7th Feb, 2015

Biometric Research

Question

Asked 6th Feb, 2015

I have a dataset with 48 sites, 6 site co-variates, 5 sample covariates modelled for 5 occasions. I think I can fit up to at least 20 parameters. Whenever I run a base model psi(.)p(.) I get a estimates with SE. However, when I increase parameters estimates go haywire (infinity etc). Even running a correlated detection base model I get the same, does anyone have a fix for this in PRESENCE or is it that the data is bad?

A couple thoughts....

The number of parameters you can reliably estimate will also depend on the true occupancy and detection probability. How many sites do you actually detect animals? At those sites, how many of the replicates are they detected in? If the answer is few, you could still have a low sample size issue.

I would build on the base model gradually by adding one parameter at a time and running the model to see when the fitting starts to fail.

I would also run a model fit test on the models that do run to see if you are meeting the basic assumptions of the model. Poor model fit could be causing your issue. If there is poor model fit you might consider a model that models the detection intercept as a random effect across sites. Or you might consider a model that evaluates non-independence among individuals or detection occasions.

Good luck,

Dan

3 Recommendations

**Get help with your research**

Join ResearchGate to ask questions, get input, and advance your work.

A couple thoughts....

The number of parameters you can reliably estimate will also depend on the true occupancy and detection probability. How many sites do you actually detect animals? At those sites, how many of the replicates are they detected in? If the answer is few, you could still have a low sample size issue.

I would build on the base model gradually by adding one parameter at a time and running the model to see when the fitting starts to fail.

I would also run a model fit test on the models that do run to see if you are meeting the basic assumptions of the model. Poor model fit could be causing your issue. If there is poor model fit you might consider a model that models the detection intercept as a random effect across sites. Or you might consider a model that evaluates non-independence among individuals or detection occasions.

Good luck,

Dan

3 Recommendations

Thankyou Dan!

I agree, I think model fit is the first way to go one by one parameter, then probably stop at the time it begins to fall. For the dataset I have non-independent detection and also false positive detection show good fit. Makes sense ecologically.

Hi Salvador,

Not sure if you have had success in solving your issue or not. If not there are likely still some things you can do. The first question from me would be similar to Dan's. I would be curious to know how those captures and sampling occassions broke down. Often my surveys include more than 50 sampling occassions for poorly detected species. This requires me to have to collapse my capture histories to diminish the number of 0's and this improves model convergence. With only 5 sampling occassions (assuming you have not collapsed down to 5 days) I wouldn't think this would solve the problem. I fully agree that starting with modeling detection one parameter at a time is your best best. I think any expectations of having more than 5-7 parameters in your models is going to be a real long shot.

If you still need/want someone to look over the data feel free to let me know.

Best of luck

1 Recommendation

Article

- Mar 2015

1.Occupancy models are employed in species distribution modeling to account for imperfect detection during field surveys. While this approach is popular in the literature, problems can occur when estimating the model parameters. In particular, the maximum likelihood estimates can exhibit bias and large variance for datasets with small sample sizes,...

Conference Paper

Full-text available

- Jan 1997

In this work I present a Prolog program MARK, I developed, for parameter estimation in the simulation of ecological systems, capable of qualitatively explaining the simulated system. To generate qualitative explanation, the program combines a numerical and a qualitative simulator. The program was tested on a number of simple ecological models. The...

Get high-quality answers from experts.