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

The global rise in homelessness calls for urgent and alternative policy solutions. Non-profits and governmental organizations alert about the many challenges faced by people experiencing homelessness (PEH), which include not only the lack of shelter but also the lack of opportunities for personal development. In this context, the capability approach (CA), which underpins the United Nations Sustainable Development Goals (SDGs), provides a comprehensive framework to assess inequity in terms of real opportunities. This paper explores how the CA can be combined with agent-based modelling and reinforcement learning. The goals are: (1) implementing the CA as a Markov Decision Process (MDP), (2) building on such MDP to develop a rich decision-making model that accounts for more complex motivators of behaviour, such as values and needs, and (3) developing an agent-based simulation framework that allows to assess alternative policies aiming to expand or restore people’s capabilities. The framework is developed in a real case study of health inequity and homelessness, working in collaboration with stakeholders, non-profits and domain experts. The ultimate goal of the project is to develop a novel agent-based simulation framework, rooted in the CA, which can be replicated in a diversity of social contexts to assess policies in a non-invasive way.

💡 Research Summary

The paper tackles the growing global homelessness crisis by proposing a novel computational framework that merges Amartya Sen’s Capability Approach (CA) with Markov Decision Processes (MDP), reinforcement learning, and agent‑based modeling (ABM). The authors argue that traditional resource‑oriented policies overlook the essential question of what individuals are actually able to do and be; CA shifts the focus to real opportunities—capabilities—derived from resources, conversion factors, and personal choice.

First, the authors formalize the CA’s core constructs—resources, conversion factors (personal, social, environmental), capabilities (central and specific), choice factors (values and needs), and functionings—into the components of an MDP. States encode an individual’s resource endowment and conversion‑factor profile; actions correspond to specific capabilities (e.g., riding a bike, seeking shelter); transition probabilities capture how conversion factors probabilistically enable or block a capability; rewards are derived from the satisfaction of values (long‑term importance) and needs (short‑term urgency). By treating values as drivers of long‑term rewards and needs as drivers of immediate rewards, the model captures the tension between urgent survival actions and value‑guided life choices.

Second, the paper extends the basic MDP with a multi‑objective reinforcement‑learning layer that allows agents to balance competing objectives. The authors propose a hierarchy where values can override needs, reflecting the hypothesis that people may sacrifice immediate comfort for higher‑order aspirations (e.g., security, autonomy). This design choice is justified with references to Schwartz’s value theory and Maslow’s hierarchy, yet the authors acknowledge that the interaction is simplified and plan to explore more nuanced dynamics in future work.

Third, the authors embed the enriched decision‑making model within an ABM that simulates a heterogeneous population of stakeholders: people experiencing homelessness (PEH), social workers, non‑profit organizations, and municipal authorities. The ABM enables both top‑down policy experiments (e.g., new housing legislation, cash transfers) and bottom‑up processes (e.g., diffusion of social norms, resource distribution). By tracking each agent’s central capabilities (such as bodily integrity, health, affiliation) and specific capabilities (e.g., mobility via a bike), the simulation can evaluate how policies affect the overall capability distribution and identify interventions that most effectively restore or expand central capabilities, aligning with the UN Sustainable Development Goals.

The framework is illustrated through a concrete case study in Barcelona, Spain, where the authors collaborate with local NGOs (Arrels Fundació, Caritas, Salut Sense Llar) and academic experts. They present a “bike example” to demonstrate how conversion factors (disability, legal restrictions, weather) modulate the probability that a resource (a bike) translates into the capability of free movement, and how personal values influence the choice to ride despite adverse conditions.

While the conceptual integration is innovative, several limitations are evident. The paper provides limited detail on data collection, parameter estimation, and validation. The probabilities governing conversion factors and the weighting of values versus needs are not empirically calibrated, raising concerns about reproducibility and external validity. Moreover, the model’s performance is not benchmarked against real‑world policy outcomes, leaving the practical predictive power untested. The simplification that values always dominate needs may not hold in many homelessness contexts where survival pressures are extreme.

In conclusion, the study offers a compelling interdisciplinary bridge between normative welfare economics and computational social science. By operationalizing the Capability Approach as an MDP and embedding it in a multi‑agent reinforcement‑learning environment, the authors provide a flexible platform for non‑invasive policy experimentation on homelessness and related inequities. Future work should focus on rigorous empirical calibration, sensitivity analysis, and validation against longitudinal policy data to transform this promising prototype into a robust decision‑support tool for policymakers worldwide.