Capture probability by trap type

[jCeradini]

Hi folks,

I am using Huggins robust design with 2 years of small mammal data. I had 2 trap types on my trapping grids, shermans and havaharts. I am trying to estimate capture probability by trap type but am having trouble choosing the best covariate for the job.

Trapping grids were not split 50/50 between the trap types but were ~2/3 shermans and ~1/3 havaharts. Grids had 10 columns with 12 traps per column (120 total traps). Each column had 8 shermans (S) and 4 havaharts (H) , arranged as S S H S S H etc. There are some issues with what traps are available to each individual since a trapping station only had 1 type of trap, but grid-level inferences may be reasonable.

I initially used an individual covariate that was the ratio of captures in shermans to havaharts when shermans are coded as 1 and havaharts as 0, but I'm not sure how to interpret the beta since the number of S and H are not equal on a grid....maybe this is not an issue?
I also tried a "trap preference index", where the S:H ratio for each individual is divided by the proportion of S on that grid. If it's greater than 1 there's a S pref, < 1 is a H pref (with some buffer around 1 representing no pref). I then coded the preferences as 1's and 0's for the covariate. However, this gives me capture probability for individual's with a S pref, H pref, or no pref, which is not the same as capture probability by trap type.

I appreciate any advice.
Thanks!
Joe

[murray.efford]

This looks to me like a spatially explicit capture-recapture problem - simply use trap type as a detector-level covariate in 'secr'.
Murray

[jCeradini]

Thanks so much for the response, Murray.

I'm in the midst of completing my analysis in MARK without a trap type covariate (decided to analyze trap preference outside of MARK). I will keep SECR in mind for the future - it would be great to get a trap type specific capture prob estimate.

Joe

[gwhite]

Joe:

Capture probabilities are a characteristic of the animal, not of the type of trap. What you are really asking is which trap is more effective, i.e., probability trap catches an animal given the trap type.
, or P(success|type).

The way to estimate trap effectiveness is to turn your data upside down and use the single-season occupancy data type. Each trap is now a site and you build an encounter history for each trap, with either 2 groups to model the 2 trap types, or an individual covariate of trap type. Each trap has an encounter history with 1 representing a small mammal capture, 0 trap empty. Then p in the occupancy model is now trap effectiveness, and you can compare the p estimates across trap type, even with your unbalanced number of traps of each type.

The parameter psi is now the probability that a trap can capture an animal. At first I was thinking that you would want to fix psi = 1, because every trap has a chance of catching a small mammal. But I could imagine a case where with very low density of small mammals and LOTS of traps, this assumption might not be true. So I would consider models with both psi estimated and psi fixed equal to one. In either case, the ratio of the p estimates represents an estimate of the relative efficiency of the trap types.

Gary

[jCeradini]

Thanks for coming up with another alternative Gary!

I definitely was thinking of trap type as a characteristic of the animal (as the proportion of captures in a certain type), but obviously that covariate won't accomplish what I need.

Thanks,
Joe