A Note on Covariances for Categorical Data

摘要

Generalization of the covariance concept is discussed for mixed categorical and numerical data. Gini's definition of variance for categorical data gives us a starting point to address this issue. The value difference in the original definition is changed to a vector in value space, giving a new definition of covariance for categorical and numerical data. It leads to reasonable correlation coefficients when applied to typical contingency tables.