incorporate age without violating assumptions?

[lcp729]

My dilemma is that I've only got two years worth of data, but I'm really interested in survivorship (and recapture probability) of hatch-year turtles to their second-year (as well as subadult and adult phi and p).

I know now that I sampled incorrectly for using many (all?) of the models available in MARK because I checked traps approximately every three days for six months out of each of the two years.

From what I've gathered, in order to really incorporate age, the intervals between encounter events are supposed to coincide with transition from one age to the next, but with only two years, I can't get the parameters I need estimated if I only use yearly encounters...not to mention lumping six months worth of recapture data doesn't really seem valid.

If I tried to use body size as an individual covariate, it would change so dramatically over the course of the study, that it wouldn't really be meaningful.

So, should I give up on trying to fit my data into a particular model in MARK or is there something you can suggest?

Thanks,
Laura

[conroy]

Laura-
Whereas sampling intervals coincident with age transition are the 'usual' case, there is nothing to prevent one from building a model in which the sampling intervals are shorter than age transition (as in your case) or longer. The real issue then becomes 'what parameters are and aren't identifiable, given the data."

Forgetting about age for a moment, with2 years of data, if you were sampling once per year, there would be a single time interval, and complete confounding between survival and recapture. With sampling once a month for 2 years there are now 23 monthly intervals. Introducing age, potentially one could estimate both monthly and age-specific effects. Finally, if you are sampling every 3 days, there is now the potential for a richer analysis based on the Robust Design.

My point is that MARK has a great deal of flexibility and that at this point you should think in terms of building models that allow you to use most of your data (and address your important questions), rather than squeezing your data into 1 or 2 standard model (and probably tossing out data/ losing information). A lot of this depends on how much data (ie. recaptures) there are, and some pooling of sampling occasions may certainly be warranted.

Mike Conroy

My dilemma is that I've only got two years worth of data, but I'm really interested in survivorship (and recapture probability) of hatch-year turtles to their second-year (as well as subadult and adult phi and p).

I know now that I sampled incorrectly for using many (all?) of the models available in MARK because I checked traps approximately every three days for six months out of each of the two years.
From what I've gathered, in order to really incorporate age, the intervals between encounter events are supposed to coincide with transition from one age to the next, but with only two years, I can't get the parameters I need estimated if I only use yearly encounters...not to mention lumping six months worth of recapture data doesn't really seem valid.

If I tried to use body size as an individual covariate, it would change so dramatically over the course of the study, that it wouldn't really be meaningful.

So, should I give up on trying to fit my data into a particular model in MARK or is there something you can suggest?

Thanks,
Laura

[lcp729]

Hi Mike,
Thanks for responding. Unfortunately, my sample sizes are pretty small in some groups, and my recapture rates are really low and hetergeneous within groups, so a lot of parameters aren't estimable.
What I attempted to do was break my turtles up into species*sex*age groups, which I hope is valid, and pooled by whether or not they were caught in a particular month to try to eliminate some of the spareness of the data.
I've got western pond turtles (WPT) and red-eared sliders (RES) at my site. I made 7 groups (sample size): WPT-male (11), WPT-female (5) (there were no juvenile WPT), RES-juvenile (17), RES-male (86), RES-subadult female (20), RES-small adult female (67), RES-large adult female (50).
I decided to run these data through POPAN to get survivorship, recapture probability, and population size. I thought about robust design, and right now I can't remember why I ruled it out. I think it was because I'd violate the assumptions of long intervals between the short sampling periods...not that I'm not violating the instantaneous capture and release assumptions anyway.
Because these are long-lived organisms and the duration of the study was so short relative to their lifespan, I thought the most biologically relevant quesitons would be to look at whether there are differences in survivorship and recapture probability among groups, which is why I organized my data this way. Do you think I'm on the right track?
Thanks,
Laura