Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: probit "backed up"
From
"JVerkuilen (Gmail)" <[email protected]>
To
[email protected]
Subject
Re: st: probit "backed up"
Date
Tue, 12 Mar 2013 11:44:27 -0400
On Tue, Mar 12, 2013 at 7:12 AM, Maarten Buis <[email protected]> wrote:
> On Mon, Mar 11, 2013 at 5:45 PM, Helen Markelova wrote:
>> I'm having problems with a simple probit not converging---the output says "backed up" after 4-5 iterations. Im using Medical Expenditure Panel Survey from 1996-2010 (about 217,000 observations) and running a probit on the probability of being privately insured (private (0,1) as the dependent variable). The explanatory variables include a number of dummy variables, but even if I limit the list to only a few dummy and continuous variables (family size, age, male, black, employed), it still won't converge. Any suggestions on what else to try? Im using Stata 12.
>
> Maybe you are in a situation with near perfect prediction, -logit- and
> -probit- models don't work well in such situations. My first guess
> would be to inspect the relationship between private and employed.
> i.e. look at -tab private employed-. In general I would look at many
> such tables and look for near empty cells.
Yes, agreed. Dummy predictors in binary regression models can be quite
troublesome and I suspect that's where the issue lies. Work with the
continuous only first to see if that runs. If so that suggests that
the problem lies in the binaries.
I'd also seriously consider subsampling the data. That's a HUGE
dataset, so large that it might well be possible to tell the
difference between logit and probit. You can do a real
cross-validation so you might as well.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/