On Thu, 28 Apr 2005, Speybroeck, Niko wrote:
Dear Professor King,
My name is Niko Speybroeck and I started working for WHO a few months
ago. One of the persons I work with is David Evans.
I am trying out Amelia and compare it with some other software for
imputing data.
I have some questions (the first related to Amelia)?
-- Is the main difference between Amelia and NORM, that NORM uses data
augmentation and Amelia EM Importance Sampling as a method? Is the IP
method in your paper with Honaker et al. equivalent to NORM? I got some
problems when using the IP option in Amelia (the Windows version, I do
not have Gauss). Do you have any advice? Do you know about a paper
comparing these two? Furthermore do you know about the mice software of
van Buuren? How does it compare to Amelia? Can you use the same
variables to impute as the ones you use in the analysis?
NORM and Amelia's EMis will give almost identical answers, but Amelia is
much faster. it also requires less expertise to use well since NORM
requires that you understand MCMC diagnostics. I don't know the details
of MICE. the paper we wrote explains these things. see
http://gking.harvard.edu/files/abs/evil-abs.shtml
-- Do you agree with the following statement (based on Schafer (2003)
Statistica Neerlandica 57, 19-35): "In the early years, many thought
that imputing data under one model and analysing the imputed datasets
under another model was ludicrous and potentially harmful but in some
settings, there might be advantages when imputation and analysis models
differ."?
Do you also feel that the imputation model should at least contain all
the variables in the model used to analyse the imputed data?
its ok for the two models to differ, but the imputation model should
include all the variables (and squares and such) from the analysis model.
our paper talks a lot more about this; I suggest you have a look.
Thanks a lot in advance for your answers and advice. If you have any
references related to these questions, do not hesitate to let me know.
best of luck with your research.
Gary King
---
Gary King
Institute for Quantitative Social Science
Harvard University, 34 Kirkland St, Cambridge, MA 02138
http://GKing.Harvard.Edu, email: King(a)Harvard.Edu
Direct 617-495-2027, Assistant 495-9271, eFax 812-8581
Kin
-
Amelia mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive:
http://lists.hmdc.harvard.edu/?info=amelia