🤖 AI

Can Poverty Be Reduced by Acting on Discrimination? An Agent-based Model for Policy Making

This paper introduces the Aporophobia Agent-Based Model (AABM), which uses real-world data from Barcelona to computationally demonstrate that discrimination against the poor (aporophobia) exacerbates wealth inequality, thereby advocating for policy shifts that address poverty as a societal issue rooted in discrimination rather than solely through traditional redistributive measures.

Original authors: Alba Aguilera, Nieves Montes, Georgina Curto, Carles Sierra, Nardine Osman

Published 2026-02-06

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Alba Aguilera, Nieves Montes, Georgina Curto, Carles Sierra, Nardine Osman

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 a giant, digital city where thousands of tiny, computer-generated people live their lives. This isn't a video game about building cities; it's a "crystal ball" used by researchers to test how laws affect real people's wallets before those laws are ever passed in the real world.

Here is the story of that paper, broken down into simple concepts.

The Big Problem: The "Poverty Trap"

For a long time, the world has tried to fix poverty by simply giving people money or sharing wealth (like a parent splitting a pie). But lately, this isn't working as fast as it used to. The researchers noticed something else might be blocking the path: Aporophobia.

Think of Aporophobia as "poverty-phobia." It's the fear, dislike, or rejection of poor people. It's like if a neighborhood decided that anyone with empty pockets was "bad" or "lazy," and the rules of the town were written to punish them rather than help them. The paper asks a simple question: If we change the rules to stop punishing the poor, will poverty go down?

The Tool: A Digital City Simulator

To answer this, the team built a Agent-Based Model (ABM).

The Agents: These are the digital people. They aren't random; they are built using real data from Barcelona, Spain. They have jobs, they are homeless, they have families, and they have bills.
The Brain: These digital people have a "brain" based on Maslow's Hierarchy of Needs. Imagine a pyramid. At the bottom, you need food and a bed. If you don't have those, you can't worry about the top of the pyramid (like making art or feeling respected). The computer people act to fill their most urgent "empty cup" first. If they are hungry, they go to the grocery store. If they have no home, they look for shelter.
The Rules (The Laws): The researchers programmed the city with real laws from Barcelona. Some laws are kind (Non-Aporophobic), like giving unemployment benefits or free housing. Others are mean (Aporophobic), like fining people for sleeping on the street or throwing them in jail if they can't pay.

The Experiment: Running the Simulation

The researchers ran the simulation 64 different times. In some runs, they turned on the "kind" laws. In others, they turned on the "mean" laws. In some, they turned them all on at once.

They measured the result using a Gini Coefficient.

Think of this as a "Fairness Score."
A score of 0 means everyone has exactly the same amount of money (perfect fairness).
A score of 1 means one person has all the money and everyone else has nothing (total unfairness).

What They Found

The results were like a lightbulb moment for policy-making:

The "Mean" Laws Make Things Worse: When the simulation included laws that punished the poor (like fining homeless people for sleeping on the street), the "Fairness Score" got much worse. The gap between rich and poor grew wider. It was like trying to put out a fire by throwing gasoline on it.
The "Kind" Laws Help: When they used laws that provided support (like unemployment checks or free housing), the "Fairness Score" improved. The gap between rich and poor shrank.
The Specific Impact: One specific "kind" law (giving unemployment benefits) was the most effective at stopping people from hitting rock bottom. Conversely, a law that turned unpaid fines into jail time made the situation for the poorest people significantly worse, pushing more of them into bankruptcy.

The Conclusion

The paper argues that poverty isn't just a problem of "not having enough money." It's also a problem of how society treats the people who don't have money.

If a city's laws are written with fear or hatred toward the poor (aporophobia), those laws actually make poverty harder to solve. But if the laws are written to support and protect the vulnerable, the whole society becomes more equal.

In short: You can't fix a leaky boat just by bailing out water (redistributing wealth); you also have to stop punching holes in the bottom of the boat (discriminatory laws). This computer model proves that stopping the "punching" helps the boat stay afloat.

Technical Summary: Can Poverty Be Reduced by Acting on Discrimination? An Agent-based Model for Policy Making

Problem Statement
The paper addresses the deceleration in global poverty reduction rates, suggesting that traditional redistributive approaches may be losing effectiveness. The authors posit that discrimination against the poor, termed "aporophobia," acts as a significant impediment to poverty mitigation. While literature suggests that aporophobia makes poverty-mitigating policies unpopular and difficult to implement, there is a lack of empirical evidence demonstrating a causal link between aporophobia and increased poverty or wealth inequality. The core problem is the absence of computational tools to evaluate how discriminatory legal frameworks influence wealth distribution before policies are enacted.

Methodology
The authors propose the Aporophobia Agent-Based Model (AABM), a multi-agent simulation designed to computationally evidence the correlation between aporophobia and poverty. The model is built using the Python library Mesa and integrates three distinct components:

Agent Profiles: Agents represent citizens with unique socio-economic and demographic profiles (age, gender, status, wealth, income, location) instantiated from real-world data for four districts in Barcelona (Sarrià-Sant Gervasi, Gràcia, Les Corts, and Eixample).
Autonomous Decision-Making (Needs-Based Model): Agents possess "free will" and make decisions based on a modified version of Dignum et al.'s needs-based model, inspired by Maslow's hierarchy.
- Needs: Agents manage "deficiency needs" (physiological, safety, belonging, esteem) which decay over time. Growth needs (self-actualization) are excluded to focus on basic survival.
- Deliberation: At each time step (starting at 5 PM), agents calculate a score for available actions based on the urgency of unmet needs, the importance of need categories, and the expected satisfaction an action provides. They select the action with the highest score.
- Constraints: Actions are constrained by the agent's status (e.g., only employed agents can work), financial capacity, and physical location.
Regulatory Environment: The model incorporates a set of legal norms (policies) as preconditions and post-conditions. These norms are tagged as either Aporophobic (Apo) or Non-Aporophobic (Non-Apo) based on collaboration with specialized NGOs.
- Non-Apo examples: Unemployment benefits, minimal vital income, and provision of dignified living space for the homeless.
- Apo examples: Fines for sleeping on the street or stealing food, commutation of fines to imprisonment, and eviction due to bankruptcy.
- The simulation assumes perfect adherence to these rules by agents and flawless implementation by the system, abstracting away the legislative process to focus on policy effects.

Experimental Setup
The study utilizes a "proof of concept" scenario with $N=100$ agents simulated over $T=2880$ timesteps (equivalent to 4 months). The researchers tested the power set of six specific norms ( $2^6 = 64$ combinations). For each combination, 10 runs were executed to account for stochastic variability, resulting in a sample of 1,000 agents per configuration. The primary metric for evaluation is the Gini coefficient, which quantifies wealth inequality, alongside the proportion of agents ending in a "bankruptcy" state (wealth $\le$ 0).

Key Results
The simulation results provide computational evidence linking aporophobic policies to increased wealth inequality:

Individual Norm Effects: Most individual norms increased the Gini coefficient compared to a baseline with no rules. Only Norm 1 (unemployment benefits) decreased inequality. Norm 5 (commuting fines to imprisonment) caused the most significant increase in inequality (+18.3%).
Collective Effects (Apo vs. Non-Apo):
- Implementing the Non-Apo subset (Norms 1, 2, 3) reduced the Gini coefficient by 18.8%, leading to a more egalitarian distribution. It significantly reduced the proportion of agents in bankruptcy (from ~24% to ~7%).
- Implementing the Apo subset (Norms 4, 5, 6) increased the Gini coefficient by 20.8%, leading to higher inequality and a sharp rise in homelessness and bankruptcy (39.3% of agents).
Correlation with Aporophobia: A clear positive correlation was observed between the proportion of aporophobic norms in a policy mix and the resulting Gini coefficient. The lowest Gini (0.405) was achieved with the Non-Apo subset, while the highest (0.600) resulted from a mix containing 60% aporophobic norms. Norm 1 was identified as a critical driver for egalitarian outcomes.

Significance and Claims
The paper claims to be the first to provide empirical, computational evidence of the relationship between aporophobia and the increase of wealth inequality levels. Its significance lies in:

Novelty: It introduces a new perspective on poverty reduction by treating it as a societal problem influenced by discriminatory beliefs encoded in law, rather than solely a problem of resource redistribution.
Policy Tool: It demonstrates the utility of Agent-Based Modeling as a non-invasive tool for predicting the societal effects of public policies and identifying unforeseen negative consequences (e.g., how fines for homelessness exacerbate poverty).
SDG Contribution: The findings support the UN Sustainable Development Goal of poverty eradication by suggesting that effective policy-making must address institutional discrimination (aporophobia) alongside economic redistribution.

The authors maintain a modest tone, acknowledging the model is a "proof of concept" with limitations, such as the need for a more robust synthetic population and the current exclusion of multidimensional poverty indicators beyond wealth distribution. They frame the work as a foundation for a new generation of poverty reduction policies that actively mitigate discrimination.