Regression and Correlation A Level Maths Revision Guide
Master regression and correlation for A-level Maths with PMCC, regression lines, hypothesis tests, worked examples, exam questions, answers and revision tips.

Regression and correlation A level maths questions examine how two variables are related, how to model that relationship and whether sample evidence supports correlation in the population. You need to interpret scatter diagrams, understand the product moment correlation coefficient, use regression lines and carry out hypothesis tests.
This guide explains these skills through regression and correlation worked examples, common mistakes and an exam-style question with answers.
What are regression and correlation?
Bivariate data consists of paired observations for two variables, such as each student's revision time and test score.
The explanatory variable is used to explain or predict the response variable. These are also called the independent and dependent variables.
| Concept | Purpose |
|---|---|
| Correlation | Measures the direction and strength of a linear relationship |
| Regression | Models the relationship to estimate one variable from the other |
Strong correlation does not prove causation. Another factor may influence both variables, or the association may be coincidental.
Interpreting scatter diagrams
A scatter diagram plots each pair of observations as a point. Inspect its direction, form and unusual values before interpreting numerical results.
- Positive correlation: larger values of one variable tend to accompany larger values of the other.
- Negative correlation: larger values of one variable tend to accompany smaller values of the other.
- Strong linear correlation: points lie close to a straight line.
- Weak linear correlation: points are widely scattered around a straight line.
Look for outliers, clusters and curved patterns. A clear curve can have a small correlation coefficient, so little linear correlation does not necessarily mean no relationship.
Separate clusters may represent different groups. Investigate their context before treating them as one population.
The product moment correlation coefficient
The product moment correlation coefficient, or PMCC, measures linear association. The sample coefficient is rrr; the population coefficient is ρ\rhoρ.
Its value satisfies
−1≤r≤1.-1\leq r\leq1.−1≤r≤1.The sign indicates direction; ∣r∣|r|∣r∣ indicates strength.
| Value of rrr | Interpretation |
|---|---|
| Close to 111 | Strong positive linear correlation |
| Close to −1-1−1 | Strong negative linear correlation |
| Close to 000 | Weak or no linear correlation |
| Exactly 111 or −1-1−1 | Perfect linear correlation |
There are no universal boundaries separating weak, moderate and strong correlation. Interpret the coefficient alongside the scatter diagram.
One formula is
r=SxySxxSyy,r=\frac{S_{xy}}{\sqrt{S_{xx}S_{yy}}},r=SxxSyySxy,where
Sxx=∑x2−(∑x)2n,Syy=∑y2−(∑y)2n,S_{xx}=\sum x^2-\frac{(\sum x)^2}{n},\qquad S_{yy}=\sum y^2-\frac{(\sum y)^2}{n},Sxx=∑x2−n(∑x)2,Syy=∑y2−n(∑y)2,and
Sxy=∑xy−∑x∑yn.S_{xy}=\sum xy-\frac{\sum x\sum y}{n}.Sxy=∑xy−n∑x∑y.A calculator can find rrr from paired data in statistics mode. Check that the lists have equal lengths and that each pair remains correctly matched.
Regression lines
A linear regression model predicting yyy from xxx is
y=a+bx,y=a+bx,y=a+bx,where bbb is the gradient and aaa the intercept. Here, xxx is explanatory and yyy is the response.
For the least-squares regression line of yyy on xxx,
b=SxySxx,a=yˉ−bxˉ.b=\frac{S_{xy}}{S_{xx}},\qquad a=\bar{y}-b\bar{x}.b=SxxSxy,a=yˉ−bxˉ.The line therefore passes through (xˉ,yˉ)(\bar{x},\bar{y})(xˉ,yˉ). Least squares minimises the sum of squared vertical differences between observed and predicted responses. Rearranging this line does not generally give the regression line of xxx on yyy.
Your specification may emphasise interpreting supplied or calculator-generated coefficients rather than calculating them manually. Follow your exam board's requirements.
Interpreting coefficients
The gradient gives the predicted change in the response for a one-unit increase in the explanatory variable. Include both variables and their units when interpreting it.
The intercept predicts yyy when x=0x=0x=0. It may have no practical meaning if zero is impossible or outside the observed range.
Regression and correlation worked example
Revision time xxx hours and test score yyy marks have summary statistics
xˉ=4,yˉ=62,Sxx=10,Sxy=35,Syy=140.\bar{x}=4,\quad\bar{y}=62,\quad S_{xx}=10,\quad S_{xy}=35,\quad S_{yy}=140.xˉ=4,yˉ=62,Sxx=10,Sxy=35,Syy=140.Find the regression line and predict
Calculate the gradient and intercept:
b=3510=3.5,b=\frac{35}{10}=3.5,b=1035=3.5, a=62−(3.5)(4)=48.a=62-(3.5)(4)=48.a=62−(3.5)(4)=48.The regression line is
y=48+3.5x.y=48+3.5x.y=48+3.5x.The model predicts an increase of 3.53.53.5 marks per additional hour of revision.
For 555 hours,
y=48+3.5(5)=65.5.y=48+3.5(5)=65.5.y=48+3.5(5)=65.5.The predicted score is 65.565.565.5 marks. Retain precision during calculations before rounding appropriately for the context.
Calculate and interpret the PMCC
r=3510×140≈0.935.r=\frac{35}{\sqrt{10\times140}}\approx0.935.r=10×14035≈0.935.There is strong positive linear correlation between revision time and score in this sample. This does not establish causation, since other variables may affect performance.
Interpolation and extrapolation
Interpolation predicts for an explanatory-variable value within the observed range. Nearby observations generally make this more reliable.
Extrapolation predicts outside that range. The relationship may change beyond the data, so explain this limitation whenever assessing a prediction's reliability.
A high ∣r∣|r|∣r∣ does not justify unlimited extrapolation. Even interpolation requires a suitable model: a strong curved pattern or influential outlier can undermine a linear prediction.
Hypothesis testing for correlation
A hypothesis test assesses evidence about population correlation using the sample PMCC. Write hypotheses using ρ\rhoρ, not rrr.
For positive correlation:
H0:ρ=0,H1:ρ>0.H_0:\rho=0,\qquad H_1:\rho>0.H0:ρ=0,H1:ρ>0.For negative correlation, use H1:ρ<0H_1:\rho<0H1:ρ<0; for either direction, use H1:ρ≠0H_1:\rho\neq0H1:ρ=0. Choose the alternative from the question before examining the result.
Standard PMCC critical-value tests assume independent paired observations from a bivariate normal population. Use the stated modelling assumptions and consider whether the sampling method supports population conclusions.
Worked hypothesis test
A sample of 101010 pairs gives r=0.680r=0.680r=0.680. The supplied critical value for a one-tailed positive test at the 5%5\%5% significance level is 0.5490.5490.549.
Since
0.680>0.549,0.680>0.549,0.680>0.549,reject H0H_0H0. There is sufficient evidence at the 5%5\%5% significance level to suggest positive population correlation. Name the variables in your conclusion.
For a negative test, use the negative critical boundary. For a two-tailed test, use the appropriate two-tailed threshold. Select the correct sample size and significance level from the table.
If the result is not significant, do not reject H0H_0H0. This means insufficient evidence of correlation, rather than proof that population correlation is zero.
Transforming non-linear relationships
Some relationships become linear after transforming variables. For a power model,
y=axn⇒logy=loga+nlogx.y=ax^n\quad\Rightarrow\quad\log y=\log a+n\log x.y=axn⇒logy=loga+nlogx.Plotting logy\log ylogy against logx\log xlogx gives gradient nnn and intercept loga\log aloga.
For an exponential model,
y=abx⇒logy=loga+xlogb.y=ab^x\quad\Rightarrow\quad\log y=\log a+x\log b.y=abx⇒logy=loga+xlogb.Plotting logy\log ylogy against xxx gives gradient logb\log blogb and intercept loga\log aloga. Use the corresponding inverse logarithm to recover parameters. Logarithms require positive arguments, and the same logarithm base must be used throughout.
Common exam mistakes
| Mistake | Correction |
|---|---|
| Saying correlation proves causation | Describe an association only |
| Writing hypotheses using rrr | Use the population parameter ρ\rhoρ |
| Reversing explanatory and response variables | Predict the response using the appropriate regression line |
| Ignoring curves or outliers | Inspect the scatter diagram |
| Extrapolating without comment | Compare the input with the observed range |
| Giving a context-free test conclusion | Name the variables and significance level |
| Rounding intermediate results | Retain calculator precision until the final answer |
Exam-style question and answers
Temperature xxx in degrees Celsius and daily drink sales yyy are recorded for 121212 days. The regression line is
y=84+7.6x.y=84+7.6x.y=84+7.6x.Observed temperatures range from 12∘C12^\circ\mathrm{C}12∘C to 26∘C26^\circ\mathrm{C}26∘C, and r=0.742r=0.742r=0.742.
- Interpret 7.67.67.6 in context.
- Estimate sales at 20∘C20^\circ\mathrm{C}20∘C.
- Explain why prediction at 35∘C35^\circ\mathrm{C}35∘C may be unreliable.
- Test for positive correlation at the 5%5\%5% significance level using critical value 0.4970.4970.497.
Answers
- Each increase of 1∘C1^\circ\mathrm{C}1∘C predicts an increase of 7.67.67.6 drink sales.
- Substitution gives y=84+7.6(20)=236y=84+7.6(20)=236y=84+7.6(20)=236, so estimated sales are 236236236.
- This is extrapolation beyond the observed temperatures; the relationship may change.
- Write H0:ρ=0H_0:\rho=0H0:ρ=0 and H1:ρ>0H_1:\rho>0H1:ρ>0. Since 0.742>0.4970.742>0.4970.742>0.497, reject H0H_0H0. There is sufficient evidence at the 5%5\%5% significance level of positive correlation between temperature and daily drink sales.
Continue with MathsGenie's regression and correlation topic practice. Individual questions are available for Edexcel, AQA and OCR.
Revision and conclusion
Practise entering paired data and finding coefficients on your examination calculator. Clear old lists before starting, then check interpretations against the question's context. Review errors in hypotheses and conclusions as carefully as arithmetic errors.
For earlier foundations, consult GCSE Maths topics by exam board and tier.
Regression predicts a response; correlation measures linear association. Successful answers combine accurate calculations with suitable models, cautious predictions and contextual conclusions. Use MathsGenie's revision lesson and topic questions to practise these skills, then check the available solutions.
Sources and referenced URLs
- https://mathsgenie.co.uk/a-level/maths/edexcel/a-level-statistics-1-regression-correlation-and-hypothesis-testing
- https://mathsgenie.co.uk/a-level/maths/edexcel/a-level-statistics-1-regression-correlation-and-hypothesis-testing/questions/q172~8b07c46c-fa85-4336-ad1d-f109d0474ecd
- https://mathsgenie.co.uk/a-level/maths/aqa/a-level-statistics-1-regression-correlation-and-hypothesis-testing/questions/q66
- https://mathsgenie.co.uk/a-level/maths/ocr/a-level-statistics-1-regression-correlation-and-hypothesis-testing/questions/q4
- https://mathsgenie.co.uk/blog/gcse-maths-topics-by-exam-board-and-tier



