Abstract An important part of critical thinking is analyzing the logical structure of inferences made in text, such as checking whether if–then rules appearing in legal regulations or scientific arguments are applied correctly. Metacognitive sensitivity, defined as feeling confident about one's correct answers and uncertain about incorrect ones, is an indispensable component of such assessments because it supports the reliable identification of unstated assumptions and flawed conclusions. In this paper, we explore how visual cues can foster metacognitive sensitivity during the different phases of reasoning in logical inference tasks. We propose a visual cueing framework based on findings from deductive reasoning psychology and multimedia learning. The cues highlight premises and conclusions of problems stated in natural language, and visualize the extracted sequences of conditional statements as diagrams to guide viewers' attention to common invalid inferences. Results from an empirical study (N =164) demonstrate that users exhibit significantly greater metacognitive sensitivity with such visual cues than without.

In the visual analysis of hierarchical data, a main challenge is comparing data attributes both within and between different levels of the hierarchy. Such tasks are typically addressed using aggregated data, where the attribute of a parent node is calculated from the attributes of its children. Our review of existing literature on visualization methods that encode the hierarchy implicitly (e.g., icicle plots, treemaps) shows that most approaches rely on additive aggregation. Proportional aggregation, where a parent is assigned a weighted average of the values of its children, remains unexplored although relevant in practice. We introduce stalactite plots, a visualization technique that displays proportional aggregation and supports visual value comparison. Our empirical evaluation (N = 148, N = 50) shows that, with some explanation, stalactite plots are as easily understood as established visualization techniques for hierarchical data. Furthermore, for large datasets, participants are faster and more accurate using our approach.