[R-sig-ME] repeated measures random effects structure

[Lionel]

Dear Kevin,

The first model correspond to a nested nested random intercept of 
replicates within generation, this means that the model will estimate 
the standard deviation of the intercept between each replicates within 
generation, I guess it will also estimate the overall standard deviation 
of the intercept between the generation.
The second model is a random slope (of generation effect on your 
response)+intercept of replicates, this time the model will estimate the 
standard deviation of the intercept between the replicates as well as 
the standard deviation of slopes of response vs generation between the 
replicates. This modelling structure is similar to a fixed effect linear 
model with one continuous variable, one grouping variable plus their 
interaction.

These two models are therefore rather different, I would argue that your 
first model would be more appropriate for your experimental design, you 
should make sure that the replicate ID are unique over the generation.

Sincerely yours,
Lionel

On 03/21/2014 12:05 AM, Kevin Burls wrote:
> Hello,
>
> I am analyzing the results from an artificial selection experiment and have a question regarding the random effects structure. My experimental design is fairly simple: there are two treatments, with 5 replicates for each treatment, measuring a response from each replicate each generation for 20 generations. I am interested not only in the effects of treatment, but in a main effect of time, well as the interaction (looking for a difference between treatments over time). I have looked over quite a few of the examples regarding the structure for such an experiment and have found two alternative models that seem appropriate, but I am unclear what the difference is. These models are:
>
> lmer1<- Response~treatment*generation + (1 | generation/replicate); a more typical nested effects structure
>
> lmer2<- Response~treatment*generation + (generation | replicate) ; following the sleepstudy design described in section 2 of the 2012 lme4 implementation guide by Dr. Bates.
>
> I am sure part of my confusion is not really understanding what it means to put something in front of the grouping factor, versus a nested design. It seems as though the second one could be more appropriate for a repeated measures framework, but I am at a loss to explain why.
>
> Thanks for your help,
>
> Kevin Burls
>
> Kevin Burls
> Ph.D. candidate
> EECB Program
> University of Nevada, Reno
> kburls at unr.edu<mailto:kburls at unr.edu>
> http://wolfweb.unr.edu/~kburls
> www.nevadabugs.org<http://www.nevadabugs.org>
>
>
>
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models