Hypothesis tests for binomial and Poisson

Hypothesis tests for binomial and Poisson

Spec 4.6 · Edexcel International A Level Further Maths: Revision notes

Studying this course? Keep your progress in one place.

Key facts

  • Binomial test: under H0:p=p0\mathrm{H}_0:p=p_0, X∼B(n,p0)X\sim\mathrm{B}(n,p_0). Poisson test: under H0:λ=λ0\mathrm{H}_0:\lambda=\lambda_0, X∼Po(λ0)X\sim\mathrm{Po}(\lambda_0).
  • Scale a Poisson mean to the interval observed.
  • Use the critical region method or the probability method; both give the same decision.
  • In a two-tailed test compare each tail with half the significance level.
  • Conclude in context: "sufficient" or "insufficient evidence".

Setting up the test

The number of successes (binomial) or events (Poisson), with its distribution under H0\mathrm{H}_0.

To test a binomial proportion pp, the test statistic XX is the number of successes in nn trials. Under H0:p=p0\mathrm{H}_0:p=p_0, X∼B(n,p0)X\sim\mathrm{B}(n,p_0).

To test a Poisson mean λ\lambda, XX is the number of events in a fixed interval. Under H0:λ=λ0\mathrm{H}_0:\lambda=\lambda_0, X∼Po(λ0)X\sim\mathrm{Po}(\lambda_0).

Scale the mean to the interval observed: 2.52.5 per hour becomes λ=10\lambda=10 for 44 hours.

  1. 1

    State the hypotheses

    H0\mathrm{H}_0 and H1\mathrm{H}_1 about pp or λ\lambda

  2. 2

    Give the distribution

    Under H0\mathrm{H}_0

  3. 3

    Find the region or probability

    Critical region or p-value

  4. 4

    Compare

    With the significance level

  5. 5

    Conclude

    In context

A test, step by step

A shop receives calls at a mean rate of 33 per hour. A test is based on a 55-hour period. What is the Poisson mean under H0\mathrm{H}_0?

Binomial tests

Find the critical region, or the probability of a result at least as extreme, and compare with the significance level.

The critical region method finds the largest region in the relevant tail with probability at most the significance level. The probability method finds the probability of a result at least as extreme as observed and compares it with the level.

Tables usually stop at p=0.5p=0.5. For p>0.5p>0.5, let Y=n−X∼B(n,1−p)Y=n-X\sim\mathrm{B}(n,1-p). For B(20,0.7)\mathrm{B}(20,0.7), P(X≤10)=P(Y≥10)=1−P(Y≤9)=0.0480\mathrm{P}(X\le10)=\mathrm{P}(Y\ge10)=1-\mathrm{P}(Y\le9)=0.0480.

00.050.10.150.20.0040.0078910111213141516171819rProbability
X∼B(20,0.7)X\sim\mathrm{B}(20,0.7). The lower tail X≤10X\le10 is the critical region at 5%5\%.

Worked example

X∼B(20,0.7)X\sim\mathrm{B}(20,0.7) and H1:p<0.7\mathrm{H}_1:p<0.7 at the 5%5\% level. Find the critical region. If 1010 seeds germinate, what is the conclusion?

For X∼B(30,0.4)X\sim\mathrm{B}(30,0.4) and H1:p>0.4\mathrm{H}_1:p>0.4, which probability do you calculate for an observed value x=18x=18?

Poisson tests

Define λ\lambda for the interval observed, then find the tail probability.

The hypotheses must define λ\lambda for the interval observed, and the probability uses the tail that H1\mathrm{H}_1 points to.

Calls arrive at a mean of 2.52.5 per hour. In 44 hours there are 1616 calls, and we test at 5%5\% whether the Poisson mean has increased.

00.020.040.060.080.10.120.0070.00445678910111213141516171819rProbability
X∼Po(10)X\sim\mathrm{Po}(10). The upper tail X≥16X\ge16 has probability 0.04870.0487.

Worked example

Calls arrive at a mean of 2.52.5 per hour. In 44 hours there are 1616 calls. Test at the 5%5\% level whether the rate has increased.

If X∼Po(10)X\sim\mathrm{Po}(10) and P(X≤15)=0.9513\mathrm{P}(X\le15)=0.9513, what is P(X≥16)\mathrm{P}(X\ge16)?

Two-tailed tests

The significance level is shared between the two tails.

For H1:p≠p0\mathrm{H}_1:p\ne p_0 or λ≠λ0\lambda\ne\lambda_0, the significance level is shared, so at 5%5\% each tail has 2.5%2.5\%.

Find the tail probability on the same side as the observed value and compare it with 0.0250.025, not 0.050.05. Alternatively, double that probability and compare with 0.050.05.

00.050.10.150.20.00020.00290.00290.00020123456789101112XP(X = x)
B(12, 0.5): each tail of the 5% two-tailed critical region (X ≤ 2 and X ≥ 10) has probability 0.0193, below 0.025
  • Two-tailed, 5% leveleach tail 0.025
  • Two-tailed, 10% leveleach tail 0.05

In a two-tailed test at the 10%10\% level, what should a one-tail probability be compared with?

Large samples and conclusions

For large nn use a normal approximation with a continuity correction, then conclude in context.

For large nn, such as 150150, use X≈N(np0,np0q0)X\approx\mathrm{N}(np_0,np_0q_0) under H0\mathrm{H}_0 with a continuity correction, checking npnp and nqnq exceed 55. For a critical region, solve an inequality like c+0.5−μσ<−1.645\frac{c+0.5-\mu}{\sigma}<-1.645 and take the largest integer cc.

Always state whether H0\mathrm{H}_0 is rejected, then say "sufficient" or "insufficient evidence" in the language of the question. Never say the claim is proved.

4045505560657075800.010.020.030.040.050.060.070.08xy71.5N(60, 36)
N(60,36)\mathrm{N}(60,36): P(X≥72)≈P(Y>71.5)\mathrm{P}(X\ge72)\approx\mathrm{P}(Y>71.5) is the shaded upper tail.

Worked example

X∼B(150,0.4)X\sim\mathrm{B}(150,0.4) under H0\mathrm{H}_0. The observed value is X=72X=72 and H1:p≠0.4\mathrm{H}_1:p\ne0.4 at the 5%5\% level. What is the conclusion?

Why is a continuity correction needed when approximating B(150,0.4)\mathrm{B}(150,0.4) by a normal distribution?

Try an exam question

A call centre receives calls at a mean rate of 66 per hour. After a change of staffing, 22 calls are received in one hour. Test, at the 5%5\% level, whether the mean rate has decreased.

[5 marks]

That's the notes covered.

Carry on to the next subtopic.

Try a question

Assuming H0\mathrm{H}_0 is true, what is P(X≤10)\mathrm{P}(X\le10)?

1 mark, marked against the mark scheme

Revision notes