Hypothesis test for the mean of a Poisson distribution

Hypothesis test for the mean of a Poisson distribution

Spec 4.1 · Edexcel A-Level Further Maths: Revision notes

Studying this course? Keep your progress in one place.

Key facts

  • The null hypothesis H0:λ=λ0\mathrm{H}_0:\lambda=\lambda_0 and the alternative hypothesis H1\mathrm{H}_1 are about the parameter λ\lambda, never about XX.
  • Scale λ\lambda to the length of the observation period before you start.
  • Under H0\mathrm{H}_0, X∼Po(λ0)X\sim\mathrm{Po}(\lambda_0). Use P(X≥x)=1−P(X≤x−1)\mathrm{P}(X\geq x)=1-\mathrm{P}(X\leq x-1) for an increase and P(X≤x)\mathrm{P}(X\leq x) for a decrease.
  • For a two-tailed test put half the significance level in each tail.
  • Conclude in context: reject H0\mathrm{H}_0, or there is insufficient evidence; never "proved".

The Poisson setting

A single observed count is used to decide whether the mean rate λ\lambda has changed from its stated value.

The Poisson distribution X∼Po(λ)X\sim\mathrm{Po}(\lambda) models the number of events in a fixed interval of time or space. It needs events that occur randomly, independently and at a constant average rate. The parameter λ\lambda is the mean number of events in the interval.

A hypothesis test for the mean uses a single observed count to decide whether the population parameter λ\lambda has changed from its stated value.

If the sample covers several intervals, scale the mean first. A rate of 1.21.2 errors per page gives Po(6)\mathrm{Po}(6) for 55 pages.

00.050.10.150.20.0081012345678rP(X = r)
X ~ Po(3): P(X = r), to 4 d.p. Events occur randomly at a constant average rate.
  • Rate λ\lambda per interval, tt intervalsX∼Po(tλ)X\sim\mathrm{Po}(t\lambda)

Calls arrive at a mean rate of 33 per hour. What is the distribution for a 44-hour period?

Stating the hypotheses

Write H0:λ=λ0\mathrm{H}_0:\lambda=\lambda_0 and an alternative in terms of the parameter λ\lambda.

The null hypothesis says the parameter has its stated value: H0:λ=λ0\mathrm{H}_0:\lambda=\lambda_0. The alternative hypothesis H1\mathrm{H}_1 says how it differs. Write both in terms of λ\lambda, not the random variable XX or the observed value.

Define λ\lambda in context, for example "the mean number of calls per hour". Then state the distribution of the test statistic under H0\mathrm{H}_0, for instance X∼Po(3)X\sim\mathrm{Po}(3).

Writing H0:X=3\mathrm{H}_0:X=3 is a common error: hypotheses are about the parameter.

012345678910λ₀ = 3

One-tailed test

  • Increase: H1:λ>λ0\mathrm{H}_1:\lambda>\lambda_0
  • Decrease: H1:λ<λ0\mathrm{H}_1:\lambda<\lambda_0

Two-tailed test

  • Change only: H1:λ≠λ0\mathrm{H}_1:\lambda\neq\lambda_0

A shop usually gets 5 complaints a day. A manager thinks the mean has fallen. Which alternative hypothesis is correct?

One-tailed tests

Find the probability of a result at least as extreme as the observation and compare it with the significance level.

Assume H0\mathrm{H}_0 is true, so X∼Po(λ0)X\sim\mathrm{Po}(\lambda_0). Find the probability of a result at least as extreme as the one observed and compare it with the significance level. If it is smaller, reject H0\mathrm{H}_0.

The critical region is the set of values that leads to rejecting H0\mathrm{H}_0. For X∼Po(6)X\sim\mathrm{Po}(6) and H1:λ<6\mathrm{H}_1:\lambda<6 at 5%5\%: P(X≤1)=0.0174<0.05\mathrm{P}(X\leq1)=0.0174<0.05 but P(X≤2)=0.0620>0.05\mathrm{P}(X\leq2)=0.0620>0.05, so the critical region is X≤1X\leq1. The actual significance level is 0.01740.0174.

00.020.040.060.080.10.120.140.160.0025012345678xP(X = x)
Probabilities for Po(6). The critical region for H1: λ < 6 at 5% is X ≤ 1.
  • H1:λ>λ0\mathrm{H}_1:\lambda>\lambda_0P(X≥x)=1−P(X≤x−1)\mathrm{P}(X\geq x)=1-\mathrm{P}(X\leq x-1)
  • H1:λ<λ0\mathrm{H}_1:\lambda<\lambda_0P(X≤x)\mathrm{P}(X\leq x)

Worked example

Calls arrive at a mean rate of 33 per hour. In one hour 77 calls are received. Test at the 5%5\% level whether the rate has increased.

For X∼Po(6)X\sim\mathrm{Po}(6), P(X≤3)=0.1512\mathrm{P}(X\leq3)=0.1512. Is X=3X=3 in the critical region for H1:λ<6\mathrm{H}_1:\lambda<6 at 5%5\%?

Two-tailed tests

Split the significance level equally between the two tails.

For H1:λ≠λ0\mathrm{H}_1:\lambda\neq\lambda_0, split the significance level equally between the two tails. At 10%10\% use 5%5\% in each tail: the lower tail P(X≤c1)≤0.05\mathrm{P}(X\leq c_1)\leq0.05 and the upper tail P(X≥c2)≤0.05\mathrm{P}(X\geq c_2)\leq0.05.

For an observed value in the upper tail, compare P(X≥x)\mathrm{P}(X\geq x) with 0.050.05 (or double it and compare with 0.100.10).

Two-tailed critical regions have two parts. Give both, and add the two tail probabilities to get the actual significance level.

00.020.040.060.080.10.120.0023234567891011121314151617rP(X = r)
X ~ Po(10), r = 2 to 17, 10% two-tailed: P(X ≤ 4) = 0.0293 and P(X ≥ 16) = 0.0487, so reject H₀ if X ≤ 4 or X ≥ 16.
  • Significance level α\alpha, two-tailedα/2\alpha/2 in each tail

Worked example

X∼Po(10)X\sim\mathrm{Po}(10) under H0\mathrm{H}_0 and H1:λ≠10\mathrm{H}_1:\lambda\neq10 at the 10%10\% level. Find the critical region and the actual significance level.

A two-tailed test is carried out at the 5% level. What probability goes in each tail?

Writing the conclusion

State what happens to H0\mathrm{H}_0, then write a sentence in context.

Finish in two steps. First state what happens to H0\mathrm{H}_0 ("reject H0\mathrm{H}_0" or "do not reject H0\mathrm{H}_0"). Then write a sentence in context, such as "There is sufficient evidence, at the 5%5\% level, that the mean number of calls per hour has increased."

If the result is not significant, say there is insufficient evidence that the rate has changed, rather than stating that the rate is unchanged. A non-significant result does not prove H0\mathrm{H}_0.

If asked to comment on the model: bursts or a varying rate break the Poisson conditions, so counts vary more than expected and the test may be unreliable.

  1. 1

    Hypotheses

    State H0\mathrm{H}_0 and H1\mathrm{H}_1 using λ\lambda, defined in context

  2. 2

    Distribution

    Write X∼Po(λ0)X\sim\mathrm{Po}(\lambda_0) under H0\mathrm{H}_0

  3. 3

    Probability

    Find the tail probability or critical region

  4. 4

    Compare

    Compare with the significance level

  5. 5

    Conclude

    Reject or do not reject, then a sentence in context

Structure of a hypothesis test.

The p-value is 0.12 at the 5% level. Which conclusion is best?

Try an exam question

Calls to a helpline arrive at a mean rate of 44 per hour. In one hour 99 calls are received. Test, at the 5%5\% significance level, whether the rate of calls has increased.

[5 marks]

That's the notes covered.

Carry on to the next subtopic.

Try a question

Which of the following gives the correct hypotheses, where λ\lambda is the mean number of calls per hour?

1 mark, marked against the mark scheme

Revision notes