|
Activity Number:
|
428
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 9, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistical Computing
|
| Abstract - #306321 |
|
Title:
|
SAS Estimation of Standard Errors for Partial Least Squares Regression
|
|
Author(s):
|
April Grant*+ and David K. Williams and Zoran Bursac and Geoffrey M. Curran
|
|
Companies:
|
University of Arkansas for Medical Sciences and University of Arkansas for Medical Sciences and University of Arkansas for Medical Sciences and University of Arkansas for Medical Sciences
|
|
Address:
|
4301 W. Markham Street, Little Rock, AR, 72205,
|
|
Keywords:
|
partial least squares regression (PLS) ; standard error calculation
|
|
Abstract:
|
Partial least squares (PLS) regression is a useful predictive modeling tool when complex correlation structures exist in data. PLS has well documented applications in the fields of econometrics and chemometrics. While the SAS PROC PLS is able to perform partial least squares regression, estimates of the standard error (SE) are currently unavailable through the procedure syntax or the output. Empirical (bootstrap and jackknife) and closed form solutions are among the possible methods to calculate SE. This research will present a SAS Macro used to estimate SE and show the comparability of the SE with some standard analytical methods through simulation and application examples. By making this macro available to practitioners, PLS can further be applied to other fields and prove a valuable tool to researchers.
|
