** Send unsubscribe or help commands to mailto:majordomo@hsphsun2.harvard.edu **
LSMEANS in SAS is a bit tricky. The default setup ives you adjusted means
assuming orthogonality. That is, if the design is unbalananced,
you get an adjusted set of means under an assumption of orthoronality.
This has been the default for years. If the design is orthogonal it is
of no consequence, but it is a serious issue if you analyze survey data.
Many users have not appreciated this, thinking they are getting a
standard ANCOVA adjustment when they were getting something rather
different.
In version 6.12 of SAS the problem has been corrected by allowing
an OBSMARGINS statement on LSMEANS which "is reasonable when you want the
inferences to apply to a population that is not necessarily balanced bu
has the margins observed in your data set." (SAS/STAT Software" Changes
and Enhancements trough Release 6.12, p. 353). Note, however, that the
default is to assume a balanced design.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Richard T. Campbell
Professor of Sociology
Director, Research Methods Core
Health Research and Policy Centers (HRPC)
E-mail: mailto:dcamp@uic.edu
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dept. of Sociology M/C 312 HRPC M/C 275
1007 W. Harrison St. 850 W. Jackson Blvd
Chicago IL 60607-7140 Chicago IL 60607-3025
312.413.3759 312.413.0480
FAX: 312.996.5104 FAX: 312.996.2703
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
