Fundamentals of Empirical Research | empirio

Ask three students what “empirical” actually means and you get three answers: running your own survey, working with numbers, or simply the opposite of theory. All three fall short.

Empirical research answers a research question systematically with data from the observable world instead of with literature and argument alone. Whether those data are numbers or transcripts, and whether you collect them yourself or pull them from an archive, does not decide whether your work is empirical. By the end you will know which route fits your question and how to defend that choice in your methods chapter.

📌 Key takeaways

• Empirical research answers a research question with data instead of with argument.
• Empirical is not the same as quantitative. Interviews are empirical too.
• Ready-made data count: ICPSR calls itself the largest social science data archive.
• Before your first question come consent and, in most cases, an IRB determination.
• A literature review is not an emergency exit but a method in its own right.

What is empirical research? Definition and boundaries

Empirical research is a way of working in which a research question gets answered by collecting and analyzing data about the observable world in a systematic way. The opposite pole is not literature, but pure argument.

The word goes back to the Greek empeiria, meaning experience. Right there sits the risk of confusion, because anyone who asks around in a group chat also gathers experience. What separates the empirical approach from everyday observation is not the experience itself but the handling of it.

Four points mark the difference. Everyday knowledge piles up by accident, while empirical research follows a plan set down in advance. Everyday knowledge keeps whatever fits the picture, while empirical research records the cases that spoil it. Everyday knowledge gets by with fuzzy terms, while empirical research pins every term down until it can be measured. And everyday knowledge stays private, while empirical research is written up so that somebody else can follow the path and, ideally, walk it again.

Turn the sentence around and it matters just as much: not every piece of scholarly work is empirical, and nobody should read that as a defect. A mathematical proof, a legal interpretation and a philosophical clarification of terms are scholarly work without a single number collected. Empirical methods are a tool for a certain kind of question, not a seal of quality for scholarship.

Two myths that turn up almost everywhere

Two sentences about empirical research get repeated so often that hardly anyone checks them. Both sound plausible, and both push students toward a more expensive design than their question calls for.

Taking them one at a time makes sense, because they do damage in different places. The first myth costs you suitable methods, the second costs you suitable data.

The myth that empirical means numbers

Empirical and quantitative are not the same thing. Six interviews with nurses gather data about the observable world, analyze them systematically and count as fully empirical, even though no percentage ever appears. The axis from quantitative to qualitative describes the kind of data, not the question of whether data are collected at all.

Side effect worth naming: because so many guides declare objectivity, reliability and validity to be preconditions for any empirical work, qualitative projects end up measured against a yardstick built for something else. All three criteria come from quantitative test theory. Qualitative research has developed catalogs of its own, and a student who reaches for the wrong one concludes that the work is methodologically weak when it is not.

The myth that empirical research is the opposite of a literature review

Setting empirical work against literature work is a simplification that falls apart on close inspection. A meta-analysis and a systematic review draw exclusively on existing material and are empirical all the same, because they gather, code and analyze findings from published studies. That is why the same guides that claim the contrast list meta-analysis as an empirical method two paragraphs later.

The clean dividing line runs not between “your own data” and “literature” but between two questions: are you working through a body of material by rules fixed in advance, or are you ordering and weighing the state of research? The former is empirical, even without fieldwork of your own. The latter is a theoretical paper, even if hundreds of studies get read.

💡 Tip

Test your planned project against a single question: at the end, is there a body of material you worked through by rules fixed in advance? If the answer is yes, your work is empirical. Whether that material consists of questionnaires, transcripts or 40 coded studies changes nothing.

Empirical research methods: qualitative, quantitative and mixed

Your method follows from your research question, not from your taste. A question about frequencies, distributions or relationships calls for numbers. A question about meanings, motives or processes calls for text. Anyone who picks the method first and bends the question afterward finds out during analysis.

In practice that means phrasing your question precisely enough that the kind of data follows from it. “How satisfied are third-year students with academic advising?” calls for a standardized survey. “What makes students feel that advising helped them?” calls for interviews. Both questions cover one topic and lead to entirely different research methods. When even the vocabulary is unclear, the piece on what a survey is separates poll, survey and questionnaire cleanly.

Quantitative methods: measure and compare

Quantitative research pins a characteristic down so that it can be captured the same way across many cases. The instrument in quantitative research is fixed beforehand and stays untouched during fieldwork, because comparability breaks otherwise. Typical designs are standardized surveys, structured observation, quantitative content analysis and experiments. Comparability is the gain, depth is the price: whatever the questionnaire did not anticipate never shows up in the data.

Qualitative methods: understand and reconstruct

Qualitative research keeps the procedure deliberately open, so that the material can show something nobody expected. Qualitative research covers semi-structured interviews, focus groups, participant observation and qualitative content analysis. Your interview guide sets topics, not wording and not order. Case numbers stay small, analysis is slow, and findings hold for the cases you studied rather than for a population.

Mixed methods: both in one project

Mixed methods combines both approaches inside one design instead of splitting them across two papers. The most common sequence in a thesis starts with a handful of interviews to understand the field, then tests the resulting hunches with a standardized questionnaire. Reversing the order works just as well: survey first, interviews afterward to explain results that surprised you. Justify the combination in your methods chapter, or two half studies will read like one overloaded project.

Where your data come from: collect or reuse

Almost every guide assumes that an empirical project starts with fieldwork of your own. Collecting data yourself is the best known route, not the only one, and for many questions not the best one. The alternative is secondary analysis: you work with data somebody else gathered and put your own question to them.

Weigh the difference honestly. Fieldwork of your own costs several weeks for instrument, recruitment and cleaning, and often ends with a sample drawn from people you know. An existing dataset hands you several thousand cases with a documented sampling plan on day one. What you give up is freedom over the wording, because you can only analyze what somebody asked back then.

Four routes to your data

Four routes are open to you in practice, and they do not rule each other out. Plenty of strong theses pair a small survey of their own with a reference dataset that shows how the sample compares to the wider population.

1. Your own fieldwork. You build an instrument and run data collection yourself. Full control over the questions, full workload.
2. A social science data archive. ICPSR, part of the Institute for Social Research at the University of Michigan, holds thousands of curated studies for reuse.
3. Federal statistics. Public use files from the Census Bureau cover employment, housing, commuting and educational attainment at no cost.
4. Published findings. For a systematic review or a meta-analysis, the results of other people’s studies are the material itself.

Money decides archive access here, which is the part nobody mentions. Membership in ICPSR “is determined by your university or organizational affiliation”, and students at a member school get “institution-wide access to unlimited data downloads for students, faculty, staff”. Some collections are open to anyone. Everybody else pays an administration fee of roughly 825 US dollars per dataset, and a Researcher Passport account is required either way. Where a national archive is publicly funded, the same kind of data costs students nothing, so check your library’s status before you build a plan around a dataset.

The survey nobody may decline

Federal statistics rest on a survey with no opt-out, and the wording is unusually blunt. Answering the American Community Survey “is required by law (Title 13, U.S. Code, Sections 141, 193, and 221)”, and the Census Bureau adds that Title 13, as changed by Title 18, imposes a penalty for not responding. About 3.5 million households receive it every year, month after month. Anyone writing about housing or commuting starts from a stronger footing here than from a questionnaire of their own.

Sources in detail: membership in ICPSR, the Census Bureau on the duty to answer the ACS and its comparison of the ACS and the decennial census (as of September 2026).

Once you settle on fieldwork of your own, the online survey is the most common route in a thesis. Why it carries and whom it systematically misses is covered in the piece on the online survey as an empirical method. How it compares with mail, phone and in-person interviewing is covered in the piece on online surveys and their limits.

How an empirical study runs: six decisions

Textbooks cut an empirical study into anywhere from five to nine phases. For practical purposes the number of phases matters less than the insight behind it: at six points you make a decision that fixes everything downstream and can barely be corrected later.

Order here is not a recommendation but a chain of dependencies. Anyone who picks the analysis after fieldwork sometimes discovers that the scales do not permit the planned calculation. Step six therefore belongs, in your head, to step three.

1. Fix the research question. Your research question has to be narrow enough that fieldwork follows from it. Aim follows too: describe, explain or explore.
2. Form hypotheses where they belong. Confirmatory studies need hypotheses, exploratory ones do not. An exploratory project with forced hypotheses is a classic beginner mistake.
3. Choose the research design. Your research design sets method, timing and case selection. Whether you measure once or repeatedly gets settled here.
4. Make your terms measurable. “Student satisfaction” is a word, not a variable. Only the translation into concrete items and answer scales makes it collectable.
5. Determine the sample. Procedure and size of your sample decide what you may claim at the end and what you may not.
6. Settle the analysis. Which of the analysis methods is available depends on your kind of data and your level of measurement, not on your wishes.

Full walk-through with examples, a schedule and the question of what goes wrong in each phase sits in the piece on the empirical research process.

Privacy and IRB review: what has to be settled first

Asking people questions means handling information about them, and the rules bite before the first answer is analyzed. Almost no guide mentions the point, although the same guides send students straight into interviews. Nothing here can be repaired afterward, because a study run without consent does not become acceptable once the data are anonymized.

Federal privacy law is the part that surprises readers from abroad. The Government Accountability Office states that “The United States does not have a comprehensive Internet privacy law” and noted as of February 2026 that such legislation has not been enacted. State law fills part of the gap, and the California Consumer Privacy Act “gives consumers more control over the personal information that businesses collect”, as the state Attorney General puts it.

The Common Rule and your IRB

Research with human subjects runs under the Common Rule, 45 CFR Part 46, overseen by the Office for Human Research Protections within the Department of Health and Human Services. Two definitions decide whether the rules touch your project. Research means work “designed to develop or contribute to generalizable knowledge”, and a human subject is “a living individual about whom an investigator (whether professional or student)” gathers data. Read that parenthesis twice, because students are not carved out. Among the criteria an IRB applies is § 46.111(a)(7), which asks whether “there are adequate provisions to protect the privacy of subjects”. The full text sits in the eCFR (as of September 2026).

Class projects and two mistakes to avoid

Class projects built solely for teaching usually fall outside the definition, because they do not aim at generalizable knowledge. Ohio University publishes a handout on class projects that walks through the distinction, and most research offices keep an equivalent. Two errors show up constantly all the same, and both sound sensible.

First error: publication is not the test. The Office for Human Research Protections is explicit that “Whether or how an investigator shares results with the scientific community is not the deciding factor”, so what you intended at the outset is what counts. Second error: the Common Rule “does not specify who can make determinations about exemptions”, which makes the widespread claim that federal law bars you from deciding simply wrong. OHRP advises against investigators deciding alone, and most institutions require a submission, so treat it as institutional practice rather than federal law.

⚠️ Heads up

Exempt is a determination, not a synonym for skipping paperwork. Exempt category (2) at § 46.104(d)(2) covers “survey procedures, interview procedures, or observation of public behavior”, which is where most student surveys land, and somebody still has to make and record that call. Submit before you recruit, not after.

Your IRB office decides the individual case and your faculty advisor decides the scholarly one. Ask both before the first invitation goes out. This article is not legal advice and does not replace guidance from your own institution.

When a thesis does not have to be empirical

Plenty of students start from the decision to do something empirical and look for a fitting question afterward. Working in that order produces the familiar paper with 43 responses from one cohort, which carries nothing in the end.

An honest check runs the other way. Some questions are served better by existing literature or existing data than by fieldwork, and some programs require an empirical component in writing. Settle both before you register a topic.

A literature review is no fallback for students who cannot find participants. Working that way carries its own standards: a search strategy somebody else can reproduce, a justified selection of sources, synthesis instead of a list. Done properly, the result says more than a thinly populated survey.

Terminology trips people up here, because a thesis in the United States is the master’s document and a dissertation the doctoral one, the reverse of British usage, as Cornell’s Graduate School shows by listing the dissertation defense for doctoral and the thesis defense for master’s candidates. If your program requires an empirical component, that requirement wins. How the empirical part fits the outline is covered in the piece on the empirical thesis. Your program handbook governs in every case.

How to tell a sound empirical study from a shaky one

Whether a study holds up is decided by the fit between question and procedure, not by sample size, and by whether the procedure is laid open. Quality criteria exist for exactly that, and the most common error is reaching for the wrong catalog.

Quantitative work is judged by objectivity, reliability and validity. Objectivity means the result does not depend on who gathers and analyzes it. Reliability means a repeat under the same conditions produces the same result. Validity means the instrument measures what it claims to measure. Definition three gets garbled constantly, often into a question about whether the results answer the research question, which is a different matter and misses the term.

Qualitative work has catalogs of its own, because repeatability is no sensible yardstick there. At the center sit traceability of the procedure, fit between method and subject, and disclosure of your own role in the field. Which catalog fits your project and how to apply it in the methods chapter is shown in the piece on quality criteria in qualitative and quantitative research.

A study does not become good by having many cases. A study becomes good when somebody else can retrace it.

As a reference covering both approaches in one volume, Creswell, John W. and Creswell, J. David (2023): Research Design: Qualitative, Quantitative, and Mixed Methods Approaches, 6th edition, SAGE, is widely used in American courses. Older editions still circulate in course packets, so check which one a citation points to. The current edition is listed at SAGE.

Common mistakes in a first empirical project

Four mistakes turn up in student projects again and again. All four share a pattern: they arise early and surface late, usually once the data are in and nothing can be rescued.

Knowing them lets you steer during planning. Guarding against them costs a few days and saves an entire round of fieldwork.

The method comes before the question

Deciding to run a survey and then wondering what to ask flips the dependency. Questionnaires built that way graze many topics and answer none. The question comes first, and the method follows from it. If the method is fixed for practical reasons, because your advisor set it, then the question adapts before the questionnaire exists.

The sample is your own social circle

A survey shared through your group chat and your Instagram story reaches people who resemble each other. No disaster follows as long as the methods chapter says so and the findings stay cautious. Trouble starts when the conclusion talks about students in general. Whether a study counts as representative depends on the sampling procedure, not on the number of responses.

Level of measurement gets checked during analysis

Whether you may compute a mean, run a correlation or only count frequencies depends on the level of measurement of your variables. Which level of measurement a question carries is settled while you build the questionnaire, not during analysis. Anyone who notices in the statistics software that every item is nominal cannot run the planned calculation and has to bend the question afterward.

The methods chapter describes instead of justifying

Writing that an online survey was conducted is a description. What readers expect is a justification: why this method fits this question, which alternative existed and why it dropped out. Three sentences usually cover it, and they separate a documented project from a defended one. Almost no guide mentions the point, and almost every committee asks about it.

Conclusion

Empirical research is less a technique than a posture: you decide in advance what will settle your question, and you stick to it even when the answer turns out inconvenient. Whether numbers or transcripts come out of it, and whether you gather them yourself or reach into an existing dataset, is a question of fit rather than rank. Check first whether your question needs data at all, and pick the route afterward.

Where to go next

• Wondering how the empirical part fits your outline? Writing an empirical paper
• Unsure whether your study is descriptive or explanatory? Exploratory, descriptive, explanatory, causal
• Data collected and the right analysis still open? Descriptive and inferential statistics
• Building your questionnaire right now? Create a questionnaire
• Methods chapter written but the tone is off? Academic writing

Ready to run your own study?

With empirio.ai, an online survey tool from Germany, you build your survey on the free plan and keep every response hosted in the EU.

Create your survey for free