[R-sig-ME] Quasi-GAMM AIC?
luca corlatti
|uc@c1980 @end|ng |rom y@hoo@|t
Fri Dec 4 23:33:03 CET 2020
Dear all, a quick question regarding AIC & quasi-GAMM.
I'm investigating age-dependent variation in body mass in 2 different populations, and decided to go for a GAM approach. As my data are grouped within years & areas, these have been fitted as random intercepts. In the attempt to fix heterogeneity issues in residual variance, I fitted the model with a "quasi" family, so that it looks like:
mod1 <- gamm(mass ~ s(age, by= population) + population, data = my.data, random = list(year = ~ 1, area = ~ 1), family = quasi(link = "identity", variance = "mu"))
Now, if I try to extract the AIC from this model, I actually get a value (16620.34), and a seemingly reasonable one (if compared to a corresponding full-likelihood Tweedie GAMM, which returns the same AIC).
My question is, how is it possible that I get an AIC from a quasi-family?
Re-fitting the same model without random terms:
mod2 <- gam(mass ~ s(age, by= population) + population, data = my.data, family=quasi(link="identity", variance = "mu"))
AIC(mod2) gives, as expected, a "NA".
What allows GAMM to return an AIC value even when using a quasi-family?
Thanks in advance for your help!Luca
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models
mailing list