[R-sig-ME] Size/metric of variance components in lme and lmer

[Stuart Luppescu]

On Tue, 2010-03-09 at 10:16 +0100, ONKELINX, Thierry wrote:
> Dear Stuart,
> 
> I think that the "extra" variance is substracted from the fixed
> effects. Which indicates that some of the information in your fixed
> effects was due to the levels of tid.
> 
> But to make a fair comparison you should run both models with lmer.
> And then compare both the random effects variances and the fixed
> effect estimates.

OK. I copied function call from the cross-classified model with lmer and
removed the second random effect and reran it. Here are the random
effects tables:

Nested model:
   Data: all.subj 
     AIC     BIC   logLik deviance REMLdev
 3448318 3449229 -1724083  3448166 3448707
Random effects:
 Groups   Name        Variance Std.Dev.
 sid      (Intercept) 0.77403  0.87979 
 Residual             0.85746  0.92599 
Number of obs: 1185094, groups: sid, 122897

Total variance:  1.63
--------------------------------------------------

Cross-classified model:
   Data: all.subj 
     AIC     BIC   logLik deviance REMLdev
 3236856 3237779 -1618351  3236702 3237217
Random effects:
 Groups   Name        Variance Std.Dev.
 sid      (Intercept) 0.75066  0.86641 
 tid      (Intercept) 0.39171  0.62587 
 Residual             0.68583  0.82815 
Number of obs: 1185094, groups: sid, 122897; tid, 8939

Total variance: 1.83

Could some of the difference be due to covariance between the random
effects?


-- 
Stuart Luppescu <slu at ccsr.uchicago.edu>
University of Chicago