Module 11 Lesson: Correlation, Confounding, and Causation
Classroom Explanation
Correlation says two measures move together. It does not say which moves first, whether either moves the other, or whether something else moves both — and the third of those is the usual explanation.
In this module, the goal is to avoid claiming causation from association alone. The important part is not only the calculation. The important part is the thinking before and after the calculation.
Why This Matters
The common mistake is saying screen time caused lower scores just because the two variables are related. This mistake can make a report sound more confident than the evidence deserves.
Key Ideas
1. Correlation describes how two variables move together.
A relationship between two variables can have more than one explanation. The safer habit is to list possible explanations before choosing a causal story.
2. Confounding happens when another variable helps explain the pattern.
A relationship between two variables can have more than one explanation. The safer habit is to list possible explanations before choosing a causal story.
3. Causal claims need stronger design than ordinary observational data.
A relationship between two variables can have more than one explanation. The safer habit is to list possible explanations before choosing a causal story.
Worked Example
Higher screen hours appear with lower sleep and lower quiz scores. After-school work hours may also affect both screen time and sleep, so the pattern is not simple proof.
Here is the dataset used in this module.
| student_id | screen_hours | sleep_hours | after_school_work_hours | study_hours | quiz_score |
|---|---|---|---|---|---|
| Q01 | 1.5 | 8.0 | 0 | 6 | 82 |
| Q02 | 2.0 | 7.5 | 0 | 5 | 78 |
| Q03 | 3.0 | 7.0 | 1 | 5 | 74 |
| Q04 | 4.0 | 6.5 | 2 | 4 | 69 |
| Q05 | 5.0 | 6.0 | 3 | 3 | 63 |
| Q06 | 2.5 | 7.2 | 0 | 7 | 85 |
Look for the variable that could produce both. Where you can measure it, adding it to the comparison usually makes a spurious relationship collapse, which is a useful test. Where you cannot measure it, that limitation belongs in the write-up rather than out of it.
How To Think Through It
- Measure the strength of the association.
- List the variables that could drive both.
- Adjust for the ones you can measure.
- Say what happened to the association.
- Name the confounders you could not measure.
Common Mistake
The common mistake is saying screen time caused lower scores just because the two variables are related.
How To Write The Result
Screen hours, sleep, work hours, and quiz scores are related in this synthetic sample, but the table alone does not prove that screen time caused the score difference.
Practice Prompt
Write three possible explanations for why two variables move together.
Takeaway
Adjustment handles the confounders you thought of; randomisation handles the ones you did not.