Re: [unmarked] Accounting for survey occasions of different time lengths

118 views
Skip to first unread message

Matt Giovanni

unread,
Feb 26, 2013, 5:21:17 PM2/26/13
to unma...@googlegroups.com, luke.l...@gmail.com
Maybe not relevant, but I once used the "offset" argument to account for variation in area of sampling transects, so perhaps it can account for the variation in your sampling durations:

Null=pcount(~1 ~1+offset(log(AREA)),mixture="ZIP",umf)

Matt
__________________________________________
Matt Giovanni, Ph.D. | Research Biologist
Director, American Kestrel Partnership
The Peregrine Fund
5668 West Flying Hawk Lane, Boise, ID 83709 USA
208.362.8279 office | 402.617.3764 mobile
Check out my research website
Conserving birds of prey worldwide


On Tue, Feb 26, 2013 at 3:08 PM, Luke Powell <luke.l...@gmail.com> wrote:
I conducted a bird survey in which we used three consecutive observation periods (3 min pre-broadcast, 38 second broadcast, 5 min post-broadcast period) to model occupancy.   The current single-season occupancy modeling approach I'm using with UNMARKED (originally written by Ian Fiske) does not seem to account for those different lengths.  I know from watching the birds' behavior that they respond very aggressively to playback (thus increasing detectability), but  when I include samplingMethod as a variable (indicating that detectability varies among observation periods), the parameter estimates are nonsensical. In the best-fit model, the beta estimates for detectability during different periods are:
βbeforeBroadcast = [no estimate - reference category],
βduringBroadcast = -0.61 ± 0.39 SE, 
βafterBroadcast  = 0.75 ± 0.35 SE) and an 
intercept (β = 7.8 ± 1.9 SE).  

The estimates indicate that detectability is lowest during the broadcast, which is clearly not the case as indicated by my observations and by an independent analysis of these data.  I'm pretty sure what's happening here is that we're not accounting for the difference in survey length - the post-broadcast period is showing the highest detectability because it is the longest, and the during-broadcast is showing the lowest detectability b/c its the shortest. In the independent analysis, we statistically accounted for the difference in observation period lengths in a GLM framework (using poisson errors, as explained here: http://stackoverflow.com/questions/12208346/in-r-how-do-i-calculate-expected-values-in-a-chi-square-test-when-survey-length), but I'm not sure how to apply this technique to a single-season occupancy framework.  

FYI, the best fit model for detectability looks like this: 
fm.best <- occu(~ wind + jday + samplingMethod ~ choice, rublUMF, knownOcc = which(rubl$WetChoice == "driveby"))

Any advice on how to account for the different length of survey occasions would be greatly appreciated. 

Best, 

Luke L. Powell
PhD Candidate
School of Renewable Natural Resources
Louisiana State University
LPo...@LSU.edu or Luke.L...@gmail.com
cell: 516-313-0457;  Skype: LukeP9






--
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.
 
 

Murray Efford

unread,
Feb 26, 2013, 6:09:59 PM2/26/13
to unma...@googlegroups.com, David Borchers
Matt's right that varying duration is in the nature of an offset - we make the point in a capture-recapture paper accepted this week for Methods in Ecology & Evolution - but to be rigorous you need to keep an eye on the relevant scales. Probability of detection relates to cumulative hazard; how this plays out depends on your link scale, whether you're modelling counts or probabilities etc.
Murray
Reply all
Reply to author
Forward
0 new messages