gls for generalized linear model (ChenChun)
ChenChun <talischen at ...> writes:
[snip]
1) shall I consider the boundary issue (null:sigma=0, alternative: sigma>0) when conducting the analysis of deviance?
Sure, if you want to be more precise (note that finite-size effects aren't taken into account either).
2) When comparing two nested models with random effect (One model contains an extra random term than the other), I can directly apply function anova(). In this case, does the anova output contain the consideration of boundary issue? or I need to do it by myself?
You need to do it yourself.