BPA Ch.13: Incorrect `occ_fs` about example-models

[bob-carpenter]

@ito4303 Would you mind looking at this?

@mikemeredith Thanks for the carefully constructed issue. Is this a problem with the BUGS code in the original book or just an issue with the translation?

I have a complete replication of Dorazio and Royle's occupancy model here:

http://mc-stan.org/documentation/case-studies/dorazio-royle-occupancy.html

I hope that's a faithful translation of their cited paper, but if not, please let me know, as I understand that model better than the BPA models. In the Dorazio and Royle approach, the result gets bounded between observed and max (pseudo-sites) because you start with the occupied ones (they're observed) then add in expected occupance of other ones, which has to be between 0 and 1. So we don't need to declare constraints to enforce the constraint.

[mikemeredith]

Exceeding the constraints is a symptom, not the problem; enforcing constraints isn't the solution but is a check. The problem is that, in the call to binomial_rng, N is too big and theta is too small. That will often give a plausible result, but is still wrong.

In the BUGS code, they use a latent binary variable, z[i], for occupancy for each site, and then occ_fs = sum(z). They don't need to do the kind of posterior predictive thing. So it's an issue with the translation.

If I were really using Stan (instead of translating), I'd get the conditional (on the data) probability of occupancy for each site and sum those to get an expectation for occ_fs.

I'll look at the Dorazio et al stuff later, but what you describe is I think the right way to do it and what's needed for the BPA Ch.13 examples and the closed-capture examples in Ch.6.

[ito4303]

@mikemeredith I greatly thank you for pointing out the problems. As you pointed out, they are incorrect translations.

I will contribute corrected models as soon as possible. Or, perhaps you can create pull requests with the corrected code.

[mikemeredith]

@ito4303 I'll do pull requests for the models I've looked at so far. Give me a day or 2.

[bob-carpenter]

Great! Thanks. - Bob

…

On Mar 17, 2017, at 10:50 PM, mikemeredith ***@***.***> wrote: @ito4303 I'll do pull requests for the models I've looked at so far. Give me a day or 2. — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.

[mikemeredith]

@bob-carpenter There are issues with the write-up of the Dorazio & Royle analysis, but I think the math is all ok. Where would be a good forum to discuss this?

[bob-carpenter]

An issue in this repo---the source is in `example-models/knitr/dorazio-royle-occupancy/`

[mikemeredith]

@bob-carpenter I've done a pull request.