www.phidot.org

I am new to RMark and shrinkage estimates.

I think the data I am analyzing will benefit from a random effects model. Given that colleagues in my field ( I am a paleobiologist) mostly use R and are unfamiliar with MARK or capture-recapture approaches, I would really love to be provide R code for all the analyses I am doing so the work is more accessible for my colleagues (as an introduction).

I read a post by benaug from June this year where more code was added to var.components that goes beyond only the beta and sigma estimates.

So I am wondering if there are any plans to wrap that code (as Jeff Laake mentioned in his response benaug) and plans to implement the rest of what MARK does for random effects /shrinkage in RMark??

It would be so helpful to my entire analysis in R instead of having to toggle back and forth R and MARK. (I have 25 sampling intervals, so everything takes forever in MARK).

Thanks!

lhliow,

I think you can only do the shrinkage analysis inside of MARK as it is not part of var.components best I can tell.

Bret

Hello Jeff et al.,

As part of a larger project, we are modeling vital rates with random year effects using the var.components.reml command. One of the reasons for our usage is to attain estimates of process variance. In addition, we were wondering if RMark has options to obtain predictions for the shrinkage estimates of either the beta coefficients or real parameters for each sampling unit (e.g. each year) of the random effect? In the documentation for the var.components.reml command, the estimates from the initial fixed-effects model are then modeled in a mixed-effects format using the ‘lme’ command. Was that provided in the documentation such that one could then use the ‘predict’ command for the ‘lme’ model to obtain the associated shrinkage estimates? We figured others might be interested in how to obtain shrinkage estimates using the var.components.reml command in RMark.

Thank you,
Michael Guttery and Dave Koons

Ben Augustine implemented the shrinkage estimates in var.components and I've not done so for var.components.reml unfortunately. The lme code was in the example to show that you got the same results as var.components.reml for that example ignoring the v-c matrix. Using lme might be an option if there is a way to specify a known v-c matrix. If you work something out please post the solution.

--jeff