Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


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

RE: st: mi impute chained error messages


From   chong shiauyun <shiauyun416@hotmail.com>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: mi impute chained error messages
Date   Tue, 23 Oct 2012 14:26:18 +0800

Thanks for your response. I will try linear regression or I may use truncated regression that allows me to specify the lower and the upper limit. Actually I did manage to run the dryrun when I put my syntax like this
mi impute chained (reg) .... (ologit)... (intreg, ll(lTIQ) ul(uTIQ))TIQ, add(20),... dryrun report (just have to put the lower and upper limit in the same blanket as the (intreg) and then specify a new variable, TIQ, that have the impputed value of IQ.

I think truncated regression would be more suitable to apply to my IQ scores (which are continous variables with boundaries between 45-155).I wil also try linear regression.

Another thing that I would like to know is that does it matter if I put variables that I have collapsed (eg. for education level, I collapsed bachelor degree and postgraduate degree into a group called high education) instead of the original data into the imputation model?

Please advice.

Regards
Shiau Yun

----------------------------------------
> Date: Mon, 22 Oct 2012 09:11:41 -0400
> Subject: Re: st: mi impute chained error messages
> From: jvverkuilen@gmail.com
> To: statalist@hsphsun2.harvard.edu
>
> On Sun, Oct 21, 2012 at 8:18 PM, chong shiauyun <shiauyun416@hotmail.com>wrote:
>
> > Hi there
> > I was doing a dryrun for my imputation model using 'mi impute chained". My
> > outcome of interest is Total IQ, which is a continous variable. I censored
> > the score of Total IQ in between 45 and 151 by generating a lower limit
> > ll(lTIQ) and a upper (uPIQ). I was also thinking to put verbal IQ and
> > performance IQ, which are also censored, into the imputation model. In my
> > imputation model, I used different methods including mlogit, logit,
> > intereg, and regression. I also inserted a few interaction terms, which are
> > new variables that I created. When I specify (intreg) for Total IQ, verbal
> > IQ, and Performance IQ, I received error message that interval regression
> > can only take one new variable. I tried again with only Total IQ in my
> > interval regression, but Stata error message said the ll and ul are
> > invalid. Can anyone please help?
> >
>
> I'd simplify things substantially. You have A LOT going on here, so
> much it's not really possible to do troubleshooting. Cut down to
> simple imputation models and build up so you can see where things run
> into trouble. These intermediate models don't need to be realistic,
> but should be moving in the direction of your final model.
>
> For instance, IQ is probably going to naturally censor between those
> cuts in any typical population, so you can probably simplify just by
> switching to a linear regression.
>
> By the way I highly recommend getting the new -midiagplots- command,
> which is simple to use and of great help diagnosing imputation models'
> performances. You may not need a super complex MI model to get
> reasonable imputations. I hope this gets incorporated into the base
> product in the future!
> *
> * 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/


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