[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: multiple_imputation

From   Maarten buis <>
Subject   Re: st: multiple_imputation
Date   Sun, 12 Jul 2009 18:00:07 +0000 (GMT)

--- On Sun, 12/7/09, Victor Mauricio Herrera wrote:
> this paper shows that a hierarchical approach
> is the one I should take in order to correctly deal with
> missing values in multiple surveys.

I wouldn't take that so literally, hierarchical models 
are a pet-model of Andrew Gelman. This is a sensible
way of thinking about this issue, but not the only way.
It is a good thing to have pet-models, these are tools
of our statistical toolkit that you are very familiar 
with, so it makes sense to use them whenever slightly
possible. The total number of tools available is just
too large to be familiar with all them. However, it 
does not mean that somebody's pet model should also be 
the pet-model of somebody else.

I would try to stick with models you can estimate with
-ice-, as Gelman and collegues haven't implemented 
their model in software. In the end, it is no use to
think about models you can't estimate. (I actually
considered writing a program that implemented this
or a related model and contacted some of the authors 
about it, but I soon abandoned this project, as it would
be so complicated that finishing it would just cost too
much time against too little payoff.)

-- Maarten

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen


*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index