[R-sig-ME] Excluding random intercepts?
bosaylor at yahoo.com
Fri Dec 7 23:47:47 CET 2012
I think the comment from Tom is a good place to start thinking about your
question. In addition, Gelman and Hill (2007) address your question
specifically on page 283. Their short answer is that when a slope is allowed to
vary it makes sense to for the intercept to vary too, and that it is an
implausible scenario where intercepts do not vary but slopes for some x do.
Nevertheless, they go on to provide an example (an experimental design) where it
does make sense.
More information about the R-sig-mixed-models