Skip to content
Prev 5386 / 7419 Next

Correction: negative binomial mixed model using glmmadmb

Heather Major <heather.major at ...> writes:
long intervals
during the night, ranging from 0 (full
0 = no clouds 1 = complete overcast sky.
(1|JDOY)+TAS+MA+CC+CWS+CWH+TAS*MA+TAS*CC+TAS*CWS+TAS*CWH+TAS*MA*CC+MA*CC, 
    data = murrelet, family="nbinom")

 You don't need all those *: A*B is equivalent to A+B+A:B (in R
: means 'interaction' (* in SAS), * means 'main effects plus all
interactions'; I _think_

 (1|JDOY)+TAS*MA*CC+TAS*(CWS+CWH)

is equivalent.
You might be pushing these data too hard; what is the number of parameters
(length(fixef(fitted_model)) or 
ncol(model.matrix(~TAS*MA*CC+TAS*(CWS+CWH),data=your_data) ...)
You need 10-20 data points per parameter ...
You don't need the Q part of QAICc; quasi-AIC(c)s are only 
needed to correct for overdispersion when you're using a response
distribution (e.g. Poisson) that fixes the dispersion.

For future reference, I think that in general *something* like

sum(residuals(model)^2))/(nrow(data)-length(fixef(model))-
   (number of variance parameters)

should give you c-hat ...