282 THE FACTORS OF THE MIND

then follow King David, and define what we are to regard
as a typical or representative sample. Variance also
depends on what tests we choose : let us then lay down
certain types of test as standards, specifying if necessary
place, time, material, and the like, just as we do in defining
weight or length. Sooner or later, no doubt, a regulation
from the Board or Education will provide an official
definition of the mental year, just as it once defined educa-
tional attainments in terms of ' standards,' and just as an
Act of Parliament has defined the meaning of the yard and
pound*1

1 More recently Stephenson has argued that " Burt's reciprocity proce-
dure " (using covariances between persons to obtain the same factors as were
formally obtained by correlating tests or traits) "is a notable advance in
technique but is insecurely founded, because he has as yet put forward no
generally acceptable system of units, which is surely of first importance in any
science " ([137], p. 95). This seems to overlook the fact that from 1913 to
1927 the majority of my investigations were bound up with an attempt to
u put forward a generally acceptable system of units " in terms of which
intellectual, emotional, and (more particularly) educational differences could
be measured. Various possibilities were proposed, since for different pur-
poses different units seemed to be required. For example, with school
children it was suggested that, for rough practical purposes, the notion of the
mental year as unit could be extended from intelligence to educational and
emotional characteristics: such a method at once indicates wide differences
in the variance of different characteristics, usually corresponding with the
complexity of each (cf. [35], p. 25, [41], pp. 258 et seq.}. For more exact
measurements of general ability, the idea of ' internally graded tests' was
advocated. Unlike the Binet scale, each sub-test was to consist of a graded
series of homogeneous items increasing in difficulty or equally spaced, the
equality of the unit-intervals being determined either by the number of
children passing the items or by the number of persons judging the intervals
between the items to be equal or different: the unit-intervals in the different
sub-tests were then to be related by means of regression coefficients. Such a
method of construction was slow and cumbersome, but seemed essential for
exact theoretical work ([27], pp. 46, 151, and later L.C.C. Reports, e.g.
[41], p. 138). For more general purposes still a * standard scale,5 based on the
normal frequency curve, was drawn up ([35], p. 49) and widely used : it has,
as a matter of fact, been adopted by Stephenson himself in some of his
earlier experiments on correkting persons. With this procedure, it may be
noted, if elementary aspects of a complex process are marked in standard
measure and then summed (according to the * analytic method of marking'),
the total process will be automatically weighted in accordance with its
assumed complexity. The last of my books opens with an entire chapter
devoted to the { Choice of Units * in mental measurement (Backward Child,