Forms · Glossary

What is stratified sampling?

Stratified sampling divides a population into separate groups, called strata, that share a characteristic such as region, age band or job type, and then draws a random sample from each group independently. It guarantees every group is represented and usually gives more precise estimates than a simple random sample of the same size.

National statistics offices use it widely in household and business surveys, and the idea scales down well. A staff survey or a customer study that must hear from small groups is a natural fit, as long as you know which group every person belongs to before you sample.

Nuwan Madhusanka · Co-founder

4 min read · Published 21 October 2026

A stratified sample of 400 from 4,000 fictional staff, two ways
Stratum	Staff	Proportionate sample	Disproportionate sample and weight
Retail	2,400	240	160, weight 15
Warehouse	1,200	120	160, weight 7.5
Head office	400	40	80, weight 5
Total	4,000	400	400, weights restore the true mix

How it works

Statistics Canada describes the method in two steps: divide the population into homogeneous, mutually exclusive strata, then select an independent sample from each, using any random method within each stratum. The stratifying variable has to be known for everyone on the sampling frame before sampling, such as province, age, job type or business size. The ABS explains why it helps: by sampling within each stratum, the variation between strata is removed from the sampling error, leaving only the variation within them, so the most efficient strata are as different from each other as possible and as similar as possible inside.

Proportionate or disproportionate

In proportionate allocation each stratum gets the same share of the sample as it has of the population, as in the middle column of the table, so the sample mirrors the population and results can be combined directly. In disproportionate allocation small strata get extra places so they can be analysed on their own, as in the right column, where head office is sampled at 1 in 5 instead of 1 in 10. Results for the whole population then need weights, the inverse of each stratum's sampling fraction, so that over sampled groups do not count for more than their real share.

When to use it

Stratify when you need reliable results for particular groups, when groups differ a lot on what you are measuring, or when a simple random sample might by chance under represent a small but important group. A council surveying residents by ward, a hospital surveying patients by department or a franchise surveying customers by store all benefit. It is not worth the effort when the frame lacks the information to place people into strata, or when the groups are so alike that stratifying gains little.

Common mistakes

Confusing stratified sampling with quota sampling, which fills group targets with whoever is available rather than at random. Forgetting to weight a disproportionate sample, so small groups distort the totals. Using strata with too few people to sample. Stratifying on a variable recorded wrongly in the frame, such as an out of date department. And reporting results for a stratum with so few responses that a couple of answers swing the percentage.

Running a stratified survey with forms

The form builder does not draw samples or allocate strata, so the selection happens before the form is sent: draw the random sample within each group from your own list, then send the link to those people. A login required form can limit responses to the addresses you selected, so the link cannot spread beyond the sample. Ask the stratum, such as the work area, as a single choice question so it becomes its own column in the CSV export and you can apply weights afterwards in a spreadsheet. The form has no hidden fields, so the stratum has to be asked, not passed in the link.

Questions people ask

What is the difference between stratified and cluster sampling?

In stratified sampling you sample individuals from every group. In cluster sampling you randomly choose some groups, such as schools or stores, and survey everyone or a sample within them. Stratifying improves precision; clustering cuts travel and cost but usually reduces precision, because people in the same cluster tend to be alike.

Is stratified sampling a probability method?

Yes, as long as the sample within each stratum is drawn at random and every person has a known chance of selection. That is what lets you calculate margins of error. Filling group targets with volunteers instead is quota sampling, a non probability method that looks similar but lacks that guarantee.

How many strata should I use?

Enough to separate groups that really differ on what you measure, and few enough that each stratum still gets a workable sample. A handful of strata is common in small surveys. Every extra stratum needs the stratifying information for everyone on the list and adds weighting work.

Do I need to weight a proportionate stratified sample?

Not to correct the design, because each stratum already has its population share. You may still weight for non response if some strata answered at much lower rates than others, which would shift the mix of the responses away from the population even though the sample itself was balanced.

Can I stratify after collecting the data?

Grouping and weighting responses afterwards is called post stratification, and it is common. It adjusts the mix of the responses to known population totals but does not give the precision of designing strata in advance, and it cannot help a group that was barely reached in the first place.

Make one with forms

The button opens the generator with this use case already described. Change the wording to match your own.

Create a form with OneCraft

Sources

Written and checked by the OneCraft team. Last checked 21 October 2026.