Table I. 



in taking Niis 300 1). The probable error will then be . 67449/Vn^ . 0389, 

 saj' E, = . 04, for each individual. If we regard r/E^)2.S as significant, 

 we must look upon all constants lying between + . 100 and — . 100 

 as of questionable value for biological arguments. 



In thirteen of the cases r^o lies between and + . 100. There 

 remain seven which are possibly significant. Of these, six are negative 

 in sign and the other is only slightly greater than . 100. Of the twenty 

 constants, thirteen have the negative sign; the deviation from equality 

 of plus and minus carries little weight. 



Calculating the means of the constants for the individuals, we 

 find . 0413 for the positive, — . 1349 for the negative, and — . 0733 

 for the whole series. The standard deviation of the coefficient of 

 correlation is . 1173, and from this we find the probable error of the 

 average correlation to be . 0177. Now — . 0733 + . 0177 is 4.14 times 

 its probable error and so perhaps statistically significant, but it is so 

 low that little practical biological importance would ordinarily be 

 attached to it. 



1) This is the number of locales counted; each fruit has three locules. There 

 is always a question in the case of repeated organs as to what number should be 

 used in the determination of the probable errors. If it be considered that the actual 

 number of fruits should be used, the probable error will be higher than if the number 

 of locules is used. Possibly we should use the actual number of inflorescences instead 

 of weighting them with either the number of pods or the number of locules which 

 they produce. This would raise the probable errors still higher I have used 3C0 as 

 xV since it favors as much as possible the hypothesis that there is a relationship 

 between the position of fruits on the inflorescence and their fertility characters, i, e. 

 I have used the method which tests as severely as possible my own theories. 



