PROFESSOR Iv. PEARSON AND DR. A. LEE ON 
grey blue-green range are also not entirely satisfactory in their results, nor those 
taken for a still larger range covering tints 3, 4, 5, and 6, or blue-green, grey, hazel, 
light brown, and brown, which cover roughly about 1*5 to 1'6 times the standard 
deviation. We shall now consider the results of three methods of considering: the 
relative variability, (a) from the excesses given in columns 1 and 2 of Table VIII. ; 
(J3) from the grey blue-green range given in columns 3 and 4 of Table VIII. ; and (y) 
from the range of tints 3 to 6 inclusive given in columns 5 and 6 of Table VIII. As 
we have already indicated, these methods are not likely to give the same relative 
magnitude numerically for the variabilities ; we must content ourselves if they agree 
in making the ratio of cr,. to <x, greater or less than unity. Now, in the twenty-two 
cases 
ol and (3 disagree in 10 cases. 
(3 and y disagree in 7 cases. 
a and y disagree in 5 cases. 
Further, for the five cases in which ol and y disagree, those for father and son, 
paternal grandfather and grandson, maternal uncle and nephew, show so little 
difference of variability in the two sexes that both methods give sensibly the same 
results, i.e., equality of variability. In the cases of the paternal grandfather and 
grandchildren, the two methods diverge rather markedly. 
It will be of interest accordingly to work out the probable errors as given by the 
excess method for one, say the first of these cases. The theory is given in Part VII. of 
the present series. Here E t = 275T65, E 3 = 309'013, whence we find probable error 
of E, = 17'273, probable error of E 3 = 16*925, correlation between errors in E : and 
E. : = — *4424, probable error in oq/oq = ‘0394. 
Thus the probable error in the ratio of the variabilities is about 4 per cent., and of 
the order of the quantities by which we are distinguishing the relative size of oq 
and (To. 
Further, there is another source of error in evaluating E, and E, due to the 
method of interpolation used, and this would still further increase the probable error 
in oq/oq. We cannot therefore lay any very great stress on the manner in which the 
ratios of variabilities for the paternal grandfather and grandchildren have swung 
round from (a) to (y). 
A further examination shows us that in all five cases wherein y differs from a it is 
in accord with (3. I shall accordingly take y as the standard criterion, but in those 
cases in which it has agreement with a, its conclusions, must be given greater weight. 
(a.) On the Relative Variability of Sex in Eye-colour. —The following male groups 
are more variable than the corresponding female groups :— 
Sons of fathers than daughters of fathers. 
Sons of mothers than daughters of mothers. 
Brothers of brothers than sisters of sisters. 
