A good research question names a specific population, a single primary exposure or intervention, a comparator, and measurable outcomes, then states the direction and size of the effect you expect. The strongest questions are also answerable with feasible data, falsifiable, and narrow enough that one study can resolve them. Write yours as a structured sentence first, test it against those four criteria, and only then decide on design and analysis.

The anatomy of a testable question

A research question is not a topic. "Inflammation and depression" is a topic; it cannot fail. A question becomes scientific only when it makes a prediction that data could contradict. That requires four components: the unit of analysis (who or what you study), the contrast (what varies), the outcome (what you measure), and the expected relationship (direction and magnitude).

The PICO frame formalizes this for clinical work: Population, Intervention, Comparator, Outcome. For observational science, swap Intervention for Exposure and you get PECO. The frame matters because it forces you to commit to a comparator. A question without a counterfactual—"what happens to patients who receive drug X?"—cannot be answered causally, because there is nothing to compare against.

One level deeper, a question is a model of variation. You are claiming that change in X produces detectable change in Y, net of Z. Each component you specify constrains the design: a rare outcome demands a larger sample or a case-control structure; an exposure you cannot randomize demands adjustment for confounding; an outcome measured with error demands attention to reliability. Writing the question is therefore the first act of study design, not a preamble to it.

A practical test: could two competent researchers read your question and independently sketch the same data collection plan? If not, it is still a topic.

What makes a question good, quantified

The best evidence on question quality comes from systematic review methodology. Investigators at McMaster University and elsewhere developed the FINER criteria—Feasible, Interesting, Novel, Ethical, Relevant—and empirical checks exist for several components. A pilot or feasibility study typically needs enough participants to estimate event rates with reasonable precision; a rule of thumb for a proportion is that 12 or more observations begin to stabilize an estimate, though this is a heuristic, not a threshold with regulatory force.

Novelty is countable. Before committing, search PubMed or Web of Science for your exact constructs and count systematic reviews on the topic. If three or more recent reviews already answer your question, you need a different angle—a new population, a new comparator, or a mechanism the reviews could not test. If none exist, check whether primary studies are too sparse to synthesize; both extremes are informative.

Relevance has a quantitative shadow: the expected effect size relative to the smallest effect size of interest. If your intervention is expected to shift the outcome by less than the measurement's minimal clinically important difference, the question is technically answerable but practically trivial. State the smallest effect you would care about before you collect data.

Feasibility is where most questions die. Estimate your recruitment rate from comparable studies, multiply by your enrollment window, and subtract expected attrition. A common planning assumption is 10 to 20 percent loss to follow-up for short studies, higher for longitudinal cohorts. If the resulting number falls below your sample-size requirement, the question must shrink.

The FINER and PICO comparison

Framework Components Best for Main strength Main risk
PICO Population, Intervention, Comparator, Outcome Clinical trials, systematic reviews Forces an explicit counterfactual Can feel rigid for qualitative work
PECO Population, Exposure, Comparator, Outcome Observational epidemiology Handles non-randomizable exposures Confounding must be addressed by design
FINER Feasible, Interesting, Novel, Ethical, Relevant Early-stage question screening Screens practicality before design No outcome specification
PICOT PICO plus Timeframe Studies with time-dependent outcomes Clarifies follow-up window Adds a parameter you must justify
SPIDER Sample, Phenomenon of Interest, Design, Evaluation, Research type Qualitative and mixed methods Fits non-numeric evidence Less familiar to clinical reviewers

Use PICO or PECO to write the question, then run FINER as a gate. The frameworks are complementary, not competing. A question that passes PICO but fails feasibility is not a research question; it is a grant rejection waiting to happen.

Where good questions go wrong

The most common failure is the compound question: two outcomes, two populations, or two exposures joined by "and." "Does exercise and sleep quality affect cognition in older adults?" cannot be answered by one study, because a null result on either factor is ambiguous. Split it. Each question gets one primary outcome; secondary outcomes are supporting cast, not co-stars.

A subtler failure is the unfalsifiable hedge. Questions phrased as "explore," "understand," or "examine the relationship" often lack a stated prediction and therefore cannot be wrong. Exploration is legitimate in qualitative work, but then the question should name the phenomenon, the setting, and the analytic approach.

Watch for these red flags before you finalize:

Two or more primary outcomes, with no stated hierarchy among them. A comparator that is implicit ("compared to usual care" without defining usual care). An outcome that cannot be measured with the instruments and budget you have. A population so broad ("adults") that confounding swamps any plausible effect. A novelty claim you have not verified with a literature search. A time frame that is unstated, so the outcome's meaning shifts with follow-up length.

A professional caveat: peer reviewers and ethics boards increasingly ask for a preregistered primary outcome. If your question cannot be stated as a single primary outcome with a direction, it will struggle at both gates.

From question to protocol: the practical steps

Write the question as one sentence using PICO/PECO components. Then run this sequence.

Draft the sentence in plain language first, then restructure it into the frame. If the plain version is muddled, the structured version will be too. Search for existing systematic reviews on the exact constructs. Record the count and dates. Estimate your expected effect size from the closest comparable study, and compare it to the smallest effect you would care about. Compute a rough sample size using the expected effect, your alpha (conventionally 0.05), and power (conventionally 0.80). Adjust upward for attrition. Write the primary outcome and its measurement instrument before anything else in the protocol. If you cannot, the question is not ready. Have one methodologist and one domain expert read the sentence independently and state what data they would collect. Divergence means ambiguity.

Step 4 deserves a threshold: if your feasibility calculation yields fewer than roughly 80 percent of the required sample, revise the question rather than the power target. Lowering power to fit recruitment is how underpowered studies are born.

Strengths, limits, and what the frames cannot do

Structured frames improve clarity and comparability, but they do not generate good questions. A PICO sentence can be perfectly formed and still boring, redundant, or ethically impossible. The frames also privilege quantifiable outcomes, which can crowd out questions about meaning, process, and experience—territories where SPIDER and qualitative designs serve better.

Nor do frames settle the hardest judgment: how narrow is narrow enough. Too narrow, and the question answers itself or cannot recruit; too broad, and the study drowns in confounding. There is no formula, only the feasibility and effect-size checks above. Finally, a well-formed question does not guarantee a well-conducted study; design, measurement, and analysis can each fail independently.

The honest limit: question quality is judged partly in hindsight. A question that seemed precise can prove unanswerable once data arrive, and a modest question can yield a decisive result. Write the best question your evidence and resources allow, preregister it, and treat revision as a legitimate part of the process rather than an admission of failure.