06-19-2015 09:49 AM
I want to estimate a conditional logistic regression with many alternative that has one or more potentially endogenous explanatory variables.
Proc Qlim only seems to work with binary discrete choice models (but does have a method for handling endogenous explanatory variables by estimating separate regressions for each on the instruments.). Is there a way to do this in Proc Qlim or another procedure?
Proc MDC will estimate the mode and I can put into the model the residuals from auxillary regressions of the endogeneous explanatory variables regressed on all the instruments, but I don't see a way to correct the standard errors in the conditional logit model (with the original variables and the residuals for the endogenous explanatory variables) to account for the inclusion of variables that are predicted values. Is there a method (bootstrapping?).