520 



THE AMERICAN NATURALIST [Vol. XLVI 



— .012 units. Considering- the differences individually 

 in comparison with their probable errors 2 we note that if 

 we regard a difference at least 2.50 times its probable 

 error as statistically significant, 10 of the differences may 



be looked upon as trustworthy. Of these 5 are positive 

 and 5 are negative. If we require Diff./E<ur f .=4.00, we 

 find only 7 trustworthy cases, 4 positive and 3 negative in 

 sign. 



Diagram I shows graphically the amount and the sign 



2 It is somewhat difficult to decide just what ratio of the difference to its 

 probable error should be used to indicate trustworthiness. As already 



