Back-transforming Parameter Estimates

[Daniel]

Hi all,
This really seems like something that should be straightforward but haven't been able to find anything here that addresses it directly.

I've gotten 12 models as part of my 95% confidence set and would like to do some model averaging and had a couple questions that seem basic but I haven't been able to find good answers to.

1) Is there an easy way to back-transform individual parameter estimates from each model to normal probability scale? It's easy to do that for actual actual occupancy or detection probability resulting from each model but it didn't seem like that would work for individual parameters since you get prob values over 1 if you have binary covariates.

2) Adding on to that, can you average the logit scale parameter estimates over my 12 models and then use those as the betas for an overall averaged model? I had thought no since you would be averaging values on a non-linear scale but I tested a couple values and it seemed to work.

3) Lastly, is it even common to report individual or averaged parameter estimates in the normal probability scale? I'm mostly seen just the logit parameters reported.

Thanks in advance for any thoughts.
Daniel

[darryl]

Hi Daniel,
1. PRESENCE uses the logit-link function by default so you can get your probabilities from the betas and your covariates by using the usual appraoches as for logistic regression.
2. You could, but personally I think it's better to get the derved values from each model first, then model average those.
3. Completely depends on what it is you're trying to achieve. Eitehr could be appropriate in different situations.
Cheers
Darryl

[Daniel]

Thanks Darryl!