摘要：Contingency tables are used to summarize categorical data with multiple attributes, which frequently arise in natural and social sciences. In Tian and Li (2015), the equality of risk difference based on a $2 \times 2$ table is tested at the presence of non-response. In this paper, we derive the joint distribution of multiple contingency tables with non-response. Consequently, we propose a new homogeneity test statistic for the risk difference among multiple contingency tables. The limiting distribution of the proposed test statistic is established along with inferential procedures. Upon rejection of the global null hypothesis of homogeneity, to identify contingency tables with discordance, a multiple comparison procedure is proposed. Numerical studies corroborate theoretical results. We illustrate our method with dataset from a psychiatric study.
