part R^2 calculation has been modified. We now estimate the
difference in explained variance in the linear predictor and divide this
by the estimated total variance from the full model. The results should
be very similar, but more stable.
added option “none” to possible expct arguments to allow calculating
R2 for GLMMs without distribution-specific variation in the
denominator.