svy: logit command stuck on "(running logit on estimation sample)"

[Jonathan Watts]

Hi,
I'm trying to run a set of svy: logit commands, but for some reason, including certain variables in the model results in Stata not getting past "(running logit on estimation sample)". I left it running overnight, and the next morning it was still stuck on the same command line when it normally takes roughly 5-30 seconds for the "Jackknife replications" to appear.

Although there is not initially an error message, I noticed after breaking the command, the following two messages appear (but appear before the break in the output), which read "an error occurred when svy executed logit" and "an error occurred when jackknife executed logit". I saw a couple of threads on one of these messages where someone seemed to be making other errors like having a categorical outcome for logit, but I have a dichotomous outcome for model in question. One thread said Stata needs all inputted variables in a logit model to be numeric, which I verified is the case for mine.

I've even tried running a model with only the outcome and the covariate that seemed to be the issue, and it ran fine. Then, I continued to add my other covariates back into the model one-by-one, which all ran fine, until my final version of the model, in which adding a different covariate (than what the initial problem variable was) as my final variable, resulted in Stata getting stuck, again, on "(running logit on estimation sample)".

Any suggestions for troubleshooting this?

[Carlo Lazzaro]

Jonathan:
welcome to this forum.
The simplest fix is to kick the culprit predictor out the right-hand side of the regression equation, if it does not affect a true and fair view of the data generating process..

[Richard Williams]

Welcome to Statalist.

This handout contains various troubleshooting tips.

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

The first thing I usually do is make sure my software is up to date. If I get really lucky, the problem may have already been fixed.

The next thing I do is add the difficult option. It probably won't work but it is easy to do and sometimes works miracles.

If the easy ways don't work there other things you can try, like rescaling some variables.

[skolenik]

webuse nhanes2jknife, clear
svy, noisily : logit diabetes i.race, iter(20)
* noisily to show each regression
* iter(20) to limit the number of iterations; you may be going to hundreds of thousands failing to converge