Approximating the marginal effect of discrete regressors in logit models

Logit models are non-linear in their explanatory variables. Derivatives with respect to the explanatory variables therefore only approximate the response to discrete changes in regressor values, yet have gained some support within the literature. This note investigates such an approximation strategy and delineates the conditions under which it does or does not give satisfactory results. The conclusion is that it is a risky strategy and that therefore the correct alternative should be employed.