Big ideas from big samples.
Every year, some brewery walks away with a medal and a marketing budget built on the phrase 'Best Beer in America.' Here's what those competitions actually measure — and why the honest answer is 'a handful of thirsty judges on one afternoon.'
Open-label placebo trials keep producing a result that shouldn't exist: patients get better from pills they know are empty. Here's why that finding is messier than the headlines suggest — and what it exposes about the sacred architecture of the randomized controlled trial.
Sugar pills keep winning, and it's not because they're magical. It's because the bar they're measured against — the humble placebo response — has been quietly climbing for decades. Here's what that means for every 'breakthrough drug' headline you'll read this year.
That 4.9-star taco truck with a dozen reviews and the 3.8-star diner with three thousand are telling you less than you think. Here's how sample size, Bayesian smoothing, and a little review gaming decide where you eat lunch.
Every summer, a record-breaking heatwave or a freak blizzard gets trotted out as proof of whatever someone already believed about climate. Here's why one storm settles nothing — and how scientists stack thousands of weather stations and a century of data to find a signal that refuses to hide.
Your worst acne flare-up was always going to get better on its own — but the supplement you started that week gets the credit. Welcome to regression to the mean, the invisible force behind millions of glowing testimonials.
The math was fine. The arithmetic checked out. What failed was the question itself — and when America asks the wrong question, the answer comes back with a ten-figure price tag. Five cases where measurement design, not calculation, steered fortunes.
The 1936 Literary Digest poll gathered 2.4 million responses and still called the election wrong. The problem was never the size of the sample — it was asking the wrong Americans. Five polling disasters that prove a big sample of the wrong people is just a confident mistake.
Your sleep app says coffee is wrecking your nights, and honestly, the chart looks convincing. But there's a difference between the data of you and the science of many — and knowing which one you're holding changes everything.
'Top 10% hospital.' 'Top-rated employer.' 'Above-average school district.' America runs on ranking math, and most of it is a magic trick performed with small groups and selective slicing. A field guide to spotting the racket.
Your DNA test says you're mostly Irish. The confidence interval says it could say almost anything. Here's why ethnicity estimates are the flashiest, least reliable numbers in consumer science.
Every ten years, America counts itself — and every ten years, the count comes up short somewhere. From the 1990 undercount to the 2020 fiasco, here's what happens when millions of people vanish from the ledger.
'No significant difference found' sounds like a verdict. Often, it's just an alibi. Here's why an underpowered clinical trial can burn through billions and still tell us absolutely nothing.
Follow a hypothetical American through a single day — waking, scrolling, streaming, shopping — and count the experiments running on her without her knowledge. The A/B testing industry is the largest behavioral science program in human history. Nobody signed a consent form.
Every November, millions of Americans watch a candidate's 'lead' evaporate and conclude the polls were broken. They weren't. Here's what margin of error actually means, why 'up by two' can mean nothing at all, and how to read a poll like a data scientist instead of a doomscroller.
The fate of a drug, an election, and even a war has sometimes hinged on a sample so small it would fail an intro stats class. Five American stories where the n was tiny — and the stakes were anything but.
Ask a college grad how many people you need for a 'real' study and you'll hear the same number: 30. Here's how a piece of 20th-century textbook shorthand became one of the most stubborn misconceptions in American classrooms — and what actually determines whether a sample is big enough.