increased density estimates with addition of adjustment terms

[Karen Newlon]

Dear List,

I am comparing breeding bird densities between sheep- and cattle-grazed
riparian areas using data from point transect surveys conducted at 60 point
transects (30 points in cattle-grazed and 30 points in sheep-grazed). I am
using the MCDS engine in Program Distance to estimate the bird densities.
I initially modeled the detection function (half-normal) with one covariate
and no adjustment terms. I then incorporated one adjustment term
(cosine). This vastly improved the model fit as well as model support (via
AIC), but it also dramatically increased my density estimates. I should
also note that I am fitting a global detection function and using that to
estimate detection probability by stratum (i.e., grazing treatment). In
other words, both the global and stratum detection function boxes are
checked in the model definition properties.

Have others noticed appreciable differences in density estimates with the
addition of adjustment terms? If so, what is causing these differences?

Thanks,

Karen Newlon
k...@montana.edu

--
distance-sampling list home page:
http://www.jiscmail.ac.uk/lists/distance-sampling.html
To unsubscribe, go to the home page and click "Join or leave the list",
or email jisc...@jiscmail.ac.uk with the following message body:
leave distance-sampling
Before posting to the list, read the list guidelines at
http://www.jiscmail.ac.uk/files/distance-sampling/
distance-sampling.welcome
Any questions, email the list owner at
distance-sam...@jiscmail.ac.uk
--

[Steve Buckland]

> > Have others noticed appreciable differences in density estimates with the
> > addition of adjustment terms? If so, what is causing these differences?

Karen, you would need to look at the histograms in the Distance output, to
understand what is going on. My guess is that you have 'too many'
detections close to the point, relative to what the half-normal model
'expects'. I suspect a problem with one of the assumptions - in which
case, you may get more robust estimation with the half-normal model
(albeit with poor fits - what were the gof test statistics like?) than
with the model with adjustments. Possible causes:

1. If you record all birds detected, not just singing males, then you may
find that females (possibly young too?) tend to be detected only if
they're close to the point, whereas singing males are detected at larger
distances. If this is the case, you would be better (at least for most
species) to attempt to estimate density of singing males only, leaving out
the visual detections of non-singing birds.

2. If your field procedures are such that birds that are moving around
tend to be recorded too close to the point, this problem can arise.
The method is a 'snapshot' method, in the sense that it attempts to
quantify density at a moment in time. The longer you stay at each point,
the greater the chance that a bird that is initially some distance from
the point will approach it, and be detected. This can be made worse if
the observer is aware that a bird is moving around in the vicinity of the
point, so records a distance very close to the point as an 'average'
location.

So I would look closely at your data, and try to understand what features
of the data cause the problem. Then examine your field procedures, and
think how best to change them, to avoid the problem.

Steve Buckland
_________________________________________________________________________
Stephen T. Buckland
CREEM, The Observatory, Buchanan Gardens, St Andrews KY16 9LZ, Scotland
Tel. +44-(0)1334-461841 Secretary +44-(0)1334-461842
Fax +44-(0)1334-461800 e-mail st...@mcs.st-and.ac.uk
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
Professor of Statistics http://www.mcs.st-and.ac.uk/
Co-Director, NCSE http://www.ncse.org.uk/
CREEM http://www.creem.st-and.ac.uk/
RUWPA http://www.ruwpa.st-and.ac.uk/
_________________________________________________________________________