unpacking covariate significance/contribution to density

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

unpacking covariate significance/contribution to density

Postby cdevens226 » Wed Jul 01, 2015 9:22 am

Dear Murray,
I just posted a question to you and the phidot community about incorporating sex successfully, but I wanted to start another thread to ask you an additional question about whether there is a way to unpack how each covariate within the script contributes to the ultimate density result.

Don't want to sound like a broken record, but to add context to THIS thread... My script is currently using 21 anthropogenic an environmental covariates including human footprint, anthropogenic biomes, veg coverage, land cover, human population density, bioclims, etc. in an effort to create a comprehensive leopard density that illustrates how leopard density is affected by various land use, land cover and edge effects across a highly modified landscape in the Cape.

Unfortunately I haven't gotten a D lately as a result of the error:
Code: Select all
Error in secr.fit(PP_ALLregionsCH_allCov_sex, model = g0 ~ 1, trace = FALSE,  :
  'verify' found errors in 'capthist' argument

Sometimes I get this error for script that had been previously running perfectly fine with no issue!?

But for WHEN I do get the script to run and generate a D, are there script tricks that will allow me to look at each covariate independently within the analysis so that I may be able to see which covariates are the most significant, as well as the weight/influence of each within the generated D? Its one thing to get a D=__ result, but I want to understand this number and breakdown (if possible) how the chosen model and covariates affect leopard density overall for a density. I'm not sure what the R options for this are since I'm still new to all this. Hopefully this is possible with some crafty pro R skills that you can share!?

Appreciate your help with this (and any thoughts you have on what is going on to for me to keep getting the error in capthist argument?)

All the best,
Carolyn
cdevens226
 
Posts: 10
Joined: Mon Mar 03, 2014 1:10 pm
Location: Pretoria, South Africa

Re: unpacking covariate significance/contribution to density

Postby murray.efford » Fri Jul 10, 2015 3:03 pm

are there script tricks that will allow me to look at each covariate independently within the analysis so that I may be able to see which covariates are the most significant, as well as the weight/influence of each within the generated D?


(I think this is the question I ignored before...) I'm afraid the simple answer is No. As a non-statistician I am reluctant to offer advice on what is really a classic statistical problem : sifting a bunch of probably correlated predictors to find those with the strongest relationship to density. Comparing the AIC weights of alternative models in a model set does take you some way there.

However, I doubt you will get satisfaction from an SECR analysis with so many predictors for the simple reason that there is not that much information in the data regarding spatial variation in density, even if your sampling design was rigorous and at a suitable scale (often this is not the case). What we see in the capture-recapture data is a distinctly blurred rendition of a spatial point pattern, and the ability to resolve habitat relationships is limited by the number of detected individuals (probably a few dozens or hundred in your case) not the amount of effort you put in or number of photographs you have. This is my intuitive rationale for why I have yet to see strong predictions of spatial density gradients from SECR involving more than one or two habitat variables. That is an argument for applying biological judgement to reduce the number of predictors before the SECR model fit, rather than after.

Sorry that this is just uncitable opinion and I can't be more helpful.
Murray
murray.efford
 
Posts: 686
Joined: Mon Sep 29, 2008 7:11 pm
Location: Dunedin, New Zealand

Re: unpacking covariate significance/contribution to density

Postby cdevens226 » Mon Oct 05, 2015 10:22 am

Thanks Murray for your reply. I guess in an ideal world these programs would be able to do it all!!
Appreciate your thoughts.. lots of food for thought!

best,
Carolyn
cdevens226
 
Posts: 10
Joined: Mon Mar 03, 2014 1:10 pm
Location: Pretoria, South Africa


Return to analysis help

Who is online

Users browsing this forum: No registered users and 17 guests