Page 123 -

5.4 Probabilistic sampling 109

not answer their phones as often; there are now several phones per individual; many
people no longer have a landline phone; and government efforts in some countries
have made it so that many individuals are placed on a “do not call” list because they
do not want to receive many types of phone calls (Couper, 2005). When phone sur-
veys are used now, there are often biases in the response.
For research into HCI, the population of interest is generally a bit more focused
than just the general public, or a very broad set of criteria (e.g., registered voters).
Often, there is a much more focused set of inclusion criteria. For example, if the
inclusion criteria relate to being in a specific profession, websites, membership lists,
and social networking for that profession, are great places to start. For instance,
if the survey research targets researchers or practitioners in HCI, commonly used
lists for HCI research are membership directories and social networking groups for
professional organizations (such as SIGCHI, UXPA, and/or HFES). If inclusion
criteria for a survey study relates to having a specific disability, membership direc-
tories of organizations for people with a specific impairment (such as organizations
for people with spinal cord injuries) might be appropriate. If the survey relates to
usage of a certain software application, lists of registered software users from a
company might be appropriate. All of these types of lists may provide information
on postal mailing addresses, phone numbers, or e-mail addresses. There may also be
monthly or annual gatherings at which surveys, or information about surveys, can
be distributed (Lazar, 2006). It is also possible that a website, online community,
or social networking group might provide contact information for a group of poten-
tial respondents. Social networking applications can help recruit participants with a
shared interest, for participation in a survey study (e.g., recruitment information can
be posted on an interest group on Facebook or shared via someone on Twitter who
has a lot of followers with a common interest). However, these methods alone may
not work well for a lot of HCI research.
If the population for a survey is not easily well-defined, then the goal may be
either to get a response that is diverse and represents multiple subgroups within the
respondents or to get a survey response that matches what is known about the popula-
tion (see Section 5.5).

5.4 PROBABILISTIC SAMPLING
The classic use of a survey in sociology is to make estimates for populations. The
most accurate way to do this is by running a census, in which you attempt to get a
survey response from every member of a population. Because a census is often very
expensive and complex, they are not carried out very frequently. When a census is
done, it tends to be sponsored by a large organization or governmental entity (see the
US Census sidebar). If a population of interest is known and very small (say, up to
a few thousand individuals), you might try to organize a modified census, in which
everyone is invited to participate in the survey. However, it is not expected that every-
one will take you up on the invitation and participate (Sue and Ritter, 2007).