Central Limit Theorem: Concepts, Formulas and Examples

The Central Limit Theorem (CLT) states that the average of a large number of independent observations is approximately normally distributed, no matter what shape the original data have. Average 100 die rolls and the result behaves like a bell curve around 3.5, even though a single die roll is completely flat.

It matters because it is the single most used theorem in applied statistics. It justifies treating sample means as normal, which unlocks confidence intervals, hypothesis tests and the whole machinery of inference, even when the underlying data are skewed, discrete or strange-looking.

Key Concepts

• X̄ₙ: the sample mean of n independent, identically distributed observations with mean μ and variance σ².
• The theorem: (X̄ₙ - μ)/(σ/√n) → Z ~ N(0, 1) as n grows. In words, the standardised sample mean becomes standard normal.
• Standard error: σ/√n is the standard deviation of the sample mean; it shrinks as √n, so quadrupling n halves the error.
• Rule of thumb: n ≥ 30 is usually enough for a decent approximation, though skewed data may need more.
• Requirements: independent observations, identical distribution, and finite variance. The CLT describes the mean, not the raw data: individual die rolls never become normal.

Formulas

Approximate distribution: X̄ₙ ≈ N(μ, σ²/n). Standardised form: Z = (X̄ₙ - μ)/(σ/√n). For sums: Sₙ = X₁ + … + Xₙ ≈ N(nμ, nσ²). For proportions: p̂ ≈ N(p, p(1-p)/n). Assumptions: independence, finite σ², and n large enough that the approximation is good. The theorem says nothing about how large n must be for your specific distribution; heavy skew needs bigger n.

Worked Example

Roll a fair die n = 36 times and average the results. Here μ = 3.5 and σ² = 35/12 ≈ 2.9167, so σ ≈ 1.7078. The CLT says X̄ ≈ N(3.5, 2.9167/36) = N(3.5, 0.08102), so the standard error is σ/√n = 1.7078/6 ≈ 0.2846. What is P(X̄ > 3.9)? Z = (3.9 - 3.5)/0.2846 = 0.4/0.2846 ≈ 1.405. From the Z table, P(Z > 1.405) = 1 - 0.9199 ≈ 0.0801, about 8%. A single roll can never exceed 6 with probability logic like this, but the average behaves beautifully normally.

Common Misconceptions

• The CLT is about averages, not data: 36 die rolls themselves stay flat and discrete; only their average becomes bell-shaped. Plotting raw skewed data and expecting normality is the classic error.
• n ≥ 30 is a guideline, not a law: symmetric data can look normal at n = 10, while heavily skewed data (like incomes) may need n in the hundreds.
• Finite variance is essential: for the Cauchy distribution, which has no variance, averages never settle into a bell curve no matter how large n is.

Real-World Applications

• Polling: the margin of error, 1/√n style, comes from the CLT applied to sample proportions.
• Manufacturing: average fill levels of 50 bottles are treated as normal to set control limits, even if single fills are not normal.
• A/B testing: comparing conversion rates between two website versions relies on the CLT for the difference of two proportions.

The Central Limit Theorem turns any well-behaved data into normal averages: X̄ ≈ N(μ, σ²/n), with the error shrinking as σ/√n. It is why statisticians can build confidence intervals and tests without knowing the data's original shape, and why the normal curve rules applied statistics.