[R-sig-Geo] Doubts about GLM

[Andrew Finley]

Hi Suzana,
You can try the spGLM function in the spBayes package or similar functions in the geoRglm package which provide spatial random effects to accommodate residual spatial dependence.
-Andy

On Jan 5, 2013, at 15:34, Suzana Stutz <suzana.stutz at gmail.com> wrote:

> Hello dear list members.
> 
> First I want to wish you all Happy New Year! I Hope 2013 brings us many
> opportunities to learn and to do our best in our careers and personal life.
> After that,  I must apologize for the late response.
> 
> I am very greatful  for your useful cues.
> 
> Sr. L�pez, you're right about the uncountable variables, the formula
> accepted it and all worked well with the countable ones. Thank you a lot
> for the help. I'll keep trying to solve the problem on the correlated
> predictors.
> 
> Sr. Rowlingson, thank you for the example of spatial autocorrelation, and
> yes it is an important issue to worry about. The environmental variables
> were colected at stations far from each other, trying to avoid it. Also we
> are not using the colected values, but new ones estimated from the real
> ones. Does it help to avoid the data spatial autocorrelation too??
> 
> 
> Thank you a lot.
> 
> Cheers!!
> 
> 
> Suzana
> 
> 
> 
> 
> 
> 
> 
> 
> 2013/1/2 Barry Rowlingson <b.rowlingson at lancaster.ac.uk>
> 
>> On Sun, Dec 30, 2012 at 1:55 PM, Suzana Stutz <suzana.stutz at gmail.com>
>> wrote:
>>> Dear list members.
>>> 
>>> I've been trying to apply GLM for spatial analysis of  a marine animal's
>>> distribution (as the dependant variable) related to many environmental
>>> variables (as the independant ones). But some of them are correlated, for
>>> example, the depth increases as the distance to coast increases, and  I
>>> must test both. Does it disagree with the assumption of independence of
>> the
>>> data? If yes, how can I fix it in order to keep the two variables in the
>>> model?
>> 
>> Correlation in your explanatory variables is a classical problem of
>> variable selection, and there are methods for that. The bigger problem
>> with doing a GLM with spatial data is that the data points are going
>> to have spatial autocorrelation...
>> 
>> Two measurements of 10 fish per square meter at depth 50m taken two
>> metres apart are less likely to be independent than two measurements
>> of 10 fish per square meter at depth 50m taken twenty kilometres
>> apart. In the first case, the second measurement doesn't much extra
>> information - you already knew that this location had 10 fish/m^2 -
>> but in the second case the second measurement is much more useful.
>> There were 10 fish/m^2 even though the locations were far apart! It
>> might be due to the 50m depth!
>> 
>> You can do a GLM for starters, but its important to check the
>> independence assumption of the GLM otherwise you could be getting
>> significant effects when none are really there.
>> 
>> Barry
>> 
> 
> 
> 
> -- 
> *Suzana Stutz Reis*
> Mestranda do Programa de P�s Gradua��o em Ci�ncias Biol�gicas -
> Comportamento e Biologia Animal
> Universidade Federal de Juiz de Fora, MG, Brasil.
> 
>    [[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo