marginal means?
Robert Ellis
I'm running a mixed Anova on SPSS 10 and getting different F values from
what I get when I run the same data on SuperAnova. Further, when I ask SPSS
to display Means - even they are different from the means SuperAnova
computes. To do a check, I impoted some of the data into Excel and did a
straightforward Means calculation which resulted in the means that
SuperAnova reports. In the means that SPSS reports there is a subscript that
says "based on modified population marginal mean". Does anyone know what
this is about? Would it account for the different F values repoted by the
two stats packages?
Robert
Bruce Weaver
Hi Robert,
You can find what SPSS says about estimated marginal means on page
244 of this document from the algorithms page:
http://www.spss.com/tech/stat/algorithms/glm_u_m.pdf
Personally, I'm not mathematically adept enough to get much out of that
explanation. But I think it means that the estimated marginal means are
"predicted" means rather than observed means--and your other package may
be giving you the latter.
One other thing I've noticed is that the estimated marginal means for main
effects appear to be "unweighted" means (i.e., the mean of some cell
means), so if you have an unbalanced design, your estimated marginal means
for main effects will be different than what you get from DESCRIPTIVES or
MEANS.
Hope this helps.
Cheers,
Bruce
--
Bruce Weaver
New e-mail: wea...@mcmaster.ca (formerly wea...@fhs.csu.mcmaster.ca)
Homepage: http://www.angelfire.com/wv/bwhomedir/