Visualizing Contingency Tables

Abstract

Categorical data analysis is typically based on two- or higher dimensional contingency tables, cross-tabulating the co-occurrences of levels of nominal and/or ordinal data. In order to explain these, statisticians typically look for (conditional) independence structures using common methods such as independence tests and log-linear models. One idea behind the use of visualization techniques is to use the human visual system to detect structures in the data that may not be obvious fromsolely numeric output (e.g., test statistics). Whether the task is purely exploratory or modelbased, techniques such as mosaic, sieve, and association plots offer good support for visualization. Mosaic and sieve plots in particular have been extended over the last two decades, and implementations exist in many statistical environments.