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 on April 23, and its replacement, statalist.org is already up and running.


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

Re: st: RE: combining/integrating the results of -stepwise- and -mim- for variable selection after multiple imputation


From   Jordan Hoolachan <jihool3670@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: RE: combining/integrating the results of -stepwise- and -mim- for variable selection after multiple imputation
Date   Mon, 13 Sep 2010 15:07:05 -0400

Hello all,

As no one seems to have any more insight into this issue and the size
of my imputed data set makes model selection infeasible on my machine
due to the amount of time it takes to run commands, I've been looking
for another approach.

In their short paper, Mehta et al. (2007) discussed a two step
approach for model selection in which regular imputation using the
mean/median of each variable is first carried out, resulting in one
complete data set on which the usual array of model building
techniques can be applied.  After a model has been developed, multiple
imputation is implemented on the original data set.  The model
developed previously is then applied to each of the imputed data sets
and the resulting estimates are combined.

To those who are familiar with multiple imputation:  do any major
drawbacks to this approach jump out?

Here is a link to the paper for your reference:
www.nesug.org/proceedings/nesug07/sa/sa15.pdf

Thanks for the consideration.
Jordan



On Fri, Sep 10, 2010 at 12:19 PM, Lachenbruch, Peter
<Peter.Lachenbruch@oregonstate.edu> wrote:
>
> I think you are correct.  I did get an experimental version called mimsw from Patrick Royston and I thought he'd updated mim with that.  He may be able to let us know if it's part of mim.
>
> Tony
>
> Peter A. Lachenbruch
> Department of Public Health
> Oregon State University
> Corvallis, OR 97330
> Phone: 541-737-3832
> FAX: 541-737-4001
>
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Jordan Hoolachan
> Sent: Friday, September 10, 2010 9:07 AM
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: st: RE: combining/integrating the results of -stepwise- and -mim- for variable selection after multiple imputation
>
> Hey, Tony
>
> I'll look into -ice- further but I believe the stepwise option in the
> -ice- command allows for stepwise variable selection of the prediction
> models used within the imputation.  I'm looking for a way to perform
> stepwise regression on previously imputed data in which the Wald test
> statistics used in the stepwise regression a generated via
> -micombine-.
>
> Jordan
>
>
>
> On Thu, Sep 9, 2010 at 5:47 PM, Lachenbruch, Peter
> <Peter.Lachenbruch@oregonstate.edu> wrote:
> > Update your mim.
> > A simple matter should be to issue the adoupdate command.  I went to the help page and looked for the stepwise option, but didn't find it.  I had seen it earlier.
> > Maybe the authors can help.
> > It seems to be implemented in ice:  Here's an excerpt from the help file
> >
> >
> > Syntax
> >
> >        ice [mainvarlist] [if] [in] [weight] [, major_options
> >            less_used_options]
> >
> > <snip>
> >    options                       description
> >    ------------------------------------------------------------------------
> >    ice major_options
> >      clear                       clears the original data from memory and
> >                                    loads the imputed dataset into memory
> >      dryrun                      reports the prediction equations - no
> >                                    imputations are done
> >      eq(eqlist)                  defines customised prediction equations
> >      m(#)                        defines the number of imputations
> >      match(varlist)              prediction matching for each member of
> >                                    varlist
> >      passive(passivelist)        passive imputation
> >      saving(filename [,replace]) imputed and non-imputed variables are
> >                                    stored to filename
> >      stepwise                    constructs prediction equations by                                                                                                                                                             stepwise variable selection
> >      swopts(stepwise_options)    options for stepwise
> >
> >    ice stepwise_options
> >      forward                     perform forward-stepwise selection
> >      group(group_list)           create groups of variables for joint
> >                                    testing for addition or removal
> >      lock(varlist)               Variables to be kept in all models
> >      pe(#)                       significance level for addition to a model
> >      pr(#)                       significance level for removal from a model
> >      show                        show each stepwise regression
> >
> >
> > Tony
> >
> > Peter A. Lachenbruch
> > Department of Public Health
> > Oregon State University
> > Corvallis, OR 97330
> > Phone: 541-737-3832
> > FAX: 541-737-4001
> >
> >
> > -----Original Message-----
> > From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Jordan Hoolachan
> > Sent: Thursday, September 09, 2010 12:20 PM
> > To: statalist@hsphsun2.harvard.edu
> > Subject: Re: st: RE: combining/integrating the results of -stepwise- and -mim- for variable selection after multiple imputation
> >
> > Tony,
> >
> > I'm not sure what option you are referring to.  I've tried the command
> >
> > xi: mim, cat(combine): stepwise, pr(0.05): logistic ...
> >
> > but receive the message "prefix stepwise is not allowed after -mim-" .
> > I also don't see any mention of a stepwise command within the -mim-
> > help page.
> >
> > Can you be more specific?
> >
> > Jordan
> >
> >
> >
> > On Thu, Sep 9, 2010 at 2:51 PM, Lachenbruch, Peter
> > <Peter.Lachenbruch@oregonstate.edu> wrote:
> >> Check out mim.  It has a stepwise option.  Works well.
> >>
> >> Tony
> >>
> >> Peter A. Lachenbruch
> >> Department of Public Health
> >> Oregon State University
> >> Corvallis, OR 97330
> >> Phone: 541-737-3832
> >> FAX: 541-737-4001
> >>
> >>
> >> -----Original Message-----
> >> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Jordan Hoolachan
> >> Sent: Thursday, September 09, 2010 11:30 AM
> >> To: statalist@hsphsun2.harvard.edu
> >> Subject: st: combining/integrating the results of -stepwise- and -mim- for variable selection after multiple imputation
> >>
> >> Dear All,
> >>
> >> I am using Stata 11.1 and attempting to perform variable selection
> >> after multiple imputation.  All 10 imputed datasets are currently
> >> stacked into one large data set with "_mj" identifying the dataset to
> >> which an observation belongs and "_mi" identifying observations within
> >> a data set.
> >>
> >> In their paper "How should variable selection be performed with
> >> multiply imputed data?", Wood et al. (2008) identify a model selection
> >> approach (the "RR appoach") that utilizes Rubin's rules for estimating
> >> parameters and standard errors across imputed data sets.
> >> Specifically, "each model selection step involves fitting the model
> >> under consideration to all data sets and combining estimates across
> >> imputed data sets."  The only information that they provide in regards
> >> to actually doing this in Stata is the following: "For the RR method,
> >> -stepwise- was modified to use the Wald test statistics from
> >> -micombine- ."
> >>
> >> I am only an intermediate Stata user on my best days so I'm not even
> >> really sure where to start on this.  It seems like I need to code an
> >> iterative procedure in which the results of each -logistic- command
> >> run under -stepwise- are fed to -micombine (or -mim-) which then
> >> combines the results across the imputed data sets and finally feeds
> >> the resulting Wald test statistic back to -stepwise- in order for the
> >> next -logistic- command to be able to run.  Any advice do doing on
> >> setting up this type of program?
> >>
> >> This is the web address of the the Wood et al. paper for your
> >> reference: http://onlinelibrary.wiley.com/doi/10.1002/sim.3177/abstract
> >> Unfortunately, access to the full .pdf is only granted if you have a
> >> subscription.  I couldn't find a location in which it is available to
> >> everyone.
> >>
> >> Thanks for the consideration!
> >>
> >> Jordan
> >> *
> >> *   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/
> >>
> >
> > *
> > *   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/
> >
>
> *
> *   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/

*
*   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/


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