Empirical Research Process: 7 Steps With an Example

Try a quick test. Sentence one: “I have noticed that a lot of people sleep worse when they watch the news in the evening.” Sentence two: “In a standardized survey, sleep quality was associated with evening news use.” Both claim the same thing. Only the second one can be checked by someone else.

The difference between an everyday observation and research lies in exactly that possibility of checking. The empirical research process is the ordered sequence of steps that turns a hunch into a testable statement: you formulate a research question, derive hypotheses from it, choose a method of data collection, gather data, prepare it, analyze it and place the findings in context. By the end of this article you will have a plan you can lay directly over your own thesis.

📌 The key takeaways

• The research process splits into discovery, justification and utilization.
• Raithel (2008) divides the sequence into seven consecutive steps.
• How many phases a model names depends on the textbook.
• Sample size does not create representativeness; selection does.
• A significant result supports your hypothesis, it does not prove it.

Discovery, justification and utilization: the three contexts

Discovery, justification and utilization are the three contexts into which every empirical study can be divided: the context of discovery clarifies what is being studied, the context of justification clarifies how it is studied, and the context of utilization clarifies what the results are ultimately good for.

The three-way split comes from the philosophy of science and is set out in a compact form for students by Jürgen Raithel. The split is not a schedule, it is a map: it tells you which terrain you are standing in when you are sitting at your thesis and no longer sure whether you are still justifying or already interpreting.

For a bachelor’s or master’s thesis the context of justification matters most, because it fills the main body. The context of discovery is usually settled quickly in student work: your own academic interest or a gap in the literature is reason enough. Nobody expects you to start from a major social problem. What sits behind the term as a whole is covered in our guide to empirical research.

The context of discovery explains why you are asking. The context of justification decides whether anyone believes your answer.

The research process: the seven steps at a glance

The research process runs through seven steps according to Jürgen Raithel: research aim and research question, theory and hypothesis building, conceptualization, preparation of the fieldwork and data collection, data preparation, data analysis, and finally interpretation and dissemination, meaning the write-up of the results.

Raithel wrote his model explicitly for students of the social sciences, and it shows: the steps are cut so that almost every one of them can become its own chapter in a thesis. They are set out in the chapter “Phasen des Forschungsprozesses im Überblick” on pages 25 to 32.

The right-hand column is more useful than it looks. Anyone stuck in a thesis rarely has a flaw in their thinking, they are usually working on two steps at once. A clear stopping criterion per step helps more than any schedule.

💡 Tip

If you have to reference your approach in the methods chapter, cite the model at source: Raithel, Jürgen (2008): Quantitative Forschung. Ein Praxiskurs. 2nd edition. VS Verlag für Sozialwissenschaften, Wiesbaden, pages 25 to 32. The book is German-language and can be found via SpringerLink.

How many phases does the research process have?

How many phases the research process has depends on the textbook and not on research itself: common models distinguish between three and nine sections, because they cut the same activities at different levels of detail. Five, six, seven or nine phases are therefore not a contradiction, only a question of how finely the process is described.

A look at the range makes that tangible. The philosophy of science gets by with three contexts, Raithel (2008) names seven steps, and Rainer Schnell, Paul B. Hill and Elke Esser divide the sequence differently again in their German-language standard work Methoden der empirischen Sozialforschung.

Two things follow for your own work. Take one model, name the source and stay with it, because a methods chapter that jumps between two phase models reads as unsure of itself. And do not let a different number unsettle you when your adviser prefers another textbook. A short question about which model is standard in your field saves a lot of time at this point.

One limitation gets lost in almost every account: the sequence is not a one-way street. In practice you jump back, because the pretest kills a question or because a variable turns out to be missing during analysis. The order describes the logic of your argument, not your calendar.

From research question to data collection: steps 1 to 4

The first four steps decide the quality of your work, even though not a single number is produced in them. Anyone who cuts corners here notices only during the analysis, and by then the fieldwork is over and cannot be repeated.

The four steps hang closely together: the research question leads to the hypothesis, the hypothesis leads to the variables, the variables lead to the questionnaire. Change something in one place and check the other three again.

Step 1: set the research aim and the research question

At the start comes the decision about what you actually want to find out. Narrow your topic down far enough that the question fits into a single sentence, and settle what purpose your study serves: exploring a field that has barely been researched, describing a state of affairs, or explaining a relationship. That decision shapes the analysis later on. How to turn it into a workable question is covered under formulating a research question and hypothesis, and the four types of aim under exploratory, descriptive, explanatory and causal.

Step 2: review the theory and build hypotheses

Hypotheses are assumptions about a relationship between at least two variables, and they do not come from your gut, they come from theory. Review the literature for models that already explain your subject. A usable hypothesis names both variables and the direction of the relationship. “Frequent news consumption goes together with poorer sleep quality” can be tested. “Media influence people” cannot.

Step 3: conceptualization, operationalization and pretest

Conceptualization translates your hypothesis into something measurable. First you settle which variables you need and how you capture them, which is called operationalization (the translation of a concept into a concrete measurement rule). Then you build the instrument, usually a questionnaire, define your sample and plan the fieldwork period. At the end comes the pretest: let a handful of people from your target group fill in the draft, then change whatever they got stuck on.

Step 4: prepare the fieldwork and collect the data

Data collection starts the part you cannot undo. Sort out consent from your participants beforehand, set a fixed fieldwork period and think hard about which channels really reach your target group. With an online survey you send out the link, with an expert interview you run the conversations. How to get enough responses back is covered in our piece on finding participants for your online survey.

From raw data to the results chapter: steps 5 to 7

The last three steps turn a table full of answers into a statement. They are the part your committee reads most closely, because this is where it shows whether you have understood your own data.

Plan more time for them than the length suggests. Analysis usually takes longer than the fieldwork, and unlike the questionnaire there is no template that does the thinking for you.

Step 5: prepare and clean the data

Raw data is never ready for analysis. First you bring the answers into a data matrix, a table with one row per person and one column per variable, and transfer it into your analysis software. Then you check the dataset for obvious errors and sort out incomplete cases and ones that were visibly clicked through. What matters is the justification: removing cases because they contradict your own hypothesis is falsification. The quality criteria behind this are covered under objectivity, reliability and validity.

Step 6: analyze the data

The method of analysis follows from the type of your data. Quantitative data is analyzed through frequencies, means, measures of association and, depending on the question, significance tests, qualitative data through content-analytical procedures and coding. A significance level of 5 percent is standard, written as alpha equals 0.05. The differences between the two routes are set out under qualitative and quantitative research methods, the specific procedures under methods of analysis.

Step 7: interpret the results and write them up

Interpretation answers your research question and places the finding in the existing literature. Keep a clean line between what the data shows and what you conclude from it. Interpretation always includes the limits of your study: sample size, selection procedure, self-reporting by participants. Dissemination then means spreading the results, which for a thesis is simply the finished text with a conclusion and an outlook.

An example: how an empirical thesis runs

The sequence becomes clearest once it is played through on a single topic. The example used here is a bachelor’s thesis on the relationship between news consumption and wellbeing, because all seven steps can be shown on it.

One note first, so that no wrong impression arises: what follows is a template for the process, not a study that was carried out. Results stay deliberately open here, because without real fieldwork they would prove nothing.

Preparation in the example: question, hypothesis and questionnaire

Four of the seven steps happen before the first answer arrives, and that is exactly where it is decided whether the later analysis will yield anything at all.

1. Research aim and research question. The research question is: does the frequency of news consumption relate to self-rated wellbeing? The aim is explanatory, so it is about a relationship and not about a pure description.
2. Theory and hypothesis. Cultivation theory, developed by George Gerbner, who coined the term “mean world syndrome”, works as the theoretical frame. The hypothesis is: frequent news consumption goes together with lower self-rated wellbeing.
3. Conceptualization. Data is collected through a standardized online survey. The variables are news use, wellbeing and a few sociodemographic characteristics. The target group covers adults in the United States, and the sample comes from a convenience selection. After the pretest the wellbeing questions move to the front, so that the questions on media use do not color them.
4. Data collection. The fieldwork period runs for four weeks, with the survey link going out through campus mailing lists and relevant groups. Anyone planning a survey for a bachelor’s thesis should budget generously for this step, because responses rarely come in as fast as hoped.

Analysis in the example: from raw data to conclusion

From this point on you work only with what has actually come back. New data cannot be obtained, and that is exactly why the preparation carries so much weight.

1. Data preparation. The answers are transferred into a data matrix and coded. Drop-outs and cases with a strikingly short completion time are removed according to rules set in advance, not by feel.
2. Data analysis. Frequencies and means for both variables come first, then a measure of association and a suitable significance test. Whether a relationship shows up is open, and that is the point: a hypothesis whose result is fixed in advance does not need testing.
3. Interpretation and dissemination. The finding is placed in the existing literature and the limits are named: convenience sampling, self-reporting and a cross-sectional design that says nothing about cause and effect. The outlook names what a follow-up study could do better.

Common mistakes in the research process

Most problems in a thesis do not arise in the statistics, they arise much earlier. Five mistakes turn up so regularly that a quick check before the fieldwork is worth the time.

What the five have in common is that all of them can be fixed before data collection and almost never afterward. That is the bad news and the good news at the same time.

Mistake 1: the method is fixed before the research question

A method chosen too early narrows the question instead of serving it. Anyone who sets out to “do something with interviews” then rewrites the research question until interviews fit. The sensible order runs the other way around: first you settle what you want to know, and from that follows how it can be measured. You have to justify the choice in the methods chapter anyway, and “I liked the sound of it” will not carry there.

Mistake 2: a large sample is taken to be representative

A larger sample makes your results more precise, but not automatically representative. You often read that representativeness grows with the number of participants. That does not hold, because what counts is how the sample came about.

The Task Force on Non-Probability Sampling of the American Association for Public Opinion Research warned in its 2013 report against claiming representativeness for samples that were not drawn at random, such as opt-in panels or convenience selections. The report also makes clear that a larger number of cases does not make up for the missing random selection. For a thesis that is a relief: what is expected is a well-justified selection, not a population-representative study. Which criteria really apply is covered in our piece on the representative survey, and the report itself is available as a PDF from AAPOR.

Mistake 3: a significant result is treated as proof

A statistically significant result supports your hypothesis, it does not prove it. The p-value indicates how well the data fits an assumed model, not how likely it is that the hypothesis under study is true. The American Statistical Association set exactly that out in six principles in 2016.

Careful wording matters here. Saying that a relationship is statistically significant is correct. Saying that a hypothesis has been proven is not. And one detail slips regularly: the usual significance level sits at 5 percent, meaning alpha equals 0.05, and not at 0.05 percent. The six principles are available as a PDF from the American Statistical Association.

Mistake 4: answer categories overlap

Overlapping answer categories make a question unusable. Anyone who offers 0 to 2, 2 to 4 and 4 to 6 hours for daily news consumption forces every person with exactly two hours into a coin flip. Correct categories are mutually exclusive and leave no gaps, for example under 2 hours, 2 to under 4 hours, and 4 hours or more. The same mistake happens with age groups, and it almost always surfaces only during the analysis.

Mistake 5: the limits of the study are missing

A results chapter without limitations does not read as confident, it reads as careless. Every student study has limits, and naming them costs no points, it earns them. Say concretely what restricts your findings: the selection procedure, the size of the sample, self-reporting by participants, a cross-sectional design that allows no statement about cause and effect. Knowing your own limits shows that you have understood the method.

Sources and further reading

The following sources carry the statements in this article and work at the same time as references for your own methods chapter.

• Raithel, Jürgen (2008): Quantitative Forschung. Ein Praxiskurs. 2nd edition. VS Verlag für Sozialwissenschaften, Wiesbaden. German-language textbook, chapter on the phases of the research process, pages 25 to 32.
• Friedrichs, Jürgen (1990): Methoden empirischer Sozialforschung. 14th edition. German-language textbook, foundational for the distinction between the contexts of discovery, justification and utilization.
• Schnell, Rainer; Hill, Paul B.; Esser, Elke (2013): Methoden der empirischen Sozialforschung. German-language standard work with its own phase model.
• American Association for Public Opinion Research (2013): Report of the AAPOR Task Force on Non-Probability Sampling.
• American Statistical Association (2016): Statement on Statistical Significance and P-Values.

Conclusion

The empirical research process is less a method than a sequence of decisions that you have to be able to justify at any point. Where research question, hypothesis, variables and analysis lock together cleanly, your work is good even when the finding turns out to be unspectacular. What gets graded is the approach, not the result.

Where to go next

• Want to go over the basics again? Empirical research: definition, methods, guide
• Looking for the right method of data collection? Qualitative and quantitative research methods
• Already sitting on your data? Methods of analysis in empirical research

Still missing the data for your context of justification?

With empirio.ai, an online survey tool from Germany, you build your questionnaire for free, share it as a link and see the answers analyzed straight away.

Create your survey for free