Page 1 of 1

Equivalency of Binomial (known-fate) model and Binomial GLMs

PostPosted: Tue May 15, 2018 5:01 pm
by jbauder
I was wondering if anyone could provide some guidance on the equivalency (or lack thereof) between the Binomial model used in the known-fate model and a Binomial GLM/GLMM, particularly when individuals are monitored every day until their death with no censoring. I have some annual survival estimates modeled as a function of landscape features using multi-state models. I am using these “observed” estimates to help calibrate an individual-based model where individual movements/survival are simulated daily and survival is a Bernoulli draw based on the landscape features encountered (e.g., cross a road and your daily survival is X, don't cross a road and your daily survival is Y). Since my simulated data are “perfectly observed,” I could just use a known-fate model to estimate daily survival and then convert daily survival to annual survival for comparison with my “observed” estimates. However, this seems equivalent to fitting a Binomial GLM where the response is 0=survived that day and 1=died that day, although I would suspect that a Binomial GLMM with a random effect of individual would be preferable to account for inter-individual variation. I was wondering if this interpretation was correct, or if there is anything to be gained from fitting such simulated data to a known-fate model.

Thanks,
Javan