Central Limit Theorem: Where the Bell Comes From

"The CLT explains why standardized sums and means often approach a bell under stated conditions. It does not make every natural quantity normal."

The Statement, Plainly

The Central Limit Theorem (CLT) says: if you take the sum (or equivalently the mean) of N independent identically-distributed random variables with finite variance, and you standardize that sum (subtract its mean and divide by its standard deviation), the resulting distribution approaches a standard normal as N grows large. In symbols:

(X₁ + X₂ + ... + Xₙ − nμ) / (σ√n) → N(0, 1) as n → ∞

This is a central result in introductory statistics. The starting distribution need not be normal, but the stated finite-variance and iid conditions matter. Skewness and tail thickness determine how quickly the approximation improves. Its broad applicability helps explain why normal approximations recur.

What This Buys Us in Real Life

So many real-world quantities are the result of many small independent contributions:

Adult human heights are the sum of countless genetic and developmental factors.
Measurement errors are the sum of many small instrumental and environmental noise sources.
IQ scores are constructed to be normal by design, but the underlying cognitive abilities are themselves sums of many small genetic and experiential contributions.
Test scores on a well-designed multi-question exam aggregate many small per-question outcomes.

These examples motivate a normal approximation but do not prove it: contributions may differ in size, depend on one another, or produce bounded and discrete outcomes. The CLT directly describes a standardized sum or mean, not every observed marginal distribution.

The Demo That Makes It Click

Take a deeply non-normal distribution — like the outcome of a single die roll, which is uniform over {1,2,3,4,5,6}. The histogram of one roll is flat and obviously not bell-shaped. Now sum N independent rolls. The histogram of the sum visibly becomes more bell-shaped as N grows. For dice the shape becomes visibly smoother as N grows, but there is no universal N=30 rule or perfect finite-sample bell. This is not magic; it is the CLT operating live.

The Preconditions Are Load-Bearing

Independence, identical distribution, and finite variance are familiar sufficient conditions for the simplest CLT—not the only possible conditions. Some CLTs allow weak dependence or non-identical terms, while strong dependence, dominant terms, or infinite variance can defeat the usual √N normal approximation. Correlation alone does not imply a nonnormal sum; correlated jointly normal variables still sum to a normal variable. These conditions are part of the test for whether to trust a normal approximation — and they are exactly what Track 07 will dismantle in detail.

Code

CLT demo: sums of uniform dice converge to a bell·python

import numpy as np
rng = np.random.default_rng(40)

# CLT demo: sum N uniform-die rolls and watch the sum become bell-shaped.
M = 20_000   # number of simulated 'sum experiments' per N

for N in (1, 2, 5, 30):
    # M experiments, each summing N independent die rolls.
    rolls = rng.integers(1, 7, size=(M, N))
    sums = rolls.sum(axis=1)
    # Skewness and kurtosis collapse toward normal values (0 and 3) as N grows.
    mean = sums.mean()
    std = sums.std()
    # Standardized skewness check (rough).
    skew = ((sums - mean) ** 3).mean() / std ** 3
    kurt = ((sums - mean) ** 4).mean() / std ** 4
    print(f"N={N:>3d}  mean={mean:>6.2f}  std={std:>5.2f}  "
          f"skew={skew:+.3f}  kurt={kurt:.3f}  (normal: skew=0, kurt=3)")

# At N=1: single die, uniform — visibly non-normal.
# At N=30: sum of 30 dice — skewness near 0, kurtosis near 3.
# Same dice; the SUM is bell-shaped even though the individual roll isn't.
# That is the CLT operating in front of your eyes.

Exercise

Identify one quantity in your life that is a sum of many small independent contributions (your weekly total spending, your monthly typing volume, your daily heart rate variability average). Verify the CLT intuition: do you expect this quantity's distribution to look bell-shaped across many weeks/months/days? Then identify one quantity where the CLT preconditions are violated (correlated effects or a dominating single factor) and notice that you would NOT expect a bell shape there.

Hint

CLT yes: small + independent + many. CLT no: correlated, or one dominator, or fat-tailed individual contributions.