A Correlation Measure for Nominal Data
Kenneth H. Ives, Jean Dickinson Gibbons · The American Statistician · 1967
Where two factors or attributes may be either present or absent in the observations in a sample, and one wishes to determine the strength of the relationship between the factors, available correlation techniques for nominal data do not seem to apply. Here we have a single sample problem, but two observations on each element in the sample, analogous to a paired sample problem. The data can be recorded as tl where the factor is present, and -1 if the factor is absent. If the factors are either both present or both absent in the ith observation pair (Xi, Y,) so that there is agreement, then xi - Y, = 0 , and \Xi - Y,I = 2 for disagreement,