17. Causal Inference in Epidemiology: Pies, Counterfactuals, DAGs
After this lesson you can
- Distinguish necessary, sufficient and component causes using the sufficient-component cause model
- Explain why fractions attributable to different component causes can sum past 100 percent
- State the counterfactual contrast and explain exchangeability as what randomization buys and confounding breaks
- Weigh an association against the classic viewpoints, with temporality as the one requirement
- Classify a variable in a causal diagram as a fork, chain or collider and decide whether to adjust for it
Video coming soon
Scheduled for release on November 14, 2026.
A lesson on moving from an observed association to a judgment about cause. It covers necessary, sufficient and component causes through causal pies, the counterfactual contrast and exchangeability, reverse causation, the classic viewpoints for judging causation with their caveats, and fork, chain and collider shapes in a causal diagram, all applied to the fictional Oakhaven solvent question.
Questions and answers
Questions a learner would ask about this lesson. Think through your own answer before you open each one.
- What counts as a cause in epidemiology?
Show answer
A cause is an event, condition, or characteristic that preceded the outcome and without which the outcome would not have occurred, or would not have occurred until later. The definition builds in sequence and the counterfactual, and it counts hastening an outcome. In the lesson's fall example, ice, hurrying and worn shoes were all present, and removing any one would have prevented the fall, so none of them alone is "the cause."
- What is the difference between a necessary, a sufficient and a component cause?
Show answer
A necessary cause must be present for the outcome to occur. A sufficient cause is a complete set of factors that makes the outcome inevitable, drawn as a whole pie. A component cause is one slice within a sufficient cause, and most studied exposures, including the Oakhaven solvent, are components that are neither necessary nor sufficient.
- How can attributable fractions add up to more than 100 percent?
Show answer
In the three Oakhaven pies, the solvent is in the pie that produced 50 of 100 cases, the seasonal trigger is in pies producing 80, and the susceptibility factor is in all three pies, producing 100. Each case is fully attributable to every slice in its pie, so the fractions are 50, 80 and 100 percent, which sum to 230 percent. Nothing is wrong with the arithmetic.
- What is exchangeability and why does it matter?
Show answer
The ideal comparison is the exposed group against itself unexposed, which can never be observed, so the unexposed group serves as a stand-in. Exchangeability means the groups are swappable: each would have had the other's outcome under the other's exposure. Randomization buys exchangeability, and confounding breaks it, as when the factory's exposed workers were mostly older.
- When should I adjust for a variable in a causal diagram?
Show answer
Adjust for a fork, a common cause such as age, because it opens a backdoor path. Do not adjust for a chain variable, a mediator such as airway inflammation, when you want the total effect. Never adjust for or select on a collider, a common effect such as clinic referral. In Oakhaven the adjustment set is age alone, and the diagram, drawn from knowledge of timing and biology, decides this rather than the data.
Practice quiz
- Question 1A component that is present in every sufficient cause is a:
- Question 2One hundred cases arise from three pies, and the solvent is only in the pie that produced 50. What fraction of cases is attributable to the solvent?
- Question 3In a dose survey, risk was 40 of 400 at low exposure, 90 of 600 at medium, and 126 of 600 at high. What are the risks, and which viewpoint do they support?
- Question 4A variable caused by both the exposure and the outcome is a:
- Question 5Which of the classic viewpoints is the only one required before any causal claim?
- Question 6Early, undiagnosed illness leads residents to move into lighter indoor work that uses the solvent, and a cross-sectional survey finds exposure and illness associated. What explains the association?
- Question 7What is the main difference between exchangeability and representativeness?
- Question 8Which statement about the classic viewpoints is correct?
- Question 9In the Oakhaven diagram, which variables belong in the adjustment set for the total effect of the solvent?
- Question 10Before the classic viewpoints are weighed, which four alternatives should be examined?
Saved in your browser only: no account, no server.
Principles of Epidemiology is education in epidemiologic reasoning, not clinical, screening, or policy guidance. Every example is set in the fictional Oakhaven Health District, and every number is invented for teaching.