sleet and snow. In all three series the length of the humerus is 

 longer in the birds which survive, and in the group of adult 

 males the difference is perhaps statistically significant. The 

 same is true for the length of the femur, but the results are 

 again insignificant except in the adult males where they are per- 

 haps statistically trustworthy. In the adult males and in the 

 adult and young females the length of the tibio-tarsus seems to 

 be longer in the survivors, but the result is insignificant for the 

 young males. 



If selective elimination be a reality in nature one would not 

 expect all of the characters of a series of individuals which per- 

 ished when exposed to a given set of unfavorable conditions to 

 differ from the same characters in the individuals which survive, 

 and this for the simple reason that variations in many characters 

 may not be of vital importance to the individual — in short, not 

 of selective value. 



The constants seem to me to justify no conclusion concerning 

 the length of the sternum. For alar extent all three differences 

 individually considered are insignificant; taken comparatively 



tions in the spread of wing have under the particular conditions* 

 no significance in determining the chances of survival. The 

 young males which survived have longer skulls (tip of beak to 

 the occiput) than those which perished, and the difference seems 

 to be significant in comparison with its probable error, but in the 

 other two classes of birds the differences are not merely statisti- 

 cally insignificant but negative in sign. 



Tables IV- VI show the standard deviations and their probable 

 error. These are essential in calculating the probable errors of 

 the means and in testing the hypothesis of a reduction in varia- 

 bility by selective elimination. Bumpus has discussed this ques- 

 tion in detail in his lecture, but to me it seems that the standard 

 deviations as given here do not justify any final conclusions con- 

 cerning the relation of selection to variability : the problem is too 

 complicated and the data are too few. As in other evolutionary 

 problems we need more measurements. When these are available 



