Sums of Poisson variables and Poisson hypothesis tests

Sums of Poisson variables and Poisson hypothesis tests

Spec SB4–SB5 · AQA A-Level Further Maths: Revision notes

Studying this course? Keep your progress in one place.

Key facts

  • Independent Po(λ)\mathrm{Po}(\lambda) and Po(μ)\mathrm{Po}(\mu) add to Po(λ+μ)\mathrm{Po}(\lambda+\mu).
  • A Poisson test uses a single observation of the count; hypotheses are about λ\lambda, defined in context.
  • One-tailed: compare P(X≤x)\mathrm{P}(X\le x) or P(X≥x)\mathrm{P}(X\ge x) with the significance level.
  • Two-tailed: compare the tail probability with half the significance level.
  • Conclude in context: "evidence" or "insufficient evidence", never "proved".

Sums of independent Poisson variables

The sum of independent Poisson variables is Poisson, with the means added.

If X∼Po(λ)X\sim\mathrm{Po}(\lambda) and Y∼Po(μ)Y\sim\mathrm{Po}(\mu) are independent, then

X+Y∼Po(λ+μ).X+Y\sim\mathrm{Po}(\lambda+\mu).

This extends to any number of independent variables, and to counts over several periods: four independent weeks of Po(4)\mathrm{Po}(4) give Po(16)\mathrm{Po}(16).

State the combined distribution and its parameter before calculating. The rule is for totals only: X−YX-Y and 2X2X are not Poisson.

00.050.10.150.20.00530.00190.00060.00030.00270.01070123456789101112xP(X = x)
  • Po(4)
  • Po(4 + 4) = Po(8)
Two independent Po(4) counts add to Po(8): the distribution of the total sits further right and is more spread out.
  • Po(λ)+Po(μ)\mathrm{Po}(\lambda)+\mathrm{Po}(\mu)Po(λ+μ)\mathrm{Po}(\lambda+\mu)

Worked example

Server A gets Po(3)\mathrm{Po}(3) emails per minute and server B gets Po(2)\mathrm{Po}(2), independently. Find the probability that the total in a minute is 4.

X∼Po(1.5)X\sim\mathrm{Po}(1.5) and Y∼Po(2.5)Y\sim\mathrm{Po}(2.5) are independent. What is the distribution of X+YX+Y?

Setting up a hypothesis test

State H0\mathrm{H}_0 and H1\mathrm{H}_1 about λ\lambda in context, decided from the wording of the claim before seeing the data.

A hypothesis test about a Poisson mean λ\lambda uses a single observation xx of the count.

  • Null hypothesis H0:λ=λ0\mathrm{H}_0:\lambda=\lambda_0.
  • Alternative hypothesis H1\mathrm{H}_1: λ<λ0\lambda<\lambda_0 or λ>λ0\lambda>\lambda_0 (one-tailed), or λ≠λ0\lambda\ne\lambda_0 (two-tailed), chosen from the claim.

Define λ\lambda in context: "the mean number of complaints per week".

The significance level is the probability, assuming H0\mathrm{H}_0, below which a result is too unlikely to be chance.

  1. 1

    State hypotheses

    H0 and H1 about λ, defined in context

  2. 2

    Assume H0 is true

    use Po(λ0) for the count

  3. 3

    Find the probability

    of a result at least as extreme as observed

  4. 4

    Compare

    with the significance level (halved if two-tailed)

  5. 5

    Conclude in context

    reject or do not reject H0

Steps of a Poisson hypothesis test

A manager claims the mean number of complaints per week has fallen from 4. Which alternative hypothesis is correct?

One-tailed tests

Assume H0\mathrm{H}_0, find the probability of a result at least as extreme, and compare it with the significance level.

Assume H0\mathrm{H}_0 is true and find the probability of a result at least as extreme as the one observed in a one-tailed test.

  • H1:λ<λ0\mathrm{H}_1:\lambda<\lambda_0: find P(X≤x)\mathrm{P}(X\le x).
  • H1:λ>λ0\mathrm{H}_1:\lambda>\lambda_0: find P(X≥x)=1−P(X≤x−1)\mathrm{P}(X\ge x)=1-\mathrm{P}(X\le x-1).

If this probability is less than or equal to the significance level, reject H0\mathrm{H}_0.

00.050.10.150.20.250.30.00030.0030.01380123456xP(X ≤ x)
P(X ≤ x) for Po(8) under H₀: λ = 8. At the 5% level, P(X ≤ 3) = 0.0424 is below 0.05 but P(X ≤ 4) = 0.0996 is not.

H1:λ<λ0\mathrm{H}_1:\lambda<\lambda_0

Probability to find:
P(X≤x)\mathrm{P}(X\le x)
Reject if:
at most the level

H1:λ>λ0\mathrm{H}_1:\lambda>\lambda_0

Probability to find:
1−P(X≤x−1)1-\mathrm{P}(X\le x-1)
Reject if:
at most the level

Worked example

Complaints were thought to average 4 per week. After a change, 1 complaint is received in a week. Test at the 5% level whether the mean has fallen.

For H1:λ>3\mathrm{H}_1:\lambda>3 with observation x=7x=7, which probability do you compare with the significance level?

Two-tailed tests

The significance level is split between the two tails, so compare with half of it.

When H1:λ≠λ0\mathrm{H}_1:\lambda\ne\lambda_0 the change could be in either direction, so the significance level is split: half in each tail. This is a two-tailed test.

Find the tail probability on the side of the observation, then compare with half the significance level. At the 5% level compare with 0.0250.025.

00.020.040.060.080.10.120.140.00030.00270123456789101112131415xP(X = x)
Po(8), two-tailed 5% test: 2.5% in each tail gives X ≤ 2 (0.0138) or X ≥ 15 (0.0173). Illustrative values.
  • Two-tailed, 5% levelcompare with 0.025
  • Two-tailed, 10% levelcompare with 0.05

Worked example

H0:λ=2.5\mathrm{H}_0:\lambda=2.5, H1:λ≠2.5\mathrm{H}_1:\lambda\ne2.5. The observed count is 7. Test at the 5% level.

A two-tailed test is carried out at the 10% level. What should the tail probability be compared with?

Try an exam question

Complaints to a shop were thought to average 44 per week. After a change of staff, 11 complaint is received in a week. Test, at the 5%5\% level, whether the mean number of complaints per week has fallen.

[5 marks]

That's the notes covered.

Carry on to the next subtopic.

Try a question

Which of the following is the distribution of the total number of emails, X+YX+Y, arriving at the two servers in a minute?

1 mark, marked against the mark scheme

Revision notes