Lesson 7: Pearson's correlation coefficient, r
When r looks strong but the pattern curves
SkillsPearson's correlation coefficientDirection, form and strength
Worked example
A Brisbane start-up recorded the total number of downloads of its new app at the end of each of the first 10 weeks after launch.
| Week | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
|---|---|---|---|---|---|---|---|---|---|---|
| Downloads (thousands) | 2 | 3 | 4 | 6 | 8 | 12 | 17 | 24 | 34 | 48 |
A team member calculates and reports "a strong positive linear association between week and downloads". Evaluate this conclusion.
Step 1: Check the calculation
With week as and downloads as , technology gives . The arithmetic is right.
Step 2: Look at the form
The downloads grow faster each week — by 1 thousand between weeks 1 and 2, but by 14 thousand between weeks 9 and 10 — so the points curve upwards. The form is non-linear.
Step 3: Decide whether r applies
measures the strength of a linear association only. For a curved pattern it isn't an appropriate measure, however close to 1 it is.
Step 4: Evaluate the conclusion
The conclusion is not valid. The association is strong and positive, but it is non-linear, so it should not be described as linear or measured with . Treating it as linear would badly underestimate the downloads in later weeks.
"Evaluate" needs a judgement and a reason. Here the reason is the form you can see in the scatterplot.
Answer
The conclusion is not valid. Although , the scatterplot curves upwards, so the association is strong, positive and non-linear, and should not be used to describe it.
Try a similar one
A fresh question of the same kind, checked as you answer, with hints and a full solution.
Method
Always look at the scatterplot before using .
If the pattern is curved, describe the form as non-linear and don't rely on .
A value of near does not prove an association is linear.
Watch out for
Accepting the conclusion because is close to 1.