Utilizing Alternative Data Sources for Official Statistics

Abstract

To ensure that knowledge is not exclusive to a few, but available for the many, public agencies publish statistics as a public good - also called official statistics. Facing increasing competition about the prerogative of ’facts’, statistical agencies are expected to provide more disaggregated, frequent, granular and reliable statistics in a timely manner. Already today, enormous amounts of data are generated around us frequently as more and more parts of our lives move into the digital realm. Much of this information on human action is linked to place and time. Exploiting those spatio-temporal patterns may help to better understand underlying dependencies and trends in the socio-economic fabric of our society. Consequently, can official statistics leverage this new ”data deluge” and live up to the promises of better, more relevant statistics? In attempts to do so, the use of satellite imagery, mobile phone data, social network data and other new data sources for demographic or socio-economic mapping has drawn much attention in recent years. The thesis contributes to this field of research on applied statistics in two ways: Part I showcases how new data sources can be utilized to improve traditional statistical data collection techniques, notably censuses and surveys, especially in settings with weak national statistical systems. In contrast to the application-driven first part of the thesis, Part II focuses on overcoming methodological challenges in augmenting official statistics with new data sources, especially when combining these disparate data sources in the first place. The presented use cases and methodological contributions are just a few among many steps necessary to pave the way for alternative data sources to be utilized in the business processes of official statistics. But as enterprise data became an important pillar for measuring modern-day economic activity a century ago, as administrative data overhauled long-time census practices only recently, eventually (privately-held) non-statistical data from mobile and social networks, from satellite imagery or other sensors may find their way into mainstream official statistics one day to produce statistics in a more disaggregated, frequent, granular, reliable and timely manner after all.