Hello Allstat,
I'm teaching a course on model selection and averaging based on AIC this
semester. I can compute AIC if I have the Sum of Squares Error (SSE),
or alternatively a model likelihood. My problem is that there is no
summary SSE for MANOVA or discriminant analysis... only an error SSCP
matrix. And some of my students want to use discriminant analysis for
their projects this term.
My question is whether it is possible to summarize the SSCP matrix into
a single SSE statistic that is exactly equivalent (or at least pretty
darn close) to the SSE in a general linear model. My intuition is that
summing the diagonal of the error SSCP matrix should give me the
statistic I want... but I wasn't sure if I also needed to add (or
possibly even subtract) the cross-product terms.
I'd appreciate any light this list could share on this issue!
Thanks,
Brian
--
----------------------------------------------------------
Brian R. Mitchell
Post-Doctoral Associate
University of Vermont
The Rubenstein School of Environment and Natural Resources
81 Carrigan Drive
Burlington, VT 05405-0088
(802) 656-2496
[log in to unmask]
----------------------------------------------------------
|