CAF - Lasting effects of retaking college admission exams

C A F - W O R KI N G P AP E R # 2 0 2 6 / 0 4

F i r s t v e r s i o n : M a r c h 1 1 , 2 0 2 6

T h i s v e r s i o n : M a r c h 1 2 , 2 0 2 6

Lasting Effects of Retaking College Admission Exams Veronica Frisancho1 | Sebastian Gallegos 2 | Constanza Gonzalez3 1Chief Economist at CAF, Development Bank of Latin America and the Caribbean vfrisancho@caf.com 2Associate Professor of Economics, Business School, Universidad Adolfo Ibañez, Chile, IZA@LISER Research Fellow and member of the Human Capital and Economic Opportunity Global Working Group (HCEO) at The University of Chicago. sebastian.gallegos@uai.cl 3Business School, Universidad Adolfo Ibañez, Chile. constanzagonzalez@alumnos.uai.cl Do second chances at a high-stakes admission exam yield longterm gains? Leveraging fifteen years of Chilean administrative data and an RDD, we examine the causal effects of retaking on educational and labor market trajectories. Narrowly missing a preferred program cutoff triggers a 44% increase in retaking, leading to substantial score gains (0.27 SD) and improved placement and enrollment chances. However, these immediate gains do not persist. Retakers graduate at the same rate and from programs with similar earnings and employability profiles as their counterfactual peers. Our results suggest that retaking serves as a reshuffling mechanism yielding null net welfare gains.

K E Y W O R D S

High-Stakes Exams, Developing Countries, Latin America, Educational Policy, College, Enrollment, Graduation, Choices.

Small sections of text that are less than two paragraphs may be quoted without explicit permission as long as this document is acknowledged. Findings, interpretations and conclusions expressed in this publication are the sole responsibility of its author(s) and cannot be, in any way, attributed to CAF, its Executive Directors or the countries they represent. CAF does not guarantee the accuracy of the data included in this publication and is not, in any way, responsible for any consequences resulting from its use. ©2025 Corporación Andina de FomentoC A F - D O C U M E NT O D E T R AB A J O # 2 0 2 6 / 0 4

E s t a v e r s i ó n : 1 2 d e m a r z o d e 2 0 2 6

Efectos Persistentes de Repetir los Exámenes de Ingreso a la Universidad Veronica Frisancho1 | Sebastian Gallegos 2 | Constanza Gonzalez3 1Chief Economist at CAF, Development Bank of Latin America and the Caribbean vfrisancho@caf.com 2Associate Professor of Economics, Business School, Universidad Adolfo Ibañez, Chile, IZA@LISER Research Fellow and member of the Human Capital and Economic Opportunity Global Working Group (HCEO) at The University of Chicago. sebastian.gallegos@uai.cl 3Business School, Universidad Adolfo Ibañez, Chile. constanzagonzalez@alumnos.uai.cl ¿Brindan las segundas oportunidades en exámenes de admisión de alto impacto beneficios a largo plazo? Utilizando quince años de datos administrativos chilenos y un diseño de regresión discontinua (RDD), examinamos los efectos causales de repetir el

examen en las trayectorias educativas y laborales. El hecho de no alcanzar por un margen estrecho el puntaje de corte de un programa preferido detona un aumento del 44 % en la repetición del examen, lo que conduce a mejoras sustanciales en los puntajes (0,27 DE) y mayores probabilidades de asignación y matrícula. Sin embargo, estas ganancias inmediatas no persisten. Quienes repiten el examen se gradúan a la misma tasa y de programas con perfiles de ingresos y empleabilidad similares a los de sus pares contrafactuales. Nuestros resultados sugieren que la repetición del examen funciona como un mecanismo de redistribución que genera beneficios netos de bienestar nulos.

K E Y W O R D S

Exámenes de Alto Impacto, Países en Desarrollo, América Latina, Política Educativa, Educación Superior, Matrícula, Graduación, Elección Escolar Pequeñas secciones del texto, menores a dos párrafos, pueden ser citadas sin autorización explícita siempre que se cite el presente documento. Los resultados, interpretaciones y conclusiones expresados en esta publicación son de exclusiva responsabilidad de su(s) autor(es), y de ninguna manera pueden ser atribuidos a CAF, a los miembros de su Directorio Ejecutivo o a los países que ellos representan. CAF no garantiza la exactitud de los datos incluidos en esta publicación y no se hace responsable en ningún aspecto de las consecuencias que resulten de su utilización. ©2025 Corporación Andina de FomentoFRISANCHO ET AL. 2

1|INTRODUCTION Most educational systems around the world grade and rank students according to their performance in one or more subjects. When these ranks are used to allocate scarce university seats, standardized exams become high-stakes pivots for a student’s future. Consequently, scoring just below or above a given threshold can significantly alter educational trajectories (Kirkeboen et al., 2016; Machin et al., 2020; Chetty et al., 2023). A relatively novel—and still thin—strand of literature examines how students respond to these high-stakes outcomes through retaking behavior. To date, this research has focused almost exclusively on immediate outcomes, consistently documenting gains in test scores and admission probabilities in the U.S., China, and Greece (Goodman et al., 2020; Kang et al., 2024; Bizopoulou et al., 2024). However, the literature remains largely silent on whether these initial gains translate into meaningful long-term success, leaving a gap in our understanding of how retaking ultimately shifts enrollment persistence, graduation rates, and eventual employment prospects. This paper contributes to filling this gap by studying both the immediate and longterm causal effects of retaking on educational and labor market outcomes. We combine 15 years of administrative data from the centralized college admission system in Chile with a Regression Discontinuity Design (RDD). Because placement in this system is driven exclusively by exam performance and stated preferences for university-major combinations, we are able to isolate these two key channels and unravel the mechanisms behind our results. Beyond immediate placement, we track the impact of retaking on enrollment and graduation. Finally, by linking these records to program-level data on employability and wages, we provide some of the first evidence of the causal effects of retaking on long-term earnings and employment prospects.

Our analysis focuses on the 2010 cohort of first-time exam takers and applicants, for whom we reconstruct educational trajectories over a 15-year period. First, we use centralized admission records to observe retaking behavior, shifts in test performance, and evolving preferences, as well as final placement outcomes across successive attempts. Second, we link these files to longitudinal administrative records from the Ministry of Education (MINEDUC) for the period 2010–2025, allowing us to track enrollment and graduation across all public and private universities in Chile. Finally, we incorporate program-level labor market indicators derived from national tax records (Servicio de Impuestos Internos, SII). These data provide expected employment probabilities and wage distributions across tertiary programs, enabling us to estimate how retaking shapes long-term earning potential and labor market integration. To address the non-random nature of the decision to retake, we exploit the exogenous variation provided by admission cutoffs. Following the empirical strategy used in recent studies of centralized systems in Greece, China, and the U.S. (Bizopoulou et al., 2024; Kang et al., 2024; Goodman et al., 2020), we compare students who narrowly missed the threshold for their most preferred program to those who barely cleared it. Our first-stage estimates show that "near-miss" applicants are 4.3 percentage points more likely to retake the exam—a 44% relative increase that aligns closely with findings in other international contexts. This retaking behavior leads to a significant performance gain of 0.27 standard deviations between the first and second attempts. Consequently, threshold-induced retakers see improvements in admission and enrollment rates. Crucially, we show that these gains are driven primarily by improved test scores rather than changes in application strategy or preferences. Despite these substantial immediate gains, the advantages of retaking appear to dissipate

over the long run. Following students for fifteen years after their first attempt, we findFRISANCHO ET AL. 3 that the initial boost in placement does not translate into better educational or labor market outcomes; retakers graduate at the same rate as non-retakers and enter programs with nearly identical employability and expected earnings. These results suggest that while retaking helps marginally weaker students access their desired programs, it does not fundamentally alter their long-term economic trajectory. Notably, however, we find no evidence that these students “lag behind” once enrolled; the fact that graduation rates remain stable implies that retakers are as prepared for the rigors of university as their peers who were admitted on their first attempt. All in all, our findings provide a cautionary tale: second chances may improve admission and enrollment outcomes, but these seem to be decoupled from lasting gains in graduation rates and the labor market. Students’ responses to failure in the context of centralized admission systems remain an understudied topic. A nascent but growing literature explores the decision to retake exams and its subsequent impact on short-run outcomes. Early evidence from Turkey (Frisancho et al., 2016) and the United States (Goodman et al., 2020) suggests that the opportunity to retake leads to substantial score improvements, which consequently leads to better college admission prospects. The improvement in scores in Turkey and the US were particularly salient among less-advantaged students, thereby narrowing gaps in college access. Building on these findings, a recent set of studies has leveraged the discrete nature of centralized admission systems to implement fuzzy Regression Discontinuity Designs (RDD). By exploiting exogenous variation in retaking probabilities around admission cutoffs, these papers isolate the causal gains of a second attempt. Bizopoulou et al. (2024) take advantage of a minimum qualifying score to get access to tertiary education in Greece and

find that low-achieving students who retake national admission exams obtain better quality placements due to improved performance and more ambitious preferences the second time around. Similarly, in the competitive context of the French Grandes Écoles, Landaud and Maurin (2020) document that repeating a year allows marginal students to improve their performance by roughly 25 percentile ranks. Most recently, Kang et al. (2024) use data for China to exploit discontinuities in the retaking probability around the entry cutoff for selective universities (schools in the top two tiers, accessible to the top 25% of students). Again, retaking generates increases in exam performance (0.47 SD) and relative ranking (11 percentage points). While the evidence consistently documents immediate gains in scores and placement, our paper extends this literature by tracking these effects into graduation and the labor market to determine if such “second chances” yield lasting economic returns.1 Our study builds on recent contributions and poses at least three key advantages. First, while the pioneering work of Landaud and Maurin (2020), Krishna et al. (2018), Kang et al. (2024), and Goodman et al. (2020) has established a robust understanding of the immediate effects of retaking on academic performance and initial placement, the longerterm trajectories regarding graduation rates and labor market outcomes remain an open empirical question. Our focus on these lasting outcomes allows for a more complete assessment of the welfare implications of retaking. Second, our setting allows us to disentangle the underlying the drivers of change, i.e., performance and/or preference shifts. While the existing literature primarily focuses on test score gains, the Chilean centralized system provides a direct, unconditioned look at students’ 1Previous research highlights the complexities of these systems; for instance, Krishna et al. (2018) use a structural model of the Turkish exam system to show thatlimitingretakes can actuallyimprovewelfare for most agents.

This is largely due to a general equilibrium effect, where retakers crowd the market and impose negative externalities by driving up admission cutoffs. Similarly, a recent study in Italy finds that restricting retakes at a major Italian university improved student efficiency, leading to lower dropout rates, higher exam pass rates, increased credit accumulation, and improved on-time graduation rates (Bratti et al., 2024).FRISANCHO ET AL. 4 ranked preferences. By observing these lists across successive years, we can determine whether retakers secure better placements simply by scoring higher, or by strategically changing their university-major choices. This level of granularity is often unavailable in other contexts. Third, other studies relying on RDDs focus on minimum entry cutoffs (Bizopoulou et al., 2024) or thresholds to ration access to elite universities (Kang et al., 2024), which produces results for students at the bottom or the top of the score distribution, respectively. By relying on cutoffs defined at the individual level, our analysis sample is likely more representative of a broader segment of the performance distribution, providing more generalizable insights. The remainder of this article is organized into five sections. The following section presents the institutional background. Section 3 describes the data and defines the outcomes variables. Section 4 presents the estimation strategy while section 5 presents the results. Section 6 concludes by discussing the broader implications of our results. 2|COLLEGE ADMISSIONS AND RETAKING IN CHILE College Admissions. The Chilean college admission process is largely centralized, with coordinated application and selection procedures for the majority of higher education institutions across the country. This system was designed to ensure a standardized, transparent, and fair method of admitting students into various undergraduate programs. Central to this system is the Prueba de Selección Universitaria (PSU), similar to the SAT in the United States.2

These standardized tests are critical in determining student eligibility and placement in universities. Test takers complete exams in mathematics and language. Consistent with most OECD countries (Kirkeboen et al., 2016), the Chilean system is characterized by early specialization: students apply directly to a specific major upon entry. This stands in contrast to systems like that of the United States, where students generally enter a broad program and defer their major choice until later in their undergraduate studies. Prospective students apply to university-major combinations ranked based on personal preference, via a centralized process. The students’ application score for each preference is a major-specific weighted average of the PSU test scores that are relevant to compete for a seat in each major (about 80%) and their high school GPA (about 20%). The allocation process ranks all the students applying to any given major using their respective major-specific scores. Admission to Chilean tertiary education is governed by a centralized annual cycle. Each December, graduates take a suite of mandatory (Mathematics and Language) and elective (History or Science) exams. Crucially, the selection criteria—including the weights assigned to each exam and to high school records—are predetermined by the universities before the tests are administered. Once scores are disclosed in January, students use a centralized platform to apply to a ranked list of programs, with placement determined solely by their weighted scores and stated preferences. The allocation process ranks all applicants to a given major by their weighted scores, following a student-proposing deferred acceptance algorithm (Abdulkadiroˇglu et al., 2014; Rios et al., 2021). In the first round, students are tentatively assigned to their top-ranked preference. Each program then rejects students whose scores fall below the program’s capacity limit. In each subsequent roundr, students rejected in roundr−1 are considered

for their next highest-ranked preference. Programs then re-evaluate all currently held and 2Chile possesses one of the world’s longest-standing centralized admission systems, established in 1967 with the Academic Aptitude Test (PAA). Although the exam has undergone several structural and nomenclature revisions—evolving into the University Selection Test (PSU) in 2004, the Transition Test (PDT) in 2021, and the Higher Education Access Test (PAES) in 2023—the fundamental administrative cycle of examination, results disclosure, preference submission, and centralized enrollment has remained consistent for over five decades.FRISANCHO ET AL. 5 newly arriving applicants, again rejecting those in excess of capacity and updating their provisional admission lists. This iterative process concludes when all students are either matched to a program or have exhausted their preference list. This mechanism naturally defines a score cutoff for each program—the score of the last student admitted. Once finalized, students are matched to at most one program (their highest possible preference).At this stage, admitted students must choose whether to enroll, while those without an offer may seek placement in non-selective institutions outside the centralized system. The centralized admission process is administered by the Department of Educational Evaluation, Measurement, and Registry (DEMRE). As of the 2025 cycle, 47 universities—representing approximately 81% of all universities in Chile—participate in the system, offering roughly 1,800 academic programs. Participation has expanded significantly over time; in 2010, the system was restricted to the 25 member institutions of the Council of Rectors of Chilean Universities (CRUCH). These long-standing institutions are widely regarded as highly selective and academically rigorous. In subsequent years, additional universities have joined the centralized system as they met the necessary participation requirements. Unlike the flexible, multi-date schedule of the U.S. SAT, the Chilean system—similar

to those in Greece and Turkey—relies on fixed, nationally synchronized testing windows that have historically occurred only once per year. Furthermore, during our study period, scores were non-transferable between admission cycles; applicants were required to retake the examination to participate in a subsequent year’s process. Retaking. Over the last ten years, the number of students taking the exam has fluctuated between 225,000 and 250,000. Retaking is prevalent in the Chilean context; each year, between 15% and 20% test-takers are making their second or subsequent attempt. The vast majority of these students sit for the exam in the year immediately following their first attempt. On average, three out of four retakers in any given round are individuals who first took the exam just one year prior. Over time, the retake rate in Chile has remained relatively stable, at levels comparable to other countries with single annual admission cycles, such as Greece (15%–20%) and China (13%–15%), yet significantly lower than in systems with multiple annual testing windows, such as the United States (54%). The decision to retake the exam reflects a trade-off between expected benefits and costs and can be viewed as an investment in human capital. Students incur costs today in exchange for higher expected returns through improved college placement. In the Chilean context, the potential gain arises from obtaining a higher score in the next admission cycle, one year later, which may allow access to more selective programs with higher expected returns. There is no formal penalty for retaking. Universities do not distinguish between first-time and repeat applicants, and there are no limits on attempts or age restrictions. Retaking, however, is costly. Students delay enrollment by one year, implying foregone time and potential earnings. They must also pay a registration fee of about USD 40–50; this fee is waived for eligible first-time applicants but not for repeat takers. In addition,

many students purchase preparatory services, private tutoring, or paid online platforms.3 Students will retake if the expected returns from improved placement exceed these time and monetary costs. 3These expenditures are private and are not recorded in administrative data, either in our dataset or in any centralized source. As a result, they cannot be directly measured.FRISANCHO ET AL. 6 3|DATA 3.1|Sources and Outcomes Sources. Our primary data source consists of administrative records from the centralized admission system managed by DEMRE between 2010 and 2025, we supplement these data with further administrative records from the Ministry of Education (MINEDUC) and aggregated information from the Chilean Internal Revenue Service, as we describe below. The DEMRE data cover the population of test-takers in Chile in each exam round and provide baseline demographic and academic information, including high school GPA, type of high school, gender, household income, and parental schooling. The records also include exam scores for each mandatory subject (language and mathematics), application scores, and the submitted rank-ordered list of preferred university–major combinations, as well as admission results. We complement these records with administrative data from the MINEDUC, which compiles annual information from all higher education institutions in Chile. Using encrypted individual identifiers, we merge the DEMRE files with MINEDUC records that track students’ enrollment and graduation histories. These records allow us to follow individuals over time within the higher education system. Finally, we incorporate program-level labor market indicators from the Chilean Higher Education Information Service (SIES). The SIES aggregates individual tax records provided by the Chilean Internal Revenue Service (Servicio de Impuestos Internos, SII) to the university-program level. The data released for 2024 and 2025 report employability rates and average income levels for 1,090 distinct university programs. These indicators allow us to characterize the expected labor market prospects associated with different tertiary

programs. Outcomes. We categorize our outcomes into two primary groups: immediate academic results and long-term educational and labor market trajectories. Immediate Outcomes.Immediate outcomes include admission exam scores, preferences, admission offers, and enrollment results. We use the scores standardized within the cohort of test takers. We use students’ submitted preferences over university programs to measure application behavior proxied by the selectivity of each applicant’s portfolio using the average cutoff score of the listed programs.4 In additional analyses, we examine alternative metrics of application behavior, including the total number of programs listed, the selectivity of the top-ranked choice, the dispersion (interquartile range) of admission thresholds across the preference list, and the share of selective programs in the application portfolio. The admission outcomes come directly from the centralized matching mechanism. Because admission offers are determined by exam performance and submitted preferences, these outcomes allow us to assess whether improvements in scores or changes in application behavior translate into improved placement. Finally, we observe actual enrollment decisions using the MINEDUC administrative records. These data allow us to observe whether individuals enroll in tertiary education and the institution and program in which they enroll, providing a direct measure of realized educational choices rather than admission offers alone. 4To ensure our measure of selectivity is exogenous to the 2010 applicant pool, we rely on the 2009 cutoff scores as a fixed benchmark for program selectivity.FRISANCHO ET AL. 7 Long-Term Outcomes.Our long-term outcomes include graduation, time to graduation, and prospects of employability and earnings at the program level. We construct graduation outcomes at the individual level using the MINEDUC administrative records for all available years. These data allow us to observe whether students graduate, the exact date of graduation, and the university-program of their degree. Because the records cover the full population of higher education institutions in Chile and

are available annually from 2010 to 2025, we can follow individuals for up to fifteen years after their first attempt at the admission exam in 2010. This feature of the administrative data is particularly important for studying graduation outcomes. Graduation is inherently difficult to measure because students complete their programs at different times, and shorter observation windows may miss late graduates. By tracking the full population over a long horizon and observing the exact graduation date for each individual, our data allow us to construct precise measures of both degree completion and time to graduation. We characterize students’ labor market prospects using program-level indicators of employability and expected earnings from the SIES dataset. These measures include employment rates up to two years after graduation and average earnings in the fourth year after graduation. Because these indicators are constructed using aggregated tax records from the Chilean Internal Revenue Service (SII), they provide a consistent measure of the expected labor market outcomes associated with each tertiary program. We use these outcomes allow us to assess whether the initial gains associated with retaking translate into meaningful differences in long-run educational attainment and expected labor market prospects. 3.2|Sampling Our main analysis sample comprises applicants who took the university entrance exam in 2010 for the first time. We exclude test-takers who participated in a previous admission process. These criteria define a working sample of about 46 thousand students. We follow these students over time and classify their retaking status based on whether they retake the admission exam in 2011, the year immediately following their first attempt.5 We define our running variable as the margin between a student’s application score for their top-ranked program and the score of the last student admitted to that same program. Our RDD analysis uses a sample of 19,387 observations falling within a ±27.5point bandwidth of the admission threshold. This bandwidth is selected using the MSEoptimal selector proposed by Calonico et al. (2019) for the first-stage specification. We note that, in principle, the optimal bandwidth in a regression discontinuity design is outcomespecific, as it depends on the variance of the corresponding dependent variable. To facilitate comparability across specifications and maintain a common analytical bandwidth sample, we adopt the first-stage optimal bandwidth for all outcome estimations. Importantly, our results are robust to the use of outcome-specific optimal bandwidths and to alternative bandwidth choices (see Table B.1 in Appendix B). 3.3|Descriptive Statistics Table 1 presents the mean sociodemographic characteristics for students in both the working and the bandwidth samples, further disaggregated by retaker status. Overall, both samples exhibit similar demographic profiles and high school backgrounds. However, students in the bandwidth sample appear slightly more advantaged in terms of household income and parental education. In both the working and bandwidth samples, retakers are more likely to 5We do so because most retaking behavior occurs immediately after the first attempt, while delayed retakes represent a small and quickly diminishing share of the cohort.FRISANCHO ET AL. 8 have lower high school GPAs and come from more advantaged socioeconomic backgrounds compared to non-retakers. The last column of Table 1 presents the mean characteristics of compliers, computed following Goodman et al. (2020) and the methods developed by Abadie et al. (2002) and Abadie (2003). In our setting, compliers are the students who only retake because they narrowly missed the cutoff of their top preference. If they had barely passed, they would have enrolled instead. Overall, we find that compliers are relatively similar to all retakers along observable dimensions. There are modest differences in a few covariates: compliers are more likely to be female, live in the capital city, and have higher household income.

Table 2 presents descriptive statistics on admission exam performance, placement and enrollment, comparing the full sample (Columns 1–3) with the bandwidth sample (Columns 4–6). As expected, first-attempt scores are higher on average within the bandwidth sample, reflecting the fact that these students scored closer to the admission cutoffs of their topranked programs. TA B L E 1Baseline Characteristics (Working and Bandwidth Samples) Sample Bandwidth Sample (1) (2) (3) (4) (5) (6) (7) All Non-retaker Retaker All Non-retaker Retaker Complier Retakers 0.18 - 1.00 0.15 - 1.00Female 0.51 0.50 0.52 0.49 0.49 0.48 0.54 GPA Score 610 612 603 620 621 614 618 Type of School Public School 0.30 0.30 0.29 0.29 0.29 0.29 0.27 Voucher School 0.49 0.48 0.53 0.48 0.47 0.50 0.50 Private School 0.21 0.21 0.18 0.23 0.23 0.20 0.23 No information 0.01 0.01 0.00 0.01 0.00 0.01 0.00 Capital Region 0.31 0.30 0.33 0.34 0.34 0.35 0.41 Parents’ Schooling Mother’s years of education 12.56 12.51 12.76 1