[R-sig-eco] Liner vs. Poisson vs. Mixed effect models

[Canning-Clode, Joao]

Yes, we do have more than one panel (=plate) per region. Local richness was assessed as the mean number of observed species in 8 panels. Regional richness was estimated based on a species accumulation curve (with Jackknife 2) with other different 60 panels.


On 10/7/09 6:33 AM, "Dunbar, Michael" <mdu at ceh.ac.uk> wrote:

Dear Joao

I'm not entirely sure that I understand your design correctly, but I will try to help. Please bear this in mind though that I may have misunderstood. For example what is local? What is a panel, is it the same as a plate? One thing I couldn't get at from your description is whether there is one plate per region or more than one. If more than one, then you need an identifier for the plate in the model if you're dealing with multiple time points or if you have plate level measurements. I'm assuming this is the case, because if you don't have more than one plate per region, then it isn't really a mixed model, it's just a multiple regression and you can't say anything about the regions.

It does not seem to make sense to have LFR/region. LFR is a measure of something, so it cannot be a grouping variable.

For region being a random effect, this seems OK, but are your regions a sample of possible regions, with the estimated variation across your regions being an estimator for general regional variability?

I'd strongly suggest building your models without the time component initially.

When you decide to include it, you need to make some decisions as to what it is about time that you are trying to get at. Are you expecting temporal trends or are the time points just replicates? Forget about auto-regressive structures for the time being, you only have four time points. So long as you have more than one plate per region, you can do:

LTR.LME <- lme(LTR~RTR+RFR+OS+LFR+month, random=~1|region/plate)

regards

Mike


-----Original Message-----
From: r-sig-ecology-bounces at r-project.org [mailto:r-sig-ecology-bounces at r-project.org] On Behalf Of Canning-Clode, Joao
Sent: 06 October 2009 21:56
To: r-sig-ecology at r-project.org
Subject: [R-sig-eco] Liner vs. Poisson vs. Mixed effect models

Hi all,

First of all thanks to all for helping me with my previous post on negative binomial.
As I mentioned I got a paper back for revision requesting alternative analysis, and I am seeking advice from the forum. I really appreciate any help you could provide.

My major question was to explore the local-regional richness relationship in fouling assemblages at a global scale. I have deployed PVC sampling plates in the water at 8 different regions, and sampled the benthic communities 4 times (after 2, 4, 6 and 8 months).

Therefore, my experimental design included 8 regions, each having one estimate for regional taxonomic richness (RTR), one estimate for regional functional richness(RFR) (both estimated with a jackknife estimator), 8 measures of open space (OS) on panels, 8 measures of local functional richness (LFR)and 8 measures of local taxonomic richness (LTR).

One of my objectives is to examine how these variables affect local taxonomical richness (LTR) and a suggestion to try a linear mixed effects models where RTR, RFR and OS are fixed effects; LFR is a random effect nested within region.  I square-root transformed LTR, used the lme function and set my model as

LTR.LME <- lme(LTR~RTR+RFR+OS+LFR, random=~1|LFR/region)

A question I have - is it okay to have the random effect LFR in the left (fixed-effects) equation?  I wish to estimate the coefficients to examine the effects of LFR; maybe I should treat this as a fixed effect instead?

A second question is that I have repeated measures (over four time periods).  Is it possible to set up the above equation in a repeated fashion say using a auto-regressive 1 structure?

Thank you in advance for any help,
Joao

João Canning Clode, PhD
Postdoctoral Fellow
Marine Invasions Research Lab
Smithsonian Environmental Research Center
647 Contees Wharf Road
Edgewater, MD 21037

Email: canning-clodej at si.edu
Web: www.canning-clode.com
Tel: 443-482-2354

_______________________________________________
R-sig-ecology mailing list
R-sig-ecology at r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology

--
This message (and any attachments) is for the recipient ...{{dropped:19}}