Statistics: Hypothesis Testing
Hypothesis testing is a formal statistical method for using sample data to decide whether there is enough evidence to reject a claim (the null hypothesis) about a population parameter in favour of an alternative hypothesis. A Level Statistics covers binomial hypothesis tests for a proportion, finding critical regions, one-tailed and two-tailed tests, and testing the product moment correlation coefficient.
Before you start
Make sure you're comfortable with these topics first:
Method
- State the null hypothesis H0 (the claimed value of the parameter, e.g. p = 0.3) and the alternative hypothesis H1 (p > 0.3 for a one-tailed test, or p not equal to 0.3 for a two-tailed test).
- Identify the test statistic and its distribution under H0, e.g. X ~ B(n, p0) using the value of p claimed in H0.
- For a given significance level, find the critical region: the most extreme values of X whose total probability is as close as possible to (but not exceeding) the significance level, working from the relevant tail(s).
- Compare the observed value of X with the critical region: if it lies inside the critical region, reject H0; if not, there is insufficient evidence to reject H0.
- Always write the conclusion in context, referring back to the original claim, and state whether the result is significant at the given level.
- For a two-tailed test, split the significance level equally between both tails (e.g. use 2.5% at each end for a 5% two-tailed test) and find both critical regions.
Common mistakes
MistakeChoosing the tail from the observed result rather than from the alternative hypothesis stated in the question.
InsteadWrite H1 first. A less-than alternative uses the lower tail, a greater-than alternative uses the upper tail, and a not-equal alternative uses both tails.
MistakeUsing the sample proportion as the binomial parameter when calculating the critical region under H0.
InsteadUse the value claimed by H0 in the distribution, because the critical region describes results expected if the null hypothesis is true.
MistakeRejecting H0 merely because the observed value is unusual, without comparing it with the stated critical region or significance level.
InsteadState the critical region, locate the observed statistic, and then give a conclusion in context at the specified significance level.
Worked examples
Example 1
A machine is claimed to produce faulty components with probability 0.15. After a repair, the manager suspects the true probability has decreased. A random sample of 20 components is tested, and none are found to be faulty. Test, at the 5% significance level, whether there is evidence that the probability of a faulty component has decreased.
- State the hypotheses: H0: p = 0.15, H1: p < 0.15 (a one-tailed test, since the manager suspects a decrease).
- Under H0, X ~ B(20, 0.15). Find P(X = 0) = 0.85^20 = 0.0388 (4dp).
- Since this is a one-tailed lower-tail test, compare P(X <= 0) = P(X = 0) = 0.0388 with the significance level 0.05.
- 0.0388 < 0.05, so the observed value X = 0 lies in the critical region, and H0 is rejected.
Example 2
Under H0, X ~ B(16, 0.6). Test H0: p = 0.6 against H1: p is not equal to 0.6 at the 5% significance level. The observed value is X = 13. Find the critical regions and state the conclusion.
- This is a two-tailed test, so use 0.025 in each tail. Under H0, X ~ B(16, 0.6).
- For the lower tail, P(X <= 5) = 0.0191, while P(X <= 6) = 0.0583. Therefore the lower critical region is X <= 5.
- For the upper tail, P(X >= 14) = 0.0183, while P(X >= 13) = 0.0651. Therefore the upper critical region is X >= 14.
- The observed value X = 13 is in neither critical region, so H0 is not rejected at the 5% significance level.
- There is insufficient evidence at the 5% level that the population proportion differs from 0.6.
- Final answer: critical regions X <= 5 and X >= 14; do not reject H0 for X = 13.
Example 3
A hypothetical test uses H0: p = 0.4 against H1: p not equal to 0.4. Under H0, X ~ B(12, 0.4). For a nominal 5% two-tailed test, the supplied critical region is X <= 1 or X >= 9. Calculate the actual significance level as a percentage to 3 decimal places. Explain why neither 5% nor twice the lower-tail probability gives the actual level, and check why moving either boundary one integer towards the centre would exceed its allocated tail probability.
- Actual significance is P(reject H0 when H0 is true). Here rejection means X <= 1 or X >= 9. These events are disjoint, so add their probabilities under B(12, 0.4). The nominal 5% level allocates 0.025 to each tail.
- Calculate the lower tail: P(X <= 1) = 0.6^12 + 12 x 0.4 x 0.6^11 = 0.0196 (4dp). Keep the unrounded value for the final sum.
- Calculate the upper tail with the correct complement: P(X >= 9) = 1 - P(X <= 8) = 0.0153 (4dp). The two achieved tail probabilities are unequal, so doubling the lower-tail probability would count the wrong upper-tail probability.
- Add the unrounded probabilities: P(X <= 1) + P(X >= 9) = 0.0348583076 (10dp). Multiply by 100 to give an actual significance level of 3.486% (3 decimal places).
- Check the adjacent boundaries: P(X <= 2) = 0.0834 and P(X >= 8) = 0.0573 (both 4dp). Each exceeds 0.025, so neither tail can be enlarged by the next integer under the equal-tail allocation.
- X takes whole-number values, so enlarging a tail adds a whole point probability. The available tail sums need not equal 0.025. Here both achieved tails fall below that allowance, so their sum falls below the nominal 5% limit; the allowance is not an additional rejection event.
- Final answer: the actual significance level is 3.486%, compared with the nominal 5%. It is the sum of the two unequal rejection-tail probabilities under H0, not twice the lower-tail probability.
Practice questions
Q1[1 mark]State what is meant by the significance level of a hypothesis test.Show answer
Answer: The probability of incorrectly rejecting H0 when H0 is actually true.
Q2[1 mark]For a two-tailed test at the 10% significance level, state the significance level used in each tail.Show answer
Answer: 5% (10% divided equally between the two tails).
Q3[1 mark]A test uses H0: p = 0.2 against H1: p > 0.2. State what type of error is made if H0 is rejected when p is actually equal to 0.2.Show answer
Answer: A Type I error.
Q4[2 marks]Under H0, X ~ B(20, 0.3). Given that P(X >= 10) = 0.0480 (awrt) and P(X >= 9) = 0.1133 (awrt), state the critical region for a one-tailed test at the 5% significance level.Show answer
Answer: X >= 10 (the smallest upper-tail region with probability not exceeding 0.05).
Q5[4 marks]A gardener claims a new fertiliser has changed the proportion of seeds, historically 0.25, that fail to germinate. A random sample of 30 seeds is treated, and 2 fail to germinate. Given X ~ B(30, 0.25) under H0 and P(X <= 2) = 0.0106, test at the 5% significance level whether there is evidence of a change (two-tailed).Show answer
Answer: Compare 0.0106 with 0.025 (half of 5%): since 0.0106 < 0.025, reject H0; there is evidence the proportion has changed.
Q6[2 marks]A test statistic gives a p-value of 0.038. State, with a reason, the conclusion of the test at the 1% significance level and at the 5% significance level.Show answer
Answer: At 1%: do not reject H0, since 0.038 > 0.01. At 5%: reject H0, since 0.038 < 0.05.
Exam-style questions
Written in the style of a A Level Maths exam paper, with a full mark scheme.
Exam tips
- Write H0, H1 and the distribution under H0 before calculating any tail probability.
- For a two-tailed test, compare each tail with half the significance level and give both critical regions before judging the observation.
- Finish with the context and the significance level: say whether there is sufficient evidence for H1, rather than claiming that H0 has been proved true.
Q1[5 marks]A machine is set to produce ball bearings with a historic defect rate of 0.12. Following maintenance, the plant manager, Owen, suspects the defect rate has decreased. He tests a random sample of 25 ball bearings and finds that 1 is defective. Let X be the number of defective ball bearings, out of 25. (a) State suitable null and alternative hypotheses for Owen's test. (2) (b) Assuming X ~ B(25, 0.12), find P(X <= 1). (3)Show mark scheme
Tick each line you got.
Nothing ticked yet - 5 available
Q2[6 marks]A call centre historically finds that 18% of outbound calls result in a sale. Following a new script, the manager wants to test whether the proportion has changed (either increased or decreased). A random sample of 30 calls is monitored. Let X be the number of calls, out of 30, that result in a sale. (a) State suitable hypotheses for a two-tailed test. (2) (b) Given that, under H0, P(X <= 1) = 0.0197 and P(X <= 2) = 0.0741 (both awrt), find the lower critical region for a two-tailed test at the 10% significance level. (2) (c) State the actual significance level contributed by the lower tail. (1) (d) In the sample, 8 of the 30 calls resulted in a sale. Given that the upper critical region is X >= 10, state, with a reason, the conclusion of the test. (1)Show mark scheme
Tick each line you got.
Nothing ticked yet - 6 available
Q3[4 marks]For 10 randomly selected towns, the annual rainfall (x, mm) and average crop yield (y, tonnes per hectare) are recorded. The product moment correlation coefficient is calculated to be r = 0.712. (a) State suitable hypotheses to test, at the 5% significance level, whether there is positive correlation between rainfall and crop yield. (2) (b) Given that the critical value for a one-tailed test at the 5% level, with n = 10, is 0.5494, determine whether there is significant evidence of positive correlation. Justify your answer. (2)Show mark scheme
Tick each line you got.
Nothing ticked yet - 4 available
This topic in real past papers
We have tagged some boards' past papers question by question (not every board or year yet). This topic comes up in:
- OCR A Level Maths past papers: 8 questions across 6 papers, latest tagged June 2024
- OCR AS Level Maths past papers: 6 questions across 6 papers, latest tagged June 2023
- WJEC A Level Maths past papers: 6 questions across 6 papers, latest tagged Summer 2024
- AQA A Level Maths past papers: 4 questions across 4 papers, latest tagged June 2023
Free printable worksheet
Want more practice on paper? Download the Statistics: Hypothesis Testing worksheet pack - 8 pages of exam-style questions with a full mark scheme. Free to download, no card or account. Print it for personal and classroom use.
This topic is chapter 5 of A Level Maths Statistics Essentials, the whole course as one free printable PDF. Its worksheet is also chapter 5 of A Level Maths Statistics Topic Workbook and chapter 5 of A Level Maths 10-Minute Tests 2, free printable practice books.
Other versions of this worksheet: