Exploratory Data Analysis with Linked Dotplots and Histograms

Abstract

Data sets are currently increasing in their number of variables as well as their number of observations. Standard exploratory tools for multivariate data analysis, like the scatterplot matrix, have problems when dealing with such data. The screen space available gives only enough resolution for a scatterplot matrix of four variables. Overplotting of tied or close observations is present in scatterplots even for medium sized data sets. As an alternative the use of linked low-dimensional views has been suggested in the literature. In this paper we look at some particular data analytic problems and investigate the usefulness of linked dotplots and histograms for these problems. We restrict ourselves mainly to two-dimensional phenomena so that we can easily check our conclusions by scatterplots.