Glossary · Term

effect size

← all terms

Definition

Plain language

A number that says not just whether a difference is real, but how big it is.

As stated in the literature

A standardized measure of the magnitude of an observed difference or relationship (e.g., Cohen's d, correlation coefficient), independent of sample size; distinguishes practically meaningful effects from statistically detectable but trivial ones.

Also called: effect sizes

Why it matters: It stops people from getting excited about differences that are technically detectable but too small to make any practical difference.

For example, a new tutoring method might raise test scores by a tiny sliver that shows up in a huge study, and the effect size tells you that sliver is too small to bother with in a real classroom.

Heard on the show

“It documented two of its own instrumentation bugs with before-and-after numbers, revised its effect sizes downward, and addressed a published result that points the other way.”
Episode 233 — Why a Model Can Grade an Answer But Not Write the Answer Key

Mentioned in 7 episodes

  1. 233
    Why a Model Can Grade an Answer But Not Write the Answer Key
  2. 196
    AI Agents Reached Opposite Conclusions From the Same Data — and Passed Review
  3. 110
    How an Agent Got 44 Points Better by Mining Its Own Scratch Paper
  4. 108
    The Reasoning Cliff: Why Thinking Longer Makes Models Worse at Exact Step-by-Step Tasks
  5. 095
    Seven Wins to Zero: How Organizing AI Agents Like a Lab Changes the Search
  6. 070
    When Models Know the Answer But Say the Wrong Thing Anyway
  7. 011
    When RL Actually Teaches Agents Something New, And When It Doesn't

Related terms