Spatial autocorrelation-based information visualization evaluation

Abstract

A data set can be represented in any number of ways. For example, hierarchical data can be presented as a radial node-link diagram, dendrogram, force-directed layout, or tree map. Alternatively, point-observations can be shown with scatter-plots, parallel coordinates, or bar charts. Each technique has different capabilities for representing relationships. These capabilities are further modified by projection and presentation decisions within the technique category. Evaluating the many options is an essential task in visualization development. Currently, evaluation is largely based on heuristics, prior experience, and indefinable aesthetic considerations. This paper presents initial work towards an evaluation technique based in spatial autocorrelation. We find that spatial autocorrelation can be used to construct a separator between visualizations and other image types. Furthermore, this can be done with parameters amenable to interactive use and in a fashion that does not need to take plot schema characteristics as parameters.