THE KNOWLEDGE CONTENT OF STATISTICAL-DATA

An information-theoretic framework is used to analyze the knowledge co ntent in multivariate cross classified data. Several related measures based directly on the information concept are proposed: the knowledge content (S) of a cross classification, its terseness (Zeta), and the s eparability (Gamma(x)) of one variable, given all others. Exemplary ap plications are presented which illustrate the solutions obtained where classical analysis is unsatisfactory, such as optimal grouping, the a nalysis of very skew tables, or the interpretation of well-known parad oxes. Further, the separability suggests a solution for the classic pr oblem of inductive inference which is independent of sample size.