What is a cluster sample? (with examples)

By Jude Wallis · Updated July 31, 2026

A cluster sample splits the population into groups called clusters, each one meant to mirror the whole population, takes a simple random sample of whole clusters, and collects data from every individual inside the selected clusters. You sample all of a few groups, not some of every group.

AP Statistics: Unit 1 (topics 1.11 Random Sampling). The Fall 2026 course defines the cluster random sample in Unit 1 topic 1.11, where you are asked to identify a sampling method from a description of a study (skill 2.A) and justify its appropriateness (skill 2.B).

What a cluster sample is

A cluster sample starts by splitting the population into groups called clusters. Ideally each cluster mirrors the variety of the population, and the clusters look similar to one another. You then take a simple random sample of the clusters themselves, and you collect data from every individual inside each selected cluster.

Picture a district with 40 elementary schools. You number the schools 1 to 40, use a random number generator to pick 5 of them, and survey every student in those 5 schools. The randomness acts on schools, not on students. Once a school is drawn, all of its students are in the sample, and the other 35 schools contribute nobody.

That last part is what makes cluster sampling feel strange at first. Most of the population has zero chance of being measured once the draw is done, and yet the method has no systematic tendency to run high or low, because chance alone decides which clusters get drawn.

Cluster vs stratified: the contrast students reverse

Stratified and cluster sampling both divide the population before sampling, and that shared first step is why students swap them. The two divisions are built for opposite reasons and then sampled in opposite ways.

Strata are internally similar. You build them so the individuals inside one stratum are alike on the trait that matters: grade level, income band, region. Then you take a separate simple random sample within every stratum and combine the pieces into one sample. Every stratum is guaranteed to appear.

Clusters are internally diverse. You build them so each one holds the same mix the population has, which makes the clusters resemble each other. Then you draw whole clusters, and most clusters never appear in the sample at all.

Say it in one line and it stops being confusing: stratified means some of every group, cluster means all of a few groups. If a study reaches into every group it created, it is stratified. If it picks a handful of groups and sweeps them completely, it is cluster. The side-by-side version, including simple random and systematic sampling, is in simple random vs stratified sampling.

One more warning, because the vocabulary overlaps. Strata and blocks both create homogeneous groups, but blocking belongs to experiments rather than to sampling, and blocking vs stratifying separates the two.

A decision table for naming the sampling method

Use this table to name a method from a description. Read the middle two columns first, since how the population was divided and what the randomness picked settle almost every question.

Method	How the population is divided	What the randomness picks	Who gets measured
Simple random sample	Not divided at all	$n$ individuals from one list, where $n$ is the sample size	The individuals drawn
Stratified random sample	Into homogeneous strata	Individuals inside every stratum	A sample from each and every stratum
Cluster random sample	Into heterogeneous clusters	Whole clusters	Everyone in the selected clusters
Systematic random sample	Put in order, not grouped	A random start, then every $k$ -th unit, where $k$ is a fixed interval	The units that land on the interval

Two features give a cluster sample away. The study picks whole groups at random, and it then measures all of the individuals in the groups it picked. Wording like "every resident of the chosen buildings" or "all students at the 5 selected schools" is the tell. If instead you read "a random sample of residents from each building," the study went inside every group, so it is stratified.

Why cluster sampling exists: cost and logistics

Cluster sampling exists for reasons of money and logistics, not because it is statistically better. Two practical problems push researchers toward it.

The first is the list problem. A simple random sample or a stratified sample needs a list of individuals to draw from, and for many populations no such list exists. You probably cannot name all 25,000 households in a county, but you can name all 500 census blocks, and a list of clusters is often the only list anyone has.

The second is the travel problem. Those 25,000 households across 500 blocks work out to 50 households per block. Suppose you can afford 500 household interviews. Drawing 500 households at random from a whole county scatters your interviewers over hundreds of square miles, with a long drive between one doorstep and the next.

Sampling 10 whole blocks and knocking on every door instead puts each block's 50 interviews within a few hundred feet of each other, and $10 \times 50 = 500$ . Same 500 responses, a fraction of the fuel and the field hours.

The same logic runs through quality control. A plant that ships granola bars in cases of 24 can pull 8 cases at random and test every bar in them, because opening 8 cases is far easier than hunting down 192 individual bars scattered across a warehouse.

What cluster sampling costs you: extra sampling variability

The price of that convenience is extra sampling variability. Cluster sampling is a random method, so it does not build in bias the way a convenience sample does, but a cluster sample of 500 people usually carries more variability than a simple random sample of 500 people.

The reason is that neighbors resemble neighbors. Two households on the same block tend to be more alike on income, commute, and language than two households drawn at random from across the county, so the second interview on a block adds less new information than a fresh random household would. Your 500 responses then behave like a smaller number of independent observations, and the estimate wobbles more from sample to sample.

This is exactly why the definition insists that each cluster should mirror the population. The closer real clusters come to that ideal, the smaller the penalty. When clusters turn out to be homogeneous inside, meaning everyone in a cluster is alike, a cluster sample can lose a lot of precision, which is the mirror image of stratified sampling, where homogeneous groups are what you want.

One practical consequence: more clusters help more than more people per cluster. For the same 500 interviews, 20 blocks of 25 households beats 5 blocks of 100 households, because most of the variability comes from which clusters you happened to draw.

How to spot a cluster sample on the AP exam

Cluster sampling sits in Unit 1, topic 1.11 Random Sampling, inside a unit worth 20% to 30% of the multiple-choice section. Topic 1.11 has two learning objectives: identify a sampling method given a description of a study (skill 2.A), and justify the appropriateness of a sampling method (skill 2.B). Most questions are the first kind.

Work a naming question in two steps.

Did the study divide the population before sampling? If it did not, and every group of size $n$ was equally likely, it is a simple random sample. If it walked an ordered list at a fixed interval from a random start, it is systematic.
If it did divide, did it sample inside every group or take whole groups? Inside every group is stratified. Whole groups, everyone in them measured, is cluster.

A justification question wants a reason, not a label. Strong reasons for cluster sampling name the cost or the missing list: no roster of individuals exists, or the population is spread out and travel is expensive. Saying only that the method is random is not a justification, because all four methods are random.

To practice naming methods from scenarios, work the sampling methods practice set. The topic page is 1.11 Random Sampling, the one-sentence version is the cluster sample glossary entry, and the official course framework is at AP Central.

Name the sampling method in four plans

Name the sampling method in each plan. (a) A parks department numbers all 6,200 season pass holders and uses a random number generator to select 150 distinct numbers. (b) A cereal plant packs granola bars into cases of 24. An inspector randomly selects 8 cases from a day's production and tests every bar in those 8 cases. (c) A hospital splits its nurses into a day-shift group and a night-shift group, then takes a simple random sample of 20 nurses from each group. (d) A ferry counts passengers as they board, picks a random start among the first 12, then surveys every 12th passenger after that.

Plan (a): the department draws from one undivided list of pass holders, so any set of 150 holders is equally likely. No groups are formed, which makes this a simple random sample.
Plan (b): the population of bars is divided into cases, each case holds a mix of the day's output, the randomness picks whole cases, and every bar in a selected case is tested. Whole groups measured completely makes this a cluster random sample. The inspector ends up testing $8 \times 24 = 192$ bars.
Plan (c): the nurses are split into two groups that share an attribute (the shift they work), and a sample is taken inside each group and combined. Sampling within every group makes this a stratified random sample.
Plan (d): the passengers are in boarding order and the surveyor takes every 12th one from a random start, which is a fixed periodic interval, so this is a systematic random sample.

(a) simple random sample, (b) cluster random sample, (c) stratified random sample, (d) systematic random sample.

Cluster or stratified? Two plans for the same county

A county has 4,800 households spread across 60 neighborhoods of 80 households each, and each neighborhood contains a mix of household types. The county wants to estimate the mean number of pounds of recycling a household drops off per month. Plan A: randomly select 6 of the 60 neighborhoods and survey all 80 households in each selected neighborhood. Plan B: sort all 4,800 households into apartment, townhouse, and single-family, then take a simple random sample of 100 households from each type. Name each method, find each sample size, and give one reason a county might pick each plan.

Check the county's arithmetic first: $60 \times 80 = 4800$ households, so the neighborhoods account for the whole population with none left over.
Classify Plan A. The neighborhoods are groups that each hold a mix of household types (heterogeneous), the randomness selects whole neighborhoods, and every household inside a selected neighborhood is surveyed. That is a cluster random sample.
Find Plan A's sample size: $6 \times 80 = 480$ households. As a check on the design, $6 / 60 = 0.10$ of the neighborhoods gives $480 / 4800 = 0.10$ of the households, the same 10%.
Classify Plan B. The households are sorted into groups that share an attribute and are alike within it (homogeneous), and a sample is drawn inside every group. That is a stratified random sample.
Find Plan B's sample size: $3 \times 100 = 300$ households.
Give a reason for each. Plan A is cheaper and simpler in the field, since a crew works 6 neighborhoods instead of driving to households scattered across all 60, and it needs only a list of neighborhoods rather than a roster of 4,800 households. Plan B guarantees all three household types appear and, because the types are internally alike, tends to give a less variable estimate than a simple random sample of the same 300 households.

Plan A is a cluster random sample of 480 households, chosen for lower field cost and because only a list of neighborhoods is needed. Plan B is a stratified random sample of 300 households, chosen to guarantee every household type appears and to reduce variability.

Frequently asked questions

Is a cluster sample a simple random sample?

No. A simple random sample requires that every possible group of size n is equally likely. A cluster sample cannot produce most of those groups at all. Selecting a block brings every household on it into the sample, so a sample holding one household from block A and one from block B without their clustermates has probability zero, no matter how the draw goes. The randomness is real, but it acts on clusters rather than on individuals.

What is the one-line difference between cluster and stratified sampling?

Stratified sampling takes some of every group; cluster sampling takes all of a few groups. Strata are built to be similar inside, and clusters are built to be diverse inside so each one resembles the whole population.

Do all the clusters have to be the same size?

No. Real clusters such as city blocks, schools, or households vary in size, and that is normal. Unequal cluster sizes make the estimate a little harder to compute and add some variability, but they do not stop the method from being a cluster sample.

Is a cluster sample biased?

Not by design. Because the clusters are chosen at random, the method has no built-in tendency to run high or low, so its weakness is variability rather than bias. Bias would come from something else, such as choosing convenient blocks instead of random ones, or from people in the selected clusters refusing to respond.