What is Multistage Sampling? Definition, Methods, Examples

Key Takeaways:

• Multistage sampling selects a sample in 2 or more successive stages, moving from large clusters (such as districts) down to individual units (such as people).
• It is the method of choice for large, geographically dispersed populations where no complete list of individuals exists and where budgets limit travel.
• Every stage must use a probability method; analysis must apply sampling weights and account for the design effect.
• New researchers get better precision by sampling more clusters with fewer units per cluster, and by documenting every stage of selection.

Glossary of Key Terms

What Is Multistage Sampling?

Multistage sampling is a probability sampling method in which researchers draw a sample in 2 or more successive stages, first selecting large clusters and then selecting progressively smaller units within them. Only the units chosen at the final stage provide data.

The method is common in national health surveys, education studies, election polling, and market research. It solves 2 practical problems at once: the absence of a complete list of individuals, and the high cost of collecting data across a wide area.

A typical design looks like this:

• Stage 1: Select districts from all districts in a country.
• Stage 2: Select villages or wards within the chosen districts.
• Stage 3: Select households within the chosen villages.
• Stage 4: Select 1 adult within each chosen household.

How Does Multistage Sampling Work?

It works by dividing the population into large clusters, sampling some of those clusters, then sampling smaller units inside the selected clusters, repeating until individual respondents are reached. A frame is needed only for selected clusters at each stage.

Follow these 6 steps:

Types of Multistage Sampling Designs

Researchers combine sampling methods across stages to fit their frames and budgets. The 4 most common designs are summarized below.

When Should You Use Multistage Sampling?

Use multistage sampling when the population is large and geographically spread out, no complete list of individuals exists, and budget or time constraints rule out simple random sampling across the whole area.

It is a strong fit in these situations:

• National or state-level surveys where travel to every location is impossible.
• Studies where frames exist for clusters (districts, schools, hospitals) but not for individuals.
• Face-to-face fieldwork, where concentrating interviews in selected clusters cuts travel costs.
• Studies of institutions and the people inside them, such as teachers within schools.

Avoid it when the population is small and listed, when you can survey everyone cheaply online, or when extreme precision is required and cluster effects would inflate the error too much.

How Is Multistage Sampling Different from Cluster and Stratified Sampling?

Multistage sampling selects clusters and then samples units within them; single-stage cluster sampling surveys every unit inside selected clusters; stratified sampling draws units from every stratum rather than from a subset of groups.

Advantages and Disadvantages of Multistage Sampling

Worked Example: A 4-Stage National Education Survey

Suppose a ministry wants to estimate the average reading score of grade 10 students nationwide. There is no national list of students, but there are lists of districts, of schools within districts, and of class sections within schools.

Each student’s overall selection probability is the product of the 4 stage probabilities, and the student’s weight is the inverse of that product. Because PPS was used at stages 1 and 2 with fixed takes later, weights stay roughly equal, which keeps the design efficient.

Tips for New Researchers

Tip 1: Define Your Target Population and Units Before Touching Any List

Write 1 sentence that states who is in the population, where, and when. Then name the unit at every stage. Vague definitions cause coverage errors that no analysis can repair.

Example: “All women aged 15-49 living in private households in Maharashtra in 2026” leads naturally to districts, then villages or wards, then households, then 1 eligible woman per household.

Tip 2: Choose Stage Units That Have Reliable, Recent Frames

Your design is only as good as the lists behind it. Prefer administrative units with official, dated lists: census enumeration areas, registered schools, licensed clinics. Check each frame for 3 problems: missing units, duplicate units, and outdated size measures.

• Ask when the frame was last updated; anything older than 5-7 years deserves field verification.
• Reconcile overlapping lists (for example, public and private school registers) before selection.
• If no household list exists in a selected village, plan a quick field listing operation before stage 3.

Tip 3: How Many Stages Should You Use?

Use the fewest stages that solve your frame and cost problems: 2 stages are enough for most student projects, and national surveys rarely need more than 4. Every extra stage adds clustering, paperwork, and room for error.

Example: to survey nurses, sampling hospitals and then nurses (2 stages) beats sampling states, then cities, then hospitals, then wards, then nurses (5 stages), unless no hospital list exists without going through states first.

Tip 4: Use Probability Proportional to Size at the Early Stages

When clusters vary widely in size, select PSUs with PPS and then take a fixed number of units in each. This gives every individual a roughly equal overall probability, keeps interviewer workloads uniform, and simplifies weights.

Example: with PPS, a district of 500,000 students is 10 times more likely to be selected than a district of 50,000; taking exactly 100 students in each selected district then balances the probabilities.

Tip 5: Should You Sample More Clusters or More Units per Cluster?

Sample more clusters with fewer units in each; for a fixed total sample, this lowers the design effect and yields more precise estimates. People in the same cluster resemble each other, so the 50th interview in 1 village adds little new information.

• Rule of thumb: 25-35 units per cluster is a common ceiling in household surveys; 10-15 is safer when the ICC is high.
• Example: 100 schools x 10 students usually beats 20 schools x 50 students, even though both give 1,000 students.
• Balance this against cost: if reaching a new cluster is very expensive, a moderately larger take per cluster is a fair compromise.

Tip 6: Calculate, Check, and Apply Sampling Weights

Record the selection probability at every stage for every sampled unit, at the moment of selection. The base weight is 1 divided by the product of those probabilities. Adjust weights for nonresponse, then check them: extreme weights signal a selection or recording error.

• Keep a selection log: frame size, sample size, method, and random seed for each stage.
• Trim or investigate weights more than 4-5 times the median.
• Analyze with survey-aware tools: the survey package in R, svyset in Stata, or complex samples modules in SPSS.

Tip 7: Plan Sample Size with the Design Effect, Not SRS Formulas

An SRS sample size formula will underestimate what a multistage design needs. Multiply the SRS size by the expected design effect: DEFF = 1 + (m – 1) x ICC, where m is the number of units per cluster.

Example: an SRS calculation says 400 respondents. With 20 respondents per cluster and an ICC of 0.05, DEFF = 1 + 19 x 0.05 = 1.95, so you need about 780 respondents, spread across roughly 39 clusters.

Tip 8: Pilot the Full Selection Procedure Before Fieldwork

Run the entire chain, from stage 1 selection to the final interview, in 2-3 clusters. Pilots reveal missing village lists, gatekeeper refusals at schools, unclear household definitions, and timing problems while they are still cheap to fix.

• Test the household or respondent selection rule (for example, the Kish grid or last-birthday method) with real interviewers.
• Time each stage; travel between clusters often costs more than the interviews themselves.
• Revise the written protocol after the pilot and version it clearly.

Tip 9: Document Every Stage for Transparency and Reproducibility

Reviewers, supervisors, and future users of your data will ask exactly how the sample was drawn. Write a sampling annex that states, for each stage: the frame and its date, the stratification, the method, the sample size, and the response rate.

Example: major surveys such as the Demographic and Health Surveys publish stage-by-stage sampling appendices; using their structure as a template is a fast way for a new researcher to look professional and be reproducible.

Common Mistakes to Avoid

• Selecting clusters by convenience (the nearest 5 schools) and calling the design multistage; without probability selection it is not.
• Ignoring weights in analysis, which biases estimates when selection probabilities differ across units.
• Reporting confidence intervals from SRS formulas, which understates the true error under clustering.
• Letting interviewers substitute a refusing household with the neighbor, which quietly destroys the probability design; use documented nonresponse procedures instead.
• Using outdated cluster size measures for PPS, which distorts selection probabilities.
• Forgetting eligibility screening at the final stage, such as interviewing any adult instead of the randomly selected one.

Frequently Asked Questions

What is the difference between multistage sampling and cluster sampling?

In single-stage cluster sampling, every unit inside the selected clusters is surveyed. In multistage sampling, only a sample of units inside each selected cluster is surveyed, often through several further stages. Multistage sampling is best understood as cluster sampling repeated at 2 or more levels.

Is multistage sampling a probability sampling method?

Yes, provided a probability method (simple random, systematic, or PPS) is used at every stage and selection probabilities are recorded. If any stage uses convenience or judgment selection, the design loses its probability status and its estimates cannot be generalized statistically.

What is an example of multistage sampling in real life?

The Demographic and Health Surveys and most national labor force surveys use it. A typical example: select census enumeration areas with PPS, list households in each selected area, select 25 households per area systematically, then randomly select 1 eligible adult per household.

How do you calculate sample size for multistage sampling?

Start with the standard SRS sample size for your desired margin of error, then multiply it by the design effect, DEFF = 1 + (m – 1) x ICC, where m is the planned number of respondents per cluster. Finally, inflate for expected nonresponse and divide by m to get the number of clusters.

What are the advantages and disadvantages of multistage sampling?

Its main advantages are lower cost, feasibility without a full list of individuals, and flexibility across stages. Its main disadvantages are reduced precision from clustering, more complex weighting and variance estimation, and vulnerability to poor frames at any stage.

When is multistage sampling used in research?

It is used when populations are large and dispersed, when frames exist only for groups such as districts or schools, and when face-to-face data collection makes travel a major cost. Health, education, agriculture, and opinion research at national scale rely on it heavily.

Can multistage sampling be combined with stratified sampling?

Yes, and it usually is. Researchers stratify the stage 1 units (for example, urban vs rural districts, or regions) and then run the multistage selection independently within each stratum. Stratification guarantees subgroup representation and typically improves precision at no extra fieldwork cost.

What is probability proportional to size (PPS) sampling in a multistage design?

PPS gives each cluster a selection chance proportional to its size, such as its population or enrollment. Paired with a fixed number of final units per cluster, it produces roughly equal overall probabilities for individuals, near-equal weights, and uniform interviewer workloads.