Skip to content
Prev 16808 / 20628 Next

R Squared for GLMM with gamma distribution

One clarifying question might be how do you know that the gamma distribution properly describes the errors of your data situation? The fact that a value appears to be "gamma" when viewed as a single value does not mean that the errors around a well-constructed model will have gamma-distributed errors.

Another clarifying question: What do you expect for a "coefficient of determination" if you were performing some sort of regression that was suited for data that remained gamma-error distributed? You might look at the multiple "R^2" variants that exist for models constructed with logistic regression. They all  fail in some manner or another to live up to the expectations generated by the "original" R^2 developed for linear models.
David Winsemius
Alameda, CA, USA

'Any technology distinguishable from magic is insufficiently advanced.'   -Gehm's Corollary to Clarke's Third Law