4.15 · AHL · HL onlyPaper 1Paper 2Paper 3

Sample mean and the central limit theorem

In a minute

Add or subtract independent normal variables and the result is normal too. In particular, the mean Xˉ\bar{X} of a of size nn from N(μ,σ2)\mathrm{N}(\mu, \sigma^{2}) is N(μ,σ2n)\mathrm{N}\left(\mu, \frac{\sigma^{2}}{n}\right). The says that for a large Xˉ\bar{X} is approximately normal whatever the looks like, and in exams n>30n > 30 is large enough.

You'll be able to

  • Find probabilities for sums and differences of independent normal variables, such as a total mass or a race between two people.
  • Write down the Xˉ\bar{X} for a random sample from a normal population, and use it to find probabilities.
  • Use the central limit theorem to treat Xˉ\bar{X} as approximately normal for a large sample from any population, and say when it applies.
  • Interpret a probability about a sample mean in context, and explain why means vary less than single values.
Before you start3 quick questions on what this builds on
Check yourself
X∼N(50,62)X \sim \mathrm{N}(50, 6^{2}). Find P(X<58)\mathrm{P}(X < 58). Give your answer to 3 s.f.
Check yourself
X1X_{1}, X2X_{2} and X3X_{3} are independent, and each has variance 44. Find Var(X1+X2+X3)\mathrm{Var}(X_{1} + X_{2} + X_{3}).
Check yourself
A gym wants a random sample of 4040 of its members. Which method gives one?

Revise first: 4.1 Sampling, data quality and outliers, 4.3 Central tendency and dispersion, 4.9 Normal distribution and 4.14 Linear combinations and unbiased estimates

Sums and differences of normal variables

In 4.14 you found the mean and variance of a such as X+YX + Y or X−YX - Y. That gives the centre and the spread, but not the shape. Without the shape, you can't find a probability.

For variables with a the shape is settled too. Add or subtract independent normal variables, and the result is normal again.

In the formula booklet
E(a1X1±a2X2±⋯±anXn)=a1E(X1)±a2E(X2)±⋯±anE(Xn)\mathrm{E}(a_{1}X_{1} \pm a_{2}X_{2} \pm \dots \pm a_{n}X_{n}) = a_{1}\mathrm{E}(X_{1}) \pm a_{2}\mathrm{E}(X_{2}) \pm \dots \pm a_{n}\mathrm{E}(X_{n})
In the formula booklet
Var(a1X1±a2X2±⋯±anXn)=a12 Var(X1)+a22 Var(X2)+⋯+an2 Var(Xn)\mathrm{Var}(a_{1}X_{1} \pm a_{2}X_{2} \pm \dots \pm a_{n}X_{n}) = a_{1}^{2}\,\mathrm{Var}(X_{1}) + a_{2}^{2}\,\mathrm{Var}(X_{2}) + \dots + a_{n}^{2}\,\mathrm{Var}(X_{n})

Both rules are in the formula booklet, and the variance rule is for independent variables. The normal shape needs two things: every XiX_{i} is normal, and they are independent.

1. Define the combination in words and symbols, such as T=B+CT = B + C, the total mass.

2. Find its mean and its variance, then write its distribution, such as T∼N(3250,4500)T \sim \mathrm{N}(3250, 4500).

3. Use normal cdf on your GDC, entering the standard deviation variance\sqrt{\text{variance}}.

Worked exampleA gift hamper is packed in a basket whose mass BB grams is modelled by N(850,302)\mathrm{N}(850, 30^{2}). The food inside has mass CC grams, where C∼N(2400,602)C \sim \mathrm{N}(2400, 60^{2}), independent of BB. Find the probability that a packed hamper weighs more than 3.33.3 kg.

Try it first, then reveal one step at a time.

  1. Let T=B+CT = B + C be the total mass in grams. BB and CC are independent and normal, so TT is normal.
Worked exampleOver 400400 m, Amara's time AA seconds is modelled by N(58.2,1.12)\mathrm{N}(58.2, 1.1^{2}) and Bea's time BB seconds by N(59.0,0.82)\mathrm{N}(59.0, 0.8^{2}). Their times are independent. Find the probability that Bea beats Amara in a race, and comment on the claim that Amara always wins. Give your answer to 3 s.f.

Try it first, then reveal one step at a time.

  1. Bea wins when her time is shorter: B<AB < A, that is D=A−B>0D = A - B > 0.
Check yourself
X∼N(20,32)X \sim \mathrm{N}(20, 3^{2}) and Y∼N(15,42)Y \sim \mathrm{N}(15, 4^{2}) are independent, and D=X−YD = X - Y. DD is normal with mean 55. Find Var(D)\mathrm{Var}(D).
Check yourself
A lift carries at most 680680 kg. The masses of adults who use it are modelled by N(78,122)\mathrm{N}(78, 12^{2}) kg, independently. Eight adults chosen at random get in. Find the probability that their total mass is more than 680680 kg. Give your answer to 3 s.f.

The mean of a sample from a normal population

One bottle from a filling machine may be well off target. Average nine bottles, and the light and heavy ones partly cancel out. So the mean lands closer to the target than a single bottle usually does.

Take a random sample X1,X2,…,XnX_{1}, X_{2}, \dots, X_{n} from a population with mean μ\mu and variance σ2\sigma^{2}. Each XiX_{i} has the population's distribution, and you treat them as independent. Their mean Xˉ=1n(X1+X2+⋯+Xn)\bar{X} = \frac{1}{n}(X_{1} + X_{2} + \dots + X_{n}) is a linear combination with every coefficient 1n\frac{1}{n}.

The rules give E(Xˉ)=μ\mathrm{E}(\bar{X}) = \mu and Var(Xˉ)=n×1n2σ2=σ2n\mathrm{Var}(\bar{X}) = n \times \frac{1}{n^{2}}\sigma^{2} = \frac{\sigma^{2}}{n}. If the population is normal, Xˉ\bar{X} is a linear combination of independent normal variables, so it is normal too. This is the .

Learn this
X∼N(μ,σ2)  ⇒  Xˉ∼N(μ,σ2n)X \sim \mathrm{N}(\mu, \sigma^{2}) \;\Rightarrow\; \bar{X} \sim \mathrm{N}\left(\mu, \frac{\sigma^{2}}{n}\right)

The variance is divided by nn, so the standard deviation of Xˉ\bar{X} is σn\frac{\sigma}{\sqrt{n}}. To halve the spread of the mean, you need four times as many values.

Two normal curves centred on the same mean mu. The wide, low curve is the population X with standard deviation sigma. The tall, narrow curve is the mean of a sample of 9, with standard deviation sigma over 3. μ mean of 9 values: SD σ/3 one value X: SD σ
The mean of 99 values has standard deviation σ9=σ3\frac{\sigma}{\sqrt{9}} = \frac{\sigma}{3}, so its curve is narrower and taller.
Worked exampleA machine fills bottles of juice. The volume VV ml is modelled by N(505,62)\mathrm{N}(505, 6^{2}), and the label says 500500 ml. (a) Find the probability that one bottle holds less than 500500 ml. (b) An inspector buys 99 bottles at random. Find the probability that their mean volume is less than 500500 ml. Give your answers to 3 s.f.

Try it first, then reveal one step at a time.

  1. (a) One bottle: normal cdf with lower −1099-10^{99}, upper 500500, μ=505\mu = 505, σ=6\sigma = 6 gives P(V<500)=0.2023…\mathrm{P}(V < 500) = 0.2023\ldots
Worked exampleFinishing times at a weekly 55 km park run are modelled by T∼N(31.5,4.82)T \sim \mathrm{N}(31.5, 4.8^{2}) minutes. A random sample of 2020 runners is chosen. Write down the distribution of Tˉ\bar{T}, and find P(29.5<Tˉ<32.5)\mathrm{P}(29.5 < \bar{T} < 32.5). Give your answer to 3 s.f.

Try it first, then reveal one step at a time.

  1. TT is normal, so Tˉ\bar{T} is normal for any nn, with mean 31.531.5.
Check yourself
Battery lifetimes are modelled by N(80,102)\mathrm{N}(80, 10^{2}) hours. For a random sample of 2525 batteries, find Var(Xˉ)\mathrm{Var}(\bar{X}).
Check yourself
Bags of flour have masses with standard deviation 1515 g. Find the smallest sample size nn for which the standard deviation of Xˉ\bar{X} is at most 33 g.
Check yourself
Seedling heights are modelled by N(12,2.52)\mathrm{N}(12, 2.5^{2}) cm. Which is more likely: one seedling chosen at random is taller than 1313 cm, or the mean height of 1010 seedlings chosen at random is more than 1313 cm?

The central limit theorem

Many populations are not normal. Spending in a café is skewed: most bills are small, and a few are large. Scores on a die are all equally likely. Is the mean of a sample still normal?

Approximately, once the sample is large enough. Each mean averages many values, so the rare large ones are diluted. The bell shape appears whatever the population looks like.

Left: a population curve that rises steeply, peaks early and has a long tail to the right, with its mean mu marked. Right, drawn on a stretched horizontal scale: the distribution of the mean of random samples of 40 from that population, a symmetric bell-shaped curve centred on the same mean mu. μ μ one value X: skewed mean of 40 values: near normal population sample mean X̄
A skewed population (left) and the means of samples of 4040 from it (right). The right panel is stretched sideways: the means are far less spread out.

How large is large? It depends on the population. A roughly symmetric population needs only a small sample, and a strongly skewed one needs more. In IB exams, n>30n > 30 counts as large enough.

If the population is normal, you don't need the theorem: Xˉ\bar{X} is exactly normal for every nn. The theorem is for populations that are not normal, or whose shape you don't know.

The theorem is about Xˉ\bar{X}, not about single values. A large sample does not make the population normal: one value keeps the population's shape.

Worked exampleThe amount a customer spends in a café has mean 8.408.40 dollars and standard deviation 5.105.10 dollars, and its distribution is right-skewed. A random sample of 5050 customers is taken. Find the probability that their mean spend is more than 99 dollars. Give your answer to 3 s.f.

Try it first, then reveal one step at a time.

  1. The population is skewed, not normal. But n=50>30n = 50 > 30, so by the central limit theorem Xˉ\bar{X} is approximately normal.
Worked exampleThe number of people in a car passing a checkpoint has mean 1.61.6 and standard deviation 0.90.9. Its distribution is right-skewed. (a) Explain why a normal distribution is a poor model for the number of people in one car. (b) A survey records 6464 cars chosen at random. Find the probability that the mean number of people per car is at least 1.751.75. Give your answer to 3 s.f.

Try it first, then reveal one step at a time.

  1. (a) One car holds a whole number of people, at least 11, and the distribution is skewed. A normal model is symmetric and continuous: N(1.6,0.92)\mathrm{N}(1.6, 0.9^{2}) would put about 11%11\% of cars below 0.50.5 people, which is impossible.
Check yourself
A random sample of 1212 is taken from a strongly right-skewed population. Can you model Xˉ\bar{X} with a normal distribution?
Check yourself
A fair six-sided die gives scores with mean 3.53.5 and variance 3512\frac{35}{12}. It is rolled 4040 times. Find the probability that the mean score is greater than 3.83.8. Give your answer to 3 s.f.

Common mistakes

  • Dividing the wrong thing by nn. For a random sample of size nn, Var(Xˉ)=σ2n\mathrm{Var}(\bar{X}) = \frac{\sigma^{2}}{n} and the standard deviation is σn\frac{\sigma}{\sqrt{n}}, not σn\frac{\sigma}{n}.
  • Using the population's standard deviation for a question about a mean. A probability about Xˉ\bar{X} needs σn\frac{\sigma}{\sqrt{n}} entered as the standard deviation in the GDC; entering σ\sigma gives the probability for a single value instead.
  • Subtracting variances for a difference. For independent AA and BB, Var(A−B)=Var(A)+Var(B)\mathrm{Var}(A - B) = \mathrm{Var}(A) + \mathrm{Var}(B).
  • Saying the central limit theorem makes the population normal. It is about Xˉ\bar{X} only, and only for large nn; single values keep the population's shape.
  • Giving a GDC probability with no distribution written down. State the distribution, such as Xˉ∼N(8.40,0.5202)\bar{X} \sim \mathrm{N}(8.40, 0.5202), then the probability.

Exam tips

  • Write down the distribution of Xˉ\bar{X} with its mean and variance, such as Xˉ∼N(64,920)\bar{X} \sim \mathrm{N}\left(64, \frac{9}{20}\right). These often earn their own marks, separate from the probability.
  • Justify a normal model for Xˉ\bar{X} in one line: either the population is normal, or n>30n > 30 so the central limit theorem applies. Name which one.
  • State an assumption in context. For a total or a sample mean the key one is independence, such as one customer's spend not affecting another's, not just "the data are independent".
  • Interpret a probability about a mean in the question's words: a small value means a sample mean like that would be unusual if the model is right.

Key takeaways

Learn thisIf X1,…,XnX_{1}, \dots, X_{n} are independent and each is normal, then a1X1±⋯±anXna_{1}X_{1} \pm \dots \pm a_{n}X_{n} is normal. Find its mean and variance with the booklet rules, then enter σ=variance\sigma = \sqrt{\text{variance}} in normal cdf.
Learn thisX∼N(μ,σ2)  ⇒  Xˉ∼N(μ,σ2n)for a random sample of size nX \sim \mathrm{N}(\mu, \sigma^{2}) \;\Rightarrow\; \bar{X} \sim \mathrm{N}\left(\mu, \frac{\sigma^{2}}{n}\right) \quad \text{for a random sample of size } n
Learn thisFor a random sample of size nn from a population with variance σ2\sigma^{2}, Var(Xˉ)=σ2n\mathrm{Var}(\bar{X}) = \frac{\sigma^{2}}{n} and the standard deviation of Xˉ\bar{X} is σn\frac{\sigma}{\sqrt{n}}: four times the sample size halves the spread.
Learn thisCentral limit theorem: for a random sample of size nn from any population with mean μ\mu and variance σ2\sigma^{2}, Xˉ\bar{X} is approximately N(μ,σ2n)\mathrm{N}\left(\mu, \frac{\sigma^{2}}{n}\right) when nn is large. How large depends on the population; in IB exams n>30n > 30 is enough.
Learn thisA normal population gives an exactly normal Xˉ\bar{X} for every nn. A population that is not normal needs a large sample before Xˉ\bar{X} can be treated as normal.

Practise this

12 past-paper questions on Unbiased estimates and sample mean distribution in the AI HL bank, each with a mark scheme and a worked explanation.

Practise Unbiased estimates and sample mean distribution →