The research question determines the method, not vice versa

What it is

Research question–method fit means specifying the kind of knowledge needed before choosing how to collect evidence. Locating where errors arise calls for observation of task processes; estimating prevalence requires a defined sampling frame and denominator; attributing change to a redesign requires comparison and control of rival explanations. Deciding to “run interviews” or “send a survey” first, then reshaping the question around that instrument, substitutes organizational convenience for an epistemic requirement.

Why it happens

Every method is a selective observation device. Interviews surface meanings participants can recall and articulate; telemetry retains events instrumented in advance; controlled experiments use manipulation and comparison to identify effects. Each preserves some information and systematically loses other information. Method-first planning lets available tools narrow the inquiry: an available scale turns experience into ratings, while available logs make uninstrumented behavior appear absent. Moving from question to required evidence and only then to method exposes those losses.

Studying it

Classify the inference as descriptive, interpretive, predictive, causal, or generative, then specify its unit of analysis and required counterfactual. A causal question needs an account of what would occur without exposure, often through randomization or a credible quasi-experiment. A mechanism question needs process evidence and rival explanations. A prevalence question needs a sampling frame and denominator. A question–evidence–method–threat matrix can document what each candidate observes, misses, and biases. Mixed methods add inferential value only when each source has a defined role and an integration rule.

Where it stops holding

Time, budget, ethics, and access constrain ideal methods; question-driven selection does not erase those constraints. The defensible response is to narrow the question or weaken the claim, rather than retain a broad claim and pretend that a cheap method answers it. New instruments and found data can legitimately generate interpretive or generative questions. Independent data, held-out evidence, or another control for selection becomes necessary when the work advances a confirmatory, stable predictive, or generalizable claim. Reversible, low-risk decisions may need directional evidence rather than the strongest causal design; residual uncertainty should remain explicit.

Applying it

• Before naming an instrument, state what the team must know and which contrary result would change the decision.
• Select the inference type, then the evidence form, then the method. Treat convenient recruitment as a feasibility constraint, not the target population.
• Give the preferred method one explicit blind spot and decide whether to narrow the claim or add complementary evidence.
• During review, hide method names and ask whether the proposed data can support the conclusion; then inspect whether familiarity with a tool displaced the actual evidence need.

Related

• Same group: Q1.01.1 Research questions must be answerable with evidence · Q1.01.2 “Do users like it?” needs decomposition
• Adjacent: Q1.03 Choosing qualitative and quantitative methods · Q1.10 Three-dimensional classification of research methods
• Search terms: question–method fit · method selection · inference goal