model-averaging and c-hat>1

[efaubion21]

Hello,
I apologize if my question is answered in the literature, I've looked all over this forum and elsewhere and I can't seem to find a straightforward answer.
I have finished running models in PRESENCE and I am having some difficulty with determining the best way to present the results. I have six species of interest for three individual years and for all three years combined. I'm running a simple single-season single-species model for each species of interest. I would like to create a figure for each species with year on the x-axis and model-average estimates of psi on the y-axis with confidence intervals. I would like to only include models that had a delta-AIC<2. Would it be appropriate to run all of my models and then remove the ones with delta AIC>2 and then run the model averaging of tool? Or do I need to run the model averaging for all of the models and sparse out the desired models from that? I tried to do the model averaging and the se manually but I can't seem to get the same numbers that the tool outputs. The wgt x est for the estimate average is very close to the same, but I can't get the standard error to be the same. I'm not sure what calculations were used to create the modelavg standard error values.

My second question is that many of my species have a c-hat of greater than 1. I know that means I have to adjust the estimates by sqrt(c-hat), which I did for the beta coefficients. Does that include the estimates in the model averaging?

I would appreciate any advice I can get, I want to make sure I'm doing it correctly. Thanks!

[jhines]

I suggest using the model-averaging tool and include all models. Since the models are weighted by differences in AIC scores, models with higher AIC scores (eg., > 2) will get proportionally less weight and have little influence on the model-averaged estimates.

The model-averaged se is comprised of 2 parts: a) variance of each model estimate and 2) variance among models of each estimate. To compute it,

1) compute model-averaged estimate of ma.psi = sum[psi(i)*modwgt(i)] for model i=1,2,...
2) compute within-model var[i](psi) = se(psi)^2 for model i=1,2,...
3) compute among-model amvar[i](psi) = (psi[i] - ma.psi)^2
4) compute mod-avg var = (var[i](psi) + amvar[i](psi)}*modwgt(i)
5) compute mod.avg.var.psi = sqrt(mod-avg var)

If c-hat > 1, then variances should be adjusted by c-hat.