Re: st: mi impute chained troubleshooting

[Jacob McDermott <jpmcd87@gmail.com>]

Thank you for the response.

The error I received was:

mi impute: VCE is not positive definite
    The posterior distribution from which mi impute drew the 
imputations for ldl_all is not proper when the VCE estimated from the 
observed data is not positive definite.
    This may happen, for example, when the number of parameters exceeds 
the number of observations. Choose an alternate imputation model.
error occurred during imputation of educacion height hba1c_all 
glucose_all hdl_all ldl_all tc_all triglyc_all crp_all obesidad tension 
corazon diabetes smoke1 exercise1
on m = 1
 -- above applies to female = no

Concerning your suggestions:
tc_all is not an arthimetic sum of hdl and ldl, but your -omit- comment 
helped me come up with the following model:

mi impute chained (ologit) educacion (regress, omit( tc_all )) height 
hba1c_all glucose_all hdl_all ldl_all triglyc_all crp_all (regress)  
tc_all  (logit) obesidad tension corazon diabetes smoke1 exercise1, 
add(5) by(female)

This ran without any errors.

Thanks for the help,
Jacob


On 7/16/2013 4:26 PM, Stas Kolenikov wrote:
> First of all, we'd all be better off if you showed the output and
> reported the error message verbatim. You may have had a syntax error,
> for instance, rather than a statistical difficulty.
>
> In terms of modifying the equations, you can specify -omit( tc_all )-
> option to remove that variable from your equations, although you would
> still get it imputed since it appeared in -mi impute- statement. If it
> is an arithmetic sum of hdl and ldl, then you should -mi register
> passive tc_all- and -mi passive : generate tc_all = hdl_all +
> ldl_all-.
>
> -- Stas Kolenikov, PhD, PStat (ASA, SSC)
> -- Senior Survey Statistician, Abt SRBI
> -- Opinions stated in this email are mine only, and do not reflect the
> position of my employer
> -- http://stas.kolenikov.name
>
>
>
> On Tue, Jul 16, 2013 at 2:59 PM, Jacob McDermott <jpmcd87@gmail.com> wrote:
> > Hello Statalist,
> >
> > The following mi impute chained command (unsurprisingly) returns an error
> > mi impute chained (ologit) educacion (regress) height hba1c_all glucose_all
> > hdl_all ldl_all triglyc_all crp_all tc_all  (logit) obesidad tension corazon
> > diabetes smoke1 exercise1, add(5) by(female)
> >
> > The issue rising from collinearity due to the inclusion of tc(total
> > cholesterol) along with ldl and hdl (major components of total cholesterol)
> > Note: tc_all ldl_all and hdl_all are continuous variables with roughly 10%
> > missing values
> >
> > If I remove tc_all from the imputation command (or if I remove either ldl or
> > hdl) everything runs just fine, but I do not get imputed values for tc_all.
> >
> > Is there anything I can do to get imputed values for tc_all?
> >
> > What I have been trying is running the above imputation command twice (once
> > excluding tc_all and again excluding ldl_all), and then merging in tc_all
> > imputed values from the second run into the first imputed dataset. My only
> > real justification for this is that ldl and tc are highly correlated and the
> > resulting imputed datasets appear to be very similar . Do you think this
> > procedure is at all legitimate? Do you know of any alternatives?
> >
> > Thanks for your help,
> > Jacob
> > *
> > *   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/

*
*   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/