404 THE FACTORS OF THE MIND

Let us therefore proceed to compute the correlations
between the various traits as measured in Table I, and then
factorize this second correlation table. The correlations
and reliability coefficients, calculated in the ordinary way,
are shown in Table IV. Practically every correlation is
positive ; and it is at once obvious that the figures are
tending towards as good a hierarchical order as we could
expect with a group so small and traits so multifarious.
Such an order implies, not only the presence, but the
predominance, of a general factor common to all the
traits: this first factor is, of course, virtually eliminated
when we correlate persons. On eliminating its effects
from Table IV, at least one other factor—possibly two—
are discernible. The saturation coefficients have been
calculated for all three,1 and are given, rearranged for
purposes of comparison, in the first three columns of
Table V.

Here we are interested neither in the first factor (' the
general factor,' as it is commonly termed), nor yet in the
third factor whose nature and even existence here seem
highly dubious. Our immediate interest lies in the second.
This, if any, will be the factor that corresponds with that
discussed above while correlating persons. The trait-
weights for the two, as obtained by correlating traits and
persons (with the c abridged method' for the latter), must
accordingly be compared. The figures are given in the
second and fourth columns of Table V, headed t. ii and
p. ii. The figures in the latter column are simply the

1 The calculation was carried out in accordance with Dr. Stephenson's
own description of his method. The procedure is based on Spearman's
well-known summation formula. It differs from my own in implying a
slightly different figure for the self-correlation of each test. If this implied
figure is inserted, then a check with table-by-column multiplication shows
that the saturations obtained do not differ greatly from those that would
be obtained by the least-squares method with these self-correlations. The
choice of figures for the self-correlations is largely arbitrary ; and I myself
should prefer to insert values that would give (i) zero correlations between
the saturation coefficients for different factors and (ii) a total of zero for the
bipolar saturation coefficients of the same factor. This would incidentally
yield saturations for t. ii that were much closer to the figures given under