Reflections on Working with Older Adults in Visualization Research

Visualization research has examined how aging affects perception and cognition in areas such as working memory [28], cognitive load [26, 37], and spatial processing [22, 43]. These changes have motivated a wide range of design considerations, supporting older adults in interpreting and interacting with data. As a result, design considerations such as visualization selection [19, 20], narrative [14], and scaffolding for interpretation [24, 45] have been considered with older adults in mind. At the same time, older adults have been observed to benefit from accumulated or crystallized knowledge [44], which in a visualization context may manifest as visualization literacy [18] as well as factors regarding designs that are familiar [13, 40] or grounded in lived experiences [5]. Beyond visual encodings themselves, the ability to interpret data is also shaped by contextual information. Context for presented information is also important for proper interpretation [5, 6], which, when absent, can lead to reducing a tool’s perceived trustworthiness [23] or usefulness [13]. Specific application areas such as patient-provider communication [41], decision-making [39], and behavioral change [9] have also received some research attention.

Despite this body of work, many of these findings are fragmented across specific domains, authors, and design considerations [48], highlighting the need for broader methodological reflection.

My first study in this area was motivated by older adults’ increasing usage of wearable devices such as smart watches to track and view their own data [7]. As such, we executed a replication study of Blascheck et al. [8] examining the impact of aging on the time needed to view a smartwatch visualization and perform a simple data comparison [47]. In the study, we investigated three visualization types (bar, donut, and radial) depicting three different amounts of data (7, 12, and 24 data points), splitting the older group into young-old (65-74, $N=12$) and old-old (75 and older, $N=12$) groups. We found that older participants encountered broad difficulty with the radial chart and benefited more from donut charts than younger participants. Furthermore, we found strong evidence of differences in time thresholds between the young-old and old-old groups. My work then shifted focus to a non-data encoding component of visualization design, investigating the impact of an interface’s contrast polarity (i. e., light and dark mode) on performance (speed and accuracy) for both younger and older adults [49]. While often framed as an accessibility feature, it was unclear whether this was also the case for data visualization. This suggested that aging does not make contrast polarity more impactful on performance, and that contrast polarity impacted speed as much as visualization type. We also found that, in both groups, a majority of participants’ best-performing polarity did not match their preferred one.

After that, we aimed to dig deeper into understanding how age can affect visual analysis by replicating a study by Saket et al. [42], examining the interactions of aging with low-level visualization tasks (e. g., finding extreme values and retrieving values) and visualization type (bar charts, line charts, pie charts, scatterplots, and tables) [50]. Using Bayesian modeling to tease out both population- and individual-level differences, we observed that older participants generally required more time to complete the tasks, however accuracy between the two age groups was noticeably similar. Furthermore, the best-performing visualization types differed between the age groups, with greater variance exhibited by older adults. Finally, we turned our attention to information displays that older adults use in their daily life. Embedded information displays (EIDs) are interfaces that present information relevant to a device’s operations, now common in home appliances like ovens, washers, and thermostats that may pose usability issues for older adults due to age-related changes [17, 46] via design concerns also noted in visualization studies [25, 36]. We thus conducted a set of focus groups to qualitatively analyze older adults experiences with these displays at a nearby retirement home [51]. The thematic analysis indicated the existence of several compensatory strategies aimed at circumventing difficulties interacting with and reading from their devices. Additional information about all four papers is provided in \autoreftbl:key-papers.

During the analysis for the task–visualization study [50], we observed several instances in which older participants appeared more heterogeneous than younger ones. This was especially evident when examining the tasks where participants achieved their best performance. Whereas more than 85% of younger adults shared the same sets of top-, middle-, and bottom-ranked tasks, the corresponding agreement among older adults was substantially lower, with the highest shared proportion reaching only 67%. This difference was particularly apparent for the middle-ranked tasks, where the five most-common groups of tasks among older adults were shared by 10% (twice), 11%, 15%, and 35% of participants.

A similar pattern emerged in the glanceable visualizations study [47]. In several conditions, the confidence intervals associated with the older group were noticeably wider than those of the younger group. Similar comparisons between the young-old (ages 65-74) and old-old ($\geq$ 75) subgroups even suggested variability within the older participants. Thus, researchers should avoid treating older adults as a homogeneous group, as doing so may obscure important individual qualities that impact performance. When sample sizes permit, I recommend reporting and analyzing results separately for commonly used age groups such as the young-old and old-old, as these groups may reveal differences that are obscured when older adults are analyzed as a single category. Collecting raw age values rather than age ranges can further support this goal by providing a more granular understanding of age-related diversity and enabling analyses beyond predefined age categories. However, age groups alone may not fully capture this diversity, and attributes such as technology experience, health, and disability status can sometimes be more informative when explaining observed differences. When recruiting participants, researchers should also aim for representation from a wide variety of ages above $60$ whenever possible and report their relative proportions. This aligns with recent work spotlighting the study of individual differences [30] and could motivate the use of data analysis methods that effectively model them, such as Bayesian hierarchical modeling [12].

Participant recruitment for in-person studies involving older adults presents unique challenges, as has been noted in broader aging research [16, 34, 35] as well as in human subjects research for accessibility [31]. Based on my experiences, I highlight several practical considerations for researchers working with this population. For the glanceable study [47], recruitment involved identifying community bulletins targeted toward older adults, both digital and in-person. While both were useful, in-person bulletin boards appeared to be significantly more effective, suggesting that physical community spaces may remain valuable recruitment channels for this population. The in-person bulletin boards may have been more effective based on their proximity to where older adults in the area gathered (e. g., a community center), so it is important to consider advertisements locations. In contrast, the EID study [51] relied on collaboration with a nearby retirement home, where recruitment required interacting with both staff and residents. Providing a short introductory presentation proved particularly beneficial, allowing for direct engagement with potential participants, clarification of study goals, and informal snowball sampling. I thus highly recommend this approach to proactive communication and relationship-building.

Additionally, logistical and communication preferences may differ for older adults recruited for in-person studies. Participants for those studies often preferred phone calls for initial discussions to clarify details and emails for scheduling, and using a local phone number helped alleviate concerns about unfamiliar contact information; this aligns with the Communication dimension for recruitment described by Mack et al. [31]. Multiple participants required additional scheduling flexibility due to having a family member or caretaker driving them to and from the study. Ultimately, successful recruitment with older adults seems to benefit from creating in-person connections with the community, establishing trust through direct interaction, and adapting communication methods to participant preferences. However, this work occurred in the U.S., so some takeaways may not apply broadly to other areas and cultures.

Both the glanceable visualizations study [47] and the task–visualization study [50] demonstrated an interesting observation in slightly different ways: older adults can achieve approximately the same accuracy as younger adults while performing visual analysis tasks, however they appear to require more time to do so. This pattern appeared across several tasks, multiple visualization types, and two experimental settings, suggesting that aging may influence the efficiency of visual analysis rather than its correctness. This finding both conflicted with some existing results [29] and challenged assumptions that aging necessarily affects performance [25], instead pointing toward a more nuanced relationship between aging and visual analysis. As such, time becomes a critical, underexamined performance metric, raising methodological questions regarding study duration, pacing, and the interpretation of performance differences across age groups. Researchers should therefore collect and report completion times alongside accuracy whenever possible, rather than treating accuracy as the sole measure of performance. Studies should also account for potentially longer completion times by providing flexible pacing and allocating additional session time for older participants. When study duration becomes a concern, researchers may also need to reduce the number of tasks or conditions to avoid participant fatigue while still allowing sufficient time for task completion. Broader accessibility work has similarly emphasized flexible pacing, breaks, and accommodations for participants with varying abilities and fatigue levels [31, 32]. This motivates further work investigating the nuanced impact of aging on task speed.

The EID [51] and glanceable visualization [47] studies were conducted in-person, whereas the contrast polarity [49] and task–visualization [50] studies were conducted online using Prolific [3]. While both modalities proved valuable, they often exposed different aspects of the older adult experience. Conducting studies in-person provided contextual insights that would have been difficult to obtain remotely. For example, touring apartments and shared spaces within the retirement community before the EID focus groups improved the specificity of discussion prompts and revealed technologies and appliances commonly used by residents. Likewise, conducting the glanceable study in-person caused us to observe participant discomfort looking at one type of visualization for several minutes (shown in \autoreffig:radial24), which required modifying the procedure [47]. Furthermore, the study’s primary input peripheral (shown in \autoreffig:devices), was causing physical discomfort due to being pressed repeatedly for the entire study (45-75 minutes), requiring us to quickly find a more ergonomic replacement [47].

In contrast, online studies highlighted that participants recruited through Prolific were often highly computer literate and comfortable using technology. While this enabled efficient recruitment of older adults at scale, it also raised questions about how representative such participants were of the broader aging population. Consequently, researchers should not assume that older adults recruited online are representative of older adults more broadly. When possible, I recommend reporting participants’ technology experience, recruiting through multiple channels (e. g., online platforms and community organizations), and explicitly acknowledging which segments of the older adult population may be underrepresented. At the same time, online studies provide valuable access to older adults who are comfortable using technology, making them well-suited for rapidly collecting larger samples and studying technology-oriented populations. Online studies are also well-suited for any type of study where personal or home items are of interest (e. g., co-design studies [10]) or transportation for a participant may be difficult [31], since the study can take place via a video call. For research involving technology adoption, accessibility, or everyday use cases, supplementing online recruitment with at least a small in-person version of a study may help reveal otherwise-hidden challenges.

Both the glanceable [47] and task–visualization studies [50] were designed as conceptual replications of prior visualization research (i. e., [8, 42]), examining whether findings established with younger participants would generalize to older adults. To accomplish this, we recruited only older adults and compared their performance against the younger cohorts reported in the original studies. While this approach was efficient, it introduced differences in experimental setup, recruitment procedures, and study context that complicated age-based comparisons. In retrospect, recruiting younger and older adults within the same study would have provided a stronger basis for evaluating age-related differences while simultaneously testing the generalizability of prior findings. Therefore, when adapting existing visualization studies for older adults, I recommend recruiting both age groups whenever feasible. This allows observed differences to be more confidently attributed to age rather than methodological variation between the two studies.

Participant remuneration also presented unexpected situations, particularly for in-person studies with older adults. Our initial plan was to offer Amazon gift cards distributed digitally via email; however, several participants either did not use Amazon or expressed discomfort with online shopping, citing reasons such as lack of familiarity, distrust, or personal preferences. As a result, some participants declined compensation or attempted to transfer it to others. To better align with participant needs, we transitioned to offering physical gift cards sent via mail, with a broader selection of stores (e. g., Wal-Mart). This approach reduced confusion and improved overall acceptance of compensation. With this in mind, I warn that remuneration strategies commonly used in online studies or with younger populations may not necessarily generalize to older ones, and that offering flexible, tangible, and widely accessible compensation options is important when working with older populations.