Data Analysis Methods in Empirical Research: How to Choose

Picture two dissertations on a viva table: thirty charts in one, a single mean and three interview quotes in the other. Which one has been analysed better? Not the one with more output, but the one whose analysis matches the question.

That is where data analysis methods come in, and it is where most people get lost. Data analysis methods are the systematic procedures you use to organise, condense and test the material you have collected so that it answers your research question. Three things decide which one applies: your research question, the kind of material you have, and for numbers the level of measurement. By the end of this article you will be able to justify your choice rather than assert it.

📌 Key points at a glance

• The research question decides the method, not the sample size.
• Descriptive statistics always come before inferential ones.
• Level of measurement limits which statistic you may report.
• Qualitative material passes through recording, transcription and coding.
• Statistical significance says nothing about the size of an effect.

What are data analysis methods?

Data analysis methods are defined, repeatable procedures for ordering, condensing and interpreting collected material. They start where fieldwork ends and finish where numbers or text have turned into a reasoned answer to the research question.

Several labels describe the same thing. Data analysis, analytical techniques and analysis procedures are used interchangeably in British academic writing, and none of them is more precise than the others. A different distinction matters far more: the one between preparation, analysis and interpretation.

Preparation makes the material usable at all: checking missing values, coding answers, transcribing recordings. Analysis turns the prepared material into statistics or categories. Interpretation places those results against the research question and the existing literature. Collapsing interpretation into analysis is the quickest way to lose the transparency your marker is looking for.

Choosing a data analysis method: three questions

Choosing a data analysis method means answering three questions in order: what do you want to find out, what kind of material do you have, and at what level of measurement was it recorded? Only the third question narrows the field far enough for a single technique to remain.

The order is not interchangeable. Picking a technique first because it appeared in a module, then bending the research question to fit it, produces a dissertation that calculates correctly and answers nothing. Marking criteria at most British universities reward the justification, not the volume of output.

Write the answers down before you open any software. Those three lines become the opening paragraph of your methodology chapter, and they save you from the most common supervision comment of all: that the analysis is competent but unmotivated.

Quantitative data analysis: describing and inferring

Quantitative data analysis falls into two groups: descriptive statistics summarise the sample in front of you, while inferential statistics ask whether a finding holds in the wider population. An empirical dissertation needs both, because either one on its own leaves the argument incomplete.

The order never changes. Descriptive analysis shows how your sample is composed and whether the values look plausible at all. Only then is it worth asking whether a difference would survive outside your sample. The full comparison sits in our article on descriptive and inferential statistics.

Descriptive statistics: summarising your own sample

Descriptive statistics condense your data into a handful of figures without generalising beyond the sample. They cover frequencies and proportions, measures of central tendency such as the mean, median and mode, measures of spread such as range, standard deviation and variance, and measures of association such as the correlation coefficient, which runs from -1 to +1 with 0 indicating no linear relationship.

One error appears again and again at this point: correlation does not establish cause. When two variables move together, an unmeasured third variable may be driving both. Causal claims require a design that rules out competing explanations, not a larger correlation coefficient.

Inferential statistics and what significance really means

Inferential statistics use significance tests to judge whether a difference found in a sample is plausible in the population. The Office for National Statistics defines the term precisely: a result is statistically significant if it is likely not caused by chance or by the variable nature of the samples, and at the 5% level that means less than a 1 in 20 chance of observing the change if no underlying change exists (Office for National Statistics, accessed 23 August 2026).

The ONS also sets a writing rule worth copying. In its statistical bulletins it avoids the word significant on its own and always writes statistically significant, so readers cannot mistake a technical statement for a claim about importance. Adopt that habit in your own dissertation and one whole class of misreading disappears.

Confidence intervals and the limits of a sample

A confidence interval expresses how much uncertainty sits around an estimate. At the 95% level the ONS describes it like this: if you drew 20 random samples and calculated an interval for each, roughly 19 of the 20 would contain the true population value and 1 would not. The margin of error is the standard error multiplied by 1.96.

Two limits belong in every write-up. The coefficient of variation, the standard error divided by the estimate, should not be used for values close to zero or for percentages, and where it exceeds 50% the estimate is very imprecise. And sampling error is only part of the story: non-response, refusal, inaccurate answers and processing mistakes would still be present even if the entire population had been surveyed.

Which statistics fit which level of measurement?

Level of measurement decides which statistic and which test your data will bear. Nominal data support frequencies only, ordinal data add rank statements and the median, and scale data are the first to permit the arithmetic mean and everything built on top of it.

The table below stays deliberately short and covers the cases that appear in almost every undergraduate and taught masters project. Our article on levels of measurement explains what sits behind each step.

This is not a formality. An arithmetic mean across nominal categories, such as an average degree subject, produces a number with no meaning, and that is exactly the kind of number an external examiner spots in seconds.

⚠️ Watch out

Likert scales are strictly ordinal, because the distance between agree and strongly agree is not measurably equal. Whether means may still be taken from them is debated in the methods literature to this day. Reporting the median and the frequency distribution alongside any mean, and justifying the decision in your methodology chapter, keeps the argument defensible.

Qualitative data analysis: from interview to code

Qualitative data analysis condenses text, audio or image material into an answer to the research question by way of a coding frame. Unlike numbers, this material does not arrive in analysable form at all, which is why a preparation step always comes first.

The sequence is strikingly similar across approaches, even though the schools justify their procedures very differently. Three steps run through practically every qualitative analysis.

The three steps of any qualitative analysis

1. Recording. Interviews, focus groups or observations are recorded or logged, with documented consent and, at most British universities, prior ethics approval.
2. Transcription. The recording is written up according to rules set in advance. Whether accent, pauses and filler words are captured depends on the approach and belongs in the methodology chapter rather than being decided on the fly.
3. Coding. The transcript is assigned to categories passage by passage. Codes are either developed inductively from the material or applied deductively from theory and the interview guide.

Approaches you are most likely to use

Thematic analysis is the approach British students meet first, and it is the most widely taught in psychology, health and social sciences departments in the UK. It moves from familiarisation through initial coding to themes that are reviewed against the whole data set, and it does not commit you to a particular theoretical school.

Qualitative content analysis is the closely related, more rule-bound alternative, systematised by Philipp Mayring from the first edition of 1982 onwards, distinguishing inductive category development from deductive category application. Grounded theory, developed by Glaser and Strauss, aims instead at building theory from the material and alternates continuously between collecting and analysing. Sequence-based and interpretive approaches work minutely through single passages and suit very small samples.

Reading tip: Mayring, Philipp (2010). Qualitative Inhaltsanalyse. Grundlagen und Techniken. 12th edition. Weinheim and Basel: Beltz. An English summary by the same author from 2014 is freely available through the Social Science Open Access Repository.

Writing up your analysis in a dissertation

The write-up of your analysis belongs in the methodology chapter and answers four questions: which technique you chose, why it suits the research question, how you applied it in practice, and which software you used. Four to six paragraphs are usually enough.

Reasoning beats description. A sentence such as “the data were analysed in SPSS” reveals nothing about a methodological decision. “Because the dependent variable is ordinal, a rank-based test was used in place of a t-test” shows that you know the limits of your material, and that is what earns marks.

Write the chapter so that another researcher could repeat your analysis from it. That is precisely what the criteria of objectivity, reliability and validity are testing, and it is where dissertations with plenty of calculation and little documentation come unstuck. Your department handbook and your supervisor set the binding requirements.

Common mistakes in data analysis

Most marks lost in empirical projects go missing through four recurring errors of reasoning rather than through arithmetic. All four are avoidable if you know about them before you start calculating.

Mistake 1: fixing the method before the research question

Choosing an analysis technique before the research question is settled builds the project backwards. The result is a set of correctly calculated figures that add up to nothing. Formulate your hypotheses before you type a single line into your analysis software.

Mistake 2: calculating statistics the measurement level will not support

A mean across grades, ranks or answer categories looks harmless and remains indefensible. Check each variable separately for its level of measurement and record the result before you analyse anything. Twenty minutes of work spares you an awkward question in the viva.

Mistake 3: confusing significance with importance

A statistically significant result says nothing about whether the effect is large or practically relevant. In very large samples even a trivial difference reaches significance. Report an effect size alongside every p-value and explain what the difference means in substantive terms.

Mistake 4: planning the analysis after the fieldwork

Sending out the questionnaire first and thinking about analysis afterwards usually reveals that a decisive question was asked in the wrong format. An open question cannot be converted into a scale after the fact. Plan the analysis while you are still designing the questionnaire.

Conclusion

The right data analysis method is not found in a list. It follows from your research question, your type of material and your level of measurement. Settling those three points before fieldwork spares you the uncomfortable discovery that the best data in the world answer the wrong question.

The data do not decide the method. The question decides both.

Where to go next

• Want the difference between the two branches of statistics? Descriptive and inferential statistics
• Unsure about the measurement level of your variables? Levels of measurement
• Looking for the whole research process in one place? Empirical research

Thinking about analysis while you build the survey?

With empirio.ai, an online survey tool from Germany, you see frequencies, cross-tabulations and filters in a dashboard that updates with every new response.

Create a survey and analyse it straight away