Home▸Data Science▸Range Restriction: Why Selected Groups Show Weak Correlations (Thorndike Correction)

Range Restriction: Why Selected Groups Show Weak Correlations (Thorndike Correction)

Admit only the top scorers and the test seems to predict nothing. Slide the selected fraction, the true correlation and whether selection is made on the predictor or on the outcome, and watch the correlation among the selected collapse while the regression line stays put or tilts. Rejected points sit behind in grey, with the truncated-normal variance, the Thorndike correction and Monte Carlo checks built in.

Data Science3DModerate60 FPS📱 Mobile-adapted🟠 Blender⇄ 2D version
Range Restriction: Why Selected Groups Show Weak Correlations (Thorndike Correction) ↗ Open standalone

Fit the best non-decreasing curve through noisy points without any smoothing parameter. Watch the pool-adjacent-violators algorithm merge neighbouring points into flat blocks, compare the staircase with the raw data and a straight line, and break the monotone assumption with a wavy truth. Checks verify the min-max formula, the optimality conditions and the projection identity.

⚙ Under the hood

Admit only the top scorers and the test seems to predict nothing. Slide the selected fraction, the true correlation and whether selection is made on the predictor or on the outcome, and watch the correlation among the selected collapse while the regression line stays put or tilts. Rejected points sit behind in grey, with the truncated-normal variance, the Thorndike correction and Monte Carlo checks built in.

Isotonic RegressionPool Adjacent ViolatorsMonotone ConstraintConvex ProjectionNonparametric RegressionRange RestrictionSelection BiasThorndike CorrectionTruncated NormalCorrelation Attenuation

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install

What did you find?

Add reproduction steps (optional)