Poisson distribution and the central limit theorem AI HL

A Poisson distribution models the number of events in a fixed interval when they happen independently at a uniform average rate, and the calculator gives its probabilities once the rate has been scaled to the interval in the question. The central limit theorem says that the mean of a large sample is approximately normal with a smaller spread than a single value, which turns a question about a sample mean into a normal probability. Work through the Poisson section first, then the central limit theorem section, and read the worked example in each before trying the questions.

Formulas at a glance

Only the Poisson probability formula is in the formula booklet. The mean and variance of a Poisson variable, the sum of two independent Poisson variables, and the distribution of a sample mean are not given as results, so learn them.

Poisson probability, X∼Po(λ)X\sim\text{Po}(\lambda)

P(X=r)=e−λλrr!P(X=r)=\dfrac{e^{-\lambda}\lambda^{r}}{r!}

In the booklet

Mean and variance of a Poisson variable

E(X)=Var(X)=λE(X)=\text{Var}(X)=\lambda

Not in the booklet

Sum of two independent Poisson variables

X∼Po(λ1), Y∼Po(λ2)⇒X+Y∼Po(λ1+λ2)X\sim\text{Po}(\lambda_1),\ Y\sim\text{Po}(\lambda_2)\Rightarrow X+Y\sim\text{Po}(\lambda_1+\lambda_2)

Not in the booklet

Distribution of the sample mean

X¯∼N(μ,σ2n)\bar X\sim N\!\left(\mu,\dfrac{\sigma^2}{n}\right)

Not in the booklet

Standard deviation of the sample mean

σn\dfrac{\sigma}{\sqrt n}

Not in the booklet

Standardising a sample mean

z=x¯−μσ/nz=\dfrac{\bar x-\mu}{\sigma/\sqrt n}

Not in the booklet

Which method to use

First decide which model fits the context: a count of events in an interval is Poisson, a count of successes in a fixed number of trials is binomial, and a measurement on a continuous scale is normal. Then work out the distribution you need before touching the calculator.

The question hasUseNotes
Events occurring at an average rate, with no fixed number of trialsX∼Po(λ)X\sim\text{Po}(\lambda), where λ\lambda is the mean number of events in the intervalEvents must be independent and occur at a uniform average rate
A rate quoted for one interval, and a probability asked for a different intervalMultiply the rate by the length of the new interval to get λ\lambdaDo this before any calculation, then check the units agree
"Exactly rr events"Poisson probability density (poissonpdf) with λ\lambda and rrThis is P(X=r)P(X=r), one value
"At most", "fewer than", "at least" or "more than"Poisson cumulative probability (poissoncdf), using the complement where it is shorterP(X≥k)=1−P(X≤k−1)P(X\ge k)=1-P(X\le k-1); watch whether kk itself is included
The probability of no events, or a probability given and the rate wantedP(X=0)=e−λP(X=0)=e^{-\lambda}, then solve for λ\lambda with a logarithmUseful when the rate is not stated directly
The total of two independent counts, each PoissonX+Y∼Po(λ1+λ2)X+Y\sim\text{Po}(\lambda_1+\lambda_2)The means add; the counts must be independent
A probability about the mean of a sample of nn valuesX¯∼N(μ,σ2n)\bar X\sim N\!\left(\mu,\dfrac{\sigma^2}{n}\right), then a normal probability on the calculatorEnter the standard deviation σn\dfrac{\sigma}{\sqrt n}, not the variance
The smallest sample size for which a probability condition holdsStandardise X¯\bar X with the inverse normal value, then solve the inequality for n\sqrt nRound up to a whole number of items

Section 1 of 2 AHL4.17

Poisson distribution

A Poisson distribution models the number of independent events occurring in a fixed interval when they happen at a constant average rate. It is written X∼Po(λ)X\sim\text{Po}(\lambda), where λ\lambda is the mean number of events in the interval, and the variance is also λ\lambda. Probabilities are found on the calculator, after rescaling λ\lambda if the interval changes.

The formula and its conditions

The probability formula is in the formula booklet, but the mean and variance, E(X)=Var(X)=λE(X)=\text{Var}(X)=\lambda, are not, so learn them. If X∼Po(λ1)X\sim\text{Po}(\lambda_1) and Y∼Po(λ2)Y\sim\text{Po}(\lambda_2) are independent, then X+Y∼Po(λ1+λ2)X+Y\sim\text{Po}(\lambda_1+\lambda_2). You are also expected to justify when the model applies: the events are independent and occur at a uniform average rate. With a fixed number of trials, use the binomial distribution; with a continuous measurement, use the normal distribution.

Poisson probability

X∼Po(λ)⇒P(X=x)=e−λλxx!X \sim \text{Po}(\lambda) \Rightarrow P(X=x) = \dfrac{e^{-\lambda}\lambda^{x}}{x!}

λ\lambda is the mean number of occurrences in the interval you're working over - rescale it first if the question's interval differs from the one λ\lambda was quoted for.

Key terms

What does λ\lambda represent?

The parameter λ\lambda is the mean number of events in the interval being considered, and it is also the variance. So E(X)=Var(X)=λE(X)=\text{Var}(X)=\lambda.

e.g. A call centre receives on average 44 calls per hour, so the number of calls in one hour is modelled by X∼Po(4)X\sim\text{Po}(4).

How do you change the interval?

Scale λ\lambda in proportion to the length of the interval, since the rate stays constant. Always rewrite λ\lambda for the interval in the question before calculating.

e.g. If taxis arrive at 55 per hour, then λ=2.5\lambda=2.5 for 3030 minutes and λ=15\lambda=15 for 33 hours.

How do you find the probability of at least one event?

Use the complement, P(X≥1)=1−P(X=0)P(X\ge1)=1-P(X=0), because P(X=0)P(X=0) is quick to find. For a mean of λ\lambda, P(X=0)=e−λP(X=0)=e^{-\lambda}.

e.g. For X∼Po(2)X\sim\text{Po}(2), P(X≥1)=1−e−2≈0.865P(X\ge1)=1-e^{-2}\approx0.865.

How do you find the probability of at most kk events?

Add the probabilities for 0,1,…,k0, 1, \ldots, k, which the calculator does as a cumulative probability. Read the inequality carefully, because P(X≤k)P(X\le k) includes kk but P(X<k)P(X<k) does not.

e.g. For X∼Po(2)X\sim\text{Po}(2), P(X≤1)=e−2+2e−2=3e−2≈0.406P(X\le1)=e^{-2}+2e^{-2}=3e^{-2}\approx0.406.

Worked example, step by step

The question Calculator needed [6 marks]

A coastal rescue station receives call-outs at a mean rate of 0.8 per day. The number of call-outs on a day is modelled by a Poisson distribution.

(a) Find the probability of exactly 2 call-outs on a given day.

(b) Find the probability of at least 3 call-outs in a 5-day period.

(c) A volunteer says the model is unsuitable because call-outs come in clusters during storms. Explain which condition for a Poisson model this breaks.

Worked solution: worked example

Answer: (a) 0.1440.144 (b) 0.7620.762 (c) the rate is not uniform, because it is higher during storms

Step 1 - Name the distribution.

Let XX be the number of call-outs in one day. The mean is 0.8, so X∼Po(0.8)X\sim\text{Po}(0.8), and part (a) asks for P(X=2)P(X=2). M1 - Identify X∼Po(0.8)X\sim\text{Po}(0.8) and P(X=2)P(X=2)

Why? The parameter λ\lambda of a Poisson distribution is the mean number of events in the interval being asked about. Here the interval is one day, the same as the rate, so no scaling is needed.

Step 2 - Use the calculator for exactly 2.

Use the Poisson probability density function with λ=0.8\lambda=0.8 and x=2x=2. The calculator shows P(X=2)=e−0.8(0.8)22!=0.144.P(X=2)=\dfrac{e^{-0.8}(0.8)^2}{2!}=0.144. A1 - 0.1440.144

Why? "Exactly" means a single value, so the probability density function is the right one, not the cumulative one.

Step 3 - Scale the rate to 5 days.

The rate is constant, so over 5 days the mean number of call-outs is λ=0.8×5=4\lambda=0.8\times5=4. Let Y∼Po(4)Y\sim\text{Po}(4) be the number in 5 days. M1 - λ=0.8×5=4\lambda=0.8\times5=4

Why? Using λ=0.8\lambda=0.8 for a 5-day period is the most common error. Rewrite λ\lambda for the interval in the question first.

Step 4 - Choose the complement.

"At least 3" means 3 or more, so P(Y≥3)=1−P(Y≤2).P(Y\ge3)=1-P(Y\le2). M1 - Use P(Y≥3)=1−P(Y≤2)P(Y\ge3)=1-P(Y\le2)

Why? The cumulative probability handles 0,1,20,1,2 in one step, whereas "3 or more" has no upper end. The complement is shorter.

Step 5 - Use the calculator for at most 2.

Use the Poisson cumulative distribution function with λ=4\lambda=4 and upper value 2, which gives P(Y≤2)=0.2381…P(Y\le2)=0.2381\ldots. Then P(Y≥3)=1−0.2381…=0.762.P(Y\ge3)=1-0.2381\ldots=0.762. A1 - 0.7620.762

Why? Keep the unrounded value of P(Y≤2)P(Y\le2) in the calculator, then round only the final answer to three significant figures.

Step 6 - Judge the model.

Clusters of call-outs in storms mean the events do not occur at a uniform average rate, because the rate is higher during storms. They are probably also not independent of each other. The Poisson model may therefore be unsuitable. R1 - The rate is not uniform (higher during storms), or the events are not independent

Why? The two conditions for a Poisson model are that events are independent and that they occur at a uniform average rate. The reason must be tied to the context in the question.

Marks (6):
M1 Identify X∼Po(0.8)X\sim\text{Po}(0.8) and P(X=2)P(X=2)
A1 0.1440.144
M1 λ=0.8×5=4\lambda=0.8\times5=4
M1 Use P(Y≥3)=1−P(Y≤2)P(Y\ge3)=1-P(Y\le2)
A1 0.7620.762
R1 The rate is not uniform (higher during storms), or the events are not independent

Now you try: enter your answers to be marked

These are real practice questions. Type your answer in each box and press Mark it. The worked solution and mark scheme appear straight away, and your first attempt is the one that counts. Use the keyboard for powers, roots and trig, and the working box for your notes.

1
Easy
Calculator
[2 marks]
Calls arrive at λ=3\lambda = 3 per hour (Poisson).

Find P(X=2)P(X=2) in one hour.
[2]
Fractions & powers
Other

Worked solution: practice question 1

2
Hard
Calculator
[4 marks]
Flaws in a cable occur as a Poisson process. The probability that a 1-metre length has no flaws is 0.7.
(a) Find the mean number of flaws per metre.
[2]
(b) Find the probability that a 3-metre length has exactly 2 flaws.
[2]
Fractions & powers
Logs
Other

Worked solution: practice question 2

3
Medium
Calculator
[3 marks]
Flaws occur in cable at a mean rate of 2 per metre (Poisson).

Find the probability of exactly 5 flaws in 3 metres.
[3]
Fractions & powers
Other

Worked solution: practice question 3

Your score so far: 0 out of 9 marks

Common mistakes

  • Forgetting to rescale λ\lambda. If λ\lambda is given per hour but the question asks about a 3-hour or 30-minute period, λ\lambda must be scaled proportionally before it goes into the formula - using the original hourly value is one of the most common errors.
  • Confusing "exactly" with "at least" or "at most". P(X=x)P(X=x) uses the Poisson pdf directly, but "at least 1" needs 1−P(X=0)1 - P(X=0) and "at most" needs a cumulative sum - reaching for the wrong calculator function flips the whole answer.
  • Assuming a Poisson model without checking the conditions. The distribution only applies when events occur independently and at a uniform average rate; a quick check that the mean and variance of the data are close is good practice before committing to it.

Try some questions on poisson distribution

Section 2 of 2 AHL4.15

Central limit theorem

The central limit theorem says that the mean X¯\bar{X} of a random sample of size nn has mean μ\mu and variance σ2n\dfrac{\sigma^2}{n}, and that its distribution is approximately normal for large nn, whatever the shape of the population. You use it to turn a question about a sample mean into a normal probability calculation.

The sample mean distribution

If the population is normal, X¯\bar{X} is exactly normal for any nn. The central limit theorem extends this beyond normal populations: whatever distribution the population actually has, X¯\bar{X} approaches normality as nn grows, and the IB treats n>30n>30 as large enough to assume this in an exam.

Distribution of X¯\bar{X}

X¯∼N(μ,σ2n)\bar{X} \sim N\!\left(\mu,\, \dfrac{\sigma^2}{n}\right)

The mean of X¯\bar{X} is μ\mu, the mean of a single value; its variance is σ2\sigma^2 divided by nn, so bigger samples give a tighter spread.

Standardising X¯\bar{X}

Z=X¯−μσ/nZ = \dfrac{\bar{X}-\mu}{\sigma/\sqrt{n}}

Once you know the mean and variance of X¯\bar{X}, any probability about it is read straight off the normal distribution on the calculator. For a smallest-sample-size question, standardise with the inverse normal value for the given tail probability, then solve the inequality for n\sqrt{n} and round up.

Key terms

What is a sampling distribution of the mean?

It is the distribution of X¯\bar{X} over all possible random samples of the same size nn. Each sample gives one value of X¯\bar{X}, and these values vary from sample to sample.

e.g. If many samples of size 2525 are taken and each sample mean is recorded, the recorded means form the sampling distribution of X¯\bar{X}.

What is the standard deviation of the sample mean?

It is σn\dfrac{\sigma}{\sqrt{n}}, the square root of the variance σ2n\dfrac{\sigma^2}{n}. It is smaller than σ\sigma, so sample means are less spread out than individual values.

e.g. For σ=10\sigma=10 and n=25n=25, the standard deviation of X¯\bar{X} is 1025=2\dfrac{10}{\sqrt{25}}=2.

What is the difference between one observation and a sample mean?

A single observation XX has variance σ2\sigma^2, but the sample mean X¯\bar{X} has the smaller variance σ2n\dfrac{\sigma^2}{n}. Choose the distribution that matches what the question asks about.

e.g. For a normal population with σ=10\sigma=10, a single value has standard deviation 1010, but means of samples of size 2525 have standard deviation 22.

When is a sample size large enough?

For a population that is not normal, the normal model for X¯\bar{X} is reliable when nn is large, and the IB treats n>30n>30 as large. If the population is already normal, X¯\bar{X} is normal for any nn.

e.g. A skewed population sampled with n=50n=50 gives an approximately normal X¯\bar{X}, but n=5n=5 is generally too small.

Worked example, step by step

The question Calculator needed [6 marks]

The masses of loaves from a bakery have mean 400 g and standard deviation 12 g. A random sample of 40 loaves is taken, and X¯\bar X is the mean mass of the loaves in the sample.

(a) Find the mean and the variance of X¯\bar X.

(b) Explain why X¯\bar X can be modelled by a normal distribution even if the masses of individual loaves are not normally distributed.

(c) Find P(X¯<397)P(\bar X<397).

Worked solution: worked example

Answer: (a) mean 400400, variance 3.63.6 (b) n=40>30n=40>30 (c) 0.05690.0569

Step 1 - Find the mean and variance of the sample mean.

The mean of X¯\bar X is the mean of the masses, so E(X¯)=400E(\bar X)=400. The variance is Var(X¯)=σ2n=12240=14440=3.6.\text{Var}(\bar X)=\dfrac{\sigma^2}{n}=\dfrac{12^2}{40}=\dfrac{144}{40}=3.6. A1 - E(X¯)=400E(\bar X)=400 M1 - Use σ2n=14440\dfrac{\sigma^2}{n}=\dfrac{144}{40} A1 - Variance 3.63.6

Why? The sample mean is centred on the same value as one loaf, but it varies less, because the variance of one loaf is divided by nn.

Step 2 - Justify the normal model.

The sample size is n=40n=40, which is greater than 30. By the central limit theorem, X¯\bar X is therefore approximately normal, so X¯∼N(400,3.6)\bar X\sim N(400,\,3.6). R1 - n=40>30n=40>30, so X¯\bar X is approximately normal

Why? The syllabus treats n>30n>30 as large enough in examinations. If the loaf masses were stated to be normal, X¯\bar X would be normal for any nn.

Step 3 - Set up the probability.

The calculator needs the standard deviation, not the variance: σX¯=3.6=1.897…\sigma_{\bar X}=\sqrt{3.6}=1.897\ldots Use the normal cumulative distribution function with lower bound −1099-10^{99} (a very large negative number), upper bound 397, mean 400 and standard deviation 3.6\sqrt{3.6}. M1 - Normal probability with standard deviation 3.6\sqrt{3.6} and upper bound 397

Why? "Less than 397" has no lower limit, so a very large negative number stands in for minus infinity. Entering 3.6 as the standard deviation is a common error.

Step 4 - Read the probability.

The calculator shows P(X¯<397)=0.0569.P(\bar X<397)=0.0569. A1 - 0.05690.0569

Why? A sample mean 3 g below the mean is unusual, about 1.6 standard deviations of X¯\bar X below it, so a small probability is sensible. A single loaf would be far more likely to be this light.

Marks (6):
A1 E(X¯)=400E(\bar X)=400
M1 Use σ2n=14440\dfrac{\sigma^2}{n}=\dfrac{144}{40}
A1 Variance 3.63.6
R1 n=40>30n=40>30, so X¯\bar X is approximately normal
M1 Normal probability with standard deviation 3.6\sqrt{3.6} and upper bound 397
A1 0.05690.0569

Now you try: enter your answers to be marked

1
Easy
No calc
[2 marks]
A population has variance σ2=25\sigma^2 = 25.

For a sample of size n=100n = 100:
(a) State the variance of X¯\bar{X}.
[1]
(b) State the standard deviation of X¯\bar{X}.
[1]
Fractions & powers
Other

Worked solution: practice question 1

2
Medium
Calculator
[3 marks]
The daily output of a machine has mean 500500 units and standard deviation 3030 units. A sample of n=36n = 36 days is observed.

Find P(495<X¯<505).P(495 < \bar{X} < 505).
[3]
Fractions & powers
Other

Worked solution: practice question 2

3
Hard
Calculator
[4 marks]
A population has μ=100\mu = 100 and σ=20\sigma = 20.

Find the minimum sample size nn such that P(X¯>103)<0.05.P(\bar{X} > 103) < 0.05.
[4]
Fractions & powers
Other

Worked solution: practice question 3

Your score so far: 0 out of 9 marks

Common mistakes

  • Using σ2\sigma^2 instead of σ2/n\sigma^2/n as the variance. X¯\bar{X} is less spread out than a single observation - dividing by nn (or n\sqrt{n} if you're working with the standard deviation) is essential, not optional, and skipping it gives the wrong spread for every probability that follows.
  • Confusing XX with X¯\bar{X}. A single observation is still N(μ,σ2)N(\mu,\sigma^2); only the sample mean of nn observations gets the shrunken variance σ2/n\sigma^2/n. Read the question carefully to see which one is actually being asked about.
  • Forgetting the theorem needs a large nn for a non-normal population. If the population itself is stated to be normal, X¯\bar{X}'s distribution is exact for any nn. Otherwise, the central limit theorem gives only an approximation, once nn is large - the IB treats n>30n>30 as sufficient, so quote that threshold if a question asks you to justify using it.

Try some questions on central limit theorem

Using your GDC

The exact key presses for the TI-84 Plus CE, Casio fx-CG50, Casio fx-CG100 and TI-Nspire CX, with what the screen shows. Pick your calculator.

Poisson distribution

Used in: Poisson distribution

How to find a Poisson probability on a TI-84 Plus CE

poissonpdf and poissoncdf

For exactly 6, press 2nd vars (DISTR) and scroll down to C:poissonpdf(. With the Stat Wizards on, a screen asks for each value in turn: enter the mean μ 4 and the x value 6, then press enter on Paste.

poissonpdf(
μ:4
x value:6

Press enter to run it. If the Stat Wizards are off, type poissonpdf(4,6) with the mean first and press enter.

P(X=6)=0.104P(X = 6) = 0.104 to 3 significant figures.

poissonpdf(4,6)
0.1041956

For at most 6, press 2nd vars (DISTR) and choose D:poissoncdf(. Enter μ 4 and x value 6, or type poissoncdf(4,6), and press enter.

P(X≤6)=0.889P(X \le 6) = 0.889 to 3 significant figures.

poissoncdf(4,6)
0.8893260

For at least 5, subtract the probability of at most 4 from 1: type 1-poissoncdf(4,4) and press enter.

P(X≥5)=0.371P(X \ge 5) = 0.371 to 3 significant figures.

1-poissoncdf(4,4)
0.3711631

poissoncdf adds the probabilities from 0 up to x. It cannot start at a value other than 0, so for P(a≤X≤b)P(a \le X \le b) use poissoncdf(μ,b) − poissoncdf(μ,a − 1).

The full guide: Poisson distribution

Normal distribution probabilities

Used in: Central limit theorem

How to find a normal probability on a TI-84 Plus CE

normalcdf (between two bounds)

Press 2nd vars (DISTR) and choose 2:normalcdf(. With the Stat Wizards on, a screen asks for each value in turn: type it and press enter each time.

Enter the lower bound 140, the upper bound 165, the mean 150 and the standard deviation 12, then press enter on Paste.

normalcdf(
lower:140
upper:165
μ:150
σ:12

Press enter to run it. If the Stat Wizards are off, type normalcdf(140,165,150,12) with commas between the four values and press enter.

P(140<X<165)=0.692P(140 < X < 165) = 0.692 to 3 significant figures.

normalcdf(140,165,150,12)
0.6920218

For more than 170, there is no upper limit, so use a very large number as the upper bound: type 1 2nd , (EE) 9 9 for 1E99.

P(X>170)=0.0478P(X > 170) = 0.0478 to 3 significant figures.

normalcdf(170,1E99,150,12)
0.0477904

For less than a value, use −1E99 as the lower bound. If you give only the two bounds, the calculator uses mean 0 and standard deviation 1.

The full guide: Normal distribution probabilities

Inverse normal (find the value for a given probability)

Used in: Central limit theorem

How to find a value from a normal probability on a TI-84 Plus CE

invNorm (area to the left)

Press 2nd vars (DISTR) and choose 3:invNorm(. With the Stat Wizards on, a screen asks for each value in turn: type it and press enter each time.

Enter the area 0.9 (the probability to the left of the value), the mean 150 and the standard deviation 12, then press enter on Paste.

invNorm(
area:0.9
μ:150
σ:12

Press enter to run it. If the Stat Wizards are off, type invNorm(0.9,150,12) with commas between the three values and press enter.

P(X<k)=0.9P(X < k) = 0.9 gives k=165.4k = 165.4 to 4 significant figures.

invNorm(0.9,150,12)
165.3786188

For P(X<k)=0.25P(X < k) = 0.25, repeat with area 0.25: invNorm(0.25,150,12).

k=141.9k = 141.9 to 4 significant figures.

invNorm(0.25,150,12)
141.9061230

The area is always the probability to the left of the value. For the top 10%, use 1 − 0.1 = 0.9. If you give only the area, the calculator uses mean 0 and standard deviation 1.

The full guide: Inverse normal (find the value for a given probability)

See the full list of GDC guides for every calculator task.

Syllabus content covered

CodeSyllabus content
AHL4.15A linear combination of nn independent normal random variables is normally distributed. In particular, X∼N(μ,σ2)⇒X¯∼N(μ,σ2n)X\sim N(\mu,\sigma^2)\Rightarrow\bar X\sim N\!\left(\mu,\dfrac{\sigma^2}{n}\right). Central limit theorem. In general, X¯\bar X approaches normality for large nn; how large depends upon the distribution from which the sample is taken. In examinations, n>30n>30 will be considered sufficient.
AHL4.17Poisson distribution, its mean and variance. Sum of two independent Poisson distributions has a Poisson distribution. Situations in which it is appropriate to use a Poisson distribution as a model: events are independent; events occur at a uniform average rate (during the period of interest). Given a context, students should be able to select between the normal, the binomial and the Poisson distributions, recognizing where a particular distribution is appropriate.

Official IB source: Mathematics: applications and interpretation subject brief (PDF, ibo.org).

Quick questions

What is the formula for the Poisson distribution?

If X∼Po(λ)X \sim \text{Po}(\lambda), then P(X=x)=e−λλxx!P(X=x) = \dfrac{e^{-\lambda}\lambda^x}{x!}, where λ\lambda is the mean number of occurrences in the interval.

What does the central limit theorem say about the sample mean?

If the values have mean μ\mu and standard deviation σ\sigma, then the sample mean X¯\bar{X} of a random sample of size nn is distributed N(μ,σ2n)N\!\left(\mu,\dfrac{\sigma^2}{n}\right) - exactly if the population itself is normal, and approximately for any population once nn is large (the IB treats n>30n>30 as large enough).

How do I know when a Poisson distribution is an appropriate model?

Events must occur independently of each other and at a uniform average rate over the interval in question. A mean close to the variance of the data also supports a Poisson model.

Why does the variance of the sample mean divide by n?

Averaging nn independent observations cancels out random fluctuation: the variance of a sum of nn independent variables adds, but dividing that sum by nn to form a mean divides the variance by n2n^2, leaving σ2/n\sigma^2/n. Larger samples give a sample mean that clusters more tightly around μ\mu.

How do I choose between a Poisson and a binomial model?

Use a binomial model when there is a fixed number of trials, each with success or failure and the same probability. Use a Poisson model when events occur at an average rate in an interval with no fixed number of trials. For example, the number of heads in 1010 coin tosses is binomial, but the number of buses arriving at a stop in an hour is Poisson.

Does the central limit theorem change the mean?

No. The mean of X¯\bar{X} is still μ\mu; only the variance changes, from σ2\sigma^2 to σ2n\dfrac{\sigma^2}{n}.

What is the distribution of the total from two independent Poisson variables?

It is also Poisson, with the means added. If X∼Po(2)X\sim\text{Po}(2) and Y∼Po(3)Y\sim\text{Po}(3) are independent, then X+Y∼Po(5)X+Y\sim\text{Po}(5).

How do I find the sample size needed for a given probability?

Write X¯∼N(μ,σ2n)\bar{X}\sim N\!\left(\mu,\dfrac{\sigma^2}{n}\right), standardise to turn the probability condition into an inequality in nn, then solve it. Give a whole number of nn that satisfies the condition.

Is there an upper limit to a Poisson variable?

No. A Poisson variable can take any value 0,1,2,…0, 1, 2, \ldots, although very large values have very small probabilities. This differs from a binomial variable, which cannot exceed its number of trials nn.

Do I need to know the population is normal?

Not if nn is large, because the theorem makes X¯\bar{X} approximately normal anyway. For a small sample, the population must itself be normal for the normal model to be valid.

Ready to practise properly?

Practice questions on every section, marked instantly like the real exam.