A Hierarchical System of Interaction Measures for Multidimensional Contingency Tables
针对(d+1)维列联表，提出两个汇总度量，分别衡量r阶及更高阶交互作用的大小，基于部分调整表与完全调整表的比较，并推广了用于两个名义变量交叉分类的lambda和tau关联度量。
Summary Two summary measures are proposed of the amount of rth degree (r + 1 factor) and higher interaction displayed in a (d + 1)-dimensional contingency table, 1 ≤ r ≤ d. The measures are based on a comparison of the “partially-raked” table having the r-factor interactions and lower-order effects removed to the “fully-raked” table in which all effects are removed. The measures have a proportional reduction in error construction with reference to using d of the variables to predict the remaining one within the partially-raked and fully-raked tables, and they are generalizations of the lambda and tau measures of association for cross-classifications of two nominal variables.