Big Administrative Data for Survey Methodology in the Era of Machine Learning |
|
| Coordinator 1 | Dr Asaph Young Chun (Seoul National University ) |
This session will feature papers that theorize and implement "total administrative records error" and “total linked data error” methods and provide best practices of using administrative data tied to the survey life cycle (Chun, Larsen, Durrant, and Reiter, 2021 Wiley book) in the wake of machine learning advancements. The session invites methodological papers that discuss fundamental challenges and recent innovations involved in the collection and analysis of massive administrative records, including deep learning, and integration with surveys, censuses, or auxiliary data. We also encourage submission of papers discussing quality of the linked data, and sustainable access to the linked data. The session welcomes papers demonstrating how big administrative data would answer real-world policy questions that traditional surveys alone could not address. For example, papers in this session include, but are not limited to the following topics:
1)Innovative use of administrative data in household or establishnent surveys to improve the survey frame, reduce nonresponse follow-up, and assess coverage error.
2) Quality evaluations of administrative data and quality metrics for the data linked to administrative data, including machine learning applications
3) Recent advancements in processing and linking administrative data with survey data (one-to-one) and with multiple sources of data (one-to-many), including machine learning-driven approaches
4) Recent methods of disclosure limitation and confidentiality protection in linked data, including linkages with geospatial information
5) Machine learning or Bayesian approaches to using administrative data in surveys, censuses, small area estimation, and nonresponse control.
6) Applications demonstrating how massive administrative data would answer real-world policy questions in public health, economics, science, and education that traditional surveys alone could not address.