Page 1 of 1

Individual site estimates of Psi in RMark?

PostPosted: Tue Jan 14, 2014 3:38 pm
by ekalies
Is there a way to compute individual site estimates of Psi (and other real parameters) in RMark (I'm using model RDOccupEG)? PRESENCE does this quite nicely in the standard output. In RMark, the real estimates are only given by group when I use summary(model)$real. If I use covariate.predictions, I can start to get at what I want, but it gets pretty complicated if I have several covariates- plus, I have to do a lot of recoding for each model/covariate of interest. I'm wondering if there is a more straightforward way of producing individual site estimates of Psi (and other real parameters), based on a single model, that I am missing?

Re: Individual site estimates of Psi in RMark?

PostPosted: Wed Jan 15, 2014 1:20 am
by bacollier
ekalies wrote:Is there a way to compute individual site estimates of Psi (and other real parameters) in RMark (I'm using model RDOccupEG)? PRESENCE does this quite nicely in the standard output. In RMark, the real estimates are only given by group when I use summary(model)$real. If I use covariate.predictions, I can start to get at what I want, but it gets pretty complicated if I have several covariates- plus, I have to do a lot of recoding for each model/covariate of interest. I'm wondering if there is a more straightforward way of producing individual site estimates of Psi (and other real parameters), based on a single model, that I am missing?


If your interest is in estimating a Psi for each surveyed site then why not just insert the covariate data from your dataframe into covariate.predictions for data=? Or you could even use something as simple as plogis() to estimate Psi for each row (site) of the data? I mean, effectively it seems that all you want to do it take your models beta estimates and multiply them by your covariate info from each site to predict site-specific estimates? How is it getting complicated if you have multiple covariates, it all works the same whether its 1 or 10 covariates in a model as the site level data are fixed for each site?

for instance, if your best model for Psi is alpha + beta1*Height

and your data consists of a data frame in R (call it H) with 10 rows of covariate data on Height measured at each survey site, then your prediction for each survey site would be

Code: Select all
plogis(alpha _+ beta1*H$Height)


would predict a Psi for each value of Height in your dataframe, so on and so forth for other covariate data.

\bret