Life History to short for the number of occ specified

[CasTERN]

Hello,
For my Master Thesis about colour-ringed Caspian Terns I want to measure their stopover persistence via Mark on one particular site and later estimate the Flyway-Population. I have done daily counts (n=77) and read a total of 139 individuals with 419 resightings. I take it as an open population. So there are plenty off zeros in my data.
After reading and working through Mark´s handbook chapter 1-5 and read chapter 2-3 multiple times and also looking for posts in the forum, I did not find an answer.

I am aiming for the results like in Loonstra. et al. 2016:Staging Duration and Passage Population Size of Sanderlings in the Western Dutch Wadden Sea

I have coded the data via Notepad ++, like the handbook suggested.

Example of my data:(excluded the "." for the 2 days I could not go out because of the weather)

00100000000000000000000000000000000000000000000000000000000000000000000000 1;
00000001000000000000000000000000000000000000000000000000000000000000000000 1;
00000000000000011100011111101111000000000000000000000000000000000000000000 1;
00000000000000011100011111101111000000000000000000000000000000000000000000 1;

So now my questions are:
When I add 75 sightings, the system tells me later, when I want to RUN the current model that my encounter history too short for the number of occasions. How do I get rid of that problem?
Is it because of the bunch of zeros, so that my result is meaningless?
Shall I shorten the encounter history to two-day counts?

I appreciate any help and will provide my data if someone wants to have look on it.
Thanks, Dennis Heynckes

[cooch]

I have done daily counts (n=77)

Example of my data:(excluded the "." for the 2 days I could not go out because of the weather)

00100000000000000000000000000000000000000000000000000000000000000000000000 1;
00000001000000000000000000000000000000000000000000000000000000000000000000 1;
00000000000000011100011111101111000000000000000000000000000000000000000000 1;
00000000000000011100011111101111000000000000000000000000000000000000000000 1;

So now my questions are:
When I add 75 sightings, the system tells me later, when I want to RUN the current model that my encounter history too short for the number of occasions.

The example EH (above) are 75 time steps long. But, you start by saying you have n=77 daily counts. If you entered 77 for nmber of occasions, you would get the error you report. You would also get an error if you had dots for missing occasions, and entered only the number of occasions for which you had 0/1.

[CasTERN]

The example EH (above) are 75 time steps long. But, you start by saying you have n=77 daily counts. If you entered 77 for nmber of occasions, you would get the error you report. You would also get an error if you had dots for missing occasions, and entered only the number of occasions for which you had 0/1.

Good evening Mr cooch
I have tried it with 75 as number of occassions and without points but the problem is still existing with "encounter history too short for number of occassions".

Here is an example of the error information page which pops up if I try to let it run:

This problem will use 8 of 8 possible threads.
Static thread allocation is used.
Schedule kind is omp_sched_dynamic with chunk size 1


INPUT --- proc title rss;

For procedure title CPU Time was 0.001 seconds and System Time was 0.001 seconds.


INPUT --- proc chmatrix occasions=75 groups=1 etype=Live mixtures=2
INPUT --- Nodes=101 NoHist hist=300;

INPUT --- glabel(1)=Group 1;

INPUT --- time interval 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
INPUT --- 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
INPUT --- 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1;
Error around encounter history number 1

ERROR -- Encounter history was too short for the number of occasions specified.

Can you recommend a paper or another example which I should try out, because I have run the excersises multiple times and they are working. The only difference between them and my data is the length and the number of 0.

Thanks for your fast reply, Dennis Heynckes

[cooch]

First, what data type (model) are you trying to fit? With that many zeros (as you suggest), fitting a 'mixture' model (which it looks like you're trying?) won't likely work - at all. Mixture models tend to be pretty data-hungry.

Second, dots count as occasions. So, for example, 10.1101 would be 7 occasions. So, if you try setting number of occasions to 77, what happens?

Failing that, send me your .inp and .dbf files offline, and I'll have a look.

[CasTERN]

Good evening again Mr cooch

what data type (model) are you trying to fit? With that many zeros (as you suggest), fitting a 'mixture' model (which it looks like you're trying?) won't likely work - at all. Mixture models tend to be pretty data-hungry.

Yes, I have tried it with mixture models. But I have to admit that I am certainly sure about which model to use. I have thought about testing "Pradel and Link-Barker Models" because I have an open population and a high dynamic species plus I have alo counted all individuals (ringed and non ringed and discriminated them by age); so if the "Pradel and Link-Barker Model assumes an open population plus a number of observed animals, then it could help me to estimate "phi" for the occasions, or am I wrong?

Second: I have tried the point version with n=77 beforehand (because the handbook told me about the point as occassions without countings, but to test for all mistakes there could occur, I have deleted them after some mistrying).

I am sending you the .inp and .db for the data "with" and "without".
Thank you really much for your time and help.
Greetings Dennis Heynckes

[cooch]

Well, one thing I skipped over, but you need to be aware of -- the 'dot' convention isn't suported for any of the 'reverse time' (Pradel) or the Link-Barker JS model. For reasons that are explained in the book in various places -- largely because unequal intervals/missing occasions makes interpretation of other parameters ambiguous at best). If you pull dwn the 'Help menu', at the bottom is a link to 'Data types' You'll see which data types do (and do not) support 'dots'.

So, you're only option if your interest is 'reverse-time'-type modeling is to drop the 'dots'.

Working with that file (being, the one without dots -- which still leaves you the problem of manually specifying the intervals for the 'missing occasions' -- discussed at length in Chapter 4),

1\ the number of occasions is 74, not 75. MARK assumes you can count. For example, here is the first line of the .inp file. 74 charaters in the EH string, a space, and then "1;". So, by you entering 75, MARK reported an error, because the EH is actually shorter than that.

Code: Select all

10000000000000000000000000000000000000000000000000000000000000000000000000 1; 


2\ there was an error in the file -- a missing 0 or 1 in line 102. I filled it in with a zero.

3\ your data are sufficiently sparse (in simple terms, not enough encounters, too many zeros) that it might be difficult for you to fit much of anything. Mixtures are definitely out. Even simple CJS models ignoring any other structure (like age) have trouble being fit to these data. For anything like a Link-Barker or Pradel model, I suspect you're out of luck. I couldn't even get numerical convergence without applying Draconian constraints to at least one of the parameters. If you have a strong, and informative set of environmental covariates (shorebirds = wind, tide, temperature...), you might be able to get something reasonable, by applying those covariates as linear constraints to the parameters, but....there is that pesky GOF test which needs a model closer to fully time-dependent.

[CasTERN]

Thank you Mr Cooch for your help!

I will try to work with it for now (Trying to imply the tide level, that I have also measured) But yes the GOF will become presumably harsh.

What if I lower the abundance of "0" via one count for two days or even presence for a week?
Studies about GPS-loggered Caspian Terns suggest a 40-60 stopover duration in the region. So if I drop the daily counts and exchange too weekly counts the number of occassions goes down to 11. Or if I take two days in concern its 37 (if n=74).

Then my data would concern far less "0", I could show the stopover persistence per week. Of course I have to argue with that data compression, but due to the reason for getting something reasonable and the studies with the proof of like 40-60 days staying caspian Terns I think it is a possibility to work with the data even that this was not the first idea I had in mind.

I would be very happy to hear your thouhts about it. And I thank you a lot for your time, help and fast reply!

...And I have also assumed that I could count... well.

[cooch]

In other words, what happens if you pool data into longer periods. There are pros and cons, but worth a try.

Good luck!

[CasTERN]

Dear Mr Cooch
I am in the end of my Thesis`statistics. The CJS model does not work as I thought for the estimation of the flyway-population, because it calculates negative estimes due to my not so high number of birds and the ssumption of a closed population, while its certainly open. It worked for Loonstra et al tho but they had mor sanderlings than I have Caspian Terns.

Now I have looked up different possibilities and the best one was a JS-Model (where you told me that I am out of luck ) used to examine the entering probability for new birds in the population to get an estimate for a superpopulation (Lyons J.E et al. 2016: Population Size and Stopover Duration Estimation Using
Mark–Resight Data and Bayesian Analysis of a Superpopulation Model).
As I tried it via Mark JS-Model with Phi(t)p(t)andBeta(t) the program said as you mentioned earlier "that its hard to get even numerical convergence".
If I assume that the individuals were there between the first and last time (i just not have seen them), I minimize the "0" in my study.
My question is, if you have another suggestion or have heard about another paper to estimate stopover population size?

Greetings Dennis Heynckes

[cooch]

I'm not comfortable answering 'how to finish my thesis' sorts of questions. These are questions for your advisor. There are lots of papers on stopover estimation.