Re: p-values for variance components in lme

["Douglas Bates" <bates@stat.wisc.edu>]

On 11/17/06, Roberts, J. Kyle <jkrobert@bcm.tmc.edu> wrote:

> Does anyone know how to test for the statistical significance of variance
> components in lme (actually produce the p-values)?  I have an editor asking
> for them.
Fit the model with and without the variance component and compare the
two model fits using the anova function.  Because the null hypothesis
of the variance component being zero is on the boundary of the
parameter space, the p-value obtained from a chi-square distribution
with one degree of freedom will be conservative in the sense that it
is an upper bound on the true p-value for the test.

If removing the variance component results in a model without any
random effects you need to fit the reduced model using lm and you must
list the lme model before the lm model in the call to anova.

I hope this helps.

Doug Bates