This paper is from Session 9F: Using data visualization to build statistical thinking
which comes under Topic 9: Technology and multimedia in statistics education
Paper 9F2 (Friday 13th, 14:00-15:30)
How much do model violations affect regression results: a demonstration using Shiny
Presenter
- John Walker (California Polytechnic State University, United States)
Co-author
- Harry Wu (California Polytechnic State University, United States)
Abstract
Checking the model assumptions is an important part of a regression analysis. However, other than knowing that model violations are bad and how to detect them, students often learn little about the degree to which the regression results are affected. This may leave the incorrect impression that even modest violations will ruin a regression. What negative effects should we expect if these assumptions are violated, and how are these negative effects modified by sample size and the strength of the regression association? Through the use of Shiny apps, students can see for themselves the effects of different levels of non-normality and unequal variance on the widths of confidence intervals and Type I and Type II error rates in regression.