WiP Seminar: Roujing Wu

Abstract

This study employs large language models (LLMs) to evaluate the difficulty of math exam questions in China's National College Entrance Examination (Gaokao) across provinces and years. By standardizing the assessment of exam difficulty, this method facilitates meaningful comparisons across regions and cohorts. Leveraging this newly constructed measure, the research investigates the impact of exam difficulty on critical educational and socioeconomic outcomes. In particular, the variation in exam difficulty across provinces and years generates differences in students' performance at universities and their subsequent labor market performance.