Dear Stata folks,
Many tests and statistics to control and correct for endogeneity seem to focus on linear models or, in the case of nonlinear models, on probit regression. But how can I test most efficiently if my independent variable is endogenous in a logit regression model (with panel data)? Is there an alternative to the Hausman test (which does not seem to be suitable for my model for several reasons – e.g., because I am analyzing firm-year/between and not cluster/within effects and my main effect gets eliminated in the case of a fixed-effects model, although the result of this test suggests to use a fixed-effects model)?
I would assume that there must be a more straightforward way to only test if my independent variable correlates with the error term (covariance of independent variable and error term not equal to zero) but I couldn’t find a common procedure that seems to be generally accepted in the literature. Usually, it is suggested to run post-estimation tests like the Hausman test to control for endogeneity and, then, conduct a two-stage/instrumental variable estimation procedure (2SLS or something analog for logit/probit) to correct for endogeneity.
Your help and advice is much appreciated.
Best regards,
Charlotte
Many tests and statistics to control and correct for endogeneity seem to focus on linear models or, in the case of nonlinear models, on probit regression. But how can I test most efficiently if my independent variable is endogenous in a logit regression model (with panel data)? Is there an alternative to the Hausman test (which does not seem to be suitable for my model for several reasons – e.g., because I am analyzing firm-year/between and not cluster/within effects and my main effect gets eliminated in the case of a fixed-effects model, although the result of this test suggests to use a fixed-effects model)?
I would assume that there must be a more straightforward way to only test if my independent variable correlates with the error term (covariance of independent variable and error term not equal to zero) but I couldn’t find a common procedure that seems to be generally accepted in the literature. Usually, it is suggested to run post-estimation tests like the Hausman test to control for endogeneity and, then, conduct a two-stage/instrumental variable estimation procedure (2SLS or something analog for logit/probit) to correct for endogeneity.
Your help and advice is much appreciated.
Best regards,
Charlotte
