Click the canvas to add a data point. Drag points to move them. Watch r, r², and the regression line update live — then see how one well-placed outlier can wreck (or fake) a correlation.
Presets:
0
n
—
r
—
r² (variance explained)
—
Ŷ = a + bX
Try this:
Load the Outlier demo: a cloud with r ≈ 0 plus one point in the corner. Check r — then drag that one point around and watch a single participant manufacture a "strong correlation." Now imagine it's a data-entry typo.
Load Curvilinear. The relationship is obviously strong and systematic — but r is near zero. Why? What does r actually measure?
Load Restricted range: it's the strong preset, but you only sampled the middle third of X (like studying the SAT–GPA link using only Ivy admits). Compare r to the full-range version.
Turn on Residuals and drag a point straight up: which moves more, the line or r?
Sigma says:r only speaks LINEAR. It's also a terrible gossip — one extreme point and it changes its whole story. Always, always plot your data before you trust a correlation.