Distribution-Free Two-Sample Procedures for Judgment Post-Stratified Samples

• Activity Number: 494
• Type: Topic Contributed
• Date/Time: Wednesday, August 6, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Marketing
• Author(s): Omer Ozturk*+
• Companies: Ohio State University
• Keywords: Judment post-stratified sample ; Ranked set sampling ; Imperfect ranking ; Rank-sum test ; Pitman efficacy ; Two-sample inference

Abstract

In this talk, we use a judgment post-stratified (JPS) sample to develop distribution-free statistical inference for a two-sample problem. The paper constructs a rank-sum test. This test leads to distribution-free confidence intervals and point estimates for the location shift between two distributions. The advantages of the new tests (confidence intervals and the point estimates) are that they require essentially no assumptions on ranking mechanism, they maintain their levels, and they provide adjustment for empty strata in the JPS samples. We investigate the performance of the proposed inferential procedures. For finite sample sizes, it is shown that the new procedures have efficiencies between the efficiencies of the same procedures based on a simple and ranked set sample data. For large sample sizes, the new procedures have the same efficiency as the efficiency of Bohn-Wolfe and Fligner-MacEachern procedures.