Advanced methods (e.g. model-based estimation, multivariate methods, forecasting/nowcasting and microsimulation) are already being used by statistical institutes - sometimes in regular production, other times in supporting processes or in analytical projects.
At this daWos session, we will discuss the use of advanced statistical methods in data analytics for official statistics.
Questions that could be addressed include
- Which are the advanced statistical methods currently being used for the purpose of data analytics in the ESS?
- What is the feasibility and reliability of data analytics models and methods to support the generation of official statistical products?
- Through the use of advanced data analytics, it is possible to form heterogeneous information networks from multiple interconnected (big or not so big) data sources, in which information redundancy can be explored to compensate for missing data, e.g. validating trustworthy relationships and/or uncovering hidden relationships and models otherwise inaccessible. What are the statistical and computational techniques requested?
- Model-based inference (in contrast to the, sometimes model-assisted, design-based paradigm) has been debated for year. Recently, algorithm-based inference has emerged. What are the advantages, drawbacks and areas of applicability of the various approaches?
- What are the risks of strictly algorithm-based decisions, e.g., shifting the decision process from humans to machines? Should data analytics really be adopted in all steps of the policy-driven data-informed decision-making cycle - and what would the role of statistical institues be?
With new methods, new sources of uncertainty appear. Should a framework for quality measurement and quality metrics (e.g., in terms of accuracy, uncertainty, representativeness, etc…) be commonly defined for data analytics?