Glossary
Confounding
What is Confounding?
Confounding occurs when a third variable influences both the independent and dependent variables, distorting the association used to estimate a causal effect. Confounding can occur alongside a genuine causal effect. It is an important threat to valid causal inference from observational data.
How it works
Common causes and adjustment pitfalls
Common causes of treatment and outcome can create noncausal paths between them. Association with both is not a sufficient rule for choosing an adjustment variable: conditioning on a common effect, or collider, can instead introduce bias. For an illustration, hot weather could increase both ice cream buying and time spent swimming. An association between ice cream sales and drownings would then not, by itself, show that ice cream causes drowning.
Study design and random assignment
A causal interpretation needs a study design and assumptions that address relevant confounding. Adjusting for measured variables is one approach, but different designs rely on different assumptions.
Random assignment makes treatment selection independent of baseline causes through the assignment process, though the resulting groups can still differ by chance in a finite sample. Problems such as nonadherence or selective loss to follow-up can still compromise a causal analysis.
Applied example
A study finding that coffee drinkers live longer than non-coffee drinkers may be confounded by socioeconomic status: in a hypothetical population, socioeconomic conditions could influence both coffee consumption and other determinants of health. Such confounding could explain some or all of the observed association, alongside any effect of coffee itself.
Why it matters
Hernán and Robins’s account of confounding explains how common causes create noncausal paths. Observational analyses require justified assumptions about which variables to adjust for and whether important confounding remains.