Regression line of x on y and prediction
In a minute
When you know and want to estimate , you need the . It is a different line from the one you met for predicting . You learn to find it with your GDC, predict with it, and pick the right line for every prediction.
You'll be able to
- Find the with your GDC, and use it to predict from a given .
- Choose the right regression line for a prediction, and explain why a line of on can't reliably predict from .
- Use the mean point, where the two regression lines cross, to find , or a missing constant.
- Judge whether a prediction is reliable: the right line, strong correlation, and a given value inside the data.
Before you start3 quick questions on what this builds on
Revise first: 4.4 Correlation and regression (y on x)
The regression line of x on y
A forester can measure round a tree trunk with a tape in seconds. Finding the tree's age is much harder. So she wants to predict age from circumference : here is known and is the unknown.
The line of on from 4.4 won't do. It was built to keep the errors in small. To predict , you need a line that keeps the errors in small.
Here and are constants. Exam questions may use other letters, so check which variable is on the left.
Try it first, then reveal one step at a time.
- You know and want , so use the line of on : .
- Substitute : .
- You were given a , so check it against the range of the data. For a line of on , that is the range that decides .
- kWh is inside the recorded to kWh, so this is interpolation. The mean temperature was about °C.
Your GDC finds the line of on with the same linear regression command as before. The only change is the lists: the -values go in the X list and the -values in the Y list.
Try it first, then reveal one step at a time.
- (a) You will predict from , so you need the line of on . Enter the circumferences () as the X list and the ages () as the Y list.
- Run linear regression (). The screen shows and . These are and , so the line is .
- (b) Write down the substitution, using the stored values: .
- This gives years. cm is inside the circumferences to cm, so this is interpolation.
Which line predicts which variable
Each regression line predicts the variable named first. The line of on predicts . The line of on predicts .
Rearranging doesn't help. Making the subject of gives the same line, which still minimises horizontal distances. It is not the line of on .
Mean point (x̄, ȳ) = ()
y on x:
x on y:
Pearson's r =
The ringed point starts far off the trend, so and the lines cross at a wide angle. Drag it up to : rises to and the lines close up. The weaker the correlation, the more their predictions differ. They are one line only when .
Try it first, then reveal one step at a time.
- (a) You know and want , so use the line of on . Run the regression with the ages in the X list, then use the stored values: cm.
- (b) The student rearranges to .
- At : cm, about cm below the estimate in (a).
- The line of on keeps the errors in age small, not in circumference, so it can't reliably predict a circumference. The gap is large because is well short of .
Try it first, then reveal one step at a time.
- (a) You know and want : use the line of on . litres per week.
- (b) You know and want : use the line of on . .
- Check: rearranging the line of on in (a) would give . The wrong line gives a different answer, so the choice of line matters.
Where the two lines meet
Both regression lines pass through the . So that is where they cross.
Try it first, then reveal one step at a time.
- The point lies on each line, so its coordinates satisfy both equations at once.
- Replace in the first equation by : .
- So , which gives .
- Then . Check in the first line: . Correct.
Try it first, then reveal one step at a time.
- First find .
- (a) The line passes through , so , which gives .
- (b) You know , so this is the right line. .
- is inside the -values, to , so this is interpolation.
Common mistakes
- Rearranging the line of on to predict , or the line of on to predict . Each line predicts only the variable on its left, so use the other line.
- Entering the lists in the usual order. For the line of on , the -values go in the X list.
- Mixing up the constants. Write the line out in full, such as , so it is clear which number multiplies .
- Rounding and before predicting. Use the stored values and round only the answer.
- Commenting only on when asked about reliability. Also check that the given value is inside the range of the data.
Exam tips
- Find the regression line of on : write the whole equation, with on the left and constants to 3 s.f.
- Give a reason a method is not appropriate: say which variable the line was built to predict, and that the question predicts the other one.
- Find the means from two regression lines: both pass through , so show the simultaneous equations you solve.
- Estimate on Paper 2: write the substitution with the stored values first. The method mark depends on it.
Key takeaways
Practise this
5 past-paper questions on 4.10 in the AA HL bank, each with a mark scheme and a worked explanation.