P. F. EVERITT 
389 
large. It may also quite easily happen in such cases, that the value of d/N does not 
vary for values of r between "8 and 1 and in such a case the method of the fourfold 
table fails, as the magnitude of the probable error would render the determination 
by the ordinary method quite idle as regards a quantitative result although it 
would generally be possible to say whether the correlation was large or small. 
Unfortunately it is not generally possible to classify the data, on which the fourfold 
table has to be based, in more than one way, or else by a suitable choice of 
classification it might be possible to avoid such cases. 
In conclusion I desire to express my gratitude to Prof. Pearson for many 
valuable suggestions, especially in connection with the choice of a suitable 
quadrature formula. 
