Visual Inference for Linear Mixed-Effects Models

• Activity Number: 250
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Graphics
• Author(s): Adam Loy*+ and Heike Hofmann
• Companies: Lawrence University and Iowa State University
• Keywords: Diagnostics ; Residuals ; Statistical graphics ; Nonparametric test ; Visualization

Abstract

Valid model-based statistical inference relies on proper specification of a model, making diagnostic tools central to analysis. Graphical methods are commonly used for checking the assumptions of a model; however, they are often criticized as being too subjective since decisions are based on one plot. This has lead to the reliance on conventional hypothesis tests to make formal decisions. Recently, visual inference has been established, providing a rigorous framework for graphical discovery that allows for the quantification of strength of evidence. In this paper we use visual inference as a framework to overcome common difficulties with conventional hypothesis tests and statistical graphics that are encountered in the selection and validation of linear mixed-effects models.