Some Thoughts on Survey Statistics and Its Future
This presentation examines the evolution of survey statistics within the field of official statistics, in an era characterized by administrative data, online sources, and machine learning methods. Statistical data are never simply “given” but are constructed through sampling, measurement, and adjustment processes, making extrapolation to well-defined populations the main goal of official statistics. While new and massive data sources offer unprecedented opportunities, they often lack the structure required for valid population inference, particularly regarding statistical units, coverage, and extrapolation principle. Classical survey sampling principles, such as design-based and model-based inference, calibration, weighting, and imputation, remain essential tools for restoring a coherent extrapolation framework. The talk also revisits historical debates on sampling and emphasizes that many modern methods, including machine learning and forecasting techniques, extend rather than replace traditional statistical ideas. Finally, it highlights key challenges for official statistics, including data quality, integration of heterogeneous sources, declining survey participation, and the need for continued methodological innovation and institutional capacity building.
.png&w=3840&q=75)
