Why don’t all researchers share their data openly?

With the proliferation of fake news and the growing risks from fraudulent research, being able to verify and reproduce scientific claims is more important than ever. Sharing the research data underpinning published articles supports greater transparency and reproducibility – and, thereby, innovation.

However, sharing research data publicly remains a real challenge for many scientists. Our recently published analysis of publication data from more than 30,000 articles reveals that only one in 10 researchers in the physical sciences share their data in a findable, accessible, interoperable and reusable (FAIR) way.

The FAIR principles were introduced in 2016 and have become a foundational framework for standardised data management and sharing. Developed by academic researchers, librarians and industry professionals, they are designed to foster well-managed and accessible data sets. At IOP Publishing, our articles align with the FAIR principles by promoting standardised metadata, persistent identifiers and clear usage licences, and we encourage all our authors to ensure that data underpinning their papers can be reliably located, accessed and reused with proper attribution.

Since 2022, we’ve required all authors to include a data availability statement in their articles, outlining whether and how the data supporting their research can be accessed. In 2023, we went a step further and began requiring authors who are unable or unwilling to share their data publicly to explain why.

Early results suggest that these policies are starting to have an impact. In the past year, 11 per cent of articles published in our journals shared data in a FAIR way, up from 8 per cent the previous year. Additionally, 64 per cent of articles indicated that their data was publicly accessible via some route, up from 50 per cent.

However, despite this progress, FAIR data sharing remains relatively low. What is holding researchers back?

Our analysis of the data statements provided by researchers shows that there are multiple barriers, varying by research discipline and geography. These include technical difficulties, lack of infrastructure, concerns over proprietary information, and cultural factors within different scientific communities.

For instance, we see that 81 per cent of environmental research papers reference publicly available data, with 59 per cent meeting FAIR standards. Researchers in this field often state that they can’t share their data because it is owned by third parties, such as government agencies or private organisations. Physicists, meanwhile, are often concerned that the format of their data might not be accessible because of ownership restrictions or because their data is difficult to use without specific knowledge. And in engineering and materials science, we see quite low levels of data sharing, with only 8 and 5 per cent of papers respectively meeting FAIR standards. Researchers in these fields say that commercial confidentiality is a major obstacle, particularly in projects with industrial partners.

There are also geographical variations in sharing practices. For example, 71 per cent of researchers in the UK share their data publicly in some way, followed by 67 per cent in the US, 65 per cent in China and 59 per cent in India. However, only 26 per cent of researchers in the UK and US share FAIR-compliant data, and in China and India compliance is just 6 and 4 per cent respectively.

These disparities highlight the need for better guidance, standards, tools and infrastructure to make data more accessible and interoperable across the globe. But solutions need to be tailored. A one-size-fits-all approach won’t suffice, and supportive policies are also essential.

At IOP Publishing, our recently launched open access Machine Learning journal series offers new formats to maximise the utility of data sharing, including data set, benchmark and challenge articles, all designed to incentivise and showcase open data practices. For instance, data set articles provide detailed descriptions of research data, including how it was collected and processed, and relevant metadata; the aim is to help others to understand and reuse data, rather than to test hypotheses or present new interpretations, methods or in-depth analyses.

Challenge articles bring large groups of researchers together to solve specific problems by creating new algorithms, data sets, or workflows. For example, there is a “protein structure prediction challenge” where researchers from every part of the world are encouraged to submit their own machine learning model that is applied to this problem and produces an outcome. They may use an existing data set, so there is no requirement for the data set to be created specifically for the challenge or to be open access but they usually are. The same is true for algorithms and workflows, which describe how the machine learning models are coded and used with respect to hardware, software and other technical specifications.

Benchmark articles evaluate the performance of various models, algorithms or software against a consistent problem or dataset that represents the problem (such as how to identify tumours in a set of cellular images).

IOP Publishing remains dedicated to pursuing both independent and collaborative innovations, in our publishing and other offerings, that advance the broader adoption of open practices across the physical sciences. But future progress will require closer engagement with research communities and collaboration between funders, institutions and academic publishers.

The aim must be to create an environment where data sharing becomes seamlessly integrated into the research process in ways that research communities want and need. This will be key to unlocking the full potential of open science.

Daniel Keirs is head of journal strategy and performance at IOP Publishing.