[R-lang] bootstrap vs. mixed model

[T. Florian Jaeger]

Hey R-lang folks,

does anybody know of a good reference that *directly* compares bootstrap vs.
mixed effect models as approaches to clustering for data (I'd be interested
in comparisons both for continuous and binary categorical variables)? More
specifically, I've been asked how these methods compare depending on the
cluster size distribution, e.g. how does bootstrap with random speaker
cluster replacement over an lm model compare to an lmer model with random
speaker effects? I've definitely seen for my own data that the two
approaches can yield different results (especially, if there is collinearity
between the fixed effect and the random effect in the mixed effect model).

I've seen that at least some people (Harald, are you reading this?) seem to
prefer bootstrap for lots of small clusters, presumably because it's hard to
fit good random intercepts if there's only one data point for each level of
the random effect. I'd be interested to hear your guys's (love that one)
ideas about this. are there any references?

cheers,

florian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://ling.ucsd.edu/pipermail/r-lang/attachments/20070807/2e90b4e8/attachment.htm