Household Displacement and Return in Disasters: A Review

At least 265 million people were displaced due to sudden-onset disasters (e.g., earthquakes, tropical cyclones) between 2008 and 2018, which amounts to nearly one person per second (IDMC 2019). The number of people displaced by disasters is projected to increase, driven by poorly managed urban growth in hazard-prone areas (IDMC 2017) and the effects of climate change (IPCC 2012). This scale of human impact has prompted international attention, such as within the 2015–2030 Sendai Framework for Disaster Risk Reduction, which stresses the severe consequences of displacement, including disrupted family and community life, destroyed livelihoods, interrupted education, increase in poverty, health and safety concerns, and reduced access to basic services (UNDRR 2019).

Despite this massive scale of human impact, disaster risk assessments have historically focused on direct economic losses due to physical damage. This metric tends to highlight the wealthiest as the most at risk because they have the most economic value to lose. This outcome is discordant with observations from past disaster events, which consistently demonstrate that poor and marginalized groups are disproportionately affected (Hallegatte et al. 2020; Fothergill and Peek 2004). Mitigation strategies selected primarily based on economic losses may inadvertently deepen existing inequalities, highlighting a need for more people-centered and equitable risk metrics to inform disaster mitigation (Markhvida et al. 2020; Cremen et al. 2023). Displacement could serve as a more equitable risk metric that better captures the human toll of disasters, but consideration within disaster risk assessments has been limited.

Most attempts to incorporate population displacement within disaster risk and recovery models focus on the effects of physical damage to the built environment, which primarily inform initial displacement and emergency shelter needs. Factors beyond damage to the built environment, such as psychological and social phenomena (e.g., place attachment, social capital and networks, perceived risk), demographic factors (e.g., socioeconomic status, housing and land tenure, age, race/ethnicity/caste), and policy decisions (e.g., housing reconstruction approach, design of aid programs), are rarely considered within community recovery models despite widespread evidence of their significance on displacement duration and household return following past disaster events.

This review synthesizes existing research on disaster-induced displacement, including key definitions and contextual factors, determinants of return decisions, existing models of household displacement, and opportunities for future research. The findings from this review can aid an enhanced understanding of household displacement in disasters and guide future efforts to quantify the phenomenon within disaster recovery models.

Disasters are life events that can subject households to key migration decision points (Chang et al. 2008). The relative weight of influencing factors may vary by context (e.g., currently evacuated versus already returned). Although the initial evacuation or dislocation is often a forced movement, return decisions are more voluntary, albeit constrained by realities such as housing availability, moving costs, and proximity to livelihoods. Some common household decision points within the context of disasters are illustrated in Fig. 4.

This section focuses on the decisions to return. For a review of household evacuation decisions, see Thompson et al. (2017) and Dash and Gladwin (2007). For a review of household decisions to stay or resettle, see Greer et al. (2019). For a discussion of household decisions on shelter choice, see Lee and Chen (2018).

Prevalent determinants of household return decisions are described individually herein, although these factors are often highly linked (Henry 2013; Esnard and Sapat 2014) and intersecting (Rufat 2013; Morrow 1999). Moreover, the determinants of return decisions can differ by population subgroup (e.g., renters versus owners, city dwellers versus rural inhabitants) (Henry 2013; Burton et al. 2019) and by decision choice (i.e., return versus relocate) (Henry 2013; Kim and Soo Oh 2014). For example, Henry (2013) found that homeowners after Hurricane Katrina revolved their decisions to return or relocate around housing, but only as it related to factors such as family, work, insurance, and perceived or acceptable risk. In contrast, renters were less motivated by housing matters and more influenced by work, education, and family.

Available quantitative studies focused on household return decisions after disasters were reviewed, with a summary of the top influencing factors shown in Fig. 5 and the full Table S2 provided in the Supplemental Materials. Fig. 5 indicates the frequency of the top influencing factors in two ways: across all study instances (i.e., data sample or subset presented within a given study) and across only instances that considered that factor. The former may be biased toward factors that have been heavily considered within studies, whereas factors considered in fewer studies may be more prone to extreme values in the latter case. Different metrics are used to quantify similar phenomena across some studies, but these factors are presented separately. For example, attachment to place is often measured through proxies such as tenure time or hometown status. Similarly, housing damage can be measured using property damage losses as a proxy. To help organize the results, each factor was grouped into the following categories: housing matters, financial aspects, social and community aspects, and demographics. The factors that most consistently influenced return decisions in the reviewed quantitative studies include housing damage, tenure, neighborhood damage level, and family and relationships. The following subsections elaborate on these and some additional factors, organized into the following categories: physical damage to the built environment, psychological and social phenomena, household demographics, and pre- and postdisaster policies.

This section focuses on displacement models that extend into the recovery phase rather than evacuation and shelter choice. An evaluation of models for evacuation behavior in hurricanes is available in Anyidoho et al. (2022), and an evaluation of models for shelter needs following earthquakes is available in Vecere et al. (2017). Displacement duration following disasters depends critically on housing recovery and the community context, such as socioeconomic characteristics, infrastructure and services, and resource constraints (Comerio 2006; Costa et al. 2021; Sutley et al. 2019). For a comprehensive overview of various community and housing recovery methods, see Miles et al. (2019). This review focuses specifically on the integration of population displacement estimates within these models. A summary of the key models discussed herein is shown in Fig. 6.

FEMA proposed a population displacement model in HAZUS-MH (FEMA 2003), primarily based on research following Hurricane Hugo, the Loma Prieta earthquake, and the Northridge earthquake (Lin 2009). The number of uninhabitable buildings is estimated by building type (i.e., single family or multifamily) and damage level (i.e., complete for single family and moderate for multifamily). Lin (2009) proposed two alternate approaches: the first a modified version of the HAZUS methodology (i.e., using different damage state probabilities and aggregation units) and the second an algorithmic approach incorporating social vulnerability factors (i.e., race/ethnicity) based on empirical evidence following Hurricane Andrew in 1992 in Miami-Dade County, Florida. The algorithmic approach uses a logistic regression model that estimates the probability that residents of a given structure will dislocate based on the percent loss of the structure (due to damage), housing type, and the percent of both Black and Hispanic subpopulations in the respective block group. The output of these models represents a dislocated population, but does not explicitly include a duration component nor does it capture successive population movements such as returns, relocations, and resettlements.

Regression models based on survey data have commonly been applied to identify determinants of household decisions to return or relocate (Nejat and Ghosh 2016; Kim and Soo Oh 2014; Burton et al. 2019; Cong et al. 2018; Lee et al. 2017; Gray et al. 2009; Elliott and Pais 2006; Abuhamdia 2018; Nawrotzki et al. 2014; Girard and Peacock 1997; Groen and Polivka 2010). Burton et al. (2019) used such a model to directly link community recovery model variables (i.e., housing damage, utility disruption, neighborhood evacuation level, duration of building inaccessibility) and household characteristics (income, tenure, insurance access) to household decisions to return or resettle (with or without repairs). The integration of the proposed household decision-making module within the disaster recovery model was shown to have a significant impact on the population recovery trajectories for their case study of a hypothetical earthquake in Los Angeles, emphasizing the importance of such decisions on community and household recovery. This model captures displacement duration by tracking population return (versus resettlement), but does not explicitly capture the transition of individual households through various stages of sheltering and temporary housing.

Costa et al. (2022a) proposed a comprehensive model of population displacement following earthquakes, differentiating between temporary displacement (based on structural damage, availability of water and power, and socioeconomic demographics) and permanent relocation (based on housing recovery speed, neighborhood recovery status, the recovery decisions of neighbors, a predisposition to move out, and psychological trauma). This model adopts an agent-based approach, which has previously been advocated to capture the heterogeneity and interactions across individual agents in household migration studies (Husby and Koks 2017). This initial model was later extended to integrate place attachment (Costa et al. 2022b) and to link the repair of rental units with landlord decisions (Mongold et al. 2022). Model outputs include population recovery trajectories, typically aggregated at the neighborhood level, although the approach could theoretically be applied at the household level if such detailed data were available. This framework captures the duration of displacement by tracking population return, but also explicitly differentiates dislocation (temporary displacement) versus resettlements (permanent displacement). Similar to the Burton et al. (2019) model, the Costa et al. (2022a) approach does not explicitly track household movements between stages of sheltering and temporary housing.

Similarly, Bhattacharya and Kato (2021) adopted an agent-based model approach to capture household decisions to return or relocate after a disaster, accounting for place attachment level, option level (e.g., financial standing and insurance access), and infrastructure requirements as primary household characteristics. Secondary household characteristics (i.e., elderly with or without pension; young with or without children) allow different household archetypes to prioritize different place attributes (e.g., employment, schools, hospital), influencing their perception of the disaster area’s attractiveness. Each household agent has a tolerance threshold considering the cost, waiting time, attractiveness, and (in the case of a high attachment) distance from home. Their framework allows the simulation of various recovery policy options, such as the provision of temporary housing, target infrastructure service levels over time, and the spatial location of facilities. In contrast to the other models described herein, this approach explicitly tracks relocations into temporary housing.

Grinberger and Felsenstein (2016) also employed agent-based models to capture urban dynamics and disaster recovery, incorporating household relocation decisions. Households decide to leave their residence based on utility-maximizing behavior (income relative to residential prices) and constraints (i.e., lack of space and affordability). Residential prices vary depending on the floor area and the average housing price in the neighborhood, which dynamically varies with demand, supply, and the level of community services available. Additionally, interaction with commercial buildings is considered as household agents decide to visit nonresidential buildings; without sufficient traffic, commercial buildings become unoccupied. The framework allows evaluating population recovery over time, depending on policy choices such as land-use regulations, provision of shelters, and restoration of damaged public services. Similar to most of the other models described herein, the focus is on returns and resettlements (i.e., the permanent housing stage).

Although household displacement and return is a growing research area, several vital gaps and research opportunities remain. Despite the significance of displacement duration on the affected households and the community, few studies explicitly capture this temporal component (Sutley and Hamideh 2020; Smith and McCarty 1996; Fussell et al. 2010). Household displacement duration can be integrated within existing community recovery models that explicitly account for a temporal component (Costa et al. 2021; Miles and Chang 2011; Burton et al. 2016) but require validation that is limited by the lack of authoritative time-series data from past disaster events. Moreover, existing estimates of displaced persons after disasters are typically derived using housing damage estimates (IDMC 2020), posing a validation challenge because most household displacement models rely on similar assumptions. Mobile location data (e.g., call detail records, smartphone GPS data) could serve as a valuable data proxy for displacement that explicitly captures the time and space component (Lu et al. 2012; Bengtsson et al. 2011; Yabe et al. 2019, 2020; Wilson et al. 2016), but additional research is required to benchmark past events across geographies and hazard types to understand the data’s representativeness and applicability. The suitability of housing and community recovery models also requires review across different contexts because most applications have been to high-income communities that may have unique policy contexts (e.g., market-driven, limited public assistance) and determinants of household decision-making.

Furthermore, observations from past events indicate challenges with the recovery of affordable rental housing in urban areas (Hirayama 2000; Taheri et al. 2013; Mukherji 2010), whereas explicit consideration of multifamily and rental housing is largely missing from community and housing recovery models (Miles et al. 2019; Mongold et al. 2022). Surveys and regression models have been applied to homeowner and renter decisions to return or relocate/resettle following a disaster. Yet, similar work may be required to capture landlord decisions regarding repair or abandonment. Until rental and multifamily housing are incorporated, community and housing recovery models may have a limited capacity to evaluate policy decisions that encourage broad-based recovery (e.g., rental assistance and cash vouchers, incentives for rental property owners). Moreover, these models often use housing as the sole vehicle for displacement and return. Still, livelihood factors (e.g., job loss) and related policies (e.g., postdisaster employment insurance, livelihood assistance) may also have a significant impact.

Another challenge within community recovery models is the correlation of factors that influence both social and physical vulnerability. Synthetic population distribution is often employed in disaster risk models to populate buildings with people and their relevant characteristics, typically using aggregate demographic or census data. Microdata (e.g., individual or household-level responses to statistical surveys) may help realistically capture the evident correlation of some of these characteristics (e.g., income level and housing quality) that have been observed in past disaster events (Rubinyi et al. 2022).

Last, given the conflation and inconsistency of common terminology under the scope of disaster-induced displacement, future research should aim to define the adopted definitions clearly and indicate the setting of displacement within the disaster timeline (e.g., in the emergency phase, in the recovery phase).

Despite the scale and prevalence of household displacement following disasters and the toll of protracted displacement on affected households and communities, quantification of displacement primarily remains limited to the emergency phase in existing disaster risk models. This review aimed to synthesize the knowledge from existing studies and proposed models to focus future research and improve model-based approaches to quantify disaster-induced displacement.

From a review of the existing literature, it is clear that different factors influence household evacuation in the preparedness or emergency phase versus return in the recovery phase. There is consensus that physical factors (e.g., housing damage, utility disruption, neighborhood damage) are critical during the emergency phase. However, social and psychological factors such as place attachment and social capital become more important in the recovery phase as displaced households decide whether or not to return. Furthermore, return outcomes and recovery trajectories frequently appear stratified across demographic lines such as income level and ownership status. These differential patterns of household recovery are contributed to by predisaster conditions (e.g., concentrations of socially vulnerable populations in hazard-prone areas) and postdisaster policy decisions (e.g., the design of government assistance programs). As such, aggregate metrics cannot represent all households and often obscure significant differentials across specific population subgroups (Bolin 1985; Hirayama 2000; Hamideh et al. 2021; Peacock et al. 2018).

Various approaches have been taken to integrate household displacement and return into disaster recovery models. While physical damage to the built environment has long been considered a driver of household displacement, more recent models incorporate additional factors such as financial autonomy, household demographics, and attachment to place. Although significant improvements have been made, there are ongoing challenges in modeling household displacement and return in disasters. Some of these research opportunities include model validation and review across different contexts, robust incorporation of multifamily and rental housing, and consideration of the correlations between physical and social vulnerability factors.

All data generated or analyzed during the study are included in the published paper.

This work was partly funded by the University College London Overseas Research Scholarship (ORS) and the Willis Towers Watson Research Network.