Examining Image-Based Button Labeling for Accessibility in Android Apps through Large-Scale Analysis
Anne Spencer Ross, Xiaoyi Zhang, James Fogarty, Jacob O. Wobbrock · 2018 · Proceedings of the 20th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS '18) · doi:10.1145/3234695.3236364
Summary
This paper presents the first large-scale analysis of mobile app accessibility, focusing specifically on label-based accessibility barriers in image-based buttons across 5,753 free Android apps. The researchers analyzed three classes of interactive image-based buttons — Clickable Images, Image Buttons, and Floating Action Buttons (FABs) — examining whether they had appropriate content descriptions that would make them usable with screen readers like TalkBack. The study used an epidemiology-inspired framework to structure the investigation, treating accessibility barriers as "diseases" within a "population" of apps. The dataset was drawn from the Rico repository, which contains app metadata and View hierarchies from thousands of Android apps. Three types of labeling errors were assessed: missing labels (no content description at all), duplicate labels (multiple clickable elements on the same screen sharing identical labels), and uninformative labels (labels that convey no meaningful information about the button's function, such as generic terms like "image" or "button"). The researchers also evaluated existing "treatments" — the tools, guidelines, and example code that Android provides to developers — analyzing whether discrepancies in how the three button classes are represented in these resources might correlate with differences in error rates.
Key findings
The distribution of missing labels across apps was strikingly bimodal: 35.9% of apps had 90% or more of their image-based buttons properly labeled, while 45.9% had fewer than 10% labeled, with only 18.2% falling between these extremes. This polarization suggests that apps tend to exist in environments where accessibility is either prioritized or largely ignored. At the class level, Floating Action Buttons had the worst missing-label rate at 92.0%, compared to 86.3% for Clickable Images and 54.7% for Image Buttons. This disparity correlated with gaps in developer resources — FABs were notably absent from Android's accessibility guidelines and example code, while Image Buttons were well-represented. For duplicate labels, 87.1% of apps with labeled buttons had fewer than 10% of those buttons duplicated, though Clickable Image Views had a 46.8% duplication rate among labeled elements. Uninformative labels were the least prevalent error type. The study also found only a very weak correlation between app store ratings and missing label prevalence (rho = -0.05, p = .001), indicating that current rating systems do not meaningfully capture accessibility quality.
Relevance
This research has direct implications for anyone involved in mobile app development or accessibility testing. The finding that nearly half of all analyzed apps had fewer than 10% of their image-based buttons labeled underscores the scale of the mobile accessibility problem. For practitioners, the study highlights that developer tools and guidelines have measurable influence on accessibility outcomes — the classes of buttons best represented in documentation and example code had significantly better labeling rates. This suggests that investing in clear, comprehensive accessibility documentation and integrating accessibility checks into development workflows can drive real improvements. The weak relationship between app ratings and accessibility also argues for dedicated accessibility rating systems or more prominent surfacing of accessibility information in app stores. The epidemiology-inspired framework used here offers a replicable methodology for organizations wanting to audit their own app portfolios at scale.
Tags: mobile accessibility · Android · screen readers · image labeling · large-scale analysis · automated testing · content descriptions
Standards referenced: WCAG 2.0 · Android Accessibility Guidelines