[R-sig-ME] Zero random effect variance?

[Ben Bolker]

Dave Marvin <marvs at ...> writes:

> 
> For the following dataset (described at the bottom of this email), 

  Did you mean to include an actual data set, or just the text description?
Without the data set itself, we can't do better than guessing.

> a boxplot of the data by chamber
> 
> > height=read.table("height.txt",header=TRUE)
> > boxplot(HtChg~Chamber,data=height)
 
> shows there is clearly a lot of chamber-to-chamber variation in the
> response variable. However, if I run a random intercept-only model:
 
> > lmer(HtChg~1+(1|Chamber),data=height)
> 
> I get 0 variance for the random intercept. Same is true if I then 
> include any categorical fixed effects. Does
> this seem correct, and if so why? 
> -Dave
 
> > I am analyzing the growth response (Height Change) of two plant
>  types (vines vs. trees) to different CO2 levels, for a mix of
>  species of each plant type in plant growth chambers (Chamber). CO2
>  and FT are categorical predictors, each with two levels
>  (elevated/ambient CO2, vine/tree plant Functional Types). Each
>  growth chamber had the same mix of 8 species (Spp).

  Presumably the within-chamber variation is large enough that
it adequately accounts for the among-chamber variation?
Again, hard to say without seeing the data ... you could do the 
math yourself (e.g. is variance among >= (variance within)/(n within)?),
or simulate some representative examples ...