[R-sig-eco] Help with nonlinear population trends & binomialregression

[Bret Collier]

Matt,
Check out the SemiPar package maintained by Matt Wand and see the book: 
Semiparametric Regression.  2003, Ruppert, Wand, and Carroll.  There are 
some nice code in the book and  online for Bayesian logistic modeling 
using penalized splines as well.

Bret

On 12/17/2010 3:04 AM, Alexandre Villers wrote:
> Hey Matthew,
>
> You should have a look at the Zuur et al. 2009 and the chapter 6 dealing
> with Violation of Independance (page 143). They explain how to fit GAMMs
> for different locations (species in your case). You can select models
> through AIC (but I guess you have to be pretty cautious with this type
> of model in model selection process).
> You should also consider adding a temporal correlation structure given
> you have times series (this is also explained in this book).
>
> HTH
>
> Alex
>
> On 17/12/2010 10:55, Aitor Gastón wrote:
>> Matthew,
>>
>> A more flexible model may be a better option than fitting two separate
>> lines (e.g. restricted cubic splines, rcs function of the rms package
>> or a generalized additive model, mgcv package). A restricted cubic
>> spline with 4 knots may be enough to model the response curve that you
>> describe.
>>
>> Hope that helps,
>>
>> Aitor
>>
>>
>> --------------------------------------------------
>> From: "Matthew Forister" <forister at gmail.com>
>> Sent: Friday, December 17, 2010 7:19 AM
>> To: <r-sig-ecology at r-project.org>
>> Subject: [R-sig-eco] Help with nonlinear population trends &
>> binomialregression
>>
>>> Hello, I'm hoping someone can point me in a new direction with this
>>> particular issue...
>>>
>>> I have count data for many species across 20-30 years. I know that many
>>> species are declining, probably in association with habitat
>>> destruction, and
>>> I have been using binomial regression to model the declines
>>> (e.g. glm(cbind(presence,visits-presence)~years,binomial). I have
>>> also used
>>> the glm.binomial.disp function for overdispersion. So far so good, but
>>> here's the issue: not all species decline in the same way...
>>>
>>> Some go down steadily over the years, while others will be holding
>>> steady
>>> and then suddenly start on a decline. There are other patterns, but
>>> those
>>> two are dominant and I would like to be able to say that different
>>> species
>>> have these different dynamics. But how do I quantify those different
>>> curves? I have played with fitting quadratic and cubic terms within the
>>> binomial
>>> regression (e.g.
>>> glm(cbind(presence,visits-presence)~years+I(years^2),binomial)),
>>> and then comparing models with AIC to think about the better fit of the
>>> model with the quadratic. That kinda makes sense for some species,
>>> but it's
>>> far from satisfying... In the case I described (a species holding
>>> steady for
>>> 2 decades and then going into a steady decrease for another 10), what it
>>> really looks like is two different lines, one flat and one precipitous.
>>>
>>> Is there a way to ask if a given relationship is better fit by two lines
>>> than one? any other hints on how to describe these kinds of dynamics?
>>>
>>> thanks!
>>> Matt
>>>
>>>
>>> --
>>> Matthew L Forister
>>> Assistant Professor
>>> Department of Biology / MS 314
>>> 1664 N. Virginia St.
>>> University of Nevada, Reno
>>> Reno, Nevada 89557
>>> --
>>> E-mail: forister at gmail.com
>>> Office phone: (775) 784 - 6770
>>> Lab phone: (775) 784 - 7083
>>> Fax: (775) 784 - 1302
>>> Office: room 257 Fleischmann Agriculture Building
>>> --
>>> Webpage:
>>> https://sites.google.com/site/greatbasinbuglab/
>>> --
>>>
>>> [[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
>>>
>>
>> _______________________________________________
>> R-sig-ecology mailing list
>> R-sig-ecology at r-project.org
>> https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
>
>