[R-sig-ME] Testing Random Effects--On the Boundary

[Ben Bolker]

On 13-11-21 03:21 PM, Thompson,Paul wrote:
> I have never heard of a rule of "dividing the p values in half".
> There are corrections like Bonferroni but these depend on the number
> of tests.
> 

  See my answer to this question ...  The "divide by half rule" applies
when the null value of a single parameter in a linear mixed model is on
the boundary (i.e. H0: variance=0). It's not at all related to multiple
comparisons corrections, but to a modification of the sampling
distribution expected under the null hypothesis, from likelihood theory ...

  cheers
    Ben Bolker