A Comparison of Repeated Measures ANOVA and Mixed-Effects Models in Analyzing Repeated Measures Data

• Activity Number: 556
• Type: Contributed
• Date/Time: Wednesday, August 3, 2016 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Allison M. Butler* and Gregory L. Snow and Bill McDermott and Jeffrey Ferraro and Kyle Hiroyasu
• Companies: Intermountain Healthcare and Intermountain Healthcare and The Orthopedic Specialty Hospital and Homer Warner Center for Informatics Research and Homer Warner Center for Informatics Research
• Keywords: Repeated measures ANOVA ; mixed effects model ; simulation

Abstract

Repeated measures data is common in many fields, particularly medical research. The predominately accepted method of analysis across disciplines for such data is repeated measures ANOVA; however, mixed effects modeling offers many advantages over repeated measures ANOVA, the most substantial being its ability to handle missing data. Simulated data were based on a common biomechanical study where peak vertical ground reaction force is measured as a subject jumps from increasing heights. Simulations were conducted with varying fixed and random effects and varying amounts of missingness. As expected, when no missing data were present, repeated measures ANOVA and mixed effects models produced equivalent estimates and p-values. However, as the amount of missing data increased mixed effect models produced more precise and unbiased estimates, especially for the random effects.