British Journal of Psychology (1997), 88, 355-383 Printed in Great Britain 355

© 1997 The British Psychological Society

Quantitative science and the definition of

measurement in psychology

Joel Michell*
Department of Psychology, University of Sydney, Sydney NSW 2006, Australia

^ g e d that establishing quantitative science involves two research tasks: the

scientific one of showing that the relevant attribute is quantitative; and the
instrumental on& of constructing procedures for numerically estimating magnitudes;
In proposing quantitative theories and claiming to measure the attributes involved,
psychologists are logically committed to both tasks. However, they have adopted
their own, special, definition of measurement, one that deflects attention away from
the scientific task. It is argued that this is not accidental. From Fechner onwards,
the dominant tradition in quantitative psychology ignored this task. Stevens'
definition rationalized this neglect. The widespread acceptance of this definition
within psychology made this neglect systemic, with the consequence that the
implications of contemporary research in measurement theory for undertaking the
, scientific task are not appreciated. It is argued further that when the ideological
support structures of a science sustain serious blind spots like this, then that science
is in the grip of some kind of thought disorder.

...unluckily our professors of psychology in general are not up to quantitative logic...

E. L. Thorndike to J. McK. Cattell, 1904

Psychologists resist the intrusion of philosophical considerations into their science,

as if such considerations could somehow threaten its genuine achievements.
Resistance is especially stiff in the methodological area where the tone was set by the
founder of quantitative methods in psychology, G. T. Fechner, who, to criticisms of
his psychophysical methods, responded, 'all philosophical counter-demonstrations
are, I think, mere writing in the sand' (1887, p. 215). The same resistance is noted
by Meier (1994) from a reviewer of a critical paper: ' So much for what the paper says,
I am even more concerned about what it implies; namely, that we should cease and
desist because applied psychological measurement has flaws Applied measurement
is the primary contribution that psychology has made to society' (p. xiii). The
attitude that any science should insulate itself against criticism is anti-scientific. If
principled criticism is not answered in a principled way then the doubts raised
remain.
Philosophical criticism in the methodological area has a special function. If the
* Requests for reprints.
356 Joel Michell
methods of science are not sanctioned philosophically then the claim that science is
intellectually superior to opinion, superstition and mythology is not sustained.
Psychologists have as much at stake in this as other scientists. The methods of science
(e.g. observation, experiment, measurement, etc.) involve some of the deepest of
philosophical problems, problems in which definitive solutions seem as elusive as
ever. In seeking solutions to these problems, the scientist will prefer a scientific
philosophy, i.e. one that is capable of justifying science as a rational cognitive
enterprise. The natural scientific attitude and the one that promises the most coherent
defence of science is that of empirical realism (i.e. that of an independently existing
natural world which humans are able to successfully cognize via observational
methods, at least sometimes). From this perspective, philosophical research now
enables a clear view of what some methods, especially measurement, amount to.
Taking this view seriously, it will be argued that psychology is in danger of losing
contact with the great intellectual tradition of quantitative, experimental science.
Many psychological researchers are ignorant with respect to the methods they use.
This ignorance is not so evident at the instrumental \evt\, i.e. that of using techniques
of data collection and analysis, although there is a surprising degree of ignorance
even here. The ignorance I refer to is about the logic of methodological practices, i.e.
about understanding the rationale behind the techniques. Knowing the logic of
methodological practices is not a matter of icing on the cake, icing which, if
neglected, leaves the substance unaffected. Ignorance of this logic may mean not
knowing the right empirical questions to ask or, even, that there are any in this
context. The history of science shows that the insights underlying quantification
were hard won. The history considered in this paper, by comparison, shows that they
are very easily lost and, once lost, not easily regained. When the attitude of turning
a deaf ear to criticism becomes entrenched, it can alter patterns of thinking and the
way words are used, with the result that criticism may be treated as irrelevant.

cl. The concept of scientific measurement^

1.1 Centrality of the concept of quantity
In quantitative science attributes (such as velocity, temperature, length, etc.) are
taken to be measurable. That is, it is theorized that an attribute, such as length, has
a distinctive kind of internal structure, viz., quantitative structure. Attributes having
this kind of structure are called quantities. Following a well-established usage, specific
instances of a quantity are called magnitudes of that quantity (e.g. the length of this
page is a magnitude ofthe quantity, length). Magnitudes of a quantity are measurable
because, in virtue of quantitative structure, they stand in relations {ratios) to one
another that can be expressed as real numbers.
While quantitative science has existed since ancient times, quantitative structure,
itself, was only explicitly characterized late in the nineteenth century and its best
known formulation is given by Holder (1901). Holder's set of seven axioms define
a continuous quantity and the following is a slightly more succinct definition of the
same concept (Michell, 1994). A range of instances of an attribute, £ , constitutes a
continuous quantity if and only if the following five conditions obtain (in each case
 Quantitative science and psychology 357
an attempt has been made to state first a more accessible explanation of what the
condition means, free of mathematical symbols and technical terms).
1. Any two magnitudes of the same quantity are either identical or different and, if
the latter, there must exist a third magnitude, the difference between them, i.e. for
any a and b in Q, one and only one of the following is true

(i) a = h,
(ii) there exists c m Q such that a = h-\-c,
(iii) there exists c \i\ Q such that h = a-\-c;

2. A magnitude entirely composed of two discrete parts is the same regardless ofthe
order of composition, i.e. for any a and b'\nQ, a-\-h = h-Va;
3. A magnitude which is a part of a part of another magnitude is also a part of that
same magnitude, the latter relation being unaffected in any way by the former, i.e.
for any a, b and c \n Q, a-\- {b -\- c) = {a-^ b)-\- c\
4. For each pair of different magnitudes of the same quantity there exists another
between them, i.e. for any a and b in Q such that a> b, there exists c in Q, such
that a> c > b; and
5. Given any two sets of magnitudes, an 'upper' set and a 'lower' set, such that each
magnitude belongs to either set but none to both and each magnitude in the upper
set is greater than any in the lower, there must exist a magnitude no greater than
any in the upper set and no less than any in the lower, i.e. every non-empty subset
oiQ that has an upper bound has a least upper bound.
Note that one magnitude is greater than another if and only if the latter is a part
of the former, i.e. for any a and b m Q, a > b \£ and only if (ii) above is true.
Conditions 4 and 5 ensure the density and continuity, respectively, of the quantity,
which intuitively may, thus, be thought of as containing no gaps in the sequence of
its magnitudes.
Some words of caution should be added about the use ofthe mathematical symbol,
' -f-', in the above conditions. Readers will be most familiar with the use of this
symbol in arithmetic contexts, where the terms added are numbers. My first warning,
then, is that in the above conditions the addition is not of numbers but of magnitudes
of a quantity (e.g. specific lengths, say). My second warning is this: ' + ' is often
understood as a mathematical operation and this interpretation, when applied to
magnitudes, has sometimes (e.g. by Campbell, 1920, 1928) been understood as
requiring an empirical operation of concatenation (i.e. an operation of putting
magnitudes together in some way). Such an interpretation is not intended here and
to forestall it I recommend the alternative of interpreting a-\-b = c as a relation
between the magnitudes a, b and c. The relation I have in mind is this: magnitude
c is entirely composed of discrete parts, magnitudes a and b. This interpretation is
suggested by Bostock (1979). The point of making this distinction is that just because
magnitudes stand in this relation, it does not follow that suitable operations of either
concatenation or division will obtain for objects possessing the magnitudes so
related. This may be so, as with length and other convenient quantities, or it may not,
as with density or temperature. That is, the additive relation between magnitudes is
a theoretical one and how we gain access to it may often be indirect.
358 Joel Michell

1.2 Tbe concept of quantity entails that of measurement

If an attribute is quantitative then it is, in principle, measurable. This was the main
theorem of Holder's (1901) paper. He showed that given such structure, for any a and
b in Q, the magnitude of a relative to b may always be expressed by a positive, real
number, r, where a = r.b. That is, the ratio oi a to b (a positive, real number) is the
measure of a in units of b. This fact, in turn, makes it meaningful to hypothesize the
existence of quantitative relations between attributes (like that between density, mass
and volume). The practice of measurement requires getting some grip, either directly
or indirectly, upon the additive structure of the attribute in order that ratios between
magnitudes of the attribute may be discovered or estimated. Hence, scientific
measurement is properly defined as the estimation or discovery of the ratio of some
magnitude of a quantitative attribute to a unit of the same attribute. It is invariably along
such lines that measurement is, and always has been, defined in the physical sciences
(see, for example, Beckwith & Buck, 1961; Clifford, 1882; Cook, 1994; Massey,
1986; Maxwell, 1891).
This definition of measurement is a logical consequence of the structure that
quantitative attributes are taken to possess. Given that structure, the fact that
magnitudes of a quantity stand in numerical relations to one another is a provable
mathematical theorem. Measurement is nothing more or less than the attempt to
discover or estimate such numerical relations. This is the logical basis of quantitative
science with all its mathematical beauty, conceptual scope, empirical power and
practical utility (see Appendix II).

1.3 The empirical commitments of the concept of quantity

In conceptualizing an attribute as quantitative a scientific hypothesis is proposed.
There is no logical necessity that any attribute should have this kind of structure.
Hence, accepting this hypothesis is speculative, unless there is evidence specifically
supporting it. The issue of evidence for quantity is always complex and some of the
conditions 1 to 5, mentioned in section 1.1, are never separately, directly testable
(such as 5, the continuity condition). However, in the case of some quantities (e.g.
length) conditions 2 and 3 are directly testable, at least for humanly manageable
lengths. For example, if x , j and ^ are straight rigid rods and x exactly spans the rod
entirely composed of discrete parts, j and ^, linearly concatenated in a particular
order, then if 2 is true, x must exactly span the length of the rod entirely composed
of ^ andj/, linearly concatenated in the opposite order.
It should be stressed that for many physical quantities, the existence of which is
now taken for granted (e.g. temperature and density), the evidence that they are
quantitative is entirely indirect. That is, the additive structure of the attribute is not
directly reflected via a relation of physical concatenation, as with the case of length
(at least for lengths of humanly manageable sizes). Given the fact that much of
science involves theories that are likewise only indirectly testable, such a point would
hardly be worth stressing had it not been an almost permanent source of confusion
over the last century, both in psychology and measurement theory generally. It
would seem that measurement has been mistakenly thought of by some philosophers
 Quantitative science and psychology 359

as being an atheoretical, purely observational base upon which science's more

theoretical structures stand. It is not. Measurement always presupposes theory: the
claim tbat an attribute is quantitative is, itself, always a theory and that claim is
generally embedded within a much wider quantitative theory involving the
hypothesis that specific quantitative relationships between attributes obtain. Because
the hypothesis that any attribute (be it physical or psychological) is quantitative is a
contingent, empirical hypothesis that may, in principle, be false, the scientist
proposing such an hypothesis is always logically committed to the task of testing this
claim whether this commitment is recognized or not.

1.4 The two tasks of quantification: The scientific and the instrumental
Establishing a quantitative science involves two tasks. First, there is the logically
prior scientific one of experimentally investigating the hypothesis that the relevant
attribute is quantitative. Second, there is the instrumental task of devising procedures
to measure magnitudes of the attribute shown to be quantitative. Failure to
investigate the scientific task prior to working upon the instrumental one and failure
to confirm the hypothesis that the relevant attribute is quantitative means that
treating the proposed measurement procedures as if they really are measurement
procedures is at best speculation and, at worst, a pretence at science.

2. Measurerhent in psychology
2.1 The measurement of psychological attributes and the commitment to quantitative
structure
Even a superficial perusal of relevant psychological publications reveals that
psychologists believe that they are able to measure many distinctly psychological
attributes, such as cognitive abilities, personality traits, social attitudes and sensory
intensities. These attributes are distinctly psychological in the sense that they form
part of psychology's subject matter and, also, in the sense that they do not belong to
the network of quantitative attributes measurable using the methods of the physical
sciences (see, for example, Jerrard & McNeill, 1992; Sena, 1972). While these
psychological attributes do not form part of this network, it is clear that quantitative
psychology was first modelled upon quantitative physics (Fechner, 1860). That is, in
both disciplines alike, certain attributes are supposed to have quantitative structure.

2.2 The consequent commitment to the scientific task of quantification

Psychologists, in their attempt to construct a quantitative science by analogy with
quantitative physics, hypothesize that some of their attributes are quantitative and,
furthermore, that some of these attributes relate quantitatively, either amongst
themselves or with physical quantities. For example, Fechner (1860), in proposing his
psychophysical theory, conceived of the intensity of sensations as a quantity and
hypothesized a particular functional relationship between it and the physical intensity
ofthe stimulus. Or, Spearman (1904), in proposing that level of performance on an
360 Joel Michell

intellectual task was due to a combination of the level of general ability and the level
ofthe ability specific to that task, conceived of general ability and the various specific
abilities as quantitative attributes and proposed a functional quantitative relation
between these quantities and test scores. Fechner's and Spearman's quantitative
speculations provided the model for many later developments, for example, those by
Thurstone (1938), Hull (1943), Stevens (1956), Cattell (1943) and others. In every
case, the above concept of continuous quantity was necessarily presumed, although
it must be said that this was not often explicitly acknowledged. It was presumed,
however, as a necessary concomitant of quantitative theorizing. Hence, presumed
along with it, as part of the same conceptual package, was the traditional concept of
scientific measurement.

3.^'The definition of measurement in psychology

3.1 Stevens' definition of measurement
Even though quantitative psychologists (by whom I mean those who either theorize
about or attempt to measure psychological quantities) hypothesize that their
attributes are quantitative and, so, commit themselves to the concept of scientific
measurement, the definition of measurement actually endorsed by most of them is
radically different. This definition is the one formulated by Stevens (1946):
measurement is the assignment of numerals to objects or events according to rule. It is easily
verified that this, or something very similar, is the definition they prefer. Many
psychological texts, especially those on research methods or relating in some way to
psychological measurement, offer their readers a definition of measurement. These
definitions are surprisingly uniform and, while they do not all match Stevens'
definition word for word (although many of them do), they wear the mark of
Stevens.

3.2 Its widespread acceptance within psychology

I recently surveyed the psychology library of a major European university. I was
easily able to locate 44 books (see Appendix I), each providing a definition of
measurement, ranging in publication date from the early 1950s to the early 1990s. Of
these, 39 offered a definition of this form: measurement is the assignment of ^ to Y
according to Z. Within this schema, which derives directly from Stevens, X was
typically either numerals or numbers., although other cognate terms sometimes
appeared (e.g. numerical values, scores, abstract symbols). Stevens' phrase, objects or
events, was mostly retained for Y, although a wide range of other terms appeared as
well, including things, situations, individuals, behaviour, observations, attributes,
properties and responses. Where authors departed from Stevens' preference for Z,
it was generally (in about a dozen cases) to offer a specific rule (in most cases a
representational rule). None of the 44 definitions even remotely resembled the
traditional scientific concept. These observations confirm that psychology, as a
discipline, has its own definition of measurement, a definition quite unlike the
traditional concept used in the physical sciences.
 Quantitative science and psychology 361

3.3 Its relation to the concept of quantity and to the consequent scientific task of
quantification

As will be shown in section 4, Stevens' definition of measurement entered

quantitative psychology at a particularly crucial stage of its history. The definition
was accepted by psychologists, not innocently, out of ignorance of the truth about
quantity and measurement: it was accepted and became entrenched because it
appeared to solve a conceptual problem which had existed since Fechner's time and
which in the 1940s was particularly pressing. The acceptance of this definition
involved a quite deliberate turning away from traditional concepts and it resulted in
a systemically sustained blind spot, one which has persisted to the present. These are
claims that I will support shortly by considering historical evidence.
However, first, it should be noted that (as I will show in more detail in section 5.2)
the facts about quantity and measurement are readily evident to anyone motivated to
find them. More than that, the idea that the practice of measurement is underwritten
by the concept of a quantitative attribute and the knowledge ofthe structure of such
an attribute is almost entirely absent from books on psychological measurement (I
exclude from this generalization the publications of R. D. Luce, P. Suppes and their
associates—e.g. Krantz, Luce, Suppes & Tversky, 1971; Luce, Krantz, Suppes &
Tversky, 1990; Narens, 1985; Suppes, Krantz, Luce & Tversky, 1989—which are
exceptional in this respect and, also, in having escaped the notice ofthe majority of
quantitative psychologists (Cliff, 1992)). Such books (sec, for example. Lord &
Novick, 1968; Thorndike, 1982) typically begin with an account ofthe procedures
used to measure psychological attributes and move on to a consideration of relevant
quantitative theories, but nowhere do they explicitly discuss the empirical
commitments implicit in such theories regarding the internal structure of the
attributes involved. Nor do they discuss ways in which these commitments can be
tested experimentally.
These two facts, the widespread acceptance of Stevens' definition of measurement
amongst psychologists and the failure of books on psychological measurement to
note the character of quantitative attributes, mean that the true nature of scientific
measurement and the empirical content of the hypothesis that an attribute is
quantitative are almost universally overlooked within psychology. If a quantitative
scientist (1) believes that measurement consists entirely in making numerical
assignments to things according to some rule and (2) ignores the fact that the
measurability of an attribute presumes the contingent (and therefore, in principle,
falsifiable) hypothesis that the relevant attribute possesses an additive structure, then
that scientist would be predisposed to believe that the invention of appropriate
numerical assignment procedures alone produces scientific measurement. This is
exactly the situation that exists in quantitative psychology, a situation that Stevens'
definition serves to justify.
362 Joel Michell

4. The measurement tradition in psychology and the scientific task of

quantification ^
4.1 Fechner, pythagoreanism and the scientific task of quantification
From its inception, modern quantitative psychology was more concerned with the
implementation of a quantitative programme than with the pursuit of answers to
fundamental scientific questions about its hypothesized quantities. While there had
been attempts prior to Fechner's (1860) to establish a quantitative psychology, most
notably Herbart's (1816), these attempts failed where Fechner's succeeded because (1)
at the theoretical level, he linked his quantitative psychology to quantitative physics,
via his psychophysical law; (2) at the practical level, he supplemented his law with
a range of alleged measurement methods; and (3) at the level of rhetoric, he
persuaded others that these methods were measurement in exactly the same sense as this
term is used in physics. In doing this Fechner's motives were Pythagorean, i.e., like
Pythagoras (Burnet, 1955) and, following him, many of the greatest scientific and
philosophic minds in history (Crombie, 1994), Fechner believed that reality is
fundamentally quantitative. Both the physical and mental realms, in common, were
subordinate, he believed, ' to the principle of mathematical determination' (Fechner,
1887, p. 213). As such, psychology and, in particular, psychophysics must be
quantitative.
As an exact science psychophysics, like physics, must rest on experience and the mathematical
connection of those empirical facts that demand a measure of what is experienced or, when such
a measure is not available, a search for it (Fechner, 1860, p. xxvii).
However, Fechner's commitment to 'experience' as the basis of science was not
as firm as his commitment to pythagoreanism and when this deviation from
empiricism was combined with his inadequate understanding of the nature of
measurement, the result was a dogmatic a priorism.
The pythagoreanism that Fechner had inherited from his education in the physical
sciences was only one side of the ideological legacy bequeathed to psychology by the
scientific revolution ofthe 17th century. The other was the quantity objection, the thesis
that psychological attributes are not quantitative. To some extent, this objection had
its origins in the 17th-century relegation of the secondary qualities (like colours,
flavours, odours, etc.) to the mind (while the allegedly real, physical (i.e. non-mental)
qualities of things were held to be quantitative) and in a later, 18th-century, Kantian
view (Kant, 1786) that, as a matter of fact, psychology can never be a quantitative
science. Fechner's main critic from this perspective was von Kries (1882), who
argued that sensations do not stand in additive relations to one another and, so, the
claim that one sensation is, say, ten times another in intensity, is meaningless. James
(1890) and other psychologists (see Titchener, 1905; and Boring, 1921 for reviews)
made similar criticisms.
From the scientific point of view, the only way to counter such a criticism is to
present evidence that intensities of sensations are additive. Fechner did not see the
need to do this. He believed that because he could, using his psychophysical methods,
determine a series of stimuli in which each is just noticeably different from its
immediate predecessor, the elements of the corresponding series of sensations would
 Quantitative science and psychology 363
each differ from its predecessor by one unit of sensory intensity. Thus, thought
"^Fechner, the psychophysicist is simply counting units, albeit indirectly via stimulus
intensities, in a manner analogous to the physicist counting units of some physical
quantity. In a reply to von Kries he put the matter this way.
Given several values, in any field, which may be taken to be magnitudes inasmuch as they can
be thought of as increasing or decreasing; given the possibility of judging the occurrence of
equality and inequality in two or more of these values when they are observed simultaneously
or successively; and given that « values have been/oaW equal or, ifthey can be varied freely, have
been made equal: then it is self-evident (because it is a matter of definition and therefore a
tautology) that their total magnitude,which coincides with their sum, equals « x their individual
magnitudes. It follows that each single value, or each definite fraction or each definite multiple
of the magnitudes that have been found equal (no matter which), can be taken as the unit
according to which the total magnitude, or every fraction of it, can be measured. The « equal
parts that can be thought of as composing a total magnitude of course have the same magnitude
as the « equal parts into which the total magnitude can be thought to be decomposable. All
physical measurement is based on this principle. All mental measurement will also have to be
based on it (Fechner, 1887, p. 213).

As revealed in this quotation, Fechner's understanding ofthe logic of measurement

was seriously defective. Given an ordered series of elements, a-^, a^,.,,, a^, claiming
to show that tfj^j — a^ = a^ — a^_-^ (for all i) does not amount to showing that a^^^ — a^_i
= 2(aj^_j —Sj), unless it is also shown that ^i+i—i«i_i = ('*i+i~*i) + (*i~'*i-i)> i-C
unless it is shown that the series possesses an additive structure. That is, Fechner
needed to show that differences between sensation intensities are additive in order to
justify his claim that counting jnds is counting units of measurement. Fechner,
believing all attributes to be quantitative, thought that the only task required of him
as a scientist was the instrumental one of identifying units and counting them. That
apparently done, it was 'self-evident', he thought, that a series of what he took to
be n equal and contiguous intervals would equal n of each. The flaw in Fechner's
thinking was his pythagoreanism. It caused him to presume incorrectly that
psychological attributes must be quantitative. Thus set, his mind could not see the
force of von Kries' quantity objection.
One way in which the additivity of differences can be tested experimentally was
revealed by Holder (1901) in his axioms for stretches of a straight line. Holder was
thinking of the geometric case, but it applies to Fechner's case by analogy. This test,
in fact, is a special case of the Thomson condition in conjoint measurement (see
Krantz et al.,, 1971). If (for any f> g > h and / >J > k ranging over /,..., «) a^ — a^
— a^ — a^ and a^ — «^ ~ '^}~^k^ then the hypothesis that these differences are additive
entails that a^ — a^^ = a^ — a^, If this test is represented geometrically, as in Fig. 1, it
is obvious that the additivity of the component distances entails the conclusion
stated. Sensory differences are not geometric distances, however, and what is
obviously true ofthe latter is not necessarily true ofthe former. If individuals judge
the sense differences a^ — a^ and a^ — a^ equal, and also judge sense differences a^ — a^
and a^—a^ equal, it does not follow from that alone that they must judge a^ — a^ equal
to a^ — a^. That only follows if sense differences are quantitative. This experiment
would test this proposition.
This test requires participants making direct judgments about the equality of
sensory differences, a task which Fechner allowed could be done (Fechner, 1887).
364 foel Michell

ah ag f k j

IF ag - Eh — Hi -

AND af -

THEN af - ah = ai - ak
Figure 1. If the component distances are additive then, given the two antecedent conditions, the
consequent condition follows.

The fact that it is in principle possible for a participant's judgments to violate this
prediction and, hence, falsify a consequence of additivity, means that this is an
empirical issue. It must be investigated experimentally before any claim to scientific
measurement is justified. The logic of such tests was not known to Fechner and,
indeed, was not generally accessible in the psychological literature until Suppes &
Zinnes (1963). However, as already argued, Fechner's mind was effectively closed to
the possibility of such empirical tests of additivity and the extent of this closure is
revealed in his comment in relation to the Plateau-Delboeuf method that' We simply
call a total difference twice as large as each of two equal partial differences of which
it is composed' (Fechner, 1887, p. 214).

4.2 Fechner's psychophysics as the exemplar of psychological measurement

Fechner's mind was closed because of his commitment to the doctrine of
pythagoreanism and, being thus, he presented his psychophysical methods as
methods of measurement within a milieu that shared similar views (Hacking, 1983;
Michell, 1990). As the founding father of quantitative psychology, Fechner's work
established the quantitative paradigm in psychology and became the definitive
exemplar emulated by others. In this it established the trends of {a) dismissing the
quantity objection as 'mere writing in the sand' (Fechner, 1887, p. 215) and {b)
concentrating exclusively upon the instrumental task of quantification. Psychologists
after Fechner also ignored the first task and concentrated upon the second,
constructing number-generating procedures which, they thought, measured psycho-
logical attributes.

4.3 Spearman., the quantitative imperative, and practicalism

The truth of this last claim is shown clearly in Spearman's pioneering research on the
measurement of intelligence. Spearman had been trained in psychophysics under
Wundt and, so, naturally, had been influenced by Fechner's example. As part of that
training it is almost certain that he would have been exposed to the controversy
 Quantitative science and psychology 365
surrounding the quantity objection. Spearman proposed a quantitative theory which
purported to explain intellectual performance. He carried out tests of this theory
(such as his predictions relating to tetrad differences), but these were to do with the
number of abilities involved in solving tasks of specific kinds and were not sensitive
to the issue of whether or not the postulated abilities were quantitative in structure.
Like Fechner, he believed that psychological attributes had to be quantitative and the
primary problem was to devise procedures for their measurement.
In doing this Spearman's motivation was not quite the same as Fechner's. He was
not explicitly committed to pythagoreanism, but he did endorse the view that lacking
measurement' any study is thought by many authorities not to be scientific in the full
sense of the word' (Spearman, 1937, p. 89). This view, that measurement is a
necessary feature of all science, has been called the quantitative imperative (Michell,
1990). However, Spearman was also moved to promote ability tests as measurement
instruments by another reason as well. In his epoch-making 1904 paper, he expressed
his disappointment in psychology as a basis for applied science, especially in
education and psychiatry: '...when we without bias consider the whole actual fruit
so far gathered from this science—which at the outset seemed to promise an almost
unlimited harvest—we can scarcely avoid a feeling of great disappointment' (1904,
p. 203). Spearman believed that psychology should provide a quantitative basis for
practical applications in these areas and he expressed the hope that his research would
produce 'practical fruit of almost illimitable promise' (1904, p. 206).
Practicalism, the view that science should serve practical ends, when stronger than
the spirit of disinterestedness, can corrupt the process of investigation. Science, as the
attempt to understand nature's ways of working, knows nothing of practicalism, for
scientific knowledge is neither useful nor useless, in itself. It only becomes so when
taken in relation to interests other than the scientific, interests which may shift with
the tide of social change. When scientific questions seem intractable, an impatient
practicalism may presume upon nature. Unable to admit that such questions cannot
yet be answered, the practicalist may blindly collude in the pretence that they do not
exist by ignoring them, especially when the psychological expectations and social
rewards are high. Spearman thought psychological measurement a necessity,
practicalism an imperative, and he ignored the scientific issue of quantification.
Hence, the hypothesis of a causal connection cannot be ruled out. In these attitudes
he was not alone.

4.4 Applied psychology, practicalism and the instrumental task

Practicalism became a powerful motive driving the instrumental side of quantitative
psychology. In America, the impetus for applied psychological measurement came
from the work of J. McK. Cattell and E. L. Thorndike. Cattell believed, like
Spearman, that' Psychology cannot attain the certainty and exactness of the physical
sciences, unless it rests on a foundation of experiment and measurement' (1890,
p. 373) and wrote freely about mental measurements (1893) with respect to a range of
psychological procedures, with no acknowledgement of the scientific task of
quantification, even though he endorsed the traditional scientific concept of
366 Joel Michell

measurement ('all measurement depending on ratios'; 1893, p. 321). In advocating

mental measurements, he had applied psychology firmly in mind:
Control ofthe physical world is secondary to the control of ourselves and our fellow man... If
I did not believe that psychology affected conduct and could be applied in useful ways, I should
regard my occupation as nearer to that of the professional sword-swallower than to that of the
engineer or scientific physician (Cattell, 1904, as quoted in Brown, 1992, p. 3).
That engineering and medicine should have been consistently selected as the
guiding metaphors for applied psychology (Brown, 1992) reveals a connection
between practicalism and quantification. Engineering is applied physics and physics
is the paradigm of quantitative science. If applied psychology was seen in this light,
then the pressure exerted upon psychology by practicalism to claim to be able to
measure would have been great indeed. This metaphor occurred repeatedly (e.g.
Terman, 1916; see also Brown, 1992). Medicine had also recently made great strides
through the introduction of quantitative methods derived from physiology and,
likewise, provided an example that applied psychologists strove to emulate (Brown,
1992).
Thorndike, agreeing with Fechner's pythagoreanism, had said that 'Whatever
exists at all exists in some amount. To know it thoroughly involves knowing its
quantity' (quoted in Clifford, 1968, p. 283). His view that 'Any mental trait in any
individual is a variable quantity' (1904, p. 22) is an immediate consequence.
Conjoined with the engineering metaphor it entails that 'Education is one form of
human engineering and will profit by measurements of human nature and
achievement as mechanical and electrical engineering have profited by using the foot-
pound, calorie, volt and ampere' (quoted in Brown, 1992, p. 119). While Thorndike
was aware that measurement in psychology (' by relative position') was different from
that in physics ('by amount of some unit'), he believed that 'Measurement by relative
position in a series gives as true, and may give as exact, a means of measurement as
that by units of amount' (1904, p. 19). Of course, to assert that, in a quantitative
order, a particular value, b, falls between two others, a and c, is never as exact a
specification of its relation to other values (to a, for example) as is given by specifying
its numerical relation to a unit (for the latter will always entail that b = ra, where r
is a positive real number, while the former never does). However, in asserting this
Thorndike was giving psychologists permission to use the term measurement for
practices which were not supported by any scientific evidence of quantity. The*
advantages of this terminology to applied psychologists attempting to present
themselves as applied scientists is obvious. It allowed them to present themselves as
applied scientists in what was an easily identified manner. This terminology rapidly
became standard and it was a commonplace observation that 'Tests are the .devices
by which mental abilities can be measured' (Viteles, 1921, p. 57).
This way of thinking, combined with the successful incorporation of tests in
American society following World War I, meant that the scientific task of
quantification was easily ignored. Some idea of how extensive the use of tests in
applied psychology was may be gauged by a number of indices. By 1922, three
million children a year were subjected to one form or another of mental measurement
(Thorndike, 1923). By 1937, '5005 articles, most of them reports of new tests, which
[had] appeared during the fifteen year period between 1921 and 1936' (South, 1937,
 Quantitative science and psychology 367
as quoted in Hornstein, 1988) were available for use by psychologists. Terman (1921)
reported that 'more than half of the psychological research which is being carried on
by members of the American Psychological Association (which includes practically
all the psychologists of the United States) falls in one or another of the fields of
applied psychology' (p. 3).
Some idea of the proportion of applied psychological research that was devoted to
measurement can be gained from considering the number of pubhcations reporting
mental measurements of one kind or another published in the Journal of Applied
Psychology from its inception in 1917 to 1946 (the year of the pubhcation of Stevens'
definition). From 1917 to 1926 the number was 150 (out of 338, or 44.4 per cent);
from 1927 to 1936, 234 (out of 585, or 40 per cent); and from 1937 to 1946, 253 (out
of 668, i.e., 38 per cent). In these studies, scores (generally obtained via counting
responses of a certain kind) or transformations of scores were typically treated as
measures of some psychological attribute or other. In the spirit of both Fechner and
Spearman, the scientific question of the additivity of the attributes involved was
hardly ever addressed. That is, decades prior to the publication of Stevens' definition,
the practices of apphed psychologists already conf^ormed to it.

4.5 Psychological measurement as a scientific anomaly

The logic of applied science is such that if there is no science there can be no apphed
science, so whatever these psychologists were doing it was not applied science.
Actually, they were doing what so-called 'applied psychologists' haye, in general,
always done. They were applying a methodology to what were thought of as practical
problems (Freyd, 1926; Terman, 1924). This methodology was based upon the
construction of procedures that yield numerical data. Such procedures, innocent
enough in themselves, were then packaged and marketed as forms of 'scientific
measurement' and, as such, constituted a pretence of applied science, rather than
applications of empirically confirmed scientific theories. Constructing psychological
tests for practical applications may be a useful thing to do. However, its relation to
psychology as a science needs to be clarified. Even if it is found that performance
upon a particular test (say, test A) is useful for predicting some criterion (say, success
in a training course, X), this by itself is not applied science in any meaningful sense.
The discovery that A predicts X raises a scientific issue, it does not solve one and
neither does it amount to the apphcation of a well-confirmed body of scientific theory
to a practical problem as, for example, engineers apply physics to the building of a
bridge. The scientific issue raised by such a discovery is this: why does performance
on A predict performance on X? In attempting to answer such a question, a
psychologist may theorize that test A measures intellectual ability 7 and 7, in turn, is
a cause of performance on X. This is a perfectly respectable way to theorize, but it is
only one possible theory out of an indefinitely large array of possible theories and
remains so until thoroughly investigated empirically. A part of this process of
investigation must involve testing the hypothesis that ability 7 is a quantitative
attribute. If this is not done then the claim that A measures I remains completely
speculative. In this case, the use of test A to 'measure' 7 is not apphed science and it
368 Joel Michell

is misleading to think of it in this way. Until the scientific task of quantification is

completed, claiming that a procedure measures anything is premature.
Hence, both quantitative psychology and 'applied psychological measurement'
stood as anomalies within a discipline purporting to be a science because
psychologists cither declined to or did not know how to consider fundamental
scientific issues and persistently presented their procedures as amounting to more
than was justified scientifically. This anomaly was repeatedly noted by those who
took the care to understand the character of scientific measurement. I have already
mentioned the quantity objection to Fechner's work. Again, in 1913, at a joint
symposium organized by the Mind Association, the British Psychological Society
and the Aristotelian Society, the issue of whether or not sensory differences are
quantitative was considered by Brown (1913), Dawes Hicks (1913), Myers (1913) and
Watt (1913). Boring, who was to consider the quantity objection in some detail in his
paper on the stimulus error (Boring, 1921), had written a year earlier that
psychologists were 'not yet ready for much psychological measurement in the strict
sense' (1920, p. 32), a theme he reportedly continued to maintain more than a decade
later (Newman, 1974). Psychological tests were subjected to similar criticisms (e.g.
McCormack, 1922). In the 1930s, other scholars (e.g. Adams, 1931; Brown, 1934;
Johnson, 1936) made many ofthe same criticisms again. Thus, there remained within
psychology a critical trajectory that surfaced from time to time, from von Kries until
the 1930s. The mainstream of quantitative psychologists paid it no heed, however.
As long as the criticism was internal to the discipline and consisted of only a few
voices, it could be ignored with impunity. However, when it became external, with
some kind of official backing, notice was taken of it.

4.6 The Ferguson Committee

When in 1940, a committee established by the British Association for the
Advancement of Science to consider and report upon the possibility of quantitative
estimates of sensory events published its final report (Ferguson et al., 1940) in which
its non-psychologist members agreed that psychophysical methods did not constitute
scientific measurement, many quantitative psychologists realized that the problem
could not be ignored any longer. Once again, the fundamental criticism was that the
additivity of psychological attributes had not been displayed and, so, there was no
evidence to support the hypothesis that psychophysical methods measured anything.
While the argument sustaining this critique was largely framed within N. R.
Campbell's (1920,1928) theory of measurement, it stemmed from essentially the same
source as the quantity objection.
At this juncture in the history of psychology two avenues were open. One was to
admit the vahdity of these criticisms and so admit that the scientific issue of whether
or not psychological attributes are quantitative had not been adequately addressed.
In doing this, psychologists would have been admitting that their claim to be able
to measure their attributes rested upon theory and speculation rather than upon
direct scientific evidence. They could have put a brave face to the scientific world and
claimed with Bartlett (1940, p. 441) that 'Scientific insight, as everyday perception.
 Quantitative science and psychology 369
has ever run ahead of measurement and mathematical proof. The next step would
have been to explore ways of testing the question begged throughout their history
and a good starting point would have been Holder (1901) (which Nagel, 1932, had
recently brought to the attention of philosophers).

4.7 Stevens' attempt at a rational reconstruction of psychological measurement

However, this was not the path taken. The combined pressure of scientism (in the
guise of pythagoreanism and the quantitative imperative) and practicalism was too
strong. One finds in the psychological journals from 1940 to 1950 a rash of papers
attempting to defend psychological practices and, sometimes, to redefine measure-
ment in a way that legitimized psychology's claim to the scientific high ground it had
never actually occupied (e.g. Bartlett, 1940; Bergmann & Spence, 1944; Brower,
1949; Comrey, 1950; Coombs, 1950; Cureton, 1946; GuUiksen, 1946; Nafe, 1942;
Perloff, 1950; Reese, 1943; Thomas, 1942). This process culminated in Stevens' early
papers on measurement theory (Stevens, 1946, 1951).
Like other quantitative psychologists, Stevens endorsed the quantitative im-
perative: 'It can be said that the history of science is the history of man's efforts to
devise procedures for measuring and quantifying the world around him' (1967,
p. 734). Thus, like Fechner, he presumed that psychophysical measurement was
possible. It was Stevens' sone scale of loudness that the Ferguson Committee
considered as a putative example of psychophysical measurement. He believed that
his psychophysical methods produced scales of'true numerical magnitude' (1936^,
p. 406), so the contradiction of this claim by eminent members of that committee
spurred him to defence. Short of undertaking the intellectual and scientific labour
necessary to test that claim empirically, what Stevens required was an effective
rationalization which would render such empirical tests apparently redundant. One
can admire Stevens' resourcefulness in constructing this rationalization. As an
exercise in rhetoric it displayed considerable creativity. However, in science all is not
rhetoric. It is the role of the scientist, as far as possible, to let the facts speak for
themselves. Instead, Stevens constructed a solution that obscured the facts from
view.
His rationahzation was a two-tiered ideological structure. The first layer was a
reconstruction of the representational theory of measurement (the theory that
measurement is the numerical representation of empirical relations). Russell (1897)
had already attempted to undermine the traditional concept of measurement by
trying to disengage the concept of number from that of quantity (Michell, in press).
Following this he (1903) provided the first systematic presentation of the
representational theory of measurement (Michell, 1993). However, it was not a
completely thoroughgoing representationahsm. Even less so was Campbell's (1920,
1928) later version, which had guided the thinking of many on the Ferguson
Committee, for it really attempted httle more than to translate the traditional concept
of measurement into representational terms, requiring additivity as a necessary
condition for all measurement. Stevens was, I think, the first to see clearly that basing
measurement upon the concept of numerical representation freed it from exclusive
dependence upon the concept of additivity or that of any specific relation beyond
370 foel Michell

equivalence. For Stevens, 'measurement is possible in the first place only because
there is a kind of isomorphism between (1) empirical relations among objects and
events and (2) the properties of...' numerical systems (Stevens, 1951, p. 1). From this
starting point he developed his theory of the four possible types of measurement
scales (nominal, ordinal, interval and ratio) (he later, 1959, added the log-interval
scale) and the associated doctrine of permissible statistics (see Michell, 1986).
This layer of his reconstruction was a masterstroke because it at once disarmed
Campbell and his associates on the Ferguson Committee of their most powerful
weapon. Their criticism of psychophysical measurement, that it was not based upon
the demonstration of any relevant additive relation between sensory intensities, was
made to look as if it depended upon an unnecessarily restrictive version of the
representational theory of measurement. However, this variety of liberalized
representationalism also posed a threat to psychological measurement and, especially,
to Stevens' psychophysical methods. If measurement involves the numerical
representation of empirical relational structures and such structures are understood
realistically (i.e. as structures existing independently of the scientific observer), then
measurement still requires a logically prior scientific stage in which the hypothesis is
tested that relations of the required kind hold within a particular empirical domain.
Even the humble nominal scale would require the demonstration of a reflexive,
transitive, and symmetric empirical relation of equivalence (or sameness with respect
to some attribute) and for most putative instances of psychological measurement not
even this much scientific work had been done. Hence, the second layer of Stevens'
reconstruction required a repudiation of such a realist interpretation of the empirical
structures numerically represented in measurement.
If Stevens' first layer was a masterstroke, his second was audaciously bold, for he
replaced the natural scientific attitude of realism with a form of relativistic
subjectivism. The 1930s had been a time of ferment within the philosophy of science,
with the newer views of logical positivism and operationism challenging older
positions (see Passmore, 1957). Stevens adopted both operationism (Hardcastle,
1995) and logical positivism (Stevens, 1939). His representational theory of
measurement, with its emphasis upon the numerical representation of directly
observable empirical relations and its formalist conception of numerical systems was
essentially positivistic. His bolder vision, however, was to construct an operational
interpretation of the representational theory. He took Bridgman's (1927) opera-
tionism, according to which the meaning of a concept is synonymous with the
operations used to identify it and applied it to psychology (Stevens, 1935a, b, \9?)(>a, b,
1939), agreeing with Bridgman that 'the meanings ofour words can never transcend
the operations which went into their determination' (1936a, p. 93). Operationism has
more in common with Berkeley's ideahsm than with empirical realism (Michell,
1990) and this is not a judgment with which Bridgman would have disagreed (see
Bridgman, 1950) and Stevens (1936a) also endorsed its implied subjectivism. What
is of interest here, however, is the use to which Stevens put Bridgman's philosophy.
In the first place, operationism was woven by Stevens into an elaborate, self-
serving, philosophy. Across a number of papers in the 1930s, he argued (1) that
because operationism implies a relativity to the observer in all science, psychology is
the 'propaedeutic science', the study of the observer (1936a); (2) all scientific
 Quantitative science and psychology 371
operations are reducible to the operation of sensory discrimination (1935a); (3) the
operational methods of psychophysics are central to the study of sensory
discrimination (1935^); and (4) the scaling methods advanced by Stevens allow the
central question of psychophysics to be decisively answered (1936^). Operationism
enabled Stevens to believe that his research was the rock upon which all science
stood.
Secondly, operationism enabled him to believe that his psychophysical methods
yielded ratio scale measurement of the intensity of sensations without research into
the scientific task of quantification being necessary. If the meaning of a scientific
concept is given by the operations (i.e. procedures) used to identify it, then it follows
that the empirical relations numerically represented in measurement must hkewise be
defined by the operations used to identify them (Bergmann & Spence, 1944). Thus,
for Stevens, it could just as truthfully be said that measurement is possible because
of an isomorphism between empirical relations and numerical ones, as because of an
'isomorphism between the formal system and empirical operations' (Stevens, 1951,
p. 23, my italics). The point here is that ordinarily a distinction is made between a
relation into which things enter (e.g. the relation of object x being heavier than
object j ) and an operation or procedure used to identify such a relation (e.g. by
placing X andj in the pans of a balance and observing that the arm of the balance
supporting X tilts down). The fact that x is heavier thanj would normally be thought
of as a necessary condition for this outcome of the operation. However, because of his
operationism, Stevens would confuse two such facts, asserting that the operation
with its outcome is all that is meant by asserting that the relation holds. If ordinal
numerical assignments were made to x andj they could, on Stevens' view, be taken
to represent the above operation as easily as the above relation. When the operation
itself involves making numerical assignments (as, say, with Stevens' psychophysical
methods or with mental testing), these assignments may be taken, according to this
operationist logic, to both define the relation represented and to represent it. It is
only if this point is grasped that one can appreciate the conceptual unity between
Stevens' definition of measurement as the assignment of numerals to objects or events
according to rule and his representationahsm.
Given operationism, any rule for assigning numerals to objects or events could be
taken as providing a numerical representation of at least the equivalence relation
operationally defined by the rule itself. Hence, any rule for making numerical
assignments always defines at least a nominal scale, according to Stevens' view. This
is why Stevens was able to rephrase his definition as ' the assignment of numerals to
objects or events according to rule—any rule' (1959, p. 19) without feeling that he
had shifted his ground one inch. Of course, to the realist, this way of thinking is
viciously circular and there appears to be an hiatus between Stevens' definition of and
theory of measurement (Michell, 1986). This is because the realist views the relations
represented in measurement as having an existence independent of human
observations or operations.
When this operational way of thinking was applied to his psychophysical methods
it gave Stevens what he wanted. In the first place, it enabled him to reject the concept
of' private or inner experience for the simple reason that an operation for penetrating
privacy is self-contradictory' (1936a, p. 95) and to conclude that what psycho-
372 foel Michell

physicists had hitherto thought of as 'a subjective scale is a scale of response' (1936^,
p. 407). Then, further applying operationism, the relation of one tone's sounding half
as loud, say, as another will be defined by the operation used to determine it, i.e. by
a subject judging it to be so. Thus, Stevens was able to believe that'... the response
ofthe observer who says "this is half as loud as that" is one which, for the purpose
of erecting a subjective scale, can be accepted at its face value' (1936^, p. 407). T"hen
it follows, he thought, that such a scale is additive because 'With such a scale the
operation of addition consists of changing the stimulus until the observer gives a
particular response which indicates that a given relation of magnitudes has been
achieved' (1936^, p. 407). According to Stevens, an additive (or ratio) scale is
obtained because the person is both instructed to judge and taken to be judging
additive or numerical relations. In this case, the operation by which the numerical
assignments are made was taken by Stevens to define the ratio scale that he believed
was produced.

4.8 Its implications for the scientific task of quantification

Thus Stevens had both deflected the criticisms of the Ferguson Committee by
adopting a thoroughgoing representationalism and protected his own psychophysical
methods from scientific challenge by adopting an operationist interpretation of
representationalism. Both Fechner and Stevens thought that their methods could be
taken as methods of measurement without any further scientific justification. In so
doing, Fechner established the psychological tradition of regarding number-
generating procedures as measurement, a tradition strengthened by Spearman and set
in concrete by applied psychometrics. Stevens then enshrined the practices of that
tradition within an explicit definition of measurement. Prior to its formulation
psychologists were already in the habit of regarding as measurement procedures for
making numerical assignments to objects or events and were in the habit of ignoring
the scientific issues relating to quantification. Stevens' definition perfectly matched
this practice and appeared to legitimize it. For this reason it was rapidly absorbed into
psychology's ideological support structures, soon after publication being cited in
major texts as the only definition of measurement (e.g. Green, 1954; Guilford, 1954;
Lorge, 1951), being referred to in leading journals as the 'classical' theory of
measurement (Coombs, RaifFa & Thrall, 1954), a characterization that persisted
(Fraser, 1980), and eventually projected back upon Fechner himself (Adler, 1980).
That it blinded the majority of psychologists to the scientific necessity of testing
via experiment that psychological attributes are quantitative was dramatically
revealed over the subsequent four decades. What may be described as a broadly
realist interpretation of Stevens' thoroughgoing representational theory of measure-
ment was worked out only just over a decade later. Suppes & Zinnes (1963)
considered a wide range of numerically representable empirical relational structures.
In many cases they stated necessary and sufficient empirical conditions for numerical
representations of one kind or another. Their theory implied a much stricter
definition of measurement than did Stevens'. Measurement was a homomorphism
between independently existing empirical and numerical relational systems. Because
of the required independence of the two systems involved, such a homomorphism is
 Quantitative science and psychology 373
never automatically obtained merely by having a rule for making numerical
assignments. Furthermore, if the empirical relational system is quantitative (as
required within quantitative theories), then the scientific task of quantification is
made exphcit. This made clear to the psychological scientific community the fact that
measurement required attention to fundamental empirical issues and, by imphcation,
condemned the non-empirical measurement tradition of Fechner, Spearman and
Stevens. Then, a year later, perhaps the most important development in measurement
theory since Holder (1901) occurred when Luce & Tukey (1964) published the
theory of conjoint measurement (which was anticipated to some extent by Adams &
Fagot, 1959, and Debreu, 1960). This theory and its subsequent developments (see
Krantz et al., 1971) revealed a range of decisive, indirect tests for the hypothesis that
attributes are quantitative. This research programme culminated in the publication of
volume three of Foundations of Measurement (Luce et al., 1990). It makes explicit the
conditions under which apparently non-additive empirical structures are really
additive at a deeper level (viz. when automorphisms of such structures constitute a
simply ordered, Archimedean group; see also Luce, 1987). While this research was
directly relevant to the scientific task of quantification, it inspired only a relatively
small number of empirical studies towards that end, leaving that task still seriously
incomplete. For the most part, mainstream quantitative psychology remained
oblivious to these revolutionary developments in measurement theory despite the
fact that they occurred within the discipline and were published in one of its leading
journals {Journal of Mathematical Psychology). It is of interest to note, also, that Stevens'
preferred methods of psychophysical measurement (magnitude estimation and cross-
modality matching) were analysed theoretically by some of those associated with this
programme (e.g. Krantz, 1972; Luce, 1990; Narens, 1996). These analyses state
empirical conditions which must be true if these methods are actually measuring the
psychological attribute of sensation intensity.
Almost at the inception of this research programme, Stevens, still oblivious to the
scientific issues involved, belittled its achievements, claiming that 'measurement
models drift off into the vacuum of abstraction and become decoupled from their
concrete reference' (Stevens, 1968, p. 854) and, thereby, demonstrated his persisting
blind spot. So complete is this failure to see the obvious that a recent commentary
(Cliff, 1992) upon the failure of mainstream quantitative psychology to absorb the
conceptual breakthroughs of this programme effectively laid blame upon that
programme itself (a charge, incidentally, rejected by Narens & Luce, 1993). Cliff
argued that its lack of influence resulted primarily from the difficulty level of the
mathematics used in reporting its research results, the lack of demonstrated empirical
power of the results obtained, and the difficulty of dealing with error. He admitted
that ' factors that have to do with the Zeitgeist and the habits of work and thought
among psychologists' (p. 189) also played a role, but there was no recognition ofthe
fact that those who understand measurement in the way defined by Stevens must of
necessity regard ways of addressing the scientific task underpinning quantification as
not only irrelevant to psychological measurement but as placing an unnecessary
obstacle in the way of its development and application.
It should be noted that the minority tradition of those critical of psychological
measurement, spawned a small but continuous series of experimental studies
374 Joel Michell

engaging the scientific task, especially in psychophysics (e.g. Beck & Shaw, 1967;
Gage, 1934^,^; Garner, 1954; Gigerenzer & Strube, 1983; Levelt, Riemersma &
Bunt, 1972; Reese, 1943; Zwislocki, 1983). Here is not the place to review this work
but, in general, its thrust does not warrant its consistent neglect by those endorsing
Stevens' definition of measurement and advocating psychological methods as
measurement procedures.

5. Psychological measurement and methodological thought disorder

5.1 The concept of thought disorder
By methodological thought disorder, I do not mean simply ignorance or error, for
there is nothing intrinsically pathological about either of those states. Ignorance has
many causes and not all indicate a cognitive fault. Likewise, when error occurs under
certain conditions, it can be construed as part of the normal functioning of the
cognitive system (e.g. the standard geometric, visual illusions). Ignorance and error
are only pathological when some mechanism within that system sustains them under
external conditions favourable to their correction. Hence, the thinking of one who
falsely believes he is Napoleon is adjudged pathological because the delusion was
formed and persists in the face of objectively overwhelming contrary evidence. I take
thought disorder to be the sustained failure to see things as they are under conditions
where the relevant facts are evident.
Hence, methodological thought disorder is the sustained failure to cognize relatively
obvious methodological facts. It is well known that many psychologists are ignorant
of important methodological facts and their methodological thinking is often
erroneous (e.g. Rosnow & Rosenthal, 1989; Zuckerman, Hodgins, Zuckerman &
Rosenthal, 1993). That, itself, is sufficient cause for deep concern and a subject
worthy of serious scientific research. I am interested, however, not so much in
methodological ignorance and error amongst psychologists per se, as in the fact of
systemic support for these states in circumstances where the facts are easily accessible.
Behind psychological research exists an ideological support structure. By this I mean a
discipline-wide, shared system of beliefs which, while it may not be universal,
maintains both the dominant methodological practices and the content of the
dominant methodological educational programmes. This ideological support
structure is manifest in three ways: in the contents of textbooks; in the contents of
methodology courses; and in the research programmes of psychologists. In the case
of measurement in psychology this ideological support structure works to prevent
psychologists from recognizing otherwise accessible methodological facts relevant to
their research. This is not then a psychopathology of any individual psychologist.
The pathology is in the social movement itself, i.e. within modern psychology.

5.2 The systemic character of methodological thought disorder in psychological measurement

Stevens did not only give modern psychology his definition of measurement. He also
gave it his theory of scales of measurement (e.g. Stevens, 1946, 1951, 1959). As with
his definition of measurement, this theory has been largely absorbed into the
 Quantitative science and psychology 375
collective wisdom of psychology and most psychological researchers and students are
familiar with its details. One thing that is clear from this theory is that postulating
quantitative relations between attributes presumes, at least, what he called, ' interval
scale' measurement. This is why most quantitative theories in psychology, if they pay
heed to the issue at all, stipulate this level of measurement (in Stevens' sense). The
following comment is typical:
The level of measurement most often specified in mental test theory is interval measurement, which
yields an interval scale. This scale presupposes the ordering property of the ordinal scale but, in
addition, specifies a one-to-one correspondence between the elements of the behavioral domain
and the real numbers, with only the zero point and the unit of measurement being arbitrary. Such
a scale assigns meaning not only to scale values and their relative order but also to differences
of scale values (Lord & Novick, 1968, p. 21).

Now, if an interval scale requires an isomorphism between elements of the

behavioural domain (by which, I take it. Lord and Novick simply mean the
psychological attribute) and the real numbers, then it should be obvious that an
additive structure within the psychological attribute is being presumed. That is,
Stevens' theory of measurement scales (a theory widely accepted within psychology)
together with the fact that quantitative theories require at least interval scale
measurement (a position implicitly endorsed by most quantitative psychologists)
entails the conclusion that the relevant attributes must have a structure similar in
some respects to that of the real number system. (Strictly speaking, with an interval
scale, it is differences between magnitudes of the attribute that are quantitative.)
However, while the majority of psychologists accept these two premises, something
of a general, systemic nature within psychology prevents them from seeing the
conclusion entailed.
Now, of course, grasping this implication does not involve recognizing the precise
character of quantitative attributes. Quantitative science existed for thousands of
years before the structure of the real numbers was finally articulated (Dedekind,
1872), enabling quantitative structure itself to be correctly characterized (Holder,
1901). However, the nature of the structure of the real number system is now
described in the most elementary of university algebra texts (e.g. Birkhoff &
MacLane, 1965) and there is an extensive mathematical and philosophical literature
available on the structure of quantitative attributes (e.g. Behrend, 1953, 1956;
Mundy, 1987; Nagel, 1932; Suppes, 1951; Swoyer, 1987; Whitney, 1968^,^), the
relevant portions of which have been available in specifically psychological hterature
since WeitzcnhofFer (1951) (see also Suppes & Zinnes, 1963). That is, the relevant
information has been available in the psychological literature on measurement
theory, even if not in the hterature on psychometrics, since the time Stevens'
definition was first proposed. Thus, other things being equal, the interested
psychologist could easily have gleaned the relevant facts and, having done so,
recognized the contingent character of the hypothesis that a psychological attribute
is quantitative (i.e. at least measurable on an interval scale). Then the interested
psychologist who happened to be also a committed empiricist, would have
recognized that claims to measure such a psychological attribute in the absence of
independent evidence for its quantitative structure have, like resorts to mere
postulation in any area of science, all of 'the advantages of theft over honest toil'
376 Joel Michell

(Russell, 1920; as quoted by Stevens, 1951, p. 36). Thus, the interested empirical
psychologist would have come to see that Stevens' definition of measurement is
nonsense and the neglect of quantitative structure a serious omission by quantitative
psychologists. Of course, some psychologists did follow something like this route
and periodically critiques of Stevens' definition were published (e.g. Ross, 1964;
Rozeboom, 1966). If some psychologists were able to travel this route, why not all?
The widespread acceptance of Stevens' definition within quantitative psychology
created an intellectual environment in which measurement seemed easily attainable
without travelling this route. This prevented the recognition of these otherwise
evident facts. That is, we are dealing with a case of thought disorder, rather than one
of simple ignorance or error and, in this instance, these states are sustained
systemically by the almost universal adherence to Stevens' definition and the almost
total neglect of any other in the relevant methodology textbooks and courses offered
to students.
The conclusion that follows from this history, especially that of the last five
decades, is that systemic structures within psychology prevent the vast majority of
quantitative psychologists from seeing the true nature of scientific measurement, in
particular the empirical conditions necessary for measurement. As a consequence,
number-generating procedures are consistently thought of as measurement pro-
cedures in the absence of any evidence that the relevant psychological attributes are
quantitative. Hence, within modern psychology a situation exists which is accurately
described as systemically sustained methodological thought disorder.

5.3 Paradigms of measurement

The traditional view of scientific measurement and the view represented by Stevens'
definition are di^iSet&nt paradigms in Kuhn's (1970) sense (sec also Michell, 1986).
However, this does not mean that there is no basis for making an informed and
rational choice between them. The operationism behind Stevens' definition
contradicts the realist view that the subject matter of science is logically independent
of the observer. If the subject matter of science (the quantitative attributes studied,
for example) is constituted by the operations used to study it, as Stevens clearly
thought was the case with his psychophysical scales, then there is no it to study.
Science simply reduces to the study of our operations and cannot be construed as the
study of an independently existing world whose secrets we penetrate via these
operations. Only a realist view docs justice to the concept of scientific discovery.
On this basis then, the widespread acceptance of Stevens' definition within
psychology is an aberration for it does not mesh with an empirical reahst view of
science. It was accepted within psychology, not because psychologists were
converted to the subjectivism of Bridgman's operationism, but because at the time
it seemed the only way forward. Stevens' definition seemed to justify the quantitative
practices that had developed within psychology. Such a misperception could only
have been made in the first place and sustained for half a century because
psychologists, generally, remain ignorant about the logic of science and, in particular,
about the logic of quantification.
 Quantitative science and psychology "hll

5.4 The logic of science and the category of quantity

This ignorance is due, not only to the systemic cause detailed above, but also, it
should be said, to a loss of intellectual nerve on the part of philosophers of science.
For whatever reason, many have been mesmerized by the relativism deriving from
Kuhn (1970), Feycrabend (1975) and others and nervous of asserting a logic of
science that does justice to its enormous achievements, a point also noted elsewhere
(e.g. Stove, 1982; Theocharis & Psimopoulos, 1987). This vacuum has given
psychologists the intellectual space to utilize Stevens' definition as a convenient
rationalization without criticism from philosophers of science. A final cause of this
ignorance is the fact that methodological education within psychology not only
reproduces Stevens' definition but, also, has been maintained at a particularly low
level by the social and managerial structures regulating academic psychology (Aikcn,
West, Sechrest & Reno, 1990).
If psychologists are to be rc-cducated in the logic of science then the concept of
quantity needs to be seen as one of science's fundamental categories and the manner
in which both the concept of measurement and its logic unfolds from this concept
emphasized. The conceptual and mathematical foundation upon which such re-
education could be based has been developed this century, from Holder (1901) to
Luce et al. (1990), in ways that are genuinely insightful and revolutionary. This body
of mathematical and philosophical knowledge provides an opportunity for
quantitative psychologists to remake their discipline as a science rather than a
pretence. A narrowly conceived empiricism that ignores relevant conceptual and
philosophical issues has here been shown to be intellectually bankrupt. This chapter
in the history of psychology shows that
if the work of inquiry is to be carried on, it must be at once scientific and philosophic, that if, in
particular, the scientist is not philosophic, he will fall into confusions, he will rebuff philosophic
criticism—he will lack a theory of categories, of sorts of problem, of'method'—especially he will
be carried away by practical interests, by interest in producing something or implementing a
programme instead of in finding something out (Anderson, 1962, p. 183).
Acknowledgements
Research for this paper was done in 1995 while at Centrum voor Mathematische Psychologie en
Psychologische Methodologie, Department Psychologie, Katholieke Universiteit Leuven, Belgium. I
am grateful to the Board of Administration of KUL for appointing me Visiting Professor and to
Professors Luc Delbeke and Paul De Boeck for arranging my appointment and for generous help while
there. I also thank the University of Sydney for granting the period of study leave from normal duties
that made this visit possible. This research has been supported financially by ARC Grants from the
Federal Government of Australia and I gratefully acknowledge this assistance. Valuable help was
received from my research assistants, Fiona Hibberd and Kate Toms. I would also like to thank both
the reviewers and editors of this journal for their constructive comments. Other useful comments were
given by Dr Agnes Petocz and the members of the Psychology IV, Conceptual Foundations of
Quantitative Methods Seminar, University of Sydney.

References
Adams, E. W. & Fagot, R. F. (1959). A model of riskless choice. Behavioral Science, 4, 1-10.
Adams, H. F. (1931). Measurement in psychology. Journal of Applied Psychology, 15, 545-554.
Adler, H. E. (1980). Vicissitudes of Fechnerian psychophysics in America. In R. W. Rieber & K.
Salzinger (Eds), Psychology: Theoretical-historical Perspectives, pp. 11-23. New York: Academic Press.
378 Joel Michell
Aiken, L. S., West, S. G., Sechrest, L. & Reno, R. R. (1990). Graduate training in statistics,
methodology, and measurement in psychology. American Psychologist, 45, 721-734.
Anderson, J. (1962). Studies in Empirical Philosophy. Sydney: Angus & Robertson.
Bartlett, R. J. (1940). Measurement in psychology. Advancement of Science, 1, 422-441.
Beck, J. & Shaw, W. A. (1967). Ratio-estimations of loudness-intervals. American Journal of Psychology,
80, 59-65.
Beckwith, T. G. & Buck, N. L. (1961). Mechanical Measurements. Reading, MA: Addison-Wesley.
Behrend, F. A. (1953). A system of independent axioms for magnitudes. Journal and Proceedings of the
Royal Society of NSW, 87, 27-30.
Behrend, F. A. (1956). A contribution to the theory of magnitudes and the foundations of analysis.
Mathematische Zeitschrift, 63, 345-362.
Bergmann, G. & Spence, K. W. (1944). The logic of psychophysical measurement. Psychological Review,
51, 1-24.
Birkhoff, G. & MacLane, S. (1965). A Survey of Modern Algebra. New York: MacmiUan.
Boring, E. G. (1920). The logic ofthe normal law of error in mental measurement. American Journal of
Psychology, 31, 1-33.
Boring, E. G. (1921). The stimulus-error. American Journal of Psychology, 32, 449-471.
Bostock, D. (1979). Logic and Arithmetic, vol. 2, Rational and Irrational Numbers. Oxford: Clarendon
Press.
Bridgman, P. W. (1927). The Logic of Modern Physics. New York: MacmiUan.
Bridgman, P. W. (1950). Refiections of a Physicist. New York: Philosophical Library.
Brower, D. (1949). The problem of quantification in psychological science. Psychological Review, 56
325-333.
Brown, J. (1992). The Definition of a Profession: The Authority of Metaphor in the History of Intelligence
Testing, 1890-1930. Princeton, NJ: Princeton University Press.
Brown, J, F. (1934). A methodological consideration ofthe problem of psychometrics. Erkenntnis, 4,
46-61.
Brown, W. (1913). Are the intensity differences of sensation quantitative? IV. British Journal of Psychology,
6, 184-189.
Burnet, J. (1955). Greek Philosophy: Thales to Plato. London: MacmiUan.
Campbell, N. R. (1920). Physics, The Elements. Cambridge: Cambridge University Press.
Campbell, N. R. (1928). An Account ofthe Principles of Measurement and Calculation. London: Longmans,
Green.
Cattell, J. McK. (1890). Mental tests and measurements. Mind, 15, 373-380.
Cattell, J. McK. (1893). Mental measurement. Philosophical Review, 2, 316-332.
Cattell, J. McK. (1904). The conceptions and methods of psychology. Popular Science Monthly, pp.
176-186.
Cattell, R. B. (1943). The measurement of adult intelligence. Psychological Bulletin, 40, 153-193.
Cliff, N. (1992). Abstract measurement theory and the revolution that never happened. Psychological
Science, 3, 186-190.
Clifford, G. J. (1968). Edward L. Thorndike: The Sane Positivist. Middletown, CT: Wesleyan University
Press.
Clifford, W. K. (1882). Mathematical Papers. London: MacmiUan.
Comrey, A. L. (1950). An operational approach to some problems in psychological measurement.
Psychological Review, 57, 217-228.
Cook, A. (1994). The Observational Foundations of Physics. Cambridge: Cambridge University Press.
Coombs, C. H. (1950). Psychological scaling without a unit of measurement. Psychological Review, 57,
145-158.
Coombs, C. H., Raiffa, H. & Thrall, R. M. (1954). Some views on mathematical models and
measurement theory. Psychological Review, 61, 132-144.
Crombie, A. C. (1994). Styles of Scientific Thinking in the European Tradition. London: Duckworth.
Cureton, E. E. (1946). Quantitative psychology as a rational science. Psychometrika, 11, 191-196.
Dawes Hicks, G. (1913). Are the intensity differences of sensation quantitative? IL British Journal of
Psychology, 6, 155-174.
 Quantitative science and psychology 379
Dedekind, R. (1872). Stetigkeit und Irrational^ahlen. Braunschweig: Vieweg.
Debreu, G. (1960). Topological methods in cardinal utility theory. In K. J. Arrow, S. Karlin & P.
Suppes (Eds), Mathematical Methods in the Social Sciences, 1959, pp. 16-26. Stanford, CA: Stanford
University Press.
Ellis, B. (1966). Basic Concepts of Measurement. Cambridge: Cambridge University Press.
Fechner, G. T. (1860). Elemente derpsychophysik. Leipzig: Breitkopf & Hartel. (English translation by
H. E. Adler, Elements of Psychophysics, vol. 1, D. H. Howes & E. G. Boring (Eds). New York: Holt,
Rinehart & Winston.)
Fechner, G. T. (1887). Uber die psychischen Massprincipien und das Weber'sche Gesetz. Philosophische
Studien, 4, 161-230. (English translation of pp. 178-198 by S. Scheerer (1987). My own viewpoint on
mental measurement. Psychological Research, 49, 213-219.)
Ferguson, A., Myers, C. S., Bartlett, R. J., Banister, H., Bartlett, F. C, Brown, W., Campbell, N. R.,
Craik, K. J. W., Drever, J., Guild, J., Houstoun, R. A., Irwin, J. O., Kaye, G. W. C, Philpott, S.
J. F., Richardson, L. F., Shaxby, J. H., Smith, T., Thouless, R. H. & Tucker, W. S. (1940). Final
report of the committee appointed to consider and report upon the possibility of quantitative
estimates of sensory events. Report of the British Association for the Advancement of Science, 2, 331-349.
Feyerabend, P. (1975). Against Method: Outline of an Anarchistic Theory of Knowledge. London: New Left
Books.
Fraser, C. O. (1980). Measurement in psychology. British Journal of Psychology, 71, 23-34.
Freyd, M. (1926). What is applied psychology? Psychological Review, 33, 308-314.
Gage, F. H. (1934^). An experimental investigation of the measurability of auditory sensation.
Proceedings of the Royal Society, Series B, Biological Sciences, 116, 103-122.
Gage, F. H. (1934^). An experimental investigation ofthe measurability of visual sensation. Proceedings
ofthe Royal Society, Series B, Biological Sciences, 116, 123-138.
Garner, W. R. (1954). Context effects and the validity of loudness scales. Journal of Experimental
Psychology, 48, 218-224.
Gigerenzer, G. & Strube, G. (1983). Are there limits to binaural additivity of loudness? Journal of
Experimental Psychology, Human Perception and Performance, 9, 126—130.
Green, B. F. (1954). Attitude measurement. In G. Lindzey (Ed.), Handbook of Social Psychology, vol. 1,
pp. 335-369. Reading, MA: Addison-Wesley.
Guilford, J. P. (1954). Psychometric Methods. New York: McGraw-Hill.
Gulliksen, H. (1946). Paired comparisons and the logic of measurement. Psychological Review, 53,
199-213.
Hacking, I. (1983). Representing and Intervening. Cambridge: Cambridge University Press.
Hardcastle, G. L. (1995). S. S. Stevens and the origins of operationism. Philosophy of Science, 62, 404-424.
Herbart, J. F. (1816). Lehrbuch z'"' Psychologie. (English translation by M. K. Smith, 1897, A Textbook
in Psychology. New York: Appleton.)
Holder, O. (1901). Die Axiome der Quantitat und die Lehre vom Mass. Berichte iiber die Verhandlungen
der Koniglich Sachsischen Gesellschaft der Wissenschaften•s^uLeipzig, Mathematisch-Physische Klasse, 53, 1-46.
Hornstein, G. A. (1988). Quantifying psychological phenomena: Debates, dilemmas, and implications.
In J. G. Morawski (Ed.), The Rise of Experimentation in American Psychology. New Haven, CT: Yale
University Press.
Hull, C. L. (1943). Principles of Behavior. New York: Appleton-Century-Crofts.
James, W. (1890). The Principles of Psychology, vol. 1. London: MacmiUan.
Jerrard, H. G. & McNeill, D. B. (1992). Dictionary of Scientific Units. London: Chapman & Hall.
Johnson, H. M. (1936). Pseudo-mathematics in the mental and social sciences. American Journal of
Psychology, 48, 342-351.
Kant, I. (1786). Metaphysical Foundations of Natural Science. (J. EUington transl., 1970.) Indianapolis, IN:
Bobbs-Merrill.
Krantz, D. H. (1972). A theory of magnitude estimation and cross-modaUty matching. Journal of
Mathematical Psychology, 9, 168-199.
Krantz, D. H., Luce, R. D., Suppes, P. & Tversky, A. (1971). Foundations of Measurement, vol. 1. New
York: Academic Press.
Kuhn, T. (1970). The Structure of Scientific Revolutions. Chicago, IL: University of Chicago Press.
380 Joel Michell
Levelt, W. J. M.,. Riemersma, J. B. & Bunt, A. A. (1972). Binaural additivity in loudness. British Journal
of Mathematical and Statistical Psychology, 25, 1-68.
Lord, F. M. & Novick, M. R. (1968). Statistical Theories of Mental Test Scores. Reading, MA: Addison-
Wesley.
Lorge, I. (1951). The fundamental nature of measurement. In F. Lindquist (Ed.), Educational
Measurement, pp. 533-559. Washington, DC: American Council of Education.
Luce, R. D. (1987). Measurement structures with Archimedean ordered translation groups. Order 4
165-189.
Luce, R. D. (1990). 'On the possible psychophysical laws' revisited: Remarks on cross-modality
matching. Psychological Review, 97, 6()-ll.
Luce, R. D., Krantz, D. H., Suppes, P. & Tversky, A. (1990). Foundations of Measurement, vol. 3. San
Diego, CA: Academic Press.
Luce, R. D. & Tukey, J. W. (1964). Simultaneous conjoint measurement: A new type of fundamental
mc^suKment. Journal of Mathematical Psychology, 1, 1-27.
McCormack, T. J. (1922). A critique of mental measurements. School and Society, 15, 686-692.
Massey, B. S. (1986). Measures in Science and Engineering: Their Expression, Relation and Interpretation
Chichester: Ellis Horwood.
Maxwell, J. C. (1891). A Treatise on Electricity and Magnetism. London: Constable.
Meier, S. T. (1994). The Chronic Crisis in Psychological Measurement and Assessment: A Historical Survey. S
Diego, CA: Academic Press.
Michell, J. (1986). Measurement scales and statistics: A clash of paradigms. Psychological Bulletin 100
398-407.
Michell, J. (1990). An Introduction to the Logic of Psychological Measurement. HiUsdale, NJ: Erlbaum.
Michell, J. (1993). The origins ofthe representational theory of measurement: Helmholtz, Holder, and
Russell. Studies in History and Philosophy of Science, 24, 185-206.
Michell, J. (1994). Numbers as quantitative relations and the traditional theory of measurement. British
Journal for the Philosophy of Science, 45, 389-406.
Michell, J. (in press). Bertrand Russell's 1897 critique ofthe traditional theory of measurement. Synthese.
Mundy, B. (1987). The metaphysics of quantity. Philosophical Studies, 51, 29-54.
Myers, C. S. (1913). Are the intensity differences of sensation quantitative? I. British Journal of Psychology
6, 137-154.
Nafe, J. P. (1942). Toward the quantification of psychology. Psychological Review, 49, 1-18.
Nagel, E. (1932). Measurement. Erkenntnis, 2, 313-333.
Narens, L. (1985). Abstract Measurement Theory. Cambridge, MA: MIT Press.
Narens, L. (1996). A theory of ratio magnitude estimation. Journal of Mathematical Psychology 40, 109-129.
Narens, L. & Luce, R. D. (1993). Further comments on the "nonrevolution" arising from axiomatic
measurement theory. Psychological Science, 4, 127-130.
Newman, E. B. (1974). On the origin of 'scales of measurement'. In H. R. Moskowitz et at. (Eds)
Sensation and Measurement, pp. 137-145. Dordrecht, Holland: Reidel.
Passmore, J. (1957). A Hundred Years of Philosophy. London: Duckworth.
Perloff, R. (1950). A note on Brower's 'the problem of quantification in psychological science'.
Psychological Review, 57, 188-192.
Reese, T. W. (1943). The application of the theory of physical measurement to the measurement of
psychological magnitudes, with three experimental examples. Psychological Monographs, 55, 1-89.
Rosnow, R. L. & Rosenthal, R. (1989). Statistical procedures and the justification of knowledge in
psychological science. American Psychologist, 44, 1276-1284.
Ross, S. (1964). Logical Foundations of Psychological Measurement. Copenhagen: Munksgaard.
Rozeboom, W. W. (1966). Scaling theory and the nature of measurement. Synthese, 16, 170-233.
Russell, B. (1897). On the relations of number and quantity. Mind, 6, 326-341.
Russell, B. (1903). Principles of Mathematics. Cambridge: Cambridge University Press.
Russell, B. (1920). Introduction to Mathematical Philosophy. New York: MacmiUan.
Sena, L. A. (1972). Units of Physical Quantities and their Dimensions. Moscow: Mir.
South, E. B. (1937). An Index of Periodical Literature on Testing, 1921-1936. New York: The
Psychological Corporation.
Spearman, C. (1904). General intelligence, objectively determined and measured. American Journal of
Psychology, 15, 201-293.
 Quantitative science and psychology 381

Spearman, C. (1937). Psychology down the Ages, vol. 1. London: MacmiUan.

Stevens, S. S. (1935 a). The operational definition of psychological terms. Psychological Review, 42,
517-527.
Stevens, S. S. (1935^). The operational basis of psychology. American Journal of Psychology, 47, 323-330.
Stevens, S. S. (1936a). Psychology: The propaedeutic science. Philosophy of Science, 3, 90-103.
Stevens, S. S. (1936^). A scale for the measurement of a psychological magnitude: Loudness.
Psychological Review, 43, 405-416.
Stevens, S. S. (1939). Psychology and the science of science. Psychological Bulletin, 36, 221-263.
Stevens, S. S. (1946). On the theory of scales of measurement. Science, 103, 667-680.
Stevens, S. S. (1951). Mathematics, measurement and psychophysics. In S. S. Stevens (Ed.), Handbook
of Experimental Psychology, pp. 1-49. New York: Wiley.
Stevens, S. S. (1956). The direct estimation of sensory magnitudes—loudness. American Journal of
Psychology, 69, 1-25.
Stevens, S. S. (1959). Measurement, psychophysics and utility. In C. W. Churchman & P. Ratoosh
(Eds), Measurement: Definitions and Theories, pp. 18-63. New York: Wiley.
Stevens, S. S. (1967). Measurement. In J. R. Newman (Ed.), The Harper Encyclopedia of Science, pp.
733-734. New York: Harper & Row.
Stevens, S. S. (1968). Measurement, statistics, and the schemapiric view. Science, 161, 849-856.
Stove, D. C. (1982). Popper and After: Four Modern Irrationalists. Oxford: Pergamon.
Suppes, P. (1951). A set of independent axioms for extensive quantities. Portugaliae Mathematica, 10,
163-172.
Suppes, P., Krantz, D. H., Luce, R. D. & Tversky, A. (1989). Foundations of Measurement, vol. 2. New
York: Academic Press.
Suppes, P. & Zinnes, J. (1963). Basic measurement theory. In R. D. Luce, R. R. Bush & E. Galanter
(Eds), Handbook of Mathematical Psychology, vol. 1, pp. 1-76. New York: Wiley.
Swoyer, C. (1987). The metaphysics of measurement. In J. Forge (Ed.), Measurement, Realism and
Objectivity, pp. 235-290. Dordrecht: Reidel.
Terman, L. (1916). The Measurement of Intelligence. Boston, MA: Houghton MifHin.
Terman, L. (1921). The status of applied psychology in the United States. Journal of Applied Psychology,
5, 1-4.
Terman, L. (1924). The mental test as a psychological method. Psychological Review, 31, 93-117.
Theocharis, T. & Psimopoulos, M. (1987). Where science has gone wrong. Nature, 329, 595-598.
Thomas, L. G. (1942). Mental tests as instruments of science. Psychological Monographs, 54, 1-87.
Thorndike, E. L. (1904). An Introduction to the Theory of Mental and Social Measurements. New York:
Science Press.
Thorndike, E. L. (1923). Measurement in education. In G. M. Whipple (Ed.), Twenty-first Year Book
of the National Society for the Study of Education, part 1, pp. 1-9. Bloomington, IN: Public School
Publishing.
Thorndike, R. L. (1982). Applied Psychometrics. Boston, MA: Houghton Mifflin.
Thurstone, L. L. (1938). Primary Mental Abilities. Chicago, IL: University of Chicago Press.
Titchener, E. B. (1905). Experimental Psychology: A Manual of Laboratory Practice, vol. II. London:
MacmiUan.
Viteles, M. S. (1921). Tests in industry. Journal of Applied Psychology, 5, 57-63.
von Kries, J. (1882). Uber die Messung intensiver Grossen und uber das sogenannte psychophysische
Gesetz. Vierteljahrsschrift fur wissenschaftliche Philosophie, 6, 257-294.
Watt, H. J. (1913). Are the intensity differences of sensation quantitative? III. British Journal of
Psychology, 6, 175-183.
Weitzenhoffer, A. M. (1951). Mathematical structures and psychological measurements. Psychometrika,
16, 387-406.
Whitney, H. (1968«). The mathematics of physical quantities, I: Mathematical models for measurement.
American Mathematical Monthly, 75, 115—138.
Whitney, H. (1968^). The mathematics of physical quantities, II: Quantity structures and dimensional
analysis. American Mathematical Monthly, 75, 227-256.
Zuckerman, M., Hodgins, H. S., Zuckerman, A. & Rosenthal, R. (1993). Contemporary issues in the
analysis of data: A survey of 551 psychologists. Psychological Science, 4, 49-53.
382 Joel Michell
Zwislocki, J. J. (1983). Group and individual relations between sensation magnitudes and their
numerical estimates. Perception and Psychophysics, 33, 460-468.

Received 16 November 1995; revised version received 21 March 1996

Appendix I
The survey of texts was done in the Psychology Library at the Katholieke Universiteit Leuven, Belgium
in June 1995. The books surveyed were:
Allen, M. J. & Yen, W. M. (1979). Introduction to Measurement Theory. Monterey, CA: Brooks/Cole.
Andreas, B. G. (1960). Experimental Psychology. New York: Wiley.
Black, J. A. & Champion, D. J. (1976). Methods and Issues in Social Research. New York: Wiley.
Borgatta, E. F. & Bohrnstedt, G. W. (1981). Levels of measurement once over again. In G. W.
Bohrnstedt & E. F. Borgatta (Eds), Social Measurement. Beverly Hills, CA: Sage.
Calfee, R. C. (1975). Human Experimental Psychology. New York: Holt, Rinehart & Winston.
Cliff, N. (1973). Psychometrics. In B. J. Wolman (Ed.), Handbook of General Psychology. Englewood Cliffs,
NJ: Prentice-Hall.
Coombs, C. H. (1953). Theory and methods of social measurement. In L. Festinger & D. Katz (Eds),
Research Methods in the Behavioral Sciences. New York: Holt, Rinehart & Winston.
Crano, W. D. & Brewer, M. B. (1973). Principles of Research in Social Psychology. New York: McGraw-
Hill.
Dominowski, R. L. (1980). Research Methods. Englewood Cliffs, NJ: Prentice-Hall.
Eagly, A. H. & Chaiken, S. (1993). The Psychology of Attitudes. Fort Worth, TX: Harcourt Brace
Jovanovich.
Eysenck, H. J., Wurzburg, W. A. & Berne, R. M. (1972). Encyclopedia of Psychology. London: Search
Press.
Francis, R. G. (1967). Scaling techniques. In J. T. Doby (Ed.), An Introduction to Social Research. New
York: Appleton-Century-Crofts.
Galtung, J. (l967). Theory and Methods of Social Research. London: George Allen & Unwin.
Green, B. F. (1954). Attitude measurement. In G. Lindzey (Ed.), Handbook of Social Psychology. Reading
MA: Addison-Wesley.
Guilford, J. P. (1956). Fundamental Statistics in Psychology and Education. New York: McGraw-Hill.
Haimson, B. R. & Elfenbein, M. H. (1985). Experimental Methods in Psychology. New York: McGraw-
Hill.
Harris, C. W. (1960). Encyclopedia of Educational Research. New York: MacmiUan.
Hays, W. L. (1967). Quantification in Psychology. Belmont: Brooks & Cole.
Hilgard, E. R. (1953). Introduction to Psychology. New York: Harcourt, Brace & World.
Kantowitz, B. H. & Roediger, H. L. (1978). Experimental Psychology. Chicago, IL: Rand McNally.
Kaplan, A. (1964). The Conduct of Inquiry. San Francisco, CA: Chandler.
Kerlinger, F. N. (1966). Foundations of Behavioral Research. New York: Holt, Rinehart & Winston.
Kurtz, K. H. (1965). Foundations of Psychological Research. Boston, MA: AUyn & Bacon.
Lemke, E. & Wiersma, W. (1976). Principles of Psychological Measurement. Chicago, IL: Rand McNally.
Lemon, N. (1973). Attitudes and their Measurement. London: Batsford.
Lord, F. M. & Novick, M. R. (1968). Statistical Theories of Mental Test Scores. Reading, MA: Addison-
Wesley.
Manning, S. A. & Rosenstock, E. H. (1968). Classical Psychophysics and Scaling. New York: McGraw-
HiU.
Marlowe, L. (1971). Social Psychology. Boston, MA: Holbrook Press.
Meyers, L. S. & Grossen, N. E. (1974). Behavioral Research. San Francisco, CA: Freeman.
Miller, G. A. (1962). Psychology: The Science of Mental Life. New York: Harper & Row.
Morgan, C. T. & King, R. A. (1966). Introduction to Psychology. New York: McGraw-Hill.
Newcomb, T. M., Turner, R. F. & Converse, P. E. (1965). Social Psychology. New York: Holt, Rinehart
& Winston.
Nunnally, J. C. (1967). Psychometric Theory. New York: McGraw-Hill.
 Quantitative science and psychology 383
Peck, D. F. & Shapiro, C. M. (1990). Measuring Human Problems. New York: Wiley.
Phillips, B. S. (1971). Social Research. New York: MacmiUan.
Plutchik, R. (1968). Foundations of Experimental Research. New York: Harper & Row.
Sartain, A. Q., North, A. J., Strange, J. R. & Chapman, H. M. (1973). Psychology: Understanding Human
Behavior. New York: McGraw-Hill.
Selltiz, C , Jahoda, M., Deutsch, M. & Cook, S. W. (1959). Research Methods in Social Relations. New
York: Holt, Rinehart & Winston.
Shaw, M. E. & Wright, J. M. (1967). Scales for the Measurement of Attitudes. New York: McGraw-Hill.
Siegel, S. (1956). Nonparametric Statistics for the Behavioral Sciences. New York: McGraw-Hill.
Simon, J. L. (1969). Basic Research Methods in Social Science. New York: Random House.
Sjoberg, G. & Nett, R. (1968). A Methodology for Social Research. New York: Harper & Row.
Sommer, B. & Sommer, R. (1991). A Practical Guide to Behavioral Research. New York: Oxford
University Press.
Wiersma, W. & Jurs, S. G. (1985). Educational Measurement and Testing. Boston, M A : AUyn & Bacon.

Appendix II
Put as succinctly as possible, measurement is the numerical estimation of the ratio of a magnitude
of a quantitative attribute to a unit of the same attribute.
The concept of measurement is embedded within a matrix of closely related concepts. There is no
adequate definition of it that avoids implicating them, as well. The associated concepts used here are
briefly explained as follows.
Attribute. An attribute is a range of properties or relations that may vary from instance to instance. E.g.
length is an attribute of objects, different objects often having different lengths; so is sex, some creatures
being female, others male; nationality is another example, some people being British, some French, etc.
Quantitative attribute. A quantitative attribute (or quantity) is an attribute the instances of which are
related to one another both ordinally and additively. One version of (continuous) quantitative structure
is given by Holder's (1901) axioms, another is given in section 1.1 of this paper. Not aU attributes are
quantitative. E.g. length is quantitative, but neither sex nor nationality is.
Magnitude. A magnitude is a specific instance of a quantity. Thus, each instance of length (say, the
length of this page) is a magnitude of the quantity, length.
Ratio. As intended in this context, a ratio is a special kind of relation holding between magnitudes of
the same quantity. The ratio of one magnitude of a quantity to another is the size of the first relative
to the second. Thus, ratios are relative magnitudes. The most useful way to express a ratio is as a
number. E.g. the ratio of the length of a cricket pitch to one yard is 22.
Units. A unit is a more or less arbitrarily selected magnitude of a quantity, singled out to be the instance
against which any other is to be compared. If the unit is known, then expressing the magnitude of any
other instance of the quantity relative to the unit means that the magnitude of that instance is defined
and, so, known via the unit. E.g. one widely used unit of length is the metre.
Numerical estimation. In the first instance, a numerical estimation or number specifies how many? or how
much?. E.g. knowing that there is some whole number, say, three, books on the desk is knowing that
there is a book, another and one more at that location. The range of possible answers to the question,
how many?, is the sequence of whole (or natural) numbers, the meaning of each being definable via the
concepts of one and one more than. Using the natural numbers, the real numbers may be specified. They
answer the question, how much?. It foUows from the axioms for eontinuous quantity that some multiple
(say, «) of an instance of some quantity (call it a) is always less than, equal to, or greater than any other
multiple {m) of another instance of that quantity, b. If less than, then the ratio oi a to b is less than the
ratio oi m to n (i.e. a/b < m/n); if equal to, then a/b = m/n; and if greater than, then a/b > m/n. Thus,
each ratio of magnitudes has a location relative to each ratio of natural numbers and, so, is given by
a real number (i.e. the real number specified by the class of all ratios of natural numbers equal to or less
than it). Thus, the ratio of each instance of a quantity to the unit selected is expressed by a unique
numerical value. In general, measurement procedures only permit the (approximate) estimation of such
unique numerical values.
British journal of Psychology (1997), 88, 385-387 Printed in Great Britain 385
© 1997 The British Psychological Society

Commentary on Michell, Quantitative Science

and the definition of measurement in
psychology

Paul Kline*
Department of Psychology, University of Exeter, Washington Singer Laboratories,
Exeter EX4 4QG, UK

Michell has developed some powerful arguments which correctly cast doubt on the scientific validity
of much ofthe measurement procedures in psychology .Jn this reply I shall restrict myself to answering
t;he objections relevant to psychometrics. The attack on psychophyslcal measurement I shall leave others
to rebut.
There are two key points in the paper which are entirely accepted: that scientific measurement may
be defined as the estimation or discovery of the ratio of some magnitude of quantitative attribute to a^
unit of the same attribute; and that quantitative science involves two tasks—the investigation of the
hypothesis that the relevant attribute is quantitative and the instrumental task of devising procedures
to measure magnitudes of the attribute shown to be quantitative. The burden of the paper is that in
psychometrics neither of these two tasks has been done. It is simply assumed that the variables are
quantitative and the measurement techniques fail by the criteria of scientific measurement. An example
will clarify his arguments. It is assumed, in the case of extraversion, despite the work of Jung, that this
is a quantitative variable, without prior proof, and the tests contain no clear unit of measurement, in
comparison with, to cite Michell, a cricket pitch where the ratio of the pitch to the unit is 22 (if yards
are still permissible). Finally, Michell claims, where measures depart from the scientific model, then
there is no justification for using them in the kinds of mathematical arguments which have played so
large a part in developing the natural sciences. All this leads Michell to conclude that psychometrics is
not scientific either in its theory or application.
While accepting these premises of scientific measurement, I shall argue that psychometrics is not as
entirely flawed as Michell claims although some aspects of the field should be abandoned or modified
considerably.
i^irstl shall discuss the psychometrics of intelligence, as typified by the factor analytic work of
Spearman, Burt and Cattell (Cattell, 1971). There are three points which need to be made. First the
original question, which the factor analysis of abilities was designed to answer, was essentially this: what
accounts for the positive correlations between human abilities? It turns out that two g factors,
crystallized and fluid intelligence, account for much of the covariance. The intelligence test score,
measuring these factors, may be regarded as a good index of the level of difficulty which an individual
has reached in cognitive problem solving, ^econdly, measures of these factors have construct validity,
that is they behave as one might expect of intelligence tests. They predict academic success, occupational
success and differentiate between groups, as is well documented. All this means that instead ofthe vague
term intelligence psychometrists can use these tests o( g, despite their admitted defects as scientific
measures, as operational measurements of intelligence which lead to useful predictions and theorizing.
Sternberg (1982) summarizes much ofthis work.
The third point and by far the most germane to this argument is not mentioned by Michell. This
concerns the fact that the leading psychometrists, such as Cattell and Eysenck, figures whom Cattell is

* Requests for reprints.

386 P. Kline
keen to separate from mere itemetric moles, regard factors only as starting points for their
investigations. Before the recent development of confirmatory analysis, factor analysis was a deliberately
multivariate exploratory technique, to indicate the variables to study. In the field of intelligence great
efforts are being made to investigate the underlying nature of the ^ factors. Intelligence tests are not the
end but the means of investigation. Thus as Barrett (1996) has discussed there is an emphasis on process
rather than test refinement. This includes nerve conduction variability and velocity (Deary & Caryl,
1993) and Weiss (1995), for example, invokes gene biochemistry, and quantum statistics to account for
EEG and psychometric findings. Lehrl & Fischer (1990) conceptualize intelligence as information-
processing capacity and have developed a measure of this, the BIP, the basic period of information
processing, which fits the scientific criteria mentioned in this article. This is derived from the speed of
reading letters, but investigations by the present author (Draycott & Kline, 1994) have shown it to be
related to crystallized rather than fluid intelligence.
Thus in the field of intelligence I would argue that psychometric testing is leading to a real scientific
understanding of the phenomenon and that, in time, we shall look upon intelligence tests much as the
modern astronomer at Jodrell Bank looks at Galileo's telescope.
As regards personality tests there is more force in the arguments of Michell. The factor analysis of
personality questionnaires has yielded, to the satisfaction of most investigators, two clear factors,
extraversion and anxiety, and three others over which there is still some argument, tough-mindedness,
conscientiousness and openness. As with intelligence these factors correlate but only moderately with
a variety of external criteria and are widely used in occupational psychology (see Kline, 1993).
Immediately a severe problem arises as to what is the unit of measurement. A score on these tests
consists ofthe number of items endorsed by an individual but these are clearly not units of measurement
in any meaningful sense. At best each item can be considered to be a sample from the universe of items
measuring that variable (the true score). Thus the more items in the sample of items which are endorsed,
the higher the fallible score and, by inference, the true score. This is the classical psychometric model.
Of course the universe of items is notional and there is no method of ensuring that the items sample
the universe although high reliability ensures that the items are consistent and sample some universe.
Again the leading workers in the psychometrics of personality, especially Cattell (1981) and Eysenck
(1967) regard these tests simply as starting points for the study of personality. By the experimental study
of these factors the nature of extraversion and anxiety and their physiological bases are being explicated
as Barrett (1996) has summarized. Gray (1981) and Zuckerman (1991) have contributed further to the
investigation of the processes underlying these factors. It should also be pointed out that biometric
investigation of these personality factors, as is the case with intelligence, has demonstrated a
considerable genetic component in their variance (Eysenck, 1994), a finding which suggests that despite
their imperfections as scientific measures, they are far from worthless.
However, it is in the field of applied and social psychology, for example, the construction of
questionnaires to measure variables which are of interest to researchers in health and education that the
strictures of Michell bite hard and, in my view, render the work of little scientific value. As I have
argued previously (Kline, 1993) locus of control exemplifies these problems. Here items which have face
validity, e.g. 'When I get sick, I am to blame' and 'No matter what I do, I am likely to get sick', are
factored and items loading a particular factor are regarded as scales named from the high-loading items.
With such a scale the unit of measurement is unknown. Often with only six items per scale it is difficult
to see what universe of items they might purport to represent. That they factor together indicates
nothing more than that they mean the same thing. This type of blind factoring is bound to yield factors
if enough items which are essentially paraphrases of each other are included in a test. With this
methodology, there is literally no end to factors which can be produced.
These scales, and there are many such, are then used as variables in further factor analytic studies with
other variables thus derived. This kind of psychometrics in which the scales are the variables, simply
because their items load a factor, does seem to be measurement gone mad as described by Michell—a
systematic thought disorder.
,To conclude, it is true that psychometric measurement is not scientific in the sense defined by Michell.
On the other hand, the true score model has yielded measures which are certainly better than no
quantification at all and psychological variables are hard to fit to the strictly scientific measurement
model.^It is only when psychometric scores are regarded as end-products rather than as guides for
scientific investigation that the full force of Michell's arguments obtains.
 Commentary on Michell 387

References
Barrett, B. (1996). Process models in individual differences research. In C. Cooper & V. Varma (Eds),
Processes in Individual Differences. London: Routledge.
Cattell, R. B. (1971). Abilities, their Structure, Growth and Action. New York: Houghton Mifflin.
Cattell, R. B. (1981). Personality and Learning Theory. New York: Springer.
Deary, I. & Caryl, P. (1993). Intelligence, EEG and evoked potentials. In P. A. Vernon (Ed.), 'Biological
Approaches to Human Intelligence. Norwood, NJ : Ablex.
Draycott, S. & Kline, P. (1994). Further investigations into the nature of BIP: A factor analysis ofthe
BIP with primary abilities. Personality and Individual Differences, 17, 201—209.
Eysenck, H. J. (1967). The Biological hasis of Personality. Springfield, IL: Thomas.
Eysenck, H. J. (1994). PersonaUty and intelligence: Psychometric and experimental approaches. In
R. J. Sternberg & P. Ruzgis (Eds), Personality and Intelligence. Cambridge: Cambridge University Press.
Gray, J. A. (1981). A critique of Eysenck's theory of personality. In H. J. Eysenck (Ed.), A Model for
Personality. Berlin: Springer-Verlag.
Kline, P. (1993). Handbook of Psychological Testing. London: Routledge.
Lehrl, S. & Fischer, P. (1990). A basic information psychological parameter (BIP) for the reconstruction
of the concepts of intelligence. European Journal of Personality, 4, 259-286.
Sternberg, R. J. (Ed.) (1982). Handbook of Human Intelligence. Cambridge: Cambridge University Press.
Weiss, V. (1995). Memory span as the quantum of action of thought. Cahiers de Psychologie Cognitive, 14,
387-408.
Zuckerman, M. (1991). The Psychobiology of Personality. Cambridge: Cambridge University Press.
British Journal of Psychology (1997), 88, 389-391 Printed in Great Britain 389
© 1997 The British Psychological Society

A critique of a measurement-theoretic
critique: Commentary on Michell,
Quantitative science and the definition of
measurement in psychology

Donald Laming*
Department of Experimental Psychology, University of Cambridge, Downing Street,
Cambridge CB2 3EB, UK

Over the past 40 years theire has been much theoretical progress in the understanding of what it means
to make measurements. If numbers are assigned to objects or events, the kinds of arithmetical operations
(such as averaging or calculating ratios) which it is thereafter meaningful to carry out on the numbers
depend on the rule of assignment. Measurement theory, roughly speaking, is concerned to identify what
conditions need to be satisfied to make this or that arithmetical operation meaningful. Measurement
theorists, generally, feel that psychologists have disregarded their work, to the detriment of the
development of psychology as a natural science. Michell's article is a polemic—a very scholarly and
well-argued polemic—addressing this issue. It would have helped his argument, however, to have
explained why measurement theory should matter to psychologists, and I endeavour, first of all, to
remedy that deficit.

Intelligence test data

There is an intelligence test, X. If respondent A answers 40 questions correctly, whereas B answers only
20, does that mean that A is twice as intelligent as B? Psychologists do not make that mistake because
there is another test Y (Y is test X with 20 very easy questions added) on which A answers 60 questions
correctly and B 40. And on test Z (Z is test X with its 10 easiest questions removed) A answers 30
questions correctly and B only 10. It would manifestly be arbitrary to say that A is twice as intelligent
as B. For a similar reason psychologists do not say that the difference in intelligence between A and C
(C answers 30 questions correctly on test X) is the same as that between C and B. But they do make a
very similar error.
It is conventional to normalize intelligence test scores by so transforming the scores of a very large
sample of respondents that the transformed score is normally distributed with a mean of 100 and a
standard deviation of 15. The score (IQ) of any subsequent respondent is expressed as a normal deviate
on that scale. It is thereafter common to regard the difference between an IQ of 100 and of 110 as equal
to the difference between 110 and 120. But that is to say that the difference in intelligence between the
50th and the 75th percentiles in the normalizing population is the same as that between the 75th and
the 91st, and there is no basis in the test scores for any such assertion. Psychologists do not draw
inferences about intelligence (or any other matter) which are obviously arbitrary—obviously arbitrary
because the test can be rigged to give a different result. But there are other errors, not obvious but
nonetheless arbitrary. Such errors frequently derive from some past assumption (in the present example,
that intelligence is normally distributed in the population) which is now no longer recognized for what
it is. Measurement theory is needed to protect psychologists against those other errors.

* Requests for reprints.

390 D. Laming
To develop the present example, different mental tests correlate less than perfectly, and the practice
(factor analysis) has developed of calculating linear combinations of the normalized percentile {rank
order) scores, in order to seek some simpler description ofthe pattern of correlation. That is a nonsense.
The joint test scores are analysed as if they were multivariate normal, systematizing exactly the error
described above. That multivariate normal distribution of normalized test scores is pure assumption.
Measurement theory has an important role to play in exposing deceits of that kind.
Even worse is the long-continued controversy concerning the structure of mental abilities, most
notably between Spearman's (1927) idea of general and specific abiUties and Thurstone's (1938) idea of
primary mental abilities, with Burt (1940), Guilford (1967), Thomson (1939), and Vernon (1950)
contributing yet other suggestions. Mental test data do not speak at all to this controversy. Generations
of psychology students have been scratching their heads over an issue which is entirely artificial. While
there have been many improvements in the precision and reliability of mental tests, the contribution
from test data over the past 90 years (from, say Spearman, 1904, to Sternberg, 1994) to our
understanding of the nature of intelligence has been nil! Measurement theory, had it been far enough
developed, could have said as much in 1904. But 'intelligence' is still what intelligence tests measure!
There is an important function for measurement theory to fulfil as a critique of contemporary work in
psychology.

Category judgment
MicheU, however, does his cause a disservice by placing as much as half of his emphasis on the 'proper'
use ofthe word 'measurement'. For Michell, 'measurement' only means ratio-scale measurement ofthe
kind met with in the physical sciences. This will seem quite beside the point to most psychologists
because so many psychological experiments yield merely rank order or categorical data. Would Michell
have us throw all those data away?
Twelve years ago I investigated the thesis that human judgment is little better than ordinal (Laming,
1984). That idea made sense of a variety of results from magnitude estimation and category judgment
experiments—the Umited transmission of information in category judgments (Garner, 1962, chapter 3);
the interaction of the order of stimulus presentation with resolving power (Luce, Nosofsky, Green &
Smith, 1982); the correlation between successive magnitude estimates (Baird, Green & Luce, 1980), and
other results besides—results which otherwise appeared quite baffling to theorists in that field. Much
psychological data comes from respondents who make judgments about stimuli, and the possibility must
seriously be considered that the analysis of categorical and rank order data is essential to psychology.
That does not mean that psychology cannot be scientific; nor that the treatment of such data is arbitrary.
It means that measurement theory must address the question how to treat that kind of data.
Michell does indeed take S. S. Stevens to task; and while he pays lip-service to the overriding role
of scientific theory (pp. 358-359)—real theory, not just measurement theory—seems not then to know
what to do with it. So far as sensory scales are concerned: we can know about internal sensations only
through the medium of participants' judgments of stimuU. If human judgment is no better than ordinal,
there can never be any empirical basis for establishing a subjective scale distinct from the natural
physical measure of the stimulus (Laming, 1977). Any theory of magnitude estimation or category
judgment will have to be formulated in those terms.

Fechner's law
One of the favourite targets of measurement theorists is Fechner's scheme for measuring sensation by
cumulating just noticeable differences, and Michell is no exception. The criticism originated with Luce
& Edwards (1958). The gist of their argument is that if the empirical data consist of no more than the
marking off of successive just noticeable differences along a continuum, many more schemes of
numerical assignment are possible besides Fechner's law. Fechner's procedure of replacing a finite
difference (the jnd) with a differential is unjustified and, except in the special case of Weber's law, actually
gives the wrong answer. But Michell's comments about Fechner's scheme will again seem beside the
point.
That point is that any scheme of measurement necessarily presupposes a theory relating the actual
observations to that which is to be measured (the scientific theory which Michell acknowledges on
 Measurement-theoretic critique 391
p. 359). Since Fechner's day our understanding of sensory discrimination has developed beyond all
recognition, the crucial step in that development being the introduction of signal-detection theory by
Tanner & Swets (1954). Treating the signal-detection model purely as a vehicle for the analysis of data,
categorical responses in a two-alternative discrimination task can be used to estimate a continuous
index, d', which has all the properties of an interval measure. Given a discrimination between two
separate stimuli, d' increases in proportion to the stimulus difference (Laming, 1986, chapter 4; also
Mountcastle, Talbot, Sakata & Hyvarinen, 1969), so that the condition of additivity (Michell, p. 363) is
empirically satisfied. Luce & Edwards' objections to Fechner's passage to a differential do not apply to
d'. Experimental psychologists have substantially solved the problem which Michell addresses and he
writes in apparent ignorance of that solution. He is, as it were, directing the critical power of modern
measurement theory against the psychology of 1860. It will not do.
That is a pity because otherwise Michell has a powerfully written argument that has much to
contribute to present-day psychology. Measurement-theoretic critique—specifically the weeding out of
those artificial questions to which the data do not speak—is much needed in psychology today.

References
Baird, J. C, Green, D. M. & Luce, R. D. (1980). VariabiUty and sequential effects in cross-modaUty
matching of area and loudness. Journal of Experimental Psychology: Human Perception and Performance,
6, 277-289.
Burt, C. L. (1940). The Factors of the Mind. London: University of London Press.
Garner, W. R. (1962). Uncertainty and Structure as Psychological Concepts. New York: Wiley.
Guilford, J. P. (1967). The Nature of Human Intelligence. New York: McGraw-HiU.
Laming, D. (1984). The relativity of'absolute' judgements. British Journal of Mathematical and Statistical
Psychology, 37, 152-183.
Laming, D. (1986). Sensory Analysis. London: Academic Press.
Laming, D. (1997). The Measurement of Sensation. Oxford: Oxford University Press.
Luce, R. D. & Edwards, W. (1958). The derivation of subjective scales from just noticeable differences.
Psychological Review, 65, 222-237.
Luce, R. D., Nosofsky, R. M., Green, D. M. & Smith, A. F. (1982). The bow and sequential effects in
absolute identification. Perception <& Psychophysics, 32, 397-408.
Mountcastle, V. B., Talbot, W. H., Sakata, H. & Hyvarinen, J. (1969). Cortical neuronal mechanisms
in flutter-vibration studied in unanesthetized monkeys. Neuronal periodicity and frequency
discrimination. Journal of Neurophysiotogy, 32, 452-484.
Spearman, C. (1904). "General intelligence", objectively determined and measured. American Journal of
Psychology, 15, 201-293.
Spearman, C. (1927). The Abilities of Man. London: MacmiUan.
Sternberg, R. J. (1994). Intelligence and cognitive styles. In A. M. Colman (Ed.), Companion Encyclopedia
of Psychology, vol. 1, pp. 583-601. London: Routledge.
Tanner, W. P. Jr & Swets, J. A. (1954). A decision-making theory of visual detection. Psychological
Review, 61, 401-409.
Thomson, G. H. (1939). The Factorial Analysis of Human Ability. London: University of London Press.
Thurstone, L. L. (1938). Primary Mental Abilities. Chicago: University of Chicago Press.
Vernon, P. E. (1950). The Structure of Human Abilities. London: Methuen.
 British Journal of Psychology (1997), 88, 393-394 Printed in Great Britain 393
© 1997 The British Psychological Society

Commentary on Michell, Quantitative science

and the definition of measurement in
psychology

A. D. Lovie*
Department of Psychology, University of Liverpool, Eleanor Rathbone Building,
PO Box 147, Liverpool L69 3BX, UK

\ completely reject the hard-nosed (and very outdated) positivist and empiricist/realist line adopted in
c the paper. It is as if the upheavals of the last 30 or more years in the philosophy, history and sociology
of science had not happened, and that the powerful new insights provided by this work amounted to
no more than a kind of crass anti-scientism (see the usual reaUst knee-jerk reactions to Kuhn and
Feyerabend at the end of the paper). By adopting such a reactionary approach to the history and
philosophy of science, the, author has reduced what could have been an insightful account of a crucial
and pivotal episode in i:heTiistory of psychological statistics into a stale and superficial morality tale, with
S. S. Stevens and others cast as the villains, and Holder and his modern followers as the heroes.
Although such a knockabout tale might please the groundlings, the serious, historical and conceptually
embedded business as to just what Stevens actually did in the late 1930s and early 1940s (and why he
did it) are all left essentially untouched because the author cannot bring himself to look at this episode
in its historical and intellectual context. Instead,,^what we get is a non-reflexive tour deforce on the sins
of the Popes written by one who is in possession of the final truth (note the strategic use of the words
'reality' 'truth' and 'true measurement' throughout the paper). But there are no absolute, ahistorical
mathematical truths or methods, only locally developed and locally maintained collective commitments
and practices: what the ethnomethodologist, Eric Livingston, has termed the 'Uved-work' of the
practising mathematician (Livingston, 1986). Starting from this general position (as most historians of
science now do) would have generated a very different kind of story from the one before us.
I am also both intrigued and unhappy about the rhetorical deployment ofthe so-called 'ideological
support system' which is used to account for Stevens's (and others') inability to see the obvious.
Intrigued because suddenly the author seems happy to embrace an explanatory device which smacks a
Uttle of the social constructivist approach that I have advocated earUer, only to notice unhappily that
this notion is never appUed to Holder et al. whose ideas are emphatically not generated by such a social
constructivist process: blatant ontological gerrymandering of this kind cannot go unchallenged. Would
it not be ultimately more insightful to argue that everybody works within such a support system (the
author and myself included!)?
Let me add something more about this matter of conclusions or points which are described as
'obvious' but which, in some mysterious way, people like Stevens were too blinkered or perhaps too
wilful to see. Nothing is obvious ^er se: it is rather a matter of which interests you are committed to
and which assumptions and presuppositions you buy. If you do not buy into the current author's (or
Holder's or whoever's) position and all that that entails, then such conclusions are not necessarily
obvious. So the term ' obvious' is used here as a contingently phrased rhetorical device to undermine the
opposition, as is the linked term 'methodological thought disorder'. Hence the crudely contingent
argument used to account for psychology's perverse and continuing attachment to measurement a la
Stevens: viz. if my conclusion (that is, that Stevens's approach to measurement is unscientific) is not

* Requests for reprints.

 394 A. D. Lovie
'obvious' to you, then you must be suffering from a 'methodological thought disorder' grounded on
an essentially irrational 'ideological support system'.
Overall, I regard the paper as representing one further twist in a very long running piece of positivist
argumentation, one whose case should have been strong enough within its own particular domain of
'discourse not to require this authoritarian and regressive recasting of the historical record.

Reference
Livingston, E. (1986). The Ethnomethodological Foundations of Mathematics. London: Routledge & Kegan
Paul.
British Joumat of Psychology (1997), 88, 395-398 Printed in Great Britain 395
© 1997 The British Psychological Society

Quantification and symmetry: Commentary

on Michell, Quantitative science and the
definition of measurement in psychology

R. Duncan Luce*
Institute for Mathematical Behavioral Sciences, Social Science Plai^a, University of California at Irvine,
Irvine, CA 92697-5100, USA

Several of Michell's points are amplified and emphasized and the following
additional point is made. Most quantitative attributes can be measured in more than
one way, and there are interesting questions about how they relate. Among other
things, units of measurement and symmetries of the underlying structure may or
may not agree.

Because I agree with almost everything Michell says, my commentary is restricted to some amplification
and to an added observation.

1. Quantification and scale type

The British Association for the Advancement of Science Subcommittee on Measurement claimed
quantification to be possible only when there is an empirical operation satisfying Holder's (1901)
conditions (see Michell's article). In this they were wrong, although at the time there were no specific
examples to prove it. In response S. S. Stevens (1946,1951,1975) claimed that what counted was having
an interval or ratio scale type. Subsequent research has given meaning to this assertion (see §2), but
given his attempts to invoke scale type ideas it is doubtful if he understood it himself. The issue hinges
on his meaning of a 'rule' in his famous definition of measurement. Stevens (1975, pp. 46-47) said:
'Measurement is the assignment of numbers to objects or events according to rule (Stevens, 1946). The
rule of assignment can be any consistent rule. The only rule not allowed would be random assignment,
for randomness amounts in effect to a non-rule'. So, to him, the rule had to do with the person
performing a measurement, not with an empirical law involving the attribute being measured. In
particular, he must have viewed IQ measurement as forming an interval scale because the scientists
involved rescale the counts of questions answered so that IQ is normally distributed. Nothing empirical
forces this choice. No measurement theorist I know accepts Stevens' broad definition of measurement.
In our view, the British committee got it wrong by being far too narrow and Stevens got it wrong by
being far too broad, extending the concept of measurement much beyond the empirical. In our view,
the only sensible meaning for 'rule' is empirically testable laws about the attribute.
This aspect of Stevens' views makes it very difficult to understand what is involved in, for example,
his method of magnitude estimation. Is it more than ordinal? He confounded two meanings ofthe word
' ratio' in discussing it. He instructed his participants to preserve subjective ratios and he assumed that
the resulting numerals formed a ratio scale, but he did not make at all clear what empirically justified

* Requests for reprints.

396 R. Duncan Luce
that belief. Recently, Narens (1996) has worked out an interpretation of what he might have meant,
including a plausible empirical law, which if true indeed justifies believing this is measurement at the
level of ratio scales. The subtle argument is, however, very different from any that Stevens gave.

2. Quantification of symmetry
One of the most significant discoveries of contemporary representational measurement theory is that a
very broad class of structures can be quantified indirectly (Alper, 1987; Cohen & Narens, 1979; Luce
& Narens, 1985; Narens, 1981a, b). An isomorphism of an ordered empirical structure with itself is
called a symmetry (or automorphism) of the structure. A symmetry with no fixed point is called a
translation. All familiar examples of measurement structures are rich in translations—they are homogeneous
in the sense that any point can be mapped into any other point by a translation. So the elements of the
structure are structurally identical. What is remarkable is that if the structure forms a continuum, then
the set of translations plus the identity forms a group under function composition, has an order induced
from the structural order, and together these form a quantitative system in the classical sense of Holder.
Moreover, the group of translations can be mapped onto the multiplicative positive real numbers. And
in the homogeneous case the structure itself can be mapped isomorphically into the translations and so
into a numerical representation in which the translations form a multiplicative ratio scale. This enlarges
enormously the potential domain of measurable structures.
For example, consider the class of structures involving an ordering and a binary operation (e.g.
receiving two things at once, placing two objects on a pan of a pan balance, etc.). For such structures
that are homogeneous and continuous, the possible numerical representing operations ® are of the
form
x®y=yf{xly), (1)
where/is strictly increasing and/(:^)/;j is strictly decreasing. This forms a ratio representation because
for /fe > 0,
kx@ky = kyf{ky/kx) = k{x ®y). ' (2)
The classical case of measurement in physics is the special case where y(;j) = 1-1-^.
This development opens many possibilities which simply have not been explored by empirical
scientists. It enlarges much beyond additive systems ones that, in a principled way, can be viewed as
quantifiable. Of course, it does not extend to the non-empirical generalizations of Stevens.

3. Units and symmetries

ImpUcitly Michell seems to assume that an attribute either is or is not quantifiable and he does not
discuss the possibility that it might be quantifiable in several ways. In that case, which in fact is common,
the question can be raised in what sense do the quantifications agree.
For example, in classical physics many attributes are quantifiable both extensively and conjointly. A
case in point is mass where the conjoint structure involves manipulating mass by varying the volumes
and substances forming the masses. For this and the usual extensive measure to agree, as they do in
classical physics, a certain type of interlocking distribution law must be satisfied, as indeed it is. The
details of this, which are somewhat complex, can be found in Krantz, Luce, Suppes & Tversky (1971)
and Luce, Krantz, Suppes & Tversky (1990). Also, as we saw in the previous section, the extensive ones
are quantifiable both in terms ofthe structural operation and the group of translations. These too agree.
This means that changes of units in the representation are the same as the translations of the
representation.
Relativistic velocity is a similar but interestingly different kind of example. Constant velocities can be
concatenated—a person walking in a train that is moving relative to an observer. The velocity of the
walker relative to the observer is the concatenation of the walker's velocity relative to the train and the
train's velocity relative to the observer. Such a velocity—concatenation structure is extensive and so has
additive representations, called rapidity; in that measure the speed of Ught maps to oo. Of course,
velocities form a conjoint structure with elapsed distance, j-, and time, t, so that v = s/t. This velocity
 Quantification and symmetry 397
measure is not rapidity but rather a transform of it which has the well-known representation of
concatenation

1 -I- uv/c'
where c denotes the velocity of light in the same units. Note that both the rapidity and v are invariant
under multiplication by positive constants. The former corresponds to the translations ofthe extensive
structure whereas the latter does not; it is just a change of units. Such changes of units, however, do
eorrespond to translations of the two components of the conjoint structure. It is simply the case that
the two groups of translations are not the same.
A somewhat similar example has arisen recently in utility theory. For half a century utility has been
studied using uncertain alternatives (gambles) for which the representation is a weighted average of
numerical utilities over pure consequences—the famous subjective expected utility model and its
modern generalizations. Luce & Fishburn (1991, 1995) introduced a second way to measure utility
based on a binary operation ® of joint receipt, i.e. receiving two valued objects at the same time. They
assumed for gains relative to a status quo that it forms an extensive structure. Assuming an interlock
called segregation, which is highly rational and has been sustained empirically (Chd & Luce, 1995; Cho,
Luce & von Winterfeldt, 1994), they show that the usual utility function U derived from gambling
behaviour has the following non-additive representation of ® :
7/.A

(4)

where C is the asymptotic maximum of the utility function. Note that this representation is invariant
under change of unit, but that Uke the velocity case these transformations do not correspond to
translations of the extensive structure although they do of the gambling structure.
I believe that these two cases suggest that if we examine sensory intensity measurement from both
the perspective of conjoint structures and extensive ones based on the physical operation of
concatenation, we may find an interlock leading to somewhat similar non-additive, bounded
representations.

References
Alper, T. M. (1987). A classification of all order-preserving homeomorphism groups ofthe reals that
satisfy finite uniqueness. Journal of Mathematical Psychology, 31, 135-154.
Cho, Y. & Luce, R. D. (1995). Tests of hypotheses about certainty equivalents and joint receipt of
gambles. Organisational Behavior and Human Decision Processes, 64, 229-248.
Cho, Y., Luce, R. D. & von Winterfeldt, D. (1994). Tests of assumptions about the joint receipt of
gambles in rank- and sign-dependent utility theory. Journal of Experimental Psychology: Human
Perception and Performance, 20, 931-943.
Cohen, M. & Narens, L. (1979). Fundamental unit structures: A theory of ratio scalability. Journal of
Mathematical Psychology, 20, 193-232.
Holder, O. (1901). Die Axiome der Quantitat und die Lehre vom Mass. Berichte iiber die Verhandlungen
der Koniglich Sachsischen Gesellschaft der Wissenschaften ^u Leipzig, Mathematisch-Physische Klasse, 53, 1-
Krantz, D. H., Luce, R. D., Suppes, P. & Tversky, A. (1971). Foundations of Measurement, vol. 1. San
Diego: Academic Press.
Luce, R. D. & Fishburn, P. C. (1991). Rank- and sign-dependent linear utility models for finite first-
order gambles. Journal of Risk and Uncertainty, 4, 29-59.
Luce, R. D. & Fishburn, P. C. (1995). A note on deriving rank-dependent utility using additive joint
receipts. Journal of Risk and Uncertainty, 11, 5-16.
Luce, R. D., Krantz, D. H., Suppes, P. & Tversky, A. (1990). Foundations of Measurement, vol. 2. San
Diego: Academic Press.
Luce, R. D. & Narens, L. (1985). Classification of concatenation measurement structures according to
scale type. Journal of Mathematical Psychology, 29, 1-72.
398 R. Duncan Luce
Narens, L. (1981a). A general theory of ratio scalability with remarks about the measurement-theoretic
concept of meaningfulness. Theory and Decision, 13, 1-70.
Narens, L. (198U). On the scales of measurement. Journal of Mathematical Psychology, 24, 249-275.
Narens, L. (1996). A theory of ratio magnitude estimation, lournal of Mathematical Psycholo£v 40
109-129. J &y' '
Stevens, S. S. (1946). On the theory of scales of measurement. Science, 103, 677-680.
Stevens, S. S. (1951). Mathematics, measurement and psychophysics. In S. S. Stevens (Ed.), Handbook
of Experimental Psychology, pp. 1-49. New York: Wiley.
Stevens, S. S. (1975). Psychophysics. New York: Wiley.
British Journal of Psychology (1997), 88, 399-400 Printed in Great Britain 399
© 1997 The British Psychological Society

Measurement in psychology: Commentary on

Michell's Quantitative Science and the
definition of measurement in psychology

Michael Morgan*
Institute of Ophthalmology and Department of Anatomy/Developmental Biology, University College London,
Gower Street, London WC1E 6BT, UK

Michell's argument is that the branches of psychology that use numbers are vitiated because they do so
without properly validated scales of measurement. In particular, they do not obey the axioms of Holder
(1901) which appear to describe the behaviour of continuous quantities like mass.
The two fields of quantitative psychology which come in for the greatest drubbing are psychometrics
(mental testing) and psychophysics. In the case of psychometrics I am not competent to judge whether
Michell's criticisms are damaging or not. In the case of psychophysics the criticisms seem to me to be
largely irrelevant to actual practice. If a mainstream journal of visual psychophysics is opened, for
example, it will be found that the measurement scales in use are visual angles (arcmin), luminances
(cd/m^), contrast (a ratio scale) and time. In other words, these are physical scales.
The fundamental dependent variable is the probability of detection/discrimination, which is a
continuous, well-behaved scale. Thus results are expressed as the physical magnitude at which some
criterion level of probability is met, often 75 or 84 per cent correct. The reason for this is that the
majority of psychophysicists have followed the tradition of E. H. Weber in measuring the limits of
sensory discrimination power in physical units. The main task of psychophysics is to determine the
relevant physical dimension that limits performance, which is not always obvious.
For example, the best vernier acuity is in the region of 5 arcsec. This is interesting because the
diameter of the foveal photoreceptor is ~ 20 arcsec. This may appear mysterious but when optical
blurring is taken into account, it is found that a 5 arcsec shift in the position of a line implies a 2 per
cent contrast shift in the photoreceptor best placed to detect the shift. This is similar to the eontrast shift
required to detect the presence of a very thin line. Vernier acuity for Unes is reduced when they move.
Is velocity the relevant measurement scale? No, because if sinewave gratings are used instead of lines
the loss of acuity is predictable not from their velocity but from their temporal frequency (Hz).
The Weberian tradition preceded that of Fechner, which Michell aptly criticizes. Fechner wanted to
measure the magnitude of sensation, and thought he could do so rationally by totting up jnds. There may
be people out there who still believe this, but I doubt it. The enterprise of measuring the magnitude
of a sensation appears to have been doomed because there is no such thing. Most psychophysicists
believe, on the other hand, that it is reasonable to rank order sensations by magnitude. Astronomers
had been doing this with the brightness of stars long before psychology had been invented. Michell has
little time for ordinal scales, and presents them as desperate inventions of S. S. Stevens to deflect
attention from the failure of ratio scales. I do not follow the reasons for this severity. Other sciences
have worked quite happily with scales that are not continuous, and which therefore fail to satisfy the
Holder axioms. For example, classical genetics used a unit called (as it happens) the morgan or
centimorgan. This was the distance apart between two genes measured by the probability of
recombination. The fact that centimorgans turned out in many cases to be additive was taken as
evidence for the Unear arrangement of genes on chromosomes. But it is clear that this scale never had

* Requests for reprints.

400 M. Morgan
a hope of being continuous, since genes are discrete. Similarly, molecular biologists now happily
measure the genome in kilobases. Again, this cannot be continuous because the number of bases is
discontinuous. The requirement of continuity for a scientific measurement scale is far too restrictive.
And what about quantum mechanics?
There are some points on which it would have been interesting to have Michell's comments. First
psychophysicists have spent some time in trying to devise measurement scales for comparing
performance on different tasks. For example, is vernier acuity better or worse than wavelength
discrimination? The best solution proposed so far is to compare relative efficiencies on the two tasks.
Relative efficiency is a dimensionless ratio deriving from Fisher, which compares the statistical efficiency
ofthe real observer to that of an ideal observer. Are there any objections in principle to dimensionless
scales like this?
Another class of problem arises when asking whether vernier acuity is worse for chromatic stimuli
than for achromatic. Here we can measure acuity in the same units (arcsec) but the problem is that acuity
IS affected by contrast and we don't know precisely how to equate the contrast of chromatic and
achromatic stimuli. One solution is to define contrast as cone contrast: the ratio of photon captures
between the two relevant cone classes. The problem is that cone contrast is known not to predict the
detectability of chromatic stimuli: equivalent cone contrasts are better detected when they are
chromatic, at low spatial frequencies. The alternative solution has been to equate stimuli for
detectability, by expressing their contrast as a multiple of their contrast at detection threshold. This is
getting dangerously close to Fechner's equal-magnitude jnd conjecture. Is there, I wonder, a case here
for conjoint measurement? I was first introduced to conjoint measurement by Stephen Lea, who applied
It to problems of measurement in animal learning (Lea & Morgan, 1972). Michell gives a tantalizing
glimpse into this recondite field without explaining how it is useful. Are there any cases in
psychophysics where it could solve existing problems?
In summary, I thought Michell's article rather cruel to ancient horses, or to take another equine
analogy, very good at shutting stable doors after the flight of their occupants. Psychophysicists do not
do what Michell thinks they do.

References
Lea, S. E. G. & Morgan, M. J. (1972). The measurement of rate-dependent changes in responding. In
J. MiUenson & R. Gilbert (Eds), Reinforcement: Behavioral Analyses. New York: Academic Press.
British journal of Psychology (1997), 88, 401-406 Printed in Great Britain 401
© 1997 The British Psychological Society

Reply to Kline, Laming, Lovie, Luce

7
and Morgan

Joel Michell*
Department of Psychology, University of Sydney, Sydney, NSW 2006, Australia

My paper proposed first, that psychology is committed to the scientific task of testing
the hypothesis that its supposedly measurable attributes really are quantitative;
second, that from its inception, modern quantitative psychology has, with few
exceptions, ignored this task, concentrating instead upon the instrumental task of
quantification; and third, that modern psychology has adopted an intellectually
pathological defence mechanism against recognizing the existence of this scientific
task. The commentaries by Kline, Laming, Lovie, Luce and Morgan either amplify
or criticize my arguments for these theses. I will consider each thesis in turn,
presenting a sketch of my argument and then assessing the force of the criticisms.

1. The logical thesis

In summary, my argument here is as follows.
Premise 1. All measurement is of quantitative attributes.
Premise 2. Quantitative attributes are distinguished from non-quantitative attributes by the possession
of additive structure.
Premise 3. The issue of whether or not any attribute possesses additive structure is an empirical one.
Conclusion 1. The issue of whether or not any attribute is measurable is an empirical one.
Premise 4. With respect to any empirical hypothesis, the scientific task is to test it relative to the evidence.
Premise 5. Quantitative psychologists have hypothesized that some psychological attributes are
measurable.
Final thesis. The scientific task for quantitative psychologists is to test the hypothesis that their
hypothesized attributes are measurable (i.e. that they possess additive structure).
Morgan rejects premise 1; like Stevens, he advocates the inclusion of ordinal scales as measurement.
This view obscures a crucial difference between the measurement of a quantitative attribute and the
numerical coding of an ordinal structure (an ordinal scale, so-called). On the one hand, ratios of
magnitudes of continuous quantities instantiate real numbers (Holder, 1901) and so are intrinsically
numerical. (Luce makes a similar point in different terms. Holder's ratios being extensionally equivalent
to Luce's translations in the case of continuous quantity.) By contrast, merely ordinal structures are not
intrinsically numerical, although of course it may be convenient to use a numerical code for any
structure known to be ordinal. The essential point is that if an attribute is merely ordinal, the scientific
task is to discover this and not to pretend that it is quantitative.
Morgan also claims that not all quantitative attributes are continuous and in this he is correct. I
emphasized continuous quantities because psychologists have generally theorized in those terms.
However, discrete and merely Archimedean quantities carry empirical commitments similar to

* Request for reprints.

402 / . Michell
continuous quantities with respect to the requirement of additive structure and, so, had psychologists
theorized, instead, in these terms, premises 1, 2 and 3 would have applied just the same.
The scope of premise 5 is challenged by Morgan's claim that modern psychophysics pays little
attention to the measurement oi psychological attributes being more concerned with the measurement of
jfiAy.r/ira/attributes because research now follows the tradition of Weber rather than that of Fechner. This
is true of some psychophysical research, but interest in Fechner's question (the form of the
psychophysical function) remains strong (see Krueger, 1989, and Murray, 1993). Attempts to measure
sensations presume that there are sensations to measure and this presumption is questionable. An
alternative approach is to interpret psychophysics as 'the capacity ofthe organism to respond correctly
to stimuli' (Boring, 1921, pp. 459-460) and this falls clearly into the Weberian tradition (see also Cattell,
1893, and Luce, 1972). For those within the Fechnerian tradition, my logical thesis applies.
Lovie's claim that 'there are no absolute, ahistorical mathematical truths or methods, only locally
developed and locally maintained collective commitments and practices', implies that my thesis is false.
On his view, there is the 'local' practice of regarding certain procedures (e.g. psychological testing) as
measurement and if these procedures are not measurement according to other 'local' groups of
scientists then so be it, for there is no generally applicable standard by which measurement can be
judged.
Lovie is a relativist (see Lovie, 1992) and suffers the philosophical parallax of that position. More than
once he mistakes my realism iot positivism. Realism and positivism are largely contraries and positivism
is really a species of relativism, a fact clearly apparent in S. S. Stevens' (1936) operationalist version. Not
only did Stevens explicitly reject realism but he anticipated much of modern relativism: the thesis that
scientific truth is socially constructed and relative; that there is no data language 'given' by nature; that
there is no viable distinction between data and theory; and that mathematics and logic are arbitrary
human inventions. More than this, Bridgman's doctrine of operational definitions, with its confinement
of science to laboratory concepts, is echoed in Lovie's view of science confined by local practice and
it is now established that Carnap's (1950) conventionalism and ontological relativism anticipated Kuhn's
(1970) (e.g. Earman, 1993, and Irzik & Grunberg, 1995). Lovie's position, not mine, abuts positivism.
If science is taken realistically (i.e. as the attempt to understand the ways of working of natural
systems), and its successes leave us no reasonable alternative, then a major task for the philosophy of
science is to specify the kind of place the world must be, in its most general features, for it to be possible
that some scientific theories are true (where by true is meant absolutely true, i.e. things being just as stated
in those theories). Applying this to quantitative science (exemplified paradigmatically in physics), the task
is to specify the character which quantitative attributes must have if they are both measurable and
interrelated continuously. An intellectual tradition stretching from Euclid in ancient Greece to modern
measurement theorists has unfolded the logic of quantity and measurement. If, in formulating their
theories and practices, psychologists copy established quantitative science (and they do), then their
efforts must accord with this logic.
Laming's comments about the role of theory in establishing measurement could be misunderstood as
suggesting that theory alone somehow delivers quantification without empirical tests specifically
sensitive to additivity within the hypothesized attribute being necessary. We can distinguish measurement
theory from substantive theories (which Laming curiously dubs 'real theory') as follows. Different
substantive theories in quantitative psychology (e.g. signal-detection theory, Rasch's item-trait theory,
etc.) have features in common. One such feature is the hypothesis that the attributes involved are
continuous quantities. Because of this, the theory of continuous quantities (i.e. measurement theory)
applies to each. The empirical hypothesis that an attribute is quantitative is not metaphysically
privileged, miraculously immune from test. When Luce, referring to Narens' (1996) theoretical
reconstruction of Stevens' method of magnitude estimation says that it incorporates 'a plausible
empirical law, which if true indeed justifies believing this is measurement at the level of ratio scales' (my
emphasis), the word true has not suddenly changed its meaning because the subject is quantification. If
Narens' ' plausible empirical law' is true, then that is because things are as it states them to be, and the
only way we have whereby such matters can be known is observation, imperfect as it is.
Complex theories, such as Narens' or, say, signal-detection theory (to take Laming's example), can be
tested in many different ways. Not all tests of quantitative theories are sensitive to the issue of whether
or not the hypothesized attribute (in this case, sensation intensity) is quantitative. Unless the test is
sensitive to the presence of additivity within the attribute, the issue of quantity is not tested. For example.
 Reply to Kline, Laming, Lovie, Luce and Morgan 403
Laming's comment about the signal-detection theory index, d', being already established by experimental
psychology as a psychological measure of some kind, requires the support of tests of this kind.
d' may be calculated from a participant's hit rate, H, and false alarm rate, F, by taking the difference
between the two ^ scores on the unit normal curve above which exactly proportions H and F,
respectively, lie. That is,

where ^ is the inverse of the usual normal distribution function. If experimental psychologists have
discovered that for some set of quantitative stimulus attributes, d' increases in proportion to the physical
difference between stimuli, that is a significant discovery, but it is one which no more establishes
psychophysical measurement (or additivity) than does the fact that transformations or combinations of
test scores correlate systematically with some criterion. In the absence of a theory, it is just an
observation about the relationship between physical stimulus differences and a numerical function of hit
and false alarm rates, these rates being directly observable frequencies (not measures of anything). As
with test scores, results such as these raise scientific questions, they do not solve them. The scientific task
is to explain why the observed regularity occurs. A proposed explanation may involve the postulation
of distinctly psychological quantities, or it may not. If it does, then it may be possible to interpret d'
as a measure of something psychological, but such an interpretation will only be correct if the proposed
theory is true and that, in turn, will only be the case if the postulated psychological attributes really are
quantitative. The theory must be tested and the test must be sensitive to the presence of additivity. Only
when supported by such tests does the claim of measurement become more than speculation.
Laming knows this. For example, he (1986) refers to empirical conditions specified by Levine (1970)
and Falmagne (1971) for pair comparisons to be scalable in Fechner's sense. (Falmagne, 1985, provides
further empirical conditions relevant to psychophysical measurement.) In my paper I drew attention to
the psychophysical research tradition that has attempted to investigate the quantitative hypothesis.
However, Laming's judgment that this tradition has 'substantially solved the problem which Michell
addresses' (i.e. the question of quantity) is more sanguine than mine and not supported by other
commentaries mentioning psychophysics. Furthermore, Laming knows only too well that many
researchers in psychophysics neglect these questions, a point he has eloquently made for the general
reader elsewhere (Laming, 1987). Despite his throw-away hne, that I am simply 'directing the critical
power of modern measurement theory against the psychology of I860,' his true attitude to psychological
measurement is, I believe, more faithfully revealed in his other, deftly delivered, salvoes at modern
targets indistinguishable from mine.
The logic that applies to Laming's comments applies equally to Kline's. Kline accepts that ' it is true
that psychometric measurement is not scientific in the sense defined by Michell' and accepts the force
of my logical thesis in relation to social measurement, but with respect to personality measurement and,
more so, abihty measurement, cannot bring himself to swallow its full implications. As scientific
measures, he thinks, 'personality scales are far from worthless' and, better still, 'psychometric testing
is leading to a real scientific understanding ofthe phenomenon' of intelligence. Indeed, Kline appears
convinced that intelligence tests will be to some future scientific apparatus for measuring ability as
Galileo's telescope is to the modern apparatus at Jodrell Bank. On present evidence, I cannot say
whether this view is false or otherwise, of course, but I do know that the best defence against our
seemingly limitless capacity for self-delusion is to try to put our ideas to empirical test. If Kline's faith
in psychometrics as scientific measurement proves true, it will only ever be shown thus by researchers
who, taking a more critical view, find ways to test the hypothesis that abilities are quantitative, ways
specifically sensitive to additive structure. None of the evidence that Kline mentions bears upon the
scientific task of quantification. For example, factor analytic studies are insensitive to the falsity of this
hypothesis because they already presume that the relevant psychological attributes (i.e. abilities) are
quantitative.
Because all research makes assumptions, I have no quarrel with any particular researcher who
candidly assumes that the abilities postulated are quantitative, one at the same time aware of the extent
to which this assumption qualifies any conclusions drawn. My quarrel is with a discipline that while
invariably begging the same central assumption, rarely acknowledges this fact, never explores the
implications of this fact for conclusions drawn and evinces little interest in empirically investigating the
truth of this assumption. In believing that the evidence he adduces moderates the force of my critique.
404 /. Michell
when in fact that evidence is logically irrelevant to it, Kline betrays a resistance, typical of psychologists
in general, to facing the implication that psychological attributes may not be quantitative at all and that
psychological measurement, in the scientific sense, may thus never be able to be realized. Perhaps this
is because such psychologists cannot quite believe that if such measurement proves impossible,
psychology will be no less a science than if quantification is realized. In Galileo's words, 'We must not
ask nature to accommodate herself to what might seem to us the best disposition and order, but must
adapt our intellect to what she has made, certain that such is the best and not something else' (quoted
in Crombie, 1994, p. 45).
Luce's comments on units and symmetries reflect a slant towards measurement theory differing
slightly from mine, but the difference is not substantial. As I have discussed this complex matter
elsewhere (Michell, 1993, 1994), I will confine my reply. Luce is correct to note that any quantitative
attribute ' might be quantifiable in several ways' and that these might result in quite distinct relations
of additivity being identified within the same quantity, as his velocity/rapidity example shows. What
this implies is that for any two magnitudes there is no unique ratio (as measurement theorists from
Euclid to Holder seem to have thought). In order to define a ratio between two magnitudes and, thus,
in order to measure anything, a relation of addition between magnitudes of the relevant kind must also
be specified. At least in the first instance, such a relation can only be discovered empirically (via
extensive or conjoint measurement procedures, for example), although subsequently further additive
relations can always be defined via transformations of the resulting ratios (e.g. logarithmic
transformations). This observation reinforces the need to discover additive structures within an
attribute by empirical methods.
In summary, the force of my logical thesis remains: psychology is committed to the task of testing
the hypothesis that its supposedly measurable attributes are quantitative. Its force is not moderated by
the fact that, to some extent, this task has been pursued already in psychophysics and, of course, in some
other areas (e.g. Luce's research on utility).

2. Historical thesis
My historical thesis consists of the following components.
Part 1. Because oi practicalism and scientism, the founders of modern quantitative psychology simply
assumed that psychological attributes are quantitative.
Part 2. Consequently, they concentrated upon the instrumental task of quantification and ignored the
scientific task.
Part 3. The report of the Ferguson Committee displayed the fact that, as a result, psychological
measurement was anomalous relative to the logic of quantity and measurement.
Part 4. Institutional endorsement of Stevens' definition of measurement deflected this criticism and
satisfied most psychologists that measurement was attainable via the instrumental task alone.
Lovie expresses a global disagreement with this thesis. According to him, I have neglected even to
attempt 'the serious historical and conceptually embedded work of finding out what Stevens actually
did in the late 1930s and early 1940s (and why he did it)'. Lovie believes that there is only one level at
which the history of science can be undertaken (viz., that ofthe 'lived-work' ofthe scientist). This is
a reductionist fallacy. There is no ultimate level of inquiry in any scientific field, and the serious scholar
adapts his or her focus to suit the questions at hand. This inevitably means ignoring what one adjudges
to be irrelevant detail. The historical questions that interested me concerned certain broad conceptual
trajectories relating to quantification and measurement, passing through late 19th and early 20th century
thought, converging upon Stevens and thereafter affecting the way measurement was defined in
psychology. I was not interested in more microscopic details of Stevens' life. If I am mistaken in
identifying the lines of development that I think I have detected, it will be shown by historians studying
this history at the same level as I have done and not by historians pursuing Stevens' 'lived-work'.

3. Sociological thesis
My argument here is as follows.
Premise 1. Resistance to readily available facts is symptomatic of thought disorder.
 Reply to Kline, Laming, Lovie, Luce and Morgan 405
Premise 2. The logic of quantity and measurement constitutes a set of readily available methodological
facts.
Premise 3. Modern psychology resists the incorporation of these methodological facts into its ideological
support systems (i.e. its methodology courses and texts, and its research programs).
Conclusion 3. Methodological thought disorder is systemic in modern psychology.
Here, premise 2 is the proposition that the body of knowledge constituting the logic of quantity and
measurement is readily available (or 'obvious') to psychologists. Lovie labours the obvious point that
the concept of the obvious is a relative one. I did not suggest otherwise. If this body of knowledge is
readily available in the psychological literature (and it is) and if quantitative psychologists are largely
unaware of this body of knowledge (and they are), then it is because it has been excluded from this
science's ideological support structures. Prima facie this is odd because this body of knowledge is highly
relevant to attempts at quantification. Drawing the reader's attention to this anomaly is hardly a
rhetorical device: something is radically amiss in a scientific discipline when crucial methodological
matters are available, but ignored.
Lovie was intrigued by my use of the sociological concept of an ideological support structure, but
unhappy because, in his eyes, it amounted to gerrymandering ontologicalty (or ' fatally undermining the
other's position by treating their givens as problematic, while not applying the same (honest or
consistent) analysis to my own' (Lovie, 1992, p. 32)). This accusation is false. I have attempted a
socio/historical explanation for the current situation in quantitative psychology, but not in order to
undermine anyone's position. The acceptance of any set of beliefs will always have its socio/historical
causes, but these causes can never undermine those beliefs because these causes are always logically
independent of the issue of the truth or falsity of those beliefs. I have argued that the current
understanding of measurement in psychology contradicts the logic of quantity and measurement and
is mistaken just for that reason. I have presented the socio/historical facts in an attempt to understand
the causal circumstances of his mistake.
Only when the role of social interests is made explicit, can the error be understood in context. It is
important to do this. I know that the message that most psychologists have erred about the concept of
measurement will be resisted. This is evident in Laming's claim that I am 'directing the critical power
of modern measurement theory against the psychology of I860' and Morgan's, that I am 'shutting the
stable doors after the flight of their occupants', as if the logical problems identified have been
miraculously removed by time alone. One way to deal with such resistance is to uncover the deeper
socio/historical dynamics. By exposing the ideological forces and social interests at work, the way is
opened for a more objective view of one's science.
Pertinent to this point is Lovie's rejection of my diagnosis of modern psychology's methodological
thought disorder. Lovie may have another name for it, but anyone who takes science seriously will
appreciate that when a particular discipline adopts its own, peculiar definition of a fundamental concept
and is thereby able to continue avoiding for another 50 years the same central empirical issue it had
already avoided for almost a century, that discipline is malfunctioning as a scientific enterprise. At first
sight, it might appear that relativism's intellectual scorched-earth policy offers exemption from this
diagnosis. However, a thoroughgoing relativism about truth is unsustainable and Lovie, inevitably,
makes claims (about my paper, for example) which I am encouraged to think he believes state the way
things"are objectively.
One such is his description of my paper as 'a stale and superficial morality tale, with S. S. Stevens
and his acolytes cast as the villains, and Holder and his modern followers as heroes'. However, he is
mistaken. My paper makes no moral judgments and it identifies no villains or heroes. It does identify
certain errors in thinking on the part of individuals and it identifies an instance of thought disorder in
modern psychology, but these are not moral judgments, nor are the defects identified the defects of
villains. Lovie's colourful categories trivialize the issue and only serve anti-intellectual interests. The
debate will be advanced only by taking the issue seriously and either showing where my argument is
factually or logically mistaken or, failing this, taking my conclusions as a basis for action. A necessary
condition for reversing the diagnosed situation is a strengthening of education in the conceptual
foundations of methods in psychology (i.e. in the mathematics, logic, and the history and philosophy
of science necessary for critical thinking about the methods used by psychologists). Because, at present,
I only see a weakening of the curriculum in our universities, I am not optimistic.
406 /. Michell

Conclusion
Have the criticisms of my paper, raised here, deflected its aim or diminished its force? No critic has
explained why the hypothesis that an attribute is quantitative, alone amongst scientific hypotheses, must
be exempt from empirical test. No critic has explained why psychology, alone amongst the sciences, is
entitled to its own definition of measurement, a definition that leaches the concept of all content. And
none has submitted a rational defence of the present, anomalous, situation in quantitative psychology.
Readers of this journal have been given no adequate reason, yet, to avoid the conclusion that
methodological thought disorder is systemic in modern psychology.

Acknowledgements
I am grateful to Benjamin Barnes, James Dalziel, Scott Gazzard, Fiona Hibberd, George Oliphant and
Andrew Rantzen of the Department of Psychology, University of Sydney for their comments upon
earlier versions of this reply.

References
Boring, E. G. (1921). The stimulus-error. American Journal of Psychology, 32, 449-471.
Carnap, R. (1950). Empiricism, semantics, and ontology. Kevue Internationale de Philosophie, 4, 20-40.
Cattell, J. McK. (1893). On errors of observation. American Journal of Psychology, 5, 285-293.
Crombie, A. C. (1994). Styles of Scientific Thinking in the European Tradition, vol. 1. London: Duckworth.
Earman, J. (1993). Carnap, Kuhn, and the philosophy of scientific methodology. In P. Horwich (Ed.),
World Changes: Thomas Kuhn and the Nature of Science, pp. 9-36. Cambridge, MA: MIT Press.
Falmagne, J.-C. (1971). The generalized Fechner problem and discrimination. Journal of Mathematical
Psychology, 8, 22-43.
Falmagne, J.-C. (1985). Elements of Psychophysical Theory. Oxford: Oxford University Press.
Holder, O. (1901). Die Axiome der Quantitat und die Lehre vom Mass. Berichte iiber die Verhandlungen
der Koniglich Sdchsischen Gesellschaft der Wissenschaften ^u Leip^^ig, Mathematisch-Physische Klasse, 5
(English translation of Part I by J. Michell & C. Ernst (1996), 'The axioms of quantity and the theory
of measurement'. Journal of Mathematical Psychology, 40, 235-252.)
Irzik, G. & Grunberg, T. (1995). Carnap and Kuhn: Arch enemies or close allies? British Journalfor the
Philosophy of Science, 46, 285-307.
Krueger, L. E. (1989). Reconciling Fechner and Stevens: Toward a unified psychophysical law.
Behavioral and Brain Sciences, 12, 251-320.
Kuhn, T. S. (1970). The Structure of Scientific devolutions. Chicago, IL: University of Chicago Press.
Laming, D. (1986). Sensory Analysis. London: Academic Press.
Laming, D. (1987). Psychophysics. In R. L. Gregory (Ed.), The Oxford Companion to the Mind, pp.
655-657. Oxford: Oxford University Press.
Levine, M. V. (1970). Transformations that render curves parallel. Journal of Mathematical Psychology 7
410-443. '
Lovie, A. D. (1992). Context and Commitment: A Psychology of Science. New York: Harvester-Wheatsheaf.
Luce, R. D. (1972). What sort of measurement is psychophysical measurement? American Psychologist
27, 96-106.
Michell, J. (1993). Numbers, ratios, and structural relations. Australasian lournal of Philosophy 71
325-332. J^J' '
Michell, J. (1994). Numbers as quantitative relations and the traditional theory of measurement. British
Journal for the Philosophy of Science, 45, 389-406.
Murray, D. J. (1993). A perspective for viewing the history of psychophysics. Behavioral and Brain
Sciences, 16, 115-186.
Narens, L. (1996). A theory of ratio estimation. Journal of Mathematical Psychology, 40, 109-129.
Stevens, S. S. (1936). Psychology: The propaedeutic science. Philosophy of Science, 3, 90-103.