no. 2.] PSYCHOLOGICAL EXAMINING IN THE UNITED STATES ARMY. 457 



or more indicate an operation, in addition to the purely chance variations of sampling in choosing 

 the groups, of selective factors that are correlated to some extent with the type of selection 

 achieved in the alpha and heta tests. We said above that if two tests were highly correlated 

 then a change ha the difference between two sample groups would not tend to produce a change 

 in the difference of x 2 - We are now reversing the argument. If in two cases we get very large 

 differences in x 2 (s a y» greater than 10) we conclude that the great size of this difference indi- 

 cates that a variation from one sample to the other is not merely a chance variation, but 

 includes an actual difference in capacities specifically measured by the tests and not common 

 to both (i. e., those capacities not tending to produce correlation). 



Each "best private," "poorest private," and noncommissioned officer group in the Camp 

 Travis Infantry and Artillery regiments, as well as the "superior" and " inferior " groups from 

 Camp Kearny, has been compared with the standard group, and several intercomparisons of 

 these groups have been made (Table 127). 



The Camp Meade data are not of such a kind that the use of the product-moment method 

 of correlation is justifiable, but two different biserial correlation coefficients have been calcu- 

 lated as follows: (1) By treating the combined "military value" classes a and b (the two 

 highest classes) as the subgroup of the total sample, and (2) by treating similarly "military 

 value" classes d and e (the two lowest classes) as the subgroup. These two sets of coefficients 

 are indicated in Table 126 by the symbols ab/cde and abc/de, respectively. The correlation 

 ratios of score on military value, mT rj sc , and mditary value on score, scimv, are also given in this table. 

 Finally, correlation with military value of total score on various combinations of four alpha 

 tests have been calculated, as indicated also in Table 126. 



Table 1213. — Comparison of alpha tests. 



[Coefficients of relationship between performance in each test and in certain combinations of four tests and officers' estimates of military value. 

 Camp Meade experiment. The biserial coefficients of correlation are calculated with the two highest officers' ratings, a and b, as a subgroup 

 (column headed ab/cde), and also with the two lowest ratings, d and e, as a subgroup (column headed abc/de). The correlation ratios are 

 given for test score on military value, m »»?io and for military value on test score, M vmv] 



Table 127. — Comparison of alpha tests. 



[Degree in which the pairs of groups (at left of table) are differentiated by performance in every test. The first row of figures for every pair of 

 groups gives the difference between the means of the performances of the two groups in every test. The second row gives the ratio of this 

 dillerence to the standard deviation of the dilference. The third row (figures in parentheses) gives the rank order of every test with respect 

 to its effectiveness in differentiation as measured by this ratio. Data are from Camps Custer, Kearny, and Travis.] 



121435°— 21- 



-30 



