Designing Belief-driven Interactions with Data

Abstract

One's beliefs play a critical role in interpreting new information and making decisions. However, conventional visualization frameworks rarely consider users' beliefs in designing and evaluating visualizations. The main goal of this dissertation is to formalize visualization interaction in light of beliefs to inform visualization design and evaluation. As a first step, I introduce belief elicitation techniques and evaluate them on whether the elicitation act can impact how much users process data and reason with uncertainty. The multiple controlled studies reveal that the belief elicitation act has positive impacts on people's ability to reason with data and its uncertainty. With the understanding on the effect of elicitation, this dissertation presents a Bayesian modeling approach to understand people's belief updating process during visualization interaction. The analysis demonstrates that people's belief updating process slightly deviates from the Bayesian standard when they examine small data and severely deviates when they examine large data. To mitigate the deviation, I also introduce and evaluate personalized data presentations formulated using one's prior beliefs. By working toward formalizing visualization interaction in light of beliefs, this dissertation sheds light on how designers and researchers can take into account one's beliefs in understanding visualization interactions.