Clearer Thinking with Spencer Greenberg Should science stop worshiping statistical significance? (with Andrew Gelman)
50 snips
Mar 5, 2026 Andrew Gelman, Higgins Professor of Statistics and Political Science at Columbia, is a leading critic of sloppy statistical practice. He discusses why flawed studies persist, the limits of p-values, the need for better measurement and priors, how replication and criticism should work in science, and why Bayesian approaches and transparency can improve reliability.
AI Snips
Chapters
Transcript
Episode notes
Plot Data And Admit Uncertainty
- When running experiments, plot raw data and embrace uncertainty instead of ritualistically converting everything into p-values to claim certainty.
- Gelman suggests graphs often convey the result better and encourages honest statements like 'this looks promising but I'm uncertain.'
P-Values Exist Partly To Deter Noisy Studies
- P-values serve a social role: they reduce moral hazard by setting a publication bar, discouraging intentionally tiny, noisy studies geared to produce occasional positive findings.
- Gelman argues removing p-values without a replacement could incentivize low-quality noisy experiments.
Fragile Findings Can Imply Cynical Politics
- Some unreplicable research often assumes small manipulations can strongly alter behavior, reflecting a politically cynical view that voters and citizens are easily swayed.
- Gelman notes such findings, if taken literally, would undermine trust in democratic behavior and have anti-democratic implications.