[R] Bug in lmer?

[Martin Maechler]

>>>>> "Mark" == Mark Lyman <mlyman at byu.edu>
>>>>>     on Thu, 29 Sep 2005 14:44:38 -0600 writes:

    Mark> I am relatively new to R so I am not confident enough in what I am doing 
    Mark> to be certain this is a bug. 


    Mark> I am running R 2.1.1 on a Windows XP 
    Mark> machine and the lme4 package version 0.98-1. 

lme4 nowadays is heavily based on "Matrix" which version are you
using there?

    Mark> The following code fits the model I want using the
    Mark> nlme package version 3.1-60.

      < .............. >  {see a script at the end}

    Mark> The problem is that when I try fitting the model using
    Mark> the lmer function with the following code:

    Mark> lmer(adg~trt+(1|loc)+(1|block:loc)+(1|loc:trt),mltloc)

    Mark> I get this message from Windows and R closes.

    Mark> >> R for Windows GUI front-end has encountered a problem and needs to 
    Mark> >> close.  We are sorry for the inconvenience.

That definitely means there is a bug.
The question is *where* the bug is:  "lme4", "Matrix", "R", "Windows".

One first thing coming to mind is a mismatch of "lme4" and "Matri

    Mark> This same code works on a Macintosh. So it doesn't
    Mark> seem that I have made an error in my code. 

correct; I can also fit the model nice and quickly on Linux,
and summary() confirms the same fit {with the "usual problem" of
different estimates for the degrees of freedoms 'df'}.

So currently the bug only shows on the Windows platform. 
Could it be that you have a mismatching package "Matrix" version
there, but not on the Mac?

    Mark> Also if anyone of the random effect terms is removed there is
    Mark> no problem. Is this something that is being looked at?

not yet, AFAIK.

    Mark> Or I have I made a mistake somewhere? I have included
    Mark> the data that I am using below.

I'm putting the data and an R script up for FTP,
so that you or others can run this ``from anywhere'' via

 source("ftp://stat.ethz.ch/U/maechler/R/mltloc-ex.R", echo = TRUE)

Maybe this helps diagnosis,
Martin Maechler