[R] Doubts about mixed effect models

Bert Gunter gunter.berton at gene.com
Wed Mar 21 20:52:28 CET 2012

Post this on the r-sig-mixed=models list rather than here.

However, fwiw, it is nonsense to estimate a random effect with a
sample size of 3. That's trying to estimate variance with a sample
size of 3. You can't do it with any meaningful precision. Whether or
not the effect really **is** conceptually random is beside the point.
I suggest you cross off your list of statistical advisers anyone who
says otherwise.

Entropy cannot be denied!

-- Bert

On Wed, Mar 21, 2012 at 11:01 AM, Lívia Dorneles Audino
<livia.audino at gmail.com> wrote:
> Hi everyone!
> I have some doubts about mixed effect models and I hope someone could help
> me. I´m trying to analyze a dataset coming from samples of dung beetles in
> the same forest fragments along 3 consecutive years (1994, 1995 and 1996)
> and 14 years after (2010). I sampled dung beetles in 18 different fragments
> with different sizes and different degrees of isolation. My aim is to
> determine whether total species richness change over time in forest
> fragments and to verify the influence of fragment size and isolation on
> species richness. However, I'm trying to find a way to consider in the
> analyses the temporal pseudo-replication in the data. So, I decided to use
> mixed effects models to analyze this data, but I still have doubts about
> how I should construct the models. When I asked for some help I received
> three different answers about how to construct the model.
> The first suggestion was to treat year as a fixed rather than a random
> effect because it won't be practical to estimate the variance of a
> random effect
> with only four levels. So, I constructed the model like cited below:
> m1<-lmer(riqueza~área*ano+isolamento*ano(1|fragmento),family=poisson
> The second suggestion proposed to treat year as a random effect, as cited
> bellow:
> m1<-lmer(riqueza~área*ano+isolamento*ano(ano|fragmento),family=poisson
> And the third suggestion also proposed to treat year as a random effect,
> but to consider it *as continuous variable rather than categorical*. In the
> models above I consider year as a categorical variable.
> m1<-lmer(riqueza~área*ano+isolamento*ano(ano|fragmento),family=poisson
> When I analyze my dataset using the second and the third model I always
> face with a singular convergence warning: *In mer finalize(ans): singular
> convergence (7)**.*   What is that mean? Does anyone have some idea about
> how can I solve this issue?
> I also need to know which one of these models is more appropriate to the
> dataset available. Does anyone have some suggestions?
> Thanks in advance!
> Lívia.
>        [[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.


Bert Gunter
Genentech Nonclinical Biostatistics

Internal Contact Info:
Phone: 467-7374

More information about the R-help mailing list