Deceptive visualizations and user bias: a case for personalization and ambiguity in PI visualizations

In Personal Informatics (PI) systems, users can obtain information about themselves and the way they live their lives. Such data can be difficult to interpret: what constitutes a high value for one person may be normal for another. And what if the data does not match the user's self-image? Our study shows that participants' interpretations of feedback about their stress level were biased by their expectations of what the graph 'should' show. Nevertheless, participants were susceptible to value interpretations of their data suggested by the visualization. This latter finding may be problematic if the standardized interpretation suggested by such a system is not accurate for the individual user in question. We therefore argue that value interpretations of PI data should be personalized and future designs of PI data visualizations should incorporate ambiguity and visualize uncertainty.