Hi class,
Here is an interesting example from Ben Bolker using the owls dataset. He shows how to use an observation level group effect (to effectively change to a Poisson-lognormal distribution), how to predict the expected values of a lme4 model, and an interesting way to plot expected vs. observed with confidence intervals. I might have to learn ggplot2! Enjoy!
Here are the pdf and the script I did in class:
Jack
Hi class,
I worked on some more models today, and found one that is better. All issues are not resolved, but some are.
New script is now Chapter13_jtf_v2.r
Jack
Hi class,
Here is my modified script for Chapter 13 – Chapter13_jtf.r
Can’t quite get to everything we might want to do. Here is a paper that talks about GLMMs: Bolker et al. 2009 . It goes into some detail about how each pacakge (or program) solves the problem. Apparently, there is real difficulty in solving GLMMs and GAMMs. The difficulty seems to be that the solution to the fixed effects have to be integrated over the random effects. This is not exact, but there are various approximations. PQL (penalized quasi-likelihood) is worst, LaPlace Approximation is better, etc.
Also, look at this link (http://glmm.wikidot.com/pkg-comparison#GLMMs), which lists the various commands and what you can do with them. For example, it looks like glmmPQL is the only command that can do correlation structures.
Have fun!
Jack