There are pooling rules in both Rubin (1987) and King et al. (1999), described in the
documentation for Amelia II. The short answer, however, is that your university computing
center is entirely wrong. If you have "m" datasets, you will have to perform the
analysis m times (on each one) and then use the combination rules to arrive at a final
parameter estimate.
If you have a fairly standard model, Zelig should be able to handle this for you without
too much trouble. It has an way of passing imputed datasets to the model pretty easily
(but I can't recall the syntax off the top of my head).
Best,
-Nathan A. Paxton
----------
Nathan A. Paxton, Ph.D.
Lecturer
Dept. of Government, Harvard University
napaxton AT fas DOT harvard DOT edu
http://www.fas.harvard.edu/~napaxton
=========================================================
The most courageous act is still to think for yourself. Aloud.
-Coco Chanel
=========================================================
On 24 May 2010, at 2:00 PM, Mitzi Lewis wrote:
Good Afternoon,
My name is Mitzi Lewis and I'm working with Amelia for the first time to
help deal with missing data for my dissertation. I've been able to use
Amelia to get the five imputed files, outdata_. However, I'm running into
problems pooling the results of subsequent regression analysis run on each
of these files. Some of the results can be manually pooled, but not all (at
least I don't know how to do it; e.g., calculating a pooled p value for 5
Wald statistics or pooled confidence intervals for odds ratios). When I
called a research and statistical support center at my university, it was
suggested that all I need to do is run the statistical analysis on the last
of the imputed files (outdata5 in this case) as each of the imputed data
files holds increasingly better imputed data. Is this correct? If so, is
there some sort of documentation I could cite to support this decision? The
information I've found (before making this call) has centered around how to
pool the data. My next step was going to be exploring Zelig, but if the last
imputed file from Amelia would be appropriate for analysis, that would save
some time (and I'm pressed for time on this particular project).
Thanks so much for any guidance you can provide.
Mitzi
-
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 -
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