Hi Sivan,
Chain lengths that differ that dramatically different usually indicate
a problemin the likelihood. It can often mean that there are multiple
modes and different chains are ending up at different nodes. You can
try to use the "overdisperse" diagnostic to see if it confirms this
intuition. This could be happening due to an extreme dependence in the
data--perhaps an almost linear transformation exists in one of the
measures, for instance. Or, two of the measures are almost always the
same.
Cheers,
matt.
On Wed, Feb 9, 2011 at 10:37 PM, Sivan Rotenberg
<sivanrotenberg(a)gmail.com> wrote:
Hi Everyone,
I used amelia to run an imputation and I was wondering if there was a
standard on how consistant the chain lengths should be for each imputation.
I did 10 imputations, and some ranged between 10-12 iterations while other
took between 180-400. The tolerance was .001 and I set the ridge prior to be
5% of the observations. Are these differences ok or do I need to change
something to try and get the imputations more equal. The data set has 127
observations with 6 measures that are repeatedly measured over 3 time
points.
Thanks for your help!
Sivan
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive:
http://lists.gking.harvard.edu/?info=amelia
More info about Amelia:
http://gking.harvard.edu/amelia