Harris, A quantitative study of the factors influencing etc. 5 



is not to be given the same weight as in the case of coefficients 

 of moderate values. Again, the N used in the calculation of the 

 probable errors is the number of seeds weighed, not the number 

 of pods involved, many of the pods furnishing two or more seeds. 

 Possibly, the number of pods rather than the number of seeds 

 weighed should bave been employed. This would give higher 

 probable errors. In view of these facts, much stress cannot be 

 laid upon the ratio of the constants to their probable errors. De- 

 ductions must be drawn rather from the general run of the constants. 

 Consider, flrst, the relationships for the number of ovules per 

 pod. Three of the constants are negative and two are positive. 

 The highest is only — .123. while the average of the five is — .047. 

 Thus the influence of the number of ovules upon seed weight is 

 very slender indeed. Expressing it in terms of regression as has 

 already been done, x ) we find: 



Thus the highest absolute change in seed weight for a Var- 

 iation of one ovule per pod is .28 units, or 70/10,000 th gram! 

 The mean value, regarding signs, is but 24/ 10,000 th gram. 



When one takes into consideration that the number of ovules 

 and the number of seeds per pod are correlated, 2 ) one can hardly 

 assert on the basis of the present materials, extensive though they 

 may be, whether there is any relationship at all between the 

 number of ovules in a pod and the weight of the seeds which it 

 matures. 3 ) 



For number of seeds per pod, the results are steadier. In 

 all, the sign of the correlation is negative. In 4 of the 5 cases, 

 the constant is nominally significant in comparison with its probable 

 error. The mean correlation is — .096. 



*) Harris, J. Arthur, On the Relationship between the Bilateral Asym- 

 metry of the Unilocular Fruit and the Weight of the Seed which it Produces. 

 (Science. N. S. 36. 1912, p. 414—415.) 



') For actual constants in many series see "On the Relationship between 

 Bilateral Asymmetry and Fertility and Fecundity." (Arch. f. Entwicklungs- 

 mech. d. Organ. Bd. 35. 1912. S. 500-522.) 



3 ) Apparently, the numercial smallness of these correlations and their 

 diversity in sign cannot be attributed to regression of a higher order than 

 linear. Diagram 1 shows these lines and the empirical means which they 

 smooth. Dr. Roxana H. Vivian of Wellesley CoUege has kindly worked out 

 the correlation ratios and applied Blakeman's test for linearity of regression 

 in four of the cases, and (bearing in mind the difficulties involved in testing 

 for linearity when r is low) there is no clear evidence that a curve of a higher 

 order would be better than a straight line for expressing the change in seed 

 weight due to Variation in number of ovules. 



