There is no single universal number. For quantitative research on a large population, at a 95% confidence level and a margin of error of ±5%, the figure usually quoted is around 400 respondents; at ±10%, roughly 100 will do. In qualitative research you do not count respondents at all. Instead, you watch for theoretical saturation. The exact number depends on the type of research, the precision you need and how homogeneous your population is.
Population and sample: what is the difference
Get this distinction clear before you collect a single answer.
The population is everyone your results are supposed to apply to. If your thesis examines the study motivation of first-year undergraduates, the population is every first-year undergraduate you want your conclusions to cover.
The sample is the specific group of people you actually reached. Penn State's introductory statistics course defines the population as the entire set of possible cases and the sample as the subset from which data are collected. The sample is always smaller than the population.
The difference matters in practice too. In your methodology chapter you must first define the population and only then describe how you selected from it. Skip the first step and your examiners have nothing to compare your sample against.
What a representative sample means
A representative sample is a scaled-down version of the population: the same essential characteristics, just in smaller numbers. With a representative sample you can cautiously generalize your results to the whole population.
Representativeness depends on two things at once: the size of the sample and the way it was selected. A sample that is too small cannot be representative. The reverse holds too. A large sample gathered the wrong way, say a thousand responses from one Facebook group, will not be representative either.
Most bachelor's and master's theses never reach a representative sample in the statistical sense. That is not automatically a flaw. Claiming otherwise is.
How respondents are selected
Sampling methods fall into two groups. In probability sampling, every member of the population has a known, non-zero chance of ending up in the sample. In non-probability sampling, that chance is unknown, so the results cannot be statistically generalized to the whole population.
| Sampling method | How it works | When it fits | Generalizability |
|---|---|---|---|
| Simple random | a lottery draw or random number generator applied to a list of the whole population | quantitative research when you have a list of the population | highest |
| Stratified | you divide the population by characteristics (age, gender, year of study, region) and draw from each stratum | when the sample must keep the population's proportions | high |
| Systematic | you take every k-th element from an ordered list | when you have a list with no hidden ordering | high |
| Purposive | you select by criteria you defined and justified yourself | qualitative research, specific groups | only to the studied group |
| Convenience | you work with whoever you can reach (one class, one company) | when random selection is not feasible | lowest |
| Snowball | one respondent refers you to the next | hard-to-reach or hidden populations | only to the studied network |
With stratified sampling you preserve proportions. If the population is 52% women and 48% men, your sample must keep the same ratio. This is how professional opinion polls are typically built.
Convenience sampling is the most common choice in theses because it is feasible. Do not treat it as a mistake, but do not pretend it is random selection either. Penn State's statistics course warns that a convenience sample is not random and may not represent the intended population: your conclusions will apply only to the people and places you actually studied.
If you are still choosing your overall approach and methods, start with the article on research methodology in your thesis.
What determines the sample size you need
Sample size is not a matter of gut feeling. Four factors determine it.
Type of research. Quantitative research needs enough data for statistical tests. Qualitative research needs enough depth, not enough people.
Required precision. The smaller the error you allow yourself, the more respondents you need. The difference between ±10% and ±5% means roughly four times the sample.
Homogeneity of the population. If people in the population are similar, fewer are enough. A very diverse population needs more respondents to capture the spread.
Number of variables and subgroups. This is what students underestimate most. If you want to compare men and women across three age groups, you need enough respondents in each of those six cells, not just overall. And the more variables you examine, the larger the sample you need.
If you work with hypotheses, sample size decides whether you will have the statistical power to test them at all. How to formulate them properly is covered in the article on hypotheses and research questions.
Quantitative research: how sample size is calculated
The principle is simpler than it looks. You fix two numbers in advance:
- The confidence level (typically 95%), meaning how often the interval would capture the true value in the population.
- The margin of error, meaning how wide an interval around the result you can accept (for example ±5%).
These two inputs determine the minimum sample size. The classic formula for a proportion is given by Glenn D. Israel in the methodological publication Determining Sample Size from the University of Florida (IFAS Extension) as n = Z²pq / e² (for a large or unknown population), where Z corresponds to the chosen confidence level, p is the assumed proportion of the studied attribute, q equals 1 minus p, and e is the margin of error.
The same publication lists rough figures for a large population at a 95% confidence level: about 400 respondents at ±5% error, about 204 at ±7% and about 100 at ±10%.
If the population is finite and not too large, the required number drops. Israel therefore also includes a table for finite populations: with a population of 500 people, a ±5% error calls for a sample of 222 respondents, with 1,000 people it is 286 and with 10,000 people 385. Treat such tables as rough minimums only, since different sources use slightly different formulas and rounding.
You do not have to compute this by hand. The freely available sample size calculator for a population proportion from the British firm Select Statistical Services also accounts for population size. For a population of 10,000, 95% confidence and a ±5% error, it returns a required sample of 370 respondents. The small gap versus the table above is just a difference in calculation and does not change the order of magnitude.
For context: large national surveys work with sample sizes and selection procedures that student research almost never has, which is why thesis findings generalize far less.
Qualitative research: theoretical saturation instead of a headcount
With interviews, focus groups or case studies, counting percentages makes no sense. The criterion is theoretical saturation: the point at which another interview no longer brings a new category or insight, only confirms what you already have.
So you do not declare saturation in advance, you document it. Describe in your thesis after which interview new themes stopped appearing, and show it in your coding.
Rough numbers do exist. A systematic review of empirical saturation tests by Monique Hennink and Bonnie N. Kaiser, published in 2022 in Social Science & Medicine, found that studies reached saturation within a narrow range of 9 to 17 interviews or 4 to 8 focus groups. The authors also caution that these ranges apply mainly to fairly homogeneous groups and narrowly defined research aims.
With a heterogeneous group or a broad topic, plan for more interviews.
What to do when your questionnaire response rate is low
The response rate is the share of completed questionnaires out of those sent. You send out 500, 320 come back, the response rate is 64%.
No clear-cut minimum threshold exists. As a general rule, though, the lower the response rate, the less your data can tell you. And it is not just about the count but also about the structure: the people who answered should look like the people you approached. If only the most motivated filled it in, the results will be skewed no matter how many came in.
When your data falls short, proceed like this:
- Send a polite reminder after a week and possibly another later. With paper questionnaires, a cover letter explaining the aim of the research and thanking respondents for their time also helps.
- Shorten the questionnaire. Overlong, unclear or overly personal questions, along with unguaranteed anonymity, are the most common reasons for a low response rate.
- Widen your channels. Add another school, another department, another group, but describe the new source of respondents in your methodology.
- Reword the scope of your conclusions. If you have 120 answers instead of the planned 400, do not write about "teachers' attitudes" in general. Write about the teachers at the specific schools you approached.
- Admit it in the limitations. A low response rate that is stated and explained in the limitations section is a far smaller problem than one kept quiet.
How to build a questionnaire people actually answer is covered in a separate article on creating a thesis questionnaire.
How to describe and defend your sample in the methodology chapter
Your examiners do not judge whether your sample is large. They judge whether you understand what your sample allows you to claim. So include six things in the text:
- The definition of the population. Who exactly the population is and how large it is.
- The sampling method and its justification. Not "I selected 120 respondents", but "I used convenience sampling at three secondary schools because no list of all teachers in the region was available".
- Inclusion and exclusion criteria. Who could enter the sample and who could not.
- The resulting structure of the sample. Counts and proportions by relevant characteristics, ideally in a table.
- How the data was collected. When, where, through which channel, how many approached and how many responded.
- The limitations. What your sample does not allow. For example, that the conclusions apply only to the schools studied.
Limitations show that you know where the boundary of your data lies. A thesis that claims less and backs it up is easier to defend than one that claims a lot and backs up nothing.
How the description of your sample connects to the rest of the empirical part is covered in the article on the empirical section of a thesis. And if you need expert source material for your methodology chapter or feedback on your sampling plan, you can place an order with us.
The most common sampling mistakes
- A sample without a defined population. Without a population, representativeness has no meaning.
- Convenience sampling passed off as random. Sharing a questionnaire on social media is not random sampling.
- Generalizing beyond the data. A hundred students from one faculty are not "university students" in general.
- Sample size decided after the fact. You set the target before collecting data, not by how many responses happened to arrive.
- Ignoring subgroups. A total of 150 respondents may be enough, but split them into eight cells and some will hold five people.
- Saturation merely asserted. In qualitative research it is not enough to write that saturation occurred. You have to show what it is based on.
- Missing limitations. A chapter without research limitations comes across as less credible, not more.
Frequently Asked Questions
How many respondents do I need for a bachelor's thesis?
There is no universal standard. The size follows from the type of research, the size of the population and the precision you require. Student theses usually cannot match the sample sizes of national surveys, so their findings generalize less. Always check specific expectations with your supervisor and your department's guidelines.
Is 50 respondents too few?
It depends on what you want to do with them. For a descriptive account of a small, clearly defined group, 50 answers can be enough. For testing hypotheses across several subgroups it is usually too few, because too few people remain in each cell. It always comes down to how strong a claim you want to draw from that number.
How many interviews are enough in qualitative research?
The criterion is theoretical saturation, not a count. The systematic review by Hennink and Kaiser (2022) found that the studies examined reached saturation at 9 to 17 interviews or 4 to 8 focus groups, mainly in homogeneous groups with narrow aims. With a diverse group or broader topic, plan for more and document saturation in your thesis.
How do I calculate sample size when I do not know the population size?
For a very large or unknown population, its size is dropped from the calculation and you work only with the confidence level and the margin of error. The publication Determining Sample Size (University of Florida, IFAS Extension) lists roughly 400 respondents at a ±5% error and roughly 100 at ±10% for a large population at a 95% confidence level. Common online sample size calculators will give you similar numbers.
Does the sample in a thesis have to be representative?
No, as long as you do not promise it. Most bachelor's and master's theses work with convenience or purposive sampling, and that is legitimate. The condition is that you describe your selection truthfully and keep your conclusions within what your sample allows. The problem starts only when you present a non-representative sample as representative.
Can I change the size or composition of my sample during the research?
Yes, but you must document it. If you originally planned 300 respondents and got 140, or you added another school, describe the original plan, the reason for the change and the resulting structure of the sample. An undocumented change to the sample is a methodological error. A documented change is a normal part of research practice.