Date: Thu, 29 Sep 2005 15:34:55 -0400
Reply-To: Sigurd Hermansen <HERMANS1@WESTAT.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Sigurd Hermansen <HERMANS1@WESTAT.COM>
Subject: Re: Selection of a good statistical Model
Content-Type: text/plain; charset="us-ascii"
While David and Peter have dealt with statistical issues in fine style,
I might add a note on why time series require special methods. Early
studies of time series in economics found that many have extraordinarily
high degrees of serial correlation over intervals of a year or, in some
cases, longer. Trends in oil prices, for example, tend to persist month
to month. The impacts of Katrina and Rita on oil prices will propagate
across several months. A model that predicts continuing high prices next
month may have a low prediction error, but it doesn't tell us anything
that we don't already know. A better model predicts how the impact of an
event or intervention plays out over several weeks or months at least.
An event or intervention may start a longer term trend or a shorter or
longer cycle. An even better model separates seasonal trends and ongoing
cycles. A great model would also separate out 'cross-section' effects
From: email@example.com [mailto:firstname.lastname@example.org]
On Behalf Of Deepak
Sent: Wednesday, September 28, 2005 10:20 AM
Subject: Selection of a good statistical Model
I need some help in selecting the right model for the set of data I
have. I have the following data: Dependent Variable: Sales (Continous
Variable) Independent Variables: price (Continous Variable) and
Promotions (Categorical Variables - On/Off type) My objective is to
predict the Sales (dependent varible) using the Independent variables.
Currenly I am using Linear Regression technique which is not giving good
results (which may be due to the mixed type of Independent variables
(Categorical & continous)). I might not be able to use Logistic or
Probit models because of the dependent variable nature.
It would be of great help if anyone can direct/help me in approaching
this modeling problem.