Abstract:

The talk will discuss the role of the odds ratio as a measure of association between categorical variables. From among the several measures of association proposed for categorical data, the odds ratio is the only one which is variationally independent from the marginals of the table. The odds ratio can be generalized to describe various aspects of the association structure among several variables and log-linear models, which are the most widely used tools in the analysis of categorical data, can also be interpreted in terms of the conditional odds ratios. In spite of all these attractive properties, odds ratios should be interpreted with care. A fact which may cause difficulties to the naive user, Simpson's paradox, will also be discussed.