overadjustment

Overadjustment is a bias that occurs when statistical adjustment in a study includes variables that should not be controlled for when estimating the causal effect of an exposure on an outcome. In causal analysis, variables are typically categorized as confounders, mediators, or colliders. Overadjustment most often arises when a mediator—an intermediate variable on the causal path from exposure to outcome—is included in the adjustment set, or when a collider is conditioned on.

The consequence is that the estimated effect can be distorted. If a mediator lies on the pathway from exposure to outcome, adjusting for it can remove part of the causal effect, often biasing the estimate toward the null or changing the magnitude of the effect. In some cases, adjusting for a collider can open a spurious association, introducing bias rather than reducing it. The direction and magnitude of bias depend on the underlying causal relationships among the variables.

Examples include adjusting for post-exposure biomarkers or intermediate clinical outcomes when estimating the total effect of a treatment on recovery. Such adjustments can understate the treatment’s overall impact. Conversely, inappropriate adjustment for variables that are confounders is not overadjustment but underadjustment if important confounding is left uncontrolled.

Preventing overadjustment involves careful study design and analysis planning. Researchers use causal diagrams (DAGs) to distinguish confounders from mediators and colliders, aiming to estimate the total effect when that is the goal. When a direct effect is of interest, mediation analysis with appropriate methods is used.