02 Section A, Key question 2 - Non-probability sampling (part 3)
How people enter a sample affects whose experiences a sociologist can hear and what conclusions they can draw. This lesson explains snowball, volunteer, opportunity, purposive and quota sampling, including why a sample can be useful without representing a whole population statistically. All scenarios and numbers below are original teaching examples, not reported research findings.
What makes selection non-probability
Non-probability sampling selects participants without a probability procedure that gives each member of the population a known, non-zero chance of selection.
The target population is the group the research concerns; a sample is the smaller group actually studied. A researcher investigating adults who care unpaid for a relative must decide which carers to approach, rather than treating anybody willing to talk as automatically relevant. The decision involves both who qualifies and how those people are found.
Non-probability sampling
Selecting cases through a non-random procedure, such as availability, volunteering, referrals, deliberate criteria or category quotas, rather than known selection probabilities.
This differs from a probability sample drawn randomly from a suitable sampling frame, a list or other record from which cases can be selected. Non-probability sampling can work without a complete frame, but people outside its recruitment channels may have no chance of entering. Selection bias occurs when the recruitment process systematically favours some relevant experiences over others. It does not mean that participants are lying.
Representativeness concerns how far a sample reflects relevant characteristics of its target population. Statistical generalisation uses sample findings to make estimates about that population. Without known selection probabilities, a non-random sample does not by itself justify a claim such as “this percentage applies to all carers”. A large sample can still systematically omit carers with little free time. A probability design provides a stronger basis for population estimation, although a poor frame or uneven refusal can also damage its achieved sample. Statistics Canada explains the selection and inference distinction.
A different aim is to understand how caring responsibilities are experienced. Carefully chosen interviews might reveal how people negotiate paid work or explain a pattern worth investigating elsewhere. This can support a reasoned explanation of a process without estimating its prevalence. Non-probability does not mean unplanned, and qualitative does not mean automatically valid: claims must still fit the evidence collected.
Opportunity sampling and immediate access
Opportunity sampling recruits people because they are readily available at the time and place where the researcher is working.
Suppose a sociologist stands outside a town library on Tuesday afternoon and asks available adults about using council services. People enter because they happen to be there and can be approached. This is opportunity sampling, also called convenience sampling. Saying the interviewer asks “people at random” in everyday language does not establish a statistical random procedure.
The practical advantage is immediate access: a researcher with little funding can begin collecting responses without assembling a complete list. This can be appropriate for a pilot that checks whether a question is understandable. However, the setting and time shape inclusion. Adults working elsewhere that afternoon, people unable to travel and people who do not use the library are absent. The sample may therefore give a misleading picture of barriers to council services, precisely because those facing the greatest barriers may never arrive.
Here is the reasoning chain: library attendance determines availability; availability influences selection; experiences of those unable to attend are missing; conclusions about every adult in the town are therefore insecure. Moving to several venues and different times can widen access, but it does not turn selection into a probability sample. Approachable people may still be favoured at each venue.
Permission to stand in a building is also different from an individual's agreement to participate. An adult approached at a counter should be able to decline without believing that service provision depends on answering.
Volunteer sampling and self-selection
Volunteer sampling recruits people who put themselves forward in response to an invitation to participate.
A researcher advertises interviews about unpaid caring on a community website. Interested carers contact the researcher, who checks that they meet the study's stated eligibility criteria. The defining feature is self-selection: people decide to respond to the invitation. It differs from opportunity sampling, where the researcher approaches people because they are available. Recruitment can combine procedures, so identify what happens at each stage rather than assuming that every design has only one label.
All ethical research normally seeks voluntary participation. That does not make every consenting participant part of a volunteer sample. A person randomly selected from a list can agree freely; their initial selection still came from a probability procedure. “Volunteer sampling” describes how respondents are recruited, not whether coercion is acceptable in the other methods.
An invitation can reach interested people across several locations and make detailed interviews feasible. Yet carers who have time to respond, see the website or feel especially strongly about their experience may be more likely to enrol. Self-selection bias arises if the motivations or resources behind volunteering also relate to what the research measures. For example, an invitation to discuss difficulties may attract carers wishing to describe serious problems. This is a possible bias, not proof that every volunteer has an extreme opinion.
The researcher should state where the invitation appeared and who could realistically see and answer it. Offering several accessible routes to respond may reduce exclusion. It still does not establish that respondents resemble non-respondents in all relevant respects. Their accounts can reveal important meanings without supporting an estimate of how common each account is among all carers.
Snowball sampling through referral networks
Snowball sampling uses contacts with initial participants to reach further people through referrals.
Suppose researchers want to understand how adults doing informal cash-paid work experience insecurity. There may be no complete accessible list of this population. After interviewing initial eligible participants, researchers invite them to pass information about the study to other eligible people they know. Those people can contact the researchers and, if they participate, pass the invitation onward. Recruitment develops through a chain of social connections.
An introduction from a trusted acquaintance may make contact possible where an unfamiliar researcher would receive no response. Hidden populations are groups difficult to identify or contact through ordinary research channels; they are not necessarily involved in illegal activity. Atkinson and Flint's methodological discussion (2001) explains both the access benefits of referral networks and the risk of missing isolated people. It is an account of research strategies, not evidence that every snowball sample represents its population.
The main limitation follows from the same mechanism as the advantage. People often know others with similar work, locality or outlook, so one chain may repeatedly recruit similar cases. This is network bias. Someone with no connection to the initial contacts cannot be reached by that chain. A finding that participants rely heavily on friends could partly reflect selecting people through friendship networks in the first place. Starting several chains from different relevant settings can broaden the experiences heard, but cannot guarantee inclusion of people outside all those networks.
Referrals also create ethical risks. Giving a researcher another person's name can disclose sensitive circumstances before that person has agreed. A safer approach in this scenario is for participants to pass on a neutral invitation so potential recruits can make private contact themselves. Researchers should explain that participation is optional and avoid telling the referrer who accepted or what they said. The referrer's cooperation is not the referred person's consent.
Continue this lesson.
You’re previewing NovaLearn without an account. Create a free account to read the rest.
Create free account