Perhaps even more to the point, "covariate adjustment" and "classification" should not be separate. One should fit the appropriate model that does both.
-- Bert On Thu, Sep 2, 2010 at 11:34 AM, Ben Bolker <bbol...@gmail.com> wrote: > On 10-09-02 02:26 PM, James Nead wrote: >> My apologies - I have made this more confusing than it needs to be. >> >> I had microarray gene expression data which I want to use for >> classification algorithms. However, I want to 'adjust' the data for >> all confounding factors (such as age, experiment number etc.), before >> I could use the data as input for the classification algorithms. Since >> the phenotype is known to be effected by age, I thought that this >> would be a fixed effect whereas something like 'beadchip' would be a >> random effect. >> >> Should I be looking at something else for this? >> > > Sounds to me as though you should use residuals() rather than fitted() > if you want to "adjust for confounding factors". > > But since you've made up a nice small example, I think you should look > at the results > of fitted() and residuals() > for your example and see if it's doing what you want. >> >> >> ------------------------------------------------------------------------ >> *From:* Ben Bolker <bbol...@gmail.com> >> *To:* r-h...@stat.math.ethz.ch >> *Sent:* Thu, September 2, 2010 2:06:47 PM >> *Subject:* Re: [R] Linear models (lme4) - basic question >> >> James Nead <james_nead <at> yahoo.com <http://yahoo.com>> writes: >> >> > >> > Sorry, forgot to mention that the processed data will be used as >> input for a >> > classification algorithm. So, I need to adjust for known effects >> before I can >> > use the data. >> > >> > > I am trying to adjust raw data for both fixed and mixed effects. >> > The data that I >> > > output should account for these effects, so that I can use >> > the adjusted data >> > >for >> > > further analysis. >> > > >> > > For example, if I have the blood sugar levels for 30 patients, >> > and I know that >> > > 'weight' is a fixed effect and that 'height' is a random effect, >> > what I'd want >> > > as output is blood sugar levels that have been adjusted for these >> effects. >> >> What's not clear to me is what you mean by 'adjusted for'. >> fitted(lm.adj) will give predicted values based on the height >> and weight. I don't really know what the justification for/meaning >> of the adjustment is, so I don't know whether you want to predict >> on the basis of the heights, or whether you want to get a >> 'population-level' >> prediction, i.e. one with height effects set to zero. Maybe you want >> residuals(lm.adj) ...? >> >> I suggest that follow-ups go to r-sig-mixed-mod...@r-project.org >> <mailto:r-sig-mixed-mod...@r-project.org> >> >> ______________________________________________ >> R-help@r-project.org <mailto:R-help@r-project.org> mailing list >> https://stat.ethz.ch/mailman/listinfo/r-help >> PLEASE do read the posting guide >> http://www.R-project.org/posting-guide.html >> and provide commented, minimal, self-contained, reproducible code. >> > > > [[alternative HTML version deleted]] > > ______________________________________________ > R-help@r-project.org mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide http://www.R-project.org/posting-guide.html > and provide commented, minimal, self-contained, reproducible code. > -- Bert Gunter Genentech Nonclinical Biostatistics 467-7374 http://devo.gene.com/groups/devo/depts/ncb/home.shtml ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
