Occupancy removal design and time varying covariates

[patk]

Hi: I understand that for occupancy data collected from a removal design, models with fully survey-specific detection probabilities cannot be fit. But I am wondering if this also applies to covariates that vary in time. Can t be added as structure to detection if it is associated with a covariate?

Thanks
patk

[bacollier]

patk,

I think the answer is yes. I assume you are modeling the detection process as a function of some value relative to when the survey occurred, perhaps time since arrival for a breeding bird, or time since precipitation, or some other time varying covariate. If so, then using that covariate in your detection model, relative to each survey occasion is fine even under a removal model.

Bret

[patk]

Hi Bret: Actually the covariate is observer and I want to examine survey-specific observer effects on detection in a removal design. Do you think that is still kosher?
Thanks
pat

[bacollier]

Pat,
No, its not (unless I missed something our there in the literature). The issue is that you will not have any repeated positive observations by any observer, when they see it once, they are done collecting P/A data, so you get encounter histories that look like (1...) or (01..) or (0001), so estimating survey-specific detection probability is out as there are no histories that go 1010, or 1101 on which to base the survey-specific, observer specific detection estimates (this is detailed in MacKenzie et al. 2006 on 102-103, 175-178).

You might (untested) be able to set up a constraint, maybe setting constant and estimate a detection value at the observer level using observer treating observer as a individual covariate, but it would not be survey-specific.

Bret