The term “sampling” refers to the selection of a subset of a population for research.

Since
only in very rare instances can an entire population be studied, sampling has an impact on
virtually all the research discussed in this book. As we will see below, how the sampling is
conducted determines what information is acquired and influences the researcher’s
interpretation of the results in important ways. The key terms used in sampling are
explained in Box 7.1. We’ll begin with a discussion of sampling in studies involving
structured interviews or questionnaires, and then consider the techniques used for other
methods of quantitative research. Finally, ways to increase response rates are presented.

Suppose you are wondering about your fellow students’ attitudes on certain matters, or
perhaps about their behaviours or backgrounds. To examine any of these areas, either
structured interviews or questionnaires would be appropriate. However, let’s say that
there are around 30,000 students at your university. It would be a huge undertaking for you
to arrange for all of them to fill out a questionnaire, and the time and energy required to
interview 30,000 people would be prohibitive. It’s almost certain that a sample of students
would have to suffice. In fact, the need to sample is almost universal in quantitative
research. But would any sample do? What if you gave questionnaires to everyone in your
classes? Or if you stood in a busy location on campus and then interviewed anyone who
agreed to speak with you? If your goal is simply to get some experience in data gathering
and analysis, or to conduct a pilot study, these samples would be fine. However, if the
object is to get a representative sample, one that can be used to make inferences about all
30,000 students, these samples would not do. Why? There are various reasons, but the
following stand out:

•If you distribute questionnaires in your classes, anyone not taking a course with you will
be excluded. How many music or engineering majors take sociology?

•If you go out looking for people to interview, your sample will exclude anyone who
happens to be somewhere else at that particular time. Some students will be at work,
others will have skipped classes that day, and many will have a timetable different from
yours. Some students may never need to go where you are.

• Your decisions about which people to approach may be influenced by your

judgments about how friendly or cooperative they appear to be. Gender, “race,” and age
are only a few of the factors that could have an impact on your selection of prospective
respondents.

The problem with these sampling methods is that they make the choice of possible
participants dependent on non-universal criteria: availability in one case and personal
judgment in the other. In the language of sampling, this means that the sample would be
biased: it would not represent the population from which it was selected. It is extremely
difficult to remove bias altogether and get a truly representative sample, but steps should
be taken to keep it to a minimum.

Three sources of bias can be identified:

- Not using a random method to pick the sample. Essentially, a random method is
one in which each element has the same chance of being selected, like numbers
from a lottery cage (although in some circumstances that principle may be violated
for technical reasons; for example, in order to get a sufficiently large number of
people from a small subgroup into the sample). As we saw above, if a random
method is not used, there is a risk that the selection process will be affected by
human judgment, so that some members of the population are more likely to be
selected than others. This source of bias is eliminated through the use of probability
sampling, described later in this chapter.
- The sampling frame or list of potential participants is inadequate. If the sampling
frame excludes some cases or is otherwise inaccurate, the sample derived from it
may not represent the population, even if a random sampling method is employed
to select participants from it.
- Some people in the sample refuse to participate or cannot be contacted—in other
words, there is non-response. The problem with non-response is that those who
agree to participate may differ from those who refuse or cannot be reached in ways
relevant to the subject matter being investigated. If that is the case, the sample may
not be representative. For example, if you are studying rates of criminal
victimization and people who have been victims of violent crime decline to
participate, that would skew the results.

If the data are available, a researcher can check to see how non-respondents differ from
the population. It is often possible to do this in terms of characteristics such as gender or
age, but these are not the only potentially relevant factors; in the case of a sample of
university students, for example, you might want to find out whether non-respondents
reflect the entire student body in terms of faculty of registration. However, it is usually
impossible to check characteristics that go beyond basic demographic categories, such as
attitudes toward separatism or patterns of smoking behaviour, because population data
are generally not available on these topics.

Sampling error
To appreciate the significance of sampling error for achieving a representative sample,
consider Figures 7.1 through 7.4. Imagine a population of 200 people and a sample of 50.
Imagine as well that the topic is whether people watch reality TV programs, and that the
population is divided equally between those who do (100) and those who do not (100). If
the sample is representative, the sample of 50 should also be equally split in terms of this
variable (see Figure 7.1). If there is a small amount of sampling error, so that the sample
contains one person too many who does not watch reality TV and one too few who does,
the distribution would look like Figure 7.2. Figure 7.3 shows a more serious over-
representation in the sample of people who do not watch reality programs. This time there
are three too many who do not watch them and three too few who do. Finally, Figure 7.4
shows a very serious over-representation of people who do not watch reality TV: 35, as
opposed to 25 if the sample were perfectly representative. It’s important to appreciate that
the possibility of sampling error can never be completely eliminated. Even with a well-
crafted probability sample, a degree of sampling error is likely to creep in, just as flipping a
coin 100 times is likely to yield more 49Heads/51Tails and 51Heads/49Tails results
combined than 50Heads/50Tails outcomes, even though the latter is the most probable
single result. Ask your instructor to take a few minutes for an in-class experiment in which
a coin is flipped 20 times, several times over, and you’ll see the surprisingly large number
of outcomes that are not 10Heads/10Tails. However, probability sampling stands a better
chance than non-probability sampling of minimizing sampling error, so it’s very unlikely
that the sample will end up looking like Figure 7.4 when it is used. Moreover, probability
sampling allows you to employ tests of statistical significance that will permit you to make
inferences about the population from which the sample was selected, with a known
probability of error. This topic will be addressed in Chapter 8.

Simple random sample

The simple random sample is the most basic form of probability sample. With it, each unit
of the population has an equal probability of inclusion in the sample.

The key steps in devising a simple random sample are as follows:

1. Define the population. Here N is all full-time students at the university: 9000.

2. Select or devise a comprehensive sampling frame. In this case, let’s assume that (a)
the office that keeps student records has a list of all full-time students and (b) you have
access to it (and all proper ethics protocols have been followed).

3. Decide on the sample size (n) you will use; here n is 450.
4. List all the students in the population and assign them consecutive numbers from 1
to N, here 1 to 9000.

[Link] a table of random numbers, or a computer program that can generate random
numbers, select n (450) different random numbers from 1 to N (9000).

[Link] students who match the n (450) random numbers constitute the sample. Two points
are worth noting here. First, there is almost no opportunity for bias, since in this case the
selection process is entirely random: students are not selected on the basis of subjective
criteria such as whether or not they look friendly and approachable. Second, selection
does not depend on availability: students do not have to walk by the interviewer in order to
be included in the sample. The process of selection takes place without their knowledge,
and they do not know they will be asked to be part of a social survey until they are
contacted by someone associated with the study.

Systematic sample

A systematic sample is selected directly from the sampling frame, without using random
numbers. In the present case, 1 student in 20 is to be selected (450/9000). That “1 in 20”
figure is called the sampling interval. A random start begins the process, which in this case
can be achieved by selecting at random a number from 1 to 20, possibly by using the last
two digits in a table of random numbers. With the 10 random numbers above, the first
relevant one is 54016, since it is the first one where the last two digits yield a number in the
desired range (1–20), namely 16. This means that the sixteenth student in the sampling
frame is the first one to be included in the sample. Thereafter, take every twentieth student
on the list; the sequence of case numbers would be: 16, 36, 56, 76, 96, 116, and so on.
Who is the last person in the sample? Case number 8996. In systematic sampling, it is
important to ensure that there is no inherent ordering or pattern in the sampling frame, a
feature called periodicity. For example, imagine that the sampling frame was set up such
that case 1 was male, case 2 female, case 3 male, case 4 female, and so on for all cases.
In the example above, that would yield a sample of 450 females and 0 males, in all
likelihood a very unrepresentative sample. If there is any pattern to the list, arrange the
cases in random order or choose a different sampling method. Note that with systematic
sampling, not every possible combination of cases has an equal chance of being selected.
Once that first case number (here 16) is chosen, every case number in the sample must
end in 6, and no even-numbered prefixes will qualify, such as 26, 86, or 166.

Stratified random sampling

In the imaginary study of university students discussed earlier, the student’s discipline
may be relevant to what you will be researching. Generating a simple random sample or a
systematic sample may yield one in which, for example, the proportion of kinesiology
students in the sample is the same as that in the student population, but usually it is not an
exact match. Thus, if there are 1800 students majoring in kinesiology, a sampling ratio of 1
in 20 should produce 90 students in the sample from this faculty. But because of sampling
error there might be, say, 85 or 93. So long as data are available on each student’s faculty
of registration, it’s possible to ensure that students are exactly represented in terms of
their faculty membership by using stratified random sampling. In the language of sampling,
this means stratifying the population (dividing it into subgroups) by a criterion (in this case
faculty) and selecting either a simple random sample or a systematic sample from each of
the resulting strata. In the present example, five faculties would mean five strata, with the
sample in each stratum being one-twentieth of the total for each faculty, as in Table 7.1,
which also shows a hypothetical outcome of a simple random or systematic sample, one
that does not exactly mirror the population. The advantage of stratified sampling is clear: it
ensures that the sample is distributed in the same way as the population in terms of the
stratifying criterion. Using a simple random or systematic sampling approach may result in
a distribution exactly like that of the stratified sample, but it is unlikely. On the other hand,
this strategy requires that the relevant criteria for stratification be known in advance of the
research, which is not always the case. Two more points are relevant here. Stratified
sampling is feasible only when it is relatively easy to identify and allocate units to strata.
Five strata based on physical activity level, from “competitive athletes” at one end to
“couch potatoes” at the other, would be very hard to set up, because it would require an
initial study just to get the population data on activity level that you would need to create
these strata. Second, more than one stratifying criterion can be used at the same time. In
our example, we could stratify by faculty and gender, or faculty, gender, and whether
students are undergraduates or postgraduates, and so on, as long as the criteria are easy
to use and relevant to the research question. But this approach is practicable only when
data on the relevant stratifying criteria are available for the population. Does your school
have an accessible list of female undergraduate engineering students?

Multi-stage cluster sampling

In the previous example, the students to be interviewed all attend the same university.
However, imagine a national sample of students. A major problem with large populations
is the general lack of an adequate sampling frame. Is there a list somewhere of all students
registered at Canadian universities? No. Without a sampling frame, it’s impossible to
select a simple random or systematic sample. One way to deal with these problems is to
employ cluster sampling. With cluster sampling, the primary sampling unit (the first stage
of the sampling procedure) is not the individuals or units of the population to be studied,
but an aggregate of them, known as a cluster. Imagine you want a nationally representative
sample of 5000 students. One solution is to sample universities and then students from
each of the sampled universities, with a probability sample at each stage. For example,
you could randomly sample 10 universities from the entire population of universities,
yielding 10 clusters, and then interview 500 randomly selected students at each of the 10
universities selected, to create a sample of 5000 students. (This would require “probability
proportionate to size” sampling because each university has a different number of
students, but that technique goes beyond the scope of our discussion here.)

BOX 7.2 Generalizing from a random sample to the population

If an average of 9.7 units of alcohol is consumed in the previous seven days by

respondents in a probability sample, would a similar figure be found in the population? The
answer is complex but it is sketched out here and in Box 7.3. Assume for the sake of
illustration that it would be possible to take an infinite number of random samples of the
same size from a population. Not all the resulting sample means would be perfect
estimates of the population mean. We can never know the population figure with 100 per
cent certainty, but we can imagine another study with a sample mean of 9.6 units, another
with 9.8, and so on. These outcomes, as long as there are enough of them and the sample
size is sufficiently high, will take the form of a bell-shaped curve known as a normal
distribution (see Figure 7.5) with the sample means clustering around the population
mean. Of the sample means not exactly at the population mean, half will be below the
population mean and half above it. Moving to the left or to the right (away from the
population mean), the curve tails off, reflecting the smaller and smaller number of
samples generating means that depart considerably from the population mean. The
variation of sample means around the population mean is the sampling error and is
measured using a statistic known as the standard error of the mean. This is an estimate of
the amount by which a sample mean is likely to differ from the population mean.
(Technically, it is the standard deviation of the sampling distribution of the mean; see Box
7.3). The standard deviation of a set of scores is a measure of how dispersed or spread out
the scores are around the mean, and is calculated by taking each score, subtracting the
mean, squaring that number, adding up all the resultant scores, dividing that sum by the
number of scores, and then square rooting that number. (See Chapter 8 for more on
standard deviation.)

Sample size

“How large should a sample be?” “Is the sample big enough?” No issue in sampling raises
more questions than sample size. Decisions about sample size are not straightforward, as
we will see below. At the same time, the quest for the perfect sample is usually
constrained by the practical considerations of cost and time.
Absolute and relative sample size

People unfamiliar with the field may be surprised to learn that it is the absolute size of a
sample that is most important, not the proportion of the population that it makes up. This
means that a national probability sample of 1000 Canadians has about as much validity as
a national probability sample of 1000 Americans, even though the latter involves a much
larger population. Increasing the size of a sample increases the precision of the estimates
derived from it; for instance, a larger sample size means that the 95 per cent confidence
interval referred to in Box 7.2 is narrowed. However, a large sample cannot guarantee
precision, so it might be better to say that increasing the size of a sample probably
increases the precision of the estimates it can create. In other words, as sample size
increases, sampling error tends to decrease. Common sample sizes are 100 (minimum)
and then 400, 900, 1600, and 2500. These sample size increases cut the sampling error by
half, then by one-third, then by one-fourth, and then by one-fifth respectively. (The reason
has to do with the square roots involved in the denominator of the calculation; thus the
original square root, 10, for a sample of 100 becomes 20, 30, 40, and 50 [the square root of
2500], yielding the one-half, one-third, one-quarter, and one-fifth.) Notice that the
increases in precision become less pronounced (going from 100 to 400 reduces sampling
error by half, but 400 to 900 reduces it by only one-third) and thus the rate of reduction in
the standard error of the mean declines. Sample size is likely to be profoundly affected by
considerations of time and cost at such a juncture, since striving for smaller and smaller
increments of precision becomes increasingly uneconomic as ever-larger samples
become less and less cost efficient. Fowler (1993) warned against simple acceptance of
this size criterion and argued that it is not normal for researchers to be able to specify in
advance “a desired level of precision” (p. 34). Moreover, since sampling error is only one
component of any error entailed in an estimate, the notion of a desired level of precision is
not realistic. Instead, to the extent that this notion affects decisions about sample size, it
usually does so in a general rather than a calculated way. It is usually cost that ultimately
determines sample size.

Non-response

Considerations about sampling error do not end with issues of sample size. It’s also
important to bear in mind the problem of non-response. The response rate can be defined
as the percentage of the sample that actually participates in the study. It is often a good
deal less than 100 per cent, partly because there are usually some people in the sample
who will either refuse or be unable to participate, who cannot be contacted, or who turn
out to be unsuitable for the study. Others will be excluded if they leave a large number of
questions unanswered, or if they do not take the interview or questionnaire seriously.
Unfortunately, survey response rates have been in “steep decline” for decades (Johnson &
Wislar, 2012), with the added problem that there has been no consensus among
researchers regarding how to define and calculate response rates. In their research on
sustainable consumption in the Edmonton area, Kennedy, Krogman, & Krahn (2013)
reported a response rate of 69 per cent. Schieman & Narisada (2014), who used a national
sample of the Canadian workforce to examine people’s sense of mastery over their lives,
achieved a response rate of about 40 per cent. However, non-response bias (the extent to
which people who are selected for the sample and participate in the study differ in relevant
ways from those who are selected but do not participate) is increasingly considered to be
more important than the response rate itself (Johnson & Wislar, 2012). Are the people who
have been selected but cannot be contacted richer and on vacation? Are they ill and in
hospital? Such differences may be important (depending on the research topic) and may
have to be taken into account in the analysis. Researchers sometimes take steps to
determine the level of non-response bias, as Schieman & Narisada (2014) did when they
concluded that it was not an issue in their study. Given that 100 per cent response rates
are rare, sample size may have to be adjusted. For example, if the goal is a sample of 450
students and earlier surveys suggest that the response rate will be approximately 80 per
cent, it may be advisable to sample 575 individuals, on the assumption that approximately
115 (20 per cent) will not participate, leaving 460 and still room for some additional
refusals. How far should researchers go to boost response rates? Smith (1995) showed
that response rates are affected by things like the subject matter of the research, the type
of respondent, and the effort expended on obtaining the cooperation of prospective
respondents. Later in the chapter, a number of steps that can improve response rates to
questionnaires (which are particularly prone to poor response rates) are discussed.
However, boosting response rates can be expensive. Teitler, Reichman, & Sprachman
(2003) discussed the steps taken to increase the response rate of a US sample that was
hard to reach, namely both unwed parents of newborn children. They found that although
increasing the sample from an initial 68 per cent to 80 per cent of all couples meant that
the final sample more closely resembled the population from which it had been taken,
diminishing returns set in. This does not mean that steps should not be taken to improve
response rates, however. For example, following up on people who do not respond to a
questionnaire usually results in an improved response rate, at little additional cost.

Heterogeneity of the population

Yet another consideration is the heterogeneity of the population from which the sample is
to be taken. When a population is heterogeneous with regard to characteristics relevant to
the topic under study, as a whole country or city normally is, the samples drawn are likely
to be highly varied. When it is relatively homogeneous, such as a population of students or
members of a particular occupation, the amount of variation is less. Generally, the greater
the heterogeneity of a population, the larger a sample should be in order to maximize the
chances that all groups will be adequately represented.

Kind of analysis

Finally, researchers should bear in mind the kind of analysis intended. An example is a
contingency table showing the relationship between two variables. In a 2 × 2 table there
are four cells into which the cases can fall, let’s say, gender (M/F) × whether the person
voted in the last election (Y/N). Suppose there were only 10 males and 10 females in the
sample, and that 50 per cent of the men and 50 per cent of the women voted, showing no
relationship between gender and voting. But suppose the sample by chance included one
more woman who voted and one more man who did not. Now the figures show that 40 per
cent of men voted versus 60 per cent of women—quite a difference. To prevent such a
small change in the sample from having such a major impact on the estimate, a larger
sample is required. You would probably want to have at least 50 males and 50 females.
Because the population is approximately half male and half female, this is easy to achieve.
But what if the variable is religion? Fifty Roman Catholics (or 50 Anglicans) are easy to find
in any random sample of 1000 Canadians, but what about 50 Jehovah’s Witnesses or 50
Hindus? One strategy to achieve statistically meaningful results is to randomly over-
sample some of these small groups, as Dinovitzer, Hagan, & Parker (2003) did to guarantee
sufficient social class variation in their study comparing educational attainment among
immigrant and non-immigrant Canadian youth. Similarly, the Canadian Election Study
usually over-samples the less populous provinces (such as Prince Edward Island) in order
to make interprovincial comparisons. (A statistical adjustment can be made to
compensate for the over-sampling when the data are used for purposes other than
between-group comparisons.) This illustrates how sample size depends in part on the
analysis to be conducted.

Types of non-probability sampling

Non-probability sampling includes all forms of sampling that are not conducted according
to the canons of probability sampling outlined earlier. It covers a wide range of different
types of sampling strategy, at least one of which—the quota sample—is claimed by some
practitioners to be almost as good as a probability sample. This section covers three
common non-probability samples: the convenience sample; the snowball sample; and the
quota sample.

Convenience sampling
A convenience sample is one that is used simply because the elements are readily
available to the researcher. Imagine a professor in a faculty of education who is interested
in the qualities that teachers want in their principals, and who has several classes made
up of teachers working part-time toward master’s degrees. Now suppose that she decides
to administer a questionnaire to those students—a sample of convenience. Chances are
that she will receive all or almost all of the questionnaires back, thus achieving a high
response rate. However, the use of the convenience sampling strategy makes it impossible
to generalize the findings to teachers as a whole: the very fact that her students are taking
this degree program marks them off as different from teachers in general. (In addition, of
course, it’s ethically questionable for a researcher to ask people who are dependent on
them for a grade to take part in their research.)

RESEARCH IN THE NEWS The perils of polling

An article in the Winnipeg Free Press (Taniguchi, 2020) cites a 2020 poll conducted by a
local survey research company indicating that a greater number of Manitobans believe that
racism is a problem in the province than was the case five years earlier. A survey done by
the company in 2015 suggested that 44 per cent of Manitobans thought that racism was a
problem in their province, compared to 61 per cent in 2020. However plausible these
figures may be, the samples were not selected randomly, so there is no way to estimate
how accurate the results were. To their credit, both the author of the newspaper article and
the research firm on its website state that because of the sampling procedures used, “no
error rate can be calculated.” This is not to say that the results have no value—it’s quite
possible that more Manitobans are becoming aware of racism as a social problem.
According to Lori Wilkinson, a sociologist at the University of Manitoba, “more people are
seeing it [racism]. You ask people their opinions about anything, and it really depends on
what’s newsworthy or what’s being discussed online” (Taniguchi, 2020). Wilkinson
maintains that more people are becoming aware of racism because of recent worldwide
anti-racism protests and the extensive media attention they have received. Although
representative sampling may not be feasible in a number of research situations, the results
produced by non-probability sampling should be interpreted with caution. At a minimum,
the researcher should, as was done here, state explicitly that there is no way to gauge their
accuracy. This is not to suggest that convenience samples should always be avoided. In
fact, they are used more often than you might think. Mairs & Bullock’s (2013) research on
sexually risky behaviour and HIV testing on the part of Canadian “snowbirds” in Florida
was based on a convenience sample, as were MacKinnon & Luke’s (2002) study of cultural
change, and Beagan’s (2001) analysis of medical students mentioned in earlier chapters.
However, studies involving convenience samples should acknowledge that they are of
limited use in estimating population characteristics. Convenience samples are often
utilized in pilot studies. Imagine that part of the education professor’s research involves
developing a battery of questions to measure the leadership preferences of teachers that
she plans to use later, with a different sample. Since it is highly desirable to pre-test such a
research instrument before actually using it, administering it to a group that is not part of
the main study is a legitimate way of exploring issues such as whether respondents will
reply honestly to questions on sensitive topics, and whether the questions are clear and
comprehensible. A convenience sample may also be used to test newly created scales for
reliability and to generate ideas for further research. Box 7.4 contains another illustration
of how a convenience sample may be used.

BOX 7.4 A convenience sample

Suppose you want to study university undergraduates to find out the extent of their part-
time employment while at school. To save time and effort you decide to conduct the study
at your own university. You could use probability sampling to get participants, but you think
the research is not important enough to justify the cost. Instead, you choose five faculties,
and within each, one or two of the specific degree programs it offers: anthropology, visual
arts, chemistry, electrical engineering, French, kinesiology, media studies, and
psychology. This choice of participants is designed to maximize variety in the types of
degree programs represented while providing similar numbers of males and females (since
in particular degree programs one gender frequently predominates). Questionnaires can
then be given to students in first-, second-, third-, and fourth-year courses. Together, these
procedures represent a good attempt at generating a varied sample. It is a convenience
sample, because the choices of university, faculty, degree program, and course are made
purposively rather than randomly. Because of the way the questionnaires are administered
(in class), there is a very high response rate. On the other hand, students who are absent
from class do not get a chance to fill out a questionnaire. An important question is whether
absence from class is connected in some way to part-time work, the latter variable being
the focus of your study. In other words, is absence higher among students who work part-
time, perhaps because they are working at the time of the class? Are some students too
tired to go to class because of their part-time work? If so, you will probably underestimate
the proportion of students who work part-time. Also, because you have selected a
convenience sample, you must say so in your research report, acknowledging that the
results cannot be generalized to Canadian students as a whole or even to the student
population at your university.

Snowball sampling

Snowball sampling is a form of convenience sampling, but worth distinguishing because it

has attracted quite a lot of attention over the years. With this approach, the researcher
makes initial contact with a small group of people who are relevant to the research topic
and then uses them to establish contact with others. Tastsoglou and Miedema (2003) used
this approach to create a sample of immigrant women in the Maritimes. Bowen (2015)
used it to recruit informants for her study of people in the Vancouver area who either had
experienced the transition from off-street sex work to “square” (non-sex) work, or were
simultaneously employed in both. And Whalen and Schmidt (2016) chose snowball
sampling to recruit participants for their study of Newfoundland women whose male
partners were away from home, working in the Alberta oil industry. Box 7.5 describes the
generation of a snowball sample of marijuana users for what is often regarded as a classic
study of drug use. Becker’s comment on creating a snowball sample is interesting: “The
sample is, of course, in no sense ‘random’; [indeed] it would not be possible to draw a
random sample, since no one knows the nature of the universe from which it would have to
be drawn” (Becker, 1963, p. 46). What Becker is saying here is that there is no accessible
sampling frame for the population from which the sample is to be taken. The difficulty of
creating such a sampling frame means that a non-probability approach is the only feasible
one. Moreover, even if it were possible to create a sampling frame of marijuana users or
British visitors to Disney theme parks, it would almost certainly become obsolete very
soon, because those populations are constantly changing. People become and cease
being marijuana users all the time, while new theme park visitors arrive every minute. The
problem with snowball sampling is that it is very unlikely to be representative of the
population, although, as Becker’s experience suggests, the very notion of a population is
problematic in some circumstances. However, snowball sampling is generally used not
within a quantitative research strategy, but within a qualitative one: both Becker’s work on
marijuana users and Bryman’s (1995, 1999) studies on theme parks used a qualitative
research framework, and concerns about the ability to generalize do not loom large in
qualitative research. Indeed, Tastsoglou and Miedema (2003) warned their readers not to
generalize from their 40 women to all immigrant women in the Maritimes, much less to
those in the rest of Canada. This is not to suggest that snowball sampling is entirely
irrelevant to quantitative research: when the researcher needs to focus on relationships
between people, tracing connections through snowball sampling can be a better approach
than conventional probability sampling. Statistically small groups—gay francophone
lawyers in New Brunswick, for example—are researchable with a snowball approach.
Taking a random sample from that population would not be feasible. Quota sampling.
Quota sampling is not a predominant form of sample selection in academic social
research, but it is widely used for commercial purposes in market research and in political
opinion polling in some countries. However, the advent of the online opt-in survey panel
has made quota sampling more common in academic research in recent years. Opt-in
panels are made up of people who are willing to participate in online surveys, often in
exchange for credit points or small amounts of money. Third parties who need participants
for their research contact the people in charge of the panel and pay a fee to have their
surveys administered to people on the panel. The advantages of using opt-in panels
include fairly quick turnaround times and a relatively low cost.

PRACTICAL TIP Sample size and probability sampling for students

For most quantitative researchers sample size does matter: the bigger the sample, the
more representative it is likely to be (provided the sample is randomly selected), regardless
of the size of the population from which it is drawn. However, students have to do their
research with very limited resources, and a truly random sample may not be feasible for
them. The crucial point is to be clear about your sampling decisions and to justify what you
have done. Explain the difficulties that you would have encountered in generating a
random sample. Explain why you could not increase your sample size. Above all, though,
don’t make claims about your sample that are not warranted. Never claim that your
sample is representative, or randomly selected, unless that is true. Also, even if your
sample is small, it may have some good features: the range of people included, a good
response rate, high levels of honesty. Make sure to play up these positive features while
being honest about the sample’s limitations. This is what professional researchers do. The
aim of quota sampling is to produce a sample that reflects a population in terms of the
relative proportions of people in different categories (gender, ethnicity, age,
socioeconomic status, region of residence, etc.), or in combinations of these categories.
Unlike a stratified sample, however, a quota sample is not random, since the final
selection of people in each category is left up to the researcher or the organization in
charge of the opt-in panel. For Canadian studies, information about the stratification of the
Canadian population or about certain regions can be obtained from sources such as the
census and from surveys based on probability samples such as the General Social Survey,
which can then be used to establish the proportion of the sample that is to be made up of
people in the various categories. Once the categories and the numbers of people to be
surveyed within each one (the quotas) have been decided, the researchers or those
contracted by them to select the sample try to pick people who fit those categories. The
quotas are typically interrelated. As in stratified sampling, the population may be divided
into strata covering several characteristics (for example, gender, employment, and age) all
at once (see Table 7.2). For example, fifty 25- to 44-year-old unemployed females may be
required to meet the quota. If necessary, statistical adjustments can be made to the
sample data such that the relative sizes of the categories match those of the population.

Nominal variables. These variables, also known as categorical variables, are composed of
categories that have no relationship to one another except that they are different. In the
case of religion, for example, the categories might be Roman Catholic, Protestant, Jewish,
Muslim, Hindu, Other, and No Religion. Any two research participants will be either in the
same category or in different categories: no other kind of comparison is possible. This
means that the order of the categories is arbitrary: it could have begun with No Religion
and ended with Roman Catholic. Switching the order has no implications for the way the
data are interpreted.

• Ordinal variables. With these variables the different categories can be rank ordered.
That means that “greater than” (>) and “less than” (<) statements can be made about the
categories and the people in them. For example, in question 5 “always” indicates greater
frequency than “usually,” which indicates greater frequency than “rarely,” and so on. Note
also that it would be illogical to re-order the categories as, say, “always, never, sometimes,
usually.”

The distance or amount of difference between the categories, however, may not be equal
across the range. Thus, the difference between the categories “always” and “usually” is
probably not the same as the difference between “usually” and “rarely,” and so on. This is
because no unit of measurement is used for this question. Ordinal variables are similar to
words that have either a comparative “er” or a superlative “est” attached to them, as in
“taller” and “tallest.”

• Interval/ratio variables. These are variables for which a unit of measurement exists
and thus the distances or amount of difference between the categories can be made
identical across the range of categories. The values take the form of actual numbers (such
as 1, 2, 7.5, etc.). In the case of variables var00010 to var00012, the unit is the minute and
the difference between the categories is a one-minute interval. Thus, a person who spends
32 minutes on cardiovascular equipment is spending one minute more than someone who
spends 31. That difference is the same as the difference between someone who spends 8
minutes and another who spends 9 minutes on the equipment. At the interval/ratio level of
measurement, it is possible to say that two scores are equal or not equal (as at the
nominal level), for example, 32 ≠ 31; that they have greater or lesser values (as at the
ordinal level), for example, 32 > 31, 8 < 9; and that the distance or amount of difference
between them can be specified, for example, 32 − 31 = 1. This is the highest level of
measurement, the one that allows the widest range of analysis techniques, including
addition, subtraction, multiplication, and division.

Credibility

The idea of credibility is tied to the notion that different people may interpret the social
world in different ways. Since there can be several possible ways of experiencing or coming
to terms with a particular social situation or event, the researcher has to ensure that the
interpretations presented in the study ring true to the people observed. The establishment
of credibility entails both following proper research procedures and submitting the findings
to the people studied for confirmation that the account is consistent with the way they see
their world. The researcher is, after all, describing their reality. This practice of getting
feedback on the study from the participants involved is referred to as respondent
validation, member validation, or member reflection. The aim of respondent or member
validation is to seek corroboration or criticism of the researcher’s observations and
interpretations. It can take several different forms:

• Each research participant may be provided with an account of what he or she said
to the researcher or others. For example, in a study that explored issues of trust between
mothers of children with disabilities and school principals, Shelden. Angell, Stoner, &
Roseland (2010) gave all respondents a summary of the findings by either telephone or
email. They also asked each participant if their views were represented in the findings, and
whether verbatim quotes were accurate.

• Researchers may show pre-publication versions of their articles or books to their

participants. They may also provide the study participants with an opportunity to ask
questions about the methods and findings of the research and offer feedback and
criticism. Taken further, Lassiter (2005, p. 15) makes a case for collaborative ethnography,
which entails researchers and participants collaborating in all aspects of a study, “from
project conceptualization, to fieldwork, and, especially, through the writing process.”

In each case, the goal is to confirm that the researcher’s findings and impressions are
congruent with those of the people on whom the research was conducted and, if there are
discrepancies, to find out why. However, the idea is not without practical difficulties:

• Respondent validation may lead some research participants to respond defensively

and even demand censorship.

• Bloor (1997, p. 45) observed that research participants who develop relationships of
“fondness and mutual regard” with the researcher may be reluctant to criticize the study.

• It’s doubtful that research participants can validate all of a researcher’s analysis,
since it will be written for social scientists and will therefore include many specialized
references, concepts, theories, and contextual issues unfamiliar to them. Skeggs (1994, p.
86), for example, reported that the “most common response” among her participants was
“Can’t understand a bloody word it says.”

Transferability

Because qualitative research typically entails in-depth, intensive study of a small number
of people, qualitative findings tend to flow out of the context in which the observations are
made. As Lincoln and Guba (1985, p. 316) put it, whether the findings “hold in some other
context, or even in the same context at some other time, is an empirical issue.” Instead of
trying to come up with findings that can definitely be applied to other times, places, and
people, qualitative researchers are encouraged to produce what Geertz (1973) called thick
description: rich, detailed accounts of a group’s culture or people’s experiences. Lincoln
and Guba argued that thick description provides others with the database they need in
order to assess the possible transferability of findings to other milieus.

RESEARCH IN THE NEWS What’s missing in media portrayals of violent crime

In 2017, a Canadian veteran of the war in Afghanistan shot and killed his wife, his mother,
and his daughter, and then killed himself. Many of the media accounts of the tragedy
focused on the perpetrator, in particular on the fact that he had suffered from post-
traumatic stress disorder (PTSD) brought on by his combat experience. A common
narrative in the media portrayals of the event involved the idea that Canadian soldiers
suffering from PTSD and other combat-related ailments typically cannot get the help they
need, despite the significant sacrifices they have made in Canada’s military operations.
Lost from view in these media accounts was the fact that there were victims besides the
person who committed the murder-suicide. The lives and circumstances of the other three
people who died were largely overlooked and forgotten. Another thing that was missing
was an acknowledgment that the murders were part of the larger societal issue of deadly
violence against women (Renzetti, 2017). As Mount Allison University sociology professor
Ardath Whynacht put it, “There were four victims that day and we’re talking only about the
services that could have helped him, and not, for example, services that might have helped
his spouse be safer, in trying to leave that relationship and get space” (Renzetti, 2017).

Dependability

Lincoln and Guba proposed that to establish dependability, researchers should adopt an
“auditing” approach. That is, they should keep complete records of all phases of the
research process—problem formulation, selection of research participants, fieldwork
notes, interview transcripts, data analysis decisions, and so on—and ensure that the
records are accessible. Peers would then act as auditors, possibly during the course of the
research and certainly at the end, to make sure that proper procedures have been followed
and to assess the degree to which the study’s interpretations and theoretical inferences
can be justified. Auditing has not, however, become a popular approach to enhancing the
dependability of qualitative research. A rare example is a study of behaviour at an
American “swap meet,” where second-hand goods are bought and sold (Belk, Sherry, &
Wallendorf, 1988). A team of three researchers collected data over four days through
observation, interviews, photography, and video records. The researchers conducted
several trustworthiness tests, including respondent validation. In addition, they submitted
their draft manuscript and data set to three peers whose task it “was to criticize the project
for lack of sufficient data for drawing its conclusions, if they saw such a void” (1988, p.
456). The same study highlighted some problems with this approach, among them the fact
that, because qualitative research generates extremely large data sets, validation is very
demanding for the auditors. That may be the major reason why auditing has not been
widely used. Confirmability

Objectivity is a difficult if not impossible standard to live up to in social research. For that
reason, the confirmability criterion is designed to ensure that the researcher has acted in
good faith. In other words, it should be apparent that personal values or theoretical
inclinations did not blatantly and unduly sway either the conduct of the research or the
findings derived from it. Lincoln and Guba suggest that establishing confirmability should
be one of the objectives of auditors.

The main goals of qualitative researchers

Seeing through the eyes of the people being studied

Many qualitative researchers try to view the social world through the eyes of the people
they study. To accomplish this, Blumer (1969) advocates striving for a sense of “intimate
familiarity” with research participants’ everyday lives. Achieving a sense of intimate
familiarity means “gaining an in-depth knowledge of the research participants, their setting
or settings, and their situations and actions” (Charmaz, 2008, p. 53). This approach is in
keeping with qualitative researchers’ attempt to achieve verstehen (empathetic
understanding) and their focus on studying the meanings people attribute to the social
world in situ. Qualitative researchers maintain that “(1) … face-to-face interaction is the
fullest condition of participating in the mind of another human being, and (2) … you must
participate in the mind of another human being (in sociological terms, ‘take the role of the
other’) to acquire social knowledge” (Lofland & Lofland, 1995, p. 16). This tendency reveals
itself in frequent references to seeking and privileging an insider perspective. Here are
some examples:
• Armstrong (1993) carried out research on British soccer hooliganism through
participant observation. He described his work as rooted in “Verstehende sociology—
trying to think oneself into the situations of the people one is interested in … in this case
the ‘hooligan.’ This approach involves recognizing social phenomena as due not to any
single or simply identifiable cause and attempting to make sense from the [multi-causal]
social actors’ viewpoint” (Armstrong, 1993, pp. 5–6).

• For their research on teenaged girls and violence, Burman, Batchelor, & Brown
(2001, p. 447) “sought to ground the study in young women’s experiences of violence,
hearing their accounts, and privileging their subjective views.”

The goal of seeing through the eyes of the people under study is often accompanied by the
closely related goal of probing beneath surface appearances. After all, by taking the
position of those under study, researchers may be able to see things in a way that an
outsider with little direct contact would not. This insight is revealed in:

• Hallgrimsdottir, Phillips, & Benoit (2006) study of sex-trade workers, which revealed
that, contrary to media depictions, these workers often see their occupation in a mundane
light (for example, as simply a way to make a living);

• Kleinknecht’s (2011) research on computer hackers, which found that popular

portrayals of hackers as “computer criminals” were often at odds with hackers’ activities
and how they defined themselves: those who take a passionate, creative approach
towards problem-solving; and

• McDonough and Polzer’s (2012) examination of the daily struggles of municipal

employees following the creation of the “mega-city” of Toronto, which showed that rather
than creating a leaner, more efficient workforce, the restructuring led many manual
workers to question their commitment to excellence on the job, while white-collar
employees felt they had to do extra work just to maintain professional standards of
service.

On the other hand, sometimes qualitative research confirms popular notions of a

particular group. For example, Bell (2007) found that media depictions of the western
Canadian separatist movement as decidedly right wing were essentially correct.
The empathetic effort to see the world through the eyes of their participants can present
researchers with practical problems. For example: How far should one go to develop
intimate familiarity and empathy? What if the research participant’s experience includes
illegal or dangerous activities? There is also the risk that the researcher will be able to see
through the eyes of some of the people in a particular social scene but not those of others,
such as people of another gender or culture. These and other practical difficulties will be
addressed in later chapters.

Description and the emphasis on context

Qualitative researchers are much more inclined than their quantitative colleagues to
include descriptive detail when reporting their research. However, they are not exclusively
concerned with description. They also try to explain, and what they endeavour to explain
may pertain to issues of power and exploitation. For example, Skeggs (1997, p. 22) tried to
answer this question: “Why do women who are clearly not just victims of some ideological
conspiracy, consent to a system of class and gender oppression which appears to offer
few rewards and little benefit?”

Many qualitative studies do provide a detailed account of what goes on in the setting under
investigation. On the surface, some of this detail may appear irrelevant; indeed, there is a
risk of becoming too embroiled in descriptive detail. Lofland and Lofland (1995, pp. 164–
165), for example, warned against what they called “descriptive excess” in qualitative
research, whereby the amount of detail overwhelms or inhibits the analysis of the data. But
one of the main reasons for providing descriptive detail is that it permits a contextual
understanding of social behaviour. This implies that we cannot understand the behaviour
of members of a social group without some knowledge of the specific environment in
which they operate. Behaviour that appears odd or irrational may make perfect sense
when understood in the context within which it takes place. The emphasis on context in
qualitative research goes back to the classic studies in social anthropology, many of which
demonstrated how seemingly illogical practices (e.g., a magical ritual that accompanies
the sowing of seeds) make sense when they are understood as part of the society’s belief
system. The provision of descriptive detail is also a manifestation of qualitative
researchers’ desire for naturalism: that is, their belief that researchers should study the
social world as it actually is, rather than relying on contrived settings such as formal
interviews or experiments.

Emphasis on process
Qualitative research tends to view social life in terms of processes. This tendency reveals
itself in a number of ways, but the main one is the qualitative researcher’s concern with
showing how events and patterns unfold over time. Qualitative evidence often conveys a
strong sense of change and flux. As Pettigrew (1997, p. 338) put it, process is “a sequence
of individual and collective events, actions, and activities unfolding over time in context.”
Qualitative research using participant observation is particularly well suited to the study of
process. Ethnographers are typically immersed in a social setting for a long time—in many
cases, for years. Consequently, they are able to observe how events develop over time and
how the different elements of a social system (such as values, beliefs, and behaviour)
interconnect. Tracing interdependent streams of actions and events allows the researcher
to present social life as a process (see Box 9.2).

BOX 9.2 Process in youth shelters

Karabanow (2002) described his experiences in two Canadian shelters for homeless and
runaway youth. As a participant observer, he was able to monitor routine activities there
and describe the shelter culture. In addition, he carried out in-depth interviews with three
levels of shelter workers and used agency archival materials. The dramatic
transformations in the shelters’ external environment and internal operations that took
place over a period of years provided the basis for an analysis of the evolution of their
organizational processes. This example shows the development of a sense of process in at
least two ways. First, observation of the shelters over time made it possible to bring out
developments and interconnections between events. Second, connecting these events
with historical and other data made it possible to show how the shelters were affected by
the larger society. A sense of process can also be developed through semi-structured and
unstructured interviewing in which participants are asked to reflect on the activities
leading up to or following an event. Shaffir and Kleinknecht (2005), for example, used
informal interviews with 45 former Canadian politicians to explore how they coped with
their involuntary exit from political life. The enormity of the loss was often described as
being akin to a social death. A former Liberal parliamentarian described the process of
coming to terms with the defeat as follows: “It’s like the phases of death. You have loss,
anger, sadness, and then you come to accept it” (Shaffir & Kleinknecht, 2005, p. 714). The
life history approach is another form of qualitative research that can be used to show
process. One of the best-known studies of this kind is Lewis’s (1961) classic study of a
poor Mexican family, in which he conducted extended taped interviews with family
members to reconstruct their life histories.

Flexibility and limited structure

Many qualitative researchers are disdainful of approaches that impose a predetermined
set of assumptions on the social world. This position is closely related to the preference for
seeing the world through the eyes of the people under study. After all, a structured method
of data collection must be planned ahead of time on the basis of the researcher’s prior
ruminations and expectations regarding the nature of a social reality that they may never
have encountered before. This structure limits the degree to which the researcher can
genuinely adopt the world view of those being studied, and can lead to serious
misunderstandings. Keeping structure to a minimum is supposed to enhance the
likelihood that the research will reveal the perspectives of the people being observed. This
approach allows aspects of people’s social world that are particularly important to them to
come to light—aspects that might never even cross the mind of a researcher using a more
structured method. For that reason, qualitative research tries to avoid limiting areas of
inquiry, and the research questions it asks tend be fairly general (see step 1 in Figure 9.1).

Ethnography, with its emphasis on participant observation, is particularly well suited to an

unstructured approach. It allows researchers to immerse themselves in a social setting
with a general research idea in mind and then gradually, through observation, narrow it
down to a more specific topic. Another advantage of the unstructured nature of most
qualitative inquiry is that it offers the prospect of flexibility. The researcher can change the
direction of the investigation much more easily than in quantitative research, which tends
to have a built-in momentum once the data collection is underway. (If you send out
hundreds of postal questionnaires and realize only after getting some back that you left out
an important issue, it’s not easy to rectify the situation.) Structured interviewing and
structured observation can offer some flexibility, but it will be limited by the requirement
that interviews be as comparable as possible. Having conducted ethnographic fieldwork in
sites as disparate as a psychiatric institute in Quebec to a day shelter for individuals
experiencing homelessness in Iceland, Dolson (2018) found that the fieldwork experience
was often unpredictable and uncertain. To respond to these situations, he advocates
fostering skills in adversity management: “attempting to o be comfortable with the
unexpected, and trying to make the best of ambiguous situations in which you have no idea
what will happen next” (Dolson, 2018, p. 22). Qualitative researchers, especially
ethnographers who confront some of the most precarious research situations, must
remain open, flexible, and ready to improvise their research plans. As noted previously,
some qualitative researchers use sensitizing concepts as a starting point for their
research. While this might be seen as removing some of the flexibility and unstructured
character of qualitative research, Charmaz (2003) states, “We may use sensitizing
concepts only as points of departure from which to study the data” (p. 259, emphasis in
original). When used in this fashion, concepts offer tentative analytical direction to a
project. Qualitative researchers believe it’s important to remain receptive to new and
alternative understandings as they emerge from the data. See Box 9.3 for an illustration of
how a conceptual angle can help give direction to a study, while the unstructured data
collection style of qualitative research can suggest alternative avenues of inquiry or ways
of thinking.

BOX 9.3 Emerging concepts

In an update to Hochschild’s (1983) work (discussed in Box 9.1), Canadian researchers

Santin and Kelly (2017) used participant observation and informal interviews with flight
attendants to revisit the concept of emotional labour following the 9/11 terrorist attacks.
Consistent with Hochschild’s findings, Santin and Kelly found that flight attendants
continued to engage in “surface acting:” they would conceal their true emotions in order to
please their employer or customers. However, they also discovered that changes to airline
security protocols post-9/11 provided cabin crew with more discretionary power and
emotional agency in responding to unruly passengers. The “9/11 effect” led to a shift in
how emotional labour was carried out in the new pro-security context. It meant that cabin
crew were not only providing a service to passengers, but also protecting them. Their new
role as guardian partially shielded them from workplace repercussions that traditionally
followed from not treating passengers in a passive, courteous manner. In line with
Hochschild’s findings, institutional norms continued to structure and constrain flight
attendants’ behaviour, but now new safety protocols offered them more control over their
emotional labour. By exploring flight attendants’ emotional labour post-9/11, Santin and
Kelly expanded on Hochschild’s ideas, adding three new concepts to our social science
lexicon: the 9/11 effect, emotional agency, and role shields. Their study demonstrates how
existing theoretical ideas can be used as sensitizing concepts in subsequent research and
then built upon to develop new ways of understanding changing social contexts.

Ultimate goals

The goals of qualitative research mentioned above—seeing through the eyes of others,
bringing out a sense of process, and having a flexible and unstructured method of inquiry—
are essentially immediate or proximate goals that are used as means of achieving a deep
understanding of the people or groups being studied. But qualitative researchers’ goals
often go beyond simply understanding. They may also wish to do research in order to
pursue social justice and bring about social change, as mentioned in our discussion of
critical approaches to science in Chapter 1. Box 9.4 illustrates how people doing
participatory action research can use qualitative methods to address a serious social
problem. Indeed, qualitative studies have shed light on a wide range of societal injustices
in Canada, including racism (Branker, 2017), homophobia (Abramovich, 2016), and poverty
(Coloma & Pino, 2016).

BOX 9.4 The use of qualitative methods in participatory action research

Type 2 diabetes among Indigenous children and youth is a growing problem, one with
important implications for the quality of life among Indigenous people in Canada. Sharma,
Reimer-Kirkham, & Meyerhoff (2011) sought to do something about it, but rather than
approach the issue from the perspective of conventional social science, they used the
methods of participatory action research (PAR) to address it. Instead of devising the
methodology themselves and then seeking consent from prospective research
participants, they first consulted members of the community where the research would
take place to get their permission to do a study on the topic, and then asked for their views
on what an appropriate methodology might be. After gaining permission to go ahead with
the project from the band and chief of a rural Indigenous community in British Columbia,
Sharma et al. held focus groups with several community groups, including Elders with
diabetes, the Chief and Council, a health-care services team, and some Indigenous youths
to determine how the study would be conducted. This was done in an attempt to neutralize
the power differential between the researchers (who were non-Indigenous and had PhDs,
secure middle-class incomes, and Western cultural outlooks) and the members of the
community. It was especially important to recognize the potential for exploitation in this
research, given the history of colonialism in Canada and the tendency of past researchers
to marginalize Indigenous experiences and ways of knowing. The goal was to make the
members of the community equal partners in the research, and in so doing allow them to
bring their own issues and perspectives to bear on it.

The various participants decided collectively that the project would take the form of a short
film to be created by five Indigenous youths aged 14–16. The film featured interviews with
Elders suffering from diabetes, a description of what diabetes is, and some tips on healthy
eating. The production of the film involved a constant negotiation of the relationship
between the professional researchers and the youths. The former tried to stay in the
background as much as possible, although they did end up editing the first version of the
film. In order to enhance the collaborative nature of the project, the professional
researchers then showed the edited version to the youths and removed any parts that the
young people thought were inappropriate. This illustrates several aspects of PAR: explicit
recognition of the power differentials that often characterize conventional research; an
attempt to minimize those differentials in order to address the subject matter from the
perspective of the people participating in the research; and full participation by the group
that is the focus of the inquiry in order to produce knowledge and actions that are
beneficial to them and not just to the professional researchers.

Some contrasts between quantitative and qualitative research

Several writers have used tables to contrast quantitative and qualitative research (for
example, Hammersley, 1992). Table 9.1 highlights the main differences. As with any
summary of a large body of material, the table outlines general tendencies. The full picture
is more nuanced: there are exceptions to these general points, as we will see in Chapter
14.

• “Numbers” versus “Words.” Quantitative researchers set great store by precise,

numerical measurements of social phenomena. They use those measurements to
calculate statistics to understand social life, whereas qualitative researchers use mainly
words in their analyses of society.

• “Point of view of researcher” versus “Points of view of participants.” In quantitative

research, investigators are in the driver’s seat; their concerns structure the investigation. In
qualitative research, the investigation is structured by the people being investigated: their
perspective—what they see as important—provides the point of orientation.

• “Researcher is distant” versus “Researcher is close.” In quantitative research,

researchers tend to be less involved with their subjects and in some cases, as in research
based on online or mailed questionnaires, they may have no direct contact at all.
Quantitative researchers often consider this distance desirable, because they want to be
as objective as possible and feel their objectivity would be compromised if they were to
become too involved with the people they study. By contrast, qualitative researchers seek
involvement with the people they study in order to see the world through the eyes of their
participants and promote a more equalitarian relationship.

• “Theory and concepts tested in research” versus “Theory and concepts developed
from data.” Before data collection begins, quantitative researchers typically have a theory
in mind that they want to test. In qualitative research, concepts and theories develop as
the data are collected.

• “Structured” versus “Unstructured.” Quantitative research is typically highly

structured to maximize validity and reliability; in qualitative research, the approach is less
structured in order to allow the researcher to get a sense of the meanings people derive
from their everyday lives, and to permit the development and elaboration of concepts and
theories as the data are collected.
• “Generalizable knowledge” versus “Contextual understanding.” Whereas
quantitative researchers want their findings to be applicable to some larger population,
qualitative researchers focus on understanding the behaviour, values, and beliefs of the
people in their study. Qualitative findings usually develop out of the context these people
find themselves in; hence the insights gained may or may not apply to other people.

• “Hard, reliable data” versus “Rich, deep data.” Quantitative researchers often
describe their data as “hard”—robust and unambiguous—because they strive for precision
when devising their measurement techniques. Qualitative researchers, by contrast, claim
that their contextual approach and often prolonged involvement in a setting mean that the
information they collect is rich, nuanced, and very detailed.

• “Macro” versus “Micro.” Quantitative researchers often seek to uncover large-scale

social trends and connections between variables; as we noted above, their research is
generally designed to be applicable to large populations. Qualitative researchers are
usually concerned with in-depth analysis of small-scale aspects of social reality, such as
personal interaction, although these may be linked to larger social and political issues.

• “Behaviour” versus “Meaning.” It is sometimes suggested that quantitative

researchers focus more on people’s behaviour, and qualitative researchers more on the
meaning of that behaviour for the actor.

• “Artificial settings” versus “Natural settings.” Whereas quantitative research is

often conducted in a contrived context, such as an experiment or a formal interview
setting, qualitative researchers investigate people in their natural environments, such as
social gatherings or workplaces.

TABLE 9.1 Common contrasts between quantitative and qualitative research

Quantitative Qualitative

Numbers Words

Point of view of researcher Points of view of research participants

Researcher distant Researcher close

Theory testing Theory development

Structured Unstructured

Generalizable knowledge Contextual understanding

Hard, reliable data Rich, deep data

Macro Micro

Behaviour Meaning

Artificial settings Natural settings

Qualitative interviewing is quite different from interviewing in quantitative research:

• In quantitative research, the approach is highly structured to maximize reliability

and validity in measuring key concepts. The researcher usually has a clearly specified set
of research questions that the interview is designed to answer. Qualitative interviewing, on
the other hand, is much less structured and tends to be more open-ended. It also allows
greater freedom to modify and add research ideas once the investigation has begun.

• There is greater interest in the interviewee’s perspectives and concerns in

qualitative interviewing; in quantitative research, the interview is driven mainly by the
research agenda of the person conducting the study.

• In qualitative interviewing, going off on a tangent is often encouraged. The

interviewer may vary both the order and the wording of questions, and may even ask
unplanned questions to follow up on interviewees’ replies; the additional replies may
provide new insight into what the interviewee sees as relevant and important. In
quantitative research, departing from the interview schedule is usually discouraged
because it diverges from the specific research plan and may affect reliability and validity.

• Qualitative interviewing tends to be flexible, responding to the direction in which

interviewees take the interview and perhaps adjusting the emphases in the research as a
result of the issues that emerge during the interview. For instance, in describing her
approach to conducting qualitative interviews with social movement actors in Ontario,
Maiolino (2019) writes, “A set of interview questions was used to guide the interview, but I
also tried to be flexible and create space for interviewees to direct the discussion as it
developed” (p. 170). By contrast, structured interviews are typically inflexible, because of
the need to standardize the interactions with all interviewees.

• In qualitative interviewing, the researcher wants rich, detailed answers; in

structured interviewing the interview is normally designed to generate specific answers
that can be coded and processed quickly.

• In qualitative interviewing, it is not uncommon for the interviewee to be interviewed

more than once. For example, in his research on “clubbers” (people who frequent night
clubs), Malbon (1999) conducted an initial interview to put the participant’s mind at ease
about the research, help plan his approach to observation, and begin collecting
preliminary data. In a follow-up interview, he discussed club visits, the experience of
clubbing, and occasionally, wider aspects of participants’ lives. In structured interviewing,
unless the research is longitudinal in character, the person is usually interviewed only
once.

Unstructured and semi-structured interviewing

As we have noted, there are two main types of qualitative interview:

• In an unstructured interview the researcher uses at most a memory aid: a small set
of self-prompts to investigate certain topics. The interviewer may ask just a single
question; then the interviewee is allowed to respond freely, with the interviewer pursuing
points that seem worthy of follow-up. Unstructured interviewing tends to be similar in
character to a conversation (see Box 11.1 for an example).

• In a semi-structured interview the researcher has a list of questions or fairly specific

topics to be covered (an interview guide), but the interviewee still has a great deal of
leeway in deciding how to reply (see Box 11.2 for an illustration). Questions may not follow
the exact order in the guide, and some questions not included on the list may be asked in
response to what the interviewee says. Sometimes lines of thought identified by earlier
interviewees are taken up and presented to later ones. Nonetheless, all the questions on
the list are usually asked and similar wording is used from interviewee to interviewee. Box
11.3 illustrates how these features of qualitative interviewing are used in life history
research.

BOX 11.1

Unstructured interviewing

Torelli (2019) adopted a largely unstructured interview approach when conducting field
research on the “dirty work” of Canadian homeless shelter caseworkers. His introspective
reflections offer insights into how qualitative interviewers must listen attentively, think on
their feet, and carefully follow lines of inquiry presented to them by participants. Torelli
(2019) described his strategy for interviewing caseworkers as follows: I perceived the
interview as a social event, a process whereby, through our interactions, me as a
researcher and the researched were collaborating jointly in knowledge production…. This
interchange produced unique conversational avenues which could only be achieved
naturalistically rather than with strict, albeit aloof, adherence to a structured interview
guide. I therefore saw the interview process contingently: unpredictable, spontaneous
conversational instances emerging out of our interaction that would have not occurred had
I stuck necessarily to my interview guide and structured the interview in a determinative
way. (Torelli, 2019, p. 76)

BOX 11.2 Semi-structured interviewing

Wright, Alaggia, & Krygsman (2014) used qualitative interviewing to assess the success of
an after-school arts program that was offered to Canadian youth living in low-income
communities. The study indicated that the program, which involved having the youth
participate in theatre productions, had a number of long-term benefits for the participants
that went beyond developing their theatre skills. Here are some sample interview
questions, along with sample probe questions that were used to get the respondents to
elaborate on their initial replies:

Interview questions Probe questions

Did you enjoy the art program? What activities did you like?

Did you like the instructors?

Do you have many friends/get along with others your age? Do you sometimes get into
arguments with others your age?

What do you do when you get into arguments?

Have you been getting along with your parents? Do you sometimes get into arguments?
About what?

BOX 11.3

Life history interviews

One special form of interview associated with qualitative research is the life history
interview. It is often combined with personal documents such as diaries, photographs, and
letters. Participants are invited to look back in detail across their entire life course and to
report their experiences and how they understood their world. Valuable insights may be
gained from a life history regardless of whether other people have had the same
experiences as the person described in the work. It also has the advantage of illustrating
process: how events unfold and interrelate in people’s lives over long periods of time. An
example of the life history interview approach is provided by Lewis (1961) in his research
on the Sánchez family and their experiences in a Mexican slum: I asked hundreds of
questions of [the five members of the Sánchez family]…. While I used a directive approach
to the interviews, I encouraged free association, and I was a good listener. I attempted to
cover systematically a wide range of subjects: their earliest memories, their dreams, their
hopes, fears, joys, and sufferings; their jobs; their relationship with friends, relatives,
employers; their sex life; their concepts of justice, religion, and politics; their knowledge of
geography and history; in short, their total view of the world. (Lewis, 1961, p. xxi) Miller
(2000) points out that interest in the life history method is reflective of the popularity of
“narrative interviews” (discussed in Chapter 13) in which the person studied, instead of
responding to an interviewer’s questions, “tells a story.” In the past this approach was
associated with the study of a single life, or perhaps two, but today it is increasingly
applied to several lives in a single project. M. Atkinson (2002; 2004), for example,
conducted narrative interviews with women tattoo enthusiasts. P. Atkinson (2004)
observed that the duration of life story interviews varies considerably from study to study,
but that it usually takes two or three sessions lasting 60 to 90 minutes each. He provided a
catalogue of questions that can be asked, and divided them into the following groups
(1990, pp. 43–53):

• birth and family of origin (e.g., “How would you describe your parents?”)

• cultural traditions (e.g., “Was your family different from others in town?”)

• social factors (e.g., “What were some of your struggles as a child?”)

• education (e.g., “What are your best memories of school?”)

• love and work (e.g., “How did you end up in the work you do or did?”)

• inner and spiritual life (e.g., “What are the stresses of being an adult?”)

• major life themes (e.g., “What were the crucial decisions in your life?”)

• vision of the future (e.g., “Is your life fulfilled yet?”)

• closure questions (e.g., “Have you given a fair picture of yourself?”)

A variant of the life history interview is the oral history interview, a technique often used by
historians. It is usually somewhat more specific in tone than the interviews described
above in that the participants may be asked to reflect on certain historical events or eras
they have lived through. The emphasis is often on how the individual’s life was affected by
those things. The information gathered is sometimes combined with other sources of data,
such as documents. The chief problem with the oral history interview (a problem it shares
with life history interviews in general) is the possibility of bias caused by memory lapses
and distortions. Sugiman (2004) pointed out in her work on Japanese-Canadian women
interned during the Second World War that remembering is a social and even political act,
involving recalling, forgetting, transforming, and shaping of recollections. On the other
hand, oral history testimonies can give a voice to groups that are typically marginalized in
historical research (this is true of life history interviews in general), either because of their
lack of power or because they are regarded as unimportant. In semi- and unstructured
interviews, the process is designed to bring out how the interviewees themselves interpret
and make sense of issues and events. Bell (2007), for example, wanted to understand what
western Canadian separatists thought of Canadian federalism and the federal party
system. Tastsoglou and Miedema (2003) sought to understand the meaning of community
from the perspective of immigrant women, and Smith (2008) examined the meanings of
pain for professional wrestlers. Projects like these, unlike quantitative studies, are usually
not designed to test hypotheses or theories. In short, qualitative research is not just
quantitative research without the numbers. There is a growing tendency to refer to semi-
structured and unstructured interviews collectively as in-depth interviews or qualitative
interviews, although the semi- and unstructured types can produce very different results.
The choice of one type rather than the other is affected by a variety of factors:

• Researchers who feel that using even the most rudimentary interview guide hinders
genuine access to the world views of members of a social setting are likely to favour an
unstructured approach. But even here the interview is rarely completely unstructured: the
researcher usually has at least a general topic that is to be discussed.

• If the researcher is beginning the investigation with a fairly clear rather than a
general focus, it is likely that the interviews will be semi-structured, so that they can
address more specific issues.

• If more than one person is to carry out the fieldwork, semi-structured interviewing
may be preferred, to ensure that interviewing styles are comparable.

• Multiple-case study research generally needs some structure to ensure cross-case

comparability.

Preparing an interview guide

An interview guide is much shorter and less detailed than a structured interview schedule.
In fact, it is often simply a brief list of memory prompts for areas to be covered in
unstructured interviewing and a somewhat more elaborate list of issues to be addressed or
questions to be asked in semi-structured interviewing. What is crucial is that the actual
questioning is flexible, allowing interviewers to pursue leads offered by research
participants as they begin to open up and reveal their view of the social world.
Ricciardelli’s (2019) approach to interviewing Canadian correctional officers about the
realities of prison work is illustrative of the flexibility qualitative researchers attempt to
build into their interviews and interview guides: A semi-structured interview guide—a list of
open-ended questions I constructed—was available during interviews, but was often only
used as a sort of checklist to ensure that conversational paths included select topics….
The conversation tended to be determined by the interviewee because my objective was to
hear what was on each participant’s mind, what stood out in his or her experiences, and
how he or she interpreted the correctional officer role. Although I rarely directed
conversation, I did at points ask interviewees for clarification or for more detail.
(Ricciardelli, 2019, pp. 159–160) In preparing for qualitative interviews, Lofland and Lofland
(1995, p. 78) suggested that researchers ask themselves “Just what about this thing is
puzzling me?” This question can be applied to each of the research questions generated; it
can also serve as a mechanism for generating research questions. Lofland and land
suggested that puzzlement may be stimulated in various ways: by recording random
thoughts in different contexts (writing them down as quickly as possible); through
discussions with colleagues, friends, and relatives; and, of course, by reading the existing
literature on the topic. The formulation of the research question(s) should not be so
specific as to close off the alternative avenues of inquiry that may arise during fieldwork.
Premature closure of the research focus would be inconsistent with qualitative research’s
prime purpose: to explore the world view of the people being studied, rather than to test
the researcher’s own ideas about that world.

The interview guide should:

• establish a certain amount of order, so that questions flow reasonably well, but still
allow for changes in the order of the questions and the impromptu asking of different
questions;

• include questions or topics that address the research questions (without being too
specific);

• use language that is comprehensible and familiar to those being studied;

• not ask leading questions—that is, questions that imply a “correct” or socially
acceptable answer (for example, “Have you ever done something really stupid, like taking
crack?”); and

• include prompts to remind the researcher to record basic information about the
participant (name, age, gender, etc.) as well as more specific information that is relevant to
the research questions (position in company, number of years employed, number of years
involved in a group, etc.); such information is useful for putting people’s comments into
context.

There are also some practical details to attend to before the interview begins:

• Become familiar with the research setting, which in many cases will be the everyday
surroundings of the interviewees. This will help to put their comments in context.

• Get a reliable recording device. Qualitative researchers (with the possible exception
of ethnographers and people doing participant observation research) nearly always record
and then transcribe their interviews. This is important for the detailed analysis required in
qualitative research, and to ensure that the interviewees’ answers are captured in their
own terms. Simply taking notes makes it too easy to lose specific words and phrases.

• Make sure as far as possible that the interview takes place in a setting that is quiet
(so that the quality of the recording will be as good as possible) and private (so that the
interviewee won’t worry about being overheard).

• Prepare for the interview by cultivating the traits of a quality interviewer suggested
by Kvale (1996) (see Box 11.4).

BOX 11.4

Kvale’s ten traits of an effective interviewer (plus three others)

Kvale (1996) proposed that a successful interviewer has ten characteristics:

• Knowledgeable: is thoroughly familiar with the topic of the interview

• Structuring: tells the interviewee the purpose of the interview; asks if the
interviewee has questions

• Clear: asks simple, easy, and short questions; no jargon

• Gentle: lets people finish; gives them time to think; tolerates pauses

• Sensitive: listens attentively both to what is said and to how it is said; is empathetic

• Open: responds to what the interviewee considers important; is flexible

• Steering: knows what needs to be found out

• Critical: is prepared to challenge what is said when, for example, there is an
inconsistency in the interviewee’s replies

• Remembering: relates what is said to what has previously been said

• Interpreting: clarifies and extends the meaning of interviewees’ statements but

without imposing meaning on them

To Kvale’s list of traits, please add the following three:

• Balanced: does not talk too much, which can make interviewees passive, but does
not talk too little, which can cause interviewees to feel that their responses are not
satisfactory

• Ethically sensitive: for example, assures the interviewee that all answers will be
treated confidentially

• Non-judgmental: does not communicate (even subtly) a moral judgment about

what an interviewee has said or done; negative judgments may cause people to withhold
information that could be useful to the study, while positive judgments may encourage
them to focus on pleasing the interviewer rather than providing authentic commentary

After the interview, make notes about:

• how the interview went (obstacles, ethical issues, surprises, the interviewee’s
demeanour, what worked well and what can be improved upon);

• how you felt during the interview and how this may have impacted the process
(relaxed, nervous, in control, sad, cheerful);

• where the interview took place and what the setting was like (noisy/quiet, many/few
other people in the vicinity); and

• tentative insights gleaned about your research question and potential new lines of
inquiry.

Focus groups: An introduction

Most people think of an interview as an exchange between one interviewer and one
interviewee. The focus group technique, however, involves speaking with more than one
(usually at least four) interviewees at the same time. Essentially it is a group interview in
which the interviewees can speak to and interact with one another.
• Most focus group researchers work within a qualitative research tradition. They try
to provide a fairly unstructured setting in which the person who runs the focus group,
usually called the moderator or facilitator, guides the session but does not intrude. A focus
group offers several advantages. One of the most important is that it allows the researcher
to develop an understanding of why people feel the way they do. In individual interviews,
participants are often asked about their reasons for holding a particular view, but a focus
group allows participants to probe one another’s reasons. This can be more informative
and revealing than the question-followed-by-answer method of ordinary interviews. For
example, an individual who has already answered a certain question may decide, after
hearing others’ answers, to qualify or modify that response in some way. Some people may
want to voice agreement with a view they might not otherwise have thought of. These
possibilities permit focus groups to elicit a wide variety of perspectives on an issue.

• In conventional one-to-one interviewing, interviewees are rarely challenged; they

may say things that are inconsistent with earlier replies or that patently cannot be true, but
interviewers are often reluctant to point out such deficiencies. In a focus group,
participants may argue and challenge one another’s views. Arguing often produces more
realistic accounts of what people think, because it forces them to defend and possibly
revise their views.

• A focus group offers an opportunity to study how individuals collectively make

sense of a phenomenon and construct meanings of it. It is a central tenet of theoretical
positions such as symbolic interactionism that meanings and understandings are not
derived by individuals in isolation: rather, they develop out of interactions and discussions
with others. In this sense, focus groups reflect the processes through which meaning is
constructed in everyday life, which allows them to be more naturalistic than individual
interviews (Wilkinson, 1998). On the other hand, it is not clear whether these new
meanings persist beyond the focus group session.

Conducting focus groups

A number of practical aspects of conducting a focus group are considered below.

How many groups?

How many focus groups are needed? Table 11.1 provides some data on this question, and
on some other aspects of focus groups (cf. Deacon, Pickering, Golding, & Murdock, 1999).
As it suggests, there is a good deal of variation in the number of groups required for a
particular study, but it generally ranges from 10 to 15. A single group is unlikely to be
sufficient, since the responses may not be typical of other groups. Nonetheless, there are
good reasons for limiting the number of groups (besides saving time and resources). Once
the moderator is able to anticipate fairly accurately what the next group is going to say,
enough groups have participated. This notion is similar to theoretical saturation in
grounded theory, discussed in Chapters 10 and 13. One factor that can affect the number
of groups required is whether the researcher thinks that the range of views is likely to be
affected by socio-demographic factors such as age, gender, or social class. Many focus
group researchers like to ensure that different demographic groups are included, which
usually requires a large number of groups. In connection with Kitzinger’s (1994) research
on audience responses to media messages about AIDS (see Table 11.1), she wrote that a
large number of groups is preferred in order to capture as many different perspectives as
possible. On the other hand, conducting more groups increases the complexity of the
analysis. For example, Schlesinger, Dobash, Dobash, & Weaver (1992, p. 29; see Table
11.1 and Box 11.7) reported that their 14 tape-recorded hour-long sessions produced more
than 1400 pages of transcription for analysis.

Limitations of focus groups

Focus groups are particularly useful in illustrating how meaning is jointly constructed.
What, then, are their chief limitations?

• The researcher probably has less control over the proceedings than in individual
interviews. As we have seen, not all writers on focus groups perceive this as a problem;
indeed, researchers concerned with reflexivity often see it as an advantage. However, the
question of control raises questions about the extent to which a researcher should allow
the participants to “take over” the discussion. There is clearly a delicate balance to be
struck in deciding how involved moderators should be and how much a set of prompts or
questions should be allowed to influence the conversation. What is not clear is the degree
to which it is appropriate to surrender control, especially when there is a set of research
questions to be answered.

• An unwieldy amount of data is sometimes produced. For example, Bloor,

Frankland, Thomas, & Robson (2001) suggested that one focus group session can take up
to eight hours to transcribe—somewhat longer than an equivalent personal interview,
because of variations in voice pitch and the need to identify who says what. Also, group
recordings often include inaudible comments from participants sitting too far from the
microphone, and these also affect transcription.
• The data may be difficult to analyze. Developing an analysis strategy that
incorporates both the themes of the discussion and the patterns of interaction is not easy.

• Focus groups are difficult to arrange. It may be hard to get people to agree to
participate, and harder still to get them to show up. Very small payments, such as gift
cards, are sometimes offered to induce participation, but it is still common to find that
some people will not turn up.

• Group effects may be a problem. This category includes the obvious challenge of
dealing with participants who either are too reticent to speak up or hog the stage. Krueger
(1998) suggested that the people running the focus group should make it clear to
participants that other people’s views are definitely required; for example, he suggested
saying something like “That’s one point of view. Does anyone have another point of view?”
(1998, p. 59). As for those who are reluctant to speak, Krueger recommended that they be
actively encouraged to say something. It would be interesting to know whether agreement
among focus group participants is more common than disagreement; if that is the case,
one reason might be group pressure to conform. It’s also possible that participants are
more likely to express socially acceptable views in a group setting than in individual
interviews. Morgan (2002) cited a study in which group interviews with boys on the topic of
relationships with girls were compared with individual interviews with boys on the same
subject. Alone, the boys expressed a degree of sensitivity that was not evident in the group
context, where they tended to express more macho views. This suggests that in the group
setting the boys were trying to impress each other or avoid embarrassment, and were
influenced by peer pressure. But this does not mean that such data are tainted: in fact, it
may be the gulf between privately and publicly held views that is of interest.

•Madriz (2000) proposed that there are circumstances when individual interviews are
more appropriate. One is when participants are likely to disagree profoundly with each
other; another is when participants are not comfortable in one another’s presence (for
example, people in hierarchical relationships). Finally, because of the potential for focus
groups to cause discomfort among participants (for example, when intimate details of
private lives need to be revealed), an individual interview or even a questionnaire may be
better.

Introduction

There can be similarities between quantitative and qualitative approaches to social

research, and even where there are differences the two orientations can sometimes be
combined or at least used in a complementary fashion. Content analysis in some ways is a
blend of the two. It examines forms of communication to see what they reveal about a
society, a culture, or even the relationships between individuals. Meanings and
interpretations are important aspects of content analysis, but so is formal categorization
of the communications, which often includes some quantification. For example, suppose
you are interested in how newspapers cover crime. You come up with the following
questions:

• Do certain newspapers report more crimes than others?

• How much crime is reported? Where in the paper do crime stories appear—on the
front page, or inside?

• Do columnists as well as reporters write about crime?

• Are some crimes given more attention than others?

• Do more crime stories appear during the week or on weekends?

• What sorts of crime predominate in newspaper articles (crimes against the person,
property crime, crimes in which society is the victim)?

Most content analysis of the media is likely to entail several research questions, generally
revolving around the same five W’s that are the basis of any news report: who (does the
reporting); what (gets reported); where (does the issue get reported); why (does the issue
get reported); and when (does it get reported). But researchers are also interested in what
media coverage omits. For example, interviews with the family of the accused are rare;
such inattention is itself notable, revealing what is and is not important to writers and
publishers. Another common theme in content analysis is change in the coverage of an
issue over time. For example, Buchanan (2014) found that in the period 1894–2005 there
was a significant decrease in the amount of local news featured in the Toronto Star and the
Ottawa Citizen, and that both papers had increased their national news coverage over this
period.

Content analyses of this kind yield a predominantly quantitative description of the

characteristics of a communication. Those who perform content analysis in this way claim
to be objective and systematic, clearly specifying the rules for the categorization of the
material in advance. Researchers try to create transparency in the coding procedures so
that personal biases intrude as little as possible. The coding rules in question may, of
course, reflect the researcher’s interests and concerns and therefore be a product of
subjective bias to some degree, but once the rules have been formulated, it should be
possible to apply them without bias. Content analysis can be performed on unstructured
information, such as transcripts of semi- and unstructured interviews, and even qualitative
case studies of organizations (for example, Hodson, 1996). Until recently it was used
mainly to examine printed texts and documents, but other sorts of materials may also be
analyzed. If you were interested in gender roles or representations, for example, you might
look at:

• differences in the degree to which men and women are sexualized on the covers of
popular magazines (Hatton & Trautner, 2013);

• pictures on social media (e.g., how parents portray their children’s gender on
Instagram) (Choi & Lewallen, 2018);

• profiles on dating sites and apps (Wada, Hurd Clarke, & Mortenson, 2019);

• animated cartoons (Thompson & Zerbinos, 1995); and

• lyrics of popular songs (e.g., to look for changes in the representation of women)
(Marcic, 2002).

Here we will call all such materials “documents,” and define them as any data source that:
•can be “read” (including visual materials such as photographs); and

•was not produced specifically for the purpose of social research.

Documents are important because they are unobtrusive and non-reactive. The fact that
the subjects don’t know they are being studied removes a common threat to the validity of
the data. In discussing the different kinds of documents used in social science, Scott
(1990) distinguished between personal and official documents and, within the latter group,
between private and state documents. These distinctions are used in much of the
discussion that follows. Scott also enumerated four criteria for assessing the quality of
documents (p. 6):

•Authenticity. Is the evidence genuine and of unquestionable origin?

•Credibility. Is the evidence free from error and distortion?

•Representativeness. Is the evidence typical of what it is supposed to represent (for

example, social life at a particular time and place)? If not, is the extent of its uniqueness
known?

•Meaning. Is the evidence clear and comprehensible?

Note that this use of “authenticity” is different from how the term was used in Chapter 9 to
assess qualitative research.

What things need to be analyzed?

Obviously, what is to be analyzed depends on the specific research questions. In mainly

quantitative studies, these elements are usually specified in advance in order to guide both
the selection of the media to be assessed and the construction of the coding schedule.
They usually include words, subjects and themes, and value positions, as discussed next.

Words; Determining the frequency with which certain words are used is often the first step
in content analysis. Jagger (1998), for example, searched dating advertisements and
counted words such as “slim” and “non-smoker” to compare women and men with
respect to what each deemed desirable in a date (see Box 12.5). Simple counting of
particular words can reveal emphasis, style of writing or presentation, and even the
overplaying of certain events. For example, Dunning, Murphy, & Williams (1988) noted a
tendency for the British press to sensationalize disturbances at soccer matches by using
emotive words such as “hooligans” and “louts” to refer to audience members, and terms
such as “battle” to refer to a game; less dramatic terms such as “fans” or “contest” would
have encouraged more neutral responses among readers.

Subjects and themes; A more interpretative approach is required to code text in terms of
subjects and themes. At this point, the analyst is searching not just for the obvious or
manifest content, but also for some of the underlying or latent content as well. This occurs
if the researcher wants to probe beneath the surface to ask deeper questions about what is
happening. In reports on crime, for example, is the victim blamed along with the accused?
Is the occupation of the accused or the victim stated? If not, is it implied in an address or a
picture? Why is it that news stories about men who have been mugged sometimes mention
their marital status? In seeking to answer these sorts of questions, the quantitative
aspects of content analysis can serve as a starting point for a study in which qualitative
research becomes the central focus of the endeavour.

Value positions; A further level of interpretation is likely when the researcher is seeking to
demonstrate that the writer of a particular text has taken a certain value position. For
example, is a journalist who writes about crime sympathetic or hostile to the accused? Is
all the blame put on the accused, with the implication that punishment is appropriate? Or
is the focus on social conditions, with the assumption that less blame should be placed on
the criminal? If there is no manifest indication of the writer’s value positions, can
inferences be made from the latent content? Another way in which content analysis can
reveal value positions is through the coding of ideologies, beliefs, and principles. For
instance, Sturgeon and Morrissette (2010) examined suicide ideation among Manitoba
farmers who called a rural crisis line. The data were gathered from forms that were filled
out by the crisis line counsellors (which are similar in design to the coding manual
depicted in Figure 12.2). One of the themes that emerged as codes were developed was
“family salvation,” which involved the belief that life is worth living if one has meaningful
relationships with family members. “I just keep going because I think someday my son will
need me,” said one farmer (p. 199). Surprisingly, only a small proportion of farmers
mentioned their spouses or life partners as the people who made their lives bearable;
others such as siblings, children, and extended family members were more likely to be
cited.

Content analysis without a pre-existing coding scheme

Content analysis may be undertaken before any specific decisions have been made
regarding how the information available for analysis will be coded. In such cases the data
are searched in order to find underlying themes, making the study primarily qualitative in
nature. Like other qualitative approaches, qualitative content analysis tends to be
inductive, with themes emerging through an iterative process as the researcher interprets
the data. Kennedy (2006), for example, in his study of the Canadian fathers’ rights
movement called “Fathers 4 Justice,” used content analysis in conjunction with
participant observation and semi-structured interviews to examine how the movement
developed a collective identity. The content analysis was conducted on documents such
as position papers submitted to government agencies, newsletters, pamphlets, and
minutes from meetings, and was used mainly to make sense of the data gathered through
participant observation and interviews. Similarly, Bell (2007) used pamphlets, speeches,
newsletters, and other texts to arrive at an overall assessment of western Canadian
separatism as a neo-liberal movement. The processes by which the themes are extracted
in this kind of content analysis are often left implicit, although they are usually illustrated
with quotations from the text in question.

Lynch and Bogen (1997) examined core sociological textbooks and found that they
contained recurring themes that presented an upbeat and scientific view of the
discipline—one that, in the researchers’ view, was biased and value laden. Seale (2002)
examined newspaper reports about people with cancer. One of the phases of his analysis
entailed an “NVivo coding exercise, in which sections of text concerning themes of interest
were identified and retrieved” (p. 109). He was especially interested in gender differences
in the representation of sufferers, and demonstrated that stories about men were much
more likely than stories about women to discuss the role that the individual’s character
played in dealing with the disease.

Altheide and Schneider (2013) outlined an approach they called ethnographic content
analysis (ECA). They described this approach as differing from quantitative content
analysis in that the researcher is constantly revising the themes or categories distilled from
the examination of documents. As they put it:

ECA follows a recursive and reflexive movement between concept development-sampling-

data collection-data coding-data analysis-interpretation. The aim is to be systematic and
analytic but not rigid. Categories and variables initially guide the study, but others are
allowed and expected to emerge during the study, including an orientation toward
constant discovery and constant comparison of relevant situations, settings, styles,
images, meanings, and nuances (p. 26; emphases in original). While more quantitatively
oriented analysis typically entails applying predefined categories to the sources, ECA
allows for greater refinement of those categories as well as generation of new ones. For
example, Canadian researchers Schneider and Hannem (2019) examined how the sexual
misconduct allegations against Donald Trump in the 2016 US election were framed as a
political issue, which had the effect of minimizing victims’ experiences. Instead of
beginning with a pre-existing coding scheme, they conducted a preliminary analysis of a
few potential documents to be used in their study (articles from the Washington Post and
New York Times). This early reading of possible data informed their initial categories and
codes to be used as part of a data collection protocol. The subsequent analysis then
followed an iterative process to construct qualitative themes in the data. In addition, ECA
emphasizes the context within which documents are generated. For instance, a study of
newspaper reports of crime would require some understanding of news organizations and
the work that journalists do (Altheide, 2004).

Semiotics

Another form of predominantly qualitative content analysis is semiotics, the science of

signs. In social research, semiotics involves analysis of the signs and symbols
encountered in everyday life. It can be applied to documentary sources as well as many
other kinds of data. The main terms employed in semiotics are as follows:
• The sign is something that stands for something else, such as a yellow traffic light; it
has two components, a signifier and a signified.

• The signifier is the thing (here the yellow light) that points to an underlying meaning.

• The signified is the meaning to which the signifier points (“caution: stop if
possible”).

• The denotative meaning is the manifest or obvious meaning of a signifier and as

such indicates its function (here to regulate traffic).

• A connotative meaning is one that can arise in conjunction with the denotative
meaning (for example, “speed up to beat the coming red light”).

• Polysemy refers to the notion that signs can be interpreted in many different ways.

Semiotics seeks to uncover the hidden meanings that reside in texts, broadly defined.
Consider, by way of illustration, the curriculum vitae (CV) in academic life. The typical
academic CV contains information on matters such as degrees earned; previous and
current posts; administrative responsibilities and experience; teaching experience;
research experience; research grants acquired; and publications. It can be seen as a
system of interlocking signifiers denotatively providing a summary of the individual’s
experience (its sign function). At the connotative level, it is an indication of an individual’s
value, particularly in connection with employability. Every CV is capable of being
interpreted in different ways and is therefore polysemic, but there is a shared code among
academics whereby certain attributes of CVs (such as lists of publications) are seen as
especially desirable and are therefore less contentious in terms of their meaning.
Applicants for posts know this and design their CVs to highlight the desired qualities and
downplay others, so that the CV becomes a presentation of self, as Miller and Morgan
(1993) suggested. Box 12.6 outlines a semiotic study (Gottdiener, 1997) in which the “text”
was Disneyland. The chief strength of semiotics lies in its invitation to see beyond and
beneath the apparent ordinariness of everyday life and its manifestations. The main
difficulty with semiotic analysis is that, although it may offer a compelling exposition of
some aspect of everyday life, the interpretation may be somewhat arbitrary. For example,
Gottdiener’s association of “Adventureland” with colonialism/imperialism might not be
shared by other analysts. However, the results of a semiotic analysis may be no more
arbitrary than those of any other interpretive approach. Indeed, it would be surprising not
to be struck by a sense of arbitrariness in semiotics, given its emphasis on the principle of
polysemy.

Hermeneutics
The hermeneutic approach has been used in understanding and interpreting the Bible, but
it has also been employed in the analysis of other texts. It was influential in the formation
of interpretivism as an epistemology (see Chapter 1) and has much in common with
Weber’s notion of Verstehen. The central idea behind hermeneutics is that the analyst of a
text must seek its meanings from the perspective of its author, considering the social and
historical context within which the text was produced. Phillips and Brown’s (1993)
hermeneutic study of the corporate image advertisements of a Canadian synthetic crude
oil company was a “formal analysis of the structural and conventional aspects of the text”
(p. 1563). For them this meant examining the texts of the advertisements in terms of their
constituent parts and the writing conventions they reflected. They also relied on a large
database of magazine and newspaper articles relating to the company for additional
documentary materials. They showed how corporate image advertisements were used to
mobilize support for company activities from government and the public, and to ward off
the environmental legislation that threatened them.

Two approaches to the study of language

In this section two approaches to the study of language are examined: conversation
analysis (CA) and discourse analysis (DA). While CA and DA do not exhaust the range of
possibilities for studying language, they do represent two of the most prominent
approaches, and each includes both quantitative and qualitative aspects. Both have
developed a technical vocabulary and a set of techniques. This section outlines some of
their basic elements and draws attention to their contrasting features.

Conversation analysis

The roots of CA lie in ethnomethodology, a sociological approach to communication

focusing on the “practical, common-sense reasoning” that people use in their everyday
lives. It includes notions of cause and effect (“if I do this, then that will happen”) and the
generalizations that allow people to perform everyday tasks. This process of reasoning and
communication is presented as a way in which social order is created. Social order is not
seen as a pre-existing force constraining individual action, but as something that is worked
at and achieved by people through interaction. Garfinkel, one of the pioneers of
conversation analysis, claimed that the role of sociology is not to uncover “objective”
social facts as Durkheim suggested, but to see them as an accomplishment, as the
eventually taken-for-granted patterns established through the activities of ordinary people
going about their daily lives (Garfinkel, 1967, p. vii).
Two ideas are particularly central to ethnomethodology and find clear expression in CA:
indexicality and reflexivity. The former suggests that the meanings of words or utterances,
including pauses and sounds, depend on the context in which they are used. Reflexivity
means that talk is not a “mere” representation of the social world, standing for something
else, but is itself a reality. In these ways, ethnomethodology fits squarely with two aspects
of qualitative research: a preference for a contextual understanding of action (see Chapter
9) and an ontological position associated with constructionism (see Chapter 1).

In the years following its introduction, ethnomethodological research sought to conduct

fine-grained analyses of the sequences of interaction revealed in conversations recorded
in naturally occurring contexts. As such, CA is a multifaceted approach, part theory, part
method of data acquisition, part method of analysis. The predilection for the analysis of
talk gleaned from naturally occurring situations suggests that CA fits well with another
preoccupation among qualitative researchers: a commitment to naturalism (see Chapter
9).

Conversation analysts have developed a variety of ways to study talk. Psathas (1995, p. 1)
described them as “rigorous, systematic procedures” that can “provide reproducible
results.” Such a framework brings to mind the procedures used to generate valid, reliable,
and replicable findings in quantitative research. It is not surprising, therefore, that CA is
sometimes described as having a positivist orientation. Thus, some features that are
broadly in tune with qualitative research (contextual and naturalistic analysis without prior
theoretical commitments) are married to traits associated with quantitative research.

However, the emphasis on context in CA is somewhat at variance with contextual

understanding as it is normally thought of in qualitative research. For CA practitioners,
context refers to the specific here-and-now context of immediately preceding talk,
whereas for most qualitative researchers it has a much wider focus, encompassing
phenomena such as the broader culture of the group within which the social action
occurs, including their values, beliefs, and typical modes of behaviour. This is precisely the
kind of attribution that CA practitioners seek to avoid. To import elements that are not
specifically grounded in the here and now of what has just been said during a conversation
is to risk imposing an understanding that is not grounded in the participants’ immediate
concerns. It is no wonder, therefore, that writers like Gubrium and Holstein (1997) treated
CA as a separate tradition within qualitative research, whereas Silverman (1993) found it
difficult to fit CA into broad descriptions of qualitative research.

Assumptions of conversation analysis

Analysts often begin practising conversational analysis when they notice something
significant about the way a speaker says something, and this recognition generates
interest in the function that the turn of phrase serves. To illustrate, Clayman and Gill (2004)
referred to the way a child will often begin a conversation with an adult by saying “You
know what, Daddy [or whoever]?” The question generally produces a “What?” reply that
allows the child to find a slot in a sequence of communications, or to inaugurate such a
sequence. The use of a question, rather than a declarative statement, may reflect the
child’s lesser power.

Once such a focus has been identified, conversation analysts typically follow certain basic
assumptions. Heritage (1984; 1987) proposed three:

• Talk is structured. Talk comprises patterns, and participants are implicitly aware of
the rules that underpin these patterns. As a result, conversation analysts do not attempt to
infer the motivations of speakers from what they say, or to ascribe their talk to purely
personal characteristics. Such information is unnecessary, since the conversation analyst
is oriented to the underlying structures that are revealed in pauses, emphases, questions
followed by answers, and so on.

• Talk is forged contextually. Talk must be analyzed in terms of its context and
understood in terms of the talk that has preceded it.

• Analysis is grounded in data. Conversation analysts shun prior theoretical schemes

and argue that the characteristics of talk and social order must be derived from the data.

Discourse analysis

Discourse analysis incorporates aspects of CA, but is more flexible and can be applied to
forms of communication other than talk. It puts less emphasis on naturally occurring talk,
so the language used in research interviews can also be a legitimate subject of analysis.
According to Potter (1997), DA “emphasizes the way versions of the world, of society,
events and inner psychological worlds are produced in discourse” (p. 146). For continental
philosophers such as Foucault (1926–84), “discourse” refers to the way a particular set of
linguistic categories relating to an object frames people’s understanding of that object.

For example, a certain discourse concerning mental illness can produce particular
conceptions of what mentally ill persons are like, the nature of their illness, how they
should be treated, and who is legitimately entitled to treat them. In this way discourse
serves to justify the power of those who provide such treatment as well as the treatment
regimens they prescribe. This discourse is much more than language: it is part of the social
world of mental illness.

Several different approaches fall under the DA heading (Potter, 1997). The version
discussed here is one associated with writers such as Potter (1997), Potter and Wetherell
(1994), and Billig (1991). This version of DA exhibits two distinctive features at the level of
epistemology and ontology (Potter, 1997):

• It is generally anti-realist in that it denies the existence of an external reality

awaiting a definitive portrayal by the researcher. It therefore disavows the notion that
researchers can arrive at accurate and objective accounts of the social world.

• It is constructionist in that it gives priority to the version of reality propounded by

members of the social setting being investigated. More specifically, it recognizes that
discourse analysis entails a selection from many possible interpretations of a given
situation. For example, is a person who speaks to herself dangerous or harmless? Is she
more in need of treatment or social support? In the process of answering such questions, a
particular reality is constructed.

Discourse is not simply a neutral device for imparting meaning. After all, there are many
things that people seek to accomplish when they talk or write. DA is concerned with the
strategies they employ in order to create different kinds of effect. This is illustrated in three
basic questions that DA asks:

• What is this discourse doing?

• How is it constructed to make this happen?

• What resources are available to perform this activity? (Potter, 2004, p. 609)
This action orientation is revealed in a study of the first few moments of telephone calls to
the helpline of the UK’s National Society for the Prevention of Cruelty to Children. Through
analysis of these call openings, Potter and Hepburn (2004) showed that a variety of actions
are performed during the first few moments:

• The callers spell out the details of their concerns.

• The callers seek to establish, in collaboration with the “experts” on the line (the
child protection officers) that the incidents they are reporting do in fact require some kind
of intervention.

• The callers convey the idea that the activity they are describing is serious and that
they are concerned about it, but not so concerned that the police should be contacted
immediately.

• The child protection officers show a willingness to treat the reports as serious,
without making judgments as to their actual truth or seriousness.

Analysis of these brief moments of conversation shows that the flow of discourse achieves
a number of objectives for both parties and in that sense constitutes action. On the other
hand, as Gill (1996) suggested, what is said is always a way of not saying something else.
Either way, discourse can be seen as providing a solution to a problem (Widdicombe,
1993).

DA shares with CA a preference for locating contextual understanding in terms of the

situational specifics of talk. As Potter (1997, p. 158) put it, discourse analysts prefer “to
see things as things that are worked up, attended to, and made relevant in interaction
rather than being external determinants.” However, DA practitioners are less wedded to
this principle than are conversation analysts.

Discourse analysts resist the idea of codifying their procedures and argue that such
codification is probably impossible. Potter (1997) described his work as “a craft skill, more
like bike riding or chicken sexing than following [a] recipe” (pp. 147–148). Gill (2000)
suggested adopting a posture of “skeptical reading.” This means searching for a purpose
lurking behind the way things are said or presented. Gill proposed that DA can be usefully
thought of in terms of four main themes, outlined in Box 12.9.

Access
One of the most important and yet most difficult steps in ethnography is gaining access to
the social setting one wants to research. How to gain access depends on several things,
one of which is whether the setting is relatively open (public) or closed (private or
restricted) (see Lofland & Lofland, 1995). Closed settings generally include organizations
of various kinds, such as firms, schools, cults, and social movements. Open settings are
areas where anyone can gain access, such as libraries, parks, and sidewalks, but even
there it can be quite difficult to make observations and talk to people. “Open” does not
necessarily mean easy access to people.

Overt versus covert ethnography

One way to ease the access problem is to assume a covert role: in other words, to not
disclose that you are a researcher. Getting access can be very problematic, and the
adoption of a covert role removes some of the difficulties. These two distinctions—open
versus closed setting, and overt versus covert role—suggest a fourfold distinction in forms
of ethnography Three points should be noted about Figure 10.1. First, the “open versus
closed setting” distinction is not hard and fast. Sometimes the process of gaining access
in “open” settings can take on an almost formal quality; for instance, if you are talking to a
gang leader in the street, you may have to answer some pointed questions about your
goals. Also, “closed” organizations and social movements sometimes create contexts that
have a public character, such as meetings arranged for members or prospective recruits.
Bell (2007) attended many western Canadian separatist meetings and rallies that were
open to the public, and was able to get a lot of information on the movement and its
leaders that way. Second, the “overt versus covert” distinction can vary from context to
context even within the same research project. Although ethnographers may seek access
through an overt route, many of the people with whom they come into contact may be
unaware of their status as researchers. Atkinson (1981, p. 135) noted in connection with
his research on the training of doctors in a medical school that although he was “an ‘open’
observer with regard to doctors and students,” he was “a ‘disguised’ observer with regard
to the patients.” Also, some ethnographers move between the two roles (see Box 10.1)
Access to closed settings

Gaining access to most organizations requires strategic planning, hard work, and
sometimes luck. In selecting a case study for an ethnographic investigation, the researcher
may employ several criteria, depending on the research area of interest. One may choose a
certain case or setting because of its “fit” with the research questions, but there is no
guarantee of access. Sometimes sheer perseverance pays off. Leidner (1993) was
determined that McDonald’s would be one of the organizations in which she conducted
ethnographic research on the routinization of service work:

The company was a pioneer and exemplar of routinized interaction, and since it was locally
based, it seemed like the perfect place to start. McDonald’s had other ideas, however, and
only after tenacious pestering and persuasion did I overcome corporate employees’ polite
demurrals, couched in terms of protecting proprietary information and the company’s
image. (Leidner, 1993, pp. 234–235)

This kind of determination is necessary whenever the target is a specific organization or

group (e.g., a particular religious sect or social movement). Rejection can mean having to
find a completely different research topic.

On the other hand, for many research questions you are likely to find that several potential
cases will be suitable. Organizational researchers have developed a range of tactics, some
of which may seem rather unsystematic but are still worth considering:

• Use friends, contacts, and colleagues to help gain access; as long as the
organization is relevant to the research question, the route should not matter.

• If possible, get someone in the organization to vouch for you and the value of your
research. Such people are considered “sponsors.” If permission is granted at a lower level
of the hierarchy, you may still need clearance from top management or senior executives.
People who must grant their permission for you to gain access are sometimes called
gatekeepers.

• Offer something in return: for example, a final report (if you provide a working draft,
people may ask you to make changes that you would not consider acceptable). However,
this strategy carries a risk of making the researcher a cheap consultant. Some writers on
research methodology do not recommend offering something in return, although it is
commonplace among researchers on formal organizations.

• Provide a clear explanation of your aims and methods. Suggest a meeting to deal
with the participants’ concerns and explain what will happen in terms that people who are
not familiar with social research can understand.
• Be prepared to negotiate; almost no one gets complete access.

• Be frank about the amount of people’s time you are likely to need. Time is an
important issue for commercial organizations, and for many non-profit groups too.

Access to open settings

Gaining access to people in open settings is similar in many ways. For instance,
sometimes ethnographers have their paths smoothed by “sponsors” and “gatekeepers.” In
seeking access to a group of soccer hooligans, Giulianotti (1995) sought out someone who
could play both roles for him. Later, he was able to call on existing acquaintances to ease
his entrée into a second group. In seeking access to female intravenous drug users, Taylor
(1993) contacted a drug counsellor who introduced her to some local users and
accompanied her on her first few research visits. A “research bargain” (Becker, 1970) was
struck in which Taylor agreed in return to see any of the counsellor’s clients who preferred
to discuss issues with a woman.

“Hanging around” is another common access strategy. An example of the difficulties that
await the researcher using this approach can be found in one of Whyte’s (1955) early field
encounters in Boston’s North End in his classic case study Street Corner Society. The
following incident occurred in a hotel bar:

I looked around me again and now noticed a threesome: one man and two women. It
occurred to me that here was a maldistribution of females that I might be able to rectify. I
approached the group and opened with something like this: “Pardon me. Would you mind
if I join you?” There was a moment of silence while the man stared at me. He then offered
to throw me downstairs. I assured him that this would not be necessary and demonstrated
as much by walking right out of there without any assistance. (Whyte, 1955, p. 289)

Wolf (1991) used a hanging-around strategy to gain access to Canadian outlaw bikers. On
one occasion he met a group of them at a motorcycle shop and expressed an interest in
hanging out with them. But he tried to move too quickly in seeking information about them
and was forced to abandon his plans. Eventually, Wolf’s hanging-around strategy led to his
being approached by the leader of a biker group (Rebels MC), who acted as his sponsor. To
bring this off Wolf ensured that he was properly attired in biker garb, although this was not
difficult for him as he had been a motorcyclist and had absorbed biker culture long before
he started his research. Attention to dress and demeanour can be a very important
consideration when seeking access to either public or closed settings.

Complete participant. The ethnographer is a fully functioning member of a social setting

but one whose true identity is unknown to members: in other words, a covert observer. In
such cases the ethnographer is engaged in regular interaction with people and participates
in their daily lives but assumes the researcher’s role in private to write down notes once
the situation has unfolded. This is what Humphreys (1970) did when he studied male
homosexual encounters in public washrooms.

• Participant-as-observer. The ethnographer participates as described above, except

that members of the social setting are aware that the ethnographer is studying them, such
as in van den Scott’s research (see Box 10.3).

• Observer-as-participant. The researcher is mainly an interviewer and observer, and

participates only marginally in the group’s activities. This was Bell’s (2007) role in his study
of western Canadian separatists. Ethnographic research on the police is often similar,
since the opportunities for genuine participation are limited by legal and safety
restrictions. Norris (1993) described how as an observer-as-participant, he concentrated
on gathering two types of data: “naturally occurring inter-officer talk” and “detailed
descriptions of how officers handled ‘live’ incidents” (1993, p. 126). For other examples of
situations in which minimal participation may be appropriate, see Boxes 10.4 and 10.6.

• Complete observer. The final possibility is no interaction with the people observed.
Most writers do not include this as a form of ethnography, since by definition there is little
or no involvement or participation. There may be less risk of reactivity (participants
behaving unnaturally because they know they are being observed) because the researcher
is at a distance; but there is also greatly reduced potential for understanding because the
researcher does not ask questions or try in any way to get into the heads of the people
under study.