Problem SE and CV with bootdht

[Verity Miles]

Hi DS community,

I have recently obtained an odd SE/cv resultS with my CT-DS data when bootstrapping and I'm wondering if you can help me figure out why this has happened?

The data are from a CT survey with 149 cameras deployed for approx. two weeks. 66 (44%) of the cameras had detections.

When running dht2, I get some sensible results:

Estimate    se       cv         LCI        UCI         df

4.6328       1.213  0.262  2.7861  7.7036   168.649

But when I run bootdht I get these results, with the high SE and CV suggesting something isn't right:

median  mean   se           lcl       ucl       cv
5.19       17.4      380.73   3.07   7.58    73.42

Does anyone know why this might be? I'm happy to provide code and data files if that's useful.

Many thanks for your help!

Verity 

[Stephen Buckland]

Verity, you’ll have noticed that your bootstrap confidence limits are more-or-less OK.  Most (more than 97.5%) of your bootstrap estimates are probably perfectly OK, but in a small number of bootstrap resamples, you’re getting an exceptionally large estimate of population size.  You would probably need to look through the bootstrap estimates, and identify those very large ones, and look at the data from the corresponding resample to understand what is going on.  Perhaps 1 or 2 of your cameras recorded many detections very close to the camera – possibly a result of animal response to the camera, or of a well-used track passing very close to the camera.  Resamples that select such cameras more than once may cause the difficulty.  If you can identify the cause, it may suggest a solution (e.g. left-truncation if the cause is animal reaction to the camera).  Or you might just opt to stick with the analytic estimates …

 

Steve Buckland

--
You received this message because you are subscribed to the Google Groups "distance-sampling" group.
To unsubscribe from this group and stop receiving emails from it, send an email to distance-sampl...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/6db7e4c4-b99e-45c1-8e75-0e4a71aa54a3n%40googlegroups.com.

[Verity Miles]

Thanks so much, that was relly helpful. I had a look at the bootstrap resamples and did indeed have one riduculously high density estimate. Looking at the data, I noticed it could have been because I left-truncated the data at 2m but one camera had many detections at around 2.2, so perhaps I left-truncated by too much? I tried truncating at 1.5m instead and got a much more sensible result from the bootstrap:

median  mean   se       lcl     ucl    cv

4.51       4.57     1.02   2.81  6.74  0.23

Appreciate your help,

Thank you,

Verity 

[Stephen Buckland]

Verity, if it was just the one resample with a very high estimate, then it’s quite likely that simply re-running with the same left-truncation value would also have worked. If the animal at 2.2m had reacted to the camera by approaching it and staring at it for example, then you would need to left-truncate more, not less. E.g. at 2.5m.  You should be able to tell from the images whether you’re getting reactions to the camera.

Steve 

Stephen T. Buckland

CREEM, The Observatory, Buchanan Gdns, St Andrews KY16 9LZ, Scotland

e-mail st...@st-andrews.ac.uk

The University of St Andrews is a charity registered in Scotland:No SC013532

---

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Verity Miles <verity...@gmail.com>
Sent: Thursday, November 2, 2023 9:54:19 PM
To: distance-sampling <distance...@googlegroups.com>
Subject: Re: [distance-sampling] Problem SE and CV with bootdht

 

To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/21aaf0b1-22dc-4179-867e-5846402d42e6n%40googlegroups.com.