Statalist The Stata Listserver


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

RE: st: RE: mim: reg command


From   Richard Williams <[email protected]>
To   [email protected]
Subject   RE: st: RE: mim: reg command
Date   Fri, 27 Apr 2007 18:11:19 -0500

At 05:55 AM 4/27/2007, Maarten Buis wrote:
3) Finally, Stata's excellent simulation tools can help you assess how bad or good
Richard's Rule works compared to Rubin's Rule. The point estimates seem fine, but the
coverage seems to be off. To get a better view of the coverage, you would probably have
to run the simulation longer, but even with 500 simulation it seems pretty clear that with
Richard's Rule you reject a true null hypotheses in more than 5% of the samples, while the
coverage for Rubin's Rule is closer to the nominal 5%.
"The point estimates seem fine" might suggest that Richard's Rule may sometimes have some redeeming social value! We often aren't concerned with p-values and we often don't need the numbers to be exactly right. For example, a student of mine working with a mim dataset wanted multicollinearity diagnostics, so she just ran the analysis pooling all the imputed data sets. Sometimes, too, it is sufficient to know whether a diagnostic test clearly indicates that there is or is not a potential problem. And, then there are things like the adjust command or Scott Long's spost routines, which help you to interpret results. It may be sufficient for their output to be in the ballpark.

That is not to say that you should deliberately use numbers you know are wrong when you can easily get the right ones, but if the choice is no numbers at all versus ballpark numbers the ballpark numbers may be worth having.

It seems, too, that you could also do separate analyses on each imputed data set, to see how wildly the numbers fluctuate.

I'm not as good as Maarten is at figuring these things out and getting them programmed in 15 minutes, so my Stata 10 wishlist now includes extensive support for a mim: type command. Ideally, it would be very similar to the svy: command, and do all the appropriate calculations, and know when, say, a Wald test is more appropriate than a likelihood ratio test, and so on.

Again, great post, Maarten.


-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME: (574)289-5227
EMAIL: [email protected]
WWW: http://www.nd.edu/~rwilliam

*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/




© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index