RE: st: mi impute chained

[Richard Williams <richardwilliams.ndu@gmail.com>]

At 03:28 AM 11/2/2012, chong shiauyun wrote:
> I used -force- in my MI model it resulted in a number of missing 
> values which can't be imputed. For example, I have 7000 missing data 
> on IQ and only 1800 are imputed.

Do you have hard missing or soft missing? From the manual:

hard missing and soft missing. A hard missing value is a value of .a, 
.b, : : : , .z in m = 0 in an imputed variable. Hard missing values 
are not replaced in m > 0. A soft missing value is a value of . in m 
= 0 in an imputed variable. If an imputed variable contains soft 
missing, then that value is eligible to be imputed, and perhaps is 
imputed, in m > 0. Although you can use the terms hard missing and 
soft missing for passive, regular, and unregistered variables, it has 
no special significance in terms of how the missing values are treated.

> I know this is because of the missingness of predictors in my model 
> but I don't understand why is happens because I have already 
> specified to use -mi impute chained- to impute other predictors as well.
>
> is there any ways to overcome this problem?
>
> This is how the IQ conditional model looks like:
>
> truncreg totaliq birthweight i.smkpreg i.marist i.matdepr i.homeown3 
> i.alcpreg3 i.hhcrowd i.ednmatpat i.findiff i.ethnicity i.scmatpat3 
> i.mumhealth i.breastfed i.social verbiq perfiq mumiq tempcatwlc_i 
> tempcatclc_i sex mumage babygestation , ll(45) ul(151) noisily
>
> Many thanks
>
> Shiau
>
> > Date: Wed, 31 Oct 2012 08:48:39 -0400
> > Subject: Re: st: mi impute chained
> > From: jvverkuilen@gmail.com
> > To: statalist@hsphsun2.harvard.edu
> >
> > On Wed, Oct 31, 2012 at 4:54 AM, chong shiauyun 
> <shiauyun416@hotmail.com> wrote:
> > > Hi,
> > >
> > > thanks for your advice.
> > > I simplified my MI model by excluding some interactions and 
> reduced some of my variables. It works fine. However, I am concern 
> that I have to use the -force- option  to make the model works. It 
> am quiet reluctant to drop all of the interactions seeing that it 
> may affect the relationship between the exposure and the outcome 
> which I am interested in.>
> >
> > I've used -force- and I think it works OK but check using
> > -midiagplots-, which you can download.
> >
> > As to the interactions, if they're very collinear with the other
> > variables you have in the model they're not adding anything. You can
> > experiment with dropping or adding variables and keep checking with
> > -midiagplots- to determine how things are working. Remember, the idea
> > in MI is not to have a perfectly reconstructed dataset but to
> > optimally preserve insofar as is possible the information you do have.
> >
> > *
> > *   For searches and help try:
> > *   http://www.stata.com/help.cgi?search
> > *   http://www.stata.com/support/faqs/resources/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/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW:    http://www.nd.edu/~rwilliam

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