topic Re: Should I suppress the intercept when using the strata statement in proc surveylogistic? in Statistical Procedures

Should I suppress the intercept when using the strata statement in proc surveylogistic?

rachard11 — Sat, 08 Jul 2017 15:58:25 GMT

Do I need to suppress the intercept in proc surveylogistic when using the strata statement to avoid the dummy variable trap (SAS 9.4)? I see that when I use the strata statement with proc logistic, the model automatically suppresses the intercept. When I use the strata statement with proc surveylogistic, the intercept is not suppressed, and I get very different coefficient estimates on other variables of interest. Why do proc logistic and proc surveylogistic handle the intercept differently under the Strata statement?

Re: Should I suppress the intercept when using the strata statement in proc surveylogistic?

SAS_Rob — Fri, 21 Jul 2017 13:50:06 GMT

The STRATA statement in the two procedures have very different functionality attached to them. In Proc SURVEYLOGISTIC, it is used to identify the strata for a complex survey design. In Proc LOGISTIC, it is used to idenitfy the matched pairs for running models with n:m matching.

Put another way, the STRATA statement in LOGISTIC runs a stratified logistic regression while in SURVEYLOGISTIC it runs a logistic regression from a stratified sample.

Unless you had reason to restrict the intercept to zero in SURVEYLOGISTIC, you would not want to use the NOINT option.

Re: Should I suppress the intercept when using the strata statement in proc surveylogistic?

rachard11 — Fri, 21 Jul 2017 15:04:18 GMT

Thank you. My understanding is that the strata statement in proc logistic is what I should use to implement a fixed effects model with panel data. Is there a way to also cluster standard errors with this model in SAS? This was my intent when I originally tried using the strata statement in proc surveylogistic (along with the cluster statement).

Re: Should I suppress the intercept when using the strata statement in proc surveylogistic?

SAS_Rob — Fri, 21 Jul 2017 17:20:27 GMT

You might also consider looking at GENMOD with the REPEATED statement because GEE models do not require estimation of a parameter for each set of correlated observations (i.e. panels).

LOGISTIC with the STRATA statement uses a conditional model (or GENMOD with the STRATA and EXACT statements in GENMOD for exact, conditional estimation for other models--this approach obviously wouldn't work with the REPEATED statement)

Either way SURVEYLOGISTIC would not be what you would want to use.as can be done by using the STRATA statement in PROC LOGISTIC for logistic models