Confounding
The short answer · 17 words
In short
Distortion of a causal interpretation because the observed relationship is influenced, among other things, by common causes.
Flow-alerta.cab (2026): Confounding. https://flow-alerta.cab/en/glossar/confounding. Editorially reviewed on: 09 October 2026.
How much detail would you like?
Practical
Relevant for analyses, experiments and ML evaluations, as soon as causal statements or measures are to be derived from data.
Hypothetical example: A company sees higher sales among customers with app use; however, overall customer activity may affect both app use and purchases.
Identify possible common causes before making causal decisions, examine study design and use appropriate causal methods instead of mere prognosis metrics.
At a glance
- Core
- In confounding, an observed relationship between two variables can be partially or completely explained by common causes. As a result, a statistical correlation cannot be interpreted as an effect without appropriate assumptions and causal analysis.
- Use
- Relevant for analyses, experiments and ML evaluations, as soon as causal statements or measures are to be derived from data.
- Example
- Hypothetical example: A company sees higher sales among customers with app use; however, overall customer activity may affect both app use and purchases.
- Limits
- Interpreting correlation as causality; selecting confounders based only on statistical correlation; introducing additional distortion by adjusting to colliders; ignoring domain knowledge.
Questions about this term
Is every correlated factor a confounder?
No.
Does confounding affect pure prognosis just as strongly?
For causal interpretation stronger than for pure prediction.
Do DAGs help?
Yes.
All details and background
In confounding, an observed relationship between two variables can be partially or completely explained by common causes. As a result, a statistical correlation cannot be interpreted as an effect without appropriate assumptions and causal analysis.
ELI5 – Explained simply
Two things seem connected, but a third factor influences both and distorts the impression.
Why it matters in practice
Relevant for analyses, experiments and ML evaluations, as soon as causal statements or measures are to be derived from data.
Example
Hypothetical example: A company sees higher sales among customers with app use; however, overall customer activity may affect both app use and purchases.
Pitfalls and limits
- Interpreting correlation as causality; selecting confounders based only on statistical correlation; introducing additional distortion by adjusting to colliders; ignoring domain knowledge.
Origin & history
Confounding is a fundamental problem of epidemiology, statistics and modern causal inference and is now often formalized with causal charts and counterfactuals.
Use in mid-sized companies
- Identify possible common causes before making causal decisions, examine study design and use appropriate causal methods instead of mere prognosis metrics.
Related terms
- Causal inference
- DAG
- Distortion (bias)
Matching Flow-alerta service
AI opportunity analysis & roadmapAI consulting: AI potential analysis & roadmap on tasks, data and prerequisites of a possible first AI use case.Matching tool / use case
View use casesUse Case: Examination of whether an observed connection is a reliable basis for a business measure.Sources
- pubmed.ncbi.nlm.nih.govVanderWeele & Shpitser, On the definition of a confounder, Originalabstract
Quellenprüfung: 09.10.2026. Definition und dokumentierte Einordnung abgeglichen; Beispiele beschreiben mögliche Anwendungen.
Editorially reviewed on: