By Ross Metusalem (JMP Statistical Discovery LLC)
Information
A regression model can be more useful to students when they possess an intuitive understanding of the trends or relationships that the model has captured. In simple regression, that intuition is typically fostered with a simple visualization: a scatterplot overlaid with the regression line. This visualization is so ubiquitous that it’s easy to take it for granted, yet its very ubiquity underscores how important it is to provide students with visualizations that help them develop more intuitive understandings of statistical models.
But beyond simple regression, what about multiple regression models with interactions or other higher-order terms? Thankfully, useful visualizations exist for fostering intuitive understanding of these more complex models, too.
This Beyond session will present model visualization techniques suitable for teaching multiple regression or other intermediate-to-advanced modeling techniques: interaction traces, surface plots, contour plots, and marginal model plots. For each, a specific example will be presented along with discussion prompts that ask students to use the visuals to intuit what relationships the model has captured, as well as exercises that foreground the correspondences between multiple visualizations of the same model to foster higher-level understandings that are independent of any one visualization. These discussion exercises would be appropriate at the intermediate undergraduate level and above and could be applied in classes of any size.