Reverse-time for sensitivity...transitioning from known fate

[Moynahan]

I’ve recently completed a known-fate (KF) analysis of hen sage grouse survival in Montana. I’d like to now look at sensitivity of lambda to changes in hen survival by using a Pradel model with reverse-time analysis (RTA). I’m new to the reverse-time world, so have a few questions.

Background: My KF analysis was structured with 10 intervals across three years: 2 in year 1 and 4 each in years 2 and 3 (April 2001 to April 2004). The best approximating model in the known fate analysis included interval specificity plus two seasonal effects: nesting status (yes, no; only applied during breeding season intervals) and a site-level hunting effect (yes, no; only applied during fall intervals). I have 221 individuals in the data set, with slightly more in years 2 and 3 than in year 1.

I’ve begun the transition by restructuring the LDLDLD input file for KF to LLLL, and will constrain detection probability to 1.0 given the KF source of the data. I’d like to keep the structure of the best-approximating model from the KF analysis, but believe I may have to collapse that data into annual survival (4 occasions, 3 intervals) which would force me to abandon the seasonal effects of nesting and hunting.

My most basic question regards the process. My read on the Nichols et al. 2000 Ecology paper is that I would run the Pradel model once with real or forward capture histories to get survival and lambda, then run it again with an input file that includes reverse capture histories. In this second run, the survival estimates are really the gamma estimates which can then be used to evaluate relative to lambda to get at sensitivity. Is this correct? If so, I believe I can get around the issue of abandoning my important seasonal effects by using the survival outputs from my KF analysis with the gamma estimates from the RTA. Thoughts?

Also, the known fate module allows for censoring individuals during particular intervals (e.g., 1010000010 in the LDLD format); will this cause a problem when I use the LLLL format while constraining p = 1.0? This would be a rare situation, but there are a few individuals that were marked in year one, never heard in year 2, then recaptured and fitted with a new radio in year 3 (e.g., 1001 in the CJS LLLL format).

Finally, how do I constrain p = 1.0 – in the design matrix? Will MARK read this as a “blue 1” beta indicator, or as a “red 1” intended to constrain p to 1.0?

Thanks in advance for your help.

Brendan Moynahan
Wildlife Biology Program
The University of Montana
Missoula, Montana