dhewitt wrote:I am no expert on RD analyses and should probably shut up at this point. Nonetheless, it doesn't surprise me that the main challenge is disentangling survival and temp. emigration -- that's a tough beast. In general, it sounds to me like your task is to decide on what makes a reasonable simplified model to start with.
The problem I got myself into is that there are better models than those from simple model sets, and the model selection process yields a different answer with respect to the question(s) at hand depending on whether I start with this large set of 60, or just stick to models with constant survival and temp. emigration.
I've read in some papers people say things like "we only tested models where p varied between primary periods, not within, to reduce the number of parameters estimated." Well that's all good and dandy. But in my case, if I do that, the result is a model where p does not vary by time. And the question at hand is, does p vary over time? Well with a larger model set where p is allowed to vary within AND between primary periods, the best model is a time varying p model.
Two (legitimate?) model selection procedures, with completely different results. How does one reconcile this?
Thanks for your thoughts on this BTW.
dhewitt wrote: I can't help with the GOF and median c-hat stuff for these models. Can you coerce Bill Kendall into being interested?
Bill Kendall, are you interested at all in this?