Of all the stakeholder groups, calls to improve metadata quality fall perhaps most frequently on publishers. Publishers could “clean up the information that they provide to vendors about their items, which would help vendors create higher quality records” (Flynn 2013, p. 30), and thus also benefit those who use vendors’ services. Fault is also found with vendors’ and publishers’ semiautomated metadata collection techniques, which can leave out core information. One case study profiles the difficulties encountered by librarians due to the lack of ISSN as a standard piece of metadata in “unsolicited submissions or in collected aggregator/publisher notifications”; without it, it is much more difficult to determine the status of access restrictions (Bull and Schultz 2017, p. 14). Another finds that publishers “could hire professionally-trained catalogers on staff to create high-quality cataloging records for their items,” though there is perhaps a more direct incentive to do so for vendors than publishers (Flynn 2013, p. 30). 

On the other hand, there is an observable trend of publishers giving metadata more serious consideration. According to a 2017 survey of industry leaders, 90% of all publishers are planning to invest in metadata over the next three years (Imbue Partners 2017, p. 11). Publishers are looking to take this action despite budget concerns and the fact, acknowledged by Kemp et al. (2018) that “... enriching existing metadata records can be difficult and time consuming” (p. 208).

The reason for this shift in focus is in part due to the incentives that have been associated with good metadata. Two white papers by Nielsen, a publisher with a presence in the US and UK, found increased sales to be associated with quality metadata. Specifically, a quantitative analysis on the sales of Nielsen’s top 100,000 book titles in the period between July 2015 and June 2016 showed higher sales for titles which included basic metadata, a set of 9 elements from the Book Industry Communication standard: ISBN, title, format/binding, publication date, Book Industry Standards and Communications (BISAC) subject code, retail price, sales rates, cover image, and contributor. Titles with these elements had sales 75% higher than those with incomplete metadata. The study also found that supplemental descriptive metadata such as author biography, reviews, and title description, give a boost to sales. On the whole, each element of descriptive metadata boosts sales, with an increase of 72% for titles having 3 descriptive elements over those having none. (Walter 2016, p. 9). The fact that “metadata was considered of highest importance in digital transformation – a 4.6 out of 5” by industry leaders, combined with the fact that metadata “had the second lowest rating for current organizational ability – a 2.0 out of 5,” suggests that publishers will invest more in metadata in coming years (Imbue Partners 2017, p. 6). 

Moreover, investments in developing technologies for metadata promise substantial rewards. Some publishers are exploring AI-generated keywords and abstracts as well as chapter-level metadata which may lead to more granular search functionality (Tan 2018, p. 20). Ruth Pickering of Yewno sees possibilities in the developing area of full-text semantic analysis. Line-by-line subject analysis of publications, also known as semantic analysis, will allow publishers to generate categorization and keyword metadata for backlogs of poorly cataloged publications (Silverchair Information Systems 2018, 18:00). While semantic analysis will likely not replace traditional metadata generation and search techniques, it might also help to overcome inconsistencies among publishers (Silverchair Information Systems 2018, 20:00). In a test of this technology, Westchester Publishing Services applied semantic analysis to a backlog of 200 academic book titles, hoping to “‘... create discoverability-focused metadata tags and other supporting metadata and content to add value for these titles when they are placed online’” (Tan 2018, p. 30). The results of this effort have not been reported. Similarly, AIP Publishing is using a semantic tool from Access Innovations to generate keyword metadata for articles focused on the physical sciences (Society for Scholarly Publishing 2018, 16:17). Semantic analysis may have wider implications for the publishing industry and scholarly communications as a whole. In a panel discussion at Davos 2016, Stewart Russell of University of California at Berkeley predicted that “If the search engine industry is worth a trillion dollars roughly right now, this new level of technology [semantic analysis] could be worth 10 trillion because it will have so many more applications” (World Economic Forum 2016, 7:52). 

New technologies pose interesting opportunities for metadata, but their effectiveness is not yet explored in the literature possibly because of their proprietary nature. The extent of the use of semantic analysis within publishing as a whole has not yet been surveyed. Equally lacking is a deeper look at the roles of traditional catalogers and metadata experts in publishing companies, a study of which could provide more clarity regarding the differences between various publishers in the way that metadata is created and used. Again, proprietary concerns might be a cause, in addition to publishers who think of themselves more as self-contained units rather than part of a larger metadata ecosystem. The same concerns may prevent adoption of share controlled vocabularies which would aid in making discoverability easier across publishers. 

Metadata problems can arise when indexing practices differ between publishers and service providers. Publishers may index content at a different level than service providers. For instance, a database vendor might index several small articles under one title when a publisher indexes them discreetly. Service providers may not index some but not all of the articles in a journal issue, or they may index each article in an issue but not the issue itself (ExLibris 2017). 

Problems also arise in the area of linking to content. Service providers create platforms, such as Integrated Library Systems (ILS), that allow users to discover bibliographic resources. These platforms feature links to full-text versions of content housed on external websites. The act of clicking on a link and arriving at the correct resource, known as link resolution, can be adversely affected by inconsistencies in metadata between content vendors, publishers, and providers of platfoms. ExLibris (2017) provides the following example: "You find an Article from Provider A, and go through a link resolver to link to Full Text from Provider B, it can happen that the metadata for the same Article from Provider A and Provider B are not the same, so Provider B cannot match the article, and the link fails." In the absence of a direct stable URL, the OpenURL standard (NISO Z39.88-2004) is used to facilitate linking from central indices like Primo or Summon to a target (e.g. article or ebook) on a publisher or aggregator platform. But when an OpenURL contains typographical errors or is poorly formatted, links fail to resolve.

Inconsistent or incomplete use of metadata fields also makes it a challenge to disambiguate similar titles, journals, and authors across distribution systems. If the metadata utilized in these systems is not normalized to authority sources or linked to unique identifiers, it is difficult to tell if two similarly-described resources are in fact the same. This serves as an inconvenience for individual researchers but also impacts our knowledge of the scholarly ecosystem as a whole. One research study detected significant problems in researchers’ use of bibliometrics to analyze author networks because of lack of disambiguation. Analyses which falsely equate authors with the same or similar names distort our understanding of author networks and “may result in ill-informed decisions about research policy and resource allocation” (Kim 2017). Disambiguation could remain an obstacle, especially for text and data mining and the continued use of bibliometrics, though the adoption of ORCID is providing a solution. 

Organizations such as the Scholarly Publishing and Academic Resources Coalition (SPARC) and the Confederation of Open Access Repositories (COAR) have developed best practice principles for data repositories which can also be applied to vendor services. Among the principles are calls for open metadata and usage data, transparent pricing and contracts, allowance for community input, and use of well-established international standards to promote interoperability (Confederation for Open Access Repositories (COAR) and Scholarly Publishing and Academic Resources Coalition (SPARC) 2019). Several organizations have fulfilled aspects of these guidelines through policy changes and initiatives. In 2014, EBSCO opened metadata in its 139 databases for use by third parties (Enis 2014). In 2015, a partnership of health information vendors, health information management professionals, and other medical professional associations formulated standards for ensuring interoperability between different health information systems (Information Technology Infrastructure (ITI) Planning Committee 2015). The standards include statements as to the importance of quality metadata in maintaining vital information about data provenance (ibid., Table 8, Table 10.C). These two major initiatives demonstrate the rewards of improving metadata and access to metadata among services providers in scholarly communications.

One challenge for other stakeholders in understanding the metadata practices of service providers is the large number of providers and tools offered. The literature would benefit from a comparison of different kinds of platforms and tools offered by service providers. Platforms and tools could be classified according to the impact of metadata for collection, curation, or consumption. Librarians in particular benefit when service providers are willing to adopt open metadata and usage data, transparent pricing and contracts, allowance for community input, and use of well-established international standards to promote interoperability. 

While researchers understand the value of making their work available to others, literature on the subject draws attention to the fact that the quality of their metadata does not always follow suit. Authors may be required to submit metadata along with traditional articles or monographs. Author-supplied metadata may benefit from the author's expertise, but is also, writes Tan (2018), prone to typos and outdated or incorrect information. One common problem, for instance, is the rate at which email addresses and other contact information becomes dated (p. 28). An earlier study gives weight to this claim, concluding that authors and article contributors are in the ideal position to provide valuable metadata for their articles given their deep knowledge of the subject (Wilson 2007, p. 16). Problems arise however, in the area of formatting metadata consistently. For 104 articles in the discipline of music, researchers identified 201 structural metadata issues, or mistakes in formatting (ibid., p. 21). While contributors expedite the metadata submissions process and lend expertise, “efficient, usable systems and interfaces . . . guiding correct structural entry of metadata” are needed (ibid., p. 26).

On the whole, however, recent literature has focused more on researcher-generated metadata for data sets than on metadata submitted for published articles. On the subject of metadata for research data, a consensus emerges that researchers, while invested in the idea of sharing their data and findings, often lag behind in practice. Two studies of researcher perspectives on publication of data find that “researchers frequently fail to make data available, even when they support the idea or are obliged to do so” (Kratz and Strasser 2015, p. 2) and that “despite its potential and prominent support, sharing data is not yet common practice in academia” (Fecher, Friesike, & Hebing 2015, p. 18). One reason for this phenomenon could be the time and effort required to do so. Researchers most frequently cited time investment as a barrier for sharing research in a number of surveys (Kratz and Strasser 2015, p. 2).

The ability to manage large sets of data is increasingly important for researchers across disciplines. In the sciences, however, educational and training programs for researchers are still catching up. In one case, researchers published their experiences adopting new practices to manage data for the Ocean Health Index project. They found that “the need to improve practices is common if not ubiquitous” among environmental scientists who work with large data sets (Lowndes et al. 2017, p. 1). Echoing this concern, Barone et al. (2017) found, among a set of principal investigators in the environmental sciences who received National Science Foundation grants, 87% work with big data sets. When surveyed, the PIs reported that their research institutions were not meeting their training needs. 78% stated that their needs went unmet in the area of “data management and metadata” (p. 4). 

The learning curve which researchers must negotiate to manage research data is not the only barrier for researchers who want to publish their data with appropriate metadata. Researchers must also manage the requirements and standards, sometimes conflicting, of different repositories, journals, and funders. In “Data management assessment and planning tools," Andrew Sallans and Sherry Lake find that “researchers’ current practices appear fragmented largely because funding agencies propagate broad requirements and provide few resources” (Sallans and Lake 2014). Similarly, McQuilton et al. (2016) find that “the proliferation of content standards and databases creates a barrier for researchers and database maintainers, creating confusion over which standard they should use to format their data, or which database to submit their data to” (p. 2).

Interestingly, researchers’ reliance on informal sharing might create a disincentive for quality metadata. In exploring how researchers share and obtain data, direct contact rates higher than all other methods, including sharing and retrieval via data repository (Kratz and Strasser 2015, p. 7). On the one hand, these numbers could be somewhat inflated because researchers are more likely to remember incidents of sharing which took place as a result of direct contact. On the other, the ubiquity of personal contact could explain a finding in the same study that researchers have relatively low expectations that data sets be accompanied by formal metadata. 

Again, specifically in regard to data sets, a lack of shareable metadata has contributed to relatively low citation rates for data sets in published articles. While 49% of researchers thought that citation would be the most appropriate way of giving credit for data consulted (Kratz and Strasser 2015, p. 10), citation rates for data sets continue to remain low, with two previous studies finding that authors cited the data sets they consulted as little as 19% and 17% of the time (p. 16).

Researchers have a strong incentive to create good metadata. The quality of metadata associated with their scholarly outputs will affect the number of citations they receive, while utilization of quality metadata early in the research process will facilitate organization and save time down the road. A more selfless incentive, researchers also want to see knowledge advanced in their fields through better findability and wider access. These interests can be seen at work in the wide adoption of the Open Researcher and Contributor ID (ORCID), an initiative to assign a unique identifier to every author. These unique identifiers prevent confusion when authors change names or share a name with another individual: if a person publishes under the name Michaela Schaal in 2019 and then publishes an article under the name Michaela Petsch in 2020, it will remain clear through use of an ORCID that these names belong to the same person. Similarly, Michaela could publish as Michaela M. Petsch, M. M. Petsch, or any other configuration while remaining unambiguously associated with all her work. Searching by ORCID, researchers can be more certain that they are retrieving all the works of a given author and none from unintended authors. 

A similar effort applied to the citation and sharing of data could yield powerful results, though not perhaps without shifting away from the model of traditional citation (Kelly 2015, p. 12). Efforts to improve researcher metadata should not be limited to researchers. Poole (2016) finds that researchers show an interest in creating shareable metadata but are often unaware that help is available from metadata experts, most often in the form of librarians at their home institutions (p. 969). Metadata experts in academic libraries, research data management institutions, and data publishing organizations all have a role to play, especially in developing a practice for research data sharing. Further discussion is offered in another section of this review, Data Curators and Repositories.

A robust collection of literature profiles the problems that researchers encounter when managing research data, supplementing research on similar issues in the context of published books or articles. However, it remains difficult to attain a bird’s eye view of the subject given the variety in the ways that researchers approach metadata: Metadata for research data varies based on funder requirements, data management software, data repository specifications, and research discipline and training. Likewise, author-supplied metadata for published works will vary according to the fields required by journals/publishers, type of metadata submission form or tool, and, research discipline. For both research data and publications, researchers’ level of involvement in creating metadata may vary based on the requirements of their institutions or publishers. 

Studies that employ methodologies such as that of Wilson (2007) -- surveying a large number of publications to evaluate metadata supplied by authors and contributors -- would make it possible to more accurately depict a state of practice for researcher metadata. Given that community efforts for evaluating research data quality already exist (such as the widely deployed FAIR principles), readers would also benefit from an evaluation of research data quality across repositories or platforms. This evaluation could resemble a study by Neumaier et al. (2016) which developed criteria to measure metadata quality for open government/civic data across multiple sites and platforms. There is also a lack of understanding of the degree of consistency between institutions when it comes to practices for collecting metadata. A study of such practices would illuminate the researchers' training needs. 

Though research policy makers (mostly in government institutions) support the accessibility of research data (Harvey et al. 2017, p. 19), more could be done to understand “the discipline-specific barriers and enablers for data sharing in academia in order to make informed policy decisions” (p. 20). More generally, there is a concern that “the ways in which data travel among scholarly fields remains understudied” (Poole 2016, p. 966).

Government funders have strengthened requirements for researchers to make their research publicly available. For example, in 2008, NIH began to require researchers to submit the peer-reviewed, manuscript versions of their papers to PubMed Central, an open access database for medical research (Dietrich et al. 2012, p. 1). In 2011, NSF began requiring that all researchers include a data management plan in their funding applications to facilitate data sharing (p. 2). While data management requirements in humanities-focused funding bodies may be less stringent, the NEA, for example, requires that grant proposals include a data management plan (National Endowment for the Arts 2019). Despite these and similar funder requirements, one survey found that data sets were recoverable in only 26% of cases where sharing was required, or 81 of 315 projects. Of these 81, the authors found that 60 data sets contained sufficient metadata for formal archiving while 21 did not (Couture et al. 2018, p. 5). Though this low rate can be explained in part by the fact that the studies considered spanned 1989-2010, Poole (2016) finds that funder policies still lack sufficiently clear language and that they focus on access at the cost of preservation (p. 972). Mentioned above in the Researchers section, the finding of Sallans and Lake (2014) that “current practices appear fragmented largely because funding agencies propagate broad requirements” also applies. There is a consensus that funders acknowledge data sharing to be valuable but leave researchers without resources to ensure that their data is published and findable through quality metadata.

Beyond the wider benefits to the scientific community from access to open data, both private and public funders stand to benefit from more rigorous metadata requirements. For example, to measure the output of its Very Large Telescope (VLT) instruments, the European Southern Observatory (ESO) mandates that any article using data gathered from a VLT instrument include the ID of the project under which it was gathered in a standardized format ((ESO) 2018, 2018). This requirement allows ESO administrators to determine the number of citations (“science impact”) generated by their instruments on the whole as well as how many citations are generated by each of their 12 VLTs (Leibundgut et al. 2017, p. 11). Over time, this data will allow the ESO to better understand the impact of its instrumentation, what kinds of research are performed on which instruments, and to make more informed decisions on future instruments. The judicious use of metadata in this case, which has an impact on the fiscal health and long-term success of an organization, could be extended to other organizations, working in different disciplines, which have an interest in measuring the impact of their labs or instruments.

Foremost, the literature would benefit from studies which provide a systematic comparison of funder requirements for research data metadata as well as any guidelines provided, schemas recommended, or platforms specified. Additionally, government funding agencies are better studied than corporate funders even though, at least in the United States, corporate entities were responsible for 72% of research and development funding in 2015 (Boroush 2017, p. 1). Corporate funding bodies should be studied in the same manner as governmental bodies.

In “How libraries use publisher metadata,” Steve Shadle demonstrates how library discovery systems are impacted by publisher-generated metadata. (Shadle 2013) In the current environment, major academic libraries are paying third party suppliers, or library system vendors, for access to hundreds of thousands of electronic resources. Metadata records are not generated ‘by hand’ for most of these resources. Instead, metadata, most often created by publishers and content aggregators and reused by library software vendors and provide online access (via either direct linking or OpenURL link resolvers), is also used by librarians who download or receive MARC records in bulk for the packages they have purchased. For an electronic journal article, this metadata might include title, author, date, page numbers, ISSN, issue number, volume number, issue title, and DOI. In this way, writes Shadle, the choices of a publisher impact the librarian and end user: “In order for link resolvers to function properly, citation metadata coming from a source must be accurate. In most cases, this data is originally coming from the publisher” (Shadle 2013, p. 293). Kemp et al. (2018) echo these issues, writing that “... libraries face multiple external obstacles such as insufficient metadata provision upon the delivery of content from publishers (e.g. data about article license information), prohibitive costs of vendor systems and services, the absence of single-service options that provide multiple solutions” (p. 208).

Poor metadata quality brings into question the value of working with large numbers of publisher- and vendor-supplied metadata records. Records are generated by a variety of automatic methods, all of which are aimed at increasing efficiency when working with a large number of resources. Overall efficiency is somewhat diminished, however, if librarians are left without reliable ways to access a resource and if users of library databases see search results that are less than optimal. Wiersma and Tovstiadi (2017), for instance, found widely varying inconsistencies for search results based on which paid platform was used to search for ebooks. Librarians can encounter roadblocks when trying to confront vendors, library software vendors, or publishers about these problems (see the example in Bascones and Staniforth 2018, 2018, p. 13), and, when they do have the ability to make improvements to metadata themselves, “enriching existing records can be difficult and time consuming” (Kemp et al. 2018, p. 2018). Compounding the problem, Flynn argues that libraries lack entry-level cataloging positions (Flynn 2013, p. 30) at a time when they are very much needed. Given this and other problems, Bascones & Staniforth ask, “how many institutions are paying for items end-users will not be able to find in the library catalogues/discovery systems? How many disappointed end users are there?” (2018, p. 19).

Given the nature of these challenges, some see librarians as the best advocates for the end user and the health of the system overall because they “represent end-users who experience specific resources not ‘value for money’ packages” (Bascones and Staniforth 2018, p. 15). Furthermore librarians should not shy away from an advocacy role due to the fact that “Libraries account for a majority of vendors’ business, giving librarians leverage for demanding high-quality, full-level cataloging records” (Flynn 2013, p. 30).

Librarians are also perceived as experts who can benefit other stakeholders. Kemp et al. (2018) argue for the value of librarians who connect with publishers to create taxonomies and to researchers in describing research data sets (p. 209). Similarly, Poole (2016) calls for librarians to create direct relationships with researchers who might not think of the library as a collaborator (p. 969). Mercer discusses her direct experience in acting as an advisor to two nascent journal editorial boards as a model for others to do the same (Mercer and Dyas-Correia 2011, p. 239). Librarians can also channel their expertise into advocacy for new initiatives. Librarians were “instrumental” in the development of ORCID (Kemp et al. 2018, p. 2019). The Shared Access Research Ecosystem (SHARE), an initiative “to maximize research impact by making research widely accessible, discoverable, and reusable,” is a collaboration between the Center for Open Science and the Association of Research Libraries (Association of Research Libraries In press). Librarians have also had a significant role to play in research data management. Poole (2016) writes optimistically of their potential impact on the quality and preservation of data sets, and is most optimistic about the potential of archivists in particular “...archivists are increasingly well-placed to address pressing digital curation needs” (p. 969).

A useful contribution to the literature in the confluence of librarianship and metadata would be a study that takes up the aforementioned question of Bascones and Staniforth (2018), “how many libraries are paying for items that end users will not be able to find…?” Indeed, the question raised in their title, “Is wrong metadata really bad for libraries and their end users,” though answered with a resounding yes by the authors, is only answered from the perspective of the limited cases they were able to compile. More surveys like that performed in Wiersma and Tovstiadi (2017) would not only put librarians in a better position to make their case with publishers and vendors, but also serve to establish what in particular is wrong with the metadata in most cases.

As with researchers, publishers, and funders, librarians are a large group working in many different kinds of institutions with different subject knowledges. Hearkening back to the alternative metadata roles first mentioned in the Stakeholders section of this review, librarians are metadata creators, consumers, and custodians all rolled into one. While the differences between a cataloger, a health science librarian, and a reference librarian may be clear to those within the profession, the metadata needs of these various roles are not immediately apparent to those coming from the outside. Use cases and narrative examples written as far as possible in plain language would serve well in elaborating metadata problems for other stakeholders. Organizations like Metadata 2020, the Dublin Core Metadata Initiative (DCMI), and the North American Serials Interest Group (NASIG) each attempt to build our understanding of the metadata needs of different roles within the profession.

A movement toward public access to data has followed and has attempted to create best practices for the infrastructure necessary to make data available. Data infrastructure is the collection of hardware and software used to support the preservation and retrieval of data, and relies heavily on good metadata. For instance, storing data with an eye toward long-term preservation requires technical metadata elements such as file inventory, file format (FITS, SPSS, HTML, Stata, Excel, tiff, mpeg, 3D, Java, CIF), file structure including “organization of the data file(s) and layout of the variables,” version information including a date/time stamp for each version, checksum values, and the software and hardware in which the data were created (DMPTool In press). The Metadata Encoding and Transmission Standard (METS) is one metadata standard suited to capturing this information and is maintained by the Library of Congress (Library of Congress 2019). In terms of retrieving data, metadata is required which can accurately describe the data with subject-specific language while still facilitating access for the general public. For these reasons, “… activities involved in developing a data infrastructure are highly sophisticated and require various types of expertise (including discipline-specific knowledge and information science)” (Leonelli 2016, p. 394).

A closely related issue is the challenge of citing data. As referenced in the Researchers section, proper metadata plays a role in preserving the most important incentive for researchers to publish data, that of certainty that their data will be cited. In Kratz and Strasser (2015), 83% of respondents indicated that a citation was the best way of giving credit for use (p. 6), but rates of citation fall well below this benchmark now as in the past (p. 16). Echoing this finding, Fecher et al. (2015) write that researchers donating their data to open access repositories “do not receive enough formal recognition to justify the individual efforts and that a safeguard against uncredited use is necessary” (p. 11). Moreover, traditional citation formats do not leave authors with enough options for citing all of the people, software, and tools (often discipline-specific) that contribute to the research process.

Proper methods for storing and citing research data are still in development, but a number of initiatives have emerged with best practices to guide researchers and publishers when creating robust metadata for citation by others. With regard to citation, The Research Data Alliance’s Data Citation Working Group published a list of recommendations for citing data*4 (latest revision 2015) while DataCite assists with creating unique identifiers in the form of DOIs for data sets and data set updates (Kelly 2015, p. 15). CRediT from Casrai has brought nuance to the way that people are given credit, if not cited traditionally, for their contributions to research: authors and publishers can now choose between 14 different roles ranging from data curation to project administration to visualization (Casrai 2018). The Research Resource Identification Initiative (RRID) from PLOS is providing researchers, primarily biologists and geneticists, with a means for citing previously neglected sources of information such as models of organisms, antibodies, software, and databases (Hill 2015). Finally, the Software Citation working group at Force11 has developed recommendations for citing software used in academic research and software developers (Smith et al. 2016) at a time when authors of research software are increasingly looking to document, preserve and cite their software: At the time of writing, over 72,000 software DOIs have been minted via Zenodo, an open access repository (DataCite In press, accessed 2019-04-11), since the publication of “Making Your Code Citable,”*5 a github guide.

With regard to creating metadata for research data management, Force11 published the Findable, Accessible, Interoperable, and Re-usable (FAIR) principles in 2016 with associated metrics for evaluating machine actionable metadata and data quality. DataCite has created a metadata schema for data sets used by many repositories such as Elsevier’s Mendeley-based repository (Kelly 2015, p. 16). Still, with over 2,000 data repositories worldwide (Witt 2018, 2018) a consensus has yet to form on the most appropriate platform for hosting data.

In “A metadata-driven approach to data repository design,” Harvey et al. outline the process for creating a repository in alignment with the FAIR (Findable Accessible Interoperable Re-usable) principles. DOIs are first assigned to collections of data, e.g. groups of "logically related material" (DSpace In press), as well as to data sets, individual data entries, and even individual instances when data was queried by a researcher. The authors recommend starting by assigning a DOI to the data collection which, in DSpace, will automatically inherit the ORCID of its creator. Another step in the workflow employed by the authors is to register the data via r3data, a global registry of research data repositories that enables one to describe the repository content and obtain a persistent identifier (DOI) for the repository itself (Harvey et al. 2017). The principles for research data management have been built from the ground up and cross institutional and disciplinary lines. The collaboration around this issue serves as a model for improving the rest of scholarly communications metadata.

The promise of new initiatives has generated excitement but leaves questions as to their effectiveness which can only be answered as time passes. There is an opportunity for future research to quantify the adoption rates and impact of practices recommended by these initiatives. For instance, while Harvey et al.’s recommendation that DOIs be assigned to data collections, data sets, and individual data points would usher in a new level of transparency in scholarly research, a significant question remains as to whether researchers and data curators have the resources to implement it. The world of research data management would benefit from a service such as Crossref Participation Reports, which provides member organizations with an instant evaluation of their metadata quality and suggests ways to make improvements (Bartell 2018). 

Moreover, it has yet to be seen whether adoption of initiatives by some researchers and organizations, but not others, will create a discordant state of practice that is itself a barrier to access for researchers who seek to study scholarly outputs in the aggregate. It may be difficult to ascertain which citation practices were in effect for whom and at what time.