356 On the Error of Counting imth a Haemacytometer 
In this way it was possible to count at leisure without feai" of the cells straying 
from one square to another owing to accidental vibrations. A few cells stuck here 
and there to the cover glass, but as they appeared to be fairly uniformly distributed 
and were very few compared with those that sank to the bottom they were 
neglected : had the object of the experiment been to find the number of cells 
present they would have been counted by microscope fields, and correction made 
for them ; but in our case they were considered to belong to a different "population " 
to those which sank. 
Those cells which touched the bottom and right-hand lines of a square were 
considered to belong to the square ; a convention of this kind is necessary as the 
cells have a tendency to settle on the lines. 
There was some difficulty owing to the buds of some cells remaining undetached 
in spite of much shaking. In such cases an obvious bud was not counted, but 
sometimes, no doubt, a bud was counted as a separate cell, which slightly increases 
the number of squares with large numbers in them. 
In order to test whether there was any local lack of homogeneity the correlation 
was determined between the number of cells on a square and the number of cells 
on each of the four squares nearest it ; if from any cause there had been a tendency 
to lie closer together in some parts than in others this correlation would have been 
significantly positive. 
Distributions 3 and 4 were tested in this way (Table II), with the result that 
the correlation coefficients were + "016 ± 037 and 'Olo + 037. This is satisfactory 
as shewing that there is no very great difficulty in putting the drop on to the 
slide so as to be able to count at any point and in any order ; as good a result may 
be expected from counting a column as from counting the same number of squares 
at random. 
The actual distributions of cells are given below, and compared with those 
calculated on the supposition that they are random samples from a population 
following the law which we have investigated : the probability P of a worse fit 
occurring by chance is then found. 
I. Mean =-6825 : 1113 = -8 11 7 : /X3 = 1-0876. 
Containing 0 1 2 3 4 5 cells 
Actual 213 128 37 18 3 1 
Calculated 202 138 47 11 1-84 -24 
2 
Whence x^ = 9-92 and /'=-04. 
Best fitting binomial (1-1893 - •1893)-3-6"m x 400 for which F=-b± 
II. Mean =1-322.5 : ;x2 = 1-2835 ^3 : =1-3574. 
0 1 2 3 4 5 6 
Actual 103 143 98 42 8 4 2 
Calculated 106 141 93 41 14 4 1 
"Whence x' = 3-98 and P = -68. 
Best fitting binomial (-97051 -|--02949)''«-2»s* X 400 for which i'=-72. 
