ESRA 2027 sessions by theme

Back to Overview of Sessions

Session Details

The increasing availability of AI-based methods raises fundamental methodological questions for survey research: to what extent can these methods improve the quality, efficiency, and validity of official statistics relative to established approaches, and under what conditions can their use be considered reliable, transparent, and methodologically sound? This session brings together experimental evidence and methodological experiences examining AI across three complementary stages of the survey process.
First, we focus on survey data quality, with particular attention to missing and incomplete information. Contributions will investigate whether Machine and Deep Learning methods can improve imputation and estimation compared with established statistical approaches. Beyond predictive performance, experiments will assess whether gains in prediction translate into improvements in statistical quality and inference, considering potential implications for bias, uncertainty, and the robustness of estimates.
Second, the session examines AI-enabled approaches to data collection. Conversational AI, AI-powered survey agents, and AI interviewers offer new possibilities for interacting with respondents and adapting the data collection process to individual responses. Contributions will investigate their effects on respondent engagement and measurement, while critically examining methodological challenges related to response behaviour, measurement error, comparability, and the reliability and consistency of AI-mediated interactions.
Third, we consider the role of AI in disseminating and interpreting survey data. Conversational interfaces and natural-language systems can enable users to interact directly with official statistics, formulate queries, and obtain explanations tailored to their needs. This raises methodological questions concerning the accuracy, transparency, interpretability, reproducibility, and responsible use of AI-generated statistical information.
The session aims to move beyond the general promise of AI and provide evidence-based insights into whether, when, and under what conditions AI can genuinely add methodological value to official statistics. Particular emphasis will be placed on empirical evaluation, benchmarking against established methods, methodological validity, limitations, and best practices.