15. Confounding Explained: Criteria, Stratification, Adjustment
After this lesson you can
- State the three criteria for a confounder and restate confounding as a failure of comparability and as a backdoor path
- Distinguish a confounder from a mediator and decide which one to adjust for
- Detect confounding by comparing crude and stratum-specific risk ratios
- Name the direction of confounding, positive or negative, from the confounder's two associations
- Compare randomization, restriction, matching, stratification, standardization and regression, with the cost of each
- Pool agreeing strata into a Mantel-Haenszel adjusted risk ratio
Video coming soon
Scheduled for release on November 12, 2026.
A lesson on confounding, the third alternative explanation for an association, defined three ways: as three criteria, as a failure of comparability, and as a fork in a causal diagram. It works through a fictional Oakhaven factory table where a crude risk ratio of 2.0 shrinks to 1.0 within each age group, then covers the design and analysis methods that control confounding.
Questions and answers
Questions a learner would ask about this lesson. Think through your own answer before you open each one.
- What is confounding, in plain words?
Show answer
Confounding is a mixing of effects, in which a real third variable distorts the association between an exposure and an outcome. The crude estimate is correct arithmetic but leads to the wrong inference. Removing the third variable's influence gives a fair comparison between exposed and unexposed.
- What are the three criteria for a confounder?
Show answer
The variable must be associated with the exposure in the study population, it must be an independent risk factor for the outcome, and it must not be on the causal pathway from exposure to outcome. All three are required. In the Oakhaven factory, age meets all three: older workers are far more common among the exposed, older age raises risk among the unexposed, and the solvent does not change anyone's age.
- How do I tell a confounder from a mediator?
Show answer
Ask whether the exposure could have caused the variable. A confounder is a common cause that comes before the exposure (a fork), so you adjust for it. A mediator is caused by the exposure and then causes the outcome (a chain), so adjusting for it removes part of the total effect you want to measure. The answer comes from causal knowledge of timing and biology, not from the table.
- How do I detect confounding in a table?
Show answer
Split the table by the third variable and compare the stratum-specific measures with the crude measure. When the strata agree with each other but differ from the crude, that is the signature of confounding. In the lesson, the crude risk ratio was 2.0 while both age strata gave 1.0, so age was confounding the comparison.
- Why not just test the confounder with a p-value?
Show answer
Confounding is a question about how much the exposure estimate changes, not about chance. The usual convention is that a change of about ten percent or more between the crude and the adjusted estimate matters. A p-value can hide a large confounder in a small study and flag a trivial one in a large study.
Practice quiz
- Question 1Which of these is NOT one of the three criteria for a confounder?
- Question 2In the Oakhaven factory table, 240 of 2,000 exposed workers and 120 of 2,000 unexposed workers became ill. What is the crude risk ratio?
- Question 3In the same factory, the risk ratio is 1.0 among workers under 50 and 1.0 among workers 50 and over, while the crude risk ratio is 2.0. What is the most likely explanation?
- Question 4Eighty percent of the exposed workers and twenty percent of the unexposed workers were 50 and over, and older age raised the risk of illness. What kind of confounding is this?
- Question 5The solvent causes airway inflammation, and inflammation causes the illness. If you want the total effect of the solvent, what should you do with inflammation?
- Question 6Which strategy is the only one that can balance confounders that nobody measured or thought of?
- Question 7In the partial set, both age strata give a risk ratio of 1.5 and the crude risk ratio is 3.0. Which adjusted risk ratio should be reported?
- Question 8Studying only workers under 50 so that age cannot vary is an example of which control method, and what is one of its costs?
- Question 9In a matched case-control study, what must happen in the analysis?
- Question 10When is it appropriate to pool stratum-specific estimates into one adjusted estimate?
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.