| Session Title: Evaluating Regression Models: Understanding and Using Regression Diagnostics to Improve your Analyses |
| Demonstration Session 872 to be held in Centennial Section B on Saturday, Nov 8, 1:20 PM to 2:50 PM |
| Sponsored by the Quantitative Methods: Theory and Design TIG |
| Presenter(s): |
| Charles Collins, Michigan State University, colli43@msu.edu |
| Steven Pierce, Michigan State University, pierces1@msu.edu |
| Abstract: Regression models are frequently used to analyze evaluation data. Examining whether the underlying statistical assumptions are met is a critical task for evaluators employing this technique. Using data from a community change initiative, we will demonstrate how to use regression diagnostics to diagnose and solve problems with regression models. This session will introduce the audience to a variety of graphical and statistical tools for performing regression diagnostics, discuss how to interpret and use the resulting output to detect problems such as non-linearity and non-normality, and how to identify influential and outlying data points that may distort analysis results. Diagnostics allow evaluators to understand in what way the data violate the assumptions of regression models, which can then guide decisions about how to fix the violations and improve the analysis, thereby yielding more accurate and valid conclusions. A complete example from a real evaluation study will be presented. |