Using raw abundance as a detection covariate

[bmitchel]

Hello PhiDot,

I've been working with some colleagues on a multi-season occupancy analysis, and we're trying to think through some options related to modeling detection probability. We are aware that abundance can have a large effect on whether a species is detected (few individuals --> low p; lots of individuals --> high p). Is there any reason why the raw abundance at a point count should not be used as an index of abundance and a covariate for detection probability? What about metrics derived from raw abundance, like a binary variable representing "few" or "many" detection, or the average abundance across a season at a site? I think that there might be some confounding with other detection covariates (to the extent that they also correlate with abundance), but I'm not sure if this is "fatal" or not.

Thanks!

Brian Mitchell

[darryl]

Hi Brian,
You could do that, but essentially you're double dipping and using the same bit a information as both a response and predictor variable, particularly when you're going to use the observed count as a survey specific covariate. Also think about the biological interpretation of such a model... Personally I'd advise against such an approach.

If you're happy to define categories (eg, few vs lots of individuals) you could use these categories with the multi-state occupancy models. It would be a much more explicit way of using that data and also accounts for the fact that you may only ever observe few individuals at a place even though it actually had lots of individuals.

Cheers
Darryl