[R-sig-eco] Repeated Measures PERMANOVA

[Steve Brewer]

Keith,

I've posted a similar question before and got no helpful response. I think
that's  because no one has developed a straightforward way to do this in
R. Hope I'm proven wrong. Having said that, I CAN  tell you how to
properly test the within-subjects effects (including time and the time x
treatment interaction, which you said you were interested in). Just
realize that the p-value for the between-subjects treatment effects will
be wrong, because it uses the residual error term to get it. The issue is
that, as far as I know, no one has developed code to enable you to use two
different error terms in adonis. Again, I hope someone proves me wrong.

Try this code.

First, you need to define a variable that represents the observation upon
which you're taking repeated observations, I.e., the "subject." In your
case, these would be your "plots." The plots should be nested within the
treatments.

adonis(nema.troph.bc~Treatment*Time + plot,data=nema,strata=plot)


Strata treats your plots like blocks. Again, note that you can only
interpret the p-values for the within-subjects effects (time,
treatment*time). The p-value will be wrong for the treatment effect.
Unfortunately, you cannot solve this problem by simply averaging the
responses among years within plots and then running a separate analysis to
determine between-subjects treatment effects because the analysis uses
pairwise distances, rather than distance from a centroid. When you use
bray-curtis distance, the average of responses among years is not the
centroid.

Hope this helps.

Good luck,

Steve

 
J. Stephen Brewer 
Professor 
Department of Biology
PO Box 1848
 University of Mississippi
University, Mississippi 38677-1848
 Brewer web page - http://home.olemiss.edu/~jbrewer/
FAX - 662-915-5144
Phone - 662-915-1077




On 1/17/13 1:02 PM, "Keith Post" <kpost at rams.colostate.edu> wrote:

>Hello,
>
>I have a simple BACI design with four plots exposed to herbivory and four
>without herbivores.  I quantified the densities of nematode trophic groups
>before and after manipulating herbivory.  So there is one treatment factor
>(+/- herbivory) and a time variable (before/after), and the data are
>balanced.
>Here is a sample of my data:
>
>Plot Time Treatment   BF        FF         PP        Om
>
>1    A      Cont 17.211259 2.0320259 11.1151816 3.9218100
>
>2    A      Cont 26.991283 5.1102956 10.0989175 2.8187742
>
>3    A        GH  6.295530 0.6860513  2.0985099 1.9169081
>
>
>Since I want to see whether herbivory affects how nematode communities
>change, I'm interested in the herbivory * time interaction.
>
>I've tried to analyze a Bray-Curtis dissimilarity matrix of my nematode
>data using the following code:
>
>adonis(nema.troph.bc~Treatment*Time,data=nema,permutations=999)
>
>
>Although this returns an interaction (and main effects), it ignores the
>repeated measures aspect of my design.  How do I incorporate this into
>adonis() or another function.
>
>
>Thanks,
>
>
>
>Keith
>
>	[[alternative HTML version deleted]]
>
>_______________________________________________
>R-sig-ecology mailing list
>R-sig-ecology at r-project.org
>https://stat.ethz.ch/mailman/listinfo/r-sig-ecology