Toward Filling a Critical Knowledge Gap: Charting the Interactions of Age with Task and Visualization

TL;DR

This study investigates how aging interacts with data analysis tasks and visualization types by comparing late adults with younger adults using a Bayesian regression framework and counterfactual simulations. It replicates a prior task-visualization study across five visualizations and ten low-level tasks, using two datasets (Cars and Movies) and 200 PLA participants (final n=148) to model accuracy and timing. The findings show PLA exhibit substantially slower processing yet maintain comparable accuracy overall, with pronounced inter-task heterogeneity and age-dependent shifts in preferred visualizations. These results yield concrete design implications for aging-inclusive visualization, such as incorporating more numerical representations, annotations, and options for personalization to accommodate PLA's diverse needs and time constraints.

Abstract

We present the results of a study comparing the performance of younger adults (YA) and people in late adulthood (PLA) across ten low-level analysis tasks and five basic visualizations, employing Bayesian regression to aggregate and model participant performance. We analyzed performance at the task level and across combinations of tasks and visualizations, reporting measures of performance at aggregate and individual levels. These analyses showed that PLA on average required more time to complete tasks while demonstrating comparable accuracy. Furthermore, at the individual level, PLA exhibited greater heterogeneity in task performance as well as differences in best-performing visualization types for some tasks. We contribute empirical knowledge on how age interacts with analysis task and visualization type and use these results to offer actionable insights and design recommendations for aging-inclusive visualization design. We invite the visualization research community to further investigate aging-aware data visualization. Supplementary materials can be found at https://osf.io/a7xtz/.

Figures (11)

• Figure 1: Demographic information for the 148 PLA participants in the study: (a) age range, (b) gender, (c) visualization familiarity (scale 1-10), and (d) education level. Reported values are the percentage of participants of a given demographic attribute.
• Figure 1: Examples of tasks and visualizations used in the study, replicated from the prior study saket2018task.
• Figure 2: Vision-related information for the 148 PLA participants in the study: (a) use of corrective glasses, (b) use of reading glasses, and (c) reported corrected-to-normal vision impairments. Reported values are the percentage of participants of a given attribute.
• Figure 2: This figure shows the distribution of mean accuracy, aggregated by task. Data shown is the distribution as well as the 50th (median), 66th, and 95th percentiles of all mean accuracies for the 12000 counterfactual participants, aggregated across all tasks for YA and PLA . Tasks are color-coded and sorted highest-to-lowest by median value, and arrows connect tasks across age groups.
• Figure 3: This figure depicts which sets of tasks were most commonly the three highest (Top 3), four middle (Middle 4), and 3 lowest (Bottom 3) rankings for accuracy across all visualizations for both YA and PLA . Sets are presented in alphabetical order, i. e., YA may have had different rankings between Find Extremum, Filter, and Retrieve Value, however overall $91.1\%$ of counterfactual participants had those tasks as their three highest-accuracy tasks. Arrows between sets from YA to PLA indicate a shared set of tasks.