Skip to content
Prev 10935 / 20628 Next

Zero random effect variance?

On 13-10-17 08:28 PM, Dave Marvin wrote:
Yes.  This is a case, I think (referred to from time to time in
threads on this list), where the classical method of moments estimates
would give a negative among-group variance, or a compound symmetry model
would give negative within-group correlations; the framework used in
lme4 can't do either of those things easily.  (It would be entertaining
but totally impractical to try to figure out what kind of imaginary- or
complex-valued values one would need in the computations to get this to
work out).

  Ben Bolker