ubms model averaging

[Natalia Martínez]

Hi everyone,  I’m seeking help regarding models fitted with ubms.

Due to my relatively small sample size, I analyzed several seasons using single-season models and fitted occupancy models in three steps. First, I fixed detection, then tested individual habitat variables to fix a parameter on occupancy, and finally tested additive occupancy models. I limited it to three variables because, in some cases, it improved the estimation of some betas. In general, the best model is the one that fits best and contains variables that do not cross zero. However, in some cases, several models in the ranking are competitive (diff ELPD < SE diff ELPD), but only the habitat variable does not cross zero, while the rest do. For these cases, how could I perform model averaging with models fitted in ubms? Is it possible to use only the best model to estimate mean occupancy and interpret covariates?

example:

[Ken Kellner]

Unfortunately ubms does not have any model averaging functionality built in. I may look into model-averaged prediction sometime in the future.

I would recommend either (1) fit one "global" model containing all the covariates relevant for your hypotheses/predictions (what are they?) then address those hypotheses in the results based on the parameter estimates and 95% CI. Or (2) fit a small handful of models, one corresponding to each specific hypothesis, then rank the models and interpret from there. If a few of these models have similar support, that uncertainty just becomes part of your discussion of the results.

Ken