Research Design Selection and Matching Design to Research Question

Core Idea

Different research questions require different research designs, and no single design is universally superior; rather, designs differ in strengths regarding internal validity, external validity, practical feasibility, and cost. Experimental designs best answer questions about causation; correlational designs examine relationships and predictors; descriptive and qualitative designs explore phenomena. Researchers must consider fundamental trade-offs: experimental designs sacrifice external validity and ecological authenticity for internal validity and control; field studies gain external validity while losing control. Matching design to research question ensures the study can adequately address the question with appropriate evidence.

How It's Best Learned

Select several different research questions and determine which design(s) would be most appropriate for each, justifying design choices based on the specific question.

Common Misconceptions

Experimental designs are always superior to other designs (actually, experiments cannot answer all types of research questions). A poorly designed experiment is better than a well-designed correlational study (actually, design appropriateness depends on the research question).

Explainer

Your prerequisites on experimental, correlational, and survey designs each introduced a specific tool. This topic steps back to address the prior question: given a research question, how do you choose the right tool? The answer turns on two fundamental properties that every design trades off: internal validity — the degree to which you can attribute your findings to the variable you manipulated rather than to confounds — and external validity — the degree to which your findings generalize beyond the specific sample, setting, and conditions of your study.

Experimental designs maximize internal validity by randomly assigning participants to conditions and controlling all other variables. If you randomly assign people to read either fear-arousing or neutral health messages and then measure their intentions to exercise, you can confidently attribute any difference to the message content — random assignment equates the groups on everything else. But the lab setting may be artificial, your sample may be a convenience sample of undergraduates, and the measured intentions may not reflect real behavior. The tight internal control that makes the causal inference possible is often purchased at the cost of ecological realism. This isn't a flaw to apologize for — it's the correct tradeoff when your question is "does X cause Y?"

Correlational designs sacrifice causal claims but gain breadth and naturalism. Measuring conscientiousness and job performance in real employees at real organizations tells you something about how these variables co-occur in the world — that's external validity. But any observed correlation could reflect a third variable: maybe high-conscientiousness people also come from higher socioeconomic backgrounds that provide better jobs, and the SES is the actual driver. Correlational designs are the right choice when you want to characterize relationships in the natural world, when random assignment is impossible (you can't randomly assign people to be extroverted or to have experienced childhood trauma), or when you want to build predictive models rather than test causal theories.

Survey and descriptive designs are best when the research question asks "what is the distribution or prevalence of X in a population?" rather than "does X cause Y?" or "are X and Y related?" Prevalence of depression, distribution of political attitudes, frequency of health behaviors — these are answerable by surveys with careful sampling and not by experiments. Qualitative designs serve yet a different purpose: when you don't yet know what categories or variables matter, and need to discover the structure of a phenomenon before you can measure it, qualitative methods generate hypotheses rather than test them.

The practical skill is translating a research question into design requirements. Start by identifying the strongest form of your question: is it a causal claim, a relational claim, a prevalence claim, or an exploratory mapping? Then ask: what kind of evidence would definitively answer this question? What are the ethical and practical constraints on data collection? If a true experiment is feasible and the question is causal, run the experiment. If random assignment is impossible but causal inference is still needed, consider quasi-experimental designs with matched comparison groups. If the question is genuinely descriptive or exploratory, don't force an experimental frame onto it — you'll introduce constraints that produce artificial answers to a question nobody was asking. Good research design is the art of choosing the method that fits the question, not the method that sounds most rigorous in the abstract.