Stratified Sampling: How to Get Precise, Representative Results (2026)

Stratified sampling in 30 seconds

Stratified sampling is a probability method that divides your population into mutually exclusive subgroups — called strata — and then draws a random sample from within each one. Instead of hoping a single random draw happens to represent every segment, you guarantee representation by sampling each segment deliberately: every region, every plan tier, every persona gets its fair share of the sample.

Why it matters: for the same number of participants, stratified sampling generally produces more precise estimates than simple random sampling, and it ensures small-but-important subgroups are never left out by the luck of the draw (Statistics By Jim). The cost is that you must know which stratum each person belongs to before you sample. It is the method of choice whenever a population contains meaningful subgroups that might behave differently — and modern research platforms like Koji apply the same logic through per-segment quotas and screener routing, so you get stratified rigor without a spreadsheet full of manual math.

---

What is stratified sampling?

Stratified random sampling is a probability technique in which the population is first divided into mutually exclusive, internally homogeneous subgroups (strata), and a random sample is then drawn from each stratum — either proportionately or disproportionately.

Two properties define a good stratum:

• Mutually exclusive: every member of the population belongs to exactly one stratum. No overlap.
• Internally homogeneous, externally heterogeneous: members within a stratum are similar on the stratifying variable, while the strata themselves differ from one another. You stratify by things like region, age band, gender, industry, or customer plan tier — variables you expect to correlate with the answers you care about.

Because it uses random selection within each stratum, stratified sampling is a probability method — unlike purposive, snowball, or convenience sampling, it supports margins of error and generalization. For the full landscape, see qualitative research sampling methods.

---

Proportionate vs disproportionate stratification

The key decision in stratified sampling is how to allocate your total sample across the strata.

Proportionate stratified sampling

Each stratum contributes to the sample in the same proportion it holds in the population. If history majors are 40% of the population and English majors are 60%, the final sample mirrors those percentages. Proportionate allocation gives the best overall population estimate because the sample is a faithful miniature of the whole.

Disproportionate stratified sampling

You deliberately break from population proportions to oversample smaller strata you want to study precisely. Classic example: if Indigenous participants are only 20% of the population but you need a reliable read on their perspective, in a sample of 100 you might take 50 Indigenous and 50 non-Indigenous participants. This yields precise subgroup estimates at the cost of some whole-population accuracy (QuestionPro).

Rule of thumb: use proportionate when you care most about the total population, and disproportionate when you need to compare subgroups or study a small group in depth.

---

How to calculate a stratified sample

The proportionate calculation is simple arithmetic — take each stratum’s share of the population and multiply by your total sample size.

Suppose your total sample size is 200 and your population breaks down like this:

Then draw a simple random sample of the required size from within each stratum. For disproportionate allocation, you override these numbers deliberately — for instance, bumping Enterprise from 30 to 60 because that segment drives most revenue and you need tighter estimates for it.

To choose your overall sample size before allocating it, start with the survey sample size guide.

---

Why stratified sampling beats simple random sampling

The central advantage is precision. In a simple random sample, chance alone might hand you too few enterprise customers or too many free-tier users, skewing your estimates. Stratified sampling removes that risk by fixing each subgroup’s representation in advance. The result: for the same sample size, stratified sampling generally delivers lower sampling error and more precise estimates than simple random sampling — the gains are largest when the strata genuinely differ from one another.

There are two more benefits:

• Guaranteed subgroup representation. Small but critical segments (a tiny enterprise tier, a rare persona) will appear in the sample rather than being missed by chance.
• Segment-level analysis. Because each stratum is sampled properly, you can compare segments with confidence — exactly what you need for customer segmentation research.

The main requirement is knowledge: you must be able to assign each population member to a stratum before sampling, which means you need that attribute in your sampling frame.

---

How to run a stratified study: step by step

1. Choose a stratifying variable that matters. Stratify by the characteristic most likely to drive different answers — plan tier, region, role, or usage level. Stratifying on an irrelevant variable adds work without adding precision.

2. Define exhaustive, mutually exclusive strata. Every person must fit exactly one stratum, and the strata together must cover the whole population.

3. Determine your total sample size. Base it on your desired confidence and margin of error.

4. Allocate — proportionate or disproportionate. Apply the calculation above, choosing allocation based on whether you prioritize the whole population or subgroup comparisons.

5. Randomly sample within each stratum. This random step inside each group is what keeps the method a probability sample.

6. Weight when analyzing disproportionate samples. If you oversampled a small stratum, apply weights to recover accurate whole-population estimates.

---

The modern approach: stratification through quotas and AI

Traditional stratified sampling assumes you have a clean sampling frame with every person tagged by stratum, plus the time to draw and manage separate random samples from each group. In real product and UX research, that frame is often incomplete, and the manual allocation math is error-prone.

AI-native platforms like Koji operationalize stratified logic without the overhead:

• Quotas enforce your strata automatically. Set a target count per segment — 30 enterprise, 50 mid-market, 80 small business — and Koji fills each quota, closing a stratum once it is complete so no group over- or under-fills.
• Screeners route participants into the right stratum. Screener questions identify each participant’s segment up front and direct them into the correct quota, replacing the sampling frame you may not have.
• Structured questions keep cross-segment comparison clean. Koji’s six structured question types (open_ended, scale, single_choice, multiple_choice, ranking, yes_no) ensure every stratum answers identically-structured questions, so segment comparisons are apples-to-apples — see the structured questions guide.
• Reporting breaks results down by stratum. Instead of manually re-weighting spreadsheets, you get automatic per-segment analysis and can compare strata side by side in minutes.

The methodological principle is unchanged — divide the population, guarantee representation, sample within groups. What changes is the effort: stratified rigor becomes a few quota settings rather than a statistician’s afternoon. You do not need a background in survey statistics to run a properly stratified study; you need to know your segments and set your quotas.

---

Common stratified sampling mistakes

Even a sound method fails when applied carelessly. Watch for these:

• Stratifying on an irrelevant variable. If the stratifying characteristic does not correlate with your outcome, you add complexity without gaining precision. Stratify by what actually drives different answers.
• Overlapping or incomplete strata. Strata must be mutually exclusive and collectively exhaustive. If a person could fall into two strata, or none, your allocation breaks.
• Too many strata, too few people. Slicing a small sample into a dozen strata leaves each one too thin to analyze. Keep strata few and meaningful.
• Forgetting to weight a disproportionate sample. If you oversampled a small segment, apply weights before reporting whole-population figures — otherwise the small group is over-counted.
• Confusing stratified with quota sampling. True stratified sampling uses random selection within each stratum. If you fill strata with whoever is convenient, you have quota sampling — a useful non-probability cousin, but not equivalent.

Avoiding these keeps stratification delivering its core promise: precision and guaranteed representation of every subgroup.

Key takeaways

• Stratified sampling divides a population into mutually exclusive strata and randomly samples within each — a probability method.
• Proportionate allocation mirrors population shares for the best whole-population estimate; disproportionate oversamples small groups for precise subgroup reads.
• Calculate proportionate sizes by multiplying each stratum’s share by the total sample size.
• For the same sample size, stratified sampling is generally more precise than simple random sampling and guarantees subgroup representation.
• Koji applies stratified logic through per-segment quotas and screener routing, delivering the rigor without the manual math.

---

Related Resources

• Structured Questions Guide — the six question types that keep every stratum comparable
• Purposive Sampling Guide — targeted non-probability selection
• Snowball Sampling Guide — reaching hard-to-find populations
• Qualitative Research Sampling Methods — the full map of sampling options
• Customer Segmentation Research — where stratified thinking pays off
• Survey Sample Size Guide — set your total sample before allocating it