[FernandoRios]

Hi
unfortunately I never added an option to the command that would also export the first stage model. Here the logit.
for dummy variable, there is nothing in particular. Perhaps the only thing to consider, if y is a dummy, the range of statistics you can use to estimate is limited (no quantile not ginis)
hth

[Odongo Kodongo]

Many thanks, FernandoRios . This is very useful.

I have encountered an unrelated problem with the oaxaca_rif command. I'm estimating the O-B decomposition on a number of binary dependent variables (using the same set of regressors):

oaxaca_rif  depvar  x1 x2 x3 . . . xN  , by( treatvar ) wgt(1) rif(mean) rwlogit ( z1 z2 z3 . . . zN ) cluster( clustervar ) relax

The code runs well, without -relax-, for all the dependent variables, except one. For the one dependent variable, Stata reports the following error.

score variables for model __00000J contain missing values

I have tried to check whether this is the result of missingness in the data, but there is none.

It happens that this particular dependent variable has very few 1's relative to 0's, almost akin to "overdispersion" in count outcomes.

Question: I'm wondering what's causing the error and how I can correct it to enable me to continue with the analysis.

I have seen some answers on related problems on Statalist; I have tried them with no success.

When I specify -noisily- I get some not-so-useful results with an additional error message:

dropped coefficients or zero variances encountered
specify -noisily- to view model estimation output
specify -relax- to ignore

Thanks again for your help.

[FernandoRios]

You have a problem of specification
I would start with modeling the logit problem first to see if it works
after that work on the on decomposition. Seems you are trying to use variables that are not observed for one of the two groups.
f

[Odongo Kodongo]

Thank you very much, FernandoRios. I'll try this approach.