CRITICISMS OF METHODS 379

" has put forward the claims of covariance analysis. He works
with unstandardized scores and covariances. Unfortunately, in
psychology, there are no generally accepted units (other than the
standard deviation). In our intelligence tests, for example, we can
obviously make the raw scores and variances what we like : both
are purely arbitrary matters, dependent upon the mere whim of the
psychologist. Clearly any principle involving such arbitrary founda-
tions can scarcely merit a moment's serious consideration." This
is a criticism which I have already endeavoured to meet (cf. above,
pp. 282-6). As regards traits, I readily agree that in psychology
the variances may be, and perhaps very commonly are, " purely
arbitrary matters " ; and I should equally maintain that, from the
statistician's point of view, this is c unfortunate.' But I do not
admit that it is entirely unavoidable. Stephenson cites the analogy
of intellectual tests. But here it is by no means difficult to show
that the amount of variance must differ appreciably from one
intellectual process to another, since it evidently depends on the
complexity of the process and also (it would seem) on the degree to
which it has become automatic (cf. pp. 283 f.).

When we are working with temperamental assessments, the
matter is more obscure. Yet even here I should have thought that
no one (except for convenience of calculation) could really suppose
that the variance for every trait or every person was precisely the
same. In describing the particular data with which my article was
concerned, I gave special reasons why it seemed unnecessary, and
indeed undesirable, to treat the variances and the standard devia-
tions as equal for all the traits. The exceptionally wide range of the
marks for ' fear' and ' anger,' for instance, was due, not to the
accidents of the psychologist's tests, but to the fact that exception-
ally wide differences of observed behaviour had been actually
recorded in the different persons studied. We had, in fact, picked
out a number of fearless delinquents and over-anxious neurotics ;
and it would seem very difficult to assume that the range of be-
haviour and the variance of the measurements were the same for
' fear' and * anger' as they were for (say) ' curiosity/ Since,
however, it was also of interest to consider what might happen in a
' selected standard population ' (i.e. a group in which the individual
variation might conceivably be regarded as approximately equal for
every trait, so that the standard deviations could be set at unity
throughout), I went on to show that the factors derived by my
method would fit the correlations quite as well as the covariances.

When we turn from traits to persons, it is often difficult to
discover what precise meaning is to be attached either to * standard-