Stratified random sampling is a probability sampling method that divides a population into mutually exclusive subgroups called strata and randomly selects participants from each group. Researchers commonly create strata based on characteristics such as age, income, location, or customer type. This approach helps ensure important population subgroups are represented and can improve sampling precision.
Summary:
|
Stratified random sampling is a probability sampling method that divides a population into mutually exclusive subgroups called strata and randomly selects participants from each group. Researchers commonly create strata based on characteristics such as age, income, location, or customer type. This approach helps ensure that important subgroups are represented and can improve sampling precision.
A simple random sample can look fine on paper and still miss the point. By chance, it may include too few younger customers or almost none from a small but valuable segment. Teams that depend on consumer insights cannot afford that gap, because decisions made on a skewed sample are only as good as the sample itself.
This guide covers the definition, step-by-step process, formula with a worked example, types, comparisons with other methods, and when stratified sampling is the right choice.
What Is Stratified Random Sampling?
Stratified random sampling splits a population into distinct groups, called strata (singular: stratum), and then draws a random sample from within each group. The results are combined into one final sample.
Strata must be mutually exclusive, so every member of the population belongs to exactly one group. A customer cannot be both "new" and "returning" in the same stratification. Strata should also be collectively exhaustive, so no one is left out.
The point of stratification is representation. If a population has meaningful segments, such as age bands or customer tiers, stratifying makes sure each one shows up in the final sample in a planned way instead of by luck. It is one of several probability sampling designs, which all give every member of the population a known, non-zero chance of selection. For a broader view, see this overview of probability sampling.
How Does Stratified Random Sampling Work?
The process has five steps:
Define the population. Be specific about who is included, such as "active customers in the past 12 months."
Choose the stratification variable. Pick characteristics related to what you are measuring, such as age, income, region, or customer type.
Divide the population into strata. Make sure the groups do not overlap. Defining strata well often builds on demographic segmentation.
Decide how many to select from each stratum. Use proportional allocation or another approach (see below), then randomly sample within each group.
Combine the selected participants into the final research sample.
The random selection inside each stratum is what keeps the method a probability design. Without it, you would be hand-picking respondents, which introduces selection bias.
Stratified Random Sampling Formula
For proportionate allocation, the formula is:
nₕ = (Nₕ / N) × n
Where:
nₕ is the sample size for stratum h
Nₕ is the population size of stratum h
N is the total population size
n is the total desired sample size
Proportional allocation keeps each stratum's share of the sample consistent with its share of the population. If a group makes up 40% of the population, it makes up 40% of the sample.
Stratified Random Sampling Formula Example
Suppose a brand has 10,000 customers and wants a sample of 500. The customers fall into three segments:
Stratum | Population (Nₕ) | Share of population | Sample size (nₕ) |
Segment A | 4,000 | 40% | (4,000 / 10,000) × 500 = 200 |
Segment B | 3,500 | 35% | (3,500 / 10,000) × 500 = 175 |
Segment C | 2,500 | 25% | (2,500 / 10,000) × 500 = 125 |
Total | 10,000 | 100% | 500 |
The final sample of 500 mirrors the population: 40% from A, 35% from B, and 25% from C. Within each segment, the 200, 175, or 125 respondents are chosen randomly, for example with a random number generator applied to a customer list.
Types of Stratified Sampling
1. Proportionate Stratified Sampling
In proportionate stratified sampling, each stratum's share of the sample matches its share of the population, as in the example above. It is simple and works well when all segments are large enough to analyze.
2. Disproportionate Stratified Sampling
In disproportionate stratified sampling, some strata get larger or smaller allocations than their population share. This is useful when a small but strategically important group would otherwise yield too few responses to analyze.
Take the same 10,000 customers, split into Segment A (6,000), Segment B (3,500), and a high-value Segment C (500). A proportionate sample of 500 would give Segment C only 25 respondents, too few for reliable analysis. A disproportionate design might allocate 250, 150, and 100.
Stratum | Population | Proportionate sample | Disproportionate sample | Weight (Nₕ / nₕ) |
Segment A | 6,000 | 300 | 250 | 24.0 |
Segment B | 3,500 | 175 | 150 | 23.3 |
Segment C | 500 | 25 | 100 | 5.0 |
The trade-off is that the sample no longer mirrors the population, so overall results must be weighted. Each Segment C respondent represents 5 customers, while each Segment A respondent represents 24.
Large public surveys use this approach in practice. The CDC's National Health and Nutrition Examination Survey oversamples groups such as non-Hispanic White and other people at or below 130% of the poverty level and adults age 80 and older. Because the total sample size is fixed each year, adding more of one group means sampling fewer of another.
Stratified Sampling vs Simple Random and Cluster Sampling
Feature | Simple random | Stratified random | Cluster |
How selection works | Draws directly from the whole population | Divides into strata, then samples every stratum | Randomly selects whole groups (clusters) |
Subgroup representation | Left to chance | Guaranteed by design | Not guaranteed |
Best when | The population is fairly uniform | Subgroups differ on the outcome | The population is spread out and costly to reach |
Main drawback | May miss small subgroups | Needs accurate population data | Higher sampling error for the same size |
Simple random sampling selects directly from the overall population without dividing it first. It is easy to run but can underrepresent small groups.
Stratified sampling samples from every defined stratum. Use it when subgroup representation or comparison matters. Cluster sampling instead selects some groups and surveys within them, which saves cost but can reduce precision. This guide to cluster sampling explains when that trade-off pays off.
Other approaches differ in how they select. Stratified sampling is random within strata, while quota sampling fills set targets without random selection.
A systematic sample picks every nth member from a list.
When Should You Use Stratified Random Sampling?
Use it when:
The population has identifiable subgroups that differ on a characteristic relevant to your research.
You need reliable representation or comparisons across segments such as age, region, income, customer type, or market.
Subgroup differences can help improve the precision of your overall estimates.
You have accurate information on the population, such as a customer database with segment tags.
Skip it when the population is fairly homogeneous, you lack data to define strata, or speed matters more than precision. In those cases, convenience sampling is quicker but carries much greater bias risk.
When you deliberately want information-rich participants, purposive sampling is the better fit.
Examples of Stratified Random Sampling
Market research. A retailer wants to understand satisfaction across its customer base. It divides customers into loyalty tiers (new, regular, VIP) and regions, then randomly samples within each combination so every tier and market is represented.
Employee or education research. A company running an engagement survey divides employees by department and job level, then samples randomly from each. A university does the same by academic year so first-year and final-year students are both heard.
Healthcare and population research. A public health study divides a population into age groups, because disease risk varies by age, and samples within each.
Government and national surveys. The U.S. Current Population Survey covers about 60,000 eligible housing units each month, using a stratified two-stage sample design to produce reliable national and state estimates.
Education assessment. The National Assessment of Educational Progress targets 4,100 fourth graders and 4,200 eighth graders per jurisdiction, and schools are deeply stratified before sampling so the sample reflects the school population and minimizes sampling error.
Advantages and Limitations of Stratified Sampling
Advantages
Stronger subgroup representation. Each important group appears in the sample by design.
Potentially greater precision. When strata are internally similar but differ from each other, estimates can be more precise than a simple random sample of the same size.
Easier subgroup comparison. Planned allocations give you enough respondents per group to compare them.
Precision matters because sampling error shrinks with sample size. Pew Research Center notes that a survey's margin of error is driven primarily by sample size, and illustrates the risk of chance misses with 590 national polls in the 2012 election, where roughly 5% (about 30) would be expected to fall outside the margin by chance alone.
Understanding sampling error helps you decide how large each stratum needs to be.
Limitations
Requires accurate population information. You need to know each member's stratum before sampling.
Strata must be defined correctly. Poorly chosen or overlapping strata undermine the benefits.
More complex setup than simple random sampling, especially with several stratification variables.
Needs weighting when allocation is disproportionate.
Even a well-designed sample can go wrong in execution. This guide to types of sampling biases covers common pitfalls.
Applying Stratified Sampling in Consumer Research
In consumer research, strata often include demographics, customer segments, markets, product usage, or behavioral characteristics. Customer groups are a natural basis for strata, and this guide to customer segmentation explains how to build them.
The sampling source matters as much as the design. Pew Research Center compared six online surveys and found that opt-in samples had an average error of 5.8 points, about twice the 2.6-point average error of probability-based panels.
Stratifying an unreliable panel does not fix it, so recruit carefully, as explained in this guide to online panels for surveys.
Stratified designs let you compare responses across relevant consumer groups while keeping an appropriate overall sample structure. A brand might stratify by age group and region, then compare how each reacts to a new product concept. Many teams run such studies on a single consumer research platform so that sampling, fieldwork, and analysis stay connected.
This panel guide helps plan participant recruitment.
Representative samples are only half of the picture. Stated answers do not always match real reactions, so pairing them with behavioral and emotional signals adds depth. Decode by Entropik brings these together for consumer research, with verified capabilities that include:
Facial coding with 90%+ accuracy across 62 facial expressions.
Eye tracking with 96% accuracy.
Support for 70+ languages.
17 patents and 150+ global brands served.
Teams that track brand health across segments over time can pair stratified samples with brand tracking.
Facial coding relies on facial emotion AI to capture how each segment reacts moment by moment.
To compare tools for running segmented studies, see this roundup of the best consumer research platforms.
Frequently Asked Questions (FAQs)
1. What is stratified random sampling?
Stratified random sampling is a probability sampling method that divides a population into mutually exclusive strata and randomly selects participants from each one, so important subgroups are represented in the final sample.
2. What is the formula for stratified random sampling?
The proportionate allocation formula is nₕ = (Nₕ / N) × n, where nₕ is the sample size for a stratum, Nₕ is the stratum population, N is the total population, and n is the total sample size.
3. How do you calculate sample size for each stratum?
Divide the stratum's population by the total population and multiply by the total sample size. For a 4,000-person stratum in a 10,000-person population with a sample of 500, the result is (4,000 / 10,000) × 500 = 200.
4. What is an example of stratified sampling?
A retailer divides 10,000 customers into three segments of 4,000, 3,500, and 2,500, then randomly samples 200, 175, and 125 from them to reach a total of 500.
5. What is the difference between stratified and simple random sampling?
Simple random sampling selects directly from the whole population, leaving subgroup representation to chance. Stratified sampling divides the population into strata first and samples from each one.
6. What is the difference between proportionate and disproportionate stratified sampling?
Proportionate sampling matches each stratum's share of the sample to its share of the population. Disproportionate sampling gives some strata larger or smaller shares, usually to get enough responses from small but important groups, and requires weighting.
7. When should stratified random sampling be used?
Use it when the population has meaningful subgroups that differ on what you are measuring, when you need reliable subgroup comparisons, and when you have accurate data to define the strata.
8. What are the advantages and disadvantages of stratified sampling?
Advantages include stronger subgroup representation, potentially greater precision, and easier comparisons. Disadvantages include the need for accurate population data, careful strata definition, and a more complex setup.
9. Build Representative Samples, Then Go Deeper
Connect representative consumer samples with behavioral and emotional research across meaningful audience segments.
Explore Decode, or request a demo to see how 90%+ facial coding accuracy, 96% eye tracking accuracy, 62 facial expressions, 70+ supported languages, 17 patents, and 150+ global brands support clearer consumer research.
