What aspects of data management are worth addressing before it is too late? Researcher and data steward Klára Kosová shares her experience from the INTERFER project.

I think the dual role of data steward and researcher is a major advantage. I know first-hand how research works in practice, that things change as a project progresses, and that researchers have many other responsibilities to deal with. I therefore understand why data management can sometimes feel like yet another administrative burden. At the same time, I believe that technological development and digitalisation now make it possible to do many things much more efficiently. By setting up appropriate data management processes, we can not only make researchers’ work easier but also save them time. I see this as one of the important roles of a data steward: helping to find solutions that make sense not only from an Open Science perspective, but also in everyday research practice.

Good data management is not just about what happens to the data at the end of a project, as it might initially seem. When data are well organised and documented from the outset, and we know how they were created and what happened to them throughout the research process, this makes the research itself easier and facilitates its subsequent verification and reuse. My own experience as a researcher helps me identify approaches that are, I hope, genuinely useful to my colleagues. The aim is not to add more obligations, but to help ensure that their research is transparent and trustworthy.

INTERFER brings together several institutions, disciplines and research approaches – from experiments focused on public opinion to research on foreign policy or emerging technologies. What does this diversity mean from the perspective of research data management?

The diversity of research approaches means, above all, that there is no single universal approach to research data management that works equally well for every team. Experimental data are handled differently from interviews or text corpora. Each type of data has different requirements in terms of collection, documentation, storage, protection and subsequent sharing. Each discipline also has its own established practices. And when collaboration between institutions is added to the mix, it is important to ensure that the individual teams have a shared understanding of how different types of data are handled, while also complying with the internal rules of the institutions involved.

That is why, particularly at the beginning, it was important for us to establish common practices while still respecting the specific requirements of individual disciplines and methods. The aim was not for everyone to follow exactly the same steps or produce identical data outputs, but rather to ensure that the data were easy to find, accessible under appropriate conditions, interoperable and reusable by other members of the team and potentially by third parties. In practical terms, this meant agreeing on shared principles for file naming and documentation. These may seem like minor details, but in practice they make a considerable difference, especially when several people are working on a project at the same time or when someone who was not directly involved in creating the data needs to return to them later.

This diversity is also reflected in data sharing. Different types of data have different audiences, so for each dataset we try to identify a repository that reflects both the nature of the data and the community for which they are intended. These range from more general platforms such as OSF to disciplinary repositories and platforms such as GESIS or LINDAT/CLARIAH. This, too, cannot be resolved through a single decision at the beginning of a project; the most appropriate solution needs to be considered separately for each dataset.

What types of data do you work with in the project, and which of them are the most challenging from a data stewardship perspective?

In INTERFER, we work with a fairly broad range of data. At one end of the spectrum are purely quantitative data, while at the other are various types of qualitative data: recordings or transcripts of interviews, field notes from participant observation, and text corpora compiled from publicly available documents. In practice, however, it is becoming increasingly difficult to maintain a clear distinction between quantitative and qualitative approaches to data. Textual data, once associated primarily with qualitative research, are increasingly taking on a quantitative dimension. The same text corpus can be used for in-depth discourse analysis, automated text classification or the training of a language model. This creates quite specific data management requirements, because documentation and metadata that are sufficient for one type of use may not be useful for another.

From a data stewardship perspective, however, the most challenging data for me are clearly those collected through interviews, primarily because of the need to protect personal data and the safety of the respondents themselves. INTERFER focuses thematically on foreign interference, so some interviews are conducted with people in environments that are sensitive or even dangerous for various reasons. In such situations, it is not always possible to obtain written informed consent, because signing a document may itself pose a risk to the respondent. In these cases, we use oral consent, which is recognised as standard practice in research methodologies for sensitive contexts and is approved in advance by the ethics committee. Protecting these individuals is our absolute priority and takes precedence over any considerations of data openness. Such data are not shared, even in anonymised form. In these cases, anonymisation may simply not be sufficient. Even seemingly neutral details, when combined, can lead to the identification of an individual, which is absolutely unacceptable. From my perspective, this type of data is therefore the most difficult to manage, not so much because it is technically complex, but because every decision carries significant ethical weight.

Protecting respondents therefore clearly takes precedence over data openness. More generally, where do you think Open Science reaches its limits? And how do you decide in practice which data can be made available and when sharing is no longer appropriate?

It is certainly not the case that all research data must automatically be made fully public. Open Science does not mean openness at any cost; rather, it is about taking a responsible approach to research. In practice, we make these decisions on a case-by-case basis together with the researchers involved. There are several reasons why data may not be shared at all, or may only be shared under certain restrictions. First and foremost, it is necessary to consider the protection of the people from whom the data originate, whether this concerns the protection of personal data in accordance with the GDPR or, in more sensitive cases, the safety of respondents. There are also legal and contractual considerations. For example, data obtained from social media are subject to the terms of use of individual platforms, which generally do not allow such data to be freely redistributed to third parties.

I think it is important to distinguish between openness and transparency in the research process. For a wide range of reasons, it may sometimes be impossible to make the data themselves publicly available. However, it may still be possible to share documentation, such as analytical or methodological procedures, coding manuals or replication scripts. This information can make it much easier to understand how the research was conducted and how its conclusions were reached. Personally, I therefore see openness in science not simply as a question of whether everything is made public, but of whether the research process is sufficiently transparent and verifiable.

Do you encounter situations where data management is addressed too late when working with researchers? What can still be fixed along the way, and what, on the other hand, needs to be considered already at the research design stage?