Convergence not achieved (error 430)

[Martin Paul]

Dear STATALIST,

I am estimating an endogenous switching regression model using the 'movestay' command.

The first regression runs very well without any convergence issues. The remaining two regressions iterate 16000 times and then arrive at no convergence. I have tried the difficult option but the problem still persists.

What in your opinion could be happening?

[FernandoRios]

There could be many reasons for that happening. Among those, it is possible that the assumptions for MLE are not fullfilled. Perhaps you should try two step estimator
Fernando

[Martin Paul]

....the treatment effect model you mean?

[Richard Williams]

See pp. 2-3 of the following for some suggestions:

https://www3.nd.edu/~rwilliam/stats3/L02.pdf

Rescaling especially seems to often help, at least if the variables are in very different metrics.

[FernandoRios]

In addition to Richards helpful comments, My suggestion is to do something like follows:

use http://fmwww.bc.edu/RePEc/bocode/o/oaxaca.dta, clear
* This is the MLE estimator
movestay lnwage educ exper tenure, select( single= age agesq educ exper tenure)
* This is an alternative MLE estimator using the command oheckman:
oheckman lnwage educ exper tenure, select( married= age agesq educ exper tenure)
* And this is the two step estimator:
oheckman lnwage educ exper tenure, select( married= age agesq educ exper tenure) two
** First and second command should give you the same results, but if there is a problem fitting the joiint density, it is often easier to use an equivalent to the heckit procedure (third line of code).

HTH
Fernando

[Martin Paul]

Thanks very much for your help!

I just rescaled all my income and asset variables as they were in thousands by generating new variables where they are divided. However, the iteration issue still continues.