Collinear covariates question

[Josephine Smit]

Hi, I have a question about collinear covariates.

I have two collinear covariates, Habitat and Cover, and I'm interested in exploring their effect on occupancy and detection.

Would it be wrong to include Habitat and Cover in the same model if one was specified as a site use covariate, and the other as a detection covariate?

Thanks in advance

[Ken Kellner]

It's not uncommon to see an identical covariate included in both occupancy and detection models, so I don't see a major issue in having collinear covariates as long as they're not both on occupany or both on detection.

Ken

[Marc Kery]

Dear all,

I agree with Ken, of course. One of the big advantages of these "explicit" hierarchical models over, say, a simple GLMM (Barker et al. 2016 etc) is exactly that we can usually put the same covariate into the state part of the model and into the detection part. In an implicit hierarchical model (such as a GLMM), this is not possible.

I showed this in my Auk paper in 2008 for an Nmix model and we show this also in chapters 6 and 10 of the AHM1 book for the Nmix and the occupancy model.

Best regards  --- Marc

---

From: unma...@googlegroups.com <unma...@googlegroups.com> on behalf of Ken Kellner <con...@kenkellner.com>
Sent: Tuesday, October 18, 2022 17:49
To: unmarked <unma...@googlegroups.com>
Subject: [unmarked] Re: Collinear covariates question

 

--
*** Three hierarchical modeling email lists ***
(1) unmarked (this list): for questions specific to the R package unmarked
(2) SCR: for design and Bayesian or non-bayesian analysis of spatial capture-recapture
(3) HMecology: for everything else, especially material covered in the books by Royle & Dorazio (2008), Kéry & Schaub (2012), Kéry & Royle (2016, 2021) and Schaub & Kéry (2022)
---
You received this message because you are subscribed to the Google Groups "unmarked" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/21e3fabe-8f76-4a46-b545-f9a6e841d94bn%40googlegroups.com.

[Jim Baldwin]

I have maybe a minority viewpoint on including (I suppose you mean "highly" rather than "perfectly") collinear covariates in models (whether they be in the occupancy or detection part or both):

1. If you're only interested in prediction and the software doesn't run into numerical convergence issues, use all the variables you have available or are cheap to measure or you simply have access to (but with a rational model building process - i.e., starting with simpler or theoretically reasonable models).

2.  Throwing out highly collinear covariates doesn't fix a collinear problem:  doing so just ignores the problem.  Interpretation of the coefficients of those variables included in the model would still be suspect if the excluded covariates aren't explained away with explicit justification.  (Throwing out a more expensive covariate that is replaced by a relatively cheaper-to-measure or more-readily-available covariate where both covariates are highly correlated would be one kind of justification.)

Jim

--