Skip to content

Correlation & Regression Lab

Drag points. Every (x − x̄)(y − ȳ) rectangle is one term of the covariance; blue rectangles push r up, red ones pull it down. The line minimises the sum of the orange squares.

025507510002468101214x̄ = 8.08ȳ = 62.65ŷ = 21.46 + 5.10x
Scatter: n = 20; r = 0.910, b = 5.10, R² = 0.828
Points (x, y, residual)
1e = -8.2
2e = 1.6
3e = -4.5
4e = -13.1
5e = 7.9
6e = 10.5
7e = 7.2
8e = 6.6
9e = -5.8
10e = -1.4
11e = -3.9
12e = 12.2
13e = -0.5
14e = -1.1
15e = 10.6
16e = 7.1
17e = -17.7
18e = -0.3
19e = -9.2
20e = 1.8

Pearson r and its t-test, Spearman ρ on average ranks, OLS by normal equations with classical SEs (statsmodels-equivalent), logistic by IRLS. Synthetic data, seeded; the 'outlier' preset plants one high-leverage point by hand.

r = 0.910, b = 5.10, R² = 0.828

Formula

r = Σ(xᵢ−x̄)(yᵢ−ȳ) / √(Σ(xᵢ−x̄)² Σ(yᵢ−ȳ)²)b = r · s_y/s_xa = ȳ − b x̄t = r√(n−2)/√(1−r²)

One far point in the corner can dominate both r and b: leverage grows with (x − x̄)². Spearman only sees ranks, so it barely moves.

Method: Pearson product-moment r; t = r√(n−2)/√(1−r²); Fisher-z CI; Pearson r on average ranks (ties averaged); p via t approximation (SPSS) · Synthetic, seeded data. Illustrative, not an empirical finding.

Validation: Pearson, Spearman, OLS and logit match the SciPy/statsmodels reference (tests/unit/engine.test.ts) · numerically validated