Juan Morales writes:

I am currently fitting a multilevel model to data of fruit removal rates which I model using binomial distributions for the number of removed fruit out of total fruits available. I would like to estimate the proportion of variance explained and the amount of pooling at each level (tree, forest stand and so on). You show how to do such things for the radon example and mention that something similar could be done for generalized linear models using deviances. Has this been done somewhere?

My reply: R-squared for multilevel linear models is discussed in our book and in my paper with Pardoe. I think it would make sense to do this with logistic regression also (perhaps using the latent variable formulation with residual s.e. of 1.6, as Jennifer and I discuss in chapter 5). But I haven’t done it yet. A good research paper, I think!