Page 131 -

5.5 Nonprobabilistic sampling 117

Theoretically, it might be possible to do a random sampling of current screen
reader users (people who are blind or low vision and are already using a screen
reader). However, to do so would require the major screen reader companies
(companies such as Freedom Scientific, AI Squared, and Apple) to collaborate,
share sales data, and any data that they have on users, to come up with a list
of current screen reader users which probabilistic sampling methods could be
applied to. Because the screen reader market is highly competitive, there have
been lawsuits over intellectual property infringement, and there are a number
of partnerships in place (e.g., AI Squared partnered with Microsoft to allow
users of Microsoft Office to download free versions of the Window-Eyes screen
reader), the likelihood of these companies collaborating on research and sharing
data is not very high. Because companies want the Screen Reader Survey to
report favorably on their market share, there can be pressure to “get out the
vote.” While the Screen Reader Survey has a fair number of methodological
flaws, it is the best set of data out there, over a 6-year period, about screen
reader usage. For more information about the most recent Screen Reader
Survey, see http://webaim.org/projects/screenreadersurvey6/. Chris Hofstader
provides an in-depth criticism of the methodology of the screen reader survey at
http://chrishofstader.com/screen-reader-market-figures-my-analysis-of-webaim-
survey-6/. The inherent conflict between the need for the data and the wish for
highly valid data when none is available can be seen in Chris's comment that
“I love numbers and, while the WebAIM survey has some major flaws, it is by
far the best data we have available to us regarding the questions it covers” right
before he provides pages and pages of criticism. ☺

Finally, it is important to note that self-selected, nonprobability-based surveys
may be the most natural data collection method for investigating new user popula-
tions or new phenomena of usage. For instance, if no data exists about a certain user
population or usage pattern, then a self-selected survey of users, asking about usage,
might make the most sense, just as a starting point. The population of interest can be
informed about the survey by posting a message about the survey to a social network-
ing group, listserver, or chat room where members of the population are known to
congregate (Schmidt, 1997).

5.5.5 UNINVESTIGATED POPULATIONS

Surprisingly, there are user groups that have still not been investigated in much de-
tail. For instance, people with certain types of cognitive impairments have yet to
receive much attention, if any, in the HCI research literature (see Chapter 16). For
these populations where no baseline data exists, not enough is already known to
develop hypotheses, experimental design, or well-structured time diaries. Population
estimates may exist on how many people are living with a certain impairment within
a certain country; however, no data exists on how many individuals with a certain