46.063 - Multiple Regression: Principles and Practices

Session Type: Symposium

Abstract

In practice, there are several regression models which are overlooked or not understood. Curve estimation is possible when fitting a regression line; regression models do not have to be linear. Another modeling issue is predictor variable selection in the presence of several predictors, multicollinearity, and suppressor variable effects; how do we decide? The dependent variable also does not have to be a continuous measure. Logistic regression permits probability estimates for a dichotomous dependent variable given a set of weighted predictor variables. Ordinal regression permits ranked values for the dependent variable. Loglinear regression models permit nested chi-square tests in lieu of running numerous individual chi-square tests. This symposia will therefore present modeling principles and practical applications of these different regression models.

Sub Unit

• SIG-Multiple Linear Regression: The General Linear Model

Chair

• Randall E. Schumacker, The University of Alabama

Papers

• Dangerous Curves Ahead: A Practical Approach to Curve Estimation in Regression - Lisa Raymon Besnoy, The University of Alabama

• Multiple Regression Using R: A Comparison of All Possible Subset Regression, Beta Weights, Structure Coefficients, and Commonality Analysis - Brigette Winkles, The University of Alabama

• Logistic Regression and the Chi-Square Test Statistic - Alan Luther Webb, The University of Alabama; Robin Harvey, The University of Alabama

• Ordinal Regression Analysis in Educational Research - Erin O'Connor, University of Alabama

• Practical Application of Log-Linear Analysis - Sijia Zhang, The University of Alabama