438 



EDWIN CARLETON MacDOWELL 



in figure 3 and table 3, groups 14 to 20 show considerably re- 

 duced correlation. It seems probable that if the environment 

 had been ruled out still further by making all the matings on 

 the same day, as was done for all the generations beginning 

 with the 36th, there would be found no more correlation in this 

 period than in the period beginning with generation 36. 



Although lumping data often conceals much that is impor- 

 tant, it also tends to balance out fluctuating differences. After 

 making the above study of individual generations in a some- 

 what detailed manner, the results of correlating the genera- 

 tions in groups of five may be presented. All the same data 



TABLE 4 



Correlation coefficients, parents by sons and daughters; generations put together in 

 groups of five. The same primary data as used in table 1 



used for the correlations for single generations were used. The 

 constants are given in table 4. 



This grouping gives much larger numbers and accordingly 

 smaller probable errors; it greatly increases the variations in 

 the environment involved in each coefficient, so there is more 

 chance of their being balanced out than when shorter periods 

 were included. The regression straight lines and empirical 

 means in figure 4 and table 5 indicate that the correlation co- 

 efficients for these groups give a somewhat more accurate 

 description of the general relationship of parents and children; 

 the means lie more closely along the regression lines as would 



