Distance R issue with uniform key

37 views
Skip to first unread message

Stephanie Augustine

unread,
Sep 6, 2022, 1:57:10 PM9/6/22
to distance-sampling
Hi all,

New to distance sampling at a new job and I'm trying to move into the R framework, having never used Distance software and pretty comfortable in the R environment. I have data that was analyzed through Distance and am attempting to replicate results before moving on to new analyses. Data are bird detections in line transects. For context, it is open grassland surveys (histogram of detection distances attached). 

In the original Distance analysis, Uniform key had the lowest AIC, with pretty much identical AIC between hermite poly, simple poly, and cosine adjustments. Half normal was within 2 AIC units. 

In R, I was able to get the half normal to work and replicate the Distance software results. But when the key is uniform in R, it gives me a very low detection probability equating to an insanely high density and abundance estimate. 

The density under half-normal is 0.1 birds/ha (reasonable, abundance estimate corresponds well to expectations), vs. uniform says 14 birds/ha (far too high). 

Where I think the low detection probability comes from 1/w, in this case, with the right 10% truncated, is 149.86. 

Therefore, 1/149.86 = 0.00667 = average P for uniform key.
In Distance, the p = 1 for uniform key.

Is there a setting in Distance to indicate this perfect detection because distance doesn't change with detection that has to be differently specified in R? I've been up and down this forum and the Distance R package information without success, and I suspect it's because uniform detection is not usually a good thing and can be a sign of flawed design. 

Any assistance with parsing this issue is greatly appreciated! 


DistHist.PNG

Eric Rexstad

unread,
Sep 6, 2022, 2:12:09 PM9/6/22
to Stephanie Augustine, distance-sampling
Stephanie

Rather than focus upon differences between the contrasting pieces of software. I think the telling point is your next-to-last statement: data that you provide in your figure isn't great.  It doesn't appear from the histogram that detectability is influenced by distance from the transect.  I would be surprised that the half normal has an AIC close to the AIC for the uniform.  I don't expect the adjustment terms are having any influence upon the shape of the detection function; hence "pretty much identical AIC values".

There isn't a model being fitted when using a uniform key without adjustments; it is effectively a strip transect.  So density is simply number detected divided by area covered by the sampling effort.

Happy to discuss off-list if you want to discuss.

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Stephanie Augustine <saugu...@ecoinst.org>
Sent: 06 September 2022 18:57
To: distance-sampling <distance...@googlegroups.com>
Subject: [distance-sampling] Distance R issue with uniform key
 
--
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/89cbfde2-8dd2-4158-b71a-b94ac9add72dn%40googlegroups.com.

Stephen Buckland

unread,
Sep 6, 2022, 5:15:22 PM9/6/22
to Eric Rexstad, Stephanie Augustine, distance-sampling
Adding to Eric's comment, f(0)=1/w, whereas p= 1/(w.f(0)) which is 1 for the uniform, as in the original analysis. Not sure how these 2 quantities have been confused in the analysis you have run.

Steve Buckland 

Stephen T. Buckland
CREEM, The Observatory, Buchanan Gdns, St Andrews KY16 9LZ, Scotland
Professor of Statistics www.st-andrews.ac.uk
The University of St Andrews is a charity registered in Scotland:No SC013532

From: 'Eric Rexstad' via distance-sampling <distance...@googlegroups.com>
Sent: Tuesday, September 6, 2022 7:11:52 PM
To: Stephanie Augustine <saugu...@ecoinst.org>; distance-sampling <distance...@googlegroups.com>
Subject: Re: [distance-sampling] Distance R issue with uniform key
 
Reply all
Reply to author
Forward
0 new messages