Recommended tests and confidence intervals for paired binomial proportions

Area of Science:

• Biostatistics
• Statistical Methods

Background:

• Analysis of paired binomial proportions is crucial in various scientific fields.
• A wide array of statistical methods exists, necessitating clear recommendations.

Purpose of the Study:

• To evaluate and recommend optimal statistical methods for analyzing paired binomial proportions.
• To provide guidance on selecting appropriate tests for association, difference, ratio, and odds ratio.

Main Methods:

• Consideration and evaluation of 24 distinct statistical methods.
• Application of methods to real-world data from pediatric airway reactivity and a case-control study.

Main Results:

• Asymptotic McNemar test and McNemar mid-p test are recommended for association.
• Specific confidence intervals (closed-form and asymptotic score) are recommended for differences and ratios.
• Transformations of Wilson score and Clopper-Pearson mid-p intervals are recommended for odds ratios.

Conclusions:

• The study provides evidence-based recommendations for statistical analysis of paired binomial proportions.
• The findings aid researchers in selecting robust methods for their specific analytical needs.