[The directionality of measurement bias: a directed acyclic graph-based structural perspective]

Area of Science:

• Causal inference
• Measurement bias
• Epidemiology

Context:

• Measurement bias (MB) complicates causal inference, particularly concerning the substitution estimate (SE) of effects.
• Existing understanding of MB often relies on assumptions of bidirectionally non-differential misclassification.
• The precise structure and mechanisms of MB within causal frameworks require further elucidation.

Purpose:

• To propose a directed acyclic graph (DAG)-based framework for understanding single-variable measurement bias.
• To differentiate the influences of the measurement system versus external factors on MB.
• To define reverse causality at the measurement level within causal structures.

Summary:

• This paper introduces a DAG model to analyze measurement bias (MB) stemming from imperfect measurement systems.
• It distinguishes how the measurement system's properties ensure non-differential bias in the substitution estimate (SE), while external factors can introduce differential misclassification.
• The framework incorporates reverse causality and temporal relationships to clarify MB's structure, mechanisms, and directionality.

Impact:

• Provides a structured approach to understanding and potentially mitigating measurement bias in causal inference.
• Enhances the theoretical foundation for evaluating the validity of substitution estimates.
• Offers a tool for researchers to critically assess measurement error in observational studies.