58 



THE AMERICAN NATURALIST [Vol. L 



\Yc think it must he ci mclndcd that for samples of 50 the usual theory 

 of the probable error of the standard deviation holds satisfactorily, and 

 that to apply it for the case of n = 25 would not lead to any error which 

 would be of importance in the majority of statistical problems. 



Tho original papers should be read by those who are dealing 

 with coefficients 1 >* i n near the limits of the range of correlation, 

 i.r win. must work with small samples. Those who can by extra 

 lahur olitain larger series of data should do so, for no knowledge 

 of the theory of the probable error can ever take the place of 

 widened series of data, although it may be essential to the inter- 

 pretation of eonstants based of necessity on a limited number of 



t(s Measurable on a Quantitative Scale; Regression 

 !■ or eases in which the rate of change in the y char- 

 be described by a straight line, the proper measure 

 dence is Pearson's 24 correlation ratio, -q. The value 

 ition ratio is two-fold, (a) It furnishes a measure 

 ^pendence of two variates in cases in which the use 

 ition coefficient is not fully justified. (6) It affords 

 testing, by the use of Blakeman's criterion, 25 for 



ins in deciding between the correla- 

 •elation ratio, the calculation of each 



68, be necessary. 



This method, which involves the 

 test for goodness of fit, should have 



dness of fit of regression curves has 



