Understanding research design
Research design is your plan for how you'll collect, analyze, and interpret data to answer your research questions. It must align with your field, your questions, and your access to resources. Choosing a design is not a formality to get past before the "real" writing begins. It is the decision that everything downstream depends on. Get it wrong, and no amount of careful writing in the results or discussion chapters can compensate, because the data itself won't be capable of answering what you asked.
Quantitative research design
Quantitative research uses numerical data and statistical analysis. Common in sciences, engineering, business, economics, psychology.
Characteristics
Quantitative designs typically rely on larger sample sizes than qualitative work, because the statistical tests underlying them need enough data points to detect an effect reliably. A small sample doesn't just weaken a quantitative study, it can make the entire analysis underpowered and effectively useless. They collect numerical data that can be statistically tested, whether that's survey responses on a Likert scale, test scores, physiological measurements, or financial figures. The goal is objective measurement: using instruments and procedures designed to minimize the researcher's own influence on what gets recorded, in contrast to the interpretive stance of qualitative work. Quantitative studies are typically built around hypothesis testing: a specific, falsifiable prediction stated before data collection, then tested against the data rather than derived from it afterward. And they aim for reproducibility: the expectation that another researcher following the same procedure on a similar sample would obtain broadly similar results, which is part of why quantitative methods sections demand such precise, step-by-step procedural detail.
Qualitative research design
Qualitative research explores meaning, context, and human experience through text, interviews, observations, and thick description. Common in sociology, anthropology, education, cultural studies.
Characteristics
Qualitative designs use small, purposeful samples: participants are deliberately selected because they can speak richly to the phenomenon under study, not randomly selected to represent a broader population statistically. The data itself is text or observational rather than numerical: interview transcripts, field notes, documents, images. Analysis is interpretive rather than purely mechanical. A qualitative researcher is actively constructing meaning from the data, which is why reflexivity and an audit trail (see our qualitative dissertation help guide) matter so much for trustworthiness. Qualitative work prioritizes depth over breadth, favoring a small number of cases explored thoroughly over a large number explored superficially. And its findings are understood as context-dependent: transferable to similar settings through detailed description, but not statistically generalizable to a wider population the way a quantitative sample is.
Mixed methods combine quantitative and qualitative approaches. Use quantitative data to answer "how much" and qualitative to answer "why." Common in public health, education policy, organizational research.
Choosing your design
| Research Question | Best Approach |
|---|---|
| Does X predict Y? (correlation) | Quantitative |
| What does X mean to participants? | Qualitative |
| How much and why? | Mixed methods |
Need help designing your study?
Our dissertation advisors review your methodology and ensure it aligns with your research questions.
Common design pitfalls
- Misalignment: Quantitative questions paired with qualitative design (or vice versa)
- Scope creep: Mixed methods that tries to do too much without clear purpose
- Sampling problems: Sample size too small (qualitative) or biased (quantitative)
- Unclear procedures: Methods section too vague to be replicated
The Four Core Quantitative Designs. And When Each Fits
Once you've settled on a quantitative approach, the next decision is which of four core designs actually answers your research question, because each one supports a different kind of claim. Choosing the wrong one is the single most common reason a committee sends a proposal back with "your design doesn't support this conclusion" written in the margins.
Experimental design
An experimental design randomly assigns participants to conditions (treatment vs. control) and manipulates an independent variable directly. This is the only design that can support a causal claim, "X causes Y", because random assignment rules out most alternative explanations for a difference between groups. It fits research questions phrased as "does this intervention cause this outcome?" It is also the hardest design to execute in applied doctoral research, since random assignment is often impractical or unethical (you usually cannot randomly assign real students, employees, or patients to receive or not receive something beneficial).
Quasi-experimental design
A quasi-experimental design compares groups or conditions without random assignment. For example, comparing outcomes at two organizations, one of which already adopted a new policy and one of which didn't. It's the practical workhorse of applied dissertations because it allows a treatment-versus-comparison structure in real-world settings where randomization isn't possible. The trade-off is that it can only support a claim of association or a cautious "the intervention appears associated with," not a clean causal claim, because pre-existing differences between the groups can't be fully ruled out.
Correlational design
A correlational design measures two or more variables as they naturally occur and examines the statistical relationship between them, with no manipulation or grouping at all. It fits "is X related to Y" and "does X predict Y" questions. It cannot establish causation on its own. This is the single most commonly misused claim in doctoral quantitative chapters, where a candidate runs a regression and then writes as though the predictor "caused" the outcome. A correlational design supports language like "predicts," "is associated with," or "explains variance in," never "causes."
Descriptive design
A descriptive design summarizes characteristics of a population or phenomenon without testing relationships between variables. Frequencies, means, proportions. It fits exploratory questions like "what is the prevalence of X among this population?" It's a legitimate and sometimes appropriate choice, particularly for a first study in an under-researched area, but it is also occasionally chosen by mistake when a candidate actually has a relational question and simply hasn't framed it that way yet.
How to Justify Your Design Choice
A design justification that will actually satisfy a committee has three components, and most weak justifications are missing at least one of them.
- Fit to the research question: state explicitly what the question requires, a causal claim, an associative claim, or a description, and show that the chosen design is the type capable of delivering that.
- Feasibility given real-world constraints: acknowledge why a "stronger" design (e.g. a true experiment) wasn't feasible, rather than ignoring the possibility and hoping no one asks.
- Explicit comparison to at least one alternative: naming a design you considered and rejected, and why, is far more convincing than defending your chosen design in isolation. It shows the choice was deliberate rather than default.
Say what your design cannot claim, not just what it can. A correlational design section that states up front "this design cannot establish causation, and findings will be interpreted as associative relationships only" pre-empts the single most common committee objection before it's even raised.
Common Committee Pushback on Design. And How to Answer It
"Why didn't you use a true experimental design?"
Answer with feasibility and ethics, not convenience: explain specifically why random assignment wasn't possible or ethical in your setting (e.g. you can't randomly deny some employees access to a training program), and name the quasi-experimental or correlational alternative you chose instead, with its narrower but still valid claim.
"Isn't your sample too small/large for this design?"
This is answered with a stated rationale, not a number in isolation. A power analysis for quantitative designs, or a saturation argument for qualitative ones (see our statistics help guide for power analysis specifics).
"How do you know this design will actually generate usable data?"
This is best answered by referencing prior studies that used a similar design successfully on a similar population, or by describing a small pilot test of your procedure before full data collection begins.
"Your research questions and your design don't match."
When this comes up, it usually means the wording of the questions needs to change to match what the design can actually deliver, or the design needs to change to match what the questions are actually asking. We check this alignment first, before any other part of the methodology is drafted.
A Worked Example
A candidate wants to study whether a new onboarding program improves new-hire retention at a mid-sized company. True experimental design is ruled out immediately. The company won't randomly deny onboarding to half its new hires. A quasi-experimental design comparing hires onboarded under the new program against a historical comparison group (hires from the prior year, before the program existed) is chosen instead, with retention at 12 months as the outcome. The justification names the ruled-out experimental design and the ethical/practical reason, states the claim the design can support ("associated with," not "causes"), and reports a power analysis behind the sample size. That is a design a committee will approve on the first pass, because every choice is pre-justified rather than left for the committee to question.
How This Differs at the Doctoral Level
At the master's level, naming a reasonable design and describing it competently is often sufficient. At the doctoral level, committees expect the choice to be argued, including an explicit account of designs considered and rejected, and a precise statement of what kind of claim (causal, associative, descriptive) the chosen design licenses. A doctoral committee is far more likely to challenge overreaching language. A correlational study that talks about "impact" or "effect" as if causation had been established, because doctoral research is expected to model methodological precision for the field, not just answer the question at hand.