surveyfreq chi square

[EddieJackson]

Quick question about the chi square from the surveyfreq command,

I'll enter something like the following:

PROC SURVEYFREQ DATA= code;
CLUSTER xxxx;
STRATA xxx;
weight xx;
TABLE obesity*gender/chisq;
RUN;

I get a chi square result. However if I double check this result with any online 2x2 table calculator, the p-value is different.

However if I do a regular PROC FREQ chisquare instead of surveyfreq, the p-value checks out with all those online calculators.

Am I doing something wrong with the surveyfreq command? Or is the calculation intrinsically different for a surveyfreq such that it won't check out with the online calculators like regular proc freq does?

Thanks

[Watts]

Yes, it's a different statistic. The CHISQ option in PROC SURVEYFREQ provides a Rao-Scott chi-square test, which is a design-adjusted version of the Pearson chi-square test. The Rao-Scott chi-square test takes into account the complex survey design that you specify (in design options and statements -- e.g., STRATA and CLUSTER statements).

The CHISQ option in PROC FREQ provides the (unadjusted) Pearson chi-square.

For more info, the PROC SURVEYFREQ documentation is here .

[Reeza]

The marjority of online calculators do not take into account the weight and strata components. So the question is really, do you need those components and are you using them correctly.

[EddieJackson]

Thanks!