Students’ performance and typical errors in filling empty probabilistic visualizations with probabilities or frequencies

<jats:title>Abstract</jats:title> <jats:p>It has been established that, in Bayesian tasks, performance and typical errors in <jats:italic>reading</jats:italic> information from filled visualizations depend both on the type of the provided visualization and information format. However, apart from reading visualizations, students should also be able to create visualizations on their own and successfully use them as heuristic tools in modeling tasks. In this paper, we first want to broaden the view on Bayesian reasoning to probabilistic tasks with two binary events in general and embed the whole process of solving these tasks using probabilistic visualizations in a modified modeling framework. Thereby, it becomes apparent that most of the steps remained untouched by existing research. Second, in the present empirical study, we focused on one part of the largely unexplored creation process and examined <jats:italic>entering</jats:italic> statistical information into empty visualizations as heuristic tools. <jats:italic>N</jats:italic> = 172 participants had to enter conditional and joint probabilities or the corresponding frequencies into empty visualizations in a paper-and-pencil test. We analyze (a) students’ performance when entering information in visualizations and (b) typical errors, both dependent on the <jats:italic>information format</jats:italic> (probabilities vs. natural frequencies), which <jats:italic>empty visualization structure</jats:italic> (2⨯2 table, double tree, net diagram) was provided, and <jats:italic>type of information</jats:italic> (conditional vs. joint information). The well-known positive effect of natural frequencies on participants’ performance was evident when entering conditional information into 2⨯2 tables and net diagrams. However, with respect to joint information, no superior effect of frequencies was observed. Furthermore, the theoretical implementation of our research in a modeling cycle allows us to identify desiderata for future research.</jats:p>