Category Profiling and Unusual Category Detection Based on Estimated Marginal Means (EMMEANS)

• Activity Number: 81
• Type: Contributed
• Date/Time: Sunday, July 31, 2011 : 4:00 PM to 5:50 PM
• Sponsor: Social Statistics Section
• Author(s): Jing Shyr and Sier Han and Jane Chu*+
• Companies: IBM Research and IBM Research and IBM Research
• Keywords: EMMEANS ; Analysis of variance ; Unusual category

Abstract

Analysis of variance is a frequently-used modeling technique for identifying significant factors or factor interactions. The task of understanding how a significant factor or interaction impacts the target variable usually relies on interpretation of the parameter estimates. In this paper we show how to use EMMEANS to understand the impact of a significant factor or interaction on the target. EMMEANS are model-based predicted means that are computed after the model is built. A significant factor or interaction implies its categories have different impacts on the target. We propose a grouping analysis to highlight those different impacts. In addition, we propose an outlier detection method to detect categories with unusually high impacts on the target. One simulated dataset and one real dataset are used to demonstrate usefulness of the two proposed methods.