130 NATURAL INHERITANCE. [chai\ 



ties, so all were disregarded that contained fewer than 

 four individuals. In (2) and (3) I could not with 

 safety use large Fraternities. In (4) the method of 

 selection was, as we have seen, quite indifferent. This 

 makes the accordance of the results derived from the 

 Special data all the more gratifying. Those from the 

 E.F.F. data accord less well together. The E.F.F. 

 measures are not sufficiently exact for use in these 

 delicate calculations. Their results, being compounded 

 of b and of their tendency to deviate from exactness, 

 are necessarily too high, and should be discarded. I 

 gather from all this that we may safely consider the 

 value of b to be less than 1*06, and that allowing for 

 some want of precision in the Special data, the very 

 convenient value of l'OO inch may reasonably be 

 adopted. 



Trustivorthiness of the Constants. — There is difficulty 

 in correcting the results obtained from the E.F.F. data, 

 though we can make some estimate of their general 

 inaccuracy as compared with the Special data. The 

 reason of the difficulty is that the inaccuracy cannot 

 be ascribed to an uncertainty of equal ± amount in 

 every entry, such as might be due to a doubt of 

 "shoes off" or "shoes on." If it were so, the Prob. 

 Error of a single value of the E.F.F. would be greater 

 than that of one of the Specials, whereas it proves to 

 be the same. It is likely that the inaccuracy is a com- 

 pound first of the uncertainty above mentioned, whose 

 effect would be to increase the value of the Prob. Error, 



