1974 Volume 1 Issue 1 Pages 25-37
Covariation analysis is a simple method of analysing the structure of data in the form of frequency tables. In this paper, the argument is limited within the context of 2×2 frequency tables for ease of introducing the basic principles. In this context, it is shown that the deviation of the observed frequency (in any cell) from its marginal estimate, the familiar quantity in the 2×2 X2-test, can be interpreted as an estimate of covariation of the two conditional probabilities corresponding to the pair of observed binary random variables. By using this value as an indicator of structural information contained in the data, the contribution of each of the condition variables to the determination of the nature of the given data can effectively be evaulated. This is demonstrated in this paper both in theory and through an example of actual analysis.