Cross validation of averaged binomial GLMs ("model.avg" from package MuMIn)

Question

I am spatially predicting binomial probabilities (using proportional data; cbind in R) across a spatial domain. I use the functions get.models followed by model.avg in the R package MuMIn to get averaged coefficients of models with delta AIC less than 2. This returns an object of class model list showing the component models and averaged coefficients. I would now like to use the cross validation cv.glm function (or any similar methods) from the R package boot to obtain leave-one-out cross validation prediction accuracy of this averaged model. However, I get the following warning when doing cv.glm(data, model.avg output):

Error in eval(expr, envir, enclos) : could not find function model.avg.default.  

I would greatly appreciate it if anyone could provide suggestions on how to obtain leave one out cross validation on averaged models from the MuMIn package. Thank you.

Answer 1

It seems that cv.glm cannot handle any of the model averaging objects provided by either MuMIN or bbmle. I am looking for a similar answer, yet not for exactly the same question as yours. During my search, I found a blog post where a function is presented that allows cross-validation of MuMIN models. I thought it might be handy for you.

Check out the link here Noam Ross` Blog

in the bottom you should find the proposed Cross function. I only works with 2 models so you will have to adjust it.

Did you find another solution already? If so, please post it here.