Density estimation by trap covariate groups

[leopard1]

Dear Murray and Phidot Forum users

As I do not have a soil map, I have opted to use Murray’s “addCovariates” to mask command, and this has extrapolated orientation and soil type from the nearest traps to my habitat suitability mask. This has worked, however I get the following error when I try get a density estimate using the “derived(dataset, groups=Orientation)” command :

Error in `[.data.frame`(covariates(capthist), , groups) :
undefined columns selected

From my understanding it seems that SECR is trying to use my capture history to find the Orientation and Soil covariates. When I try model sex and age specific density using the “derived(dataset, groups=Sex)” command, this works fine but when I replace Sex with Orientation or Soil, I get the above error.

I’ve also read through the forum and am worried I don’t have enough recaptures (+-7 for my baited survey). Nonetheless, I’d appreciate any help on the above error, and whether my logic to run this type of analysis is even valid.

Thank you Murray and forum users

Alex

[murray.efford]

Hi Alex

It would help to see your actual call to secr.fit, but I think you may be confusing different types of covariates. Mask covariates are used to model variation in density across space. Your use of 'derived' suggests that you fitted a model by maximising the conditional likelihood (CL = T), which is incompatible with modelling density. Specifying group=Orientation without quotes in derived() is never going to work (did group=Sex really work?) because there is no object called Orientation, but even group = 'Orientation' will make sense only if 'Orientation' is the name of an individual covariate by which you want to post-stratify the Horvitz-Thompson-like density estimate. As I read your post 'Orientation' is the name of a mask covariate.

I guess you may be able to distinguish exclusively northern vs southern animals and include that as an individual covariate. Probably best to do this by manually adding a column to the capture file (the R route is complicated). If the northern and southern populations really are distinct they could also be treated as 'sessions'.

A couple of other comments: as you probably sense, with so few cameras and recaptures you have very little power to detect population-level differences (elaborate analysis cannot make up for lack of data). The name of the package is 'secr' not 'SECR' (R disciplines us to be case-sensitive!)

Murray

[leopard1]

Thanks a ton Murray! To be frank I think I was a bit ambitious with the data I had at hand!

I've followed the simple model options provided in your guide with good success. Thanks again for your help.

Bests

Alex