Beyond the Cutoff: Remodeling Persistence in Elite Admissions

Executive Summary

Secondary learning institutions in the United States operate under three competing models for identifying academically elite students, none of which has produced reliable indicators predicting whether those students will persist after admission. The SHSAT in New York operates with a single-day crossing of a psychometric threshold at which Stuyvesant’s 2026 cutoff of 561 out of 800 was set by employing approximately 26,000 candidates in a competitive context for its 800 spots. On the other hand, the 2020 reformation of Thomas Jefferson High School has succeeded in substituting testing for geographic quotas and contextual factors, which do not render race as a consideration when implemented in the Coalition for TJ v. Fairfax County School Board case, which was upheld in 2024 by the US Supreme Court, declining to conduct its review. Finally, regional portfolio approaches, such as NCSSM, reliant on congressional district structure or holistic university review, which were implemented in the aftermath of the case of SFFA where race as a criterion was prohibited, also encounter the same flaw in that they cannot predict if the student would stay beyond the admission stage.

This essay claims that the solution lies not in developing another admissions approach but in adopting a different paradigm of thinking. The suggested model is a two-step statistical analysis in which grades and test scores are normalized with the help of cohort measurements conducted locally first to account for grade inflation, and then survival analysis techniques are utilized to evaluate the dropout risk during the first stage for the STEM gateway class. When used as a diagnostic tool rather than a ratifying mechanism, the model can connect the process of admission with staying in school as well as providing evidence for K-12 and higher education governance.

The Gatekeepers of Elite American Mobility

Every August, three admissions systems that seemingly have nothing in common yield the same outcome. Stuyvesant High School in Manhattan publishes its cutoff for the Specialized High Schools Admissions Test, which indicates that approximately 561 out of 800 candidates have passed the test out of the 26,000 who took it. Thomas Jefferson High School for Science and Technology in Virginia unveils its admission list according to updated rules that changed its previous system of evaluation of candidates with a single test to one that takes into consideration quotas stipulated by the Fourth Circuit in Coalition for TJ v. Fairfax County School Board. North Carolina School of Science and Mathematics also announces the new cohort using a model that involves geographical quotas with the use of a math exam and a holistic review of applicants.

Thus, there are three institutions with different systems, yet producing the same result.

The public debate surrounding these processes has shifted into a binary argument. Standardized testing has become both “objective merit” and “structural bias,” while holistic evaluation has transformed into either “context-sensitive equity” or “unaccountable subjectivity.” Both concepts conflate a legal question with a structural question. The Supreme Court’s decision in the case of Students for Fair Admissions v. Harvard in June of 2023 forced elite universities to change their evaluation models. The decision also appears in the January 2024 refusal to grant certiorari in Coalition for TJ, thus extending the Fourth Circuit’s ruling for K-12 selection. With that, both components face less legal security and more analytical exposure than at any point within the last four decades.

This essay analyzes the argument that no single mechanism is currently in use to achieve what selective institutions claim to be choosing: the ability to persist long-term in academic programs. The purely psychometric channel works to predetermine an applicant’s signal value in one day. The geographic quota model, in turn, has been selecting subjects for access. The holistic evaluation attempts to interpret a candidate’s application in a particular way. What none of these models consider is the calculation of applicants’ probability of graduation and entering the labor market with the accreditation promised by an educational establishment. The essay presents different types of the mechanisms mentioned and suggests a diagnosis based on the idea of survival analysis.

Archetype I: The Pure Psychometric Single Gate (SHSAT)

New York City has a system for determining its high school admissions, consisting of a single test, the Specialized High Schools Admissions Test. The use of this test was formalized by the Hecht-Calandra Act of 1971 and must be carried out by law in the case of Stuyvesant, Bronx Science, and Brooklyn Tech, although the utilization of something else instead of the mentioned test could still be allowed by a specific piece of legislation. For the Class of 2030, about 26,000 students had taken the exam. Stuyvesant’s cutoff score was 561 out of 800, which allowed about 800 students to be admitted to the school, and it meant that the acceptance rate stood at about 3%, so one could say that such a school is in line with some Ivy League institutions rather than with several ordinary public schools.

This very architectural model of admission has its benefits. The use of a single exam makes it genuinely unlikely that it gets corrupted, while other selection methods require an interview, an essay, and even a portfolio. It does not involve a possibility for a family with some connections, a professional coach, and a reader with personal preference exerting their influence. The use of the exam is also good in that it does not require any specific criteria for establishing the criteria for passing the test.

Its weaknesses are just as real. The most thorough empirical research into what takes place with an SHSAT admission comes from Will Dobbie and Roland Fryer’s study conducted in the American Economic Journal: Applied Economics. The authors use regression discontinuity to compare students admitted to the school through an SHSAT score of just above or just below the cut-off. Their findings, confirmed in parallel studies conducted by Josh Angrist, Parag Pathak, and Atila Abdulkadiroğlu, reveal that admission into an elite school has no causal connection with college attendance, college graduation, or later job market outcomes. The peer achievement effect on which admission officers rely does not work.

Such a finding is fatal for the idea of the psychometric explanation in its strongest form. If the exam measured the latent academic ability of students and translated it into practical persistence, then the students just above the cut-off would obtain better results than their peers in terms of relevant outcomes. That is not the case. What SHSAT measures is the outcome of a one-day test subject to variations considering the test date and prior preparations. It shows who did better on the test conducted on one Saturday rather than who will pass AP Physics C in eleventh grade.

This gap between a one-day outcome and a multi-year outcome is what has driven Fairfax County to stop using a single examination for admission to Thomas Jefferson School.

Archetype II: Race-Neutral Geographic Quotas (TJHSST)

In December 2020, the Fairfax County School Board executed the new admission policy for Thomas Jefferson High School for Science and Technology that removed the use of the standardized entrance assessment and application fee. This, in turn, was replaced by the distribution method that would allot seats to the best-performing 1.5 percent of students in the eighth grade in every middle school in the region of Fairfax. The Washington Post reported the process of the deliberation and previously published the entire range of options that were offered by the superintendent to solve the situation. The aim of this step was simple: to expand the geographic and demographic boundaries of admission to this high school since its entering students seemed to be represented mainly by those from the few middle schools in the area.

The parents’ association called the Coalition for TJ filed the lawsuit against the Equal Protection Clause, claiming that the new admissions system was unfair to the Asian American applicants since they dominated the admissions process before and made up the majority of those who were successful in passing the exam. Initially, the district court was in solidarity with Coalition, but the case was reconsidered by the Fourth Circuit, which overturned the ruling made by the district court in Coalition for TJ v. Fairfax County School Board, claiming that no violation of the principle of equal justice occurred because the new policy was neutral in terms of race. Following an appeal from NBC News, the Fourth Circuit ruling turned out to be the first significant ruling in the field of K-12 admissions after the SFFA case was resolved.

Justice Samuel Alito and Justice Clarence Thomas issued a dissent against the non-acceptance of the case. Their argument suggests the “proxy discrimination” thesis in a manner that now seems decisive. This theory is being used by a variety of organizations planning to take similar legal action to the courts. Regardless of opinions about its merits, this dissent has made the arguments on the issues surrounding quota policies much more complicated than indicated by the decision to reject the case. A different court could come up with different solutions.

Structural differences are, however, quite distinct from constitutional ones. Geographic quotas may be successful in terms of accessing more individuals. At the same time, they reduce the psycho-signal density of the students admitted, since the top 1.5% of a poor school cannot be compared to the top 1.5% of a good school. Whether the mentioned trade-off brings about positive or negative results in terms of persistence is a question that is neither clarified by the Fourth Circuit nor by the dissenting opinion. However, the importance of this very issue is one of the most vital objectives of the third type of study under consideration.

The Higher-Ed Mirror

The North Carolina School of Science and Mathematics has introduced a model that is not in use in either Stuyvesant or TJHSST. Their admissions criteria are structured based on the congressional district of a student’s residence, require an assessment in mathematics, and involve a holistic review process of a student’s records—which includes transcripts, personal essays, recommendations from teachers, and proof of involvement in STEM activities—which adds up to the regional-quota floor criterion that they have already laid down. WUNC investigated legislation which has changed the nature of their applicant pool based on redistricting, but only to the extent that the overall effect of the geographic-quota design is dependent on the districts that make up the design.

The NCSSM model is a simpler version of an extremely selective model adopted by most elite universities after SFFA. The June 2023 Supreme Court decision in Students for Fair Admissions v. Harvard has banned the use of race in admissions but also allowed students to provide their racial background in the essays.

As a result, the holistic review has become the most widely used method of assessment in both K-12 and higher education settings. However, this review process is not as reliable as many believe. Those who assess the quality of an application process for applicants, the one who helps applicants with their recommendation letters, and those who are involved in the process of establishing the so-called “demonstrated intellectual curiosity” of the candidates create different rankings and assessments of applicants even while evaluating the same file/dossier. Similar issues are widely discussed in various studies on admissions.

Finally, we come to the common thing about the three described types of assessments: none of these assessments provides us with information about what it is trying to predict or estimate. The SHSAT is aimed at examining its candidates’ performance for only one day, and many researchers believe that the test cannot predict the results of the candidates’ performance in the future. The use of geographical quota helps applicants to get access to certain colleges better, but it does not measure the results of academic persistence among the admitted candidates. The application process assesses applicants’ ability to make judgments about the candidate. However, the issue is that the three methods do not use the proper quantitative methodology and thus do not assess the likelihood of a candidate persisting successfully.

Statistical Modeling for Persistence

A well-formulated admissions analysis would achieve two goals that the current systems do not achieve. First, it would consider the grade inflation dynamic, which makes raw GPA increasingly incomparable across different high schools, and secondly, it would make use of persistence probability directly via the use of the tools of survival analysis used by biostatisticians for clinical results.

Stage 1: The Rigor-Persistence Index

Grade inflation cannot be dismissed as simple rhetoric and is now corroborated by measurements. As elaborated in the National Bureau of Economic Research working paper “Easy A’s, Less Pay,” graduating from colleges that tend to inflate grades negatively affects the graduates’ income. This implies that bad grades mean that the differences between degrees are less substantial. Similarly, the ACT Research Report examines grade inflation and finds the same level of impact at the level of high schools.

A Rigor-Persistence Index (RPI) is developed to help understand in more detail the data about each applicant’s GPA, taking into consideration the average GPA from the applicant’s high school. Thus, having an unweighted GPA of 3.9 in a high school where the average GPA is 3.7 means a less substantial outcome than having the same GPA in a school with an average GPA of 3.2. The purpose of RPI is to provide a single numerical index that will enhance the school admission process.

Stage 2: Cox Proportional Hazards for Persistence Modeling

The second layer estimates the probability of persistence directly. The Cox proportional hazards regression technique, which was initially developed for medical survival analysis, enables us to consider discontinuance in a given academic program as a time-to-event process. RPI, standardized test scores, coursework difficulty, and demographic variables comprise the set of independent variables, and as a result, we obtain estimated hazard ratios for the probability that candidates persist in STEM gateway courses with the highest attrition rate. Weston and Seymour’s study entitled “Weed-out courses” and an NCBI review of STEM students’ persistence confirm that most STEM attrition occurs within the first three semesters and can be predicted by pre-enrollment data.

To illustrate this, we refer to the case study of Opportunity Insights in terms of Ivy League universities’ admissions. In their analysis of scores on standardized tests and in their broader NBER paper dedicated to diversification of leadership, Raj Chetty, David Deming, and John Friedman conclude that SAT and ACT results tend to be an excellent predictor of GPA at Ivy League universities as compared to high school GPA or subjective assessment of students’ capabilities, but only concerning students’ background information. The authors of the paper and The Chronicle of Higher Education’s commentary similarly present their findings: it is better to use standardized tests for contextual analysis than abandon them completely.

Governance: Diagnostic, Not Automation

The factor that limits governance crucially concerns how to apply the results obtained from the survival analysis process; the outputs generated should serve only as input information for further human examination rather than an automated resource that helps in rejecting candidates. If a candidate has a low persistence probability, it means that further examination of their application should be performed instead of engaging in the automatic rejection process. When viewed from this perspective, the architecture of RPI, along with the Cox model is able to reduce the gap in prediction brought by the three existing models.

Strategic Leadership and Policy Takeaways

The connection between K-12 and higher education selection discussions is closer than most educational administrators would recognize. Stuyvesant, TJHSST, and NCSSM produce the same candidate pipelines that are going to be used by Ivy-Plus admission officers 4 years later, and they employ frameworks that share structural DNA. The same predictive validity questions can be utilized in both phases, and the American Psychological Association’s amicus brief in SFFA explained the science of predictive validity thoroughly.

There are three conclusions. First, standardized testing ought to be upheld and reorganized rather than abolished. The Opportunity Insights and NBER data are enough to fully reject both the “abolish the SAT” and the “make the SAT decisive” ideas.

Second, geographic quota methods like those used by TJHSST provide real access benefits and are tradition-resilient, yet they do not help solve the problem of persistence prediction, and they should be combined with a persistence model. Third, the retention system used for the admitted students is as important as the admission filter. Weed-out courses, an unadjusted grading system, and a lack of an early warning system will create the same attrition processes irrespective of whatever admissions model was utilized before.

The main point is that after the cutoff—after the one number, one quota, and one person’s judgment—lies the place where institutional management really influences results. The gatekeepers have been discussing the gate, while the persistent things occur on the other side of the gate.