422 EDWIN CARLETON MacDOWELL 



of data that would otherwise be difficult, if not impossible, to 

 summarize. The relationship of every fly to its parents has its 

 effect on the results. However, correlation tables afford only 

 superficial descriptions, and accordingly great care is necessary 

 in determining their meaning. A positive correlation coefficient 

 indicates that higher parents had higher-grade offspring; a 

 negative coefficient indicates that higher-grade parents had 

 lower-grade offspring. But from any single coefficient little 

 can be said of the underlying genetic significance. In any par- 

 ticular generation the high-grade parents mia^y have had, on the 

 whole, better conditions than the low-grade parents and conse- 

 quently produce higher-grade offspring; this would give a posi- 

 tive coefficient. The reverse might be equally possible in an- 

 other generation; the lower-grade parents might, by chance, 

 have found better conditions and so have produced higher- 

 grade offspring than the high parents; in such a case the correla- 

 tion would be negative. Although a single plus or minus coeffi- 

 cient would not bear much evidence, a series of one or the other 

 would indicate that real genetic phenomena were involved. 

 Even such a series cannot be authoritative unless the experi- 

 mental procedure has been the same for all families included. 

 This is one of the greatest difficulties with results based alone 

 on mathematical treatment. In many such cases the material 

 has lacked homogeneity. In this respect the bristle data have 

 one great advantage, being derived entirely from experimental 

 procedure ; their origin and the nature of the families put together 

 are fully known. 



The selection of low-grade flies tends to isolate the small 

 ones and therefore to carry on any conditions that tend to make 

 small flies, such as weakness or disease. Such weakness, apart 

 from any germinal cause, would tend to make the offspring of 

 these low-selected parents lower than the offspring of parents 

 selected for grades not associated with weakness. If such 

 differently selected lines be united in one correlation table, 

 even though they are all the result of brother and sister mat- 

 ings and are all the same number of generations away from 

 common ancestors, the finding of a positive coefficient would 



