Data Mining Accessibility

Pages tested after deployment have more than twice as many accessibility issues. This talk presents the data-driven case for shifting accessibility testing left.

#1about 4 minutes

How large-scale data was collected for accessibility research

The research methodology involved analyzing 14.6 million errors from 6 million URLs to establish a statistically significant dataset.

#2about 8 minutes

Common accessibility issues found through automated testing

Automated analysis reveals that 90% of issues fall into five categories, with missing alt text and navigation problems being the most frequent.

#3about 9 minutes

Analyzing the results of manual accessibility audits

Manual audits on representative samples show that keyboard accessibility and color contrast are top issues, with nearly half being high-severity problems.

#4about 8 minutes

The strong correlation between automated and manual testing

Data shows a significant overlap between automated and manual test findings, with seven of the top ten failing success criteria being identical.

#5about 3 minutes

Why testing before deployment is twice as effective

Pages tested before deployment have less than half as many issues as those tested after, proving the ineffectiveness of a reactive audit-fix cycle.

#6about 8 minutes

Applying extreme programming principles to accessibility

Sustainable accessibility is achieved by integrating practices like early automation, specific acceptance tests, pair programming, and treating accessibility as a core quality problem.

#7about 17 minutes