STUDIES OF CEKEBKAL FUNCTION. IX 17 



the other hemisphere. The area of overlap between the two 

 lesions was then outlined, measured, and expressed as per- 

 centage of the total area of the neocortex. These percentages 

 have been used as a basis for computing the constants for 

 comparison with those obtained by consideration of the total 

 extent of lesion. In later discussions the measurements ob- 

 tained in this way are referred to as 'lesions common to both 

 hemispheres. ' 



In computing correlations we have used the method of rank 

 order, 5 in preference to the Pearson r, since our data do not 

 follow a random distribution. All methods of measuring the 

 degree of association between variables are based upon as- 

 sumptions concerning random sampling which are not ful- 

 filled by our data on brain lesions, and the use of correlation 

 methods in such cases can be justified only as a crude method 

 of expressing the presence or absence of a significant associa- 

 tion. The relation between extent of lesion and retardation 

 is probably not rectilinear, so that the correlation ratio would 

 be a more suitable measure of the association. In most cases 

 it would give a somewhat higher figure than the correlation 

 coefficients reported, but in the present state of our knowledge 

 slight differences in the magnitude of the coefficients have no 

 significance. We are dealing with differences between indi- 

 viduals and groups which are many times greater than the 

 range of normal variation, so that refinement of statistical 

 treatment is of relatively less importance than if we were 

 trying to measure smaller differences. 



Special controls 



There is little doubt that the experimenter may influence 

 the maze records of his animals by slight differences in pro- 

 cedure of which he is scarcely aware. Somewhat more gentle 

 handling of one than of another animal, deviations in the al- 

 lowance of food, and personal variations in the criteria of 

 errors may influence the data and are likely to do so in a 



5 p = 1 -~- d ~,) P-E.p = l ~f 0.7063. 

 n(nM)' p Vn 



THE JOURNAL OV COMPARATIVE NEUROLOGY, VOL. 57, NO. 1 



