Lesson 3 of 8 · 15 min

The central limit theorem

See sample means from skewed, uniform and two-peaked populations become approximately normal as nn grows, and standardise a sample mean with sn\frac{s}{\sqrt{n}} for n≥30n \geq 30.

What you will be able to do

  • Describe how the shape of the distribution of Xˉ\bar{X} changes with nn for populations of different shapes.
  • State the central limit theorem and decide when a normal model for Xˉ\bar{X} is reasonable.
  • Standardise a sample mean with xˉ−μs/n\frac{\bar{x} - \mu}{s / \sqrt{n}} when σ\sigma is unknown and n≥30n \geq 30.

Last lesson's batteries had normally distributed lifetimes, so their sample means were normal too. Most real populations are not normal. Waiting times on a helpline are right-skewed: most calls are answered quickly, a few wait a very long time. Incomes, house prices and insurance claims are skewed too. Can you still use the normal distribution for their sample means?

Interactive

Sample means from a skewed population

Draw many samples with n=1n = 1, then n=2n = 2, 55, 1010 and 3030. When does the histogram of xˉ\bar{x} start to look like the normal curve? Then switch the population to uniform or bimodal and repeat.

Whatever population you start from, the same three things happen as nn grows: the histogram of sample means stays centred at μ\mu, it narrows like σn\frac{\sigma}{\sqrt{n}}, and its shape becomes more and more like a normal curve. Skewed populations take longer than symmetric ones, but they get there.

Key idea

The central limit theorem

For random samples of size nn from any population with mean μ\mu and standard deviation σ\sigma, when nn is large

Xˉ  ∼˙  N(μ,σ2n)approximately\bar{X} \;\dot\sim\; N\left(\mu, \frac{\sigma^2}{n}\right) \quad \text{approximately}

The larger nn is, the better the approximation. If the population itself is normal, the result is exact for every nn.

How large is "large"? A working guide.
PopulationDistribution of Xˉ\bar{X}
NormalExactly normal for every nn.
Symmetric with one peak (e.g. uniform)Close to normal for quite small nn, often by n≈10n \approx 10.
Two peaks, or moderately skewedNeeds a larger nn; usually fine by n=30n = 30.
Strongly skewed (e.g. waiting times)Needs n≥30n \geq 30, and more for extreme skew.

Household water bills in a city are strongly right-skewed. A random sample of 100 bills is taken. Which statement is correct?

Choose one answer

There is a practical snag. The formulas use σ\sigma, the population standard deviation, which you almost never know. For a large sample, the sample standard deviation ss is a good estimate of σ\sigma, so you replace σn\frac{\sigma}{\sqrt{n}} by sn\frac{s}{\sqrt{n}}. The syllabus form of the central limit theorem says that for large samples (n≥30n \geq 30)

Z=Xˉ−μsnZ = \frac{\bar{X} - \mu}{\frac{s}{\sqrt{n}}}

is approximately standard normal, N(0,1)N(0, 1). Here ss varies from sample to sample too, so this is a different statistic, and the widget below lets you check it.

Interactive

Standardising with s

With n=30n = 30 the standardised means sit close to N(0,1)N(0, 1). Reduce nn to 5: the tails get heavier, because ss from a small sample is an unreliable estimate of σ\sigma.

Worked example

Using the central limit theorem for a skewed population

  • Complex familiar

Waiting times for a council's phone line are strongly right-skewed, with mean 88 minutes and standard deviation 66 minutes. A random sample of 3636 calls is taken.

(a) State the approximate distribution of the sample mean waiting time Xˉ\bar{X}, and justify it.

(b) Find the approximate probability that the sample mean is more than 1010 minutes: (i) without technology, using the 68–95–99.7% rule, and (ii) with technology.

(c) Explain why the same method can't be used to find the probability that one call waits more than 1010 minutes.

5 steps

Try it yourself first, then check the working one step at a time.

Method

  1. Find μ\mu and σn\frac{\sigma}{\sqrt{n}}.

  2. Justify normality: large nn (central limit theorem) for a non-normal population.

  3. Count standard errors for the 68–95–99.7% rule, or use normCdf.

Watch out for

  • Claiming the waiting times themselves become normal.

  • Using σ=6\sigma = 6 instead of 636=1\frac{6}{\sqrt{36}} = 1.

A random sample of 3636 values from a population with mean μ=50\mu = 50 has xˉ=52.4\bar{x} = 52.4 and s=9s = 9. Calculate the standardised value z=xˉ−μs/nz = \frac{\bar{x} - \mu}{s / \sqrt{n}}.

Common mistake

The population does not become normal

The central limit theorem is about the distribution of sample means. One helpline call still has a right-skewed waiting time, so you cannot use a normal model to find P(X>10)P(X > 10) for one call. Only probabilities about Xˉ\bar{X}, with nn large, use the normal approximation.

For which of these is a normal model for the sample mean Xˉ\bar{X} least justified?

Choose one answer

Exam tip

Justify the normal model

When you use a normal distribution for Xˉ\bar{X}, say why in a few words: "XX is normally distributed, so Xˉ\bar{X} is normal" or "n=64n = 64 is large, so by the central limit theorem Xˉ\bar{X} is approximately normal". If neither holds, say that the normal model may not be reasonable.

Summary

  • For large nn, Xˉ\bar{X} is approximately N(μ,σ2n)N\left(\mu, \frac{\sigma^2}{n}\right) whatever the population shape: the central limit theorem.

  • The more skewed the population, the larger the sample needed; n ≥ 30 is the usual guide.

  • For n≥30n \geq 30, Xˉ−μs/n\frac{\bar{X} - \mu}{s/\sqrt{n}} is approximately N(0,1)N(0, 1), so ss can stand in for an unknown σ\sigma.

  • Individual values keep the population's shape; only the sample mean becomes normal.

  • Always justify the normal model: normal population, or large nn.

Finished the lesson?

Try the 3 quick checks above and the lesson completes itself, or mark it complete now.

Lesson 3 of 80% read

Save your progress with a free account