Errors in causal inference: an organizational schema for systematic error and random error.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Systematic and random errors are critical in estimating causal measures.
• Existing frameworks may not fully clarify error types from a causal inference perspective.

Purpose of the Study:

• To propose an organizational schema for systematic and random errors in causal inference.
• To clarify the concepts of errors, enhancing understanding of accuracy, validity, and precision.

Main Methods:

• Dividing systematic error into structural and analytic error.
• Identifying four major sources of random error: nondeterministic counterfactuals, sampling variability, exposure event generation, and measurement variability.
• Utilizing directed acyclic graphs to illustrate structural error.

Main Results:

• Structural error is defined via counterfactual reasoning, encompassing nonexchangeability bias (confounding, selection bias) and measurement bias.
• Analytic errors stem from small-sample properties or misspecified models/methods.
• Nonexchangeability bias highlights a lack of exchangeability between exposed and unexposed groups.

Conclusions:

• The proposed schema clarifies the relationship between systematic and random error.
• This framework enhances the understanding of accuracy, validity, and precision in causal estimations.
• Provides a novel perspective on error classification in epidemiological and biostatistical research.