750 



THE AMERICAN NATURALIST [Vol. XLVII 



to develop. The number of negative differences is sig- 

 nificantly higher than the number of positive differences. 

 The mean of the negative differences is larger numer- 

 ically than that of the positive differences. The propor- 

 tion of negative differences is higher among the constants 

 which are more probably trustworthy, being only 1.5 : 1 

 among those < 2.5E but 3:1 among those > 2.5£\ The 

 average ratio of the difference to its probable error is 

 only 1.65 for the positive differences, but reaches 3.04 

 for those which are negative in sign. 



Thus these results are in excellent agreement with 

 those of the field experiments. 



Turn now to the same question with regard to the 

 seeds giving abnormal germinations, B-C: 



Plus differences 23 +.235 +12.14 



M i 1 1 u - diti. P.-m-,.- . 27 -.331 -15.32 



All difference s 1 50 1 -.071 ' - 2.69 



For those differences at least 2.5 times their probable 

 error, the results are: 



Plus differences 3 j +.466 ' +28.14 



\ . 11 -^294 



Perhaps the evidence for a reduction in variability is 

 not so strong when seeds germinating abnormally are 

 compared with those not germinating at all. This is 

 precisely what one would expect if such seeds may be 

 regarded as in some degree intermediate between those 

 which produce perfect seedlings and those which produce 

 no seedlings at all. 



I now turn to the question of a possible reduction in 

 variability as one passes from seeds germinating ab- 

 normally to those germinating normally. The answer 

 is given by the comparison A-B : 



