44 THE FACTORS OF THE MIND

even when collected with, little or no intelligible plan, if
only he can apply some fool-proof criterion. On the
contrary, as I tried to point out in an early paper, it is
requisite in all such researches to see that the tests or traits,
as well as the persons tested or observed, constitute a fair
and systematically selected sample : according to the
nature of the investigation, it was argued, they should either
be " chosen so as to represent, so far as possible, all the
typical aspects or levels of the mind" ; or else (this
holds most frequently of the persons, but may also hold of
the traits) they should be expressly selected so as to form
what is called a ' random' sample.1 The two principles
are really the same : for ' random ? selection does not mean
blind or careless selection, but " selection according to some
precept or method which ensures that the mode of choice
shall be irrelevant to the probability of the generalization
to be established." And " our knowledge of this irrelevance
is prior to the empirical establishment of the generalization/3
and therefore a priori, as the phrase has been used here
(cf. [43], pp. 41 f, 281 f.).2

(B) The Weakness of the Factorial Link.—My second
point, it may be remembered, was that the introduction of
factors, deduced from correlations, does not of itself (as is

1 In my first investigation on c Experimental Tests of General Intelli-
gence ' I sought to lay down the former principle as essential to researches
based on ordinary experimental tests ([16], p. 98): the latter principle is
often more appropriate in researches based on observations or impressions,
since these may be more numerous than tests,

2 The special problems of sampling tests or traits are not unlike those that
have arisen more recently in connexion with sampling persons or populations.
Most factorists appear to assume that in both cases the method of random
sampling is the only available or legitimate procedure; and by random
sampling they understand the method of * simple sampling' as described
by Yule and others ([no], p. 350] rather than the method of systematic ran-
domization as described by Fisher ([109], p. 20 f,). This view is taken, for
example, by Stephenson in a recent paper ([136], p. 20). As I have indicated
elsewhere, in studying the existence of mental or socialt types' or tendencies
among the general public, some method of representative or stratified
sampling has often to be substituted for the simple effort to sample the whole
population at random (e.g. in surveys of school populations). Similarly, in
selecting tests or traits, the method of simple random sampling must in
general give way to some more elaborate method of systematic sampling by
strata, levels, or other representative scheme, in which, no doubt, randomiza-