The Future of Indirect Evidence

Area of Science:

• Statistics
• Biostatistics
• Data Science

Background:

• Traditional statistical inference relies on direct observation (e.g., clinical trials).
• Frequentist and Bayesian approaches traditionally differ in their use of direct vs. indirect evidence.
• Historical focus on direct evidence prioritized objectivity, limiting the use of indirect data.

Purpose of the Study:

• To explore the utility of indirect statistical evidence in modern scientific research.
• To introduce Empirical Bayes methodology as a bridge between direct and indirect evidence.
• To discuss the evolving standards of statistical objectivity in the era of big data.

Main Methods:

• Discusses the conceptual differences between direct (frequentist) and indirect (Bayesian) statistical evidence.
• Highlights the challenges and opportunities presented by large-scale data from technologies like microarrays.
• Presents Empirical Bayes as a practical approach to integrate both direct and indirect evidence.

Main Results:

• Empirical Bayes methodology provides a valuable compromise, leveraging indirect evidence alongside direct observations.
• Modern scientific devices generate massive datasets where ignoring indirect evidence is increasingly impractical.
• There is a discernible shift towards less rigid objectivity standards, enabling better utilization of indirect evidence.

Conclusions:

• Empirical Bayes offers a powerful framework for analyzing complex datasets by incorporating indirect evidence.
• The increasing volume and complexity of scientific data necessitate a re-evaluation of traditional statistical objectivity.
• Future statistical practices will likely embrace more flexible approaches that effectively integrate diverse sources of evidence.