[AniMov] Interesting paper

Adam T. Ford atford at gmail.com
Thu Apr 26 12:55:57 CEST 2012


Hello Animove,

I was trying to work through a fairly obvious but elusive problem in
the world of resource selection using animal movement data. I would
like to include contextual variables in the model, information that
doesnt change within a cluster, but does change among them. Here,
cluster refers to the matched observed-random samples used in a step
selection function (eg Fortin et al 2005; Forester et al 2009).

Consider the following example data, with 4 randoms for every observed GPS fix:
FID,     Cluster,  OBSERVED ,   % Cover, Time,    Frag,
1       1       1       10      14:00   40
2       1       0       20      14:00   40
3       1       0       30      14:00   40
4       1       0       30      14:00   40
5       2       1       25      14:30   20
6       2       0       50      14:30   20
7       2       0       80      14:30   20
8       2       0       4       14:30   20
9       3       1       5       15:00   80
10      3       0       25      15:00   80
11      3       0       39      15:00   80
12      3       0       80      15:00   80
13      4       1       1       15:30   10
14      4       0       6       15:30   10
15      4       0       50      15:30   10
16      4       0       80      15:30   10

The FID is the unique observation for each sample line in the SSF model.
CLUSTER is the unique ID from the observed data (GPS fixes)
OBSERVED is the binary response variable for the SSF model (eg, to use
in a clogit function in R)
% COVER is the classical SSF sample variable, it changes among samples
within a cluster
TIME and FRAG are variables that change across Clusters but not within
it. For example, the time if day when the sample was taken or the
amount of landscape fragmentation in the area surrounding the observed
fix.

I would like to explicitly incorporate contextual variables like TIME
and FRAG within the model to see if there is an interaction with
habitat selection. This cannot be achieved with the traditional coxPH/
matched case-control logistic regression approach. I have heard that some people
are moving to GEE (geepack) for modelling habitat selection now, has
anyone this problem or know of ways to get at it ?

Best,

Adam T. Ford
Ph.D. Candidate
Department of Zoology
University of British Columbia
Biological Sciences Building
Room 4200, 6270 University Boulevard
Vancouver, B.C.
Canada V6T 1Z4

"Plants are what prey eat."
                 -An ecologist



On Thu, Apr 26, 2012 at 1:14 PM, Paolo Cavallini <cavallini at faunalia.it> wrote:
> Quantitative comparison and selection of home range metrics for
> telemetry data
> Graeme S. Cumming and Daniel Cornélis
> Article first published online: 19 APR 2012 | DOI:
> 10.1111/j.1472-4642.2012.00908.x
>
> http://onlinelibrary.wiley.com/doi/10.1111/j.1472-4642.2012.00908.x/abstract
>
> All the best.
>
> --
> Paolo Cavallini
> See: http://www.faunalia.it/pc
>
> _______________________________________________
> AniMov mailing list
> AniMov at faunalia.it
> http://lists.faunalia.it/cgi-bin/mailman/listinfo/animov


More information about the AniMov mailing list