WorkshopsMaximization of t-statistics based on semiparametric inference
74
reads
In the case control studies or the disease screenings, one population can be assumed to be homogeneous in the sense of its distribution of characteristics to be examined. On the other hand, the other population often has heterogeneity in the characteristics. In such situation, statistical methods that effectively use the homogeneity have been widely used and recognized their applicability in classification problems or quality control procedures. In this paper, we focus on t-statistics that utilize the homogeneity of one population and propose a semiparametric inference procedure to capture the heterogeneity of the other population. To extend the t-statistics, U function is considered and investigated its statistical properties. When U is taken to be a linear function, the extended t-statistics reduces to the original one. The consistency of the estimator based on U function is proved and the optimal U function that attains the minimum asymptotic variance is derived under some assumptions. The performance of the proposed method is illustrated in simulation studies.
reads
Osamu Komori
2011-12-20
14:10:00 - 15:00:00
101 , Mathematics Research Center Building (ori. New Math. Bldg.)
In the case control studies or the disease screenings, one population can be assumed to be homogeneous in the sense of its distribution of characteristics to be examined. On the other hand, the other population often has heterogeneity in the characteristics. In such situation, statistical methods that effectively use the homogeneity have been widely used and recognized their applicability in classification problems or quality control procedures. In this paper, we focus on t-statistics that utilize the homogeneity of one population and propose a semiparametric inference procedure to capture the heterogeneity of the other population. To extend the t-statistics, U function is considered and investigated its statistical properties. When U is taken to be a linear function, the extended t-statistics reduces to the original one. The consistency of the estimator based on U function is proved and the optimal U function that attains the minimum asymptotic variance is derived under some assumptions. The performance of the proposed method is illustrated in simulation studies.