Cochran (1952): none of the expected cell frequencies 
should be less than one. No more than 20 % of the expected 
cell frequencies should be less than 5.

Cochran, W. G. (1952). The chi-square goodness-of-fit test. 
Annals of Mathematical Statistics, 23, 315-345.

The test statistics is approximately Chi-squared, under the 
assumption that all expected cell frequencies are at least 1.5,
in which case this approximation is satisfactory. Morris H. 
DeGroot Probability and Statistics. Addison Wesley.


Application of the chi-square formula. Let O be the observed 
value of each cell in a table. Let E be the expected value 
calculated in the previous step. For each cell, subtract E 
from O, then square the result and then divide by E. Do this 
for every cell and sum all the results. This is the chi-square 
value for the table. If Yates' correction for continuity is 
to be applied, due to cell counts below 5, the calculation is 
the same except for each cell, subtract an additional .5 from 
the difference of O - E, prior to squaring and then dividing 
by E. This reduces the size of the calculated chi-square value, 
making a finding of significance less likely -- a penalty 
deemed appropriate for tables with low counts in some cells.