Regression Playground

Fit and diagnose regression models interactively: adjust predictors, interactions, and transformations with live output

Purpose

Students of regression rarely see what happens to the fitted line and the residual diagnostics when a single observation is moved, a predictor is log-transformed, or a high-leverage point is removed. The Regression Playground makes every such change instantaneous and visible.

User inputs

• Dataset (built-in or uploaded)
• Outcome and predictor(s) selection
• Model family: OLS, logistic, Poisson, negative binomial
• Transformations: log, square, polynomial, spline (df selector)
• Interaction toggles between pairs of predictors
• Subset filters: exclude rows by leverage, residual, or user click

Outputs

• Model summary table with coefficients, SEs, t/z, p-values, and confidence intervals
• Fitted plot with confidence and prediction bands
• Residual-vs-fitted, Q-Q, scale-location, and leverage plots
• Marginal effects plot (via ggeffects) for any predictor
• The R code to reproduce the current fit
• A “model comparison” panel for nested-model likelihood-ratio tests and AIC/BIC

Didactic value

The app teaches what goes wrong, not just what goes right: the visible effect of a single high-leverage point on the slope estimate; the way residual patterns betray a missed non-linearity; the inflation of standard errors when two predictors are highly collinear. These phenomena are the substance of a regression course but rarely experienced hands-on.

Embedded in

• regression-modelling/simple-linear-regression.md
• regression-modelling/diagnostics-and-assumptions.md
• regression-modelling/interactions.md

Source code

Local: apps/07-regression-playground/

Run with:

shiny::runApp("apps/07-regression-playground")