Adaptive Merit Framework: Merit-Anchored Fairness via SES Correction

TL;DR

<AMF addresses the merit-equity tension in admissions by introducing a non-displacing, merit-anchored framework with dynamic thresholds and direct SES measurement.> The method uses a correction rule $C_i = \alpha(0.5 - S_i)$ on a percentile-normalized SES index and maintains a fixed merit threshold defined by the top-k raw scores, enabling additive opportunity expansion without displacing high-merit regular admits. The empirical validation on PISA 2022 Korea data shows only a small number of additional admits (4, 6, 9 for $\alpha=5,10,15$), all surpassing the merit threshold and originating from the bottom half of SES, highlighting targeted recognition of suppressed potential. The paper further provides a design blueprint for unified admissions architectures, discusses implementation pathways (Hybrid and Expansion), and outlines robustness, policy implications, and future research directions. Overall, AMF demonstrates that fairness can be operationalized as a transparent, threshold-based design that expands access while preserving academic standards.>

Abstract

College admissions systems worldwide continue to face a structural tension between meritocracy and equity. Conventional fairness interventions--affirmative action, categorical quotas, and proxy-based targeting--often rely on coarse indicators (e.g., race or region), operate within fixed quotas that induce zero-sum trade-offs, and lack transparent decision rules. This paper introduces the Adaptive Merit Framework (AMF), a policy-engineered mechanism that recognizes latent potential while preserving merit-based thresholds. AMF integrates three components: (1) a merit-anchored architecture in which conditional admits must exceed the same threshold as regular admits, (2) a dynamic threshold anchored to the raw score of the last regular admit, and (3) direct, continuous SES measurement verified through administrative data. Empirical validation using the full PISA 2022 Korea dataset (N=6,377) shows that AMF identifies 4, 6, and 9 additional admits under alpha = 5, 10, and 15 respectively (0.06-0.14% of cohort). Population-weighted estimates using OECD sampling weights suggest that the real-world scale of conditional admits is modestly larger than the raw sample counts, yielding approximately 491, 603, and 760 additional admits under alpha = 5, 10, and 15. All conditional admits exceed the merit threshold by 0.16 to 6.14 points, indicating that AMF recognizes suppressed performance rather than relaxing standards. Beyond SES-based corrections, AMF provides a design template for unified admissions architectures that replace fragmented equity tracks and support multi-dimensional evaluation frameworks.

Figures (5)

• Figure 1: AMF Mechanism Overview. Four-step process: (1) Input (M, S), (2) Correction C = $\alpha(0.5-S)$, (3) Adjusted score $M^* = M+C$, (4) Selection by $M^* \geq T$. Regular admits ($M \geq T$) are never displaced.
• Figure 2: Empirical Results from PISA 2022 Korea ($N=6,377$). (a) Additional admits scale linearly with $\alpha (R^2 = 0.987).$ (b) 100% originate from bottom 50% SES under baseline conditions (Q1-Q2). (c) All exceed threshold T=666.62 by 0.16-6.14 points.
• Figure 3: Robustness to Perturbations. (a) Linear scaling with $\alpha$. (b) Stability under 5-10% SES noise. (c) Consistent targeting across threshold percentiles (5%, 10%, 15%).
• Figure D.1: Comparison of sample counts and population-weighted estimates for conditional admits, along with SES and merit gap distributions.
• Figure E.1: DBN Long-term Trajectories