Data science training for official statistics: A new scientific paradigm of information and knowledge development in national statistical systems

Abstract

The ability to incorporate new and Big Data sources and to benefit from emerging technologies such as Web Technologies, Remote Data Collection methods, User Experience Platforms, and Trusted Smart Statistics will become increasingly important in producing and disseminating official statistics. The skills and competencies required to automate, analyse, and optimize such complex systems are often not part of the traditional skill set of most National Statistical Offices. The adoption of these technologies requires new knowledge, methodologies and the upgrading of the quality assurance framework, technology, security, privacy, and legal matters. However, there are methodological challenges and discussions among scholars about the diverse methodical confinement and the wide array of skills and competencies considered relevant for those working with big data at NSOs. This paper develops a Data Science Model for Official Statistics (DSMOS), graphically summarizing the role of data science in statistical business processes. The model combines data science, existing scientific paradigms, and trusted smart statistics, and develops around a restricted number of constructs. We considered a combination of statistical engineering, data engineering, data analysis, software engineering and soft skills such as statistical thinking, statistical literacy and specific knowledge of official statistics and dissemination of official statistics products as key requirements of data science in official statistics. We then analyse and discuss the educational requirements of the proposed model, clarifying their contribution, interactions, and current and future importance in official statistics. The DSMOS was validated through a quantitative method, using a survey addressed to experts working at the European statistical systems. The empirical results show that the core competencies considered relevant for the DSMOS include acquisition and processing capabilities related to Statistics, high-frequency data, spatial data, Big Data, and microdata/nano-data, in addition to problem-solving skills, Spatio-temporal modelling, machine learning, programming with R and SAS software, Data visualisation using novel technologies, Data and statistical literacy, Ethics in Official Statistics, New data methodologies, New data quality tools, standards and frameworks for official statistics. Some disadvantages and vulnerabilities are also addressed in the paper.