Data Shape and Response Modalities Can Bias Estimations of Average Data Location in Visualizations

Our visual system is powerful at extracting summary statistics from data visualizations (Alvarez & Oliva, 2008; Szafi et al., 2016). Yet, our visual recall of summary statistics, for example, recall of average y-position value of line charts, can be systematically biased (Xiong et al., 2020). Potential factors that drive this bias might include data distribution, which changes the resulting shape of the chart, and the modality of the memory representations, which can be visual (e.g., remembering where the values are spatially located), or verbal (e.g., remembering a specific number). To determine how these factors might bias the recall of data, we asked participants (N = 13) to view line charts with a labeled y-axis. The lines were randomly generated following either a positive or negative parabolic data distribution (U-shaped ‘bowl’ versus inverted U-shaped ‘hat’). After a brief delay, participants estimated the average y-position value of the line either verbally, by reporting a number, or visually, by moving a horizontal line representing the average on a blank chart. We found main effects of data shape and response modality, but no interaction between the two factors (F(1, 12) = 0.01, p = 0.9). Positive parabolic ‘bowl’ charts were significantly more underestimated than negative parabolic ‘hat’ charts (t(12) = 4.26, p < 0.01). Verbal reports of average values were significantly overestimated compared to visual reports (t(12) = 5.41, p < 0.01). These results suggest that both the shape of the line and the response modality can bias a viewer’s recall of average positions in visualizations. Future researchers should consider these factors when designing data visualizations and visualization experiments.