GCRI · Labs
Reproducibility and negative results
Science corrects itself only when failures are recorded: replications that did not reproduce, experiments that found nothing, errors found after publication. Many of these records never appear, so the published literature overstates effects and certainty. This page sets out how irreproducibility arises, how replication and correction work, which records exist, and what editors, integrity offices and funders rely on when they correct or retract.
Reproducibility and negative results on the record.
Reproducibility and negative results: how the risk develops.
Failure records and corrections. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.
Publication bias filters the record
Significant, novel results are more likely to be written up, accepted and cited than null ones. The literature then shows a biased sample of what was studied, and meta-analyses inherit the bias.
Errors propagate through citation
A flawed or retracted paper keeps being cited, often without mention of its status. Reviews, guidelines and later studies that build on it carry the error forward, sometimes long after the notice.
Reproduction fails for ordinary reasons
Missing code, undocumented software versions, unshared reagents, unclear methods and lost raw data stop others from repeating an analysis or experiment, even when the original work was honest.
Correction is slow and uneven
Investigations take time, journals differ in how they flag problems, and retraction notices can be vague. Readers may not learn that a paper was corrected, and databases may not carry the notice forward.
Protocol and replication board.
The screen a team would use to follow this subject: its measures, states and decisions. Names are examples and values are illustrative.
Protocol and replication board
Reproducibility and negative results · Labs · Quarter 1
Protocols
| PR-594 sensor drift | Field experiment | Yes | Approved | Running |
| PR-714 AI triage | Randomised | Yes | Conditions | Setup |
| PR-551 message recall | Survey experiment | Yes | Approved | Complete |
| PR-896 heat exposure | Cohort | No | Pending | Draft |
Replication R-20
- Original effect
- 0.3 SD
- Replication effect
- 0.3 SD
- Verdict
- Smaller, same sign
- Published
- with data and code
Testbed availability
Negative results matter in every domain.
Negative results are published in the Labs Evidence series. The National Academies’ report (2019) distinguishes reproducibility from replicability; DORA (2012) supports assessing research on its merits.
- Negative results.
- Replication attempts.
How reproducibility and negative results is modelled, measured and tested.
The model, its assumptions, its data and its uncertainty, in the form a scientist or engineer would check. Structures follow published methods. The values drawn are illustrative, not findings.
Model, equations and assumptions
The model structure used for reproducibility and negative results, the assumptions it makes and how each is tested.
Ri = f(Hi, Ei, Vi, Ci)hazard, exposure, vulnerability and coping capacityEAL = ∑k pk · Lkover events k with annual probability pkxt+1 = A xt + B ut + εtA carries the couplings, u the shock, ε the error| Assumption | Why it is made | How it is tested |
|---|---|---|
| Hazard and exposure are independent within an event | Keeps the loss integral tractable | Compare with a copula fit on joint records |
| Couplings are linear near the operating point | Lets shocks propagate through A | Perturb beyond the linear range and compare |
| Vulnerability curves transfer between places | Local curves are rarely available | Validate against observed losses where they exist |
- Inputs
- versioned datasets with provenance
- Outputs
- distributions, never single numbers
- Code
- released with the model card and tests
- Reproduction
- same data, code, environment and seed
What is measured, and how well.
Each indicator carries its unit, its cadence, the kind of source behind it and the evidence label a value would carry. No values are shown: a programme supplies them under its own authority.
| Indicator | Unit | Cadence | Source type | Label |
|---|---|---|---|---|
| Computational reproducibility | percent of sampled studies regenerated from shared code and data | per audit | Reproducibility audits | OBSERVED |
| Replication outcome | proportion of replications meeting the prespecified criterion | per programme | Coordinated replication projects | OBSERVED |
| Retraction notices | count by field and reason | annual | Retraction Watch database via Crossref | Published statistics |
| Time from concern to correction | months | per case | Journal and integrity office records | REPORTED |
| Post-retraction citations | citations per retracted article | annual | Citation data linked to retraction records | OBSERVED |
| Null and negative results deposited | count of records | annual | Repositories and registries | REPORTED |
| Data available on request | percent of articles whose data could be obtained | per audit | Data request audits | OBSERVED |
Where the evidence comes from.
Named public datasets and institutional records, each with its licence class. Sources by agreement are used only under their terms and never republished.
| Source | What it gives | Licence class |
|---|---|---|
| Retraction Watch database (Crossref) | Retractions, corrections and expressions of concern with reasons | Open, attribution |
| Crossref metadata and Crossmark updates | Links from articles to their corrections and retractions | Open metadata, no restrictions |
| OpenAlex | Works, citations, authors and institutions across disciplines | Open, CC0 |
| FORRT Replication Database | Replication attempts and outcomes, mostly in the social and behavioural sciences | Open, attribution |
| OSF projects of coordinated replication programmes | Protocols, data and code of large-scale replications | Open, licence per project |
| PubPeer | Post-publication comments identifying possible problems in papers | Free access, terms of use |
| Zenodo (CERN) | Citable deposits for null results, data and code | Open, licence per record |
Who decides, and what the evidence serves.
Labs prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.
| Who decides | The question | Evidence this page serves |
|---|---|---|
| Journal editors and publishers | Whether to correct, retract or issue an expression of concern | Evidence of error or misconduct, author response, COPE guidance, reproduction attempts |
| Research integrity offices | Whether an allegation is substantiated and what follows | Raw data, records, code, interviews and an independent analysis |
| Research funders | Whether to fund replications and require data and code sharing | Reproducibility audit results, replication evidence by field, cost of sharing |
| Guideline developers and systematic reviewers | Whether to rely on a study that has been questioned or corrected | Retraction and correction status, replication evidence, risk-of-bias assessment |
Outputs this topic produces
- Replication protocols registered before data collection, with success criteria fixed
- Reproducibility records listing code, data, environment and the outcome of each rerun
- Deposits of null and negative results with full methods
- Correction logs that link each notice to the works affected
What experts ask about reproducibility.
- Can an independent team re-run the analysis from the published data and code?
- Has the finding been replicated with new data, in a new setting or by a different team?
- Which null or negative results exist, and where are they recorded?
- When an error is found, how is it corrected and how are users of the result told?
- Which parts of a workflow, such as software versions or random seeds, could change the answer?
Why results fail to reproduce.
Many published findings in several disciplines have proved difficult to replicate. The causes are well understood: small samples, flexible analysis, selective reporting, undocumented data processing and incentives that favour novel positive findings. In computational work, results can also depend on software versions, hardware, random seeds and data that change after publication.
Negative results are under-reported because they are harder to publish and seem less interesting. This distorts the evidence base: if only successful trials of an intervention are visible, the intervention looks more effective than it is. Systematic reviews and models built on that evidence inherit the bias.
Corrections matter for the same reason. An error in a widely used dataset or model can propagate into many downstream analyses before it is found.
Practices that support reproducibility.
- 01
Open data and code
Sharing the materials needed to re-run an analysis, within privacy and licensing limits. Some data cannot be shared and must be described instead.
- 02
Computational environments
Recording software versions and environments, a practice developed with reproducible modelling. Long-term preservation remains hard.
- 03
Registered reports
Peer review of the protocol before results exist, so publication does not depend on the outcome.
- 04
Correction records
Dated, visible corrections linked to the original result, rather than silent replacement.
A failed replication does not by itself show the original was wrong; differences in setting, sample or method must be examined before conclusions are drawn.
How it is studied, and what the methods cannot say.
- 01Probability and statisticsSample size and power planning · Meta-analysis with bias assessmentApplied here
- 02Uncertainty quantification and model validationMeasurement uncertainty budgets · Sensitivity analysis of simulation modelsApplied here
- 03Causal inference and impact evaluationRandomised pilots and stepped rollouts · Quasi-experimental programme evaluationApplied here
- 04Artificial intelligence and machine learningAssisted screening for systematic reviews · Surrogate models to guide experimentsApplied here
Where this page’s references lead.
The bodies and documents this page is read against. Status and editions change; each link goes to the issuing body.
Questions and answers.
What is the difference between reproducibility and replicability?
Why publish negative results?
Does retraction always mean misconduct?
Can a retracted paper still be cited?
Will Labs publish a study that found nothing?
How do I report an error in a Labs result?
Are corrected results removed?
Contribute evidence from this page.
Reports is the Consortium’s quarterly publication. What you know about this subject can become evidence, a case or a correction in the next edition, credited to you and published under CC BY 4.0 where it is your own work.
- EmailGateway GW-GCRISend it from this pageEvidence, a correction, a case, a method, a review, a translation or profile input. It arrives tagged with this page, so the editor who owns it receives it.
- ZenodoOpen depositDeposit it and submit to the communityPublish your own record on Zenodo with your ORCID iD, then submit it to the Reports community there. Curators review it; the DOI stays yours.
- ReportsGCRISee every way to contributeThe contribution types, how each is reviewed, what credit you receive, the licence and what is never accepted.
Never send personal data, sensitive infrastructure detail or anything you have no right to share. A contribution is reviewed before use; sending it confers no authorship or editorial rights, and sponsorship never buys a finding.
Bring a method, a dataset or a question.
Information and enquiries are open today. An enquiry receives a reference and names the next step; applications, payments, compute allocations and registrations open only when their team, terms and review capacity exist.
Legal documents.
Maps, data and figures on this site are shown for context and general information. They are not advice, forecasts or warnings; the limits of the information are set out in the Disclaimers and Data Notices.