Administrative Report Or Publication

 

A proposed "partitioned" ridge regression procedure for estimating models unsuited for regular ridge regressions Public Deposited

Downloadable Content

Download PDF
https://ir.library.oregonstate.edu/concern/administrative_report_or_publications/6m311q215

Descriptions

Attribute NameValues
Alternative Title
Creator
Abstract
  • Earlier papers by Brown [1973] and Brown and Beattie [1975] showed, for the common case of positively interrelated explanatory variables, that the ridge regression estimator of Hoerl and Kennard [1970] gave much larger bias and mean square error (MSE) for models with unlike signs. To avoid this excessive bias and MSE in such cases, a method for partitioning ridge estimation into compatible blocks is proposed in this paper. The expected values and variances of the "partitioned" ridge estimates are derived, and the procedure is applied to a simple experimental model, giving partitioned ridge estimates that were two to three times more accurate in terms of }lSE than OLS or regular ridge regression. The partitioned ridge method was also applied to a nonorthogonal subset of data from an empirical produc­tion study. Both regular and partitioned ridge regression estimates had about one-half the MSE of OLS for certain ranges of k values. However, the partitioned ridge estimates were superior to the regular ridge estimates in that a lower MSE was obtained over a wider range of k values. Also, a much larger proportion of the partitioned MSE was composed of easily measured variance, as compared to a high percent of bias squared component for the MSE of the regular ridge regression estimator. These results indicate that the partitioned approach should substantially in­crease the usefulness of biased linear estimation for many regression problems.
  • Published June 1975. Facts and recommendations in this publication may no longer be valid. Please look for up-to-date information in the OSU Extension Catalog: http://extension.oregonstate.edu/catalog
Resource Type
Date Available
Date Issued
Non-Academic Affiliation
Series
Subject
Rights Statement
Publisher
Language
File Format
File Extent
  • 1898196 bytes
Digitization Specifications
  • Master files scanned at 600 dpi (256 Grayscale) using Capture Perfect 3.0 on a Canon DR-9080C in TIF format. PDF derivatives scanned at 300 dpi (Black and White) using Capture Perfect 3.0 and OmniPage Professional 15.0 for textual OCR on a Canon DR-9080C.
Replaces

Relationships

Parents:

This work has no parents.

In Collection:

Items