[R-sig-eco] explained variation in GAM/GAMM
Zoltan Botta-Dukat
bdz at botanika.hu
Tue Mar 9 08:58:19 CET 2010
Dear All,
I would like to estimate the variation explained by different sets of
predictors. My response is binary, and I want to capture the possible
non-linear relationships, thus first I chooose GAM with binomial family.
Summary of the model gives adjusted R-squared and explained deviance (in
%). The latter seems to be what I need, and I'm not sure what the
R-squared is in this situation. Could someone explain it?
Then, I recognzed that residuals are highly spatially autocorrelated, so
I changed to GAMM. It handle the spatial autocorrelation, but the
summary of gam part of the fitted model gives adjusted R-squared only,
but not the expalined variance. R-squared is lower in GAMM than in GAM,
so I think considering spatial autocorrelation is necessary. My
question is whether I can use this adjusted R-squared or I should
calculate something else, for example Efron's pseudo-R-squared, from the
predicted values.
Thanks for your advice
Zoltan
More information about the R-sig-ecology
mailing list