Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

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

st: Multiple-imputation estimates and log-likelihood

From   James Bernard <>
Subject   st: Multiple-imputation estimates and log-likelihood
Date   Fri, 25 Oct 2013 20:40:14 +0800

Hi all,

The -mim-command which performs multiple-imputation estimates across
imputed observations does not produce the typical test-statistics
(log-likelihood ratio etc). Perhaps, the reason is that averaging doe
snot make sense across these statistics.

This seem a bit strange, given the extent that research on imputation
has developed in the past 20 years.

Is this a methodological issue or a computation issue? Any solution. I
do need the typical test statistics.

If so, then would it make sense to use the first imputation (let's say
m=5) and run the model on that. How bad this approach is? How biased
the results would be?

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index