How People with Low and High Graph Literacy Process Health Graphs: Evidence from Eye‐tracking

Abstract

Graphs facilitate the communication of important quantitative information, often serving as effective decision support tools. Yet, graphs are not equally useful for all individuals, as people differ substantially in their graph literacythe ability to understand graphically presented information. Although some features of graphs can be interpreted using spatial-to-conceptual mappings that can be established by adults and children with no graphing experience (e.g., higher bars equal larger quantities), other features are linked to arbitrary graph conventions (e.g., axis labels and scales). In two experiments, we examined differences in the processes underlying the comprehension of graphs presenting medical information in individuals with low and high graph literacy. Participants eye movements were recorded while they interpreted graphs in which information in conventional features was incongruent with that conveyed by spatial features. Results revealed that participants with low graph literacy more often relied on misleading spatial-to-conceptual mappings and misinterpreted the data depicted. Higher graph literacy was often associated with more time spent viewing the conventional features containing essential information for accurate interpretations. This suggests that individuals with high graph literacy are better able to identify the task-relevant information in graphs, and thus attend to the relevant features to a larger extent. Theoretical, methodological, and prescriptive implications for customization of decision-support systems are discussed.Keywords: Graph comprehension, eye movements, medical decision making, individual differences, graph literacy GRAPH LITERACY AND PROCESSING OF HEALTH GRAPHS 3 How People with Low and High Graph Literacy Process Health Graphs: Evidence from Eye-TrackingGraphical displays such as line plots, bar charts, and icon arrays can serve as highly valuable tools for overcoming difficulties in the comprehension of numerical concepts, thus serving as highly effective decision support tools (Ancker, Senathirajah, Kukafka, & Starren, 2006; Garcia-Retamero & Cokely, 2011, 2013 Lipkus, 2007). Unfortunately, graphs are not equally useful for all individuals, as people in the general population differ substantially in their ability to understand graphically presented information (Galesic & Garcia-Retamero, 2011; Kutner, Greenberg, Jin, & Paulsen, 2006). These differences can affect the extent to which individuals benefit from visual displays (Gaissmaier et al., 2012; Garcia-Retamero & Cokely, 2014; Garcia-Retamero & Galesic, 2010; Okan, Garcia-Retamero, Cokely, & Maldonado, 2012). Yet the processes underlying graph comprehension in individuals with varying levels of graph literacy are not well understood. We used eye-tracking methodology to investigate this issue. Individual Differences in Graph LiteracyGraph literacy refers to ones ability to understand graphically presented information and includes general knowledge about making inferences from different graphic formats ...

The experience‐based format of probability improves probability estimates: The moderating role of individual differences in numeracy

“…Second, visual aids containing statistical information tend to be most helpful for people with moderate to high graph literacy—“the ability to evaluate and extract data and meaning from graphical representations of numerical information” (Garcia‐Retamero & Cokely, , p. 7). For instance, people with high graph literacy process icon arrays more efficiently, enabling them to overcome biases, probably due to their more accurate spatial‐to‐conceptual mapping and more time spent on the processing of visual aids (Okan, Galesic, & Garcia‐Retamero, ). Importantly, even among people with low numeracy who benefit from visual aids there is a group of individuals with low graph literacy for whom visual aids are not as helpful in risk comprehension.…”

Designing Graphs that Promote Both Risk Understanding and Behavior Change

“…(29)(30)(31)(32)(33) Unfortunately, there is increasing evidence that people often fail to incorporate information from such conventional features in their interpretations. (34)(35)(36) Eye-tracking data suggest that one of the reasons for this failure is that people do not allocate sufficient attention to such features. (36) In foreground-only graphs, information about the number of people at risk is generally provided in features such as graph titles, (18) legends, (16) or accompanying textual information.…”

“…A key process for accurate graph interpretations is the so‐called integration phase, which involves inferring information from conventional features in graphs (e.g., axes labels, numerical values on scales, legends, or titles) and integrating this information with that inferred from the visual pattern . Unfortunately, there is increasing evidence that people often fail to incorporate information from such conventional features in their interpretations . Eye‐tracking data suggest that one of the reasons for this failure is that people do not allocate sufficient attention to such features …”

“…Unfortunately, there is increasing evidence that people often fail to incorporate information from such conventional features in their interpretations . Eye‐tracking data suggest that one of the reasons for this failure is that people do not allocate sufficient attention to such features …”

Visual comparison of two data sets: Do people use the means and the variability?

“…Indeed, bar graphs also suffer from a 'within the bar' bias, where data points falling within the area of the bar itself are seen as more likely than those appearing outside the bar (Newman & Scholl, 2012). As one would predict, the particular layout and details of the graph play an important role in how the data are interpreted, as does participants' graph-related prior knowledge (Okan, Galesic, & Garcia-Retamero, 2016;Okan, Garcia-Retamero, Galesic, & Cokely, 2012;Shah & Freedman, 2011).To our knowledge, only one study has displayed raw data (that is, each value separately) visually. Fouriezos, Rubenfeld, and Capstick (2008) difference had a large effect on participants' responses, while the sample sizes and standard deviations showed statistically significant, but far smaller, effects on decisions.…”