Hi Iza,
Its important to run your chosen model on each of the 5 or 10 imputed
datasets separately and then to recombine any given quantity of interest
using the formulae in the lecture and section notes (on the latter, the
formula for standard errors had a typo originally but it is fixed now, so
update your notes accordingly if you printed them out). While it might be
possible to average imputations for purposes of getting a point estimate,
you would be underestimating considerably the uncertainty over the imputed
data by acting as though your averaged imputation were fully observed data.
The formula for the standard error of your quantity of interest
incorporates the extra uncertainty in your estimates generated by
uncertainty about the imputed data.
Iain
On Tue, Apr 27, 2010 at 12:23 AM, Iza Ding <yding at fas.harvard.edu> wrote:
Hey class,
I have two quick questions regarding amelia. Apologies if they have
already been answered. First, is the result going to be different if I
average my imputed values of the 5 data sets first and then run the
models on them? Second, given my results on the 5 data sets, what is
the proper way to get the "averaged standard error", do I take the
average of the standard errors?
Thanks,
Iza
_______________________________________________
gov2001-l mailing list
gov2001-l at
lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l