268 E. C. MACDOWELL AND E. M. VICAEI 



5. Comparison of the tests and controls on the basis of the numbers 



of errors 



a. All types of errors together. For the present purposes an 

 error has been defined as a departure that extends more than 

 half an inch from the true path as measured on the record sheets 

 (approximately the length of a rat on the maze itself); it maj^ 

 involve passing a door it should enter, it may involve turning 

 back in the true path. All types of errors, regardless of their 

 length have been put together. Table 16 presents the error 

 data summarized in the same way as the distance data in table 

 7; the numbers in the body of the table are averages of each 

 rat's average per trial for the period and group indicated. 



The table gives sixty-five 'plus' ratios and seven 'minus' 

 ratios (six of which are due to the females in strain A) ; in sixty- 

 five pairs of averages the tests have more errors than the con- 

 trols, while in seven the controls have more errors than the 

 tests. Table 17 gives the differences between the averages and 

 the probable errors of the differences when all strains and the 

 sexes are put together. First half of training, omittiyig the first 

 day, all training, and training and retention give differences that 

 are over three times their probable errors. For the second half 

 of training the difference is only a little below three times its 

 probable error (2.52 times). Figure 14 shows the distributions 

 of the individual averages in the different groups of trials. 



These results are in accord with all that has gone before and 

 they follow those for distance so closely, quantitatively as well 

 as qualitatively, that the comparisons based on each day by 

 itself may be assumed to give the same results. Although these 

 averages have been calculated, they are not presented; we can 

 state that the supposition that they would give the same results 

 is borne out by the averages themselves. It has not seemed 

 necessary to even make the calculations of the averages of all 

 rats in each group for each of the thirty-six trials by itself. 



The variability of the tests, judged by the standard deviations 

 of the average number of errors for each rat, is significantly 

 greater than that of the controls in the first half of traiimig 



