Given the growing interest of government bodies in the use of big data, the importance of conducting an appropriate audit increases. According to sector studies^[1], big data audits include three interrelated levels: technological, managerial, and strategic. Each of them is oriented towards the assessment of the following parameters:

• Effectiveness of the innovative technologies introduction to the work of a government body. The key assessment criteria are the quality of data collection, the variety of processing methods, and the development and availability of infrastructure.
• Effectiveness of organization's structure in the context of the big data use. Particular attention is paid to assessing the quality of professional training, interaction between departments on the use of big data, and the work of technological centers for processing big data.
• Correspondence between the expected and obtained results from the introduction of big data analysis technologies. The assessment criteria are the total number of identified inaccuracies in the operation of data mining algorithms (performance), the social effect of increased transparency of information (legitimacy), and the accuracy of risk management (forecasting).

^[1] Appelbaum D, Kogan A, Vasarhelyi M A. Big Data and Analytics in the Modern Audit Engagement: Research Needs. Auditing A Journal of Practice & Theory, 2017, 36(4):1-27; Al-Sai Z, Abdullah R, Husin H. Critical Success Factors for Big Data: A Systematic Literature Review. IEEE Access, 2020:1-1

Despite the undeniable advantages of big data analytics technologies, there are certain risks and limitations to their use. One of the main problems here is ensuring the necessary quality of data, which is usually understood as the sum of the following features:

• Complete (there are no gaps in the data preventing their analysis or use);
• Comprehensive (all elements describing the target object/event/situation are included in the dataset);
• Timely (regular updates);
• Understandable (including metadata and machine-readable format);
• Accurate (the available data is correct and clearly reflects the current state of the target object, process and/or phenomenon);
• Consistent (the available datasets do not contradict each other; the use of different datasets does not affect the accuracy of conclusions);
• Unique (items are not repeated within the same dataset);
• Discoverable (raw data and data catalogues are obtained from open government information systems and/or within the state department);
• Machine-readable (data are presented in formats that can be used by machine algorithms at once and/or with minimal human processing);
• Inter-operable (standards, semantics, and common data identifiers are available for processing on the most common technological platforms);
• Protected (all requirements for the protection of personal and/or other confidential data are met).

Effective monitoring and control of data quality requires government bodies to develop a clear methodology for data management, as well as to assess its reliability. To achieve the required data quality, in addition to data-cleaning, the OECD experts recommend regular and random data audits. The objectives of such audits are to assess data for compliance with generally accepted standards and set goals, ethical standards (including ensuring privacy), and legislation. Such measures are intended to ensure that actual data are not fabricated to meet any expectations.

Guidance on Conducting Audit Activities with Data Analytics (published by the INTOSAI Working Group on Big Data, WGBD) includes the following methods for data verification during an audit:

• Checking the completeness of data by comparing the quantity and volume of partially structured and unstructured data within the arrays provided by the auditee;
• Checking integrity constraints of relational models, including primary key constraints, referential constraints, user-defined integrity constraints, etc.;
• Checking the compliance of the data with the initial information (financial statements);
• Checking the total amount of data, statistical indicators of the main variables, as well as the authenticity of values in data arrays by calculation and aggregation. For example, the range of the main variables is checked to be corresponding to the range presented in the operating report by means of calculating the maximum and minimum values of the main variables and aggregation;
• Checking the correctness of reporting (availability of intermittent and repeated values, date range, etc.).

A specific set of big data issues is associated with the provision of open access to governments’ data for citizens. The policy to improve the legal regulation of the OECD countries information policy is aimed at reducing administrative barriers and increasing the availability of government data. In 29 of the 32 OECD Member States, central/federal governments require data to be available free of charge, in machine-readable formats, and with appropriate metadata. In 28 of the 32 OECD Member States, data is required to be available with an open license.

In addition, many OECD members have committed themselves to publicly promoting the principle of transparent public administration. Some focal areas include increasing the number of programs to raise awareness of how beneficial the open government data and their reuse are, and increasing the digital literacy of civil servants.

Within the European Union (EU), the EU Directive 2019/1024 dated June 20, 2019, plays a central role in supporting government efforts to improve data transparency by promoting innovation and proper governance. Article 16 explicitly calls on States to “promote the creation of data which are open by design and by default."

Despite the difficulties in ensuring open data and source code in the public sector, these measures contribute to transparency, accountability, and public control over the decisions and results of public policy. In these circumstances, the following seems appropriate:

• To promote the establishment of quality data ecosystems by providing people with unrestricted access to data sources and helping to ensure the equitable distribution of information in society;
• To provide public access to disaggregated data in accordance with the requirements of confidentiality, security, and respect for property rights;
  anonymous and granular^[1] open data can be used to identify relevant social and economic problems and make fact-based decisions; in turn, such demonstrative results will increase confidence in data analysis tools both within the public sector and in society;
• To make the source code open to public control and auditing, especially when personal data or datasets are processed through digital government projects.

^[1]Level of detail based on available data. The detail includes data one level below the previous one (in particular, hours, minutes, seconds, etc.). The maximum level of detail implies the maximum level of dataset detail.