Online Program

Saturday, February 22
PS3 Poster Session III & Continental Breakfast Sat, Feb 22, 7:30 AM - 9:00 AM
Bayshore II-IV

Confidence Intervals of Differences Between Correlated Proportions: An Empirical Comparison Among Three Estimation Methods (302818)

Eun Sook Kim, University of South Florida 
Jeffrey D. Kromrey, University of South Florida 
Diep Thi Nguyen, University of South Florida 
*Thanh Vinh Pham, University of South Florida 
Patricia Rodriguez de Gil, University of South Florida 
Jeanine Romano, University of South Florida 

Keywords: Interval estimates, Correlated Proportions, Dichotomous Outcomes

This study compared the accuracy and precision of three methods for computing confidence intervals around differences between correlated proportions: Wald’s method, the adjusted Wald method proposed by Bonett and Price (2011), and the likelihood-based method proposed by Tango (1998) using Monte Carlo methods. The design factors examined in this simulation study included sample size, magnitude and direction of population correlation between the two proportions, and difference between population proportions. The results showed that the adjusted Wald CI provided the best coverage across the conditions. The original Wald method did not provide adequate coverage for smaller sample sizes. The Tango method became less adequate (likely over coverage) when the difference in two population proportions became larger and the correlation between two proportions was positive. Finally, only minimal differences in precision were associated with the choice of method.