Quantitative Methods

Regression to the Mean and the Illusion of a Turnaround

Compute how much a selected group will improve on its own, so you stop paying for turnarounds that arithmetic was going to deliver anyway.

  • Advanced
  • 11 min total
  • 13 chapters

What decision this helps you make: Whether to credit an intervention you applied to your worst performers, and whether to roll a coaching programme, performance plan, or turnaround out across the company.

What this topic is

Regression to the mean is what happens when you select a group because its measurement was extreme and then measure it again. Because any real measure contains both signal and noise, the extreme group was extreme partly through luck, and the luck does not repeat. The group moves back toward the average with no intervention at all, and the amount it moves is calculable in advance.

Why it matters

Businesses systematically apply interventions to selected extremes: the bottom decile of reps, the worst stores, the accounts with the lowest satisfaction, the region that had a terrible quarter. Every one of those groups improves next period whether or not the intervention does anything. The improvement is then attributed to the intervention, the intervention is rolled out at scale, and the company spends real money on a programme whose measured effect was arithmetic.

Who should learn it

Sales leaders running performance plans, multi-unit operators launching turnarounds, HR teams evaluating coaching, and acquirers pricing a business off its best year.

What you will understand

  • The formula for how far a selected group will move back, and how to get its one input from your own data
  • Why the same effect makes praise look harmful and criticism look effective
  • How to design an evaluation that separates a real effect from the regression it sits on top of
  • When the extremes are genuinely real, and refusing to act on them is its own mistake

Prerequisites

Common misconception

"Six of the eight reps we put on the plan improved, so the plan works." Six of eight would have improved without the plan. Selecting the bottom eight of forty selects reps who are genuinely weaker AND reps who had a bad quarter, and the second group bounces back on its own. With a quarter-to-quarter correlation of 0.5 and a group sitting 35 points below the mean, the expected recovery with no intervention is 17.5 points. Any programme applied to that group inherits 17.5 points of free credit, and any programme applied to your best performers inherits an equivalent free blame.