Omitted Variables and Misspecified Disturbances in the Logit Model

In binary discrete regression models like logit or probit the omis- sion of a relevant regressor (even if it is orthogonal) depresses the re- maining <font face="Symbol">b</font> coefficients towards zero. For the probit model, Wooldridge (2002) has shown that this bias does not carry over to the effect of the regressor on the outcome. We find by simulations that this also holds for logit models, even when the omitted variable leads to severe misspecification of the disturbance. More simulations show that es- timates of these effects by logit analysis are also impervious to pure misspecification of the disturbance.