What is a research sample? Understand the concept and main types
18 min read
What is a research sample? This is one of the main questions for those who want to understand how companies, institutes, and researchers can make decisions based on the opinion of only a part of the population.
The concept may seem complex, but it is present in daily life and helps researchers explain everything from electoral forecasts to the development of new products and services.
In this complete article, we will talk about everything you need to know about sampling. Grab your coffee and let's go!
What is a sample in a survey?
To answer the central question, what is a sample in a survey?, we can think of a trip to the supermarket. Imagine you are in the fruit section and see a huge box of strawberries.
You don't need to eat the whole box to know if the strawberries are sweet or sour. Just pick one or two, taste them, and from that small experience, you conclude the quality of the entire box.
In statistics and scientific methodology, a sample works in exactly the same way. A sample is a small group (or subset) of individuals, objects, or data taken from a much larger group. The idea is to study this small portion to understand the behavior, opinions, or characteristics of the entire group.
When a research institute wants to know the opinion of Brazilians about a new bill, it is impossible (and very expensive) to knock on every citizen's door.
Instead, they select a sample, for example, 2.000 people spread across different states, ages, and social classes, and ask them questions. If the sample is well chosen, the answers will accurately reflect what the public thinks.
What does sampling mean?
While the “sample” is the group of people or elements you will research (the strawberries you tasted), sampling is the process or technique used to choose who will be part of that group.
Sampling is the action plan. It is the stage where you define the rules of the game: “Will I choose people on the street randomly?”, “Will I send a questionnaire only to customers who bought in the last month?”, “Will I select half men and half women?”.
There are several sampling techniques, and choosing the right technique ensures that your research will yield a result that matches reality, and not a result distorted by your own preferences or by pure chance. In summary: sample is the “who”, and sampling is the “how”.
What is the difference between sample and population?
This is one of the most common questions, but the difference is quite easy to grasp. In research language, these two terms always go hand in hand.
Population (or Universe): It is the complete group. It is everyone who fits the characteristics of what you want to study.
If you want to know what students at a specific school think about the school lunch, your population is all students enrolled in that school. If you want to know the satisfaction level of your customers, the population is all active customers of your company.
Sample: It is the slice of that population that you will actually investigate. It is the reduced group that will answer your questions.
| Characteristic | Population | Sample |
| Definition | The total and complete group of elements. | A representative part taken from the total group. |
| Size | Generally very large, difficult to measure. | Smaller, practical, and manageable. |
| Cost and Time | High cost and very time-consuming to research. | Cheaper and much faster. |
| Result generates | Parameters (absolute data). | Statistics (estimates about the population). |
Why is the sample important in a survey?
You might be asking yourself: “If the population is the whole group, why not survey everyone at once to get the exact result?”. The answer comes down to three words: time, money, and feasibility.
Cost-benefit
Surveying an entire population (what we call a census) is a very expensive process. IBGE, for example, spends billions of reais and mobilizes thousands of census takers to conduct the Demographic Census of Brazil.
Most companies and academic researchers do not have that budget. The sample allows you to reach conclusions very close to reality by spending a fraction of the cost.
Speed
Currently, information ages quickly. If it takes you six months to interview all your customers about a new product, by the time you finish, the market has already changed. Using a sample allows you to collect data and make decisions in a matter of days or weeks.
Practical feasibility
In many cases, it is literally impossible to access the entire population. How would you test the water quality of an entire river? Or how would you find out the opinion of all the people in the world who like coffee? In these cases, the sample is not just a smart choice, it is the only possible option.
How is a sample chosen?
The choice of a sample is not made haphazardly, like pulling names from a hat (unless that is the chosen mathematical method!). The selection process requires planning and follows some fundamental steps to ensure that the final data is useful.
Define the target population
Before choosing the sample, you need to know who your universe is. Who do you want to understand? Women aged 20 to 30 who live in São Paulo? Men who bought sports cars in the last year? IT professionals who work from home? The clearer your population is, the easier it will be to choose the sample.
Choose the sampling method
Here you decide whether to use a probabilistic method (where luck and mathematics rule, as in a lottery) or a non-probabilistic method (where the researcher chooses participants for convenience or profile). We will detail these types below.
Calculate the sample size
How many people do I need to hear from? 100? 500? 2.000? This number is calculated based on the size of your population, the margin of error you accept, and the confidence level you want to give to your research.
Collect the data
With the plan ready, it's time to go to the field (or to the internet) and apply questionnaires, conduct interviews, or observe the behavior of the people who fell into your sample.
What are the main types of sampling?
The way you choose who will participate in your research changes everything. Basically, the world of sampling is divided into two large families: probabilistic sampling and non-probabilistic sampling. Let's understand how they work and learn about the most common types within each.
Probabilistic sampling
Here, the golden rule is randomness. In probabilistic sampling, every single member of the population has a known (and greater than zero) chance of being drawn to participate in the research. It is the safest method to avoid biases and ensure that the result represents the whole.
Simple random sampling
Simple random sampling is the famous lottery. If you have a list of 1.000 customers and need 100 for the research, you put the 1.000 names into a system (or a giant hat) and draw 100. Everyone has exactly the same chance of winning.
Systematic sampling
It is similar to a lottery, but with a mathematical rule. Imagine you have a list of patients. You decide that you will choose patient number 1, then skip 10 names and pick patient 11, skip another 10 and pick 21, and so on. It is fast and maintains randomness.
Stratified sampling
Used when the population is very divided into different groups (strata). For example, if your company has 60% female customers and 40% male customers, your sample needs to maintain this exact proportion. You divide the list by gender and conduct a random draw within each group to ensure that the sample is a mini-mirror of reality.
Cluster sampling
Used when the population is geographically dispersed. Instead of randomly drawing people across Brazil (which would cost a fortune in travel), the researcher draws some cities (clusters) and interviews people within those drawn cities.
Non-probabilistic sampling
In this group, selection is not done by drawing lots. The researcher chooses participants based on practical criteria, convenience, or personal judgment.
It is widely used in qualitative research, product testing, or when there is no complete list of the population. The point of attention here is that the results cannot be taken as an absolute truth for the entire population, but they provide excellent directions.
Convenience sampling
The name says it all. The researcher approaches people who are easiest to reach. You know that person at the mall entrance with a clipboard asking if you have a minute? That's convenience sampling. It's quick and cheap, but not very statistically precise.
Quota sampling
It is the “non-probabilistic” version of stratified sampling. The researcher defines that they need to interview, say, 50 young people and 50 elderly people. They go to the street (or internet) and interview whoever appears, until they meet the quota for each group. Very common in street electoral surveys.
Purposive sampling (or Judgment sampling)
The researcher uses their own knowledge to handpick who will participate, seeking “experts” or very specific profiles. For example, if the research is about the luxury market, the researcher will intentionally invite people they know consume designer products.
Snowball sampling
Used when the target audience is very difficult to find (such as rare coin collectors or people with a very specific disease). The researcher finds a person with this profile, interviews them, and asks them to refer a friend who also has the profile. One refers the other, and the sample grows like a snowball.
What is a representative sample?
You know when a survey comes out saying that “80% of Brazilians prefer the beach to the countryside,” and you think: “But no one asked me anything, and all my friends prefer the countryside!”.
This happens because the research was done with a representative sample. A representative sample is one that can reflect, in miniature, all the important characteristics of the total population. It is a faithful portrait, but in reduced size.
For a sample to be considered representative, it needs to respect the diversity of the larger group. If a city's population has 52% women, 30% young people, and 15% people with higher education, the sample needs to have, proportionally, 52% women, 30% young people, and 15% people with higher education.
If the researcher goes in front of a university to interview people, the sample will have many young people and many people with higher education, ceasing to be representative of the entire city. It would become a “biased” (distorted) sample.
How to know if a sample is reliable?
The reliability of a sample depends on two main pillars: how people were chosen and how many people were chosen. To know if you can trust the results of a survey, ask the following questions:
Was the selection method adequate?
If the research wants to talk about all of Brazil, but only interviewed people in downtown São Paulo on a Tuesday afternoon, it is not reliable. The probabilistic method (with random drawing) is always the one that brings the greatest mathematical reliability.
Is the sample size sufficient?
Interviewing 10 people to try to understand the behavior of 1 million will not work. The sample needs to have a statistically valid size.
3.What is the margin of error? Every serious research discloses its margin of error. If a survey says that a candidate has 30% of voting intentions with a margin of error of 2% plus or minus, it means that the real value is between 28% and 32%. The smaller the margin of error, the more reliable the sample.
What is the confidence level?
Generally set at 95%. This means that if the same research were repeated 100 times with different samples, in 95 of them the result would fall within the predicted margin of error.
How many people are needed for a survey to be valid?
“How many responses do I need?” This is every researcher's million-dollar question. And the answer is: it depends on the math. There is no universal magic number, but there is a “Sample Size Calculation” formula. The ideal number depends on three factors:
Population size
How many people are there in total? Interestingly, for gigantic populations (above 100 thousand people), the sample size does not need to grow proportionally. A sample of 400 to 1.000 people is usually sufficient to represent millions, if well selected.
Desired margin of error
How much error do you accept? If you want a super precise survey (margin of error of 1%), you will need a huge sample (thousands of people). If you accept a margin of error of 5% (the market standard), a sample of about 380 to 400 people usually works for large populations.
Confidence Level
The market standard is 95%. If you want to increase to 99% statistical certainty, you will have to significantly increase the number of respondents.
Rule of thumb: For most market and opinion surveys aimed at the general public, a sample of between 300 and 500 complete responses is considered an excellent starting point for statistically valid data with an acceptable margin of error (around 4% to 5%).
Does every survey need a sample?
Believe it or not, the answer is no. Not every survey needs a sample.
There are surveys called Census surveys. In this type of study, the researcher goes after 100% of the population. There is no selection, no drawing. Everyone is heard. The census is used in two situations:
When the population is very small:
If you have a team of 15 employees and want to know where they want to have their year-end party, it doesn't make sense to calculate a sample. You send the question to all 15 of them.
When the government requires absolute data:
The IBGE Demographic Census is the biggest example. The government needs to know exactly how many people live in each municipality to allocate funds, so they try to count all inhabitants of the country.
However, as we have seen, censuses are expensive, time-consuming, and difficult to manage. Therefore, 99% of market, academic, and opinion surveys choose to work with sampling.
How does the sample influence survey results?
The sample is the foundation of the building. If the foundation is crooked, the building falls. The way you build your sample directly dictates the quality of your results.
If you choose a large enough and highly representative sample, your results will be a mirror of reality. You will be able to make million-dollar business decisions, launch products, or create public policies with the certainty that you are meeting the will of the majority.
On the other hand, if your sample is poorly chosen, it will generate what we call bias. Bias is a distortion. If you want to launch an app for seniors, but your sample only has 20-year-olds testing the interface, the survey result will say the app is perfect. When it is launched to the real public (seniors), it will be a failure.
The wrong sample creates the illusion that you are on the right track, when in fact you are heading for the abyss.
What errors can occur in sample selection?
Even with a lot of planning, errors happen. Knowing the main ones is the best way to avoid them:
Coverage error
Occurs when the list used to draw the sample does not include the entire population. Example: conducting a survey by landline phone nowadays. You automatically exclude millions of people who only use cell phones, biasing the survey towards an older audience.
Non-response error
You selected the right people, but a specific group refuses to respond. If you conduct an employee climate survey and dissatisfied employees are afraid to respond and be fired, your final sample will only have happy employees, masking the real problem.
Insufficient Size
Wanting to save money and interview too few people, resulting in a margin of error so large (e.g., 15%) that the data becomes useless.
Examples of samples in market research
Market research is the engine of modern companies. See how sampling is used in practice:
New product testing
A soda brand wants to launch a new lemon flavor. The population is all soda consumers in the country. The chosen sample consists of 600 people selected in supermarkets in 4 different capitals to conduct a blind taste test.
Customer Satisfaction (NPS)
An e-commerce with 50.000 active customers sends a monthly satisfaction questionnaire to a random sample of 1.000 customers, to monitor the health of the business without constantly filling everyone's inbox.
Price Research
A software startup wants to know how much small businesses are willing to pay for its system. They extract a sample of 300 small businesses from a Federal Revenue database and call the managers.
Examples of samples in opinion polls
Opinion polls measure the pulse of society on various topics, from politics to pop culture.
Electoral Survey
The most classic example. To find out who will win the presidency, institutes like Datafolha or Ipec select a stratified sample of about 2.000 to 3.000 voters, ensuring that the proportion of men/women, income brackets, education levels, and regions of the country is identical to the official TSE data.
Opinion on Bill Projects
A news portal wants to know if the population is in favor of or against reducing the workweek. They hire an online research panel that sends questionnaires to a representative sample of 1.500 Brazilian workers.
Government Evaluation
Monthly, governments commission telephone surveys with samples of 1.000 citizens to assess the approval of the mayor or governor and understand which areas (health, security, education) need more attention.
Examples of samples in academic research
At the university, the sample is the heart of theses, dissertations, and scientific articles.
Health research
A group of scientists wants to test the effectiveness of a new migraine medication. The population is all migraine sufferers. The sample will be a group of 200 volunteers diagnosed with the problem, divided into two groups (one takes the medication, the other takes a placebo).
Education research
A master's student wants to investigate the impact of tablet use on child literacy. Since she cannot visit all schools in the country, she selects a convenience sample of 4 public schools in her city and follows 120 students for one semester.
Psychology research
A study on anxiety levels in university students during exam week. The researcher sends an online questionnaire to all students on campus, and the 450 who respond voluntarily form their final sample.
The future of sampling with Artificial Intelligence
We couldn't conclude without talking about how AI is revolutionizing the way we choose samples and analyze data.
If in the past researchers relied exclusively on clipboards, calculators, and manual draws, today the scenario is completely different.
AI in sample selection and segmentation
One of the biggest challenges of probabilistic sampling is ensuring that the draw is truly random and free of human biases. Today, AI and Machine Learning algorithms can scan gigantic databases (such as CRM bases with millions of customers) and select perfect stratified samples in milliseconds.
AI analyzes variables that a human would take days to cross-reference. For example, instead of stratifying a sample only by “age” and “gender,” AI can create strata based on “website browsing behavior,” “purchase frequency,” and “email reading time,” creating hyper-representative samples.
Dynamic and synthetic research panels
One of the most fascinating (and somewhat frightening) innovations is the creation of synthetic samples. Some technology companies are using generative AI to create “digital personas” that simulate the behavior of real humans.
If a company needs to test a marketing campaign, it can first run the test on a “sample of AIs” programmed to act as middle-class mothers from the Southern region, for example.
This does not replace research with flesh-and-blood humans, but it works as an incredible initial filter, saving a lot of money before going into the field with a real sample.
Qualitative analysis at a quantitative scale
For a long time, open-ended questions, those where the respondent freely writes what they think, instead of just selecting an option, represented a great challenge for researchers. After all, if a survey had a sample of 2.000 participants, it would be necessary to manually read, interpret, and categorize 2.000 responses. This process demanded time, increased costs, and limited the use of open-ended questions in large-scale surveys.
For this reason, many quantitative surveys prioritized multiple-choice questions, neglecting the richness of detail, opinions, and perceptions that only open-ended responses can reveal.
Today, this scenario has changed. With AnáliseTAP, it is possible to analyze thousands of open-ended responses quickly and organized, identifying patterns, recurring themes, sentiments, and relevant insights in just a few minutes.
Thus, researchers and companies can combine the depth of qualitative research with the scale of quantitative research, obtaining more complete results to support decisions with more confidence.
Will AI replace traditional sampling?
No. Artificial Intelligence is a powerful tool, but the mathematical basis of statistics (population, margin of error, confidence level) remains the same as developed over a century ago. What changes is the speed, precision, and ability to process data volumes that were previously unimaginable.
For those who work with data, the message is clear: deeply understand the human concepts of sampling, but embrace AI tools to do the heavy lifting. This is the combination that will dictate the success of market research and opinion polls in the coming decades.
How to find the sample of respondents?
Finding a reliable sample of respondents can be a challenge, but specialized platforms make this process much simpler.
With the PainelTAP, your research is directed to people who match the defined profile, increasing data quality and result reliability. Want to know how it works? Schedule a demonstration with our team.