What Would Fisher Do? A Framework for Promoting a Rich Understanding of Generalized and Mixed Model Construction

• Activity Number: 143
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Education
• Author(s): Julie Couton* and Walter Stroup
• Companies: University of Nebraska - Lincoln and University of Nebraska - Lincoln
• Keywords: Teaching Tool ; GLMM ; ANOVA ; Study Design ; Fixed/Random Model Effects ; Probability Distributions

Abstract

Stroup (2013) in his GLMM textbook commented, "Most modeling problems are really failure to understand design problems in disguise." Students in introductory design courses struggle converting study descriptions to plausible statistical models. Consultants struggle determining which model to use for data analysis and how to explain model rationale to their clients. This presentation demonstrates a tool to assist teachers of linear models, statistical scientists planning modeling research, and consultants. Inspired by R.A. Fisher's comments following Yates (1935) "Complex Experiments," What Would Fisher Do (WWFD) is a technique that organizes elements associated with the design structure, treatment structure, and re-envisions the resulting ANOVA table as the basis for constructing plausible GLMMs. The WWFD framework aids in understanding fixed/random effects, the impact of the response variable distribution on the model, and the role of the residual term in the context of different distributions. Finally WWFD provides a basis for planning simulation studies.