[R-sig-eco] Re : guidance required (GLM?)

[Dunbar, Michael J.]

Chris is unlikely to need a binomial model if the response variable is biomass.

glm is generalised linear model: are you sure you don't just want lm (linear model with normal residuals) Chris?

You're new to stats, but you are trying to do something quite complex, you do need to build up you knowledge gradually. Can you fit some models to simpler subsets of the data with fewer factors, e.g. to one species at a time, or one shore at a time. Further you mention "post-hoc adjusted pairwise comparisons". A. This is not something for a beginner. B. They power to detect differences with these methods will be very dependent on the replication you have in your dataset - you don't mention anything about that. C. You seem to have a designed experiment but by going for post-hoc tests you are ignoring some basic principles of the scientific method associated with designed experiments: i.e. design an experiment to test a specific hypothesis or at least a small number of hypotheses. Many pairwise comparisons plus lots of interactions sounds like a poorly designed experiment to start with.

I'm afraid I couldn't understand how you created your raked and fished factors. I think you want a treatment factor and a raked factor. Making them interact with everything sounds like overkill unless you have prior evidence for particular interactions. 

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 momadou sow
Sent: 24 August 2011 11:17
To: Christopher Cesar
Cc: R-sig-ecology at r-project.org
Subject: [R-sig-eco] Re : guidance required (GLM?)

Hi,
With glm, you add the binomial family:
model<-glm(biomass~Shore*Raked*Species+Shore*Fished*Species,family=binomial)
model


De : Christopher Cesar <C.Cesar at apemltd.co.uk> À : "r-sig-ecology at r-project.org" <r-sig-ecology at r-project.org> Envoyé le : Mercredi 24 Août 2011 12h02 Objet : [R-sig-eco] guidance required (GLM?)

Hi all,

I have carried out experimental removal of bivalves at 2 intertidal shores. Bivalves were removed by raking of surface sediments. I wish compare the biomass values of for a total of 8 species between the 2 shores

My 3 treatments are: Undisturbed Controls (Cont), Procedural Controls (Proc) and Experimetnally Fished (Fished).

As Fished and Proc have both experienced disturbance, I set the model using 2 factors as follows:

    Controls    Procedural    Fished
Raked    0        1        1
Fished0        0        1

As a newcomer to R (& stats!), I am unsure as to how to proceed.

i am currently adopting the approach

model<-glm(biomass~Shore*Raked*Species+Shore*Fished*Species)

And then run post-hoc adjusted pairwise comparisons between signifcant terms.

Does this look OK to you guys?

Many, many thanks

Chris



This message has been scanned for viruses by BlackSpider...{{dropped:10}}