[R] Robustness of linear mixed models

From: John Maindonald <john.maindonald_at_anu.edu.au>
Date: Wed 28 Jun 2006 - 07:54:50 EST

I'd use mcmcsamp() to examine the posterior distribution, under a relatively uninformative prior, of of the parameter estimates. For estimates that are based on four or five or more "degrees of freedom", I'd surmise that the prior will not matter too much. With estimates where the number of "degrees of freedom" is one or two or three, the posterior distribution may vary greatly from one run of mcmcsamp() to another. Of course, the definition of "degrees of freedom" becomes quite fraught if there is severe imbalance; e.g., some of the items that contribute to the estimate based on much more data than others.

Subject to such caveats as just noted, I'd expect that credible intervals derived from the posterior distributions would be close to the usual frequentist confidence intervals. The main effect of the non-normality may be that the estimates are "inefficient", i.e., the variance may be larger, or the distributions more dispersed, than for "true" maximum likelihood estimates, were you able to obtain them!
John Maindonald

John Maindonald email: john.maindonald@anu.edu.au phone : +61 2 (6125)3473 fax : +61 2(6125)5549 Mathematical Sciences Institute, Room 1194, John Dedman Mathematical Sciences Building (Building 27) Australian National University, Canberra ACT 0200.

On 27 Jun 2006, at 8:00 PM, r-help-request@stat.math.ethz.ch wrote:

> From: "Bruno L. Giordano" <bruno.giordano@music.mcgill.ca>
> Date: 27 June 2006 11:21:25 AM
> To: r-help@stat.math.ethz.ch
> Subject: [R] Robustness of linear mixed models
>
>
> Hello,
>
> with 4 different linear mixed models (continuous dependent) I find
> that my residuals do not follow the normality assumption
> (significant Shapiro-Wilk with values equal/higher than 0.976;
> sample sizes 750 or 1200). I find, instead, that my residuals are
> really well fitted by a t distribution with dofs' ranging, in the
> different datasets, from 5 to 12.
>
> Should this be considered such a severe violation of the normality
> assumption as to make model-based inferences invalid?
>
> Thanks a lot,
> Bruno

        [[alternative HTML version deleted]]



R-help@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Wed Jun 28 08:00:46 2006

Archive maintained by Robert King, hosted by the discipline of statistics at the University of Newcastle, Australia.
Archive generated by hypermail 2.1.8, at Wed 28 Jun 2006 - 10:13:38 EST.

Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-help. Please read the posting guide before posting to the list.