Visual Analysis for Multi-Attribute Choice

This dissertation presents findings from problem-driven research that centers around the design of visualization tools to assist experts in making data-informed choices. Identifying the most preferred solution among many alternatives is a common task in our everyday and professional lives. Pivotal information is usually hidden in the data and visualization research has long treated decision-making as a data comprehension task. To arrive at a decision, however, the understanding of patterns in the data needs to be synthesized with subjective judgments. Existing visualization tools do not target this synthesis and many approaches focus on simplified decisions tasks. As a result, their relevance and applicability in real-world settings might be limited. This dissertation promotes field research to investigate the cognitive processes underlying real-world decisions and to operationalize them for the design and validation of decision support tools. Being based on a close collaboration with real decision-makers, it provides an emphasis on decision processes that problem-driven visualization research did not have before. By synthesizing the collected real-world experience with concepts from human science, it also contributes to making decision models and theories usable for visualization design. This dissertation refines the existing multi-attribute choice definition by describing it as a constructive problem where preferences are incrementally formed at the actual time of choice. It further proposes a characterization scheme to help visualization researchers concretize the decision problem to design for. Finally, going into the field revealed a novel type of constructive decision problem, which this dissertation defines as co-dependent choices. As theoretical contributions, these formalizations make the design space of decision tasks more tangible. A knowledge elicitation method is adapted from cognitive science to systematically detail the knowledge, experience, and cognitive tasks underlying current decision-making practices. As a methodological contribution, this introduces a decision-oriented way of conducting problem characterizations. As technical contributions, this dissertation presents two design studies. Their results demonstrate the relevance and applicability of the proposed concepts within and beyond the studied decision contexts. PAVED provides a simple yet effective means for decision-makers to construct and apply preferences as they learn what level of performance is achievable. Its extension COMPO*SED is the first tool that helps decision-makers explore the side effects of co-dependent choices. Their usefulness has been confirmed with domain experts on their day-to-day decision problems. The long-term benefit of PAVED is indicated by the adoptions recorded after four years. Through user-centered design, this dissertation addresses the lack of discourse on validated visualization tools that are dedicated to assist expert choices in the wild. Its theoretical, methodological, and technical contributions shape the understanding of decision-related activities on large data sets and how to support them with visualization. As task clarity, design guidelines, and real-world experience with decision support tools evolve, more rigorous claims regarding decision-making as a core goal of visualization research will be possible. The research presented in this dissertation is an important step in this direction.