Improved Confidence Interval of the Difference of Paired Binomial Using Weighted Profile Likelihood

• Activity Number: 620
• Type: Topic Contributed
• Date/Time: Thursday, August 5, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Vivek Pradhan*+
• Companies: Boston Scientific Corporation
• Keywords: Coverage ; Expected Lengths ; Likelihood-ratio statistic ; P-value ; Score statistic ; Small-sample

Abstract

Confidence interval of the difference of two binomial proportions based on paired data is required in many areas of applied statistics. For example, in observational study where patients are not randomized; for treatment comparison patients are selected based on matched propensity scores, subsequently for efficacy or safety, matched binomial samples are used to compute confidence intervals for the difference of proportions. Newcombe (1998) proposed a method based on score intervals, and compared several methods. Along with profile likelihood method based on tail probabilities, he recommended two other methods (Method 5 and 6) in his paper. Our proposed method is also profile likelihood based, however the likelihood is weighted by a non-informative Jeffreys' prior. We have carried out extensive simulations finding the coverage and expected lengths. The proposed method has been applied on