Vous êtes sur la page 1sur 10

Fundamenta/ measurement concepts

A review of the fundamental concepts


of measurement
L. Finkelstein* and M. S. Leaningt
* Chairman, I MEKO Technical Committee for Higher Education (TC-1), Professor of
Measurement and Instrumentation, The City University, London EC1
tAcademic Department of Medical Physics, Royal Free Hospital School of Medicine,
University of London, London NW3, and Department of Systems Science,
The City University, London EC1

The paper surveys the current state of the theory of the fundamental concepts of
measurement which is based on the model theory of logic. A brief review is given of the
historical development of measurement theory. The model-theoretic definition of
measurement is presented, together with a discussion of representation and uniqueness
conditions. Nominal, ordinal, extensive and interval measurement structures are out-
lined. The classification of scale types and the problem of meaningfulness are con-
sidered. A survey is given of conjoint and derived measurement. A brief review is made
of the applications of measurement theory. Consideration is given to the treatment of
uncertainty. The setting-up of systems of scales of measurement for a domain of
science and its relation to theories for that domain are discussed. It is argued that
measurement as defined is related to other forms of symbolic representation such as is
involved in computer data representation and natural language.

1 Introduction should be additive, and this implied that measurement of


non-additive psychological attributes was fundamentally
Measurement, the objective representation of our em-
impossible. The same conclusion was arrived at for non-
pirical knowledge of the world by numbers, is an essential
physical properties in general, in a report to the British
element of all science. The techniques of measurement are
Association (Guild, 1938), which was a substantial
highly developed in the physical sciences and are gaining an
challenge to the scientific basis of social and behavioural
increasing role in the behavioural and social sciences. The
sciences.
logical and philosophical concepts underlying measurement
In the social sciences, Pareto, in the nineteenth century,
are fundamental to its validity in all sciences. The explicit
originated an ordinal theory of measurement for economic
application of these fundamental principles is essential for
preferences, or utility. Armstrong (1939) showed that
the establishment of measurement methods, in particular in
numerical utility functions could be constructed if utility
those sciences at a formative stage of development.
could be compared by intervals, thereby refuting the criti-
There exists an organised body of foundational concepts
ques of Cambell and Guild. The axiomatic basis for utility
and principles, which is generally referred to as measure-
theory, propounded in the classic work of yon Neumann
ment theory. The present paper presents a review of this
and Morgenstern (1944), was a great stimulus to the con-
theory.
tinuing development of both theory and practice of
The growth of information processing technology has
measurement in the social sciences.
led to representation of the real world by non-numeric
The work of Stevens (1946), from the point of view of
symbols and the paper also addresses this aspect.
psychological measurement, directed the whole field of
measurement theory into new lines, particularly in the
2 History of measurement theory
definition of measurement and classification of scale types.
The study of fundamental concepts of measurement can There was generally much work on psychological scaling
be traced to the Ancient Greeks - Pythogoras, Aristotle, (Thurstone, 1947; Torgerson, 1958; Coombs, 1964). This
Euclid, Cnidus. However, it was not until the first studies in period of measurement theory is well summarised in the
the logic of mathematics and physics at the end of the nine- book of Churchman and Ratoosh (1959).
teenth century that the now classical approach to measure- Since the 1950s, measurement theory has been pre-
ment theory began. Most notable are Helmholtz's analysis dominantly concerned with the development of a formal
of counting and measurement (1887) and HOlder's formu- logical framework suitable for a wider variety of measure-
lation of the axioms for additive properties (1901). This ment structures than simply extensive and relating to
work was extended by Campbell (1920, 1928) into a measurement in many areas of science.
detailed theory of measurement for the physical sciences. H61der's axioms for additive properties (or extensive
For Campbell (1938), a necessary condition for measure- measurement) were modified by Suppes (1951) to make
ment was that the manifestation of properties or attributes them less restrictive. Tarski's concept (1954) of a relational
Measurement Vol 2 No 1, Jan-Mar 1984 25
Finkelstein and Leaning Fundamental measurement concepts
system in the model theory of logic proved to be a natural The conditions under which this theorem holds are the
vehicle for expressing and further developing measurement representation conditions. If M satisfies the representation'
theory (Scott and Suppes, 1958). Issues have ranged from conditions, there exist other mappings M', which also
teclmical aspects concerned with axiomatisation (for satisfy these conditions, related to M by a transformation f
example, Holman, 1974;Narens, 1974; Falmagne, 1975)to such that M' = f(M). It follows from (1) that:
the establishment of new measurement structures such as
conjoint measurement (Luce and Tukey, 1964) and that of Ri(q, r . . . . ) ~* Pil f[M(q)] , f[M(r)] . . . . 1t ... (2)
senti-order (Luce, 1956). Four major books devoted to All such f comprise the class of admissible transformations,
measurenrent theory are those by Ellis (1966), Pfanzagl also known as the uniqueness condition. S = (Q, N, M )
(1968), Krantz et al (Vol 1, 1971), Roberts (1979). may be termed the scale of measurement.
While there has been an explosive development of The form of a representational theory of measurement
measurement and instrumentation technology in the last outlined above also has application to the more general case
40 years, there has been virtually no interest in measure- of representation by symbols. This will be discussed further
ment theory on the part of those involved with this work. in section 10.
Since a review of measurement theory by one of us at an
IMEKO Congress (Finkelstein, 1973), there has been a
distinctive growth of work on fundamentals of measure- 4 Some fundamental measurement structures
ment by technologists. The measurement structures below are concerned with
Although there has been a general awareness of the the case of direct measurement, where only one attribute or
problem of error or uncertainty in measurement, most property is being measured. In this sense they are funda-
developments in measurement theory have been concerned mental measurement. Uniqueness conditions for these
with the case of precise measurement. A major exception is structures will be considered in the next section, in regard
the development of semi-order theories of measurement to scale type.
(Luce, 1956) which are based on the intransitivity of the in-
difference relation as reflected in the observed 'just notice-
able difference' or 'j.n.d.'. In recent years there have been 4.1 Binary systems
attempts to use probabilistic formulations to represent Nominal Let Q consist of the binary equivalence relation
measurement uncertainty (Domotor, 1969; Leaning and which is reflexive, symmetric and transitive. The em-
Finkelstein, 1980 ; Falmagne, 1980). pirical relational system is ( Q, - ) and the representation
Measurement theory, therefore, has been the interest not theorem may be stated as, for q, r ~ Q,
only of philosophers of science, but of scientists and
engineers working in a wide variety of disciplines. q ~ r ¢~ M(q) = M(r) ... (3)
An example of nominal measurement is colour
3 The representational form of measurement matching.
theory Ordinal It may be argued that a minimum binary relation
for fundamental measurement of an attribute is the
Measurement is the assignment of numbers to properties weak order relation (H61der, 1901; Campbell, 1928;
of objects or events in the real world by means of an objec- Ellis, 1966). The ability to order a set of objects/events
tive empirical operation, in such a way as to describe them. is related to the notion of the 'amount' of a property
The modern form of measurement theory is representational manifested in each object/event. In ordinal measurement
numbers assigned to objects/events must represent the R contains >-which is transitive and reflexive. Then,
perceived relations between the properties of those objects/
events. (We use the terms property and attribute inter- q > r og(q) > M(r) ... (4)
changeably.) Such a theory comprises three main sections:
the description of an empirical relational system, a Moh's scale of hardness is typical of ordinal measure-
representation theorem, and a uniqueness condition. ment (for a brief discussion, see Finkelstein, 1975a).
Measurement theory defines measurement formally in
the following manner. Let Q = (Q, R ) be an empirical 4.2 Extensive systems
relational system where Q is a set of manifestations q, r,
s, ... of the property or attribute and R = {R1, Ri, Rn} is An extensive attribute is one for which the empirical
a family of relations on Q. Let N-- (N, P ) be a numerical operation of concatenation exists on the object set Q. Con-
catenation may be formally defined as the ternary relation
system where N is the set of real numbers and P = {P1, Pi,
Pn} is a family of relations on N. Then measurement is an Re on Q x Q x Q such that, for q , r , s ~ Q,
objective-empirical operation which maps attribute mani- R0(q,r,s) ~ qor- s ... (5)
testations into numbers M: Q-+ N, in such a way that the
relations between numbers also hold between property where the symbol 'o' denotes concatenation. The funda-
manifestations. This is expressed as the following theorem. mental physical properties of mass, length and time are
extensive. Another example is probability.
Representation theorem
Let Pi be a relation on N corresponding to Ri; then, if Representation theorem
q,r .... ,eQ,
1 q ~ rC~M(q) = M ( r )
Ri(q, r . . . . ) o Pi[M(q),M(r) . . . . ] ... (1)
2 qor-s~g(q)+M(r)=M(s)l ...(6)
for all i = 1, n. In other words, if M is a homomorphism.
(The order of relations Pi and R i does not in general exceed The theorem may also be formulated with the weak order
four.) relation (Krantz et al, 1971; Holman, 1974). It may be
26 Measurement Vol 2 No 1, Jan--Mar 1984
Fundamental measurement concepts Finkelstein and Leaning
shown that the representation theorem for (closed) exten- the scale type. Stevens' (1946) original identical classifica-
sive measurement holds under the following conditions: tion was based on the applicable arithmetic on each scale.
(Note that in Table 1, M' = f ( M ) does not give all possible
1 ~ is the equivalence relation (or > is weak order)
representations for the interval or extensive property being
2 Associativity: q o (r o s) ~ (q o r) o s
measured. There are alternative isomorphic numerical
3 Monotonicity: q ~ r ~ q o s ~ ros
representations in which addition operation + is replaced by
4 Archimedean postulate: if q >- r, then an integer n exists
a more general operation ® (Ellis, 1966; Krantz et al,
such that nr > q , where 'nr' is the n-fold concatenation
1971).)
of r with itself. Detailed presentations of the theory of
The class of admissible transformations and the notion
extensive measurement and of particular extensive
of applicable arithmetic may be extended to the more
properties, such as length and velocity, can be found in
general concept of meaningfulness. This is concerned with
Ellis (1966), Krantz et al ( 1971) and Finkelstein (1975 a).
the status of statements concerning measurements on a
Alternative axiomatisations have been developed to particular scale. For example, the statement that q is twice
account for certain problems in satisfying these conditions: as long as r, where q and r are measured on an extensive
limited concatenation on Q (Krantz et al, 1971); elimina- scale is meaningful, whereas if they are measured on an
tion of Archimedean postulate with stronger structural ordinal scale it is not. One approach to meaningfulness is to
axioms (Falmagne, 1975), an upper limit q' on Q such that examine the invariance of statements under the transforma-
q' :~ q for all q' ~ Q (Krantz et al, 1971). tion M' = f ( M ) (eg, Roberts, 1979). It is clear that the
meaningfulness of a statement is independent of its truth.
An equivalent, but more direct approach is to test a state-
4.3 Quaternary (or interval) systems ment by tracing it back to the fundamental empirical rela-
In difference measurement, intervals between attributes tions. If it can be expressed in these terms, then it is
of objects are compared. Let R contain the quaternary rela- meaningful.
tion ~ where (q, r) >~ (s, t) means that the interval (q, r) A statement, which on the above basis is not meaningful,
between q and r is discriminated as being larger than or is not always discarded automatically if pragmatic con-
equal to the interval (s, t). siderations dominate. For example, it is common to take
the arithmetic mean of a set o f examination marks, which
can be assumed to be ordinal, even though the median is
Representation theorem
the strongest 'meaningful' indicator location that can be
1 (q, r) ~ (s, t) ~* M(q, r) > M(s, t) t made on such a scale.
(7)
J
Q • J

2 M(q, s) = M(q, r) + M(r, s)


This linear interval tbrm of representation is appropriate to 6 Derived and conjoint measurement
measured attributes in psychophysics (Stevens, 1946)'and An attribute or property of an object or event is said to
in utility theory (Armstrong, 1939), for example. Many be directly measurable on a ratio or interval scale if it is
results for extensive systems can be applied to interval extensive or ordered by intervals. There is a large class of
systems owing to the close similarity between a series of physical and non-physical attributes that cannot be directly
adjoint intervals and a set of concatenated objects, as was measured in such a way. Each of these attributes, however,
originally noticed by H61der (1901). is capable of inducing a weak ordering on the object/event
set. If it did not, then it can be argued that the attribute
5 Scale types and meaningfulness concept has no empirical significance (Campbell, 1928) and
is redundant.
The uniqueness condition specifies the class of M for There are two options for the measurement of this class
valid representation, viz, that the mapping M is homomor- of attribute: first, restrict measurement to an ordinal scale;
phic. Table 1 summarises the direct measurement structures or, secondly, search for subsets of interdependent attributes.
presented in section 4 together with the admissible transfor- Ordinal measurement is often satisfactory, for example, in
mations M' = f ( M ) defining the uniqueness condition and Moh's scale of hardness or in tests of mental ability. Of
course, the theory of meaningfulness (Roberts, 1979), des-
TABLE 1: Direct measurement structures, admissible transforma- cribed in the preceding section, severely limits the number
tions (uniqueness conditions) and scale types (where Z is a set of of meaningful statements that can be made with such
symbols, ordered in the case of ordinal measurement) measures.
The existence of an interdependent group of attributes,
Empirical Uniqueness class
relational of admissible Scale or property manifestations, endows the empirical relational
system Representation transformation type system Q with a rich structure that allows at least an inter-
val representation, even though each attribute may be only
( Q, ~ ) ( N, = } M' = f(M) Nominal weakly ordered independently. This is well known, in non-
or where f is any
/Z, ~ ) one--one function
formal terms, to physical scientists and engineers who
'derive' a measurement of a primary attribute, such as
( Q,~ ) iN, >) M' = f(M) Ordinal density, by measuring its component attributes, of mass
or where f is monotonic and volume, and combining the measures in an appropriate
(Z, ~) increasing
numerical law (Campbell, 1920; Bridgman, 1922; Ellis,
(QXQ, ~) iN,>,+) M'=~M+fJ, Linear 1966; Causey, 1969). In the Social Sciences and Psychology,
> 0 interval and axiomatic approach to such measurement has become
known as 'conjoint' measurement (Luce and Tukey, 1964);
iQ, ~,o) iN,>,+) M ' =c~M, Ratio
an example is the construction of a single-valued utility
a>O
function on a number of attributes (Roberts, 1979).
Measurement Vol 2 No 1, Jan-Mar 1984 27
Finkelstein and Leaning Fundamental measurementconcepts
Although there is a formal equivalence between physical M is therefore of an interval scale type. This may be proved
derived measurement and conjoint measurement (eg, for the empirical relational structure above, where only
Krantz et al, 1971), as we show below, there is a different ordinal relations are assumed on Q, QI, and Q2.
feel or emphasis in the two types which is probably related In this formalisation of conjoint measurement there is a
to the existence of a body of well tested and fully quanti- symmetry between the components such that any two of
fied theories in the former. the three attributes could be chosen for the measures MI
and M2. On this view, conjoint measurement is concerned
with the setting up of a structure for the simultaneous
6.1 Basic terms measurement of all the properties concerned.
Let Q be the manifestation set of the primary property,
the weak order relation on it, and let Q1 . . . . . Qi . . . . , Qn
6.3 n-component derived physical measurement
be the component property sets. If p, q, r , . . . are elements
of Q corresponding to individual objects/events then Pi, qi, By contrast, in derived physical measurement, there is
r i , . . . , ~ Qi are the equivalent elements in the component the notion of an asymmetry between the attribute to be
sets. If R i, i = 1, n are sets of relations on the component measured indirectly and the components' attributes for
sets, then the lull empirical relational system is given by: which measures (often on ratio scales) exist. In addition,
physical quantities are related through monomial functions,
Q= (Q, Q1,R1, Q2,R2 . . . . . Qi, Ri . . . . . Qn, Rn, >~) . . . ( 8 ) a property which has given rise to the science of dinren-
The axioms for Q required to prove the representation sional analysis (Bridgman, 1922; Palacios, 1964). To
theorems below include two which expresss the indepen- capture these properties, the more general empirical rela-
dence and essentiality of each Qi in the measurement of Q tional system of subsection 6.2 is used.
(see eg, Krantz et al, 1971). Thus the empirical relational
Q = ( Q , QI . . . . . Qn, ~ ) ...(13)
system may also be regarded as a weak order relation on the
Cartesian product : where Q I , . . . , Qn are empirical relational systems for the
components, which may be extensive as well as ordinal.
Q=(Qix...xQn, >~) ...(9)
If p, q ~ Q, p i , q i ~ Q i , i= 1,n then
Hence this type of measurement structure is referred to as
n
a 'product space' (Roberts, 1979).
From the axioms of independence and essentiality it is P ~ q "~ 1--I M'[i(pi) > I-I MTi(qi) ... (14)
i=! i-1
easy to show that the weak order relation on the product
space induces a weak order on all the component sets Qi. This representation theorem may be proved by genera-
Other axioms of importance are strong transitivity of the lising the two component additive system into n dimensions
indifference relation (or the Thomsen condition) and a (assuming the conventional conjoint structure of a weak
version of the Archimedean axiom (Krantz et aI, 1971). order relation on an n-dimensional product space). Next, a
Let M1 . . . . . Mn be real-valued functions on Q1, . - . , Qn class of isomorphic numerical representations (N, >, ® ) is
(the component measures), and let M be a real-valued func- defined through a o n e - o n e function f such that, for y 1. . . . .
tion of M 1 , . . . , M n (the indirect measurement of Q). Yn ~ N

6.2 Two-component additive conjoint measurement Yl + . . . +Yn = f - l [ f , ( Y l ) + " " + fn(Yn)] ... (15)

In the simplest tbrm on conjoint, or derived, measure- It remains to find f and fi such that the left-hand side is
ment, there are two component properties Q1 and Qe, and monomial, f = log e and fi = Ti loge satisfy this condition
M is an additive function of M1 and Me. An illustration of (Krantz et al, 1971).
such measurement would be an additive utility function
over two commodities. The empirical relational system is:
6.4 Some generalisations
Q = (Q1 × Qe, ~ ) ... (10) The n-component monomial representation theorem
(Full axioms and proof of a representation theorem may be immediately suggests a generalisation where, if ~b is a
found in Krantz et al (1971), or Holman (1971), Roberts single-valued function over M1, . . . , Mn, then
(1979) gives a proof of a simpler system in which an equal
spacing assumption is made.) p ~ q "~ ¢ [ M I ( p , ) . . . . . Mn(pn)] ~- ¢ [ M , ( q l ) . . . . . Mn(qn)]
...(16)
Representation theorem This appears to be an appropriate scheme for pattern recog-
For p , q ~ Q , p l , q l c Q a , p z , q 2 EQ2, nition (Duda and Hart, 1973; Finkelstein, 1977) and for
model-based indirect measurement in the life and social
p ~ q ~ ' M x ( p l ) +Mr(p2) > Ml(ql) +Mz(q2) ...(11) sciences. However, the problem remains of proving such a
theorem for given classes of 4~ and establishing uniqueness
Note that the LHS may be expressed as: conditions on the Mi.
(p,, pe) ~ (ql, qe) In multidimensional measurement, an attribute of an
object/event is represented by an ordered set of symbols or
numbers. The elements of this set are measurements of
Uniqueness condition component attributes as in the case of conjoint (and
derived) measurement, but the relations on the primary
m'l = ~MI + 3x, 342 = ~m2 +/32, a>O and
...(12) attribute(s) are given by the geometrical properties of the
M'=o~M+/3 where /3=/31+32 symbol or number space rather than a mapping into a single
28 Measurement Vol 2 No 1, Jan--Mar 1984
Fundamental measurement concepts Finkelstein and Leaning
symbol (number) set. This is a feature of methods of scaling presented is taken as valid and the uncertain results of
as used, for instance, in multidimensional scaling (Torger- practical measurement are analysed with the statistical
son, 1958) and Coombs' theory of data (1964). The theory of errors. The second approach attempts to include
generalisation of measurement theory to multidimensional uncertainty in the basic formulation of measurement
scaling was initially pointed out by Suppes and Zinnes theory.
(1965). The formalisation of such a system requires the
inclusion in the empirical relational system Q of a relation
R on Q (or, equivalently, QI x ... x Qn which in the geo- 8.1 Statistical theory of errors
metric sense is an m ry relation. The statistical theory of errors is concerned in the error
as the difference between a true value of a measured attri-
bute and the value of that attribute as measured (for
7 Areas o f application
example, Beers, 1953; Kendall and Stuart, 1970). It
Although a detailed review of the applications of assumes there is a defined 'true' value which, from the
measurement theory is beyond the scope of this paper, point of view of measurement theory, must be regarded as
some general analytical remarks will be made. (To some inherently uncertain.
extent applications have been covered in the historical
review, section 2.)
In the physical sciences, measurement methods have 8.2 Semiorder
been developed without a basis of the theory of the founda-
tions of measurement. The interest of physicists (such as In discussing the measurement of physical quantities,
Helmholtz, 1887; Campbell, 1920, 1928) was directed Campbell (1928) formalised the uncertainty range of the
towards providing a logical and philosophical basis for indifference relation (and hence its intransitivity) by intro-
established practice. There are no practical issues in classical ducing the concept of distance of ordering. If - is the
physical measurement which are addressed by measurement indifference relation and > is the strict order relation, then
theory. Dimensional analysis (Bridgman, 1922; Palacios, for q, r ~ Q:
1964) which is related to measurement theory is important
q - r ~ ] m ( q ) - r e ( r ) I = e < E;m t
in practice. A good discussion of the relation between the / (17)
two is given by Krantz et al (1971). q > r ~* ] M ( q ) m ( r ) [ = C > C m I
In the social and behavioural sciences there are practical
problems in the establishment of measurement scales which The effect of the intransitivity of indifference on the
call for the application of measurement theory. In particu- measurement of utility was examined by Armstrong
lar, the establishment and validation of empirical represen- (1939). The idea that intransitive indifference is related to
tation conditions is of great importance. An example is the closeness of ordering of a set of objects is reflected in the
development of the theory and practice of conjoint notion of a just-noticeable-difference (or jnd). Goodman
measurement (Luce and Tukey, 1964; Krantz et al, 1971; (1951) and Galanter ( 1956) produced methods for ordering
Emery and Brown, 1979). The strong relation between jnd data which were combined by Coombs (1964) in the
measurement theory and practice is well illustrated in the Goodman-Galanter model, where objects are ordered in a
literature (Torgerson, 1958; Coombs, 1964; Luce et al, way compatible with jnd data rather than attempting to
1965; Blalock, 1974). assign numerical values. Luce (1956) axiomatised such a jnd
Another important area of application of measurement structure, calling the resultant order system a semiorder,
theory is in decision and utility theory. Good references to There is a weak order induced by such a semiorder, for
this work are Fishburn (1964, 1968) and Roberts (1979). which there is an order-preserving real-valued function.
In technological measurement, it is not infrequent to Below we give a definition of a semiorder and a represen-
measure properties which differ from classical physical tation theorem for ordinal measurement (for more details,
quantities. For example, it may be necessary to characterise refer to Luce, 1956, 1973; or Roberts, 1970).
the quality of a surface of steel strip, or the impact strength
of coal, to give examples from the authors' experiences.
These are obvious areas to which the application of funda- Definition o f a semiorder
mental concepts of measurement theory is beneficial. It is
Let >- and - be the relational symbols for the semiorder,
not commonly used because the methods are not widely
and P and I be those in the induced weak order. (Q, >, - ) i s
known to technologists.
a semiorder iff, for a, b, c, d ~ Q:
It is an argument of this paper that measurement is
closely related to the representation by symbols in general 1 - is reflective
(for example, data in computers). This is possibly an impor- 2 a>-b,b~c,c>.d~a>-d
tant future trend which is discussed below. 3 a~-b,b >-c,b-d~not(a-dandd~c)

8 U n c e r t a i n t y in m e a s u r e m e n t Induced weak order


The theory which has been outlined in the preceding Let the relation (P, I) induced on Q be defined, for all
sections is idealistic; in practice there is no such thing as an a, b, c, d ~ Q: for all aPb iff either
exact empirical equivalence or weak order relation, or a
1 a > b , or
numerical assignment, which is free from scatter or bias.
Fundamentally, all uncertainty in direct measurement can 2 a-bandc~Qexistssuchthata-candc>b, or
be traced to the uncertainty of the basic empirical relations. 3 a- bandd~Qexistssuchthata>-dandd-b.
In measurement theory there have been two approaches to If neither aPb or bPa then aIb. It can then be proved
the problem. In the first, the theory we have so far that < Q , P , I ) is a weak order.
Measurement Vol 2 No 1, Jan--Mar 1984 29
Finkelstein and Leaning Fundamental measurement concepts
Representation theorem for ordinal measurement Finkelstein and Leaning, 1982)when the random function
m is described by discrete and continuous distributions.
There exist real-valued functions M and 8 on Q such More recently, Falmagne (1980) has worked at a proba-
that : bilistic theory of extensive measurement in more detail.
1 M is order-preserving over ( Q, P, I ); for a, b e Q: This starts from the axioms tbr a regular (or deterministic)
extensive structure and then, as above, regards aRr as an
M(a) = M(b) o aIb event in a probability space such that P(q, r) is the proba-
M(a) >=m(b) o aPb bility of observing qRr. Then, if q, r e Q:

2 a>b iff M(a)>M(b)+6(b) P(q,r)> ½ iff m(q)>m(r) ...(21)

3 a~b iff not (a >- b or b >- a), i e . . . . (18) if the concatenation q o r e Q, then
iff M(b) - 6 (a) < M(a) < M(b) + 6 (b) m(q or) = m(q) + re(r) ... (22)
6 is the numerical value corresponding to the jnd threshold. (Falmagne, 1980, uses the notation qlq2 for ql °q2), sub-
ject to the following axioms:
Further developments of the semiorder concept 1 (Q, R ) is weak order. This follows from weak stochastic
A semiorder analysis of extensive measurement was transitivity: if both P(q, r) > % and P(r, s)2_ ½, then
undertaken by Krantz (1967), where concatenation is P(q, s) ~_ ½.
regarded as a set-theoretic union on the o-algebra of Q. A 2 If P(q, r)>½, P(s, t) >½ and q o s ~ Q , thenroteQ
finitely-additive structure for utility functions was axio- with p(q o s, re t) > ½.
matised by Fishburn (1968). The graph-theoretic approach Falmagne considers two special cases. The first is a
proposed by Harary (1964) was applied by Roberts (1970, probabilistic difference extensive system:
1979) to various models of intransitive indifference struc-
tures. Axiomatisations and representation theorems for P(q, r) = F[m(q) - m ( r ) ] ... (23)
extensive, probability and conjoint systems were presented
by Luce (1973). with F an increasing function, and the second is a probabi-
Semiorder is a formal approach to measurement uncer- listic ratio extensive system:
tainty which has great power, yet there remain significant
P(q, r) = F [ m ( q ) - m(r)] ... (24)
problems. In the representation theorems, the axioms are
expressed in terms of the ideal induced relations, rendering
them fundamentally untestable. Further, the empirical 8.4 Other approaches
division between >- and ~ on a class of objects is most
likely to be blurred in a stochastic or fuzzy fashion. Fuzzy set theory (Zadeh, 1965) offers a framework for
formalising measurement uncertainty but does not appear
to have been used for this purpose. A possible reason for
8.3 Probabilistic theories of measurement
this is the problem of interpreting the empirical meaning of
In the classic theory of measurement and those based on the fuzzy set membership function. In quantum mechanics,
semiorder, the empirical relations are perceived exactly. Prugovecki (1974) proposed that measurements in QM may
This does not allow for situations where the relation aRr be expressed as a probability measure on fuzzy events.
holds at one instance but not the next. Such a relation may
be characterised by its relative frequency of occurrence.
9 Measurement considered in terms of a theory
Demeter (1969) developed a theory of probabilistic
relational systems and applied this to ordinal measurement, The setting up of scales of measurement is commonly
outlined below. presented in the literature of measurement theory in terms
Suppose ((2, R ) is an ideal empirical relational system of the establishment of scales for individual quantities in
and ( ~ , A, k ) is a qualitative probability structure where A isolation.
is a Boolean algebra over which (Q, R ) is defined and R is The scales for some quantities are established on the
the weak order relation. ~ is the whole sample space and~ basis of direct or fundamental measurement. That is, for a
the weak ordering of probability. Measurement is defined as quantity Q = (Q, R ) we find a homomorphic mapping M
the random function into a numerical system N = (N, P) giving us a measurement
scale S - - ( Q , N, M) without any other quantities than Q
m: Q-+N being involved in the establishment of the scale.
such that, for all q, r e Q: Scales of measurement are thus directly or fundamen-
tally defined for a set of quantities, commonly termed
[qRr] ~ Ira(q) > m(r)] ... (19) basic, of a domain of science, say Ql . . . . . (21.. Scales for
other quantities Qj+I,...,Qn a r e then established in-
where the square brackets indicate events and - the in- directly in terms of the basic quantities, on the basis of
difference of probability. This theorem may be interpreted empirical relations between the indirectly measured quan-
as a probabilistic homomorphism. Its proof requires axioms tity and its components.
for ( ~ , A, ~ ) as well as (Q, R). I f P is a real valued proba- This account correctly represents the way in which
bility function, the theorem may be expressed as: scales of measurement are established for a domain of
P[q 1R qa] = P[m(q 1) > m(q~)] ... (20) science at an early stage of its development. Thus for an
early stage of development of physics we have accounts of
The authors investigated such representations for ordinal, establishment of a scale of measurement for mass in terms
extensive, internal and conjoint structures (Leaning, 1977; of an equiarm balance and the construction of a set of

30 Measurement Vol 2 No 1, Jan--Mar 1984


Fundamental measurement concepts Finkelstein and Leaning
standard weights. Accounts of the construction of scales for Let{Q = qi} be a set of entities and R ={R1 . . . . .
the measurement of length are given in terms of a standard R i . . . . . Rn} be a family of relations on Q, so that Q =
rod, and the concatenation of rods. Scales for indirectly <Q, R> represents a source relational system. Let Z = {zi}
measured quantities such as density are given in terms of be a set of entities to be used as symbols and let P =
empirically observed relations between volume, and its {P1 . . . . . ei . . . . . Pn} be a family of relations on Z so that
components mass and volume. Z = <Z, P > is a symbol system. Let M: Q ~ Z be a mapping
However, as a domain of science becomes developed, it that is a relation that assigns elements of Z to elements of
acquires a comprehensive body of theory, and scales of Q. Let M b e such, that i f P i is a relation on Z corresponding
measurement are defined in terms of this theory. Thus, in to the relation Ri on Q and qj, q k ~ Q and zj, Z k ~ Z , then:
modern physics, scales of measurement are defined in terms
of a set of base quantities - length, mass, time, electric R i ( q i . . . . . qk) ~* Pi(zi . . . . . zk) ... (26)
current, thermodynamic temperature, amount of substance
and luminous intensity. Other quantities are derived from where z i =M(q/) and zk = M ( q k ) . We can then say that Z
and M provide a symbolisation for Q. We can term e = <Q,
them (Page and Vigoureux, 1973). The definition of the
scale of measurement of temperature is established on the Z, M > as a code of the symbolisation. We can then say that
I = <zj, c > constitutes information about q/ (where zj =
basis of accepting a theory of thermodynamics and defining
M(qj)). zj is termed the symbol of or for q / a n d q/is termed
the scale in terms of this theory. The scale of measurement
of electric current is defined in terms of a force per unit the meaning o f z/.
length acting between two infinitely thin, infinitely long We can see how this formal description of symbolisation
conductors. It is thus based on the acceptance of a theory corresponds to the formal definition of measurement. It is
of electrodynamics. possible by considering M not as a mapping but as a many-
Formally, we may express the establishment of scales of to-many relation to describe such aspects of symbolisation
measurement in terms of a theory, as follows: as synonymy, homonymy, ambiguity and vagueness.
Let there be a theory for a domain of science. Let ~2 be Assignment of symbols as names of entities is a simple
the set of all objects of that domain and Q1 . . . . , Qn the set example of the above definition of symbolisation. Nominal
of all quantities describing phenomena in that domain; and ordinal measurement scales, which may have measures
where quantities Q~ are in the form Qj = <Qj, R/> with the other than numbers, are clear examples that measurement
notation described above. Let there be for each of the may be treated as a special case of general semiotics.
Finkelstein (1977) has described indirect symbolisation
quantities a scale of measurement of the form Ij = <Qi, N ,
codes, that is, a symbol code for an entity in terms of
Mj> where N is a numerical system N = <N, P> and M i is a
symbol codes for related entities, in the same way as
mapping from Qj into N. Then the theory takes the form
of a set of statements relating empirical and numerical general indirect measurement.
relations. Consider any object cop e ~2p where 12p c ~2. Let Language is a system for combining symbols into groups.
A formal language is one in which the symbols and their
~ p be characterised by quantity manifestations q~, . . . , p f
groups do not have a meaning. In an interpreted language
of quantities Qk . . . . . Ql (where qf
is a manifestation of Qj
the symbols and their groups have a meaning that is a cor-
occurring in object p. Then for all coe~ ~ e we have:
respondence to extralinguistic entities. We can describe an
S p ( q P , . . . , q~) ~ (ap [Mk(qPk) . . . . . M l ( q q ) ] ... (25) interpreted language formally as L = <Z, G, c> where Z is
the set of meaningful symbols of the language; G, the syntax
where S t, is an empirical relation between the quantities or grammar of L, is the system of rules which generates
involved and ~bp a numerical relation (generally an equa- permissible groupings of the symbols of L; and ¢ is the
tion). system of rules which establishes an isomorphic or homo-
To establish a scale of measurement for the domain of morphic correspondence between the symbols and the rela-
science with the above theory, a number of base quantities tions between them, and the extra-linguistic entities and
Qx, Qy . . . . are chosen and the mappings M x , M y . . . . the relations between them.
defined. The mapping for the other quantities then follows It will be seen that there exists a close relation between
from the statements of the theory. The base quantities are measurement and interpreted languages. The results of
chosen because of the centrality of the quantities in the measurement are stated in a linguistic form such as qi = hiS,
theory, the convenience of the practical realisation of scales which is a statement that the manifestation qi of property
and the like. Q which is specified in terms of space and time or of some
object, has the value n i on the scale of measurement S. The
10 Measurement, semiotics and information symbols qi, =, ni and S have a defined correspondence to
the extra-linguistic entities of relations they represent.
Measurement can be viewed as a special case of represen- Given that measurement can be considered as a special
tation of aspects of the real world by symbols. aspect of semiotics, it is necessary to identify the special
A symbol is an object or event which bears a defined features which give it importance. First, the representation
relation to a source entity. The symbol, together with the relation M is objective and empirical. Secondly, the source
above relation, which may be termed the representation relational system Q to be represented is, in the case of
relation, carries information about the source entity. measurement, an empirical system. Measurement is not
The general term for the study of symbols, their use, and naming - relations between measures represent relations
relation to the object they represent is known as semiotics. between the entities they describe. Some authors confine
Finkelstein (1975b) has formulated some of the the term measurement to representation by numbers;
principal concepts of semiotics in terms of representational however, some representation by non-numeric symbols on
measurement theory. A briefer and broader statement is (say) a nominal or ordinal scale does not appear in its essen-
given in Finkelstein and Leaning (1982). These formulations tials to differ from measurement on numerical scales.
are used in this paper as a way of demonstrating the rela- The formal representation of reality by symbols is being
tion between measurement and general semiotics. developed in connection with computer information
Measurement Vol 2 No 1, Jan--Mar 1984 31
Finke/stein and Leaning Fundamenta/ measurement concepts
processing and databases. The essential aspects of this 12 Conclusions
work are concerned with the development of conceptual This paper has renewed the fundamental concepts of
models of the sub-set of the real world represented and measurement in terms of the representational approach.
with the establishment of structures of information-carrying The representational treatment of measurement, which
symbol systems (such as relational tables, hierarchical tree has its basis in model theory, evidently provides a second
structures, or networks convenient for information storage foundation for the logic and philosophy of measurement.
and processing). References may be made to Bobrow and On the argument that an understanding of underlying
Collins (1975); Deen (1977); Kent (1978); Date (1981); theory is essential for effective application, measurement
Tennant (I981). theory should be more widely studied and understood by
practitioners of measurement.
11 Measurement and description by natural Certain important trends and issues in measurement
language theory have been identified.
There exists a close relation between measurement and The form of a theory of uncertain measurement, which
the description of objects or events of the real world in has deterministic and probabilistic components, is now
natural language. Thus, to describe an object as red is taking shape.
related to measuring colour on a nominal scale. Similarly In examining systems of scales of measurement for a
the distinction in language between 'hot' and 'hotter' is domain of science for which there are reasonably compre-
related to ordinal measurement of temperature. hensive theories, it is clear that such theories must form the
The present paper is not a vehicle for a review of the basis of such scales.
relevant literature of linguistics but reference may be made Measurement is a special case of representation by
to Lyons (1968) for general linguistics; Leech (1981) and symbols, with strong relation to description by language,
Palmer (1976) for semantics; and Alston (1964) for the computer data representation and the like. There will be
philosophy of language. significant advances in the areas of database and intelligent
In comparing description by natural language with des- knowledge-based systems and one can confidently expect
cription by measures we may note the similarities and that measurement theory will progress in strong relation
differences outlined below. with them.
Descriptive adjectives are similar to measures of property
manifestations. They may be viewed as non-numeric Bibliography and references
symbols corresponding to property manifestations as Adams, E. W. 1965. 'Elements of a theory of inexact
described in section 10. measurement', Phil Sci, 32,205 228.
Lexical items when viewed as symbols have linguistic Alston, W. P. 1964. Philosophy of language, Prentice Hall,
relations on them. Typical relations are: binary taxonomy, Englewood Cliffs.
e g, 'dead' or 'alive'; multiple taxonomy, e g, colour names; Armstrong, W. E. 1939. 'The determinateness of the utility
polarity, eg, small, large, larger. For a discussion see Leech function', Econ J, 40,453 467.
(1981) and the literature cited therein. Beers, Y. 1953. Introduction to the theory of errors,
Componential semantics, in which one lexica] item - say, Addison-Wesley, Reading (Mass).
'man', is expressed in terms of its components 'human', Black, M. 1937. 'Vagueness', Phil Sci, 4 , 4 2 7 - 4 5 5 .
'adult', 'male' - is similar to conjoint or indirect measure- Black, M. 1970. 'Margins of precision', Essays in logic and
ment. Again, see Leech (1981) and the literature cited language, Cornell University Press,
therein. Blalock, H. M. 1974. Measurement in the social sciences
However, the meaning of such lexical items is not (theories and strategies), Macmillan.
defined, unlike measures, in terms of an explicit correspon- Bohrow, D. and Collins, A. 1975. Representation and
dence to its referent. The correspondence is subjective, en- understanding studies in cognitive science, Academic
coding and decoding being dependent on communicator Press, New York.
and communicatee, respectively. Correspondence with Bridgman, P. W. i922. Dimensional analysis, Yale Univer-
referent is only one interpretation of meaning in natural sity Press.
language. Other ways of looking at meaning in natural Byerly, H. C. and Lazara, V. A. 1973. 'Realist foundations
language are in terms of correspondence with the concepts of measurement', Phil Sci, 40 (10).
or ideas of the communicator and communicatee or, alter- Campbell, N. R. 1920. Physics." the elements, Cambridge
natively, in terms of behaviour of the communicator- University Press.
communicatee system (Alston, 1964; Leech, 1981). Lexical Campbell, N. R. 1928. An account of the principles of
items of natural language, unlike measures, have, in addi- measurement and calculation, Longmans Green,
tion to conceptual or denotative meaning, also associative London.
meaning, which, for example, conveys information about Campbell, N. R. 1935. 'The statistical theory of errors',
the feelings or views of the communicator or else communi- Proc Phys Soc, 4 7 , 8 0 0 - 8 0 9 .
cates something by virture of other uses of the same item. Campbell, N. R. 1938. 'Measurement and its importance
In the development of scales of measurement for parti- for philosophy', Arist Soc Suppl, 17.
cular properties, it is common to start with a qualitative, Campion, P. J., Bums, J. E. and Williams, A. 1973. 'A code
verbally described, concept of the property. The aim of of practice for the detailed statement of accuracy',
measurement is to give the description of the property National Physical Laboratory, HMSO, UK.
objectivity, an empirical basis, precision and conciseness. In Causey, R. L. 1969. 'Derived measurement, dimensions and
the process of establishing a scale of measurement, that dimensional analysis', Phil Sci, 36,252.
concept is often considerably refined and altered. The Churchman, C. W. and Ratoosh, P. 1959. Measurement.
verbally described qualitative concept may have a richness Definition and theories, Wiley.
which the scale of measurement does not have. In that case Colonius, H. 1978. 'On weak extensive measurement', Phil
the latter needs to be adapted to reflect the concept better. Sci, 45,303 308.
32 Measurement Vol 2 No 1, Jan-Mar 1984
Fundamental measurement concepts Finkelstein and Leaning
Coombs, C. H. 1964. A theory of data, Wiley, New York. H61der, O. 1901. 'Die Axiome der Quantitat und die Lehre
Date, C. J. 1981. An introduction to data base systems, von Mass', Math-Phys Klasse, 53, 1-64.
Addison-Wesley, Reading (Mass). Holman, E. W. 1971. 'A note on conjoint measurement
Deen, S. M. 1977. Fundamentals of data base systems, Mac- with restricted solvability', J Exp Psychol.
millan Press, London. Holman, E. W. 1974. 'Extensive measurement without an
Desai, M. 1976. Applied econometrics, Philip Allen. order relation', Phil Sei, 41, 361.
Destouehes, J. L. 1975. 'About measurement theory', Proc Jacob, M. 1961. 'L'exactitude dans les mesures', Aeta
IMEKO Colloquium on Nature and Scope of Measure- IMEKO, I, 160 175.
ment Seience, 66-67. Jeffreys, H. 1938. 'Measurement and its importance for
Domotor, Z. 1969. 'Probabilistic relational structures and philosopy', Arist Soc Suppl, 17, 143-151.
their applications', Technical Report No 144, Stanford Kendall, M. and Stuart, A. 1970. The advanced theory of
University, Institute for Mathematical Studies in the statistics, Griffin.
Social Sciences. Kent, W. 1978. Data and reality. Basic assumptions in data
Domotor, Z. and Stelzer, J. H. 1971. 'Representation of processing reconsidered, North-Holland, Amsterdam.
finitely additive semiordered qualitative probability Krantz, D. H. 1967. 'Extensive measurement in semiorders',
structures',J Math Psych, 8 , 1 4 5 - 1 5 8 . Phil Sci, 34,348-362.
Duda, R. O. and Hart, R. E. 1973. Pattern classification and Krantz, D. H. 1973. 'Fundamental measurement of force
scene analysis, John Wiley. and Newton's first and second laws', Phil Sci, 40,481.
Ellis, B. 1964. 'On the nature of dimensions', Phil Sci, 31, Krantz, D. H., Luce, R. D., Suppes, P. and Tversky, A.
357-380. 1971. Foundations of measurement, Vol. 1, (Additive and
Ellis, B. 1966. Basic concepts of measurement, Cambridge polynomial representations), Academic Press, New York.
University Press. Leaning, M. S. 1977. Foundational studies in measurement,
Emery, D. R. and Brown, F. H. 1979. 'Axiomatic and MSc thesis, The City University, London.
numerical conjoint measurement: an evaluation of diag- Leaning, M. S. and Finkelstein, L. 1980. 'A probabilistic
nostic efficacy', Psychometrika, 44, 195-210. treatment of measurement uncertainty in the formal
Falmagne, J. C. 1975. 'A set of independent axioms for theory of measurement', Acta IMEKO 1979, G. Striker,
positive HOlder systems', Phil Sci, 42, 137. ed, Elsevier, Amsterdam, 73-81.
Falmagne, J. C. 1980. 'A probabilistic theory of extensive Leech, G. 1981. Semantics, (2nd edn), Penguin Books.
measurement', Phil Sci, 47,277-296. Luee, R. D. 1956. 'Semiorders and a theory of utility
Finkelstein, L. 1973. 'Fundamental concepts of measure- discrimination', Econometrica, 24, 178-191.
ment', Acta IMEKO 6th Congress, Dresden, 1973, Luce, R. D. 1973. 'Three axiom systems for additive semi-
North-Holland, Amsterdam. ordered structures', SIAM J Appl Math, 25, 41-43.
Finkelstein, L. 1975a. 'Fundamental concepts of measure- Lute, R. D. 1978. 'Dimensionally invariant numerical laws
ment: definitions and scales', Measurement and Control, correspond to meaningful qualitative relations',PhilSei,
8,105-111. 45,1 16.
Finkelstein, L. 1975b. 'Representation by symbols as an Lute, R. D. and Tukey, J. W. 1964. 'Simultaneous conjoint
extension of the concept of measurement', Kybernetes, measurement: a new type of fundamental measurement',
4, 215-223. J Math Psychol, 1, 1-27.
Finkelstein, L. 1977. 'The relation between the formal Luee, R. D., Bush, R. F. and Galanter, E. 1965.Handbook
theory of measurement and pattern recognition', Acta of mathematical psychology, Wiley.
IMEKO 7th Congress, London, 1976, North-Holland, Lute, R. D. and Marley, A. A. J. 1969. 'Extensive measure-
Amsterdam. ment when concatenation is restricted and maximal
Finkelstein, L. and Leaning, M. S. 1982. 'Foundations of elements may exist'. In Essays in honor of Ernest Nagel,
symbolic representation systems', in European Meeting S. Morgentesser, P. Suppes and M. G. White, eds, St
on Cybernetics and Systems Research, 5th, 1980, Martin's Press, New York.
Vienna. Progress in Cybernetics and Systems Research, Lyons, J. 1968. Introduction to theoretical linguistics,
R. Trappl, N. V. Findler and W. Horn, eds, Hemisphere, Cambridge University Press.
Washington, DC, 501-505. Maeh, E. 1896. Die Prineipien der Warmelhre, Leipzig.
Fishburn, P. C. 1964. Decision and value theory, John (Partial trans 'Critique of the concept of temperature',
Wiley, New York. in Ellis, B., 1966.)
Fishburn, P. C. 1968. 'Utility theory', Management Sci, Morgenstern, O. 1963. On the accuracy of economic obser-
14,335-378. vations, Princeton University Press.
Fishburn, P. C. 1968. 'Weak qualitative probability on Narens, L. 1974. 'Measurement without Archimedean
finite sets', Ann Math Stat, 40, 2118-2126. axioms', Phil Sei, 41,374.
Galanter, E. H. 1956. 'An axiomatic and experimental study Page, C. H. and Vigoreux, P. 1973. SI, the International
of sensory order and measure', Psychol Rev, 63, 16-28. System of Units, NPL, HMSO, London.
Goodman, N. 1951. The structure of appearance, Harvard Palaeios, J. 1964. Dimensional analysis, Macmillan.
University Press, Cambridge (Mass). Palmer, E. R. 1976. Semantics." A new outline, Cambridge
Guild, J. 1938. 'Quantitative estimation of sensory events University Press.
(Part III)', Interim Report, Rep Br Ass Advan Sci, 2 9 6 - Pfanzagl, J. 1959. 'A general theory of measurement: appli-
328. cations to utility', Naval Res Logistics Q, 6 , 2 8 3 - 2 9 4 .
Harary, F. 1964. 'A graph-theoretic approach to similarity Pfanzagl, J. 1968. Theory of measurement, Wiley, New
relations', Psychometrika, 29, 143-151. York.
Helmholtz, H. V. 1887. Zdhlen und Messen Erkenntis - Prugovecki, E. 1974. 'Fuzzy sets in the theory of measure-
Theoretisch Betrachet, Philosophische Aufs~itz Eduard ment of incompatible observables', Found Phys, 4, 9.
Zeller Gewidmet, Leipzig. (trans by C. L. Bryan, Count- Roberts, F. S. 1970. 'On non-transitive indifference', J
ing and Measuring, Van Nostrand, Princeton, 1930.) Math Psych, 7,243-258.
Measurement Vol 2 No 1, Jan--Mar 1984 33
Finkelstein and Leaning Fundamental measurement concepts
Roberts, F. S. 1979. 'Measurement theory with applica- Tennant, H. 1981. Natural language processing: An intro-
tions to decision making, utility, and the social sciences', duction to an emerging technology, Petrocelli Books,
Encyclopedia of mathematics and its applications, Vol New York.
7, Gian-Carlo Rota, ed, Addison-Wesley, Reading Thurstone, L. L. 1947. Multiple-factor analysis, Chicago
(Mass). University Press.
Robinson, F. E. 1965. 'Measurement and statistics: towards Torgerson, W. S. 1958. Theory and methods of scaling.
a classification of the theory of "permissible statistics" ', John Wiley, New York.
PhilSci, 32,229 243. Tsichntzis, D. and Kiug, A. 1977. The ANS1/X3/SPARC
Rosen, R. 1978. Fundamentals of measurement and repre- DkMS framework. Report of the Study Group on
sentation of natural systems, North-Holland, Amster- data base management systems. AFIPS Press, New
dam. York.
Russell, B. 1923. 'Vagueness', Aust J Psych Phil, 1 , 8 4 - 9 2 . yon Neumann, J. and Morgenstern, O. 1944. Theory of
Scott, D. and Suppes, P. 1958. 'Foundational aspects of games and economic behaviour, Princeton University
theories ofmeasurement',JSymb Logic, 23, 113-118. Press.
Shapere, D. 1982. 'The concept of observation in science Whitney, H. 1968. 'The mathematics of physical quantities.
and philosophy', Phil Sci, 49,485-525. Part I: Mathematical models for measurement. Part II:
Stevens, S. S. 1946. 'On the theory of the scales of Quantity structures and dimensional analysis', Am
measurement', Science, 103,677 680. Math Monthly, 75, 115 t38,227 256.
Stevens, S. S. 1951. Handbook of experimental psychology, Wiener, N. 1921. 'A new theory of measurement: a study
Wiley. in the logic of mathematics', Proc Lond Math Soc, 19,
Suppes, P. 1951. 'A set of independent axioms for exten- 181-205.
sive quantities', Portugaliae Mathematica, 10, 163-172. Zadeh, L. A. 1965. 'Fuzzy sets', Inform Contr, 8, 338
Suppes, P. and Zinnes, J . L . 1965. 'Basic measurement 353.
theory', in Handbook of mathematical psychology, Zadeh, L. A. 1968. 'Probability measures of fuzzy events',
R. D. Luce, R. R. Bush and E. Galanter, eds, Wiley. J Math Anal Appl, 23, 421-427.
Tarski, A. 1954. 'Contributions to the theory of models', Zadeh, L. A. 1971. 'Similarity relations and fuzzy order-
Indagationes Mathematicae, 16,572-588. ings',Info Sciences, 3,177 200.

International Measurement Confederation Tech nical Corn m i t t e e - 1 2


Temperature Measurement

Temperature Measurement in Industry and Science

The TC-12 will organise, in co-operation with the • Industrial temperature measurements
national IMEKO member organisation of the GDR, • Measurement of thermophysical properties, and
the W G M A (Scientific-Technical Society for heat flow measurement.
Measurement Technology and Automation) in the • Poster section.
KDT (Chamber of Technology), its 2nd Symposium
Temperature Measurement in Industry and Science Interesting papers that go beyond the scope of the
in the German Democratic Republic. It will take programme may also submitted for discussion.
place from 16-18 October 1984 in Suhl, the capital of
this district in the heart of the Thuringian Forest.
The working language used during the Sympo-
sium is English. The conference fee amounts to
Main topics DM 400, including the preprints and abstracts.
• Improvement of the International Practical
Temperature Scale Please address inquiries concerning the organisa-
• Temperature sensors and their calibration tion of the symposium, and provisional application,
• Dynamic behaviour of temperature sensors to: Kammer der Technik, Dipl.-Ing. U. Hartung,
• Radiation temperature measurement Thilmannplatz 4, DDR-6000 Suhl.

34 Measurement Vol 2 No 1, Jan--Mar 1984

Vous aimerez peut-être aussi