Ask Mallard

Confounding

Confounding occurs when a factor or set of factors creates a noncausal association, or distorts a causal effect estimate, because exposure groups differ in causes of the outcome.

What question does this help answer?

“Could the observed exposure–outcome association be explained partly or entirely by differences between the comparison groups rather than by the exposure itself?”

Why it matters

Confounding is one of the central threats to causal interpretation in observational research. Statistical adjustment can reduce confounding by measured variables, but it cannot guarantee elimination of bias.

How to think about it

A confounder is not simply any variable associated with both exposure and outcome. Temporal order and causal structure matter. Variables caused by the exposure may be mediators, and variables caused by both exposure and outcome-related factors may be colliders. Adjusting for the wrong variable can create bias.

What confounding looks like as a diagram

A causal diagram with three nodes. A confounder has arrows into both the exposure and the outcome, and the exposure has an arrow into the outcome labelled the effect of interest. The path from exposure back through the confounder to the outcome is not the effect of interest.Confounding: a common cause of exposure and outcomeEffect of interestConfounderExposureOutcomeExposure and outcome are associated through the confounder even with no effect at all. Adjusting for theconfounder closes that path.
Conceptual diagram. It shows the structure of the idea, not data from any study.

How Mallard may use this concept

Mallard may flag confounding when a question compares naturally occurring treatment or exposure groups. It may recommend a causal diagram, thoughtful covariate selection, design restriction, matching, standardization, propensity-score methods, or sensitivity analysis.

Good practice

  • Start with the causal question and time zero.
  • Use subject-matter knowledge to identify causes of exposure and outcome.
  • Prefer prespecified covariates over data-driven p-value selection.
  • Avoid automatically adjusting for post-exposure variables.
  • Distinguish variables needed for precision from variables needed for confounding control.
  • Consider residual and unmeasured confounding explicitly.

Common mistake

“Adjusted for everything available” is not automatically better. Overadjustment and collider adjustment can increase bias.

Clinical example

If sicker patients are more likely to receive a new treatment and also more likely to have poor outcomes, disease severity can confound the treatment–outcome association.

Recent studies that used this
Searching PubMed…
Part of the learning path Plan an observational study from existing records (step 5 of 7).
Part of the learning path Make a causal claim carefully (step 2 of 7).
Work it out
Related Learn pages
Sources & further reading
  1. Principles of confounder selection (opens in a new tab) · Eur J Epidemiol · doi 10.1007/s10654-019-00494-6
  2. Tutorial on directed acyclic graphs (opens in a new tab) · J Clin Epidemiol · doi 10.1016/j.jclinepi.2021.08.001
Last source review: 2026-09-05
AI-generated educational content
Mallard uses reputable methodological and reporting sources, but this material may contain errors or omit important context. Verify important methodological decisions with the cited sources and, when appropriate, a qualified research methodologist or statistician.