Validating 'secr' model. R2?

questions concerning anlysis/theory using program DENSITY and R package secr. Focus on spatially-explicit analysis.

Validating 'secr' model. R2?

Postby ctlamb » Mon Dec 05, 2016 2:52 pm

Hello,

Is there a procedure that I could use to validate/evaluate a 'secr' model's fit to the data? I have included two spatial covariates in my model and thus a beta and associated error are evaluated for these relationships, as in other linear models. Like a GLM, is there a measure such as R^2 that I can use to test how well my model is fitting the observed data?

I see previous posts (2010 and 2011) regarding goodness of fit procedures using sim.secr, and the potential issues with these approaches. But, unlike what I'm looking for, I believe these folks are looking for measures of overdispersion such as the C-hat used in MARK where the user is testing for differences from C=1.

Cheers,
Clayton
ctlamb
 
Posts: 56
Joined: Mon Nov 04, 2013 9:44 pm

Re: Validating 'secr' model. R2?

Postby murray.efford » Mon Dec 05, 2016 4:17 pm

Even R^2 has its detractors, but it _is_ nice to have a single number on a straightforward scale (proportion of variance 'explained'). I don't know of anything similar for SECR models, which of course doesn't mean it doesn't exist.

I assume, you are talking about models for spatial variation in density, and that you want specifically to compare a null model for spatially uneven detections (due to both spatial randomness of unobserved Poisson centres and stochastic detection) to a model with those effects plus spatial covariates. I see from a quick search that deviance-based versions of R^2 do exist, and perhaps these might be adapted... Not a straightforward task.

Murray
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand

Re: Validating 'secr' model. R2?

Postby ctlamb » Mon Dec 05, 2016 4:26 pm

Thanks, Murray!

My interest is in providing a description of the reliability of my density predictions through space. If I related density to spatial covariates, and interpolate density (using these covariates) across the trap grid, I would like to be able to quantify my confidence in these predictions. Clearly the density predictions across space are only as good as my competing models and the variables I use. Would be nice to know how well my "best" model actually predicts variability in density, as realized on the landscape. Without collecting independent data. Perhaps these deviance based approaches may help.

CL
ctlamb
 
Posts: 56
Joined: Mon Nov 04, 2013 9:44 pm


Return to analysis help

Who is online

Users browsing this forum: No registered users and 10 guests