[R] the numimum number of fixed factors lme package can deal with

[Douglas Bates]

On 4/3/07, Fang, Yongxiang <y.fang at lancaster.ac.uk> wrote:
> Dear Douglas,
>
> Thanks for your help.
> the error message is:
> 'Error in eval(expr, envir, enclos) : Object "dx.200" not found';
>
> In fact, dx is the design matrix and now in a data frame: dframe
> I have checked if dx.200 in dframe. dframe$dx.200 does exist and in correct form.
>
> In addition, when the number of columns of dx is smaller than 200, the lme runs no prolem.
>
> Cheers
>
> Yongxiang

Without a *reproducible* example we cannot be of any assistance.

> -----Original Message-----
> From: dmbates at gmail.com on behalf of Douglas Bates
> Sent: Tue 4/3/2007 5:57 PM
> To: Fang, Yongxiang
> Cc: r-help at stat.math.ethz.ch
> Subject: Re: [R] the numimum number of fixed factors lme package can deal with
>
> On 4/3/07, Fang, Yongxiang <y.fang at lancaster.ac.uk> wrote:
>
> > In my study, mixed effects model is required and the number of fixed
> > effects is very large. When lme package is employed, a model error is
> > displayed once the number of fixed  factors in the formula reaches 200. Is
> > this the maximum number of fixed factors can be handled by lme package?
> > If not, what is possible reason of the error message?
>
> What error message?  You didn't tell use what you did and what
> happened.  Please read and follow the instructions in the posting
> guide - otherwise we will only be able to guess at what the problem
> may be.
>
> There is no specific limit of 200 fixed factors (perhaps you mean 200
> columns in the model matrix for the fixed effects?) in lme.  However
> there will be limits on the amount of memory available to store the
> model matrices and associated structures needed to fit the model.
>
> The first thing I would suggest is determining why you want to fit a
> model with 200 (or possibly more if you really meant 200 fixed
> factors) fixed effects.  It is rare to want to examine such a large
> number of coefficients.  Frequently the number of coefficients gets to
> this order because you have a factor with a large number of levels, in
> which case why not model such a factor with random effects?
>
> You could also try using the lmer function from the lme4 package
> instead of lme to fit a linear mixed model.  It is generally more
> efficient than lme in both time and storage.
>
> However, before we can help you much you will need to be much more
> specific in your question and provide us with the recommended
> background information.
>
>