Stephenson, W. (1935). Correlating persons instead of tests.
Character & Personality; A
Quarterly for Psychodiagnostic & Allied Studies, 4, 17–24.
https://doi.org/10.1111/j.1467-6494.1935.tb02022.x
CORRELATING PERSONS INSTEAD OF TESTS
W. STEPHENSON
Psychological Laboratory, University College, London
This is not going to be a statistical article. Instead, I am to
introduce a new general technique to readers of this journal, one
which should be of greatest interest to them, because its widest use
will be in the study of human personality. The technique is a com-
plete inversion of all previous factor techniques.^ It is a very simple
notion indeed, but it brings a new tool to the aid of psychology, gen-
eral, social, or individual.
It is necessary, first, to give a brief account of the present-day
factor technique.
T H E PRESENT-DAY FACTOR TECHNIQUE
Today we correlate tests or estimates. The work of Professor
Spearman and those he has so greatly influenced in England and
America, whether along statistical or experimental lines, has been
concerned with group tests and estimates, such as tests of intelligence
and estimates of character qualities. These are used with large
populations of persons, and correlated.
The correlations for a number of tests, as is well known, are
next analyzed in terms of factor theorems, like the Two Factor
Theorem of Professor Spearman.
All this serves two ends. The analysis can only be, or should
only be, a scientific check or test of psychological hypotheses. All
science proceeds along this general line, and psychology cannot be
an exception. Next, science is concerned not with unique events (an
axiom that some workers in psychology overlook), but with repeat-
able processes, and correlations are above all things indices of this
repeatability.
' It was described as such in a letter to Nature, June 30, 1935.
18 CHARACTER AND PERSONALITY
Present-day factor work, then, can be truly scientific. Its result
is to tell us first how people are universally alike, and thereupon to
allow us to measure them for the processes in which they are alike,
because only with such a warranty could we compare them. Thus
we can measure all individuals, normal or otherwise, babies or
adults, for a few universal factors such as Spearman's g, c, and w,
which represent, it so happens, the broadest features respectively of ,
cognitive, affective, and volitional behavior.
Large populations of persons, however, are required in ex-
periments that are made on these factors, and as a consequence the
quality of the testing suffers. The size of the population sets a
limit on the tests that can be applied, and on the apparatus and con-
trols that can be used. All the refinement and delicacy, the intricacy
and subtlety of the laboratory, are lost just when it is most needed
in the study of the most complex of all activity, human behavior.
Experimental work is slowed down, almost to the point of being
burdensome. A research student may spend two years isolating a
single factor. All one's controls and hypotheses have to be sand-
wiched into the one experiment, to be performed on one and the
same group of persons. One cannot perform an experiment today
and use its results for another tomorrow. Tests cannot be changed
in the way that a chemist changes the mixtures in his test-tube to try
out his hypotheses. In short, the present-day technique lacks the
pliability that the energetic experimentalist wants at his command.
It is a device for massive field work, and not for the clinic, the
laboratory, or for rapid and subtle experimentation. On the other
hand, this massive technique has features that no other can replace.
But the problem arises, and it is one that has bothered me for
some time, can the above disabilities be removed? Can we make
factor studies on a few individuals, and bring the methods of cor-
relations and factor analysis into the laboratory?
T H E N E W TECHNIQUE
Fortunately, for a large part of psychology, these questions can
be answered affirmatively. The solution is to correlate persons in-
stead of tests. In my letter to Nature I indicated that all present-day
factor theroms could be so inverted.^
Whereas previously a large number of people were given a small
number of tests, now we give a small number of people a large num-
'Ibid.
CORRELATING PERSONS INSTEAD OF TESTS 19
ber of tests or test-items, or require a large number of responses
from them. Previously individuals obtained scores; now the tests
get them instead, due to the operation of the individuals upon them.
By the present-day technique we obtain the factor saturations or
loadings of tests, but by the new one we can obtain saturations for
individuals. By the older technique we could estimate a person's g,
or c, or w, or other factor; now we can obtain factor estimates for
test-items.
But a concrete example of the new technique in use will do more
than any further description to illustrate some of the above points,
and in the following section of this article I describe one such study,
made in a few hours, concerned with those highly personal things,
our predilections and tastes.
The whole domain of psychology, however, is open for the
technique to explore: introspective psychology in general, our pref-
erences, sentiments, motives, and ideals; studies in aesthetics, such
as investigations into the nature of beauty; studies in industrial
psychology, on fatigue, sensory and motor acuities, and the like;
studies in educational psychology, such as examinations of the abil-
ities of teachers to mark essays or drawings. The technique reaches
into all these, no less than into those somewhat rarified regions of
sensory and psycho-physical psychology,
II
EXPERIMENTS ON TASTE OR PREDILECTION FOR COLOR
We are not, of course, here concerned with those tastes which
require the tongue for their perception, but with the predilections
that we have for some things rather than for others. The new fac-
tor technique described in the preceding short note could be used to
conduct experiments on both kinds of tastes, but here I am to be
concerned with the latter only.
Tastes and matters of tastefulness make up a good part of the
stock-in-trade of the educated person. It is not that the person
likes one thing more than another, or that it gives him greater pleas-
ure only. I may like old pipes, even to the extent of being sen-
timental about them, but my taste may be for choice cigars. To have
taste is to be discerning, to relish things intellectually no less than
affectively or by way of sentiment.
We have to distinguish, too, between tastes and active participa-
tion in them. I may have a taste for etchings or for Giotto-like
20 CHARACTER AND PERSONALITY
paintings, for old books or for fashionable suits, and yet neither
possess any of these nor take any active interest in them. If I had
the opportunity, then no doubt I might possess these things for which
I have a taste. Tastes are merely tools to be used when an appro-
priate occasion arises.
Taste is all a matter of opinion, so popular opinion avers, and
one man's taste might be the abomination of another. Tastes are
subject to fashion, and are highly partial too. We have tastes for
clothes and luxuries generally, but I have yet to hear of the engineer
with a taste for A.C. generating motors.
Finally, no one would deny, I hope, that tastes are part of one's
personality. Obviously their importance in a thousand and one
walks of life, for success in social and commercial activity, is of no
little moment, and one feels that psychologists should give them
far more of their attention than they have received hitherto. For
indeed very little has been done on the scientific study of tastes. All
that I have written above may be true of them, but woeful gaps are
left to be filled in by experiment. What kinds of tastes are there?
How specific in fact are they? Are they connected one with
another ? Have tastes a pleasure content, or are they intellectualized
only ? Are the things we like the same as those we have a taste for ?
The new factor technique can help us to develop and solve these
questions. It is outside my present purpose, of course, to answer
them all. I am merely to point to the problems and to illustrate the
efficacy of the new technique to solve them. To this end, then, I
describe the following short experiment on taste for color.
T H E EXPERIMENT
I took from an artist's colorman a sample of every homogeneous
colored paper that he stocked, sixty different colors in all. The
sample contained some brilliant poster colors, vivid reds, greens,
blues, and oranges; there were pure and delicate blue-green hues
and salmon tints, a range of tinted browns, greens, and greys. No
blacks or pure greys were included, and no very light creams or
whites. All the papers had the same "matt" surface.
As subjects I took 12 young women attending lectures, and 8
men of the same range of age and social status.
Each subject graded the colors for me on a scale ranging from
10 for the color he liked best, to 0 for that he liked least. No
reference had to be made to the use to which any color might be
CORRELATING PERSONS INSTEAD OF TESTS 21
put; judgments had to be taken quickly and as spontaneously as
possible. To ease my later calculations, the subjects graded the
colors in such a way as to supply a pre-arranged frequency distribu-
tion as follows:
Score (x) : 0 1 2 3 4 5 6 7 8 9 10
Frequency: 1 2 4 7 10 12 10 7 4 2 1
Thus one least pleasing color was given mark 0, the two next least
got mark 1 each, and so on up the scale.
Each of the sixty colors, then, received a score (x) from each
of the twenty subjects. To obtain reliability coefficients the whole
grading was repeated.
Now just as would happen if we had 60 people scored on 20
tests, correlations can be calculated for the 60 colors scored by 20
people. In the former case ri2 would be the correlation between
tests No. 1 and No. 2; in the latter case r^^ would be the correlation
between the scores given to the colors by subjects A and B.
The experiment should help us to answer several questions. If
each individual has his or her own specific likes and dislikes, then
the individuals will not correlate except by chance alone. But if
significant correlations appear, they could be factorized. It might
then be possible to describe as separate tastes any unitary factors
that emerge, to determine their range, and to measure individuals
for them.
T H E RESULTS
The facts in a summary form are as follows. They, and others
like them, are to be published in detail elsewhere.
Twelve individuals, 10 women and 2 men, correlated positively
among themselves a mean amount 0.36 ± 0.065, 0.66 being the
highest correlation. The rest correlated a mean amount —0.15 with
these twelve, and -j- 0.20 on the average among themselves.
That is, to a first approximation two distinct factors have
emerged, one among the twelve individuals, and the other among
the remaining eight, the one tending to be the obverse of the other.
Incidentally we can see from this example how useful the technique
will be to put "type" psychology on a scientific footing.
Next, the factor saturations or loadings of these individuals can
be determined, using for convenience the Spearman theorems. The
factor among the 12 persons we shall name factor I, and the other
factor II. The saturations of the individuals with these factors.
22 CHARACTER AND PERSONALITY
that is, an index of how well they serve to grade the colors for the
factor concerned, are as follows:
Subject No. 12 34 56 78 9 10
Saturation with
factor I: 70 -.65 .44 .63 .61 .63 .60 .57 .55 .63
Saturation with
factor I I : — — — — — — — — — —
Subject No. 11 12 13 14 15 16 17 18 19 20
Saturation with
factor I: 68 .72 — — — — — — — —
Saturation with
factor II: — — .50 .44 .32 .53 .44 .34 .43 .55
That is, subject No. 1 had a factor saturation 0.70 for factor I,
but none (except obversely) for factor II; and so on along the table.
Finally the factor content of each color can be calculated, just as
we can estimate the g of an individual by the old technique, now for
either factor I or for factor II, in terms of the two groups of persons
respectively.
EXPLANATION OF THE FACTORS
What, then, can be said about these two factors? Factor II is
for those persons who rated the pure bright colors as most pleasur-
able. Factor I, on the other hand, is for those who chose the subtle
tones and hues, the subdued colors, as most pleasing. The former
colors would be considered to be vivid but crude by a person of
taste, while the latter are nowadays regarded as the most tasteful.
Now an advantage of this technique is that there are only a few
persons concerned, and I know them all extremely well. Those
with factor I are certainly the more artistic and tasteful in general;
two are art students, and all dress with taste, although not neces-
sarily "smartly." (This raises the problem, indeed, whether those
who dress in ultra-smart fashions are lacking in taste.) Those
with factor II are frankly inartistic; some dress crudely; one hor-
rified me with his taste, or lack of it, for furniture; and in all of
them the bright colors have appealed to them as they do to children.
The exquisite blue of a sugar bag is merely a crude color to these
people, while the brilliant vermillion, that shocks the more tasteful
person by its very intensity, is highly esteemed. It may well be, too,
that women are usually more tasteful for color than men—a matter
for no surprise to judge by the clothes that men wear, and the fash-
ions that women have to attend to all their lives.
CORRELATING PERSONS INSTEAD OF TESTS 23
Thus, not only do the subjects with factor I apparently obtain
more pleasure from the colors they all tend to rate alike, but these
colors are by modern standards the most tasteful. Pleasure and
taste agree in this instance, although, to be sure, experiments can be
performed to check up thoroughly the point.
EXTENSION OF THE WORK
Now all this is just a beginning. To extend the work, we could
use the same series of colors to compare individuals from different
social groups, from different professions and occupations. Children
can be compared with adults, and children of one age with those of
another. From the factor estimates obtained for the colors new
tests can be devised, with more uniform colors for the factors con-
cerned. This, indeed, is one of the great services performed by the
new technique of correlating persons, for it allows us to construct
tests with known factor contents, and these, again, can be applied in
due course by the older technique of correlating tests.
Or the range of colors could be extended. One could even devise
interesting tests of color-blindness involving the new technique.
Individuals could be asked to select those that they believe to be
the most tasteful, instead of merely the most pleasing, to see whether
knowledge, pleasure, and taste are associated.
Again, we could correlate the same individuals, or a specially
selected group or groups, for other series of tests, designed to in-
vestigate other tastes. Tastes for vases, cloth materials, furniture,
perfumes, pictures, literature, and so forth can be factorized. It
could be determined whether these various taste factors are them-
selves interconnected. So with intimate control of the experiments
a science of tastes in general can be built up.
Nor would this work cease with the discovery of factors and
the investigation of their range. Since only a few individuals at a
time need be examined, it should be possible to enter into great de-
tails, to find out still more about the nature of the various factors.
Are there, for instance, deeper unconscious processes underlying our
preferences and predilections ? Is it intelligence, opportunity, hered-
ity, or environment that determines one's tastes ?
CONCLUSION
Sufficient has been said, I hope, to give an introduction to the
scope of the new factor technique. Its use on small numbers of
24 CHARACTER AND PERSONALITY
persons and its amenability to rapid experimentation are noteworthy.
I performed the above experiment within an hour and worked out
the results in two more. Its use as an essential tool to check the
theories of "type" psychologists, and as an aid in the construction of
homogeneous tests for further application by the new or the older
techniques, as well as its wide range of practical usage, is illustrated
in the above experiment and in the extended work that it merits.
Finally, one can factorize the correlations as one finds it con-
venient or appropriate to do, using Spearman's or anyone else's
factor theorems.
