💻 computer science

Agent-based Modeling meets the Capability Approach for Human Development: Simulating Homelessness Policy-making

This paper proposes a novel agent-based simulation framework that integrates the Capability Approach with reinforcement learning to model and evaluate homelessness policies aimed at expanding individuals' real opportunities for personal development.

Original authors: Alba Aguilera, Nardine Osman, Georgina Curto

Published 2026-01-29

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Alba Aguilera, Nardine Osman, Georgina Curto

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to help a group of people who are struggling to survive on the streets. Traditionally, policymakers might look at a map and say, "We need to give them more blankets and food." But this paper argues that just handing out resources isn't enough. It's like giving someone a bicycle but forgetting to check if they have legs to pedal it, if the roads are blocked, or if they are afraid to ride because of a sign that says "No Bikes Allowed."

This paper proposes a new way to design policies using a "digital sandbox" (a computer simulation) to see what really helps people get back on their feet. Here is how they are doing it, broken down into simple concepts:

1. The Core Idea: It's Not About the Gift, It's About the Ability

The authors use a concept called the Capability Approach. Think of it like this:

Resources are the tools you have (a bike, a job offer, a shelter).
Capabilities are what you can actually do with those tools.

If you give a bike to someone who doesn't know how to ride, or if the law forbids them from riding, they don't really have the "capability" to travel. The paper argues that homelessness isn't just a lack of money; it's a lack of opportunities to live a dignified life. They want to measure whether a policy actually gives people the ability to do things like stay healthy, find work, or feel safe, rather than just counting how many blankets were distributed.

2. The Tool: A "Digital Sandbox" for People

To test policies without hurting real people, the team is building a Agent-Based Model.

Imagine a giant video game where every character (agent) is a real person experiencing homelessness, a social worker, or a non-profit organization.
These characters aren't just following a script; they have their own personalities, values, and urgent needs.
The computer runs the simulation to see what happens when you change the rules (the policies). For example: "What happens if we change the law so that people without ID cards can see a doctor?"

3. The Engine: How the Characters Decide

The paper introduces a clever way to make these digital characters think like real humans using a math framework called a Markov Decision Process (MDP).

The Map (States): Where the person is right now (e.g., sick, homeless, registered citizen).
The Moves (Actions): What they can try to do (e.g., go to a clinic, ask for help).
The Roadblocks (Conversion Factors): Sometimes a move is blocked. Maybe the clinic is too far (environment), or the law says "No" (social factor).
The Motivation (Choice Factors): This is the special part. The characters don't just act randomly. They are driven by two things:
1. Urgency (Needs): "I am in pain right now." (Short-term)
2. Values (Long-term): "I want to be a good father" or "I value my dignity." (Long-term)

The simulation tries to figure out how these characters balance immediate pain with their long-term dreams. For instance, a person might skip a free meal (ignoring an urgent need) because going there would violate their personal values or safety.

4. The Real-World Test: Health in Barcelona

The team is testing this on a specific problem in Barcelona, Spain: Why are homeless people dying younger and getting sicker?

The Problem: Many homeless people are "non-registered" (they don't have official ID papers). Because of this, the law blocks them from seeing a primary care doctor. They only get help when they are in a life-or-death emergency at a hospital.
The Simulation: They are creating a digital version of this neighborhood. They will create thousands of "digital homeless people" with different backgrounds and health issues.
The Experiment: They will run the simulation twice:
1. Current Reality: The law blocks non-registered people from doctors.
2. New Policy: A proposed law that lets non-registered people see doctors immediately.

They will watch to see if the new policy actually restores the "capability" of these people to stay healthy, or if other barriers (like fear or lack of trust) still stop them.

5. Why This Matters

The goal isn't just to make a cool computer program. It's to create a non-invasive testing ground.

Instead of trying a new law on real people and hoping it works (which could fail and hurt them), they can "break" the policy in the computer first.
They can see if the policy actually expands people's freedom to live with dignity, or if it accidentally creates new problems.

In short: This paper is about building a "flight simulator" for social policy. Instead of guessing if a new law will help homeless people, they are using math and computer science to simulate the lives of these people, ensuring that the policies they design actually give them the power to live good, healthy lives, not just the stuff to survive.

Technical Summary: Agent-based Modeling meets the Capability Approach for Human Development

Problem Statement
The paper addresses the urgent need for alternative policy solutions to the rising global scale of homelessness. Traditional development models often focus on resource distribution (e.g., utilitarian frameworks), whereas the authors argue that homelessness and poverty should be understood as severe forms of capability deprivation. This perspective, rooted in the Capability Approach (CA), posits that individuals are unable to access "central capabilities" (essential human entitlements for a dignified life, such as health, bodily integrity, and control over one's environment) due to a lack of resources and adverse conversion factors (personal, social, and environmental).

Current policy-making lacks tools to non-invasively assess how specific interventions (legal norms, resource allocation) impact these capabilities. Existing agent-based modeling (ABM) literature rarely utilizes the CA as a conceptual framework, and those that do often oversimplify human motivation by relying solely on hierarchical needs (e.g., Maslow) rather than the complex interplay of values and needs inherent in the CA.

Methodology
The authors propose a novel framework that operationalizes the Capability Approach using Agent-Based Modeling (ABM) and Reinforcement Learning (specifically Markov Decision Processes - MDPs). The methodology involves three core technical steps:

Formalizing CA as an MDP: The authors map CA concepts to MDP elements to enable computational simulation:
- States ( $s$ ): Represent the agent's current situation, including personal characteristics, available resources, and individual need/value preferences.
- Actions ( $a$ ): Differentiated into possible actions (specific capabilities), impossible actions (deprived capabilities), and realised actions (functionings).
- Transition Probabilities ( $P(s'|s, a)$ ): Represent Conversion Factors. These determine the probability of an agent moving from a resource state to a capability state. Social conversion factors are modeled as legal and social norms that may render certain actions impossible (e.g., a non-registered citizen cannot access primary healthcare).
- Rewards: Represent Choice Factors. The framework distinguishes between:
  - Short-term rewards: Driven by need urgencies (e.g., immediate pain relief).
  - Long-term rewards: Driven by value preferences (e.g., long-term health or security).
- Policy ( $\pi$ ): The agent's decision rule aggregates short-term and long-term expected rewards ( $Q_s$ and $Q_l$ ) to select actions, allowing for the modeling of complex behaviors where values may override immediate needs.
Enhanced Decision-Making Model: Unlike prior works relying on rigid need hierarchies, this model integrates human values (based on Schwartz's value framework) and needs as distinct motivators. The system allows for dynamic updates where realized actions can alter an agent's state (resources/conversion factors) and their internal choice factors (e.g., an accident making an agent more risk-averse).
Simulation Framework: The MDP serves as the behavioral engine for agents representing diverse stakeholders (People Experiencing Homelessness - PEH, social workers, non-profits). The simulation includes a physical environment (resources) and a regulatory environment (legal norms). It is designed to be populated with synthetic data derived from real-world anonymized data and stakeholder interviews.

Key Contributions

Computational Operationalization of CA: The paper provides a concrete mapping of the abstract Capability Approach (Resources, Conversion Factors, Capabilities, Choice Factors, Functionings) into the technical language of MDPs and ABM.
Integration of Values and Needs: It introduces a decision-making model where "values" (importance/long-term goals) and "needs" (urgency/short-term goals) act as distinct choice factors, allowing for a more nuanced representation of human behavior than previous CA-based simulations.
Policy Assessment Tool: The framework is designed to assess the impact of policies not just on resource distribution, but on the expansion or restoration of central capabilities. It evaluates equity by analyzing the set of possible vs. impossible actions available to agents under different regulatory scenarios.
Interdisciplinary Case Study Design: The authors outline a specific application in Barcelona's Raval neighborhood, focusing on health inequity. They detail how to construct agent profiles using real data from non-profits (e.g., Arrels Fundació) and domain experts to simulate the effects of proposed laws on access to primary healthcare for registered vs. non-registered citizens.

Results and Current Status
As this is a proposal and methodology paper, it does not present final simulation results or empirical validation data. Instead, it presents:

A Conceptual Prototype: A simplified MDP example (Figure 2) demonstrating how a legal norm (registration status) acts as a conversion factor, rendering the action "receive medical attention" impossible for non-registered citizens, thereby depriving them of the central capability of "bodily health."
Implementation Roadmap: A plan to scale from this simple example to a complex simulation involving multiple central capabilities, utilizing synthetic population data and iterative feedback from stakeholders.
Evaluation Metrics: A defined set of indicators for future evaluation, including housing states (ETHOS terminology), health states, registration status, economic expenses, and institutional discrimination.

Significance
The paper claims significance in providing a non-invasive, replicable simulation framework rooted in the Capability Approach. Its primary goal is to support the evaluation of policies aimed at expanding human capabilities, aligning with the UN Sustainable Development Goals (SDGs). By shifting the focus from resource allocation to the opportunity to achieve a dignified life, the framework aims to identify policies that effectively restore central capabilities for marginalized groups. The authors emphasize that this approach requires close interdisciplinary collaboration to define context-specific elements (conversion factors, values) and aims to address ethical challenges such as "aporophobia" (rejection of the poor) and institutional discrimination. The work is positioned as a step toward a robust, computationally grounded tool for social justice and human development policy-making.