clogit in svy package does not converge.

[Yen Lee]

Hello everyone,

I am fitting a conditional regression model by using clogit in svy since the sampling structure of my sample is complex. I would like to fit the asclogit model but since I have complex sampled data, I used the clogit in svy with data manipulation instead. The code of my fitted model is as follow:

svyset psu [pw=psweight], strata(str) bsrweight(bsw*) vce(bootstrap) singleunit(scaled)

svy: clogit choice alt1 alt2 alt3 sub1-sub20, group(id).

The "choice" variable indicates the choice the subject made. The alt1-3 variables are the alternative variables and are dummy coded. The sub1-20 are the interactions of the subjective variable and each alternative.

The model code should be correct since I have conducted the same code with different data which converged.

However, stata showed the model was un-conveged and returned the following error messages:

convergence not achieved
convergence not achieved
an error occurred when svy executed clogit
an error occurred when bootstrap executed clogit
r(430);

I wonder why it happens and if there is a way to make it converge, such as changing the starting values or other techniques. I appreciate any suggestions and possible error/bug detected method. Any reply is appreciated. Thank you for your precious time!

Best,
Yen

[Richard Williams]

Generic advice:

• Consider rescaling variables, e.g., measure income in thousands of dollars rather than in dollars. This can help with numerical precision problems.
• The difficult option will sometimes work miracles. There is no guarantee it will work (sometimes it makes things worse) but it is very easy to try.
• Try simplifying the model (use fewer variables) and then gradually build back up. You may be able to identify a variable that is causing you problems. I would be suspicious of having so many interactions.
• Try doing this without svy and see if it will work. This may help you isolate the problem

[Yen Lee]

Hello Richard,

Thank you for your precious suggestions. Your suggestion helps me resolve the problem.
The non-convergence occurs since there is one category which can perfectly predict the outcome.

I appreciate your time and priceless help!

Best,
Yen