[R-sig-ME] residual variance estimates fixed to 1

[Jake Westfall]

Ah, I missed the detail about family=binomial. That is definitely what's
going on.

Jake

On Wed, Jan 24, 2018 at 5:02 PM, Viechtbauer Wolfgang (SP) <
wolfgang.viechtbauer at maastrichtuniversity.nl> wrote:

> For glmer() with family=binomial, there is no 'residual variance'. If you
> use sigma() to extract it, it will always return 1.
>
> Best,
> Wolfgang
>
> >-----Original Message-----
> >From: R-sig-mixed-models [mailto:r-sig-mixed-models-bounces at r-
> >project.org] On Behalf Of Jake Westfall
> >Sent: Wednesday, 24 January, 2018 23:41
> >To: Lindner, Melanie
> >Cc: r-sig-mixed-models at r-project.org
> >Subject: Re: [R-sig-ME] residual variance estimates fixed to 1
> >
> >Hi Melanie,
> >
> >As far as I know, lmer never fixes the residual variance to 1 or any
> >other
> >value, and in fact this isn't even possible to do with lmer (at least not
> >without resorting to add-on packages). My guess is that in your loop you
> >accidentally grabbed the wrong field, not the variable giving the
> >residual
> >variance estimate. If you give us a sample of the code you used, we could
> >help figure out what happened.
> >
> >Jake
> >
> >On Wed, Jan 24, 2018 at 12:02 PM, Lindner, Melanie <
> >melanie.lindner at helsinki.fi> wrote:
> >
> >> Hi again,
> >>
> >> I use lme4 to model methylation count data. I specify my response as
> >> cbind(methylation count, unmethylation count) and use the argument
> >> family=binomial.
> >> In my data set I have count information for 500,000 CpG sites over 64
> >> samples (so, each sample contains count information for all 500,000
> >sites).
> >> 32 per treatment group. I model each site separately to see if there is
> >a
> >> significant difference between the treatments and therefore use a loop.
> >> Since I cannot look at the summary of each site, I saved different
> >> estimates from the loop and recognised that the residual standard
> >deviation
> >> is always 1. To evaluate the model fit, I would like to understand why
> >the
> >> residual variance is fixed to one. It would be great if someone can
> >tell me
> >> where to find information on that.
> >>
> >> Thanks in advance,
> >> Melanie
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>

	[[alternative HTML version deleted]]