Measuring the Sustainable Development Goals (SDGs), and ultimately the success of the whole SDG initiative, will depend on the availability of authentic public sector data. By the time it reaches policy-makers, either as baseline data or as comparative data indicating progress, it will have been assembled through one or more methods from sources as varied as paper or digital records, management information systems, people and scientific instruments. After collection, it will be curated, cleaned, analysed, augmented and remixed. It will be subjected to formulas in spreadsheets, algorithms in apps and the intervention of people in various roles with various priorities and agendas. It will be distributed, published, cited and (it is hoped) preserved through numerous channels, platforms and systems.

At every stage, the authenticity of these data is potentially at risk. If public sector information is going to be available and authentic, there must be a regulatory environment and information culture that supports openness. From at least the 1960s, the international open government movement has been working towards laws, policies and standards that have supported the availability of public sector information, for instance through freedom of information laws. What it has not yet addressed is the need for technical and procedural controls to establish authenticity. This chapter argues that the principles and techniques developed over centuries in the field of recordkeeping for the purpose of assuring the authenticity of the records documenting decisions and actions also can be used to improve data quality, so that the information needed for implementing and monitoring the SDGs is not only available but authentic.

This chapter presents a case study of the Kenya Open Data Initiative (KODI). It examines the level of control presently in place for establishing and maintaining the authenticity of information released through KODI. While KODI is noteworthy as the first government open data portal in sub-Saharan Africa, it is studied here only as an example; problems identified with KODI data can be seen in open government datasets worldwide, to varying extents.¹

This chapter begins by defining authenticity in information before providing an overview of KODI (http://www.opendata.go.ke). To identify issues relating to the authenticity of information released via KODI, the chapter then examines a KODI dataset relating to land use. This analysis is discussed in relation to Kenya’s provisions for managing land information. The chapter goes on to describe the processes of preparing, publishing and using the land dataset. It maps the lifecycle of the data and identifies strengths and weaknesses in the controls for protecting the data’s authenticity. Questions about authenticity raise questions about the data’s contribution to implementing and monitoring the SDGs.

Data authenticity

In records and archives management literature, an authentic record is one ‘that is what it purports to be and is free from tampering or corruption’.² Authenticity depends on ‘integrity’, which is the ‘wholeness and soundness’ of a record, and on ‘identity’ – ‘the attributes of a record that uniquely characterize it and distinguish it from other records’.³ Over many centuries, records and archives professionals have developed principles and techniques for assuring information authenticity – from medieval chancery practices to technical standards for digital information management systems.

What these controls have in common is an emphasis on documenting the provenance and custodianship of records through metadata captured in auditable systems. For instance, the registry system used to control paperwork in the British empire required specific officers to capture specific information in registers and on file covers and flyleaves in such a way that traces of the registered record existed in multiple places and could not easily be erased or doctored. In this way, records became wrapped in metadata that described their management and their movement within and between government offices. The obligations on custodians and the system requirements were mutually reinforcing, so that custodians were encouraged to comply with and give effect to the system through its oversight mechanisms. These same principles can be seen in standards for digital records management systems, such as MoReq and ICA-Req,⁴ which require that metadata be captured to document all actions in relation to records.⁵

The Kenya Open Data Initiative

In July 2011, the KODI portal was launched by President Kibaki to provide public access to Kenyan public sector information.⁶ This was a landmark moment in the history of state secrecy and openness in Kenya. Nevertheless, KODI experienced a number of challenges in its first years of operation, as the World Wide Web Foundation’s Open Data Barometer (ODB) reports showed. The ODB ranks countries on three criteria:

The ODB methodology draws on government self-assessment, peer-reviewed expert survey responses, detailed dataset assessments and secondary data.⁸ The ODB Regional Report for Africa (3rd edition, 2016) examined 21 sub-Saharan African countries, including Kenya.⁹ Key findings were:

Within the ICT Authority, which is responsible for KODI, there is a good deal of enthusiasm for improving and expanding the open data programme.¹¹ Members of the KODI team participate in the Data Science Africa research network and its annual conferences.¹² The KODI team has a network of ‘fellows’ across the public sector who identify relevant datasets, and there is a Data Science Team that cleans data released by government agencies.¹³ Although the African Data Consensus had not yet been ratified by Kenya at the time of writing,¹⁴ the staff of the ICT Authority recognised the challenges it identified, which informed the priorities for KODI’s development.¹⁵

KODI staff are aware of problems with the data they receive for publication through the portal. They cite many of the same problems reported by civil society and the ODB, including the questionable accuracy and completeness of the data, the risk of introducing errors during the cleaning and curation of data, problems of timeliness and infrequent updates, and ongoing resistance to data release within the public sector because of the long-standing culture of secrecy in the Kenyan government. Several of the staff noted in an interview in September 2016 that ‘the quantity of data received does not seem to reflect the enactment of the FOI law’; though the law had only just received Presidential Assent, there was a sense that it should have resulted in an increase in proactive information release.¹⁶ At the time of our interview, KODI team members had not considered the issue of data authenticity and the significance of contextual information, but they noted that they ‘hardly ever get data with metadata’.¹⁷

Nevertheless, KODI staff have put in place a number of quality control measures. They take a snapshot of the portal with every new data upload. They keep copies of original datasets received from ministries, departments and agencies, so that sources can be checked in the event that the curated data are queried. In addition, they have created a ‘data release calendar’ that they use to schedule and monitor updates, and they have created templates for datasets. All of this encourages more complete and consistent data.¹⁸

Land data

Examining a sample dataset from KODI should help to illustrate what controls and processes for authenticity are in place. In view of the significance of land in Kenyan political and economic life, a dataset relating to land use is examined here in relation to Kenya’s approach to land information management.

Land information management

Kenya, which was among the ‘first [independent African] countries to experience comprehensive land reform’,¹⁹ has experienced problems in managing information about land and related resources. For instance, in October 2015, the Thomson Reuters Foundation reported, in relation to the Kenya Groundwater Mapping Programme:

Since the colonial period, Kenya’s land management system has functioned largely through the creation, transmission and exchange of paper records. Under the Registration of Titles Act (1982), a Central Registry was established in Nairobi and a Coastal Registry in Mombasa, for managing paper land registration records. Under the Registered Land Act (1989), a registry was established in every land registration district. The 2012 Registration Act aimed to rationalise and devolve the land registration process but, at the time of writing, regulations were still being developed. Recordkeeping in Kenya has faced numerous challenges, including a lack of cohesive policies, lack of compliance with procedures, ad hoc systems and lack of adequate staffing and other resources.²¹

In 2010 and 2011, during research into the readiness of Kenyan government recordkeeping for e-government and freedom of information, Justus Wamukoya and I found that what was then the Ministry of Lands (now the Ministry of Lands and Physical Planning) had experienced a period of recordkeeping reform.²² At the time of our study, its registries were well-functioning and monitored, and sanctions were imposed for infringements:

At that time, a digitisation project was underway, motivated by a sense that computerisation and digitisation would, among other benefits, reduce delays in work processes. These, according to one member of staff, accounted for the great majority of the written complaints that the ministry received.²⁴ The digitisation project was led by the Land Management Systems Technical Working Group in the Lands Reform Unit, which oversaw the ministry’s target under the government-wide Vision 2030 strategy in relation to improving land title acquisition.²⁵

Wamukoya and I found that while records staff were confident that the organisation of paper records was adequate for land title process improvements, members of the working group had identified significant gaps in the paper records being digitised. Digital surrogates were comprehensively and regularly backed up, but no digital preservation measures had been developed and no consideration had been given to the need to eventually transfer digital records to the Kenya National Archives and Documentation Service (KNADS). Moreover, there seemed to be no planning for moving from creating and scanning paper records to managing born-digital records.²⁶ When I again visited the Ministry of Lands and Physical Planning in September 2016, paper records were still being digitised, including maps. The staff could conduct online searches for Nairobi, and the ministry was setting up a central database of land titles as part of the Kenya National Spatial Data Infrastructure, which was expected to expand the ministry’s capacity to conduct online searches.²⁷

Today, recordkeeping practices in the ministry continue to be guided by the standard public service manual on records management, with no separate guidance on managing land title records.²⁸ The Records Management Procedures Manual for the Public Service provides procedures for registering and managing records, with provisions for mail management, filing, indexing, cross-referencing, classification, file tracking, ‘bring-up’, storage, survey, appraisal and disposal (transfer to archives or destruction).²⁹ It also discusses disaster management, capacity building and the institutional framework for recordkeeping, including ministerial responsibility for compliance. The ‘Security of Records’ chapter states that access to classified records should be on a ‘need to know’ basis, reflecting the long-standing civil service bias towards secrecy. However, it also notes that ‘Access to public records shall be provided within the existing legislative and regulatory framework’, which, since 2016, includes freedom of information legislation.³⁰ The manual warns staff to ‘guard against a natural tendency to over classify documents’.³¹

There is also a chapter on digital records management that sets out responsibilities and provides guidance on naming conventions, media handling and storage. It empowers the National Archives (KNADS) to set procedures for digital records management and to authorise the destruction of digital records. It states that all public institutions are required to install the Integrated Records Management System (IRMS) developed and introduced by the Personnel Office in the Ministry of Public Service. Although the IRMS has not been audited against international standards such as ICA-Req or MoReq, it covers core records system functionalities relating to registration, file tracking, ‘bring up’ and reporting.³² It does not, however, address the need for a system to preserve born-digital records or digital surrogates. Kenya’s second Open Government Partnership National Action Plan included a commitment to establish ‘a central digital repository to provide lasting access to government records and data and all information of public interest’ by June 2018.³³ However, this has not yet happened.

Examining the land dataset

Although authentic information is crucially important for the SDG initiative, difficulties in accessing or understanding data can make it virtually inaccessible to potential users, including government and civil society actors involved in SDG work. Kenyan open data specialists, Leonida Mutuku and Christine Mahihu, have reported that ‘data quality’ is a key issue for open data in Kenya. They have highlighted low relevance of data to citizens, the irregularity of data updates or dataset releases and the questionable utility of the data (for instance, incomplete data and data that are poorly structured or formatted) as key issues.³⁴ However, their study did not examine authenticity when considering data quality.

In the field of recordkeeping, authenticity requires that metadata, documented provenance and custodianship and auditable systems work in concert. For the purpose of examining how these three prerequisites are applied in producing and publishing data released through KODI, I studied the dataset ‘Proportion of Parcels Using Fertiliser 2006’. Not only is the dataset relevant to the contentious issue of land use, but the fact that it was accompanied by some contextual information suggested that it might be possible to determine if the prerequisites for authenticity were addressed in its production and publication. The dataset was uploaded by a user called ‘Knoema’ in 2015 and the original data source was identified in its metadata as the Kenya National Bureau of Statistics.

I first looked at the dataset in October 2016. Returning to KODI on 11 January 2017, the site rendered a login interface (see Figure 8.1).

Creating an account and signing in brought me to an error message (Figure 8.2) indicating that permission was now required to access the data.

On querying the staff of the ICT Authority, I was directed to use a mirror site while the portal was migrated between service providers. A search of the interim platform for the title of the dataset on 16 January 2017 produced no results. A search for ‘parcels’ produced a short list of results that included a dataset called ‘Proportion of Parcels Using Fertiliser County Estimates 2005/6’. The metadata for the dataset indicated that the dataset was uploaded by ‘kodipublisher’ rather than ‘Knoema’, with a publication date of 20 December 2016, which was probably the date of migration to the interim platform. There was nothing to indicate that the dataset was first published in 2015.

The dataset, viewed as a CSV file, comprised seven columns:

Columns D to F provide figures ranging from 0 to 0.94. Column G provides figures ranging from 0 to 428581.8. There is no data or metadata within the CSV file to help interpret these figures, such as the meaning of ‘using_1’, ‘using_2’, etc., and no formulas that provide a key to the relationship between the figures in columns D through F and the figures in column G. Columns D through F for Wajir County each have a value of 0, as does column G. Columns D through F for Mombasa County each have a value of 0, but column G has a value of 2425.2. This means that the content, context and structure of the dataset are insufficient for the average user to interpret the data.

Instead, with regard to this particular dataset, KODI does the work of interpretation for its users by visualising the data as a map. The portal includes three tabs: ‘Overview’, ‘Data’ and ‘Visualization’:³⁵

The ‘Proportion of Parcels Using Fertiliser County Estimates 2005/6’ dataset lacks essential assurances of authenticity. In terms of metadata, the content of the CSV file itself lacks sufficient metadata to enable users to interpret the data. The fact that human interpretation of the CSV is not possible does not necessarily undermine an assumption of authenticity. However, since visualisation is necessary to understand the data, questions need to be asked about the mechanisms of visualisation. What formulas and algorithms are used to render the dataset as a map? The formulas and algorithms that render the data interpretable to users are a significant component of the system for managing the information and should be capable of being audited to support authenticity. However, there is no technical information available through KODI to help users understand the process that produces the map. This effectively constitutes a gap in the chain of custody.

Of more significance for demonstrating authenticity, KODI offers little metadata about the file itself. Looked at in isolation, this dataset appears to have ‘identity’ (the attribute of a record that distinguishes it from other records).³⁶ This ‘identity’ is supplied by its metadata, in particular its unique title, author and dates. However, when compared with the dataset first viewed in October 2016, the discrepancy in publishers (Knoema and kodipublisher) and dates (uploaded 2015 and last modified 20 December 2016, without reference to an upload date) raises questions about the identity of the dataset. Is it the same dataset? If identity is called into question through inconsistent metadata, one of the two fundamental elements of authenticity (identity, with integrity) is absent.

In addition, the provenance of the dataset is obscured by the lack of metadata documenting the custodianship of the data from the point of collection. Working backwards, there is metadata about the publisher of the data (though some ambiguity about the identity of the publisher when the two versions of the dataset are considered together) and about the source of the dataset, which is given as the Kenya National Bureau of Statistics, but it is not possible to see the sources that the Bureau used.

In this dataset, there are two types of data:

To be assured of the integrity of the dataset it must be possible to know the sources of the data. In this case, assurance would require users to seek this information from the Bureau of Statistics. The recordkeeping system of the Ministry of Lands and Physical Planning functions well and conforms with the system set out in the manual, which, being government-wide, is also likely to guide the recordkeeping of the Ministry of Agriculture, Livestock and Fisheries.³⁷ The recordkeeping system documents custodianship throughout the lifecycle of the ministries’ records, but KODI users do not have a way to be aware of this. Therefore, there is a break in the documentation of the chain of custody between the creation or capture of the data by the ministries and the aggregation of the data by the Bureau of Statistics. Moreover, the user does not know what processes and controls the Bureau followed to prepare the data and assign and document responsibility for the data.

The data pass through a series of systems, and these need to be auditable if authenticity is to be assured. From the point that it is created, and throughout its use within the ministry, land parcel information is managed through an auditable recordkeeping system. The data may then pass to the Ministry of Agriculture, Livestock and Fisheries, where they serve as a parameter for the ministry’s estimates of fertiliser use. There is no metadata that would help KODI users understand this. Even at the point of aggregation, when the data passes into the Bureau of Statistics’ systems, this is not transparent to KODI users.

In preparation for publication, the data then pass into the custody of KODI and its systems. Unlike the previous transitions, this is documented in publicly available metadata. As outlined above, KODI staff follow defined measures for protecting the integrity of data, and, while these measures are not brought together in a formal system that allows each action and custodian to be audited, they would, theoretically, enable published datasets to be compared with datasets as received by KODI. In this way, there is a basic accountability mechanism in place for KODI’s data custodianship. However, there is no information about these controls on the KODI portal. Rather than assurances of integrity, KODI users have ambiguous information about the sources and treatment of the data.

Although ‘Proportion of Parcels Using Fertiliser County Estimates 2005/6’ may be authentic in the sense that it ‘is what it purports to be’, the user does not have the information needed to determine its authenticity. Partial metadata, opaque provenance, undocumented custody (particularly during aggregation) and the lack of information about the systems for its management, including its visualisation, introduce doubts about the data’s identity and integrity.

Conclusion

The development of open data in Kenya has not yet been linked to records management. Connecting the two would help facilitate trust in opened datasets and support their survival through time. At present, there are problems with the data being released through KODI, and the ability to rely on their authenticity remains limited.

This study has charted the lifecycle of a specific dataset and found that, in general, strengthening the recordkeeping practices of public sector bodies lays a foundation for assuring authenticity in public sector data. This would be enhanced by the publication of recordkeeping audit reports documenting processes from the beginning of the data lifecycle. Later in the lifecycle, when data are aggregated and then published on KODI, authenticity is called into question by partial metadata about data’s provenance and management. Leaving aside problems that may be unique to KODI’s transition between platforms, there are unanswered questions about the management of the data throughout their lifecycle.

Authenticity is at the greatest risk at the point that data are aggregated. The data aggregation practices of the Bureau of Statistics are not documented for the public. In this space between creation and publication, provenance is obscured, and custody cannot be audited. After publication, most users will need to rely on KODI’s visualisation of the dataset in order to interpret the data, making KODI an essential component of the data management system. Accountability requires that information management systems must be transparent, and custodianship must be public. At present, KODI’s mechanisms for visualisation are not published. The chain of custody of the dataset appears to be broken in several places.

These issues all undermine assurances of the authenticity of the data. Not only is there a negative impact on government openness and community and commercial reuse, but there are repercussions for the ability to pursue the SDGs. Implementing and monitoring the goals depend on access to authentic data.

KODI is taking steps to build controls into its processes. It takes snapshots of the portal with every new upload, retains original datasets received from government and standardises templates for datasets. All of this contributes to building an audit trail of the accuracy and completeness of data and metadata. If these steps can be brought together, documented and published, KODI will have taken important steps towards assuring authenticity.

The lack of technical controls for information authenticity is not unique to Kenya – it is widespread, and arises from the disconnect between the communities of practice involved in open data on one hand and records management on the other.³⁸ To improve data quality generally and data authenticity in particular, records management principles and know-how need to be incorporated into open government data curation. Records management has supported previous efforts to establish government openness (particularly freedom of information), by organising and making information available for release. The major contribution of records management to open data is, however, not in providing records for data mining but in offering techniques for improving data quality by making datasets more like records.