← Index | ABM Research Notes › Assayed 2023–2025 ABM Notes

ABM Literature Review: Assayed et al. and Related College Admissions Simulation Papers

Source: abm_assayed_2024_notes.md

ABM Literature Review: Assayed et al. and Related College Admissions Simulation Papers

Overview

This document covers the body of agent-based modeling (ABM) research on college admissions, centered on Assayed's work (2023-2025) and the foundational Reardon et al. (2016) paper that most subsequent work builds upon.

1. Assayed & Al-Sayed (2025) — "Student Behaviors in College Admissions: A Survey of Agent-Based Models"

Full citation: Assayed, S. K. & Al-Sayed, S. (2025). Student Behaviors in College Admissions: A Survey of Agent-Based Models. International Journal of Emerging Multidisciplinaries: Computer Science & Artificial Intelligence, 4(1). DOI: 10.54938/ijemdcsai.2025.04.1.385

Also on SSRN: SSRN 5223687 (posted December 31, 2024)

Authors: Suha Khalil Assayed (The British University in Dubai, UAE) and Sana'a Alsayed (Philadelphia University, Jordan)

Indexed in: Crossref, Scopus, Google Scholar

What This Paper Is

A survey/review paper (not an original simulation). It catalogs and classifies ABM approaches used by international universities to study secondary education pathways and student behaviors in college admissions.

Key Themes

Keywords

Agent-Based Model, ABM, Agents, Behavior, Education, Equality, University, Simulation, Survey

Relevance to Our Simulation

This paper confirms the importance of modeling: - Behavioral heterogeneity in application decisions (not just academic stats) - Information asymmetry — students vary in how well they understand the admissions landscape - SES-driven self-selection — low-resource students systematically under-apply - Advisory access as a factor in application portfolio quality

2. Assayed & Maheshwari (2023a) — "Agent-Based Simulation for University Students Admission: Medical Colleges in Jordan Universities"

Full citation: Assayed, S. K. & Maheshwari, P. (2023). Agent-Based Simulation for University Students Admission: Medical Colleges in Jordan Universities. Computer Science & Engineering: An International Journal (CSEIJ), 13(1). February 2023.

Also available: BUID institutional repository, SSRN 4692509, ResearchGate

Model Description

Key Findings

  1. When low-income, high-GPA students are prioritized over same-GPA higher-income students, college reputation becomes determined by cutoff GPA and student preferences rather than purely merit-based ranking.
  2. High-ranking universities are mainly allocated students with high cutoff GPA scores after multiple simulation rotations.
  3. Colleges most interested in attracting new students may not have the highest cutoff — cutoff marks are emergent from college experience across iterations.

Limitations

Relevance to Our Simulation

3. Assayed & Maheshwari (2023b) — "A Review of Agent-based Simulation for University Students Admission"

Full citation: Assayed, S. K. & Maheshwari, P. (2023). A Review of Agent-based Simulation for University Students Admission. Computer Science & Engineering: An International Journal (CSEIJ), 13(2). April 2023.

Also on SSRN: SSRN 4692455

What This Paper Is

A review article classifying several ABMs deployed by different admission offices from international universities. Models are classified by: - Level of educational attainment modeled - University selection behaviors - Main simulation contribution

Key Takeaway

Very few studies have used agent-based models to study college sorting or admissions. The review confirms that Reardon et al. (2016) remains the most influential ABM in this space, with limited follow-up work.

4. Reardon, Kasman, Klasik & Baker (2016) — "Agent-Based Simulation Models of the College Sorting Process"

Full citation: Reardon, S. F., Kasman, M., Klasik, D. & Baker, R. (2016). Agent-Based Simulation Models of the College Sorting Process. Journal of Artificial Societies and Social Simulation (JASSS), 19(1), 8. https://www.jasss.org/19/1/8.html

Affiliation: Stanford CEPA (Center for Education Policy Analysis)

Model Architecture

Agent types: 1. Students — two core attributes: - Resources (composite SES capital) - Caliber (observable academic markers valued by colleges) - Bivariate normal distribution with specified correlation (baseline r = 0.3) 2. Colleges — single quality attribute (= avg enrolled student caliber, recent years weighted more)

Three-stage process per year: 1. Application: Students estimate admission probabilities using perceived college quality and own caliber. Select portfolio to maximize expected utility. Higher-resource students submit more applications (baseline: 4 + 0.5 x resources). 2. Admission: Colleges rank applicants by observable caliber, admit enough to fill seats based on expected yield. 3. Enrollment: Students enroll at highest-perceived-utility institution. College quality updates for next cohort.

Key Parameters

Parameter Baseline Value
Students 8,000
Colleges 40 (150 capacity each)
Caliber distribution Normal, mean 1000, SD ~200 (SAT-scale)
Resource-caliber correlation 0.3
Information reliability 0.7 base + 0.1 x resources (bounded 0.5-0.9)
Application enhancement 0.1 x resources coefficient
Applications per student 4 + 0.5 x resources
Iterations 30 years to equilibrium
Data source ELS:2002 (nationally representative)

Five SES Mechanisms Tested

  1. Resource-caliber correlation (achievement gap)
  2. Application enhancement (coaching, prep, polish)
  3. Information quality (knowledge of colleges and own chances)
  4. Number of applications (portfolio breadth)
  5. Differential valuation of college quality (how much students value selectivity)

Key Findings

Limitations (as noted by authors)

Relevance to Our Simulation

This is the foundational paper for our project. Our simulation extends Reardon in every dimension they identified as a limitation: - We have 30+ student attributes (GPA, SAT, ECs, essays, hooks, demographics, income) - 8 student archetypes with behavioral variation - Multi-round process (ED, EA/REA, EDII, RD, waitlist) - Financial aid and yield modeling - Hook multipliers (athlete, legacy, donor, first-gen) - Real college data (not stylized) - Post-SFFA demographic considerations

5. Reardon, Baker, Kasman, Klasik & Townsend (2018) — Follow-up: SES-Based Affirmative Action Simulation

Full citation: Reardon, S. F., Baker, R. B., Kasman, M., Klasik, D. & Townsend, J. B. (2018). What Levels of Racial Diversity Can Be Achieved with Socioeconomic-Based Affirmative Action? Evidence from a Simulation Model. Journal of Policy Analysis and Management, 37(3), 630-657.

CEPA page: cepa.stanford.edu

Extension of the 2016 Model

Uses the same ABM framework from Reardon 2016 but adds race/ethnicity to examine whether SES-based affirmative action can substitute for race-based affirmative action.

Key Findings

  1. SES alone cannot match race-based policies for producing racial diversity at selective institutions.
  2. Combined SES-based AA + race-targeted recruiting can approach race-based AA outcomes, but is likely more expensive to implement.
  3. Spillover effects: Affirmative action adoption by some colleges reduces diversity at comparable-quality colleges without such policies.
  4. Academic matching: SES+recruiting approach produces fewer academically-overmatched Black and Hispanic students than race-based AA, but enrolled student academic achievement is also lower at schools using both policies.

Relevance to Our Simulation

Directly relevant to post-SFFA modeling. Our simulation could test SES-based priority scenarios similar to what Reardon 2018 explored, using our richer agent attributes and real college data.

6. Allard, Beziau & Gambs (2023/2026) — ReScience Replication

Full citation: Allard, T., Beziau, L. & Gambs, S. (2023). [Re] Simulating Socioeconomic-Based Affirmative Action. ReScience. HAL: hal-04328511

What This Is

A computational reproducibility study reimplementing the Reardon 2016/2018 model in Python. Confirms reproducibility of the original findings and provides an open-source Python implementation of the college sorting ABM.

Relevance

7. Lee, Harvey, Zhou, Garg, Joachims & Kizilcec (2024) — ML-Based Admissions Decision Support

Full citation: Lee, J., Harvey, E., Zhou, J., Garg, N., Joachims, T. & Kizilcec, R. F. (2024). Algorithms for College Admissions Decision Support: Impacts of Policy Change and Inherent Variability. 4th ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO '24), October 29-31, 2024, San Luis Potosi, Mexico. arXiv: 2407.11199

Affiliation: Cornell University

Not an ABM, but Highly Relevant

This paper uses ML ranking algorithms (not ABM) to study how policy changes affect admissions outcomes at a selective US university.

Key Findings

  1. Omitting race from consideration reduces the proportion of URM applicants in top-ranked pool by 62% — without increasing academic merit of that pool.
  2. Race omission reduces diversity more than omitting any other variable (test scores, intended major, etc.).
  3. Inherent arbitrariness: All admission policies contain substantial randomness; removing race increases outcome arbitrariness for most applicants.
  4. Test-optional: Limiting standardized test data further reduces predictive power.

Relevance to Our Simulation

Validates that our model's +-25% randomness term is realistic — even ML-based decision support shows substantial inherent variability. Also confirms that race/ethnicity is among the most consequential variables for admission outcomes, supporting our inclusion of demographic factors.

8. Summary Table: ABM Papers on College Admissions

Paper Year Type Agents Scale Platform US Context?
Reardon et al. 2016 Original ABM Students (2 attrs) + Colleges (1 attr) 8K students, 40 colleges Custom Yes (stylized)
Reardon et al. 2018 Extended ABM + Race/ethnicity Same Custom Yes (stylized)
Assayed & Maheshwari 2023a Original ABM Students (GPA, income) + Colleges Small scale NetLogo 6.3 No (Jordan)
Assayed & Maheshwari 2023b Review N/A (survey) N/A N/A International
Assayed & Al-Sayed 2025 Survey N/A (survey) N/A N/A International
Allard et al. 2023 Replication Same as Reardon Same Python Yes (replication)
Lee et al. 2024 ML ranking N/A (not ABM) Real univ. data ML pipeline Yes (real data)

9. Gaps in the Literature That Our Simulation Fills

The literature review reveals several gaps that our college-sim project addresses:

  1. No existing ABM models the full US admissions cycle (ED/EA/REA/EDII/RD/waitlist) — Reardon uses a single application-admit-enroll round.
  2. No ABM includes hooks (legacy, athlete, donor, first-gen) as explicit agent attributes with calibrated multipliers.
  3. No ABM uses real institutional data — Reardon uses stylized colleges; Assayed uses Jordan medical schools.
  4. No ABM models post-SFFA dynamics with agent-level demographic attributes and income-based SAT offsets.
  5. No ABM models behavioral archetypes — existing work treats student decision-making as homogeneous (utility maximizing) rather than archetype-driven.
  6. No ABM incorporates yield modeling with income-bracket-specific yield rates (a la Chetty/Opportunity Insights).
  7. Financial aid and net cost are absent from all existing ABMs.

Our simulation is, to our knowledge, the most detailed agent-based model of US selective college admissions in the literature.

10. Key References for Further Reading