Research

Working Papers

AI Policy in Education: A Randomized Control Experiment

(with Jinghao Jia, Xu Zhang; slides available upon request)

Abstract We investigate the optimal design of institutional rules for integrating Generative Artificial Intelligence (AI) into middle school education. Using a Randomized Controlled Trial (RCT) across two urban middle schools in China, we causally estimate the effects of two policy components: AI Guidance (providing instructional information) and AI Conditional Use (a policy-imposed restriction on the AI rate in writing). While we find insignificant average treatment effects on academic performance, the interventions yield significant heterogeneous effects by gender. Specifically, the Conditional Use restriction generated a positive academic spillover: female students experienced substantial improvement in math scores, with the largest benefits observed in the combined Guidance and Conditional Use group. We propose that this differential impact is driven by a compliance mechanism. Descriptive evidence shows that girls, who had higher baseline AI engagement, demonstrated greater adherence to the guidelines during the experiment, successfully steering their engagement toward more productive AI tool use. This suggests that the combination of guidance and use restriction is a powerful, gender-specific intervention that leverages existing student self-regulation to realize positive returns from AI.

Closing the Treatment Gap in Mental Healthcare: Evidence from China

(with Yuzhi Hao, Xu Zhang; slides available upon request)

Abstract We evaluate a large-scale, nationwide policy in China that aimed to improve population mental well-being by simultaneously expanding the supply of mental healthcare resources and reducing public stigma. Using a difference-in-differences design, we find that the policy significantly improved mental health outcomes on average; however, the benefits were concentrated among individuals with better mental health. We then provide descriptive evidence of an increase in mental healthcare adoption following this policy. To investigate the channels driving this enhanced adoption, we conducted a discrete choice experiment (DCE) focusing on the costs of care: monetary costs (price), time costs (commuting time), and psychological costs (stigma). The DCE reveals crucial heterogeneity: individuals with better mental health are more responsive to reductions in time costs but less sensitive to monetary costs. These findings suggest the policy’s average success was driven by the reduced time costs for these with better mental health conditions. We conclude that effective mental health interventions require a broad approach to improve general access, coupled with targeted efforts to remove financial barriers and provide privacy-sensitive support for those with severe conditions.

Nudge for Sanitation: Experimental Evidence from China (Under review)

(with Xu Zhang)

Abstract Information provision has been suggested as a cost-effective approach to engaging the target poor in anti-poverty campaigns. However, empirical evidence shows that providing information about available public services or entitled rights alone is often insufficient to motivate behavior change. In this study, we conducted a randomized controlled trial experiment to compare the behavioral impacts of different informational nudges on villagers’ participation decisions in China’s Rural Toilet Revolution (RTR), which aims to boost rural development through improved living conditions. We designed six RTR advocacy videos as informational nudges based on local residents’ information demand, highlighting the benefits of RTR participation, the harms of poor sanitation, and techniques for toilet upgrades, either with or without a former RTR participant calling for engagement. Our findings suggest that any relevant policy information can induce a higher self-reported willingness to participate in RTR, while highlighting the benefits or harms is more effective in motivating long-term enrollment decisions. The results reveal the importance of engagement benefits for the target audience and highlight information barrier removal as a useful nudge to promote anti-poverty campaigns like RTR.

Rewiring Opportunity: How Improved Internet Infrastructure Reduces Intra-City Income Inequality in China (Under review)

(with Xu Zhang, Yuzhi Hao, Aoqing Lyu, Masaru Yarime)

Abstract This study investigates the impact of enhanced internet infrastructure on intra-city income inequality in China. Employing a staggered difference-in-differences methodology, our analysis reveals that the Broadband China Strategic Program from 2014 to 2016 resulted in a notable reduction in the income Gini coefficients of the demonstration cities, with substantial income gains among low- and middle-income households. We argue that the improved internet infrastructure mitigated income inequality mainly by creating new employment opportunities in the service sector for low-skilled workers. Our findings underscore the pivotal role of technology in fostering inclusive economic growth and provide valuable insights for policymakers seeking to harness investments in internet infrastructure and improve social equality in developing countries.

Risk Tolerance, Soft Skills, and Sectoral Sorting: Lab-in-the-field Experiments in Urban India’s Dual Informal Labor Market (Under review)

(with Raja Rajendra Timilsina, Dil B Rahut, Tetsushi Sonobe)

Abstract The persistence of informal employment in developing economies raises central economic questions: why do some workers engage with digital platforms while others remain in traditional informal markets? What are the underlying economic preferences, despite the rapid expansion of the informal gig economy? This paper addresses this puzzle by using a lab-in-the-field experiment in urban India. We develop a framework in which workers are ex-ante sorted into digital or non-digital sectors. We then conduct experiments to collect data on risk tolerance, soft skills, and working conditions, while holding mobility to be limited. Our results reveal that younger non-digital workers exhibit higher risk tolerance than digital participants, while older digital workers tend to be more risk-neutral or risk-averse. Moreover, digital workers exhibit greater self-control and openness, but lower grit and conscientiousness, relative to their traditional-sector counterparts. These findings suggest that occupational sorting in secondary labor markets reflects heterogeneous preferences over income risk, flexibility, and community-based security, rather than simple technological adoption. This sorting mechanism in a utility-maximization model with entry frictions can be formalized to a broader understanding of labor market segmentation and the dynamics of informality in developing economies. The results have implications for policies aimed at improving welfare in dual labor markets, including the design of risk-sharing mechanisms, skill-building initiatives, and regulatory frameworks for digital platforms to scale up in the future.

Selected Works in Progress

Demand for Social Security Program among Informal Workers in Urban India (with Raja Rajendra Timilsina, Dil B Rahut, Ngawang Dendup, Tetsushi Sonobe)