Sidra,

I don't know of any specific programs, but the approach I have taken to doing 
logistic regression on paired observations is simply to do a logistic 
regression on the following two of the four possibilities for each pair: 1,0 
and 0,1. You can disregard the 1,1 and 0,0 options, and the corresponding 
observations can be omitted from the analysis. This approach is based on the 
same principle that applies to chi-square tests of paired, non-independent 
samples (i.e., McNemar Asymmetry tests). Done this way, the analysis is just a 
simple logistic regression, wherein the 1,0 combination can be treated as one 
of the response categories and the 0,1 combination as the other.

I'm sure there will be some statistical hotshots on the forum who will 
criticize this approach and have better (more sophisticated) suggestions, but 
this approach seems reasonable to me and is familiar and easy to understand. 

Good luck,

Steve Brewer
________________________________________
From: Ecological Society of America: grants, jobs, news 
[[email protected]] on behalf of Sidra Blake [[email protected]]
Sent: Friday, December 09, 2011 3:08 PM
To: [email protected]
Subject: [ECOLOG-L] Paired Logistic Regression for Resource Selection

Ecologgers,

I wondered if any eco-statters could provide their opinion in statistics if 
they don't mind.  I require a statistical program that conducts matched 
(paired) logistic regression , which I believe is equivalent to conditional 
logistic regression (case-control) for a resource selection study.   Most of 
the ecological papers that use this statistical approach cite STATA or other 
software that I  , nor any local lab I know of, do not have licenses for. I 
have not seen anyone cite R for this approach in the literature I consult.  And 
online I am unable to find examples, in R,  that are in similar context to my 
own data.  I live and learn from examples via scientific literature and online 
code examples, so I am a bit discouraged at this point, hung up I guess.  
However, conditional logistic regression  does appear possible in R, from one 
source I found online via the survival package.  Though, that example was very 
limited in depth. This leads me to a few questions.


1) What are my statistical software options for matched logistic regression 
(with categorical and continuous data) - and which do users seem to prefer?
2) Has anyone used R for this statistical approach?
3) And,  has anyone been able to incorporate random effects (or mixed effects 
see Duchesne et al. 2010) by the experimental unit (ie-individual) into this 
design?

I admit I am new to logistic regression and resource selection analysis.   This 
means, I would deeply benefit from detailed examples for this approach.   I 
appreciate any feedback.

Please feel free to email me off the listserve at the email address below, and 
please use the subject heading of this post.

Thanks,

Sidra







Sidra Blake
Land Management and Demonstration Program
Mid-Columbia River NWRC
US Fish and Wildlife Service

MS Student
Natural Resource Sciences
Washington State University
sidra.blakeATwsu.edu

"We shall never achieve harmony with land, any more than we shall achieve 
absolute justice or liberty for people. In these higher aspirations, the 
important thing is not to achieve but to strive. "
~Aldo Leopold

Reply via email to