Open data for official statistics: History, principles, and implementation

1.1History of the open data movement

The open data movement has its roots in the open-source, open science, and government transparency and accountability movements. The first use of the term open data comes from On the Full and Open Exchange of Scientific Data [47], which called for making environmental data available to the public so that scientists could study the global environment that transcends borders. The Berlin Declaration on Open Access to Knowledge in the Sciences and Humanities called for “a free, irrevocable, worldwide, perpetual right of access to, and a license to copy, use, distribute, transmit and display the work publicly and to make and distribute derivative works, in any digital medium for any responsible purpose” [39].

The Fundamental Principles of Official Statistics (FPOS) set out the professional and scientific standards for NSOs and provides the foundation for the open data for official statistics movement. The United Nations Statistical Commission in 1994 adopted the FPOS, which had been adopted by the European Conference of Statisticians in 1992. The first of the ten principles anchors the need for statistical information in democratic governance and recognizes the citizens’ entitlement to public information:

The commentary on the first principle anticipates the subsequent discussion of open data as requiring dissemination of data in usable formats and with suitable metadata:

Open data was also discussed in the open-source and computer science communities, which saw benefits to sharing code and data for reuse. In December of 2007, two prominent thinkers on computer science and the internet, Lawrence Lessig and Tim O’Reilly, organized a meeting in Sebastopol, California, to make the case for open government data. The preamble to the meeting’s list of eight principles articulated a fundamental claim about open data: “Information becomes more valuable as it is shared, less valuable as it is hoarded. Open data promotes increased civil discourse, improved public welfare, and a more efficient use of public resources… By embracing the eight principles, governments of the world can become more effective, transparent, and relevant to our lives” [83]. The need for data to better understand and solve global problems has become a common theme in support of open data. Trevor Manuel, then Minister of Finance of South Africa, noting the need for good data to guide development policies, quoted Peter Drucker: “What can’t be measured can’t be managed” [38].

In 2009 the United States adopted open data as official government policy [49], creating a new repository for United States government data of all types. A year later the United Kingdom followed suit, announcing a “One-stop shop for Government data” [100]. In 2013 the United Kingdom Cabinet Office published a policy paper proposing an “open data charter” for the G8 group of nations [97] that was adopted in June of that year.

With national governments working to implement open data, there was renewed pressure on international organizations to do the same. One of the first movers was the World Bank, which announced open access to their statistical databases in April of 2010 in advance of a new access to information policy taking effect that July [110, 111]. The World Bank’s adoption of open data was spurred by a hero of the open data movement, Hans Rosling, whose advocacy and work to make data applicable, interesting, and useful to policy makers and citizens alike, provided a huge service to the community [33]. Opening the World Bank’s database, particularly the World Development Indicators, which contained data from many of the United Nations’ specialized agencies, would help to normalize open data in the international realm.

The G8 Open Data Charter provided the model for the seven principles of the International Open Data Charter [57]. Open data was included as a central part of the Cape Town Action Plan for the Sustainable Development Goals in 2017, which further cemented open data’s role in international governance [22]. To enshrine the concepts of open data for official statistics, a working group was established to add open data concepts to the Fundamental Principles of Official Statistics at the 50th meeting of the United Nations Statistical commission [101]. That work is ongoing.

1.2Benefits of open data

The case for open data has often been stated in political or ethical terms: citizens are entitled to the products of their government and to use that information to hold governments accountable. The benefits are realized through increased government efficiency and responsiveness to citizens’ needs. But there may also be substantial economic benefits to making data open.

The economic case for open data is rooted in the theory of public goods [86, 90]. Public goods have two attributes: they are non-rivalrous, meaning they are not reduced by use; and they are non-excludable, so they are available to everyone. The classic examples of a public good are national defense or a lighthouse whose benefits are shared by all without diminishing anyone’s use. Other examples are air pollution – a public bad – and over-the-air television broadcasts. In all these cases, it would be inefficient or impossible to charge individuals for the use of the good.

Information produced by governments is clearly non-rivalrous: one person’s knowledge of census statistics or use of meteorological information does not reduce the amount of information available to others. However, unlike national defense, access to information can be made exclusive by law or technical means, thus limiting access, and potentially allowing the producer to discriminate among users. At this point, the theory of public goods confronts the general principle of economic efficiency: goods – including intangible goods like statistical information – should be priced at their marginal – not average or total – cost of production. Information once it has been created costs very little to reproduce and disseminate. Indeed, the hallmark of the data revolution is that digital technology and global networks have reduced the marginal cost of delivering data to near zero [23, 26]. For NSOs and other public agencies that must bear the cost of producing statistics, this presents a conundrum: the efficient price for open data will not cover their cost. Budgets for the statistical agency or for the publications department may include revenues from the sale of paper publications and physical media such as CD-ROMs, often at significant mark-ups. With the advent of internet dissemination, the need for these modes of publication have largely disappeared and the shortfall in revenue must be made up in other ways. Governments could still charge for access to their information, but the welfare gains to the public can more than justify a loss of revenues.

Economists have quantified the value generated from opening datasets. One of the early examples is Pollock [81] who estimated the value of providing free access to public sector information that had been previously sold to be GBP 1.6–6 billion, 4–15 times the forgone sales revenue in the United Kingdom. One of the biggest examples of the economic impact of open data is the United States opening GPS data for civilian use, which is estimated to have created around USD 100 billion in economic value [121]. Research on the opening of Landsat satellite data and weather data from the National Oceanic and Atmospheric Administration (NOAA) in the United States points to similar economic and societal benefits [120]. Annual savings from the open Landsat data for non-governmental organizations, the federal government, and the private sector are estimated at between USD 350 and 436 million per year [34]. Beyond the direct financial benefits, the opening of data can produce a wave of innovation and new products and services as is evidenced by the many apps and products that have been developed and built from the opening of GPS data.

There are other benefits, besides financial, in international development to opening data. Open data has been critical to saving lives in natural disasters and emergency situations. Humanitarian Open Street Map has been used for the earthquakes in Haiti and Nepal, and the Typhoon in the Philippines to provide first responders with the open data they needed to find and rescue survivors [40]. Open data, accessed via computers or through text messages on cell phones, can provide farmers with the most up-to-date information on prices and data on weather patterns that could affect their harvest [25]. Open data is a central part of the strategy for achieving the Sustainable Development Goals (SDGs) and is a critical part of the effective allocation of resources to solve international development challenges.

The case for open data is usually based on the economic and social benefits and contribution to good governance described above, but there are also benefits for NSOs that take the lead on open data. Opening official statistics can increase their use and raise the profile of the NSO. Opening data can help the NSO to start a virtuous cycle of use in which, through the provision of high-quality data, the NSO enables decision makers to increase their impact, and they, in turn, provide more funding for the NSO [98]. In more simplistic terms, open data gives NSOs the opportunity to prove their inherent use to the public and to people in positions of power who may control funding.

1.3Defining open data

In 2005 Rufus Pollock and the Open Knowledge Foundation [80] proposed a comprehensive definition of open data, built on the foundations of the Open-source Definition [72]. Pollock’s proposal distinguished requirements for open access, which he called “social openness,” from open licensing. Social openness meant: “The work shall be available as a whole and at no more than a reasonable reproduction cost, preferably downloading via the Internet without charge. The work must also be available in a convenient and modifiable form.” The subsequent 10 principles all specify the terms of an open license. Originally titled Open Definition version 1.0, it would undergo multiple revisions before arriving at version 2.1 in 2015 [82], but the core distinction between technical openness and legal (or licensed) openness is characteristic of all subsequent definitions of open data [2].

In the introduction to the Open Definition 2.1, the Open Knowledge Foundation defines open as: “Open means anyone can freely access, use, modify, and share for any purpose (subject, at most, to requirements that preserve provenance and openness).” And defines open data as: “Open data and content can be freely used, modified, and shared by anyone for any purpose” [72]. In its current form the Open Definition 2.1. has four elements or principles that define an open work:

The second part of the Open Definition presents nine criteria for an open data license. Many data publishers have adopted a Creative Commons license that satisfies these criteria. However, licensing terms must also be consistent with local and national laws governing copyright and intellectual property and the contractual authority of the national statistical office, so many variations are possible. Open data licenses and the Open Definition criteria are discussed in more detail below.

A second important document defining open data is the International Open Data Charter’s Principles of Open Data [56]. The Open Data Charter is a set of principles and best practices for how and why government data, including official statistics, should be opened. The six principles and their accompanying commentary are described as “aspirational norms” for the dissemination and use of open data:

The elaboration of these principles includes useful advice for implementation of open data. The first principle has caused the most debate. “Open by default” is generally understood to mean that data (and other government information) should be open except where there are specific reasons for limiting access. These reasons include the protection of security, privacy, confidentiality, and intellectual property. Thus “open by default” requires a “negative list” of data that are not open; all other data may be assumed to be open. Put simply, “open by default” means that producers, users, and policymakers should presume that data will be published as open, unless there is a justifiable reason for it not to be.

The International Open Data Charter has been adopted by 74 national and local governments and endorsed by 55 organizations and non-state actors. Taken together the Open Definition and Open Data Charter provide a working definition of open data that has been well adapted to the needs of many official statistical agencies.

The open science community has embraced a similar set of guidelines for the sharing of scientific research materials: the Findable, Accessible, Interoperable, Reusable (FAIR) principles. The FAIR principles for data were formally outlined in 2016 as a part of a movement to improve the infrastructure supporting the reuse of scholarly data [107]. Mons, Barend et al. [42] make the point, equally applicable to the Open Data Charter, that these are guiding principles, not standards and their implementation may be context specific. They argue that data should not be “Open by default” but “Accessible under well-defined conditions,” as they take into account some of the reasons noted above for limiting public access to certain data.

2.1Planning for open data

Planning for open data can take place through various mechanisms. Developing a National Strategy for the Development of Statistics (NSDS) is how some low- and lower-middle-income countries have planned for including open data in their official statistics [76]. PARIS21 recommends incorporating open data into each phase of the development of an NSDS [77]. As of May 2019, nearly sixty percent of all International Development Association (IDA) borrower countries were implementing their NSDS, although not all included open data as part of their strategy [78].

The World Bank offers the Open Data Readiness Assessment (ODRA) as a diagnostic and planning tool to help statistical offices and their senior management prepare for the adoption of an open data program [116]. Statistical agencies that have already taken steps to implement open data may also find the Open Data Institute’s Open Data Maturity Model a useful guide for maintaining improvements to their systems and organization [59].

An ODRA can be conducted by internal staff but more often an outside expert works with a client team. The process described by the ODRA User’s Guide is intended to be carried out over a period of nine weeks and involves eight dimensions [113]:

All eight components are important. They complement one another, and weakness in one can compromise others. When Vietnam completed an ODRA in 2018, the assessors found that while demand for open data was present, the legal framework did not provide a supportive environment for open data implementation [115]. In the years since, Vietnam has put in place new digital data management, sharing, and open data policies [35] and recently launched its open data portal [41].

The cost of implementing open data depends on existing capacity, both human and technical, and the scale of the open data program. The Open Data Institute [58] provides advice on planning and budgeting an open data initiative. Whether the plan is to develop a separate portal for official statistics or to implement a whole of government open data program, implementing open data requires staff training, software and hardware acquisition, and communication strategies, all of which require funding, again underscoring the importance of assessing the political and financial support a government is able to offer for open data implementation.

2.2Implementing open data

Having assessed and planned for open data within the NSS, implementing agencies must put in place suitable policies to “aid both data consumers and data producers by clearly outlining the standards, processes and requirements for offering and acquiring public information” [117]. They can guide very high-level processes, such as the EU Implementation of the G8 Open Data Charter [12], or describe detailed activities for implementing open data as in the case of Ireland’s Open Data Strategy 2017–2022 [19].

Open data initiatives will need a champion to lead the initial foray into open data assessments and represent the interests of open data in budgetary and political discussions. The World Bank recommends including the eventual implementer in the ODRA process, which in turn will require a pre-ODRA scoping process to ensure any resulting recommendations can be implemented by the members of the ODRA team [114].

Technical partnerships between the open data initiative and the responsible IT infrastructure team, either in-house or hired from outside, are crucial to open data implementation. Adoption of international standards can help countries struggling with the complexity of making their data open and interoperable. The Generic Statistical Business Process Model (GSBPM), while not explicitly focused on open data, nevertheless can give open data initiatives the structure they need to organize their program within the NSO and across line ministries. It is already used by 39 NSOs. Open data and further guidelines on interoperability can be added to the GSBPM, as described by the Friends of the Chair Working Group on Open Data [109]. The World Bank also makes a series of recommendations for implementing open data standards at a technical level, such as improving technical documentation and ensuring that public APIs and endpoints are interoperable [112].

There are many choices that must be made to implement an open data program: about the data to be included, the format of their publication, the user interface and tools provided, and the terms of use. There is no single standard for open data and particularly in the case of microdata, there may be reasons not to make data fully open. The following section describes some of the practical choices that satisfy the core principles of open data.

2.3Maintaining open data

To sustain an open data program governments and statistical agencies should foster an open data culture that ensures open data remains a priority through changing administrations [122]. One way to conceptualize the adoption of an open data culture and its benefits is through the data value chain, which describes the progression of statistics from raw data to the publication of reliable statistics and their uptake to their ultimate role in providing evidence for policies that have an impact on peoples’ lives [64]. In this framework, adopting an open data culture can help move NSOs farther along the data value chain to contribute to both uptake and impact of their data [109].

Self-assessment is an important practice for maintaining an open data program. The Open Data Maturity Model allows an organization to assess how well they publish and consume open data, identifying actions for improvement [59]. It was developed by the Open Data Institute and the United Kingdom’s Department for Environment, Food and Rural Affairs for a public sector audience, but it can be applied to a wide range of organizations publishing or consuming open data. Organizations can use the model to assess operational and strategic activities around open data and compare themselves against others to highlight strengths and weaknesses. Through this, they can identify potential areas for improvement, adopt best practices, and improve processes.

Organizations conduct a self-assessment using the Open Data Pathway, an online app that facilitates the mapping of open data practices [60]. The model is based on five themes, each representing a broad area of activity:

Organizations can identify their own level of maturity on a scale of 1 to 5 for each of the themes, which charts an organization’s move from “ad-hoc uncontrolled processes to those that are repeatable, standardized and well-managed; [a move from] a reactive to a proactive approach within a particular area of study; [and a move from] isolated expertise [on open data…] through to wider organizational support” [59]. The maturity model has been used, among others, by the Queensland Department of Transport and Main Roads in Australia to meet its improvement targets for open data.

The insights derived from self-assessment tools and other review processes can be used to work towards an open data culture alongside monitoring the legal and policy environment of a statistical office. Policies shape behavior, so policies need to evolve with the open data program. For example, Ghana recently replaced its 35-year-old mandate with one that now gives the NSO greater control over data gathering, setting standards across the national statistical system, and coordinating and cooperating with international partners to further increase capacity [16].

Leadership and internal capacity are vital for launching an open data initiative and are equally important for maintaining open data. New positions such as a Chief Data Officer [93] can enshrine the importance of open data in the management structure. Though data literacy itself may be high in many of the agencies charged with producing data, open data literacy may not. Including open data in internal trainings and in onboarding materials can weave open data into the daily operations of data producing and publishing entities.

Engaging with civil society and other partners outside government is key to creating open data feedback loops from champions outside the government, as well as from within. The World Bank in its review of open data practices suggests that “[E]xternal stakeholders, unfettered by the constraints of official practice, can sometimes be powerful allies in lobbying for action, and it is the use of data by users that can best demonstrate the value of an Open Data initiative” [114]. There are a multitude of tools available to increase the accessibility of open data, such as those listed above, and they should be implemented in the course of adopting open data practices in order to increase awareness of open data. Public engagements such as hackathons, briefings, press releases, blogs, and digital feedback mechanisms are also useful tools for establishing contacts with outside partners and receiving recommendations to improve open data practices. The United States Government, for example, publishes guidance for hosting open data engagement events that can be adapted to other countries’ situations [95].

Monitoring and measuring data use in conjunction with a data dissemination program can create a positive feedback loop that builds support for open data and the work of the NSO. With metrics on data use derived from website tracking programs such as Google Analytics, NSOs can make the case that their open data portals are more than just fancy websites: they are tools that are used by the public they serve. These metrics help NSOs to better understand the public’s data demands. Used to evaluate data dissemination efforts, they can point to potential gaps in dissemination strategies. Open Data Watch’s Measuring Data Use report [63] demonstrates some of the tools available and provides guidance for NSOs on monitoring data use on their sites. Monitoring and measuring data use can, in turn, improve data use and dissemination and strengthen a country’s open data culture by providing more value and use cases for open data.

3.1Copyright and open data

Copyright is a legal right to reproduce original works, license them for reuse, or otherwise assign a right over them to another person or entity. In many jurisdictions, the idea that data are copyrightable is highly contentious. Data are often interpreted in law as being mere facts, not original works. The proportion of girls in a school grade for instance is a fact, not an original work, but assembling and documenting a database on girls’ education might be considered a creative work worthy of copyright; even in this instance, it would be the structure of the database itself that might receive sui generis protection and not the underlying data themselves. However, some countries consider all work by governments to be in the public domain and not subject to copyright under any circumstances.

If data are not copyrightable, then no license is necessary, but an explicit statement that the data are license-free should be included. Determining whether data are copyrightable can be complicated. Copyright laws vary greatly by country. Although there are international treaties between some countries that agree to uphold each other’s copyright laws, there is no international copyright law [6]. The oldest such treaty, the Berne Convention for the Protection of Literary and Artistic works, outlines standards for determining whether a work is eligible for copyright, but they are not legally enforceable [108].

In the United States and many European countries, data are considered “facts” not covered by copyright law because they do not meet the threshold of originality, a criterion of the Berne Convention. In these countries, there are no legal barriers to releasing the data into the public domain license-free, especially if the data are owned by the government. However, in other countries not all data may be considered facts, or the data may be considered facts but “may be eligible for other appropriate or applicable forms of intellectual property protection, for example by virtue of sui generis database rights, depending on its jurisdiction of collection or creation” [10]. When possible, data producers should waive any of the following rights that may apply.

Moral rights, unlike other types of protection, are designed to protect the reputation of the creator rather than the economic value of a product. The specific aspects of moral rights vary by country, but generally include the right to attribution, how the work is displayed, and controlling how the work is altered or adapted [84]. In many cases, moral rights are not applicable to data. In the United Kingdom for instance, moral rights are only applicable to “literary, dramatic, musical and artistic works and film, as well as some performances” [27]. However, in other countries moral rights may apply to statistical products, even those created as government works. For example, article 5 of Oman’s Open Government License states “You are not permitted to take any act of compromising the data that would harm the honor and reputation of the data provider,” a provision added to protect the creator’s moral rights [52]. Taiwan’s Open Government Data License, Version 1.0 also protects moral rights, stating in Section 3.1, “By utilizing the Open Data provided under the License, User… shall make the reasonable efforts with respect to moral right protection of the third parties involved” [45]. Many open data licenses address moral rights by waiving them, but in most countries where moral rights exist, they cannot be waived. The most problematic aspect of moral rights as they pertain to statistical data is the protection of a data creator’s reputation. If data are reused in a way that counters the producer’s findings or a user criticizes the quality of the data, would this be considered damaging to the producer’s reputation? If moral rights cannot be waived and they are applied to statistical data according to a country’s laws, specific language should be included in the data license as to what would be considered an infringement of these rights.

In some countries, even when the data are not copyrightable, the databases that hold the data are, particularly when the design of the database or arrangement of data is an intellectual creation [53]. Database rights also protect the creator’s rights in cases when its creation involves originality or complex design. In most cases, these rights do not apply to the content of the database but may, in certain cases, apply to items such as metadata to protect a creator’s intellectual property rights [32]. If the creator’s rights do apply to the content of a statistical database or its metadata, these rights should be waived in an open license.

In contrast to database rights, sui generis database rights may apply to databases and their content that qualify for copyright protection not due to originality in creation, but because the investment in obtaining and presenting the data was substantial [14]. This right is most notably recognized in the European Union, United Kingdom, and Russia. In 2017 a public consultation was held by the European Commission [13] to review both database rights and sui generis rights, but no changes to either resulted. These rights can be waived by adopting an open license.

3.2Criteria for an open data license

While recognizing that the legal debate continues about whether data can or even need to be licensed, an open license or explicit statement that data are license-free is an important mark of open data. A data license is an opportunity for data producers to encourage public use of their data by specifically addressing how people can use data, the form of attribution requested, and what types of use, if any, are prohibited. Even where data are “open by default,” licensing affirms policies pertaining to authorized use, alleviating user concerns about legal consequences of unapproved use. For example, a journalist accessing freely published official data might fear prosecution for using the data to criticize a government program. This fear has increased in recent years with a rise in “fake news” laws – often used to stifle reporting that government officials disagree with. Between 2016 and 2019, 65 journalists were imprisoned for publishing “fake news” worldwide [4]. A clear and unambiguous license can therefore confer a degree of legal certainty and protection for users that in turn gives them confidence in their use of the data.

The Open Definition lists nine criteria that a license must have to be considered fully open along with certain conditions that can be added to a license without affecting its openness [73]. An open license must irrevocably permit the following:

The Open Definition also lists seven conditions or limitations that may be included in an open license. The most common is requiring attribution to the original source. The license may also include an integrity condition, requiring that modified versions carry a different name or version number. Many organizations specify that the user may make no claim of endorsement by the original source. While these provisions are discretionary, they may increase confidence in the data and prevent confusion about their provenance. For example, an attribution clause in a license means that data that might be reused several times over will always be traceable back to its original source. That may help the ultimate user determine whether the data are trustworthy.

Open licenses may include provisions, such as share-alike conditions, that require derivative works to be licensed under the same terms as the original. This condition does not alter the license’s openness, though it has sparked controversy in the open data community. It has been argued that share-alike licensing imposes a restriction on data use that conflict with the core principle of open data [106]. While share-alike licensing is intended to ensure the further dissemination of open data, it may have the unintended effect of inhibiting creative use of the data. For example, problems may arise when combining open data from several sources: if just one dataset has a share-alike condition, it can make it very difficult to pull together the various open data streams and re-publish them under a separate open data license.

A licensor can create a bespoke license that meets the criteria of the Open Definition. In practice, many data publishers choose a standard open license. Table 1 lists the most common open data licenses used for statistical data. Creative Commons is a non-governmental organization that has developed a set of licenses to facilitate the sharing of knowledge [7]. Open Data Commons is a project of the Open Knowledge Foundation [74]. Open Data Commons licenses are used less frequently for statistical data. In general, it is advised to not license data that do not fall within the domain of a country’s copyright law, unless there is some uncertainty that may warrant applying a Creative Commons CC0 or Public Domain Dedication and License (PDDL) [91]. Otherwise, a statement that the data are license-free or in the public domain is sufficient.

The CC BY (attribution) license is frequently used to ensure that the original source of the data is identified in subsequent products. The Share-Alike license imposes an additional restriction by requiring that the same licensing terms be used for any derivative products. It is still considered an open license. However, the CC BY NC license, which is restricted to non-commercial use of data, fails to meet the standard for a fully open license. Creative Commons licenses have changed over time. In version 4.0, for example, moral rights are waived to the limited extent necessary to exercise the licensed rights. An extensive discussion of the terms of the licenses and the specific rights granted by them can be found on the Creative Commons website [8].

3.3Restrictions on the misuse of statistics

As noted above, Creative Commons licenses waive the moral rights of the licensor to the extent possible under local law. But clauses prohibiting the misuse of statistics, or similar conditions, are sometimes included in bespoke data licenses, often at the expense of openness. They are problematic because the term “misuse” is generally undefined in these licenses, creating an unknown and unknowable liability for the data user.

Principle 4 of the Fundamental Principles of Official Statistics states that “The statistical agencies are entitled to comment on erroneous interpretation and misuse of statistics,” which has been interpreted by some statistical offices as a mandate to restrict data use by aggressively pursuing broadly defined “misuse” of statistics The Implementation Guidelines from the UN Statistical Commission recommend principle 4 be achieved through clear language in national and legal frameworks [103]. The specific examples of misuse discussed by the FPOS Implementation Guidelines include “overgeneralization,” “misreporting or misunderstanding of estimated error,” “false causality,” and “data manipulation.” When data are open, such misuse is more likely to be discovered, and as suggested by principle 4, the best response to misuse is to draw attention to it and criticize the perpetrators.

The Open Data Inventory (ODIN) reported in 2017 that 31 countries had data licenses (or terms of use that included data use provisions) that prohibited misuse, but used vague language, such as “misleading use is prohibited,” without specific examples to describe misuse [63]. Many government agencies, in absence of a data license, use the terms of use of their websites to address data use provisions that would typically be included in a data license. Early versions of some countries’ licenses, such as version 1 of United Kingdom’s Open Government License, included similarly vague language about misuse, but this language has since been removed from later versions [44]. In later versions of their open government licenses, most countries have generally removed problematic language, but some that used the UK’s license as a model have retained the original language prohibiting misuse.

To address instances of misuse, as defined by the FPOS Implementation Guide, such as claims of false causality and overgeneralization, many countries include a non-endorsement clause and disclaimer. For example, the Open License Agreement for the Central Statistics Office of the Commonwealth of Dominica states, “This license does not grant you the right to use the data in a way that suggests an official status or endorsement of you or your use of the data” and that they “will not be held responsible for damages resulting from its use or interpretation” [5]. These types of clauses do not necessarily prevent misuse but release the data producer from any responsibility for the misuse. Other countries have gone further, giving recommendations to data users that may help prevent misuse, such as Senegal’s National Agency for Statistics and Demography, which suggests that users not only provide attribution, but take into account the indicator’s relevance before using, do not modify the specific figures, and provide explanatory information about the indicator in order to prevent misunderstanding [43].

4.1Statistical disclosure control

Despite international agreement on the importance of opening microdata, there is no straightforward definition of open microdata analogous to the Open Definition that applies to aggregate indicators or macrodata and to other non-personal government records. The Open Definition provides the basis for assessments of open macrodata (such as the Open Data Inventory and the Global Data Barometer), but there are no similar assessments of the openness of microdata. The lack of standards for open microdata is due in part to uncertainty inherent in statistical disclosure control (SDC). The International Household Survey Network explains the problem this way:

To maintain the privacy of survey respondents, the right level of SDC needs to be found, but microdata from different surveys, geographical regions, and countries may have separate SDC requirements based on the sensitivity of the information, the presence of outliers in the data that may be de-anonymized, and a host of other factors. At the bare minimum, personally identifiable information such as names, email, addresses, and phone numbers should be removed from the data before making them public. A review of SDC practices and procedures is beyond the scope of this paper, but guidance on best practices for SDC is provided by the International Household Survey Network’s SDC Practice Guide [30].

Because disclosure may cause harm to individuals or establishments, it may be necessary to exclude certain uses of the data or users from access to portions of the data. In many instances, the data have been collected with an explicit promise of confidentiality and that commitment may be legally enforceable. In some instances, data sharing, or providing access to data to a limited number of people, might be more suitable than open release of a dataset. Data sharing agreements are often used when data are particularly sensitive because of personal privacy, intellectual property, or other reasons. Microdata sharing differs from open data because the data are not made open and public to all but to a select group of people or organizations.

The World Bank’s Microdata Catalog [118] defines five categories of access for surveys and censuses. (See Table 2). The same convention has been adopted by many national data archives. Open access is the only category that is indisputably open. Other categories imply restrictions that would disqualify them under the Open Definition [73] or by the standards applied by the Open Data Inventory to macrodata [65]. Another mechanism for protecting data confidentiality is the creation of data enclaves: physically controlled environments in which researchers may submit data requests but are not able to access the data directly. The National Opinion Research Center (NORC) at the University of Chicago, for example, offers a secure environment where public and private organizations can store data and provide controlled access to authorized users [48].

4.2Choosing a microdata license

Licenses for microdata can vary for a number of reasons, but the main determining factor is usually privacy concerns and statistical disclosure control. Different types of statistical data have different levels of disclosure risk and the level of risk determines the degree of openness the license will offer. Licenses, like data, are not open or closed, but rather exist on a spectrum. Data producers should strike a balance by pursing the maximum level of openness appropriate to counter disclosure risk. Figure 1 shows the disclosure risks of various statistical outputs, along with the suggested access levels (or openness) of their applicable data licenses.

Figure 1.

Disclosure risk and access levels of different statistical outputs.

The movement to open microdata is at a similar place that the open data movement was years ago. Important work is being done to develop standards, document the availability of the microdata sets, identify data gaps, and advocate for better funding. Much of this work is expected to be led by the Inter-Secretariat Working Group on Household Surveys and its member organizations [109]. As these data play an important role in many different policy and government decisions, the process of overcoming these challenges and making these data open and available, while protecting the privacy of the survey respondents, will be rewarded with increased use of this data leading to large societal benefits.

5.1Global Data Barometer

The Global Data Barometer (GDB), a research project underway, aims to track data governance, use, and availability across over 100 countries using country-level expert surveys combined with secondary data. About half of the thematic data categories covered in the GDB pertain to official statistics. The Global Data Barometer is a successor to the Open Data Barometer [17, 119]. When completed it will provide an index of data governance, capabilities for data use, data availability, and impact. Through this research, the Global Data Barometer seeks to assess government, private sector, and civil society capacity to harness data for the public good. And it will highlight uses and abuses of data sharing, open data, and algorithmic decision-making.

Discussions for the creation of the Global Data Barometer began at the Open Government Partnership global summit in May 2019, and the study analysis and report are scheduled to be launched in late 2021. Prior to 2020, the Global Data Barometer was known as the Open Data Barometer [119], but its new name reflects a focus on more than just open data concepts to include data collaboratives, use of algorithms, and the evolving landscape of data and development.

The index will be based on a country-level expert survey, combined with secondary data to produce an overall index, sub-index, and individual scores. The primary peer-reviewed, expert survey will be conducted every two years. It will track data governance policies and practices in over 100 countries and across 8-15 thematic modules that address key data gaps. Each thematic module will be co-created with partners. Themes currently under consideration include: Gender and inclusion, Land, Telecoms, Energy, Transit, Climate, Crime and Justice, Agriculture, Construction and Infrastructure, Procurement, Fiscal Transparency, Education, Urban Development, Local Government, Extractives, and Geodata.

5.2OECD OURdata Index

The OECD OURdata Index measures the availability, accessibility, and reuse of government data through an official survey sent to public officials in 32 OECD countries and one accession country (Colombia) [51]. The public results of this index do not allow for an in-depth analysis of data gaps or recommendations, but the report offers many examples from OECD countries and best practices. The OURdata Index seeks to increase data flows, challenge data monopolies, and inform automated decision-making models based on emerging technologies such as AI. The pilot version first launched in 2015 with the latest set of results having been shared in 2019. It has been developed by the OECD Open and Innovative Government Division, within the Directorate for Public Governance.

The results of the OECD Open Government Data Survey sent to public sector officials are structured around three pillars: data availability, data accessibility, and government support for data reuse. For data availability, the index measures the scope of datasets available on open data portals, the extent to which governments promote open government data at the national level, and user involvement in data policy processes. The data accessibility pillar covers requirements for unrestricted access, such as terms of use and metadata, and the role of the ecosystem and portal in ensuring data quality. And under the government support for data reuse, the index measures how governments promote the reuse of government data within government and beyond.

5.3European Open Data Maturity Assessment

The European Open Data Maturity Assessment covers the maturity of open data policies, portals, impacts, and quality through a questionnaire sent to national open data representatives in 32 EU28+ countries. The European Data Portal, a project of the European Commission, conducts this assessment as an annual exercise to provide the EU28+ countries with an overview of their maturity and progress [15]. The exercise assesses maturity across four dimensions, including policy, portal, impact, and quality. It categorizes countries into four groups: trend-setters, fast-trackers, followers, and beginners. These assessments have been conducted since the European Data Portal was launched in 2015 and document year-on-year progress since then.

Data for this research is collected through a questionnaire sent to the national open data representatives that work in collaboration with the European Commission and the Public Sector Information Expert Group. The questionnaire is structured to collect detailed metrics for each of the four dimensions. These dimensions and metrics were developed during a major methodology revision in 2018. The government survey completed by officials receives further validation and analysis from the European Data Portal team in cooperation with government officials. The 2019 version of the Open Data Maturity Assessment included 32 countries.

5.4Open Data Inventory

The Open Data Inventory (ODIN) independently rates the coverage and openness of official statistics available from or linked to national statistical office websites in 180 countries. These assessments help identify gaps and help improve access to open data produced by national statistical offices (NSOs) [66]. The ODIN scores provide an indicator of how complete and open the data provided by the NSO are. The summary scores for social, economic, and environmental statistics and summary scores for coverage and openness provide a picture of the national statistical systems’ strengths and weaknesses. Open Data Watch published the first round of ODIN assessments in 2015. The most recent version was launched for 2018/19 and included 178 countries. The next ODIN round is expected to be released in early 2021.

The statistics on NSO websites are assessed by trained researchers on ten elements of coverage and openness. Coverage scores are based on the availability of key indicators and appropriate disaggregations over time and for geographic subdivisions. The openness assessment is based on the first four principles of the Open Definition. It tests whether data available in 22 topical categories can be downloaded in machine-readable and non-proprietary formats; are accompanied by metadata; can be selected directly by users or through APIs or for bulk download; and whether the terms of use satisfy requirements of an open license.

The ODIN assessment process assumes NSOs to be at the center of national statistical system, taking charge of the dissemination of official statistics as open data. Even for decentralized national statistical systems, ODIN looks for NSO websites to provide links to other ministries’ data sites. Feedback from government officials is solicited to engage NSOs in improvements to the coverage and openness of official statistics.