Testing for zero correlation

Testing for zero correlation

Spec 3.3 · Edexcel A-Level Further Maths: Revision notes

Studying this course? Keep your progress in one place.

Key facts

  • Hypotheses use the population coefficient: H0:ρ=0H_0:\rho=0 (or ρs=0\rho_s=0), never the sample value.
  • H1H_1 is ρ≠0\rho\ne0 (two-tailed), ρ>0\rho>0 or ρ<0\rho<0 (one-tailed).
  • For a two-tailed 5% test, use the one-tailed 2.5% column of the table.
  • Compare ∣r∣|r| (or rr in the right direction) with the critical value, then conclude in context.
  • Not rejecting H0H_0 means insufficient evidence, not proof of no correlation.

Hypotheses

H0H_0 says there is no correlation in the population; H1H_1 says what you are testing for.

A sample rr or rsr_s comes from sample data, but we want to know about the population, with coefficient ρ\rho (product moment) or ρs\rho_s (Spearman).

The null hypothesis is H0:ρ=0H_0:\rho=0 (or ρs=0\rho_s=0). The alternative hypothesis is ρ≠0\rho\ne0 for any correlation, ρ>0\rho>0 for positive, or ρ<0\rho<0 for negative.

Hypotheses are about the population parameter, never the sample value.

−1−0.8−0.6−0.4−0.200.20.40.60.81-1 perfect negative0: H₀ (none)1 perfect positive

Any correlation

  • H1:ρ≠0H_1:\rho\ne0
  • Two-tailed

Positive correlation

  • H1:ρ>0H_1:\rho>0
  • One-tailed

Negative correlation

  • H1:ρ<0H_1:\rho<0
  • One-tailed

A teacher tests whether revision time is positively correlated with marks. Which alternative hypothesis is correct?

Critical values

Find the row for nn and the column for the significance level, halving the level for a two-tailed test.

Tables give critical values for each sample size nn, for one-tailed tests at several levels.

For a two-tailed test at the 5% level, use the one-tailed 2.5% column. The critical region is ∣r∣≥|r|\ge the critical value for a two-tailed test, r≥r\ge it for H1:ρ>0H_1:\rho>0, and r≤−r\le- it for H1:ρ<0H_1:\rho<0. The significance level is the probability of rejecting H0H_0 when it is true.

−1−0.8−0.6−0.4−0.200.20.40.60.81-0.63190.6319

Product moment

Sample size n:
10
One-tailed 5%:
0.5494
Two-tailed 5%:
0.6319

Spearman

Sample size n:
8
One-tailed 5%:
0.6429
Two-tailed 5%:
0.7381

Product moment, n=10n=10, two-tailed test at 5%. Which critical value do you use?

Carrying out the test

State the hypotheses, find the critical value, compare, then conclude in context.

Follow the same four steps every time:

  1. State H0H_0 and H1H_1.
  2. Find the critical value for nn and the significance level.
  3. Compare the sample coefficient with it.
  4. Write a conclusion in two parts: whether to reject H0H_0, and what it means in context.
−1−0.8−0.6−0.4−0.200.20.40.60.81-0.6319r = -0.60.6319

Worked example

n=10n=10, r=−0.60r=-0.60. Test for any correlation at the 5% level.

n=10n=10, r=−0.60r=-0.60, test for negative correlation at 5%. Is H0H_0 rejected?

Product moment or Spearman?

The product moment test needs bivariate normality; Spearman's only needs a monotonic relationship.

The product moment test requires data from a bivariate normal distribution. Formal checking is not required, but the scatter should look roughly elliptical.

The Spearman test makes no such assumption and only needs a monotonic relationship, so it suits ranked data and small samples.

The tests can disagree near a critical value: for n=8n=8, rs=0.786r_s=0.786 exceeds 0.73810.7381 (reject H0H_0), but r=0.703r=0.703 is below 0.70670.7067 (do not reject).

0.511.522.53510152025xyy = x³
A monotonic but non-linear relationship (y = x³): the ranks agree perfectly, so Spearman's coefficient is 1, but the points are not on a straight line.

Product moment

Assumption:
Bivariate normal data
Data:
Elliptical scatter
Detects:
Linear association

Spearman

Assumption:
No normality needed
Data:
Ranks or measured data
Detects:
Monotonic association

Which test makes no assumption of bivariate normality?

Try an exam question

A sample of 8 students gives a product moment correlation coefficient of r=0.65r=0.65 between height and arm span. Test, at the 5% level, whether there is positive correlation in the population. The critical value is 0.62150.6215.

[4 marks]

That's the notes covered.

Carry on to the next subtopic.

Try a question

Which pair of hypotheses is correct for this test?

1 mark, marked against the mark scheme

Revision notes