While you are right that the technique should be what is emphasized,  
many people are not familiar with generalized linear models (as  
opposed to general linear models), and even less are familiar with  
how to use them with their stats software of choice.  Many pieces of  
software that are more 'user friendly' don't even accommodate these  
tests.  While some of us have stumbled upon them in working with R  
(glm), SAS (proc genmod), etc, because these are not methods often  
taught in ecological stats classes, it's useful to know just what is  
out there.

But the method is general, and they are not black boxes (and in the  
case of R, it's all open source, so it's really neither proprietary,  
nor a black box).

For those who want a brief primer on these methods,  Ben Boelker's  
book is an excellent place to start, and I second that  
recommendation.  Also see Guisan et al 2002 in Ecological Modeling  
( good intro), Parker et al 1999 Human and Ecological Risk Assessment  
(great flowchart). VerHoef Boveng 2007 Ecology (quasipoisson v.  
negative binomial distributions in gzlms), Schneider 1992 Limnology  
and Oceanography (why generalized linear models are a better  
alternative to data transformation).

I've also heard good things about Hardin's Generalized Linear Models  
and Extensions, but have not yet seen it myself.

In addition, a number of books on R do cover generalized linear  
models, and are also highly practical as they have code and  
examples.  See also Gelman and Hill's book on multilevel modeling, if  
you like that sort of thing.

-Jarrett



----------------------------------------
Jarrett Byrnes
Population Biology Graduate Group, UC Davis
Bodega Marine Lab
707-875-1969
http://www-eve.ucdavis.edu/stachowicz/byrnes.shtml

On Jan 13, 2008, at 1:57 PM, William Silvert wrote:

> One point about the various replies to this and other posts that  
> disturbs me
> is the focus of the responses. It used to be that statistical  
> questions were
> answered in terms of statistical techniques, such as regression or  
> ANOVA or
> t-tests. Now the answers are phrased in terms of software - SAS, R,  
> SysStat,
> etc. I am not confident that relying on proprietary black boxes is  
> the best
> way to analyse data.
>
> Bill Silvert
>
>
> ----- Original Message -----
>
>> If you have access to SAS, ...

Reply via email to