Asymptotic statistical theorem
Central Limit Theorem
Under broad conditions, properly standardized sums or averages of many independent contributions approach a normal distribution.
(X_bar - mu) / (sigma / sqrt(n)) -> N(0, 1)
For independent, identically distributed variables with finite variance, the standardized sample mean converges in distribution to the standard normal as n grows.
The sampling machine repeatedly averages draws from a fixed right-skewed exponential population. The reported standard-error relation is sigma/sqrt(n); convergence speed depends on the source distribution and assumptions.
(%)
AVERAGE
REPEAT
Change sample size, then repeat the experiment. The source remains skewed; it is the distribution of averages that tightens and becomes more symmetric.
- CHANGE
- Sample size
- WATCH
- standard error
- MEANING
- The sampling machine repeatedly averages draws from a fixed right-skewed exponential population. The reported standard-error relation is sigma/sqrt(n); convergence speed depends on the source distribution and assumptions.
Averages stabilize before everything is normal.
Individual observations may be skewed or lumpy. Repeated averages become tighter and often more bell-shaped as sample size grows.
What it actually says
The Central Limit Theorem explains why normal approximations appear so often in measurement, polling, quality control, and statistics. It is a theorem about standardized sums or sample means, not a claim that all real-world variables are normally distributed.
The theorem has many versions. The familiar iid finite-variance case is only the entry point; dependence, unequal variances, heavy tails, and finite samples require more careful forms or different tools.
"A useful law compresses a pattern. It does not erase the conditions that make the pattern true."
How the idea developed
The modern form emerged through observation, argument, and later refinement. The timeline separates the first insight from the version now used in textbooks and practice.[1]
De Moivre derives a normal approximation to the binomial distribution.
Laplace extends normal approximations for sums and errors.
Lyapunov proves an important general central limit condition.
Lindeberg, Levy, Feller, and others refine modern versions and conditions.
How the pattern works
The relation becomes useful only when its mechanism, measurement process, and operating range are visible.
Summing independent contributions blurs many distributional details.
Center by the mean and scale by the standard error to compare across sample sizes.
The approximation improves with n, but speed depends on tail weight, skew, and dependence.
For independent, identically distributed variables with finite variance, the standardized sample mean converges in distribution to the standard normal as n grows.
Where it earns its keep
Applications are strongest when the law changes a decision, measurement, model, or experiment rather than merely providing an analogy.
Estimate uncertainty
ApplicationSampling distributions of averages and proportions often support confidence intervals and margins of error.
The sample design must justify the independence approximation.
Monitor process means
ApplicationAverage measurements can be tracked with normal approximations even when individual noise is imperfect.
Autocorrelation and drift can break the model.
Power and precision
ApplicationStandard errors shrink roughly with the square root of sample size.
Four times the data gives about half the standard error, not four times the certainty.
Where it stops working
The theorem is asymptotic. Small samples from skewed, discrete, bounded, dependent, or heavy-tailed distributions may converge slowly or not follow the familiar finite-variance version.
Heavy-tailed variables with infinite variance can converge to stable laws instead of the normal. Clustered or dependent observations can make the effective sample size much smaller than the count suggests.
"The data are normal because n is large"
Better: The sampling distribution of a mean may be approximately normal; the raw data need not be."Thirty is always enough"
Better: Required n depends on skew, tails, dependence, and the desired accuracy."More observations always solve it"
Better: Biased sampling and dependence do not vanish by arithmetic alone."CLT justifies every p-value"
Better: Inference also needs design, measurement, and model assumptions.Sources and further reading
Original publications and serious secondary scholarship are prioritized over summaries.
- Encyclopedia of Mathematics - Central limit theoremFormal statement and mathematical variants.https://encyclopediaofmath.org/wiki/Central_limit_theorem
- NIST/SEMATECH e-Handbook - Central Limit TheoremApplied statistical explanation for process monitoring.https://www.itl.nist.gov/div898/handbook/pmc/section5/pmc51.htm
- Feller - An Introduction to Probability Theory and Its ApplicationsClassic probability text with rigorous CLT treatment.https://archive.org/details/introductiontopr0002fell
- Le Cam - The Central Limit Theorem around 1935Historical and technical perspective on modern CLT development.https://projecteuclid.org/journals/statistical-science/volume-1/issue-1/The-Central-Limit-Theorem-Around-1935/10.1214/ss/1177013818.full