Different research questions require different designs. Experiments test causality through manipulation and random assignment; quasi-experiments approximate causal inference without full randomization; correlational studies examine relationships without manipulation; qualitative studies explore mechanisms and experiences. The logic of each design—what it can and cannot conclude—flows from its structure.
Map research questions to designs: 'Does X cause Y?' (experiment); 'Are X and Y related?' (correlational); 'How do people experience X?' (qualitative). Compare how the same question answered via different designs yields different insights.
You know from your prerequisites that empirical questions are formulated with measurable variables, and that operationalization defines how those variables will be captured. But having a well-formed question and well-defined variables doesn't specify the design — it only opens the door. Research design selection is the process of asking: given what I want to conclude, what logical structure does the study need to support that conclusion? The answer flows from the type of question, not from preference or habit.
The key insight is that different question types have different minimum logical requirements. "Does X cause Y?" requires manipulation — you must assign participants to X versus not-X conditions — and random assignment, which ensures the groups are equivalent before the manipulation. Without both, you cannot rule out that some third variable explains any observed relationship. "Are X and Y related, and how strongly?" requires measuring both variables in the same participants but doesn't require manipulating either; correlational designs serve this question. "How do people experience X?" calls for rich qualitative data — open-ended interviews, narrative description — that quantitative measurement would suppress. Each design type is the right tool for its corresponding question and the wrong tool for others.
The practical skill is ruling out alternative explanations. When a design is selected, the question becomes: what confounds remain open? Correlational designs leave third-variable confounds open — maybe a lurking variable causes both X and Y. Experimental designs close those confounds via random assignment but cannot be used to study variables that cannot be ethically or practically manipulated (you cannot randomly assign people to poverty, abuse, or genetic conditions). Quasi-experimental designs (with a comparison group but without random assignment) partially address confounds but leave some open. Qualitative designs produce deep mechanistic understanding but not quantifiable generalizations. No design closes every alternative explanation; the task is choosing the one that closes the alternatives most relevant to the specific inference you want to draw.
Matching design to question also requires practical judgment: available sample size, feasibility of manipulation, time, resources, and ethical constraints. A researcher studying long-term effects of early adversity cannot run a controlled experiment; they work with longitudinal observational data, natural experiments, or quasi-experimental comparisons — the strongest design that is actually feasible. The goal is not the most sophisticated design, but the simplest design that logically supports the intended inference given real-world constraints. Applying a more complex design than necessary doesn't improve validity; it adds cost and analytic complexity without corresponding gain. Starting from the question, identifying what it requires logically, and then working within constraints is the sequence that leads to well-matched designs.