[R] Interpretation of VarCorr results

[Kingsford Jones]

On Tue, Oct 20, 2009 at 5:09 AM, r-quantide <r at quantide.com> wrote:
> Dear all,
>
> I'm working with a model to estimate components of variance by using the
> lme() function.
>
>
>
> The model is as the following:
>
> model=lme(fixed=X~1+as.factor(station),data=myData,na.action=na.exclude,rand
> om=~N+1|spliceOrHoming)
>
>
>
> Where X is the response measured variable, station is treated as fixed
> effects factor, N is a continuous variable, and spliceOrHoming is a
> (ordered) factor.
>
> The idea is that the response variable (X) varies almost linearly with N (in
> a decreasing fashion) , but the slope and intercept of this variation change
> with the levels of spliceOrHoming random factor.
>
>
>
> Now, the REML estimated standard deviation components are:
>
>            StdDev       Corr
>
> (Intercept) 5.274544e-01 (Intr)
>
> N           7.912717e-05 -0.58
>
> Residual    1.508647e-01
>
>
>
> My questions are:
>
> .         Is the model correctly specified?


This depends on the questions you are trying to approach with the model...


>
> .         If yes, how should I read the standard deviation estimate for N
> (7.912717e-05)? Is this  the standard deviation for a unitary variation of
> N? Or is it the standard deviation due to the "global" contribution of the N
> independent variable? (note that I have a strong feeling that N contributes
> a lot to the final variance; so, the value that I obtain here seems "too
> small" for my beliefs)


Generally if you have a random slope you would also include that
covariate in the fixed component, producing an estimate of the
population (or marginal) slope in the fixed effects output, and an
estimate of the variation in slopes around that population slope among
the levels of the grouping variable (here spliceOrHoming) in the
random effects output.

>
> .         Is there any methods/functions to obtain the variance components
> for the station factor too? (please, give me some references, at least.)
>

With the proper design (i.e. enough data in the appropriate levels)
you might specify something like:
random = ~N + station - 1|spliceOrHoming
To get estimates of the between group variation for each of the levels
of the station factor, as well as for the N slope.


hth,

Kingsford Jones



>
>
> Thank you a lot in advance for every suggestions you'll give me.
>
> Enrico
>
>
>        [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>