Estimating habitat selection function for a population (multivariate model)

[Victória]

Hi everyone,

I have a methodological question. I am working with a population of 9 animals and 4 environmental covariates. For each individual, I first fitted univariate habitat selection functions, and then I used ctmm:mean to estimate population-level coefficients.

Then, I wanted to adjust multivariate models. So, I fitted several candidate models for each individual (starting with 4 covariates, then 3, then 2…), and ranked them using AICc and BIC to identify the best-fitting models. This gave me multivariate models that best explained habitat selection at the individual level (each individual had different covariates associated with their movement).

My question is: is there a way to extend this approach to obtain a multivariate model estimated at the population level, rather than only at the individual level?

Thanks a lot for your help!

Best,
Victória

[Christen Fleming]

Hi Victória,

The recommended way to do this is to first run rsf.select() on the individuals, which will select for the individual-level covariates. After this, mean() will select for the population-level covariance parameters, which, in total, are often too many for the number of individuals tracked. E.g., 4 covariates would produce 14 population-level parameters (a 4-component mean + a 4x4 covariance matrix) - which cannot be estimated from 9 animals.

Best,

Chris

[Victória Dedavid Ferreira]

Hi Chri, thank you so much for your response!
What I did was essentially that : I ran rsf.fit for all individuals using AICc as the selection criterion, and then used mean() to average the best estimates (Betas) for each parameter.
Given my small sample size, as I understand it, estimating the full covariance matrix with 4 covariates (14 parameters) is not feasible.
In this case, can I still interpret the population-level estimates from the 4-component mean (the averaged Betas) as population-level RSF parameters, even without estimating the full covariance structure? 

Thank you 

--
You received this message because you are subscribed to the Google Groups "ctmm R user group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ctmm-user+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ctmm-user/27a04343-c7cd-403b-9bf4-5e4e9ad58fe0n%40googlegroups.com.

--

Victória D. Ferreira

[Christen Fleming]

Hi Victória,

Yes, if I understand you, this is the purpose of the automated model-selection algorithms behind mean().

Best,

Chris