Guiding Visualization Users Towards Improved Analytic Strategies Using Small Interface Changes Radu Jianu * David H. Laidlaw † Brown University Figure 1: Subtle, non-functional interface changes in an analysis support module (top, three left panels) generated significant changes in users’ analysis of a visual problem solving task (top right). A first set of changes nudged subjects to increase their use of the analysis module by 27% (bottom left, p=0.02) in an attempt to expand users’ working memory. It also caused them to switch between hypotheses 27% more often (bottom center, p=0.03), indicating more consideration of alternative hypotheses. A second set of changes then lead subjects to gather 30% more evidence per hypothesis (bottom right, p=0.02). ABSTRACT We provide quantitative evidence that subtle changes in a visualiza- tion system’s interface can be used to alter users’ analytic behaviors in targeted ways. In a controlled study subjects completed three analyses, at one week intervals, using a system consisting of a visu- alization and an analysis support module. A control group used one interface for all three analyses. A test group started with the same * e-mail: jr@cs.brown.edu † e-mail:dhl@cs.brown.edu interface but then used modified versions in the following two ses- sions. A first set of changes, included before the second session, aimed to increase subjects’ use of the system and increase their consideration of alternative hypotheses. The second set of changes, added before the last session, aimed to increase the amount of evi- dence collected. After the first set of changes, test subjects used the interface 27% more and switched between hypotheses 35% more than a control group. After the second set of changes test subjects collected 27% more evidence than control subjects. All observed increases are significant (p1=0.02, p2=0.03, p3=0.02). We hypoth- esize this approach can be used to guide visualization users unob- trusively towards improved analytic strategies. Index Terms: Visual Analytics, Analytic Biases