Differential Weighting for Prediction and Decision Making Studies: A Study of Ridge Regression.

摘要

This paper is another in a series exploring the conditions under which either differential or simple unit weighting of predictor variables in prediction and/or decision studies will be appropriate. Some of the difficulties of applying the ordinary least squares (OLS) regression analysis to practical problems are described and an alternative regression model called ridge analysis (RIDGE) is offered as a substitute to OLS. Several empirical studies were conducted using computer simulated data for various prediction situations. The OLS and RIDGE models were compared as to their efficacy in prediction and both models were compared against the simplest model possible, that of unit weighting (UNIT), in which no weighting is performed; the variables are simply added up and the sum used for prediction. The results of these studies indicate that OLS and RIDGE, with one exception, always outperformed UNIT with respect to producing smaller errors of prediction and, what is more important, RIDGE always did better than OLS.