Jackknife, Ridge and Ordinary Least Squares Estimators of Regression Parameters: A Monte Carlo Comparison.
Abstract
This study reports the results of a Monte Carlo evaluation of the small sample performance of ordinary least sqares (OLS), ridge and jackknife estimators of regression coefficients. The primary criteria of evaluation are the mean square error (MSE) of the regression coefficients and the size of the t-statistics associated with these coefficients. The conditions studied are derived from a design in which two factors are varied: the sample size to number of predictors (N/p) ratio and the metric quality of the data (dichotomous and polychotomous). Results from 500 replications of each of six cells showed that OLS and jackknife estimators did not differ appreciably on either criterion. Ridge weights had substantially smaller MSE at the lowest levels of N/p (5:1). At high levels of N/p (20:1), ridge weights had a slightly greater MSE than either OLS and jackknifed weights. A dilemma in the use of t-statistics for ridge weights is presented. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1979
- Accession Number
- ADA077724
Entities
People
- Michael K. Lindell