Hi ,
I want to test the significance of various fixed effects and their interactions
in a lme model. Following Pinhero and Bates, I used a conditional anova test -
anova(object.lme) - rather than testing the difference in log- likelihood of a
full model against a model from which the tested factor is removed. As
anova(object.lme) is a sequential anova, the significance of the added terms
changes according to the order in which I specified them in the model...I find
it therefore difficoult to interpret my tresult and understand which factors I
should actually consider significant and which not. I thought to use
anova(object.lme,type="marginal"), but I understand that the use of marginal
type anova is controversial.
I would be glad for any help you can provide!
bset regards,
Achaz von Hardenberg
|