Dear s-plus users,
Can someone give a recommendation as to how to select fixed effects in a
linear mixed-effects model? I have several variables that may be
(biologically) plausible as fixed covaribles in a lme-model, and need to
choose which to include.
Pinheiro and Bates (2000, p. 87 ff) advice not to use LR-test for
hypothesis testing when models with different fixed effects are compared,
since the chi-square reference criterium is anti-conservative. Does a
similar problem arise if fixed effects structure is determined based on
AIC-values? Is a stepwise selection procedure based on P-values from the
approximate F-tests in lme a better approach?
Any comments would be greatly appreciated!
Thanks in advance!
Leif
---------------------------------------------------------------------------------------------
Dr. Leif Christian Stige
Biogéosciences, UMR-CNRS 5561
Université de Bourgogne
6, Bd. Gabriel
21000 Dijon, France
|