136 



SCIENCE 



[N. S. Vol. XXXVII. No. 943 



of the "Goods" are shown similarly in 

 Table I. 



TABLE I 



Percentages of the "Poor" Group Reaching Cer- 

 tain Points of the "Good" Group's Ability 



It is obvioits that the opposite tast is 

 highly symptomatic of something which 

 differentiates these two groups ; that the 

 tests in discrimination of length are not; 

 and that the four tests are — in order of 

 anerit as symptoms — in the order named, 

 ■giving opposites, memorizing words, mark- 

 ing A's and comparing lengths. 



When a series of groups, each diiJering 

 slightly from the next in order in the 

 amount of the T ov 8 taken as the basis of 

 comparison, are used, the method of testing 

 a symptom's significance is by its correla- 

 tions — by the closeness of the correspond- 

 «nee of the rankings of the individuals in 

 fthe S"s with their rankings in the T. Thus 

 twelve of the "Good" group in the preced- 

 ing illustration were ranked, each by alP 

 the rest (they being somewhat acquainted 

 with the others' abilities), for general in- 

 tellectual ability. It is then possible, by 

 means which I will not rehearse here, to 

 measure the closeness of correlation or 

 oeorrespondence between the measure for 

 each man that would have been so obtained 

 had this "imputed intelligence" been de- 

 termined from 12,000, instead of 12, judges 

 of each man, on the one hand, and the 

 measure obtained for him by repeated test- 

 ing with any of the tests. These corre- 

 spondences as calculated for the four tests 



' Average of three series. 



-" Or nearly all. 



mentioned were: 96 hundredths of perfect 

 correspondence in the case of the opposites 

 test, 93 hundredths in the case of the mem- 

 ory of words, 21 hundredths in the case of 

 the A test, and apparently zero in the case 

 of discrimination of length. 



This method allows a very delicate choice 

 amongst tests, provided adequate data are 

 at hand. 



Suppose, for example, that arithmetical 

 ability is defined as the ability required by 

 a given selected set of a hundred arith- 

 metical tasks or problems. We can find 

 which ten of these will serve best as a test 

 of the ability measured by the entire hun- 

 dred, by measuring a suitable group in re- 

 spect to the entire hundred, and choosing 

 that set of ten whose combined score corre- 

 lates most closely with the combined score 

 for the hundred (the pairing being by in- 

 dividual pupils). This method not only 

 demands much time in experimentation 

 and calculation, but also a thorough under- 

 standing of the general logic and technique 

 of measuring relations, including the pe- 

 culiar relation of resemblances or mutual 

 implication. For example, suppose that 

 twenty children, who had in a given year 

 been absent, respectively, 100 days, 95 days, 

 90 days, etc., down to 5 days, attained, re- 

 spectively, average academic grades of 74, 

 75, 77, 111, 77i, 78, 78^, 784, 79, 79i, 79|, 

 80, 80i-, 80f, 81, 81i, 81f, 82, 83 and 85. 

 The correlation by any of the stock methods 

 comes out as 1.00, and the investigator 

 might fancy that he had proved that for 

 the school in question the mark to be re- 

 ceived could be perfectly prophesied from, 

 and was determined by the same causes as, 

 the attendance record. Nothing of the 

 sort, I assure you, is proved by this perfect 

 correlation. It would perhaps puzzle some 

 of us to tell just why. 



The method of observing the correspond- 

 ence of an individual's status in the trait 



