Date: Tue, 19 Oct 2010 15:17:26 -0400
Reply-To: Suzanne McCoy <Suzanne.McCoy@CATALINAMARKETING.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Suzanne McCoy <Suzanne.McCoy@CATALINAMARKETING.COM>
Content-Type: text/plain; charset="us-ascii"
Some of you statistical types might be interested in this. A comment I got from someone on our analytics team: "I've looked into RELR in the last couple of months. Interesting claims, but difficult to verify through the marketing hype."
You and your team are invited to view a brief and free web Primer Course on RELR. Reduced Error Logistic Regression (RELR) is a killer predictive modeling method that includes Survival Analysis and Forecasting applications and is coded as a SAS macro called MyRELR. When used by a skilled analytics professional, RELR will automatically generate reduced error models that are highly stable and interpretable. This is a tremendous advantage over the standard algorithms that are readily available in open source software and are also now sold to be run automatically by non-professionals in spreadsheet and other applications that claim to "emulate data mining experts". In contrast to these crude attempts to automate the standard algorithms in "pop analytics" for use by non-professionals, RELR is definitely not designed for non-professionals. Instead, RELR is "the automated tool for the skilled analytics professional" that greatly enhances rather than crudely emulates highly skilled model building.
A RELR model built from a sample size of 1000 observations can be as accurate as standard algorithms using a sample size of 50,000-100,000 observations. You will see examples of this error reduction in this course. The reason for the performance lift is that RELR accurately models and subtracts error as part of the regression. It also allows the rapid modeling of very large numbers of correlated variables and interactions without concern about standard rules that limit numbers of variables and accuracy. Users have been especially impressed with RELR's automatic variable selection which returns parsimonious models with exceptional face validity and accuracy. Unlike Stepwise Regression, RELR's variable selection is a highly stable and most probable optimization solution, so that a different modeler or independent training sample will generate the same or very similar variable selection. RELR is an automated, optimal, one-step modeling approach; this is why there is a large time savings in building a predictive model compared to standard methods. RELR has been presented at a number of major conferences and proceedings over the past several years, including several invited addresses. You can download some of these papers and read other successful Business Case Studies on our website.
This pre-recorded RELR Primer course is a three part course taught by Dan Rice, Ph.D., the inventor of RELR and the developer of the MyRELR software. You can always gain immediate access to this pre-recorded web course at the Training & Licensing page at www.riceanalytics.com<http://www.riceanalytics.com>.
You may wish to view the brief 9 minute Executive Overview only or you may wish to view all three sections. Please let me know if you have any questions.
Ted R. Wroblewski
Vice President of Business Development
Rice Analytics - Automated Reduced Error Predictive Analytics
10805 Sunset Office Drive, Suite 300
St. Louis, MO 63127
Ph. (314) 962 - 2394