Welcome!

Abstract

This paper examines how the textual context within questions on Brazil’s ENEM—the world’s second-largest college admission test—affects performance gaps across socioeconomic status (SES), gender, and ethnicity. Using data from over 3.8 million test-takers across 13 years (2010–2022), I analyze question-specific gaps, linking them to words and topics in each question through bag-of-words and topic modeling, combined with penalized regressions, to identify key drivers of these gaps. Based on top predictive words and topics, I generate hypotheses on how specific contexts influence disparities. Testing these hypotheses with a rich set of fixed effects that help rule out alternative drivers, I find that textual context matters: SES gaps grow by 0.8 percentage points (p.p.) –13% of the SES gap—when questions feature financial language but decrease by 0.3 p.p. (5% of SES gap) in everyday contexts. Gender gaps are sensitive to negative effects only among high-ability test-takers: for these students, abstract scientific contexts widen the gap by 0.3 p.p. (17% of the high-ability gender gap), while creative and social contexts reduce it consistently across the entire ability distribution by 0.6 p.p. (16% of gender gap). Additionally, questions featuring female characters help close gender gaps, while references to underprivileged individuals widen SES disparities. Ethnic gaps, however, show no significant text-based predictors except in the language subject. These findings show that textual context affects testing equity, highlighting paths to fairer design.

Abstract   

This paper evaluates a pro-competition reform in the Chilean pension system that shifted the enrollment process from individual choice to a default option set by competitive auctions held every two years. The reform consistently lowered fees and modestly improved the rate of participants switching to more cost-effective fund managers. Notably, only one of the six auctions analyzed significantly influenced participant behavior, with a slight increase in both the switching rate and the likelihood of selecting cheaper managers. While fee awareness and broader systemic improvements remained limited, reduced fees provided extra disposable income, which led to increased voluntary retirement savings among some members, suggesting that savings from lower fees are partly reinvested into retirement. Additionally, the study underscores a stark contrast in participant response to fee increases post-auction. While most firms maintained stable fees, one manager sharply raised them by 182% after the mandatory two-year price freeze, prompting approximately 16% of active affiliates to exit their plans—a reaction that highlights members’ heightened sensitivity to losses over equivalent gains. This finding reflects a strategic exploitation of contributor inertia and underscores the challenges of securing long-term benefits from competitive mechanisms alone, revealing the potential for managerial exploitation without sustained regulatory oversight  

Abstract

Breastfeeding is widely acknowledged as beneficial for both mother and child, yet empirical evidence establishing causal links remains limited. This study investigates the causal relationship between breastfeeding duration and subsequent child development outcomes, leveraging the timing of a major earthquake in Chile as an exogenous shock. Results indicate that proximity to the earthquake event negatively affects breastfeeding duration, with each additional week away associated with an increase in 1.5 days of extra breastfeeding. Employing this exogenous variation as an instrument, a two-stage least squares (2SLS) model reveals significant positive effects on child cognitive development indicators measured two years after the event, corresponding to approximately 0.035 standard deviations per additional month of breastfeeding. In turn, the impacts on cognitive domains seven years after show a positive point estimate that is not statistically significant. However, no statistically significant impacts are observed on non-cognitive and health outcomes during this timeframe. These results are consistent with controlled trial findings. Notably, analyses of maternal outcomes yield null effects across dimensions considered, including attachment quality and employment likelihood two and seven years post-quake, suggesting that the observed effects stem primarily from neural formation mechanisms rather than mother-child interaction channels. These findings carry significant implications for public policy, suggesting the efficacy of early breastfeeding promotion interventions in fostering child development

Abstract

Centralized admission systems are designed with the expectation that welfare improvements can be achieved through rule-setting, competition, and coordination. These systems, which are becoming increasingly popular, along with other mechanism design interventions, rely heavily on the assumption that participants can understand the rules and play optimally. However, recent evidence suggests that real-world applications of centralized admission systems often fail to meet theoretical benchmarks, leading to significant fairness implications. In this paper, we leverage a policy change in the Chilean system, which made previous years' cutoffs—an essential element in application strategies—less informative. This distortion affected students differently depending on their position in the ability distribution. Using comprehensive data on individuals who participated before and after the change, we show that increasing the complexity of the application process led to a reduction in the likelihood of being admitted to their first reported preference, with enrollment rates decreasing by 12 percentage points. We propose a model in which agents differ in their ability to predict the new cutoffs, explaining the observed behaviors based on differences in overconfidence. While all students applied with overconfidence, those with lower baseline overconfidence performed relatively better.