[R-sig-ME] mixed-effects model specification question

[Andrew Robinson]

Hi Tony,

On Tue, May 06, 2008 at 03:42:07PM -0500, Antonio_Paredes at aphis.usda.gov wrote:
>    How were animals allocated to treatments (at random)? What about the
>    housing of the animals. Both of these factors could be influential in
>    selecting the unit of study.

you're absolutely right.  My response was based on the information
presented, but there could have been relevant information omitted.

>    You also need to get an ideal at to why a p-value is enough to support
>    the objective of the study. Why not estimation of treatment effects?
>    Tony.

again, you are correct.  I'm kicking myself for missing that angle.
Treatment effects and confidence intervals should also be considered.

Cheers

Andrew

>    Andrew Robinson <A.Robinson at ms.unimelb.edu.au>
>    Sent by: r-sig-mixed-models-bounces at r-project.org
> 
>    05/06/2008 03:19 PM
> 
>                                                                         To
> 
>    Mark Kimpel <mwkimpel at gmail.com>
> 
>                                                                         cc
> 
>    r-sig-mixed-models at r-project.org
> 
>                                                                    Subject
> 
>    Re: [R-sig-ME] mixed-effects model specification question
> 
>    Mark,
>    You should talk to a local statistician about this, but I think that
>    you can probably average across the measurements within each rat, if
>    all you are interested in is a treatment effect.  For the analysis of
>    treatment the relevant unit of replication should be the rat, in any
>    case.
>    (Does anyone have any thoughts on why that might be a bad idea?)
>    Also if I understand your design, there are three batches per rat.  I
>    suspect that using Rat/Tissue would lead to an over-parametrized
>    model, if my interpretation is correct.  My guess is that Rat should
>    be adequate.
>    Final points
>    1) My speculations would be better-informed if you showed us the
>      output of the model that you proposed - fyi :) .
>    2) You could try all three configurations and see if it makes any
>      difference to the inference of interest.  I suspect that it will
>      not.
>    I hope that this helps,
>    Andrew
>    On Tue, May 06, 2008 at 10:23:08AM -0400, Mark Kimpel wrote:
>    > Perhaps a bit of a newbie question, but I need to get this right. I
>    need to
>    > make sure I am specifying a model correctly. Here is our
>    less-than-perfect
>    > experimental design:
>    >
>    > 36 rats divided into two treatment groups, i.e. 18 per group
>    >
>    > each rat has measurements taken on 3 brain regions, but each brain
>    region
>    > was analyzed in a separate batch (there are strong batch effects) so
>    we
>    > really can't compare the regions per se, but do recognize that the 3
>    regions
>    > from a single rat have variance above and beyond that accounted for
>    by
>    > technical factors. Since, because of the batch effect we are not
>    going to
>    > look at the effect of brain region, I "think" this should be a
>    considered a
>    > random effect.
>    >
>    > So I have rat, treatment, and region(batch) as variables. The only
>    thing I
>    > am interested in getting a p-value for is treatment.
>    >
>    > Is the model below correct and can I squeek by with using nlme to get
>    a
>    > p-value (notwithstanding recent threads on this list)?
>    >
>    >  myModel <- lme(fixed = geneExpression ~  Treatment,  random = ~1 |
>    > Rat/Tissue)
>    >
>    > Thanks,
>    > Mark
>    > --
>    > Mark W. Kimpel MD ** Neuroinformatics ** Dept. of Psychiatry
>    > Indiana University School of Medicine
>    >
>    > 15032 Hunter Court, Westfield, IN 46074
>    >
>    > (317) 490-5129 Work, & Mobile & VoiceMail
>    > (317) 663-0513 Home (no voice mail please)
>    >
>    > ******************************************************************
>    >
>    >                  [[alternative HTML version deleted]]
>    >
>    > _______________________________________________
>    > R-sig-mixed-models at r-project.org mailing list
>    > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>    --
>    Andrew Robinson
>    Department of Mathematics and Statistics            Tel:
>    +61-3-8344-6410
>    University of Melbourne, VIC 3010 Australia         Fax:
>    +61-3-8344-4599
>    http://www.ms.unimelb.edu.au/~andrewpr
>    http://blogs.mbs.edu/fishing-in-the-bay/
>    _______________________________________________
>    R-sig-mixed-models at r-project.org mailing list
>    https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

-- 
Andrew Robinson  
Department of Mathematics and Statistics            Tel: +61-3-8344-6410
University of Melbourne, VIC 3010 Australia         Fax: +61-3-8344-4599
http://www.ms.unimelb.edu.au/~andrewpr
http://blogs.mbs.edu/fishing-in-the-bay/