ANALYSIS OF ILLUSTRATIVE GROUP 401

by each, factor to the eleven emotional traits. However, this pro-
posal raises several questions about which different investigators
are not altogether agreed.

To discover the nature of the ' positive type ' (as we may call it
for the moment) the simplest method of all would be to compare
the figures for each trait, as shown by the final column of sums in
Table III; to discover the nature of the ' negative type ' we should
take the same figures with the signs reversed. But, since the
calculation of these figures rests on an argument which may not
be generally accepted, it will be better first to inquire what other
means of determination may be available.

For those who treat the problem of comparing and correlating
persons as merely another case for the application of Spearman's
technique (first* inverting ' or, as I should prefer to say, transposing
the initial matrix of measurements) the most logical procedure
would be to employ Spearman's regression equation or * weighting
formula '—a method which, as it happens, is based on the principle
of least squares: only, instead of constructing a regression
equation to get factor-measurements for the persons, we shall
now, of course, construct one to get factor-measurements for
the traits.1 The calculations would be long ; but they were
apparently2 carried out by Beebe Center ; and his methods appear
in the main to have been followed by Stephenson in his recent
article. At all events Stephenson himself states that the figures he
deduces to describe his two temperamental types should be regarded
as Łtrue regression estimates' ([92], p. 304). Elsewhere, however,
he explains that " the person with the highest saturation is the most
useful for predicting (by correlation) the [measurements of the]
type to which all persons of one factor are approximating " ; and to
save labour it would seem that he generally employs simple inspec-
tion in order to pick out the most typical person or persons in his
batch.8 If so, to single out one or two persons from a small batch
of twenty or thirty would appear a somewhat precarious method of
setting up a standard.

1 D>6]> e<l- 3r> P- xx- -^n illustration of the calculations as adapted to
obtain factor-measurements for traits from correlations between persons was
given in my previous paper ([114], p. 175).

2 [8°1 P- 2O7* (Actually Center's final values are expressed in terms of
simple ranks.)

3 [97]> P* 351* ^n tne only case i*1 which, figures are actually given for
e predicting' or describing the type ([92], p. 303) the figures are simple
integers from o to 10, and obey his prescribed frequency distribution: so
that they are presumably merely the assessments of the one most typical
person in his batch of 21.

26