
Causation vs Correlation: Key Distinctions
Causation vs correlation: the five things a correlation can mean, Bradford Hill's criteria, and the hormone therapy reversal that proved it.
What was corrected
Article inflated the Framingham Heart Study's real publication count from the widely-reported 'over 3,000' peer-reviewed papers to an unverified 'over 4,000.' It described a specific 2019 Airbnb pricing-methodology publication and price-nudge experiment that could not be verified anywhere - no such publication exists in Airbnb's real public causal-inference work, which is dated 2016 and 2022 and describes different methods. It also understated the real Gordon/Zettelmeyer/Bhargava/Chapsky Facebook ad-measurement study's finding, describing observational ad-effectiveness estimates as overstated 'by an average of two to three times' when the real finding is a factor of three or more.
Framingham publication count corrected to the real, widely-cited 'over 3,000.' The unverified Airbnb case study was replaced with an accurate general description of the same real confounding problem (price correlating with demand in marketplace pricing) without inventing a specific unverifiable Airbnb publication or experiment. The Facebook ad-attribution figure corrected to 'a factor of three or more,' matching the real study's actual finding.
Why this is better
This article was substantially cleaner than others in this batch - independent verification found 3 of 6 major citations (Angrist's 1990 draft lottery paper, the Nurses' Health Study/Women's Health Initiative HRT reversal, and the Walmart Pop-Tarts anecdote) were fully accurate with correct figures, and only one citation (Airbnb) was outright unverified rather than merely imprecise.