Illustrating bias due to conditioning on a collider

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Selection bias, specifically collider-stratification bias, arises when researchers condition on a common effect of an exposure and an outcome.
• This phenomenon is often counterintuitive, making it difficult to recognize and avoid in observational studies.

Purpose of the Study:

• To explain the concept of collider-stratification bias using clear, hypothetical examples.
• To illustrate how conditioning on a common effect can induce an association between independent variables.

Main Methods:

• Presented two hypothetical scenarios to demonstrate bias due to conditioning on a collider.
• Example 1: Fever as a common effect of influenza and a tainted sandwich.
• Example 2: Case-status as a common effect of genotype and environmental factors.

Main Results:

• In both examples, conditioning on the common effect (collider) created a statistical association between variables that were initially independent.
• This induced association is termed selection bias in the context of collider-stratification.

Conclusions:

• Conditioning on a common effect of exposure and outcome can introduce selection bias.
• Recognizing and avoiding conditioning on colliders is essential for accurate causal inference in epidemiological research.