"They Aren't Built For Me": A Replication Study of Visual Graphical Perception with Tactile Representations of Data for Visually Impaired Users

TL;DR

It is found that the visually impaired subjects read charts quicker and with similar and sometimes superior accuracy than in those replications of the Cleveland and McGill study on graphical perception using swell form printing.

Abstract

New tactile interfaces such as swell form printing or refreshable tactile displays promise to allow visually impaired people to analyze data. However, it is possible that design guidelines and familiar encodings derived from experiments on the visual perception system may not be optimal for the tactile perception system. We replicate the Cleveland and McGill study on graphical perception using swell form printing with eleven visually impaired subjects. We find that the visually impaired subjects read charts quicker and with similar and sometimes superior accuracy than in those replications. Based on a group interview with a subset of participants, we describe the strategies used by our subjects to read four chart types. While our results suggest that familiar encodings based on visual perception studies can be useful in tactile graphics, our subjects also expressed a desire to use encodings designed explicitly for visually impaired people.

Figures (9)

• Figure 1: Examples of tactile graphics produced using the swell form machine, illustrating a 50% data ratio across four chart types: bar, pie, bubble, and stacked.
• Figure 2: Participants interacting with different types of tactile graphics during the follow-up group interview: (a) Bar Chart, (b) Pie Chart, (c) Stacked Bar Chart, (d) Bubble Chart.
• Figure 3: Midmean errors by chart type for ours (solid lines) and Heer and Bostock (2010, sighted participants). Both datasets show the same general trend, but midmean errors are lower in Heer and Bostock for bar charts and stacked bar charts, a reversal of the overall errors in Figure \ref{['fig:error_rates']}, suggesting outliers may have had an effect in our study.
• Figure 4: Error scores for our study (red) compared with two previous comparable studies: Cleveland and McGill cleveland1984graphical and Heer and Bostock heer2010crowdsourcing. Error bars indicate 95% confidence intervals via bootstrapping. Pie Chart and Bubble Chart were not measured in Cleveland and McGill. Previous studies' results were derived via published graphics rather than raw data.
• Figure 5: Cumulative Distribution Functions (CDFs) of the hierarchical model comparing our data (left) to data from Davis et al (right) davis2022risks. Davis et al. introduced this model to allow for additional effects between participants and visualization types and to learn distributions of error. This plot illustrates that errors within our experiment had similar rankings to Davis et al (H1), but the difference in shapes suggests that our study resulted in fewer small error estimations, indicated by the flat left side of the curves in our plot.