Visualization of cluster structure and separation in multivariate mixed data: A case study of diversity faultlines in work teams

Abstract

In organizational management, researchers and managers study separations or faultlines that occur in diverse teams when members form subgroups based on the alignment of multiple demographic characteristics. The team faultline concept is operationalized using multivariate cluster analysis—analysts use faultline measures to identify subgroups/clusters in a team and to quantify how subgroups/clusters are separated. Unfortunately, these measures have limited capacity to enable users to observe and explore faultlines and subgroup structure across the examined attributes efficiently. We address this problem and make three contributions. First, we propose a visual representation for communicating faultline information that is based on multiple linked, stacked histograms in an axis-parallel layout. Second, we evaluate the effectiveness of the proposed technique in a controlled user study, comparing it to the two other common multivariate representations of clusters: parallel coordinates and scatter plot matrices. While we chose faultline-related tasks based on the requirements by domain experts in organizational management, the study findings can be generalized to representations and tasks involving distributions of clusters of multivariate objects in mixed-type data. Finally, inspired by geological faultlines, we propose several visual enhancements to stacked histograms to further facilitate the task of identifying “cracks” within work teams.