Law of Large Numbers: More Data, Truer Average

THE MENTAL MODEL: The more times you repeat an experiment, the closer the observed average will get to the true, underlying average. A handful of coin flips might result in 80% heads by pure chance, but after 10,000 flips, the result will be extremely close to 50% heads. The law guarantees that a large enough sample will reflect the true reality.

HOW IT WORKS: The law relies on two conditions: the trials must be independent (one coin flip doesn't affect the next) and identically distributed (you're using the same coin each time). With each new sample, the total average is recalculated. While any single result can be random and noisy, its impact on the overall average shrinks as the number of samples (N) grows. The random noise cancels itself out, and the sample mean converges toward the true mean, or expected value.

WHEN TO USE IT: This principle is the bedrock of many data-driven fields. First, in A/B testing, it allows you to trust that a 2% lift in conversions is a real improvement and not just random noise, provided you've collected enough data. Second, in insurance, companies use vast amounts of historical data to accurately predict the average number of claims in a year, allowing them to set premiums that guarantee profitability. Third, in quality assurance, a factory tests a large sample of products to reliably estimate the defect rate of the entire production batch.

WHEN NOT TO USE IT: The law is not a magic bullet. It does not apply to small sample sizes; making a business decision on ten data points is dangerous because the average is not yet stable. It also fails when events are not independent. For example, you cannot use it to predict tomorrow's stock price by averaging past prices, because market events have memory and influence each other. Finally, it assumes the underlying probability is constant. If the system you're measuring changes mid-experiment, the law no longer applies.

ONE CANONICAL EXAMPLE: A casino's business model is a direct application of the Law of Large Numbers. A single spin on an American roulette wheel is unpredictable. But the casino knows that for a 1 bet on red, the player has an 18/38 chance to win 1 and a 20/38 chance to lose 1. The expected value is -0.0526. Over millions of spins at thousands of tables, the casino can be statistically certain that it will earn an average of 5.26 cents for every dollar bet, locking in its house edge.