Sampling and participant recruitment
Introduction to Social Research Methodology
From population to sample
A critical part of any study is deciding what to observe and what not to observe. The process of selecting those observations is called sampling, and it is the bridge between the people we actually study and the people we want to draw conclusions about. We almost never study everyone: surveying every customer or every citizen would be impossibly slow and expensive. Done well, sampling is astonishingly efficient — election polls of only 1,000 respondents routinely estimate national results with impressive accuracy. Done badly, it quietly wrecks a study, because no analysis, however sophisticated, can repair a sample that was selected in a biased way.
Population, element, sample
Three terms anchor everything that follows. An element is the unit about which information is collected — usually a person, though it can equally be a store, a team, or a transaction. The population is the entire aggregation of elements about which we want to draw conclusions. The sample is the subset of elements we actually study, chosen to stand in for the whole.
Defining the population sounds trivial but rarely is. “Meridian employees” could mean current staff only, current and former staff, permanent contracts only, or everyone including self-employed gig couriers. For a study of why turnover has doubled, the population that matters arguably includes the people who have already left — the very group a current staff list excludes. Because every claim a study makes is a claim about its population, the population must be defined explicitly before anyone is selected.
The sampling frame
A sampling frame is the actual list of elements from which a probability sample is drawn: a university’s student register, a company’s HR roster, a customer database, or Poland’s PESEL register of citizens. The crucial and easily forgotten point is that a properly drawn sample provides information about the population composing the frame — not necessarily the population the researcher had in mind. Good organisational frames exist for employees, members, and registered customers; they rarely exist for informal or transient groups. The frame also constrains the contact mode: a frame of postal addresses suits PAPI, e-mail addresses suit CAWI, and telephone numbers suit CATI — and each mode reaches some kinds of people better than others.
When the frame is not the population
The gap between frame and population is one of the commonest hidden flaws in applied research. Meridian’s HR roster lists everyone employed today, but “everyone who worked at Meridian this year” includes dozens of leavers who no longer appear on it. A turnover study sampled from the roster therefore systematically excludes the people with the most direct knowledge of why staff leave. Other classic gaps follow the same logic: a landline telephone frame under-covers young people; an e-mail frame under-covers warehouse staff without company accounts; a loyalty-card frame excludes occasional customers. The discipline to adopt is simple: state the population, name the frame, and list — in writing, before fieldwork starts — who falls through the gap, and whether their absence biases the answer.
Representativeness
A sample is representative of the population from which it is selected if the aggregate characteristics of the sample closely approximate those same aggregate characteristics in the population. If the workforce is 60% female, a representative sample contains close to 60% women; if 73% of staff work in retail stores, so should roughly 73% of the sample. The basic principle of probability sampling is that a sample will tend to be representative if all members of the population have an equal chance of being selected. No sample is perfectly representative; the operative question is whether its deviations from the population are small, random, and quantifiable, or large, systematic, and invisible. For a manager, the stakes are practical: an unrepresentative employee survey does not merely miss the truth — it hands the board a confident, precise-looking wrong answer.
Probability sampling designs
In probability sampling, every element in the frame has a known, non-zero chance of selection; in random selection, chance alone decides who is chosen. There are two reasons to let chance choose. First, randomness removes researcher bias — conscious or unconscious — in picking cases that might support a preferred conclusion. Second, and more important, random selection unlocks probability theory, which allows us to estimate the characteristics of the population and to say how precise those estimates are. Non-probability methods can be quick and useful, but only probability samples let researchers quantify their own uncertainty. (The mechanics of that quantification — margins of error and confidence — are the subject of session 13.)
The four main designs
| Design | How it selects | Best when | Watch out for |
|---|---|---|---|
| Simple random | Number every element; select by random numbers | A complete frame exists; baseline precision is wanted | Laborious on large frames |
| Systematic | Every k-th element from the list, random start | Long ordered lists | Hidden periodicity in the list |
| Stratified | Random selection within key subgroups | Subgroups matter and are recorded in the frame | Choosing irrelevant strata |
| Cluster | Sample groups first, then members within them | No element-level frame exists | Extra sampling error at each stage |
Simple random sampling is the baseline method assumed by the statistical computations of social research. Once a frame has been established, the researcher assigns a number to every element and uses random numbers — once a printed table, now a random-number generator — to select elements until the sample is full. Every element, and every combination of elements, has an equal chance of selection. In practice it is used less often than one might expect, because it demands a complete numbered frame and can be laborious to execute.
Systematic sampling is a practical shortcut that is, in practice, virtually identical to simple random sampling. The researcher computes the sampling interval k by dividing the population size by the desired sample size, and then selects every k-th element from the frame: a list of 10,000 with a target sample of 1,000 gives k = 10, so every tenth person is taken. To guard against human bias, the starting point within the first interval must be chosen at random — this is what keeps the method a probability design. The one caution is hidden periodicity: if the list has a recurring structure (for instance, every seventh roster entry being a Sunday shift lead), the interval can align with it and bias the sample, so the ordering of the frame should be checked first.
Stratified sampling is a method for obtaining greater representativeness by decreasing probable sampling error. Rather than sampling from the population at large, the researcher divides the frame into homogeneous strata — subgroups that matter for the question — and samples randomly within each. In a proportionate design, each stratum receives a share of the sample equal to its share of the population, and sampling error on the stratifying variable is reduced to zero. Organisations are naturally stratified — by division, site, seniority, and contract type — and HR data usually records all of these. With 330 retail, 70 e-commerce, and 50 HQ staff at Meridian, a simple random sample of 90 could by chance represent e-commerce poorly; stratifying by division guarantees each its correct share. A disproportionate design deliberately over-samples small strata about which precise statements are needed, and re-weights them at the analysis stage.
Cluster sampling is used when it is impossible or impractical to compile an exhaustive list of the elements composing the target population, but the elements come naturally packaged in listable clusters. The procedure is multistage: sample the clusters first, then list and sample elements within the selected clusters. The classic textbook example is church members: no list of all members exists, but a list of churches does, so researchers sample churches and then members within them. The organisational equivalent: there is no convenient frame of Polish retail workers, but there is a frame of retail companies — sample firms, then staff within them. Meridian itself cannot list its customers, but it can list its 32 stores, sample eight of them, and interview customers within those. The price of the convenience is that each stage adds sampling error, so a cluster sample is less precise than a simple random sample of the same size — a deliberate trade of precision for feasibility. Real large-scale surveys usually combine designs: stratifying regions, clustering by locality, and sampling systematically within clusters.
Non-probability sampling designs
Much social research is conducted in situations that do not permit probability sampling. Some populations have no frame and never will — homeless people, informal gig workers, users of a competitor’s product. Sometimes probability sampling is possible but not appropriate: a five-person exploratory study gains nothing from a random draw. Qualitative research in particular usually needs informative participants rather than statistically representative ones. The non-probability designs below are legitimate tools for these cases, provided the researcher stays honest about what they can and cannot support.
| Design | How it selects | Legitimate use | Key weakness |
|---|---|---|---|
| Convenience | Whoever is available | Cheap pilots; when the convenient group is the object of interest | No control over representativeness |
| Purposive | Deliberate choice of information-rich cases | Qualitative work; comparative designs | Describes chosen cases, not the population |
| Quota | Fill a matrix mirroring known population proportions | Commercial market research; online panels | Non-random selection within each cell |
| Snowball | Participants refer further participants | Exploratory work on hidden, networked populations | Reaches one social network, not the population |
Convenience sampling — sometimes called “haphazard” sampling — relies on available subjects, such as people passing a street corner, visitors to a website, or whoever happens to be in the canteen. It is the journalist’s person-on-the-street method, and an extremely risky one for social research, because it permits no control over the representativeness of the sample. It is justified only when the researcher genuinely wants to study the characteristics of the people passing the sampling point, or when less risky methods are not feasible — and even then, generalisation demands great caution.
Purposive (judgmental) sampling selects cases on the basis of the researcher’s knowledge of the population, its elements, and the purpose of the study. Participants are chosen because of who they are: the most experienced store managers, the newest hires, the teams with the best and worst retention. A comparative design can work well even without representativeness — sampling the members of contrasting groups suffices for comparing those groups, even if it describes neither group as a whole. Purposive sampling is the workhorse of qualitative research, where a handful of information-rich cases is worth more than a random scatter of uninformative ones. To understand exceptional stores, for instance, one might deliberately select Meridian’s three lowest-turnover and three highest-turnover stores and study the contrast.
Quota sampling begins with a matrix describing the characteristics of the target population: what proportion is male and female, and what proportions fall into each category of age, division, or contract type. Recruiters then collect data from people having the characteristics of each cell until every quota is filled, so that the finished sample mirrors the population’s known structure; weighted appropriately, the data should then provide a reasonable representation of the population. This is how much commercial market research and online panel work is done. The catch is that selection within each cell remains non-random — whoever was easiest to recruit fills the quota. Quota sampling superficially resembles stratified sampling, but the family is different: stratified sampling is random within strata; quota sampling is convenient within cells.
Snowball sampling is appropriate when the members of a special population are difficult to locate but know one another — homeless individuals, migrant workers, undocumented immigrants, niche professionals. The researcher collects data on the few members of the target population who can be located, then asks each to provide the information needed to locate other members they happen to know. Because the resulting samples have questionable representativeness — the researcher reaches one social network rather than the population — the method is used primarily for exploratory purposes. In organisational research, ex-employees, freelancers, and gig workers often form exactly this kind of referral-reachable population: the three couriers a company can contact each know other couriers.
The choice among all these designs is dictated by the question. Precision questions demand probability samples; depth, speed, and hidden-population questions justify non-probability ones. The cardinal sin is not using a non-probability sample — it is using one and then claiming population-level precision it cannot deliver.
Error, bias, and sample size
Sampling error and sampling bias
Even a perfect random sample will not match its population exactly. The random, quantifiable wobble between sample and population is sampling error; it shrinks predictably as samples grow, and probability theory tells us by how much. Sampling bias is different in kind: it means that those selected are not typical or representative of the larger population from which they were chosen — a systematic tendency, built into the selection procedure, to over- or under-represent certain sorts of people. Selecting people who are convenient, or allowing personal preference to affect who is studied, produces bias. Crucially, bias does not shrink as the sample grows: a bigger biased sample is just a more confident wrong answer. Sampling error is the price of sampling; sampling bias is a defect of design. Researchers budget for the first and work to eliminate the second.
Self-selection bias
The most common bias in organisational practice arises from letting participants choose themselves. The classic illustration is a newspaper polling its own readers: the results (a) reveal nothing about the national mood, because only a small proportion of citizens read that newspaper, and (b) probably reveal little even about the newspaper’s readership, because only a certain type of reader fills in polls. The modern equivalent is a company polling its newsletter subscribers and reporting the result as customer opinion: subscribers are the most engaged customers, and those who respond are the most opinionated subscribers. Two filters stack — who is in the frame, then who bothers to answer — and each tilts the sample further from the population. Volunteers systematically differ from non-volunteers: they are more engaged, more opinionated, and often more extreme in both directions. An open survey link in Meridian’s staff newsletter would over-hear the happiest and the angriest employees, and miss the quietly disengaged middle who are actually drifting towards the exit.
How big does a sample need to be?
Intuition says a sample must be a large percentage of its population; statistics says otherwise. What governs precision is the sample’s absolute size, not its share of the population: a well-drawn sample of 1,000 describes 38 million Poles about as precisely as it describes a city of 100,000. Precision also shows diminishing returns: growing a sample from 100 to 400 respondents improves it substantially, while growing it from 1,000 to 4,000 helps far less — as a rule of thumb, quadrupling the cost roughly halves the error. Two practical consequences follow. First, a small sample drawn properly beats a huge one drawn badly: 90 random employees are worth more than 300 self-selected volunteers. Second, sample size should be planned around the smallest subgroup that must be reported on — a sample of 90 at Meridian leaves only about 14 e-commerce staff, which is thin ground for any division-level claim.
Response rates and nonresponse bias
Selecting a sample is not the same as obtaining one: some of those invited will never respond. The response rate is the proportion of invited, eligible people who actually take part, and what counts as “taking part” — for instance, whether partial completions are included — should be decided and stated in advance. A low response rate is dangerous not in itself but through nonresponse bias: the risk that responders differ systematically from non-responders on precisely the things being measured. An engagement survey answered mainly by the engaged will overstate engagement, because the disaffected are exactly the people who ignore it. Good practice is to report the response rate honestly, compare responders with the frame on known characteristics (division, age, tenure), and send reminders — every additional reluctant respondent makes the sample a little less skewed. If 300 CAWI invitations at Meridian yield 96 completes, that is a 32% response — and the first question to ask is which third of the workforce answered.
Recruiting participants
Selecting people on paper is sampling; getting them to take part is recruitment — a craft of its own, and the place where surveys are won or lost in practice.
Recruitment channels
| Channel | Reach | Characteristic bias or cost |
|---|---|---|
| Internal comms / intranet | Cheap and broad within the organisation | Misses non-desk staff; reads as “management asking” |
| E-mail invitations (CAWI) | Targeted and trackable | Only reaches those with accounts; competes with a full inbox |
| Access panels | Pre-recruited pools of willing respondents; fast for customer research | Panellists are professional survey-takers |
| Social media | Reaches beyond the organisation — ex-staff, customers | Severe self-selection |
| In-store / on-site intercepts (PAPI) | Reaches customers and shop-floor staff no list covers | Real fieldwork cost |
The governing rule is to match the channel to the frame: a channel the target group does not use is a bias generator, not a recruitment tool.
Crafting an invitation that works
The invitation is the single highest-leverage text in the project, because most nonresponse happens at this hurdle rather than mid-questionnaire. A working invitation does four things: it says who is asking and why this person was selected; it states the topic and an honest time estimate (a promised “10 minutes” that turns into 30 poisons the well for every future survey); it explains what happens to the answers — anonymity or confidentiality stated plainly, with who will see what; and it names what the study is for — the concrete decision the results will inform, and when results will be shared back with participants. Beyond content, three practicalities matter: keep it short and personal where possible; send reminders, since two polite reminders routinely add more respondents than the original invitation; and choose the sender carefully — an invitation from a neutral research team draws franker answers than the same text signed by the CEO.
Incentives and their trade-offs
Incentives raise response rates; the real question is what they do to the sample and the answers. In rising order of cost and complication: no incentive relies on goodwill and topic salience, and is fine for short, clearly useful internal surveys; a lottery or prize draw is cheap per respondent but has a mild effect and attracts the prize-motivated; a guaranteed small incentive (a voucher, or a donation to charity) produces the most reliable lift and rewards everyone rather than just the lucky; and paid participation is standard for interviews, panels, and busy or external groups — expect to pay for an hour of someone’s time. The trade-offs are twofold: incentives can recruit people interested in the reward rather than the topic, and paying employees for an internal survey can feel odd where working time already covers it. Inside an organisation, time is the honest incentive: guarantee that participation happens on paid time, and much resistance dissolves.
Reaching hard-to-reach groups
Some of the most decision-relevant voices are the hardest to recruit, and skipping them silently biases the study. Senior managers are guarded by gatekeepers and diaries: recruit top-down through a sponsor, offer scheduling flexibility, and keep interviews short and sharply focused. Gig workers and contractors are absent from the HR roster and paid per task: recruit through the app or dispatch channel they actually use, pay for their time, and consider snowballing through referrals. Ex-employees appear in no internal channel: use exit-interview consent lists, professional networks such as LinkedIn, or personal e-mail addresses retained with consent, and consider a neutral external interviewer, who owes the company nothing and is easier to be frank with. Shift and shop-floor staff have no desk and no quiet hour: go to them, with intercepts in the back office, paper or tablet options, and kiosk time during shifts. The general principle covers all four: meet people in their own channel, on their own time, at a real value for their effort.
Recruitment ethics
Recruitment is where the research ethics of session 4 meet organisational hierarchy — and hierarchy can turn an invitation into an instruction. Consent must be genuinely voluntary, yet an e-mail from one’s line manager saying “please complete the survey” is, in practice, hard to refuse. Practical safeguards keep participation free: invitations come from the research team rather than the chain of command; there are no participation lists that managers can see, and no chasing of named non-responders by supervisors; and the statement that declining carries no consequences is made explicitly — and made true by organisational behaviour. Anonymity promises must also survive the analysis stage: reporting results for a five-person team identifies individuals as surely as printing their names. Finally, incentives must not shade into coercion — a reward large enough that a low-paid worker cannot afford to decline undermines voluntariness. And selection itself must stay out of interested hands: if store managers hand-pick which staff are interviewed about turnover, the study inherits every manager’s interest in looking good.
Conclusion
Sampling is the bridge between the people studied and the people about whom conclusions are drawn, and every claim a study makes is only as strong as that bridge. The discipline begins before fieldwork: define the population, name the frame, and account for the gap between them before a single invitation is sent. Probability designs — simple random, systematic, stratified, and cluster — buy generalisation and quantifiable precision; non-probability designs — convenience, purposive, quota, and snowball — buy feasibility, speed, and depth. The sin is not choosing the second family but claiming the precision of the first while doing so. On size, the rule is that absolute numbers beat percentages of the population, with sharply diminishing returns — but no sample size cures bias, and self-selection and nonresponse do the quiet damage. Recruitment is sampling’s last mile: the right channel, an honest invitation, fair incentives, and a hierarchy kept at arm’s length. The next session turns to what we actually ask the people we have recruited: the design of the questionnaire itself.