[R-sig-ME] p values for random effects in an unbalanced nested mixed model

[Lionel]

Dear Stacey,

The classical null hypothesis of a question like: "is the variation in 
richness significant at each spatial scale. " is that the variation (or 
standard deviation) equal 0. You can have a look at the Box3 in Bolker 
et al (2009) Trends in Ecology and Evolution, there it is explained that 
such a null hypothesis make little sense since you cannot have 0 
standard deviation, then this test will be too conservative (high Type 
II error).

You could look at ?pvalues after having loaded the lme4 package, there 
are various options you could use, the easiest being a likelihood ratio 
test using anova(model1,model2), model1 having a random term like: 
random=~1|Ecoregion/Area/Site  and model2=~1|Ecoregion/Area, the p-value 
you will get correspond to the null hypothesis: The likelihood ratio is 
equal to one, in other words there is no difference in likelihood 
between the two models. In this context significant difference in 
likelihoods can be interpreted as one model fitting better the data due 
to its particular random structure.

Yours,
Lionel


On 04/05/2014 00:20, Stacey Williams wrote:
> Hi List-serv community,
> I would like to calculate the p values for the random effects in an
> unbalanced nested mixed model (lme or lmer). There are three factor levels
> (all random), ecoregions, area nested in ecoregion, and sites nested in
> area and ecoregion.
>
> My model looks like something like this
> coral_richness<-lme(Richness~1, random=~1|Ecoregion/Area/Site)
>
> My goals are to examine 1) how much richness varies at different spatial
> scales (ecoregions, areas, sites) and 2) is the variation in richness
> significant at each spatial scale.
>
> Any help would be greatly appreciated!!!!!
>
> Thanks,
> Stacey
>