Exploratory data analysis: data visualization or torture?

Abstract

Exploratory Data Analysis offers a set of graphical and statistical tools to find the full meaning from data sets. The user visualizes, analyzes, and transforms data distributions with these tools. Graphs reveal relationships between variables; the residuals left after fitting data show the adequacy of the model. Without this careful examination and understanding of the data, rote data analysis using standard statistical tests can give misleading results. Exploratory Data Analysis has its own set of pitfalls and must be used with confirmatory statistics and studies. Increasing power and resolution in personal computers enables modern statistical software to make these methods widely accessible. By easily moving between data and their graphic representation, analysis can be comprehensive without being tedious. Exploratory Data Analysis can add an exciting and useful tool to the epidemiologist's repertoire. This article illustrates several tools from an evolving list.