Pseudo R-squared of averaged model

Question

I am working on model averaging of data collected about bird species and habitat vegetation. I have been using the MuMIn package in R and have taken a subset of all possible models and then averaged the variables in those models to create the coefficients from the subset of models but now I need to find the pseudo r-squared of this averaged model. Does anyone know how to accomplish that?

Answer 1

You can calculate the r-squared for every model like so:

dredged <- dredge(model, extra = "R^2")

I don't believe there is an agreed-upon method to calculate an averaged r-squared, however you can report the range with:

range(dredged$'R^2')

Answer 2

I'm not sure about the pseudo R2 for the averaged model, but if you are simply looking for a way to measure the 'fit' of the averaged-model, I believe you can use AUC or Cross-Validation. You can get the area under the receiver operating curve with function roc(), from package pROC on averaged models, or you can get cross-validation error rates for the averaged model (it can be altered slightly depending on how many models are averaged); link for CV averaged model.

Another way is to report the deviance explained, by using ((null deviance – residual deviance)/null deviance), although I'm not sure how common this is used.