Stat Myth Hunt

Stat Myth Hunt is a hypothesis testing game built around one question that governs every scientific study, clinical trial, and A/B test ever run: if nothing was actually happening, how surprising would this data be? A scenario arrives. A casino claims their coin-flip machine is perfectly fair — but in their public demonstration, they flipped 62 heads out of 100. A coffee company claims their blend boosts alertness — they tested 30 customers, who averaged 0.4 points above the population mean. An app team claims their redesign increased sign-ups — old design converted 10 percent, new design converted 14 percent, each tested on a thousand users. The player must decide which of these claims are backed by real evidence and which could simply be explained by chance. The decision is made through a bell curve. Every scenario produces a test statistic — a score summarising how extreme the data is relative to what chance alone would produce. A glowing ball slides from the centre of the curve to the test statistic's position. The outer five percent of the distribution is shaded in red — the zone where results are too surprising to dismiss as luck. If the ball lands there, the data is statistically significant. If it stops in the safe centre zone, the result could easily be a random fluctuation. The player sees this directly, then taps their verdict. Six cases span three difficulty levels. Easy cases are clear: sixty-two heads in a hundred flips is obvious, and a 0.4-point boost across thirty people is obvious in the other direction. Medium cases introduce two-group comparisons and reveal how sample size changes everything — the same four percentage point difference that means nothing across thirty users means a great deal across a thousand. Hard cases are designed to be wrong in both directions: a drug that reduces blood pressure by twice the placebo amount but still fails to reach significance because the variability is high, and a diet comparison that lands at 5.7 percent — just above the threshold — when the claim was certainty. After each verdict, the p-value is shown in plain language. A result at 1.6 percent means the data would appear by chance one time in sixty-two. A result at 14.4 percent means it would appear one time in seven. The explanation panel shows what drove the outcome — whether the sample was too small, the variability too high, or the effect simply too large to explain away. The game never shows a formula.