METAPHYSICAL STATUS OF FACTORS 235

thing more real than the covariancej whereas the reverse is
the case.

Thus, although most psychologists treat factor-analysis as
essentially concerned with the analysis of correlations, I
myself believe that it is better regarded as a mode of
analysing variance. Variance is admittedly a statement of
variation—that is, of differences, and of differences only.
And this mode of treatment, imposed on us at almost every
turn, brings out an important peculiarity of psychological
measurements as contrasted with physical measurements,
or at any rate with the common conception of physical
measurements. When we measure children's heights and
weights, our figures have the form of absolute measurements
taken from a zero point that is extraneously determined.
In psychology a few of our measurements, it is true, also
appear in an absolute form, e.g. the mental age or the speed
of reaction. But, whatever be the shape of the initial
measurements, in calculating a correlation our very first
step is to convert those measurements into deviations from
an intrinsically determined zero, namely, the arithmetic
mean.

The same is true of the factor-measurements and the
factor-saturations derived from those correlations. The
final factor-measurements as ordinarily computed are
expressed in standard measure : that is, they too are devia-
tions about their own mean. If, as I believe, the factor-
measurements for traits (determined by correlating persons)
are equivalent to the standardized factor-saturations for
traits (determined by correlating persons), then even the
saturation coefficients should in general be expressed as
positive and negative deviations. Thus, a factor-measure-
ment can never specify the absolute quantity of ability
that a man is supposed to possess, as we might specify the
pints of blood in his system or the ounces of brain-tissue
in his skull: it can only express the relations between differ-
ent persons' performances. And similarly the factor-satura-
tions can only express the way individual differences in test-
performances or measurable behaviour depend on individual
differences of a more general kind.

Sometimes, it is true, correlations are said to measure