GENERAL. AND GROUP-FACTOR METHODS 313

first three tests from the next three. On turning to their
description, we find that the first six tests are all verbal,
whereas the content of the remainder is essentially non-
verbal. This factor therefore represents a dichotomous
classification of the fifteen tests into two main categories—
verbal and non-verbal.

The third factor is, as usual, c doubly bipolar * : that is
to say, its saturation coefficients for the first six tests (which
all had positive saturations with the second factor) are half
negative and half positive, and so add up approximately to
zero ; and again the saturation coefficients for the remaining
nine tests (which all had negative saturations with the second
factor) are four of them positive and five of them negative,
and so once again add up approximately to zero.1 This
third factor, therefore, really indicates two distinct dicho-
tomous sub-classifications : (i) first of all, it divides the
nine non-verbal tests into (a) a sub-group of four, which
turn out to be all numerical, and (ft) a sub-group of five,
which turn out to be all of the performance type, i.e. visuo-
kinassthetic ; (ii) and secondly it also divides the six verbal
tests into two antithetical kinds, and so explains the second
gap already noted. But the principle a/ division is no longer
the same. A glance at the nature of the tests reveals the true
reason. The first three tests, which are here given negative
saturation coefficients, are all labelled ' Opposites tests' ;
the remaining three are verbal intelligence tests of a more
miscellaneous character (Analogies, Definitions, etc.). It
would be absurd to argue that the Opposites tests are all
visuo-kinaesthetic and the remainder numerical.

Now let us compare this mode of classification with that
disclosed by the new factorial matrix reached after rotation,
We see that the new classification embodied in the c simple
structure ? agrees precisely in regard to two of the lines of
subdivision ; but the last of them has disappeared, or at

1 The totals are not exactly zero, but about -03. On turning back,
however, to the mode of computation, we discover that the discrepancy is
evidently due to the insertion in the leading diagonal of the highest correla-
tion instead of the figure demanded by the general pattern (see below,
Appendix I, p. 450). We shall find in a moment that this has apparently
influenced the interpretation offered.