4 CAPTURE questions - Chao Mh, p-hat, Mo and truncating data

questions concerning analysis/theory using software from the Patuxent Software Archive

4 CAPTURE questions - Chao Mh, p-hat, Mo and truncating data

Postby shannonbarbermeyer » Thu Jun 25, 2009 11:39 am

Dear CAPTURE users -

I have 4 questions. I'm looking at 4 different datasets and have run up against similar issues in them.

Sampling frame for each dataset is 15 days - but this represents trapping over blocks for 15 days in one area and then an adjacent area, etc. So the 15 days are pooled - day 1 indicates the 1st day of trapping in each area - not the same date for all - though they all occur within a theoretically closed sampling period.

First question - when running CAPTURE sometimes Mh is selected but it always says Jackknife estimator is suggested - even when Chao may be better because of low capture probability (p-hat). Is the program default set to suggest Jackknife when Mh is the model selected or is it actually considering Chao's Mh as well and indicating that Jackknife is better?

Second question - how low is low enough to use Chao's Mh over Jackknife? The CAPTURE2 manual indicates that Chao's should be used when working with low p-hat. Though this is a bit of a subjective decision with each species / situation in the Capture Recapture and Removal Methods for Sampling Closed Populations book chapter by White et al. 1982, LA-8787-NERP they report a p-hat of 0.11 “fairly low” (see chapter 4, page 144, figure 6.4c.). However, in this same manual 0.30 is ranked as "high". Anyone know a reasonable rule of thumb for middle values or is there any reason why NOT to just always use Chao's if p-hat <0.20 as a safety measure? Or should one always use Jackknife unless p-hat is <0.15 for example?

Third question - I've read and have been told Mo is not robust to violations (in the same manual above and elsewhere) especially with small samples (in some cases we have only 4 or 7 animals captured with just 18 and 25 captures, respectively) and that Mh is often used instead of Mo even when Mo is selected because of this reason. Is there general agreement on not using Mo with small samples (say under 50 animals)? Of course, with any small sample situation everything has to be interpreted with caution. Is there ever a time to use Mo over Mh? I've seen repeatedly Mo selected but Mh used unless there was really good evidence to suggest that there was indeed no heterogeneity although it seems in most animal populations that there is often heterogeneity.

Fourth question - In one of the datasets Mo was selected with Mh as a close second but the test 3 showed time specific variation was a problem. I examined the data and saw that for some reason on Day 15 far fewer animals were captured. There isn't a clear reason for this because the procedure was the same in each block on all last days (day 15 of trapping in each block) and the same as earlier days in each block and they are on different dates - so I'm not sure what is going on there. But I reran the dataset just without the last sampling session (day 15 from each block) and that took care of the time violation and then Mh came out as the top model with the Mh results between the two runs were very similar just that the run without Day 15's didn't have the time violation issue. But the Mo results compared to Mh are very different independent of whether data were truncated or not.

In another dataset I saw that there were very few captures on day 1 and day 15 - I reran the set without those days and got a much better result with respect to the closure test. Similar results in terms of population estimation.

Similarly, in another dataset I saw that there were no captures at all between days 7-10. The model selection indicated Mh but the GOF for Mh failed. I reran the dataset with just days 1-6 and got similar results as when the whole set was run - except that the GOF passed this time.

My question is - I don't want to be data manipulating without good reason. I don't know what the biological reasons behind these issues would be. I do know that with or without the datasets being manipulated the estimates are largely the same - but in the cases where I've truncated the data to shorter sampling frames the assumptions aren't violated. Is it okay to truncate data in some situations in order to meet the assumptions of the models selected? I don't want to take the model selection results blindly without making sure they fit with the tests 1-7 I'm not sure about doing it in my 3rd example above - but I feel pretty comfortable with the 1st two - especially in the 1st case because I don't feel comfortable selecting a model that doesn't work if the data are truncated by just 1 day - it seems no single day should have that much leverage if the model really is describing the data as a whole well.

Thank you very much for your help!
Shannon
shannonbarbermeyer
 
Posts: 12
Joined: Tue Aug 05, 2008 1:32 pm
Location: Ely, MN

stuck with CAPTURE on this one

Postby shannonbarbermeyer » Thu Jun 25, 2009 5:16 pm

Thank you John for the papers and suggestions. Unfortunately I'm stuck with CAPTURE on this one. I'm helping a team that doesn't have experience with MARK to interpret their CAPTURE outputs. The data aren't mine. I do enjoy the flexibility of MARK though (which I did find hard to learn at first but then rewarding indeed!). So any help on my CAPTURE specific questions would be much appreciated! Thanks, Shannon
shannonbarbermeyer
 
Posts: 12
Joined: Tue Aug 05, 2008 1:32 pm
Location: Ely, MN

Re: 4 CAPTURE questions - Chao Mh, p-hat, Mo and truncating

Postby Angelica » Fri Aug 24, 2012 7:24 pm

I would like to know the papers and suggestions of John.

Thanks!
Angelica
 
Posts: 6
Joined: Wed Jun 06, 2012 12:15 pm


Return to analysis help

Who is online

Users browsing this forum: Google [Bot] and 1 guest