Understanding Screen-Reader Users' Experiences with Online Data Visualizations
Ather Sharif, Sanjana Shivani Chintalapati, Jacob O. Wobbrock, Katharina Reinecke · 2021 · ASSETS '21: The 23rd International ACM SIGACCESS Conference on Computers and Accessibility · doi:10.1145/3441852.3471202
Summary
Data visualizations are ubiquitous on the web, communicating everything from health statistics to financial trends, yet their inherently visual nature creates profound barriers for the approximately 7.6 million screen-reader users in the United States. Sharif et al. conducted two complementary empirical studies to systematically document these barriers. The first was a qualitative study with nine screen-reader users who participated in contextual interviews while interacting with real-world online data visualizations embedded in websites. Participants used their preferred screen readers (JAWS, NVDA, Fusion, or VoiceOver) and were observed via shared screens as they attempted to extract information from visualizations created with three common JavaScript libraries: D3, ChartJS, and Google Charts. The second study was a controlled quantitative experiment with 36 screen-reader users and 36 non-screen-reader users who answered multiple-choice questions about bar, line, and scatter plot visualizations at varying complexity and difficulty levels. The researchers examined both accuracy of extracted information and interaction time across the three visualization libraries, three complexity levels, and three difficulty levels, using mixed logistic regression and mixed-effects ANOVA models to analyze the results.
Key findings
The quantitative results reveal stark disparities: screen-reader users extracted information 61.48% less accurately (34% vs. 87% correct) and spent 210.96% more time interacting with visualizations compared to non-screen-reader users (84.6 seconds vs. 27.2 seconds on average). Among visualization libraries, Google Charts performed dramatically better for screen-reader users (73% accuracy) compared to D3 (17%) and ChartJS (11%), because Google Charts automatically appends a data table to the visualization element. However, even Google Charts' 73% accuracy remained significantly below the 86-89% accuracy non-screen-reader users achieved across all libraries. The qualitative study revealed three core challenges: (1) 33% of visualizations were completely invisible to screen readers — participants interacted with web pages without knowing a visualization was present; (2) when detected, visualizations were labeled meaninglessly as "blank," "graphic," "frame," or "object"; and (3) even when partially comprehensible, the data within visualizations was often inaccessible. Screen-reader users identified four preferred strategies for improvement: tabular data representation, descriptive alternative text, non-visual trend summaries, and multi-modal approaches including sonification and braille printouts.
Relevance
This paper provides the first empirical quantification of the accessibility gap screen-reader users face with online data visualizations, making it essential reading for web developers and visualization library maintainers. The four design recommendations are directly actionable: make visualizations discoverable and comprehensible via proper ARIA attributes and semantic markup; provide both holistic overviews (via alternative text or summaries) and detailed drill-down access (via data tables); auto-generate alternative text from underlying data rather than relying on developers to write it manually; and offer multiple exploration modes (tables, text summaries, sonification) rather than a single approach. The finding that visualization library choice has an enormous impact on accessibility — with Google Charts vastly outperforming D3 and ChartJS — has immediate practical implications for technology selection decisions. For organizations committed to accessibility, this research underscores that data visualizations are a significant blind spot that current WCAG compliance testing may not adequately catch.
Tags: screen readers · data visualization · web accessibility · blind users · low vision · alternative text · information access
Standards referenced: WCAG 2.0 · ARIA