A grid-based approach to exploratory data analysis

Abstract

A grid-based approach to exploratory data analysis is described wherein the size of a grid’s cells may be altered. The primary tool of exploratory data analysis described here is cluster analysis. Background information is provided regarding the different types of clustering algorithms and their effectiveness. A grid-based approach for exploratory and cluster analysis is then described. Both exploratory and cluster analyses are performed for a series of different sizes of grid cells. Two marketing data sets are utilized to illustrate research questions often faced by market researchers. The first is a cluster analysis of cities used to create strata for selecting a representative probability sample. The second example is a segmentation of customers into groups for the purpose of targeting advertising and product insight. These examples illustrate related issues such as the elimination of “background noise” observations, significance tests for departures from uniform density, and metrics for the shape of a cluster.