P-, Q-, AND R-TECHNIQUES 189

Our methods of analysis must therefore be elastic and
eclectic, adapted specifically to the requirements of each
particular research. Whether we use a multiple-factor or
two-factor procedure, seek * non-fractional' group-factors
or c fractional? common factors, trust to c simple summa-
tion ' or prefer the ' method of least squares'—these are
questions to be settled afresh for the problem under investi-
gation, and not by a priori, cast-iron principles of Pro-
crustean universality. In short, I hold, as stated in my
original Memorandum, " that in principle, though not
perhaps in. detail, the statistical technique which has been
worked out for factorizing the results of a test is equally
valid for factorizing the operations of a person" ([93],
p. 275). For this reason, in summarizing the available
factor theorems, I described them as applicable, without
essential change, either to correlations between tests or to
correlations between persons.

That is precisely the view that Stephenson would con-
trovert. " All earlier investigations based on correlating
persons or extracting person-factors come far short of a
general psychometric technique, avowedly defined as such*
for the plain reason that it has refused to break with
conceptions that are part and parcel of the old R-technique.
Burt . . . still retains the standpoint of earlier psycho-
metry ; his work on P-technique, therefore, appeared to
Stephenson to have entirely missed the essential possi-
bilities implicit in the idea of correlating persons " ([138],
p. 276). Stephenson thus insists that his own technique

Eysenck has used paired comparison instead of ordinary ranking, and a
correlational technique instead of studying mean variations (cf. Guilford,
* The Method of Paired Comparison as a Psychometric Method/ Psych. Rev.,
XXXV, p» 494 f.) ; in such a case, if we take the usual table, whether we
start by correlating rows or columns, the results will be the same, because the
initial matrix is itself essentially symmetrical, and the variables for both
are the same series ofc tests,' in the one case serving as * standard ' or' judge,'
in the other case asi variable ' or item * judged*' On the other hand, where
we have a number of psychologists assessing a number of children, the
variables in both cases are c persons.' Hence, as I have argued elsewhere,
" the convenient antithesis between test and person is inexact and at times
misleading: what appears as a test in one investigation may be treated as
a person in another, and vice versa. The real antithesis is between comparing
rows and comparing columns in one and the same matrix " ([101], p. 67).