[R-sig-ME] 2-level null versus 1 level null model

[Ben Pelzer]

Hi all,

This is probably a very basic question. I would like to test, using 
lmer, whether a 2-level null model (no covariates, only random 
intercept) has a better fit (lower deviance) than a 'flat' 1-level null 
model. My thought was to estimate both null models with lmer and then 
with anova(M02level, M01level) test for the significance of the 
chisquare difference. However, with lmer one cannot simply estimate a 
1-level model, as far as my knowledge goes... What I did to circumvent 
this was the followin, with Y being the dependent and country the 
level-2 identifier:

one <- rep(1, length(Y))

M01level <-  lmer (Y ~ (1 | one), REML=FALSE)
M02level <-  lmer (Y ~ (1 | one) + (1 | country), REML=FALSE)
anova(M02level, M01level)

This seams to work properly. However, I don't feel really convinced by 
this solution yet. Is it to be trusted, or is there a much simpler way 
to test this using lmer? Thanks for any help!!!

Ben.