THE CRITERION OF LEARNING 67 



trials preceding the first which was made without error. This 

 comparison is best made by correlating the number of trials 

 preceding the first errorless run with the number preceding 

 ' ' perfect learning ' ' for all the animals. The former varied from 10 

 to 75 with a mean at 23.8±.977, the latter from 10 to 150 with 

 the mean at 47. 3± 2.99; the correlation in the variations of the 

 two is 0.632±0.061. The coefficient of regression of the varia- 

 tions in trials preceding the first errorless run over those pre- 

 ceding " perfect learning " is 1.304, that of variations in " per- 

 fect learning " over first trial is .306. This means that if we 

 are dealing with fairly large numbers of animals and have found 

 a given difference between two groups, as measured by the aver- 

 age number of trials required to make one perfect run, we may 

 expect that the difference in the number of trials required for 

 " perfect learning " will be in the same direction and 1.304 times 

 as great. Conversely, if we know the difference in trials re- 

 quired for " perfect learning " we may predict a difference .306 

 times as great in the number of trials required for one error- 

 less run. 



It follows from this correlation that that group of animals 

 which has made the most rapid progress up to the time when 

 the first errorless run is made will continue in the lead until the 

 limits of training are reached; will, indeed, increase that lead. 

 As a test of the application of this principle, the groups of ani- 

 mals which were treated differentially in the experiments have 

 been graded in the order of the average number of trials re- 

 quired by them to attain to each of the two standards. The 

 results of this are shown in table 1. The different methods of 

 rating result in an interchange in the order of some of the groups 

 but in no case is the position of any one group changed by more 

 than one place. 



The groups included in the table are not all strictly compar- 

 able. The methods of training were the same in every case 

 but some of the groups differed in the heredity and age of their 

 members, in the season during which they were trained, as well 

 as in certain drugs administration during training. In the sepa- 

 rate experiments, all these factors were controlled and the groups 

 a, J, g, and i, c, and d and b, c, h, and j are mutually comparable 

 and differ only in the drugs administered. The order of these 

 by the two criteria of learning is — 



