Date:         Thu, 7 Nov 2002 13:20:54 -0800
Reply-To:     Dale McLerran <stringplayer_2@YAHOO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Dale McLerran <stringplayer_2@YAHOO.COM>
Subject:      Re: help: how to make graphs after multiple imputatioin
In-Reply-To:  <OFC3859FD8.4EA40439-ON88256C6A.0060A0F0@rtp.epa.gov>
Content-Type: text/plain; charset=us-ascii

While I would agree with David regarding a choice made among these
three approaches, I would think that an altogether different plot
could be very informative.  I presume that in each imputation
dataset, there is some sort of indicator of which values have
been imputed and which data values are real.  Concatenate all of
the imputation datasets and use the real/imputed indicator to
select a plot symbol to be employed.  For each imputation dataset,
you may want to have a different value for the imputed data
indicator.  Now, using the concatenated dataset and an appropriate
set of SYMBOL statements, plot all of the real data and all of the
imputations simultaneously.  Real data from each imputation
dataset will overlay.  If you select appropriate symbols, then
the imputed data will be visually distinct not only from the
real data, but also from one imputation to another.  Then you can
get the full picture conveying both structural relationships in
the real (and imputed data) as well as variance in the real and
imputed data.

Dale


--- "David L. Cassell" <Cassell.David@EPAMAIL.EPA.GOV> wrote:
> Andrew Sun <andrew70912@YAHOO.COM> wrote:
> > Could anybody give your thoughts on how to make graphs after
> multiple
> > imputation (SAS MI procedure)? please comment on the three options:
> > 1. only use raw dataset without imputed values to make graphs.
> > 2. use just one imputed dataset (for example, the first dataset)
> from
> > the multple imputed datasets which are generated from the multiple
> > imputation procedure  to make graphs.
> > 3. take the average of the multiple imputed datasets to make
> graphs.
>
> Go with option #1.  The imputed values are not *real* values, and
> may give a misleading picture, particularly of the variability in
> the data and the missing-data patterns.  The average of the missing
> values in MI will be particularly unhelpful, so *please* don't do
> that.
>
> What's so troublesome about making a graph showing the actual data,
> leaving any holes where it is truly missing?  That shows what the
> data really look like, and that's what a graph should do.
>
> HTH,
> David
> --
> David Cassell, CSC
> Cassell.David@epa.gov
> Senior computing specialist
> mathematical statistician


=====
---------------------------------------
Dale McLerran
Fred Hutchinson Cancer Research Center
mailto: dmclerra@fhcrc.org
Ph:  (206) 667-2926
Fax: (206) 667-5977
---------------------------------------

__________________________________________________
Do you Yahoo!?
U2 on LAUNCH - Exclusive greatest hits videos
http://launch.yahoo.com/u2