Bridging observational studies and randomized experiments by embedding the former in the latter

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Observational studies often face challenges in establishing causal relationships.
• Standard frequentist statistical measures (p-values, confidence intervals) require assumptions about data generation.
• A robust method is needed to validate causal claims from non-randomized data.

Purpose of the Study:

• To present a rigorous, multistage framework for drawing valid causal inferences from observational data.
• To embed observational studies within hypothetical randomized experiments to ensure statistical validity.
• To provide a method for scientifically meaningful analysis of observational data.

Main Methods:

• A four-stage approach: conceptual formulation, design approximation, statistical analysis, and conclusion summarization.
• Embedding observational data within a hypothetical randomized experiment.
• Utilizing scientific argumentation in the conceptual stage and computational methods in later stages.

Main Results:

• The proposed framework ensures the validity of point estimates, p-values, and confidence intervals.
• It guards against analyses being perceived as "scientifically meaningless arithmetic calculations."
• The method was illustrated using data on parental smoking and children's lung function.

Conclusions:

• Rigorous causal inference from observational data necessitates embedding the study in a hypothetical randomized experiment.
• The conceptual stage is crucial for scientific plausibility and meaningfulness.
• This approach enhances the credibility and interpretability of statistical findings from observational research.