What does the central limit theorem actually say

The central limit theorem (CLT) is a cornerstone of inferential statistics. It tells us that when we take many independent samples from any population with a finite variance, the distribution of the sample means will look increasingly like a bell‑shaped curve. This happens even if the original data are highly skewed or discrete, as long as we have enough observations. Because the normal shape is fully described by its mean and variance, the CLT gives us a simple way to approximate probabilities for sums and averages.

Formal Statement

Mathematically, if ($X_1$,\dots,$X_n$) are i.i.d. with mean $\mu$ and variance $\sigma^2$, then $\sqrt{n}(\bar{X}-\mu)/\sigma \xrightarrow{d} N(0,1)$ as $n\to\infty$. In plain language, the sample mean $\bar{X}$ becomes normally distributed around the true mean, with a spread that shrinks like $\sigma/\sqrt{n}$. This convergence holds no matter how non‑normal the original population may be, as long as its variance is finite.

The CLT holds under three main conditions:

• Observations are independent and identically distributed (i.i.d.).
• The population has a finite mean $\mu$ and finite variance $\sigma^2$.
• The sample size $n$ is sufficiently large; larger $n$ is needed for more irregular original shapes.

To apply the CLT in a problem, follow these steps:

1. 1Check that the data meet the i.i.d. assumption and have finite variance.
2. 2Choose a sample size $n$ that is large enough for the desired accuracy.
3. 3Compute the sample mean $\bar{X}$ and the standard error $\sigma/\sqrt{n}$.
4. 4Standardize: $Z = (\bar{X}-\mu)/(\sigma/\sqrt{n})$ and use the standard normal table.

Worked example: suppose a factory produces bolts that are either defective (1) or good (0) with a true defect rate $p=0.30$. Draw a random sample of $n=30$ bolts. The sample proportion $\hat{p}=\frac{\sum X_i}{30}$ has mean $0.30$ and standard error $\sqrt{p(1-p)/30}=0.084$. Using the CLT, $Z = (\hat{p}-0.30)/0.084$ is approximately standard normal, so the probability that $\hat{p}$ falls between 0.22 and 0.38 is about 0.95, matching the normal prediction. The approximation is already good, illustrating how the CLT lets us treat the proportion as if it were drawn from a normal distribution.

Effect of sample size on the sampling distribution of the mean for a Bernoulli(0.3) variable:

Why the CLT matters: it provides a practical bridge from unknown, messy data to the well‑understood normal model, enabling confidence intervals and hypothesis tests without knowing the exact population shape. Even with a highly skewed exponential or a discrete binomial source, the average of a modest number of observations behaves predictably. Thus the CLT is the engine behind most statistical inference taught in introductory courses.