414 THE FACTORS OF THE MIND

instead of tests was not a legitimate procedure, and was likely
to be 'both futile and misleading' (p. 178). Accordingly,
to test the differences between us, it was suggested that,
with the aid of our joint research students, the problems
chiefly affected should be worked over afresh by both
procedures.1 Their investigations, we hope, will shortly
be published.2 Meanwhile, Stephenson himself has given
a preliminary account of the highly interesting results
obtained by his modified technique in several short and
illustrative inquiries. So far, however, as I have already
indicated, I fail to see any serious discrepancy between
either our figures or our deductions.

Here we may confine ourselves chiefly to those studies that bear

1 At first it was suggested that the data already obtained with the
various tests of -p and / should themselves be correlated by persons.
Stephenson, however, held that such a procedure would be invalidated by
differences in the unit of measurement: ' since each of our tests has a different
unit, they cannot be correlated by columns and the factors would be meaning-
less/ This did not appear to me to be fatal, since, if they were first expressed
in standard measure for tests, the original measurements could still be
correlated for persons.

A more serious objection, as it seemed to me, was that the data supplied by
the tests, being confined to -p and /, did not cover a sufficiently wide field of
human behaviour : in correlating tests, our persons must be a fair sample of
the population of persons; and similarly, in correlating persons, our traits
must be a fair sample of the population of traits. Accordingly, for a valid
use of the new method it appeared essential to take, not a number of tests for
just a pair of important qualities, but a more comprehensive list of traits
(such as McDougall's catalogue of the primary emotions) professing to repre-
sent the entire emotional life. In his later studies Stephenson has, I think,
tacitly accepted this point of view, since the list that he has used, though
drawn from Kretschmer's psychobiogram instead of McDougall's catalogue,
includes (with a slight change of nomenclature) practically every trait in my
own list except sex. He himself, however, lays stress on a somewhat different
reason, namely, not so much to make the list a better sample of the total
universe of individual actions, but rather to reduce the probable error by
increasing the number of items over which the correlation is carried. Accord-
ingly, since number is his main aim, he does not object to repeating estimates
of the same trait in slightly different forms—a procedure I should be tempted
to criticize on the ground of the irregular distribution that the sample of
traits would then exhibit.

2 Some of their earlier results are described by Stephenson [97]. Fuller
accounts will be found in their M.A. and Ph.D. theses (University of
London Library).