ESTIMATION AND TESTING OF FINER GRADATIONS 139 



other? Plainly, this is possible only when the rank- 

 orders obtained by the tests deviate from the rank- 

 order obtained by estimation in contrary directions 

 in some portions of the series. 



In illustration : if a pupil obtains Station 10 in the 

 estimated rank-order, Station 8 in the order of Test 

 A and Station 12 in the order of Test B, and if a simi- 

 lar thing occurs with other pupils, then the above- 

 mentioned differences in the correlations follow of 

 necessity. But it is equally evident that the combin- 

 ing of the two stations for tests, 8 and 12, gives the 

 so-called ' ' resulting rank-place for tests, ' ' 10, a value 

 that now coincides with the station for estimated 

 rank-order. The two tests therefore mutually com- 

 pensate one another and thus form, when combined, 

 a measure of intelligence that comes much nearer 

 the estimated intelligence than either test by itself. 

 Put psychologically, the tests demand the activity of 

 aspects of intelligence so different as to be very un- 

 equally developed in one and the same person, but 

 which, taken together, do characterize his degree of 

 intelligence. 



And, as a matter of fact, the correlation computed 

 from Bies' data did figure out so that the amalga- 

 mated rank-order for the two tests presents the ex- 

 traordinarily high correlation with estimated intelli- 

 gence of 0.98. 



In this way, then, the mutual compensation of tests 

 that we have already set forth as a requirement, be- 

 comes a controlling principle of the test-series, and 

 the correlation method gives us a numerical device 

 for discovering that combination of tests in which we 

 approach most nearly to perfect compensation. I 



