Analysis of Navigability of Web Applications for Improving Blind Usability
Hironobu Takagi, Shin Saito, Kentarou Fukuda, Chieko Asakawa · 2007 · ACM Transactions on Computer-Human Interaction (TOCHI), Vol. 14, No. 3, Article 13 · doi:10.1145/1279700.1279703
Summary
This TOCHI article is a two-study investigation into why modern web applications — especially visually dense online shopping sites — remain slow and frustrating for blind users even when they pass automated accessibility checks. The authors argue that current usability theory, such as the Cognitive Walkthrough for the Web, is implicitly built around visual parsing of a page at a glance, whereas blind users must linearise each page and scan it one fragment at a time through a voice browser. Study 1 introduces the Nonvisual Usability Visualization Method, implemented in the aDesigner tool, which models a voice-browser user as a virtual reader that must either play through content line-by-line or jump via heading tags and skip links. The tool produces a 'reaching time' metric — the shortest listening time needed to arrive at the target content — and colour-codes each region of the page accordingly. The authors survey 30 major online shopping sites across the US, UK, and Japan. Study 2 focuses on fine-grained behaviour using the Logging Home Page Reader (LogHPR), a modified IBM voice browser that records every keystroke, navigation command, and TTS output, and a Voice Access Behavior Diagram (VABD) visualisation. Five blind participants performed realistic shopping tasks on Rakuten, Amazon, and accessibility-enhanced versions of the same pages; five sighted participants performed the same tasks for comparison.
Key findings
Only four of the 30 shopping sites in Study 1 reached search results within 20 seconds; 21 exceeded 40 seconds, and the average Japanese site took 178 seconds. Across all sites, just 13.2% of pages had skip links and 39.3% had heading tags, with the UK ahead of the US and Japan, consistent with the effect of the UK Disability Discrimination Act. In Study 1's task comparison, blind users completed a DVD purchase in 563.6 s on average versus 46.3 s for sighted users — more than ten times slower. In Study 2, surprisingly, adding heading tags and skip links to pages did not significantly reduce reaching time or task-completion time. Blind participants mostly ignored logical navigation: 90% of their commands were Next/Previous Link and Next/Previous Item, and none noticed the skip links that had been added to the enhanced pages. Instead they relied on 'scanning navigation' — gambling jumps through the content, then slowing down when they detected an information scent matching the task. They built mental models around landmarks they had discovered incidentally (prices, 'Tax included' phrases, footer text) rather than around heading structure, and reached information scents after sampling only 1–5 words (300–700 ms) per fragment. The authors propose a landmark-oriented navigation model in which both intended landmarks (headings, ARIA roles) and unintended landmarks (scented text) serve as waypoints.
Relevance
This paper is a foundational critique of 'checklist' accessibility — it demonstrates that technical conformance with heading and skip-link guidelines does not translate to measurable gains for real blind users if those features are unfamiliar, inconsistently used, or buried in busy commercial templates. For practitioners, the key takeaways are: (1) measure navigability empirically (reaching time, VABD traces) rather than trusting automated checkers alone; (2) place heading tags and skip-link destinations directly on the main content, not 'just before' it; (3) assume blind users scan on information scent and therefore ensure that distinctive, scent-rich text marks the boundaries of search results and product details; and (4) treat training and consistency across a site as integral to the design, not afterthoughts. The work also seeded the case for ARIA landmark roles and for transcoding-based remediation. Limitations include the small sample in Study 2 (n = 5 blind), reliance on HPR rather than modern screen readers, and the restricted task domain of online shopping.
Tags: web accessibility · screen readers · voice browser · navigation · usability testing · blindness and low vision · accessibility evaluation · assistive technology · accessibility research
Standards referenced: WCAG 1.0 · Section 508 · UK Disability Discrimination Act