Learn · Research methods
A study design tells you which question was tested—not whether the answer is automatically right.
Randomization, comparison groups, blinding, preregistration, analysis, and replication each address a different source of uncertainty. Read them separately before deciding what a result means.
Start with the question
Different designs limit different biases.
A design creates opportunities to answer a question. Execution, missing data, analysis choices, reporting, and replication determine how much confidence the result deserves.
Can a controlled change alter a process?
Useful for mechanism and measurement. A controlled system can omit the complexity, exposure, duration, and tradeoffs present in a whole organism.
What happens in a living model?
Can test integrated biology and long follow-up. Model fit, dose, outcome, and species differences limit translation to people.
What travels together in people?
Can reveal patterns in real populations. Confounding, selection, measurement error, and reverse causation can remain.
What changed under an assigned comparison?
Randomization can balance some causes of bias. It does not guarantee good execution, complete follow-up, meaningful endpoints, or applicability.
What does a defined body of studies show?
A transparent search can reduce cherry-picking. Its conclusion still depends on the included studies, synthesis choices, and publication record.
Does the signal survive a new test?
Repeated results can strengthen confidence, but only when the new population, method, outcome, and independence are understood.
Keep the fields apart
Ten judgments, not one score.
- Study designHow participants, models, comparisons, and interventions were assigned or observed.
- Population or modelWho or what was studied, including eligibility, setting, and relevant differences.
- Evidence maturityHow directly the evidence addresses the population and outcome readers care about.
- ExecutionProtocol adherence, blinding, missing data, measurement, and analytic choices.
- Publication statusPreprint, peer review, correction, withdrawal, or unavailable status—not correctness.
- Endpoint typeBiomarker, function, disease outcome, healthspan, mortality, or lifespan.
- Development stageDiscovery, preclinical, trial phase, approved use, off-label use, or unknown.
- Replication stateNot assessed, preliminary, replicated, conflicting, or unavailable.
- Conflicts and fundingWho supported the work and which interests readers should see.
- Practical relevanceWhether the size, duration, population, outcome, safety, and uncertainty matter to a real decision.
A six-question reading check
Move from headline to method.
- What was the exact question?
Name the population or model, exposure or comparison, outcome, and time window.
- What could the design limit?
Look for assignment, comparison groups, blinding, preregistration, and prespecified outcomes.
- What happened after the design?
Check adherence, attrition, missing values, deviations, analysis, and selective reporting.
- What was actually measured?
Do not translate a biomarker into function, healthspan, mortality, or lifespan.
- Was it reproduced?
Publication and statistical significance are not replication or certainty.
- Who does this apply to?
Compare the studied population, setting, exposure, duration, and outcome with the claim being made.