Vous êtes sur la page 1sur 20

Power motivation and risk-taking

David C. McClelland and Robert I. Watson, Jr., Harvard

Why do people work on tasks of different levels of difficulty,

set higher or lower levels of aspiration for themselves? Why do
some people choose goals which are easy to achieve whereas
others choose goals in pursumg which they run a considerable
risk of failure? The classic explanation erf nsk-takmg behavior
was formulated by Atkmson m 1957 "The strength of motivation
to perform some act is assumed to be a multiphcative function
of the strength of the motive, the expectancy (subjective prob-
abihty) that the act will have as a consequence the attamment
of an mcentive, and the value of the incentive Motivation = f
(Motive X Expectancy X Incentive)." Although the formula-
tion IS given m general terms as if it applied to any motive mter-
actmg with any mcentive, in fact Atkmson and his associates
have worked exclusively with achievement motivation and gen-
erally assumed that to the extent that the model of nsk-takmg
behavior works for achievement motivation it will work for other
motives as well
The key to the success of the model m prechctmg choice be-
havior in task situations hes m the beautifully economical as-
sumption that the amount of achievement mcentive (la) in a
task situation is an inverse function of the probability of accom-
plishmg (Pa) the task successfully That is, the more difficult a
task (l-Pa), the greater the incentive value of succeedmg at it.
la = l-Pa. Atkinson writes Is or Ps to refer to the incentive value
of success or probability of success, but we have substituted la
and Pa to make it dear at the outset that more precisely the
model IS formulated to deal with the incentive value of achieve-
1 Hie researdi reported hare was made p<»sible hy a grant from the Na-
tional Institute of Mental Health, MH-02980, and horn ^ Natimal Sdeaee
Foundation, GS-31914X
122 David C McClelland and Robert I. Watson, Jr.
ment and probabihty of aeihievement. The term success is more
general and may refer to a successful outcome m terms of any
incentive, not just completmg a task successfully. Atkinson used
the more general notation because he beheved the model was
generally applicable, but we believe whether it is or not is an
empincal question to be settled by further research. So we prefer
the more precise notational system
Table l Appre>ach to tasks of different diflScnilty as a jomt function of
achievement motive (M,), expectancy (P,) and mcentive (la) where

Proportion of
lines distant from Predicted actual ciioices
peg* in nng toss M. X P. X 1. ~ approach out of 400 throws

Task A 1-3 1 99 01 01 00
Task B 4-6 1 88 12 11 05
Task C 7-9 1 70 30 21 09
TaskD 10-12 1 55 45 25 33
TaskE 13-15 1 38 62 24 34
ToskF 16-18 1 .25 75 19 .18

Note —After Atkinson (1957) and litwin (1966)

*Lines started 10" from peg and subsequent iines were 10" apart

Table l shows how the model works fairly well m predicting

the distnbution of actual choices m an achievement situation
in which college students had to decide how far away to stand
from a peg in trying to throw rmgs over it. Subj"ective estimates
of probability of gettmg ringers from various chstances (Pa) X
the assumed incentive value (la) of rmgers made from those
distances predicted a bow-shaped distribution of choiejes in which
most students should choose tasks of moderate difficulty (throw-
mg from lines 10-15). This in fact turned out to be the case,
though considerably fewer throws were made from the easy
distances (lines 7-9) than the model predicted, and in general
the students t<X)k greater risks than would be expected from the
theoretical me>del This discrepancy could be explained on the
grounds that they had higher n Achievement than an average
group. For other studies have shown that people with high n
AeJiievement tend to overestimate their ch£uic«s of succeeding
They find tasks to be most attractive for which the general esti-
mate of probability of succeeding may be as low as .33 (Task E in
Power motivation and risk 123
Table l ) because they thmk their personal chances erf success are
more like .50, which yields maximum attractiveness accordmg to
the model (Atkmson, 1957). Further research has shown the gen-
eral utihty of the model m explainmg how people react who are
trymg to avoid failure, why some people persist at a task more
than others, how an actual success or failure modifies the motiva-
tional determmants of subsequent choices, etc. Most recently
Wemer and Kukla (1970) have uncovered some of the reasons
why a subjective probabihty of acxiomplishment of around 50
IS so attractive to people with high n Achievement more than
most people they attnbute success to personal factors (effort
and abihty), and performance on tasks of intermediate diffi-
culty "is more hkely to yield mformation about one's capabihties
than selection of tasks which are very easy or very difficult"
(P- 19)
But obviously all of this extensive research effort restncts
its focus largely to workmg on tasks, achievement mcentives,
probabihties of accomplishmg somethmg yourself, etc Yet Atkin-
son's model of choice behavior which is based on earher formu-
lations of Lewm, Tolman, and Rotter is completely general So
it seemed highly desirable to us to see how well it fits choice
behavior where the dommant motive and mcentive are for power.
In this way its generality might be explored empirically.
Recent studies (D. G. Wmter, m McClelland et al, 1972)
support the conclusion that the mcentive for a power-oriented
person is to "have impact," to "stand out" m some way, or to
be considered important. He may pursue this mcentive m a
vanety of ways—by trymg to wm arguments (Veroff, 1957), by
collecting prestige possessions (Winter, in McClelland et al,
1972), by nurturmg others, by bemg aggressive, or even by
dnnldng to increase fantasies of personal power (McClelland
et al, 1972). Domg well at a task, the achievement mcentive,
may help a person "stand out" or attain a power incentive, but
only in a pubhc situation where others will notie» the superior
performance and probably only where the person can stand out
relative to othrars—that is, can be judged superior to others. Thus
pubhc competition involves both achievement and power in-
centives, but it is known that the person with high n Achieve-
ment is if anythmg "thrown off" (Atkmson & Reitman, 19^) by
124 David C. McClelland and Robert I. Watson, Jr
power mcentives (pubhc competition) and it might be assumed
that if the person with high power motivation can find a way to
attam his goal of "standmg out" without workmg or achieving,
he will do so Contrariwise, a pure achievement situation will
have no special attraction for him, since it gives him no chance
for the public recognition he wants This analysis suggests the
first hypothesis to be tested
Hypothesis l In che>osmg tasks to work on m pnvate, persons
with high n Achievement will che>ose tasks of mtermechate
difficulty, but persons with high n Power will show no particular
preferences for tasks as a function of their difficulty. The hy-
pothesis smiply tests the assumption that while the Pa X l-Pa
model will predict task choices for people with high n Achieve-
ment, it will not predict task choices for people with high n
Power, because very simply l-Pa (the difficulty of the task)
IS not an incentive for them They are mterested m standing out
m some way, and the choice of tasks differmg m difficulty is
irrelevant to that goal when no one will know what levels of
cJifficulty they have chosen
Next it IS essential to find a game situation in which people
with high n Power can pursue their incentive and pee)ple with
high n Achievement cannot. Cambling is an obvious choice be-
cause previous studies have shown n Achievement is not as-
sociated with moderate nsk taking in chance situations, pre-
sumably because the mcentive to demonstrate success by per-
sonal effort is not present (Littig, 1963) Furthermore public
gamblmg, as m roulette, should be attractive to people with
high n Power because they have a chance to "stand out" m
front of others by bemg "big nsk takers'* or "big winners." Mc-
Clelland et al (1972) m fact report that high n Power among
lower-middle-class men is significandy associated with the ten-
dency to gamble more. If the analysis of the power mcentive is
correct, however, people with high n Power should not be as
drawn to "plungmg," taking big risks, m private, where no one
will notice, as they are m pubhc. The second hypothesis foUo-w^
from this analysis.
Hypothesis 2 In a public, but not in a private, gambling
situation, people with high n Power will take more extreme
Power motivation and risk 125
nsks, whereas people with high n Achievement will show no
particular preferences for bets as a function of their risk
The current study was designed to test these two hypotheses
but it also permitted checking to see whether people with other
dommant motive dispositions would seek other goal satisfac-
tions either under working or gamblmg conditions That is,
smce the subjects were given a Thematic Apperception Test to
get the n Achievement and n Power scores, it could also be
scored for the two mam subtypes of power motivation, p (for
personal) and s (for social) Power (cf McClelland et al, 1972)
and for n Affiliation No strong hypotheses could be made as to
how people with these motive dispositions would react to the
experimental conditions, but previous work has suggested that
p Power might be more related to extreme nsk takmg and
gamblmg behavior than s Power and that high n Affiliation
might be associated with strategies of trymg to avoid or with-
draw from mterpersonally competitive situations, as m public
gamblmg Terhune (1968) found, for mstance that teams made
up entirely of people with high n Affiliation seemed to be trying
to refuse to participate in a competitive interaction game simula-
tion by showing a lower frequency of all types of acts (cooper-
ative, competitive, etc )


Seventy-two volimteers participated m the study (36 males and 36
females) They were all Harvard summer school students, recniited
through advertisements posted on campus Their ages ranged from
16 to 45 with a mean age of 20 6 They were paid $2 cx) for attendmg
two sessions, eacii of which lasted about an hour
Firs* Session Motive Measurement
In the flrst session, the suhjects who appeared m groups of 20-30
were first shown four shdes of standard Thematic Apperception Test
pictures and asked to wnte a brief imagmative story to each one As
part of another study, they also wrote a fifth story to Homer's fear of
success verbal cnie (1970). The four pictures were- Man at drawmg
hoard, ship's captam, two lawyers, man and woman m bar. They were
known to cemtam cues which would ehcit a wide range of power and
126 David C. McClelland and Robert I. Watson, Jr.
achievement concerns although they were weak m affihation cues
The stones were scored blmcUy for n Achievement (sconng —l for
unrelated imagery), n Power, usmg Wmter's revised system (1972),
p Power, s Power, and n Affihation Tlie coders had previously estab-
lished their expertness by demonstratmg agreement with expertly
coded stones of 88 percent for n Achievement, 86 percent for n
Power, 81 percent for s Power and p Power, and 88 percent for n
Affibation Each TAT story booklet was numbered beforehand, and
the subjects used these code numbers throughout the experiment to
preserve their anonymity After completmg the TAT, each subject
filled out a biographical information and activities questionnaire, as
part of another study Then they were asked to return the next day
or the day after that for the final session which mvolved the roulette
bettmg game and the achievement task

Second Session The Two Tasks

When the 14-20 subjects amved for the second session, they were
divided randomly mto two groups, one to take part m the bettmg game
first while the other performed the achievement task Upon complet-
mg the first activity, the groups switched rooms and proceeded with
the remammg task Seven to ten subjects went through each of the
tasks as a group This meant that order was systematically varied so
that half the subjects gambled first, and half worked on Ae achieve-
ment task first. There were ten groups m all, two run at eac^ of five
Achievement task The achievement task segment of the expenment
mvolved two different test-hke operations used by Homer (1968)
The first was a long set of mental anthmetic problems while the second
consisted of three sets of triangular puzzles that the subjects were to
trace After mducmg an achievement onentation by descnbmg the
tasks as reflectmg "abstract and mtellectual abihty," the expenmenter
told the subjects that performance on the two were highly correlated
so that after seemg how well they did on tiie first task, he could de-
termme what tasks he should give a person whicii would correspond
to the level of difiBculty they wanted to work at. In his mtroduction to
the first task, the expenmenter said, "111 use the first test, which youil
now work on, to tell you how difficult the next task will be for you
m tenns of tius particular skill Do your best" Once the subjects
finished the first task (anthmetic problems) the experimenter said.
Now well go on to the test mvolving abstract reasonmg As you
see, tiiere are many different stacks of puzzles on the table
Power motivation and risk 127
What I'd hke you to do is choose how difficult a set you want
to work on 111 then use your score on the test you just finished
to determme which stack of puzzles matches that level of diffi-
culty for you personally
The expenmenter then explamed how to do the tnangular puzzles by
tracmg all lmes without retracmg, usmg a Ime segment more than
once, or hftmg the pencil from the paper He then asked them to
choose the level of difficulty of the puzzle they would hke to try
tracmg first, makmg it clear they would also have two more such
choices to make They chose the difficulty level by circlmg a number
between l ("very easy for you") and 9 ("very difficult for you")
spaced across a hne segment He then matched their choice of chffi-
cnilty level with a puzde set aceordmg to a conversion table which
he had told the subjects used their score from the first test to de-
termme what puzzle difficulty would correspond for them personally
to the level of difficulty at which they wanted to work After the sub-
jects received their puzzle sets and worked for two mmutes, the same
procedure was followed for the next two sets of puzzles Thus each
subject made three successive choices of level of chfficnilty of tasks
to work on, when he believed that the difficulty was defined in terms
of his demonstrated abihty Actually the expenmenter did not adjust
what puzzles he gave them on the basis of demonstrated abihty, but
simply gave each subject the same puzzle for the level of difficulty
chosen Note that all of the choices were made pnvately so that no
subject knew how well other subjects were domg or what levels of
difficulty they had chosen
Roulette game. A group settmg was deliberately used for the
bettmg game since it was hypothesized that m a pubhc bettmg condi-
tion subjects with high n Power would attempt to impress the others
by takmg high nsks at the roulette table The setting, a very nicely
furnished, comfortable lounge, was as naturalistic as possible The
expenmenter acted as croupier and attempted to keep the game
movmg durmg the actual play After all tiie subjects were seated
aroimd the lounge with the roulette table m the middle of the room,
the expenmenter explamed the game and how to play it
We are now gomg to play a bettmg game based on roulette
As most of you know, roulette is a game of chance based on
numbers and color. There are two colors, red and black, and 36
numbers on the wheel and the board where you place your
bets There are many different ways to bet in this game. One
can bet cm a single number, which is the hardest bet, or on a
128 David C. McClelland and Robert I. Watson, Jr.
color (red or black so that you have a 50 50 chance of wmnmg),
or on different combmabons of numbers. I h e odds range from
J2 to 1 up to 35 to 1 On the next page is a bnef descnption of
the different types of bets, the odds for each bet, and tiie pay-
off for wmnmg that bet Read this over carefully There is
also a diagram of the roulette playmg board which shows
you how to place the vanous bets
As soon as he had read through the vanous odds and methods for
placmg bets, the expenmenter demonstrated the bets on the table
Each subject was then given a dittoed sheet which was a rephc^ of
the bettmg doth they would be playmg on The subjects were then
directed to place 15 bets on the sheet by markmg a dot m pencil m
the same position a chip would occupy for that particular bet This
was the pnvate bettmg condition Once they had marked these 15
bets, they were asked to step up to the table Eacdi subject was given
10 chips for ten bets. It was explamed that they could place only
one-chip bets and that they should choose one bet at a bme from their
ongmal 15 Those they finally chose to play were circled so that the
expenmenter would have a record of them. The game began with the
expenmenter actmg as the bank, paymg out the chips to the subjects
when they won and collectmg all the losmg bets Obviously some
subjects lost most of their cibips while others won On the whole,
interest m the game was high m most sessions and many subjects
were cx>mpebbvely mvolved m whether they won or lost even though
no money actually passed hands After the subjects had placed ten
bets, the expenmenter collected their chips and either dismissed them
or took them to another room to work on the achievement task


As expected the major motive scores were uncorrelated (Table

2). The p and s Power subscores are moderately related to the n
Power total, but remain sufficiently distinct for possible differ-
ences m relationships to the nsk-takmg variables to appear.
Also as expected the pictures yielded considerable variability in
individual scores for n Achievement and n Power. The mean
for n Achievement appears low because —1 is scored for each
unrelated story so that the lowest possible score is —5 whereas
it is o for n Power and n Affiliation. The pictures did not evoke
madx affiliation imagery. Only 15 of the 72 subj'ects included any
thoughts about affiliation m their stories.
Power motivation and risk 129
Table 2 Intercorrelabons and means of the mobve measures (N = 72)

n Ach n Power p Power s Power n Aff

n Adi , , - 14 -01 - 12 11
n Power 55*** 37** .00
p Power — 26* 01
s Power — -06
Mean 82 6 29 1 11 64 54
SD 4 04 3 25 I 00 91 1 50

*p < 05
**p < 01
***p < 001

Table 3 Correlabons between mobve scores and nsk takmg m work

and gamblmg (N = 72).

Level of risic ciwsen

(Figure tracing) Gambling (Roulette)

MoHve type Moderate* High» low° Moderate<i High'

B Power -03 .13 - 20 -22 24*

n Achievement 31** - 14 05 32** - 45***
n Affiiiation 19 -04 31** - 17 -09

^Number out of three problems chosen of moderate difficulty (levels 4-6 on scale of 1-9)
''Sum of difficulty levels of three problems chosen
'Number out of 10 bets placed at odds of Vz or 1 to 1
•'Number out of 10 bets placed at odds of 8 or 11 to 1
•Number out of 10 bets placed at odds of 17 or 35 to 1
*p < 05
**P < 01
***p < 001

Table 3 presents the data necessary to test the two main

hypotheses. A fuU correlational analysis showed that neither sex
nor order (roulette or achievement task first) bore any significant
relationships to the variables involved m the two mam hypothe-
ses. So more complex analyses of vairance seemed unnecessary.
The first hypothesis is clearly confirmed The subjects high in n
Achievement significantly more often chose tasks of moderate
difficulty (levels 4-6 on a scale of 1-9) relative to their own
abihty. This of course is a frequently reported findmg, but it is
reassuring to know that under the conchtions of the present ex-
penment the n AeJiievement scores obtamed are vahd to this
extent. The n Power score did not relate significantly to moderate
130 David C McClelland and Robert I. Watson, Jr.
nsk-takmg m a work situation, as predicted, nor did it relate to
a tendency to choose extremely difficult (risky) tasks. These
results are consistent with the hypothesis that tie mcentive for
people with high n Power is to stand out or be considered im-
pressive m some way, rather than to work on somethmg which
they could do well at. Smce difficulty levels were chosen in
private, those with high n Power could not even get credit in
the eyes of others for trymg to solve very diflBcult puzzles. The
subjects with high n Affihation were also not particularly moved
to choose high or moderate difficulty levels m the achievement
The picture is qmte different for nsk takmg m roulette.
Hypothesis 2 is only partiy confirmed, for quite unexpectedly n
Achievement did relate significantly to moderate nsk takuig
(odds of 8 or 11 to 1) m a game of chance. It also led to avoid-
ance of extremely risky odds (17 or 35 to 1) One can only assume
that this situation somehow evoked the charactenstic tendency
of people with high n Achievement to choose moderate nsks,
even though they clearly could not affect the outcome by person-
al effort. It may be that in other gamblmg situations, like Littig's
(1963), the probability preferences were stated under such
complex con(iitions (betting for poker dice hands) that the
tendency to generalize from achievement to nonachievement
situations was elimmated
However, the more crucial aspect of Hypothesis 2 was con-
firmed: subjects with high n Power did significantly more often
place the most nsky bets (17 or 35 to 1) m the public gamblmg
situation The tendency was especially marked for bettmg on
single numbers where the odds were known to every one to be
35 to 1. Of the 32 subjects above the median in n Power (score
7 or more) 69 percent placed at least one such bet as compared
with only 43 percent of the 40 subjects at or below the median
in n Power (jt* = 5.01, p < .05).
What about choosmg risky bets in private? There the results
are not striking, but they are definitely m the expected direction.
The n Power score correlates only msignificantiy (f = .18) with
the number erf extremely risl^ bets chosen privately to be among
the 15, 10 of which they would actually play m pubhc. The
reason the corrdiation is not lower (in fact it is fairly close to tlie
Power motivation and nsk 131
.24 correlation with nsky bets placed) seems probably due to
the fact that those with high n Power knew they had to choose
nsky bets if they were gomg to be able to play them later
A more direct test of the hypothesis mvolves observmg how
many of the extremely nsky bets chosen were actually used
under conditions of pubhc competition To make the test, it is
necessary to omit those 12 subjects who chose no extreme bets
(17 or 35 1), smce they couldn't then place any later. Of those
who chose extreme bets, 61 percent of those with high n Power
used all or all but one of them, contrasted with only 34 percent
of those with low n Power (x^ = 4 09, p < .05). In other words
under the stress of public competition, or under conditions where
they had a chance to be observed as being 'Tsig gamblers," those
with high n Power were more likely to use the extremely risky
bets they had available than those with low n Power
No significant correlations were obtamed for the p and s
Power scores and risk takmg under work or chance conditions
However one near significant findmg for p Power is worth re-
porting because it helps delmeate the kind of person who sees
power m terms of personal dommance. Of the 49 subjects at
or above the median m p Power (score 1 or more) 63 percent
chose more than the median number (4 or more) of very risky
bets (17 or 35:1) as contrasted with 39 percent of the 23 subjects
With a zero p Power score (^^^ = 3.74, p < 06). However, this
difference practically disappears for very risky bets placed under
conditions of pubhc competition. Forty-mne percent of those
with high p Power scores placed two or more very risky bets
as contrasted with 39 percent of those with low p Power scores
(5C^ = 57, p = n.s.). Nor is there any tendency for those with high
p Power scores, like those with high n Power scores, to use more
of the very risky bets they have chosen What is interesting about
this findmg is that it is the reverse of what was found for people
wiiii high n Power who tended not to chcrase more very nsky
bets in private but to place more of what they had chosen in
pubhc. It looks as if people high in p Power might be boasters,
ten(hng to imagine or represent themselves as bigger risk takers
than in fact they tum out to be under the stress of public com-
petition. Such an interpretation is consistent with other findings
which show the person high in p Power to be oriented more
132 David C. McClelland and Robert I. Watson, Jr.
toward self-aggrandizmg actions (drmking, collecting prestige
supphes, aggressive thoughts, McClelland et al., 1972) than
toward actual social competition or leadership behavior m the
real world
Fmally Table 3 shows a strong tendency for those subjects
who give affiliative responses to these picture cues to place bets
with the lowest risk {% or 1.1) in the gamblmg situation. This
finding is quite consistent with others whicJi show that people
with high n Affihation tend to dislike and avoid interpersonal
competition. Smce they have to play roulette m this experiment,
they avoid competition the only way they can—by choosmg bets
with as httie nsk as possible
Figure 1 has been constructed to give a somewhat better
descriptive picture of the results m the roulette expenment which
he behmd the correlations m Table 3. Since we are primarily
interested in the goals or incentives sought m the situation by
individuals high in one motive or another, we decided to simplify
the pictoxo'iSy eliminating subjects who were high in more than
one motive and whose risk takmg would therefore be under the
competing influences of two or more mcentives Thus we found
15 subjects who were above the median in n Power but below
the me(han m n Achievement and in n AflSliation. To smooth
the curves of how they distnbuted the 150 bets made among
them (10 by each subject), we combmed adjacent bets (}^ and
1:1, 2 and 5 1, 8 and 11 1, and 17 and 35 1) and plotted the
proportion of the total number of bets placed at each of these
four levels of nsk takmg. As can be seen in Figure 1, those who
scored high only in n Power placed more of their bets toward
the more risky end of the contmuum, those with only high
n Achievement concentrated their bets at the 8 or 11.1 level
and those with high n Affihation only (just three subjects,
or 30 bets) concentrated their bets at the low-risk end of the
(X)ntmuum These curves simply show m another way the more
statistically sound findings presented in Table 3 However they
do suggest an interpretation of the incentive charactenstics erf
a social or task situation whicJi is somewhat different from the
one given often by socaal psychologists. The curve for all bets by
all aibjects in Figure 1 is definitely bow shaped with a greater
proportion of the bets being placed at moderate odds of 2, 5,
Power motivation and nsk 133
. 6 0 I-

.50 Bets by Ss with Ji

high n. Ach only <*

"i .40 Bets by Ss with
high n. Aff only


Bets by Ss with
.10 high n Power only

.60 .50 .40 .30 .20 .10 .00
Probobility of success

t t f t
8 or II I 17 or 35 I
or I I 2 or 5-1
Figure 1 Proportions of roulette bets placed at vanous odds by all
subjects and subjects high m n Power, n Achievement, and n AflSliation
8 and l i . i rather than either J* and l . i or 17 and 35.1. The
obtamed distnbution of bets differs significantiy from a chance
distribution which would lead one to expect an equal number in
all categoric (x* = 107.3, df = 4, p< .001). So one might be
inclined to infer that the demand characrfieristics of roulette are
sucdi that it (X>ntains a greater m(entive for moderate risk taking
rather than low or high risk takmg ^ However, it seems clear
from companng the curves for those high in n Achievement
2 Actually die smoothed curve is a htde misleading m detail because & e
proportion choosing the 17 1 altemative is actually lower ( 5 3 percent) than
tibe 35 1 alternative ( 1 7 3 percent) But this does not aSect the general argu-
134 David C. McClelland and Robert I. Wotson, Jr.
and in n Power that the overall distnbution of bets could just
as well be a simple function of the sample of people playing the
game If they happened to be generally high m n Power and
low in n Achievement and n Affihation, the over-all curve would
lose its bow-shaped characteristic. If they were high m n Affiha-
tion and n Power, it might became approximately flat The fact
that it is bow shaped here might well be attributed to the fact
that this particular sample of students, gomg to Harvard sum-
mer school, IS probably fairly high m n Achievement Professional
soldiers and actual roulette players might yield quite a different
distribution of bets, since they are probably more power-oriented.
So many of the findmgs of experimental social psychology may
be seen not as relationships between a situational vanable (de-
gree of nsk) and a typical human bemg's response (choice of
moderate risk) but as a statement about the (usually unknown,
or unspecified) personahty characteristics of the sample of people
participating m the expenment. This is another way of saymg
that the Pa Xia function m Atkinson's model (Table l) obviously
has no power to ehcit differential approach behavior if the Ma
term is reduced to zero, l e, if there are no motivational dis-
positions toward achievement m the population sampled. Yet
many generalizations m psychology are made lgnormg the
samplmg of personahty charactenstics involved, just as one
might mcorrectly conclude on the basis of the overall nsk-taking
curve m Figure l that there is a general tendency for people
playmg roulette to choose moderate risks


The results confirm the general hypothesis that people vdth

different dominant motives seek out mcentives m game situations
which satisfy those motives. Those with high n Achievement
seek moderately difficult tasks in work situations as the outcome
will be most rea(hly attributed to their effort to do well. Those
with high n Power take extreme risks in roulette because doing
so satisfies the incentive they seek—to stand out in some way.
Let us examme the n Power case in more detail to see whether
or not it fits readily into the Atkinson model statmg that choice
is a function of Motive X Expectan(;y X Incentive value.
Power motivation and risk 135
If the incentive of people with high n Power only is just to
be noticed or admired for being big gamblers, why don't they
make all 35 1 bets, when m fact only 27 percent of their bets
are m this category? Three explanations come to mmd
1. Our motive measures contain error Even those high m n
Power only have some residual concerns (and possibly higher
than measured) for achievement and affiliation which lead them
to distnbute their bets m other categones. Still the proportion of
extremely nsky bets is so low that one may legitimately wonder
if it can all be attributed to measurement error
2. Those high in n Power only may be attracted primarily
to makmg very risky bets but they are unsure as to whether m
fact they will be noticed by others when they make them From
this viewpomt the key probability for them is not the probability
of wmmng but the probability of bemg noticed and admired
for being a "big gambler " Unfortunately we have no measure
of how the latter probabihty varies m this expenment and we
can only note that it may be a factor which should be taken
mto account m future work.
3 Fmally it seems highly likely that the mcentive for those
with high n Power only mvolves not only "risking big" but
"winning big " So such people may take the probability of wm-
mng mto account just as the Atkmson model predicts they should
m placing bets at (hfferent odds.
Table 4 has been prepared to check on the adequacy of
vanous models for predictmg the actual (hstnbution of choices
for bets of different nsks by subjects high m n Power only
Obviously the Atkmson model for people high m n Achievement
does not work here because Ps X l-Ps yields maximal approach
tendencies m the low- to moderate-nsk range, and the subjects
high in n Power only are more attracted to extreme nsks. This
IS just another way of saymg that difficulty or "challenge" (l-Ps)
is not an incentive to people with high n Power, either in the
work or the gambhng situation.
But of course the E X I mcxiel might still hold if we could
identify some oiher incentive which is operating for subjects
with high n Power only. One obvious possibihty is that they
are attracted by the size erf the payout—the number of chips to
be won if their bet is su(xessful. Table 4 shows the calculations
136 David C McClelland and Robert I Watson, Jr.

Table 4 Approach to bets of diflFerent nsk as a function of expectancy

(Ps) and mcentive (I) for subjects high m n Power only

Percentage of bets placed

1 2 3 4 5 6 7
Payout from
Odds Ps X 1* = valence Predicted Obtained l-Ps only

Vi and 1 1 .58 X 75 = 435 13 05 .14

2 and 5 1 25 X 35 = 875 26 27 25
8 and 11 1 096 X 95 = 912 28 29 30
17 and 35 1 042 X 26 = 1 092 33 39 31

X" = 9 0 n.s 12 1
P < <D5

'Average payout in chips per chip successfully wagered

which predict the distnbution of bets placed usmg this mcentive

m the E X I model The first column gives the odds or payouts
per chip bet combmed for adj'acent categones For example, if
he places a chip on red, he has a fifty-fifty chance of vraonmg
smce half of the numbers are red and half are black If he wins,
he receives one chip for each chip bet. This is explamed as a
1 1 bet. If he bets on a smgle number, he has one chance m
36 of winning, and receives 35 chips if he wins Column 2 gives
the average probabihty of success (Ps) for combined odds
categories which is close to but not identical with the odds, smce
the house keeps a percentage Column 3 defines the mcentive
vanable (I) not as l-Ps but as the average payout in chips for
a win (size of mcentive) The product of columns 2 and 3 gives
the attractiveness or valence of various bet opportunities usmg
this E X I model To translate these mto predictions of the
proportion of bets placed m various categones, it is necessary
to assume that all bets have some attractiveness at a given
moment and that the frequency with which one type wQl be
chosen will be a function of its attractiveness in proportion to
the sum of the attractiveness of all bets Thus the valence of the
most extreme bets (1092) is 33 percent of the sum of the
valences of all bet opportunities. So the model pre(hcts that 33
percent of the bets placed will be of the extremely risky variety.
The predicted distribution of bets (column 5) can then be com-
Power motivation and risk 137
pared with the obtamed distribution (column 6) and a Chi-
square test computed as an index of the goodness of fit of the
predictions by tiie model. As Table 4 shows, the fit is rather
close, the Chi-square value actually being below the 05 level,
showmg that the obtained values might well have deviated to
this extent from the prechcted values by chance It looks as if
the subjects with high n Power only were functionmg accordmg
to an E X I model m which "winnmg big" was important to
them. That is, their bets were not only determined by the size
of the payoff but by the probabihty of attainmg it as well
Unfortunately there is another possibihty that predicts the
results almost as well As noted earher, they might also be mter-
ested m "nskmg big" since that will also gam attention from
others If this were their main incentive, the probabihty of
wmmng would not enter m to modify their choices How well
does such a model predict their bettmg behaviorf* Column 7
shows It makes the simple assumption that nsk (l-Ps) alone is
determmmg their choices and shows how their bets would be
distnbuted proportionally based on such an assumption. The
fit with actual choices is not as good as for the E X I model, the
X^ test actually showmg a significant deviation from chance, but
it IS surpnsmgly good. It certainly is not so much worse than the
E X I pre(hctions that one would feel much confidence in m-
sistmg tiiat for people with high n Power only, it is essential to
assume that they take probabihty of v^onmng into account m
making their choices, the way people with high n Achievement
do. It IS possible that people vsath high n Power are attracted
to nsky situations per se regardless of the expectancy of win-
nmg, if taking a nsk bnngs them the attention and re(X)gmtion
they seek. As noted earher, the crucial factor here may be the
profcabihty of bemg noticed rather than the probabihty of win-
Incidentally, it clearly is nsk that attracts them here rather
than size of payout. One c;an also attempt to predict the distribu-
tion of bets based on (xjlumn 3 alone (payouts), Imt the ex-
pected distribution is then very different from what is obtained
(x" = 76.8, df = 4,p< .001) For example if payout alone were
determining their (jioices one would expect 65 percent of their
bets to be in tiie riskiest category instead of the 39 percent ob-
138 David C. McClelland and Robert I Watson, Jr.
tamed. So subj'e(rt5 with high n Power seem not to be mterested
m mcentive size per se, but rather m the attention they get from
taking the big nsk that goes with trymg for the large payouts.
While the gamblmg expenment clarifies what subjects high
m n Power are seeking, it introduces considerable confusion
into the picture of what governs the choices of subjects high m n
Achievement only m such a situation The E X I model predicts
for them that they should prefer the shortest odds For example,
m the lowest category of risk, the probability of success is 58,
which multiphed by l-Ps or 42 yields a value that is 43 percent
of the sum of the attractiveness of aU odds alternatives usmg
this model Thus 43 percent of the bets should be at this level
whereas m fact only 11 percent were actually placed at this level
by subjects high m n Achievement only As previously noted,
other investigators have reported results more m lme with pre-
diction, m chance situations subjects with high n Achievement
have preferred the shortest nsks they can take, which is not the
case here. We can only assume that our subj'ects somehow
generalized their tendency to choose intermediate risks m work
situations In terms of the E X I model, this could be taken into
account only by assummg that they more or less ignored the
stated probabilities of winning and superimposed their ovm
subjective probabihties of success, such that short odds were
considered "easy," middle odds "moderately nsky," and long
shots "difficult." Thus while the objective Ps for odds of H and
1 1 IS .58, maybe the subjects with high n Achievement acted
as if it was .80 (easy) meaning that the E X I attractiveness of
these bets (.80 X .20 = .16) would only be in the low range,
explaining the fact that they placed only l i percent of their
bets in this category. Similarly the 8 and 11 1 bets may have
been thought to be moderately risky ( P s ~ 40) whereas in fact
Ps = .10 hi this way the strong preference of the subjects with
high n Achievement for these bets (50 percent placed in this
category) could be explained (.40 X .60 = .24). Nevertheless
It IS a httie hard to believe tiiat the subj'ects with high n Achieve-
ment have so wildly distorted the actual probabilities of success
and we must simply confess that their behavior m the gambling
experiment is not readily explainable according to the E X I
model. At any rate further speculation seems pointiess without
Power motivation and risk 139
more data on what subjects really feel their chances are of
Manning when they place each bet
To summarize, the expenment has clanfied the contrastmg
incentives which influence nsk-takmg behavior for subj'e(rts
strongly motivated for achievement, for power, or for affiliation.
Those high in n Achievement set goals pnvately for themselves
of mtermediate difficulty so that they are hkely to get a sense
of personal accomplishment Those high in n Power seek to stand
out pubhcly and do so m gamblmg by taking more extreme
nsks Those high in n Affihation try to avoid public competition
by takmg low risks. The Atkmson model predicting choice of
risk by multiplying Expectancy X Incentive works well when
an achievement mcentive is sahent, improves prediction of
behavior shghtly when a power incentive is salient (eg nsk
takmg in gamblmg), but does a poor job of predicting how
people onented toward achievement incentives will react when
a power mcentive is sahent, as in gambling
Atkinson, J W Mofavational determmants of nsk-takmg behavior Psychologicd
Revtew, 1957, 64, 359-372
Atkinson, J W , & Reitman, W R Perfonnance as a function of motive strengA
and expectancy of goal attainment Joumd of Abnormd and Sodd Psy-
chology, 1956, 53, 361-366
Homer, M S Sex difFerences m achievement mobvataon and performance m
competitive and noncompetitive situations Unpublished doctoral dissertation.
University of Michigan, 1968
Littig, L W. Effects of motivation on probabihty preferences Joumd of Per-
sonahty, 1963, 31, 417-427
Litwin, G H Achievemrait motivation, expectancy of success, and risk-taldiig
behavior In J W Atkinson and N T Feather (Eds ), A theory of achieve-
ment mottviOton New York Wiley, 1966
McClelland, D C , Davis, W N , IKahn, R, & Wanner, H E The dHiiking num.
New Yoric Free Press, 197a
Terhune, K W Studies of motives, cooperation and conflict widun laboratory
microcosms In G H Snyder ( £ ^ ), Studies tn interruOiorud confhct Buffdo
Studtes, 1968, 4, 29-58
Veroff, J Development and vahdation dF a projecttve measure of power motiva-
tion Joumd of Abnormd and Sodd Psychology, 1937, 54, 1-^
Wemer, B , & KvJda, A An attributional analysis of adiievement motivation
Joumd of Personality and Sodd Psychology, 1970, 15, l - a a
Winter, D G. Hie need for power in coOege men In D C McQeDand. W N
Davis, R. KaUn, & H E. Wanner, The drmking num. New York. Free Fiess,
Manuscript received November 30, 1971