Independence Tests for Uncertain Data with a Frequentist Method

摘要

In this paper, the analysis of contingency tables and the problem of independence of two variables are tackled, when information concerning one or both variables is missing or uncertain. In a first approach, the relations between the variables are described with the belief function theory. A second approach, a "frequentist" one, takes into account all the possible probabilistic distributions of contingency tables. With the use of classical statistical tests as the chi-square test, several dependence criteria adapted to the data are proposed and analyzed. A simulation and a real data example are presented to illustrate the methods.