[R-sig-ME] Error "mu[i] must be positive" with Gamma family

[Stefan Frank]

Hi,

I'm trying to fit a LMER to Gamma distributed data using the identity 
link function, but keep running into the problem that a negative mean is 
estimated:

 > lmer(y~x+(1|s), data, family=Gamma(link="identity"), verbose=TRUE)
   0:     303489.72: 0.0118042  5.95694 -0.0360192
   1:     206240.18: 0.990220  6.16357 -0.0381219
Error in mer_finalize(ans) :
   mu[i] must be positive: mu = 2.63294e-309, i = 1788870688


Fitting a GLM works fine. I tried taking its coefficients as start 
values for the LMER, but that made no difference:

 > m <- glm(y~x, data, family=Gamma(link="identity"))
 > lmer(y~x+(1|s), data, family=Gamma(link="identity"), 
start=coefficients(m), verbose=TRUE)
   0:     303489.72: 0.0118042  5.95694 -0.0360192
   1:     206240.18: 0.990220  6.16357 -0.0381219
Error in mer_finalize(ans) :
   mu[i] must be positive: mu = 2.63293e-309, i = 1778647072


What could be going on, and what can I do about it?

R version 2.9.2 (2009-08-24) on Windows
lme4 version 0.999375-32


Best,

Stefan Frank
Institute for Logic, Language and Computation
University of Amsterdam
URL: staff.science.uva.nl/~sfrank
Tel: +31 20 5256054

Visiting address:                    Postal address:
Science Park 904, Room C3.123        P.O. Box 94242
Amsterdam                            1090 GE Amsterdam
The Netherlands                      The Netherlands