# Normal Distribution ![[mm-normal-distribution.svg]] *The bell curve: most values cluster near the mean, with the 68–95–99.7 rule giving the share of data falling within 1, 2, and 3 standard deviations.* **The model:** The bell-curve probability distribution that appears with remarkable frequency in statistics. Characterized by symmetric shape around the mean, with most data clustered near the center and progressively rarer values toward the tails. Defined fully by two parameters: the mean (where the peak is) and the standard deviation (how spread out the distribution is). Roughly 68% of data falls within one standard deviation of the mean; 95% within two; 99.7% within three. The normal distribution dominates introductory statistics, much of inferential testing, and much of practical risk management. It's also one of the most overused distributions in real-world analysis, because many phenomena that look approximately normal in normal conditions exhibit fat tails (extreme values much more common than normal predicts) under stress. The 2008 financial crisis was a high-profile case where normal-distribution-based risk models failed catastrophically. ## When It Applies The normal distribution shows up when many small independent factors combine additively to produce an outcome. Heights, measurement errors, IQ scores (by design), and sums of many independent random variables tend toward normality per the Central Limit Theorem. The mathematical machinery of much classical statistics (t-tests, ANOVA, regression, confidence intervals) assumes normality. ## When It Fails Many real-world phenomena are *not* normally distributed and don't approach normality at any sample size. Financial returns, social-network influence, war casualties, viral content spread, pandemic deaths, earthquake magnitudes, and city sizes all follow power laws or other fat-tailed distributions — extreme events are far more common than the normal distribution predicts. The political importance: many risk-management and policy decisions assume normality where it doesn't hold, producing systematic underestimation of tail risk. The "100-year flood" that arrives every 10 years; the "rare" financial crisis that hits every decade; the "unprecedented" political event that resembles past unprecedented events; all are symptoms of normal-distribution reasoning being misapplied to fat-tailed phenomena. ## KB Applications ### Polling and the Normal-Distribution Assumption Poll margins of error are computed under normal-distribution assumptions about sampling. The math is sound when samples are truly random and aggregated estimates approximate normality (per the CLT). When response rates differ systematically by demographic (which is now the norm in U.S. polling), the underlying distribution becomes non-normal in ways the standard math doesn't account for. Polling errors in 2016, 2020, 2022, and 2024 are partly normal-distribution-assumption failures. ### IQ Scores: A Designed Normal Distribution IQ scores are explicitly *designed* to be normally distributed with mean 100 and standard deviation 15 — psychometric tests are constructed and calibrated to produce this distribution. Whether the underlying construct is normally distributed is a separate question (probably not, in some ways); the score distribution is by design. The right-wing race-IQ literature (Murray, Lynn) often presents distribution arguments as if they reflected underlying reality, when the distribution is partly an artifact of test construction. ### Why Normal-Distribution Thinking Misleads About Wealth Wealth distribution is dramatically non-normal — it's power-law-distributed with extreme right-tail concentration. Median wealth is much lower than mean wealth (in the U.S., mean wealth is about 5x median wealth). Reasoning about "the average American's wealth" with normal-distribution intuitions produces wrong conclusions. The fractal nature of wealth concentration is one reason political debates about inequality often talk past each other — one side uses mean, the other median, and the difference is enormous. ## See Also - Central Limit Theorem — explains when and why normality emerges - [[Fractals]] — fat-tailed alternative for many phenomena - [[Black Swan]] — Taleb's critique of normal-distribution overreach