Correlation and regression
Studying this course? Keep your progress in one place.
Key facts
- Correlation can be positive, negative or none; it never proves causation.
- The correlation coefficient is between and : the sign gives direction, gives strength.
- Use technology to find the line of best fit ; the gradient is the change in for each in .
- Interpolation is usually reliable; extrapolation is usually not.
- Comment on strength, range, whether the answer is sensible, and the amount of data.
Types of correlation
Positive, negative or none, depending on how the points rise or fall from left to right.
A scatter graph plots two variables for the same items, with one point for each item. Correlation describes the relationship between the variables.
Correlation describes a pattern; it does not prove that one variable causes the other.
The points on a scatter graph fall from left to right. What type of correlation is this?
The correlation coefficient r
The sign of gives the direction and its size gives the strength.
The correlation coefficient is found with technology. It is always between and .
The size gives the strength: close to is strong, around is moderate, close to is weak or none. or means all points lie exactly on a straight line.
The strength depends on how close the points are to a line, not on how steep the line is.
What does tell you?
Line of best fit using technology
Enter the data in two lists and ask for linear regression: it gives and .
Using technology (a graphic display calculator or spreadsheet) enter the data in two lists and ask for linear regression. For temperature and drinks sold the result is with .
Interpret the numbers in context. The gradient is about 3 more drinks for every extra degree. The intercept is the value when , which may have no meaning in practice.
What does the gradient mean in ?
Interpolation and extrapolation
Predictions inside the range of the data are usually reliable; outside it they are not.
Interpolation is predicting a value inside the range of the data you collected. This is usually reliable, especially when the correlation is strong.
Extrapolation is predicting outside the range, which is usually unreliable because the pattern may not continue.
Data for from to : at the equation gives (reliable); at it gives (unreliable, outside the data).
The data covers from to . Is a prediction at interpolation or extrapolation?
Commenting on reliability
Mention the strength of the correlation, interpolation or extrapolation, whether the value is possible and how much data you have.
When you comment on a prediction, give a reason with every comment.
Check for impossible values too: a predicted car price of thousand AED shows that the model has failed.
Write a short conclusion, for example: "The prediction for 30 °C is reliable because it is within the data and ."
- 1
Strength
how close is r to 1 or −1?
- 2
Range
interpolation or extrapolation?
- 3
Sensible
is the value possible?
- 4
Amount of data
more points make a better line
A prediction at is outside the data and . What is the best comment?
Try an exam question
The correlation coefficient between the age of a car and its value is . The line of best fit is , where is the value in thousands of AED and is the age in years, from data for cars aged 1 to 8 years. (a) Describe the correlation. (b) Predict when . (c) Comment on a prediction at .
[4 marks]
- [1]Strong negative correlation.
- [1] thousand AED.
- [1]At , : a negative value is impossible.
- [1] is outside the data (extrapolation), so the prediction is unreliable.
That's the notes covered.
Carry on to the next subtopic.
Try a question
Which description of the correlation is correct?
1 mark, marked against the mark scheme
Which description of the correlation is correct?
1 mark, marked against the mark scheme