Individuals are often asked to make predictions of the future value of a variable. For example, dispatchers use estimates of patient transfer times to support medical triage and operational decisions. Presenting decision-makers with historical data may help with this task, but the best way to present this information in a decision-aid is still unknown. Furthermore, decision-makers often have contextual information about the current situation that may influence how they interpret these decision-aids. Previously, we compared two types of visualizations of historical data (median-only vs. boxplot) and three types of contextual information (no context information, consequence information, and likelihood information) in an online experimental study of 79 participants. This previous study showed that prediction behavior was significantly impacted by the type of contextual information provided. In this poster, we provide an analysis of the reasons why participants chose these predictions by examining participant self-reports of the type of prediction that they chose (i.e., optimistic, average, or pessimistic cases). We then describe the design of a new experiment using eye-tracking and a think-aloud methodology to further explore participants prediction strategies when using visualizations of historical data. The results of these studies can help us design future decision-aids that present uncertainty information.