"Assessing the Quality of Personality Measures Generated by Large Language Models" ready for submission · PNAS
with Shuaizhang Feng, Yujie Han, and James J. Heckman
LLM-based Big Five ratings from student letters show internal consistency, construct validity,
and predictive validity; in a multi-informant framework, LLM ratings have predictive power
comparable to conventional human raters.
"The Effects of China–US Tensions on Science: Evidence from the arXiv Dataset" work in progress
with Shiyu Bo · first author
Continuous DID around Aug 2018: high overseas-Chinese-diaspora topics experienced a sharper
citation decline with little drop in publication counts; mainland-author placebo shows no
comparable effect.