Robust design in MARK -> how?

[Todd]

I am comfortable with the terminology of most of mark-recapture modeling but I am new to using program MARK or any of the capture analysis programs for that matter.

I have read all of the recent Robust design papers by Larissa Bailey on salamanders and some of Pollock's work on the topic as well, including his introduction.

However, when it comes to implementing the analysis on program MARK, I'm at a loss. Here is the scenario.

I collected data on a highly isolated population of animals with presumably no migration in/out of the population. I had 56 total trapping days collected in a robust design pattern on monthly intervals.

Thus, I had 6 primary periods separated by month-long intervals. The first 5 primary periods each had 10 consecutive days of trapping (secondary periods) and the last primary period had only 6 days of trapping.

I am most interested in understanding/estimating:
a) whether animals exhibit a trap response after first capture
b) whether some animals are more likely to be captured than others and what can account for this (age, sex)
c) total population size, and
d) whether the animals can be temporarily unavailable for capture (gamma' and gamma'' have non-zero values).

My data is formatted as a series of individual capture histories representing all 56 capture days for each animal and an identifier as to whether or not the animal belongs to one of three cohorts: Male, Female, Juvenile.

I have reason to believe that the probability of capturing an animal during consecutive days of trapping varies depending on the weather/climate of that particular day. It is also a safe assumption that survival over the five intervals is constant. Population size however may increase as the year trapping progresses due to reproduction and birth of offspring.

Where do I begin? No amount of literature searching or manual reading has proven productive so far.

Every model I try gives me a population estimate that is hundreds of animals less than the number of novel captures I made in the population and I am convinced I am failing miserably at something.

[cooch]

I am comfortable with the terminology of most of mark-recapture modeling but I am new to using program MARK or any of the capture analysis programs for that matter.

Start with

http://www.warnercnr.colostate.edu/~gwh ... bustDn.PDF

A little outdated, but it covers the basics. A full-blown RD chapter is under development.

[Todd]

Thanks for the link. This too is a paper that I have read more than once. The trouble is that I am not any closer to solving my problems after reading that paper than I was before reading it. When I fire up Program MARK and make a few click choices, I am still left with only a vague sense of what I am controlling.

It seems to me that when MARK is left on its own to generate parameters it picks a group of them that best fits the data but grossly misconstrues the population size among other things.

[Bill Kendall]

Unfortunately there is not yet a chapter in Evan's book to help you with the robust design in MARK. That is my fault. In the absence of that, I would start with something close. The Bailey et al. paper does not refer to the 'closed population' robust design that you seem to have. There is a paper on the MARK website by myself that discusses using MARK for the robust design. It discusses the estimation of the gamma's which the early papers by Pollock do not. It also refers to the first papers that include temporary emigration. There is also a paper by myself in Ecology in 1999 that deals with the issue of the possibility of closure violations.

Since Evan's book is available for closed population models and the CJS model, I will assume that you are somewhat familiar with MARK in general. The robust design model you probably need is basically a series of closed models nested within an open model. With that in mind, I'll offer some comments on your questions.

a) There are trap response parameters (c) available in each primary period. You can basically model them in the same way you would do so in a closed model. If you feel there might be permanent trap response (that persist into the next primary period), that would require something trickier, which has its own disadvantages.
(b) Modeling capture probability as a function of predictors like age or sex is discussed in other parts of Evan's book. It can be done using groups and the PIM's.
(c) If by total population size you mean the number of animals in the study area at a given time, you can get that out of the regular robust design model or the Huggins robust design model (for the latter you have to click on derived parameters, invoked through the output menu). I believe most folks have had more luck with the Huggins model than the regular robust design, which seems to be more stable especially with smaller sample sizes.
(d) Hopefully by reading my paper on the MARK website, you'll have a better idea about how to deal with the issue of estimating the gamma's.

As a final comment, it sounds like in some cases you have 10-day primary periods, separated by only 20 days. That is a pretty broad primary period. If you expect mortality over 20 days, you probably expect mortality over 10 days. If you have mortality during that time, and you have trap dependence as well, you would expect some bias in your p's and population estimates at least. Depending on your daily sample sizes, you might consider shortening those periods. That 1999 Ecology paper I mention discusses this issue.

Good luck!

[cooch]

Thanks for the link. This too is a paper that I have read more than once. The trouble is that I am not any closer to solving my problems after reading that paper than I was before reading it. When I fire up Program MARK and make a few click choices, I am still left with only a vague sense of what I am controlling.

Bill Kendall just replied at length about your options.


But, it also *sounds* as if you have some basic uncertainties about using MARK in general, not specifically the robust design. If that is correct, then you really should work through chapter 4 -> 8 first, before going much further.

It seems to me that when MARK is left on its own to generate parameters it picks a group of them that best fits the data but grossly misconstrues the population size among other things.

This is incorrect. MARK does very few things 'on its own'- much to the consternation of folks used to 'pointing-and-clicking' their way through things. If MARK seems to be giving you answers which are illogical (as noted, N < number of unique captures), this is a pretty clear indication that you are probably doing something wrong - not MARK.

[cooch]

Unfortunately there is not yet a chapter in Evan's book to help you with the robust design in MARK. That is my fault.

Nah, my fault, actually. Problem is that (1) Bill keeps pushing the envelope on what the RD can do, and (2) Gary is too fast at implementing Bill's latest ideas for me to keep up.

So, indirectly, its Bill and Gary's fault (kidding, kidding...) - but the larger blame is mine. Now that my teaching is done for the year, I'm hoping to finish up a couple of 'lagging chapters' - including the RD chapter, which I'll work on with Bill.

[Todd]

Bill I got your email and I would like to investigate some relationships of the data futher before you commit any more time to helping me. I could also stand to brush up on the aforementioned chapters of the gentle guide.

One quick question though. Is it a reasonable approach to condense the secondary trapping periods into one day and analyze the 6 primary periods as a closed model to determine the feasibility of constant survival? I'm assuming no because phi and p both factor into whether an animal was observed. Thus, collapsing the secondary periods will inflate p and phi will decrease as a result. Is this correct? Should I not do this to investigate whether survival is constant or varies with time?

Or should I just wait and rely on the robust-design analysis to figure that out.

[Todd]

I'm back and although I've read about 10 papers and reread the MARK Gentle Intro chapters, I'm still having problems.

I've shortened my dataset into 6 monthly intervals of 5 day trapping sessions.

I have no way of knowing what N, S, p, c, and gammas are a priori. I want to know how big my population is, what survivial is like, whether males and females differ in catchability, if there is temporary emigration or trap responses, etc.

I tried running program CAPTURE from within MARK and ticking the Appropriate model box to try to learn more about what scenarios are favored within each primary period and then to build RD models around that. But, it usually gets stuck in an endless loop and crashes the program. Occasionally it will give me output but it often skips the first session or two and inevitably ends up telling me that the values are too small to do a few of the tests, usually the first one.

[Todd]

Bailey et al. paper does not refer to the 'closed population' robust design that you seem to have.

I'm not sure how this is possible. All robust design options in program MARK require closed captures except for the Huggins Heterogeneity but Larissa said explicitly in her paper that "[we decided] to not use models constaining heterogeneous capture probabilities.

[Todd]

Bailey, L. L., T. R. Simons, and K. H. Pollock. 2004. Estimating detection probability parameters for plethodon salamanders using the robust capture-recapture design. Journal of Wildlife Management. 68(1):1-13.