Junk CIs with Robust Design, GIGO? or something else?

questions concerning analysis/theory using program MARK

Junk CIs with Robust Design, GIGO? or something else?

Postby Kyle » Sat Jan 27, 2018 7:57 pm

Hello everyone,

I am working on a conservation project with an endangered beetle and am trying to find a more accurate and reliable method for population modeling instead of just a Schnabel index.

The beetle is a strong flyer (regularly moving move then 1km in a night) so all of the local populations are open. They also go under ground multiple times a season to reproduce, making a model that can include the temporal aspect essential. The Robust Design model in MARK seemed to be perfect since it fits the biology of the beetle so well. We sampled 3 nights in a row then took 4 nights off, rinse and repeat though the summers.

However, I am getting bad AICs and total junk confidence intervals (if n is estimated at 7, the CI will be 7, if n is 21, the CI will be 21). I have played around with the models to the best of my ability (I have read the MARK book and have taken some ecology and population modeling classes...but I am an entomologist, not an ecologist or statistician).

My data is extremely 0 heavy since recapture rates are so low (~10% of beetles were recaptured) and we only caught individual beetles more then twice a few times. Is this a simple case of GIGO?

I have been reading through any forum posts that seem similar to my problem or have worked with Robust design over the last couple days with no success. You all seemed really helpful so I wanted to take a chance and ask before I tried to find a different option for my analysis.

Thanks for your time,
Kyle
Kyle
 
Posts: 3
Joined: Sat Jan 27, 2018 1:49 pm

Re: Junk CIs with Robust Design, GIGO? or something else?

Postby DuaneD » Fri Feb 23, 2018 4:50 pm

Kyle,

If you are getting population estimates with no confidence interval then I suspect you have the closed capture model portion of the robust design modeled incorrectly. For example, if you are using a time-dependent model and do not set the last p and c equal to one another (those parameters are confounded) you will get an estimate of N with no CI.

Given the data you describe seems to be sparse, I would recommend starting with a closed capture model with a constant capture probability (p) and set the recapture probability (c) equal to p.

I think the "bad" AICc values may be negative? If so, that is because the combinatorial portion of the likelihood is ignored (because it never changes) to make make the computation of maximizing the likelihood easier. Nothing to worry about and delta_AICc is just fine.

Good luck!

Duane
DuaneD
 
Posts: 4
Joined: Wed Feb 05, 2014 9:58 pm

Re: Junk CIs with Robust Design, GIGO? or something else?

Postby Kyle » Sun Feb 25, 2018 9:11 pm

Thanks you so much for the reply Duane! I am going to start working on this now.

I appreciate your help!
Kyle
 
Posts: 3
Joined: Sat Jan 27, 2018 1:49 pm

Re: Junk CIs with Robust Design, GIGO? or something else?

Postby Kyle » Thu Mar 01, 2018 2:01 pm

So I have been working on making p=c.

My plan was to do this through the DM, seemed like the easiest method. Now that I'm looking at my DM, I can't seem to figure out how to do this. It looks like they may already be equal? Here is a shot of the meaty portion of my DM.

https://i.imgur.com/dSBLwA0.jpg

In the book is says to set y' = y" I need to eliminate the difference between groups. Not only is my DM much simpler than the example in the book (maybe because I only have 1 group?) but it refers to a column in the DM to delete that isn't even showing up in mine.

Any ideas?

Thanks for your time,
Kyle
Kyle
 
Posts: 3
Joined: Sat Jan 27, 2018 1:49 pm


Return to analysis help

Who is online

Users browsing this forum: No registered users and 1 guest

cron