[R-sig-ME] Interpreting the variance components with crossed random effects

[ONKELINX, Thierry]

Hi Nick,

I agree with your interpretation as long as you are only talking about random intercepts. IMHO the reasoning would not longer hold with random slopes because then you would need to take to covariate into account.

Best regards,

Thierry

> -----Oorspronkelijk bericht-----
> Van: r-sig-mixed-models-bounces at r-project.org [mailto:r-sig-mixed-models-
> bounces at r-project.org] Namens Nick Isaac
> Verzonden: woensdag 19 oktober 2011 17:57
> Aan: Help Mixed Models
> Onderwerp: [R-sig-ME] Interpreting the variance components with crossed
> random effects
> 
> Dear list members,
> 
> I have a dataset with crossed random effects. I am interested in describing how
> variation in the response variable is attributable to each RE.
> 
> The only worked examples I have found involve nested, rather than crossed,
> REs. My own dataset is large and unbalanced, but the issue can be illustrated
> easily using the Assay dataset from MEMSS.
> 
> library('MEMSS')
> data(Assay)
> (rmod <- lmer(logDens~1 + (1|Block) + (1|sample) + (1|dilut), Assay))
> vc<-VarCorr(rmod)
> vars <- c(unlist(vc), attr(vc,'sc')^2)
> 100*vars/sum(vars)
> 
> I conclude that most variation (91%) is attributable to the different levels of
> dilution (dilut), most of the rest (6%) to the different wells (sample), hardly any
> at all (0.2%) to the different replicates (Block)and the remainder (3%) to residual
> error.
> 
> I would be greateful for a second opinion on whether this approach is
> appropriate.
> 
> Best wishes, Nick
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models