Confounding can distort a pooled comparison because the groups differ in another factor relevant to the outcome. Effect modification means the association varies across levels of another variable, on the particular scale being used. Stratifying the data can help tell the two patterns apart.
Read a confounding example
Imagine an observational study that compares injury rates for two job assignments. Workers in one assignment typically have less experience, and experience is related to injury. Suppose the crude risk ratio is 1.8, but within both the newer and experienced worker strata it is about 1.1. The two group-specific ratios agree with each other and differ from the pooled number. That invites a check for confounding by experience, with adjustment depending on the study design and assumptions.
Read an effect modification example
In a second hypothetical study, a training program is associated with a risk ratio of 0.50 among new workers and 0.95 among experienced workers. A single combined ratio can obscure the difference. Reporting the two estimates with their uncertainty is more informative. It also matters that the comparison here uses a risk ratio. Risk differences can tell a different story when baseline risks differ.
Decide what to report
Write down the exposure, outcome, and candidate third variable. Compare stratum-specific estimates on the same scale using compatible groups. Then compare them with the crude estimate. Check sample sizes and uncertainty before treating a difference as meaningful.
These examples illustrate what the patterns might look like. They do not establish causation: study design, measurement, and how the third variable relates to the exposure and outcome still matter.
Continue with Oakgen Agent
Open in Oakgen Agent
The prompt is filled in for you. Review it before sending.


