142 PSYCHOLOGICAL METHODS OF TESTING INTELLIGENCE 



schoolroom and to arrange them in order on this 

 basis. The first column of figures in Table XX shows 

 the correlations of the several tests with the esti- 

 mated rank-order. The correlations are arranged in 

 order of their magnitude and run from 0.88 to 0.51. 

 Mile. Descoeudres also calculated the amalgamated 

 rank-order for all the tests and found a correlation 

 of 0.99 between it and estimated intelligence almost 

 complete correspondence, then, between the two se- 

 ries. I have not myself checked up this value to see 

 if it is absolutely correct, but I have from the original 

 data calculated the correlation with intelligence for 

 each 5 tests, taken in combination (Column 2) and in 

 each case I found confirmation of the rule that the 

 amalgamated correlation was considerably higher 

 than the highest correlation of any single one of the 

 tests of which it was compounded. 



A few hints may be added concerning certain other 

 points to be observed in working at the problem of 

 ranking by tests. 



(a) Measur ability. It must be possible to ex- 

 press the performance in the test conveniently and 

 unequivocally by a numerical value : and these numer- 

 ical values must make sufficient differentiation with- 

 in a group that a rank-order of performance can be 

 drawn up. 



(b) Reliability. A test is reliable only when its 

 outcome is a true expression of abilities and is not too 

 much affected by variable and temporary conditions. 

 Eeliability is tested by applying the same (or an 

 analogous) test several times to the same group of 

 subjects. Only if these repeated testings show a high 

 degree of inter correlation is the test valid scientific- 

 ally. 



