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



465 



tests and ability as noncommissioned officers, but due to a largely fortuitous combination of 

 difference in general levels of ability of these two groups with the qualitative differences between 

 the two groups — a combination in which tests 3 and 8 are the points of contact. 



We may summarize our results for the Travis data: (1) Artillery organizations as a whole 

 seem to constitute a moderately highly selected group as compared with the Infantry organiza- 

 tions. Certain phases of value to the service, correlated differently with different alpha tests, 

 have been important factors in the primary selection, or more accurately, elimination. The 

 tests apparently most highly correlated with the primary selection of artillerymen are 1 and 2. 

 (2) The selection with which we are immediately concerned is superimposed upon this basic 

 selection and this fact apparently introduces new factors, especially those factors correlated 

 with tests 6, 7, and 8. (3) In the Infantry there is not much indication of primary selection 

 of a type that can be distinguished from the secondary selection. The diagnostic values of 

 tests in the secondary selection are probably somewhat obscured in the comparison of low- 

 grade groups by the shortness of range of the tests; but tests 1, 2, and 3 apparently con-elate 

 most highly with the factors of the secondary selection. (4) Generally speaking tests 1,2, and 

 3 appear to show sharper discrirmnation than other tests. (5) Test 5 almost invariably has 

 the lowest rank. 



The hypothesis outlined above is not presented as a definite conclusion established by 

 statistical analysis of the data, but is a frank attempt to fit the large number of rather dis- 

 cordant results into some kind of a system. The attempt has not been very successful, and the 

 one safe conclusion that can be drawn is that the standards of military value according to which 

 the several groups have been selected are so variable that an attempt to estimate the relative values 

 of the alpha tests must necessarily fail until such variability of standards is adequately controlled. 

 Since the alpha tests have rather high intercorrelations it is evident that there can be no very 

 great differences in the way in which they differentiate individuals. 



The attempt to provide a statistical criterion of reliability has evidently been beside the 

 point since this criterion of reliability takes account only of the individuals in the selected 

 groups for constant selective factors and does not give a measure of the variability of the selective 

 agents. As already pointed out, the selective agents have been so few in number that their 

 resultant is subject to a large variability. 



If we take into account the results of comparing the three groups from Camp Kearny, we 

 merely add more variation to our results. 



The results obtained from the Camp Meade data, which required treatment by methods 

 different from those employed with the Travis and Kearny data, increase the chaos. This 

 case, however, is the only one in which there is a classification of the individuals of a complete 

 (though selected) group (if we neglect the classification by alpha and beta examinations). 

 In the other cases we had a selection of a special subgroup to the total neglect of all individuals 

 in the other subgroups. It is thus theoretically possible to measure the degree of correlation 

 between the standard of classification and the scores in separate tests or in combinations of 

 them. Practically, however, such work is unsatisfactory, since we have to break up into two 

 parts (alpha-beta segregation) the group that is complete from the point of view of army 

 selection (i. e., of permanent assignment to military units). Treatment of the two fragments 

 separately is bound to lead to very small correlation coefficients. The complete group of 773 

 cases is divided between the alpha and beta examinations in the proportion of 362 to 411. If 

 we consider the relation of the alpha-beta segregation (Camp Meade never gave both examina- 

 tions to the same man) and military value we have the following distribution: 



We can calculate a biserial correlation ratio and its value proves 0.3426. 



