128 



CARNEGIE INSTITUTION OF WASHINGTON. 



same time, it is possible to obtain a fair opinion as to the general nature of the 

 final conclusion from these figures. The basis of the interpretation of these 

 tables must be the comparison of the test and control htters within one experi- 

 ment. In averaging different experiments together, the differing numbers of 

 individuals from different lines introduces a source of error that may seriously 

 modify the figures obtained. Such averages are only given to help obtain a 

 rough general impression of the nature of the results, but only by following 

 one set of rats through the 10 criteria of comparison for the results of the 

 multiple-choice problem and the 6 criteria for the maze results can a correct 

 impression of the meaning of these numbers be obtained. It is obvious that 

 all these criteria do not bear equal weight; 'training proper,' for instance, 

 bears much more weight than either 'test set-ups' of 'memory,' on account of 

 the larger numbers of trials involved. In the maze all criteria are based on 

 the same numbers of trials, but it is a question whether speed or the number 

 of perfect trials should have more or less weight, etc. 



Table 3. — Summary of the comparisons given in tables 1 and 2 oj the four main ex-periments. 

 [The capital letter signifies the superior litter. T =tests superior; C =controls superior.] 



"Table 3 has been made to assist in deciphering the confusing mass of com- 

 parisons in the other tables; it shows which htter in each of the 5 experiments 

 was found to be superior for each of the 16 criteria. This takes no account of 

 the numbers of individuals included, nor of the amount of difference, but 

 merely indicates the directions of the differences found. In the case of 

 experiment 20 there are only two exceptions to the superiority of the controls ; 

 it would seem safe to conclude that there was a difference between the two 

 litters in this experiment. But in experiment 19 the case is by no means so 

 clear. Does the superiority of the tests in the 'time' and 'errors' criteria of 



