[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: Best method for imputing dichotomous variables

From   René Wevers <>
To   <>
Subject   RE: st: Best method for imputing dichotomous variables
Date   Thu, 14 Feb 2008 13:35:14 +0100

Maarten, again thanks for your reply.

The basis of the statement is twofold, indeed one reason is coming from the
hard to explain results I got from -ice- for the (continuous) variables I
mentioned yesterday. However, another reason comes from a simple test I
performed with -ice-. I randomly created missing values (25%) for a
dichotomous variable where there were none missing and imputed these
'missing' values with -ice-. Afterwards approx. 700 out of 3000 imputed
values proved to be different from the original values. When I used -impute-
and rounded the results only 350 out of 3000 imputed values were different
from the original values. Naturally this is a very weak test, but 700 out of
3000 'faulty' imputed values does not give me a lot of confidence in -ice-
for my case.

Of course I will take a closer look at the data and possible assumptions I
might be neglecting for -ice-, but especially for the case I presented
yesterday I doubt whether this can provide a reasonable explanation for the
highly unexpected/unreliable results.

By the way, are there any additional assumptions for -ice- in the case of
dichotomous variables? Distribution should not be a real issue here.

Furthermore I would like to add that I understand that for reliable
imputation often requires a lot of effort. Nevertheless, the case of
imputing employment in FTE from employment in absolute numbers does not seem
like a very problematic case. The explanatory variable is non-missing and
explains over 80% of the variance (R2 = 0,84) of the dependent variable...
Also applying imputation for my analysis is not of extremely high priority,
but more a control mechanism and a method to include observations that would
be lost due to missing values for control variables (of limited importance).



-----Oorspronkelijk bericht-----
[] Namens Maarten buis
Verzonden: Thursday, February 14, 2008 10:44 AM
Onderwerp: Re: st: Best method for imputing dichotomous variables

--- René Wevers <> wrote:
> If I want to impute missing values of dichotomous (dummy) variables,
> what is the best method to apply? The -ice- command here seems to be
> a bad option, since running this command has shown to provide
> unreliable results.

What is the basis for that statement? The software comparisons I have
seen (Horton & Kleinman 2007, Yu et el. 2007) do not show -ice- to be 
unreliable. The weird results you mentioned yesterday are most likely
the result of a problem with your data, and not with -ice-. I would
strongly recommend working harder with -ice-, because, as far as I
know, it is the only Stata multiple imputation package that has been
independently reviewed. 

Any real imputation project, whatwever software you use, will take a
lot of time (don't think days, think weeks) because it will involve
making a lot of mistakes, consequently a lot of puzzeling results, and
a lot of checking, before you can be moderately convidend that you have
done something reasonable. That is the nature of multiple imputation,
not the nature of -ice- or any other program.

Sorry for not being more cheerful,

Horton, Nicholas J.  and Ken P. Kleinman (2007) "Much ado about
nothing: A comparison of missing data methods and software to fit
incomplete data regression models" The American Statistician, 61(1):

Yu, L.-M. , Andrea Burton and Oliver Rivero-Arias (2007) "Evaluation of
software for multiple imputation of semi-continuous data" Statistical
Methods in Medical Research, 16: 243-258.

Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

Sent from Yahoo! Mail - a smarter inbox

*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index