ITHACA, N.Y. — New research from Cornell University suggests that thousands of New York City eighth-graders may be missing opportunities to attend academically competitive public high schools—not because they lack the qualifications, but because the city’s admissions system is so complex that many students do not apply to schools where they could succeed.
The phenomenon, known as “undermatching,” occurs when students enroll in institutions that are less selective or less academically competitive than the options available to them. In New York City, the problem is amplified by the sheer scale of the high school application process. Students and families must navigate roughly 900 programs, each with different academic profiles, admissions criteria, locations, application requirements and levels of competition. The result is a choice environment in which having access to many options does not necessarily mean being able to identify the best ones.
“It’s a crazy complex process for how students apply and then match with high schools in New York City,” said Nikhil Garg, an assistant professor of operations research and information engineering at Cornell Tech and Cornell’s Duffield College of Engineering. Garg co-authored the study “Connecting Application Behavior to Undermatching in New York City School Choice,” published Aug. 18 in Nature Cities. The research examines how students make decisions when they have incomplete information about their chances of admission and how those decisions can shape educational outcomes.
The research team analyzed application and admissions data from nearly 59,000 students who participated in New York City’s high school admissions process during the 2022–23 school year. Rather than examining only the schools students ultimately attended, the researchers reconstructed alternative possibilities. They assessed where each student might have been admitted if the student had applied to different programs, allowing the team to distinguish between a student’s academic eligibility and the choices made during the application process.
To do this, the researchers combined administrative admissions data with a behavioral model of school choice under uncertainty. Such models are designed to account for the fact that families do not know precisely how competitive a program will be in a given year, how many students will apply, or how their own preferences will interact with the city’s matching mechanism. A student may avoid a highly desirable school because the family believes admission is unlikely, even when the student’s academic record would make admission realistic. Conversely, a student may concentrate applications on familiar or nearby programs without realizing that other schools offer stronger academic opportunities.
The results revealed a substantial gap between the schools students attended and the schools they might have accessed. On average, students enrolled in programs that admitted approximately 64% of applicants. Yet the researchers estimated that those same students could, in many cases, have been admitted to programs that accepted only about 37% of applicants. Because admission rates provide an indirect measure of competitiveness, the difference suggests that many students were not applying to the most selective programs within their reach.
The pattern was not evenly distributed. Undermatching was greatest among students whose application profiles were strongest, meaning that the students most capable of gaining admission to competitive programs were often the least likely to apply to them. The researchers also found larger gaps among Black, Hispanic and lower-income applicants. These disparities raise the possibility that unequal outcomes are being produced not only by differences in academic preparation or school quality, but also by unequal access to information, guidance and confidence during the application process.
The findings point to a technical problem with a human dimension. In a conventional matching system, a student’s final placement may appear to reflect demand, capacity and eligibility. But the matching algorithm can evaluate only the schools that appear on an application. If a student never lists a competitive program, the system cannot consider that option, regardless of whether the student would have been admitted. In this sense, application behavior functions as a hidden stage of the admissions process, one that can reproduce social and informational inequalities before the formal matching algorithm begins.
A second Cornell-led study tested whether personalized recommendations could help address that hidden stage. Led by doctoral student Erica Chiang, the research team developed a recommendation system and worked with New York City Public Schools to deploy it during the 2025–26 admissions cycle. The system sent individualized email messages to students attending middle schools that had historically sent relatively few students to high-performing high schools. Each message highlighted nearby programs for which the student was predicted to have a strong chance of admission.
The recommendations were generated using student and school information together with models of admission likelihood. The objective was not simply to direct every student toward the most selective school, but to identify programs that combined academic opportunity, geographic accessibility and a realistic probability of admission. This distinction is important in a capacity-constrained system, where an overly aggressive recommendation strategy could create new problems by sending too many applicants toward the same limited number of seats.
In a randomized controlled trial, students who received the personalized messages were 59% more likely to apply to a recommended program than students who did not receive them. The result suggests that relatively simple interventions—delivered at the right time and tailored to individual circumstances—can change application behavior. It also provides evidence that some students may not be rejecting competitive schools after carefully weighing their options; they may simply lack reliable information about which programs are within reach.
Yet recommendation systems in school admissions cannot be evaluated only by whether they increase applications. Every school has a fixed capacity, and an increase in applications can alter the competitive environment for everyone else. If a recommendation system directs too many students to one program, the program’s effective selectivity may rise, making the original prediction less accurate. Students who apply because they were told they had a strong chance could then face a lower probability of admission than expected, potentially damaging trust in both the recommendation system and the admissions process.
“Since each school has a limited number of seats, one student’s choices can affect outcomes for others,” Chiang said. “Recommending the same school to too many students could make the school more competitive than students were led to expect, and that could harm students or decrease trust in the match process.” This makes the design of the system a problem in mechanism design and resource allocation as well as a problem in educational guidance. Recommendations must account for individual fit while also modeling the collective consequences of thousands of students changing their applications at the same time.
The researchers say the broader goal is to help New York City align the formal promise of equal access with the practical reality of how families navigate the system. A centralized admissions process can eliminate some forms of direct selection, but it cannot guarantee equal outcomes if families differ sharply in the information available to them. By connecting behavioral data, predictive modeling and carefully tested communication, the Cornell team is attempting to make the application process more transparent without replacing student preferences or dictating where students should enroll.
For New York City families, the findings offer a potentially powerful message: a student’s current school, neighborhood or social network should not determine the boundaries of their ambitions. For policymakers, the research suggests that improving equity may require more than changing admissions rules. It may also require giving students accurate, personalized and understandable information before they submit applications. If recommendation tools can help students recognize opportunities they would otherwise overlook—while preserving fairness in a system with limited seats—they could become an important new layer of public education infrastructure.
The ultimate test, Garg said, is whether these tools close the distance between the stated goals of the admissions system and students’ real experiences. The researchers aim to ensure that the complexity of applying to high school does not prevent qualified students from reaching programs they are entitled to consider. In a city where a single application can shape a student’s academic trajectory, helping families see the full range of realistic choices may prove as consequential as the algorithm that makes the final match.
Subject of Research: New York City high school admissions, student application behavior, educational undermatching and personalized school recommendations.
Article Title: “Connecting Application Behavior to Undermatching in New York City School Choice”
News Publication Date: August 18, 2026
Web References: https://www.nature.com/articles/s44284-026-00486-0; https://news.cornell.edu/stories/2026/08/research-helps-nyc-students-aim-higher-public-high-school-applications
References: Nature Cities; Cornell University research conducted in partnership with New York City Public Schools.
Keywords: New York City schools, high school admissions, educational equity, undermatching, school choice, personalized recommendations, machine learning, operations research, student mobility, admissions algorithms, education policy, randomized controlled trial
