Basic Types of Manu/ac turing Firms
47
Mass production is a continuous specialized manufacture of identical
products and is characterized by very high production rates, special-purpose
equipment, and very high demand rates for the product.
Typically, variable operating costs per unit decrease and total fixed
costs increase as the degree (leve!) of automation increases. As batch size
increases, production flow becomes more continuous and total costs are
spread over more units. When batch sizes are quite large, more specialized
fixed or " hard" automation typically is chosen because it increases capacity
and minimizes total costs.
Figure 3-3 shows typical total costs per unit for various production
quantities and levels of automation/technology. Note by the dashed por-
tions of the curves how a given leve! might be more costly than others for a
particular production quantity. Current and future production volumes and
processes must be understood to determine the type of automated system
desired.
Figure 3-4 reflects the "classic" conflict between capacity and flexibil-
ity that can be resolved by the application of a mid-volume, mid-variety
manufacturing systems concept. The productivity versus flexibility arrows
denote the divergent paths to greater achievement of those respective crite-
ria. Flexible manufacturing systems (in the middle) make possible an opti-
mum combination of these criteria for a wide range of production volumes
and varieties.
'
j
0
~
t;
8
'
::::>
Manuel teehnology,
generel purpose machines
Semiautomated teehnology
------------- Lower li mit (materials cott)
Production quantity (units/year)
Figure 3-3. Unit costs vs. production quantities for various levels of technology/
automation. Source: Reference [ 1]. page 50. Reproduced by permission of the
publisher.
"'-J
~
ti
48
.8
E

t
..
a.
..
a.
c:
0


o.
.,-
E
" 0
>
t
Computer-lntegrated Manufacturing: the New Technologies Chap.J
High


1l a.
:.0 -

manufacturlng
-+i ---,
1
1 1 1 cell
Standard and
convention al
machinery
Low
Low Medium Hlgh
----------------- Variety, part numbert per system-----------------
Figure 3-4. Productivity versus ftexibility. Source: Reference [5] . Reproduced
with permission of the publisher.
3.4 COMPONENTS/ELEMENTS OF COMPUTER-
INTEGRATED MANUFACTURING*
A computer-integrated manufacturing (CIM) system provides an automated
link between product design, manufacturing engineering, and the factory
floor. The major components of a CIM system include computer-aided engi-
neering, management, and control systems; manufacturing processes; and
information integration. Each of these components is described in greater
detail below.
3.4. 1 Computer-Aided Engineering
Computer-aided engineering (CAE) includes those hardware and soft-
ware tools, such as interactive graphies systems, that can be used in the
product design and manufacturing engineering functions. The functional
systems associated with CAE require integration of both design and manu-
facturing data bases. A description of these systems and the required inte-
gration follows.
This section was written substantially by Earle Steinberg and James A. Brimson and
excerpted with minor changes from A Framework for Computer Jntegrated Manufacturing,
Touche Ross, lnc .. Houston. Tex .. 1985. Reprinted with permission of the authors.
j
1lt
Sec. 3.4 Computer-lntegrated Mamifacturing 49
3.4.1.1 Computer-Aided Design
Computer-aided design (CAD) involves the use of computerized tools
to improve productivity in the areas of design, drafting, and testing. CADis
actually a means of accumulating, manipulating, and displaying related
graphical data electronically. The se graphical data can then be recalled for
duplication or modification.
A typical CAD system includes:
CRT workstations where engineers can create or modify a part by
"drawing it" on the screen using a number of different deviees, such
as:
A standard computer terminal keyboard
An electronic stylus that the engineer uses like a pen to draw on the
screen
An electronic sensitive menu tablet of commands that the engineer
can select
A digitizer that can con vert an existing print to X-Y coordinate points
which can be stored in the CAD system and retrieved on the terminal
Automatic drafting machines that produce high-quality drawings from
data stored in the system
A CAD system can provide engineers, designers, and drafting person-
nel with a number of significant benefits:
Time and effort required to develop designs is reduced because the
computer software can be utilized to draw figures, calculate dimen-
sions, rotate or section views, check interferences, and revise existing
designs.
The quality of designs, and their associated products, are improved as
the increased productivity of design personnel allows more designs to
be modeled, tested, and evaluated before engineers select the final,
production version.
The costs associated with the design of tooling and fixtures can be
reduced by designing them on the CAD system, in conjunction with the
design of the part.
Communication with engineering and production personnel is im-
proved by taking advantage of CAD's ability to display any view of an
object quickly and correctly and provide high-quality drawings.
CAD systems coming to market are increasingly powerful, offering
more sophisticated capabilities while being casier for design personnel to
use. The central reason behind the improvement in CAD systems is the
-.1
-.......!
U1
\'
r
1(;
.::
~ ~ .
~ :
lJ;
'


1
'
i'
,.
'
d
:' 1
50 Computer-lntegrated Manufacturing: the New Technologies Chap. 3
rapid development of computer technology, in both the hardware and soft-
ware areas . This development will probably continue, perhaps at an increas-
ing rate, into the foreseeable future. The effect of this on companies using
CAD will be:
Significantly increased display speed and pictures/drawings of higher
quality
More complex designs, tests, and simulations which will be made pos-
sible through more powerful computers with more real memory
Improved priee/performance ratios which will facilitate the support of
more workstations, extending the power of CAD to more areas of the
company, without significantly increasing costs
3.4.1.2 Design Analysis System
In most CAD systems, a wide variety of anal y ti cal software is available
for mathematical tes ting and modeling of part designs. Computer-aided test-
ing (CAT) uses a CAD-generated mode! of a product design to simulate the
operating conditions and performance of different materials. The design can
be interactively improved on the basis of the results of the simulations.
Another important element of a CAD system is the capability to ana-
lyze the impact of product design on manufacturing cost. It is extremely
important to conduct cost trade-offs earlyin the design process since it is at
this point that most of the production and logistic support costs are locked
in. and the economie leverage to reduce cost is the greatest. The manufac-
turing cost design subsystem should provide the following capabilities:
Cost comparisons with differing manufacturing processes
Performance/manufacturing cost trade-off information
Cost trade-offs of materials chosen for the product
3.4.1.3 Group Technology
Group technology is a method of classifying parts into families accord-
ing to similar shapes, or common manufacturing process operations. This is
accomplished through a coding mechanism that assigns specifie codes to
each significant part characteristic. These codes are then combined to form
a unique identifier for each part, describing it from an engineering and manu-
facturing perspective. Thus group technology facilitates a systems approach
to the redesign and teorganization of the factory into a flexible or cellular
manufacturing system.
Su ch a system consists of a number ofwork cells, which can be viewed
as a cluster or collection of machines designed and arranged to produce a
specifie group of component parts. Group technology plays a key role in
Sec. 3.4 Computer-Jntegrated Manufacturing 51
setting up a cellular manufacturing system because it provides a computer-
oriented tool that the manufacturing engineer can use to identify and design
the work cell, and the product engineer canuse to determine if similar parts
are in existence, thereby reducing the number of parts and process plans.
For standard families of parts, standard routings can be prepared that
need only slight modification for a newly designed part. The result is typi-
cally faster, more correct routings at Jess cost. Such standard routings can
then be used to group production machines in logical cells dedicated to the
manufacture of one or more families of parts. In addition, similar parts or
families of parts requiring approximately the same tooling and machinery
can be grouped so that production setup time is reduced.
Benefits of group technology and related work cells include:
Reduced machine setup times because work cells are designed to per-
form the "common" operations identified by group technology
Less work-in-process inventory because lot sizes can usually be re-
duced, smoothing the production flow
Less material handling because the flexible nature of the work cell
allows more operations to be performed within the confines of the cell
Easier analysis of production costs of parts in a given family because
outliers can be readily reviewed to determine whether the cost is wrong
or unusually high
Rapid development of cost estimates for new parts based on the costs
associated with other parts in the family
Easier tracing of the effects of priee increases on components and raw
materials back to the part families, and identification of the associ ated
impact on product cost
The end re suit of the work ce li approach to manufacturing is increased
throughput, lower costs, and better cost control.
3.4.1.4 ComputerAided Process Planning
Computer-aided process planning (CAPP) refers to the method of pre-
paring routings or process plans using computer assistance. Recently, two
approaches have evolved for this tas k. The first and most popular of the se is
the variant approach where process plans are generated for families of parts
that have been classified under the group technology concept. In this tech-
nique, the process planner calls up existing routings based on similarly
coded families of parts. Then, using a "same as-except for" technique, the
family's process plan is quickly amended to cover the individual part in
question. To create a plan from scratch in the variant approach, the process
planner can select from a computerized menu of operations based on part
characteristics. Here the process plan is built, line by line, using company-
v
'-J
0)
1
'1
t
,,
.,
; '

H\:1
'lJ
1
r i
: .,,



:'<rl j
;_!::
52 Computer-lntegrated Manufacturing: the New Technologies Chap. 3
oriented standard text. The procedure is somewhat similar to a manual
process. but the standard text, of course, is stored on the company's data
base ready for retrieval, modification, or hard-copy duplication.
The second, far more sophisticated computer-aided process planning
technique is the generative method. Here, the computer analyzes the part
under consideration and, based on part geometry, material, and so on, gen-
erales a process automatically. Not only is the sequence of operations
generated, but the computer also selects the company's best suited machine
tools and calculates the probable machine time for each operation. From
this base it is a simple matter for process plan ners to review and edit the
plan, and revise time allowances after initial sample parts are run.
CAPP typically increases the product planner's productivity and im-
proves the accuracy of the resulting plans. Both the variant and generative
methods use a "building block" approach to developing plans based on
common attributes within the part's family, reducing the planning effort and
thereby improving productivity. Accuracy is increased because the new
plan is "assembled" from pieces of already existing plans, which have been
tested in actual production.
The generative method is just beginning to be implemented in U.S.
manufacturing plants . The key to its more universal adoption will be the
development of better three-dimensional solid geometry modeling soft-
ware. In addition, much work remains to be done on the logic of process
planning based on this part geometry. However, the generative method is an
appropriate area for further investigation in modemization programs.
3.4.1.5 Machine Program Generators
Once the part's geometry is defined, it then becomes the basis for
future part machining operations . In the past, a design engineer wou Id finish
a blueprint, and the NC Programmer would have to take that blueprint and
redefine the geometry before he/she could start programming. With an inte-
grated system, the programmer starts with something in digital format.
Machine program generators pi-ovide a bridge between CAD and
CAM, where designs can be turned into the instructions necessary for com-
puter controlled machines. The programmer does not have to create new
instructions from scratch, but as with CAPP, he or she can build new in-
structions from the instructions of older family members. This is called
family-of-parts programming and increases the programmer's flexibility and
productivity.
3.4.2 Factory Management and Control Systems
Factory management and control systems provide for the flow of pro-
duction information through the integration of data from both engineering
and manufacturing functions . The planning and control function includes
"
l
Sec. 3.4 Computer-lntegrated Mamifacturing 53
production planning for product families, master production scheduling for
customer order promising and shipping schedule development, master re-
quirements planning for component scheduling and purchasing plans, pro-
duction activity control for shop operations scheduling, capacity require-
ments planning for shop loading, priority planning for daily dispatch, and
inventory planning for material control. These systems address the produc-
tion questions of wh en and how many, and facilitate the solution of produc-
tion problems, including machine utilization, control techniques, planning
capabilities, quality control, excess work in process, and poor use of skilled
tabor.
To be effective in this area, certain requirements must be met in the
areas of planning, facility loading, and dispatching.
3.4.2.1 Production Planning
Production planning translates the dollar goals expressed in the busi -
ness plan into Ievels of production for product families. This aggregate-level
plan is tested against available plant resources through the process of re-
source requirement planning and represents top management's key "control
knob" on the business.
3.4.2.2 Master Production Scheduling
A master production schedule represents a disaggregation of product
family production plans into individual end items (or major components
when "planning bills" are used).
Mas ter production scheduling must also simulate the resources , includ-
ing machine types, and personnel and inventory levels, required to accom-
modate the proposed production schedule or product mix. The simulation
capability uses forecast work load profiles of each product and projected
resource loads for the existing master schedule. Net changes in each type of
resource are identified, as are underloads and overloads of existing resource
capacities. This process is referred to as rough-cut capacity planning.
3.4.2.3 Material Requirements Planning
Material requirements planning (MRP) involves computerized tech-
niques for scheduling the replenishment of material, based on the master
production schedule and the components needed to build the product. MRP
provides the ability to update priorities and schedules frequently white pro-
viding suggested start and completion dates for ali open and planned work
and purchase orders.
3.4.2.4 Production Activity Control
Operations scheduling establishes projected start and completion dates
for al! activities on open or planned work orders and is required to automati-
j
"-..)
"-..)
'1 .,
i
54 Computer-lntegrated Manufacturing: the New Technologies Chap. 3
cali y verify availability of tools and materials for generating shortage lists. It
calls for delivery of required tools and materials when a selected Joad is
released, generates initial shop fioor Joad, and identifies bottleneck condi-
tions . This function provides for interactive intervention to undo bottle-
necks and a simulation capability for use in interactive intervention.
Production activity control also includes the dispatch function which
releases work to specifie machines and operators. While the dispatch
function can be either automatic or manual, a simulation and optimization
capability to assist in assigning jobs and operators to machines is re-
quired for either approach. Furthermore, if a manual system is being
used, it must have the capability to make detailed recommendations upon
request.
3.4.2.5 Capacity Requirements Planning
The information produced in capacity requirements planning is used by
manufacturing management, quality assurance management, shop fioor su-
pervisors, purchasing, accounting, and personnel departments. At a mini-
mum, capacity planning must: access inventory files and process planning
files, convert production requirements for personnel, identify overload con-
ditions, and provide the capability to evaluate alternatives .
3.4.2.6 lnventory Control
The activities and techniques of maintaining the stock of items at de-
sired levels, whether raw materials, work in process, or finished products is
generally referred to as inventory control. These activities involve stocking
levels, safety stocks, and lot sizes; and focus on managing the company's
investment in inventory assets.
3.4.2. 7 Purchasing Systems
Purchasing systems serve a dual role. First, they provide information
to buyers to aid them in negotiating with vendors and placing/expediting
orders. Second, the systems interface with MRP systems to provide feed-
back on actual order delivery schedules, order quantities, and lead times.
This information must be both timely and accurate, if MRP is to schedule
material replenishments correctly.
3A.2.8 Maintenance Planning and Control Systems
Scheduling preventive maintenance and collecting maintenance statis-
tics are now computerized. These systems allow the company to monitor
such maintenance activities as:
Sec. 3.4 Computer-lntegrated Manufacturing 55
Mechanics' productivity
Work order processing for requested jobs
Requirements for parts and special tools
These systems also allow the capacity planning systems to include
scheduled downtime in their calculations and to adjust work center loads
accordingly.
3.4.2.9 Engineering Data Base and Change
Control Systems
The engineering data base is also frequent! y referred to as the engineer-
ing bill of materials. It allows engineers to maintain their product engineer-
ing information on the computer in terms of parts lists, specifications, and
process requirements. The power of this tool is achieved when it is inte-
grated with the manufacturing bill of materials to facilitate engineering
changes, and reconciliation of product information.
The engineering change control capability enhances efficiency by main-
taining the history of product changes as they are released from engineer-
ing. However, its greatest values may lie in its capability to facilitate the
incorporation of the changes into the manufacturing bill of materials and its
ability to effectively manage inventory investment by controlling effectivity
dates .
3.4.2.10 Tool Control Systems
The relationship between planning and controlling production and the
need to coordinate tooling requirements is being addressed by many of to-
day's more advanced manufacturing and control systems. Time-phased re-
quirements for specifie tools are generated, based on production schedules.
This information is supplied to the tooling design and fabrication department
to identy the tool requirement dates necessary to support the production
schedules.
3.4.2.11 Energy Management
The increasing cost of energy in factories has led to the development of
computer software to manage energy resources. These systems can monitor
energy consumption, turn lights on and off, adjust heating or air conditioning
and provide management with data to include energy costs in their business
decisions.
3.4.2.U Warehouse Systems
Software packages designed to help warehouse managers maxmze
space utilization, make more efficient use of labor, and improve customer
'-J
<Xl
);
! '
..
t ;
{;
j
~ :
~ ':
,:j, .
l,
r
l
!'
;
"
J() Computer-lntegrated Manufacturing: the New Technologies Chap. 3
service are now available for a wide range of warehouses. These systems
provide a variety of functions, including:
Order processing
Forecasting
Finished goods inventory control
Systems offering these capabilities are available for companies from
one warehouse location to a worldwide warehousing network. Systems for
the latter generally are capable of playing a coordinating role in the manage-
ment of inventories and purchasing activities.
Properly implemented, the integrated planning and control system
eliminates the problem of missing resources. It ensures that ali resources
needed to perform a given job are identified, coordinated, and available
when thatjob arrives at its workstation. This reduces the amount ofwork in
process, normally held as butTer stock in input queues. Control of material
movement reduces the size of the output queues. These capabilities support
reductions in the amount of material held as in-process inventory. Flow
times for material moving through the production process are reduced ac-
cordingly. Utilization rates for machines and workers increase, since the
workers are doing productive work rather than waiting for tools, parts, and
instructions.
One of the m<\ior challenges in implementing factory management and
control systems is the difficulty in linking together different hardware and
software packages. There are currently many vendors and little standard-
ization of technical specifications. Additionally, a company that is modern-
izing must decide on the value of off-the-shelf versus custom-tailored sys-
tems . The modernization process should assess projected costs involved in
changing existing organizational procedures to fit the software systems being
offered rather than modifying the software to reflect organizational needs.
3.4.3 Computer-lntegrated Manufacturing
Processes
The automated shop floor includes computerized production machines
and equipment that facilitate the production process. CIM requires highly
automated and intelligent production facilities which are controlled by sen-
sors, minicomputers, and microcomputers. A CIM system has many bene-
fits from the machinists' point of view. Among these are:
Increased accuracy
Increased repeatability (1,000 parts, for example, can ali be created
alike)
Sec. 3.4 Computer-lntegrated Manufacturing 57
Decreases in the time it takes to go from a part print to the finished
product
Decreases in scrap due to better utilization of raw materials
Shorter production time frames
Increased machine flexibility
3.4.3.1 Computer-Aided Manufacturing
There are many definitions of CAM in use toda y. For purposes of this
discussion, we define CAM as NC, CNC, and DNC machine tool hierar-
chies, robotics, AS/RS, FMS, PCs and related technologies.
1. NC. Numerical control (NC) refers to a category of progrmmable
machine tools. In NC, a machine is programmed with a set of instruc-
tions designed for a particular work part or job. Based on the instruc-
tions, the machine will move its cutter tool through a series of X, Y,
and Z coordinates, performing the desired machining operations. One
of the advantages ofNC is that, when thejqb changes. the program can
be changed. This feature gives NC its flexibility . Instead of making
major changes to production equipment, relatively simple changes are
made to a program. The first NC machines were programmed by a
punched paper tape that contained the instructions for the speci fi e
job. Many of these machines are in use toda y and they are weil sulted
for a number of applications. Advances in computer power hove led 10
an extension of the NC concept to computer numericnl control (CN
and direct numerical control (DNC).
2. Standardized CNC controls . Computer numeri cal control (CNC:)
places a dedicated computer in or alongsi de the NC machin
Paper tapes are no longer necessary and machine tool instructi ons con
be created and stored electronically in the computer's memory or on
tape cassettes or diskettes. In modern CNC machines , programs can
b created, edited. or changed ifnecessary on the machine. CNC ellm-
inates problems of .physical tape storage or deterioration . Thus CN
machines perform the same function as NC machines, although their
method of receiving operating instructions is different .
3. Direct NC. Direct numerical control (DNC) offers real -time computer
control of more than one NC machine at a time. Many NC programs
are stored in a central computer's memory, on tape or disk. Not onl y
can the computer control multiple NC machines simultaneously, but it
can gather feedback from each machine asto part production rates and
machine status. With DNC and CNC capability at each machine, pro-
grams can be downloaded from the DNC to the CNC machine for
running. This saves the amount of memory needed for each machinr
1
l./
'-J
1..0
58
Computer-lntegrated Manufacturing: the New Technologies
Chap. 3
and means that the program only has to be created and loaded in one
machine (the host DNC)-not every machine that can run the job.
Over the past decade, NC machining has been proven to allow a pro-
ductivity gain of approximately 3 to 1 in most applications. Other benefits of
NC machining are, of course, reduced setup time, better part quality, Jess
scrap and rework, and reduced operator attention.
3.4.3.2 Robotics
Robots are computer-controlled deviees that automatically perform a
programmed sequence of operations. Most industrial robots in use today
were designed for specifie work such as painting, welding, plastic molding,
and assembly, and were, for the most part, laborsaving deviees utilized
sole! y for the performance of noncomplex production work. In addition to
such specific-purpose robots, many others perform material handling opera-
tions such as workpiece loading/unloading and workpiece transfer in con-
junction with other automated machinery.
The benefits of robots include:
Cost. Robots are becoming quite economical relative to the wages of
workers they displace.
Precision. Robots have a higher degree of precision than humans, and
with proper preventive maintenance can be highly reliable. Precision
re fers both to the concept of accuracy and repeatability.
Availability. Robots are tireless and can work three shifts with only
ti me off for preventive maintenance; th us robots are excellent for over-
loaded or complex structured work.
Safety and environment. Robots have reduced the exposure of hu man
workers to undesirable environments such as noise, heat, dirt, and
machine hazards.
The trend in robot technology is toward universality and versatility to
produce a standard unit that can handle diverse manufacturing problems.
The real future usefulness of robots will be determined by the power of the
computers that operate them, the sensors (touch and sight) by which they
receive real-time data, and the software that con trois them. In the future, as
pattern recognition software, artificial intelligence, sensory equipment, and
adaptive feedback mechanisms become available at even lower costs, robots
should be weil qualified to perform many assembly and parts selection tasks
now commonly performed by human beings.
3.4.3.3 Programmable Controllers and Microprocessors
Programmable control/ers (PCs) are microprocessor-based deviees
that integrate automated machines and equipment to provide a factory with
Il
Sec. 3.4 Computer-Integrated Manufacturing 59
real-time control. Unlike old fashioned, hard-wired, relay-based control-
lers, PCs are relatively inexpensive, solid-state deviees that can be repro-
grammed quickly to perform new tasks. Better still, different task se-
quences or programs can be stored in the PC's memory or on tape so that
process control changes can be effected with a change of tape cassette or
with a typed-in command.
The ultimate advantage ofPCs and microprocessors is that any number
of them can be controlled by central computers. These deviees can then be
used to report back on factory floor or process status and can be repro-
grammed quickly for different jobs when necessary.
3.4.3.4 Flexible Material Handling
Automated material handling systems permit an improved integration
of material flow to and within manufacturing functions . Sorne of the avail-
able technologies include wire guided vehicles, smart carts, and dedicated
transfer tines. The automated material handling system must be flexible to
permit changes in lot sizes, design, and product mix.
3.4.3.5 Automatic Storage-Retrieval System
Automatic storage and retrieval systems (ASIRS) are computer-oper-
ated part pickers and stockers. Generally, the Iarger pallet-sized systems for
warehouses are called AS-RS while smaller stockroom bin sized systems are
labeled automatic part pickers-stockers. Working either from direct com-
puter-generated and Iinked pick or stock instructions, or from manually
generated pick or stock instructions, these machines either will deliver a
batch of parts to an open or random location, while recording the location for
future reference, will stock a given part in a preprogrammed location, or will
go to a selected location to pick a part.
3.4.3.6 Computer-Aided Inspection
Computer-aided inspection (CAl) is another new use of computerized
. engineering design data in manufacturing, but this time in the quality control
area. In CAl, coordinate measuring machines (CMMs) controlled by soft-
ware that draws data from the company's engineering design data base auto-
matically measure parts to determine if they have been manufactured to
design tolerances specified on the part's drawing. These machines have a
probe that automatically moves to a programmed point, takes a measure-
ment, and displays or records the result. Sorne of the more advanced ma-
chines, becoming available or under development, replace the probe with a
laser, significantly increasing the machine's accuracy. The computer can
also print out both the required and actual measurement or dimension and/or
the deviation, if significant, for the part under consideration. More impor-
tant, the CMMs also enable these measurements to be stored automatically
,..
"-J
~
0
60 Computer-lntegrated Manufacturing: the New Technologies Chap. 3
in a data base, to be used as a basis for further statistical analysis on the part,
or perhaps as an earl y indicator of machine tool wear. These data also serve
to record permanent) y for the company the data necessary to meet traceabil-
ity requirements and demonstrate that a part was manufactured correctly.
3.4.3.7 Flexible Manufacturing Systems
Flexible manufacturing systems (FMS) permit the continuous manu-
facture of different items within a family of parts in small batches within a
dedicated machining facility. Flexible manufacturing systems, as shown in
Fig. 3-5, use the concept of integrated raw material storage, part picking,
part transportation, and DNC machining that are linked together in such a
way that the parts being worked on can travel from raw material storage to
finished goods storage in different sequences under the control of com-
puters.
The central FMS computer schedules and tracks ail production and
material movement in the FMS center. Based on a family of similar parts,
an FMS can be reprogrammed quickly through downloaded instructions
from a central computer to individual machines, conveyors, and part pickers
to perform a new set of tasks. The major benefit of an FMS. is fiexibility of
manufacturing resource assignment along with computer-controlled oper-
ation.
3.4.3.8 Bar Coding
The expanding role of the computer as a tool for management and
production is directly tied to the company's ability to provide it with timely,
Unload
Horizontal
boring
mill
Vortlc.ol
drill
Lat ho
Figure 3-S. A typical flexible manufacturing system.
Horizontal
....
Sec. 3.4 Computer-/ntegrated Manufacturing 61
accurate data. Many businesses are discovering that bar code scanning is
the fastest, most accurate, !east labor intensive, and lowest-cost method for
delivering that data. Companies are using bar codes to track work in proc-
ess, aid in physical inventories, control stockroom inventories and lots, and
track shipments to customers. Wherever data entry is required by produc-
tion, shipping, receiving, or inventory personnel, a potential application for
bar coding exists.
3.4.3.9 Laser Machining
Where extremely precise and/or delicate machining is required, lasers
are receiving much attention. Laser machining is capable of maintaining
very tight tolerances, in part because they are unaffected by the tool wear
inherent in more conventional types of machining. Recent advances in laser
machining systems, especially in the areas of laser positioning and process
control, have significantly enhanced the capabilities of laser machining.
Lasers can be used to machine complicated or unusual shapes, and are
excellent solutions to machining delicate components which dis tort easily,
because there is no machine tool contact with the workpiece.
lt is important to note one central feature of ali CAM applications that
we have touched on so far . With CAM, the skill requirement for an opera-
tion is transferred from the operator to the programmer who creates ali the..
machine instructions. No longer must a person running an NC lathe be a
skilled machinist. lnstead, the person can be a Jess expensive and more
readily-available machine operator. The same trend is evident in ali applica-
tions of computers in manufacturing processes.
On the other hand, CIM is promoting the integration of design and
manufacturing. With CIM these two previously separate groups work from
the same data base almost concurrently in the design process to design the
part, the tools and fixtures needed to manufacture the part, the bill of mate-
rials for the part, and the process plan for the manufacturing of the part . In
fact, sorne companies have created cells of people who work as a team
through one to three CRTs to carry out the design and manufacturing engi-
neering process.
Each cell is responsible for a product family or product group. People
in each team gain energy from their continuous interaction and not inciden-
tally come to have considerable pride in their group efforts . Such a team
approach to manufacturing design and operations is the way sorne people
view manufacturing in the future.
3.4.4 Information Integration Dimension
The proliferation of automation throughout engineering and manufac-
turing processes is making the factory extremely information intensive. The
~ ~
1
1
1(; \
!ill

i
1
l
1
1
il
! ~
.t
\i
li
.,
.,
:;
r
]
l
1
1
i
Il
:;
; ~
'f
~ ~
1
:i
j:!.
il
n ~
1. :.
'W
,,
"'..J 1 ~
,!
. .,
f-I. ~
f-I. ! ~
; ~
: j
! ~
r:J
.
'
<
i
,j
:j
'i
'
62 Computer-lntegrated Manufacturing: the New Technologies Chap. 3
successful system incorporates software links capable of allowing various
systems to communicate directly with each other. This is crucial to the
integration of a new system with other in-place systems. The task of proc-
essing and controlling information in the factory of the future requires a
reconceptualization of the role of data processing and the use of new tech-
niques .
3.4.4.1 Data Base Approach
As companies migrate toward the factory of the future, data bases that
support the production processes are becoming more electronic and less
paper oriented. Common shared data is a critical element in this process .
Much of the potential advantage of factory automation lies in integrating
engineering with manufacturing functions. The key is minimizing the need
for manual intervention between functions and processes.
1. Data base management systems (DBMS) . File-oriented information
systems have been the mainstay of computer applications since their
original ion. These traditional approaches are inadequate for the fac-
tory of the future since they hinder data sharing. The concept of data
base management systems (DBMS) is to overcome this limitation by
making data independent of application programs.
2. Information network. While a centralized data base is technologically
possible, the factory of the future will probably remain a distributed
data base environment due to the nature of the automation process.
The advanced manufacturing technology explosion has occurred con-
currently with the development of minicomputer and microprocessor
technology . This technology, along with local area networking which
permits efficient transfer of files between data bases, has provided
greater ftexibility and responsiveness at the machine and process levels
where the transformation process occurs.
The adoption of a decentralized data base concept will not by itself
ensure the benefits of this approach. ln order to maximize the value of a
DBMS, the design of data structures must be organized by logical function
rather than by narrow application. The value of a functional orientation is
that data can be shared by everyone in the organization, as required.
3.4.4.2 Decision Support Systems
Although very few management fonctions have been automated, ad-
vances in information retrieval, processing, and display technologies have
led to significant computer applications that help people perform manage-
ment functions. Si nee the purpose of these systems is to support managers
responsible for making and implementing decisions rather than to replace
them, these applications are often called decision support systems.
Sec. 3.4 Summary 63
Traditionally, data processing systems were designed to expedite and/
or automate transaction processing, record keeping, and business report-
ing. Decision support systems are designed to aid in making and implement-
ing decisions. These systems facilitate development of computer models
which retlect "real-world" situations, simulations, and answering "what if"
questions. In this regard, they allow the manager to assess the impact of a
decision before implementing it within the company.
3.4.4.3 Local Area Network Approach
A local area network is a collection of computer hardware and software
designed for the specifie purpose of facilitating communications between
computer deviees within an area su ch as a factory . ln a distributed process-
ing environment, the local area network links various deviees together,
vastly increasing the total available computer capabilities.
1. Distributed data processing. One of the key challenges of the factory
of the future will be the design of a networking system. As the trend
toward distributed automation continues with systems such as CAD
and programmable controllers, it will be the role of the local area
network to tie the se potential "islands of automation" together. So-
phisticated users must design the network with enough flexibility to
facilitate future factory modemization efforts.
The local are a network will allow the creation of a more complete
data base. This will be a distributed data base built up through the
various nodes on the network, not imposed through a centralized com-
puter system.
2. Data communications. One of the major problems with designing a
local area network system today is that most commercially availabl e
systems are closed systems where the equipment that provides the data
communications function uses vendor specifie protocols. This closed
system restricts a company's freedom to choose the best possible
equipment to meet specifie needs. Open systems architecture appears
to be emerging in importance. The Manufacturing Automation Proto-
col (MAP) is a popular evolving industry standard for communications
among shop floor deviees .
For these reasons, it is important that the local area networks be de-
signed to maximize ftexibility for future expansion.
3.5 SUMMARY
In this chapter we have described the nature and major components of CIM
within the context of developing an understanding of how technology can
-..._J
f--'
N
il
Il
J,
:i
1
1
. 1
' 1
::
il
;l
1
.,
l'
g (
F
~
'
't':
5<\
4 ~
~
.. }j
~ ~ J
~ ~
~ ~
64 Computer-/nregrated Manufacturing: the New Technologies Chap.J
improve critical performance, which contributes to the competitive posture
of a manufacturing firm.
Simply acquiring technology for the sake of sophistication without
establishing performance measures and targets which support strategie ob-
jectives is both ex pensive and foolhardy. A clear understanding of how
advanced manufacturing technology contributes to better performance in
ter ms of cost, quality, delivery, ftexibility, reliability, and other factors
should precede, not follow, consideration of new technology.
REFERENCES
1. Ayres, Robert U., and Steven Miller, "Robotics, CAM and Industrial Productiv-
ity," National Productivity Review, Winter 1981-1982, pp. 42-60.
2. Curtin, Frank T. , "Planning and Justifying Factory Automation," Production
Engineering, May 1984, pp. 46-51.
3. Groover, Mikell P., Automation, Production Systems and Computer-Aided Man-
ufacturing, Prentice-Hall, Inc., Englewood Cliffs, N.J., 1980.
4. Harrington. Joseph, Understanding the Manufacturing Process, Marcel Dekker,
Inc., New York, 1984.
5. Hegland , Donald E . "Flexible Manufacturing-Y our Balance between Produc-
tivity and Adaptability," Production Engineering, May 1981, pp. 169-221.
6. National Research Council, Committee on the CAD/CAM Interface, Manufactur-
ing Studies Board. Complller Integration of Engineering Design and Production,
National Academy Press , Washington, D.C., 1984.
7. Steinberg, Earle, and James A. Brimson, A Frameworkfor Computer lntegrated
Manufaclllring, Touche Ross, Inc. , Houston, Texas, 1985.
8. Teicholz, Eric, "Computer Integrated Manufacturing," Datamation, March
1984. pp. 169-174.
EXERCISES
3-1. Briefty identify reasons why the benefits presented in Table 3-1 might be
achieved through information integration within a firm.
3-2. Explain sorne of the difficulties one might expect to encounter in using a
computer to achieve computer-integrated manufacturing in the context of Fig-
ure 3-1.
33. Outline sorne important characteristics that distinguish job shop production
from mass production.
34. What are sorne of the key benefits of CAD, and how might one go about
quantifying them?
3-S. Why is group technology particularly useful in the design and management of
flexible manufacturing systems?
"'
Chap. 3 Exercises
65
3-6. Describe briefty the generative method of computer-aided process planning
(CAPP).
37. Draw a block diagram to tie together the various components of factory man-
agement and control systems. Identify the types of information that pass from
one block to another.
3-8. List sorne of the technologies that comprise computer-aided manufacturing.
Summarize sorne of the costs and benefits of each.
3-9. What is a decision support system, and how does it relate to the generic
subject of "information integration"?
3-10. Data communications systems are an integral part of the "factory of the
future ." Research the topic "manufacturing automation protocol" (originally
proposed by General Motors) and describe how this communication protocol
is likely to reduce the costs of data-driven automation in manufacturing and
service industries.
'-J
......
w
7.14
VALUATION DES NOUVELLES TECHNOLOGIES: UNE
APPROCHE MULTICRITRE
1- Introduction:
Dans la vie de l'entreprise, il est rare que les critres d'valuation des solutions alternatives soient
uniques. L'objectif sera souvent multiple: par exemple, les gestionnaires cherchent, par le biais
d'une implantation d'une nouvelle technologie, rduire les cots, amliorer la qualit, rduire
les retards de livraison et maintenir la stabilit d'emploi. Frquemment, ces objectifs sont
contradictoires. En effet, les alternatives qui rduisent les cots peuvent, au contraire, affecter la
qualit ou le temps de livraison. C'est pourquoi des mthodes d'aide la prise de dcision face
des critres multiples ont t dveloppes, particulirement durant les vingts dernires annes.
Ces mthodes visent fournir au dcideur une solution satisfaisante en prenant en compte tous ces
facteurs simultanment.
Il importe de rappeler que dans un environnement mulitcritre, il n'est pas possible de parler
rellement d'optimisation. En effet une solution indiscutablement optimale selon un ensemble de
critres devrait tre la meilleure selon chacun d'entre eux. Un tel cas ne peut tre envisag quand
on sait, tant donn que les diffrents critres sont souvent conflictuels.
Par contre, lorsque 1' on veut comparer plusieurs options entre elles en se basant uniquement sur
des facteurs tangibles (fmanciers), on peut appliquer les mthodes d'valuation conomiques
(V AN, RSI, TRI, etc.) puis choisir la moins coteuse. La principale difficult est d'obtenir de
bons estims car il faut prvoir les cots venir et non pas seulement ceux qui sont en vigueur au
moment de l'tude. Mais l'introduction des facteurs intangibles, difficiles quantifier, vient
compliquer grandement le choix de la meilleure option. L'utilisation d'un modle d'aide la
dcision susceptible de considrer simultanment les facteurs tangibles et intangibles peut remdier
cette lacune. A cette fin, plusieurs modles ont t expriments jusqu' prsent.
7.15
TI-Difficults rencontres par les mthodes traditionnelles:
Dans le processus d'valuation, ces mthodes ne prennent en considration que les aspects
financiers ce qui rend l'tude incomplte. D'une part, le cot de l'investissement risque d'tre
sous-estim, d'autre part, les conomies sont parfois difficiles estimer. En effet, outre les gains
directs de la main-d'oeuvre (gains de fonctionnement), on retrouve presque toujours un effet
d'entranement de plusieurs avantages indirects, tous provoqus par l'implantation d'une
technologie approprie. En plus de collaborer la diminution des temps de cycle de production,
la flexibilit du systme de production augmente, les en-cours sont rduits, le temps de rponse
est diminu. Ces avantages sont particulirement intressants quand on peroit le processus de
fabrication d'une faon dynamique et non pas comme un simple processus de fabrication statique.
A cela, il faudrait ajouter la multiplicit des facteurs, souvent contradictoires, dont on doit tenir
compte simultanment.
Les mthodes d'valuations traditionnelles ne tiennent pas compte de tous ces lments et plus
des aspects stratgiques du projet.
Ds lors qu'il s'agit d'une dcision stratgique, comme c'est souvent le cas des nouvelles
technologies et des systmes intgrs, le recours d'autres mthodes devient ncessaire TI faudrait
dvelopper d'autres critres pour justifier ces technologies.
Huber [1985] souligne qu'en 1984, une enqute a rvl que 91% des compagnies utilisaient la
mthode du dlai de rcupration comme outil principal pour justifier les efforts d'automatisation
de leurs entreprises. Selon lui, le modle de justification des systmes manufacturiers avancs
doivent toujours inclure les bnfices anticips qui ne pouvaient pas tre directement dtermins
tel que l'impact de l'amlioration de la qualit. Ignorer ces bnfices parce qu'ils sont difficiles
valuer serait, selon l'auteur, l'exclusion de toute tentative d'adoption de ces nouvelles
technologies. Meredith et Suresh [1986], de leur ct, constatent que beaucoup de projets mritant
d'tre retenus ont t abandonns parce que les avantages qualitatifs ne pouvaient pas tre inclus
dans la procdure de justification tandis que les conomies ralises sur les cots directs taient
insuffisantes pour rpondre aux obstacles financiers tablis par l'entreprise.
-ces technologies sont plus flexibles que les autres quipements ordinaires. Les avantages de cette
flexibilit ne sont pas facilement capturs par les procdures traditionnelles;
7.16
-leurs synergies c'est--dire leurs effets conjugus.
En d'autres termes, elle suppose que le procd actuel continuera fonctionner dans les mmes
conditions que dans le pass, et sans aucun risque. Certains auteurs sont mme alls jusqu'
qualifier la justification conomique d'obstacle Nl pour l'implantation de ces nouvelles
technologies. En fait, les entreprises ont tendance d'incorporer deux fois le risque d'un projet
comportant des nouvelles technologies:
-par l'utilisation d'un taux preuve lev;
- et l'exclusion de certains bnfices, difficiles mesurer, dans les mthodes des estims
conservateurs.
Remdier cette double pnalit constitue un dfi toute approche de justification de ces
nouvelles technologies.
rn- L'approche multicritre:
ll.l- Df"mition:
(1
)
1


l
amc&a!nspriEs
(2) 1
1

l
tvamdatc&s:tm
al b'dialcls&lls
l
(5) 1 Ofmnrt rra
l
1 IDll

Exl3rTpe: O'dx dlJl
l'l'BIIEU
rrert
Exenpe: Ca1, Fledb
dlai, ErHx:us, etc.
&errpe: Q>Fiexi)
>[Wai etc.
7.17
Les diffrentes mthodes dveloppes sont souvent classes sous l'appellation "aide la dcision
multicritre" (Vincke 1989). Celle-ci dsigne gnralement un ensemble de mthodes permettant
d'agrger plusieurs critres afin de slectionner une ou plusieurs "actions". Elle vise fournir au
dcideur des outils qui lui permetteront de progresser dans la rsolution d'un problme de dcision
o plusieurs objectifs, souvent contradictoires, doivent tre pris en compte.
La premire constatation, qui doit tre faite, lorsqu'on aborde un tel problme est qu'il n'existe
pas en gnral une dcision qui soit la meilleure simultanment pour tous les points de vue. Le
7.18
mot "optimisation" n'a donc plus de sens dans un tel contexte
Contrairement aux techniques de la recherche oprationnelle, les mthodes multicritres ne
fournissent pas de solutions "objectivement les meilleures". La divergence des objectifs ncessite
la recherche d'une solution de meilleurs compromis possible.
Pour appliquer ces mthodes, on doit ncessairement suivre les tapes suivantes:
- dresser la liste des critres prendre en considration;
- dresser la liste des actions potentielles, ou envisageables;
- juger chacune des solutions aux yeux de chacun des critres;
- enfin, agrger ces jugements pour choisir la solution la plus satisfaisante.
A l'exception de la dernire tape, les trois premires sont communes toutes les mthodes
multicritres.
Le lecteur pourra trouver dans les livres de Schrlig [1985] et de Vincke [1989] un panorama des
diffrentes approches proposes jusqu'ici en analyse multicritre, englobant aussi bien les
mthodes d'inspiration amricaine que celles de "l'cole franaise". Pour ce qui est des modes
d'emplois dtaills et les justifications mathmatiques des diffrentes mthodes dcrites, les
ouvrages de Roy [1985], Zeleny [1982] et Saaty [1980] rpondent parfaitement ce besoin.
Un boil systme d'aide la dcision doit tre:
- simple comprendre. Seuls les phnomnes importants doivent tre inclus dans le modle.
-Robuste:
- Facile contrler
- Adaptatif: Le modle peut tre mis jours en cas de prsence de nouvelles donnes.
- Exhaustif quant aux points importants
- Ais pour ce qui est de la communication.
- pouvoir traiter diffrentes solutions alternatives;
- pouvoir considrer de nombreux critres et objectifs souvent contradictoires;
-permettre au dcideur de structurer (de modliser) le problme lui-mme;
7.19
- permettre au dcideur d'incorporer des donnes objectives mais aussi des donnes subjectives
bases sur son exprience.
m.2- Les mthodes de pondration (weiwtin2 methods):
La mthode la plus simple, et sans doute la plus utilise, est la moyenne pondre qui peut prendre
la forme suivante:
n
G( a ) = L . [g . (a)]
j =1 J J
Cette mthode consiste en les tapes suivantes:
1- dresser la liste des critres ou caractristiques des options;
2- affecter chacun un poids qui mesure l'importance que lui attache le dcideur;
3- attribuer un ordre numrique qui indique quel degr l'quipement prsente la caractristique
recherche;
4- dterminer la cote poids-rang de chaque quipement et choisir le mieux cot.
C'est la forme la plus ancienne, et donc qui a vu beaucoup d'applications dans divers domaines,
en particulier le domaine manufacturier. Par exemple, Nelson [1986] suggre un "modle
scoring", une variante importante de cette catgorie de modles, pour dterminer les sections
d'une usine qui devraient tre modernises. Les modles "scoring" ont t galement appliqus
par Sullivan [1984] pour effectuer un choix entre un systme manuel et un Robot en incluant les
critres non-montaires et par Udoka et al. [1990] pour valuer les CIMS. Huang et
Ghandfouroush [1984] procdent en cinq tapes dans leur application du modle "scoring" pour
l'valuation et la slection de robots [p.44-47]:
1- dvelopper une liste de robots disponibles sur le march;
2- dterminer les facteurs critiques ventuels susceptibles d'carter certains robots;
3- dfinir les mesures des facteurs objectifs: OFMi;
7.20
4- dfmir les mesures des facteurs subjectifs: SFM;;
On assigne chacun des facteurs subjectifs un score sur une chelle de 0 10.
5- Dterminer)a mesure de performance en considrant les 2 mesures objectives et subjectives
ainsi calcules aux tapes 3 et 4:
PM
1
= (a).OFM
1
+ (1-a).SFM;.
Le coefficient (a) dpend de la prfrence du manager accorde aux mesures subjectives et
objectives. Nous nous limiterons cette description sommaire, pour plus de dtails, nous
recommandons de consulter l'article en question. (voir l'exemple trat la fm de ce document).
Il importe de rappeler qu' ce niveau de l'analyse, on suppose qu'on a dj effectu un premier
trie et qu'on est en prsence d'actions efficaces, dans le sens o elles ne sont domines par aucune
autre action.
Par dfinition, une action domine une autre si elle fait un score au moins aussi bon que la seconde
sur tous les critres, et meilleur sur un critre au moins.
TIL 3- Pondration des critres:
Dans la plupart des mthodes multicritres, l'importance relative des critres est reprsente par
des poids numriques. Ceci soulve le problme de dtermination de ces poids et la traduction en
termes quantitatifs la notion qualitative d'importance des critres. Ces paramtres ne possdent
pas de signification claire et ont gnralement une grande influence sur les rsultats fournis par
les mthodes multicritres. qui les utilisent.
De faon gnrale, les poids permettent de traduire l'importance qu'accorde le dcideur aux
diffrents critres. L'importance de ces poids varie selon la mthode utilise.
IV- La mthode AHP (Analytic Hierarcby Process):
AHP se distingue des autres mthodes d'aide la dcision multicritre dans sa faon de dterminer
des poids. En effet, elle procde par comparaisons binaires de chaque niveau de la hirarchie par
7.21
rapport aux lments du niveau suprieur. Les comparaisons se font sur une chelle dfinie ( 1
9). Par ailleurs, elle est capable d'identifier et de prendre en considration les incohrences des
dcideurs. Un logiciel "expert choice" a t conu cet effet. Nous y reviendrons plus en dtails
un peu plus loin.
AHP, dveloppe en 1971 par Saaty [1980], est l'une des mthodes qui a reu une attention
particulire. Elle a t applique dans plusieurs domaines en particulier dans le domaine
manufacturier pour la justification et le choix d'une nouvelle technologie avance (Arbel et
Seidman 1984, Varney et al. 1985, Sullivan 1986, Wabalickis 1988, Kleindorfer et Partovi 1990,
Frazelle 1985).
Les deux premiers articles dveloppent des hirarchies spares pour les bnfices et les cots,
dterminent les vecteurs de priorit pour chaque hirarchie avant de les combiner ensemble, et
enfin calculent le ratio bnfice-cot. Le classement se fera par ordre dcroissant du ratio obtenu.
Kleindorfer & Partovi [1990] ont suggr l'utilisation de la mthode AHP pour choisir une
technologie [p.215], alors que Cambron et al. [1991] s'en servent pour choisir le meilleur
amnagement possible au niveau d'une usine.
Frazelle [1985], pour sa part, propose deux mthodes multicritres (technique de pondration, et
mthode AHP) pour choisirun systme de manutention sur la base de cinq critres globaux
retenus:
. RSI (Retour Sur Investissement)
. flexibilit
. scurit
. compatibilit
. maintenabilit
Ces deux mthodes d'valuation ont t juges trs satisfaisantes par l'auteur, en particulier AHP
qui prend en considration les incohrences dans les jugements tous les niveaux de la hirarchie.
Nous prsenterons l'exemple d'application utilis par Frazelle un peu plus loin.
7.22
Le succs remarquable ralis par cette approche est d au fait qu'elle prend en considration aussi
bien les aspects quantitatifs que qualitatifs des critres tout en assurant une certaine cohrence dans
lesjugements. En effet, AHP reprsente une faon pratique de traiter quantitativement les aspects
qualitatifs des critres de dcision tels que la flexibilit, la qualit, la fiabilit, la scurit, etc.
De plus, AHP ne force pas le dcideur tre parfaitement cohrent, mais elle offre l'avantage
dterminant de pouvoir fournir une mesure de cohrence de ses jugements.
Dans AHP, la quantit de donnes requise par l'analyste est relativement limite. L'input est
principalement bas sur l'exprience et le jugement de celui qui a conu le systme. Les articles
de Zahedi (1986), Shim (1989) et Vargas (1990) prsentent une excellente revue de la littrature
ce sujet.
La figure suivante montre un exemple de hirarchie utilise dans AHP.
1 1 1 1 1
~ ~ ~ ~ g
~ ~ f ~ ~
......
- ';11\
Dcomposition hirarchique (objectif, macro-critres,
critres, solutions).
7.23
Exemple d'application:
1
AGJS
usine Entire
Il
Olaicts
lvatars
~
~ B & C .
Systrre de rrm.iertiCll
sc:tisfaisa1t
Ill
Olaicts
lvatars
usine altire
Hirarchie utilise par Frazelle 1985 p.44.
Analyse de sensibilit:
IV
l'vtn:rail
~ A
AGJS
~ B & C .
7.24
Quelle sera la raction des ratios des changements dans les
jugements du dcideur.
Analyse de sensibilit:
0.38
0.32
0.30
1
1 0.402
i,-
1
0
0.2
0.4 0.6
Valeur dea
Discussion:
Option I > Option II pour tout a;
Option I > Option III pour a > 0.4;
7.25
0.377
0.8
1.0
Point rrort pour
les opt 1 et Ill.
0.297
Le point more entre opt.I et opt.III comme suit:
a (0.377) + (l-a)(0.320) = a (0.297) + (l-a)(0.378) = = = >
a = 0.42
7.26
Bibliographie:
- Arbel A. and Seidmann A. [ 1984),
11
Performance evaluation of
flexible manufacturing systems", IEEE 1984, pp.118-129.
- Cambron, K.E. and Evans, G.W. [1991], "Layout Design using
Analytic Hierarchy Process," Computers ind. engng, Vol.20, N . 2,
pp.211-229.
-Canada J.R. and Sullivan W.G. [1989J,"Economic and multiattribute
evaluation of advanced manufacturing systems", Prentice Hall Inc.
1989.
- Frazelle Ed. [1985], "Suggested techniques enable multi-criteria
evaluation of material handling alternatives", Industrial
Engineering Feb. 1985, pp.42-48.
- Huang P. Y. and Ghandfouroush P. [ 1984), "Procedures given for
evaluating, selecting Robots", Industrial Engineering, April 1984,
pp.44-48.
Huber R.F. [1985], "Justification-Barrier to competitive
manufacturing", Production, Sep. 1985, pp.46-51.
Kleindorfer, P.R. and Partovi, F.Y. [1990],"Integrating
manufacturing strategy and technology choice", European Journal of
Operational Research, 47, pp.214-224.
- Madu, C.R. & Georantzas, N.C. [1991), "Strategie thrustnof
manufacturing decisions: a conceptual framwork", IIE Transactions,
June 1991, pp. 138-147.
-Meredith, J.R. and Suresh N. [1986], "Justification techniques
for advanced manufacturing technologies", Int. J. Prod. Res.,
vol.24, N5, pp.1043-1057 (1986).
-Nelson C.A. [1986], "A scoring model for flexible manufacturing
systems project selection", European Journal of Operations
Research, vol.24/N3 1986 ,pp.346-359.
- Partovi, F. Y. and Burton, J. [ 1992], "An Anlytical Hierarchiy
Approach to Facility Layout
11
, Computers Ind. Engng Vol. n
0
4,
pp.447-457,1992.
-Roy, B. [1985], "Mthodologie multicritre d'aide la dcision",
Economica, Paris.
- Saaty T.L. [1980], "The Analytic Hierarchy Process" McGraw-Hill,
New-York, 1980.
- Saaty T.L. [1982],
Inc., 1982.
7.27
"Decision Making for Leaders", Wadsworth,
- SchSrlig, A. (1985], "dcider sur plusieurs critres", Presses
Polytechniques Romandes.
- Shim, Jung P. ( 1989], "Bibliographical Research on Analytic
Hierarchy Process (AHP) ", Socio-Econ. Plann. Sei. vol. 23, N3,
pp.161-167, 1989.
- Tversky A. [1972], "Elimination by aspects: a theory of choice".
Psychological review, 79, 4 (1972), pp.281-199 .
- Udoka S.J., Nazermetz J.W. (1990], "Development of a methodology
for evaluating Computer Integrated Manufacturing (CIM)
implementation performance", Computer Industrial Engineering N
5
1-4,
pp.145-149 (1990).
- Vargas, L.G. (1990], "An overview of the Analytic Hierarchy
Process and its applications", E.J.O.R. 48, pp.2-8, (1990)
Varney M. s. , Sul li van W. G, and Cochran J. K. [1985],
"Justification of flexible manufacturing systems with the
Analytical Hierarchy Process", Proceedings of 1985 IIE Spring
Conference, IEE, Norcross,GA, 1985, p.181-190.
- Vincke, Ph. [1989], "L'aide multicritre la dcision", ditions
de l'U.L.B., Ellipses, 179 pages.
- Wabalickis R.N. [1988], "Justification of FMS with the Analytic
Hierarchy Process", Journal of Manufacturing Systems, vol.7, N3,
1988, pp.175-182.
- Zahedi F. [ 1986], "The Analytical Hierarchy Process - A survey of
the method and its applications", Interfaces, 1986, 16(4), pp.96-
108.
- Zeleny, M. [1982], "Multiple criteria decision making", McGraw-
hill, New York.
210
Line Item
( t 9) Depreciation
(20) Net. before-tax savings
121) Net after-tax savings
(22l Net after-tax cash savings
(23) Net after-tax cash flow
~ ~ ~ ~ Discount factor at 20%
125) PW at JO%
(26) Discount factor at 20%
127) PW at 20%
(2RI Discount factor at 40%
(291 PW at 40%
ln upper riJ!hl cnr11er:
Net PW
Payback period
Economie Analysis Methods: After Taxes Chap. 7
Table 7-A-1 continued
Description
Yearly depreciation calculated using ACRS,
straight line, or whatever method; line (4) multi-
plied by the yearly percentage [Note: If appli-
cable, should reduce line (4) by part or ali of
line (5) .]
Total savings minus depreciation [line (18) minus
line (19)]
Line (20) multiplied by 1 - tax rate
Line (21) plus line (19)
Line (22) minus line (Il) plus line (10)
For calculation of present worth
Line (23) multiplied by line (24)
For calculation of present worth
Line (23) multiplied by line (26)
For calculation of present worth
Line (23) multiplied by line (28)
Total net PW laken from the "total" column on
right-hand side for the appropriate interest
percentage [Note: If want net PW for sorne
interest rate other than 10%, 20%(..Dt 40%, can
till in appropriate discount factors (PIF), in line
like (24), then calculate totals for line like (25).)
Use li ne (23) to fi nd number of years b.efore cu-
mulative savings will just equal the investment;
may want to interpolate to gel answer to near-
est 0. 1 year
The form normally can also be used to facilitate calculations of the
approximate discounted cash flow rate of retum (also "internai rate of re-
turn" ) for the project. Ali that needs to be done is to interpolate between the
two closes! rates which have positive and negative total net PWs, respec-
tively. In this case the reis no negative net PW for the three rates shown, bu.t
the net PW would become 0 at about 60%.
The form can also be used to calculate the payback period by adding
the cumulative net cash ftows (li ne 23) over lime. The payback period is the
number of years until the cumulative net cash How switches from negative to
positive.
multiattribute
decision analysis:
introduction
and basic techniques
chapter
8.1 INTRODUCTION
Much has been written regarding the inadequacies of traditional economi e
analyses in evaluating the prospective results (benefit s and. sometimes. th
problems and costs) of advanced manufacturing systems. Multi a!! ribut
(multiobjective, multicriterion, multifactor, etc.)* decision techniques secm
to provide an easily understood, yet comprehensive , set of quantit ative/
qualitative approaches to justify advanced manufacturing systems.
Classic advice on the subject of decisions involving multipl e factors
was rendered in a letter from Benjamin Franklin to a friend, Joseph Pri estl y.
in 1772.
The word "attribute" is used to describe what is important in a decision problcrn and
is often used interchangeably with "objective" and "criterion. " A liner distinction can he
made as follows: an "objective" represents direction of improvement or preference for one
or more attributes, while "criterion" is a standard or rule that guides decision mak ing.
211
"-J
N
co
d
if
; : ~
~ ....
:t
~ t
v
l\
i .,,
1
~
Multiallribute Decision Analysis
Chap. 8
Dear Sir,
ln the affair of so mu ch importance to you, wherein you ask my ad vice, 1
cannot. for want of sufficient premises, advise you what to determine, but if
you please 1 will tell y ou how. Wh en tho se difficult cases occur, they are
diflicult. chiefly bec a use wh ile we have them und er consideration, ali the rea-
sons pro and con are not present to the mind at the same time; but sometimes
one set present themselves, and at other times another, the first being out of
sight. Hence the various purposes or information that alternatively prevail,
and the uncertainty !hat perplexes us. To get over this, my way is to divide
half a sheet of paper by aline into two columns; writing over the one Pro, and
over the other Con. Then, during three or four days consideration, 1 put down
und er the different heads short hints of the different motives, thal at -different
times occur to me. for or against the measure. When 1 have thus got them ali
togcthcr in one view. 1 endeavor to cstimate their respective weights; and
where 1 find two, one on each side, thal seem equal, 1 strike them both out. lfl
find a reason pro equal to sorne two reasons con, 1 strike out the three. If 1
judge sorne two reasons con, equal to three reasons pro, 1 strike out the five;
and thus proceeding 1 find at length where the balance lies; and if, after a day or
Iwo of further consideration, nothing new thal is of importance occurs on
either side, 1 come to a determination accordingly. And, though the weight of
the reasons cannot be taken with the precision of algebraic quntities, yet
when each is thus considered, separately and comparatively, and the whole
lies before me, 1 think 1 canjudge better, and am Jess liable to make a rash step,
and in fact 1 have found great advantage from this ki nd of equation, and what
might be called moral or prudential a]gebra.
Wishing sincerely that you may determine for the best, 1 am ever, my
dear friend. yours most affectionately.
B. Franklin
Although the term "moral or prudential algebra" does not seem to
have tlourished, the spirit of Franklin's "rational" or "systematic" ap-
proach continues. lndeed, in the last two decades severa! very powerful
techniques or methodologies have emerged. The methodologies to be de-
scribed herein ditfer considerably regarding their assumptions and limita-
tions . Awareness of su ch limitations will make the user more effective in the
selection and proper use of these methodologies as decision guides.
8.2 SELECTION OF ATTRIBUTES
The selection of what attributes should be considered in the comparison of
alternatives is critically important to the eventual outcome (alternative se-
lected). lndeed, it is sometimes observed that the mere articulation of what
attributes are important in the context of particular alternatives under con-
sideration can lead one to a rational (presumably at !east somewhat close to
Sec. 8.2 Selection of Allributes 213
the best) choice without the forma! application of sorne quantitative or semi-
quantitative methodology.
The most important usual theoretical restriction in the na ming of auri-
butes is that they be independent of one another. In brief, but somewhat
oversimplified words, any two attributes are said to be independent if the
preference order and the trade-offs for different levels of those attributes do
not depend on the levels at which ali other attributes occur.
Independence of attributes is difficult to achieve (or to know that one
has achieved) literally. Rather, normally one specifies important attributes
to be as independent as practicable (intuitively), and theo stands ready to
discount the results of the analysis somewhat if he or she feels thal Jack of
independence affects the results significantly.
In selecting attributes, one might weil identify scores of attributes that
may be important. As a practical matter, one normally would want to keep
the number of attributes finally selected within manageable limits by drop-
ping or not otherwise considering those having little or no chance of signifi -
cantly affecting the choice among alternatives. There is no ideal number of
attributes to consider. But it should be recognized thal too few means that
sorne attribute(s) which are important have been ignored, while too many
means that one may become bogged down with futile details . This results in
Joss of val ua ble ti me and energy, if not possible opportunities forgone. due
to diversion of attention away from the discovery of better alternatives or
designs for the problem at hand.
One way to reduce the number of attributes is to identify any attribute
fo[.which each alternative in a particular decision-making situation is equally
good (or bad) with respect to that attribute. Since only differences among
alternatives are important in their comparison, such attrihutes contrihute
nothing to the analysis and, in general, can be eliminatcd .
Example 8-1
A firm may have decided upon, say, 15 significant attributes for capital inves1men1
comparisons . Further, it may have narrowed thal down to, say, 10 significanl attri
butes for replacement comparisons ("keep old" versus "bu y new" ) in general. For
a particular situation, involving possible replacement of a manufacturing system, il
may be determined quickly that, say, 4 out of the 10 have the sa me outcomes for ali
alternatives. Th us the 4 can be eliminated, leaving only 6 attributes which need to be
considered for th at particufar situation.
Rather than starting with a large number of attributes and theo narrow-
ing them down (as discussed above), the opposite might be more workablc in
particular situations. Thal is, one might start with one or a few attributes
(e.g. , only economie etfects and strategie impact), do an initial analysis for
familiarization and exploration purposes, and theo add more attributes and
analyze further as believed to be justified by the nature of the decision
'-.)
N
I.D
1 ij ~
si

;
.. ~
J
~ ' i
: ~ ~
~
~
~
li
'
t
l
214 Multiattribute Decision Analysis Chap. 8
situation. Indeed, attributes can be added and further studies undertaken in
successive increments until one is satisfied thal the lime and expense re-
quired for further study are not sufficiently justified by the potential added
information which might affect the choice(s) among alternatives.
8.3 MISCELLANEOUS SEOUENTIAL
ELIMINATION METHODS
Sequential elimination methods were first categorized by MacCrimmon [1]
and are decision rules (sometimes arbitrary) whereby an individual might be
able to eliminate one or more alternatives to narrow the choice and perhaps
even be led to a decision.
These methods are applicable when one can specify values: (outcomes)
for ali attributes and alternatives. Those values should be scalar (measurc
able) or at !east ordinal (rank orderable). The methods do not consider
weighting, if any, of attributes. A Iso, they are "noncompensatory," which
means that they do not recognize possible trade-offs among attributes across
alternatives. Thus the method in Ben Franklin's letter (above) violates the
mechanics of sequential elimination methods.
Table 8-1 port rays an example showing. outcomes (values) for four
attributes and four alternatives. Note the two right-hand columns state the
"ideal" and the "standard (minimum acceptable)" values for each attribute.
8.3.1 Alternative versus Standard: Comparison
across Attributes
This method takes two forms as described below:
1. Disjunctive constraint. An alternative will be retained (not eliminated)
if it meets the standard for at /east one attribute.
Allribule
A. Quality
B. Ftexibility
C. Serviceabili 1 y
D. Cost savings
Table 8- 1
EXAMPLE MULTIAITRIBUTE PROBLEM
FOR ILLUSTRATION OF "SEQUENTIAL ELIMI NATION METHODS"
Eslimaled Oulcome for Allernative: Standard
(Minimum
2 3 4 Ideal Acceptable)
75 90 80 60 too 70
Very good Good Poor Excellent Excellent Fair
50 38 35 30 50 30
8 6 5 6 10 7
Higher numbers are belier.
Sec. 8.3
Misce/laneous Sequential Elimination Methods
215
2. Conjunctive constraint. An alternative will be retained only if it meets
the standard for al/ attributes.
Example 82
For the problem in Table 8-1 .(and using the standards in the right-hand column), set
up a table thal indicates which alternatives do not meet the standard for which
attributes. Then condude which alternatives mee! the following constraints.
(a) Disjunctive constraint.
(b) Conjunctive constraint.
Solution
X for Allernalives
Not Meeting Standard:
Attribute
A
B
c
D
x
x x
4
x
x
(a) Ali alternatives meet the standard for at !east one attribute.
(b) Only alternative 1 meets the standard for ali attributes.
8.3.2 Alternative versus Alternative:
Comparison across Attributes
The method is normally called a "dominance" check. If one alterna-
tive is better tl:lan or equal to sorne other alternative with respect to ali
. attributes , and better for at !east one attribute, the other alternative is said to
be dominated and.can be eliminated.
Example 83
For the problem in Table 8-1, comparison of ali pairs of alternatives across ali
attributes will reveal thal alternative 2 dominates alternative 3. That is, alternati ve
2 :>alternative 3 because, in the given attribute order, 90 > 80; good > poor ; 38 > 35;
and 6 > 5. Further inspection by the interested reader will lind no other alternative
that is dominated.
8.3.3 Alternative versus Alternative:
Comparison across Alternatives
This method is of two types:
1. Lexi.c.agraphy. This involves first ranking attributes by importance.
For the most important attribute, choose the alternative, if any, which
'-J
w
0
.;.
; t.
216 Multinltribute Decision Ana/ysis Chap. 8
is best. If there is a tie between two or more alternatives, one should
go to the second most important attribute and choose which of those
" remaining" alternatives, if any, is best. This process continues in
this way until a single alternative emerges , or until ali attributes have
been examined.
2. Elimination.by Aspects. This is like lexicographybecause it examines
one attribute at a time, making comparisons among alternatives. How-
ever. it eliminates alternatives thal do not satisfy the standard (mini-
mum acceptable for thal attribute) and continues un til ali 'alternatives
except one have been eliminated, or until ali attributes have been ex-
amined .
Example 8-4
For the problem in Table 8-1 , assume thal attribute importance rankings are A > B >
C > D* and determine what choice or alternatives remain after applying (a) lexicog-
raphy and (b) elimination by aspects, using the "standards" given on the right-hand
side of the table.
Solution
(a) For attribute A, alternative 2 is best and would be the choice. (Note : Ir, for
instance. alternative 2 and, say, alternative 3 had been tied for best with
respect to attribute A, then the choice would have been based on attribute B.
Alternative 2 is better than alternative 3 with respect to attribute B and thus
would have heen the choice.)
(b) Refer to Table 8-1 (or the table in the solution for Example 8-2).
Allribute Alternatives Eliminated Alternatives Left
A 4 1' 2, 3
B 3 l , 2
c None 1, 2
D 2 1
Thus alternative 1 is the sole survivor.
Any of the methods above can be tempered/altered with good judg-
ment rcgarding the decision-making circumstances. In most cases it makes
sense to allow what might be called "bands of imperfect discrimination" so
thal one alternative is not judged better than another just because it has a
slightly higher value for one attribute wh en using "lexicography," or the
other alternative just barely feil below one standard when using "elimination
by aspects ."
A > B mea ns thal A is preferred to B, etc.
Sec. 8.4
Graphical Techniques
Ll1
8.4 GRAPHICAL TECHNIQUES
Graphical techniques are very powerful because they allow one to describe
the nature of a multiattributed decision problem so thal decision makers can
readily understand and absorb large amounts of information. They often do
not include specifie "weighting" of attributes. which may be advantageous
because attribute weights are often nebulous and/or differ greatly among
various decision makers.
8.4. 1 Alternatives-Attributes Score Ca rd
A scorecard is a matrix of alternatives versus attributes together with
numbers and/or other symbols to represent the outcome expected for each
alternative with respect to each attribute. Table 8-2 is an example in which
qualitative and quantitative results are shown. Ease of interpretation of the
scorecard results to facilitate decision making can be obtained by such de-
vices as symbols and/or colors for "best" and "worst" alternatives for each
attribute.
8.4.2 Shaded Circles to Portray
Scorecard-Type Results
Figure 8-1 shows the use of shaded circles to portray visually the
relative evaluations of alternatives with respect to the five attributes as given
in Table 8-2. As shown in the key, the evaluations for each attribute except
net present worth were categorized five ways from "exceptional" (full shad-
ing) down to "poor" (no shading) . Net present worth is shown to be "ex-
ceptional" (with full shading) for $500M, with proportionately less shading
down to $0M. Wh ile the use of shaded ci rel es does cause loss of the specifie
language and quantitative information ~ i v e n in the original scoreboard in
Fig. 8-1, it is ea sy to scan for relative comparisons.
In examining the scorecard information in Table 8-2 and/or Fig. 8-1,
one might be led to conclude thal alternative P-4 (and perhaps P-3) are
definitely not as desirable as alternatives P-1 and P-2. Perhaps a final deci -
sion can be made u..ing this method, but it should be recognized thal any
relative weights/imi5ortances of the attributes have not been assigned for-
mally. We will consider only alternatives P-1 and P-2 in demonstrating sorne '
other multiattribute methods.
8.4.3 Graphs of Evaluation Ratings for
Attribute Outcomes
Figure 8-2 shows sample graphs for depicting what are commonly
called "evaluation ratings" or "seo ring functions" (arbitrarily on a scale of 0
'-.!
w
f-4
218
NV>
' "' 00 l-
u :>
- .,
.0 -
"' " .... 1:
<
"' >
:
c

V>
"' >


!::;
<
" :>
12
"

c
" <
u
"'
"
8
en
"'
.J
..
:.;
<
)(
w
"
;;;
E

<
d.
"'
rl.
N
rl.
rl.
u
:;
.0
'5
<
r-1
'""'
: e:
1 u
'>'
:
i

,-
___,
+
r ... ,


:N:





lA
+
, .........,
: .... .
: .E
8
t::J r-
,--...
.... ' ....
.c 1 ..c 1


..2
-1
_,
.... -'
...
..c::


u
>
<



;;
Ol
3

en


<
..c::
t:
0
:l:
" :::

o.

z
ai

:
"'
" .!.!

" en
Li
t:
g
u
::l'
c
"
0
" ' tl>
"

" " E

:;j

0
'


<

1 ""'
1 1:!1
1 Ut
'>
i
1 o

l .... .J

e
u
>
<
'
..><
]
"'
"'
"
" :;;!
iii
ILi
0
.;
- "' "' .0
'5 '
- "'
"' ...
0 .g
"" 0
0 >
> -

E "
u -
- Ol
v;
"' ...
" 0
CQ
on
....
Sec. 8.4 Graphical Techniques
Alternati""
----
A. CIMS tactical ai ms
B. Net present worth
C. Serviceability
O. Management/engineering
effort
E. Rlskiness, Jack of
Key: (a) For net present worth:
P- 1

--
-
--
'it
P-2
'1
a
'it


e For $500M, with shaded area decrening proportionately to
0 For$0M
P- 3



0

P-4

"



219
(b) For ali other ortributes:
e ExceptioN!
'it Abo"" average

Balowaverage
0 Poor
Figure 81. Graphical portrayal (using shaded circles) for 4 alternatives and
5 auributes. j
to 10) correspond1tg to possible outcomes (from worst to best, respectively)
for each of four attributes. The evaluation ratings could be "utility" values
(usually on a scale of 0 to 1), as shawn in Chapt er 9. Although such graphs
are not recommended as sufficient by themselves for comparing alternatives,
they do help by portraying how ratings or scoring values vary with outcome
changes for whatever attributes to facilitate the use of other decision-making
techniques.
8.4.4 Polar Graphs
Figure 8-3 is an example polar graph showing evaluation ratings for
alternatives P-1 and P-2. Note that four of the rays drawn from the center
-.....]
w
N
220
10
"'
8
~
c 6
0
~
:>
~ 4
w
0
8
C1.
10
8
8'
- ~
c 6
0
~
:>
;;;
4
>
w
2
0
.J:
M
6
5
0
C1.
t
~
'!;
0
-.;
al
CIMS tactical ai ms
{how weil met)
8.
i ~
<1:
..
~
.8
<1:
Serviceability
{expected repair ti me,
.J: E E
M N
~
' '
!::! ~
8, '
:::
!!
"
~
8,
;;
!!
'!;
" ~ 0 >
<1: -.;
"
al >
.8
<t
~
.2
.
8
)(
w
.c
-
~
n;
c
.g
a.
1l
x
w
Multiallribute Decision Ana/ysis
8
"'
- ~
~ 6
.g
..
:>
~ 4
w
2
10
8
. ~
~
c 6
0
- ~
:>
~ 4
w
0
~
61
5
0
c..
8,
~
;;
3:
0
..
al
Riskiness, lack of
f
~
<1:
f
!
~
.8
<1:
Management/engineering
effort (hours required)
~ ~

~
1
8
8
::!
co
-
~
11.
~
!!
~
" ~ ;;
> <1:
..
~
'!;
.8 0
~
<1:
Figure 82. Sample graphs relating attribute outcomcs to evaluation ratings.
Chap. 8
1
o.
8
)(
w
8
...
~
;;;
c
.g
a.
1l
)(
w
Sec. 8.4 Graphical Tech11iques
CIMS tactical ai ms
10
10
Management-
engineering
effort
" Ideal " or "be$t" for
each attribute
221
--... ~ !) --- i Net P. W.
$500M
Figure 8-3. Pof graph for 2 alternatives and S attributes .
correspond to evaluation ratings on a scale of 0 to 10 (such as might be
obtained from Fig. 8-2) and the rightmost ray corresponds to the attribute
"net PW" (on a scale of $0 to $500M). Any attributes thal can be measured
or given an evaluation rating can be included if the maximum and minimum
for each are defined.
Note that after the appropriate evaluations or other measures are plot-
led (using linear scales) on each ray, those points are connected to form
polyhdrons for each alternative. Thi s aids one in judging differences be-
tween alternatives. However, one should not be misled into thinking that
the overall desirability of each alternative is in proportion to the areas of
their respective polyhedrons . This technique as weil as the other graphical
techniques up to this point have not considered the relative weights or
importances of the attributes, which is the subject of the next section.
'-J
w
w
222 Multiallribute Decision Analysis Chap. 8
8.5 RANKING OF ATIRIBUTES (OR ALTERNATIVES)
8efore attribut es are weighted (or alternatives are evaluated) it is often
desired to rank them in order of decreasing preference. This might be done
by presenting the decision maker with a list of factors (usually
they could be alternatives) and asking him or her to rank them in order of
preference. There is, however, a procedure that may make this task easier
and provide a check on the internai consistency of the value judgments
obtained. This is ca lied the method of paired comparisons and is illustrated
by example below.
Consider the five factors (attributes) in Table 8-2, which we now want
to rank order. They are designated as follows:
A. CIMS tactical aims
B. Net present worth
C. Serviceability
D. Management/engineering effort
E. Riskiness, lack of
The method of paired comparisons suggests that the factors be submit-
ted to the decision maker two at a ti me for preference judgments. In gen-
eral. ifthere areN factors. N(N- 1)/2 pairs must bejudged. 'Assume that
the results of this process are indicated by the following list of preference
statements (as normal, the symbol > means "is preferred to" and = means
is "equally preferred to." etc.).
1. A < 8 6. 8 > D
2. A > C 7. B > E
3. A > D 8. C > D
4. A > E 9. C = E
5. 8 > c 10. D < E
A good way to depict the pairwise comparisons above and then deter-
mine rankings is shown by the matrix in Table 8-3. In that matrix, "P" is
shown for each pair in which the row factor is preferred to the column
factor. Note thal the diagonal of the matrix is empty (since a given factor
cannot be preferred to itself). A good way to make sure thal ali pairs of
factors have been considered is to recognize thal there should be a P either
above or below the diagonal for ali pairs, or in case of equal preferences, an
" = " is shown both above and below the diagonal. Note thal on..the right-
hand side of the mat rix is shown the number of times the factor in each row
is preferred . Thus, for this example, it is found that the rank order is 8 >
A > C = E > D.
Sec. 8.6
Weighting of Attributes
A
A -
B
p
c
D
E
Table 8-3
I LLUSTRATION OF PREFERENCE CoMPARISONS'
Number
or Times
8 c D E Prererred
p p p 3
-
p p p 4
-
p ; Il
-
0
; p
-
Il
"P" irrow ractor is prererred to column ractor or"; .. ir
they are equally prererred.
223
The foregoing scheme of deducing rankings assumes rransiriuiry of
preferences. Th at is, if X> Y and Y > Z, th en X must be> Z. If the number
of times a gi ven factor is preferred to another is equal fort wo or more factors
(except in the case of ties), there is evidence of lack of consistency, which
suggests the need for questioning preference judgments for the factors af-
fected.
8.6 WEIGHTING OF ATIRIBUTES*
Weighting of attributes involves quantifying the relative importance of
each. Thefollowing metho ofweighting attributes (again called " factors")
is based on two assumptions:
1. It must be possible for the.Jecision maker to consider and judge the
relative weight of any combination of factors. That is, it must be possi-
ble to cons id er not only the weight of F 1 , but also the weight of both F
1
and F2.
2. W_ights are assumed to be additive. That is, given the weight of F
1
and
the weight of F
2
, the weight of both F 1 and F1 is the sum of their
individual weights.
The decision maker could proceed in weighting the factors as follows :
1. Rank the factors according to decreasing preference or importance by
any method, such as ordinal scaling illustrated above. Once this is
This section is adapted rrom William T. Morris, Tire Analysis of Management Deci-
sions, Richard D. Irwin, lnc. , Homewood, Ill . , !964, pp. 417-424. with permission or the
au thor.
-....J
w
+::>-
L24
Multiauribute Decision Analysis
Chap. 8
done, ir mighr be desirable to assign or reassign subscripts to indicate
the rankings. Thus F, is ranked first, F
2
is ranked second, and so on.
. 1
2. Let the weight of F1 equal lOO [i.e., W(F
1
) = 100] and weight the(other
factors so asto retlectjudgment oftheir weight relative to the weight of
F
1
Th us the weights of F2 F N will range from a possible .high
value of lOO to a possible low value of O. Ail further steps (3 through 6)
serve ro retine these initial weightingjudgments and to ensure they are
internally consistent.
3. Compare F
1
with the combination of F
2
+ F
3
(the plus sign here means
" and"). If F1 is considered to have Jess weight than F
2
+ F
3
, the
weighrs assigned in step 2 must satisfy the relation W(F
1
) < W(F
2
) +
W( FJ). If !he assigned weights do not satisfy this relation, W(F
1
)
should be adjusred until the relation above corresponds to thejudgment
on relative weights. If, on the other hand, F
1
is considered to have
greater weight than F2 + F3 , the inequality sign of the relation above
would be reversed, and, ifnecessary, the weight of F
1
may be adjusted.
4. Compare F, with the combination F
2
+ F
3
+ F
4
, and repeat the process
of adjusting the value of W(F
1
) if necessary.
5. The process of comparison ilnd adjustment is continued according to
rhe following pattern (for a given problem, it is to be expected that
many of these comparisons will not be needed, thus shortening the
process):
Compare F, with F
2
+ F
3
+ F4 + F
5
Compare F1 with F2 + F
3
+ F4 + F
5
+ + FN
Compare F2 with F
3
+ F4
Compare F2 with F
3
+ F4 + Fs
Compare F2 with F
3
+ F4 + F
5
+
. + FN
Compare FN- z wirhFN- 1
+ FN
~
6. Once !he weights [ W(F1) + W(F
2
) + + W(FN)] have been ad-
justed and checked for consistency to the satisfaction of the decision
! maker, it is common to "normalize" them to sum to 100 points by
multiplying each individual weight by
Sec. 8.6 Weighting of Allributes 225
100
N
2,: W(F;)
i= l
Example 8-5
Suppose that management desires to establish weights for the five factors or attri
butes identified above. The following are the identifications and initial weighting
assignments for the factors in rank order.
Factor (Attribute) Name Identificati on
Net present worth B
CIMS tactical aims A
Serviceability C
\
Riskiness, lack of E
Management/engineering effort D
Initial
Weighting
Asignment
1{)()
65
40
40
15
It is desired to illustrate the determination of weights in the spirit of, but not exactly
following, the consistency checking procedure in steps 3 through 5 (above).
The first consistency check will be to compare B with A + C. Suppose that the
decision maker feels that B is slightly less important than the combination of A and
C. This judgment agrees with the initial assignment of weights. That is,
Judgment: W(B) < W(A) + W(C), where W(B) is weight of factor B, etc.
Weights: 100 < 65 + 40; 100 < 105; .'. OK
The next comparison could be bJtween A and C + E. Suppose the decision
maker feels that A is more important fuan the combination of C and E. A check of
this compared to the assigned weights shows
Judgment : W(A) > W(C) + W(E)
Weights: 65 < 40 + 40; 65 < 80; .' . NOT OK
Thus there is an inconsistency to be resolved. This might be done by increas-
ing W(A) alone by m o ~ than 15 points (i.e., 80 - 65), by decreasing W(C) and/or
W() by a sum of more than 15 points, or by sorne combination ofthese adjustments
for consistency. Suppose it is judged that W(A) should be increased by 5 points and
that W(C) and W(E) each should be decreased by 10 points . With these changes, the
weights are fou nd to be consistent with the judgment.
Judgment : W(A) > W(C) + W(E)
Weights : 70 > 30 + 30; 70 > 60; . . . OK
Ifadjustments in point assignments are significant, it may be necessary to recto prior
consistency checks. This process may be continued until one is satisfied with tht
-
"'-J
w
V1
i
i
1':
( .
i '
l
'
1
1
' {;
:I:
.
t'
1
t
:/,
1
!
226
Mu/tialtribute Decision Ana/ysis
Chap. 8
8-4
C. .. LCULATION OF NORMALIZED FACTOR WEIGHTS
Factor (Attribute)
B. Net present worth
A. CIMS tactical aims
C. Serviceability
Weight
1= W(-)J
100
70
D. Management/engineering eiTort
E. Riskiness , Jack of
30
20
30
LW( ) = 250
Nonnalized Factor Weight
[
= W() x 100 w,]
LW()
40
28
12
8
12
L = 100
w_eights . Table 8-4 shows the normalization (to sum to 100) of the factor weights,
which presumably have been checked sufficiently for consistency.
8.6. 1 Formula Methods for Weighting
Many numerical formula methods for assigning weights exist that are
easier but generally much less defensible than the procedure advocated
above (even if checking for consistency is not done completely). Severa! are
described below, and example calculations are shown in Table 8-5 for the
same five factors (altributes) as shown in Table 8-4. As in Table 8-4, the
weighls are expressed as percentages.
1. Uniform or equa/ weights. Given N altributes, the weight for each,
W1 == liN x 100%
(8-1)
2. Rank sum weights. If R1 is the rank position of attribute i (with 1 the
highest rank, etc.) and there are N attributes, rank sum weights for
each altribute, W;, may be calculated as
W == N - R; + 1 x 100%
' N .
L (N - R
1
+ 1)
(8-2)

3. Rank reciproca/ weights. Rank reciprocal weights using the same no-
tation as above may be calculated as
IIR
1
W; == N x 100%
(8-3)
L (t i R;)
j ::z l
"
-
\
V) "'

oj
"::>

:;;l
1:-< 0
l
...
2
;:!
"'
Il
1 VI
"'
;;;

>
"
"'
"'
en
.:.:
)o c:
.. ..
3

"' ::c
1
..,
11:::0 -0000 N
;.:;
6 Il

...
""'
0
z
0

1=
j
::>
x;; 1 .., r- 0.., ,_,8
u
...
..- 1"'"\NN-
< -
u
1.
"'
1 H
...

o.
E
::>
< Vl
)(
.:.:

+

1 Il
:C:, H

3:
jj
1 oooool8 - NNNMN-
Il Il
H 1
v

t:
li
@
" 00
c:
c:
"' g
-5
"' .z: E ...., '60
'5 5 ...
c:
... u -
li
< - (,) 0
C :::: - ,.. U .z;;
'-J
"'
"' g bO .:.:
w
"'"' -:.;;! ..
c:
-:::n:
..
0'\

"'
227
--
;it
{
''l

.,_(

) ,

L L /l
Multiatlribute Decision Analysis Chap. 8
0
Comparing the methods in Table 8-5, note that the rank recipmcal
method gives the highest weight for the first-ranked attribute. One might
arbitrarily choose among the weighting methods above according to which
provides the closest approximation to the independently judged weight for
the highest ranked attribute. If we take that independently judged weight to
be the 40% in Table 8-4, it is seen to be closest to the 44% for the rank
reciprocal method.
8.7 WEIGHTEO EVALUATION OF ALTERNATIVES
Once weights have been assigned to attributes, the next step is to assign
numerical values regarding the degree to which each alternative satisfies
each attribute. This is generally a difficult judgment task using an arbitrary
scale of, say, between 0 and 10 or between 0 and 1,000 to refl,ect relative
evaluations for eac!h alternative and each attribute.
Example 8-6
Suppose thal we are comparing two alternatives on the basis of how.well-they satisfy
the live attributes with the weights developed in Table 8-4. The attributes, together
with the subjective evaluation of how weil a particular alternative meets each on the
ba sis of a sc ale of 0 to 10, are shown in Table 8-6. [Note: The se evaluation ratings
roughly correspond to the " scorecard" (Table 8-2), and sample graphs in Fig. 8-2.
Additionally, the " net present worth" amounts have been converted to a scale of 0 to
10. ].
Once the evaluations have been made, the results can be calculated as in Table
8-7 to arrive at weighted evaluations of attributes for each alternative. Thus the
summed weighted evaluation is 70.0 for alternative P-1 and 74.2.for. alternative..P-2,
(using Equation 8.4 at the bottom of Table 8-7), which indicates that alternative P-2 is
the better even though it happened to have the lower evaluation rating for three out
of the live attribut es.
Table 8-6
ExAMPLE EVALUATION RATtNG OF How WELL
Factor
Net present worth
CIMS tactical aims
Serviceability
Riskiness, Jack of
EACH ALTERNATIVE SATISFIES
EACH FACTOR (ATTRIBUTE)
Allernative
P-1 P-2
6 7
7.5 9
10 7.5
8 6
Management/engineering effort
8 6

\
,_ "'
. "'
CIO >

..0 <
"'z
.... ..:
"' !::;
<(
...
0
"' z
0

::>
...
<
>
Ill
c
"'
....
:.:
0
j;j

...
0
z
0

...
::>

0
N
tl..
.,
.::

E

<


'il il

c
.g CIJ
"' c ,;;
<3 "'
>QG
Ill
.,
.2
1il,'ii
- :l
.. -

;,::


<
c
.g
=

Ul

"E


:if-o
z:ge
- 0
;:s
u
:;
..0
s
<
C? oql "!
OOV"\0\f-'"11' "'f'
N N t-
"' r-0\ r-: \CI...o
Il
v-I

M
N N- r-
"' ..or---=:2oo tX>

OOO NN OO
"'t N --
t:
.g
0
QI)
c
c:
0
u
"' c
' 'o
0 .>< u

B]:;
s
Q,Vl .\!] "'
- :,z t "';

tl
v-I
...

.5

.SI!=
'iii
:l
il
>
u
x
.E
00
'

u
:;
..0
s
'iii
.,



0
c
Il
c
.!2
'iii
:l
il
>
.,
.,
.,
.E
co
'

229
"'-J
w
"'-J
Ait. P- 1
figure 8-4. Sample rormat to racilltat e calcul ation or weighted evaluation.
Format adapted substantially rrom "objectives matrix" ror measuring productivity
performance by Oregon Productivity Center, Corvalli s, OR; adapted with permission or ils
highl y productive direct or, the tate Dr. James E. Riggs.
230
-.
t -

Sec. 8.7
Weighted Evaluation of Alternatives
Project P- 1
10
"'
8
c
. .,
:!
c
s 0

::>
;;;
>
4 w
-\
1
r_l _____ ;.---
1 1
L __
r---------
1 1
1 1
1
1--- --
1
1 1 1 1 t:l
1
1
1
1 1;:; 2 1

1
V) i 1 :0 1 1 1
1
h 1 as-n 1

1 lE!!' 1
!5
1
u:l 1
t&;: 1
1

1 ij
\. 1
l!l 1 cl!
1 1
z
1 1
1 a:
1
1
1
1 1 L
1 1 ii 1
li
I O 10 20 30 40 50 60 70 80 90 100
Normalized weight , cumulative
Figure 8-S. Graphical portrayal or weighted evaluations ror two allemalives.
4_
8.7.1 Tabular and Graphical Ways to Show
Weighted Evaluation of Alternatives
231
Figure 8-2 showed sorne typical graphs relating attribute outcomes to
evaluaijpn ratings . Now we illustrate in Fig. 8-4 a tabular means which
-shows the same information and in addition provides a format* for easily
calculating weighted evaluations. The illustration is for project P-1 only,
and we see once again that it has a weighted evaluation of 700 (out of 1,000) ,
which is the same as 70 out of 100. The same form can be used for compar-
ing alternatives, together if desired.
Figure 8-5 is a graphical way of depicting the weighted evaluation of
alternatives P-l and P-2 as calculated in Table 8-7. The attribute weights are
shown to scale on the horizontal axis , and the evaluation ratings are shown
to scale on the vertical axis. Hence the areas for each alternative are in
proportion to their respective weighted evaluations. ln this case the differ-
ence does not stand out , but project P-2 does have a somewhat greater
area. Figure 8-6 shows the differences in weighted evaluations for the two
lfernatives, which may facilitate graphical interpretation better than
Fig. 8-5.
--..
'-1
w
co

BL Multiattribute Decision Ana/ysis Chap. 8
"'
c:
2
d. d. 1 ----
0
K -1
<]- -2
P-2 > P-1
40
P-2 < P-1
Normalized
weight,
cumulative
Figure 8-6. Graphical portrayal of difference in weighted evaluations for Iwo
alternatives.
8.8 VARIATION OF WEIGHTED EVALUATION METHOD
THAT CONSIDERS OBJECTIVE (ECONOMIC)
MEASURES AND SUBJECTIVE MEASURES*
The following variation of the weighted evaluation mode! involves first
transforming any objective (economie) measure of merit for each alternative
into a score between 0 and 1 so that the sum of the scores totals
1. For subjective attributes, such as lack of riskiness, nominal ratings are
th en transformed into nu me rica! scores, and the relative importances .of the
attribi.Jtes are weighted so thal the scores of each attribute over ali alterna-
tives and the weighted (subjective) scores over ali attributes each sum to 1.
Next, weights are assigned to the objective and the subjective score.s such
that those two weights sum to 1. The combined weighted scores (which also
sum to 1) show the relative desirability for each alternative.
Expressed in formula form, the combined measure (weighted evalua-
tion) for each alternative k, WEt. is
WEk = (a)(OMk) + (1 - a)(SMk)
where OMk = objective measure for alternative k
SMk = subjective measure for alternative k
a = relative importance weighting for OMk
and where OMk, SMk, and a are each 0 and :s 1.
(8-5)
We will demon strate determination of OMk, SMk, and WEk by exam-
pies below. The examples also include the use of certain formulas or ap-
proaches for determining weights and evaluation ratings which are merely
different than the more general methods for determining weighted evalua-
tions shown previously.
Method tirst developed in 1972 as the Brown-Gibson mode!. One signiticant version
with generalized modifications was b.y . .E' . -Y. Huang-and P. Ghandfo.{oush,
for Evaluating, Robots ," Indus trial Engineering, M-48. Adapted
with permission of publisher. . . . ..... .. . .
"'
Sec. 8.8
Variation of Weight ed Evaluation Method
J3
Objective Measure
The objective measure for each alterative k, OMk, can be calculated by
one of the following:
(a)
If measure is returns or profits for each alternative k, cali it OFPt
(such as net PW), then
\ OM
_ OFPk
k- K
L OFPk

where K = total number
of alternatives
(8-6)
(b) Jf measure is costs, for each alternative k, cali it OFCk (such as PW-C
or AC), then
OMk = OFC
1
x S
(8-7)
where
s=L.
k=t OFCk
Example 8-7
Alternative CIM systems are expected to have the following total equivalent a nnual
costs.
....
Alternative k
Il
Ill
OFC,
$130,000
150,000
165,000
Determine the OM for each alternative.
Solution
and
/
Similarly,
1 1 1
s = 130,000 + 150,000 + 165,000
= 0.000020418
1
OMt = 130,000 x 0.000020418 =
0

377
OMu=
OMw =
= 0.326
= 0.297
L = 1.00
'-.J
w
1.!)
234

00 z
i=
.0 <

z
0
=
<
:::>
....
<
>
t1l
0
z
<
"' ...
"'
0
;;;

w


0!
:::>
Cil
il:


>

E

<
...
.8
.

.. 00
3 .5
..
> ..
tl.JO::
H
"'
"" .5
..
a::
c
.Q
..
::s
...
>
t1l

:::
" "
>
::: ::s
u.o
" s 'E_
,;j<
il:-,
.. -

E::



... " <
0
U.tllO




tllU.O
c

NON

8
O>O
"'
.:;



gg
0000

>>
5

Soo
... >.
8 t;
o..>
:!.



U.tll
"' >.
e o o " >-
c....; .iJ:o
.... {l
.-:: 0 c. > ti.

:::::1 0 ")(
Cl u <"-l
0 V\0
N--
H -

Sec. 8.8 Variation of Weighted Evaluation Method
235
Subjective Measure
The subjective measure for each alternative k, SM*, can be calculated
for N subjective attributes, as we demonstrated earlier for weighted evalua-
. tions in Section 8. 7\
N
SMk = 2: (subjective attribute weight1) x (subjective evaluation rating;t)
1= 1
1
(8-8)
For the example below, let us assume thal subjective attribute weights are
assigned with a 20 being the highest, and that the subjective evaluation
ratings are assigned as follows:
Example 8-8
excellent = 4
very good = 3
good = 2
fair= 1 ..J_
poor = 0
For the same alternative CIM systems for which objective measures were shown
above, Table 8-8 gi'les mw subjective weights and evaluation ratings for each. Table
8-9 shows those numbers each divided by their respective totals, so they will add to
1. These we might cali "unitized" weights and ratings . Calculate the SM for each.
Solution
20 (2) 15 (3) 10(2) 10 (1) 5 ( 4 )
. = 60 7 + 60 8 + 60 6 + 60 4 + 60 10 = 0.3
20
Similarly (as shown on bottom of Table 8-9),
SMn =
SM111=
= 0.302
= 0.378
L = 1.000
Table 89
UNITIZED SUBJECTIVE WEIGIITS AND EVALUATION RATINGS'
Unitized /
Weight
20160 \
' 15/60 \
10/60 1
10/60 /
5160/
L a J.OQ
Subjective
Attribute
Quality of results
Ease of use
Competitive
Adaptability
Expandability
SM1
U nitized Evaluation
Ratings for Alternatives
1 Il Ill
2n 417 117
3/8 1/8 4/8
2/6 2/6 2/6
114 0 3/4
4/10 3/10 3/10
0.320 0.302 0.378
Final SMs are shown at bottom for each alternative .
l: Evaluation
Ratings
1.0
1.0
1.0
1.0
1.0
L SM, = 1.0
.....
'-.!

0
l36 Multiattribute Decision Analysis Chap. 8
Weighted Evaluation
The final weighted evaluation results for each alternative k (WE*)
for any a ( 2!: 0 and :5 1) can now be calculated as WE* = (a)(OM*) +
( 1 - a)( SM*) .
Example 8-9
If a = 0.3, calculate the WEs.
Solution
WE
1
: (0.3)(0.377) + (1 - 0.3)(0.320) = 0.337
WE
11
: (0.3)(0.326) + (1 - 0.3)(0.302) = 0.309
WE
111
: (0.3)(0.297) + (1 - 0.3)(0.378) = 0.354
L = 1.000
Thus. for a = 0.3, alternative III is slightly better than alternative 1, which is some-
what better than alternative IL
Very often it is useful to depict the effects of differing weights on the
weighted evaluation as illustrated in Fig. 8-7. The vertical dashed line at
0.38
0.377
0.37
0.36 -
c 0.35

0.34

'0
Il 0.33
.c
0> 0.326

0.32
0.31

0.30 0.42 = " breakeven" for
alts. 1 and Ill "'1 0.297
0.29
0 0.2 0.4 0.6 0.8 1.0
a '"' weight of objective measure
1.0 0.8 0.6 0.4 0.2 0
( 1 - a) a weight of subjective measure
Figurr 8-7. Weighted evaluation for ali possible weights ofOM and SM in Exam-
ple 8-9.
t
Chap. 8 References 237
a = 0.3 helps 'onfirm the results above for the three alternatives. The graph
shows, among other things, that alternative 1 is preferred to alternative li for
ali values of a , and that alternative 1 becomes preferred to alternative Ill for
ali a slightly greater than 0.4. lndeed, the " break-even" value of a can be
calculated by equating the WEs for alternatives 1 and Ill as follows:
{a)(0.377) + (l - a)(0.320) = (a)(0.297) + (1 - a)(0.378)
a = 0.42
8.9 SUMMARY AND PROLOGUE
This chapter has concentrated on basiq{raphical, tabular, and additive
weighting techniques for aiding in the analysis and resolution of multipl e
attribu._te deeisin problems. In the following four chapters we explain ot her
more detailed, quantitative methodologies; namely, use of the multiattribut e
utility method, analytic hierarchy process, goal programming, and expert
system technology.
It should be recognized th at presentation of most of the methodologies
in this and other chapters is based on "assumed certainty. " Thal is, we
normally assume that required input estimates (such as weights, ratings,
trade-offs, etc.) are known quantities . Of course, most real
decision-making situations involve uncertainties. The main recommended
approach to dealing with uncertainties is to perform sensitivity studies to
determine the relative effect on the indicated decision(s) due to changes in
important input estimates for whatever methodology and to display the
results in easy-to-understand ways .
The determination of what methodology(ies) to use depends on the
nature of the decision problem and the preferences of the decision
maker(s). Although this determination is a very inexact science, it can be
shown J)lat the use of widely differing models will often have Jess effect on
the probable quality of the solution than does the unintended omission of
possible outcomes, alternatives, or important criteria. Whatever methodol -
ogy is chosen should make the decision maker more confident and ultimately
make his or her job easier.
REFERENCES
1. MacCrimmon, Ken R., Decision Making among Multiple Attribute Alternatives :
A Survey and Consolidated Approach, Memo RM-4823-ARPA, Rand Corpora-
tion, December 1968.
2. Canada, John R. and Robert L. Edwards, Should We Alllomate Now? Evaluation
of Computer lntegrated Manufacturing Systems, Industrial Extensiotl Service,
North Carolina State University, 1986.
'-.1
-t:>.
f-'-
1
:1
. 1
,,
1 i
238
Multiallribwe Decision Analysis Chap. 8
EXERCISES
8-1. !Section 1!.4) Select a class or type of significant decision problem in your work
or personallife. Name the three or more significant auri butes to be considered
for thal class of problem. Assume that the weighting of these attributes is
nebulous, so you think il desirable to develop only a matrix of alternatives
versus attributes, with entries to show how weil each alternative meets each
attribute in whatever measures are appropriate, such as in dollars, lime, rank,
and so on. Th en use co lors or other symbol codings to facilitate ease of under-
standing of the differences by the persons making the final selection among the
alternatives .
8-2. (Sections 8.5 through 8.7) Select a class or type of decision problem in your
work or personallife involving multiple objectives (not necessarily same as for
Problem 8- 1 ). Na me three to live most important objectives for that class of
problem. as you see them, and do the following:
(a) Using the method of paired comparisons, rank the objectives in order of
decreasing importance.
(b) Weight them and check for consistency and then normalize the weight to
sum to 100.
(c) ldentify two to four alternatives for the decision problem and evaluate how
weil each alternative meets each objective on a scale of 0 to 10. Then
multiply the evaluation ratings by the objective weights and sum these for
each objective to compute a weighted evaluation for each alternative.
8-3. (Sections 8.5 through 8.7)
(a) Weight the relative importance of the three or four most important aUri-
butes you would consider in selecting a job. Assume that these are the
only attributes you will quantitatively consider. Show how you make
comparisons for internai consistency and normalize the factor weights to
sum to 100.
(b) Using the attribute weights developed above, obtain a weighted evaluation
of two alternative jobs in which you might be interested by evaluating how
weil each job satisfies each attribute using a scale from 0 to 10.
8-4. (Sections 8.5 through 8.7)
(a) Weight the relative importance of the three to six most important auri-
butes you would consider in selecting a persona! car. Assume that these
are the only attributes you will quantitatively consider. Show how yo6.
make comparisons for internai consistency and normalize the attribute
weights to sum to 1,000.
(b) Using the attribute weights developed above, obtain a weighted evaluation
of three alternative makes of cars you might consider for purchase by
evaluating how weil each make of car satisfies each attribute. Use a scale
from 0 to 20.
8-5. (Sections 8.4 through 8.7) Describe a significant multiattribute decision prob-
lem of existing or potential meaningful interest to you. Examples are: housing,
car, job, spouse, !ravel, and equipment. ldentify at !east four most .relevant
attributes which are as independent of each other as possible and at !east two
Chap. 8 Exercises
i 0
, , L 239
'
.. .--
or three mutually exclusive alternatives. (Note: ln the process of further anal-
ysis, you might weil add or delete alternatives and/or attributes .)
(a) Describe the problem as meaningfully as possible using two or more
graphical techniques . What conclusion (choices), if any, can you make
from these?
1
(b) Rank order the attributes. Can you combine this with one or more of the
graphical techniques in part (a) to make choices? If so. describe .
(c) Weight the aUri butes by the method or formula(s) of your choice and show
at !east two checks and/or revisions of weights for consistency between
initial judgments and preferences.
(d) For each attribute show a graph of attribute out come (x axis) versus evalu-
ation rating (y axis) .
(e) Using parts (c) and (d), show calculations weighted evaluations for
each alternative and display the results for ease of comparison.
8-6. (Section 8. 1) For' multiattribute decision problem of existing or
potential meaningful interest to you (preferably an entirely different problem
from Exercises 8-1 through 8-5), apply "Ben Franklin's method" of listing
pros and cons and canceling out combinations thereof. His description as-
sumed just one alternative or course of action is being considered. If two or
more altematives are considered, his process can be used for each alternative
by il self, and further cancellations can be do ne if one can fi nd one or more pros
which ar-e-equal for ali alternatives, or one or more cons which are equal for ali
alternatives .
8-7. (Sections through 8.7) De scribe a significant multiattribute decision prob-
lem of exasting meaningful interest to a "client" (friend, business person, or
relative). Describe the problem as clearly as possible using words, tables,
graphs, and so on, in sufficient detail to facilitate analysis by two or more
methods described in this chapter. [Note: Problem should consist of at !east
three (independent as possible) attributes and at !east two alternatives . ln the
process of analysis you might weil add or del ete alternatives and/or attributes ,
1
\ but do not go below the minimums above.)
\ 8-8,1 (Section !r.'3) Given the following matrix of outcomes for alternatives and
, attributes (with higher numbers being better), show what you can conclude
using the following sequential elimination methods:
(a) Conjunctive constraint.
(b) Dominance.
(c) Lexicography, with rank order of attributes D > C > B > A.
Alternative
Standard
Attribute t 2 3 "Ideal" (Minimum Acceptable)
A 60 75 90 100 70
B 7 7 8 JO 6
c Poor Excellent Fair Excellent Good
D 7 8 8 10 6
'.J

N
240 Multiallribute Decision Analysis Chap. 8
8-9. (Sections 8. 3, 8.6, and 8. 7) Management has asked you to look into moving the
"cutting" operation from where it is now to a new location. Management has
given you three alternative areas, which are mutually exclusive, to consider.
After inspecting the three areas and considering which factors reHect signifi-
cant differences among the alternatives, you have decided on five independent (
attributes by which to evaluate the alternatives, listed in descending order of
importance:
A. Distance traveled from cutting operation to next operation [more distance
is worse]
B. Stability of foundation [strong (excellent) to weak (poor)]
C. Access to loading and unloading [close (excellent) to far (poor)]
D. Cost of the moving operation
E. Storage capacity
(Note: The cost for operating the cutter once the operation has been moved is
independent of the area chosen and hence is the same for each area.) The data
of these factors for the three alternatives are as follows:
Alternative
Anribute Area 1 Area Il Area Ill Ideal Standard Worst
A 5000 ft 3000 ft 750ft Oft 3000 ft 10,000 ft
ll Good Very good Good Excellent Good Poor
c Excellent Very good Good Excellent Good Poor
D $7500 $3000 $8500 so $5000 $10,000
E 60,000 ft' 85,000 ftl 25,000 ft' 10,000 ftl 25,000 ft' 150,000 ftl
(
Determine which alternative is to be chosen using:
(a) Alternative versus aliernative: comparison across alternatives.
(l) Lexicography.
(2) Elimination by aspects .
(b) Weighted evaluation of alternatives using the rank reciprocal method for
weighting attributes.
(c) Weighted evaluation of alternatives using uniform or equal weights for
attributes .
For parts (b) and (c) the following are evaluation ra ting functions on a scale of 0
to 10.
r- '
"
241
Chap. 8 Exercises
Range for
A.uribute Attribute Outcome Evaluation Rating
A 0 s x s ft -x/600 + 10
3,000 s x s 10,000 ft -xli ,400 + 5017
Band Excellent 10.0
Very good 7.5
Good 5.0
Fair 2.5
Poor 0
D $0 s x s $10,000 -x/1,000 + 10
El - 15,000 s x s 25,000 ftl 35/2 - x/2.000
25,000 s x s 100,000 20/3 - x/15 ,000
:r Two highway alignments have been proposed for access to a new manufactur-
\_./ ing plant. Based on the data and information given below, comparison can be
made of the two alignments. The better one must be chosen as the proposed
highway alignment that will be the connector between an interstate highway
and tho proposed site. The route length for Alignment A is 4.7 miles and for
Alignment B, 5.1 miles. The monetary costs are as follows:
Item Alignment A Alignment B
Land s 4,044,662 s 4,390,000
Bridges 10,134,000 8,701,000
Pavement 4,112,500 -4,-;2,.500
Grade and drainage 7,050,000 7,650,000
Erosion control 470,000 510,000
Clearing and grading 188,000 204,000
/
Total $25,999,162 $25,917,500
Nonmonetary attributes are:
Alignment A Alignment B
Maintenance High
Noise pollution Very good Good
Cost savings (on gas) Excellent Poor
Accessibility to another U.S. Highway 41 None
roadway
Impact on wildlife Little Little
Relocation of residences 2 3
Road condition Flat Hill y
Which highway alignment would you recommend? Show ali work.
""'-J

w
242 Multiattribute Decision Analysis Chap. 8
811. (a) Use the weighted evaluation mode! to make a selection of one of the three
automobiles for which sorne data are given below. State your assumptions
regarding miles driven each year; !ife of the automobile (how long you
would keep it); market (resale) value at end of life; interest cost; priee of
fuel; cost of annual maintenance; and attribute weights and other subjec-
tively based determinations.
(b) Use the data developed in part (a) and the weighted evaluation method in
which objective and subjective measures are combined (OMk and SMk, as
in Section 8.8) to make a. selection. Do your answers in parts (a) and (b)
agree? Explain why they should (or should not) agree.
Alternative
Attribute Domestic 1 Domestic 2 Foreign
Priee $8,400 $10,000 $9,300
Gas mileage 25 mpg 30mpg 35 mpg
Type of fuel Gasoline Gasoline Diesel
Corn fort Very good Excellent Excellent
Aesthetic appeal 5 o/11 of 10 7 OUI of JO 9 OUI of JO
Passengers 4 6 4
Ease of servicing Excellent Very good Good
Performance on road Fair Very good Very good
Stereo system Po or Good Excellent
Ease of cleaning Excellent Very good Poor
upholstery
Storage space Very good Excellent Po or
8-12. Utilize any three multiattribute techniques described in this chapter to analyze
the following important si tuation; Y ou need to determine which job to take,
given acceptable offers from firms A, B, and C. Y ou first determine a set of
attributes thal are most important in evaluating and then construct a decision
matrix with alternatives as columns and attributes as rows, and fil! in. (Results
are given in the matrix below.) Analyze the alternatives with the different
techniques according to your perceptions (be complete-il look years to gel to ~
this point!) .
Alternative (Finn)
All ri bute A B c
Starting salary $30,000 $28,500 $33,000
Opportunity for advancement Excellent Excellent Fair
Management attitude Very good Excellent Good
Location Good Fair Poor
Type of work Excellent Poor Very good
Opportunity for continuing Very good Good Fair
education
1-
"
243
Chap. 8 Exercises
8-13. (Section 8.3) The follow)ng are evaluation ratings (on a scale of 0 to 20) of how
weil each of two alternatives satisfies each of four attributes.
Alternative
Attribute 1 2
A 15 12
B 8 20
,c
14 14
ID
13 12
If the attributes are ranked A > B > C > D in order of importance, the
minimum standard evaluation rating is 10 for each. Show which alternative
would be best by;
(a) Lexicography. -('
(b) Elimination by aspects .
(c) Dominance cheekt (assuming attribute B is eliminated from consideration).
8-14. (Sections 8.6 and 8.7) Given the same attributes and alternatives with evalua-
tion ratings as in Exescise 8-13, show which alternative would be best using the
weighted evaluation methodology and each of the following attribute weighting
formulas . (Note: Normalize the summed weighted evaluations to equal 100.)
(a) Uniform.
(b) Rank sum, with A > B > C > D.
(c) Rank reciprocal, with A > B > C > D.
8-15. (Section 8.4) Construct a polar chart depicting the evaluation ratings for the
.two alternatives in E ~ r c i s e 8-13.
8-16. (Section 8.7) Construct a graphical portrayal of Exercise 8- 14(b) with cumula-
tive weights on the x axis and evaluation ratings on the y axis .
8-17. (Section 8.8) Volunteer alumni A and B have contributed $100M and $150M,
respectively (higher is better!). However, evaluations of their leadership suc-
cess (on a 1 to 5 maximum scale) are 4 and 3, respectively . Contributions
(objective) are considered to be half as important as leadership success (sub-
jective). Using the weighted evaluation method combining these measures,
which should be the" Alumnus of the Year"? What part of 1.0 total "weighted
8-18.
brown ies" does he score?
(Section 8.8) For Exercise 8-17, assume the relative weight, a, for the objec- ,
tive measure (contributions) is not known. Plot the weighted evaluations for
each of the alumni for a ranging from 0 to 1.0 on the x axis .
'-.1
~
~
CHAPITRE 8
Dcider face la complexit.
In traduction
Our present complex environment calls for a new logic-a new way to
cope with the myriad factors that affect the achievement of goals and
the consistency. of the judgments we use to draw valid conclusions. This
approach should be justifiable and appeal to our wisdom and good sense.
lt should not be so complex that only the educated can use it, but should
serve as a unifying tool for thought in general.
That we are ali very much creatures of the moment needs little debate.
Try to reconstruct what was said two minutes before in a conversation and
you will soon discover that, even for a short duration, your recollection is
fuzzy. Even when you remember, the precision of your recollection is
usually less than exacting. Our understanding of the world not only needs
repetition to improve our recoll.ection and precision, but it also depends
very much on the in/ensi/y of our participation. lt is an exaggeration to
conclude that humans are logical creatures. It is more accurate to say that
our judgment depends on the totality of our impressions, even if they
cannot ,be logically and rigorously justified. For better understanding we
need to deal with experience as an ongoing process . We need concentra-
tion, repetition, diversity, debate, and, when necessary, consensus.
Severa! good !essons have been learned in recent years from political
leadership. Having a clearly defined goal has again been demonstrated to
be the core of successful statecraft. Being consistent is another vital ingre-
dient . Persuasion and support for one's view are still considered two of the
main attributes of a great leader. What ensues is largely designed to help
leaders get their points across .
More and more people are finding it hard to put ali their trust in the
unspoken, unjustified, and intuitive thinking of their leaders' decisions on
complex matters. Whatever internai mechanism leaders have needs to be
1
00
1-'-
2 INTRODUCTION
articulated and understood. Just as language itself and the rules of thought
had to be organized in a formai manner long ago, so must we now organize
our thought processes so they lead us to good decisions. We_should be able
to say th at, given the information, we agree on the methq_g._ o(maklng the
decision (but not necessarily on the quality that Th! matter
will then become a common concern rather than a mystical phenomenon.
The following pages offer the reader a method for organizing the in-
formation an.<J:judgments metfiofeflts
of feelings and logic bearing on the issues and then synthe-
sizes these diverse judgments into an outcome that agrees with our intui-
tive expectations as represented in ali the judgments we give.
The process contributes to solving complex problems by structuring a
hierarchy of criteria, stakeholders, and outcomes and by eliciting judg-
ments to develop priorities. lt also leads to prediction of likely outcomes
according to these judgments.
The outcome can be used to rank alternatives, allocate resources, con-
duct benefitlcost comparisons, exercise-ontrol in the systemh.y---;;;filating
the se!'lsitivity of the outcome to changes in judgment, __()ut
planning of projected and desired futures . A useful by-product is the
measurement of how weil the leader understands the relations among
factor': . Although people generally are not consistent, the main concern
here is the strength of their inconsistency. Is their understanding close to
capturing the interactions observed? Oris it a random understanding that
only hits the target now and then?
The process described in this book has had many applications so
far-to energy rationing, the Middle East conflict (1972), Sudan transport
planning (1973-1975), mineral exploration in Mauritania (1976), planning
for higher education in the United States (1976), the presidential election
(1976), the conflict in Northern Ireland (1977), planning for a research insti-
tute (1977), terrorism (1978), predicting the outcome of a world chess
championship match (1978), product portfolio selection (1979), the stock
market (1980), the presidential election (1980), oil priees in 1990 (1980), the
conflict in South Africa (1981), and severa! specifie corporate applications.
AU the applications have contributed to refinement of the principles set
forth in tlle following pags:-They nave a!so
us how to deal with larger and more complex decisig!l pr_q.blems. The
theory has been enriched and immensiyolfered by the scope and variety
6f these applications.
...
1
Making
Decisions in a
Complex World
This chapter deals with the following questions:
How can we understand com.plex problems
involvi_ng a great many factors?
What are the basic processes of thought and
behavior involved in making decisions?
What role should ethics play in the
decision-making process?
Why do we need a new way of thinking about
complex problems?
How might a group of people holding
varying opinions debate an issue in search of an
acceptable compromise?
What can they do about their differences?
3
<Xl
N
MAKING DECISIONS IN A COMI'I.EX WlliU.D
COPING WITH COMPLEXITY
To the best of our understanding, the world is a complex system of
interacting elements. The economy, for example, depends on energy and
other resources; the availability of energy depends on geography and poli-
tics; poli tics depends on military strength; military strength depends on
technology; technology depends on ideas and resources; ideas depend on
politics for their acceptance and support; and so on. In such an intricate
network of factors, first causes and final effects cannot be identified easily.
Our minds have not yet evolved to the point where we can clearly see these
ultimate relationships and readily resolve important issues like nuclear
energy, world trade, and environmental regulations.
ln our complex world system, we are forced to cope with more prob-
lems than we have the resources to handle. To deal with unstructured
social, economie, and political issues, we need to order our priorities, to
agree that one objective outweighs another in the short term, and to make
tradeoffs to serve the greatest common interest.
But it is often difficult to agree on which objective outweighs
another-'-particularly in complex issues where a wide margin of error is
possible in making tradeoffs. Leaders may be confused by the diverse
information provided by their assistants; they may need help in identify-
ing differences of opinion and seeing positions where compromise can be
reached . They may need to know which important issues must be re-
searched in depth to obtain better information and how sensitive the out -
come is to slight or drastic changes in opinion and judgments. Intuitive
thought processes that serve us weil in the familiar routine of daily )ife can
. mislead us on complicated matters where sources of information and opin-
ions are varied. lncreasingly we need to articulate and map out the issues
to see whether what we think and what we feellead us to the same kind of
answers .
Most of us be lieve !ife is so complicated thal in order to solve problems
we need more complicated ways of thinking. Yet thinking even in simple
ways can be taxing. If we struggle to examine collections of only a few
ideas at a lime, how can we understand complex problems involving a
great many factors? Simple thinking about such problems leads to combi -
nations of ideas whose structure is not unlike a dish of spaghetti in which
ali strands are separate- but tangled.
RATIONALE FOR A NEW FRAMEWORK
What we need is not a more complicated way of thi nking, since it is
d ifficult enough to do simple thinking. Rather, we need to view our prob-
lems in an organized but complex framework thal allows for interaction
and int erdependence among factors and still enables us to think about
ORGANIZING KNOWLEDGE FOR DF.CISIONS 5
them in a simple way. This new way of thinking should be accessible to ali
without straining our innate capabilities.
The analytic hierarchy process (AHP) described in this book provides
such a framework. lt enables us to make effective decisions on complex
issues by simplifying and expediting our natural decision-making pro-
cesses. Basically the AHP is a method of breaking down a complex, un -
structured situation into its component parts; arranging these part s, or
variables, into a hierarchie order; assigning numerical values to subj ect ive
judgments on the relative importance of each variable; and synthesizing
the judgments to determine which variables have the highest priority and
should be acted upon to influence the outcome of the situat ion.
The AHP also provides an effective structure for group decision mak-
ing by imposing a discipline on the group's thought processes. The neces-
sity of assigning a numerical value to each variable of the problem helps
decision makers to maintain cohesive thought patterns and to reach a
conclusion. ln addition, the consensual nature of group decision making
improves the consistency of the judgments and enhances the reliability of
the AHP as a decision-making tool.
In this chapter we consider the human behavioral and thought pro-
cesses involved in decision making. These natural processes form the basis
for the analytic hierarchy process, which is described in Chapter 2 and
explained in greater detail in Chapters 3, 4, and 5. The rest of the book
shows how the AHP can be applied to a variety of decision-making situa-
tions and problems .
ORGANIZING KNOWLEDGE FOR DECISIONS
The AHP can best be appreciated by first looking into how the human
mind organizes knowledge for decisions . The Iwo fundamental ap-
proaches humans have developed so far for analysis are the deductive
approach and the inductive or systems approach.
The Deductive Approach and the Systems Approach
We can analyze a system logically by representing it as a network and
structuring it into chains and cycles. ln analyzing natural systems, for
example, biologists break down networks into food chains, hydrologie
cycles, and so on. After structuring the network, we look for explanations
for the functioning of ils parts . Then, by an act of imagination, as no rules
of logic exist for combining these piecemeal explanations, we synthesize
an explanation for the whole network . But thi s sci entific, deductive ap-
proach ignores the feedback mechanisms among the parts and betwee n !he
parts and the environment thal affect the whole system.
00
UJ
- - - - - ~ - - ~ - - - - - - - - - - - - . - - - - - - - - - - ~ - - ~ -
MAKING DECI SIONS IN A COMPLEX WORLD
Systems theorists have pointed out that we can better understand an
en tire system by examining it from a general, holistic perspective thal does
not give as much attention to the function of the parts. For example, a caris
better understood by observing how il functions in the environment than
by studying the operation of its mechanical parts . In this way we see it as a
whole; we can simultaneously perceive how the car runs and how it inter-
acis with other cars, road conditions, traffic signais, and so on.
Clearly both the deductive and the systems approach contribute to our
understanding of complex systems. We can benefit by combining the two
within an integrated, logical framework: the anaJytic hierarchy process.
The AHP enables us to structure a system and its environment into mutu
ally interacting parts and then to synthesize them by measuring and rank-
ing the impact of these parts on the entire system. We will see how this
process works with systems as diverse as a transport network in the Sudan
and the steel industry in the United States. By providing a new logic for
synthesis, this structured approach to decision making eliminates much of
the guesswork and confusion of our ordinary method of synthesizing an
overall explanation for a system from piecemeal explanations arrived at
through deduction.
The Role of Logic, Intuition, and Experience
The everyday way we proceed to understand and 5olve problems is to
use logical deduction to argue through familiar matters. For example, it is
easy to reason that to increase capital one must obtain a good retum on it
by investing it or by obtaining a good interest rate. In this case we know
that in order for capital to increase from size A to size B, money has to be
added to A somehow. That much we can say with bold certainty. But we
tend to treat larger, imprecise perceptions of a problem by relying on
feeling, experience, emotion, appeal to other people's understanding, and
sometimes even force . Many political problemsin advanced as weil as Jess
developed countries are handled in such a fashion . In unstructured situa-
tions people often act on the ir" gut feelings" rather th an strict! y on rational
grounds . Logic plays a role mostly in arranging words and ideas after the
conclusions have been reached.
People in the public and private sectors tend to cooperate in defining
and structuring their problems broadly and richly so that ali their ideas can
be included. But when they need to explain which factors have the greatest
impact on the outcome of a decision, not even experts with the clearest
logic can hold fast to their positions in the face of objections. As a result
they are willing to compromise. Thus decisions are based not so much on
the clarity of ideas or amount of information exchanged as on the persis-
tence of sorne participant in the decision-making process and on that per-
ORGANIZING KNOWLEDGE FOR DECI SIONS 7
son's ability to persuade others to accept his or her ideas, like a politician
selling himself in a campaign.
People, then, not only have different feelings about the same situa-
tion, but their feelings change or can be changed by discussion, new
evidence, and interaction with other experienced people. Usually the out
come is a compromise of many viewpoints involving substantial change in
individual attitudes . The fact is that when we make decisions, persona!
preference and persuasion usually prevail over clear and straight logic.
Our actual d.ecision-making processes have been illuminated by recent
studies conducted by behaviorists, other psychologists, and brain re-
searchers. Let us examine their findings .
Behaviorist Theories
Those who study and explain ways, reasons, and consequences of
human and animal behavior are finding it difficult to uphold the idea that
humans are rational animais. Their theories are helping to create an atmo-
sphere in which people are accepted as they really are rather than as they
were idealistically portrayed during the Renaissance and the Age of Ra-
tionalism. Human behavior is enormously complex; the many theories
explaining human action are deep and multilayered, and probably ali con-
tribute to our understanding of human behavior .
Instinct-Drive Theory. Sorne theorists consider rational thinking to be
but a thin veneer over human behavior. Much of our action is driven by
instinct-patterns woven into the mind, bone, and muscle. Just as wasps
have an instinct for nest building and birds have their characteristic songs,
humans also follow certain unleamed patterns of behavior, such as seeking
food, mating, avoiding pain, caring for young, and so on. Although
instinct-drive theory describes such patterns, it does not explain them. lt is
inadequate to account for most adult behavior, including sentiment, value,
ambition, attitude, taste, and inclination.
Reason-Impulse Theory. We tend to regard ourselves as rational animais
capable of making choices based on objective, or real , criteria. We fee! thal
most of our decisions flow from logical necessity, not from whim and
caprice. Although we may acknowledge that needs and persona! motives
are the driving forces behind human behavior, we contend that we use
reason to attain our goals efficiently and without harm or injury. Through
reason we get what we want within the limits of available resources . And
many of us ultimately do learn to apply rational techn iques to deci sion
making, regardless of what our persona! wishes may dict ate.
But critics of this view say that our so-called reason is an abyss of
(X)
_p.
8 MAKING DECISIONS IN A COMPLEX WORLD
unconscious or barely conscious urges and habits that overwhelm the intel-
lect . They argue that human relationships are essentially governed by irra-
tional, emotional forces; rational appeal in most cases plays only a minor
role. These reason-impulse theorists hold that our actions ar bas.ed on
imitation, habit, suggestion, or other subrational forms of thinking and
are rarely due to pure logic. Planned actions are the result of analysis based
on preferences as to which objectives are served best-and preferences are
strongly influenced by habit and training rather than by rational thinking.
Dynamic Field Theory. Other behaviorists point to the influence of envi-
ronmental factors on human behavior. We act in response to a "dynamic
field" of stresses and tensions when we perceive the environment to deny
or fulfill the satisfaction of our wants and needs. The hierarchy of human
needs that motivate behavior has been examined by Abraham H. Maslow
and others; these needs range from the most basic physiological and safety
and security needs to sophisticated self-actualization and esthetic needs.
Leaming Theories
Most people tend to assume that the way we think and
use to develop our thinking are innately human _and that the basics of
human knowledge have come down tous as a packa.ge from heaven: But
recent learning theories argue that we learn mainly by triaLmd error and
through feeling rather than through logic. Stimulus-response theory, .. for
example, maintains that we gravitate somewhat randomly in certain direc-
tions to satisfy our needs and desires; acts that bring satisfaction are rein-
forced. Gestalt theory tells us that even what we consider to be
" insight" -perceiving the necessary relationships in a situation-results
from feeling.
Generally we have vague feelings and inklings about what we think
we have experienced, but we are not attentive enough to register ideas and
feelings sharply. We have no systematic way of reconstructing from mem-
ory that which we did not learn, understand, or memorize consciously.
Most of our daily experience goes before our senses and passes through our
feelings like a hazy cloud that slightly moistens the environment but
makes little difference to the growth of our understanding.
Our understanding does grow when a particular experience connects
weil with our earlier experiences, not mere! y with our knowledg;-or when
it shocks us, capturing our attention involuntarily as it intrudes into our
being (sometimes unconsciously) either pleasantly or forcefully.
can be defined as the ability to recognize a specifie act in the light _()f
1
,.
,.

1
r
1
(:
:
1
1
ORGANIZING KNOWLEDGE FOR DECISIONS 9
previous experience. lt is an iterative, or repeated, process of adding
knowledge thatelaborates on or expands existing knowledge. Leaming
can" be conscious and intentional, as in memorizing facts, or it can be
unconscious and unintentional, as in discovering physically from experi-
ence that eating green apples results in a stomachache. We generally agree
that people who experience a phenomenon firsthand are the ones who can
best shed light on our understanding of it; indeed, knowledge derived
from experience is basic to ali understanding.
Brain Research Findings
The importance of intuition, feeling, and experience in human behav-
ior and decision making is further underscored by the findings of brain
researchers. In particular, they have discovered a distinction in the func-
tiors of the two halves of the neocortex of the brain-the left is the logical,
rational, and calciiiating member; the right is the intuitive, creative, and
inarculate half. The verbal half's job is to interpret for ourselves
and the world the dcisions of its mute brother. Note that the decisions are
actually made by the intuitive, not the logical, half. The left hemisphere
mere! y arranges and puts into words the insights of the right.
Studies of human perception show that our senses both condition and
limit whatever enters our consciousness. Older people who have experi-
enced !ife longer often recognize that illusion is primary experience. (Per-
haps this awareness results from the aging of the senses, but more likely it
is due to the wisdom of age.) Our senses shape our world; thus we can
never interpret the universe with absolute accuracy. The qualities we
study are simply those we can perceive, and the laws we develop are
concoctions of our sense-limited brains.
Another !ife form could have many more senses than we have; for
example, it could have a magnetic eye, could perceive the colors of the
spectrum in white light, or could see through objects. As a result, its
consciousness would be different from ours. Of course, our senses have
expanded through human inventions such as the microscope, telescope,
and X ray. We can perceive many qualities not directly accessible to our
senses, and these are only a small portion of the potential total.
If we had more senses acting at once, we would probably have diffi-
culty sorting out our perceptions and understanding their relations, given
the present stage of the evolution of our brains. Although sense data can be
organized in chronological sequence, brain events are not rigidly bound
by time. Ideas can occur before or after other ideas . Because the way we
think is fluid, we are freer to arrange our ideas in the manner we desire-
for good or evil purposes.
co
lJl
!
i
t
l
~
- .. -- ------ -- ---................. - - - - - - - - - - ~ - - - - - - - ..
MAKING DECISIONS IN A COMPLEX WORLD
ETHICAL CONSIDERATIONS
When complexity makes normal !ife appear difficult and even hopeless
to many, a strong individual may emerge to take over leadership charis-
matically by declaring that ali the pain and confusion are simply due to a
single problem that must be solved; Hitler is an infamous example. Such
leaders draw attention to one problem and persuade or force others to
believe in them. By oversimplifying the situation, they banish ali other
problems to narrow the perspective and make themselves appear logical in
their explanations; they linearize the workings of a system in a deductive
fashion .
We accept new ideas as truths on the basis of the manner in which
they cohere with knowledge we already possess. A system of beliefs could
be perfectly consistent, and yet each belief could be false . Severa! examples
come to mind. The Gregorian calendar (1582) was based on a set of consis-
tent assumptions that the earth is the center of the universe. Yet it was so
accurate that the accumulated error was one day every 3323 years . Our
calendar is in error every 20,000 years . This is an example of a consistent
prevailing view that was in error according to modern astronomy but
produced good results. An example of a consistent theory of our ti me that
produced bad results until recently involves the humble golf bali. People
used to believe that it is the perfectly spherical golf bali thal travels farthest
when hit-until it was discovered that the more a bali is used, the more
dents and dimples it develops, which serve as wings countering drag and
sustaining it longer in the air. Now golf balls are made according to pre-
cisely dimpled patterns.
Consistent thinking with no real validity is frequently espoused by
lunatics and other mental cases (sometimes of a seemingly respectable
genre) . Beliefs can be deduced from one another in perfect consistency
depending solely on the observance of formai, linear relationships be-
tween them. Thus one may develop a consistent deductive system with no
real validity in this world.
Because it is possible in complex, unstructured situations to present
convincing arguments that have little correspondence to reality and may
harm society, one must apply certain ethical standards to the decision
making process. The philosopher Alasdair Maclntyre of Boston University
has identified four qualities thal should characterize a decision maker's
approach to dealing with social issues :
Truthfulness by not oversimplifying complexity . Our political and legisla-
tive processes demonstrate thal it is easier to consider issues such as
environmental protection or health care in a narrow, piecemeal way
than to look at ali the critical variables, fit them together, and determine
their priorities and implications. ln the short run a simplistic approach
PERSPECTIVE 11
may satisfy local contending parties, but it is no way to get at the
answers to complex problems.
Justice by evaluating costs and benefits and assig11ing costs to those who get
the benefits. Everyone involved in a decision-making situation- family
members deciding whether to purchase a home computer or corporate
executives deciding which companies to invest in-should have a
chance to weigh costs and benefits. Those who receive the benefits
should be the ones who pay the costs, and vice versa. Justice demands
not only thal everyone have a voice and a vote but also thal those who
will bear the risks and dangers have more of a voice and vote than
others. (People should also be informed about where and how they are
paying for benefits, particularly for commodities whose priees are con-
trolled or subsidized.)
Ability to plan for the unknown by calculating changes, determi11il18 where
they are likely to occur, and deciding which przorities should di ctate actio11 .
Leaders must be able tb plan and deal both with projected futures, such
as higher energy priees in 1985, and with desired and less predictable
futures, such as energy independence by the year 2000.
Flexibility in adapting to cha11ge by plallni11g , implementi11g , a11d, i11 re-
spOIISe to new conditions , repla1111ing and reimplementing . This iterative
approach is essentially a learning process; it tempers our tendency to let
immediate needs dictate short -lerm solutions . For example, fle xibility is
necessary in planning strategies for the use of alternative resources in
dealing with the energy situation.
As we will see, the analytic hierarchy process encourages socially re-
sponsible decision making by helping leaders to avoid oversimplification,
to identify and evaluate costs and benefits, to plan for the future, and to
adapt to change.
PERSPECTIVE
Most of us have trouble coping with ordinary problems of society that
cannot be understood bya deductive, linear, cause-and-effect explanation.
Our respect for the scientific method, which relies on deduction, has led us
to try to solve ali our problems through logical debate. As a result of our
scientific education and because science usually deals with thing-s we can
observe through our physical senses, we are made to fee! that there is
precis_ion in what we do. Our senses are trained to be consistent in focus -
ing on their abjects; thus our minds have a sense of consistency in syn-
thesizing and interpreting sense data. But when we deal direct! y with ideas
rather than with sense perceptions, things tose their precision. The reason
(X)
0"'1
... --- -- -----
MAKING DECISIONS IN A COMPLEX WOHLD
is that we use words whose meanings are imprecise. Philosophers have
long recognized that primary language does not express thoughts or ideas
but feelings and affections. Moreover, we cannot be exact in describing
abstract relationships, and our understanding is conditioned by our states.
of mind, feeling, and imagination at the moment we are thinking.
Thought without language is impossible. Abelard says that "language is
generated by intellect and generales intellect." According to the Chan-
dogya Upanishad, "The essence of man is speech."
We are very much creatures of the moment. At any given time our
attention is captured by whatever our senses perceive. We cannot re-
member the past de arly, not even what the tomorrow of our persona! lives
will be like, despite the fact that sorne of us may have lived more than ten
thousand days on this earth, repeating the same pattern that many times .
With ali this experience we are still unable to see the immediate future
with adequate clarity. But sorne would try to venture predictions on seri-
ous issues of politics or economies in which they have had little expe-
rience.
To understand and deal with what is going on in the world, we need
to improve our recollection of events and the precision of our knowledge
by reviewing the facts and organizing them in a logical framework. If we
are to make decisions thal are rational and effective, we must participate
intensely in the act of understanding the world around us. It is an exagger-
ation to say thal humans are logical creatures. More accurately, our under-
standing is filtered through our senses, and our judgment relies on often
hazy impressions of reality. With experience and through the perceptions
and opinions of other people, our views of reality may change and become
more precise. For a better understanding of the world, we need to perse-
vere in thinking matters through carefully and to debate with others who
hold different views.
But the complexity of social systems cannot await a full, logical analy-
sis of situations on which our health, safety, and even survival depend. We
need to rethink the traditional use of logic to derive knowledge. We also
need to expand our analytic procedures to improve our understanding of
situations in which not only time and space but also human behavior pla ys
a fundamental role in determining the outcome.
The analytic hierarchy process enables decision makers to represent
the simultaneous interaction of many factors in complex, unstructured
situations. lt helps them to identify and set priorities on the basis of their
objectives and their knowledge and experience of each problem. As we
have seen, our feelings and intuitive judgments are probably more rep-
resentative of our thinking and behavior than are our verbalizations of
them. The new framework organizes feelings and intuitive judgments as
weil as logic so that we can map out complex situations as we perceive
them. lt reflects the simple, intuitive way we actually deal with problems,
KEY CONCEPTS 13
but it improves and streamlines the process by providing a structured
approach to decision making.
KEY CONCEPTS
0 In our complex world system, we are forced to cope with more prob
lems than we have the resources to handle.
0 What we need is not a more complicated way of thinking but a frame
work thal will enable us to think of complex problems in a simple way.
0 There are two fundamental approaches to solving problems: the de
ductive approach and the systems approach. Basically, the deductive
approach focuses on the parts whereas the systems approach concen-
trates on the workings of the whole. The analytic hierarchy process,
the approach proposed in this book, combines these two approaches
into one integrated, logical framework.
0 Humans are not often logical creatures. Most of the time we base our
judgments on hazy impressions of reality and then use logic to defend
our conclusions.
0 The analytic hierarchy process organizes feelings and intuition and
logic in a structured approach to decision making.
co
'-1
i
!
l
2
The Analytic
Hierarchy
Process
. - - - - - - --
This chapter deals with the following questions:
What are the three basic principles of logical
analysis?
How do these three principles relate to a new
approach to decision making-the analytic .
hierarchy process?
What can we do when the usual scales of
measurement-dollars, time, tons, and so
forth-fail to measure intangible qualities?
Why is the analytic hierarchy process such a
powerful method for tackling complex political
and socioeconomic problems?
What can you expect to gain by using the
analytic hierarchy process?
14
WHAT IT IS AND HOW IT WORKS 15
WHAT IT IS AND HOW 1T WORKS
To introduce the analytic hierarchy process, consider the following
example of a decision problem. The Brandywine River Region in Pennsyl-
vania faces possible urbanization and its environmental effects. What
actions should the people of the region take to maintain environmental
quality? Should they allow development and invest money to prevent
environmental deterioration or should they limit development?
Planners who used the AHP to study this problem first defined the
situation carefully, including as many relevant details as possible. Then
they structured it into a hierarchy of levels of deil (Figure 2-1) . The
highest leve! was the overall objective of protecting environmental quality.
The lowest included the final actions, or alternative plans, thal would con -
tribute positively or negatively to the main objective through their impact
on the intermediate criteria. The alternatives were (A) to leave the area
nonurbanized, (B) to allow partial urbanization, and (C) to allow total
urbanization. The intermediate levels of the hierarchy comprised the Iwo
basic criteria for evaluating environmental quality: (1) esthetic criteria,
which were further structured into properties of vividness, intactness, and
Environmental Quality
Esthetics Hydrology
~ ~
Vividness
~
No Noise ,
No Disturbances
lntactness
No
Flooding
Water
Quality
~
~ ~ ~
Nonurbanized
(A)
Partly
Urbanized
(B)
Totally
Urbanized
(C)
Channel
Naturalness
~
Figure 2-1 Hierarchy for Brandywine River Regiof1
(X)
(X)
THE ANAL YTIC lliERARCHY PROCESS
no noise or disturbances; and (2) hydrologie criteria, subdivided into no
flooding, water quality, and channel naturalness. This hierarchy graphi-
cally depicts the interdependence of elements in the problem; it both iso-
laies the relevant factors and displays them in the larger context of their
relationship to each other and to the system as a whole.
After developing the hierarchy, the planners judged the relative im-
portance of ali the elements. They quantified these judgments by assigning
them numbers from 1 to 9-and sometimes they disagreed. On many
major issues where an impasse in the judgment of different people occurs,
careful assessment of the differences in the intensity with which these
people defend their preferences and opinions is necessary. Often words
alone or logical argument cannot express the subtleties of deeply felt differ-
ences. But these differences can be measured by numbers, as we will see
later on. After debate and compromise, the planners determined priorities
for the elements of the hierarchy. Through a sequential process the judg-
ments were synthesized and the desirability of each of the three alternative
plans was estimated mathematically. The plan with the highest numerical
value, and therefore priority (in this case, plan B), was the obvious best
choice.
Judgments on the relative importance of each element in the hierarchy
were made by people who were knowledgeable about the Brandywine
River Region and about problems of urbanization and environmental
quality. Yet even experts can make mistakes in setting up a hierarchy or
discriminating between pairs of elements to judge priorities. The AHP also
tests the consistency of judgments; too gre at a departure from the perfectly
consistent value indica tes a need to improve the judgments or to restruc-
ture the hi erarchy. Suppose we take a doser look at the question of consis-
tency.
The consistency is perfect if ali the judgments relate to each other in a
perfect way . lf y ou say thal you pre fer spring three times more to summer
and thal you prefer summer twice more to winter, then when you give the
judgment comparing your preference of spring to winter it should be 6 and
not anythi ng else. The greater your deviation from 6, the greater your
inconsistency. This observation applies to relations among ali the judg-
ments given . We would have perfect consistency, then, if ali the relations
checked out As we will see, there is a rather simple way of
verifyi ng inconsistency and how much it deviates from perfect consis-
tency. There is also a good way for interpreting what inconsistency means
in practical terms . When we are revising judgments, this method is useful
and necessa ry .
;; This a pproach to the Brandywine River problem illustrates the basic
(1 principles of the analytic hi erarchy process . Now let us take a doser look at
\i these principles .
!
PRINCIPLES OF ANAL YTIC THIN KING 17
PRINCIPLES OF ANALYTIC THINKING
In solving problems by explicit logical analysis, three principles can be
distinguished: the principle of constructing hierarchies, the principle of
establishing priorities, and the principle of logical consistency. As sug-
gested in the Brandywine River example, these natural principles of analy-
tic thought underlie the AHP.
Structuring Hierarchies
Humans have the ability to perceive things and ideas, to identify
them, and to communicate what they observe. For detailed knowledge our
minds structure complex reality into its constituent parts, and these in tum
into their parts, and so on hierarchically. The number of parts usually
ranges between five and nine. In the Brandywine River Region study, the
idea of environmental quality was structured into six elements: vividness,
intactness, no noise or disturbance, no flooding, water quality, and chan-
nel naturalness. By breaking down reality into homogeneous dusters and
subdividing these clusters into smaller ones, we can integrale large
amounts of information into the structure of a problem and form a more
complete picture of the whole system. (This process is explored further in
the next chapter.)
Setting Priorities
Humans also have the ability to perceive relationships among the
things they observe, to compare pairs of similar things against certain
criteria, and to discriminate between both members of a pair by judging
the intensity of their preference for one over the other. Then they synthe-
size their judgments- through imagination or, with the AHP, through a
new logical process- and gain a better understanding of the whole system.
In the Brandywine River Region study, the planners established rela-
tionships between the elements of each leve! of the hierarchy by compar-
ing the elements in pairs. These relationships represent the relative impact
of the elements of a given leve! on each element of the next higher leve!. ln
this context the latter element serves as a criterion and is called a pro pert y .
The result of this discrimination process is a uector of priority , or of relative
importance, of the elements with respect to each property. This pairwise
comparisoh is repeated for ail the elements in each leve!. The final step is to
come down the hierarchy by weighing each vector by the priority of its co
property. This synthesis results in a set of net priority weights for the
bottom leve!. The element with the h ighest weight-plan B (part ial ur-
[,
- . -
THE ANAL YTI C HI ERARCHY PROCESS
banization) in our example- is the one that merits the most serious con-
sideration for action, although the others are not ruled out entirely. This
principle and the next are fully explained in Chapter S.
Logical Consistency
The third principle of analytic thought is logical consistency. Humans
have the ability to establish relat ionships among objects or ideas in such a
way that they are coherent-thal is, they relate well to each other and their
relations exhibit consistency. Consistency means two things. The first is
that similar ideas or objects are grouped according to homogeneity and
relevance. For example, a grape and amarble can be grouped into a homo-
geneous set if roundness is the relevant criterion but not if flavor _is the
criterion. The second meaning of consistency is that the intensities of rela-
tions among ideas or objects based on a particular criterion justify each
other in sorne logical way. Thus if sweetness is the criterion and honey is
judged to be five times sweeter than sugar, and sugar twice as sweet as
molasses, then honey should be taken to be ten times sweeter than molas-
ses. If honey is judged to be only four times sweeter than molasses, then
the judgments are inconsistent and the process may have to be repeated if
more accurate judgments could be obtained .
ln utilizing these principles, the anal ytic hierarchy process incorpo-
rates both the qualitative and the quantitative aspects of human thought :
the qualitative to define the problem and its hierarchy and the quantitative
to express judgments and preferences concisely. The process itself is de-
signed to integrale these dual properties. lt clearly shows thal for better
decision making the quantitative is basic to making decisions in
complex situations where it is necessary to determine priorities and make
tradeoffs. To calculate priorities, we need a practical method of generating
scales for measurement.
MEASUREMENT
People are generally wary, if not distrustful, when numbers are intro-
duced into the traditional process of decision making. But appropriately
chosen numbers can represent variat ions in feelings more faithfully than
can words or rhetoric. In the face of complexity, we run out of words to
express adequately our full awareness of what we sense to be taking place.
Words limit the perspectives of our feelings.
Numbers are used in many different ways in our civilization to mea-
sure ali kinds of physi cal experience. We find this application acceptable.
The question is whether we can extend and justify the use of numbers in
sorne reasonable , easily underst ood way to reflect our feelings on various
social, economie, and political matters . We need to look into whether
MEASUREMENT 19
numbers are si mply artifacts that gi ve us the illusion of greater precision
than we are capable of feeling, or whether we are missing a great deal by
not realizing that numbers are a creation of our minds to reflect feelings
and distinctions. Perhaps we have not yet recognized and appreciated
their value in solving complex, unstructured problems.
In a moment we will briefly consider how numbers have come to be
used in our lives to measure our perceptions of physical stimul i. Later we
will see that numbers can also be used to reflect accurately our subj ective
judgments and their intensity; they can be used to distinguish among
intangible as well as physical stimuli. Chapter 5 describes a simple way to
use numbers to synthesize outcomes that faithfully represent our intuitive
feeling and understandi ng of what we perceive the outcomes to be. The
advantage is that finer shades of differences in judgment can be identified
for their effect on the outcome and that we can accommodate different
opinions in the decision-making framework .
Evolution of Scales
Our highly organized civili zation depends on scales to measure such
qualities as time, length. temperature, and money. Such measures were
not handed down through the burning bush but evolved hi storically.
Time. Time is a fundamental quality of nature; ils measurement is at the
foundation of science. The Sumerians were the first to divide the year and
the day into units . Their year contained twelve months; and each montlt,
thirty days. Egyptian priests divided the year into 365 days . The full sun-
light period of a day was divided into ten hours, corresponding to ten
fi,ngers; dawn and dusk were each allotted one hour, bringing the total to
twelve. The night was also allotted twelve hours, making twenty-four
hours in a day. The daylight hours were marked by shadow docks, precur-
sors of sundials, and the night hours were marked by the appearance of
stars. Thus neither daylight nor night hours were uniform in length; they
depended on the seasonal transit of the sun and the rising of the stars.
The real beginning of time measurement was the depsydra, a water
dock that measured time by the emptying and filling of a vessel. As the
depsydra became more corn mon, so too did the notion of lime as a thing in
itself, a flowing reality measured independently of the heavens . ln the
tenth century Arab scientists developed an improved sundial thal marked
off the hours accu ratel y year round and was the first use of fixed lime units.
In the thirteenth century mechanical docks measured lime by uniform 00
periodic motions . Europeans developed precision in meas uring lime dur- ;......
ing the sev{mteenth and eighteenth centuries. In October 1960 the 0
Eleventh General Conference on Weights and Measures, in Paris,:
THE ANAL YTI C HIERARCIIY PROCESS
refined chronological precision by defining the second as the duration of
9,192,631,770 cycles of the radiation associa led wi th a specified transi lion,
or change in energy leve!, of the cesium atom.
Length. The Babylonians, Egyptians, Greeks, and Chinese ali had their
own un ils and subunits of length and other physical qualities such as area,
weight, and liquid volume. In the thirteenth century the English i n ~ r o
duced a standard yard and divided it into three feet of twelve inches each.
They also defined a rodas five and a half yards and a furlong as one-eighth
of a mile. In 1878 the yard was redefined as "the straight line or distance
betwe:en the centers of two gold plugs or pins in the bronze bar ...
measured when the bar is at the temperature of sixty-two degrees of
Fahrenheit's thermometer, and when it is supported by bronze rollers
placed und er it in such a manner as best to a void flexure of the bar."
The French Revolution contributed the metric system. In 1792 Louis
XVI issued a proclamation directing two engineers to determine the length
of the meler. They set out to measure the distance on the meridian from
Barcelona, Spain, to Dunkirk, France, but civil war intervened and the task
took years to complete. A provisional meler was established in 1795 and
was adopted by the French Assembly four years later: The meler was "one
ten-millionth part of a meridional quadrant of the earth." Again in Oc-
lober 1960 the meler was redefined as 1,650,753.73 wavelengths in a vac-
uum of the orange-red line of the spectrum of krypton-86.
Temperature. The mercury thermometer was invented by the German
physicist Gabriel Daniel Fahrenheit in the eighteenth century. He consid-
ered body temperature to be 100 on his scale (later found to be 98.6, as he
erred in determining his fixed points) and the temperature of the coldest
thing he could produce in his laboratory- a mixture of salt and ice--to be
0. The freezing and boiling points of pure water at sea leve! had to be 32
and 212, respectively. Although the thermometer has been improved,
Fahrenheit's scale survives unchanged. It is sometimes preferred to the
Celsius scale, devised by the Swedish astronomer Anders Celsius in 1742,
because it has smaller subdivisions. For highly accurate temperature read-
ings, Lord Kelvin (William Thompson) first suggested the use of the gas
thermometer in the nineteenth century. AbS()lute zero is equal to -273
Celsius, the temperature at which ali atomic activity stops.
Money. Monetary standards differ from the other units of measure in
thal they are neither uniform nor consistent. The monetary value we attrib-
u te to goods and services fluctuates according to supply and demand and
perce ived desirability or utilit y. The evolution of money is closely inter-
woven with thal of ci vili zation . In sorne cases, as in ancient Egypt, the
preva iling monetary sys tem evolved from the political, social, and eco-
MEASUREMENT 21
nomic systems, whereas in ancien! Lydia these institutions followed the
progress of monetary evolution. Although goods and services can be ex-
changed directly, the adoption of a monetary system simplifies trade and
promotes higher productivity.
The Need for a New Scaie
Just as we can distinguish and measure physical relationships-
meters for length, for example, and seconds for time--we are capabl e of
doing the same with abstract relationships. We have the capacity for a
range of feeling and discrimination thal permits us to develop relat ion-
ships among the elements of a problem and to determine which element s
have the greatest impact on the desired solution. In dealing with concret e
matters, such as repairing a car, we perceive the varying intensity of im-
pact through our senses, by hearing a faulty motor or seeing a leak, or
through their refinement by scientific instruments such as a voltmeter or
pressure gauge. We carry out this process of measuring the priorities of
impacts in order to solve problems.
Soto determine the intensity of impact of the various components of a
system, we must perform sorne type of measurement on a scale with units
such as pounds, seconds, miles, . and dollars . But these sc ales li mit the
nature of ideas we can deal with. Social, political , and other qualitative
factors can in no reasonable way be assessed in terms of physical or eco-
nomie measurement . What then can we do?
We can devise a scale thal enables us to measure intangible qualities,
just as scales evolved for measuring physical qualities . Chapter 5 presents
such a scale to measure priority impacts in unstructured systems . This new
way of assessing intangible qualities should hold up in areas where we
already know the unit of measurement, which can then be used to validate
the method. And in fact examples show that this approach to measuring
priorities can be used to generale results conforming to classic ratio scale
measurement in physics, economies, and other fields where standard mea-
sures already exist.
To measure priorities, we compare one element with another . The old
adage thal one cannot compare apples and oranges is false . Apples and
oranges have many properties in common: size, shape, !aste, aroma, color,
seediness, juiciness, and so on. We may prefer an orange for sorne proper-
ties and an apple for others; moreover, the strength of our preference may
vary. We may be indifferent to size and color, but have a strong preference
for !aste,. which again may change with the time of day. It is my thesis thal
this sort of complicated comparison occurs in real !ife over and over again,
and sorne kind of mathematical approach is required to help us determine
priorities and make tradeoffs. This approach is the analytic hierarchy
process:
(X)
........
........
j
j
l
,;
l'
.
THE ANi\l.YTIC IIIFRARCIIY PROCESS
AHP: A FLEXIBLE MODEL FOR DECISION MAKING
These basic observations on human nature, analytic thinking, and
measurement have led to the development of a useful mode! for solving
problems quantitatively. The analytic hierarchy process is a flexible mode!
that allows individuals or groups to shape ideas and define problems by
making their own assumptions and deriving the desired solution from
them. lt also enables people to test the sensitivity of the solution, or out-
come, to changes in information. Designed to accommodate our human
nature rather than force us into a mode of thinking that may violate our
better judgment, the AHP is a powerful process for tackling complex poli ti-
cal and socioeconomic problems.
The AHP incorpora tes judgments and persona! values in a logical way.
lt depends on imagination, experience, and knowledge to structure the
hierarchy of a problem and on logic, intuition., and experience to provide
judgments. Once accepted and followed, the AHP shows us how to con-
nect elements of one part of the problem with those of another to obtain the
combined outcome. lt is a process for identifying, understanding, and
assessing the interactions of a system as a whole.
To define a complex problem and to develop sound judgments, the
AHP must be progressively repeated, or iterated, over time; one can hardly
expect instant solutions to complicated problems with which one has wres-
tled for a long time. The AHP is flexible enough to allow revision-
decision makers can both expand the elements of a problem hierarchy and
change their judgments. lt also permits them to investigate the sensitivity
of the outcome to whatever kinds of change may be anticipated. Each
iteration of the AHP is like hypothesis making and testing; the progressive
refinement of hypotheses leads to a better understanding of the system.
The many practical applications of the AHP have generated sample hierar-
chies, sorne of which are presented in Chapter 4. With mi nor modification,
sorne of thse paradigms can be used to structure new problems.
Another feature of the AHP is that it provides a framework for group
participation in decision making or problem solving. We have seen that
ideas and judgments can be questioned and strengthened or weakened by
evidence that other people present. The way to shape unstructured reality
is through participation, bargaining, and compromise. Indeed, the con-
ceptualization of any problem by the analytic hierarchy process requires
one to consider ideas, judgments, and facts accepted by others as essential
aspects of the problem. Group participation can contribute to the overall
validity of the outcome, although perhaps not t the ease of implementa-
tion if the views diverge widely. Th us one could include in the process any
information derived scientifically or intuitively.
The process can be applied to real problems and is particularly useful
for - " ""Ca ting resources, planning, analyzing the impact of policy, and

AHP: A FLEXIBLE MODEL FOR DECISION MAKING
Unlty:
The AHP provides a single.
easily understood, flexible mode!
for a wide range of unstructured
problems
Process Repetition:
The AHP enables people to re fine
their definition of a problem
and to improve judgment
and understanding through
repetition
Judgment and Consensus:
Complexlty:
The AHP integrales deductive
and systems approaches
in solving complex
problems
23
on consensus but synthe- The AHP can deal w1th the
The AHP does not insist \ 1 Interdependance:
sizes a representative 1nterdependence of elements
outcome from d1verse ......______ 1n a system and does not
judgments ------------e --------- tnSISt on hnear th1nk1ng
AHP
Tradeoffs: \....___________ Hierarchie Structurlng:
The AHP takes into The AHP rellects the
consideration the natural tendency of the
relative priorities of mind to sort elements of
factors in a system and a system into different
enables people to select levels and to group like
the best alternative elements in each ievel
based on the ir goals
Synthesls: Measurement:
The AHP leads to
an overall estima te of the
desirability of each
alternative
The AHP provides a scale
for measuring intangibles
and a method lor
establishing priorities
Conslstency:
The AHP tracks the logical
consistency of judgments used
in determining priorities
Figure 2-2 Advantages of the Analytic Hierarchy Process
j
1
1

'
j

!
.j
:1
1


i
:0

J
.; ,
co i:
. :
t-" ,.
N
1
1
- .. .... -- - - - - - -- - .. --. - - - .
1 ANALYTIC lllERARCHY PROCESS
resolving conflicts. Social and physical scientists, engineers, policymakers,
and eyen laypersons can use the method without intervention by so-called
experts; those who have a problem are ordinarily best informed about that
particular problem. Currently the AHP is being widely used in corporate
planning, portfolio selection, and benefit/cost analysis by government
agencies for resource allocation purposes . And it is being used more
widely on an international scale for planning infrastructure in developing
countries and for evalating natural resources for investment. Figure 2-2
summarizes the advantages of using the AHP as a new approach to prob-
lem solving and decision making.
PERSPECTIVE
The analytic hierarchy process reflects the way we naturally behave
and think. But it improves upon nature by accelerating our thought pro-
cesses and broadening our consciousness to include more factors than we
would ordinarily consider. The AHP is a process of "systemic rationality":
lt enables us to consider a problem as a whole and to study the simulta-
neous interaction of its components within a hierarchy. This process is
rather sophisticated. If we observed an individual, such as a child, who
does not have to deal with complex issues or do severa! tasks at the same
time, we most likely would not note a talent for hierarchie organization.
Over long periods of time, however, the mind learns shortcuts and groups
activities into clusters. This mental process gradually develops into a style
for looking at the world and organizing it in such a way that we can deal
with it efficiently.
Sorne people have gone so far as to point out that nature itself or-
ganizes matter and !ife hierarchically. But it is difficult to separate this
observation from the fact that to understand the complexity of nature, it is
we who have to sort and arrange what we perceive and see hierarchically.
Moreover, the interactions among elements of a hierarchy do not always
have an inherent structure apart from what we can observe. We must
identify and synthesize these interactions based on our objectives aild the
knowledge and experience we have of each problem.
The AHP addresses complex problems on their own terms of interac-
tion . It allows people to lay out a problem as they see it in ils complexity
and to refine ils definition and structure through iteration. To identify
criti cal problems, to define their structure, and to locale and resolve con-
fli cts, the AHP calls for informati on and judgments from severa! partici -
pants in the process. Through a mathematical sequence it synthesizes their
judgments into an overall est imate of the relative priorit ies of alternative
courses of acti on. The pri orities yielded by the AHP are the basic units
used i n ali types of analysis; for example, they can serve as guidelines for
all ocati ng resources or as probabilities in making predictions.
KEY CONCEPTS 25
The AHP can be used to stimulate ideas for creative courses of action
and to evaluate their effectiveness. lt helps leaders determine what infor-
mation is worth acquiring to evaluate the impact of relevant factors in
complex situations. And it tracks inconsistencies in the participants'
judgments and preferences, thereby enabling leaders to assess the quality
of their assistants' knowledge and the stability of the solution .
The next three chapters on this new approach to decision
making. Here is what one can expect to gain by using it:
1. A practical way to deal quantitatively with different kinds of functi onal
relations in a complex network.
2. A powerful tool for integrating forward (projected) and backward (de-
sired) planning in an interactive manner that reflects the judgment s of
ali relevant managerial personnel. The output of this process is explicit
rules for allocating resources among current and new st rategy
offerings--or to satisfy a specifie set of corporate objectives-or under
alternative environmental scenarios .
3. A new way to:
Integrale hard data with subjective judgments about intangible fac-
tors.
lncorporate judgments of severa! people and resolve conflicts among
them.
Perform sensitivity analysis and revision at low cost .
Use marginal as weil as average pri orities to guide allocation.
Enhance the capacity of management to make tradeoffs explicitly.
4. A technique complementing other on es (benefitlcost, priority, risk
minimization) for selecting projects or activities.
5. A single replacement for a variety of schemes for projecting the future
and protecting against risk and uncertainty.
6. A vehicle for monitoring and guiding organizational performance ta -
ward a dynamic set of goals.
KEY CONCEPTS
0 There are three basic prin ci pies of the analytic hierarchy process:
1. Hierarchie representation and decomposition, whi ch we cali h ierar-
chie structuring-that is, breaking clown the problem int o separa te
elements.
2. Priority discrimination and synthesis, which we cali pri orit y
setting-that is, ranking the element s by relative importance.
(X)
1-'"
w
,;
~ !
- . ----- -------....
TIIE ANAL YTIC HIERARCHY PROCESS
3. Logical consistency-that is, ensuring that elements are grouped
logically and ranked consistently according to a logical criterion.
0 We cannot measure without a scale, but traditional scales such as ti me
and money limit the nature of ideas we can deal with. Thus we need a
new scale for measuring intangible qualities.
0 The analytic hierarchy process is a flexible mode! that allows us to
make decisions by combining judgment and persona! values in a logi-
ca! way.
3
Analyzing and
Structuring
Hierarchies
This chapter deals with the following questions :
Why are hierarchies fundamental to human
thinking?
How can they be used to understand complex
systems?
What is the difference between a structural
hierarchy and a functional hierarchy?
How do we go about constructing a
hierarchy?
77
_j
co
!---"
.p..
ANALYZING AND STRUCfURING IIIERARCHIES
HIERARCHIES: A TOOL OF THE MIND
Complex systems can best be understood by breaking them down into
the ir constituent elements, structuring the elements hierarchically, and
then composing, or synthesizing, judgments on the relative importance of
the elements at each leve! of the hierarchy into a set of overall priorities.
This chapter explains how to structure problems hierarchically.
Hierarchies are a fundamental tool of the human mind. They involve
identifying the elements of a problem, grouping the elements into homo-
geneous sets, and arranging these sets in different levels . The simples!
hierarchies are linear, rising or descending from one levet to another, such
as physics hierarchies that rise from atoms to molecules to compounds and
so on; the most complex are networks with interacting elements, such as
systems representing the learning process of a child.
CLASSIFYING HIERARCHIES
Hierarchies can be divided into two kinds: structural and functional.
ln structural hierarchies, complex systems are structured into their con-
stituent parts in descending order according to structural properties such
as size, shape, color, or age. A structural hierarchy of"the universe would
descend from galaxies to constellations to solar systems to planets, and so
on, down to atoms, nucleii, protons, and neutrons . Structural hierarchies
relate closely to the way our brains analyze complexity by breaking down
the abject s perceived by our senses into clusters, subclusters, and still
smaller clusters.
ln contras!, functional hierarchies decompose complex systems int o
their constituent parts according to their essential relationships. A conf!ict
over school busing to achieve integration can be structured into a cluster of
major st akeholders (majority and minority communities, city officiais,
board of education , federal government) ; a cluster of stakeholders' objec-
tives (education for children, retention of power, and the like) ; and alter-
native outcomes (complete, partial, or no busing) . Such functional hierar-
chies help people to steer a system toward a desired goal- like conflict
resolution, efficient performance, or overall happiness. For the purposes of
this book, functional hierarchies are the only ki nd thal need be considered.
Each set of elements in a functional hierarchy occupies a leve! of the
hierarchy. The top leve! , called the foCIIs , consists of only one element : the
broad, overall objective . Subsequent levels may each have severa! ele-
ments, although thei r number is usually small- between five and nine.
Beca use the element s in one leve! are to be compared with one another
against a crit er ion in the next higher leve!. the elements in each leve! must
be of the same order of magnitude. If the d isparity between them is great,
they should belong to different levels. For example, we cannat make a
~
CONSTRUCflNG HIERARCHIES 29
precise comparison between two jobs whose performances differ in diffi-
culty by a factor of 100 because our judgment would be subject to signifi-
cant error. lnstead, we first group simple jobs into a cluster and compare the
cluster with a job one arder of magnitude more difficult to perform than a
simple job. We then compare the jobs in the cluster among themselves
according to difficulty of performance. When we compare the results of the
two comparison processes, we obtain a net comparison of a simple job
with the more difficult one. To avoid making large errors, we must carry
out this process of clustering. By forming hierarchically arranged cl usters
of like elements-simple jobs in this case--we can efficiently compl e te the
process of comparing the simple with the very complex. Similarl y, to com-
pare small stones with boulders or atoms with stars, we must int etvene
between them severa! levels of abjects of slightly differe(1t magnitude to
make the transition and comparison possible.
Because a hierarchy represents a mode! of how the brain analyzes
complexity, the hierarchy must be flexible enough to deal with that com-
plexity. The levels of a hierarchy interconnect like layers of cell tissue to
form an organic whole that setves a certain function . A spiraling effect is
noticeable when we move from the focus expanding the hierarchy to the
leve! of simple elements. The expansion may be continued to the leve! of
elements of minutest concern. The next chapter illust rates the flexibility of
hierarchies with a variety of examples. Ali are functional hierarchies, but
sorne are complete-that is, ali the elements in one leve! share every prop-
erty in the next higher level- and SOrne are i11 C0 111 p /ete in thal Sorne ele-
ments in a leve! do not share properties.
CONSTRUCTING HIERARCHIES
No inviolable rule exists for constructing hi erarchies . The sample
hierarchies offered throughout the book are presented not to prescribe
certain frameworks but to stimulate thinking about what types of hierar-
chicallevels to choose and what kinds of elements to include in the levels.
The numbers of levels and elements may be more or less than those in the
examples.
The great variety of examples thal !end themselves to hierarchies sug-
gests thal the Sl.\,bjects thal can be approached with the an.,Jytic hierarchy
process are infinlt&. ln ali these areas we are limited only by our experi -
ences and feelings as represented by words from the di cti onary. Lan-
guages that are limited in vocabulary may pose problems of ambiguit y or
may no.t represent human experience adequately. Recogni zing these lim-
itations, we may be encouraged to innovate by creating the needed vocab-
ulary .and other symbols, such as computer languages, to represe nt feelings
and ideas that we may become aware of i n the process of identificati on and
structuring.
co
~
l1l
ANALYZING AND STRUCTURING HI ERARCHIES
One' s approach to constructing a hierarchy depends on the kind of
decision to be made. If it is a matter of choosing among alternatives, we
could start from the bottom leve! by listing the alternatives. The next leve!
would consist of the criteria for judging the alternatives. And the top leve!
would be a single element, the focus or overall purpose, in terms of which
the criteria can be compared according to the importance of their contri-
bution.
Suppose we wanted to decide whether to buy one of five sports cars
(Figure 3-1) . These alternatives form the bottom leve! of the hierarchy. The.
criteria in terms of which the alternatives will be judged form another leve!
and might include adequacy of salary, prestige, basic necessities, comfort,
satisfaction of other needs, large savings account, and freedom from
worry. The priorities of these criteria will be judged in terms of their
contribution to the focus of the hierarchy: our overall happiness.
Note that once we construct this hierarchy, it is not necessarily cast in
bronze. We can always alter parts of it later to accommodate new criteria
that we may think of or that we did not consider to be important when we
first designed it . The computer programs that assis! us in the task are
constructed with this flexibility in mind. After we rank the criteria and
arrive at overall priorities for the alternatives, we might still have sorne
doubts about the final decision. In that case we would simpl y go through
the process again and perhaps change sorne of our judgments on the rela-
tive importance of the criteria. If the same alternative is still significantly
ahead of the others in overall priority, we know that it is the right choice
for us .
Sometimes the criteria the mselves must be examined in detail, so a
leve! of subcriteria should be inserted between those of the criteria and the
alternatives. To select a school from three possible choices, for example,
severa! criteria might be used-such as educational , cultural, and social
Sa lary

Prestige
Good Choice for a Sports Car
Basic
Needs
Comfort
Satisfaction
ofOther
Needs
Large
Savings
Account

Freedom
from
Worry

Porsche Mercedes Triumph Dats un Corvette
Figure 3-1 Hierarchy for Choosing a Sports Car
CONSTRUCTING HIERARCHIES
Educational
Quality

T eachers \ \ "-
General

of Students \ '\. ...__

Preparation \ "-,

College \,

Environment '--.,_
School
A
Selecting a Best School
Cultural
!\

School
B

School
c
Figure 3-2 Hierarchy for Selecting a School
31
Social
J
(Figure 3-2) . The educational criterion might be broken down into subcri -
teria of (1) quality of teachers, (2) general standard of students, (3) disci -
pline, (4) preparation for college, and (5) learning environment ; the other
criteria could similarly be broken down as weil. In this case the subcri teria
would be .compared in terms of the criterion to which they belonged and
not to the other criteria. Such a hierarchy would then be called incomplete
because the subcriteria are not ali compared in terms of ali the criteria of
the next higher leve!.
There is no limit to the number of levels in a hierarchy. If one; ., unable
<Xl
t---<
0'1
ANAL YZING AND STRUCTURING HIERARCHIES
to compare the elements of a leve! in terms of the elements of the next
higher leve!, one must ask in what terms they can be compared and then
seek an intermediate leve! that should amount to a breakdown of the
elements of the next higher leve!. Th us a new leve! has been introduced to
facilitate the analysis for comparisons and to increase the precision of the
judgments. Now we can answer the main question: How much more does
one element contribute than another to satisfying a criterion in the next
higher leve! of the hierarchy?
Hierarchies may be more complicated than the ones described above.
Those related to projected planning and repeated later, for example, in-
elude the following levels:
Uncontrollable environmental constraints
Risk scenarios
Controllable systemic constraints
Overall objectives of the system
Stakeholders
Stakeholders' objectives (separate ones for each stakeholder)
Stakeholders' policies (separate ones for each stakeholder)
Exploratory scenarios (outcomes)
Composite or logical scenario (outcome)
But decision makers do not have to pursue a problem to the full leve) of
detail indicated here. The depth of detail depends on how much knowl-
edge one has about the problem and how much can be gained by using
that knowledge without unnecessarily tiring the mind.
PERSPECTIVE
Although hierarchies have been known for a very long ti me, the anal y-
tic hierarchy process makes it possible to generale new levels and arrange
them in a logical fashion so they relate to each other naturally. By making
paired comparisons of the elements in a level in terms of the elements of
the next higher level, it is possible to decide on an appropriate choice of
thal upper level. Moreover, when the elements of a level cannot be com-
pared except in terms of finer criteria than identified so far, a new leve!
must be created for this purpose. Thus the analytic aspects of the AHP
serve as a stimulus to create new dimensions for the hierarchy. It is a
process for inducing cognitive awareness . A logically constructed hierar-
chy is a by- product of the entire AHP approach. lndeed, experience indi-
cates that there are a few patterns that ali decision hierarchies seem to
follow. ln the next chapter we will see how this concept of hierarchies can
be appli ed to a broad range of real situations.
KEY CONCEPTS
33
KEY CONCEPTS
0 In a functional hierarchy, complex systems are broken down into their
constituent parts according to their essential relationships.
0 The top level of the hierarchy-the focus--consists of only one ele-
ment: the overall objective. The other levels contain severa! elements
(usually between five and nine) .
0 There is no limit to the number of levels in a hierarchy.
0 When the elements of a level cannot be compared readily, a new level
with finer distinctions must be created.
0 Hierarchies are flexible. We can always alter them to accommoda te
new criteria.
HOW TO STRUCTURE A HIERARCHY
The basic principle to follow in structuring a
hierarchy is to see if one can answer the question: "Can
you compare the elements in a lower leve! in terms of sorne
or all the elements in the next higher level?"
A useful way to proceed is to come down from the goal
as far as one can and then go up from the alternatives
until the levels of the two processes are linked in a way
to make comparison possible. Here are sorne suggestions.
1. Identify overall goal. What are you trying to
accomplish? What is the main question?
2. Identify subgoals of overall goal. If relevant,
i d e n ~ i f y time horizons that affect the decision.
3. Identify criteria that must be satisfied to fulfill
subgoals of the overall goal.
4. Identify subcriteria under each criterion. Note t hat
subcriteria may be intervals of numerical values; or
intensities such as high, medium, low; or excellent,
very good, good, average, poor, and very poor.
5. Identify in descending levels, as needed, actors, actor
objectives, and actor policies in this order.
6. Identify alternatives or outcomes.
7. For yes-no decisions include for example doing and not
doing the alternative.
8. It is often useful to construct two hierarchies, one
for benefits and one for costs to decide on the best
alternative, particularly in the case of yes-no
decisions. Ratios of benefits to costs, or better,
marginal benefits to costs, are formed and the
alternative with the largest ratio is chosen. Answer
the question: Which alternative yields the greatest ~
benefit (for benefits) or costs the most (for costs)7 ~
'J
1
~
r
t
!
1
l__
. -- . - .. -- ---- ..... ------------ -----------------------------
4
Practical
Examples of
Hierarchies
This chapter deals with the following questions :
What is the main purpose of arranging goals,
attributes, issues, and stakeholders in a
hierarchy?
How can hierarchies be applied to business
decisions--choosing equipment, deciding
whether to buy or lease, making financial
decisions, and so on?
How can hierarchies be applied to persona!
and domestic decisions--choosing a car,
choosing a career, purchasing a house, and so
on?
How can hierarchies be applied to public
policy decisions--choosing areas for R&D
programs, allocating resources in a juvenile
34
J
AN APPROACH TO HIERARCHIES
correction program, analyzing a school busing
conflict, and so on?
How can hierarchies be applied to planning
economie policies-for example, planning
economie strategy for an underdeveloped
country?
How can hierarchies be used for estimating
and predicting-predicting the presidential
election, estimating the popularity of a rock
group, and so on?
How can hierarchies be used for measuring
influences- for example, parental influence on a
child's psychological we!l-being?
How can hierarchies be used to represent
systems networks-the network of a child' s
leaming system, the network of a volleyball
team, and so on?
AN APPROACH TO HIERARCHIES
35
ln later chapters we will discuss in detail specifie applications of the
AHP to planning, conflict resolution, benefit/cost analysis, and resource
allocation. This chapter offers examples of different kinds of hierarchies to
suggest ways of approaching problems and structuring them into their
constituent parts with enough detail to make reasonable decisions .
Most prob1ems arise because we do not know the internai dynamics of
a system in enough detail to identify cause-and-effect relationships. If we
were able to do so, the problem could be reduced to OJ1e of soci al engineer -
ing, as we would know at what points in the system interventio n is neces-
sary to bring about the desired objective. The crucial contribution of the
analytic hierarchy process is that it enables us to make practical decisions
based on a "precausal" understanding-namely, on our feelings and
judgments about the relative impact of one variable on another.
ln sum, when constructing hierarchies one must include enough rele-
vant detail to depict the problem as thoroughly as possible . Consider thE
environment surrounding the problem. Jdentify the issues <' ribut e!
i .
~
i'i
l i ~
; t
1
1
[
f
1
i
~
' 1
!
l
\
~
l
;
il
:1
i ~
"
1
1
!j
" :1
i\
\
l
..
!,
.,,
l'
~
,
1
)
l
t
1
(X)
,_...
(X)
1
r
H
::
i=
li
; ,
; ,
.,
,,
,,
l '
i
. ... ---. ---------...... - - - - - - - - - - - - - - - - - - ~
PRACTICAL EXAMI'LES OF IIIERARCIIIES
that you feel contribute to the solution. ldentify the participants associated
with the problem. Arranging the goals, attributes, issues, and stakeholders
in a hierarchy serves two purposes: lt provides an overall view of the
complex relationships inherent in the situation, and it permits the decision
maker to assess whether he or she is comparing issues of the same arder of
magnitude in weight or impact on the solution.
The elements should be clustered into homogeneous groups of five to
nine so they can be meaningfully compared to elements in the next higher
levet. The only restriction on the hierarchie arrangement of elements is that
any element in one level must be capable of being related to sorne elements
in the next higher leve!, which serves as a criterion for assessing the rela-
tive impact of elements in the level below.
The hierarchy does not need to be complete; thal is, an element in a
given level does not have to function as a criterion for al/ the elements in
the leve! below. Thus a hierarchy can be divided into subhierarchies shar-
ing only a corn mon topmost element . Further, a decision maker can insert
or eliminate levels and elements as necessary to clarify the task of setting
priorities or to sharpen the focus on one or more parts of the system.
Elements that are of Jess immediate interest can he represented in general
terms at the higher levels of the hierarchy and elements critical to the
problem at hand can be developed in grea ter depth and specificity.
ln addition to identifying within a hierarchie structure the major fac-
tors that influence the outcome of a decision, we need a way to decide
whether these factors have equal effects on the outcome or whether sorne of
them are dominant and others so insignificant they can be ignored. This .is
accomplished through the process of priority setting. The task of setting
priorities requires that the criteria, the subcriteria, the properties or fea-
tures of the alternatives being compared, and the alternatives themselves
are gradually layered in the hierarchy so. thal the elements in each level are
comparable among themselves in relation to the elements of the next
higher leve!. Now the priori lies are set for the elements in each level
severa! times--<Jnce with respect to each criterion of the upper level. These
in tum are prioritized with respect to the elements of the next higher level
and so on. Finally a weighting process is used to obtain overall priorities.
This is done by coming down the hierarchy and weighting the priorities
measured in a level with respect to a criterion in the next higher Ievel with
the weight of that criterion. The weighted priorities can then be added for
each element in the leve! to obtain ils overall priority.
Finally, a ft er judgments have been made on the impact of ali the
elements, and priorities have been computed for the hierarchy as a whole,
the Jess important elements can be dropped from further consideration
because of the ir relatively small impact on the overall objective. Now let us
exa mine sorne of the real-world applications of the hierarchy.
:"!
BUSINESS DECISIONS 37
BUSINESS DECISIONS
Hierarchy for Choosing Urethane Equipment
To decide on the purchase of urethane manufacturing equipment
(Figure 4-1), three principal considerations are priee, technical features,
and service. "Technical features" is really a cluster of severa! specifie de-
sired qualities to which it is further decomposed. The overall priorities
obtained through successive priority setting represent the relative impor-
tance of the considerations bearing upon the purchase decision.
The three products under consideration are then ranked with respect
to each criterion or subcriterion. The overall priority gives the relative
superiority of the brands.
Hierarchy for Choosing Word Processing Equipment
For the selection of word processing equipment for an office (Fi gure
4-2), the benefit and cost hierarchies are considered separately and
benefit/cost ratios are obtained. The qualities desired of a word processing
machine form the second leve! in the benefit hierarchy, and thei r pri or it ics
represent the relative weights assigned by the user. The eq ui pment chara
Level1 :
Focus
Level2:
Criteria
Priee
Level3:
Subcriteria
Leve14:
Alternatives
Purchase of Urethane Machinery
Technic_al Features
~
Ratio
Control
Recirculat ion
Control
Temperature
Control
~ ~ ~
Brand X Brand Y Brand Z
Figure 4-1 Hierarchy for Choosing Urethane Equipment
Servce
1
~
rt.;
}
" )
. I
1
;1
t
{
t
' l
1
!
1
' 1
'
li
! i
,.
1
1
(
1
1
1
1
1
~ i l
~ I l
1
!
. .. - ....... ...-----------------
Level1 :
Foc us
Level2:
Criteria
Level3:
Features
Level4:
Alternatives
Level1 :
Focus
Level2:
Criteria
Level3:
Alternatives
PRACTICAL EXAMPLES OF HIERARCHIES
Ti me
Saving
Filing
Benefits
Ouality of
Document
Accuracy

Training Screen Service Space
Required Capability Ouality Required
Printer
Speed

l'

Lanier Syntrex Qyx
Costs
Capital Supplies Service Training

Lanier Syntrex Oyx
Figure 4-2 Hierarchies for Choosing Word Processing Equipment
teristics that contribute to these qualities fall in the third level. They are
ranked with respect to each de si red quality, and the overall priority repre-
sents the relative importance ofeach characteristic. The overall priorities of
the brands under consideration represent the relative superiority regard-
ing benefits expected from each.
In the cost hierarchy, likewise, the overall priorities represent the rela-
tive weights of the costs . The benefit/cost ratio for each indicates the rela-
tive superiority of one machine over the others.
"
BUSINESS DECISIONS 39
Hierarchy for Deciding on Buying or Leasing
The decision regarding company ownership or leasing of a piece of
capital equipment (Figure 4-3) depends on its contribution to the com-
pany's profitability. This profitability has two dimensions : economie and
intangible. The benefits depend on a number of factors that, in turn,
depend on certain characteristics of the company. Buying or leasing would
promote characteristics to a varying extent .
By setting priorities for the factors at a certain leve! with respect to the
relevant factors at the previous leve! and finding the composite priorities,
we can find to what extent, relatively speaking, the factors in the same
leve! contribute to the firm's overall profitability. Extending this logic to
the question of company ownership or leasing, we can say, in the judg-
ment of the decision maker, which alternative is preferable.
In this example we take the intangible benefits explicitly into consid-
eration for a decision , so the subjective judgments of the decision maker
Level1 :
Focus
Level2:
Bene lits
Level3:
Factors
Level4:
Company
Status
Level5:
Alternatives
Profitabrlity
EHicient
Allocation
ofFunds
Economie Benefits
lnerease lnerease
Current Total
and Quick Re sources
Ratios ol Company
Intangible Benefits

EHicient
Allocation
of Personnel
and Time
Economie
Security
lmprove
Company
Image

Financial
Stalus of
Company

Availability
of Working
Capital
r-
Company Ownership
Centralizalion
or Tax Advantagas
Decentralrzatron

Leasing
Figure 4-3 Hierarchy for Deciding on Buying or Leasing
co
N
0
Level1:
Focus
Level2:
Education
and Skill
Level3:
Abilities
Leve14:
Alternatives
PRACTICAL EXAMI'LES OF IIIERARCIIIES
Education
Select Manager
Management
Skills
Technical
Skills
Persona!
Skills


L . Probl em Job Organize Respon Decision J d
1
eadershp Solving Knowledge and Plan sibilily Making u gmen


Candidate 1 Candidate 2 Candodale 3
Figure 4-4 Hierarchy for Choosing a Management Candidate
are also considered. This is unlike a conventional exercise where only the
hard economie data are considered and then managerial judgment is used
in a qualifying manner at the end.
Hierarchy for Choosing a Candidate for Management
To select the proper person to fill a management position, we first
identify the four areas of evaluation and then the specifie traits to which
these areas contribute (Figure 4-4). Successive stages of priority setting
give the overall priorities for the relevant traits in leve) 3, which represent
the relative weights believed to be associated with them for a person to
function effectively in that position.
Once ali the candidates have been reviewed and the choice has been
narrowed clown to a few seemingly equal candidates, they may be ranked
with respect to each trait . The composite priority for each candidate then
represents his or her relative superiority on the basis of overall judgment
and is useful in comparing candidates and ranking them in order of pref-
erence .
Hierarchy for Choosing a Beverage Container
To evaluate the desirability of different containers to be used by the
soft drink beverage industry (Figure 4-5), we first consider the criteria for
0) .
;:
BUSINESS DECISIONS 41
evaluation and then rank them according to their relative importance on
the final outcome. Next we judge the alternative containers with respect to
each criterion per unit of beverage delivered. This prioritization shows the
desirability from the point of view of each criterion, and the composite
priorities show the overall superiority of the containers in relative terms.
Hierarchy for Staggering lndustry Hours for Energy
Conservation
To decide how to stagger industry work hours, we firs t examine the
relevant consequences of staggered work hours and judge ho w importan t
they are with respect to one another. This is clone by ranking the re levan t
criteria with respect to the focus (Figure 4-6).
Next the various shift patterns under consideration are ranked with
respect to each of the criteria above to see how much they wou id affect it in
relative terms. The composite priorit ies indicate the overall desirability of
the considerations in relative terms. The highest-priority shift is the most
desirable decision .
Hierarchy for Choosing a Site for Combustion Turbines
The problem of site selection for an electric utility company' was nar-
rowed to four alternatives after preliminary screening (Figure 4-7). The
company identified nine relevant factors they would have to consider in
the selection process. Since many of the considerations were in conflict,
they were first ranked to find their relative importance bearing upon the
Level1 :
Focus
Level2:
Criteria
Level3:
Alternatives
Energy to
Produce
Best Bever age Container
Cosl
Environmental
Wasl e
Customer
Service

Glass Bimelall ic Aluminum Steel
Figure 4- 5 Hierarchy for Choosing a Beverage Container
(X)
N

1
1
1
1
- -- --------------------------------------------
.. .
;
' 1
;
'
i
i
;
Level1 :
Focus
Level2:
Criteria
Level3:
Alternatives
PRACTICAL EXAMPLES OF HIERARCHIES
Work Staggering Goats
Energy Economie Employment Transportation Social
Conservation Consequences Consequences Consequences Consequences

Shiftof Shiftot
Shoft of Shiftof 80% 20%
Historical
2 Hours 8Hours Production Production
Pattern
Earlier Earlier Shifted Shifted
Earlier Earlier
Figure 4-6 Hierarchy for Staggering Work Hours
!:: decision. Next the sites were ranked with respect to each factor and com-
posite priorities were found . These overall priorities indicate the relative
merits of selecting the sites in question when ali the factors are taken into
consideration.
i
!.
:J:,
':\;1
{
'i
i
;l "
: ;:1
. ';
Hierarchy for Allocating Resources Among R&D Projects
in a B-ank
This decision regarding resource allocation is done as a benefit/cost
exercise involving the benefits expected to accrue from the projects and the
costs expected to be increased thereby (Figure 4-8) .
In the benefits hierarchy, the benefits are ranked according to their
impact on the bank' s performance. The projects are ranked according to
how far they can generale that benefit. The composite priorities represent
their overall benefit contributions on a ratio scale. In the costs hierarchy,
Iikewise, the various costs are ranked by their severity and the projects are
ranked with respect to their contribution to that cos!. The resulting com-
posite priorities represent their overall costs. The benefit/cost ratios mea-
sure the superiority of benefit for cost incurred and also each R&D project's
expected attractiveness . Ratios comparing the greatest marginal benefits to
costs are often more useful than simple benefitlcost ratios. Sometimes dis-
counting of benefits and costs is more realistically done before forming
such

r:
BUSINESS DECISIONS
43
ii

ill VI
E a
c-

e
su..
c
w
<ii
C VI
0
.;:; 0


iiiu..
...
E !'!
i
u

(
0 0
c 13
8 ..
w"-

ill
t-
u ill
Vi
E
g
<Il
M
c::
ill E :
(( 0 !'!

u i
..

ill .

i'i
'"VI
...
J!l
. 0 tiC

.s
0 ..
(ii"-
1/)
0
0
.c:
c

u
0 VI VI
-g
..

"'
0
.....
VI VI !'!
c c-o
>.

i .c:

...

8
..
<Il

:
:>.0
a. c t--
0 iii
O. a
'

c
0
t-
-
::1
.;: r./) !'!
tiC
'l!
i iZ
r=


t=
>- ,

e
-o
>-
I
,
VI
Cl>
co

-VI N
. .
<Il;:) -o
(")<tl
N
g
<Il- Q;E
> 0
N
...JU.
<Il <tl
><Il
...JU.
Q)=
...J<(
-
:!
1
l
d

H
!1
:1
li
,.

1
1
I
l '
,,
,, :
i
. ' 1
t' ;
.
l l
. --- -------------------------------....
Level1 :
Focus
Level2:
Benefits
Level3:
Alternatives
Level1 :
Focus
Level2 :
Costs
Level3:
Alternatives
PRACTICAL EXAMI'LES OF IIIERARCIIIES
Cost
Savings
lncome
Generation
Benefils
Customer
Satisfaction
Legal
Aequiremenls

Corp.
DDA/Aecon.
Personnel
Costs
Cuslomer
Analysis
Hardware
Cosls
Costs
EFTS
Training
Cosls
"-......_
Data
Entry
lnquiry
Operating
Costs

Cor p.
DDA/Aecon.
Customer
Analysis
EFTS
Data
Entry
lnquiry
Figure 4-8 Hierarchies for Alloc.ating R&D Resources
Hierarchy for Making Financial Decisions
To select one from a number of financial projects, we may consider the
benefits and cos ts separately (Figure 4-9). In the benefits hierarchy, three
possible scenari os are considered for the future . The company would like
to base its decision on a number of considerations whose impact depends
on the scenario. Hence the overall priorities will reflect the relative impor-
Level1 :
Focus
Level2:
Scenarios
Level3:
Factors
Level4:
Alternatives
Level1 :
Focus
Level2:
Scenarios
... ,
Level3:
Factors
Level4:
Alternatives
BUSINESS DECISIONS 45
Financial Decision Making
Pessimislic StatusQuo Oplimislic

lncreased
Diversification
'V
Growth
Potenlial
Strong
Mar1<el
Position
(Early)
Currenl
Demand

Independance
lrom Internai
Economie
Slrenglh

Projecl A Projecl B Projecl C Projecl 0
Cos! Hierarchy
Pessimislic Slalus Quo Opl imislic

lncreased
Competition
induslrial
Accidents
Regulation
by
Governmenl
Unmanageabie
Cyciicaiity

insuHicienl
Knowledge
-?/
Projecl A Projecl B Projecl C Pro je cl 0
Figure 4-9 Hierarchy for Financial Decisions

:l
C/J
N
w
i
1
!
1
l:
.;

1
'!3
l


i
'
' :
1
l' i
. - - -
PRACfiCAL EXAMPLES OF IIIF.RARCIIIES
tance of these factors . The projects are ranked according to their contribu-
tion to each factor. The composite priorities give relative measures of the
benefits accruing from them.
ln the costs hierarchy, in a similar way, we find the relative impor-
tance of a number of factors the company would like to avoid or minimize.
The overall priorities of the projects in this hierarchy, th en, give the rela-
tive measure of the negative contribution of these projects. The benefit/cost
ratios give the superiority of the benefits over costs on a ratio scale. The
project with the highest marginal benefitfcost ratio is the best selection.
Hierarchy for Choosing a Marketing Strategy
The decision regarding a company's product/market posture depends
on a number of externat factors thal determine how far the company can
strive to maintain the status quo or expect an optimistic or a pessimistic
environment. The company' s objective regarding economie growth and
Level1 :
Mission
Level2:
Risk Factors
Level 3:
Scenari o
Leve14 :
Objectives
Level 5:
Acti on
Regulat ory
Standards
Weii -Being of Company
Competitive
lntensity
Inflation

StatusOuo

Optimislic
Environment
-;/"\
Pessimistic
Environment

Market
Profitability
Share
Sal es
Growth
Re duce
Vulnerability

lncrease
Promotion
l or
Producl A

Oelete
Product B

A aise
Priee of
Producl C

Enter
New Market
c;s;ure 4-10 Hierarchy for Choosing a Marketing Strategy
......
PERSONAL AND DOMESTIC DECISIONS 47
risk depends on the scenarios envisioned and will be achieved in varying
degrees by following different alternatives of product and market . The
decision mode! is thus represented in the form of a complete hierarchy
(Figure 4-10) .
By prioritizing the factors in one leve! with respect to each factor in the
preceding leve! and finding the overall priorities, we can find the relative
influence, feasibility, importance, or contribution, as the case may be, of
the factors in a leve! with respect to the focus : the company's well -being.
The priority of each course of action is therefore a relative measure of how
far that product/market posture would achieve the desired well-being.
Hierarchy for Evaluating a Division' s Performance
There are severa! dimensions of the performance of a division in a
corporation. The principal dimensions to be considered in this evaluation
are government dealings, management, imports, and customers (Figure
4-11). There are severa! factors for each dimension. Leve! 3 of the hierarchy
shows those pertaining to management alone; other factors can be simi -
larly included for the other dimensions.
The overall priorities of the factors at level3 are the relative weights by
which the evaluators would view performance in that area. Composite
priorities of the various divisions with respect to ali the factors at this leve!
show the relative performance rating of the division on an overall basis.
PERSONAL AND DOMESTIC DECISIONS
Hierarchy for Choosing a Car
The problem of choosing from a number of cars, bath new and old, is
structured into a hierarchy of three levels (Figure 4-12) . ln the second leve!,
the various factors-<osts as weil as benefits-going into the decision
maker's judgment are prioritized. Next the specifie alternatives are com-
pared with respect to each factor in leve! 2. The overall priority of each
alternative indicates its ranking and strength of preference as far as the
buyer is concerned. (This is an example of a complete hierarchy because ali
the factors at any leve! relate to ali the factors in the next higher leve!.)
Hierarchy for Choosing a Home Computer
There are both benefits and costs in having a home comput er. Three
choices are prioritized in two separate hierarchies for benefits and costs
(Figure 4-13) . In each hierarchy, the priorities are determined
\ l
:1
' .,
1
' i
i
'j
CIJ
N
_p.
!
t
q
.,

;,j

;;
!1
' 1
ii ,.

j ;
1
I
l '
,,
,, :

1
1
. ;
, . !
' j j
---"-----------------------------....
Level1 :
Focus
Level2:
Benelits
Level3:
Alternatives
Level1 :
Focus
Level2:
Costs
Level3:
Alternatives
PRACTICAL EXAMI'LES OF HIERARCHIES
Cost
Savings
lncome
Generat ion
Benelits
Cuslomer
Satisfaction
Legal
Aequiremenls

Corp.
DDA/Aecon.
Personnel
Costs
Customer
Analysis
Hardware
Costs
Costs
EFTS
Training
Costs
.............
Data
Entry
lnquiry
Operating
Costs

Corp.
DDA/Recon.
Customer
Analysis
EFTS
Data
Entry
lnquiry
Figure 4-8 Hierarchies for Allocating R&O Resources
Hierarchy for Making Financial Decisions
To select one from a number of financial projects, we may consider the
benefits and costs separately (Figure 4-9). In the benefits hierarchy, three
possible scenari os are considered for the future . The company would like
to base ils decision on a number of considerations whose impact depends
on the scenario. Hence the overall priorities will reflect the relative impor-
Level1 :
Focus
Level2:
Scenarios
Level3:
Factors
Level4:
Alternatives
Level1 :
Foc us
Level2:
Scenarios
.. ;
Level3:
Factors
Level4:
Alternatives
BUSINESS DECISIONS 45
Financial Decision Making
Pessimistic StatusQuo Opl imislic

lncreased
Diversification

Growth
Potenlial
Strong
Mali<el
Position
(Eariy)
Currenl
Demand

Independance
from Internat
Economi e
Strenglh

ProjectA Projecl B Project C Projecl 0
Cos! Hierarchy
Pessimistic Slalus Quo Oplimistic

lncreased
Competition
indu striai
Accidents
Regulation
by
Government
Unmanageable
Cyciicality

lnsuHicienl
Knowledge

Project A Project B ProjectC Pro je cl 0
Figure 4-9 Hierarchy for Financial Decisions

:'.t
'l
i
1
1
f
(X)
N
w
' t
5
.i
: :
1
!
\

!
.-!

1
1
1
JI
1 -
1
H

!
1'
1 .
j:
'
1:
j!
l
1
,
t.
'1
'
:1'
1
i
, ,1


&
c
0
'iii
;;
i5
"'iii
a;
0
- V>

_l U.
PRACTICAL EXAMPLES OF HIERARCHIES
Jt
"' t
8.

E
.,
E
.,
:?
1ij
:::1!

N (1)

> 'E
.3u
g;


.,
'ii
(/)
"2
"'
QI
a:

ii
::>
0



.
., .,


1
c 0
U:i
[l_
\0
r
<

z

t U!
r

wn_
(1)
c:
.. Q)


Q) :::1

u

"' s:
6
co
5
'iii
;;
6
<(
c:
.Q
"' :2:
0
.. c:
"<!' 0

Cll -
_JQ
. - - -
cu
v
c:
j
t
Il.
c:
.Sl
!Il

Cl
1:>1)
c:
.c
"'
= -;;
>

t5
-
>.
.c
v
:;
t
:E
....
....
.
"" QJ
...


Level1 :
Focus
Level2:
Factors
Level3:
Alternatives
PERSONAL AND DOMESTIC DECISIONS
Initial
Cos!

Maintenance
Cos!
Choosing an Automobile
Fuel
Economy
Comfort Status

49
Uncertainties
-?/'
Alternative 1 Alternative 2 Alternative 3
Figure 4-12 Hierarchy for Choosing a Car
intennediate levels of factors and subfactors contributing to the benefit or
cost of home computers. Finally, the benefit/cost ratio for each alternative
is calculated to find out where the benefits stand relative to the costs. The
alternative with the highest marginal benefit/cost ratio is the preferred
choice for the decision maker.
Hierarchy for Choosing a Career
In choosing a career, a person desires satisfaction in severa! dimen-
sions : intellectual, financial, social, and persona!. However, one's sources
of satisfaction are from expectations of learning, growth, leisure, friends,
and prestige. Each source of satisfaction may derive fulfillment from sev-
era! dimensions, which therefore occupy a superior leve! (Figure 4-14) .
The priority of each career with respect to any criterion of satisfaction
reflects its desirability with respect to that criterion only. The overall prior-
ity of the career shows the overall preference for that career; the highest-
priority career is the one preferred. (This is an example of an incomplete
hierarchy, as factors in leve! 3 do not relate to ali the factors in leve! 2.)
Hierarchy for Choosing an Ideal lnvestment
The different criteria affecting an investment are prioritized to find out
how important they are in relative terms to the investor in question (Figure
4-15) . In the next step, the different investment alternatives are ranked
with respect to each criterion to examine how weil they satisfy the criterion
in question. The overall priority will show the relative superiority of an
:t
,;
1
,;
,;:



!_ 1
:,f
:,11
' '
.,
;a


t!l
!
1
co
-
N
Vl
!
'
..,
Level1 :
Foc us
Level2:
Factors
Level3:
Subfactors
Leve14:
Alternatives
Level1 :
Focus
Level2:
Factors
Level3:
Subfactors
Level4:
Alternatives
PRAC riCAL EXAMPLES OF IIIERARCIIIES
Home Computer Benefits
Business
Education Entertainment
Persona!
~ ~ ~ ~
Me mory
Capaclty
~
Spending
Monay
~
Software Expansion
Languages Special Avallabllity Capacity
r
~ -:?\
~
Choice A
Choice B
Cholce C
Home Computer Cosls
Economie
Psychologie al
ldeological
/{\
{\
""'
Wasting Wastlng . Lack of lack of
Monay Tlme Frustration Fa!igue Entertainment Education
~ ~ ~ ~ ~
~
ChoiceA
Choice B
ChoiceC
Figure 4-13 Hierarchy for Choosing a Home Computer
. t
Level1 :
Focus
Level2:
Factors
Level3:
Subfactors
Level4:
Alternatives
Level1:
Focus
Level2:
Criteria
Level3:
Alternatives
PERSONAL AND DOMESTIC DECISIONS
51
Career Choice
lntelleelual Financial Social Persona!
1\ 1 \' -?!\,
Learnlng Growth leisure Friands Prestige
~ ~ ~ ~ ~
CareerA .Career B CareerC Ca reer D Career E
Figure 4-14 Hierarchy for Choosing a Career
Appreciation
~
Bank
Savings
ldeallnvestment
Rate
of Risk Experience
Return
~ ~ ~
Real
Estate
Stocks
Easeof
Monitoring
liquidity
lnvestment
on One's Own
7/\ ~
Troasury
Bonds
Figure 4-15 Hierarchy for Choosing an Ideal lnvestment
Q)
N
0'1
' :1
i
f
,.
1
1
l
1
-- ---- __,. ."!."'--..... ---
Level1 :
Foc us
Level2:
Criteria
Level3:
Subcriteria
Level4:
Alternatives
Educalional
Qualrty of
Teachers
General Standard
Discipline
Preparation for
Co liege
leammg
Environment

a:-:- . .. ...... ..,. _ _ .., ___ ...__ __
Selection of School
Vocational Spons Cultural
Facilities for Facilities Music
Training
Instruction Fine Arts
Ouality of
lnstructors Reputation Dramat1cs
1


SchooiA SchooiB
Figure 4-16 Hierarchy for Choosing a School
Social
Friends
Atmosphere

SchooiC
Spiritual
Religious
Instruction
Inculcation
of Desired
Values

lib. ki UlkX!kU lk!l .!LCJilX .... ( -.,M4QtUJMJt.Ut LAL$2621 145
Level1:
Focus
Purchasing a House
Ul
N
"0
s;
(
-i
n
>
m
x
>
3::
"0
r
m
IJl
0
"Tl
:;
m
;>:l
>
;>:l
Q
...
m
IJl
- ---
54 PRACTICAL EXAMI'LES OF IIIERARCHIES
alternative as viewed by the in vestor. For a large investor, the priorities
.also indicate the proportions in which the total investment could be dis-
tributed among the various alternatives available-that is, a representation
of the investor's portfolio.
Hierarchy for Choosing a School
First we establish priorities for the criteria of the student and the
parents to establish their relative importance (Figure 4-16). ln the next
leve!, we resolve the criteria in greater detail and prioritize the subcriteria.
Each of the schools under consideration is then prioritized with respect to
the subcriteria. The overall priorities show how strongly a school is rated
by student and parents in relation to the others.
Hierarchy for Purchasing a House
Considerations that apply to purchasing a house are the criteria,
which are prioritized to find their relative importance (Figure 4-17). ln the
next leve!, these criteria are decomposed into further subcriteria, which are
similarly prioritized. ln the next stage, the alternative houses under con-
sideration are prioritized with respect to each criterion or subcriterion,
and their overall priorities indicate the buyer's preferences for the houses
in question. This is a comprehensive model in which the criteria pertain-
ing to the neighborhood and those pertaining to the qualities of the house
are dealt with simultaneously.
PUBLIC POLICY DECISIONS
Hierarchy for Choosing a Mode for Crossing a River
To decide which mode to use for crossing a river would be beneficiai to
the community as a whole. We consider the nature of benefits envisaged
and enlist under each the details (Figure 4-18) . Setting priorities for the
benefits gives an idea which ones the community regards as important.
We can also establish priorities for the costs .
Hierarchy for Choosing Areas for R&D Programs
This hierarchy (Figure 4-19) is concerned with the selection of areas for
research and development to ensure adequate power and electricity in the
future. At the first stage, the focus is on the range of planning by prioritiz-
ing the time horizon of the plan. Next we set priorities for the potential

... .
l i
5
>
i
<U
g>
2i
e
u
"5
V>
:t::
a;
c


>
.3u.
PUBLIC POLICY DECISIONS
.j
"'

8

c
w
ut

EE
g
8 Cl>
wco
.. o.l!l
C\1 1/)-

-H- J!


.2!


.!.!
a;
\0
J:
ii
w

.51
E
t'
8



(
ea.
8
c)ll<Uij
a:
"' c

r
0

::J
E
E
8
c
<
0


L
E Ci!

8- [/

e
E o

YI
f
1=
" .l!l

Cl> Cl>
> c
.3J!
--------- ------ ---..
t
Cl>
u.
g>
;:;
"'
)(
w
ai
c
c
:>
t-
Cl>
g'
c
co
"'
Cl>
>
..

t
>

..
llO
c:

0
u
._
0
"'



..
0
._
;...,

v
..
..

::r::
ao

0


::s
llO
iZ
55
co
N
,
J
.-:i

i
1
1.

l
:!
l
...
:i
. 1
!
1'
' 1
,l
.i\

i '
!;!
.\
1
56
j
8
1
.:
j
1

-ll.L

PRACTICAL EXAMPLES OF HIERARCHIES

i
1? / UOijeiUIISUO::J pue puUWOQ
i
u. IIW8,&AS
5
ii(
j'.


:g


! L- f h8 / ::::::,
i l 8PGJ13 411eOH
UJ
! p
w
69!J9UV8

S9:l!A9Q
UQ!I8'JI6UOW9Q IIO::J IBn;j
IU9wd0,9A9Q pe:lUBAPVII8:) 19n;j
UO!GJ91\Uo:t A6J9U3
li
\ . e
....
U{)!IIII!WIUBJ! :JV

UQ!IIS' JW$U'IJ1 JO
UOfSS!WSU8Jl punoJ6J9pUO
.
IIIOIJP913
IOJIUO:) pue 6u!UU81d


IBWJ941081 SjUBUeJI tlMl
l:l8::1t'4l
N g'

Q)- fa
-l O..CI:

li
Q.
. . ij !J)


> 0 c 0
..3a..w(f)

i

..,. ra
- t:
Q)Q)
> :E
..3o

j
. . 1Q
Il) . Si!
- c :Q
Q)
Q) a; ..
-li-<
uatsn::1
&IUBid W881$ j80::J
UOIISnqUJ<Y.) P98 PBZ!P!"I::I
IBO:) PB!l!SWQ
P<i<>S
1
8
0:> uee1::> pue UO!Paoni:J!l
6U!t00:)

pue U!Ie&H 9:lUei&IS91;l
UOUlJ81\U00
'"lOS
.. -5
<D'-
-"':g
:;: Q)
Q) Q) ....
-lCI:<
.,
a
e
Il.
0

a:



till
c::
ii
0
.8
u

;..,
1
QI
:E
0'1
....
.,j.


i.i:
PUBLIC POLICY DECISIONS 57
areas of energy resources to determine which one holds the most promise.
With respect to each energy source, several criteria are considered that
require attention; these criteria are prioritized according to the importance
they command for the energy resource.
For each criterion, several technical aspects are identified and then
prioritized for relative importance. The overall priority at this levet shows
the share of effort and resources that should be devoted to the technical
field concerned. As a continuation of the process, each technical field is
further subdivided into research areas and subareas that need attenti on.
The overall priorities at each levet again represent the extent to which
resources should be devoted to those areas according to the best overall
judgment of the decision makers.
Hierarchy for Choosing a Program to increase Harbor
Capacity
Deciding on means to improve harbor capacity in a small country with
three harbors is viewed as a joint political process. The three principal
stakeholders are the evaluation group of specialists, the transportation
committee of Jegislators, and the harbor bureau (Figure 4-20). Their priori-
ties represent their influence in the matter.
The objectives and considerations to be followed by each stakeholder
are identified and prioritized. At the next stage, the three existing harbors
are prioritized with respect to each objective to find out how much they
contribute to the objectives. The overall priority at this levet shows the
extent to which resources and attention should be devoted to each harbor
development .
The specifie programs to increase harbor capacity are next identified
and prioritized with respect to each harbor . In the context of each harbor,
these priorities show the effectiveness of each action in achieving the goal.
The overall priorities, obtained by weighting the harbor priorities, reflect
the effectiveness for the whole country. The actual policy that is adopted
will consist of several individual programs pursued with varying degrees
of emphasis as represented by these priorities.
Hierarchy for Evaluating Energy Storage Systems
To evaluate four advanced energy storage systems, six feasibility crite-
ria are set .(Figure 4-21) . They are then prioritized to see which would be
dominant over others in relative terms .
Next we prioritize the energy storage systems with respect to each
criterion regarding their suitability. The overall priorities, obtained by
weighting them with criteria priorities, show how they are expected to
<Xl
N
1.0
:
.i
...

,,
,.
t.'
' '
Level1:
Foc us
Level2:
Stakeholders
Level3:
Objectives
Level4:
Harbors
LevaiS:
Action
Programs
_.-.,. _ .... - ---- --.. ............ _ - -- --
PRACTICAL EXAMPLES OF HIERARCHIES
How to lncrease the Harbor's Capaclty
Evaluation
Group
(Speclalists)
Transportation
Commlttee
(Congress)
Harbor
Bureau

Economie
Equlty and
Stabilization

lncrease
Berths
Evaluation
of
Eco nom leal
Elficiency
and
Benelit
Cost
Desire of
Shipowners
Desire of
Cargo Owners


Keelung Taichung Kaushung
Harbor Harbor Harbor
(North) (Middle) (South)

'if\
lmprove Push
Promote
Conlrol Port Are a Procedure
Loading
Pattern of and of
Efficiency
Ship lnland Customers
Arrivai Transport and
System Administration
Figure 4-20 Hierarchy for Increasing Harbor Capacity
satisfy the criteria. The highest-priority storage system is the one pre-
ferred.
Hierarchy for Allocating Resources in Juvenile Correction
Programs
A group of public officiais were interested in juvenile law enforcement
and wanted to allocate resources in five programs the staff had suggested.
To with, they considered three principal areas of correction and
.CTICAL EXAMI'LES OF HIERARCHIES
prioritized them to find out how much attention they should receive (Fig
ure 4-22).
Each principal area could be associated with the programs identified,
so the programs were prioritized regarding their effectiveness with respect
to each area. The overall priorities obtained after weighting by the area
priorities show how much relative importance each program commands
for the optimum juvenile correction system.
Hierarchy for Analyzing School Busing Conflict
The introduction of busing due to the 1954 Supreme Court ruling has
been a source of friction in a certain school district. The minority commu-
nity wants to introduce complete busing for racial integration at school;
the majority community wants segregation to protee! the privileged posi-
tion they now enjoy. To analyze the situation and judge the potential
outcome, we rank the stakeholders according to their relative influence on
the political scene (Figure 4-23). Then we prioritize the objectives of each
stakeholder to see which objectives weigh more and should thus be pur
sued in preference to others. The overall priorities give a picture of the
relative strengths of the forces at work on the scene.
The outcomes under consideration here are three scenarios spanning
the complete spectrum of possibilities. They are prioritized with respect to
each stakeholder objective to find out which outcome is favored by thal
objective. The overall priorities indicate the relative likelihood thal each
possible outcome will occur. This exercise shows the interaction of various
Level1 :
Focus
Level2:
Correction
Are a
Level3:
Programs
Resident iel
Prevention
Juvenile Correction
Preadjustory
Programs
Postadjustory
Programs
~ ~ ' l j \
Educational Vocational Family
lntake and
Referral
Figure 4-22 Hierarchy for Allocating Resources in Juvenile Cor-
rection Programs
Level1 :
Focus
Level2: Majority
Stakeholders Community
Education
for
Children
Level3: Re tain
Objectives
Power
Ensure
Public
Safety
~
PUBLIC POUCY DECISIONS
School Busing Conflict
Minority City Board of
Community OHicials Education
Education
Higher
for
Compensatory
Chlldren
Reputation Wages
Equal
Win Public
Rights
Election Safety
Ensure
Retain Maintain
Public
Power Standard
of
Safety
Education
~ ~
~
61
Federal
Government
En force
Law
and
Or der
lmprove
Quality
of
Education
Enforce
Busing
lor Ail
7
Level4:
Outcomes
Complete Busing Part ial Busing No Busing
Figure 4-23 Hierarchy for Analyzing School Busing Conflict
factors and forces at work, so thal if one stakeholder wanted to influence
the outcome, he or she could decide accordingly on a course of action such
as coalition or persuading others to change their objectives .
Hierarchy for Analyzing Health Administration Conflict
This hierarchy for resolving conflict in health administration finds the
likelihood of various health plans being adopted as national po licy (Figure
4-24) . First we prioritize the principal actors regarding their relative influ
ence on the issue. Next their objectives are identified and prioritized to
indicate the relative extent to which the actors are motivated by various
considerations.
At the next stage, we establish which policies of the actors would
satisfy the objectives. The policies thal relate to the same objective are


Il>
J:

c;:
c:
8
::V)

....JU.
PRACTI CAL EXAMPLES OF HIERARCHIES
cr
.
!Xl-

cr
cr
cr
I
<L-.1.----
.
<( ..C:-+-
cr
cr
!Xl-

,!;


ij
8

Qi 0
>t;
3<

<
<(-
<(0
"'
Q)
.. >

> .!t
38
cr

fl
!Xl.
cO

cr
V)
...,. Q)
"ijj (;)
>'-=
Q) 0
....Jil..

..JO..
a; c:
t: "'
(JC::

c:o..

"'


QiE

..Je(
't
...
<:;:l
c::
0
u
c::
"
li:
= -;

= 1:10
c::
...
N
.
" c::
-(
!S
-......
"
"
t
:E
-.:!'
N
0
...,.

1:10
iZ
PLANNING ECONOMIC POLICIES 63
prioritized to find the importance of the policies to serve the objective in
question. The composite priorities show the overall influence of the poli-
des in national health administration.
There are three main health plans being debated nationally. By
prioritizing the plans with respect to each policy, we find to what extent a
plan satisfies the po licy in q uestiono Th us the overall priorities of the plans
show the relative extent of support for the plans from various interested
quarterso The overall priorities are also the likelihoods of the plans being
adopted nation ally o
The elements of that hierarchy are defined as follows :
A
0
: comprehensive health insurance
A
1
: mixed private-public financing
A
2
: containment of health care costs
A
3
: minimum government involvement
A
4
: expanded health coverage
A
5
: control of health care costs
A
8
: comprehensive health services
A
1
: containment of the total costs of health ca re
B
1
: employer-employee contribution
8 2 : limited public funding
8 3: tax subsidy
B
4
: deductibles/coinsurance
B
5
: planning/integration
B
8
: utilization control
B
7
: advisory federal agency
B
8
: state regulation of N HI
Bv: no provider's charge control
B
10
: increased spending of public funds
B
11
: employer only paid premium
B
12
: control of provider's charges
B
13
: administrative reorganization of Medicare and Medicaid
B
14
: wage-related premiums
B
16
: non-wage-related premium
B
18
: premium pa id from public funds
B
17
: competition between providers and insurers
PLANNING ECONOMIC POLICIES
Hierarchy for Planning Economie Strategy for an
Underdeveloped Country
To map the strategy of economie development in an underdeveloped
country where oil is the chief source of revenue, we draw up a plar
0
"'g
Level1 :
Foc us
Level2:
Premises
Level3:
Sectors of
Economy
Level4:
Policies
Level5:
Strategies
PRACfiCAL EXAMPLES OF HIERARCHIES
Developed
Agriculture
~
More Equal
lncome
Distribulion
~
Trade-Balanced Economie Oevelopment-Year 2000
Minimum Foreign
Exchange
Requirements
~
Developed
Capital
and
tntermediate
Goods
0'
Fuller
Employment
of Labor
and
Capital
StatusOuo
(lmport
Substitution
Policies)
/'\---
Oras!lc
Reduction
in
Protection
Maximum Availabilily
of Foreign
Exchange
~
Oeveloped
Consumer
Goods
and
Services
Developed
Tourism
~ 'l1
More
Rational
Allocation
of
Resources
~
Timetable
Endto
Protection
More
Productive
New
lnvestment
~
Figure 4-25 Hierarchy for Planning Economie Strategy
hierarchy (Figure 4-25) . Although ol plays a very important role in the
economy now, it is expected to be depleted by the year 2000. The country's
planners must develop a strategy with that eventuality in view.
First, Iwo principal underlying premises for a future oil-free economy
are identified and then prioritized to indicate which one will have a more
domi nant influence. Next the important sectors of the future economy are
prior iti zed to indicate the degree to which they will become important
according to the premises. Specifie rational policies are then prioritized
wi th respect to each sector of the economy to determine which would be
ESTIMATING AND PREDICTING 65
more effective. Finally, the strategies in question are prioritized with re-
spect to the policies regarding their beneficiai role on each.
The overall priorities for the policies as weil as strategies show the
national planners which ones will be more effective and hence should be
actively promoted in an overall context.
ESTIMATING AND PREDICTING
Hierarchy for Estimating Popularity of Rock Groups
To estimate the popularity of various rock groups, we structure the
overall performance of a music group into three main components by
which they are judged (Figure 4-26) . These components are then
prioritized regarding their relative influence.
Eah component is further judged in light of :.;everal characteristics of
performance, and the characteristics are prioritized regarding the ir relative
influence. Finally, we prioritize the rock groups with respect to the charac-
teristics to give their relative standing. The composite priority we obtain
gives a relative mea:.;ure of overall superiority and hence an index of thei r
popularity.
Level1 :
Foc us
Overall Preference for Rock Group
Level2:
Music Lyrics Pertormers
Factors
~
(\
~
Level3: Compfexity Smoothness Emotional Signilicance Se x
Personality
Subfactors
of Style of Sound Appeal of Message Appeal
Level4:
Alternatives
Various Rock Groups
Figure 4-26 Hierarchy for Estimating Popularity of Rock Groups
())
w
w
,,
l
,,
1
1
;
j.;
t
'
):



f
j :



\
ji
':
!
l '
, J
i l


il
:;







c
c

- V)
(!) ::l
8
-lU..
PRACTlCAL EXAMPLES OF HIERARCHIES
1(-



:l
<llo::
.J.E

]l


.::
:::>
.!!!

cu
0>
..
_
j

:-e
O
8
(") !:?
- 0

<1l t1l
-lU..


c
I
c ,g
[
.,

0 !E
E

n.
8
<

cl,
u u

5
a: n.
r j
w

'o
w u.ll.
J!l
1


l 1
.c Q)
n. n.
.,
O


w
6


!:?
0
'<t
-rn
<ll-
>.0
<1l ::l
.J(/)
"'

..



.c
u

c
8_
)(
w
.,

Q)

w
(


l
fi
1ii
m
a:
;
t

u
V)
<1l
.. iii
10'0


c
,g
t
" Q
-;;
:D
5
,
...
"

l
..
CIO
c




s
i
"
ti
1
.,


...
"'
ESTIMA TING AND PREDICTING 67
Hierarchy for Predicting the Election
To predict a presidential election, we begin by identifying the princi-
pal political strains influencing ali coalitions and allies-liberais, moder-
ales, and conservatives-and prioritize them regarding their influence on
the contemporary political scene (Figure 4-27) . Next we consider the
characteristics of the candidates and prioritize them to find to what extent
they are decisive. These factors are subdivided into more detailed elements
and then prioritized. The overall priorities we obtain assess how far these
individual elements influence popular choice.
Finally, the three principal candidates are prioritized with respect to
each subfactor to estimate their relative superiority. The overall priority we
obtain is an index of the predictability of the candidate' s winning the
election.
Hierarchy for Predicting Likelihood of Technical
Innovation
We can investigate the likelihood of technical innovation associated
wHh corporate planning with respect to three forms of corporate control :
traditional public ownership, employee ownership, and government own-
ership. We study the relationship in three representative industries (Fig-
ure 4-28) .
For each industry, the related corporate factors are prioritized with
respect to each form. the actors are prioritized with respect to each
corporate factor to obtain their relative impact on the factor in question.
The objectives of each actor are prioritized to find their relative strengths
in influencing decisions, and, finally, the three forms of corporate control
are prioritized to indicate how far each would facilitate the objectives in
question. Therefore, the overall priorities at this leve! show the relative
likelihood that technical innovation will be fostered by the different forms
of corporate control.
Hierarchy for Predicting Number of Children in an
Average Family
This exercise is carried out to anticipate the number of children an
average family is likely to have. Severa! criteria are considered to influence
the number of children in a family: the availability of birth control mea-
sures, working mother, older age at motherhood, education of mother, cost
of raising children, and social pressure (Figure 4-29). These criteria are oo
prioritized to show the relative degrees of influence they exert . w

l
!
1
l
1.
i'
i
1
J
1
}

1
' l
1
l
i
!1


.E

c
'S

0

Gi
-"
::J
- Vl

...J IL
. . ..
PRACTICAL EXAMPLES OF HIERARCHIES


...




Il(
Vl
.. 9!

- Vl
Q) ::l

_._

!!l
-g c
-ti



(.-) Vl
- 0

Ql ro
...JIL

i

Q: (3
a:-
ra
e::J (/) J!
w-


:l
()
g."' .;::; Q)

"'
8
.eni


(.!) 0
a..
i

...,::J
w



::
::l

a:

c o.
.2 Q) -
ti

(/) (.!)
(

f
t:."
>-Cl>


:
- E
g> c
c
i!
c -"'
g ... "'

;: ..., "'
g>
w (/)
Cl
a(!
a:

C1l 0
>u
.3<



"' 0 "' , 0
en

i?n
:l "' LLLL
(/)
C1l
.. >
Ill:.=
Qi al

.g
p
"'
(/)
e:.
(/)
C1l
<DE
Qi 8
>-
Cil ::l
...JO
1
.s
-;;
"'

.c
"'

.....
0
-a
0
0
3
Cil

,_j
()[)
.s


Cil
..

..
0
......
>.
.c

"'
..

::r::
CIO
N
.;. g
QI ...
.. ..
::s l'il
oc>
0
Il.. c:
Level1 :
Focus
Level2:
Criteria
Level3:
Possibilities
Availability
ot
Birth Control
MEASURING INFLUENCES
Working
Mother
FamilySize
Older
Age at
Mothemood
Education
o!Mother
Costot
Raising
Children
69
Social
Pressure

No
Children
1 Child 2 Children 3 Children 4 Children
Figure 4-29 Hierarchy for Predicting Average Number of Chil-
dren .
The number of children considered here runs from zero to four . (We
assume that the number of families having more children is very small.)
With respect to each criterion, the priority of the number of children is
found by ranking on the basis of best judgment . The overall priorities
obtained after weighting by the priorities of the criteria reflect the distribu-
tion of the number of children in an average family . The expected value of
this distribution is the average number of children a family is likely to
have.
MEASURING INFLUENCES
Hierarchy for Measuring Parental Influence on
Psychological Weii-Being
,,
A study of an individual's psychological well-being traces the extent
to which it is influenced by the mother and the father separately and by
both jointly. Psychological well -being depends on self-respect, a sense of
security, and the ability to adapt to others (Figure 4-30) . The factors that
contribute to these qualities in one's developing years are in the third
leve); ali are nfluenced by the parents separately and jointly.
The overall priorities of the factors at any leve) show their relative
contribution to the person' s well -being. The overall priorities at the lowest
leve) represent the influences that the parents brought to bear separately
and jointly in the person' s formative years .
L
V1
1

r
1
1
f
t:
f


;:


;; .:
.!,
' ;
"
'
'

l1
J I

ti
1 .
r .
i'
l
li
i.
i
J.
l
'f
J':
.,.

i
. - ---
PRACTICAL EXAMPLES OF HIERARCHIES
Level1 :
REPRESENTING SYSTEMS NElWORKS
Focus
Level2:
Criteria
Level3:
Contributory
Factors
Visible
AHection
for
Subject
Self-Respect
Ove rail Weii-Belng
Sense
of
Security
Ability
toAdapt
toOthers

Idees of
Strict ness,
Ethics
Actual
Disciplining
ofChild
Emphasis
on
Persona!
Adjustment
wlth Others

A' ?/
Level4:
Influences
Influence
of Mother
Influence
of Father
Influence
ofBoth
Jointly
Figure 4-30 Hierarchy for Measuring Parental Influence
REPRESENTING SYSTEMS NETWORKS
Network of a Child's Learning System
A child's leaming process in the formative years is influenced by
many factors. These factors may be classified into certain groups. What is
significant, however, is that the various factors interact with and influence.
one another. Hence we represent them in the form of a system consisting of
subsystems and further divisions interacting with one another (Figure
4-31). .
By prioritizing the factors in a group with respect to each factor in
each group separately, weighting them by the relative importance of the
groups themselves, and finding how they influence one another, we can
come to a conclusion regarding the intrinsic importance of the various
factors bearing upon the child's upbringing.
The system depicted here consists of five subsystems, which ali impact
on the child' s leaming elements (CHL) (motivation, M; creativity, C;

u.1
()

en
t-d-Cl.
E

"'
>.
Vl
1>0
c
'
l'li
cu
....J
Il)
5:!
:a
u
l'li
'""' 0

15
!
cu
z
....
<"')
'
cu
..
6b
ii:
co
w
(J'\
PRACTICAL EXAMI'LES OF HIERARCHIES
reflection ability, RA; retention and assimilation, R; character and person-
ality, CHP)
1. The school (S)
2. The family (F)
3. The community (C)
4. The family atmosphere (FA)
5. The home (H)
The most important components associated with the school subsystem (S)
are:
1. Teachers (T)
2. Classmates (CL)
3. Installations (1)
4. Study programs (P)
Those for the family (F) are:
1. Parents (PS)
2. Brothers and sisters (B)
3. Close relatives (those who interact directly with the family) (CR)
Those for the community (C) are:
1. Family friends (FF)
2. The child's friends (CHF)
3. The environmental living characleristics (EL)
Those for the family atmosphere (FA) are:
1. The relation between the parents (RBP)
2. The relation between the parents and the child (RPS)
3. The relation between the child and brothers and sisters (BR)
4. The overprotection of people older than the child (OVP)
5. The family's economie status (ES)
And those for the home (H) are:
1. Physi cal characteristics (PCH)
2. The child's toys (TS)
3. The communication media (radio and television) (CM)
This mode! has been evaluated for a certain community, at an average
s o c i o e c o n o ~ i c leve!, in a democratie regime, and for children eight years
old with normal psychological characteristics.
PERSPECTIVE 73
Figure 4-32 Network of a Volleyball Team
Network of a Volleyball Team
Volleyball is agame of teamwork and simple skills, but the play.ers are
required to change positions and possess al/ the skills . Since few players
have ali the skills to the same extent, the coach has to use players with the
proper mix of skills. Th us the coach has to know the relative importance of
skills and play the proper combination of players. Since the various skills
depend upon one another, we use systems analysis and nol hierarchical
representation (Figure 4-32) .
By prioritizing the skills for a good game of volleyball and prioritizing
the players with respect to one another's skills, we get the relative stand-
ings of the players and skills in the ultimate analysis . This information
helps the coach to select players with basic skills .
PERSPECTIVE
The hierarchies just presented are only a few of those thal have actu-
ally been used to make decisions with the AHP. But they suggest the wide
range of problems to which the AHP can be applied-from choosing a car
to crossing a river. Specifically, the AHP can be used for the following
kinds of decision problems:
o Setting priorities
o Generating a set of alternatives
co
w
-....,]
\
l
1
. ~
j
T
~ ~
:i'
i)
:j
, ..
J
i
~
~ :
8
m
1
~
1
~
~~ .
. ' ,,
' :.
~
~
i
~
m
PRACTI CAL EXAMPLES OF HIERARCHIES
Choosing the best policy alternative
Detennining requirements
Allocating resources
Predicting outcomes and assessing risks
Measuring performance
Designing a system
Ensuring system stability
Optimizing
Planning
Resolving conflicts
The sample hierarchies in this chapter should give sorne insight into
the need for various levels in different kinds of hierarchies . Different
people may have their own idea about how to deal with a problem, and
that is how they should carry it out . Only when a group must act together,
as in a corporation, do people need to agree on the structure of their
problem.
In forming a hierarchy, one should include as much detail as seems to
be needed to understand the problem; the prioritization process will
eliminate elements that are unimportant . A new leve! should be added to
the hierarchy if it facilitates the comparison and evaluation of the elements
in the leve! immediately below and contributes to improving precision in
the judgments. One contribution of a good hierarchy is thal it enables
people to make better guesses about the effects of the unknown by laying
out its components and studying each separately instead of Jumping ev
erything together and making one big guess at the consequences of deci
sions made in the face of that unknown. The hierarchy can provide an
effective buffer between reason and worry.
Clearl y the design of an analytic hierarchy-like the structuring of a
problem by any other method-is more art than science. There is no pre
cise formula for identification or stratification of elements. But structur
ing a hierarchy does require substantial knowledge about the system or
problem in question. A strong aspect of the AHP is that the experienced
decision makers who specify the hierarchy also supply judgments on the
relative importance of the elements-which brings us to the next topic:
establishing priorities.
....
5
Establishing
Priorities
This chapter deals with the following questions:
How do we establish priorities in a decision
problem?
Why is the matrix useful in setting priorities?
How do we synthesize our judgments to geta
set of overall priorities?
How can we check the consistency of our
judgments? And how important is consistency?
What do we do when the elements we are
ranking overlap?
75
--
CX>
w
CX>
l
f
- . ..
EST I'RIORITIES
THE NEED FOR PRIORITIES
ln this chapter we complete the analytic hierarchy process by estab-
lishing priorities among the elements of the hierarchy, syn.thesizing our
judgments to yield a set of overall priorities, checking the consistency of
these judgments, and coming to a final decision based on the results of this
process.
Systems theorists point out thal complex relationships can always be
analyzed by taking pairs of elements and relating them through their
attributes . The abject is to find from many things those that have a neces-
sary connection. This causal approach to understanding complexity is
complemented by the systems approach, whose abject is to find the sub-
systems or dimensions in which the parts are connected.
We have seen thal the analytic hierarchy process deals with both ap-
proaches simultaneously. Systems thinking is addressed by structuring
ideas hierarchically, and causal thinking, or explanation, is developed
through paired comparison of the elements in the hierarchy and through
synthesis.
The judgments we apply in making paired comparisons combine logi-
cal thinking with feeling developed from informed experience. The math-
ematical sequence described in this chapter is a more efficient method of
arriving at a solution than the intuitive means we usually employ, but the
end result is not necessarily more accurate. If the solution reached through
the AHP does not fee) right to an experienced, well-informed decision
maker, then he or she would do weil to repeat the process and restructure
the hierarchy or improve the judgments. On the other hand, the AHP
provides its own check on the consistency of judgments, and experience
has shawn that the results of the AHP closely approximate decisions
reached more laboriously in the business world.
lt is important to note thal the calculations described here can be
ca rried out by computer. My aim is not to dwell on the mathematics of the
process-a mathematical supplement is available for those interested-but
to ex plain how subjective judgments can be quantified and converted into
a set of priorities on which decisions can be based.
SETTING PRIORITIES
Tlw first step in the priorities of elements in a decision
problem is to make pairwise comparisons-that is, to compare the ele-
ments in pairs against a given criterion. For pairwise comparisons, a ma-
tri x is the preferred form. The matrix is a simple, well-established tool that
offer s a fr amework for tes ting consis.tency, obtaining additional informa-
ti o n thr ough making ali possible comparisons, and analyzing the sensi-
ti vi ty of overall priorities to changes in judgment. The matrix approach
uniquely reflects the dual aspects of priorities: dominating and dominated.
SEHING PRIORITIES 77
To begin the pairwise comparison process, start at the top of the
hierarchy to select the criterion C, or property, that will be used for making
the first comparison. Then, from the level immediately below, take th
elements to be compared: A v A 2, A 3, and so on. Let us say there are se ven
elements. Arrange these elements in a matrix as in Figure 5-1 .
c
A,
A2
A,
A,
A A,
Figure 5-1 Sample Matrix for Pairwise Comparison
In this matrix compare the element A, in the column on the left with
the elements A 1, A2, A
3
, and so on in the row on top with respect to
property C in the upper left-hand corner. Then repeat with colurrin ele-
ment A2 and so on. To compare elements, ask: How much more strongl y
does this element (or activity) possess- or contribute to, dominate, influ-
ence, satisfy, or benefit- the property than does the element with which it
is being compared?
The phrasing of the question is important. Il must reflect the proper
relationship between the elements in one leve! with the property in the
next higher leve!. If time or another probabilistic criterion is used, then
ask: How much more probable or likely is one element than another? If the
elements are dominated by the property rather than vice versa, ask how
much more strongly the element is possessed, dominated, affected by, and
so on, this property. In projecting an outcome, ask which element is more
likely to be decisive or to result in the outcome.
To fill in the matrix of pairwise comparisons, we use numbers to repre-
sent the relative importance of one element over another with respect to the
property. Table 5-1 contains the scale for pairwise comparisons. lt defines
and explains the values 1 through 9 assigned to judgments in comparing
pairs of like elements in each leve! of a hierarchy against a criterion in the
next higher leve!. Experience has confirmed that a scale of nine units is
reasonable and reflects the degree to which we can discriminate the inten-
sity of relationships between elements. When using the scale in a social ,
psychological, or political context, express the verbal judgments first and
then translate them to numerical values . The numerically translated judg-
co
w
1.0
L
lntmsity of
Importance
3
5
7
9
2, 4, 6, 8
Reci procals
Rationals
ESTABLISHING PRIORITIES
Table 5-1 The Pairwise Comparison Scale
Definition
Equal importance .
Moderate importance
of one over another.
Esaential or
importance .
Very stronq impor-
tance .
Extreme importance .
Intermediate values
between the two ad-
iacent judgments.
Exp/an a ti on
Two elements contribute
equally to the property
Experience and judgment
slightly favor one element
over another
Experience and judgment
strongly favor one element
over another
An element is strongly
favored and its dominance is
demonstrated in practice
The evidence favoring one
element over another is of the
highest possible order of
affirmation
Compromise is needed
between two judgments
When act1v1ty 1 copared to j 1s ass1gned one
of the above nubers, then act1v1ty j compared
to 1 1s ass1gned its reciprocal.
Ratios aris1ng from forcing cons1stency of
judgents.
ments are approximations; their validity can be evaluated by a test of
consistency, which will be described later, and by real-life applications for
which the answers are already known.
When comparing one element in a matrix with itself-for example, A
1
with A
1
in Figure 5-1-the comparison must give unity (1), so fill in the
diagonal of the matrix with 1s. Always compare the first element of a pair
(the element in the left -hand column of the matrix) with the second (the
element in the row on top) and estimate the numerical value from the scale
in Table 5-1. The reciprocal value is then used for the comparison of the
second element with the f1rst. For example, if the two elements are stones
and the first is five times heavier than the second, then the second is
one-fifth as heavy as the f1.rst .
Why not simply use arbitrary numbers for ranking the elements ac-
.IIi :,
. .'
SYNTHESIS 79
cording to their impact on a criterion? If the problem concerns simple
ranking, and the degree to which the elements being ranked reflect the
criterion is obvious, then one can simply assign numbers. To discriminate
the relative strength with which each element possesses or contributes to
the criterion (property), numbers can be used directly by starting with the
smallest element and perhaps using it as a unit . This procedure may be
useful in organizing one's thinking, but the logic is not clear and, more-
over, feeling is not integrated into the process. For fine distinctions, the
pairwise comparison matrix and scale provide a more satisfactory frame-
work.
When tradeoffs must be made among severa! criteria, the problem of
ranking becomes complex. lt is no longer sufficient simply to assign arbi-
trary numbers. We must select with care the numbers used to express the
strength with which each element possesses or contributes to the property
in question. Such care ensures that in the end we obtain the correct overall
priorities for the elements by considering ali tradeoffs . (These priorities can
also then be used to allocate resources.)
SYNTHESIS
To obtain the set of overall priorities for a decision problem, we have
to pull together or synthesize the judgments made in the pairwise
comparisons-that is, we have to do sorne weighting and adding to give us
a single number to indicate the priority of each element. The following
example explains how to synthesize.
Suppose we want to decide which of three new cars- a Chevrolet, a
Thunderbird, and a Lincoln-ta buyon the basis of comfort . We draw a
matrix with the criterion "comfort" listed in the upper left-hand corner and
the cars listed in the column on the left and in a row on top (Figure 5-2) . We
Comfort c T L
Chevro/et (C) 1 112 1/4
Thunderbird (T) 2 1 112
l.i 11 co/n ( L) 4 2
Figure 5-2 Simple Matrix Comparing Three Cars for Comfort
then put ls in the diagonal positions as indicated. This matrix has nine
entries to fiJI . Three are already committed to ls. Three of the remaining six
are reciprocals. This leaves three jud_gments to make. ln -,J, if the
<Xl

0
i
1
l:
l:
'
~
j .
ill.'
:
.H
',1
'l -
ii
~ - .
~
... . - - - - - - - - ~ - - -- ___________ ....,........, ________________ _
EST ABLISHJNG PRJORITIES
matrix deals with, say, seven elements, the number of judgments needed
to fiJI the entries is 7 x 7 - 7 + 2 = 21. We subtract the seven unit entries
down the diagonal and divide by 2 because half the judgments are recip-
rocals th at are entered automatically.
We then ask: How much more comfortable is an average new Chev
rolet than an average new Thunderbird and an average new Lincoln?
Based on our experience and persona! preference, our judgment is that a
Chevrolet is one-half as comfortable as a Thunderbird and one-fourth as
comfortable as a Lincoln . To state these judgments in terms of the quan-
tifiers in the scale (Table 5-1), a Thunderbird is slightly more comfortable
than a Chevrolet, and a Lincoln is between slightly and strongly more
comfortable than a Chevrolet. Thus we enter the values 2 for the Thunder-
bird over the Chevrolet and 4 for the Lincoln over the Chevrolet. These
num bers are the reciprocals of the two judgments comparing the Chevrolet
with the other cars.
Remember that the element that appears in the left-hand column is
always compared with the element appearing in the top row, and the value
is given to the element in the column as it is compared with the element in
the row. If it is regarded less favorably, the judgment is a fraction . The
reciprocal value is entered in the position where the second element, when
it appears in the column, is compared with the first element when it ap
pears in the row.
In this example, because the Chevrolet is regarded less favorably when
compared with the other two cars, we enter 112 and 1/4 in the second and
third positions of the first row and enter 2 and 4 in what are known as the
transp ose positions in the first column. We then compare the Thunderbird
with the Lincoln and enter a value of 1/2 in the second row, third column
position, and its reciprocal 2 in the second column, third row position. We
now have the three judgments needed to complete the pirwise compari-
son matrix (Figure 5-2).
Next we want to synthesize our judgments to get an overall es ti mate of
the relative priorities of these cars with respect to comfort. To do so, we
first add the values in each column (Figure 5-3) . Then we di vide each entry
SYNTHESIS 81
in each column by the total of thal column to obtain the normalized matrix,
which permits meaningful comparison among elements (Figure 5 - ~ ) . Fi-
Comfort 1 c T L
-
c 117 117 1/7
T 2/7 217 217
L 417 417 417
Figure 5-4 Normaliz:ed Matrix
nally, we average over the rows by adding the values in each row of the
normalized matrix and dividing the rows by the number of en tries in each:
1/7 + 117 + 1/7 = 117 = 0. 14
3
2/7 + 217 + 217 = 217 = 0.29
3
4/7 + 417 + 417 = 4/7 = 0.57
This synthesis yields the percentages of overall relative priorities, or pref-
erences, for the Chevrolet, the Thunderbird, and the Lincoln: 14, 29, and 57
percent, respectively. As far as comfort is concemed, the Thunderbird and
the Lincoln are thus about twice and four times more preferable than the
Chevrolet.
The answer in this case was very simple, because ali the columns in the
normalized matrix were the same. They turned out to be the same because
the pairwise comparison matrix (Figure 5-2) was consistent . That is, from
the relationship of the Chevrolet to the Thunderbird in the first row of the
matrix,
Comfort 1 C T L 1 . C = (1/2)T
- " 1
c 1 1/2 1!4 . ~ . . . . .
and from 1ts relatlonsh1p w1th the Lmcoln,
T 2 1 112
L 4 2 1 :S... C = (1/4)L
Columrr
total 7 3.5 1.75 ' we can deduce thal
Figure 5-3 Synthesizing the Judgments . (li2)T = (1/4)L
co
~
:1
1
1
i
l
1
l
\
1
t
Il
lj
!.
li
i:
ij
l.
ii
"
1
,,
l '
:1
.,
~ 1
:;
: J
;;
d
q
i'
\;
H
li
ij
1.
(
'
'
~ .,
1
:j
'
l
!
1
!
... .. - ..... ... __ .. ___ --- -- .--.. ------------
ESTABL.ISIIING PRIORITIES
and that
T = (1/2)L
which is precisely what we have in the second row, third column entry. In
other words, if the Chevrolet is preferred half as much as the Thunderbird
and one-fourth as much as the Lincoln, then the Thunderbird must be
preferred half as much as the Lincoln. The information in the first row is
used to force judgmental consistency.
CONSISTENCY
ln decision making problems it may be important to know how good
our consistency is, because we may not want the decision to be based on
judgments that have such low consistency that they appear to be random.
On the other hand, perfect consistency is hard to live up to. Our judgments
on the relative comfort of the three cars were consistent, but in real !ife
specifie circumstances often influence preferences, and circumstances
change.
If apples are preferred to oranges, for example, and oranges are pre
ferred to bananas, then in a perfectly consistent relationship apples must
be preferred to bananas. But the same individual may sometimes like
bananas better than apples, depending on the time of day, the season, and
other circumstances. In the example of the cars, we identified a couple of
relationships that showed the strength of our preference for a Thunderbird
over a Chevrolet and for a Lincoln over a Chevrolet and fNced these rela
tionships on the comparison between a Thunderbird and a Lincoln-the
Thunderbird was preferred half as much as the Lincoln . But often such a
relationship does not hold true. Violating it, which we do ali the time,
leads to inconsistency.
How damaging is inconsistency? Usually we cannot be so certain of
our judgments that we would insist on forcing consistency in the pairwise
comparison matrix. Rather, we guess our feelings or judgments in ali the
positions except the diagonal ones (which are always 1), force the recipro-
cals in the transpose positions, and look for an answer. We may not be
perfectly consistent, but that is the way we tend to work. (lt is also the way
we grow. When we integrale new experiences into our consciousness,
previous relationships may change and sorne consistency is !ost. As long as
there is enough consistency to maintain coherence among the objects of
our experience, the consistency need not be perfect.) lt is useful to
remember that most new ideas that affect our lives tend to cause us to
rearrange sorne of our preferences, thus making us inconsistent with our
previous commitments. If we were to program ourselves never to change
.ttft'
CONSISTENCY 83
our minds, we would be afraid to accept new ideas. Ali knowledge has to
be admitted into our narrow corridor between tolerable inconsistency and
perfect consistency.
Of course, a certain degree of consistency in setting priorities for ele-
ments or activities with respect to sorne criterion is necessary to get valid
results in the real world. The AHP measures the overall consistency of
judgments by means of a consistency ratio. The value of the consistency
ratio should be 10 percent or less. If it is more than 10 percent, the judg-
ments may be somewhat random and should perhaps be revised. Let us
continue with the example of the cars and see how the AHP measures
consiste ney.
Suppose that we keep the first row of our pairwise comparison matrix
in Figure 5-2 but do not pay much attention to consistency with our previ-
ous judgments. In comparing the Thunderbird with the Lincoln, we enter
the value 1/4 in the second row, third column, and enter its reciprocal 4 in
the third r.ow, second column (Figure 5-5) . Following the steps described
Comfort 1 c T L
-
c 1 1/2 114
T 2 1 114
L 4 4
Co/um11
lolr1/ 1 7 5.5 1.5
Figure 5-5 lnconsistent Matrix
earlier, we obtain the normalized matrix, its row sums, and the percent
ages of relative overall priorities (Figure 5-6) . The percentages, 13, 21, and
66 percent, constitute the priority vector of the three cars with respect to
comfort. The value of the priority vector is approximate. (We can find the
Avrragr
Comfort c T L Row Sums Row Sum
c 1/7 1/11 1/6 0.40 0.4013 = 0. 13
T 2/7 2/11 116 0.63 0.6313 = 0 .21
L 4/7 8/11 4/6 1.97 1.9713 = 0 .66
Figure 5-6 Normalized Matrix, Row Sums, and Overall Pri r- ' ies
(X)
~
N
ESTABLISIIING PRIORITIES
exact value, but the solution is complicated. Besicles, when the judgments
are perfectly consistent, the two values are identical; nearly consislent, the
values are close.) Although the standing of the Chevrolet has not been
changed by much, the other two have changed by our reducing the value
for the Thunderbird and raising it for the Lincoln.
With inconsistency ali the values are changed. The question is: How
significant is this change? Presumably we want to compare our inconsis-
tency with the value it would have if the judgments were random. To do
this, multiply the first column of the inconsistent matrix (Figure 5-5),
changed to decimal form, by the relative priority of the Chevrolet (0.13),
the second column by that of the Thunderbird (0.21), and the third column
by that of the Lincoln (0.66). Then total the entries in the rows (Figure 5-7) .
Row
Comfort c (0 . 13) T (0.21) L (0.66) Comfort J c T L Total
- -
c 1 0.5 0.25 c 0.13 0. 11 0. 17 0.41
T 2 1 0.25 1' 0.26 0.21 0.17 0.64
L 4 4 1 L 0.52 0.84 0.66 2.02
Figure 5-7 Totaling the Entries
Now take the column of row totals and divide each of its entries by the
corresponding entry from the priority vector (figure 5-8). We can now find
the average of the three entries in the last column of Figure 5-8:
3.15 + + 3.06 = 9.;6 "" 3.09
By convention, the symbol for this number is max (lambda max) . The
consistency index (Cl) is
3.09 - 3 = = 0.045
The random value of the CI for n = 3 is 0.58. The consistency ratio (CR) is
0.045/0.58 = 0.08, which indicates good consistency.
If numerical judgments were taken at random from the scale 119, 1/8, 117, . . , 112, ... , 1,
2, . . . 9, then using a reciprocal matrix we would have the following average consistencies
for di fferent -order random matrices:
Size of matrix
Rand<'m consistency
1 2 3 4 5 6 7 8 9 10
0.00 0.00 0.58 o. 90 1.12 1.24 1.32 1.41 1.45 1.49
EXTENDING THE PROCESS oS
[
0.41] .
0.64 ..,.
2.02
[
0.13]- [3.15]
0.21 - 3.05
0.66 3.06
Figure 5-8 Determining max
A second approximation procedure is to compute the geometrie mean
of the elements in each row-that is, to multiply the elements and then
take the ir n th root. This step is followed by normalizing the resulting
vector so that its components add to unity. ln general, the geometrie mean
is a good approximation, particularly when the consistency is high . The
calculation of max can proceed as before. The geometrie mean for the
inconsistent matrix of cars with respect to comfort yields 0. 16, 0.20, and
0.64. The exact solution by computer is 0. 13, 0.21, 0.66, and max = 3.05,
which nearly concides with the results of the column normalization pro-
cess described earlier. Note that row averaging followed by a normali za-
tion of the resulting vector yields 0. 13, 0.23, and 0.64 . There are many
for which this last process yields unsatisfactory results when the
matrix is inconsistent.
One way to improve consistency when it turns out to be unsatisfactory
is to rank the activities by a simple order based on the weights obtained in
the first run of the problem. A second pairwise comparison matrix is then
developed with this knowledge of ranking in mind. The consistency
should generally be better.
EXTENDING THE PROCESS
To see how the process just described can be extended to an entire
hierarchy, consider the problem of a woman who has recently earned her
Ph.D. and is being interviewed for three jobs. Which one should she
choose? Figure 5-9 shows how she structured the elements of the problem
and arranged them in a hierarchy. Leve) 1, the focus, is overall job satisfac-
tion; leve! 2 comprises the criteria that contribute to job satisfaction; and
leve! 3 consists of the three job possibilities. The hierarchy is a complete
one: Each element in a leve) is evaluated in terms of ali the elements in the
next higher leve!.
The woman compared the leve) 2 criteria in pairs with respect to job
satisfaction and judged the relative importance of each criterion. She felt
that research, for example, would be equally as important as loca tion in
contributing to job satisfaction, but it would be slightly to strongly more
important than colleagues. Figure 5-10 shows the pairwise comparison
matrix of the criteria with respect to the focus . The last column gives the
ro

w

1
:l
' J
:!
j
j
:i
l
,.. _ --------.-..-- .--. .. .. ...._ .. ,.. ...... ...._ . . .. iloo ....... ... , r 1 a rtra._
. : ESTABLISHING PRI ORITIES
Level1:
Focus
Overall Job Satisfaction
Level2:
Criteria
Bene lit Location Repulallon Re se arch Grow1h Colleagues



Level3:
Jobs
B c A
Figure S-9 Hierarchy for Choosing Among Three Job Qffers
priorities: Reputation tums out to be the most important criterion, fol-
lowed by opportunities for growth and benefits .
Next she developed six matrices for comparing the three jobs with
respect to each criterion (Figure 5-11). Ali three entries in the vector of
priories obtained in each of the six matrices and listed in the las! column
of each are multiplied (weighted) by the priority of the corresponding
criterion. These values are shown in Figure 5-12. The results of this opera-
tion are then added to yield the overall priori lies for the jobs: A= 0.40, B =
0.34, and C = 0.26. The differences made apparent by this synthesis were
sufficiently large for the woman to accept the offer for job A.
Although we will not go into the measurement of consistency here, it
is important to note thal in a hierarchy, the highest-level elements usually
have the highest priorities. Inconsistency arising from comparison with
respect to these elements is very damaging because of their high priority.
The consistency index of a hierarchy is obtained by multiplying the consis-
tency index of each matrix. by the priority of the criterion used for the
comparison and adding ali such quanti lies . To evaluate the consistency of a
hierarchy, compare the consistency index of the hierarchy with ils coun-
terpart when the consistency indices of the matrices are replaced by aver-
age random judgment consistency indices for matrices of the same size.
INTERDEPENDENCE
So far we have considered how to establish the priority of elements in
a hierarchy and how to obtain the set of overall priorities when the ele-
ments of each leve! are independent . But often the elements are interde-
pendent. How do we account for overlapping areas, or commonalities,
among such elements? There are t_wo principal kinds of interdependence
,.,

... -
0
- 0
0

r;:
.8

::s
"-
..
c.:::
r;:
0
.z::
3
0
....J
le
::s
bo
"'
..

u
"'
't
E
<rl
_.,
fj
::
Cl
_.,
0
...
"'
..
"'
....
c.:::
..Q r;:
0 0
- .:::
- 0

t; . !(1
p -

INTERDEPENDENCE
....... 9"""'4 0 ..-4
ci ci ci ci ci ci
1:::! 1:::! 1:::!
9"""'4 9"""'4,......! ..-4-
..,.... ,...... M s .... ,..,
-.:t' "'tt' 1.1') ,...... M M
.-4 N ..-. !!2 N
........
.........
........
,.... N
.....-4 9"""'4 9"""'4 ;! ,...... N
"' "'
..s:: 1:
:: -:: .9 .!!

" ... E o o "
c.::: Cl <rl u ....J c.:::
.0
0
--
-ii

5
:e
.:!

"'
V'l
]
5
0
...
J..


"'
87
<Xl
_p.

q
h
li
i
F-lJ AC . . Sj
::_. .
Vector of Vector of
8enefits 1 Research 1 A 8 c Priorities Growth A 8 c Priori ti es
A 1 1/4 112 0.14 A 1 114 115 0.10 A
8 4 1 3 0.63 B 4 1 1/2 0.33 B
c 2 1/3 1 0.24 c 5 2 1 057 c
Vector of
Location 1
Vector of
Reputa- 1
Col/eagues 1 A B c Priorities A B c Priorities lion
A 1 113 5 0.28 A 1 1 7 0.47 A
B 3 1 7 0.65 B 1 1 7 0.47 B
c 115 1f7 1 0.07 c 117 117 1 0.07 c
Figure 5-11 Six Matrices for Comparing Three Jobs
Research Growth
8enelits Col/eagues Location
(0. 16) (0.19)
(0.19) (0.05) (0.12)
A 0. 14 {0. 16)
+ 0.10 {0. 19)
+ 0.32 (0.19)
+ 0.28 (0.05) + 0.47 {0.12)
+
8 0.63 (0.16)
+ 0.33 (0. 19)
+ 0.22 (0.19)
+ 0.65 (0.05)
+ 0.47 (0.12)
+
c 0.24 {0. 16)
+ 0.57 (0.19)
+ 0.46 {0.19)
+ 0.07 {0.05)
+ 0.07 (0.12)
+
Figure 5-12
Determining the Overall Priorities
SV' 8
A 8 c
1 3 113
1/3 1 1
3 1 1
A 8 c
1 7 9
117 1 5
1/9 115 1
Reputation
(0 .30)
0.77 (030) =
0.17 {0.30) =
0.05 {0.30) =
Vector of
Priorities
0.32
0.22
0.46
Vector of
Priori ti es
0.77
0.17
0.05
Vector of
Ovn-all
Priori ti es
0.40
0.34
0.26
m
CJl
-!
>
cc
r
Vi
::::
z
C"l
"0
;:1::1

;:1::1
::i
m
CJl

;:1::1
0
m
"'0
m
z
0
m
z
()
m
00
-o
1
. . _..:_ . ___ ; .;.._ :..:..:,:- ..... :...
0 - -- - _: _ _ _ _ __ _____ _ .. -- . ..,... . . ii'cz=n '
,.-......--ol .. ....... .....
ESTAII! .ISIIINC I'RI O RITI ES
among elements of a hierarchy leve!: additive interdependence and syner-
gistic interdependence.
Additive lnterdependence
ln additive interdependence, each element contributes a share that is
uniquely its own and also contributes indirectly by overlapping or in-
teracting with other elements. The total impact can be estimated by exam-
ining the impacts of the independent and the overlapping shares and then
combining the impacts. The effects of this simpler type of interdependence
can be computed precisely since we can usually tell how much of an ele-
ment's contribution is due toits independent properties and how much to
ils effect on other elements. 6oth mechanization and farm size con tri bute to
agricultural productivity, for example, but mechanization also influences
farm size by enabling farmers to work larger areas . ln practice, most people
prefer to ignore the rather complex mathematical adjustment for additive
interdependence and simply rely on their own judgment. ln the example
just given, mechanization would be assigned a priority than farm
size. The precise value would be determined subjectively from the descrip-
tions in the pairwise comparison scale. Such judgments can also replace
the more technical adjustment for synergistic interdependence.
Synergistic lnterdependence
ln synergistic interdependence, the impact of the interaction of the
elements is greater than the sum of the impacts of the elements, with due
consideration given to their overlap. This type of interdependence occurs
more frequently in practice than additive interdependence and amounts to .
crea ting a new entity for each interaction. Power coalitions and the biology
of marriage are examples of synergy.
The analytic hierarchy process provides a simple and direct means for
measuring interdependence in a hierarchy. The basic idea is thal wherever
there is interdependence, each criterion becomes an objective and ali the
criteria are compared according to their contributions to thal criterion.
This generales a set of dependence priorities indicating the relative de-
rendence 0f each criterion on ali the criteria. These priorities are then
weighted by the independence priority of each related criterion obtained
from the hierarchy and the results are summed over each row, thus yield-
ing the interdependence weights . By way of validation we find that this
approach is compatible with, for example, what econometricians do in
calculating input-output matrices . These ideas are illustrated in the math-
emahcal supplement.
With synergistic interdependence, one needs to introduce for evalua-
PERSPECTIVE 91
lion additional criteria thal reveal the nature of the interaction. The over-
lapping element should be separated from its constituent parts. Ils impact
is added to theirs at the end to obtain their overall impact. Synergy of
interaction is also captured at the upper levels when clusters are compared
according to their importance. Interactions within and between clusters
are better seen and judged higher up in the hierarchy.
Much of the problem of synergistic interdependence arises from the
fuzziness of words and even the underlying ideas they represent . The full
potential of interaction is never fully understood until il has taken place in
practice. The qualities thal emerge cannot be captured by a mathematical
process. Thus a set theoretical approach using Venn diagrams would not
be useful, because we do not have a simple geometrie overlap of regions.
What we have instead is the overlap of elements with other elements to
produce an element with new properties thal are not discernible in ils
parent parts.
Note that if we increase the elements being compared by one more
element and attempt to preserve the consistency of the.ir earlier ranking,
we must be careful how we make the comparisons with the new element.
For example, if we have been comparing apples, oranges, and bananas
with respect to taste and then add melons to the problem, this new addi-
tion may change the priorities . Once we compare one of the previous
elements with a new one, ail other relationships are automatically set ;
otherwise there would be inconsistency and the rank order might change.
PERSPECTIVE
The process of setting priorities captures the feelings and judgments
of informed individuals by asking them to make pairwise comparisons
among like objects as criteria. The judgments, which represent strength of
preference, are simultaneously converted to numerical values to represent
their intensity and are laid out in a matrix. The priorities are then derived
from ali the judgments and the consistency of the judgments is calculated
through the deviation of a single number from the order of a matrix . ln a
hierarchy we synthesize the priorities in a leve! by weighting the priorities
derived from each matrix by the weight of the criterion of comparison. To
obtain overall priorities, we add the results for each element . Note thal
prioritization from the top of the hierarchy downward includes less and
less synergy as we move from the larger more interactive clusters to the
small and more independent ones.
lndependence can be treated in two ways. Either the hierarchy is
structured in a way that identifies independent elements or dependence is
allowed for by evaluating in separa te matrices the impact of ali the el e -
ments on each of them with respect to the criterion being considered . As
before, a weighting process is then used to determine their prioriti es .


11
:!,
{
j
1
J


<

1
p
i
l
:l
:!
(X)
_p.
0"
1!
,.
" r,
' j
:
1
;
,,
.,.
1 ,,
J '
1!
:i
:;
i'i
::
;
;
ESTAIJLISHING PRIORITIES
If we already have an idea of the ranking of elements and wish to
derive their priorities, the judgments we give must indicate this domi
nance in rank. Otherwise we must assume that the individual does not
fully understand (or is inconsistent in) his or her subjective ranking.
KEY CONCEPTS
0 To establish the priorities of elements, we have to compare them in
pairs according to a criterion. A matrix is the best framework for this
corn parison.
0 To obtain the set of overall priorities for a decision, we have to synthe
size the results of the pairwise comparisons. That is, we have to com
bine our judgments to get an overall estimate of the relative ank of
priori ti es.
0 ln a hierarchy, the highest-level elements usually have the highest .
priorities. They are the clusters thal give rise to smaller elements at the
lower levels. Their priorities are divided by the weighting process
among their descendants .
0 The consistency of a hierarchy can be measured by multiplying the
consistency of each matrix by the priority of its criterion and adding.
This result is then compared with a similar number obtained for ran-
dom matrices of the same size . The ratio should be 10 percent or less.
Grea ter inconsistency ind ica tes lack of information or lack of under-
standing.
6
Step-by-Step
Examples of the
Process
This chapter deals with the following questions :
What are the basic steps of the analytic
hierarchy process?
How can the process be used to analyze
high-level decisions like the one to rescue the
host ages in Iran?
How can the process be used to determine
consumer preference?
How can the process be used to estimate the
economy's impact on sales?
How can the process be used to select a stock
portfolio?
How valid is the analytic hierarchy process?
93
co
~
'.J
1
~
1
(
..
..
i
;,

,_.
' ,;
l1
y
- .... ........ - - .--- .... -- '"' w .......... a ....
STEP-BY-STEP EXAMPLES OF THE I'ROCESS
AN OUTLINE OF THE STEPS
We began our study of the analytic hierarchy process by laying out the
elements of a problem as a hierarchy. We then made paired comparisons
among the elements of a leve! as required by the criteria of the next higher
leve!. These comparisons gave rise to priorities and finally, through syn-
thesis, to overall priorilies. We measured consistency and deal! with inter
dependence. These basic steps of the process can ali be condensed into a
brief outline. In broad terms, the process is stable, although certain steps
may be given special emphasis in particular problems and, as noted be-
low, repetition is generally necessary.
1. Define the problem and specify the solution desired.
2. Structure the hierarchy from the overall managerial viewpoint (from the
top levels to the leve! at which intervention to solve the problem is
possible).
3. Construct a pairwise comparison matrix of the relevant contribution or
impact of each element on each goveming criterion in the next higher
leve!. In this matrix, pairs of elements are compared with respect to a
criterion in the superior leve!. In comparing Iwo elements most people
prefer to. give a judgment thal indicates the dominance as a whole
number . The matrix has one position to enter thal number and another
to enter ils reciprocal. Thus if one element does not contribute more
than another, the other must contribute more than it. This number is
entered in the appropriate position in the matrix and ils reciprocal is
entered in the other position. An element on the left is by convention
examined regarding ils dominance over an element at the top of the
matrix.
4. Obt21in ali judgments required to develop the set of matrices in step 3. If
there are many people participating, the task for each person can be
made simple by appropriate allocation of effort , which we describe in a
later chapter. Multiple judgments can be synthesized by using their
geometrie mean.
s.
6.
7.
Having collected ali the pairwise comparison data and entered the re
ciprocals together with unit entries down the main diagonal, the priori-
lies are obtained and consistency is tested .
Perform steps 3, 4, and 5 for ali levels and clusters in the hierarchy.
Use hierarchical composition (synthesis) to weight the vectors of priori-
lies by the weights of the criteria, and lake the sum over ali weighted
priority en tries corresponding to those in the next lower levet and so on.
The result is an overali priorit y vector for the lowest leve! of the hier
archy. If there are severa! outcomes, their arithmetic average may be
ta kP"' .
$ ... .. .-rtr--""",_,.,,,..... . . , .. ,...,, , ..,...., . , , ,. , , .. , ...


ANALYZING THE HOST AGE RESCUE OPERATION 95
8. Evaluate consistency for the entire hierarchy by multiplying each con-
sistency index by the priority of the corresponding criterion and adding
the products. The result is divided by the same type of expression using
the random consistency index corresponding to the dimensions of each
matrix weighted by the priorities as before. The consistency ratio of the
hierarchy should be 10 percent or less. If it is not, the quality of informa -
tion should be improved-perhaps by revising the manner in which
questions are posed to make the pairwise comparisons. If this measure
fails to improve consistency, it is likely thal the problem has not been
accurately structured-that is, similar elements have not been grouped
under a meaningful criterion. A return to step 2 is then required, al -
though only the problematic parts of the hierarchy may need revision .
These are the basic steps we will follow in working out the examples of
this chapter. In each case we will examine the problem, set up the hierar-
chy, carry out the pairwise comparisons, determine the priorities, synthe-
size the overali priorities, and examine the consistency. We begin with a
recent application of the method to the Iran rescue operation . An intangi
.ble factor played an important role in President Carter's thinking but not
necessarily the thinking of his advisors- hence their different conclusions
regarding the rescue operation.
The second example challenges us to select the best product to man -
ufacture from three alternatives by using six criteria . The product that is
chosen costs but is more desirable ali around. The third exampl e
shows how the AHP may be used to estimate the impact of energy, reces-
sion, and inflation on a company's sales. Here percentage ranges are
prioritized according to likelihood of occurrence. The fourth example ill us-
traies how the AHP can be used to set priorities for stocks according to
severa! criteria. By way of validation, the high-priority stocks did in fa ct
appreciate in value.
ANALYZING THE HOSTAGE RESCUE OPERATION
It is customary for high-level managers and political leaders to make
their most important decisions by depending on expert recommendations
governed by their own persona! judgment and understanding. President
Carter once said thal when he had to make a really crucial decisi on be-
tween two alternatives, his experts were usually evenly spli t. ln essence ,
then, the hard decisions were still up to him .. If expert opini ons were
accurate and thorough, leaders would become superfluous. In splt e of
splitting on their judgments the experts do raise questi ons thal stimul ate
decision makers and draw their attention to issues they may have ne-
glected to consider.
1
i
1
1
j

(X)

(X)
l
... --- '- -- --- - -- - - - - - - - ~ - - - ~ - - - - - - - - - - .....
STEP-BY-STEP EXAMPLES OF THE PROCESS
Background of the Decision
A great shroud of secrecy surrounds the decision to rescue the hos-
tages in Iran. On 28 April 1980 a decision was made to send an American
air rescue team to Iran to bring out the fifty-three American hostages from
Teheran where they had been held since early November 1979. The mission
was a complicated plan involving troops, airplanes, helicopters, a long
flight, .a landing in the desert, a journey to Teheran, taking out the hos-
tages, and returning to safety.
It appears that not even the upper-leve! national security staff was
informed about how the decision was made. It is said thal President Carter
had the Joint Chiefs of Staff draw up a couple of plans with different
options. He then discussed these plans with his dosest advisors. According
to Time (5 May 1980) these advisors were Brown, Brzezinski, Christopher,
Jordan, Mondale, Turner, and Vance. Carter asked for clarification of sorne
1
of the ideas, but he made the decision on his own. There is no indication
that the president resorted to sorne kind of voting in the discussion on
whether the operation should or should not be undertaken. It has been said
thal this was a typical Carter-style decision in which he asked his advisors
for details and then made the decision himself.
There is no question that the likelihood of success of a rescue opera-
tion is a compelling factor in deciding whether to go or not to go. Of
course, it is crucial how one defines success for such a mission. High
success means no deaths among the hostages or military personnel; me-
dium success means a few military and no hostage deaths; low success
means a few military and a few hostage deaths. Greater !osses could mean
anything from failure to disaster. Sorne experts in the Department of De-
fense who are familiar with that mission have indicated that it would be
generous to assign a medium chance of success to the military operation.
They also said thal this was known prior to 28 April and is not a question
of hindsight. Time writes: "Carter himself conceded that 'the operation
was certain to be difficult and it was certain to be dangerous.' He insisted
th at the operation had ' an excellent chance of success. "'
As we will see, the president's decision was consistent with his per-
ception of the situation as it affected him; but it was not necessarily a good
decision for the nation. Suppose we do a sensitivity analysis of the subjec-
tive factors and the emphasis the president placed on them.
Analysis of the Decision
The decision problem can be divided into two parts. The first partis to
identify the best military option among those available and evaluate its
likelihood of success . A military option would also have to be examined by
expert s on foreign relations and intelligence. The second partis the process
ANALYZING THE HOST AGE RESCUE OPERATION 97
of making the go/no-go decision based on the body of knowledge provided
by the experts . We now examine these two parts in detail.
The likelihood of success was determined by military experts. They
considered such factors as:
Transferability: to the desert, to Teheran, then to the embassy
Rounding up: getting inside the compound, creating diversion, locat ing
the hostages
Rescue: subduing the captors, transfer to aircraft, departure (avoiding
Iranian forces)
It is not my intention here to analyze the possible details of their delibera-
tions. In this instance the importance of the military factors and the likeli-
hood of the mission's success were probably much more important than
the reaction of our allies or the Russians. These factors do appear in the
hierarchy for the problem, but one could have construed the medium
likelihood of success from other sources. In fact, an analysis of the first part
did produce a r11edium likelihood of success, but a Pentagon expert, when
asked, strongly supported these findings. Thus we may assume that a
medium likelihood of success was presented to the president, who then
made his go/no-go decision. How he did this is what 1 wish to dwell upon .
The hierarchie structure of both parts of the pro cess is shown in Figure
6-1. We find that the main factors that could have played an important role
in President's Carter's mind are:
Hostages' lives: the president's concem for the safe return of ali fifty -
three hostages
Carter's politicallife: the president's concern about the influence of the
decision and his chances for reelection
Military costs : The president's concem about the loss of soldiers' lives in
the operation
United States prestige: the president's concern about the influence of
the decision on relations with foreign states and their subsequent image
of the United States
These factors differ in their impact on the president's decision . More-
over, their relative importance changes as the likelihood of success is
changed. Now let us assess their priorities based on a medium likelihood
of success of the rescue operation. We can carry out a pairwise comparison
process by using the pairwise comparison scale (Table S-1) . Figure 6-2
shows, for example, that "hostages' lives" has strong dominance over
"military costs" in the president's mind. Thus we assign the value 5 in the
first row and third column position and its reciprocal value in the first
column and third row position. We always compare the row factors on the
left over the column factors on the top.
STEP-BY-STEP EXAMPLES OF THE PROCESS

<_u.
= "2 ll .>
111 "'

c-al
.= <
::>UJct
/(_



3 (
t:.
:8 i ::; 0

1
0

z
1
I <
1
i .. (
ns-
l

1

,,
f
t:!
I


t

t-
:::>
UJ
8'
:g

0
'g u
1
8

f.
a:
-'o
Qi

1
::;
(
E _

<!)-
t: "' 3
1110
ua..
- l
\
0
<!)
< "- H
i.-

0 v ::io
.=t:.
v
Qi 0

:I:-

II
f

c::
.9

0
Q
()
1
Q
z
0
()
u
=

....
;.,
-5
..
s
...
x
...
.b
u
i "
...
Ilot
ANALYZING THE HOSTAGE RESCUE OPERATION 99
2 3 4 Priori lies
1. Hostages' lives 1 113 5 113 0. 15
2. Carter's political !ife 3 1 7 4 0.54
3. Military costs 115 117 1 116 0.05
4. U .S. prestige 3 1/4 6 1 0.26
Figure 6-2 Relative Priorities of the Factors
As can be seen from the column of priorities in Figure 6-2, the two
main factors are "Carter's political !ife" and "United States prestige."
" United States prestige" is important because the United States had to do
something to assert its power despite the fact that there was a medium
chance for success. "Carter's political !ife," a subjective factor, is the
highest-priority factor.
Under each factor we consider this question: Which alternative (go or
no-go) is more favorable considering just that factor under a medium
likelihood of success? The results are shown in Figures 6-3 to 6-6.
Under "hostages' lives" the decision was evenly split . The rationale
for this judgment is that there was no immediate danger to the lives of the
hostages. With respect to "Carter's political life," we note that the presi-
dent was strongly in favor of performing the operation at that time; a few
months before the election the polis were not predicting a good chance for
his reelection. A successful operation would have resulted in a strong
influence on Carter's public image. Failure would be painful to him, but it
Host ages' Prior-
Liu es G N ities
Go 1 1 0.5
No-go 1 1 0.5
Figure 6-3 Priority of Hostages' Lives
Carter's Pol - 1 Prior-
itical Life
Go
No-go
G
1
1/3
N
3
ities
0.75
0.25
Figure 6-4 Priority of Carter's Political Life
00
V1
0
1
_..,.... .. - ... . -- ~ .. . "
STEP-BY-STEP EXAMI'LES OF Tl-lE PROCESS
Military Prior-
Costs G N ities
Go 1 1/7 0.125
No-go 7 1 0.895
Figure 6-5 Priority of Military Costs
Prior-
U.S. Prestige 1 G N ities
--
Go
1
1 4 0.8
No-go 1/4 1 0.2
Figure 6-6 Priority of U.S. Prestige
would not hurt him as much as a success would benefit him. The influ-
ences of "military costs" and "United States prestige" on the decision are
apparent . Composing the factors' weights with those of the go/no-go alter-
natives under each factor gives these results :
Go: 69 No-go: 31
lt is very easy to appear smart after the events. Yet 1 am convinced that
if the AHP had been applied bef ore the decision was made, we would have
obtained similar results. Seven people established the priorities presented
here. The variance in their judgments was small. lt is possible to argue that
if those seven people had set the priorities individually, they would .have
arrived at the same decision.
lt is clear that "Carter's political !ife" dominated that decision. Sen-
sitivity analysis shows that if the 75 percent in favor of go under that factor
is changed to 38 percent, the outcome for go/no-go would have been even.
I believe that 75 percent is a modes! estimate. Carter may not have gone
through with the decision to go if his desire to be reelected approximately
matched his concern about the hostages.
To pursue the analysis further, my colleagues and I examined the
outcome of the decision under a "low likelihood of success" recommenda-
tion by the experts. (Presumably this would have been interpreted less
optimistically by the president . Recall that he interpreted medium success
as " excellent.") Our results are shown in Table 6-1.
ANALYZING THE HOST AGE RESCUE OPERATION
Table 6-1 Relative Priorities Given a Low
Likelihood of Success
Priori/y Given Priori/y Given
Medium Likelihood Low Like/ihood
Factor of Success of Success
Hostages' lives 0.15 0.35
Carter's political life 0.54 0.39
Military costs 0.05 0. 10
U.S. prestige 0.26 0.16
101
In this case the "hostages' lives" become a more important issue and
"Carter's politicallife" less important. The influence of these factors on the
go/no-go decision is shown in Table 6-2.
The outcome of the decision under a "low likelihood of success" rec-
ommendation by the experts leads to the following results:
Go: 0.41 No-go: 0.59
The change from the previous result is due to a greater emphasis on the
hostages' lives. Only if the hostages' lives were clearly in jeopardy in Iran
would one have decided in favor of go. Moreover, "Carter's political life"
would have had to be less important because a military operation with a
low chance of success would have had a low chance of helping Carter.
Now let us pull together the observations we have drawn from this
analysis of the Iran rescue operation. For President Carter, the subjective
factor-namely, his concern with his career-accounted for 54 percent of
the total. Perhaps the dominance of this subjective factor was perceived by
Table 6-2 Influence of Factors on the
Go/No-Go Decision
Ml'dium Likelihood Law Likelihood
Factor Go No-Go Go No-Go
Hostages' lives 0.50 0.50 0.20 0.80
Carter' s political life 0.75 0.25 0.75 0 25
Military costs 0. 125 0.875 0.10 0.90
U.S. prestige 0.80 0.20 0.25 0.75
cc
lJ1
.....
i
1
!
j
'1
il
~
il
.1
:1
l\
i'
Il
~ =
i\
.1
'1
l'
~
, l ~
~ -
;,
1
'
;
. ~
~
Il
1
STEP-BY-STEP EXAMPLES OF THE PROCESS
Carter's aides; perhaps it accounts for Secretary of State Vance's resigna-
tion. Certainly "Carter's political !ife" would not have figured as impor-
tantly in Vance's analysis of the situation. He would have had to decide
against the operation. Of course he may have had other political reasons of
his own.
This application of the AHP illustrates its effectiveness in analyzing
high-level decisions. There is a value to analysis even after a decision has
been made because the method is an efficient tool for deriving a lesson
from previous mistakes. ln this example the results were consistent with
reality. ln sorne situations, as we have seen, media information may be
. sufficient to indicate what went into a decision. If used by decision makers,
the process can sharpen thinking and reveal subjective factors. lt can
show, for example, that persona! interest is weighted much higher than
what is good for the business. lt can also show that there is not a clear
commitment to the important objectives.
DETERMINING CONSUMER PREFERENCE
The AHP is basically a simple, efficient technique for problem solving.
The following step-by-step example demonstrates this simplicity; il can
also serve as a mode! for using the process to solve other problems.
A firm wants to determine consumer preferences for three different
kinds of paper towel. The attributes considered most relevant from the
consumer's perspective are (1) softness, (2) absorptiveness, (3) priee, (4)
size, (5) design, and (6) integrity. The three kinds of paper towels, X, Y, and
Z, possess ali these attributes, but at different levels of intensity: high (H),
medium (M), and low (L). Given the consumer' s "bounded rationality"-
that is, the fact that consumers do not act on perfect or complete informa-
tion and are satisfied with less than the economically most rational
choice-we can best distinguish among the attributes by dividing them
into this small number of intensity categories. The resulting hierarchy is
shown in Figure 6-7. The problem of selecting the product with the
greatest overall consumer preference is solved in the following manner:
Step 1: Determine consumer preference among the attributes by devel -
oping a matrix that compares attributes in pairs with res.pect to product
desirability (Figure 6-8).
Step 2: Determine consumer preference among the intensities of the
attributes by developing six matrices that compare intensity levels in
\c pairs with respect to each attribute (Figure 6-9) .
":Ji
Now we want to synthesize these judgments to obtain the set of over-
all priorities that will indicate which product consumers prefer. The re-
maining steps lake us through this process:
~
15
~
<ii
<1>
0
u
::J
~
a.

DETERMINING CONSUMER PREFERENCE
"
v
c
"
..
"
....
~
Cl..
..
"

<Il
c
0
u
co
c
'
'
~
"
8
Cl
'1:: ..
a. 0
....
>.
-= v
..
..
..
. ~
::r:
"'
1
\0
"
~
co
iZ
103 '
co
U1
N
1
1
i
!
l
]
~
STEP-BY-STEP EXAMI'LES OF THE PROCESS
Attributes
Product
Desirability s A
p
SI D 1
s 1 1/4 1/5 1/4 5 116
A 4 1 113 3 6 1/2
p 5 3 1 4 7 3
SI 4 1/3 114 1 5 115
D 1/5 116 117 115 1 117
6 2 1/3 5 7 1
>...,,, = 6.66; Cl= 0.12.
Figure 6-8 Matrix Comparing Attributes
Absorp-
Soft ness H M L Priority tiueness H M
H 1 5 8 0.7257 H 1 7
M 1/5 1 5 0.2122 M 1/7 1
L 118 115 1 0.0621 L 119 1/7
h"'"' = 3.15; Cl= 0.07. = 3.33; Cl = 0. 16.
l'ricr 1 1/ M L Priorily Size Il M
-
H 1 117 1/9 0.0480 H 1 3
M 7 1 117 0.1912 M 113 1
L 9 7 1 0.7608 L 1/5 1/4
= 3.33; CI= 0.16.
= 3.09; Cl = 0.04.
Design H M L Priority /ntegrity H M
H 1 1/5 2 0. 1786 H 1 7
M 5 1 5 0.7089 M 117 1
L 1/2 115 1 0. 1125 L 119 117
h,,., = 3.05; Cl = 0.03. h"'"' = 3.33; CI = 0. 16.
Figure 6-9 Matrices Comparing Intensity Levels
Priority
0.0570
0.1679
0.3837
0.1002
0.0269
0.2643
L Priority
9 0.7608
7 0.1912
1 0.0480
L Priority
5 0.6267
4 0.2797
1 0.0936
L Priority
9 0.7608
7 0.1912
1 0.0480
H
M
L
DETERMINING CONSUMER PREFERENCE 105
Table 6-3 Priorities of the Attributes
(0.0570) (0.1679) (0,3837) (0.1002) (0.0269) (0.2643)
Soft ness Absorp. Priee Sizt Design /ntegrity
0.7257 0.7608 0.0480 0.6267 0. 1786 0.7.608
0.2122 0.1912 0.1912 0.2797 0.7089 0.1912
0.0621 0.0480 0.7608 0.0936 0. 1125 0.0480
Step 3: Group the priorities of the intensities (H, M, L) for each of the six
attributes in columns and enter the priorities of the attributes, laken
from Figure 6-8, above the columns (Table 6-3). Then multiply each
column by the priority of the corresponding attribute to obtain the
weighted vectors of priority for the intensities (Table 6-4) .
Step 4: Now select from each column the element with the highest
priority to obtain the vector of desired attribute intensities:
H-Soft
0.0413
H-Absorp.
0.1278
L-Price H-Siu
0.2919 . 0.0628
M-Design
0.0190
H-llltegrity
0.2011
Then add this row and divide each entry by the total to gel the nor-
malized vector of desired attribute intensities:
H-Soft
0.0556
H-Absorp .
0.1717
L-Price
0.3924
H-Size
0.0844
M-Design
0.0256
Ff-lntegrity
0.2703
Step 5: Determine the perceived product standings by developing ma-
trices that compare the three paper towels (X, Y, and Z) in pairs with
respect to the most desired attribute intensities (Figure 6-10) .
Step 6: Group the priorities of the paper towels with respect t.o each
desired attribute intensity in columns and enter the normalized priori-
Table 6-4 Vectors of Priority for the Intensities
Soft ness Absorp. Priee Size Design
H 0.0413 0. 1278 0.0184 0 0628 0.0048 0.2011
M 0.0121 0.0321 0.0734 0.0280 0.0190 0.0505
L 0.0035 0.0081 0.2919 0.0094 0.0030 0.0127
1.
1
l
,.


1
'
(X)
lJl
w
_. ... o, -- ,. - - --- . - ........ --... - - -- -. .... v- . .. _,h_______ - ..............,,.lr"".JO:f ..........,._/,-w.oll..............,_f,, , ... . .. _, , ., .... . .. . _, .. .. -- -

=
0
r:::
Cl.
N
>-
><
"'-

..<:>
"(
l:

::
0
r:::
Cl.
N
>-
><
a- ('.. ....
LI") N .-.
l::; ;g
0 0 0
('.. <Xl -
N ,..... 22
-
N !'-.

>< >- N

- - t-: "! .
0 0 0
('..LI")-
LI")
LI") - ;::.
- !!? t::
--


:i:
STEP-BY-STEP EXAMPLES OF THE PROCESS

0
Il
0
ci
0
,.;


E
-<
g;
0
Il
0
ci
- ,.;
.
.
E
-<

r::: ::::! li)
0 N (1"')
r::: . . .
Cl. 0 0 0
N 1,........ ,..... .,.....
>- 1 N ..-< ,....
><
"'
N
Vi
:i:

i5
r:::
Cl.
N
>-
><
"' u
r:::
Cl.
...J
!:::!
-
>< >- N
('.. :il ""
0
0 N !'-.
0 0 0
Si ;3 -
.... - li) ;:::;
- ""' ('..
>< >- N
I""'
0
0
Il
0
,i)
0
...,


-<

0
0
ti
,.;
.
.
E
-<
>.
t:
0
r:::
Cl.
N
>-
><
= r:::


:r::

:::
0
r:::
Cl.
N
>-
><
"'

"'
Cl

('.. "" a-
-"' - 00 "" 00
\0 N 0
0 0 0
-c .........
.... ..... ....
;:::;

-.....
>< >- N
('.. li) <Xl
8 a-
.... N
6 0 0
..... "" -
N r-' t:?
.....
!:::!
-
>< >- N
li)
0
0
Il
0
-
-
""
'-
.
E
-<
00
.....
0
Il
0
r:;
""!
""
.

-<
Ill
:3
Ill

Cil
-
-Cil
= .0
c:
:t:
<
"Cl

;;;
Cil
0
..
.E
Ill
"ii


..
lU
Q..
.,
Q.,
Cil

..c:
(-o
:S
bO

.:
.,
Q..
El
0
u
Ill
Cil
v
c:
-:;

0
....
1
"'

bO
iZ
x
y
z
x
y
z
(0.0556)
H-Soft
0.7147
0.2185
0.0668
ESTIMA TING THE ECONOMY'S IMPACT ON SALES
Table 6-5 Overall Product Attribute Perception
(0. 1717)
H- Absorp.
0.5659
0.3727
0.0614
(0.3924)
L-Price
0.0727
0.2050
0. 7223
(0. 0844)
H-Siu
0.4126
0.2599
0.3275
(0.0256 )
M-Design
0.4067
0.3695
0.2238
107
(0.2703)
H-lntegrity
0.6817
0.2363
0.0819
Table 6-6 Weighted Overall Product Attribute Perception
H-Soft
0.0397
0.0121
0.0037
H-Absorp.
0.0972
0.0640
0.0105
L-Price
0.0285
0. 0804
0.2834
H-Si u
0.0348
0.0219
0.0277
M-Design
0.0 104
0.0095
0.0057
H-lnt egrity
0. 1842
0.0639
0.0221
lies above the columns (Table 6-5) . Then multiply each column by the
normalized priority of the corresponding attribute intensity to obtain
the weighted vectors of priority for the desired attri bute intensities for
each paper towel (Table 6-6).
St ep 7: Add each of the three rows to obtain the overall priorities of the
three paper towels. This synthesis yields the following priorit ies:
x = 0.3949 y= 0.2519
z = 0.3532
From these result s we would select product X as most desirable from the
customer's perspecti ve.
Even though low priee was the desired attribute intensity with the
highest priority, product X, whose priority was very low with respect to
low priee, was the final choice. The reason for this choice is clear: X domi -
nated Y and Z on ali other desired attribute intensities . Thus the firm
decided to market a superior but high-prieed product , a decisi on that is
not inconsistent with real-world situations.
ESTIMATING THE ECONOMY'S IMPACT ON SALES
The AHP can be used for certain special applications as weil as for
obtaining sets of priorities in decision problems. In the following examl"'"
l
;
.1
1
1
i
il
1\
~
tl
(
1
'f
:'!
' :
STEP-BY-STEP EXAMPLES OF THE PROCESS
Future Sales 1 EC R 1 Priorily
-
Energy crisis 1 7 1 0.4667
Recession 117 1 117 0.0666
Ill fla lion 1 7 1 0.4667
.\"'"' = 3.0; Cl = 0.00.
Figure 6-11 Matrix for Comparing Criteria with Respect to Fu-
ture Sales
Enl'rgy
Cr isis 0- 5 5-10 10- 15 15- 20 Pri ority
0- .'i 1 1/5 117 1/5 0.0518
.'i - 10 5 1 1/3 1/4 0. 1451
10- 15 7 3 1 1/3 0.2904
1 . ~ - 2 0 5 4 3 1 0.5127
ma ' = 4.337; Cl = 0.11.
Recessio11 1 0- 5 5-10 10-15
15 -20 ;' Priority
-
0- 5 l 2 5 7 0.5232
S- 10 112 1 3 5 0.2976
10- 15 i /5 1/3 1 3 0.1222
I S- 20 117 1/5 1/3 1 0.0570
.\ """ = 4.069; Cl = 0.02.
lnflatio1 0- 5 5-10 10- 15 15-20 Priority
0-5 1 2 5 7 0.5232
5- !Cl 1/2 1 3 5 0.2976
10- 15 115 1/3 1 3 0. 1222
IS-20 117 1/5 1/3 1 0.0570
.\ .,,." = 4.069; Cl = 0.02.
Figure 6-12 Matrix for Comparing Sales with Respect to Criteria
SELECTING A PORTFOLIO 109
the process is extended to estimate the percentage of a company's sales
affected by the energy crisis, recession, and inflation .
First the sales of the company-a manufacturer of heavy equipment
(oil drills and construction machines)-are divided into intervals : 0-5 per-
cent , 5-10 percent, 10-15 percent, and 15-20 percent. Then a matrix is
developed to compare the criteria (energy crisis, recession, inflation) in
pairs with respect to future sales (Figure 6-11). Next matrices are devel-
oped to compare sales with respect to the criteria (Figure 6-12) . The overall
priorities of the sales percentage intervals are
0.3033 0.2264 0.2007 0.2697
To get the expected value of sales affected by the energy crisis, reces-
sion, and inflation, multiply the mid point of each interval by the priorit y
weight of that interval. For example, the mid point of 0-5 percent is 2.5 and
so on. Thus
(2.5 x 0.3033) + (7.5 x 0.2264) + (12.5 x 0.2007) + (17.5 x 0.2697)
= 9.685%
As this example shows, the analytic hierarchy process can be used to
estimate numbers- in this case percentages.
SELECTING A PORTFOLIO
This application is more complex than the preceding one. Here we use
the AHP to select a portfolio. Our hierarchical mode! consists of three
separate hierarchies : one based on extrinsic factors, one based on intrinsic
factors, and a third based on the investor's objectives. The firms being
considered are ranked (weighted) relative to the criteria in each hierarchy .
The weights are then combined to get an overall preference list of firms .
Figure 6-13 gives an overall view of the mode! we will use .
The various factors and objectives that influence the selection of firms
for the portfolio include:
Extrinsic Fact ors (A) : These are the outside factors or environmental
characteristics thal affect an industry' s (or firm' s) performance. The firm,
however, has no direct influence on them. These factors are economie,
political, social , and technical. By incorporating the analysis of the ex-
trinsic variables we can determine the sensitivity of a firm to changes in
these factors .
Intri11sic Fact ors (8 ): These are the internai factors or operati on al charac-
. teristics of the firm. They may be considered as a measure of the way the
firm is making its decisions or, in general, a measure of the firm' s capac-
1
co
(J1
(J1
l
1
l
J
j
..
STEP-BY-STEP EXAMPLES OF THE PROCESS
Risk Risk Risk
!
Level Lev el Class
1
A B c
Extdnsic lntrinsic lnvestor's
Factors Factors Objectives
1
A Criteria B Criteria
1
Firms Firms Firms
1
Portfolio
Figure 6-13 Hierarchical Model for Selecting a Portfolio
ity to compete successfully. These factors are profitability, size, technol-
ogy, and philosophy.
Investor's Objectives (C): These are the values that define the actions the
investor undertakes in the business world. We are aware, of course, of
the great variety of objectives an investor may have, but to simplify the
mode! we consider four mutually exclusive objectives: profitability, se-
curi ty, excitement, and control.
Since we are dealing with a mode! that is based on future conditions,
we must consider risk-the uncertainty of future events. The mode! incor-
porates the uncertainty of the general business environment, the firm's
behavior (high, medium, low risk), and the risk class of the investor (high,
medium, low) .
As our frrst step we look at the hierarchy of extrinsic factors . At the ftrst
SELECfiNG A PORTFOLIO 111
leve! we have the primary extrinsic factors that affect a firm's behavior. At
the second leve! we have the criteria that influence each of the primary
factors:
AI: Primary Extrinsic Factors
Economie
Political
Social
Technological
Ail: Criteria
Employment conditions
Elasticity of demand
Elasticity of supply
International economy
Interest rates
Government regulations
International exposure
Employment conditions
Family disintegration
Age distribution
Educational achievement
Employment conditions
State of technology
Government involvement
The extrinsic. factors are compared pairwise for a high-risk environment, a
medium-risk environment , and a low-risk environment (Figure 6-14). ln
each case we base the comparisons on the questions: Which factor has
more impact on a firm' s behavior and by how much7 We can see that in a
high-risk environment the group constructing the matrix believed the
technology would have a generally strong influence on a firm's behavior
compared to the other factors.
We then compute the priorities for each of the factors for each risk leve!
(Table 6-7). These priorities point out that in a high-risk environment,
future technological factors have the greatest impact on a firm's behavior.
In a medium-risk environment both the economie and technological fac-
tors are most influential; in a low-risk environment the social factors are
High Risk
1
Medium Risk
1
Low Risk
Firm E
p
s T E
p
s T E
p
s T
Economie 1 3 4 1/3 1 5 7 1 1 5 1/3 4
Po/itical 113 1 3 115 1/5 1 3 1/5 1/5 1 1/5 2
Social 1/4 1/3 1 118 117 1/3 1 1/6 3 5 1 5
Te ch nological 3 5 8 1 1 5 6 1 1/4 1/2 1/2 1
Figure 6-14 Pairwise Comparison of Extrinsic Factors
q
'
(X)
lJ1
0'1
.,
'
;j
:
,
1.
( .
1
l'
r
( )
~
Risle
High
Medium
Low
STEP-BY-STEP EXAMPLES OF THE PROCESS
Table 6-7 Priorities for Each Risk Level
Economie Po/itica/
0.25 0. 12
0.43 0. 11
0.30 0.10
State of technology
Goternment irrvolvement
Social
0.06
0.05
0.54
5
1
1/4
G
4
1
Technologicn/
0.57
0.41
0.07
Figure 6-15 Comparison Matrix for Technology
most important . Economie factors have appreciable impact at ali three
levels whereas the impact of political actions is rather low:
We could carry through our analysis un der each of the three risk levels,
but for illustrative purposes we will use the average weight for each factor.
A weighted average could be used if we wished to favor one type of future
over the other. A veraging, we obtain these weights:
Economie Political
0.32 0.11
Social
0.22
Technological
0.36
Now we compare the criteria for each factor to see their order of impor-
tance. In each pairwise comparison we ask: Which criterion has more
impact on the factor and by how much?
ln the case of technology and its two criteria, we have the comparison
matrix shown in Figure 6-15, which indicates that the state of technology
has somewhat more impact on technology in general than does the gov-
ernment's involvement in it . This matrix and three others yield the follow-
ing sets of weights for the various criteria:
Employment Elasticity of Elasticity of
Conditions Demand Supply
0.16
Economie (0.32): ~
0.45 0.07
International lnterest
Economy Rates
0.05 0.27
a:
~ l ' ,
SELECTING A PORTFOLIO
Political (0.10) :
Social (0.22) :
Government
Regulations
0.24
Family
Disintegration
0.30
Educational
Achievement
0.11
State of
Technology
Technological (0.36): 0.80
International
Exposure
0.06
Age
Distribution
0.11
Employment
Conditions
0.48
Government
lnvolvement
0.20
Employment
Conditions
0. 70
d3
To get the final weights for a criterion, we must multiply the criterion
weights just found by the weight of the factor associated with that criterion:
Employment
Conditions
0.05
Economie:
International
Economy
0.02
Government
Regulations
Political : 0.02
Fa mil y
Disintegration
0.07
Social :
Educational
Achievement
0.02
Elasticity of
Demand
0.14
Interest
Rates
0.09
International
Exposure
0.01
Age
Distribution
0.02
Employmenl
Conditions
0.11
Elasticity of
Supply
0.02
Employment
Conditions
0.07
CXJ
lJl
'-]
.. - . - -- _ .. - .. ... -.... ...... -----.. -- ...
STEP-BY-STEP EXAMPLES OF THE PROCESS
Technological:
State of
Technology
0.29
Government
Jnvolvement
0.07
The weights printed in boldface are the largest ones. We will shorten our
list to include only these. In order to have the weights total1, we divide
each weight by the total of ali the weights in the shortened list. We have
our list of extrinsic criteria:
State of Employment Elasticity of
Tecltnology Cortditions Demand
0.33 0.26 0. 16
Government
lnterest Fa mil y Involvement
Rates Disintegration in Technology
0.10 0.08 0.08
The firms being considered (eight in this illustration) are now
prioritized relative to each of the six extrinsic criteria. This process is the
same pairwise comparison and weighting procedure we have been apply-
ing throughout . When comparing the firms pairwise relative to the state of
technology, for example, we ask: Which firm will respond more favorably
to the technological environment of the future? The matrix for this criterion
is sHown in Figure 6-16.
Technology T IR /C AC R DG 8 c
Tappan 1 1/3 115 4 115 117 6 5
Ing . Rand 3 1 1/4 4 1/4 116 7 6
/ . C. Industries 5 4 1 5 1/2 114 6 5
A/lied Ch . 1/4 1/4 115 1 1/6 117 4 3
Rockwe/1 5 4 2 6 1 1/3 7 6
Data General 7 6 4 7 3 1 9 7
Butler 1/6 117 116 1/4 117 1/9 1 113
Cltemetron 115 116 115 113 116 117 3
Matrix for Pairwise Comparison of Firms Relative to
State of Technology
--------.. ... - ----.. .. -- --- .... ..
SELECTING A PORTFOLIO 115
Table 6-8 summarizes the weights that the eight firms received for
each criterion. A glanee at the table shows that in a technological environ-
ment Data General, Rockwell, and I.C. Industries are heavily favored , as
one might expect . Tappan and Rockwell are beneficiaries if employment
conditions are good-Tappan's consumer market is in appliances (80 per-
cent of sales) and Rockwell's is in home and auto products (35 percent of
sales) .
To get an overall prioritized list of firms vis--vis the extrinsic fac-
tors, we multiply the weights of the firms for a given criterion by the
weight of that criterion (shown in parentheses in Table 6-8). Then we add
up these new weights for each firm:
Tappan
0.16
lng . Rand
0.07
I. C. Industries
0. 13
A/lied Ch .
0.03
Rockwe/1 Data General Butler Chemetron
0.20 0.21 0.07 0.13
For purposes of illustration we will use only the four highest ones with
their weights adjusted as we did wh