Open by default: a proposed copyright license and waiver agreement for open access research and data in peer-reviewed journals

Copyright and data

The policies and guidelines of many academic institutions advise researchers to establish intellectual property and copyrights at the start of any project (although whether the issue of data ownership is consistently addressed by researchers is unclear [19]). Copyright cannot generally be asserted in facts, only the ways in which they are presented. At a basic level raw data are merely simple, mathematical, descriptions of facts and to claim copyright a scientist would need to exert individual judgment, expression or skill in their representation. For example, Einstein could not claim copyright in the formula E = mc², but could in text explaining the theory behind it [20]. You could conclude from this that copyright and associated licenses and attribution requirements cannot legally be applied to data. However, there are many levels at which data – particularly digital data derived and integrated from different sources – and collections of data and metadata can operate and be represented, and many ways in which copyright law is applied in different jurisdictions.

In the US the law focuses on creativity (“Copyright does not protect facts, ideas, systems, or methods of operation, although it may protect the way these things are expressed”) but in Australia originality is more important – and copyright may well apply to research data “in the same way that it applies to written works like books, journal articles and reports” [21]. In the European Union “sui generis” rights exist to protect data within digital databases – effectively, copyright – which can, furthermore, be implemented differently by member states. Because of these substantial international legal differences regarding how copyright can be applied to data, there are inherent difficulties in ascertaining the extent of copyright in a dataset. A more comprehensive summary of the different approaches to copyright in data and databases can be found in [22]. All of these issues compound the uncertainty about what an individual or machine (such as a computer crawling the web) can do, legally, with information they download from the internet, including from journals.

Licenses and waivers for data

A license is a legal instrument for a copyright holder or content producer to enable a second party to use their content, and apply certain conditions and restrictions to those uses. A waiver is also a legal instrument but is designed for a rights holder to give up their rights, rather than assert them. For a comprehensive guide to the different approaches to the licensing of research data see [23].

Placing restrictions on the reuse of scientific information, particularly data, slows down the pace of research. Furthermore, legal requirements for attribution ingrained in licenses such as CC-BY can prohibit future research across large collections of content – as commonly happens in data mining research. Consider the Human Genome Project: a watershed moment for scientific data sharing and collaboration. Without the collective effort of many different research institutions, commercial organizations and individual scientists the sequencing of the human genome would not have been possible. But if a researcher wishing to query the human genome database as part of a new research project was legally required to attribute all the – probably thousands – of data contributors, by providing a link back to or citation, this would be unmanageable, and probably un-publishable in the context of a traditional research paper’s reference list.

International legal differences, described earlier, are another important reason to apply specific, appropriate legal tools to data. Also, it can be unclear what license to attach to copyright in a dataset or structure (for example a textual description of building the dataset could fall under CC-BY, but if source code were used rather than text it might not). This is an area of confusion where no licensing standard exists. Therefore, to eliminate legal impediments to integration and re-use of data, such as this stacking of attribution requirements in large collections of data, and to help enable long-term interoperability an appropriate license or waiver specific to data should be applied. There are a number of conformant licenses and waivers for open data [24], of which Creative Commons CC0 (http://creativecommons.org/publicdomain/zero/1.0/) is widely recognized. Under CC0, authors waive all of their rights to the work worldwide under copyright law and all related or neighboring legal rights they have in the work, to the extent allowable by law. Legal experts have recommended the use of standard, globally accepted licenses for data instead of developing ad hoc models [25].

The case for CC0 for scientific data

The Creative Commons’ website catalogues a number of different organizations – publicly and privately funded – which use CC0 for data [26]. These include:

Data repositories are particularly relevant users of waivers and licenses for research data. Although there are many data repositories in life sciences (for a list see http://www.datacite.org/repolist), which are growing in size and number, not all scientific domains have a common repository and journals often function as repositories when data are included as additional files (supplementary material). Dryad (http://datadryad.org/) is an international repository for the datasets supporting published, peer-reviewed journal articles across the biosciences which requires authors to explicitly place deposited data in the public domain using the CC0 waiver. An entry on the Dryad weblog sets out cogently why CC0 is the most effective solution for achieving its goals:

“By removing unenforceable legal barriers, CC0 facilitates the discovery, re-use, and citation of [that] data…

“Furthermore, Dryad’s use of CC0 to make the terms of reuse explicit has some important advantages:

Dryad’s policy ultimately follows the Science Commons’ recommendations, set out in their Protocol for Implementing Open Access Data [30].

The online laboratory notebook software LabArchives (http://www.labarchives.com/), which includes the ability to share data privately and to publish datasets publicly and permanently online, also uses CC0 for public datasets [31].

Credit where credit’s due – attribution and citation

A common concern about moving from an attribution license, such as CC-BY, to CC0 for data and waiving attribution rights is that academic credit (citations) will be lost if there is no longer a legal requirement to attribute the original rights holder (author). While attribution can sometimes be achieved in the same way as citation the two practices serve different purposes. Attribution is a legal tool designed to permit copying, distribution, and creation of derivative works such as translations. As copyright does not protect ideas (in the US), to give scientists credit for their ideas the established norm in scholarly communication is citation [32].

The examples in Table 1 demonstrate that, although they can sometimes be achieved in the same way, attribution and citation are not the same. Citations are much more important and relevant than attribution when tracking scholarly outputs and giving appropriate credit for individuals’ contributions.

Compared to legal requirements, cultural norms benefit from flexibility, and can evolve with the community which established them. In other words, using norms retains control and decision making within the research community, instead of taking it out of our control and handing it to lawyers and judges. Many scientific ideas, after a number of years, become undisputed and the community may deem it unnecessary for credit to be rigorously applied. For example, it would today be very unusual for an article describing a DNA sequencing experiment to cite the original work by Watson and Crick that elicited DNA’s structural properties. This is a cultural norm at work, where an idea is now so widely accepted, and the initial authors clearly recognized for their discoveries (in citations and prizes) a citation is not needed.

Jonathan Rees, formerly of the Creative Commons, said of community norms for influencing behavior over legal requirements: “For widest latitude of use and best scalability, and therefore greatest return to the research community, the entirety of the data set, including any incidental copyrightable elements, should be dedicated to the public domain. Note that public domain is not incompatible with a request for attribution or other terms of use following community norms. Such a request may be as effective – or more effective – at getting users to follow desired practices as any attempted legal restrictions[33].”

To our knowledge there have been no empirical studies of the citation of scholarly datasets assigned public domain dedication licenses compared to a comparable group available under attribution licenses. However, given public domain dedication – and specifically CC0 – is intended to maximize the potential for data discovery, reuse and therefore citation, it would be reasonable to hypothesize that citation potential of public domain data would be unaffected or might even increase. With CC0 there is no need for transfer agreements and preconditions, which inherently impede further (re)uses of data. Sharing of microarray experimental research data underlying journal articles has been associated with increased citation share [34], and increased reproducibility and repeatability of results [35]. In social science data collections funded by the National Institutes of Health and National Science Foundation in the US, data sharing has been associated with “many more times the publications” than collections where data were not shared [36]. Linking of publications to supporting datasets has also been associated with more citations to the linked paper in the marine science journal Paleoceanography, according to a conference abstract, and in the field of astronomy according to a pre-print paper [37].

Competition

Researchers who apply CC0 to their data, or any other product of their scholarship, waive all rights in that data allowable by law. Such a waiver has been described as an “unattractive option for data whose creators have yet to fully exploit them, academically or commercially” [23]. This is true, but a waiver or license required by a journal or publisher generally applies only in the context of data submitted for publication. If a portion of a large database was analyzed and an additional data file included for publication the larger, unpublished, body of work would retain whichever license the researchers, their employers or institutions require. In other words, researchers remain in control of what they chose to publish – what they submit to a journal – and a change in the publishers’ license does not affect this. Moreover, waivers and licenses for journal articles do not replace existing, established community norms for sharing of some data types (e.g. depositing microarray and genetic sequence data in appropriate databases) – nor do they affect requirements of many journals for sharing readily reproducible materials including raw data on request [38].

There is a trade-off between the additional opportunities which may result from transparency (such as new collaborations, secondary use) and the threat, improbable or otherwise, that opening up data may be valuable to competitors. Certain types of research, such as genetic sequencing to elucidate susceptibility to disease, generates far more data than one research team could conceivably analyze – which logically lead to sharing and collaboration. A number of companies have opened up some of their data and seen benefits [39]. A lot of data may have commercial value but much raw data, such as protein sequences of potential drug targets, are just the beginning of a knowledge-discovery process. More can be gained by “pre-competitive” sharing with the waiver of intellectual property. Such an approach is being championed by Sage Bionetworks in the US [40].

Plagiarism

Plagiarism is research misconduct and an unfortunate but ineliminable occurrence in scholarship. Plagiarism and the potential for plagiarism have increased with the proliferation of digital access to information [41]. Plagiarism is often not illegal, but it is certainly unethical, and undoubtedly damaging for the career of someone guilty of perpetrating plagiarism. Effective online tools for detecting plagiarism exist, such as CrossCheck (http://0-www.crossref.org.brum.beds.ac.uk/crosscheck/index.html), as does human detection via the peer-review process. Removal of a legal requirement for attribution for data elements of articles would be unlikely to impact on the potential for plagiarism. In addition, CC0 would not apply to the main, copyrightable text of articles.

Other safeguards

Public domain dedication of data does not mean that those who generated the data cannot express certain wishes about how the data are used. The Panton Principles frequently asked questions (FAQs) state: “You should always aim to follow any reasonable requests made by the data owners/publishers. These may be explicit or may be implicitly understood by the community. You should make an effort to understand any relevant ‘community norms’ for the data you are using[42].”

A code of conduct has been proposed for those wishing to reuse clinical trial data obtained from other researchers [43] and a clinical trial dataset published in the journal Trials, by Sandercock et al., has requested that “any publications arising from the use of this dataset acknowledges the source of the dataset, its funding and the collaborative group that collected the data.” [44].

Electronic publishing platforms provide further safeguards to ownership and authorship of published content, in the form of date stamping manuscript submissions and version control in some repositories, such as Edinburgh DataShare (http://datashare.is.ed.ac.uk/).

Citation of articles and datasets is facilitated through standard citation formats – such as those advocated by DataCite where persistent dataset identifiers, such as digital object identifier (DOI) names, are displayed as linkable, permanent URLs – and are increasingly supported by some publishers [37].

Tabular data

Data elements organized in columns and rows – a table – are extremely common in scientific publications. While attribution would be required for reproduction in whole or in part of a table as presented in a journal publication, the individual values and collection of values should be considered as data and therefore open. Data, in the course of a scientific experiment, are usually collected at a greater level of detail than are reported in a paper, with tables reporting summary or mean values. Although these data are aggregated from the raw data they remain numerical representations of a fact, and therefore data. Tables are furthermore often included as additional files in a variety of formats including PDF, HTML/XML, DOC and Excel/CSV. Ideally all tables included in the main body of a journal article should also be included as additional CSV files – an open, machine-readable format – but when they are not present as tables in journal articles CSV would represent good practice for tabular data. Proprietary file types, such as Microsoft Office, and formats which are not readily editable, such as PDF, are not recommended for tabular data provided as additional data files.

Graphs and graphical points

Graphs, graphical representations of relationships between variables, are ostensibly images and therefore not, when considered as a collective entity, data. However, the individual data points underlying a graph, similar to tables, certainly are. An example of best practice when submitting a manuscript with a graph to a peer-reviewed journal would be for authors to also submit accompanying CSV tables with the corresponding data points, so that graphs could be re-plotted. Although this practice is required by some specific journals it is not widespread. However, software tools exist that are capable of “scraping” underlying data points from graphs and images (for example http://www.chardta.com/) and can be useful, for example, for enhancing the discoverability of scientific information by exposing underlying data points to internet search engines.

XML

According to the World Wide Web Consortium (W3C), “Extensible Markup Language, abbreviated XML, describes a class of data objects called XML documents and partially describes the behavior of computer programs which process them” [48]. XML is widely used as a standard for data transfer and for creating versions of works intended for machine reading, and therefore to be used as data. Therefore for our purposes we can assume XML files are data. XML has many applications in science and is frequently published with journal articles as additional files in BioMed Central journals as well as underlying the online articles themselves. XML forms the basis of many domain-specific data standards such as Gating-ML in flow cytometry, FuGE-ML in functional genomics, GelML in gel electrophoresis and so on (see: http://biosharing.org/standards_view).

Bibliographic data

Bibliographic data have been historically described as information not included in the full text and images included with an article, which includes reference lists. “Core bibliographic data” have been further described as “data which is necessary to identify and / or discover a publication” and defined under the Open Bibliography Principles:

Therefore, these core bibliographic data should be considered open data.

RDF

Resource Description Framework (RDF) is a standard language for encoding data and metadata on the web. It is designed to indicate the relationship between online objects in a human and machine-readable way, and facilitate merging of data between different sources even if the underlying schemas of the sources are different. RDF forms the basis of the semantic web, and is a core component of achieving Tim Berners-Lee’s vision of Linked Data on the web [50]. RDF provides new opportunities for data and knowledge management in life sciences, chemistry, and publishing. All BioMed Central journal articles, for example, contain embedded RDF, which conveys harvestable information about content, such as authors, licensing information and the unique identifier for the article [51].

What aren’t data?

Although source code may be represented as data and is certainly machine readable there are a wide range of existing licensing systems and community norms that exist around software. Therefore we choose to regard software, compiled code, and source code as a separate category and not as data. Specific licenses and repositories have been developed for source code for software and Open Source Initiative compliant licenses [52] are recommended. Files pertaining to programming languages can be included as additional files with journal publications, either directly in formats such as SQL or indirectly in compressed or packaged file formats such as ZIP.

There are myriad file types which can be published as additional files but amongst the most common, in BioMed Central journals, are those usually pertaining to text and written works – PDF and DOC/DOCX and HTML (a full list of published additional data files is available on request from BioMed Central [53]). Caution is recommended in the interpretation of these objects as open data.

Setting date (CC) Zero

Creative Commons licenses (CC-BY, specifically) have provided an effective and penetrative solution for digital copyright in open access scientific works (papers). But as the nature of the published scientific paper (article) has evolved then so too should the copyright and licensing structure which authors apply to their content. Published articles are increasingly collections of different digital objects, perhaps including a few thousand words of text, half a dozen images and a similar number of CSV or XML (data) files.

The fact that CC-BY is a suboptimal license for data does not mean that the many thousands of published authors have done something wrong, as CC-BY was (and often still is) the best instrument available when copyright license agreements for open access journals were prepared (although, CC-BY version 4.0, currently in draft form, aims to tackle the issue of sui generis database rights [54], described earlier). A number of data repositories, including Dryad and FigShare, initially asked authors to make their deposited data available under a Creative Commons attribution license but have since changed their policy (https://twitter.com/#!/figshare/statuses/50241486796754944). However, licenses and waivers cannot be applied retroactively by a publisher without explicit consent of the copyright holder(s) – in the vast majority of cases at BioMed Central, the authors. A small number of datasets remain in Dryad which are not available under CC0, as explicit agreement from data depositors (rights holders) could not be obtained to change the terms of data release [29].

A change to BioMed Central’s standard license agreement to include a CC0 waiver for published data would remove ambiguities about the copyright and attribution requirement status of parts of published articles and associated data files, and enable instead the application of scientific, cultural norms that meet the needs of scientists better than an inflexible legal instrument [33, 55]. To implement open data in journal publications the new license agreement would need apply to all authors from a specific date, such that any author submitting to a journal/publisher agrees to dedicate the data elements of their article and additional files to the public domain. A proposal for how this could be reflected in published articles’ copyright license statement follows:

"© 2012 <Author> et al. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Data included in this article, its reference list(s) and its additional files, are distributed under the terms of the Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/; http://0-www.biomedcentral.com.brum.beds.ac.uk/about/access).”

This would be further indicated in article metadata, RDF and on the journal and publisher’s policy pages and author submission pages online. This addition to the statement aims to succinctly summarize that data in published articles generally originates from three sources – tabular data or text-minable factual data (e.g. numerical instances of a particular word or phrase, such as gene/protein names) in the main body of an article; additional (supplementary) files such as XML and CSV file extensions that include data; and the reference list (bibliographic data). From a legal perspective, we might need to be more comprehensive about our definitions of data (see ‘What do we mean by data?’, above) in the full legal code of a new license and in guidelines accompanying this proposed change. And practically, authors would need to agree to apply two different legal tools to content they submit for publication as part of the submission process – with links to both CC-BY and CC0, for example, in journal submission checklists.

The need for change in author behavior, as a result of this proposed modification, is minimal. CC0 would not apply to nor affect the availability of data not submitted for publication; we would instead be asking authors to apply different terms of use to a proportion of the content they already publish. By submitting a manuscript and all associated files to a BioMed Central journal authors already confirm that they agree to all terms of the BioMed Central Copyright and License Agreement [56], including that they are able to apply a CC-BY license. This proposed change to the license agreement should provide clarity to the licensing of specific components of published articles and does not represent a substantial change to the overall license agreement for authors’ published work.

Open data by default – opt out

Public domain dedication of data, while universally desirable, is not always universally possible. Authors, for the most part, prepare research articles in the context of employment by a third party and can be subject to licensing terms which supersede the standard terms of a publisher. This already happens for a small proportion of the content published by BioMed Central and undoubtedly by other publishers, such as some articles funded by the World Health Organization and US government. Therefore, any submitting author who is not able to agree to all terms of the BioMed Central Copyright and License Agreement should contact the journal editorial office at the earliest opportunity – ideally before, during or immediately after manuscript submission. The onus is on the authors and, if applicable, their employers to decide on the applicability of the publisher’s standard license agreement to their work and whether an exception is needed. Alternatives to the standard license agreement can be discussed. This process of checking suitability of the standard license agreement, and requesting alternatives where necessary, therefore already happens for all manuscripts submissions. The owners of the rights (if any) in the data must decide if their data can be made open. According to the Panton Principles FAQs “[i]n most cases, the people who make a decision to publish, and were intimately involved in the generation of the data, should be making this decision.” [42]

Publishing platform developments

Implementing a new license agreement will have technical as well as policy and procedural implications. While it is not possible to specify these changes in detail here, and is beyond the scope of this document, the following would be essential for implementation:

Furthermore, the following would be desirable to enhance the discoverability and usefulness of open data in journal articles:

Example #1 – analysis of a large clinical trial dataset

In April 2011 Sandercock et al., published in Trials ‘The International Stroke Trial Database’ [44], which “aimed to make individual patient data from the International Stroke Trial (IST), one of the largest randomised trials ever conducted in acute stroke, available for public use, to facilitate the planning of future trials and to permit additional secondary analyses.”

The “database”, including 19,000 anonymized individual patient data, is available with the journal article as a 4.3 Mb CSV file (http://www.trialsjournal.com/content/supplementary/1745-6215-12-101-s1.csv) under a CC-BY license. With the new license agreement as proposed in this article the CSV file would be available for reuse without a legal requirement for attribution engrained in CC-BY. Secondary uses for this dataset might include a novel secondary analysis by a different group of researchers, and analysis and integration of the dataset in the context of a systematic review and meta-analysis of randomized trials of heparin and/or aspirin in acute ischemic stroke. Both of these activities might conceivably result in further publications. Although there would be no legal requirement for attribution, for any secondary article about this dataset – or indeed any systematic review – to be scientifically valid it would need to cite its source(s) of data.

Example #2 – Application of magnetic resonance techniques to cross-species comparative studies

Magnetic Resonance Imaging (MRI) techniques are used to better understand the evolution of specific traits in animals and cross-species comparisons (for example in primates) are particularly important. But due to ethical, practical and funding limitations single studies typically are only able to consider one or two species. There is only one publicly available dataset that has brains from multiple primate species scanned according to a common protocol and these scans (of 11 species) were recorded (in vivo) well over a decade ago, and so do not meet the quality criteria that underpin more recent brain morphometric algorithms of the kind required for cross-species studies of brain structure. However, a review of this area of research found that “the major barrier to cross-species MR-based brain morphometry is not the lack of data nor analytical tools but barriers preventing to combine them” [57]. Open data in this field would undoubtedly drive new discoveries.

Example #3 – Research utilizing text and data from journal publications

The copyright status of data obtained though text-mining is debatable. The numerical instances of a particular gene or protein name in a full-text corpus of articles could be valuable for secondary research and, in the US at least, are likely to be considered (non-copyrightable) facts. Some scholars take the position that mining does not violate copyright law because it does not meet the statutory definition of copying which requires “fixing” the work in a permanent form [15]. Yet text mining is often restricted by commercial publishers. In the study of small angle scattering (a “technique based on the deflection of a beam of particles, or an electromagnetic or acoustic wave, away from the straight trajectory after it interacts with structures that are much larger than the wavelength of the radiation” according to Wikipedia [58]), a researcher might be interested to harvest the data used in other publications to test their analysis tools and provide better teaching aids. Generally, in this area of research, the data are only presented as a graph, the data analysis is not spelt out and there is no specific license attached if the data are available (Cameron Neylon, personal communication).

Example #4 – Open bibliography

Online scientific publishing has driven a diversification of measures of research and researcher impact, extending the focus from journal impact factors to article and individually-led metrics. Bibliographic information (rather than copyright attribution), which enables identification of scholarly work and tracks citations to scientists’ work, is central to earning of academic credit for concepts and ideas. Many services are now available which enable individual authors to calculate their citation index, known as the Hirsch or h-index. Examples include Scopus, Thomson ISI, Google Scholar and Microsoft Academic Search. However, much of the data underlying these metrics is not available openly, leading to multiple scores for the same individual or paper – depending on the tool or service used, which have different corpuses and different algorithms for calculating impact scores. A common, open bibliography, as has been established by some leading libraries would enable anyone to assess, utilize and build applications based on the data [59]. And furthermore, from a researcher’s perspective this approach is far more efficient, negating the need to maintain and report multiple sources of data from multiple impact-measuring tools. As outlined by Jones et al.[49] the motivations for and opportunities for open bibliographies are many. The negative implications of open bibliography for an author of a paper are negligible. Under the license agreement proposed in this article CC0 would apply to the article title, author names and information, unique identifying and publishers’ information, and reference list. Given a primary use of bibliographic information is to track scholarly citation activity, authors could reasonably expect these open data to increase the visibility and impact of their work.

Example #5 – Reproduction/validation of results for teaching and further research

In September 2010 Tommi Nyman and colleagues published an article in BMC Evolutionary Biology, “How common is ecological speciation in plant-feeding insects? A 'Higher' Nematinae perspective” [60]. The article included, in addition to the sequence data used to reconstruct the phylogenetic trees, the background data used in the phylogeny-based ecological analyses as additional file 1 – an Excel file. The data are well labeled and readily understandable by other scientists and fully document how they sampled their insects. This informative approach means, for example, readers would not need work through the references to discover the sampling used. These data have potential usage for reproduction and validation of the article’s findings, for teaching purposes, and conceivably uses involving the processing and integration of the data using computer software. Explicit dedication of these data to the public domain minimizes barriers to these scientifically important activities and maximizes the reuse potential of the data, as we could be more confident that all future uses of the data will not be impeded by licensing restrictions.