On Marginal Effects in Semiparametric Censored Regression Models

Abstract

This note illustrates that the typical parameter, beta, in a censored regression model can be used to calculate an interesting marginal effect even when the errors in the model and the explanatory variables are not independent. The result is relevant for cross sectional models such at the ones considered in Powell (1984), Powell (1986) and Chen and Khan (2000), as well as for panel data models such as the ones in Honore (1992) and Alan and Leth-Petersen (2006), and it applies with fixed as well as with random censoring.