A PHYSIOLOGICAL CHARACTER. . 27 



necessary in order to bring out the real significance of the material,^ 

 and has been confined to the determination of standard deviations, 

 probable errors, and a few correlations. 



Probable errors have been used in seeking to determine the 

 significance of the reaction-time differences between the selected 

 strains of the various lines. Many more probable errors have been 

 determined (and included in the tables for the benefit of the reader) 

 than have been referred to in the text discussions. On the other 

 hand, probable errors have not been determined for many differences 

 which were obviously too small to have statistical significance.^ 



For the benefit of anyone wishing to work over a portion of the 

 data statistically, the sums of the individual reaction-times and the 

 sums of the squares of the individual reaction-times are given in the 

 tables which present the data for Lines 695 and 757 by broods. 

 Hence, most of the statistical treatment for these two lines is capable 

 of verification from the data presented in this paper. Likewise the 

 data are available for any further statistical treatment. 



The standard deviations of the reaction- time means are* large, 

 whether the data is considered by shorter periods or by year periods. 

 This is largely due to the over-time individuals, the data for which 

 afforded considerable difficulty in statistical treatment. Other plans 

 of handling the data for these individuals (other than assuming 900 

 seconds as their reaction-time) were considered, but they seemed to 

 provide no better method of utilizing this very significant part of 

 the data. 



^ Such statistical treatment as has been accorded the data of this paper has been greatly 

 facilitated by suggestions from Dr. J. Arthur Harris, of the Station for Experimental Evolution. 

 However, anj' errors or insufficient treatment must be ascribed to the writer, who has conducted 

 the statistical analysis. 



''It is recognized, however, that the value of the probable error is not fully utilized when 

 one fails to determine probable errors for all the differences in a series of data — for, obviously, 

 a difference which is twice its probable error deserves more consideration than one which is 

 only a fraction of its probable error, though a difference less than 2J.^ times its probable error 

 is considered of doubtful statistical validity. Nevertheless, to conserve the computer's time, it 

 was felt advisable to omit the determination of probable errors for many of the differences 

 which did not promise marked statistical value. 



