pcount vs occuRN

262 views
Skip to first unread message

Viorel Popescu

unread,
Jan 25, 2012, 6:55:24 PM1/25/12
to unma...@googlegroups.com
Dear unmarked users,

I have a simple question: how to assess which type of model (Royle Biometrics or Royle-Nichols) is the better choice for my dataset?

I am trying to estimate bear abundance using track counts on 2-km stretches of forest roads in the Carpathians. There are 36 transects with 4 visits per transect. There are a lot of zeroes, and the track counts are low, ranging between 1 and 3 (mostly 1s). The occuRN models use the count data truncated to 1s.

So far I used model averaging for sets of pcount and occuRN models separately, and I obtain slightly different estimates of abundance. Can I select which models perform better by just comparing AIC values of occuRN and pcount models? I am basically using the same dataset for both types of models, but I was not sure if using AIC to compare across different types of models is correct.  

Thank you...

Best,
Viorel

Jeffrey Royle

unread,
Jan 25, 2012, 6:58:35 PM1/25/12
to unma...@googlegroups.com
Dear Viorel,
 you should use the pcount() function which fits the N-mixture model.  It probably isn't possible to compare the AIC because the likelihoods are scaled differently between the two models. In any case, the RN model assumes a binary observation whereas the N-mixture model the observation has support 0,1,2,3,...., which matches up with your situation.
regards
andy

Viorel Popescu

unread,
Jan 25, 2012, 7:14:36 PM1/25/12
to unma...@googlegroups.com
Dear Andy, 
This is very helpful; thanks a lot for clarifying... 

Cheers,
Viorel

Neil Carter

unread,
Feb 5, 2013, 5:02:21 PM2/5/13
to unma...@googlegroups.com
Dear Andy,
I had a similar question as Viorel.  I set traps for big cats with my observations per night ranging from 0 (1393), 1 (119), and 2 (6).  Is it appropriate in this case to use the N-mixture model or better to use the RN model?  I should add that the cats are spatially ubiquitous so there is not much heterogeneity across space in occupancy, however, there is heterogeneity in detection frequency among locations. 
I'm hoping to capture that dynamic.  Thanks.
--Neil  

Jeffrey Royle

unread,
Feb 5, 2013, 5:04:10 PM2/5/13
to unma...@googlegroups.com
hi Neil,
 if these are counts of unique individuals then the pcount() model is the right one, even if the counts are mostly or all 0/1.
regards
andy


--
You received this message because you are subscribed to the Google Groups "unmarked" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

Neil Carter

unread,
Feb 5, 2013, 5:22:09 PM2/5/13
to unma...@googlegroups.com
Hi Andy,
Great!  Thanks for the quick reply.  
--N

Pablo García Díaz

unread,
Feb 5, 2013, 5:32:02 PM2/5/13
to unma...@googlegroups.com
Just one question regarding the data of Viroel. Maybe further details about the abundance estimation that Viorel wish to do are needed; although maybe I'm not correct.

If he want to calculate bear abundance as no. of bears (or something similar), I think that he should take only presence/absence [0,1] of tracks in the transects and then use occuRN to estimate abundance (or whatever lambda should be interpreted as). Otherwise, using pcount() with track counts (no. of tracks per transect) will result in an estimation of the no. of tracks rather than the number of bears (that will require further assumptions to be derived from the track counts as no. of tracks leaved behind by each individual, etc).

Am I right with my interpretation?

Pablo





Date: Tue, 5 Feb 2013 14:02:21 -0800
From: nhcar...@gmail.com
To: unma...@googlegroups.com
Subject: Re: [unmarked] pcount vs occuRN

Jeffrey Royle

unread,
Feb 5, 2013, 6:24:26 PM2/5/13
to unma...@googlegroups.com
If the count is the number of unique individuals then pcount() is appropriate for estimating mean abundance.
If you're counting the number of tracks, as suggested by Pablo, then I agree, the model would produce something similar to an estimate of the population size of tracks (even that is questionable because tracks are not independent samples of anything!).

If it is not possible to deduce unique individuals, then reduce the data to simple presence/absence and use the RN model.

Kery Marc

unread,
Feb 6, 2013, 3:09:57 AM2/6/13
to unma...@googlegroups.com
Dear Pablo,

yes, you are right if what he counts is the number of tracks. However, I have seen people doing snow transects and trying to only count the number of unique animals that left a track. I don't know how practical that is but if this is possible, then of course the counts refer to individual animals and the abundance estimate to the total number of animals associated with the area that is sampled by the transect.

Regards  --  Marc

______________________________________________________________
 
Marc Kéry
Tel. ++41 41 462 97 93
marc...@vogelwarte.ch
www.vogelwarte.ch
 
Swiss Ornithological Institute | Seerose 1 | CH-6204 Sempach | Switzerland
______________________________________________________________
 
*** Introduction on Bayesian statistical modeling: Kéry (2010), Introduction to WinBUGS for Ecologists, Academic Press; see www.mbr-pwrc.usgs.gov/pubanalysis/kerybook
*** New book on Bayesian statistical modeling: Kéry & Schaub (2012), Bayesian Population Analysis using WinBUGS, Academic Press; see www.vogelwarte.ch/bpa
*** Upcoming workshops:
http://www.phidot.org/forum/viewforum.php?f=8

From: unma...@googlegroups.com [unma...@googlegroups.com] on behalf of Pablo García Díaz [her...@hotmail.com]
Sent: 05 February 2013 23:32
To: unma...@googlegroups.com
Subject: RE: [unmarked] pcount vs occuRN

Reply all
Reply to author
Forward
0 new messages