Understanding Regression Diagnostics: An Online Tool
I've heard people refer to linear regression as the bread and butter of data modeling. If that's true, and if we're going to stick with the food metaphor, then performing regression diagnostics is like eating your vegetables. That's because the vegetables (diagnostics) are the part of the meal (analysis process) that assures that your bones (underlying assumptions) are strong and healthy enough to support all the running, jumping, and playing (prediction/explanation) that's going to happen when you push away from the table.
The bottom line is that the health of a regression analysis depends heavily on the validity of the diagnostic assumptions. So I thought it might be useful to build a tool to help those who are new to regression learn more about how the diagnostics work. Using R and Shiny, I built an online app that lets users choose various data sets to analyze, with different tabs that show how the diagnostics, and their associated plots, apply to the chosen data set. Users can also add data points, to see the impact of the new points on the regression analysis and the diagnostics. The tool includes brief explanations of the diagnostics and their interpretations. Currently, the data sets are taken from the well known Anscombe's Quartet, but I hope to add new data sets, each demonstrating various aspects of the diagnostics, soon.
Below you'll find a screen shot from a section of the tool. Worth noting is that some additional upcoming features include a) the ability to upload your own data sets to analyze, and b) discussion of features (like multicolinearity) that are particular to multivariate regression. If you have any suggestions for improvement in the meantime, I'd love to hear them. But right now, having spent some time on broccoli and carrots, I'm going to go get some dessert. 😉
You can find the tool at https://brettamdur.shinyapps.io/regDiagnostics/. Happy Diagnosing!