Research reproducibility statistics show a persistent gap between recognizing the problem and building systems that make findings easier to check. Surveys of biomedical researchers, professors, and public-health analysts, alongside audits of hydrology and neurology papers, measure that gap from different angles. Together, they point to limited replication, scarce research artifacts, uneven open-science awareness, and weak institutional incentives.
Contents
- How researchers assess reproducibility
- What replication attempts have found
- How often data, code, and materials are available
- Where published studies fall short
- Awareness and adoption of open science
- Institutional incentives and support
How researchers assess reproducibility
The most recent biomedical survey in this set was conducted in 2024 among 1,630 researchers recruited from authors of articles published from October 2020 through October 2021. It captures researchers’ perceptions rather than a direct audit of every paper. In that survey, 72% agreed that biomedicine had a reproducibility crisis. Among respondents, 27% described the crisis as significant and 45% described it as slight. These categories describe perceived severity, so they should not be read as a measured failure rate.
Researchers were also cautious about how many papers could be reproduced. Only 5% thought more than 80% of biomedical research papers were reproducible. For clinical, in vivo, and in vitro biomedical papers separately, 11% or fewer thought that more than 80% were reproducible. The answers indicate broad concern across several kinds of biomedical work, but they do not establish that the same share of papers would fail a standardized replication test.
An earlier Nature survey provides a different time point and sample. In 2016, 1,576 researchers were surveyed: 52% reported a significant reproducibility crisis, 38% reported a slight crisis, and 3% said there was no reproducibility crisis. The 2016 and 2024 surveys used different designs and populations, so their percentages should be compared as snapshots of opinion, not as a time-series estimate.
What replication attempts have found
The 2024 biomedical survey also asked about researchers’ own replication experiences. Twenty-three percent said they had tried and failed to replicate one of their own published studies. At the same time, 31% said every replication of their own published studies had produced the same result as the original, while 46% had never tried to replicate any of their own published work. These figures concern respondents’ reported experiences and are not a controlled estimate of the probability that a biomedical result will replicate.
Replication of other teams’ studies showed a similar tension. Forty-seven percent of biomedical researchers said they had tried and failed to replicate a published study conducted by another team. Only 10% said all replications they had attempted of other teams’ studies were successful, and 43% had never tried to replicate someone else’s published research.
Publication is another constraint on what the research record can show. Among biomedical researchers who had tried to replicate their own work, 36% had published the results of their most recent attempt. Among those who had published a replication, 29% said their latest replication took about as long to publish as a typical non-replication paper. Among researchers who had attempted to replicate others’ work, 25% said they had no plans to publish the replication. In addition, 85% had never been contacted by another researcher unable to reproduce a finding they had published.
These results suggest that a failed replication can remain invisible for several reasons: a replication may not be attempted, a result may not be published, or an unsuccessful effort may not lead to direct communication with the original authors. They do not show why any particular study failed, and they do not distinguish every possible difference in methods, populations, instruments, or analysis.
How often data, code, and materials are available
Sharing practices are easier to interpret when the unit and date are explicit. A late-2017 survey of public-health analysts found that 14.4% had shared code, data, or both in a recent manuscript or report. In that survey, 11 participants reported sharing data only, six reported sharing code only, and 6.5% reported sharing both data and code. Among analysts who had not shared code in the focal report, 11.3% had made statistical code publicly available at some point.
Requirements influenced sharing decisions. Sixty percent of public-health analysts who shared data said a funder, journal, employer, or research team had required the data to be public. The comparable figure for analysts who shared code was 35%. Reproducibility training was selected by 50.9% of public-health analysts as one of their top two facilitators for greater use of reproducible-research practices.
An audit of hydrology and water-resources research examined articles published in 2017 and was published in 2019. Researchers reproduced results for 1.6% of the 360 articles tested with available digital artifacts. The study estimated that results could be reproduced for 0.6% to 6.8% of all 1,989 articles in its six-journal population, with that range reported as a 95% confidence interval. Forty-four percent of hydrology articles had only some digital research artifacts available.
| Evidence type | Result | Scope and period |
|---|---|---|
| Public-health analysts sharing code, data, or both | 14.4% | Late-2017 survey |
| Hydrology results reproduced among articles tested with available artifacts | 1.6% | 360 articles published in 2017 |
| Hydrology population estimate for reproducible results | 0.6%–6.8% | 1,989 articles; 95% confidence interval |
| Hydrology articles with only some digital artifacts available | 44% | Articles published in 2017 |
The public-health survey and hydrology audit do not measure the same outcome. Sharing an artifact is a condition that can support checking a result, while successful reproduction requires usable artifacts and a method that can be followed. The percentages therefore should not be combined into one overall reproducibility rate.
Where published studies fall short
The hydrology audit identified documentation problems as well as missing artifacts. Eighty-nine percent of hydrology articles lacked directions for using the available digital artifacts. Five percent had all artifacts available but still produced non-reproducible results. This distinction matters: availability does not guarantee that files are understandable, complete, compatible, or sufficient to recreate the published analysis.
A separate audit examined 400 neurology publications from January 2014 through December 2018. It found that 77.2% were behind a paywall, while 57.1% were publicly available through the Open Access Button. These measures describe access routes in the sampled publications and are not interchangeable measures of reproducibility.
Among the 271 neurology publications with measurable reproducibility characteristics, 3.7% provided a preregistration statement or claimed preregistration, 0.7% provided access to a protocol, 9.2% provided access to raw data, and 0.7% provided an analysis script. Of 255 eligible neurology publications, 9.4% provided access to a materials list. None of the 271 publications analyzed for reproducibility claimed to be a replication study.
The neurology figures show why reproducibility is multidimensional. A paper can be accessible but lack raw data, provide data but not analysis code, or describe a study without offering a protocol or materials list. The audit reports the presence of these features in the sampled publications; it does not establish that every available item was complete or sufficient for exact reproduction.
Awareness and adoption of open science
Awareness is widespread in the USA and India, but it varies by country and field. A survey of professors in the two countries, published in 2025, found that 83.8% of surveyed researchers in India reported some familiarity with the reproducibility crisis, compared with more than 91.5% in the USA. More than 95.23% of USA social-science researchers reported awareness, while awareness was 81.95% among USA engineering researchers, 84.93% among Indian engineering researchers, and 80% among Indian social-science researchers.
Perceptions of peer awareness were lower. In India, 26.17% of respondents believed their peers were completely unaware of the replication crisis; the corresponding USA figure was 17.62%. Awareness of the open-science movement was reported by 76.62% of USA respondents and 61.78% of Indian respondents.
| Group or practice | Reported percentage | Survey context |
|---|---|---|
| USA respondents aware of open science | 76.62% | Professors in a 2025-published USA and India survey |
| Indian respondents aware of open science | 61.78% | Same survey |
| USA social-science researchers aware of the reproducibility crisis | More than 95.23% | Same survey |
| Indian social-science researchers aware of the reproducibility crisis | 80% | Same survey |
| Indian respondents saying peers were completely unaware of the replication crisis | 26.17% | Same survey |
Awareness should not be treated as adoption. Knowing about reproducibility does not show that a researcher preregistered a study, shared data, published code, or attempted a replication. The practice measures below show how large that difference can be.
Institutional incentives and support
Preregistration was reported by 15.71% of Indian professors and 34.86% of USA professors in the 2025-published survey. The gap was especially visible in social science: 8.88% of Indian social-science professors reported preregistering a study, compared with 43.92% of USA social-science professors. Engineering showed a different pattern, with 11.11% of USA engineering professors and 17.81% of Indian engineering professors reporting that they had preregistered work.
Biomedical researchers reported limited formal support. In the 2024 survey, 16% said their institution had established procedures to enhance reproducibility. Forty-eight percent said their institution did not provide training on enhancing reproducibility, and 33% were unsure whether their institution provided such training.
Incentives also favored novelty. Sixty-seven percent of biomedical researchers felt their institution valued new research more than replication studies, and 83% thought replication funding was harder to obtain than funding for a novel study. Only 7% were aware of funders offering specific calls for reproducibility-related research.
Respondents in the USA and India identified structural changes as possible support. Publication-model changes were selected by 78.53% of Indian professors and 75.86% of USA professors as a change that could support reproducible research practices. In the USA social-science subgroup, 57.14% selected promotional-model changes as needed to incentivize reproducible practices. These are reported views about possible reforms, not evidence that the reforms have been implemented or that they would produce a particular effect.
Taken together, the statistics describe reproducibility as both a technical and institutional problem. Replication attempts, public data, executable code, protocols, and materials can make findings easier to inspect, but the evidence also shows barriers in documentation, publication, training, funding, and career evaluation. The measurement period and population matter for every percentage: a late-2017 public-health survey, a 2017 hydrology audit, neurology publications from 2014–2018, and more recent researcher surveys answer related questions without producing one universal reproducibility number.
Sources
- Biomedical researchers’ perspectives on the reproducibility of research
- 1,500 scientists lift the lid on reproducibility
- Reproducibility and replicability in research: What 452 professors think in Universities across the USA and India
- Use of reproducible research practices in public health: A survey of public health analysts
- Assessing Data Availability and Research Reproducibility in Hydrology and Water Resources
- Reproducible and transparent research practices in published neurology research