[R-sig-ME] negative estimate from sd(resid(fit))

[Rolf Turner]

On 05/10/11 21:14, Phil Smith wrote:
> Hi lmer Club:
>
> I've got a model like lmer( y ~ x , 1|cluster).
>
> I'm permuting the cluster values 1000 times to obtain the nonparametric
> distribution of sigma under the null hypothesis that sigma = 0.
>
> I'm getting alot of negative values!!
>
> Am I doing something wrong, or what is going on there??
>
> Please reply directly to Phil.Smith.Duluth.GA at gmail.com  .... I'm not too
> good at understanding listserves.

How negative is negative?  Since you are testing sigma = 0, one
would expect your estimates to be close to 0, and hence to (at
least sometimes) *be* 0 --- and hence appear equal to very
small negative values due to numerical noise.

     cheers,

         Rolf Turner