Adjusting Survey Estimates With Multi-Accuracy Post-Processing

MCML Authors

Unai Fischer Abaigar

→ Group Christoph Kern
 Social Data Science and AI
→ Co-Group Frauke Kreuter

Christoph Kern

Prof. Dr.

Associate

Social Data Science and AI

Frauke Kreuter

Prof. Dr.

Principal Investigator

Social Data Science and AI

Abstract

With the rise of non-probability samples and new data sources, survey researchers face growing challenges related to selection bias. One emerging line of work adapts algorithmic tools from machine learning to improve robustness in such settings. This talk introduces multi-accuracy boosting (Kim et al., 2019), a post-processing method that reduces subgroup-level prediction error. Originally developed in the context of fairness, it has since been explored for use in survey adjustment tasks (Kim & Kern et al., 2022). I offer an accessible overview of the method and share reflections on its potential, and open questions for future research.

inproceedings

---

ITACOSM 2025

Italian Conference on Survey Methodology. Bologna, Italy, Jul 01-04, 2025. Invited talk.

---

Authors

U. Fischer Abaigar • C. Kern • F. Kreuter

---

Links

PDF

---

Research Area

 C4 | Computational Social Sciences