Date: Fri, 29 Jul 2005 14:35:33 -0700
Reply-To: David L Cassell <davidlcassell@MSN.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: David L Cassell <davidlcassell@MSN.COM>
Subject: Re: How to Score a New Data set Based Proc NLMIXED Results
In-Reply-To: <MC8-F7sHipF1yTcsVpc002f4952@MC8-F7.hotmail.com>
Content-Type: text/plain; format=flowed
djrk0003@COMCAST.NET wrote:
>I have run proc NLMIXED and outputted my estimated parameters in an
>output file called c.est. I have a new data set that I would like to
>score. My new data set is called c.PFID827. I tried to use Proc Score
>but always got an error message. I tried to change the type= to various
>fields in my c.est data set but got the same error message shown below.
>How can I new a new data set based on the estimated coefficient from
>Proc NLMIXED?
PROC SCORE assumes a linear functional of your variables in the score
data set, with enough extra data in the score data set to adjust
according to any variance ransformation needed. Do you even have
a linear relationship? If so, then why are you using PROC NLMIXED?
If not, then why are you trying to use PROC SCORE? I'm not grasping
your plan here.
Then your errors messages look like:
>1402 id accountidx Iadjx Payamtwf transactionb chapter OwnedServiced
>SecuredUnsecured Ageadj;
>ERROR: Variable AGEADJ not found.
>1403 Var balwlog AgeAdjMean chapter OwnedServiced IssuerN34
>IssuerN15 DistCeMa DistRest
>1403! SecuredUnsecured AgeAdjMeanChap
>ERROR: Variable BALWLOG not found.
>ERROR: Variable AGEADJMEAN not found.
>ERROR: Variable CHAPTER not found.
>ERROR: Variable OWNEDSERVICED not found.
>ERROR: Variable ISSUERN34 not found.
>ERROR: Variable ISSUERN15 not found.
>ERROR: Variable DISTCEMA not found.
>ERROR: Variable DISTREST not found.
>ERROR: Variable SECUREDUNSECURED not found.
This looks like you are putting in varaibles which are not listed in
your c.est file. Is that the case? If so, then what are you trying
to achieve here? You want to have the same variables in the
score data set as in the input data set. Any extra variables serve
no purpose, and any fewer variables leave you with no way to
estimate the linear functional.
I don't think you should be doing it like this anyway. PROC NLMIXED
has its own PREDICT statement. Use it. You can't fit a non-linear
relationship using PROC SCORE, you can't do the Empirical Bayes
estimates of any random effects in PROC SCORE, and you can't
estimate the variances properly when NLMIXED is using the delta
method to get them.
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330
_________________________________________________________________
On the road to retirement? Check out MSN Life Events for advice on how to
get there! http://lifeevents.msn.com/category.aspx?cid=Retirement
