Binary Contrasts for Unordered Polytomous Regressors

The Stata Journal

In observational studies, regression coefficients for categorical regressors are overwhelmingly presented in terms of contrasts with a reference category. For unordered regressors with many categories, however, this approach often focuses on contrasting different pairs of categories to one another with little substantive rationale for foregrounding some comparisons with others. Mean contrasts, which compare categories with the overall mean, provide an alternative to the reference category, but the magnitude of mean contrasts is conflated with the relative sizes of the categories. Instead, binary contrasts compare a category with all the other categories, allowing the familiar interpretation for dichotomous regressors. Our command binarycontrast computes binary contrasts.