Hypothesis tests for correlation

Hypothesis tests for correlation

Spec O1 · AQA A-Level Maths: Revision notes

Studying this course? Keep your progress in one place.

Key facts

  • The product moment correlation coefficient rr lies between −1-1 and 11 and measures linear association only.
  • Hypotheses use the population coefficient ρ\rho: H0:ρ=0H_0:\rho=0.
  • Reject H0H_0 if rr is beyond the critical value for the sample size, level and tail.
  • With a pp-value: reject H0H_0 if pp is less than the significance level.
  • Correlation does not imply causation; conclude in context.

Correlation coefficients

rr measures how close points lie to a straight line: close to ±1\pm1 is strong, close to 00 is weak.

The product moment correlation coefficient rr measures how close the points on a scatter diagram lie to a straight line, with −1≤r≤1-1\le r\le1. Close to 11 is strong positive correlation, close to −1-1 strong negative, and close to 00 little or no linear correlation.

You do not calculate rr in this topic: it is given. It measures linear association only, so r≈0r\approx0 does not rule out a non-linear relationship.

−1−0.8−0.6−0.4−0.200.20.40.60.81-1: perfect negative0: none1: perfect positive

Which value of rr shows the strongest linear correlation?

Hypotheses about the population

H0:ρ=0H_0:\rho=0; H1H_1 is one-tailed (ρ>0\rho>0 or ρ<0\rho<0) if the direction is known, else ρ≠0\rho\ne0.

A sample rr varies from sample to sample, so we test a claim about the population correlation coefficient ρ\rho. The null hypothesis is that there is no correlation: H0:ρ=0H_0:\rho=0.

The alternative is one-tailed if the direction is known in advance, H1:ρ>0H_1:\rho>0 or ρ<0\rho<0, and two-tailed if any correlation is of interest, H1:ρ≠0H_1:\rho\ne0. Choose the direction from the question wording, before looking at rr. The test assumes a random sample.

−1−0.8−0.6−0.4−0.200.20.40.60.81H₀: ρ = 0
H₀: ρ = 0 sits at the centre; a two-tailed H₁ (ρ ≠ 0) covers both sides, a one-tailed H₁ only one
  • H0H_0ρ=0\rho=0
  • H1H_1 one-tailedρ>0\rho>0 or ρ<0\rho<0
  • H1H_1 two-tailedρ≠0\rho\ne0

A teacher wants to test whether revision time is positively correlated with test score. Which is H1H_1?

Critical value method

Compare rr with the critical value for the sample size, significance level and type of test.

A table gives the critical value of rr for each sample size nn, significance level and type of test.

  • one-tailed upper test: reject H0H_0 if r≥r\ge critical value;
  • one-tailed lower test: reject H0H_0 if r≤−r\le- critical value;
  • two-tailed test: reject H0H_0 if ∣r∣≥\lvert r\rvert\ge critical value (use the two-tailed table).

Use the number of pairs for nn, and check you are using the right table.

−1−0.8−0.6−0.4−0.200.20.40.60.810.4973r = 0.62

Worked example

n=12n=12, r=0.62r=0.62, H1:ρ>0H_1:\rho>0 at the 5%5\% level. The critical value is 0.49730.4973. Carry out the test.

n=12n=12, H1:ρ<0H_1:\rho<0 at 5%5\%, and r=−0.41r=-0.41. The critical value is 0.49730.4973. What is the decision?

The p-value method

Compare the pp-value with the significance level: smaller means reject H0H_0.

If a p-value is given, it is the probability of a sample correlation at least this extreme if ρ=0\rho=0. Compare it with the significance level (halving the level for a two-tailed test if the pp-value is for one tail). The smaller the pp-value, the stronger the evidence against H0H_0.

00.020.040.060.080.10.120.140.160.180.2p = 0.03level 0.05
Example: a p-value of 0.03 is below the 5% level (0.05), so reject H₀
  • p<p< significance levelreject H0H_0
  • p≥p\ge significance leveldo not reject H0H_0

Worked example

For r=−0.47r=-0.47 with n=10n=10 and H1:ρ<0H_1:\rho<0, the pp-value is 0.0850.085. What is the decision at 5%5\% and at 10%10\%?

p=0.03p=0.03 at the 5%5\% level (one-tailed). What is the decision?

Conclusions and cautions

State the decision in context, and add cautions: correlation is not causation, and rr is only linear.

State the decision and meaning in context: 'there is sufficient evidence, at the 5%5\% level, of positive correlation between hours revising and test score'. If H0H_0 is not rejected, say there is insufficient evidence of correlation. Do not say H0H_0 is proved.

Cautions: correlation does not imply causation (another variable may explain both); rr measures only a linear relationship; extrapolation beyond the data is unreliable; a one-tailed test is only valid if the direction was decided before seeing the data.

  1. 1

    Hypotheses

    H₀: ρ = 0 and H₁ from the question

  2. 2

    Find critical value

    from n, the level and the type of test

  3. 3

    Compare

    sample r against the critical value

  4. 4

    Decide

    reject or do not reject H₀

  5. 5

    Conclude

    in context, with a caution if needed

Correlation test in five steps

A test finds significant positive correlation between ice cream sales and drownings. What is the best comment?

Try an exam question

A teacher collects 1515 pairs of data on hours of revision and test score and finds r=0.55r=0.55. The critical value for a 5%5\% one-tailed test with n=15n=15 is 0.44090.4409. Test whether there is positive correlation, and state your conclusion in context.

[4 marks]

That's the notes covered.

Carry on to the next subtopic.

Try a question

Which pair of hypotheses is correct for the teacher's test?

1 mark, marked against the mark scheme

Revision notes