[R-sig-ME] Fwd: lme4 heteroscedasticity???

[Highland Statistics Ltd]

------------------------------

Message: 3
Date: Mon, 13 Oct 2014 10:17:54 -0500
From: Douglas Bates <bates at stat.wisc.edu>
To: David Cox <dac64 at cam.ac.uk>
Cc: R-mixed models mailing list <r-sig-mixed-models at r-project.org>
Subject: Re: [R-sig-ME] varFixed function query in lme
Message-ID:
	<CAO7JsnQpZhGyfcTf+RDUT5mDoht8L=UtvXS7+vtF0DB3LMJaZA at mail.gmail.com>
Content-Type: text/plain; charset="UTF-8"

It is best to send inquiries like this to the
R-SIG-Mixed-Models at R-project.org mailing list, which I am cc:ing on this
reply.  I am no longer the maintainer of the nlme package - I was pushed
aside by R Core many years ago and I really don't know what changes have
been made since they took over.

On Mon, Oct 13, 2014 at 8:46 AM, David Cox <dac64 at cam.ac.uk> wrote:

> Dear Professor Bates,
>
> I'm a neuroscience PhD student at Cambridge University. I saw that you
> maintain the 'varFixed' function for the 'weights' option in the nlme
> library. I have been reading the help manual but I'm a little stuck on a
> problem and wondered if you might have any quick suggestions.
>
> I'm using lme to model the peptide sequences of a set of proteins for
> cancer patients and healthy individuals. I want to use the weights function
> so that the peptide intensities are inversely weighted with the variance.
> The higher the variance, the lower the weighting etc. As the peptide
> intensities of cancer patients and healthy individuals will be different, I
> want to apply this weighting separately for each group.
>
> At the moment I've tried with a model like this for the 1st protein:
>  Peptide Intensities ~ Covariates + Peptide Sequences + Group, random =
> Sample Id, data = data, weights = ~ Group
>
> Group denotes whether each intensity is a cancer patient/healthy person.
> Sample id is the id of each cancer/healthy sample.
>
> This doesn't work though. I get an error saying: "Error in
> Math.factor(attr(object, "covariate")) :    abs not meaningful for factors."
>
> Many Thanks
>
> David Cox
>


This would be an easy exercise in JAGS or OpenBUGS. I believe there is a linear regression + multiple sigmas example (varIdent) in:

Introduction to WinBUGS for Ecologists: Bayesian approach to regression, ANOVA, mixed models and related analyses
Marc Kery

Adding two crossed random effects is not difficult neither.

Kind regards,

Alain


  





	[[alternative HTML version deleted]]