Qualitative vs Quantitative Research: How to Actually Choose

The short answer

Neither approach is better — they answer different kinds of question. Choose quantitative when your question asks how many, how much, how often or whether two things are related across a population; choose qualitative when it asks how or why something is understood, experienced or done. The decision follows from the question, and if you have already decided the method before writing the question, you have made the decision backwards.

The practical corollary matters just as much: the design you pick has to be one you can actually execute with the access, time and skills you have. A perfectly justified survey with eleven respondents answers nothing. This page covers both halves — what your question requires, and what you can finish.

What is the difference, precisely?

The usual summary — numbers versus words — is true but shallow. The real difference is in the logic of the claim.

• Quantitative research measures. You define variables in advance, collect comparable values across many cases, and use statistics to say something about the group. Its strength is prevalence and relationship: how widespread, how strong, how likely to be more than chance. Its cost is that everything must be decided beforehand — you learn only about what you thought to measure.
• Qualitative research interprets. You collect rich material from few cases — interviews, documents, observation — and analyse it systematically for meaning, process and mechanism. Its strength is depth and discovery: you can find out that the interesting variable is one you had not considered. Its cost is that you cannot say how common anything is.

A useful reframing: quantitative work is strong at testing a well-formed expectation; qualitative work is strong at generating one. That is why fields with little prior research lean qualitative and mature fields lean quantitative — not because one is more rigorous.

Which one is better?

This is one of the most common ways the question gets searched, so it deserves a direct answer: neither, and treating it as a contest is how students end up defending the wrong thing in the viva. Examiners do not assess whether you chose the prestigious method. They assess whether the design fits the question and whether you understand what your data can and cannot support.

Where the perception of a hierarchy comes from is discipline convention. In psychology or economics, quantitative is the default and a qualitative design needs more explicit justification. In education, social work or parts of management, the reverse is true. That is a fact about your department, not about the methods — so look at recently accepted theses from your own institute before you assume anything.

How do I tell which my question needs?

Read your research question and identify the verb. It usually gives the answer away immediately.

• How many, how often, how much, to what extent, is there a relationship between, does X affect Y → quantitative. These are measurement claims.
• How do, why do, in what way, how is X understood, what does X mean to → qualitative. These are interpretive claims.
• Two verbs in one question→ either you need mixed methods, or, more often, the question is doing two jobs and should be narrowed. In a bachelor's thesis, narrow it.

Then run three checks before committing:

1. Access. Can you actually reach the people or data? Not in principle — name them. A quantitative design with no distribution channel to respondents is a design that will fail in week six.
2. Existing knowledge. If your field already has established constructs and validated instruments, quantitative work is cheaper because you can adopt them. If it does not, you would be inventing a measure and then measuring with it, which is two studies.
3. Your own skills. Analysing twelve transcripts properly and running a regression properly are both learnable, but not both in the same semester alongside everything else.

If the checks contradict the verb in your question, change the question. That is normal and far cheaper than discovering it after data collection.

What can each approach actually claim?

This is the section that separates a competent methodology chapter from a weak one, and it is the distinction most often blurred: generalisation comes in two kinds, and they are not interchangeable.

Statistical generalisation infers from a sample to a population. It requires that the sample was selected in a way that makes inference valid and is large enough for the analysis. Only then may you write that a finding holds beyond the people you studied.

Analytical generalisation infers from findings to theory. Your cases demonstrate a mechanism or pattern, and the claim is that this mechanism plausibly operates in comparable settings and now warrants testing at scale. It does not require a representative sample, because it is not a claim about prevalence at all.

Two sentences that show the difference, both from twelve interviews:

• Overreach: “The findings show that most practitioners in the sector view the policy negatively.” Twelve interviews cannot support “most” or “the sector”.
• Defensible: “The interviews identify three recurring reasons for resistance to the policy, suggesting that implementation friction is driven by role conflict rather than by information gaps — a proposition that warrants testing on a representative sample.”

The second says something more useful, not less, and it cannot be attacked on sample size. Quantitative work has its own mirror-image failure, incidentally: reporting a statistically significant result without its effect size, so a negligible difference is presented as an important one. Both errors are the same error — claiming more than the design supports.

How big does the sample need to be?

The honest answer differs by approach because the two logics are different.

Qualitative: there is no fixed number.Single-digit samples are normal and defensible in bachelor's theses, and larger ones in master's work. What you must argue is coverage — that the roles, contexts or perspectives relevant to your question are represented — and whether later cases were still producing new categories. Four interviews across four relevant roles carry more than ten with the same role. Ask your supervisor what the department expects, because that expectation is real even when it is not methodological.

Quantitative: the number follows from the analysis, not from a rule of thumb. What you need depends on the test you plan, the number of variables, and the size of effect you would want to detect. The productive move is to decide the analysis first and work backwards to the required sample, rather than collecting responses and hoping. If the number that comes out is unreachable with your distribution channels, that is decisive information — get it in week two, not week twelve.

A shrunken quantitative study is worse than an equivalent qualitative one, because an underpowered analysis produces numbers that look authoritative and are not. Small qualitative samples wear their limits openly; small quantitative samples hide them behind decimal places.

Should I use mixed methods?

Sometimes — but less often than students propose it. Mixed methods means the two strands areconnected: one informs, explains or tests the other, and the methodology chapter states the sequence and the reason. The common patterns are straightforward:

• Qualitative first, then quantitative. Interviews identify the relevant factors; a survey measures how widespread they are. Use when the field is underexplored.
• Quantitative first, then qualitative. A survey finds an unexpected pattern; interviews explain why it occurs. Use when you have numbers that need interpreting.
• Both at once, then compared.The strongest design and the most work; rarely realistic in a bachelor's timeframe.

The failure mode is the undeclared version: a small survey and a few interviews with no stated relationship, each too thin to stand alone. That is not mixed methods, it is two undersized studies sharing a word count — and examiners recognise it immediately. Unless you can write one sentence explaining what the second strand does that the first could not, drop it and do one thing properly.

Choosing something you can finish in a semester

Feasibility is a methodological criterion, not a compromise, and saying so plainly in your chapter is a sign of competence rather than an admission. Some blunt guidance:

• Count the weeks backwards from submission and give data collection a hard deadline. Recruitment and distribution are what overrun, not analysis — and when they overrun, the time is taken from analysis, which is where the marks are.
• Prefer existing instruments and existing data. A validated questionnaire you adopt with citation saves weeks and strengthens the chapter. Publicly available datasets and document corpora are legitimate material and remove the access risk entirely.
• Narrow the question rather than the rigour. A small question answered properly beats a large one answered thinly, in every marking scheme.
• Have a fallback before you need one. If recruitment stalls, what is the document-based version of this study? Deciding that in week three costs an hour; deciding it in week ten costs the thesis.

Whichever way you go, the reasoning belongs in writing — the comparison you made, what the alternative would have given you, and what your choice cannot deliver. That is the substance of the methodology chapter, and we cover how to write it in How to Write the Methodology Chapter (Without Writing a Textbook).