home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!europa.asd.contel.com!paladin.american.edu!auvm!NIOBBS1.EM.CDC.GOV!SIS1
- Return-Receipt-To: <sis1@NIOBBS1.EM.CDC.GOV>
- Encoding: 63 TEXT
- X-Mailer: Microsoft Mail V3.0 (beta-2)
- Message-ID: <2B0267C5@router.em.cdc.gov>
- Newsgroups: bit.listserv.stat-l
- Date: Thu, 12 Nov 1992 18:11:00 EST
- Sender: "STATISTICAL CONSULTING" <STAT-L@MCGILL1.BITNET>
- From: sis1@NIOBBS1.EM.CDC.GOV
- Subject: Categorical data with repeated measures.
- Lines: 58
-
- Walter R. Pirie writes
-
- > Does anyone know anything that's been done on repeated measures in
- > categorical data? It's a problem that occurs fairly often and seems
- > difficult to deal with.
-
- The definitive approach, as I have heard from many sources, is Generalized
- Estimating Equations (GEE). The primary references are:
-
- Liang and Zeger "Longitudinal data analysis using generalized linear
- models" Biometrika (1986):73;13-22.
-
- Zeger and Liang "Longitudinal Data Analysis for Discrete and Continuous
- Outcomes" Biometrics (1986):42;131-130.
-
- A lot has been published since then, so you might want to check a citation
- index. SPIDA can do this analysis; and macros for GEE are written for S and
- for SAS. The GEE model can handle time-varying covariates very nicely, but
- cannot handle large numbers of repeated measures per subject (some of my data
- sets have 200 measures per subject). It can handle unequal numbers of
- repeated measures per subject.
-
- Another approach, which is actually better for clustered data than for
- repeated measures, is the random effects model. This includes the
- beta-binomial model, which has been known for a long time, and the
- normal-logistic model which was developed recently. This appears to be good
- for large number of repeated measures and for unequal numbers of repeated
- measures per subject. I don't think it allows for time varying covariates.
- I don't have a reference, but the Egret software package does these analyses.
-
- The CATMOD procedure in SAS (as mentioned by someone else on this list) is a
- weighted least squares approach that typically falls apart when there are
- small expected counts and/or zero counts. Also, I don't think this approach
- can handle unequal numbers of repeated measures per subject.
-
- Kleinbaum and Kupper give a very nice short course on this subject, and much
- of the material I am quoting I learned during the course. I hope I got it
- right.
-
- * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
- * Steve Simon, Mail Stop C-22 * Delivering on the *
- * Chief, Statistics Activity * nation's promise: *
- * National Institute for Occupational * safety and health *
- * Safety and Health * at work for all *
- * 4676 Columbia Parkway * people...through *
- * Cincinnati, OH 45226 * prevention. *
- * * *
- * 513-533-8271 (office) * NIOSH vision *
- * 513-533-8510 (FAX) * statement *
- * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
- * SMTP gateway to Internet *
- * sis1@niobbs1.em.cdc.gov *
- * X400 gateway to Internet *
- * hhscdc!sis1/o=cdc/ou=cinc/ou=niobbs1@mhs.attmail.com *
- * *
- * Views expressed here are my own. Mention of a product or *
- * company name does not constitute an endorsement by NIOSH. *
- * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
-