[R-sig-ME] Links to that "slow" lmer example I asked about

[Paul Johnson]

Dear Everybody:

I asked a couple of weeks ago about the puzzle that my colleague's
linear mixed model can be estimated in HLM6 in 3 seconds, while lmer
requires about 50 seconds.  I wondered if that was expected/known.

The discussion seemed to end with the conclusion "if you expect us to
evaluate that, give us the working example."  Due to a death in my
family, I was delayed in responding, but here are the links to the
example data and code.

The data frame is saved with R's write function

http://pj.freefaculty.org/R/MixedModel/myframe.Rdata

I believe that is workable on all platforms. That's about 26,000 rows.
Variable V33 represents the groups (in this case, country).  The
variables are generically named V1-V33.

The small simple test program to load the data and estimate the model:

http://pj.freefaculty.org/R/MixedModel/replicateMM.R

The output I get, which has system.time wrapped around the use of lmer:

http://pj.freefaculty.org/R/MixedModel/replicateMM.Rout

I get
##  user  system elapsed
##55.448   0.216  55.756

Please remember I am not saying that lmer should work faster.  I
understand it is capable of estimating models that other programs
cannot.  I'm only trying to explain to a user why this one linear
model takes more time in lmer than in HLM6.

pj
-- 
Paul E. Johnson
Professor, Political Science
1541 Lilac Lane, Room 504
University of Kansas