Percent of Variation Attributable to Model Variation

[tds]

When model averaging, one of the results is the Percent of Variation Attributable to Model Variation. Could someone explain what this is telling me? Much thanks.

[cooch]

When model averaging, one of the results is the Percent of Variation Attributable to Model Variation. Could someone explain what this is telling me? Much thanks.

p. 44 in Chapter 4.

total variation = variation due to the model (conditional variance), and variation among models. So, what percentage of the variation in the model averaged estimate is due to model selection uncertainty (which is the 'model variation' you're asking about).

[tds]

That explanation is much clearer than the book. Thank you.

OK, so a high Percent of Variation Attributable to Model Variation would mean a high model selection uncertainty. Therefore, I assume the weighted average is not as reliable (validated by the large 95% CI).

How do folks decide if the Percent of Variation Attributable to Model Variation is so high that the results are unreasonable to use? Or do folks tend to use the weighted average results, assuming the degree of reliability is illustrated by the large confidence intervals?

Thank you.

[cooch]

That explanation is much clearer than the book. Thank you.

Actually its in the book, but spread over multiple treatments of model averaging, as they arise in various parts of the book. The assumption is that you read (and understand) all the core chapters of the book. Often a question that comes up in (say) Chapter 4, is answered by the time you get to Chapter 6.

OK, so a high Percent of Variation Attributable to Model Variation would mean a high model selection uncertainty. Therefore, I assume the weighted average is not as reliable (validated by the large 95% CI).

In effect, yes.

How do folks decide if the Percent of Variation Attributable to Model Variation is so high that the results are unreasonable to use? Or do folks tend to use the weighted average results, assuming the degree of reliability is illustrated by the large confidence intervals?

The latter.