Marginal and Interaction Effects in Ordered Response Models

In discrete choice models the marginal effects of a variable that is interacted with another variable and the interaction term differ from the marginal effect of a variable that is not interacted with any variable. Standard software incorrectly estimates these marginal effects. I provide correct formulas for ordered response models that can be extended to other discrete choice models and an example using household survey data on food security in Bangladesh. Results show that marginal effects of the variables interacted and interaction term are estimated by standard software (such as STATA® 10) with large error and even with wrong sign.