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

Made Krisnanda; Raymond Chiong; Yang Yang; Kirill Glavatskiy

cs.MAFeb 23, 2026

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

Authors: Made Krisnanda, Raymond Chiong, Yang Yang, Kirill Glavatskiy

Organizations: School of Computer and Information Sciences, The University of Newcastle · School of Medicine and Public Health, The University of Newcastle · School of Science and Technology, The University of New England

Abstract

Understanding evacuation decision-making behaviour is one of the key components for designing disaster mitigation policies. This study investigates how communications between household agents in a community influence self-evacuation decisions. We develop an agent-based model that simulates household agents' decisions to evacuate or stay. These agents interact within the framework of evolutionary game theory, effectively competing for limited shared resources, which include property recovery funds and coordination services. We explore four scenarios that model different prioritisations of access to government-provided incentives. We discover that the impact of the incentive diminishes both with increasing funding value and the household agent prioritisation, indicating that there is an optimal level of government support beyond which further increases become impractical. Furthermore, the overall evacuation rate depends on the structure of the underlying social network, showing discontinuous jumps when the prioritisation moves across the node degree. We identify the so-called "community influencers", prioritisation of whom significantly increases the overall evacuation rate. In contrast, prioritising household agents with low connectivity may actually impede collective evacuation. These findings demonstrate the importance of social connectivity between household agents. The results of this study are useful for designing optimal government policies to incentivise and prioritise community evacuation under limited resources.

Figures & tables

Figure 1: Illustration of interactions between households in the model. The payoff applied depends on the decisions of the two interacting parties.

Both evacuate
$A_{ee}^{0}$ = $(1-p)P$	$B_{ee}^{0}$ = $(1-p)P$
A evacuates, B stays
$A_{es}^{0}$ = $(1-p)P-(1-r_{E})\alpha P$	$B_{es}^{0}$ = $(1-p)P-(1-r_{S})\beta P$
A stays, B evacuates
$A_{se}^{0}$ = $(1-p)P-(1-r_{S})\beta P$	$B_{se}^{0}$ = $(1-p)P-(1-r_{E})\alpha P$
Both stay

Table 1: Baseline Evacuation Payoff

Both evacuate
$A_{ee}$ = $A_{ee}^{0}+{\color[rgb]{0,0,1}\theta P}-{\color[rgb]{0,0,1}(1-r_{T})\alpha P}$	$B_{ee}$ = $B_{ee}^{0}+{\color[rgb]{0,0,1}\theta P}-{\color[rgb]{0,0,1}(1-r_{T})\alpha P}$
A evacuates, B stays
$A_{es}$ = $A_{es}^{0}+{\color[rgb]{0,0,1}\theta P}+{\color[rgb]{0,0,1}r_{T}(1-r_{E})\alpha P}$	$B_{es}=B_{es}^{0}$
A stays, B evacuates
$A_{se}$ = $A_{se}^{0}$	$B_{se}$ = $B_{se}^{0}+{\color[rgb]{0,0,1}\theta P}+{\color[rgb]{0,0,1}r_{T}(1-r_{E})\alpha P}$
Both stay

Table 2: Baseline with Government Incentive Payoff (Blue)

Notation	Definition	Value
$p$	The estimation of risk and cost of life and properties	0.5
$\alpha$	Cost fraction for evacuating group	0.4
$\beta$	Cost fraction for staying group	0.2
$r_{E}$	Cost reduction for evacuating agent when interacting with the staying agent.	$0.5$
$r_{S}$	Cost reduction for staying agent when interacting with the evacuating agent.	$0.07$
$r_{T}$	Cost reduction from government transportation services	$0.5$

Table 3: Government Incentives: Notations, Definitions, and Values

Both evacuate
$a_{ee}^{0}=0.5$	$b_{ee}^{0}=0.5$
A evacuates, B stays
$a_{es}^{0}=0.3$	$b_{es}^{0}=0.656$
A stays, B evacuates
$a_{se}^{0}=0.656$	$b_{se}^{0}=0.3$
Both stay

Table 4: Baseline Evacuation Payoff (Coefficient Values)

Both evacuate
$a_{ee}=0.3+\theta$	$b_{ee}=0.3+\theta$
A evacuates, B stays
$a_{es}=0.4+\theta$	$b_{es}=0.47$
A stays, B evacuates
$a_{se}=0.47$	$b_{se}=0.4+\theta$
Both stay

Table 5: Baseline with Government Incentive Payoff (Coefficient Values)

Figure 2: Evacuation rates in a. a randomised highest-degree scenario and b. a fixed highest-degree scenario. Each point is the average of 5 simulations, and the error band has a width of 2 standard deviations. Vertical lines indicate the fractions of nodes corresponding to the degree change, ranging from degree 9 to degree 3: 0.04%, 0.42%, 3.78%, 19.26%, 56.98%, 98.2%, and 98.8%. These thresholds mark the points at which agents with specific degrees begin to receive prioritised treatment.

Degree	Population (%)	Rate change				Rate change per agent
		-10%	0%	10%	20%	-10%	0%	10%	20%
-	Start	5.08	15.2	38.9	70.5	N/A	N/A	N/A	N/A
9	0.04	1.03	3.1	2.1	5.4	0.51	1.55	1.05	2.70
8	0.38	-0.32	-2.9	3.9	-4.3	-0.02	-0.15	0.21	-0.23
7	3.36	1.17	15.1	20.7	18.9	0.01	0.09	0.12	0.11
6	15.48	-0.12	7.8	28.2	8.0	0.01	0.01	0.03	0.01

Table 6: Contribution of evacuation rates from different degrees in the randomised highest-degree scenario. The numbers in each column add to 100%. The largest number in each column is marked in bold .

Figure 3: Evacuation rates in a) a randomised lowest-degree scenario and b) a fixed lowest-degree scenario. In contrast to the highest-degree scenarios, prioritised treatment was initiated from the lowest-degree nodes, ranging from degree 3 to degree 9: 1.2%, 1.8%, 43.02%, 80.74%, 96.22%, 99.58%, and 99.96%. The results indicate that low-degree nodes make a negligible contribution to the evacuation rates.

Degree	Population (%)	Rate change				Rate change per agent
		-10%	0%	10%	20%	-10%	0%	10%	20%
-	Start	6.78	18.8	35.3	71.8	N/A	N/A	N/A	N/A
2	1.2	-2.13	-3.2	7.8	-1.8	-0.035	-0.05	0.13	-0.03
3	0.6	1.99	1.2	-4.4	7	0.07	0.04	-0.15	0.23
4	41.22	0.19	-1.7	-1.4	-6	0.0	0	0	0
5	37.72	-0.63	2.3	0.6	0.4	0.0	0	0	0

Table 7: Contribution to evacuation rates from different degrees in the randomised lowest-degree scenario. The largest number in each column is marked in bold .

Figure 4: The degree distribution of the small-world network used in the simulations.

Figure 5: Comparison of results around the transition between 50% and 50.4% prioritisation(Plot a), and between 56.6% and 57% prioritisation(Plot b) in the 0% incentive values. Plots c) and d) show five curves with the same priority values (50% and 56.6%). This figure shows that different results occurred around the thresholds at both different and the same priority levels.

Figure 6: Evacuation rates and switches result for 0% incentive and 50% priority.The top panel displays the overall evacuation rate over time, while the remaining (arranged by descending node degree) show switching frequencies for each degree group. Each switching plot displays only non-zero switch values using circle markers to highlight discrete decision changes. The x-axis represents simulation time steps, while the y-axis denotes the number of agents switching decisions at each step.

Figure 7: Evacuation rates and switches result for 0% incentive and 56.7% priority.

Figure 8: Evacuation rates and switches result for 0% incentive and 57% priority.

Figure 9: Heatmap results for 0% incentive with a.50%, b. 56.7% and c. 57% priority values.

Disaster Response

Disaster response uses data, models, and planning tools to assess hazards and coordinate evacuation, relief, and recovery.

See all 59 papers

An LLM-powered Agent Framework for Heterogeneous Evacuation Behavior Modeling under a Moving Threat in a Public Plaza
Sep 29, 2026Jian Ma, Runxin Yu, Tianyu Tang +1Pedestrian Behavior Modeling Agent-Based Modeling
School of Transportation and Logistics, Southwest Jiaotong University, Chengdu, 610031, China · Fujian Police College, Fuzhou, 350007, China
Hierarchical Generative Agents for Simulating Sequential Human Behavior
Jun 12, 2026Maria G. Mendoza, Lucas Waldburger, Jin Lee +1Cognitive Modeling Human Behavior Simulation
University of California, Berkeley
Simulation of Crowd Egress with Environmental Stressors
Jun 3, 2022Peng Wang, Xiaoda Wang, Peter Luh +4Pedestrian Behavior Modeling Human Behavior Simulation
Department of Electrical and Computer Engineering, University of Connecticut, Storrs

Sep 29, 2026J/K move · Enter open · S save

An LLM-powered Agent Framework for Heterogeneous Evacuation Behavior Modeling under a Moving Threat in a Public Plaza

Jian Ma, Runxin Yu, Tianyu Tang, Xiaolian Li

School of Transportation and Logistics, Southwest Jiaotong University, Chengdu, 610031, China · Fujian Police College, Fuzhou, 350007, China

Pedestrian Behavior Modeling Agent-Based Modeling Large Language Model-Based Social Simulation

Open paper AskPDF arXiv

Modeling heterogeneous evacuation behavior under a moving threat is difficult because human perception, memory, and evidence evaluation are not well captured by fixed rules. We propose a novel LLM-powered agent-based framework to represent these internal decision processes. Each pedestrian agent perceives a private symbolic ASCII view, maintains a Memory-based Knowledge Graph derived solely from individual observations, and makes decisions through persona-conditioned prompts under a common sampling configuration. A compressed decision context with stateless memory preserves trial-and-error experience across turns while excluding reasoning traces, and a validation engine separates behavioral choice from physical feasibility by executing routes only over observed terrain. We evaluated eight personality compositions in eight paired randomized blocks within a simulated public plaza. Usable-exit knowledge was strongly associated with evacuation success: 89.5% of agents possessing such knowledge evacuated, compared with 1.05% of those without it. Personality compositions also differed in their evaluation of remembered threat evidence: the proportion of danger assessments varied by 0.265, while high-urgency, low-directness decisions ranged from 11.41% to 34.33%. After direct threat sightings, responses converged, with 99.6% of assessments classifying the situation as dangerous. Movement was selected in 99.8% of decisions. Overall, evacuation outcomes were strongly associated with information access, while evacuation time was jointly associated with spatial geometry, information, and affect. The framework provides an auditable approach to generating endogenous behavioral heterogeneity through persona-conditioned LLM agents in crowd-evacuation simulations.

Abstract

Disaster Response

Explore similar work

An LLM-powered Agent Framework for Heterogeneous Evacuation Behavior Modeling under a Moving Threat in a Public Plaza

Hierarchical Generative Agents for Simulating Sequential Human Behavior

Simulation of Crowd Egress with Environmental Stressors