138 THE POPULAR SCIENCE MONTHLY 



series of seeds which survive greater than that of the original distribu- 

 tion, while in 56 cases it is less. This is a deviation from equality of 

 17 ± 2.98, or over five times its probable error. One might have to toss 

 coins a long time to get 22 heads and 56 tails in a series of 78 throws ! 



Furthermore, the individual constants indicating a lowering of 

 variability by selection are statistically much more trustworthy than 

 those suggesting a decrease. To carry much weight a difference be- 

 tween two constants should be at least two and one half times its prob- 

 able error. But only 5 of the broken lines reach a straight edge laid 

 2.5 above (to the right of) the zero bar, while 22 extend beyond the 

 same limit on the other side. 



The coefficients of variation are already in relative terms — varia- 

 bility expressed in percentages of the mean — hence they may be repre- 

 sented directly by the direction and length of the lines in the figure to 

 the right. For this graph, one unit on the base scale means one tenth 

 of one per cent. The results amply confirm those secured by the pre- 

 ceding method. 



These graphs are deduced from experiments involving tens of 

 thousands of individually weighed seeds. Their evidence for a selec- 

 tive mortality can not, therefore, be lightly set aside. That the aver- 

 age weight remains unchanged while the variability is decreased can 

 only mean that there is an elimination from both the upper and lower 

 extremes of variation, that is, of both large and small seeds. 31 



Nevertheless, too great caution can not be used in the interpreta- 

 tion of the result. Purposely, the materials selected for study were 

 most varied, and while the validity of the general average result can 

 not be seriously questioned, there remains the problem of determining 

 whether, and if so to what extent, the selective mortality may not 

 fluctuate widely with different varieties and conditions. 



The whole problem of the underlying physical and chemical causes 

 of the differential mortality remains to be investigated. Finally, the 

 possible relationship of the selective mortality to organic evolution can 

 not be discussed until we have further evidences along several different 

 lines. . 



Potential Characteristics and Seed Mortality. — When we turn to 

 the question of the possible influence of the characteristics of the adult 

 plant innate but invisible in the seed upon the chances of survival, the 

 data are very scarce indeed. 



Montgomery's studies of the rise in productiveness in corn as a re- 

 sult of increased competition may perhaps be of interest here. The 



81 Montgomery, supra, found little difference in the capacity of large and 

 small seeds for producing mature plants in the cereals. Had he worked with the 

 whole range in size and taken into account variability as well as type he might 

 have found stronger evidence for selective mortality than he did. 



