Beyond Diagnostics: Tools for Consideration of Least-Squares Influential Data

• In Event: AERA Poster Session Sunday 1:45 pm
  In Poster Session: Methodological Techniques for Challenging Data and Modeling Situations

Abstract

The purpose of this paper is to suggest methods and software to more adequately consider influential data in least-squares regression. Although influence diagnostics in statistical packages are produced for each influential case, the standing herein is that one must first examine whether influential cases substantially impact the model in respect to each case and all cases, and that this must be done regarding the data context and goals of prediction. Thus, the impact on relative (r, r2, R2) and absolute (√MSE and MAE) accuracy, as well as on regression weights, is considered and automatically computed with the software offered. Some examples are provided.

Authors

• John D. Morris, Florida Atlantic University

• Mary G. Lieberman, Florida Atlantic University

• Maria D. Vásquez-Colina, Florida Atlantic University