Designing Visual Decision Support Systems to Foster Trust and Support Expertise Diversity through Human-Centered Data Science

Abstract

Decision support systems (DSSs) increasingly shape how people interpret complex data and make consequential decisions. Although these systems are often evaluated through technical performance, usability, or efficiency, their effectiveness also depends on whether users can make sense of system outputs, judge their reliability, and engage with tools in ways that reflect their knowledge, roles, and goals. In visual DSSs, design choices about how data is shown directly shape how users interpret system outputs and decide what to do with them.This dissertation examines how visual DSSs can be designed and evaluated to foster trust and support expertise diversity. Grounded in Human-Centered Data Science and Munzner's nested model of visualization design, it develops the Dual-Lens Evaluation Framework, which analyzes visual DSSs through two independent perspectives: what the system assumes about its users, and what users independently bring to the interaction. The gap between these two perspectives is treated as the site where trust holds or breaks down depending on whether expertise diversity is supported. The framework is developed and refined across three studies in distinct decision-making domains. The first, set in a traffic incident coordination platform, establishes the value of the user-side perspective and points to visualization as a promising direction for making system reasoning visible to users. The second, set in a public-facing transportation planning platform, examines how visualization features shape understanding, confidence, and engagement across users with varied expertise, and contributes the first refinement of the framework. The third, set in a qualitative coding environment, applies the full framework and surfaces tensions between system assumptions and user perspectives that would otherwise remain invisible, including a field-level divide in how the domain problem itself was understood. This dissertation contributes the Dual-Lens Evaluation Framework for visual DSSs, empirical insights from three domains into how users with varied expertise interpret, trust, and engage with visual decision tools, and four design priorities for visual DSSs that account for the expertise their users bring: system transparency, communication, layered engagement, and reflexive adaptability. The work also contributes a refinement that moves expertise diversity beyond a novice-to-expert competence range toward the different kinds of expertise users bring through their particular relationship to the domain problem.