390 



MEMOIRS NATIONAL ACADEMY OF SCIENCES. 



[Vol. XV, 



Table 53. — Correlations of beta tests with Stan/ord-Binet mertal age and with each other (Pearson product-moment 



coefficients). 



As furnishing a basis of comparison it should be noted that Rs/beta weighted = 0.731 ± 0.012, and 

 that rs/beta ra w = 0.728 ±0.012. (The latter figures refer to raw score for the 8 tests, not to 

 the later suggested form.) 



The net result of the above considerations was to indicate two possible lines of development 

 with a view to recommendations. The outstanding fact being that no appreciable improvement 

 in the correlation between Stanford and beta would be achieved by any simple reweighting of 

 tests, the following possibilities were considered: 



1. The series might be reduced to the smallest number of tests from which the maximum 

 correlation could be secured ; the tests which contributed least might be eliminated and proper 

 weights assigned to the remaining tests. 



2. Since even the best weighting of tests affects the final correlation only slightly, all 

 weighting might be eliminated. With elimination of weighting it would not be safe to assume, 

 however, that the same amount of reduction in the number of tests could be accomplished with- 

 out lowering of the correlation. A certain amount of reduction might be attempted empirically, 

 following the indications of the partial correlation coefficient with regard to value for the series. 



It was decided to follow the second line of development mentioned — namely, to try the 

 effect of use of raw scores, with elimination of tests as suggested by the results of the partial cor- 

 relations. The following variations were tried empirically, and results obtained as given below, 

 raw scores being used in every case: (1) Elimination of test 8 (spot pattern); rs/beta =0.726 

 ±0.012. (2) Elimination of tests 8 and 2 (cube analysis), rs/beta = 0.723 ±0.013. (3) Elimi- 

 nation of tests 8, 2, and 1 (maze), rs/beta = 0.723 ±0.013. 



These results indicate that dropping of tests 8, 2, and 1 could probably be accomplished 

 without appreciable reduction hi value of examination beta. Inspection of scatter tables indi- 

 cates further that the lower portion of the scale is not affected adversely by the change. There 

 remains, however, one objection to the change which seemed important enough to call for a more 

 conservative recommendation — namely, the fact that our evidence is not certain proof that the 

 correlations given above would have been obtained had the tests which appear early in beta been 

 omitted. It is possible that they contribute sufficiently to the "warming-up process" so that 

 they thereby increase the value of later tests. This argument obviously can not be applied in 

 connection with 8 (spot pattern) since this is the last test on the blank. It seemed, therefore, 

 entirely safe to recommend dropping this test. 



Though we were not prepared to urge dropping of the other two tests, the evidence indi- 

 cated that the numerical value of the maze (1) should be reduced. Since this could be accom- 

 plished quite simply, without involving the necessity of "weighting," by changing the scoring 

 rules to read that one-half point should be given for each half maze correct, instead of 1 point, 

 it seemed desirable to determine the effect of this change in conjunction with the dropping of test 

 8. This was, therefore, attempted and correlation with Stanford computed for the total English- 

 speaking group with the following results: 



rs/betn miejested form =0.727 ±0.012. 



