Fake-Data Simulation Catches the Flaws That P-Values Hide
We do modern frequentist statistics: Using fake-data simulation
Andrew Gelman revisits the notorious claim that beautiful parents are 36% more likely to have girls, a result Freakonomics swallowed whole. The p-value looked convincing, but forking paths and an implausible effect size gave it away. Gelman shows how simulating thousands of fake datasets under a null model exposes such patterns as noise — no math or subject expertise required. He argues this is what frequentism really means: treating your data as one draw from a distribution of possible realizations.
The wonderful thing about these simulation experiments is that they can reveal problems with a naive design, even if you didn't anticipate any difficulties ahead of time.
- wodenokoto
An article named “This is how we do modern frequentist statistics” is an excerpt from a book called “Bayesian workflows”
Where is the article explaining that?
Anyway, great article, thanks for sharing.
- j7ake
This should be automatic now in any statistical analysis given ubiquity of coding agents.