Human-Centered Infrastructure on Mohammad Movahedi

Uncovering Community Debris Disposal Behaviors During Post-Disaster Recovery

Thu, 17 Sep 2026 00:00:00 +0000

Research explainer

Debris does not move through a community by logistics alone.

After a hurricane, every pile of debris sits at the end of a chain of human decisions: what residents know, how they prepare, what help they can find, and which disposal options they can actually reach.

1,600Hurricane Ian respondents

13impacted Florida counties

3linked decision stages

7debris streams examined

📰 In press

Please note that the paper has been accepted and is currently in press.

Post-disaster debris management is often treated as an engineering problem: estimate the volume, locate temporary sites, schedule collection, and route trucks. Yet cleanup outcomes also depend on thousands of household-level decisions made under urgency, uncertainty, and uneven access to services.

Our research, “Uncovering Community Debris Disposal Behaviors During Post-Disaster Recovery: A Case Study of Illegal Dumping After Hurricane Ian,” reframes those decisions as a staged behavioral process. It combines a Knowledge–Attitudes–Practices (KAP) framework, survey evidence from communities affected by Hurricane Ian, and a county-level Illegal Debris Disposal Tendency Index (IDDTI) for planning and screening.

The behavioral journey

One outcome, three decision environments

Select a stage to see how behavior develops before debris reaches its final destination.

Decision question

Figure 1. Preparedness, on-property handling, and final disposal are analytically distinct but behaviorally connected. Some roadside or curbside placements remain context-dependent because emergency authorization can vary locally.

From survey response to behavioral evidence

The survey mapped nine constructs across the three stages: knowledge, attitudes or constraints, and practices. Outcome-specific models were then matched to the data: Poisson models for counts and logistic models for binary legality, placement, and material-handling outcomes. False-discovery-rate correction was applied within each outcome family.

This design matters because “illegal dumping” is not one undifferentiated act. The study distinguishes designated sites and authorized collection, locally contingent roadside placements, and clearly unauthorized locations such as highways, parks, and other cities.

The central result

Knowledge changes compliance. Access still shapes possibility.

Switch the outcome to compare standardized total associations at the final disposal stage.

Knowledge · direct

Knowledge · via access

Disposal knowledge

Access

Prior preparation

← fewer reported instances more reported instances →

Figure 2. Standardized total associations from the staged path models. Bootstrap 95% intervals excluded zero for Stage-C knowledge and access, but not for Stage-B preparation. These cross-sectional relationships are associations, not causal effects.

At the count level, a one-unit increase in standardized Stage-C knowledge was associated with a 24% increase in expected reported legal behaviors (IRR = 1.24) and an 11% decrease in expected reported illegal behaviors (IRR = 0.89). Knowledge was more strongly associated with expanding legal practice than eliminating illegal placement—a signal that information alone cannot remove physical and institutional barriers.

Evidence explorer

The answer changes with the question

Explore the reported associations for behavior counts, placement locations, and correct material handling.

Lower association1.0 · no differenceHigher association

Figure 3. IRRs are shown for count models and odds ratios for binary outcomes. Values above 1 indicate a positive association and values below 1 a negative association, holding other modeled constructs constant.

The material-specific results make the knowledge signal especially visible. Higher Stage-C knowledge was associated with correct handling of white goods (OR = 3.67), wrecked vehicles (3.34), hazardous debris (3.24), construction and demolition waste (2.58), soil (1.99), and vegetative debris (1.61).

Placement tells a different story. Concern and help-seeking were positively associated with both ambiguous and illegal off-site locations, while correct Stage-B preparation was associated with designated-site use. More knowledgeable respondents were not consistently less likely to place debris off-site. The practical interpretation is not that knowledge fails, but that knowing the right action and being able to take it are different problems.

From households to counties

What changes when hazard exposure enters the index?

Choose a validation target to compare the socioeconomic-only IDDTI-A with IDDTI-B, which adds Hurricane Ian wind exposure.

IncomeAgeHomeownershipEducationEmploymentEstablishment size

Wind exposure IDDTI-B only

Figure 4. IDDTI values rank relative screening priority; they do not predict individual behavior. Validation was internal, county samples were uneven, and performance depends on the chosen target.

Against the survey-derived county ranking, adding hazard exposure improved Spearman correlation from 0.15 to 0.26 and pairwise accuracy from 0.54 to 0.60. Against the observed illegal-disposal ratio, however, IDDTI-A and IDDTI-B were close, and the hazard-enhanced version did not lead on every metric. That target dependence is a feature worth seeing, not hiding: the indices are screening tools for prioritization, not deterministic risk scores.

Translating evidence into action

Barrier

Intervention

Evaluate with

The evidence argues for a paired strategy. Use clear, debris-specific education where correct handling is technically demanding. At the same time, reduce the friction of compliance through closer temporary debris-management sites, mobile hazardous-waste drop-offs, real-time wait information, and coordinated collection support. Enforcement messaging is most credible when residents also have a practical legal option.

Use for outreach and planning—not enforcement targeting. The survey is cross-sectional, recall-based, self-reported, and geographically imbalanced. The online panel over-represents higher-income homeowners, reported illegal behavior is not measured prevalence, and the index lacks external validation against independent complaint or cleanup records. The framework and construction procedure are transferable; the Hurricane Ian coefficients and weights are not.

The broader contribution

This research shows why post-disaster disposal should not be reduced to a single cleanup choice. Earlier preparedness and handling conditions matter, but their associations with final behavior are largely mediated. At the moment of disposal, knowledge is linked to compliance quality, while access, convenience, concern, and available support shape where debris can end up.

That distinction gives emergency managers a more useful question than “Why did residents dump debris illegally?” It asks: At which stage did the path toward compliant disposal break down—and was the missing piece knowledge, access, support, or credible alternatives?

Behavior sandbox

Watch a neighborhood respond

Change the conditions and watch the residents redistribute among authorized disposal, curbside or locally contingent placement, and clearly unauthorized off-site placement.

Disposal knowledge 65% Access to authorized options 55% Collection delay 7 days

Figure 5. This is a qualitative teaching game, not a fitted agent-based model or forecast. It translates the study’s supported directional relationships into an illustrative neighborhood scene; the character counts are not estimated population shares.

Simulating Community Behaviors with LLM Personas

Thu, 16 Apr 2026 00:00:00 +0000

Research explainer

What if a policy could meet the community before it reaches the street?

Debris planning is usually framed as a logistics problem: estimate volume, select sites, and optimize routes. Our work adds the missing human layer by grounding LLM persona agents in observed community behavior, then using them to explore how different residents may respond to an intervention.

6behavioral persona clusters

80 / 20conditioning and evaluation split

~75%persona-agent decision alignment

IanHurricane case context

After a disaster, residents make debris-disposal decisions under pressure. A legal disposal site may be too far away. Municipal pickup may be delayed. Families may weigh safety, time, cost, convenience, trust, and social expectations differently. Those choices can create illegal dumping that slows recovery, raises municipal costs, and adds environmental and health risks.

Traditional behavioral models commonly rely on static rules or aggregated demographics. They can represent an “average household,” but struggle with the heterogeneous motivations and thresholds that shape actual decisions. Our paper, “Large Language Model-Based Framework for Simulating Community Behaviors in Policy Evaluation: Application to Community Debris Management,” proposes a different approach: use survey-grounded LLM personas as transparent, testable surrogates for exploring possible community responses before a policy is deployed.

The framework

Four phases connect evidence to exploration

Select a phase to follow the information from community survey to evaluated persona agent.

Interactive Figure 1. The framework is iterative: new empirical observations can refine persona definitions, agent conditioning, and subsequent evaluation.

From demographics to behavioral personas

The process begins with localized Knowledge, Attitudes, and Practices (KAP) survey data. For this application, the survey captured residents’ disposal practices after Hurricane Ian, knowledge of debris-handling procedures, perceptions of possible interventions, prior disaster experience, service access, responsibility beliefs, sustainability attitudes, and personality-linked features such as risk tolerance.

Principal Component Analysis reduces the high-dimensional responses into interpretable behavioral components. K-means clustering then groups residents who share decision patterns rather than merely demographic labels. In the Tampa application, this produced six persona clusters with different mixtures of legal, mixed, and illegal disposal behavior.

What makes a persona?

Build the agent from behavioral evidence

Explore the four evidence layers that give a persona context beyond a demographic profile.

Evidence layer

Why it matters to the agent

Interactive Figure 2. These layers describe categories used to ground persona development; they do not expose or recreate any individual survey respondent.

A moderator around the personas

Persona agents are not left to self-validate. The conceptual framework places a moderator agent around development and evaluation to check behavioral coherence, compare simulated and empirical distributions, and flag drift using measures such as Population Stability Index and Kullback–Leibler divergence. Ensemble responses can reduce dependence on a single generation, while human reviewers and community stakeholders can correct behavior that departs from realistic expectations.

This oversight matters because an LLM’s explanation is a post-hoc rationale, not direct access to its internal reasoning. Rationales can support interpretation and auditing, but they remain prompt-sensitive and vulnerable to hallucination. The paper therefore treats them as evidence to inspect—not ground truth.

Proof-of-concept behavior

Stress-test a disposal scenario

Adjust the two conditions highlighted in the Persona 1 evaluation. The response summarizes qualitative reasoning observed in the study.

Travel time to a legal disposal site 30 min

NearbyDistant

Municipal collection delay 7 days

PromptProlonged

Persona 1 response pattern

Reasoning emphasis

Interactive Figure 3. “Disposal pressure” is a qualitative explainer, not a predicted probability. The 30-minute and 14-day thresholds reflect scenarios discussed in the proof-of-concept evaluation.

When the legal disposal site was presented as 30 minutes away or less, Persona 1 emphasized safety and family welfare and favored legal disposal, consistent with the cluster’s hazard sensitivity. Beyond 30 minutes, the agent’s rationale shifted toward inconvenience. When collection delays exceeded 14 days, frustration became more prominent. These context-sensitive shifts mirrored drivers observed in the cluster-level survey data.

Evaluation, not prediction

How closely did the proof-of-concept align?

The persona agent and conventional baseline were evaluated using the same cluster-specific data.

Persona agentGPT-4o-mini, survey-conditioned

~75%

Conventional baselineeXtreme Gradient Boosting

~60%

Interactive Figure 4. Accuracy is reported as decision alignment with observed cluster-level behavior. The baseline comparison provides context and is not a hypothesis test or claim of general model superiority.

The persona agent achieved approximately 75% decision alignment accuracy, compared with about 60% for XGBoost on the same data. The result is promising because the LLM also produced context-sensitive rationales that could be inspected alongside its choices. But it remains a proof of concept based on one persona, one survey dataset, and one disaster context—not evidence that the agent can predict a new community without recalibration.

6 clusters

PCA and K-means represented heterogeneous household, access, and KAP patterns.

1 persona

Persona 1 was selected for the demonstration because of its illegal-dumping tendency and sensitivity to convenience.

20% held out

Cluster-specific survey responses were reserved exclusively for agent evaluation.

A policy sandbox—with guardrails

The value of this framework is not automated policymaking. It is the ability to ask structured “what if” questions before spending scarce recovery resources: Which communities might respond to closer disposal access? Where could a service delay undermine an outreach campaign? Which trade-offs or unintended effects deserve direct community investigation?

Surrogate agents do not replace people.LLM personas are synthetic approximations and may inherit bias from training data, prompts, or calibration. Responsible application requires IRB approval, informed consent, removal of personally identifiable information, transparent auditing, withheld-data validation, and meaningful human and community oversight. The framework is an exploratory decision-support tool—not a substitute for engagement with affected residents.

Beyond debris management, the same four-phase workflow can be adapted to recycling, demolition planning, transportation, evacuation, and other human–infrastructure systems. Each new application would require localized behavioral evidence, domain-specific constraints, and fresh evaluation; the persona is never assumed to transfer unchanged.

Read the published paper

Citation

@inproceedings{Baldwin2026LLMPersonas,
 author = {Baldwin, Alexander J. and Movahedi, Mohammad and Choi, Juyeong},
 title = {Large Language Model-Based Framework for Simulating Community Behaviors in Policy Evaluation: Application to Community Debris Management},
 booktitle = {Construction Research Congress 2026},
 pages = {718--728},
 address = {San Antonio, Texas},
 publisher = {American Society of Civil Engineers},
 year = {2026},
 doi = {10.1061/9780784487006.071}
}