238 



E. C. MACDOWELL AND E. M. VICARI 



Unfortunately, probable errors can not be given for the group- 

 ings in which the strains or sexes are given separately, since the 

 numbers are too small for the calculation of standard deviations; 

 but such calculations have been made for the inclusive grouping 

 of all strains and both sexes. In table 4 are given the differences 

 between the test and control averages for each day of training 

 and retention, both excluding and including the 'failure' rats. 

 The probable errors of the differences and the differences in 



TIME PER DAY 



FREQUENCY DTSTRIBUTIONSOF RATIOS 



FREQUENCIES 



I.O 1.2 1.4 



SCALE OF RATIOS 



Fig. 5 Frequency distribution of the ratios of tests vs. controls, based on 

 the averages of the time for each day with the sexes and strains treated sepa- 

 rately (ratios given in table 3); the size of the ratios is shown on the base line; 

 the vertical scale gives the numbers of ratios. The broken line gives 'minus' 

 ratios (controls taking more time); the solid line gives 'plus' ratios (tests taking 

 more time). 'Completes' and 'incompletes' are included, but not 'failures.' 



terms of their probable errors (differences divided by their 

 probable errors) indicate the statistical significance of the differ- 

 ences. The first four days of training and the last day of reten- 

 tio?i for the 'completes' and 'incompletes' show differences that 

 are three or more times their probable errors. On the seventh 

 day of training and the third day of retention the differences are 

 only slightly below three times their probable errors (2.63 and 

 2.91 times). Adding the 'failures' gives the same results, with 

 the exception that the difference on the third day of training 

 falls a little below three times its probable error (2.51 times). 



