topic Re: Limiting Analysis in Logistic regression: Complex Survey Data in SAS Programming

Limiting Analysis in Logistic regression: Complex Survey Data

SAS93 — Fri, 12 Jun 2020 14:21:19 GMT

I'm using proc survey logistic to find the odds ratios for a given dichotomous outcome given a particular variable of interest while also controlling for several other confounding variables (age, etc.). Needless to say, I'm using complex survey data, which leaves me with a problem.

I'm trying to limit my analysis to a specific group of respondents (let's say people who bought ice cream in a cone and added chocolate sprinkles). For initial prevalence estimates, I'm using domain analysis. But I know with complex survey data, you're not supposed to set datasets using BY variables because it messes up the estimates; what is an option for running logistic regression while trying to limit the analysis to a certain group? Would also controlling for that specific variable--adding it alongside the others--work? Can I add it as a domain variable? I have several variables I want to control for, so running proc surveyfreq with multiple cross-tabbed domains would be messy.

Re: Limiting Analysis in Logistic regression: Complex Survey Data

ballardw — Fri, 12 Jun 2020 15:16:24 GMT

Create a class variable to identify the population of interest with an additional level of "not interested", place the variable on the Domain statement, ignore the results for the "not interested" levels.

Re: Limiting Analysis in Logistic regression: Complex Survey Data

SAS93 — Fri, 12 Jun 2020 15:24:30 GMT

Update: solved my own problem. 🙂

I wasn't aware you could also use domain statements with regression.

Note to anyone else who might have this issue in the future: you can.

Re: Limiting Analysis in Logistic regression: Complex Survey Data

SAS93 — Fri, 12 Jun 2020 15:26:15 GMT

Thank you!

I guess I wasn't aware you could also use domain statements in regression analysis. I'm still new to the inter-workings of complex survey data, but this makes sense for sure.