The practical ways to collect data, what each one costs in time and access, and how to choose between them without wasting a year finding out.
Research methods are the techniques you use to collect data: interviewing, surveying, observing, experimenting, or working with material that already exists. Choosing between them is a practical decision constrained by three things people underestimate — who will actually give you access, how long collection really takes, and what your ethics committee will approve. The method that fits your question but cannot be delivered inside your funding is not the right method.
A method is a technique: you ran interviews, you distributed a questionnaire, you sat in a classroom for six weeks. A methodology is the argument that this was the right technique for your question, and what it entitles you to conclude.
The two are examined separately. A perfectly executed set of interviews sitting inside an unjustified design will still be picked apart, because the challenge is not whether you interviewed well — it is why interviewing was the way to answer this. So decide the method with the justification in mind, rather than choosing what feels manageable and constructing the reasoning afterwards.
What follows is the toolbox, with the practical costs stated, because those are what actually decide most doctoral projects.
These give you accounts: what people say they do, think, or experienced. That is genuinely valuable, and it is not the same as what they do — a distinction worth keeping visible in your claims.
Structured interviews ask everyone the same questions in the same order, which makes comparison clean and rules out following anything unexpected. Semi-structured interviews work from a guide but allow pursuit of what emerges — the default for most doctoral qualitative work, and the right default. Unstructured interviews approach a conversation and demand real skill to keep useful.
What it costs. The interview itself is the cheap part. Transcription is the expensive part: an hour of recording commonly takes several hours to transcribe carefully, and automated transcription still needs correcting against the audio, particularly with accents, overlapping speech or technical vocabulary. Budget for that honestly, because it is the single most common source of timeline overrun in qualitative doctorates.
Where they go wrong. Questions that carry the answer inside them. Following your guide so rigidly that you miss what the participant was actually trying to tell you. And not piloting — two practice interviews will reveal which of your questions nobody understands, while it is still free to fix.
Useful when the interaction itself is the data: how a view is defended, challenged or shifted when others are present. Not a time-saving way to interview several people at once — that misunderstanding produces poor data and a hard question in the viva.
What it costs. Scheduling is the difficulty; getting six busy people into one room at one time is harder than six separate interviews. Transcription is also harder, because attributing overlapping speech to the right person is slow.
Where they go wrong. One dominant participant setting the tone for everyone. Sensitive topics where the group setting suppresses exactly what you needed to hear. Analysis that treats the transcript as several interviews rather than as an interaction.
The way to reach numbers you could not reach in person, and to measure things comparably across people.
What it costs. Design time, mostly. A questionnaire is a measuring instrument, and writing good items is a skill: every ambiguous question produces noise you cannot remove later. Using a published, validated instrument where one exists saves weeks and gives you evidence to cite.
Where they go wrong. Double-barrelled items asking two things at once. Response options that do not cover the possibilities. Deploying to a convenience sample and then writing as though it were representative. And no pilot — twenty pilot responses will show you which items are being read differently from how you intended.
These give you behaviour rather than accounts of behaviour, which is why they are worth the difficulty when the gap between the two is part of your question.
Ranges from sitting quietly with a structured checklist to full participation in the setting over months. The choice of how visible and how involved you are is a methodological decision, not a practical one, and has to be argued.
What it costs. Time and access, both heavily. Negotiating entry to an organisation can itself take months, and gatekeepers may attach conditions that shape what you can see.
Where it goes wrong. Underestimating your own effect on the setting — people behave differently when watched, and the honest response is to account for it rather than claim it did not happen. Fieldnotes written up days later, by which point the detail has gone. And no clear rule about what you are recording, which produces pages of material with nothing systematic in it.
The only designs that support confident causal claims, because random allocation is what removes the alternative explanations. Where randomisation is impossible, a quasi-experimental design with a carefully argued comparison group is the next best thing, and the argument about how the groups differed beforehand becomes central.
What it costs. Control, recruitment and ethics. You need enough participants for the comparison to be meaningful, which should be calculated in advance rather than discovered afterwards.
Where they go wrong. Conditions that differ in more than the one thing you intended, so the comparison cannot isolate anything. Attrition that is unequal between groups, which quietly destroys the equivalence randomisation gave you. And a setting so controlled that the finding does not transfer to anywhere real — worth acknowledging rather than hoping nobody raises it.
Consistently underrated in doctoral work, and often the sensible choice when access to people is the binding constraint.
Large surveys, administrative records, published datasets and repositories give you sample sizes no doctoral student could collect alone, often with sampling far better than anything you could achieve.
What it costs. Understanding someone else’s data properly. The documentation matters enormously: how variables were defined, who was excluded, what the response rate was, how missing values were handled. That reading is real work, and skipping it produces analysis built on assumptions you never checked.
Where it goes wrong. The dataset almost answers your question, so the question quietly bends to fit the dataset. Sometimes that is a reasonable trade; it should be a decision you made, not a drift you did not notice.
Policies, minutes, reports, media, correspondence, records. Particularly strong for questions about how something was justified, framed or changed over time.
What it costs. Locating and sampling. “All the documents” is rarely achievable, so the selection rule has to be explicit and defensible.
Where it goes wrong. Treating documents as neutral records of what happened. They were produced by someone, for an audience, for a purpose — and that is often the most interesting thing about them.
| If you need… | Consider | Main practical constraint |
|---|---|---|
| Depth on how people understand something | Semi-structured interviews | Transcription time |
| How views are formed and contested between people | Focus groups | Getting everyone in one place |
| Comparable measurement across many people | Survey with a validated instrument | Response rates, and who does not reply |
| What people actually do, not what they report | Observation | Access, and months of it |
| Whether something causes something else | Experiment or quasi-experiment | Control, numbers, ethics approval |
| Scale you could never collect yourself | Secondary data | Learning the dataset properly |
| How something was framed or justified over time | Document analysis | Defining what counts as the corpus |
Three practical filters are worth applying before you commit, because they eliminate more options than methodological reasoning does:
Using more than one method is common and often sensible. Two distinctions are worth keeping straight.
Using several methods within one strand — interviews plus documents, both analysed qualitatively — is usually about strengthening confidence by approaching the same thing from different angles. Combining strands with different logics, so numbers and accounts answer one question together, is a mixed-methods design and carries its own requirement: the strands have to be integrated rather than reported one after another.
Either way, the question to ask of a second method is simple: what does it let me say that the first one did not? If the answer is “it makes the study look more thorough”, it is costing you months for nothing.
Methods are the techniques you use to collect and analyse data — interviews, surveys, observation, experiments. Methodology is the reasoning that justifies those choices and sets out what your findings are entitled to claim. Examiners assess them separately: a well-run set of interviews inside an unjustified design will still be challenged, because the question is not whether you interviewed competently but why interviewing was the right way to answer your question.
Start from what the question requires: depth of understanding points to interviews, comparable measurement to surveys, actual behaviour to observation, causal claims to experimental designs, and scale to secondary data. Then apply three practical filters that eliminate more options than theory does — whether you genuinely have access, how long collection will really take once recruitment and transcription are counted, and what your ethics committee is likely to approve.
There is no fixed number, and it depends on how narrow your question is, how similar your participants are and which design you are using. What matters more for planning is the workload each one brings: an hour of recording typically takes several hours to transcribe carefully, and that arithmetic is what makes ambitious interview targets slip. Justify the number by what the design needed rather than by a figure quoted from elsewhere.
Yes, and it is often a strong choice — existing datasets offer sample sizes and sampling quality no individual researcher could achieve. The originality then comes from your question, your framework and your analysis rather than from having collected the data. The work that replaces collection is understanding the dataset properly: how variables were defined, who was excluded, and how missing data was handled.
Yes, and it is among the cheapest insurance available. Two practice interviews or twenty pilot survey responses will show you which questions are ambiguous, which produce nothing useful and how long the whole thing actually takes. Problems found at the pilot stage cost an afternoon; the same problems found after full collection usually cannot be fixed at all.
Most collection plans fail on access, timing or ethics rather than on method. Describe what you intend to collect and from whom, and a PhD in your field will tell you where it is likely to bind.
Leave an email or a WhatsApp number — whichever you prefer — and tick how we should reach you. We reach out within 30 minutes.
