Title: Semiparametric inferences for dominance index under density ratio model
Authors: Weiwei Zhuang - University of Science and Technology of China (China) [presenting]
Abstract: Two-sample problem is an over-investigated topic in statistics. Often, we are asked to test whether or not two populations have the same mean against one-sided or two-sided alternatives. The standard method is the t-test. However, the population means merely show the central tendency. They fail to reflect general relationship between two populations. The stochastic dominance index makes up some shortfall for the purpose of comparing two populations. We consider the problem of testing for the degree of stochastic dominance between two populations under a density ratio model (DRM). In many applications, the distributions of two populations under investigation are of similar nature. Modeling their density ratio provides an effective tool to improve the efficiency of statistical inference. We use the DRM-based empirical likelihood (EL) to estimate the stochastic dominance index and show that it is consistent and asymptotically normal. A bootstrap method is then used to construct test for the degree of dominance. Simulation study shows that this approach has superior performance than the existing method that does not activate the DRM assumption.