Authors: Aditeya Pandey , Peter Bex , Michelle A. Borkin
Abstract: Categorization tasks are common in everyday life, from sorting objects to a doctor diagnosing a patient's disease. In many categorization tasks, classification information is visually represented. Past work in psychology and information visualization has shown that anthropomorphic representations of data can aid in the quick understanding and recall of information. To utilize this phenomenon, categorization tasks can represent multidimensional binary classification information (e.g, symptom present/absent) as anthropomorphic glyphs. Categorization tasks can utilize this phenomenon and visually represent multidimensional binary classification information (e.g., symptom present/absent in a medical diagnosis) with anthropomorphic glyphs. However it remains to be investigated if anthropomorphic visualizations continue to be beneficial when conveying abstract information that is not directly related to parts of the human body. We study the effects of anthropomorphic and abstract glyph designs on the accuracy of abstract probabilistic categorization tasks. In our within-subject evaluation, 480 participants categorized two of four different glyph visualizations each of which encode 3 abstract probabilistic features. We hypothesized that if visual representation affects accuracy then anthropomorphic glyphs would lead to higher categorization accuracy. However, contrary to our hypothesis, subjects were significantly more accurate at categorization with the most abstract glyph design.