Comparison of unmarked and PRESENCE
jmil
I'm wondering how unmarked and the PRESENCE software (
http://www.mbr-pwrc.usgs.gov/software/doc/presence/presence.html )
compare.
For example, I just tried a couple of models using the wt data
provided with unmarked (widewt.csv) in both unmarked and PRESENCE
(i.e. I used exactly the same data in both software). I just tested
single variable models (I don't know much - anything really - about
the data so am not sure how sensble these models are from an
ecological perspective). Here are the model comparison tables (hope
the formatting survives):
i) unmarked
n nPars AIC deltaAIC AICwt
Elev 237 3 496.51 0 9.72E-01
forest 237 3 503.63 7.1261 2.76E-02
date 237 3 529.69 33.1812 6.06E-08
ivel 237 3 529.98 33.4725 5.24E-08
length 237 3 530.7 34.1937 3.65E-08
ii) PRESENCE
Model AIC deltaAIC AIC wgt Model Likelihood no.Par.
psi(elev),p(.) 515.63 0 0.9969 1 2
psi(forest),p(.) 528.49 12.86 0.0016 0.0016 2
psi(length),p(.) 528.64 13.01 0.0015 0.0015 2
psi(.),p(ivel) 563.44 47.81 0 0 2
psi(.),p(date) 582.21 66.58 0 0 2
Although some variation in AIC might be expected due to the specifics
of different algorithms in R vs PRESENCE I am surprised that the
length variable goes from being 'worst' model (according to AIC) in
unmarked to third ranked model in PRESENCE (although elevation is
clearly best performing model according to AIC in both cases).
Could someone shed some light on why this might be? Am I incorrect to
assume that results for unmarked and PRESENCE for the same data should
be very similar?
I found similar differences between models in unmarked and PRESENCE
for my own data. It's great to see this modelling capability now
available in R - just need to understand better how it compares to
what other software is doing...
Thanks for your help,
James
Andy Royle
PRESENCE might be miscounting parameters (3 is correct I think). Thats
obviously not the cause of whats going on. The two platforms should agree
in terms of log-likelihood difference which they don't, so someone is
wrong.
I modified a function from the book (see
http://137.227.242.23/pubanalysis/roylebook/panel3pt8.fn.Rd
)
and I get results that agree pretty closely with unmarked. (see table
below). Seems like they should be closer though. I think I used a log
(length) transformation and "intensity" may or may not be exactly
"ivel" (different transformation possibly). That would explain the slight
difference.
Seems like PRESENCE is the odd man out here......
andy
> panel3pt8.fn()
pconst psiconst elev1 elev2 forest length intensity
date1 date2 aic
1 1.243951 -0.7239104 0.9069355 NA NA NA NA
NA NA 496.4038
2 1.338635 -0.7555758 NA NA 0.7644742 NA NA
NA NA 503.5299
3 1.323009 -0.6693010 NA NA NA -0.1747386 NA
NA NA 529.4901
4 1.223496 -0.6563227 NA NA NA NA 0.2367342
NA NA 529.8763
5 NA NA NA NA NA NA NA
NA NA NA
6 1.274943 -0.6639075 NA NA NA NA NA
0.2095756 NA 529.5850
7 NA NA NA NA NA NA NA
NA NA NA
8 NA NA NA NA NA NA NA
NA NA NA
9 NA NA NA NA NA NA NA
NA NA NA
10 NA NA NA NA NA NA NA
NA NA NA
J. Andy Royle
Research Statistician
USGS Patuxent Wildlife Research Center
12100 Beech Forest Rd.
Laurel, MD 20708
http://profile.usgs.gov/professional/mypage.php?name=aroyle
andy_...@usgs.gov
phone: 301-497-5846
fax: 301-497-5545
Book: "Hierarchical Modeling and Inference in Ecology: The Analysis of Data
from Populations, Metapopulations and Communities" by J. A. Royle and R.M.
Dorazio.
unmarked: A very useful R package for fitting certain hierarchical models
using likelihood methods. Available from:
http://cran.case.edu/web/packages/unmarked/index.html
-----unma...@googlegroups.com wrote: -----
To: unmarked <unma...@googlegroups.com>
From: jmil <jamesdam...@googlemail.com>
Sent by: unma...@googlegroups.com
Date: 03/10/2010 05:56PM
Subject: [unmarked] Comparison of unmarked and PRESENCE
rcha...@nrc.umass.edu
I just compared occu to PRESENCE using simulated data and I got
virtually identical parameter estimates and AIC values. So, this
should ease any worry over a major difference between the two, but it
doesn't explain the discrepancy you encountered. If you want to try
and replicate this, I used the data and models from the Class4.R
script found here:
http://sites.google.com/site/hierarchicalmodelingcourse/home/r-scripts
I would make sure that you have a recent version of PRESENCE.
Richard
UMass Amherst
Natural Resources Conservation
nrc.umass.edu/index.php/people/graduate-students/chandler-richard/