www.phidot.org

Murray,

A followup from my previous and now resolved issue with Noneuc and hov....

Is there a way to get "h2" as a variable for Noneuc? Basically a sex difference (or more generally an individual covariate) in the spatial effect for Noneuc. The below model didn't work out.... "h2 not found".

secr.fit(input, model=list(D~elev,g0~h2, noneuc~elev+h2),fixed = list(sigma = 1), hcov="Sex",detectfn = 0,
details = list(userdist = fn2), mask=mask.poly)

Thanks
brian

Unfortunately not. The design matrix for each of the detection parameters (lambda0, sigma etc.) is constructed by secr.design.MS and includes finite mixtures, individual differences etc. but no mask-level spatial effects. The design matrix for density-like spatial parameters (D) is constructed by D.designdata and allows mask-level covariates. This distinction was built into 'secr' early on and has generally worked, but it does exclude some models (e.g. continuous spatial variation in sigma). 'noneuc' was added on the density side as an afterthought, and it's there that issues arise when you use it to model detection parameters (the sigma=1 trick). I'm ambivalent about finite mixtures, so don't see that as a major loss, but I wish there was a tidy way (any way?) to allow for reciprocal variation in lambda0 and noneuc sigma.

If sex is your problem (and all animals are known-sex) you can always try coding as sessions (or groups, but I use groups so seldom I fear that may expose bugs).

Murray

Thanks for the input!

I agree with your sentiment on finite mixtures. Unfortunately, not all the sexes are known in this case. I have a strong suspicion that modeling noneuc spatially is much more important than just using sex on sigma, but I know I will get some questions because it's a large cat, and there is good evidence of sex differences in movement.

My next question was about groups, but I see that might not be a path I want to go down!

Thanks again!

brian

Hi
I don't know that groups won't work and they're worth a try, but do sing out if something looks wrong. And you do have to assign all animals to a group.

Merely finding sex differences in sigma does not mean you have to model them. I still maintain that this source of heterogeneity is overrated because it is balanced (or at least moderated) by reciprocal variation in lambda0. What matters is sex difference in sigma-adjusted lambda0 (as shown e.g. in jaguars).
Murray