392 



MEMOIRS NATIONAL ACADEMY OF SCIENCES. 



[Vol. XV, 



does not invalidate the comparison of correlations.) The correlation of alpha raw score with the 

 proposed beta raw score was found to be 0.811 ±0.009, as compared with 0.806 ±0.009 for alpha 

 weighted with beta weighted, again an inappreciable difference. The correlation array is shown 

 in Table 56. It will be noted that this has the characteristics previously noted regarding the 

 correlation of alpha weighted score with beta weighted (seep. 385). Because of the curvilinear 

 relationship the correlation ratios were computed and found to be 0.847 ±0.007 and O.S40± 

 0.008. 



Table 56.— Showing correlation of beta raw score (suggested form) with alpha raw score — English-speaking whites from 



nine camps. 



r-0.811±0.709 ,(xy)=0.S4O±0.O08 

 ,(yx)=0.S47 ±0.007 

 Beta raw (suggested): Mean=62.S cr=25.7 

 Alpha raw: Mean=58.3 a=42A 



As a final basis of comparison the effect of the change in beta on the non-English-speaking 

 group was considered. For this group of 95 cases the correlation with Stanford was raised 

 by the change; correlation of Stanford with beta weighted score was 0.680 ±0.037, and for 

 Stanford with beta (suggested form) 0.760 ±0.030. 



The results of the attempt to find a way of improving the correlation between Stanford and 

 beta, or alpha and beta, by statistical manipulation of the tests may be summarized as follows : 

 (1) It was concluded that no appreciable improvement hi correlation could be secured by this 

 method. (2) Considerable reduction in the number of tests used would be possible, but seemed 

 precarious on the basis of present data alone. (3) Raw scores could be used without reducing 

 the amount of the correlation appreciably. (4) Elimination of test 8 (spot pattern) could also 

 be accomplished with no significant reduction in the correlation. Since this is the last test 

 on the blank the other tests would not be affected by its removal. (5) Reduction in numerical 

 value of test 1 (maze) by changing the scoring rules to make each half maze count one-half 

 instead of one, seemed desirable, since results of the partial correlations indicate that test 1 

 contributes least of all the tests to the correlation. (6) The suggested form of beta showed 

 no appreciable reduction in amount of correlation with Stanford, whether we consider the total 

 English-speaking group, the groups from the various camps, or the total group from non-English 

 speaking countries. It also showed no reduction in its correlation with alpha for the total Eng- 

 lish-speaking group. 



This change was therefore recommended, which has the following as its main advantages: 

 (a) It reduces the time required for beta by about ten minutes. (6) By eliminating weighting 

 of tests it saves the time of the scorers, and also reduces the chance of error which weighting 

 introduces. The recommended change was adopted and incorporated in the final edition of the 



