Types of research design: which one fits your research question

What are the main types of research design?

A design choice is really three decisions taken one after the other. The first one is the family, meaning quantitative, qualitative or mixed methods. It decides your evidence, either numbers you count and test or words and observations you interpret. The second decision is the specific design inside that family. A descriptive survey and a randomised experiment both come under quantitative, but the work in each is very different. The third decision is about time. You collect data in one round (cross-sectional) or in several rounds (longitudinal).

Many times you will also see research labelled by its purpose. Exploratory research takes a problem which is not well understood and finds out what the key variables are. Descriptive research reports what exists. Explanatory research asks why, and it does that by testing links or causes. These purpose labels come in addition to the design, so you still have to name the design separately. Some synopsis drafts I have seen call one study “exploratory” on page 2 and “explanatory” on page 9. Better you stay with one label.

Qualitative vs quantitative research: which does your question need?

Look at the question word in your research question first. Mostly that word itself decides the family.

Quantitative research answers how much, how many, how strongly related and what effect. For this you need variables measured in the same way for every participant. The sample has to be large enough for statistical tests. And usually you state hypotheses in advance.

Qualitative research is for how and why questions, where you want to understand meanings and processes and what people experienced. Here the samples are small and selected on purpose, because the aim is depth.

One family is not more scientific than the other. Where scholars go wrong is choosing the family first and then bending the question to suit it. Maybe the department always does surveys, or the scholar is scared of statistics. Objectives still not clear? Our guide on research objectives and questions will help you fix them first.

Descriptive vs experimental: the quantitative designs compared

The five quantitative designs differ in how much control you have over the cause. So the causal claim you can make also differs in strength.

Descriptive

Here you measure things as they are. Awareness surveys and prevalence studies come under descriptive, and in Indian commerce and social science theses this is the commonest design. There is no issue with it when the question is descriptive. The trouble comes only when the scholar starts writing about “impact” in chapter 5.

Correlational

You measure several variables in the same group of people and test how they are related. Correlation, regression or structural equation modelling are the usual tools. Take social media use and anxiety, and suppose your data links them strongly. Still you can’t tell which came first, or whether exam pressure is behind both of them.

Causal-comparative (ex post facto)

Here the groups were different before you came in. For example rural and urban students, men and women, or first- and second-generation entrepreneurs. You compare these groups on some outcome. Since the groups were not formed by you, the result may be due to some other difference between them.

Quasi-experimental

There is an intervention, but you can’t randomly decide who receives it. Think of a state scheme that comes to some districts first, or a new teaching method a college is trying in only two sections. The standard reference is Shadish, Cook and Campbell’s Experimental and Quasi-Experimental Designs for Generalized Causal Inference (2002).

Experimental

Here you change the cause yourself and you assign people at random. Only then can you credit the difference to the treatment. In health research this is the randomised controlled trial. In India a trial must be registered with the Clinical Trials Registry – India before the first participant is enrolled. Since 1 April 2018, CTRI has required prospective registration for all studies.

So in descriptive vs experimental, the descriptive study gives you the picture as it is now, and the experiment tells what happens after you change something. To find the test that goes with each design, see choosing a statistical test.

What are the types of qualitative research design?

Creswell and Poth’s Qualitative Inquiry and Research Design (5th edition, 2023) puts qualitative work into five approaches. Most examiners are familiar with this set.

Case study. You take one bounded case, or a few, and study it in depth using several sources. “Self-help groups in India” is too wide to be a case. “Three SHG federations in one block of Nalanda district” is a case.

Phenomenology. The lived experience of something the participants share, like caring for a parent with dementia. You talk to few people, and the interviews are long.

Grounded theory. From the data you build a theory of some process. The approach comes from Glaser and Strauss’s The Discovery of Grounded Theory (1967). Many scholars are putting this label on any interview study. Please don’t use it unless you are actually building a theory.

Ethnography. You stay inside a group’s daily life for a long time. The work is slow and hard, and it mostly suits anthropology and sociology.

Narrative research. Here you collect the life stories of one or a few people and read them for sequence and meaning.

Many good researches in India just use interviews analysed by theme, and they don’t claim any of the five. If yours is like that, say so honestly. Our thematic analysis guide explains that analysis, and semi-structured interviews explains how to collect the data.

How do you design a case study?

Scholars claim case study more often than any other design, and they plan it the least. The standard reference is Robert K. Yin’s Case Study Research and Applications: Design and Methods (6th edition, SAGE, 2018). As per Yin, case study fits when your question is a how or why question and the events are current ones. You should also have little or no control over those events. Unlike the other four approaches above, a case study can use numbers along with words. That can mean records, performance data, even a small survey done inside the case.

Single or multiple cases. If you take a single case, you need a reason for it. Yin gives several reasons. A critical case is one that tests a theory. Extreme or unusual cases are another reason, and a common or typical case is also accepted. Then there is the revelatory case, which researchers could not reach earlier. Or it can be a longitudinal case, studied at two or more points in time. For multiple cases Yin’s logic is replication, and sampling logic has no role in picking them. Each case should give either similar results (literal replication) or contrasting results for reasons you can predict (theoretical replication). For a PhD, three or four well-chosen cases usually work better than ten thin ones.

Holistic or embedded. In the first option you study the case as one whole, for example a district hospital’s response to a new referral policy. An embedded design adds units of analysis inside the case, say three wards of that hospital, each with its own data. You get more depth from embedded units, but at the end your thesis has to come back to the case as a whole. When you combine the two choices, you get Yin’s four types, which are single holistic, single embedded, multiple holistic and multiple embedded. Mention which one is yours in chapter 3.

Defining and selecting the case. Write down where the case starts and ends. Which organisation or programme is it? Which years? Which people are inside the case and who is outside? After that give the reason for picking these cases. Many drafts are just giving “the company agreed to give access” as the reason. That is only about access. It says nothing on why this case is useful, so mention what makes each case critical, typical or contrasting and write about access separately.

Data sources and triangulation. Yin lists six sources of evidence. These are documents, archival records, interviews, direct observation, participant observation and physical artefacts. A strong case study takes several of these and checks if they point in the same direction. That checking is called triangulation. Maintain case study database with transcripts, field notes, documents and coded extracts. Keep a chain of evidence also. With that the examiner can trace any claim in your findings to its source.

Generalisation. A case study does not generalise statistically to a population. From three firms you cannot say what most Indian MSMEs do. The generalisation here is analytical. Your findings support, refine or challenge a theory or some propositions, and later studies can test those. Write your conclusions this way and the examiner has less to object to.

When a case study is the wrong choice. For questions of how many, how common or how strongly related, a survey is the design to use. To show that an intervention caused an outcome, go for an experimental or quasi-experimental design. If you don’t have regular access to documents and people inside the case, what you end up with is a few interviews carrying a case study label. And if your “case” is actually a topic, like green banking in India, then it is not a case yet.

What are the mixed methods research designs?

In mixed methods you collect both quantitative and qualitative data, and you bring the two together deliberately. Two open-ended questions added at the end of a questionnaire are not enough to call it mixed methods.

Creswell and Plano Clark’s Designing and Conducting Mixed Methods Research (4th edition, SAGE, 2025) still keeps the three core designs from the earlier edition.

Mixed methods takes roughly the work of two studies, and you have to be able to defend both the analyses. I’d choose it only when one type of data really can’t answer the question. In management theses the exploratory sequential design is common, mostly where there is no scale for the Indian setting. For the second phase, our questionnaire design guide helps.

Cross-sectional vs longitudinal design: when does time matter?

A cross-sectional study collects data one time, or within a short period. Levin wrote about cross-sectional studies in Evidence-Based Dentistry (2006), and he notes that they are usually used for estimating how common something is. Most PhD surveys are cross-sectional.

A longitudinal study takes measurements more than once. In a panel the same people are followed, and in a repeated cross-section a new sample is drawn every round. Change can be shown only with longitudinal data.

Be realistic with yourself about fieldwork time. A three-wave panel can be done, but participants drop out between the waves. So begin with a larger sample (our sample size guide helps here). Many times the easier route is to use panel data that already exists from national surveys.

How do you choose a research design for a PhD?

Go through this list before the synopsis is sent to your DC or RAC.

• Write the main research question in one sentence and underline its question word, like how many, is there a relationship, does it cause, how or why.
• Decide which claim the thesis has to make, whether description, association, cause or understanding. Then take the simplest design that can support that claim.
• Check if you control who gets the intervention. When you don’t have that control, a true experiment is not possible for your study.
• Think whether time matters for the answer. If yes, plan to measure more than once, and if one time point is enough, write the reason down.
• Check that you can actually reach the people, records or sites within your registration period.
• Open a few recent theses from your area on Shodhganga and see which designs the examiners there have accepted.

A worked example

Take an illustrative topic, “the impact of smart classrooms on learning in Tamil Nadu government schools”.

Using the word “impact” means you are claiming a cause. Many first drafts propose a survey of teachers’ opinions for this, but opinions cannot support that claim. You also can’t assign schools at random. The design that can work here is quasi-experimental. You compare test scores of schools with smart classrooms and schools without them, before and after the classrooms were installed, if the records exist.

If the records are not there, change the question. “How are teachers using smart classrooms, and what gets in the way?” can be a sound qualitative or mixed methods thesis. But keeping “impact” in the title while you collect only opinions is the thing that gets picked apart at the viva.

Once the design is fixed, the next step is sampling techniques. If the design is still not decided, many times the real issue is an unclear problem. In that case go back and look at the statement of the problem.