Abstract

Interpretation in nonlinear regression models that include sets of dummy variables representing categories of underlying categorical variables is not straightforward. Partial effects giving the differences between each category and the reference category are routinely computed in the empirical economics literature. Yet, partial effects yielding the differences between each category and all other categories are not calculated, despite having great interpretative value. We derive the correct formulae for calculating these partial effects for an ordered probit model. The results of an application using data on subjective well-being illustrate the usefulness of the alternative partial effects.

Keywords: Discrete change, Factor variables, Marginal effect, Nonlinear models, Partial effect