Problems with Logistic Regression

[Girlan Oliveira]

I'm running 15 logistic regressions using the same variables of the same database, however, each of the regressions referring to different combinations of two variables (uf and year), but in all of them the stata do not provide the resulting estimates of parameters. Stata returns me some notes on some variables, reports the drop of some observations of these variables, reports that the outcome is given by a condition of a particular variable and ends with error code r (2000).
The error code must be some problem with my database, but I can not see what this problem and I am unable to send attached the database file dta format so that someone can help me.
How do I send the database as an attachment to the message?

[Clyde Schechter]

It sounds like you need a better understanding of the warning messages that Stata is giving you. The data set wouldn't be all that helpful for this. What would be very helpful is if you show us the exact commands and exact response you got from Stata by copying from your Results window or log file and pasting into a code block on this Forum.

[Girlan Oliveira]

Clyde,

Sending attached the logfile.

[Nick Cox]

The key message is numerous variations on

outcome = logrenpcdef > 3.176408 predicts data perfectly

See e.g. http://www.ats.ucla.edu/stat/mult_pk...git_models.htm

[Girlan Oliveira]

Thank very much, Nick.

The text that you showed me I am dealing with a problem of quasi-separation logistic regression. In the text there is the suggestion to use the 'firthlogit command to solve the problem.
I'm looking for more information about using the 'firthlogit' command and if questions arise again I post something.

[Richard Williams]

We can't tell much about your data set other than that it has at least 700,000+ cases. Before I went to firthlogit, I would greatly simplify my model, adding one variable at a time. See if you can isolate at what point you start having problems and then see if you can figure out why. Maybe it has something to do with your sample selection or how some variable was coded or computed. I suspect something is wrong if you have 700,000+ cases and are getting error messages about no observations left.

[Girlan Oliveira]

I am using the stata 12.0 and downloaded the 'firthlogit' to resolve the problem of quasi-separation in the logistic regression, but was unable to use the '' firthlogit 'because the stata says my version of the software can not run version 13.1 programs. The error code is r (9).
How can I solve this problem without having to buy the 13.1 version of the stata?