www.phidot.org

I am currently trying to run a single-season with false positives model on interview data from a jaguar corridor in Nicaragua. I have triple-checked the data and there are no errors in data entry. When I try to run a null model (or any model, for that matter), I get an error message that says "Presence.exe has stopped working" and then I am forced to close the program. I have no issues running regular single-season models. I am using Presence 5.7.

Perhaps this is an issue of how I coded the data. For the interview data (in response to "Have you seen a a jaguar in the last year"), we code the response as "1" (uncertain detection) if there is only one positive sighting in the sampling unit, but we code the response as a "2" (certain detection)if there are 2+ positive detections. We decided on this protocol because the detection can be considered more reliable if there is an additional person to confirm the sighting. Could this method be causing the model to crash? I would really appreciate any assistance on this.

Hi,

I don't think the way you chose the codes should cause PRESENCE to crash. Would you mind sending me a copy of the input file (whatever.pao)?

Jim (jhines@usgs.gov)

I've corrected a bug in PRESENCE which was causing problems with the FalsePositive models. Please download the new version and try the model again.

Jim

Hi again,

Thanks for fixing the bug, my models are now working well!

However, I am currently modeling collared peccary presence in a false positive framework, and the top model from standard single-season modeling (distance to protected area for Psi; hunter, residence, and farmer for p) appears not to converge in a false positive model (standard error is -1.#IND00 for DistPA coefficient, and many estimated psi have a confidence interval of 1.#QNB - 1.#QNB).

Not sure what to do when the -1.#IND00 error comes up. Should I toss out the model?

Best
Lisanne

Hi Lisanne,

Have you tried simpler models to see if the same thing happens? I would try running models with only one covariate at a time first, then try one with two covariates, etc. For example, I would start with:

Psi(DistPa),p(.),p10(.),b(.)

to see if you can get estimates of occupancy when it depends on the distance covariate. Then try each of the detection covariates with constant psi (Psi(.),p(hunter),p10(.),b(.)).

Hi there!

Yes, I have run simpler models and they worked fine. A model with Agriculture for psi and Hunter/Residence/Farmer for p ran smoothly. Adding an additional covariate for psi (SecondaryForest) also worked just fine. The only problem arose when I added DistPA (distance to protected area) as a site covariate, in either a simple or more complex model. Normally I would throw away a model with such wonky error output, but I am hesitant to do so here because this was the top model in regular single-season modeling.

One other thing that might cause problems is if the covariate has very large values and some small ones. That problem can be addressed by 'normalizing' the covariate.

Jim

Good call. Data values ranged from 0.4 to 40. Once I fixed that data range issue the model ran OK.

Funny, though, that the data range was ok for the standard single-season model but not for the false positive model.

Thanks for the help!