CRITICISMS OF RESULTS 423

These various results show clearly the dangers of ' direct judg-
ment' (cf. [114], pp- I63-4), and the risks of accepting a simple
ranking of traits as revealing the special characteristics of the
particular person judged. When our concern is primarily with
temperamental types, we can, and I think always should, eliminate
this general factor either by partial correlation or, more simply, by
requiring our judges to mark each trait, not for its comparative
strength in the individual viewed in isolation, but rather for its
variation above or below the average or normal in the entire group
or population from which he has been picked. That, however, is a
task which can be performed with accuracy only by a trained and
experienced judge who has already had wide opportunities for
observing the kind of persons to be assessed. Nevertheless, under
specifiable conditions the < general factor' introduced by simple
ranking remains constant enough for us to allow for it, at least in
some crude measure : (a simple method will be described in the next
chapter, p. 426).

necessary to tabulate the average coefficients for the two sexes separately.
For women, tenderness, submissiveness, fear, curiosity and disgust have a
higher average rank than for men, while sex and anger (pugnacity) have a
lower : this again tallies with the common notion.