Automatically Detectable Accessibility Barriers in E-Commerce: Heterogeneity and Multivariate Profiles Across 50 Websites

Abstract

E-commerce accessibility affects autonomous access to goods and services, yet automated-tool counts are often interpreted without considering their distribution, co-occurrence, and limitations. This study characterized the heterogeneity and multivariate profiles of WAVE-detectable barriers across 50 e-commerce websites and derived cautious inclusive-design priorities. We conducted a retrospective cross-sectional observational reanalysis of a public dataset audited on March 25, 2021. Eleven WCAG 2.1 criteria and six structural indicators were reconciled. Analyses included robust descriptive statistics, Spearman correlations with 5,000 permutations and bootstrap samples, Benjamini-Hochberg adjustment, robustly scaled principal components, and hierarchical clustering with stability assessment. Aggregate burden had a median of 12.5 detections per site (IQR: 7.0–62.0), totaling 3,051 errors and contrast errors. Non-text content affected 86% of websites, minimum contrast 80%, and link purpose 78%. By volume, contrast accounted for 54.36% of the 3,019 criterion-assigned detections, followed by non-text content (25.64%) and link purpose (11.59%). The first three components explained 82.55% of variance, but no two- to five-cluster solution yielded a defensible typology: the two-group pattern separated 48 websites from two outlying observations. None of 15 structural associations survived multiplicity adjustment; the largest linked criterion burden and structural elements (rho = 0.402; 95% CI: 0.161–0.611; adjusted p = 0.051). Priorities converged on contrast, text alternatives, and link purpose. These findings describe automated signals from a historical snapshot; they do not establish WCAG conformance, comprehensive accessibility, or users’ lived experience.