Abstract:
|
A Monte Carlo study was used to compare the ability of the univariate variable selection criteria that are available in either SAS or SPSS to select the "correct" multiple linear regression model. The performance of the univariate variable selection procedures was investigated for different sample sizes, intercorrelations, and interacorrelations. The results suggested that all procedures' ability to select the "correct" model improved by more than 50% when sample size increases from 25 to 50 or 100, and multicollinearity decreased the ability of all selection criteria to select the "correct" model, whereas the high correlation between the dependent variable (y) and independent variables (X) improved the ability of all procedures to select the correct model. The backward elimination criterion and Schwarz's Bayes information procedure are the best among the automatic search criteria and all-possible-regression procedures, respectively.
|