: EMPLOYMENT PSYCHOLOGY 
highly with the tests. It is not necessary for the tests to 
predict exactly what a man’s standing in the measure of 
vocational accomplishment will be; it is only necessary to 
know whether the tests predict accurately whether he will 
be a success or a failure. If the measure of vocational suc- 
cess is continuous, it remains to find where the division line 
shall be placed for success and failure in order to obtain the 
highest correlation with the tests. By calculating biserial 7 
with the numbers in the alternative categories varying (that 
is, by calling the successes the upper 10%, 20%, or 30% in 
the criterion of success), the division point may be found 
which yields the highest correlation with the test, and the 
criterion of success may then be fixed at that point. On the 
other hand, the alternative categories may be set with regard 
to test scores, the criterion being continuous, and the opti- 
mum critical score obtained by finding the dividing line the 
use of which yields the highest correlation with the criterion. 
The results will be in terms of “per cents” and will have to 
be transmuted into scores. This method of determining 
critical scores has little advantage over the method of in- 
spection. 
Ruml (160, p. 59) gives elsewhere the formula for the 
rank-tangential coefficient (¢) which is equivalent to biserial 
7 applied to ranked data. 
The correlation ratio (7) is a truer measure than 7 of the 
relationship between two variables if this relationship is 
curvilinear. The formula is 
em) a 
(23) pm Ve 
where 7. is the number of measures in a column, ¥. the arith- 
metic mean of the column, ¥ the arithmetic mean of all the 
’s in the table, o, the standard deviation of all the ¥’s in 
the table, and NV the total number of measures. The correla- 
tion ratio is greater than the correlation coefficient where the 
regression is non-rectilinear, but in rectilinear regression the 
i84