Modern computing facilities have given momentum to analysing large quantities of human-centred data in many fields of research. Whereas a high number of small individual research grants and projects has promoted the acquisition of many small-scale datasets, the collection of human-centred data can be challenging for study participants, indicating a need to collect data as parsimoniously as possible. Amongst other advantages, open data offer the opportunity to economically combine data into large-scale datasets, thereby fostering collaboration and re-use of data to treat researchers’ resources, as well as to study participants with care. We build on this idea of open data and explore the challenges and benefits of producing open data when studying humans and their behaviour through the lens of distinct research fields: life sciences, humanities and social sciences. We thereby draw on our own experience as open scholars, as well as the literature.

Below, we characterize practices, processes and challenges of opening and sharing scientific research material and we provide guidelines for researchers to make their data accessible for re-use. We aim at contributing to a growing, interdisciplinary body of literature that treats open science both as a research object and a practice (Bartling and Friesike 2014). We draw from experiences in our specific research areas and we hope this contribution is also beneficial for researchers in other fields and disciplines. To this end, we add a practitioner’s perspective on the emerging literature on open research data. Such a perspective is needed, as opening up research data might sound easy to some and hard or impossible to others. We share neither of these perspectives, but want to show a very pragmatic perspective to illustrate that producing open data has its pitfalls, but also offers rewards for various stakeholders, including the individual researcher.

In the remainder of this article, we first discuss the characteristics of open data, what qualifies as open data and which formats are reusable. Secondly, we analyse key steps in the research process to produce open data, such as gaining participants’ consent, preparing data and metadata and choosing a repository. We specifically acknowledge the effort and resources needed for these steps and highlight the availability of assistance at universities and other institutions that support the production of open data. Finally, we conclude by discussing the impact of producing open data. In doing so, we analyse long-term and short-term consequences on the societal, disciplinary and individual levels to facilitate decision-making regarding open data practices within academic institutions.

For a long time, the handling of data has been an integral part of scientific practice. More recently, data have been characterised as the key resource (Kitchin 2014) of the 21^st century. Considering the central relevance of the 'data' concept for this article, we begin with a definition of our understanding of 'data', 'research data' and 'open data'.

Data

'Datum' (sg.) and 'data' (pl.) are derived from the Latin word for 'given' or 'something given'. 'Data' typically combine three qualities: a material, symbolic and pragmatic one. From the material perspective, data can both take analogue and digital forms. The focus here is on digital data, which are pieces of information based on binary, electrical pulses. They can be sent from one location to the other as a signal or they can remain in one place, for example, when stored on a medium (Data Ethics Commission 2019: 52). On the symbolic level, data are the results of abstracting the world into categories, measures and other representational forms. These representations can be numbers, characters, symbols, images, sounds, electromagnetic waves and bits. They are understood as the raw material from which information and knowledge are created (Kitchin 2014: 1). Data are often categorised by form (qualitative or quantitative), structure (structured, semi-structured or unstructured), source (captured, derived, exhausted, transient), producer (primary, secondary, tertiary) and/or type (indexical, attribute, metadata) (Kitchin 2014: 4ff).

Research Data

According to Leonelli (2015): 2), data can be understood as "essentially fungible objects, which are defined by their portability and their prospective usefulness as evidence". In addition to the data definitions shown above, Leonelli highlights the pragmatic nature of data – the aspects of purpose and practice. 'Data' are not an entity simply existing in the environment. Entities only become data when they are treated as such. They are embedded into a certain set of practices. What counts as data and what form data assume depends on the concrete context of application. One of these contexts is scientific research, where data are shaped by comprehensive disciplinary and cross-disciplinary conventions and traditions.

For example, in neuroscience and many other life sciences, datasets are often highly modular. Methods that are used to measure brain function produce data as heterogeneous as lists of voltage values (in the case of an electroencephalogram) on the one hand and greyscale values in three-dimensional image spaces (in the case of functional magnetic resonance imaging) on the other hand. If they are to be re-used, these data points need metadata containing references to underlying technology and settings to structure the data streams in a coherent way.

In many social science fields, such as political science, public administration, organisational studies or management, data are often structured along multiple levels, referring for example to countries, organisations within these countries and individuals within countries and organisations. Such a multi-level structure is only useful when links between levels can be clearly identified (i.e. which citizen belongs to which country).

Moreover, qualitative research tends to work with highly unstructured data, which include artifacts, such as field notes, memos, official documents, images, movie clips, diagrams and tabular data. These data typically include identifiable information about concrete individuals, places or events.

To summarise, we pragmatically understand research data as any data produced in scientific processes and/or used in scientific processes. That might also include big data originally produced by social media companies and repurposed for science or data produced in private sector marketing research. Such datasets might relate to different restrictions concerning data sharing.

Open Research Data

What does 'openness' of data mean in a research context? This question has been answered in many different ways depending on one's understanding of openness in science (Fecher and Friesike 2013). The overall aim of opening research data is to make them accessible for secondary use (reusage, Steinhardt et al. 2021). This typically involves releasing datasets into a (digital) repository, structuring the data in a common, standardised way, saving data in a portable file format, adding documentation. Crucially, users are not charged when obtaining and re-analysing the datasets. In some situations, data may simply be placed online for download via a URL with no restrictions whatsoever. This contrasts with more restricted conditions of access where only a record referring to the data is published, but access to the dataset is not granted (closed), a record of the data is published and requirements are mentioned that need to be fulfilled to apply for access to the data (mediated) or a certain time point has to be reached at which the data are released automatically (embargoed). Thus, whether data are open or not is not a simple matter of yes or no.

By making data available in any form, stakeholders contribute to making research more transparent and, in the best case, reproducible. However, not all open datasets are equally useful for secondary use. While making data available in any form is conditional to making especially published work more reproducible, not all open datasets are equally useful for secondary use (Markiewicz et al. 2021). The FAIR principles (Wilkinson et al. 2016) list characteristics for maximising the impact of open data, stating that data need to be F indable (i.e. comprising a globally unique and persistent identifier and registered in a searchable resource), A ccessible (i.e. retrievable using a standard, open and free communications protocol), I nteroperable (i.e. using a standard, formal, accessible format for knowledge representation) and R eusable (i.e. enriched with accurate and relevant (meta-)attributes, a valid and accessible licence and formatted according to disciplinary standards). These FAIR principles can serve as high-level guidelines to provide information for implementation choices, independent from the domain and focused on a broad range of scholarly outputs.

Nevertheless, the intention of 'opening' is to make data accessible for the respective research community in a more controlled manner, taking into consideration specific methodological and content-related criteria. This means data access might be restricted for scientific use only or researchers need to identify their research interest or their intended methodology before gaining access. This is foremost a pragmatic decision to limit the focus of this article, because sharing data with amateur scientists and other stakeholders often implies a different quality of accessibility, for example, the need to summarise data, plot data or explain data other than just publishing a raw dataset. Such regulated access can be organised by a research data centre or an institutional repository, for example, by offering on-site access to data, via a remote desktop or sharing of just parts of a dataset. Although these restrictions might impede the notion of openness, especially when working with sensitive data, such as personal data or data from vulnerable groups, this is often the only way to open data up at least partially (Steinhardt et al. 2021). However, in our understanding, the mere claim that datasets are shared by authors upon request cannot count as open data, as authors might decide randomly who will get access and whether requests are reasonable (Tedersoo et al. 2021, Houtkoop et al. 2018). In our understanding, instead of the proactive delivery of open datasets, these claims rather represent a form of "open-washing". Therefore, stakeholders are beginning to ask for a proactive sharing of data (Morey et al. 2016).

The production of open data does not start with sharing a dataset, but usually with the beginning of each research cycle. In every step of a research project, opening datasets then requires additional considerations. Measures need to be taken and barriers overcome. The following chapter discusses these elements for producing open data along the ideal type of a linear research process. In reality, however, some of the described activities and considerations may overlap or occur in a different order.

Planning: data management plans and pre-registrations

Collecting: informed consent, opt-in and data collection

Cleaning and preparing: contextualising data and making it ready to publish

Opening up: data repositories and data publications

The production of open data is not an end in itself and especially when we revisit the resources that need to be invested to open up datasets, one might ask whether it is worth the effort. However, there are multiple rationales to open and share data in the research context (Duke and Porter 2013). In the following, we discuss some positive effects of open data production at the levels of the society, the scientific community and the individual researcher in more detail.

Societal level

On the societal level, we might, for example, focus on participants of studies. Data collection always puts a burden on participants, they need to spend their time answering questions or attending lab experiments. In the social sciences, the so-called "oversurveying" of society is highlighted (Weiner and Dalessio 2006). Especially when studies need to obtain representative samples of the population of a country or a region (for example, when validating a psychometric instrument like a questionnaire), it is often resource-consuming to obtain such a sample of a sufficient size . This can become particularly burdensome for participants who are part of a very small population when research questions are investigated that focus on rare disorders, for example. It is for good reason that research in the social and life sciences increasingly incentivises participants and builds on paid online access panels that were initially developed for private sector market studies. Yet, the quality of data collected from these professional study participants is sometimes questionable (e.g. amazon turk). Additionally, participants are incentivised to avoid panel dropouts (participants leaving a panel after only a few survey waves used, for example, for a longitudinal study). However, re-using an existing dataset for a different research question might be problematic in that a sample “overfits” a particular research question. That can lead to biases when analysing the data on a different topic. Similarly, it is also questionable in terms of robustness of findings when a whole body of literature relies on just one shared dataset.

Open data are in the long run less burdensome for society because scholars studying similar questions might build on the same dataset instead of collecting data on their own. In the social sciences, some organisations have long been producers of open datasets which can be exploited for a lot of different research questions. Socioeconomic panel studies (such as the German Socio-Economic Panel), public opinion surveys (such as the Eurobarometer) or data that are available from the national statistical offices (such as the British Office for National Statistics) are efficient ways of collecting and using representative country-level data. In psychology and neuroscience, open data are slowly gaining traction after concerns were raised regarding the replicability of several seminal findings (the so-called replication crisis). By now, extensive data-sharing initiatives, such as the Human Connectome Project, have been founded that often foster the usage of the available datasets by hosting competitions on a broad range of research questions (for example ADHD-200).

Apart from the burden on study participants, data collection is expensive and, at the same time, research organisations are often lacking resources. A more efficient use of resources - most commonly tax money - can be achieved by teaming up for data collection and sharing data. To this end, open data support the human right to information: everyone may enjoy the benefits of scientific progress and its applications (Article 15 of the International Covenant on Economic, Social and Cultural Rights). Sharing data, not only in small circles, but opening it up widely, can also help to re-distribute resources to countries in which research might not be well funded and researchers might not be able to collect specific forms of data on their own. Especially when data collection is bound to expensive technology, such as in neuroscience, worldwide use of datasets can also help to tear down knowledge barriers and distribute knowledge more equitably. In a similar way, open data may then also be valuable as an instrument increasing public trust in the results of scientific research in the face of controversy.

Disciplinary level

The most direct impact of open data, however, might be observed in the development of the discipline. Thus, open data affects both - research as well as education. In terms of education, open datasets enable teaching, based on real data and practising data analysis, based on real datasets. Realistic problem-based learning is, thus, possible and students are better prepared than working with artificial teaching datasets. Bishop and Kuula-Luumi (2017) , for example, show that open data are to a high share downloaded by students on different levels and mostly used for learning and teaching. Students might - in that - also develop new research findings from existing data and help to grow the disciplinary knowledge base, for example, in their theses or research project courses. This can enhance their motivation, based on the notion that they can make a difference. Similarly, completion times for the qualification of specifically undergraduate students can be too short to enable the acquisition of a dataset that is large enough to not be statistically underpowered. Here, open data open new possibilities to investigate a novel research question where students are commonly restricted to literature reviews or small samples.

In terms of research, open data enable innovation and make findings more robust. Whereas innovation is driven by combining datasets and answering new comparative questions, robustness of a finding can only be assessed by re-running the same analysis on different datasets. On the other hand, replications become much easier when data are opened up. The more data-intensive the research, the more dependent replicability becomes on the availability and accessibility of the original datasets for re-analysis. Specifically when using behavioural data, based on task paradigms, slight variations in the implementation of the paradigms are the norm between research groups. If these data are openly available, researchers can investigate the robustness of results across task variations, ultimately providing information for task choice for future studies.

Individual level

Open data - as discussed above - come with several costs for individual researchers or research teams. We argue that, in general, a high intrinsic motivation to publish open datasets is needed. However, we also observe a growing impact of data publications. They count as separate publications instead of just an appendix to an article and might attract additional citations. Still, the value of data publications differs between disciplines and they might be more attractive in the life sciences than in the social sciences. Apart from these metrics, which still count in the recruitment and promotion of academics, open datasets might also generate appreciation and help to build or strengthen an individual network. That again might lead to further collaboration and help in terms of career. Publishing open data might add to a positive reputation of a scholar, either because the discipline relies on shared datasets or because published results can be verified. Being perceived as an expert for open data publication might ultimately also lead to new career opportunities in research support and management.

Due to current incentive structures, producing open data offers a lot of benefit for societies and research fields as a whole, whereas individual researchers doing the work can be at a disadvantage compared to their peers working on closed data sources. Planning for, acquiring and preparing datasets for public release is a serious effort that researchers need to commit to. In comparison, producing closed datasets is less time-consuming as datasets do not need to be made easily accessible and enriched with meta-data. With hiring decisions depending on the number and visibility of publications and funding agencies incentivising the acquisition of new data instead of encouraging the use of existing, extensive open datasets, researchers investing time in optimising their datasets for public release can, thus, lose ground.

Still, apart from the obvious benefits for scientific and societal progress, open research practice is increasingly considered to be a hallmark of "science done right". To this end, individual researchers may benefit from a boosted visibility of their work. Greater transparency supports replicability, thus corroborating trust which is indispensable in scientific discourse. New developments, such as citable data publications, stronger datafication in all scientific disciplines and policy-making towards transparency and replicability by disciplinary associations, national laws and funders also act as incentives for opening up scientific work. Hence, researchers who wish to incorporate open science into their repertoire will find good conditions to do so with ample help available online and within research organisations, as well as academic networks that facilitate building routines for future projects.

Our step-by-step guide summarises the experiences of practitioners across different disciplines and strives to serve as a resource for researchers seeking orientation on how to open up their data. Given the clear benefits for society, disciplines, but also the individual researcher, with current developments in funding, policy-making and hiring decisions increasingly favouring open science approaches, opening up data should be viewed as a default – where ethically unobjectionable – with clear guidance for scholars entering a field.

We want to thank reviewers and editors for their valuable input. We also want to thank Gwen Schulte (DIPF) for proofreading this article.

The publication of this article was kindly supported by RIO. We would like to thank RIO and Wikimedia Deutschland for enabling this collection.

The authors were funded by the Open Science Fellows Programme by Wikimedia Deutschland, VolkswagenStiftung and Stifterverband in 2017/2018 (Caroline Fischer and Vanessa Teckentrup) and 2018/2019 (Simon Hirsbrunner).

The authors are presented in an alphabetical order, all authors contributing equally to the manuscript.