On Nov 11, 2008, at 9:13 PM, carla osorio wrote:
> Hello,I would like to know what is the best way to calculate and
> report bivariate correlations when multiple imputation (Amelia) is
> employed.Is it OK if I calculate the correlations between the
> variables of interest in each of the imputed datasets and just
> average them?Also, I have read that it is fine to average coeffients
> over the imputated datasets, but I do not know if standard errors
> can also be treated in that way. thanks.Sincerely,Carla Osorio
I post on this only to point out that this problem corresponds to a
lovely toy example I trot out when teaching this stuff, using a data
set that has been floating around since DLR first published on the EM
algorithm, due to Gordon Murray. For the example data set Murray
cooked up, and with a vague prior over the correlation and an
assumption of bivariate normality for the likelihood, you wind up with
a bimodal posterior density for the correlation. In cases like that
--- rare, I should think --- maybe you want to give your readers a
picture of what uncertainty over the parameter in question looks like.
see http://jackman.stanford.edu/mcmc/bivarExtremeMissingness.pdf for a
little more detail.
-- simon
Professor Simon Jackman
Dept of Political Science, and (by courtesy) Statistics,
Director, Political Science Computational Laboratory,
Director, Methods of Analysis Program in the Social Sciences,
Stanford University, Stanford CA 94305-6044
http://jackman.stanford.edu
**********************************************************
Political Methodology E-Mail List
Editors: Melanie Goodrich, <[log in to unmask]>
Delia Bailey, <[log in to unmask]>
**********************************************************
Send messages to [log in to unmask]
To join the list, cancel your subscription, or modify
your subscription settings visit:
http://polmeth.wustl.edu/polmeth.php
**********************************************************
|
