negatively biased density estimates

[asiren]

Hello,

I'm comparing live and camera trap density estimates of American marten using secr and SPACECAP (I'm not comparing software packages; rather, I'm using them to verify results).

Both packages provide nearly identical density estimates for both field methods. However, the live trapping estimates are consistently lower (~20 marten/100 km2 lower). I think that this negative bias is occurring because of higher capture heterogeneity for live trapping (marten typically either become “trap shy” or “trap happy” from live trapping) and I have evidence of this occurring (10 of 15 were recaptured during live capture compared to 14 of 15 for camera trapping and a couple radio-collared marten avoided live traps but visited camera traps). From what I’ve read density estimates can become negatively biased from “trap shyness” (Wegge et al. 2004) and that capture heterogeneity can negatively bias density estimates (Williams et al. 2002). I think that is the reason for the lower density estimate but I’d like some reassurance. FYI… The trap density is sufficient (3.4 traps/radio-collared marten home range) and the highest trap saturation that occurred during an occasion was 43% so I don’t think that trap saturation created the negative bias.

I'd like to evaluate capture probability between the 2 methods. However, because estimates of p are typically negatively biased when estimating density for single capture devices (Efford, Borchers, and Byrom 2009) I haven’t been able to compare the capture probability between the 2 methods. I've tried to calculate capture probability for the live trapping using ip.secr but it isn't converging likely because sample size are small (30 captures of 15 individuals in 6 days). For some reason I'm having trouble including the estimates from my start values. I’m concerned that the small sample size might be influencing the lower density estimates more than the capture heterogeneity.

Any insight and/or references that I can read would be appreciated.

Thank you,

Alexej

[murray.efford]

Hi Alexej
I'm not sure I can add much to your analysis - here are a few random thoughts.
1. 'Consistently lower' implies a lot of information we don't see - my impulse is to question whether that conclusion stands up statistically, or whether it's an artefact of some other design issue (differing areas sampled).
2. It's good to keep separate trap responses (learned or behavioural responses sense of Otis et al 1978) and persistent individual differences (individual heterogeneity). Heterogeneity causes negative bias, behavioural response bias can be negative or positive (I think).
3. By 'p' I suppose you mean 'g0'. But isn't it obvious that one sampling method delivers more detections per detector per day than the other method for the same set of animals? It seems what you really want is to assess whether there is (much) greater individual heterogeneity or behavioural response to traps than cameras: that will be hard. The difference you cite in recapture frequency may simply reflect the overall detection rates in the two methods. 2 radiocollared marten = interesting anecdote!
4. I would question whether you really have enough (trap) detections to be asking deep questions about the reason for any difference.
Murray

[asiren]

Hello Murray,

Thanks for your response and I appreciate your comments. I agree, 30 captures of 15 individuals is a small data set. AICc model selection of the live trapping data indicated there were no clear top models, yet for both years of the camera trapping data the top model was one that included a behavioral and trap specific response; these datasets were data rich compared to the live trapping one. Since the sample size and effort for the live trapping was similar to what I found in the literature I wanted to make a comparison with camera trapping. I conducted live trapping on the heels of the camera trapping effort primarily to compare capture history and to take pictures of their throat patches during handling.

Again, I appreciate your feedback,

Alexej