Page 127 -

5.5 Nonprobabilistic sampling 113

responses are needed. The margin of error is only valid using a true random sample.
In this example, the actual size of the population sampled is irrelevant, since there is
an automatic assumption that all populations being sampled are very large (Babbie,
2009). If the sample is relatively large compared to the population size (more than
5% or 10%), then the margin of error may be smaller, and can be calculated using the
“finite population correction,” which is beyond the scope of this book. Another way
to look at this is that, in a small population size, a smaller sample may be needed.
See Sue and Ritter (2007), Babbie (2009), or Dillman (2000) for more information
on appropriate sample sizes, confidence levels, and margins of error. The reader is
especially encouraged to read (Müller et al., 2014, pp. 238–239), which is specifi-
cally focused on sample sizes in HCI research.

5.4.3 ERRORS
Random sampling seems like an ideal method but it is subject to a number of potential
errors and biases. Careful attention to these potential problems can increase the ac-
curacy and validity of the research findings. For instance, sampling error occurs when
there are not enough responses from those surveyed to make accurate population es-
timates (e.g., if 10,000 individuals are surveyed but only 100 responses are received).
Coverage error occurs when not all members of the population of interest have an
equal chance of being selected for the survey (e.g., if you use e-mail lists or phone
lists to create the sample and not all potential respondents are on those e-mail or
phone lists) (Couper, 2000). Measurement error occurs when survey questions are
poorly worded or biased, leading to data of questionable quality.
Nonresponse error occurs when there are major differences (in demographics,
such as age or gender) between the people who responded to a survey and the people
who were sampled (e.g., if the sampling frame is split evenly by gender, but 90% of
responses are from males) (Dillman, 2000).

5.5 NONPROBABILISTIC SAMPLING
The assumption in Section 5.4 on probabilistic sampling is that the goal is to achieve
a population estimate. In HCI research, population estimates are generally not the
goal. And so, users are more often recruited in a nonprobabilistic manner. Often,
there is not a clear, well-defined population of potential respondents. There is not a
list or a central repository of people who meet a certain qualification and could be
respondents. For instance, due to requirements for patient confidentiality, it would
be very hard to create a sample frame and a strict random sample involving peo-
ple who have, for example, HIV (Müller et al., 2014). That may just be the nature
of the population that no centralized list of potential respondents exists. So strict
random sampling cannot be done. However, valid data can still be collected through
nonprobability-based samples. Nonprobabilistic samples include approaches such as
volunteer opt-in panels, self-selected surveys (where people often click on links on