Spatial and Temporal replication

213 views
Skip to first unread message

Paul J Taillie

unread,
Jun 5, 2012, 2:59:14 PM6/5/12
to unmarked
I am working with a pretty standard point-count occupancy data set: I
have 232 sites that were surveyed between 1 and 5 times in one
season. (single-season, closed). These sites are then distributed
across 108 patches of habitat with 1 - 7(or so) sites per patch. Two
of my covariates are specific to the patch (patch area and patch
isolation), but then two of them are specific to the site/point (basal
area and grass cover). I have successfully fitted models using the
site as the experimental unit, but there are issues with independence
between sites for the patch-level covariates. I am wondering if there
is a way to set this up in an unmarked frame with a different
experimental unit for different covariates (like a split-plot
design).

My only thought so far is that each spatial/temporal replicate
combination is one actual replicate. So that say one patch had 3
sites within it, two of the sites were sampled 4 times, and the third
was sampled 3 times. then the experimental unit would be the patch
with 11 replicates, averaging the site-level covariates across those
11 replicates. This method is straightforward but some information is
lost. If there is a way to accommodate this split-plot type design
please let me know. Any advice would be appreciated.

Thanks,

Paul Taillie
North Carolina State University

Matthew Krachey

unread,
Jun 5, 2012, 4:58:27 PM6/5/12
to unma...@googlegroups.com
Couldn't you just add a covariate for the patch id in the occupancy model?

Matthew

Kery Marc

unread,
Jun 6, 2012, 3:14:28 AM6/6/12
to unma...@googlegroups.com
hi Paul and Matthew,

fitting patch ID as a covariate would not account correctly for the two patch-specific covariates, but probably is the most appropriate that you can do in unmarked. To fully accommodate the nested (split-plot) structure, you would have to conduct an analysis in BUGS.

Regards -- Marc
________________________________________
From: unma...@googlegroups.com [unma...@googlegroups.com] on behalf of Matthew Krachey [aero...@gmail.com]
Sent: 05 June 2012 22:58
To: unma...@googlegroups.com
Subject: Re: [unmarked] Spatial and Temporal replication

Paul J Taillie

unread,
Jun 7, 2012, 12:58:08 PM6/7/12
to unmarked
Thanks Marc. I will try that. Is there a section in your book that
addresses fitting models with this nested design in BUGS?

Paul

On Jun 6, 3:14 am, Kery Marc <marc.k...@vogelwarte.ch> wrote:
> hi Paul and Matthew,
>
> fitting patch ID as a covariate would not account correctly for the two patch-specific covariates, but probably is the most appropriate that you can do in unmarked. To fully accommodate the nested (split-plot) structure, you would have to conduct an analysis in BUGS.
>
> Regards  --  Marc
> ________________________________________
> From: unma...@googlegroups.com [unma...@googlegroups.com] on behalf of Matthew Krachey [aerot...@gmail.com]

Kery Marc

unread,
Jun 7, 2012, 1:15:47 PM6/7/12
to unma...@googlegroups.com
Hi Paul,

not directly, but the usual way how people deal with these split-plot designs in the age of random effects (= now) is to fit a set of random effects for the patch factor. That should do it.

Kind regards -- Marc

________________________________________
From: unma...@googlegroups.com [unma...@googlegroups.com] on behalf of Paul J Taillie [pault...@gmail.com]
Sent: 07 June 2012 18:58
To: unmarked
Subject: [unmarked] Re: Spatial and Temporal replication

Dan Linden

unread,
Jun 7, 2012, 2:24:19 PM6/7/12
to unma...@googlegroups.com
Hey Paul,

A good reference for building multilevel models is Gelman and Hill (2007) - Data Analysis Using Regression and Multilevel/Hierarchical Models.  They describe exactly what Marc speaks of, where random effects at some spatial level of a nested hierarchy are defined and allow you to model the effect of covariates that are specific to that level.  All of their model code is written in R and BUGS.


http://www.stat.columbia.edu/~gelman/arm/


-----------------------
Daniel W. Linden, Ph.D.

Postdoctoral Research Associate
Dept. of Fisheries & Wildlife
Michigan State University
480 Wilson Road, Room 13
Natural Resources Bldg.
East Lansing, MI 48824

Cell: (401) 529-5675
Email: lind...@msu.edu

Paul J Taillie

unread,
Jun 13, 2012, 11:17:15 AM6/13/12
to unmarked
Thanks Matt, Marc and Dan for the help. I've got some reading to
do.

-Paul
> Email: linde...@msu.edu
Reply all
Reply to author
Forward
0 new messages