Central Limit Theorem

The central limit theorem (CLT) is a foundational principle in statistics stating that the sampling distribution of the sample mean approaches a normal (Gaussian) distribution as the sample size increases, regardless of the shape of the underlying population distribution.^[1][3] This convergence holds for any population with a finite variance, provided the observations are independent draws from that population; how large the sample must be depends on the population's shape.

The central limit theorem is the theoretical justification for many of the most widely used statistical methods, including confidence intervals, hypothesis tests (t tests, z tests), and ANOVA. These methods assume that the sampling distribution of the mean is approximately normal, and the CLT justifies treating this assumption as reasonable when the sample size is sufficiently large. Without the CLT, researchers would need to know the exact shape of the population distribution before selecting a statistical method, which is rarely possible in practice. The CLT also explains why the standard error of the mean, which quantifies the precision of sample estimates, can be turned into confidence intervals and p values using the normal or t distribution. The standard error formula itself (the population standard deviation divided by the square root of n) holds for independent observations at any sample size; what the CLT adds is the approximately normal shape of the sampling distribution.^[3]

Consider a population with any distribution (uniform, skewed, bimodal, etc.) that has a finite mean and finite variance. If you draw a random sample of size n from this population and calculate the sample mean, then repeat this process many times, the distribution of those sample means will approach a normal distribution as n increases. The mean of the sampling distribution equals the population mean, and the standard deviation of the sampling distribution (the standard error) equals the population standard deviation divided by the square root of n.^[1][3]

The rate at which the sampling distribution converges to normality depends on the shape of the underlying population distribution. For populations that are already approximately normal, even small samples (n = 10 to 15) produce sampling distributions that are very close to normal. For highly skewed or heavy tailed distributions, larger samples (n = 30, 50, or more) may be needed before the approximation is adequate.^[2] The conventional rule of thumb of n = 30 as a sufficient sample size is a rough guideline, not a universal guarantee.

The CLT justifies the use of normal theory based statistical methods (z tests, t tests, confidence intervals for means, ANOVA) when the sample size is large enough for the sampling distribution to be approximately normal. It is most relevant when the researcher is working with sample means or sums and needs to make inferences about population parameters. The CLT does not apply to statistics other than means and sums without modification (for example, medians or variances have different sampling distributions). It also does not apply when the population has an infinite variance (as with certain heavy tailed distributions).^[3]

A researcher collects data on hospital length of stay, which is heavily right skewed in the population (median 3 days, with a long tail extending beyond 30 days). With a sample of 15 patients, the sampling distribution of the mean length of stay would still be noticeably skewed, and a t test based on this sample might produce misleading p values. With a sample of 200 patients, however, the sampling distribution of the mean is approximately normal despite the skewness of the individual observations, and confidence intervals based on the normal approximation are reliable. The researcher calculates a mean length of stay of 5.2 days (95% CI: 4.6 to 5.8 days) and uses this interval for comparison with national benchmarks.