CMStatistics 2017: Start Registration
View Submission - CMStatistics
B1303
Title: Approximate confidence distribution computing (ACC): A likelihood-free method with statistical guarantees Authors:  Min-ge Xie - Rutgers University (United States) [presenting]
Abstract: Approximate Bayesian computing (ABC) is a likelihood-free method that has grown increasingly popular since early applications in population genetics. The purpose is to consider the use of ABC method in frequentist application and, in particular, its extended version based on the concept of confidence distribution. The extended version, called approximate confidence distribution computing (ACC), can overcome two defects of the traditional ABC method, namely, lack of theory supporting the use of non-sufficient summary statistics and lack of guardian for the selection of prior. It is also demonstrated that a well-tended ACC algorithm can greatly increase its computing efficiency over a traditional ABC algorithm. Related theories on frequentist coverage, both asymptotic and exact, are investigated. The method is also illustrated with both simulations and real data example of tuberculosis transmission.