# Design-unbiased statistical learning in survey sampling

**Authors:** Luis Sanguiao Sande, Li-Chun Zhang  
**arXiv:** [2003.11423](https://arxiv.org/abs/2003.11423)  
**Date:** 2020-03-25  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2003.11423v1/

## Key points

- Design-consistent model-assisted estimation is the standard practice in survey sampling
- Lack of a comprehensive theoretical framework integrating modern machine-learning techniques
- Proposed approach aims to develop a statistical learning theory for design-unbiased estimation using linear and non-linear prediction models
- Rich auxiliary information can significantly improve efficiency compared to traditional linear model-assisted methods
- Methodology ensures valid estimation for the target population and robustness against mis-specifications of assisting models at the individual level
- Sande and Zhang's work represents a significant advancement in survey sampling methodology, showcasing potential for more powerful assisting models through integration of cutting-edge machine-learning techniques

## Summary

In the field of survey sampling, design-consistent model-assisted estimation has become the standard practice. However, a comprehensive theoretical framework that integrates modern machine-learning techniques to enhance assisting models is currently lacking. The proposed approach aims to develop a statistical learning theory that enables design-unbiased estimation using both linear and non-linear prediction models. By leveraging insights from Statistical Science and Machine Learning, the authors demonstrate how rich auxiliary information can significantly improve efficiency compared to traditional linear model-assisted methods. Importantly, their methodology ensures valid estimation for the target population while also offering robustness against potential mis-specifications of the assisting model at the individual level. Sande and Zhang's work represents a significant advancement in survey sampling methodology, showcasing the potential for more powerful assisting models through the integration of cutting-edge machine-learning techniques. Their research not only contributes to enhancing the accuracy and efficiency of estimation processes but also lays the foundation for further exploration at the intersection of statistical science and machine learning within survey sampling practices.