Except for trivial cases, any accessibility evaluation has to be based on some method for selecting pages to be analyzed. But this selection process may bias the evaluation. Up to know, not much is known about available selection methods, and about their effectiveness and efficiency.
The paper addresses the following open issues: how to define the quality of the selection process, which processes are better than others, how to measure their difference in quality, which factors may affect quality (type of assessment, size of the page pool, structural features of the web site).
These issues are investigated through an experimental evaluation of thirteen sampling methods applied to 32000 web pages. While some of the conclusions are not surprising (for example, that sample size affect accuracy), others were not expected at all (that minimal sampling size obtains a high accuracy level under certain circumstances).
