[R-sig-ME] variance explained by fixed effects in mixed model

[Steven J. Pierce]

Ellen,

My take on this is that the plethora of competing formulas intended to
compute some analogue of R-square for mixed models (or even more
problematically, for generalized linear mixed models) is a sign that finding
such a measure is not simple. I have informally compared the results of
using various formulas on a set of models (fitted to the same data) and
found that they don’t all agree, and sometimes some of them yield results
that seem inconsistent with other ways of examining model fit, quality, or
comparisons between models. I'm not convinced there is a consensus that
looking for such a measure that generalizes to mixed and generalized mixed
models is even the right thing to do in the first place. 

Perhaps you could tell us what you would actually use the R-square for and
see if the folks on the list can suggest better methods to accomplish the
same goal. 

Steven J. Pierce, Ph.D. 
Associate Director 
Center for Statistical Training & Consulting (CSTAT) 
Michigan State University 
E-mail: pierces1 at msu.edu 
Web: http://www.cstat.msu.edu 

-----Original Message-----
From: Ellen Pape [mailto:ellen.pape at gmail.com] 
Sent: Wednesday, February 20, 2013 2:39 AM
To: r-sig-mixed-models at r-project.org
Subject: [R-sig-ME] variance explained by fixed effects in mixed model

Dear all,

What is the best and easiest way to calculate an R² to indicate the
variance explained by the fixed effects in a mixed model? I have read some
papers with a lot of seemingly complicated formulas, but I just want to
know how to do it in R? (because that is not always clear to me :s)

Thanks!
Ellen

	[[alternative HTML version deleted]]