Re: st: Standardized Beta and Multiple Imputation in Stata 11

[Alan Acock <acock@mac.com>]

I'm not sure what Martin suggests will work. Unless I'm missing  
something (fairly likely), what Maarten says is that it is not clear  
what mean and variance you would use since each imputed dataset will  
have a different mean and variance for each variable. One very tedious  
solution would be something like this:

mi estimate, noisily dftable: regress ln_wagem  gradem agem ttl_expm ///
     tenurem not_smsa south blackm, beta

where the noisily gives all the intermediate results and will show the  
beta values (and R-square) for each of the m solutions.

Pooling the betas and the R-squares is not something that Rubin Rules  
does. You could simply average them. You could do a log of each before  
averaging and then reverse the transformation to deal with the ceiling  
effect in the distribution of betas and R-square. I'm not sure there  
is a formally justified solution.

--Alan Acock

On Dec 31, 2009, at Thu Dec 52:21 , Martin Weiss wrote:

> <>
>
>
> "The way to get those is to standardize all variables in your model  
> prior to estimating the regression command."
>
> Catherine can find technical details in this thread: http://www.stata.com/statalist/archive/2009-03/msg00115.html
>
>
> HTH
> Martin
> -------- Original-Nachricht --------
> > Datum: Thu, 31 Dec 2009 10:04:51 +0000 (GMT)
> > Von: Maarten buis <maartenbuis@yahoo.co.uk>
> > An: statalist@hsphsun2.harvard.edu
> > Betreff: Re: st: Standardized Beta and Multiple Imputation in Stata  
> > 11
>
> > --- On Wed, 30/12/09, McDonald, Catherine wrote:
> > > There is an option in the Linear Regression
> > > estimation to check off 'standardized beta coefficients' in
> > > reporting. But when I do this, nothing changes in my
> > > coefficients. I don't see any beta coefficients reported-the
> > > coefficients are the same as if I did not check it off. Can
> > > someone give me advice on this?
> >
> > The reason is probably that the -mi estimate- dialog box
> > "inherrited" that option from the normal -regress- dialog box.
> > I am not that surprised that it doesn't work, because -mi estimate-
> > seems to work with the returned coefficients as stored in e(b),
> > and the -standardize- option does not change that matrix.
> >
> > The way to get those is to standardize all variables in your
> > model prior to estimating the regression command. The question
> > now becomes which mean and standard devation do I use, and I
> > am not sure, though I guess it doesn't matter that much.
> >
> > Hope this helps,
> > Maarten
> >
> > --------------------------
> > Maarten L. Buis
> > Institut fuer Soziologie
> > Universitaet Tuebingen
> > Wilhelmstrasse 36
> > 72074 Tuebingen
> > Germany
> >
> > http://www.maartenbuis.nl
> > --------------------------
> >
> >
> >
> >
> > *
> > *   For searches and help try:
> > *   http://www.stata.com/help.cgi?search
> > *   http://www.stata.com/support/statalist/faq
> > *   http://www.ats.ucla.edu/stat/stata/
>
> --
> Jetzt kostenlos herunterladen: Internet Explorer 8 und Mozilla  
> Firefox 3.5 -
> sicherer, schneller und einfacher! http://portal.gmx.net/de/go/atbrowser
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/