[R-sig-ME] lme4::mcmcsamp + coda::HPDinterval

Douglas Bates bates at stat.wisc.edu
Thu Apr 3 00:27:00 CEST 2008

I enclose another plot from a simulation that may show why I refer to
"the flat spot".  In this case I simulate a simple model with one
fixed effect and with random effects for a single grouping factor.  I
am simulating from the null model where the variance of the random
effect is zero.  The first two simulations produce an estimated
variance of zero - i.e. they converge on the boundary of the parameter
space.  The third model converges to a non-zero estimate.  The
profiled REML deviance, as a function of the logarithm of the relative
standard deviation has only a shallow dip for the optimum and becomes
flat for large negative values of the logarithm.  If the MCMC sampler
gets onto that plateau it has a hard time getting off again.

More information about the R-sig-mixed-models mailing list