Mohit Karnani

We study the welfare and human capital impacts of colleges’ (non)participation in Chile’s centralized higher-education platform, leveraging administrative data and two policy changes: the introduction of a large scholarship program and the inclusion of additional institutions, which raised the number of on-platform slots by approximately 40%. We first show that the expansion of the platform raised on-time graduation rates. We then develop and estimate a model of college applications, offers, wait lists, matriculation, and graduation. When the platform expands, welfare increases, and welfare, enrollment, and graduation rates are less sensitive to off-platform frictions. Gains are larger for students from lower-socioeconomic-status backgrounds.

During the Coronavirus Disease 2019 (COVID-19) epidemic, many health professionals used social media to promote preventative health behaviors. We conducted a randomized controlled trial of the effect of a Facebook advertising campaign consisting of short videos recorded by doctors and nurses to encourage users to stay at home for the Thanksgiving and Christmas holidays. We randomly assigned counties to high intensity or low intensity. The intervention was delivered to a large fraction of Facebook subscribers in 75% and 25% of randomly assigned zip codes in high- and low-intensity counties, respectively. In total, 6,998 (6,716) zip codes were included, and 11,954,109 (23,302,290) users were reached at Thanksgiving (Christmas). The first two primary outcomes were holiday travel and fraction leaving home, both measured using mobile phone location data of Facebook users. Average distance traveled in high-intensity counties decreased by −0.993 percentage points for the 3 days before each holiday compared to low-intensity counties. The fraction of people who left home on the holiday was not significantly affected. The third primary outcome was COVID-19 infections recorded at the zip code level in the 2-week period starting 5 days after the holiday. Infections declined by 3.5% in intervention compared to control zip codes. Social media messages recorded by health professionals before the winter holidays in the United States led to a significant reduction in holiday travel and subsequent COVID-19 infections.

Do messages delivered by physicians increase COVID-19 knowledge and improve preventive behaviors among White and Black individuals? In this randomized clinical trial of 18,223 White and Black adults, a message delivered by a physician increased COVID-19 knowledge and shifted information-seeking and self-protective behaviors. Effects did not differ by race, and tailoring messages to specific communities did not exhibit a differential effect on knowledge or individual behavior. These findings suggest that physician messaging campaigns may be effective in persuading members of society from a broad range of backgrounds to seek information and adopt preventive behaviors to combat COVID-19.

We study the determinants of sudden stops in capital flows to emerging markets. Using gross international asset and liability flows (from the point of view of domestic residents), we identify three types of situations: countries that do not experience any type of sudden stops; those who experience a sudden stop in inflows (liabilities), but no sudden stop in their net financial account of the balance of payments; and countries who suffer a sudden stop in inflows and in their net financial account. Based on these three events, we estimate a multinomial logit model and obtain two important results. We find that developed countries have about the same probability of experiencing sudden stops in gross capital inflows as emerging economies. Moreover, the probability of experiencing a sudden stop in gross inflows that winds up becoming a sudden stop in the financial account is affected by the behavior of a country’s international assets: countries whose agents possess assets abroad tend to repatriate them during periods of sudden stops in inflows, while the economies of countries whose agents do not possess foreign assets are much more sensitive to the behavior of foreign investors: a sudden stop in inflows can have very adverse effects on output and employment.

This paper studies the estimation of peer effects in centralized school choice systems. These centralized assignment mechanisms often ration seats by lottery, and thus classroom composition and the presence of certain types of peers in a cohort can be the result of a random assignment process. We show how this random variation can be exploited to identify peer effects. The estimator relies on two different propensity scores that can be extracted from a general class of centralized matching mechanisms, and it emulates the setting of a stratified clustered randomized experiment. We use this approach to study the impact of immigrant classmates on native children in Chile’s public school system. Our findings suggest that the presence of immigrant classmates has no significant impact on native children’s academic achievement, but it does reduce reported victimization by classmates and boosts reports of other prosocial attitudes. Additionally, we find that students already enrolled in a school whose grade receives a migrant entrant are more likely to remain enrolled in that school. These results provide insights into the dynamics of peer effects in educational settings and have implications for policy design in school assignment systems.

This paper studies the estimation of treatment effects using data from randomized controlled trials (RCTs) in the context of matching systems, such as centralized school choice systems. If treated units change their inputs (e.g. reported preferences), they can affect the allocation outcomes determined by the matching system in equilibrium. When the outcome of interest is determined in equilibrium (e.g. being matched to a target school), conventional estimators that compare a treatment group against a control group fail to account for violations of the stable unit treatment value assumption (SUTVA) induced by the matching system. Therefore, they yield biased and inconsistent estimates of the average treatment effect on the treated (ATT), even under perfect randomization designs. I propose an estimator that decomposes the observed average difference between the outcomes of the treatment and control units, and consistently estimates the ATT, along with the residual spillover effect. This nonparametric estimator relies on the matching mechanism satisfying a version of the equal treatment of equals (ETE) property. Moreover, the ATT is identified for any counterfactual fraction of treated units, which allows for the estimation of treatment effects under scaled interventions. I illustrate the applicability of this estimator using data on the Chilean centralized school choice system and the Ecuadorian teacher labor market.

We estimate community-wide effects on student achievement, as measured by performance on New York State proficiency tests, of charter school expansions in Harlem. Synthetic difference-in-differences (SDID) models estimate trends before and after enactment of the 2007 New York Charter Schools Act amendment, an exogenous supply shock that doubled the limit on the number of charter schools throughout the state, including an additional 50 in New York City. Approximately 400 weighted control schools with near identical pre-treatment trends in test scores are selected from 50 large urban districts in the state. Harlem’s charter enrollment share increased from 11% in 2007 to 47% in 2019, as compared to minimal change in the comparison districts. Post-enactment, Harlem’s ELA and Math proficiency rates, relative to those at the synthetic counterfactual, each rose by 13.8 percentage points. Both charter and district sectors contributed to these gains, though the largest gains came from the charter sector.

This paper studies screening and recruiting policies that restrict or incentivize entry to teacher-colleges. Using historical records of college entrance exam scores since 1967 and linking them to administrative data on the population of teachers in Chile, we first document a robust positive and concave relationship between precollege academic achievement and several short and long run measures of teacher productivity. We use an RD design to evaluate two recent policies that increased the share of high-scoring students studying to become teachers. We then show how data-driven algorithms and administrative data can enhance similar teacher screening and recruiting policies.