A monte carlo comparison of some ridge and other biased estimators

摘要

A comparative study of the performance of 8 ridge estimators, 5 other biased estimators, and the Ordinary Least Squares estimate is made through a Monte Carlo experiment. The control parameters used in the simulation study are:(1)A measure of the degree of multicollinearity inX,(2)The number of independent variables,(3)The variance of the error term(for unit length parameter vector #x3B2;), and(4)The most favorable and the least favorable choices of #x3B2;(as defined by McDonald and Galarneau). 100 replications are run on each combination of a 6Z#xD7;3#xD7;5#xD7;2 factorial design, using sample sizes ofn=100. The empirical mean squared error as well as some other properties of the sampling distributions of the squared lossare reported for the estimators considered.