Randomization, statistics, and causal inference

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Conventional statistical methods often rely on assumptions of randomization and random sampling.
• Epidemiologic studies frequently do not employ randomization or random sampling in cohort assembly.
• This discrepancy raises concerns about the validity of probabilistic interpretations in many studies.

Purpose of the Study:

• To review the role of statistics in causal inference.
• To highlight the necessity of randomization for causal claims and random sampling for descriptive claims.
• To address the implications of lacking these elements in common epidemiologic research.

Main Methods:

• Literature review of statistical principles in causal inference.
• Analysis of assumptions underlying conventional statistical techniques.
• Discussion of the impact of non-randomized study designs on statistical interpretation.

Main Results:

• Probabilistic interpretations of conventional statistics are rarely justified in non-randomized studies.
• The lack of randomization and random sampling can lead to misinterpretations of study findings.
• Existing statistical practices may not adequately reflect the realities of observational data.

Conclusions:

• Deemphasize inferential statistics in favor of data descriptors for non-randomized studies.
• Adopt statistical techniques grounded in more realistic probability models.
• Promote a more cautious and accurate interpretation of findings from epidemiologic research.