What Is a Sampling Method?
A sampling method is the rule you use to decide which people from a larger group answer your survey. The method determines whether the answers of a few hundred people can stand in for everyone you care about.
You almost never survey everyone. You survey a sample and treat its answers as an estimate for the whole group, called the population. Choosing the sample is one step of a study; the full process is in how to do market research.
How you pick that sample matters more than how big it is. A thousand responses from the wrong people are worse than three hundred from the right ones.
There are two families of sampling methods. In probability sampling, everyone in the population has a known chance of being chosen. In non-probability sampling, they do not.
Key point: most business surveys use non-probability sampling. That is fine as long as you know which method you are using and what it can go wrong with.
Probability Sampling Methods
These methods need a list of the whole population, called a sampling frame. They are the only ones for which the standard margin of error formula strictly holds.
1. Simple Random Sampling
Every person on the list has the same chance of being picked.
Example: you have 20,000 customers and use a random number generator to pick 400 to invite.
Use it when you have a complete list and the population is fairly uniform.
2. Systematic Sampling
You pick every nth person from the list after a random start.
Example: you invite every 50th customer in your database, starting from a random position.
Use it when you want the simplicity of random sampling without software. Check that the list has no repeating pattern that lines up with your interval.
3. Stratified Sampling
You split the population into groups that matter, called strata, then sample at random inside each one.
Example: your customers are 70% on the basic plan and 30% on the premium plan. You draw 280 basic and 120 premium customers so the sample matches.
Use it when you must be sure small but important groups are represented, or when you want to compare groups.
4. Cluster Sampling
You split the population into natural clusters, pick some clusters at random, and survey people inside them.
Example: a retail chain picks 15 of its 200 stores at random and surveys shoppers in those stores only.
Use it when the population is spread out and surveying everywhere would cost too much. It is less precise than the methods above for the same number of responses.
Non-Probability Sampling Methods
These methods do not need a list of the population. They are faster and cheaper, and they are what most online research uses.
5. Convenience Sampling
You survey whoever is easiest to reach.
Example: you post a survey link on your company's social media accounts.
Use it when you need quick, directional feedback. Do not use it to estimate what a whole market thinks: your followers already like you.
6. Quota Sampling
You decide in advance how many people of each type you need, then collect responses until each quota is full.
Example: you want 400 responses that mirror the adult population: half women, half men, and set shares for each age band and region. Collection stops for each group when its quota fills.
Use it when you buy responses from an online panel. This is how panels deliver a sample that looks like a population.
7. Purposive Sampling
You choose respondents on purpose because they have a specific trait.
Example: you survey only people who bought a competing product in the last six months.
Use it when the study is about a narrow group. In survey practice this is done with screening questions, one of the types of survey questions most studies need.
8. Snowball Sampling
Respondents recruit other respondents.
Example: you survey ten hospital pharmacists and ask each to forward the survey to colleagues.
Use it when the population is small and hard to find. The sample will be people who know each other, so opinions may be more alike than in the wider group.
9. Voluntary Response Sampling
You make the survey open and people decide for themselves whether to answer.
Example: a "give us feedback" link at the bottom of your website.
Use it when you want to hear from anyone who cares enough to respond. Expect the very happy and the very unhappy to be over-represented.
Sampling Methods Compared
Method | Family | Cost | Main risk | Best for |
|---|---|---|---|---|
Simple random | Probability | Medium | Needs a complete list | Customer bases with a full contact list |
Systematic | Probability | Low | Patterns in the list | Large lists, quick setup |
Stratified | Probability | Medium | Needs data to form groups | Comparing groups reliably |
Cluster | Probability | Low | Less precise | Populations spread across locations |
Convenience | Non-probability | Very low | Not representative | Early, directional feedback |
Quota | Non-probability | Low to medium | Hidden differences inside quotas | Market surveys through a panel |
Purposive | Non-probability | Low | Depends on your screening | Narrow, defined audiences |
Snowball | Non-probability | Low | Like-minded respondents | Hard-to-reach groups |
Voluntary response | Non-probability | Very low | Strong opinions dominate | Open feedback |
Which Method Is Your Survey Really Using?
Most teams never choose a sampling method. The way they send the survey chooses it for them.
What you do | The method you are using |
|---|---|
Email every customer and analyze whoever replies | Voluntary response from your customer list |
Email a random 10% of customers | Simple random sampling, weakened by who chooses to reply |
Post the link on social media | Convenience sampling |
Show a pop-up to every 20th website visitor | Systematic sampling of visitors |
Buy panel responses with age, gender and region targets | Quota sampling |
Buy panel responses from category buyers only | Purposive sampling with a screener, usually combined with quotas |
Naming the method tells you what to be careful about. A customer email tells you about engaged customers, not about the market. A panel sample with quotas can describe a market, as long as the quotas cover the traits that matter.
Sampling Bias and Non-Response Bias
Two problems appear with every method.
Sampling bias means some kinds of people had no chance, or a smaller chance, of being included. A survey sent only to newsletter subscribers cannot tell you about people who have never heard of you.
Non-response bias means the people who answered differ from those who did not. If only your happiest customers reply, satisfaction looks higher than it is.
You reduce both the same way: define the population first, choose a method that can reach all of it, keep the survey short, and compare who answered against who you invited. Inside the survey, attention check questions remove the people who answered without reading.
Sample Size Comes After the Method
Only once you know who you are sampling does the number matter.
For a large population, about 385 responses give a margin of error of 5 points at 95% confidence. The sample size calculator gives the number for your own population and target, and the margin of error calculator tells you how precise a sample you already have is.
Those formulas assume random sampling. With a non-probability sample, treat the margin of error as a minimum: the real uncertainty is at least that large.
If you plan to compare groups, each one needs enough responses of its own. Cross tabulation covers how many.
How This Works in SegmentOS
You can send a survey to your own audience with a link, or buy responses from the built-in panel. Panel studies use targeting and quotas, so you set who qualifies and how many of each group you need. People who arrive after a quota is full are screened out and do not count against the responses you ordered.
Key Takeaways
Method before size: who you sample matters more than how many.
Two families: probability sampling gives everyone a known chance of selection, non-probability sampling does not.
Nine methods: simple random, systematic, stratified, cluster, convenience, quota, purposive, snowball and voluntary response.
Online panels use quota sampling, usually with screening questions.
The margin of error formula assumes random sampling. With other methods, treat it as a minimum.



