Replication Data for: Overcoming Medical Overuse with AI Assistance: An Experimental Investigation

This deposit contains the complete replication package for "<b>Overcoming Medical Overuse with AI Assistance: An Experimental Investigation</b>", which examines how artificial intelligence can reduce medical overtreatment across different healthcare payment incentive structures. The research employs two randomized controlled experiments: a lab-in-the-field study with 196 practicing physicians at a hospital in Wuhan, China, and a within-subject study with 120 medical students. Both experiments use a novel medical prescription task drawn from the Chinese National Medical Licensing Examination question bank, testing physician decision-making under three incentive schemes (flat, progressive, and regressive payment structures) with and without AI assistance from ChatGPT 4.0.
<br><br>The key findings demonstrate that AI assistance increases treatment selection of recommended treatments by 25.7-28.4 percentage points, reduces overtreatment by 10.9-25.7 percentage points (15.2-80.3% relative reduction), and improves treatment accuracy by 9.8-13.3 percentage points (14.6-19.9% relative improvement). Importantly, these effects vary significantly across payment incentive structures, with AI being most effective when financial incentives are absent or aligned with optimal patient care. The deposit includes two datasets (data08.csv for the main experiment with 7,840 observations and data03.csv for the supplementary experiment with 2,400 observations), 22 Stata do-files with complete replication code, and detailed documentation enabling full reproduction of all tables and figures presented in the published paper.