topic Re: Is it easier to use SUDAAN to perform log binomial regression using weighted data? in SAS Programming

Is it easier to use SUDAAN to perform log binomial regression using weighted data?

SAS93 — Mon, 21 Jun 2021 18:57:32 GMT

I have complex survey data that I want to run log binomial regression on; from my experience so far and different documentation I've looked through, SAS doesn't handle LBR well when it comes to also accounting for survey weights and designs.

I know some papers have used LBR on weighted data using STATA, but would SUDAAN do just as good of a job?

Re: Is it easier to use SUDAAN to perform log binomial regression using weighted data?

ballardw — Mon, 21 Jun 2021 19:17:51 GMT

If you are thinking of using SUDAAN Proc LOGISTIC (or the SAS callable RLOGISTIC) then you may want to investigate SAS SURVEYLOGISTIC.

Easier?

Depends on experience I would say.

There are some differences in how variability is used so you can differences between the SAS and SUDAAN procedures though they seldom show up at greater than 0.01% of prevalence measures when set up with the similar options.

Re: Is it easier to use SUDAAN to perform log binomial regression using weighted data?

SAS93 — Tue, 22 Jun 2021 11:35:12 GMT

Proc surveylogistic does not support log binomial regression, unless I've badly misunderstood the documentation I've read. It supports a variety of different link= functions, but not link=log. Regular proc logistic does.

I've also looked into proc genmod that apparently (mixed sources) may or may not support the various weighting statements that are required for complex survey data, while also supporting log binomial regression. Which has led me to %SURVEYGENMOD, which is a privately-developed macro that I can't seem to get ahold of the creators to ask about potential sharing (even though I've seen other papers use it for complex survey data).