Dear R-Users, Am I right if I say that comparisons among gls models with differing fixed effects, but same correlation stuctures, fitted by REML is not meaningful? I need to use ML fit to get comparable AIC, BIC values? And the same is applied to likelihood ratio statistics? I don't really know what's the differece between the two fitting methods. Does anybody have a good documentation for understanding them? Would much esaier to understand restrictions of REMLfit if I see the mathematical background of it. Cheers, Balazs [[alternative HTML version deleted]]