The Accessibility of Data Visualizations on the Web for Screen Reader Users: Practices and Experiences During COVID-19
Danyang Fan, Alexa Fay Siu, Hrishikesh Rao, Gene Sung-Ho Kim, Xavier Vazquez, Lucy Greco, Sile O'Modhrain, Sean Follmer · 2023 · ACM Transactions on Accessible Computing · doi:10.1145/3557899
Summary
This comprehensive mixed-methods study examines data visualization accessibility through three complementary investigations: an accessibility audit of 87 COVID-19 visualizations, an online survey of 127 screen reader users, and contextual inquiry with 12 participants. The research addresses a critical equity issue exposed during the pandemic—while data visualizations became essential for public health communication, blind and visually impaired (BVI) users faced significant barriers accessing this vital information. The audit compared two groups: "Top Results" (76 visualizations from high-ranking Google search results including government agencies, news organizations, and research institutions) and "Born Accessible" (11 visualizations specifically designed for screen reader accessibility). Three CPACC-certified auditors evaluated visualizations against 26 criteria derived from the Chartability heuristics, organized into four categories: content, context, navigation, and interactivity. The survey recruited 127 screen reader users (43% totally blind, 32% legally blind, 25% with some vision) through blindness organization mailing lists. The contextual inquiry observed 12 participants using Born Accessible COVID-19 tracking websites, employing think-aloud protocols while they attempted to make decisions about pandemic severity across different U.S. states.
Key findings
The audit revealed stark differences: 82% of Born Accessible visualizations were rated "very" or "extremely" accessible, compared to only 14% of Top Results visualizations. Statistical analysis showed Born Accessible visualizations had nearly 30-fold higher odds of receiving higher accessibility ratings. In Top Results visualizations, only 3 of 26 criteria had majority pass rates: detectability (68%), title access (55%), and no keyboard-overriding commands (100%). Critical failures included: trends accessible to screen readers (5%), visual features (12%), specific data points (21%), accessible tables (41%), adequate summaries (28%), and approachable designs (16%). Many SVG-based visualizations had unlabeled, arbitrarily-grouped elements that were "hard to semantically connect." The survey found 73% of respondents disagreed that data-driven media is typically accessible with their assistive technology, yet 94% expressed concern about accessing accurate COVID-19 information in a timely manner. A striking gap emerged between preferred and actual access methods: 50 respondents preferred tactile graphics for data access, but only 5 actually used them—screen readers (20) dominated actual usage despite not being preferred. 55% rated themselves competent or higher in interpreting tactile graphics, versus only 23% for audio-based methods. In contextual inquiry, participants demonstrated sophisticated strategies for building mental models from accessible representations, but even Born Accessible sites had gaps—sonification couldn't be paused, tables were overly verbose, and spatial relationships remained difficult to convey.
Relevance
This research provides the most comprehensive empirical evidence to date on data visualization accessibility gaps and their real-world impact. The COVID-19 context demonstrates that inaccessible visualizations aren't just inconvenient—they can exclude people from critical public health information during emergencies. For practitioners, the findings offer concrete guidance: (1) Make visualization presence explicit—screen readers often couldn't detect when visualizations existed; (2) Provide comprehensive summaries describing trends, data range, and context before detailed data; (3) Use proper HTML table structure with headers when providing data tables; (4) Ensure SVG elements have meaningful labels and semantic groupings; (5) Connect contextual information (source, update date, methodology) explicitly to visualizations; (6) Expose interactive features to keyboard users. The gap between preferred and actual access methods highlights a systemic failure: while BVI users prefer tactile and audio-based data exploration, the web infrastructure doesn't support these modalities. The finding that only 23% feel competent with audio methods suggests sonification implementations need significant improvement alongside user education. Organizations must consider data accessibility as a core requirement, not an afterthought, especially for public-facing information during crises.
Tags: data visualization · screen readers · visual impairment · web accessibility · accessibility audit · COVID-19 · sonification · tactile graphics · data literacy
Standards referenced: WCAG · Chartability