Abstract

From Citizen Data to Policy Evidence: Rethinking Statistical Quality in the Age of AI
The integration of official statistics, big data and citizen data creates opportunities to address local information needs and inform public policy. However, greater data availability and increasingly powerful artificial intelligence (AI) tools do not, by themselves, ensure valid statistical inference. How should quality be assessed when evidence emerges from diverse sources, participatory processes and algorithmic transformations?
This paper proposes a process-based framework that treats quality as a property of the entire data-generating system and its intended use. Building on the Total Survey Error perspective, the United Nations National Quality Assurance Framework and the Copenhagen Framework on Citizen Data, it connects governance quality, production processes and technical output quality. Particular attention is given to the distinction between statistical representation—supporting inference about a defined population—and the capacity to make visible experiences, communities and phenomena that established statistical systems may overlook. Meaningful participation can strengthen relevance and interpretation, but does not automatically resolve selection bias or justify population inference.
The framework links quality principles to observable conditions and evidence, providing a basis for transparent assessment and proportionate validation. For AI-assisted data integration and indicator production, it emphasizes source traceability, reproducibility, uncertainty assessment and human accountability. Small area estimation provides a methodological setting for discussing how heterogeneous sources may support local policy indicators, while requiring explicit assumptions and careful evaluation of coverage, measurement and model error.
The paper argues that trustworthy policy evidence requires methodological rigour alongside inclusive and accountable data production. Its contribution is an integrated approach to assessing when citizen data and AI-assisted statistical processes can support defensible conclusions, and how their limitations should be communicated to policymakers and communities.