74<> 



THE AMERICAN NATURALIST [Vol. XL VII 



The first of these graphs shows the values reduced to 

 percentages of the constants for the general population 

 of seeds from which the samples used in these experi- 

 ments were drawn, i. e., the constants given in Table X. 

 The second shows the ratio of the differences to their 

 probable errors. 



The impression given by both of these charts is that 

 the mean weight of the surviving seeds has been increased 

 by the mortality, although there are one or two conspicu- 

 ously large negative values in each case. This impres- 

 sion is borne out by the numerical result, if we confine 

 our attention to the signs, merely. Of the 50 experi- 

 ments, 29 show an increase and 21 a decrease in seed 

 weight, whereas if there were no relationship between 

 mean seed weight and viability, the deviations would be 

 expected to be equally divided between positive and nega- 

 tive, except for the error of random sampling which would 

 be given by .6745 V50 X .5 X .5. Thus in the present case 

 for the whole fifty experiments, the deviation from the 

 equality which we should expect if there were no relation- 

 ship between mean seed weight and mortality is 4 ± 2.38 

 series. Surely this can not be regarded as a trustworthy 

 difference, but we note that the difference has the same 

 sign and is relatively larger as we reduce our number of 

 cases by disregarding those comparisons which are less 



