💻 computer science

Effects of Property Recovery Incentives and Social Interaction on Self-Evacuation Decisions in Natural Disasters: An Agent-Based Modelling Approach

This study employs an agent-based model grounded in evolutionary game theory to demonstrate that optimizing natural disaster evacuation policies requires balancing property recovery incentives with social network structures, revealing that prioritizing well-connected "community influencers" yields better collective outcomes than indiscriminate or low-connectivity-focused support.

Original authors: Made Krisnanda, Raymond Chiong, Yang Yang, Kirill Glavatskiy

Published 2026-02-24

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Made Krisnanda, Raymond Chiong, Yang Yang, Kirill Glavatskiy

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

🌪️ The Big Picture: The Great "Run or Stay" Game

Imagine a small town living on the edge of a volcano. One day, the ground starts rumbling. Everyone has to make a split-second decision: Do I pack my bags and run, or do I stay home and hope for the best?

This paper is about figuring out how to get everyone to run safely without causing a stampede or leaving people behind. The researchers used a computer simulation (a "digital twin" of a town) to play out this scenario thousands of times to see what works.

They focused on two main things:

The "Bribe" (Incentives): The government says, "If you leave, we'll give you money to fix your house later."
The "Gossip Chain" (Social Network): How neighbors talk to each other.

🎮 The Game: How the Simulation Works

Think of the town as a giant game of Musical Chairs, but instead of chairs, the players are fighting over safety and money.

The Players: Every house is a "player" (an agent).
The Neighbors: Players are connected by invisible strings (social networks). Some people are popular and know 9 neighbors (the "Influencers"). Some are shy and only know 2 neighbors.
The Rules:
- If you stay, you might lose your house to looters or fire.
- If you leave, you face traffic jams and crowded shelters.
- The Government's Role: They offer a "Recovery Fund" (money) if you leave. They also offer a "Shuttle Service" to make leaving easier.

The computer runs the game, and every time a player sees their neighbor getting a better deal (more money, less stress), they might copy that neighbor's decision. This is how the decision to "Run" spreads through the town.

🔑 The Big Discoveries

The researchers tested four different ways to tell people to leave first. Here is what they found:

1. The "Influencer" Effect (High-Connectivity Nodes)

Imagine the town has a few famous people: the local mayor, the church leader, and the popular bar owner. These people know everyone.

The Finding: If you convince these "Influencers" to leave first, the whole town follows quickly. It's like dropping a pebble in a pond; the ripples spread fast.
The Analogy: If the most popular kid in school packs a bag, the whole class packs up.

2. The "Wallflower" Problem (Low-Connectivity Nodes)

Now, imagine you try to convince the shy people who only talk to their immediate next-door neighbors.

The Finding: This actually slows things down. If you prioritize the shy people, the decision to leave gets stuck. They can't spread the news fast enough.
The Analogy: Trying to start a wave in a stadium by asking the people in the back row to stand up first. It takes forever to reach the front.

3. The "Tipping Point" (The 57% Magic Number)

This is the most exciting part. The researchers found that evacuation isn't a smooth line; it's like a light switch.

The Finding: If you convince 56% of the "Influencers" to leave, maybe only 40% of the town leaves. But if you convince just 0.04% more (reaching 57%), suddenly 100% of the town runs!
The Analogy: Think of a dam holding back water. You can push the water for hours, and nothing happens. But the moment you push just a tiny bit more, the dam breaks, and whoosh—everything floods out. There is a "tipping point" where the decision to leave becomes unstoppable.

4. The "Bribe" Sweet Spot

The Finding: Giving more money helps, but only up to a point. If the government offers a tiny amount, it doesn't work. If they offer a huge amount, it works, but it's wasteful.
The Analogy: It's like a sale at a store. A 5% discount might not get you to buy. A 50% discount gets you to buy. But a 200% discount (where they pay you to take the item) is just throwing money away. The researchers found the "Goldilocks" zone where the money is just enough to trigger the "Influencers" to start the chain reaction.

🚨 What This Means for Real Life

If you are a government official planning for a disaster (like a hurricane or wildfire), this paper gives you a cheat sheet:

Don't shout at the whole crowd. Instead, find the "Influencers" (community leaders, popular neighbors) and convince them to leave first.
Watch out for the "Tipping Point." You don't need to convince everyone. You just need to convince enough key people to cross that invisible line (around 57% in their model) so the rest follow automatically.
Money matters, but connections matter more. A little bit of money targeted at the right people is better than a lot of money thrown at the wrong people.
Don't waste time on the shy ones first. If you try to convince the least connected people first, you might actually delay the evacuation.

🏁 The Bottom Line

Evacuation isn't just about math or money; it's about social ripple effects. By understanding who talks to whom, governments can save lives and money by triggering a "domino effect" where one person's decision to leave inspires the whole community to follow.

1. Problem Statement

Natural disasters cause massive economic and human losses, making effective evacuation planning critical. However, designing evacuation policies is challenging due to the complex interplay between:

Household Decision-Making: Households must decide whether to evacuate or stay based on personal risk perception and property protection.
Resource Constraints: Limited government resources (transportation, shelters, recovery funds) create competition among households.
Social Dynamics: Households communicate and influence each other, creating a social network where decisions are not made in isolation.
Policy Gaps: Existing studies often overlook the specific impact of property recovery incentives (financial compensation) and how social network structures (specifically node connectivity) influence the spread of evacuation behaviors. Policymakers lack data on how to optimally prioritize incentives to maximize collective evacuation rates under limited budgets.

2. Methodology

The authors propose a hybrid modeling framework combining Agent-Based Modelling (ABM) and Evolutionary Game Theory (EGT) to simulate household interactions in a disaster-prone community.

A. Model Framework

Agents: Represent individual households. Each agent owns property with value $P$ and possesses a risk perception probability $p$ .
Network Topology: A Small-World Network with 5,000 nodes is used to mimic real-world social dynamics (high clustering with long-range connections). Node degrees range from 2 to 9.
Decision Mechanism (PMT & EGT):
- Agents use Protection Motivation Theory (PMT) to assess threats and coping costs.
- Interactions are modeled as a game where agents choose between Evacuate ( $e$ ) or Stay ( $s$ ).
- Payoff Calculation: Payoffs depend on property loss, evacuation costs, congestion, and government incentives.
  - Baseline: Agents suffer property damage ($pP$) and specific costs for unattended property ( $\alpha P$ ) or lack of neighbor protection ( $\beta P$ ).
  - Incentives: The government provides financial recovery funds ( $\theta P$ ) and services (transport coordination $r_T$ , resource allocation $r_D$ ).
Evolutionary Dynamics: Agents update decisions based on a probability-based imitation rule. An agent adopts a neighbor's strategy if the neighbor's payoff is higher, with the probability of adoption proportional to the payoff difference.

B. Experimental Setup

Control Parameter: The recovery incentive fraction ( $\theta$ ) ranges from -10% to 20%.
Scenarios: Four prioritization strategies were tested to determine which agents should receive incentives first:
1. Randomised Highest-Degree: Prioritize high-connectivity nodes; others are randomized.
2. Fixed Highest-Degree: Prioritize high-connectivity nodes; others are fixed to "Stay."
3. Randomised Lowest-Degree: Prioritize low-connectivity nodes; others are randomized.
4. Fixed Lowest-Degree: Prioritize low-connectivity nodes; others are fixed to "Stay."
Simulation: Runs over 3,000 time steps. Final evacuation rates are averaged over the last 1,000 steps and 5 simulation runs.

3. Key Contributions

Integration of Financial Incentives and Social Networks: The study uniquely models how property recovery funds interact with social network structures to drive evacuation, moving beyond simple cost-benefit analysis.
Identification of "Community Influencers": The research quantifies the disproportionate impact of high-degree nodes (highly connected households) compared to low-degree nodes in triggering collective action.
Discovery of Threshold Effects and Tipping Points: The model reveals non-linear dynamics where small changes in prioritization levels (e.g., crossing from 56.98% to 57% priority) cause discontinuous jumps in evacuation rates.
Policy Optimization Strategy: It provides a data-driven framework for policymakers to determine the optimal mix of financial incentives and social prioritization, showing that high incentives can sometimes render complex prioritization unnecessary.

4. Key Results

Diminishing Returns of Incentives: The impact of government incentives diminishes as the funding value ( $\theta$ ) increases. There is an optimal level of support; beyond this, further increases yield impractical returns.
The Power of High-Degree Nodes:
- Prioritizing high-degree nodes (influencers) significantly increases overall evacuation rates.
- Prioritizing low-degree nodes (low connectivity) often impedes collective evacuation or has negligible impact, as these agents lack the social reach to influence others.
Critical Thresholds (Tipping Points):
- In the "Randomised Highest-Degree" scenario with 0% incentive, a tiny increase in priority from 56.98% to 57% caused the evacuation rate to jump from 38.3% to 100%. This corresponds to the inclusion of degree-4 nodes, which bridges disconnected clusters.
- Similar instability and "meta-stable states" (where the population fluctuates between evacuating and staying) were observed near these thresholds.
Scenario Comparisons:
- Fixed vs. Random: Scenarios where non-prioritized agents were fixed to "Stay" (Fixed scenarios) were more resistant to change and required higher prioritization levels to achieve mass evacuation compared to scenarios with random initial seeds.
- Incentive Levels: At high incentive levels (10-20%), the need for specific prioritization strategies diminishes because the financial payoff for evacuation becomes universally attractive. Prioritization is most critical when incentives are low or negative.
Contribution Analysis:
- Degree 9 (Top Influencers): Although few in number, they have the highest "efficiency" (contribution per agent) in driving evacuation.
- Degree 8: Surprisingly showed negative contributions in some cases, likely due to overlapping influence with Degree 9 nodes.
- Degrees 2 & 3: Contributed minimally or negatively to evacuation dynamics.

5. Significance and Implications

Policy Design: Governments should focus limited resources on targeting community influencers (high-degree nodes) rather than distributing incentives uniformly or prioritizing isolated households. This approach can trigger a "cascade" of evacuations.
Resource Efficiency: Understanding tipping points allows policymakers to avoid overspending. A small, well-targeted intervention at a critical network threshold can achieve 100% evacuation, whereas massive spending without strategic targeting may fail.
Disaster Management: The study highlights that social connectivity is as crucial as monetary value in evacuation planning. Ignoring network structure can lead to suboptimal policies where resources are wasted on agents who cannot influence the collective decision.
Future Directions: The authors suggest future work should explore dynamic networks (evolving social ties), diverse network topologies (scale-free), and agent heterogeneity regarding risk perception and mobility.

In conclusion, this study demonstrates that strategic prioritization of social influencers combined with moderate financial incentives is the most effective mechanism for maximizing self-evacuation rates, particularly when resources are constrained.