Quota Sampling: Definition, Method, and Examples

Quota sampling is a non-probability sampling method in which a researcher divides a population into subgroups defined by one or more characteristics (age, sex, occupation, region, and similar traits are common) and then fills a fixed quota of participants from each subgroup using non-random, often convenience-based, selection — typically until each quota is met. It resembles stratified sampling in structure but differs in one decisive way: stratified sampling draws each subgroup’s sample randomly, while quota sampling does not. That single difference is what defines quota sampling, and it is also the most commonly confused point about the method.

This guide covers how quota sampling actually works, how to construct quota controls, the difference between proportional and non-proportional quota sampling, worked examples, how it compares to stratified and convenience sampling, and its advantages and limitations for research design.

How Quota Sampling Works

Quota sampling proceeds in four steps:

1. Identify the control characteristics. Choose one or more variables believed to be relevant to the outcome being studied — for example, sex, age bracket, or employment status — and for which the population’s known distribution (from census data, administrative records, or a prior survey) is available.
2. Set the quotas. Decide how many participants are needed from each subgroup defined by those characteristics. Quotas can be set to mirror the population’s actual proportions (proportional quota sampling) or set independently of population proportions to guarantee a minimum number in each cell regardless of how small that group is in the population (non-proportional, or “even,” quota sampling).
3. Recruit participants to fill each quota. Fieldworkers or an online panel recruit whoever is available and willing until each subgroup’s quota is filled. Selection within a subgroup is not random — this is the step that distinguishes quota sampling from stratified sampling, where selection within each stratum is done by simple random or systematic sampling from a defined sampling frame.
4. Stop recruiting once every cell is full. Once each quota is met, sampling for that subgroup ends, even if more willing participants are available.

Because selection is non-random, quota sampling does not require a sampling frame (a complete list of the population) — only reliable estimates of the population’s distribution across the control characteristics. That is also why quota sampling is a non-probability method: unlike simple random sampling or stratified sampling, not every member of the population has a known, non-zero chance of being selected, so standard sampling-error formulas do not strictly apply to a quota sample.

Proportional vs. Non-Proportional Quota Sampling

• Proportional quota sampling sets each subgroup’s quota to match its share of the population. If a population is 55% female and 45% male, a target sample of 200 would set quotas of 110 and 90 respectively, so the sample’s composition mirrors the population’s on the control variables.
• Non-proportional quota sampling sets quotas independently of population share, often to guarantee enough participants in a small subgroup for meaningful subgroup analysis. A researcher studying attitudes across four age brackets might set an equal quota of 50 per bracket even though the brackets are not equally sized in the population, deliberately over-sampling smaller groups.

Worked Example

The following is an illustrative example to show the mechanics, not a report of an actual study. A researcher wants 300 respondents for a survey on transit use in a city where, per the most recent census, the working-age population is 60% employed full-time, 25% employed part-time, and 15% not employed. Using proportional quota sampling, the researcher sets quotas of 180, 75, and 45 for those three groups. Interviewers are then sent to transit stations, workplaces, and community centers and stop each passer-by matching the day’s target category, asking screening questions to confirm employment status before proceeding, until each quota is filled. No random number table or sampling frame is used at any point — selection is by interviewer judgment and availability, constrained only by the quota targets.

A second, non-proportional version of the same study might instead set an equal quota of 100 per employment-status group, regardless of the 60/25/15 population split, specifically to ensure the part-time and not-employed groups are large enough to analyze separately with reasonable precision — at the cost of the overall sample no longer reflecting the population’s true composition unless the results are reweighted at analysis.

Quota Sampling vs. Stratified Sampling

These two methods are the ones researchers confuse most often because both start by dividing the population into subgroups defined by shared characteristics. The difference is entirely in how units are chosen within each subgroup:

Dimension	Quota Sampling	Stratified Sampling
Sampling category	Non-probability	Probability
Selection within subgroup	Non-random (convenience, interviewer judgment, or first-available)	Random (simple random or systematic sampling from a defined frame)
Sampling frame required	No — only population proportions/estimates	Yes — a complete list of the population with the stratifying variable recorded
Each unit’s selection probability	Unknown	Known and calculable
Sampling error can be formally estimated	No	Yes
Typical use case	Market research, opinion polling, exploratory or pilot work under time/budget constraints	Academic and government surveys where population-level inference is required

Because quota sampling lacks random selection within subgroups, it is vulnerable to selection bias that stratified sampling largely avoids: whoever is easiest for an interviewer to approach and recruit is systematically more likely to be included, which can skew results in ways that are difficult to detect or correct after the fact. See sampling bias for a fuller treatment of this risk.

Quota Sampling vs. Convenience and Purposive Sampling

Quota sampling is best understood as a structured variant of convenience sampling: convenience sampling recruits whoever is easiest to reach with no subgroup targets at all, while quota sampling adds population-informed subgroup targets on top of that same non-random recruitment approach. It differs from purposive sampling in intent — purposive sampling deliberately selects information-rich cases to serve a specific analytical purpose (often in qualitative research), whereas quota sampling aims to approximate a population’s demographic composition using whoever is available. The comparison of purposive sampling vs. convenience sampling covers that distinction in more depth; quota sampling sits closer to the convenience end of that spectrum, structured by demographic controls.

Advantages of Quota Sampling

• Fast and inexpensive. No sampling frame has to be built or obtained, and fieldwork can start as soon as population proportions and quotas are set.
• Guarantees subgroup representation. Unlike simple random sampling, which can by chance under-represent a small subgroup, quota sampling guarantees each defined subgroup reaches its target size.
• Useful when no sampling frame exists. For populations with no complete list available — shoppers at a mall, attendees at an event, users of an informal service — quota sampling is often the only practical way to approximate a representative sample.
• Flexible during fieldwork. Interviewers can adapt who they approach in real time to fill remaining quotas, without needing to track a pre-selected list of specific individuals.

Disadvantages and Limitations

• Selection bias within subgroups. Because selection is non-random, whoever is most accessible, most willing, or most similar to the interviewer is systematically over-represented, and this bias is not correctable through weighting the way frame-based coverage gaps sometimes are.
• No valid measure of sampling error. Because individual selection probabilities are unknown, confidence intervals and margins of error calculated on a quota sample are not statistically justified in the way they are for a probability sample, even though they are sometimes reported informally.
• Results cannot be generalized to the population with statistical confidence. A quota sample can look demographically similar to the population on the control variables while still being unrepresentative on the variables that actually matter to the research question.
• Quality depends heavily on fieldworker judgment and honesty. Because there is no external check on who gets approached, results are sensitive to interviewer behavior — including a tendency to approach easier or more agreeable respondents first.

Why Matching on Quota Variables Guarantees Nothing Else

The single most important limitation of quota sampling is easy to state and easy to forget: a quota sample is guaranteed to match the population on the variables used as controls, and is guaranteed nothing at all on every other variable. The guarantee is narrow by construction. If the quotas are age, sex and region, then the sample will match the population on age, sex and region — that is what filling the quotas means — and matching on those three says nothing about income, education, employment status, health, political attitude, or whatever the survey is actually measuring, unless those happen to be strongly determined by the three control variables.

This is not a modern objection. Jerzy Neyman made it in 1934, reviewing an Italian study that had selected sampling units purposively to match a set of population controls — an approach the American Association for Public Opinion Research’s task force on non-probability sampling later described as akin to quota sampling. Neyman’s finding was that “though the average values of seven controls used are in satisfactory agreement, the agreement of average values of other characters, which were not used as controls, is often poor,” and, pointedly, that “the agreement of other statistics besides the means, such as frequency distributions, etc., is still worse.” Two things in that sentence still apply directly to any quota sample you will design or review today. Agreement degrades as soon as you leave the control variables. And it degrades faster for distributions than for means — so a quota sample that produces a defensible average may still misrepresent the spread, the tails, and the shape, which are frequently what a subgroup analysis depends on.

The practical consequence for study design is that choosing quota variables is a substantive decision, not an administrative one. The useful question is not “which demographics are conventional?” but “which observable characteristics are most strongly related to the outcome I am measuring, such that matching on them constrains the outcome too?” Quotas on age and sex do very little for a survey about commuting if the thing that actually drives commuting behavior is distance from the workplace and that was never controlled. Variables that influence the outcome but are left out of the quotas are what Leslie Kish called disturbing variables, and they are free to differ between sample and population by any amount.

What the 1948 Polling Failure Showed

Quota sampling has the unusual distinction of being the method whose public failure reshaped an entire field, and the details of that failure are more instructive than the anecdote.

Quota sampling first made its reputation in 1936, when the Literary Digest straw poll — 2.3 million ballots returned out of 10 million distributed, an enormous sample by any standard — called the US presidential election wrongly, while a far smaller quota sample run by George Gallup called it correctly. The lesson drawn at the time, correctly, was that a small well-designed sample beats a huge uncontrolled one.

Twelve years later the method failed on the same stage. In 1948 the three leading pollsters — Crossley, Gallup and Roper — all used quota sampling, and all three predicted the wrong winner. The influential 1949 post-mortem by Frederick Mosteller and colleagues identified many contributing sources of error. On quota sampling specifically, it found that leaving final selection to interviewer discretion within the quotas produced samples of somewhat more educated and better-off respondents than the population, which biased the results in a consistent direction.

The part of that report most worth carrying forward, though, is what it did not say. It did not conclude that quota sampling caused the wrong call — it declined to blame the method per se. Its central objection was different and more fundamental: a quota sample provides no measure of the reliability of its own estimates. Because selection probabilities are unknown, there is no sampling distribution to work from, so there is no valid standard error, no valid confidence interval, and therefore no way to tell a result that is off by a point from one that is off by five. The pollsters were not merely wrong in 1948; they had no instrument that could have told them how wrong they might be. Gallup moved to probability-based methods soon afterwards and the rest of the industry followed.

That objection is unchanged today, and it is the reason a quota sample’s margin of error should be treated with suspicion whenever one is quoted. Any margin of error attached to a quota sample has been computed as though the sample were random, which it is not; the number describes a hypothetical probability sample of the same size, not the sample in hand, and it captures none of the selection bias that is the dominant error term.

Quota Sampling in Modern Online Panels

Most quota sampling encountered in research today happens without an interviewer anywhere in sight. Online research typically draws from an opt-in panel — people recruited in advance who have agreed to take surveys — and quotas are applied by closing a demographic cell to further respondents once its target is filled. The mechanics differ from street-corner fieldwork, but the statistical situation is the same one described above, with one addition: the panel itself was assembled non-randomly, so selection operates twice, once at recruitment into the panel and again at admission to the survey.

Two things follow for anyone designing or reviewing this kind of study:

• Quotas and weights are doing the same job, at different stages. Quotas match the sample to the population during fieldwork; post-stratification weighting adjusts it afterwards. Both operate only on the variables they are given, and neither repairs a difference on a variable that was not measured. Reporting the quota frame and the weighting variables together — and stating what population benchmark they were matched to, and when it was measured — is the minimum a reader needs to judge the estimates. Quotas set against a census that is several years stale will match a population that no longer exists.
• Say what the design was, and do not describe it as random. The AAPOR task force’s practical point about opt-in panels is that they are not a single method: recruitment routes, cell definitions and matching techniques vary enormously between vendors, so an evaluation has to focus on the specific sampling method rather than the panel’s name. A methods section reading “a nationally representative sample of 2,000 adults” describes a quota-matched opt-in sample as though it were a probability sample. “Quota sample of 2,000 adults from an opt-in online panel, with quotas on age, sex, region and education set to [benchmark], results weighted on [variables]” describes what happened, and lets a reader form their own view about the disturbing variables that were left uncontrolled.

When to Use Quota Sampling

Quota sampling is a reasonable choice when a study needs a sample that broadly mirrors a population’s known demographic composition, but a full probability sample is not feasible — typically because of time or budget constraints, or because no usable sampling frame exists for the population. It is common in commercial market research, opinion polling, and exploratory or pilot studies. It is a poor choice whenever the research requires defensible statistical inference to a wider population — peer-reviewed academic research, clinical epidemiology, and government statistics generally require probability methods such as simple random, stratified, systematic, or cluster sampling for exactly this reason. See the sampling methods overview for how quota sampling fits among the full set of probability and non-probability approaches.

Frequently Asked Questions

What is an example of quota sampling?

A researcher who needs 300 survey respondents and knows, from census data, that a city’s adult population is 52% female and 48% male would set quotas of 156 female and 144 male respondents, then have interviewers approach people at a public location — a shopping center, for instance — asking whoever they encounter to participate until each quota is filled. See the worked example above for a fuller version with a second control variable.

How do you define a quota sample?

A quota sample is a non-probability sample built by dividing a population into subgroups based on characteristics believed relevant to the research question, setting a target number (a quota) for each subgroup based on known or estimated population proportions, and then recruiting non-randomly — usually by convenience — until each quota is met.

What are the advantages and disadvantages of quota sampling?

The main advantages are speed, low cost, and the ability to guarantee subgroup representation without needing a sampling frame. The main disadvantages are selection bias introduced by non-random recruitment within each subgroup, the inability to calculate a statistically valid margin of error, and limited ability to generalize findings to the wider population with confidence. See the advantages and disadvantages sections above for the full list.

Is quota sampling the same as stratified sampling?

No. Both divide a population into subgroups first, but stratified sampling then selects participants from each subgroup randomly (making it a probability method with calculable sampling error), while quota sampling selects participants from each subgroup non-randomly, typically by convenience (making it a non-probability method without a valid sampling-error estimate). See the comparison table above.

Is quota sampling a type of convenience sampling?

Quota sampling is often described as structured convenience sampling: it uses the same non-random, convenience-based recruitment as convenience sampling, but adds population-informed subgroup quotas that convenience sampling alone does not use.

What are the disadvantages of quota sampling?

Three, in order of seriousness. First, no valid measure of precision: because selection probabilities are unknown, there is no sampling distribution, so any margin of error quoted for a quota sample has been calculated as though it were random and does not describe the sample in hand. Second, selection bias within cells: whoever is easiest for an interviewer or a panel to reach is systematically over-represented, and in 1948 that produced samples that were more educated and better off than the population. Third, and most often missed, the sample is matched to the population only on the quota variables — everything else is free to differ, and it differs more for distributions and tails than for means.

How do you choose quota variables?

Pick the observable characteristics most strongly related to what you are measuring, not the demographics that are conventional. The guarantee a quota buys is confined to the variables you control on, so quotas on age and sex constrain an outcome only to the extent that age and sex determine it. If the dominant driver of your outcome is something else — distance to the workplace in a commuting survey, condition severity in a patient survey — and it is not in the quota frame, the sample can match the population perfectly on paper while being badly unrepresentative on the thing that matters. You also need a credible, current population benchmark for each quota variable; quotas set against a stale census match a population that no longer exists.

Can you calculate a margin of error for a quota sample?

Not validly. Margin of error is derived from known selection probabilities, and a quota sample has none. Software will still print a number, because it computes it under the assumption of random sampling, but that number describes a hypothetical probability sample of the same size rather than your actual sample, and it excludes selection bias, which is usually the largest error component. This was the central objection the 1949 review of the 1948 polls raised against quota sampling — not that it had caused the wrong prediction, but that it offered no way to know how wrong the prediction might be.

Is quota sampling still used?

Very widely, though rarely with a clipboard. Most online survey research draws from opt-in panels and applies quotas by closing demographic cells as they fill, then weights the result afterwards. Market research, pilot studies, and any project without a usable sampling frame or the budget to build one use it routinely. What changed after 1948 is not that the method was abandoned but that claiming probability-sample properties for it became indefensible — which is why a methods section should name the design as a quota sample, state the quota variables and their benchmark, and report the weighting separately.