ESTIMATION AND TESTING OP FINER GRADATIONS 111 



of everything that he knows about his pupils, may 

 estimate their intelligence and arrange them ac- 

 cording to his estimate (see the next section) or we 

 can apply experimental tests of intelligence, the out- 

 come of which admits of arranging the pupils in a 

 series. Kank-orders of intelligence are therefore 

 divided into orders based on estimates and orders 

 based on tests. 



In the last resort the second of these divisions 

 brings us to the question: Is it possible, on the basis 

 of a short examination with a series of tests to arrive 

 at a gradation of pupils that corresponds with their 

 actual differences of intelligence and such that the 

 rank that each gains is sufficiently characteristic <jf 

 his grade of intelligence within the group? 



It is not hard to obtain a rank-order on the basis 

 of a test or of a series of tests. To be sure, the Binet 

 tests, most of which admit of a choice between the 

 evaluations 'right' or t wrong/ are not fitted for that 

 purpose, but we can obtain a gradation in all those 

 tests that bring into operation a measurable per- 

 formance, in which what is measured is the quantity 

 of the performance in a given time or the quality of 

 the performance (as indicated by the number of er- 

 rors). Every test of this sort creates an array of 

 ranks, only it remains to discover how much the ob- 

 tained order may inform us about the intelligence of 

 the examinees. 



Hence we need here, too, a device for guaging our 

 work, and this device consists in the comparison of 

 several rank-orders obtained with the same individ- 

 uals by means of the method of correlation. 



The method of calculating correlation can not, of 



