www.phidot.org

Posted: **Mon Apr 04, 2022 6:26 am**

I conducted an experiment to compare the detection probability between different observers. I did the same experiment in three different sites but some of the observers participated in only one/two sites while the others participated in all three sites. I analyzed my data, separately for each site, using closed population capture-recapture models.

From the models, I receive estimates of observer-specific detection probabilities. Now, my question is: How can I average the estimates for an observer, in particular the confidence intervals?

I am not sure if my question is almost trivial or rather complex. In any case, I couldn't find an entry in the MARK book or in the forum which deals with the same problem.

Thank you very much for any suggestions/solutions!

Posted: **Fri Apr 08, 2022 10:50 am**

I conducted an experiment to compare the detection probability between different observers. I did the same experiment in three different sites but some of the observers participated in only one/two sites while the others participated in all three sites. I analyzed my data, separately for each site, using closed population capture-recapture models.

From the models, I receive estimates of observer-specific detection probabilities. Now, my question is: How can I average the estimates for an observer, in particular the confidence intervals?

I am not sure if my question is almost trivial or rather complex. In any case, I couldn't find an entry in the MARK book or in the forum which deals with the same problem.

Thank you very much for any suggestions/solutions!

Hi,
Is there a particular reason that you analyzed the three sites separately, rather than together treating site as a factor with three levels? Running all sites in the same analysis would give you greater modeling flexibility and more statistical power. This would also likely make model averaging of your observer-specific detection probabilities more straight forward as well, although it's difficult to know for sure as you haven’t detailed how you modeled observer effects.

Cheers,

Eric

Posted: **Tue Apr 26, 2022 10:05 am**

Hi Eric,

thank you very much for your thoughts and sorry for my late reply.

My dataset is a bit special. I worked on three populations of hibernating butterfly larvae without losses/recruits during the time of the study, i.e. closed populations.
In my particular case, the time-varying capture probabilities represent the observer-specific detection rates because one observer conducted one round of survey in a population.

Some observers did surveys in all three sites while others surveyed only one site. Therefore, stitching the three datasets together would introduce some 'dots' (missing data) in the capture histories. However, the 'dot' notation is not supported by closed population models in MARK/RMark.

Posted: **Tue Apr 26, 2022 10:11 am**

butterphi wrote:However, the 'dot' notation is not supported by closed population models in MARK/RMark.

Actually, it is -- for the Huggins data type (which is typically what you should be using anyway...you can't have 'dots' for the full likelihood data type and derivatives, but otherwise...)

Code: Select all: Available data types in Program MARK (# indicates allowed with median chat, $ indicates simulation possible, _ indicates dots are allowed in encounter histories, and * indicates also available with RMark): 12. #$_*Huggins: Huggins' p and c 25. #$_*HugHet: Huggins' Heterogeneity pi and p 26. #$_*HugFullHet: Huggins' Heterogeneity pi, p, and c

Posted: **Mon May 02, 2022 4:27 pm**

That's great news.

Thanks a lot!
I used the full likelihood data type so far. I will switch to Huggins, now.

www.phidot.org

Averaging parameter estimates from independent models

Averaging parameter estimates from independent models

Re: Averaging parameter estimates from independent models

Re: Averaging parameter estimates from independent models

Re: Averaging parameter estimates from independent models

Re: Averaging parameter estimates from independent models