THE STUDY OF NATURAL SELECTION 



i37 



whole 78 series the mean difference is less than five tenths of one 

 per cent. 30 



Thus if we confine our attention to the mean, there is apparently no 

 selective elimination whatever, for within the limits of experimental 

 error there is no certain change in the mean value of the character con- 

 sidered. But an entirely new and different light is thrown upon the 

 whole question when the variabilities are examined. These are dis- 

 tinctly less on the average for the series which develop to maturity. 



This is brought out with great clearness by Diagram 3. 



Diagram 3. Differences in Variabilities of General Population of Seeds 

 and of those which Produced Plants. The figure to the left shows the ratio of 

 differences in standard deviation to their probable errors (1 space on scale = .2). 

 The figure to the right shows differences in coefficients of variation (1 space on scale 

 = .11 per cent.). The vertical lines give the points of zero difference. 



In general form these figures are similar to the one representing the 

 means, but an additional point is brought out by the one for the stand- 

 ard deviations, to the left. In this case the length of the bars indi- 

 cates not absolute nor relative values of the variabilities, but the trust- 

 worthiness of the constants. Here the base scale is in terms of the 

 ratio 



Difference 



Probable Error of Difference 



each unit being equivalent to .20. 



Instead of the light and dark line areas being approximately equal 

 they are widely different. In only 22 cases is the variability of the 



80 The mean difference in weight is more nearly zero in the 28 field experi- 

 ments than in the 50 made in the greenhouse. There may be valid biological 

 reasons for this, but they can not be discussed here. 



VOL. LXXXIV. — 10. 



