MINDS-CM2LA Seminar Series | Joseph Kang (Research Mathematical Statistician, U.S. Census Bureau) - On machine learning models for incomplete survey data
Date |
2023-12-07 ~ 2023-12-07 |
Time |
17:00:00 ~ 18:00 |
Speaker |
Joseph Kang |
Affiliation |
Research Mathematical Statistician, U.S. Census Bureau |
Place |
Math Bldg 404 & Online streaming (Zoom) |
Streaming link |
ID : 688 896 1076 /
PW : 54321 |
Topic |
On machine learning models for incomplete survey data |
Contents |
Machine learning (ML) methods have been developed to improve the accuracy of predictions for unobserved data. Unobserved data are prevalent in many social surveys conducted by US statistical agencies and pose a significant challenge to achieving unbiased results. From the perspective of survey nonresponse, ML methods can be effectively employed to reduce the bias of survey outcomes. Despite the advancement of ML, its popularity has yet to gain widespread acceptance among statistical researchers who are unfamiliar with the challenges of interpretability in ML methods. This talk will introduce a popular ML method used in survey science and provide guidance on how to avoid relying blindly on this method and instead calibrate it with other related statistical methods. The application of ML-based methods will be illustrated via a well-known simulation study design by Kang and Schafer (2007). |