284 PROF. K. PEARSON ON THE MATHEMATICAL THEORY OF EVOLUTION. 
Now the probable error of the difference of the Essex contribution and the 
remainder is •1111" for height and ‘0251 for correlation. Thus difference in height 
is nine times, and the difference in correlation more than six times the corresponding 
probable error. It is absolutely necessary therefore to conclude that the Essex con¬ 
tribution differs significantly from the remainder of the data. Now the Essex 
contribution appears to be drawn from brothers in a volunteer regiment, and I am 
inclined to think there may be two sources accounting for its peculiarities, 
(a) unconscious selection as to height by those who join the volunteers, (b) a greater 
correlation among the agricultural and working classes than among the middle classes. 
At any rate the great variation within the family to be found in the R.F.F. data does 
not repeat itself either in the Essex contribution or in other portions of the special 
data, which appear also to be drawn from military and working class sources. 
I would accordingly suggest that the R.F.F. data and the Special data give 
different results, because the latter are largely drawn from a different class of the 
population from the former (and possibly in the case of volunteer regiments by a 
method which itself tends to emphasise correlation). I should expect that the 
influence of natural selection is far greater—witness the greater infantile mortality— 
in the working classes, and that accordingly we should find the variation in a 
fraternity sensibly less, or the correlation much greater. I believe, then, that 
difference of variation in different classes of the community will ultimately be found 
to account for part, if not all, of the difference between the two values given for the 
correlation of brothers by the Special data and by the R.F.F. data. 
Considering the amount by which the elimination of a portion only of the hetero¬ 
geneity of the Special data reduces r, it does not seem likely that the R.F.F. data are 
so wide of the mark in the correlation values as might at first be thought. I doubt 
whether the correlation coefficients for collateral inheritance—at any rate in the 
middle classes—can be greater than *5. I have not at present sufficient data of my 
own to make a trustworthy determination of brother-brother correlation, but I was 
able to find the correlation of 237 brother-sister pairs from about 160 families. The 
measurements were taken without boots, and give values for the mean heights of 
brothers and sisters sensibly over 69" and 64" respectively. The families were all 
middle-class families—mostly those of male and female college students. They thus 
approximate to Mr. Galton’s R.F.F. series. The result was 
r = -4703 ± '0308 
The previous result was 
r = *3754 ± -0158 
The probable error of the difference therefore = *0346 and the difference '095, 
between two and three times the probable error. The two differ, of course, consider- 
