52 F. B. SUMNER AND R. R. HVESTIS. 



specific cross. From these facts, the inference seemed reasonable 

 that the increased variability of the second hybrid generation was 

 not in itself trustworthy evidence of mendelian segregation. 



From a study of three series of hybrids, derived from two dif- 

 ferent subspecific crosses, the senior author was later confirmed in 

 the belief that there was little if any proportionality between the 

 increased " spread "of the F., generation and the degree of differ- 

 ence between the parent races in respect to the character in ques- 

 tion. Cases were pointed out in which a considerable difference 

 between the parent races was associated with little or no increase 

 in hybrid variability, while in other cases, in which the parents 

 did not appreciably differ, the appearance of " segregation " was 

 very marked. 



For the present series of hybrids and the small number of char- 

 acters here dealt with the picture is decidedly different. There is 

 a considerable degree of proportionality between the amount of 

 difference between the parent races and the increased variability 

 of the F 2 generation. In order to test this point more precisely, 

 a correlation table (which it is not necessary to reproduce here) 

 was drawn up. The " characters " under consideration were the 

 twelve which have been dealt with in the present paper. In one 

 column were placed, not the absolute differences between the 

 means of the parental races, in respect to the various characters, 

 but the differences expressed in terms of the variability of these 

 characters. 1 In the other column were placed the differences be- 

 tween the standard deviations of the F x and F 2 generations, taken 

 as percentages of increase. 2 Thus the vast differences in the mag- 

 nitude of the original values (due to differences in the units em- 

 ployed) were largely eliminated. 



A graphic presentation of these data revealed the existence of a 

 considerable degree of proportionality between the two series of 

 figures, the only serious discrepancy relating to the behavior of the 

 '' B " hairs (see above). The extent of the correlation between 

 the two series is indicated by a coefficient of -f- .73. We readily 

 admit, however, the inconclusiveness of a correlation coefficient 



1 Each difference between the parental means was divided by the mean of the 

 parental standard deviations for the character in question. 



- I.e. each difference was divided by the mean of these two standard deviations 



