Abstract:  We examine an influence designer's optimal intervention in the presence of social learning in a network. Before learning begins, the designer implants opinions into the network to shift agents' ultimate opinions toward being as close as possible to the target opinions. By decomposing the influence matrix, which summarizes the learning structure, we transform the designer's problem into one with an orthogonal basis. This transformation allows us to characterize optimal interventions under complete information. We also demonstrate that even in cases where the designer has incomplete information about the network structure, the designer can still design an asymptotically optimal intervention in a large network. Finally, we provide examples and extensions, including repeated social learning and competition.

Abstract:  This paper explores the use of Large Language Models (LLMs) as decision aids, focusing on models' ability to learn preferences and provide personalized recommendations. To establish a baseline, we replicate standard economic experiments on choice under risk (Choi et al., 2007) with GPT, one of the most prominent LLMs, prompted to respond as (i) a human decision maker or (ii) a recommendation system for customers. With these baselines established, GPT is provided with a sample set of choices and prompted to make recommendations based on the provided data. From the data generated by GPT, we identify its (revealed) preferences and explore its ability to learn from data. Our analysis yields three results. First, GPT's choices are consistent with (expected) utility maximization theory. Second, GPT can align its recommendations with people's risk aversion, by recommending less risky portfolios to more risk-averse decision makers, highlighting GPT's potential as a personalized decision aid. Third, however, GPT demonstrates limited alignment when it comes to disappointment aversion.

Abstract:  Recent literature highlights that social choice in laboratory games can vary substantially across societies. Researchers have speculated, without direct evidence, that this may be partly due to cultural variation in focal points for decision-making. Here, we provide the first direct evidence of a focal point that influences social choice in one culture but not another. In Confucian-influenced cultures, and especially in South Korea, social structure is influenced by age hierarchies. In fact, age hierarchies are integrated directly into Korean linguistics, with age differences of even just one year dictating the respect levels, titles, formalities, and grammar that must be used between two individuals. Therefore, even small age differences are salient and relevant in virtually all Korean social contexts. This paper examines the role of age information on social choice in Korea and the US, using a coordination game, the Stag Hunt (SH), and a cooperation game, the Prisoner’s Dilemma (PD). We find that age information increased coordination in SH and cooperation in PD among Korean participants, primarily by increasing reciprocity to expected partner choices. Conversely, age information had 0% effects in American participants, thus providing the literature’s first direct evidence of a focal point for social decision-making that matters in one culture but not another. Moreover, baseline cooperation rates in PD were higher in the US (53%) than in Korea (46%), but these differences disappeared once age was provided (53% in both countries). This is because without age information, Korean participants were deprived of information that typically guides their decision-making. This cautions that cross-cultural researchers must account for culture-specific focal points in order to avoid asymmetry in external validity when comparing across societies.

Abstract:  We investigate the effects of individuals' strategic sophistication measured by level-k type on collective action in the context of social distancing during the early-stage of the COVID-19 pandemic. We build a weakest-link public goods game with a private cost of social distancing, in which agents are heterogeneous in level-k types. We find that players with higher level-k types are more likely to engage in social-distancing behaviors. We test this hypothesis with large-scale nationally representative survey data that measured level-k types through incentivized experiments. Our empirical findings provide consistent evidence with our theoretical prediction. This study sheds new light on the role of the level-k theory in understanding real-world collective action problems.

Abstract:  Promoting social capital has long been an important issue in social sciences. This paper argues that teaching practices can stimulate social capital at both the individual and the classroom levels by evaluating the impact of a student-centered teaching pedagogy program. We measure changes in students’ friendship network and directed altruism with comprehensive friendship surveys and incentivized dictator game experiments conducted before and after the intervention. We find that the project-based learning program positively affects social capital by expanding students’ friendship networks and being more generous toward their peers, especially those not in direct friendship and without homophilous traits. Moreover, structural estimations suggest that the program also reduces friendship formation costs among the students, especially for those studying in the same classroom. Our results support the idea that teaching practice focusing on student-centered learning can be considered an effective educational policy to support social capital formation among students.

Abstract:  We evaluate a program wherein an adaptive intelligent tutoring system, coupled with increased teacher engagement was applied to students’ interactive learning processes and mathematics achievement outcomes. A quasi-experimental design was utilized to analyze the effectiveness of a one-semester program implemented during regular mathematics classes for Grade 7 students in Vietnam. The intervention resulted in overall positive and significant learning gains, including evidence of direct teacher impacts on mathematics achievement, along with moderate improvement in the students’ perception of teacher efficacy and classroom interaction. Learning gains were greater for low-performing students at the baseline math performance. These results point to an effective solution for achieving quality education to improve learning outcomes.

Abstract:  We study a pandemic disease control problem in which an infectious disease like COVID-19 spreads over multiple communities and non-pharmaceutical interventions are employed to control the spread of the disease. Drawing upon a canonical susceptible-infected-recovered model, we formulate the problem as a discrete time, finite horizon, dynamic program. As a base model, we first develop an epidemic model in which one benevolent social planner uses non-pharmaceutical interventions to control the spread of disease within one community. We extend this base epidemic model to a more complex pandemic model that additionally considers the interactions between multiple communities. We derive the optimal policies for both epidemic and pandemic models and characterize their structural properties using analytical and computational approaches. For the pandemic model, we propose three different principles under which a social planner enforces intervention orders. We compare the optimal policies for the three different principles and discuss related managerial implications.