Shifting norms around open and FAIR

Norms and practices governing data management are still emerging in conventional science, and are not yet firmly established across disciplines. One important development in scientific research is the emergence of open and FAIR (Findable, Accessible, Interoperable, and Reusable) principles. Broadly, open science is research conducted in a way that allows others to collaborate and contribute (OECD 2020). As a movement or paradigm, open science can be traced to the Scientific Revolution of the late 16th and early 17th centuries when rapid dissemination of knowledge became a guiding principle for scientific research (David 2008). Contemporary advocates argue that open science strengthens research by facilitating reproducibility through transparency (Munafò et al. 2017), and makes science more accessible to stakeholders including the general public, though important power differences often remain (Levin and Leonelli 2017). Recently, open science has been accelerated by policy initiatives in Australia, the European Union, the United Kingdom, and the US (Tenopir et al. 2015).

As an umbrella term, open science encompasses a range of components, including participatory research; open access to research publications and pre-prints; open access to data and methodologies, including processes such as lab notes and code; open peer review; and open access dissemination of results and data. Within open science, much of the emphasis to date has been on open data sharing, with a strong focus on licensing. Clear data licensing helps enable open data by clarifying to third party users the status of a data set and their ability to apply the data for different purposes and under different conditions. Common ways to release open data include the Creative Commons Public Domain Dedication (CC0), the Creative Commons Attribution license (CC BY), the Creative Commons Attribution-NonCommercial license (CC-BY-NC), or the Creative Commons Attribution-ShareAlike (CC-BY-SA). These latter licenses include restrictions that can be problematic, an issue we discuss further in Section 5.

A second, related movement is emerging around making data more FAIR. Many of the ideals behind FAIR match the rhetoric around open science; guiding principles include transparency, reproducibility, and reusability (Wiklinson et al. 2016). Calls for open and FAIR data differ on a few key points. First, all FAIR data do not necessarily need to be open. FAIR is about enabling, rather than securing, access to information. Whereas open data are necessarily free of charge, FAIR data could be accessible but behind a paywall. Second, while open science can be described as a paradigm, or an approach to scientific research, FAIR is more prescriptive, offering concrete guidelines and even checklists for researchers to follow (Wilkinson et al. 2016). Practices around cataloguing and metadata documentation help make data FAIR.

The state of the data in scientific research

Understanding the current state of data management is critical for understanding and charting progress moving forward. Notably, the larger scientific community has only recently begun to adopt practices related to open and FAIR data. One benchmark study of 1,329 researchers across scientific domains explored practices and perceptions of data sharing (Tenopir et al. 2011).² At the time of publication in 2011, 29% of respondents had data management plans, while 55% did not and 16% were uncertain. Regarding data access, 38.5% of respondents stored their data in an organization-specific system. A follow-up study conducted shortly after National Science Foundation (NSF) policies went into effect reported mixed progress. Perceptions of the value of data sharing increased, but so did perception of threats, and progress on self-reported practices was mixed (Tenopir et al. 2015).

A number of factors contribute to suboptimal data management in scientific research. While researchers are generally satisfied with tools for short-term storage and documentation of their data, access to longer-term repositories may be lacking (Tenopir et al. 2011), and citizen science practitioners may not be familiar with the many domain-specific repositories—though in recent years open repositories such as Dryad and FigShare have grown in popularity. Beyond the provision of technical tools, “Barriers to effective data sharing and preservation are deeply rooted in the practices and culture of the research process as well as the researchers themselves” (Tenopir et al. 2011, p.1). Incentives are often missing for researchers to invest the time and effort required to make their data open or FAIR, since data cleaning and documentation are time-consuming activities that lack the same incentives as, for example, publication. Further, an academic culture that tethers scholarly publication to professional milestones like the tenure process may actively disincentive openness and sharing if researchers fear getting scooped. And volunteer citizen scientists are not necessarily motivated by the same incentives as researchers, but rather factors such as personal interest, learning, creativity, socialization, and the desire to contribute to scientific research (Jennett et al. 2016; Rotman et al. 2012).

Researchers have also started to study data reuse, defined as the use of data by the original data collector or third-party users, sometimes by combining the data with other data, for the same or different purposes for which they were originally collected. One study found that perceived utility of a data set was the single strongest factor leading to reuse, and concluded that the value of reuse should be more widely demonstrated to the academic community (Curty et al. 2017). Efforts to make data discoverable, promote the use of strong metadata, and improve norms and practices around data attribution and citation could all lead to more data reuse. Regarding citation, the use of persistent identifiers (e.g., Digital Object Identifiers [DOIs]) can ensure that researchers are able to refer to a unique data set produced at a given point in time by providing persistent URLs to data that are retained even if, for example, data moves from a project website to a longer-term repository. This is important for traceability in scientific findings as well as for appropriate attribution.

The state of the data in citizen science

The White House memorandum Addressing Society and Scientific Challenges through Citizen Science and Crowdsourcing (Holdren 2015) offers three core principles for citizen science: Contributions of volunteers should be 1) fully voluntary, 2) meaningful, and 3) acknowledged. Similarly, the European Citizen Science Association (ECSA)’s 10 Principles of Citizen Science (ECSA 2016) include “citizen science project data and metadata are made publicly available and where possible, results are published in an open access format.” These codes suggest that data sharing, including through publication, may be necessary to fulfill a core best practice of citizen science. Some researchers document the importance of report-backs, or the process of sharing individual and collective results with volunteers in ways that are meaningful and useful to them (Bonney et al. 2009; Morello-Frosch et al. 2009; Gallo and Waitt 2011). Related, there is often a noticeable commitment within citizen science projects to publish academic publications in open access journals (although fees can be a barrier to follow-through). However, the realities of data sharing may suggest differently: One study of open biodiversity data available through the Global Biodiversity Information Facility (GBIF) found that citizen science datasets were among the least open (Groom et al. 2016).

Beyond open data, a significant portion of research on data practices addresses data quality. The topic of data quality is a key concern in the scientific enterprise because perceptions of poor data quality can influence the willingness of scientists or policy makers to trust the results of citizen science. In the context of a research project, the construct of data quality means that data are high enough quality to serve a project’s goals: there are no universal criteria to establish quality in scientific data because it is inherently contextual. In acknowledgement of this reality, the concept of fitness for use is frequently applied in citizen science (Kosmala et al. 2016), with the focus on designing project processes with the end in mind (Parrish et al. 2018). For example, in air-quality monitoring, low-cost sensors cannot currently compete with professional instruments for achieving precision and accuracy at the levels necessary for regulation (Castell et al. 2017). Therefore, one goal of citizen science air-quality projects may be to get regulators to take notice when systematically collected data indicates a potential problem meriting further investigation. Low-cost (including commercial or open-source/do-it-yourself) sensors are of suitable quality to be fit for this, and often other, purposes.

When used to describe an individual data record, data quality typically refers to the accuracy and precision with which a data value represents a measurable parameter of an entity or phenomenon. At a whole dataset level, data quality refers to all attributes being accurately measured using a standard/common protocol and accurate instrumentation.³ Higher-quality data accurately and precisely represent reality, whereas low-quality data are a poor or inconsistent representation. Errors in measurement can be random (scattered) or systematic (always wrong or biased in the same direction), and they can arise owing to poor instrumentation (imprecise, poorly calibrated, or old) and operator errors, which usually introduce systematic biases in data. Therefore, measurement accuracy may be affected by several factors, including the training and competence of volunteers; sensitivity, calibration and construction quality of measuring instruments; establishment of a consistent sampling frame; the methods used in taking/determining measurements and their consistency over time and space and across volunteers; and delays between sample collection and measurement (in lab settings). With respect to field-based observational facts such as species occurrence recording, competency and attention to detail by citizen scientists can affect factors such as correct identification, spatial accuracy, precision and uncertainty, and date/time precision. In addition, third-party perceptions of data quality can be affected by whether records have been verified or validated by experts or if there are methods or additional data sets available to cross-validate or even triangulate results.

However, the actual quality of data has significance only in the context of usage. This is a relative concept that relates to fitness for use (Chapman 2005), i.e., for some applications, low-quality data may be acceptable. One of the underlying premises of citizen science in the field of biology, for example, is that scores of amateur scientists can collect data over much larger areas and longer periods than would ever be possible by highly trained biologists alone. Thus, in some studies, the lower quality is balanced by a far wider scope, demonstrating that almost all data has value depending on the purpose for which it is to be used. In addition, citizen science data may be analyzed along with other scientific or instrumental observations as a method of either validating or cross-validating the data, or complementing data of known quality with a larger sample size.

Researchers typically consider data quality and fitness for use in individual project design, explaining the factors affecting data quality within the text of research papers. However, such explanations are not always documented in metadata accompanying the primary raw and processed datasets used in the research, and if they are documented, it is rarely in structured, standardized formats. These cases both create a number of significant problems and constraints for secondary users of the primary data.

For a variety of reasons, researchers are increasingly turning to individual and aggregated datasets collected by other projects as primary or secondary data (e.g., to augment their own original datasets) for their research. These secondary applications of data are highly dependent on researchers having a clear understanding of the provenance, methods, data-quality constraints, and prior treatments of datasets in order to support decisions about fitness for use in their particular application of the data. For secondary users of data to be able to assess fitness for use, they must be able to efficiently filter, sort, and select particular datasets that satisfy the quality criteria for their purpose. To accomplish this, it is critical for dataset metadata to describe the quality aspects of the data as comprehensively as possible, including its provenance, treatment, constraints, and biases, in a structured, standardized way. Our results indicate that currently, well-documented data are not always the norm in citizen science.

In summary, while the citizen science community may lag slightly behind ideals, this is probably in part because of the rapid evolution of scientific norms of open data, data publication, metadata, data documentation, and data reuse over the past decade, which in fact means that many corners of the global scientific enterprise are rushing to catch up. We turn now to our methods and results, before turning to a discussion of what the evolution of norms means for the citizen science community and how the community can improve its data practices.

Sampling framework

Members of the Task Group began by reviewing a range of literature on citizen science data practices across the data lifecycle to inform development of study methods. We reviewed citizen science typologies and other classification schemes to create the sampling framework. Typologies were largely drawn from academic research, and covered aspects of citizen science including governance model (Haklay 2013; Shirk et al. 2012) and scientific research discipline (Kullenberg and Kasperowski 2016; Follett and Strezov 2015). Other classification schemes included UN regions for capturing geographic distribution, and controlled vocabularies used to document variables including type of hosting organization (e.g., university, community-based group, etc.).

The sampling framework allowed us to search for projects representing different types of diversity (e.g., in governance, in scientific research discipline, and in geographic distribution). Using this framework, we recruited participants through a three-step process. Our participants were recruited following a purposive sampling scheme designed to capture data management practices from a wide range of initiatives through a landscape sampling methodology (Bos et al. 2007).⁴ First, we pulled a random sample of citizen science projects from the SciStarter database, requesting the listed contact for each project to participate in our study. In this initial sample, we found that projects in environmental citizen science, particularly biodiversity, and projects based in the US were over-represented.

We then used our sampling framework to identify gaps in the sample and sought out projects not necessarily listed on SciStarter to fill the gaps. As gaps were filled, the research team met numerous times to discuss our evolving sample and early findings. The research team then conducted additional purposive sampling until theoretical saturation was reached (Weed 2006) at 36 interviews with citizen science projects and platforms. Note that while this sampling strategy appears successful in covering a wide range of citizen science projects, it is not intended to be statistically representative of the field as a whole, and only English-speaking projects were represented.

Data collection

We began our structured interview protocol with questions from our sampling framework. Interview questions addressed various practices related to data quality and data management (see Appendix B for the interview protocol). In addition to supporting our sampling methodology, these questions enabled us to collect valuable information to help characterize our sample. The second part of our interview protocol addressed practices related to data quality and data management. We focused on these practices because our review of the literature suggested that practices related to data quality and data management (as opposed to, for example, data security) may be unique to citizen science compared with other forms of scientific research. Grounding our protocol in the existing literature allowed us to create a structured protocol with multiple choice rather than open-ended questions. For example, rather than asking participants “Where can your data be accessed?” we asked, “Can your data be accessed from: a) Project website; b) Institutional repository; c) Topical or field-based repository; and/or, d) Public sector data repository?”

Regarding data acquisition, we asked our participants to describe the full range of data collection or processing tasks that were used in their citizen science research. For data management (including data quality), we asked about quality assurance/quality control (QA/QC) processes, including those related to data collection but also human aspects such as targeted recruitment or training; instrument control such as the use of a standardized instrument; and, data verification or validation strategies, such as voucher collection (e.g., through a photo or specimen) or expert review. We asked questions on data access, including whether access to analyzed, aggregated, and/or raw data were provided, and how data discovery and dissemination retrieval were supported (if at all). Because they relied on known practices identified through existing literature, the vast majority of our questions were multiple choice, though participants were encouraged to elaborate on their answers or provide additional information.

Members of the research team conducted interviews or surveys, either in person, by phone, by Skype, or by email.⁵ Each team member followed the same structured protocol during the interview process, although open-ended questions allowed for the collection of richer detail on selected cases.

Data analysis

Analysis was conducted through tallying responses, comparing responses with previous research, and augmenting structured responses with unstructured comments. We also compared results with prior quantitative assessments of citizen science data practices, including Schade et al. (2017) and Wiggins et al. (2011).

Characteristics of sample

The average start year of the projects in our sample was 2011, with the earliest year being 1992 and the most recent being 2017. Our sample was heavily weighted toward the environmental and biological sciences (n = 29, 81%), reflecting the early genesis of citizen science in these communities (Schade et al. 2017), but also included several health-related projects (n = 7, 19%), two VGI initiatives, a general-purpose crowdsourcing initiative, and a technology development project. Most of the projects were hosted in North America (n = 19, 53%). The remaining sample was from Europe (n = 7, 19%), Oceania (n = 7, 19%), Asia (n = 6, 17%), South America (n = 2, 6%), and Africa (n = 1, 3%). Host organizations included nonprofit organizations (n = 14, 39%); academic institutions (n = 12, 33%); government agencies, including federal, state, and tribal (n = 7, 19%); and for-profit companies (n = 3, 8%). Partnerships were plentiful, with ten projects (28%) designating one or more type of organization as host. Our sample included all participation models according to the Haklay (2013) typology, though not evenly. The sample included a majority of participatory science projects (n = 21, 58%), followed by crowdsourcing (n = 13, 36%), distributed intelligence (n=2, 6%), extreme citizen science (n = 2, 6%), and volunteered computing (n = 1, 3%). Several projects reported multiple participation models, for example offering options that included participatory science contributions as well as crowdsourcing tasks. In terms of geographic scope, 11 projects (31%) were global in reach, 11 (31%) were national, six (17%) were tied to a locality such as a city or specific site, five (14%) were regional, and three (8%) involved online-only participation with no geographic component. Most of the projects involved data collection at sites chosen by the contributors, but several involved assignments to work in specific locations.

Data life cycle

Data quality

Interview participants reported a high number of QA/QC mechanisms (Figure 1). All projects used at least one QA/QC method, while 34 (94%) used more than one method, and 22 (61%) utilized five methods or more.

Figure 1 

Number of quality assurance/quality control (QA/QC) methods per project.

First, twenty projects (56%) conducted expert review, and six (17%) leveraged human expertise through crowdsourced review. Additional data validation strategies included voucher collection (n = 9, 25%), algorithmic filtering or review (n = 5, 14%), and replication or calibration across volunteers (n = 4, 11%). Fourteen projects (39%) removed data considered suspect or unreliable, while nine (25%) contacted volunteers to get additional information on questionable data.

Second, projects focused on the human aspects of data quality through training before data collection (n = 25, 69%) and/or on an ongoing basis (n = 11, 31%). Seven projects (19%) used targeted recruiting to find highly qualified volunteers. Four (11%) conducted volunteer testing or skill assessment.

Third, many projects approached data quality through standardizing data collection or analysis processes. Twenty-two (61%) used a standardized protocol. In addition, five (14%) used disciplinary data standards (e.g., Darwin Core for biodiversity data), and five (14%) used cross-domain standards (e.g., of the Open Geospatial Consortium [OGC]).

Fourth, many projects enabled data quality through instrument control. Fourteen (39%) used a standardized instrument for data collection or measurement. Five (14%) reported processes for instrument calibration.

Finally, a handful of projects documented their data quality practices. Seven projects (19%) shared what was classified as other documentation on a project website, while one project in our sample (3%) offered a formal QA/QC plan.

In addition to reporting on practices, many participants spoke at length about their data quality practices. Some indicated that data quality was secured through “very simple protocols and instructions.” Upon reflection, one noted that the use of simple protocols led to data collection practices that were “standardized, but not deliberately standardized.” Participants also taught us that data quality practices are often rich and contextual. One explained how data were vetted according to a six-pronged approach, where all published observations must be specific, complete, and appropriate (“the content is professional and for the purpose of education, not political or to further personal agendas”). A second described how traditional data quality metrics, such as temporal accuracy, were less relevant to their work than the ability to offer a detailed reporting of a phenomenon of interest.

Adoption of Best Practices

We found projects were generally implementing best practices with regard to data quality (as described by Wiggins et al. 2011), but were not implementing, and generally not aware of, best practices with regard to aspects of data management such as data documentation, discovery, and access.

In regard to data quality, we were encouraged to see the wide range of practices that projects employed. The majority of our sample (34 projects, 94%) used more than one method to ensure data quality, and 20 projects (56%) used five methods or more. That said, many could only articulate data quality methods when prompted, and only one had a systematic documentation of QA/QC through a formal plan. This suggests that, contrary to some external skepticism (e.g., Nature 2015), the issue with citizen science and data quality is not in actual practices, but with the documentation—or lack thereof—to describe the care and consideration taken with QA/QC.

Many projects demonstrated willingness to make their data available, for example by suggesting that data would be shared upon request. But we found that such de facto attitude to open access was not always backed by the appropriate licensing required to establish the legal (and ethical) conditions required for reuse, nor was provision of access in formats accessible to human and machine users alike a dominant practice. This finding supports prior research conducted within the field of biodiversity, which found that out of different types of data hosted in GBIF, citizen science data were among the worst documented and most restrictive (e.g., especially by prohibiting commercial reuse; Groom, Weatherdon, and Geijzendorffer 2016). While seemingly egalitarian, progressive, and in keeping with the community ethos of some citizen science initiatives, the restriction on commercial uses or the inappropriate application of share-alike licenses⁶ can prevent third parties from providing value-added data and services based on raw data, and may stymie private sector research and innovation that could be in keeping with project and participant values. It may also hinder a project’s goals; for example, a primary customer of citizen science data for mosquito-vector monitoring could be commercial mosquito control groups. In addition, if citizen science data are enhanced owing to significant investments by companies, they may represent a real value proposition for all data consumers, including citizen scientists themselves. In such cases, CC-BY-NC and CC-BY-SA licenses can be viewed as regressive and not in keeping with open science principles, though the debate is nuanced and open. For example, biodiversity observations shared under an NC license cannot be used on Wikipedia (which supports a broader open data policy) to illustrate articles about species for which citizen science data may be the primary or best available records.

Some citizen science projects implemented best practices in regard to data governance, including access and control. Many solutions, such as location obscuration or masking PII, were designed to protect the privacy of humans and/or sensitive species. Further, at least a handful of the projects that did not leverage these solutions had thought about implications like privacy and made a deliberate decision to prioritize, for example, principles like notice and informed consent (see also Bowser et al. 2017).

In respect to data provenance and traceability, the use of DOIs and appropriately explicit licensing statements is an issue for establishing scientific merits. One respondent indicated that “The data could have been referenced in publications, but we don’t know about it,” a situation that could be remedied by the use of DOIs. The global biodiversity informatics community has long recognized this issue too, and has made some progress on data archiving (Higgins et al. 2014). As one example, GBIF, together with its partners and members, implemented DOI minting and tracking mechanisms to link publications citing data sources with the original source data, which include datasets sourced from citizen science projects. While users are not required to use DOIs or even to attribute to the referenced data (CC-BY is a “requirement” that may not be enforced), it is becoming an increasingly prevalent practice in the science community. This is a persistent issue related to data citation practices: It’s harder to establish impacts for fully open access data. And some practices, such as requiring registration for access to data can help to track usage, but may serve as an impediment for some users (Wiggins et al. 2018).

Finally, when datasets are not adequately described with relevant metadata, their potential for secondary uses is significantly compromised, frequently resulting in whole datasets being discounted as untrustworthy and reinforcing the perceptions of poor rigor in citizen science. Addressing this perception is critically important for citizen science–generated data to gain more trust within the research sector.

Across all aspects of data management, we found a few projects following best practices in every category, but most projects had a mishmash of practices and a clear work-in-progress narrative with respect to evolving practices as project activities progressed. As one respondent commented, “We really want scientists to use the data but we’re not at a point where we would recommend that they use the data,” and multiple projects reported plans to achieve higher levels of data management for several items we asked about. Further, many respondents, including project managers who had dedicated IT support or leveraged an external platform, often did not know details of their data management practices, as these duties were delegated to others (consistent with Wiggins et al. 2011). In a similar vein, several respondents noted that they had not written their project’s data management plans nor designed the technological workflows themselves; these tasks had been outsourced, leaving our respondents unable to fully answer the questions asked.

This was particularly notable for projects whose data access, services, and persistent identifiers were provided by a platform that offered data hosting. While this may be a reasonable option, particularly for smaller or start-up citizen science projects, and whereas taking advantage of the expertise of an interdisciplinary team is often advocated, it can clearly lead to a lack of awareness about data practices, with potential consequences for data strategies. Outsourcing may lead to, or be a sign of, inattention to the importance of decisions made by the data host. This inattentiveness could lead to issues down the road if infrastructure should fail or security be lax. As one respondent explained, each project that was affiliated with the larger program made their own data sharing decisions, but deciding to make data openly available did not mean that that the lead researcher assumed responsibility for depositing the data into an open access database with a persistent identifier. Project managers were not always sure who was responsible to carry out policy-oriented dictates for data management and preservation. While not all data need to be archived, at present probably too little are being proactively preserved for the long term.

In some cases, the adoption of best practices in citizen science data management may be similar to or lagging only slightly behind those of conventional science. For example, we found that in regard to data discovery and access, ten projects (28%) made data available through a topical or field-based repository (such as GBIF), eight (22%) through an institutional repository, four (11%) through a public sector data repository, and two (6%) through a publication-based repository. In comparison, Tenopir et al. (2015) found that 27.5% of the researchers in their survey made their data available through a discipline-based repository, 32.8% through an institutional repository, and 18.4% through a publication-based repository. Comparing these studies suggests that both citizen and conventional science lag far behind the ideal. But the consequences are more significant for citizen science. Widespread adoption of best practices in data management in citizen science would provide much needed transparency about data collection and cleaning practices and could go a long way in advancing the reputation of the field. It could also help satisfy citizen science’s commitment to ethical principles, as outlined in the Holdren Memorandum and ECSA’s 10 principles of citizen science (Holdren 2015).

While the questions on data management and discovery practices often focused on a scientific user audience, it is important to recall that the scientific research community isn’t always the primary audience for a citizen science project: Local communities, students, or other parties may be a target audience, for whom access through a project website is preferable and analyzed products may be preferred over raw data access. However, data access is also reflective of current archival practices and long-term stewardship choices. From this perspective, most of the projects in this study were not positioned to ensure long-term access to data, and in the majority of cases, data sustainability appears tenuous at best.

Infrastructure and technology impacts

Databases, software applications, mobile apps, and other e-infrastructures supporting citizen science have a significant role to play in facilitating improvements in data quality. Such infrastructures can, if they conform to appropriate standards and use good design principles, make the data more discoverable, more accessible, more reusable, more trusted, more interoperable with other systems, more accurate, and less prone to human-induced errors (Brenton et. al. 2018). Good design and open infrastructures enable efficient and simple data recording and management by using workflows, processes, and user-centered design to minimize the risk of user errors and ensure that consistent data formats and mandatory attributes are recorded correctly, along with consistent use of vocabularies, spatial referencing, and dates. At the same time, providing project managers with adequate and easily understood reference information about the default policies that apply to hosted data seemed to be a clear gap for our respondents.

At the global scale, and indeed in many countries, it would be fair to say that the e-infrastructures currently supporting the majority of citizen science projects are largely functioning independently of each other and are not often adequately ascribing metadata to describe the datasets and methods. In addition, very few e-infrastructures are currently implementing any commonly used data standards. This effectively isolates these systems from each other and from being able to share data in ways that can open doors to important new scientific insights through, for example, larger aggregated views and analyses based on spatially and temporally dense datasets.

However, there are examples in some countries where efforts are being made to bridge the e-infrastructure divide. Firstly, the Public Participation in Scientific Research-Core (PPSR-Core) project is an initiative of the citizen science associations (US, European, and Australian Citizen Science Associations) in partnership with the OGC and World Wide Web Consortium (W3C) to develop a set of standards for citizen science data, metadata, and data exchange protocols. Within each of the association regions there are separate third-party platform-based initiatives to support individual citizen science projects (e.g., CitSci.org, Zooniverse, iNaturalist and SciStarter [US]; BioCollect [Australia]; and Spotteron [Europe]). Some of these multi-project platforms are already implementing the PPSR-Core standards as they evolve and are already sharing project-level metadata amongst each other to improve the discoverability of citizen science projects. As a next step, researchers working with Earth Challenge 2020 and the Frontiers open access publication series are creating a metadata repository to facilitate the discovery and access of citizen science data.

Assuming that standards and best practices already exist in an accessible and usable form (which was not universally the case at the time of writing) to apply them in e-infrastructure and data management solutions, providers should codify them into their software to ensure consistency and offer guidance for users, particularly those inexperienced with such matters. However, one interviewee noted that adopting a third-party platform to manage their data did not allow them to direct data management practices because they didn’t have control of the technical infrastructure to impose their own field-specific or project-specific preferences. This presents a significant challenge for infrastructure providers, as it suggests that software is expected to be both highly configurable around individual user needs while applying standards, rules, and workflows that assist users to apply best practices in data collection and management. At the extremes, these are diametrically opposed concepts, but it is possible to provide flexible solutions within a standards-constrained environment. Achieving the right balance between flexibility and appropriately structured constraints will require both project owners and infrastructure providers to be aware of standards and best practices, as well as for providers to be transparent as to if or how they are applied in their platforms.

The human dimension

A fundamental rationale for improving data management practices in citizen science is to ensure the ability of citizens, scientists, and policy makers to reuse the data for scientific research or policy purposes. Mayernik (2017) explores how hard and soft incentives can help support open data initiatives. Hard incentives include requirements by funders like the National Science Foundation (NSF) in the USA for researchers to supply data management plans or requirements from publishers that mandate publishing data in conjunction with a research article. Mayernik also uses the concepts of accountability and transparency to explore additional factors that may limit reuse. Transparency includes requirements for making data discoverable and can be charted on a spectrum. For example, providing a link to data online with brief textual descriptions is less transparent than registering data in a catalogue (metadata repository) with standardized descriptions and/or tags.

Culture also has a significant role to play. In line with broader discussions of open science (David 2008; Levin and Leonelli 2017; Munafò et al. 2017), traditional academic cultures often fail to incentivize researchers for good data management to enable reuse. Here, the use of DOIs can be a technical solution that also enables cultural change if researchers can get credit when other researchers are able to find, use, and ultimately cite their data. There is also an opportunity for cultural change specifically within the citizen science community. By evoking aspirational guidelines such as those outlined in the Holdren Memo and ECSA’s 10 principles (Hodren 2015), linking good data management practices to already-articulated community values like transparency can create pressure for researchers to make their data more discoverable and accessible as an ethical imperative.