topkatz@MSN.COM wrote: >PROCs GENMOD and SURVEYLOGISTIC each have certain useful features that the >other one doesn't have. I'm trying to find out which set of capabilities >is more important for my data. This question is somewhat moot for me, >since I don't have SAS 9, so I can't use PROC SURVEYLOGISTIC. But maybe I >can learn something, anyway. > >I have data in which many subjects have repeated observations. I am >modeling a binary dependent variable, and wanted to use PROC GENMOD for >its GEE capabilities. The observations are from direct marketing contacts >to try to enroll the subjects into a program. A subject may be contacted >several times, but such contacts will cease upon enrollment in the >program. The binary dependent variable is for enrollment / non-enrollment. > >But all of these marketing contacts come from individual campaigns, each >with multi-cell design plans. Which is why I was curious about the >possibility of applying PROC SURVEYLOGISTIC. > >I'm not sure whether the effect of the designs is more or less important >than the dependence structure of multiple observations per subject. Any >advice, ideas, poems, threats, etc. are welcome. Thanks!

Okay, but remember. You asked for it.

Advice: You may need to try both PROC GENMOD and PROC SURVEYLOGISTIC in order to see how crucial the design effects really are. The design effects can be anywhere from too-minor-to-worry-about up to cannot-get-anywhere-near- correct-answers-without-taking-survey-design-into-account. If you have K survey designs, you also have to worry about whether they cover overlapping parts of the target population, or if they are part of a partition of the target population. In case #1, you have to treat them as independent re-samples of the same (or close to same) population, and you have something analogous to replicate sampling. In case #2, you have something that you may be able to model as a big stratfiied sample.

Ideas: You are going to have to think about the details of these individual samples whichever way you go. So start by pestering the data sources for mroe details on the design specs. See if you can get them to confess about their target populations, the way they built their sample frames, the (exact) way they drew the samples from those frames, how many sample points they lost and why, ...

Poem: There once was a sampler named Katz whose data were simply ersatz. The samples were icky, the math got real sticky, he threw up his hands and said "Rats!"

Threat: Today's threat potential is a soft mauve, suitable for covering that living room wall without a window.

Oh, and if you have to threaten the data sources, I've always been partial to "Ve know you haff relatives in Stuttgart!" (Gestapo officer, "Invisible Agent", 1942. It's the one Cheech and Chong made fun of.)

David -- David L. Cassell mathematical statistician Design Pathways 3115 NW Norwood Pl. Corvallis OR 97330

_________________________________________________________________ Express yourself instantly with MSN Messenger! Download today - it's FREE! http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/