"transients" in recovery data

[OlivierD]

Hi all,

With recoveries data in the LDLD format and modeled with the Brownie parameterization, how does MARK treat individuals that were banded but never contacted again (ie represented as a 10 00 00 00 )? In particular, do these "transient" individuals contribute to the estimation of first year survival when the model has a (real) age structure that separates first year survival from after first year survival?

Thanks in advance for you help

Olivier

[jlaake]

Yes and yes.

[OlivierD]

Thanks.

So basically, a 10 00 00... history is interpreted as "survived the first interval", right? If the data set has 85% of such "transient" individuals, then one should expect the first-year survival to be biased high, or even stuck to the upper limit, no? Is there a way to avoid that?

[cooch]

Thanks.

So basically, a 10 00 00... history is interpreted as "survived the first interval", right? If the data set has 85% of such "transient" individuals, then one should expect the first-year survival to be biased high, or even stuck to the upper limit, no? Is there a way to avoid that?

A history of 10 00 00 isn't interpreted as anything. It was never encountered again. Period. As such, you can't say anything about whether or not it survived.

You're missing the larger point. Transience (sensu Pradel et al.) is a a limiting form of heterogeneity, wherein a transient is defined as an individual for whom the probability of remaining in the sample is 0 after the initial marking attempt. (This is described in *much* detail for live encounter studies in Chapter 7 -- section 7.4) In the simplest case, a transient is an individual marked at time (i) which permanently emigrates the sampling area between (i) and (i+1), such that the probability of subsequent encounter is 0. The estimation of the proportion of residents is then given by (i) the survival probability for the first interval after marking (first TSM class), which is clearly biased low by the presence of some proportion of transients in the sample, (ii) divided by the estimate of survival for residents (second TSM class) for the same interval. Then, the proportion of transients is simply 1-proportion of residents.

So, in a recovery analysis, what would a transient individual be? Again, in the live encounter case, the calculation of the proportion of transients is a function of the fact that a transient is operationally defined as an individual with F=0 after initial marking event, such that the probability of subsequent encounter is 0. The assumption being that only individuals alive *and* in the sample can be encountered. For a dead recovery analysis to be comparable, you'd have to have a situation where the possibility of a dead recovery for a transient individual is 0, which could in theory be the case if you're working with a system where dead recoveries are detected/reported only in the sampling region. In that case, you'd have much the same situation as with a live encounter study, and you could, in theory, look at transience, if you make the same operational definition that Pradel did originally. The moment you can have a dead recovery outside the sampling region is the moment that your definition of transience starts to break down.

Another approach (given the data) would be to consider a Burnham live-encounter/dead recovery analysis, where you assume that dead recoveries occur outside the sampling region (and that only live encounters occur within the sampling region). This data type allows you to parse out true survival from emigration, and again is potentially an interesting way to look at the proportion of transients.

Again, the key is to think hard about what you mean by transient -- it could have a spatial definition, a temporal definition, or both. The specifics of that definition would ultimately determine whether or not you could use a dead recovery analysis alone to pull things apart.

[jlaake]

Sorry if my short answer caused confusion. What you asked was whether the 10 00 00 ... contributed to the survival estimate and yes they do contribute to the survival estimate but that does not mean that you know whether they survived or not. Not sure why you would think it implied they survived. The first 1 in CJS or in the MARK LD formulations like Barker or Burnham simply marks the release occasion. As Evan said, all you know is that it was released. You also don't know whether it is a transient or not depending on how that is defined as Evan so aptly stated. If there are transients and they leave immediately after marking then they will have that capture history but the same capture history will also represent animals that were released and were never recovered or seen again solely by chance. That capture history represents many different possible outcomes including animals that did not survive the first interval and others that did survive the first interval. The only animals with a known fate for the first interval are those with a 1 later in the capture history.

--jeff

[OlivierD]

That's actually what I thought about transient individuals but I guess being not so familiar with the LD format, I needed some confirmation from more experienced users. Thanks a lot Jeff and Evan for your clarification.

Olivier

[cooch]

That's actually what I thought about transient individuals but I guess being not so familiar with the LD format, I needed some confirmation from more experienced users. Thanks a lot Jeff and Evan for your clarification.

Olivier

Have a quick read of chapter 10 -- it covers the LDLD data type, in the context of Burnham's live-dead model.