Research transparency
Replication vs reproducibility: new evidence and repeatable analysis
Following the National Academies convention, reproducibility means obtaining consistent computational results from the same data, code, and methods; replicability means obtaining consistent findings with new data addressing the same question.
Educational content only. Not medical advice.
Reproducibility tests the analytical record
Under the National Academies definition, a result is reproducible when another analyst can obtain consistent computational results using the same input data, code, procedures, and conditions. This requires accessible data where lawful, executable code, software and package versions, random seeds, preprocessing steps, and a clear route from raw inputs to reported outputs.
Replication collects new evidence
Replication addresses the same scientific question with newly collected data. A close replication keeps methods similar to test whether the original finding recurs, while a conceptual replication varies methods or context to test the underlying claim. Results need not be numerically identical; consistency should be judged against expected sampling and measurement variation.
Robustness tests reasonable analytical or experimental choices
A conclusion is more credible when it persists across defensible model specifications, preprocessing choices, outcomes, measurement approaches, or relevant settings. This is not permission to try many analyses and report only favorable ones. The set of alternatives and decision rules should be transparent, with discrepancies treated as information about the claim's boundaries.
Failure to replicate has more than one explanation
Differences can arise from a false-positive original, a false-negative replication, hidden moderators, material or assay changes, low precision, publication bias, protocol drift, or an effect that is context-dependent. A careful response compares designs, estimates, uncertainty, and execution rather than treating one result as automatic proof of fraud or one replication as a final verdict.
Evidence limits
- Different fields use replication and reproducibility with partly reversed terminology.
- Consistent computational output does not prove that the original data or design were unbiased.
- Replication outcomes require interpretation of precision, context, and expected heterogeneity.
Sources and further reading
These sources ground the definitions and evidence boundaries on this page. A citation is a route for verification, not an endorsement of a product or personal use.
National Academies of Sciences, Engineering, and Medicine
Reproducibility and Replicability in Science
Consensus report defining computational reproducibility, replicability, uncertainty, and transparent methods.
Open sourceNational Institutes of Health
Enhancing Reproducibility through Rigor and Transparency
Official NIH principles for rigorous design, authentication, transparency, and reproducible biomedical research.
Open sourceNational Institutes of Health
Clinical Trial Reporting Requirements
Official NIH policy context for registration, results reporting, and transparency.
Open sourceCommon questions
Why do some sources define the terms differently?
Field conventions vary; this page explicitly follows the National Academies definitions.
Can a reproducible analysis still be wrong?
Yes. Code can consistently reproduce an analysis based on biased data, invalid measurements, or flawed assumptions.
Does one failed replication disprove a finding?
Not automatically. Compare design, precision, execution, context, and the combined evidence.
Continue with context
Keep building your evidence-reading skills
Explore the public PeptideSchool research library for more source-backed methods, glossaries, and evidence maps.