Date:         Wed, 5 Sep 2007 18:45:34 -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: Proc GAM
In-Reply-To:  <1115a2b00708300803p290faa4dg1b4d3c6177be75cc@mail.gmail.com>
Content-Type: text/plain; format=flowed

liuwensui@gmail.com replied:
>
>On 8/24/07, David L Cassell <davidlcassell@msn.com> wrote:
> > woods.steve@GMAIL.COM wrote:
> > >How would I predict future data/values using Proc gam. It says "PRED:
> > >predicted values for each smoothing component and overall predicted
>values
> > >at design points". Is this mean I can predict future data using this
> > >method? I guess, we can use Splue and predict future data using GAM, I
>am
> > >not 100% sure though.
> > >
> > >For example, response is Var1 and my predictor variables are var2, var3
> > >and so on. lets assume, we have data from 2005-2007, I want to see the
> > >prediction values of 2008 of var1?
> > >
> > >You must be curious why I need it! Just to find out stocks! Going to be
> > >rich soon using Proc GAM:)
> > >
> > >Thanks in advance!
> > >
> > >Cheers,
> > >SW
> >
> > Regardless of your subject, I would not use generalized additive models
> > like this.
> >
> > Here's my take on the issue.  GAMs are extremely flexible, to the point
> > that they do not make good model builders, and they do not make
> > great parameter estimators.  GAMs are excellent for data exploration
> > and structure visualization.  GAMS are good for examining the
>relationships
> > between the dependent variable and the independent variables.  But
> > once you have decided on this structure, you might prefer to step back
> > to generalized *linear* models to do the model fit and parameter
>estimation.
> >
> > HTH,
> > David
> > --
> > David L. Cassell
> > mathematical statistician
> > Design Pathways
> > 3115 NW Norwood Pl.
> > Corvallis OR 97330

>
>David,
>Long time no see since sugi.

Yes.  It was good to finally get to talk to you in person.  Even if I
couldn't
answer every question you came up with!

>I agree with what you said mostly.
>However, I'd read a paper by Hastie which shows a good use of GAM on
>time series. And personally, I've used GAM to predict daily
>utilization of call centers and it turned out to be good.
>if the purpose of model is prediction, i think it is OK not to worry
>about parameter estimate.

Well, I certainly wouldn't argue with Trevor.

GAM is certainly designed to be used in an exploratory manner, but I tend
not
to use it otherwise, unless the data are so complex that no simpler model
captures
the intricacies of the data.

And I certainly prefer to use time series methods on time series data, as
long
as the time series models provide good fits for the data.

HTCT,
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330

_________________________________________________________________
A place for moms to take a break!
http://www.reallivemoms.com?ocid=TXT_TAGHM&loc=us