eprobit margins: inconsistent estimation sample levels 0 and 1 of factor

[Andrew Perrin]

I am trying to diagnose the problem I reported here . I've reproduced the error outside of the mi environment, so am creating this new topic to reflect that fact.

The error arises after:

svy: eprobit DEPVAR IVAR1 IVAR..., entreat(TREATMENT = IVAR..., nointeract)

This command completes successfully and returns full results to eprobit. Then

margins, dydx(*) predict(pr)

The error message is:
inconsistent estimation sample levels 0 and 1 of factor IVARNAME

Note that IVARNAME is one of the predictor variables in the entreat() clause to eprobit.

Note also that this doesn't happen in all the imputations; the above happened, for example, after using

mi extract 1

, but does not happen with

mi extract 0

.

Thanks for your insights!

[Andrew Perrin]

More information: running the entreat() clause separately using

svy: probit TREATMENT IVAR...

works fine and the margins are computed properly. It also works properly under the mi environment. So I believe whatever the problem is arises from the variables used in the endogenous treatment equation in eprobit.