398 On the Frequeney DiAtrihntiom of Phagocytic Co^mti^ 
for homogeneity. Secondly, admitting the two sets of frequencies to be definite 
examples of skew variations of the types indicated, is there any distinct character 
marking one set off from the other ? This question is evidently one of much 
theoretical and practical interest, and we cannot at present furnish an answer 
which is entirely satisfactory. 
We have already I'emarked on the great diminution in skewness exhibited by 
the F. material as compared with the S. counts. Although some reduction might 
be anticipated in view of the higher modal values, which latter are in some 
measure dependent on experimental conditions (vid. sup. p. 384), we are not clear 
that this is a complete answer. To determine whether, by random sampling, such 
a reduction in skewness as we have found would be accounted for adequately 
by the change in modal value is not easy for the types involved. We have, 
however, applied a less satisfactory but somewhat easier test. 
It was noticed that, although the second moments of the F. distributions were 
larger than for the S. counts, the increase was not proportional to the mean values. 
This will be apparent from Table VII., which gives the coefficients of variation. 
TABLE VII. 
Coefficients of Variation*. 
Strangeways 
I 
90-14 
Fleming : 
Tub. Chor. 
76-46 
II s. c. 
97-86 
10 Norm. 
60-08 
Ill 
85-60 
J) 
Norm. S. A. 
62-26 
II s. & c. 
91-15 
75 
No. 2 
73-94 
IV 
94-29 
>> 
T. A. 
63-88 
V 
85-92 
)) 
N. S. B. 
60-59 
VI 
84-64 
VII 
90-38 
IV (2000) 
93-69 
Now it is well known that = — ^ — 
when ?v,M2 is the correlation between mean and second moment, and o-/,, cr^, the 
standard deviations of mean and second moment respectively. 
If therefore we calculate o-^ and o-^,, on the basis of the count of 2000 cells 
(S. IV 2000), we can form some idea as to whether the observed values of /i.^ for 
the other counts are such as we should expect, with random sampling, to be 
associated with the observed values of the mean. 
The regression of fi., on h was assumed to be linear and the probable value of 
was calculated for 7 S. counts and the results tested for goodness of fit. P was 
greater than -99. 
The same process applied to 5 F. counts gave P < '0004. 
* It need scarcely be remarked that no stress is or can be laid on the actual vahies of tlie coefficients 
since Mean and Coefficient of Variation are negatively correlated. The table merely illustrates the 
tendency discussed in tlie text. 
