514 



THE AMERICAN NATURALIST [Vol.XLVI 



suggested above, it seemed a priori improbable that a 

 selective mortality could be detected in the large non- 

 selective death rate of field cultures. But from an exam- 

 ination of the data which have accumulated during the 

 last several years, it appears that the a priori scepticism 

 which led to omitting records of the characteristics of the 

 seeds failing to develop was unjustified. There appears 

 to be, in fact, a selective mortality deteetible by proper 

 methods, even in field cultures. 



The evidence upon which this statement is based is of 

 the following kind. We know (a) the weight of all the 

 seeds weighed in any year and (b) the weights of the 

 sub-sample of seeds which developed into mature fertile 

 plants in any subsequent year. Now the sample or 

 samples which were planted were purely random draw- 

 ings from the grand population forming the entire mass 

 of seeds weighed for any culture. The physical con- 

 stants for the distribution of weights of these sub-samples 

 planted are, therefore, identical with those of the grand 

 population, plus or minus the errors of random sampling. 

 Ideally, to determine whether there be a differential mor- 

 tality, we should compare the physical constants (means, 

 standard deviations, and coefficients of variation) of the 

 seeds which produced mature plants either with the con- 

 stants of those of the same sub- samples which failed to 

 do so or with the constants for the entire sub-sample 

 planted. Practically, our end can be attained with rea- 

 sonable exactness by comparing the constants of seeds 

 which develop to maturity with those of the general popu- 

 lation from which the plantings were drawn. The only 

 objections to this procedure are two: (a) the results will 

 be vitiated if the series of seeds planted are in any way 

 selected, i. e., not a purely random drawing from the 

 general population, (6) the probable errors of random 

 sampling in the drawing of the sample for planting are 

 added to the probable errors of sampling due to the non- 

 selective (purely random) death rate. 



The first objection does not hold in the series discussed 



