EMPLOYMENT PSYCHOLOGY
on the reliability of the measure and on the validity with
which it predicts the criterion.
Tests may be weighted roughly in accordance with their
reliability. The reliability of a test can best be calculated
by giving it a second time and correlating the two series of
scores. The reliability coefficient, however expressed, may
then be used in giving relative weight to the test. Measures
of reliability have been discussed in Chapter XIII.
A rough system of weighting, when each test has been
correlated once with the criterion of success, is to assign
weights to the various test scores in accordance with the
size of the correlation of the particular test with the cri-
terion. Test scores must previously have been equalized by
weighting to overcome the weakness pointed out in the first
method in this section.
Scores in a test may be weighted in accordance with the
regression equation of the criterion on that test. If raw
scores are used, their values for the test should be multiplied
by the fraction Tule and then added to the additive con-
stant. In this fraction y refers to the criterion and x to the
test. A person’s total score will be the summation of his
scores in the separate tests after each has been weighted in
this manner. If measures are in terms of deviations from
the mean, there is no additive constant.
Another method is to weight scores in a test in inverse
proportion to the standard error of estimate of the criterion
predicted from scores in that test (see pages 199-200).
The most accurate statistical device for combining scores
in several tests is the multiple regression equation.
The multiple correlation coefficient, represented by
Ri(s...n), is the correlation between the criterion, 1, and
the best weighted combination of the various tests, 23. . .x.
The multiple correlation coefficient may be obtained by the
formula given by Yule (233, formula 15, p. 248). The
formula for three variables is as follows:
a Ri (a3) =V/ 1—(1—7%)(1—1%,)
208
43)
