America has no hereditary aristocracy, so its exclusions get called merit instead. The data on legacy admissions and resume audits says the sorting is just as real.
Uplift modeling is supposed to find people who convert because of an ad, not regardless of it. I tested the standard version on 14 million ad impressions from a real randomized experiment, and it didn't clearly beat a much simpler model.
Same-sex marriage keeps climbing while opposite-sex marriage keeps falling, which is a problem for the leading story about why Americans are pairing off less.
Prediction-market theory says more bettors should make prices more accurate. Across 73 Polymarket World Cup matches, the highest-volume games were the worst calibrated, and trader count didn't matter at all.
A striker's hot scoring streak almost never predicts the next one. I tested it across 5,498 player-seasons in Europe's top leagues.
The education gap between men and women didn't reduce college women's marriage rates. It shifted who was available to non-college women, through two distinct channels.
Analytics teams abandoned the mid-range jumper. The teams that didn't started hitting it at record rates. That pattern has a plausible explanation, and it comes from a paper about A/B testing.
PGIM manages $1.3 trillion. Their mid-year outlook tells you exactly what to buy. Read it carefully and you'll notice who isn't in the room.
The standard approach to measuring ad effectiveness is systematically biased toward false positives. A new paper from Meta Research and Northwestern trains on 2,226 real experiments and builds a better method.
The labour share of US GDP just hit a 70-year low. The same force compressing wages is inflating the assets workers need to buy to build wealth.
In 2023, average total income for Canadian men 65 and over exceeded average total income for men 25-34. The gap widened in 2024. The data is from Statistics Canada.