MEPX
Chapter 9 of 13All chapters
  1. 1What analysis is for
  2. 2Types of data
  3. 3Collecting and sampling
  4. 4Cleaning data
  5. 5Describing the middle
  6. 6Describing the spread
  7. 7Choosing a chart
  8. 8Correlation
  9. 9Getting to causation
  10. 10Hypothesis testing
  11. 11Traps and paradoxes
  12. 12Tools of the trade
  13. 13Communicating results

Chapter 9 of 13

Getting to causation

What it takes to say one thing caused another.

Experiments

A randomised controlled trial assigns the treatment by chance, which balances every other factor on average, known or not. It is the only clean way to establish cause from data.

  • Randomisation is what removes confounders; a large sample alone does not.
  • A control group is what tells you what would have happened anyway.

When you cannot experiment

Observational work can support causal claims with careful design and stated assumptions, but the claim is always weaker. Saying associated with rather than causes is not timidity, it is accuracy.