Research

Information Bias and Selection of Female Professors (PDF)

Conference version selected for publication in AEA Papers and Proceedings 

Media Coverage: State Press 

Abstract. Individuals often rely on crowd-generated ratings to form beliefs and guide decisions, yet these signals often embed bias. In higher education, students frequently consult online teaching evaluations when selecting instructors. Using survey evidence from one of the largest U.S. public universities along with a randomized control trial, I document that female professors receive ratings that are 5.5% lower than otherwise comparable male professors. Students place substantial weight on higher ratings but remain unaware of the bias embedded in them, resulting in distorted beliefs and enrollment choices. To assess whether such distortions can be corrected, I implement an informational intervention that shifts student beliefs. The intervention raises students’ valuation of female instructors by 10% relative to course cost, attenuates biased rating behavior among active raters, and increases reported female-taught chosen courses by 18% in the semester following the treatment. To interpret these findings, I develop and estimate a Bayesian updating model of belief formation and professor selection. The model shows that students who rely on professor review websites disproportionately internalize biased signals, generating systematic undervaluation of female faculty consistent with statistical discrimination. Finally, I quantify the consequences of this bias by documenting a utility loss equivalent to a 5% increase in textbook costs over a six-course semester due to suboptimal enrollment decisions. 

Assessing the Heterogeneous Effects of Remedial Education (with Esteban Aucejo) (PDF)

Journal of Human Capital 

Abstract. Using data from a large U.S. public university, we analyze the effect of remedial math education on student outcomes with a regression discontinuity approach. Students with different math preparation levels are placed in remedial courses of varying difficulty. Our study finds that remediation increases graduation rates by 22.5 percentage points for students with lower math skills in math-intensive majors. However, students with stronger math backgrounds in advanced remedial classes do not see improved graduation outcomes. Additionally, implementing an adaptive learning remedial program increased passing rates by 9.5 percentage points, leading to a 4.5 percentage point rise in five-year graduation probability. 

Assessing the Role of Study Habits on Students' Beliefs and Academic Performance (with Agustina Affonso Peyre, Esteban Aucejo, and Tomas Larroucau)

Status: Draft coming soon

Abstract. Using survey and administrative data from more than 2,300 students in intermediate economics courses at Arizona State University, we document that students often overvalue ineffective habits such as rereading and cramming while undervaluing effective strategies like retrieval practice. Students who adopt these habits perform worse and systematically mispredict their academic outcomes. We also find a strong correlation between beliefs about returns to effort and study choices, suggesting that inefficient strategies are tied to biased perceptions of how effort translates into grades. We implement a randomized intervention providing either general or personalized feedback on study strategies. Personalized feedback realigns beliefs, reduces reliance on ineffective study habits, and improves exam performance by 0.07 standard deviations, while general feedback has limited effects. A dynamic model of study habit choice under belief updating shows that personalized feedback has an equivalent effect to lowering perceived cost of adopting effective strategies by 14 percent.

College Navigation and Student Outcomes (with Agustina Affonso Peyre, Esteban Aucejo, and Jane Cooley Fruehwirth)

Status: Data collected and undergoing analysis

Abstract. This project evaluates whether personalized feedback can improve student well-being and academic performance in college. We conduct a randomized controlled trial with 2,800 undergraduates at Arizona State University. At the start of the semester, students complete a survey that measures challenges across four domains of college life: study habits, academic performance, social integration, and financial stress. Based on these scores, students are randomly assigned to treatment or control. Treated students receive personalized emails highlighting their weakest domain, comparative feedback relative to peers, and targeted video resources designed to address their primary area of concern. Outcomes are measured through follow-up surveys and academic records, focusing on exam performance. The study aims to assess whether tailored, low-cost interventions can reduce barriers to student success and improve persistence in higher education.

AI Study Habits ChatBot and Student Classroom Outcomes (with Agustina Affonso Peyre, Esteban Aucejo, and Martin Carbajal Vega)

Status: In the field

Abstract. This study evaluates whether AI-powered academic coaching can improve college students’ study habits and classroom performance. We implement a randomized controlled trial with 1,500 undergraduate students enrolled in intermediate economics courses at Arizona State University. After completing a baseline survey on study strategies and academic expectations, students are randomly assigned to one of three groups: a control group, a group receiving personalized written feedback on their study habits, and a group receiving the same feedback plus access to a custom-designed AI chatbot. The chatbot provides tailored academic support by interpreting study scores, generating personalized study plans, answering freeform questions, and processing course materials in-session. Follow-up survey data and administrative academic records will be used to measure effects on beliefs about study habits, time allocation, attitudes toward AI, and academic outcomes.

Enhancing Active Learning in Microeconomics: The Implementation of Interactive Educational Bots (with Agustina Affonso Peyre, Esteban Aucejo, and Patricia Ramirez De La Vina)

Status: In the field

Abstract. We evaluate the impact of AI-powered instructional bots on student learning and study behaviors in introductory microeconomics courses at Arizona State University. These conversational bots, aligned with core topics such as demand, supply, and elasticity, provide tailored explanations, walk students through practice problems, and offer real-time feedback. In a randomized controlled trial, students are assigned either to a treatment group that receives access to the bots or to a control group that relies solely on standard course materials. The intervention integrates the bots into course assignments, weekly study reminders, and pre-exam reviews. Outcomes are measured using academic performance scores, changes in beliefs and study habits, engagement patterns, and bot usage data. By testing whether AI tutors can substitute for or complement traditional instruction, this study provides evidence on the scalability and effectiveness of AI tools in higher education.