Skip to content
Prev 2726 / 20628 Next
Original Raw

CHOLMOD error in lmer-specifying nested random effects

Emmanuel Charpentier 
Le lundi 10 ao?t 2009 ? 10:28 -0400, Eli Swanson a ?crit :
My objection to Poisson also applies to negative binomial (a discrete
distribution, last time I looked).

As to the transformation to use, I rummaged your previous posts and came
with  possible alternative :

The difficulty is that your observation is indeed a proportion, but not,
strictly speaking, a counting process. In your original posting, you
stated : "The response variable is the number of minutes a cub was
observed nursing (Nurse.min).", which I'm tempted to rewrite as "The
response variable is the *duration* a cub was observed nursing
(Nurse.t)." This duration is bound *twice* : on the left by 0, on the
right by the *total duration* of your observation (say, Obs.t).

I *think* that your variable of interest might be better expressed as
Nurse.frac=Nurse.t/Obs.t, which is bound by 0 nd 1, like a proportion.
Here, the sins of my past return to haunt me and sorely tempt me to
treat it a bit like a logistic regression problem :

What happens when you study log(Nurse.frac/(1-Nurse.frac)) in a *linear*
(mixed) model ?

Note, however, you *cannot* treat it as a "real" logistic regression
problem : the variance of this fraction is *not* a simple function of
Nurse.frac and Obs.t (your notional "sample size") ; here you have *one*
observation per unit, not Obs.mins. (To "see" it better, think of the
different results you would get when expressing your times in minutes
and in seconds...). Whereas in Poisson or logistic regression you have
to asses only one parameter (\pi or \lambda), here, you have two : \mu
and \sigma, the later being a "nuisance" parameter (which can indeed be
real nuisance : look out for heteroscedasticity).

Therefore, use lmer(..., family=gaussian) (or maybe lme(), which would
allow you to model heteroscedasticity), rather than lmer(...,
family=binomial).

Hope this helps,

						Emmanuel Charpentier

Thread (6 messages)

Luca Borger CHOLMOD error in lmer-specifying nested random effects Aug 7 Eli Swanson CHOLMOD error in lmer-specifying nested random effects Aug 7 Emmanuel Charpentier CHOLMOD error in lmer-specifying nested random effects Aug 10 Eli Swanson CHOLMOD error in lmer-specifying nested random effects Aug 10 Douglas Bates CHOLMOD error in lmer-specifying nested random effects Aug 10 Emmanuel Charpentier CHOLMOD error in lmer-specifying nested random effects Aug 10