wrong covariate effect

[Michele Chiacchio]

Dear all,

I am currently developing a multi-season occupancy model using unmarked.

While I have no problems with p and psi, with the extinction/colonization there is something odd. Specifically, in the top-selected model there is a positive effect of a parameter (namely: precipitation) on extinction (meaning, more rains, more empty plots). However, just by looking at the raw data, it is clear that the year with the lowest occupancy estimate is also the driest one.

How could that be possible?

I obtained the top-selected model by comparing parameter combination using dredge (I kept all covariates constant while comparing models for varying detectability, that used the best one for psi and so on).

I have already checked for correlations between covariates and performed a goodness of fit and the best selected one seems reasonably good.

best

[Marc Kery]

Dear Michele,

I don't think this is necessarily a contradiction (and for the most part I would trust what unmarked says 🙂). The number of occupied sites in year t depends on:

• the number of occupied sirtes and the number of unoccupied sites in year t-1,
• the extinction probability between t-1 and t, and
• the colonization probabilty between t-1 and t.
  

So, you can't expect just one of these four to be necessarily strongly correlated with the total number of occupied sites in a given year.

If you want something in the observed data to compare with the estimates for colonization and extinction in the dynocc model then you should do logistic regressions of "apparent extinction" and "apparent colonization". That is, you do something like the following:

• reduce your full occupancy data by collapsing over the secondary occasions and only retaining the observed presence/absence status for each site and year (i.e., a 1 if there is at least one detection and a 0 otherwise)
  
• and then, for every year t=1 to T-1 (where T is the total number of years) you compute the proportion of sites apparently occupied in year t that are no longer so in t+1: this is the apparent (or detection-naive) extinction rate
• Likewise, the proportion of sites that are apparently empty in t but which are apparently occupied is a detection-naive estimator of colonization probability
• Then, you can plot these simplistic estimators of eps and gamma against covariates of choice

One of the big advantages of using an occupancy model as implemented in unmarked, MARK or PRESENCE (quite apart from being able to estimate detection) is that all these data manipulations and sepaate logistic regressions can be done in one go.

Best regards  -- Marc

---

From: unma...@googlegroups.com <unma...@googlegroups.com> on behalf of Michele Chiacchio <chiacchio...@gmail.com>
Sent: Friday, July 8, 2022 16:16
To: unmarked <unma...@googlegroups.com>
Subject: [unmarked] wrong covariate effect

 

--
*** 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/1ec4cdbd-4977-49a0-85b1-3ede00e590f4n%40googlegroups.com.