[R-sig-ME] link function with glmer and inverse.gaussian family

[Ben Bolker]

On 14-09-12 06:16 AM, Helios de Rosario wrote:
> Hello, I did this:
> 
>> mod1 <- glmer(y ~ x + (1|block), data=dat,
> family=inverse.gaussian(log))
>> getFamily(mod1)
> 
> Family: inverse.gaussian 
> Link function: 1/mu^2 
> 
> I had expected that the link function would be "log", as I entered it
> explicitly. If I fit a glm (without random effects), I do get:
> 
>> mod2<- glm(y ~ x, data=dat, family=inverse.gaussian(log))
>> getFamily(mod2)
> 
> Family: inverse.gaussian 
> Link function: log
> 
> Is there a reason for that different behaviour of the functions?
> 
> Thanks
> Helios De Rosario 

This is a bug that affects reporting (not the actual fit), which was
reported and fixed quite recently (see
http://article.gmane.org/gmane.comp.lang.r.lme4.devel/12374 for
information and workarounds)