Lab 08 / Relationships

Why extremes drift inward

What happens when we retest only the highest first scores?

7 min synthetic data fixed seed

Create two noisy synthetic measurements, select the most extreme first results, and compare the same selected group on retest. Change one control at a time, then open the values under the chart before interpreting the picture.

Interactive synthetic lab
runs in this browser

View the values behind this chart

Loading the deterministic simulation...

SEED

What this display answers

When two measurements are imperfectly correlated, an extreme first result usually combines a persistent component with unusually favorable or unfavorable noise. Selecting on that first extreme also selects on its noise.

On retest, new noise is unlikely to be equally extreme in the same direction. The selected group's average therefore moves toward the population center even when nothing intervenes.

What to notice

  1. Start with correlation 0.55 and select the highest 10%. Compare first and retest means.
  2. Raise correlation toward 0.95. More of the first ranking persists and the inward movement usually shrinks.
  3. Select a less extreme 30%. The first selected mean starts closer to the center.

The model behind it

Two standardized normal measurements have a controlled correlation and are scaled to center 50 with standard deviation 10.

Only cases with the highest first score are displayed. No treatment or event is applied between measurements.

Where the result stops

Regression to the mean is a group expectation, not a guarantee that every selected case moves inward.

A contemporaneous comparison group and a credible design are needed before attributing observed change to an intervention.