Message-ID: <20110129145812.10442al9nqj6j6sk@www.staffmail.ed.ac.uk>
Date: 2011-01-29T14:58:12Z
From: Jarrod Hadfield
Subject: Is it possible to weight residuals in MCMCglmm?
In-Reply-To: <AANLkTimcEb1emki0iUntVEcd9v5z96qe1FyuMaqkHCbv@mail.gmail.com>
Hi,
It is not possible currently - the inflexibility of the residual
structure was due to an oversight when I was writing MCMCglmm which I
haven't had time to go back and change. As I'm sure you are aware
there are three options, none of which are exactly what you want (I
think).
Lets have w as a set of values associated with each observation and
are proportional to the standard error of each measurement (m).
The three options are
a) specifying mev=w^2 which fits:
VAR(m_{i}) = w_{i}^2+VAR(e)
b) specifying random=~leg(sqrt(w), -1, FALSE):units which fits:
VAR(m_{i}) = (c*w_{i})^2+VAR(e)
where c is the square root of the variance associated with the random effects.
c) specifying rcov=~idh(units):units and prior=list(diag(w^2), fix=1)
which fits:
VAR(m_{i}) = w_{i}^2
where c is the square root of the variance associated with the random effects.
I think you would like
VAR(m_{i}) = (c*w_{i})^2
which is not possible
Cheers,
Jarrod
Quoting Szymek Drobniak <geralttee at gmail.com>:
> Is there any simple way to fit weighted residuals in MCMCglmm? I'd like to
> specify model where I could give different data points different influence
> on the fit of the model. Is there any way of doing this beside of course
> actually doing sth to the data? I was thinking of modifying the
> meta-analytical approach but I got stuck there...
>
> Cheers,
> sz.
>
> --
> Szymon Drobniak || Population Ecology Group
> *Institute of Environmental Sciences, Jagiellonian University
> ul. Gronostajowa 7, 30-387 Krak?w, POLAND
> *tel.: +48 12 664 52 19 fax: +48 12 664 69 12
>
> www.eko.uj.edu.pl/drobniak
>
> [[alternative HTML version deleted]]
>
>
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.