Design and analysis of behavioral intervention studies: A Bayesian approach.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Health Services Research

Background:

• Null hypothesis significance testing (NHST) is criticized for leading to publication bias and flawed scientific practices.
• Existing guidelines for intervention studies predominantly rely on NHST, including sample size determination.
• Bayesian hypothesis testing offers an alternative framework to address the limitations of NHST.

Purpose of the Study:

• To summarize the limitations of null hypothesis significance testing.
• To introduce and explain Bayes factors and Posterior Model Probabilities for hypothesis comparison.
• To present a Bayesian approach for a priori sample size determination in intervention studies.

Main Methods:

• Summarized shortcomings of null hypothesis significance testing.
• Introduced Bayes factor and Posterior Model Probabilities, detailing their calculation and interpretation.
• Developed and illustrated a criterion and procedure for sample size determination in Bayesian hypothesis testing using a cluster randomized trial.

Main Results:

• The study provides a comprehensive overview of Bayesian hypothesis testing methods.
• A novel methodology for a priori sample size determination within the Bayesian framework is presented.
• The methodology is demonstrated using a real-world example of a cluster randomized trial evaluating an online training for physicians.

Conclusions:

• Bayesian hypothesis testing, utilizing Bayes factors and Posterior Model Probabilities, offers a robust alternative to NHST.
• The proposed Bayesian sample size determination method facilitates rigorous study design.
• The study provides practical tools (R syntax and dataset) for replication and application in intervention research.