st: Durbin-Wu-Hausman convergence problem

[Mike Pesko <mpesko2@uic.edu>]

Dear list members,

I would like to perform an endogeneity test to see if I can include a
divorced marital status variable into a smoking model. Both the
smoking status variable and the divorce variable are binary variables.
I believe I need to conduct a Durbin-Wu-Hausman test and have run an
original regression with divorce in it. Following, I have run

'predict smoker_res, res'

and included that variable in a model with divorce as the dependent
variable. I have also included the regular smoker variable in the
divorce model. There are no new variables currently in my divorce
model that were not in the smoking model, although a few have been
deleted, such as cigarette prices.

The problem is that this second model will not converge after 25 or
more iterations. I wonder if it's because both variables are binary
variables? The dataset is also very large at 4 million observations,
although converge is even problematic when I take a small sample.
Also, many of the same variables are used in both regressions, such as
year, state, and month fixed effects, unemployment rate, income, and
employment status.

Please let me know if you have any advice on if I can do something
differently to get converge on this test, or maybe try a different
test altogether. Thanks,

Mike Pesko
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/