298 THE AMERICAN NATURALIST [Vol.LVI 



2. The test rats have been classified into seven groups 

 according to the number of right-end choices in the last 

 twenty trials of their preliminary training. The first 

 points of the lines given for the tests indicate the average 

 number of right-end choices made by the rats in each of 

 the groups in the preliminary training; the following 

 points give the average numbers of correct (right-end) 

 choices made by these same rats in successive sets of 20 

 trials in the regular training. Since the procedure in 

 the regular training is essentially different from that in 

 the preliminary trials, the lines connecting the first and 

 second points are drawn as arrows. The numbers at 

 the ends of the lines give the numbers of individuals 

 included in each group. The arrangement of the controls 

 follows the same plan. Whereas the curves for the tests 

 show a general parallelism, those for the controls are, 

 with the exception of the group of four rats whose 

 preliminary training gave between 12 and 14 right-end 

 choices, relatively independent of the preliminary records. 

 This matter can be brought out more clearly by a study 

 of the coefficients of correlation between the preliminary 

 record of each rat and the trials in the regular training. 

 When the correlation coefficients between the preliminary 

 records and the first 20 trials in regular training, and 

 between the preliminary and the second twenty trials in 

 regular training, etc., are calculated, the figures in Table 

 I are obtained. In every case the differences between 

 the coefficients of the tests and controls (fourth column 

 in Table I) show that the tests have higher correlations, 

 and in all but the correlation between the preliminary 

 trials and the last set of twenty trials in regular training, 

 the differences are statistically significant. These re- 

 sults indicate that there is a real difference between the 

 tests and controls in the way they react to the necessity 

 of using trial and error methods ; this may be due to a 

 difference in responsiveness to changes in the situation. 

 The tests appear to be less responsive to the changed 

 procedure, since they continue the same general behavior 



