average treatment effect
The mean effect of an intervention across a whole study population, which can mask wide variation in how individuals respond.
As defined in 165 | Kathryn Paige Harden on Genetics, Luck, and Fairness
Where it comes up · 2
- Andrew Gelman on Social Science, Small Samples, and the Garden of the Forking Paths2 claims
- 165 | Kathryn Paige Harden on Genetics, Luck, and Fairness1 claim
Most important claims · 3 of 3Search all claims for “average treatment effect” →
- Rather than asking whether an intervention 'works,' one should recognize that essentially everything has some effect; the meaningful questions are how large the effect is, for whom it is positive or negative, and in what specific context—since reported average effects mask heterogeneous individual effects.Andrew Gelman on Social Science, Small Samples, and the Garden of the Forking Paths
- Priming effects are real in the sense that everything has some effect, but they are so highly variable across people and contexts—sometimes slowing, sometimes speeding, sometimes irrelevant—that the notion of a single consistent 'priming effect' is essentially meaningless; the reported average effect is merely the average of heterogeneous local effects.Andrew Gelman on Social Science, Small Samples, and the Garden of the Forking Paths
- Nearly all randomized controlled trials of psychological or educational interventions show vast differences between people in how they respond, and small average treatment effects can mask both huge benefits for some and none for others; often a 'Matthew effect' occurs where the people least at risk (already doing well, e.g., rich kids learning more from Sesame Street) benefit most, and genetics adds an invariant layer to see who is actually being served by current interventions.165 | Kathryn Paige Harden on Genetics, Luck, and Fairness