Chapter 8: Nonprobability Sampling and Choosing Informants

Exam Notes — Course 103

1. When Nonprobability Sampling Is Used

A random, unbiased probability sample is the general rule in research, but there are specific circumstances where a
nonprobability sample is exactly what is called for.

• In-depth studies of a few cases: Labor-intensive, in-depth studies (such as narrative analysis or life history research)
usually work with very few cases, so cases must be chosen on purpose rather than randomly. In-depth research on
sensitive topics (sexual history, mental illness, drug use) requires informed informants rather than randomly selected
respondents.
• Large surveys where a probability sample is not achievable: If, despite best efforts, a true probability sample
cannot be obtained, researchers use a nonprobability sample instead and simply document the resulting bias openly,
rather than agonizing over it.
• Collecting cultural data: When the goal is to learn about a culture as a whole (rather than about individuals), expert
informants — not randomly selected respondents — are what is really needed, since experts know cultural norms
better than an average random person would.
• The four major nonprobability methods: Quota sampling, purposive (judgment) sampling, convenience (haphazard)
sampling, and chain referral sampling (snowball and respondent-driven sampling). Case control sampling, in addition,
combines elements of both probability and nonprobability sampling.

2. Quota Sampling
• Quota sampling: In quota sampling, the researcher decides in advance on the subpopulations of interest (e.g., gender,
age, occupation) and the proportion each subpopulation should make up in the final sample, then goes out and fills
those quotas by finding enough matching people.
• Strength and risk: Quota sampling is an art that often approximates the results of probability sampling at lower cost
and less hassle, since experienced interviewers learn not to simply pick people who are easiest or most pleasant to
interview. However, it is not truly unbiased — quota samples are biased toward people who are easy to find, which
has famously led to wrong election predictions (e.g., the 1948 Dewey vs. Truman U.S. election and the 1992 British
general election).
• Where it is appropriate: Quota sampling works well in the study of cultural domains, where interviewing a handful
of genuinely knowledgeable people can capture the relevant cultural information, and it is also commonly used for
narrative studies where only a limited number of in-depth interviews are feasible.

3. Purposive or Judgment Sampling

• Purposive sampling: In purposive sampling, the researcher decides on the purpose informants (or communities) need
to serve and then goes out to find people who fit, without any overall design specifying exactly how many of each type
is needed — the researcher simply takes what is available.
◦ Pilot studies: Studies done before running a larger study, used to develop and test research instruments before
the main data collection begins.
◦ Intensive case studies: Used to identify and describe a specific cultural phenomenon in depth, often relying on
a small, carefully chosen group of participants.
◦ Critical case studies: Cases chosen because they specifically meet an important criterion — for example,
choosing key ethnographic informants, or polling companies targeting communities that have historically
predicted election winners.
 ◦ Hard-to-find populations: Purposive sampling is almost always necessary when studying populations that are
difficult to locate or reach, such as refugees or other hidden or stigmatized groups, since a true random sample
of such populations is generally impossible to obtain.

4. Convenience or Haphazard Sampling

• Convenience sampling: Convenience sampling means simply working with whoever is available and willing to
answer questions. Studies of homeless populations or hospital intensive-care patients are often necessarily done using
convenience samples, since no better option is realistically available.
• Limitation: A convenience sample cannot be used to reliably estimate a population parameter (such as a percentage)
— it can only suggest the range of possible experiences or opinions that exist, not how common each one actually is in
the wider population.
• Turning convenience into purposive sampling: If a convenience sample is deliberately made representative of
exactly what the researcher wants it to represent, it effectively becomes a purposive sample rather than a purely
haphazard one.

5. Chain Referral or Network Sampling: Snowball and RDS

• Snowball sampling: The researcher starts with one or two informants found through key informants or documents,
then asks them to name and recommend others in the same population. The sampling frame grows with each new
interview until it becomes saturated (no new names are offered).
• Risk in large populations: Snowball sampling works well for small, tightly connected populations, but in large
populations it is risky because well-known people and people with large networks are more likely to be named
repeatedly, which biases the resulting sample toward more visible or well-connected individuals.
• Respondent-Driven Sampling (RDS): Developed by Douglas Heckathorn to fix snowball sampling’s problems. A
few paid “seed” informants are given coupons to distribute to members of their own networks; anyone who shows up
with a coupon is also paid the same amount. This method reduces bias compared to traditional snowball sampling and
also avoids the ethical problem of a researcher directly identifying and approaching people who never agreed to be
named.

6. Case Control Sampling

• Case control sampling: A purposive sample is chosen based on some criterion (such as having a certain illness,
injury, or experience, e.g., a suicide attempt or homelessness) and is then matched against a comparison group of
“controls” who share similar background characteristics (age, sex, income, etc.) but do not share the criterion being
studied. This method is widely used in public health research.
• Example: Pfeiffer et al. studied malnourished versus well-nourished children (matched on income, household type,
and water access) in Mozambique and found that the mother’s education level and the household’s protein
consumption — not household income — were the key factors distinguishing the two groups.

7. Sampling and Credibility

• Internal vs. external validity: Good measurement is the key to internal validity, while representative sampling is the
key to external validity. A well-executed nonprobability sample is actually part of good measurement and contributes
to a study’s internal validity and overall credibility.
• Extending credibility beyond the sample: If a researcher wants their conclusions to generalize beyond the specific
group studied, they must either (1) repeat the study one or more times using additional nonprobability samples, or (2)
use a true probability sample instead.
8. Choosing Informants: Key and Specialized Informants
• Respondents, subjects, and informants: These three terms tend to be used by sociologists, psychologists, and
anthropologists respectively. Respondents respond to survey questions, subjects take part in experiments, and
informants tell the researcher what they think the researcher needs to know about their culture.
• Key informants: Key informants are people who know a great deal about their culture and are, for their own reasons,
willing to share that knowledge with the researcher. In long-term ethnography, researchers develop close relationships
with a few key informants — relationships that can last a lifetime and are not simply chosen but mutually developed
over time.
• Specialized informants: Specialized informants have particular expertise in a specific cultural domain — for
example, someone knowledgeable about local marriage customs, healing practices, or religious rituals — and are
sought out specifically to speak knowledgeably about that narrow topic.

9. Finding Key Informants

• Qualities of a good key informant: Good key informants are people the researcher can talk to easily, who understand
exactly what information is needed, and who are glad to provide it or help obtain it. The most useful informants tend
to be observant, reflective, and articulate people who are somewhat cynical or intellectually critical about their own
culture, even while remaining trusted “solid insiders.”
• A note of caution: Researchers should not choose key informants too quickly — it is important to first spend time
getting to know the field setting, check the roles and social standing of potential informants, and make sure that
relying on one informant does not accidentally block access to other important people in the community.

10. Informants Sometimes Lie

• Informants can and do lie: Researchers should not be surprised when informants lie to them — this has been
documented repeatedly in the ethnographic literature, sometimes for reasons of privacy, embarrassment, mischief, or
simply because a culture places high value on skilled storytelling.
• Example: Among the Nissan of New Guinea, the most articulate informants were found to be both great truth-tellers
and accomplished liars at the same time, since community leaders were culturally expected to give persuasive
speeches and “manipulate” meaning — even if that meant telling outright lies.

11. Paying Informants

• General rule: If informants are being interviewed and their time is being used for research, they should generally be
paid at the local going rate for their time. If the people being studied are already wealthy, direct payment may be
inappropriate — in such cases, making a donation to a charity they support is often a better alternative.