Date:         Wed, 23 Feb 2005 15:24:58 -0800
Reply-To:     cassell.david@EPAMAIL.EPA.GOV
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "David L. Cassell" <cassell.david@EPAMAIL.EPA.GOV>
Subject:      Re: Multiple Imputation--Missing Data
Comments: To: "Nick ." <ni14@mail.com>
In-Reply-To:  <20050223194834.64D4F101D0@ws1-3.us4.outblaze.com>
Content-type: text/plain; charset=US-ASCII

"Nick ." <ni14@mail.com> wrote to me personally:
> I found out why SAS was not imputing. In PROC MI when listing the
> inputs (in VAR statement) one MUST also include the left hand side,
> i.e. the Y along with the inputs X1, X2, X3, etc. I still get the
> same error as before. Also, I look at some of the imputed values and
> they are absolutely false! How can SAS look at an input, say home
> market value, which ranges from $0 to $1 million+, and impute a
> missing value as negative, like, $-235,567.345? Anyway. My previous
> questions still stand. Thanks ...

This is the problem David Neal and I were talknig about the other
day.  If you use the defaults, SAS will fit distributions which
are likely to be inappropriate.  You have a distribution of market
values which (in this context) will not ever go down below zero.
(I assume.)  But you don't get that truncation if you implicitly let
SAS fit a normal distribution to your data.  You have to make sure
that a meaningful distribution of values gets used.  In the worst case,
you end up doing as Dale McLerran has discussed in SAS-L and fitting
your own distribution to the data and then letting PROC MI impute from
there.  (You can look up Dale's code in the SAS-L archives.)

Unfortunately, PROC MI in SAS 8.2 (as you are using) is not as
sophisticated as later versions.  So you don't have the flexibility
that appears in SAS 9.0 or SAS 9.1 .

HTH,
David
--
David Cassell, CSC
Cassell.David@epa.gov
Senior computing specialist
mathematical statistician

"Nick ." <ni14@mail.com> wrote on 02/23/2005 11:48:34 AM:

> David,

> NICK
> --
> ___________________________________________________________
> Sign-up for Ads Free at Mail.com
> http://promo.mail.com/adsfreejump.htm
>