Reflection on modern methods: selection bias-a review of recent developments.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Selection bias is a complex issue in observational studies, often harder to grasp than confounding or measurement error.
• Existing definitions and methods for quantifying selection bias have been insufficient.
• Endogenous selection bias, arising from conditioning on collider variables, presents unique challenges.

Purpose of the Study:

• To define and explain endogenous selection bias.
• To outline conditions for identifying causal effects in the presence of selection bias.
• To discuss generalizability and methods for correcting selection bias.

Main Methods:

• Conceptual clarification of endogenous selection bias and its sources.
• Definition of conditions for causal effect identification (exchangeability).
• Overview of methods for generalizing causal effects and sensitivity analyses.

Main Results:

• Endogenous selection bias results from conditioning on collider variables, leading to spurious associations.
• Explicit conditions involving exchangeability are required to identify causal effects in the presence of selection bias.
• Generalizing findings to a target population requires additional data or assumptions.

Conclusions:

• Understanding endogenous selection bias is crucial for valid causal inference in observational research.
• Identifying causal effects requires meeting specific exchangeability criteria.
• Sensitivity analyses can help assess the impact of selection bias and provide corrected estimates.