Regression diagnostics
About this app
Residual plots, leverage and Cook's distance on the lecture datasets and Anscombe's quartet – one wrong measurement tilts the line.
The statistics
Regression diagnostics check whether a straight line is a sensible model of the data and whether individual points dictate it. The coefficient of determination R² answers neither question. The residual eᵢ = yᵢ − ŷᵢ is the vertical distance to the line, and the standard error is s² = Σeᵢ²/(n − 2). The leverage hᵢ = 1/n + (xᵢ − x̄)²/Σ(xⱼ − x̄)² depends only on x and is large for points at the edge. Cook’s distance Dᵢ combines residual and leverage and measures how much all predictions change when point i is left out.
What you see
The scatter plot shows data points and the regression line, each point with a halo proportional to its Cook’s distance. Below sits a residual plot with optional smoothing. Data sets include lecture examples (a DC motor, the same data with a typo, mortgage rates and construction orders), the four Anscombe sets, a funnel with growing spread and an empty canvas for your own points. A diagnosis panel reports R², s and n and checks influence, curvature and funnel shape. For a clicked point the app works out leverage and Cook’s distance with the numbers filled in.
Things to try
Load the motor data and then the version with a typo, where 7 instead of 70 hp was recorded: a single point tilts the line and its Cook’s distance spikes. Switch on the line without that point to see its influence directly. Compare Anscombe II and IV: the same R², but one shows curvature in the residual plot and the other a point with leverage h = 1. In the funnel data set the spread grows with x. The tools let you drag, place and delete points.
Frequently asked questions
- What is Cook’s distance?
- Cook’s distance measures the influence of a single data point on a regression. It compares the predictions with and without that point, combining residual and leverage. As a rough guide, values above 0.5 are worth a look and values above 1 indicate strong influence.
- What is leverage in regression?
- Leverage hᵢ shows how far a point lies from the mean x̄ in the x direction, and therefore how strongly it can pull the line towards itself. It depends only on the x values, not on y. A point with high leverage and a large residual has a large influence.
- Why is R² not enough to judge a regression?
- R² only states what share of the variation in y the line explains. It shows neither whether a linear relationship fits at all nor whether single points drive the result. Anscombe’s quartet has almost the same R² four times over with completely different data; residual plots and influence measures reveal the difference.
- Which data sets are in the regression diagnostics app?
- It includes a DC motor (power against speed), the same data with a decimal error, mortgage rates and construction orders, the four Anscombe sets, a simulated funnel with heteroscedasticity and an empty canvas for your own points.
Subject: Statistics
More from Statistics
- Anscombe's quartet
- Bias-variance / overfitting
- Box plot workshop
- Geometric mean
- Law of large numbers
- Histogram workshop
- Hypothesis test & p-value
- Confidence interval
A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps