Power of Visual Statistical Inference in a Non-Normal Scenario

• Activity Number: 175
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Graphics
• Author(s): Mahbubul Majumder*+ and Dianne H Cook and Heike Hofmann
• Companies: Iowa State University and Iowa State University and Iowa State University
• Keywords: Visual Inference ; Statistical Graphics ; Inference ; Graphics ; Data Visualization

Abstract

Statistical graphics play a crucial role in exploratory data analysis, model checking and diagnosis. Recently Buja et al. (2009) introduced the lineup protocol as a means to test the significance of visual findings. Majumder et al. (2011) take this a step further by comparing the lineup protocol against classical statistical testing of the significance of regression model parameters where they demonstrated that visual statistical inference can yield power as good as the power of uniformly most powerful (UMP) tests. To examine that, they consider a worse case scenario for visual inference such as a linear regression model with normal error. In this paper, we examine the visual inference technique for linear regression models with non normal error and our results indicate that visual inference techniques manage to dramatically outperform normal tests.