Invited Commentary: Selection Bias Without Colliders.

Area of Science:

• Epidemiology
• Causal Inference
• Biostatistics

Background:

• Conditioning on colliders in causal analyses is a known source of selection bias.
• Understanding sources of selection bias is crucial for accurate epidemiological research.

Purpose of the Study:

• To describe a scenario where conditioning on a non-collider can induce selection bias.
• To highlight the conditions under which non-collider conditioning leads to bias in causal inference.

Main Methods:

• Theoretical exploration of selection bias in causal diagrams.
• Review of empirical demonstrations of selection bias mechanisms.

Main Results:

• Selection bias arises from conditioning on a collider in causal analyses.
• Conditioning on a non-collider can also cause selection bias if the exposure has a non-null effect on the outcome and the non-collider-outcome association varies across exposure levels.

Conclusions:

• Researchers must carefully consider the implications of conditioning on any variable, not just colliders, in causal analyses.
• The findings underscore the importance of understanding variable relationships to prevent biased results in epidemiological studies.