no. 2.] PSYCHOLOGICAL EXAMINING IN THE UNITED STATES ARMY. 437 



Although there is no definite information that the ratings of all men in a company in Camp 

 Meade were rated hy the same officer, the heterogeneity of ratings will certainly be considerably 

 reduced for any particular group if that group is made up exclusively of men in one company. 

 To carry out further the investigations of the effect of combining material more or less hetero- 

 geneous as to ratings, correlations between officers' ratings and separate tests of alpha have 

 been calculated for the following groups: Company B, 149 cases; Company F, 101 cases; 

 alpha before beta group — omitting all Company B men — 74 in Company H, 23 in Company K, 

 and 3 in company not specified. 



Table 100 — Correlations between officers' ratings and separate tests of alpha. 



Adjustments of these coefficients to a common basis with Company F as the standard 

 gives the following results, the correlations for the Company F group being repeated for the 

 sake of comparison: 



These results, together with the biserial correlation coefficients for officer 3, MacArthur 

 Infantry Replacement Camp, show a considerable tendency toward variation of coefficients for 

 different sets of ratings. Not only does the rank order of tests change from group to group, but 

 the range of variation of the eight coefficients fluctuates noticeably. All the coefficients for 

 the Company B group above lie between 0.621 and 0.516, a range of 0.105. The coefficients 

 for the Company F group vary from 0.495 to 0.704, a range of 0.209. Exact calculations of 

 the appropriate probable errors can not be made, but if we consider the probable errors of the 

 adjusted coefficients given above to be the same as for the unadjusted coefficients, the probable 

 error of the difference between the largest and smallest coefficients obtained from Company B 

 group, for tests 4 and 7, respectively, wall be of the order of magnitude of 0.048, or somewhat less 

 than half the actual difference. This probable error, it may be pointed out, involves not only 

 the standard deviations of sampling for the two coefficients compared, but the correlation 



