topic Chi-square tests for multi-way tables using complex surve data? in Statistical Procedures

Chi-square tests for multi-way tables using complex surve data?

SAS93 — Sat, 13 Feb 2021 15:40:33 GMT

I know how to use proc surveyfreq and specify the chi-sqare (chisq) option for looking at associations.

I usually use the Rao-Scott chi-square output, even for multi-way tables. I've never been taught/told not to, but is this actually correct? Is there some other way to specifically look at tables bigger than 2-way tables?

Re: Chi-square tests for multi-way tables using complex surve data?

StatDave — Sat, 13 Feb 2021 21:48:14 GMT

If one variable is a response variable and the others are predictors of that response, then this is usually done by logistic regression. So, you could use PROC SURVEYLOGISTIC.

Re: Chi-square tests for multi-way tables using complex surve data?

SAS_Rob — Mon, 15 Feb 2021 16:06:16 GMT

SURVEYFREQ does not have a test that is analogous to the CMH test in Proc FREQ. I have not seen much discussion of this in the survey sampling literature. Essentially it will treat each level of the third (or fourth, etc.) variable as a separate domain and perform a Chi-Square test on each unique two-way table

Re: Chi-square tests for multi-way tables using complex surve data?

SteveDenham — Tue, 16 Feb 2021 13:04:01 GMT

This in turn makes @StatDave 's suggestion regarding SURVEYLOGISTIC about the only alternative. Including an interaction term should provide an approximation to a CMH analysis.

SteveDenham