PREDICTION BY COMBINED SCORES
constant, C, for # variables is as follows (84, p. 14):
(52) C=M—bpss.. ,My—b304... 4 M3— ey. —bin.23eecn—1y Mp
If the measures are expressed as deviations from the
means of their distributions, the formulas are the same,
except that small x’s are used and the additive constant, C,
is omitted.
The formula for the probable error of a multiple correla-
tion coefficient is the same as for a zero order coefficient:
1—R%03...n
(53) PE rn. =06745— 52
The formula for the standard error of estimate for the
criterion predicted from the scores in tests, weighted in
accordance with the multiple regression equation, is, for
three variables:
(54) oL=0V1—1% Vi-r%,
and for » variables:
(55)
01.23.00 =I VI—7 VI—123, V I—740 + + VI — 7212.93. + .(n—1)
For further discussions of the multiple correlation coeffi-
cient and the multiple regression equation, see Kelley (834),
Yule (233), Rosenow (151), Symonds (178), Otis (123),
Huffaker (77), and Garfiel (60a). (See Figures 28 and 29.)
For combining scores in two tests, Thurstone’s method of
scoring tests by weighting right and wrong answers may be
used. R and W are taken to represent scores in the two
tests which are to be combined so as to obtain the highest
correlation with the criterion (see page 187).
Chapman (26) describes a procedure which may be fol-
lowed for “the investigation of the effect of adding a single
test to a battery already weighted as though complete in
itself.”
OTHER METHODS
The profile is not, strictly speaking, a method of obtain-
ing a total score, but presents graphically an individuals
211