Recent climate disasters serve as a reminder of the growing—yet overlooked—risk of climate-driven displacement in the Global North. This paper contributes to a nascent literature on disaster-induced mobility in high-income countries by extending the evidence to a new context: Australia. Applying propensity score matching to panel data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey, we conduct the first causal assessment of the impact of home damage caused by extreme weather events on residential mobility in Australia. Our findings suggest that from 2009 to 2022, an annual average of 1.6% of Australians aged 15 + (or ~ 308,000 people a year) experienced home damage caused by floods, cyclones or bushfires. Such damage increases the probability of changing address within 1 year by 56%, displacing an annual average of 22,261 Australians. Cumulatively, this amounts to ~ 312,000 people displaced by climate-induced home damage between 2009 and 2022. Importantly, this type of climate-induced mobility is not evenly spread across the population. Contrary to findings from the Global South, we find no evidence of “entrapment effects”, except for uninsured homeowners. Instead, our results indicate that over 80% of climate-displaced Australians come from the bottom two income quartiles, with the poorest 3% accounting for 14% of the displaced population. The most disadvantaged Australians thus face a double vulnerability: they are both more likely to sustain home damage from extreme weather events and more likely to be displaced. These findings bear important implications for adaptation strategies and policy responses to natural disasters.
Notes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Introduction
With climate change leading to more frequent and severe extreme weather events (AghaKouchak et al., 2020), disaster-induced population movement is becoming a topic of growing interest within academic, media and policy circles. Although the scale of the phenomenon is difficult to assess (Beyer et al., 2023; Hugo, 1996), there are numerous examples of large-scale displacement following floods, wildfires and cyclones. For example, over 500,000 people were displaced after Hurricane Katrina in 2005 (Gabe et al., 2005) and 7 million people following the Pakistani floods of 2010 (Din, 2010). There is broad consensus that the impact of natural hazards on residential mobility tends to be short-lived and occurs over short distances (Black et al., 2011; Findlay, 2011). This occurs because most disaster-affected residents express a desire to stay in their places of residence (Tinoco, 2023; Berlin Rubin and Wong-Parodi 2022; Sharygin, 2021), a preference often reinforced by reconstruction activity and policy (Huang et al., 2022). Yet there are historical cases of settlement abandonment (Greenberg et al., 2007; McLeman, 2011) and long-distance population movement (Fussell et al., 2023; Graif, 2016) following extreme weather events. This goes to show the diversity of mobility responses to natural hazards and how these can occur at a range of spatial and temporal scales (Black et al., 2013).
There is also growing recognition that exposure to extreme climate events is not randomly distributed within the population. Rather, low-income and minority populations are often concentrated in areas more prone to natural hazards (such as floodplains) or regions with poor environmental conditions (Tierney, 2020). These spatio-structural inequalities became clear after Hurricane Katrina hit New Orleans in 2005. Because of land-development and residential-segregation patterns, low-income groups and African Americans were disproportionally concentrated in low-lying neighbourhoods and consequently reported higher levels of home damage and higher odds of relocation (Fussell et al., 2010). The hazard-exposure-vulnerability framework (McLeman et al., 2021) conceptualises some of these processes by proposing that climate and environmental risks are shaped by (a) the nature of specific climatic hazards, (b) the exposure of people, resources and systems to such hazards and (c) their vulnerability. Thus, some conditions, such as poverty and inadequate infrastructure, increase the likelihood of being adversely affected by climate-induced hazards (IPCC, 2014). In the context of residential relocation, individuals from less advantaged socio-economic groups, for example, tend to have less robust housing and lack access to emergency services and information. These disadvantages compound with one another, exacerbating the risk of experiencing home damage when disaster strikes and the ensuing likelihood of subsequent relocation. In other words, disaster-induced relocations are rooted in broader social structures and power dynamics (Tierney, 2020).
Advertisement
While evidence from the Global North—largely drawn from the United States—indicates an increased risk of residential mobility following an environmental disaster, particularly among disadvantaged groups, research on the Global South is both greater in volume and more mixed in its findings. Evidence from the Global South shows that rapid-onset climatic disasters can reduce population movement by constraining households’ resources (Mueller et al., 2014), leading to the entrapment of the most vulnerable populations, especially those in the poorest regions (Black et al., 2013; Nawrotzki & DeWaard, 2018). Empirically, this manifests in a negative relationship between income and the probability of moving following extreme climate events (Mueller et al., 2020; Cattaneo et al. 2019). This again highlights the diversity of population movement responses to extreme weather events, which are contingent on the vulnerability of both people and places (Black et al., 2013). Mobility decisions are indeed known to be context-specific and shaped by broader societal conditions, which influence the resources and adaptative capacity of communities affected by climate hazards (Ronco et al., 2023). For example, floods generate higher levels of internal displacement in countries with non-democratic governance, armed conflict and low GDP (Beine & Parsons, 2017; Hoffmann et al., 2023; Vestby et al., 2024). The higher level of disaster-induced displacement in low- and middle-income countries explains the focus of most contemporary climate-migration research on the Global South compared to the Global North (Barbier & Hochard, 2018).
However, as climate change intensifies, we argue that evidence needs to be broadened to countries of the Global North. Indeed, recent weather-related disasters have highlighted substantial vulnerabilities and ensuing population displacement in wealthier and more technologically advanced nations (Black et al., 2013; Muttarak, 2021). This includes evidence from Hurricane Katrina in the United States (2005) and Hurricane Maria (2017) in Puerto Rico (Alexander, Zagheni, and Polimis 2019; Fussell et al., 2010) and the 2013/2014 floods within central Europe (Grams et al., 2014; Župarić-Iljić, 2017). Similarly, the 2017 California bushfires led to an increase in relocation intentions (Tinoco, 2023). Altogether, research on high-income countries is scarcer than on low-income countries, and it remains largely confined to the United States (Cipollina, De Benedictis, and Scibè 2023; Piguet et al., 2018). In this study, we extend the evidence base to a new high-income country: Australia.
Australia constitutes an interesting case study, as its climate is strongly affected by the surrounding oceans and the El Niño-Southern Oscillation (ENSO) phenomenon (IPCC 2022). This leads to repeated floods and prolonged droughts. Examples of recent extreme weather events in Australia include the 2019/2020 megafires that killed 450 people, both directly and indirectly as a result of smoke inhalation (Johnston et al., 2021) and cost around AUS$100 billion (Read & Denniss, 2020). Other examples include the 2021 and 2022 floods in eastern Australia—the most widespread and costly floods in the country’s recorded history (Fryirs et al., 2023). Australia faces projected increases in the intensity of cyclones, sea-level rise and localised high-intensity rainfall, which are in turn expected to lead to greater flood damage and storm-surge height in some areas and significant decreases in rainfall in others (IPCC 2022). Despite the existence and aggravation of extreme weather conditions in Australia, research has rarely considered their impacts on population displacement, and the existing evidence is largely qualitative. Here, we provide novel causal estimates of the level of population displacement associated with home damage caused by rapid-onset climate events in Australia. In doing so, we address three allied research questions: (i) which population groups are more likely to be affected by home damage due to extreme weather events?, (ii) which population groups are more likely to move as a result? and (iii) do these moves exhibit distinct spatial and temporal patterns?
To answer these questions, we draw on panel data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey—a nationally representative household panel of the Australian population aged 15 and over. We focus on the 2009 to 2022 period, when a question on home damage from extreme weather events (e.g. floods, cyclones and bushfires) was included. The HILDA Survey provides a unique opportunity to accomplish our research aims: it captures population movement at a range of spatial and temporal scales, it collects rich information on respondents’ socio-demographic characteristics and it includes an annual question rarely collected in national surveys on home damage caused by floods, cyclones and bushfires in the last 12 months. Because the question does not differentiate between different types of natural hazards, we analyse the joint impact of home damage due to floods, cyclones and bushfires on residential mobility.
Advertisement
Given the challenges associated with climate-induced population movement, providing robust evidence that enables appropriate policy response and anticipatory action for future disasters is paramount (Beyer et al., 2023). However, attempts at drawing robust estimates have been hindered by the fact that exposure to extreme weather events is not random (Fussell et al., 2010). Here, we overcome this issue by leveraging a propensity score matching (PSM) methodology (Stuart & Rubin, 2008) that adjusts the estimates for selection into experiencing home damage due to rapid-onset climate events, yielding more reliable estimates of its causal effect on residential mobility. In doing so, it is important to note that our focus is on establishing the direct impact on residential mobility of home damage caused by disasters. As we discuss later, the overall effect of disasters on residential mobility may be greater, encompassing also indirect impacts operating, for example, through damage to infrastructure, changes in risk perceptions, shifting locational preferences or job losses.
Literature review
In this section, we briefly summarise recent advances in understanding the level and determinants of disaster-induced residential mobility, starting with a discussion on variations in residential mobility responses to disasters by agency, duration and distance. We then juxtapose the hazard-exposure-vulnerability framework (IPCC, 2014) with Tierney’s (2020) critical sociological perspective to unpack the role of social structures in creating differential capabilities to respond to disasters through relocation. Finally, we review evidence from Australia, where disaster-induced mobility research remains limited. In doing so, we focus on rapid-onset climate events and do not engage with the adjacent literature on slow-onset climate change (e.g. studies on drought and sea-level rise). These slow-onset changes affect population movement through different and less direct channels than rapid-onset events (Abel et al., 2019). We focus on examples from the Global North because of similarities in socio-economic structures with Australia, but also draw comparisons research findings from the Global South where relevant.
The temporal and spatial scales of disaster-induced residential mobility
Mobility and displacement are typically conceptualised as two ends on the voluntary-involuntary population movement spectrum (Black et al., 2011), although the terms are sometimes used interchangeably (Askland et al., 2022). Often, ‘displacement’ is equated to forced moves in response to rapid-onset hazards (such as floods, cycles, wildfires and tsunamis), whereas ‘mobility’ is used to denote moves that are proactive, planned and voluntary and aim at improving movers’ livelihoods. Although differentiating voluntary and involuntary population movement is difficult in practice (Hugo, 1996; Koser & Martin, 2011), two separate literatures have emerged (de Sherbinin et al., 2022; Piguet, 2018).
Rooted in sociology and legal studies, the displacement literature typically examines the causes, processes and consequences of forced relocations and highlights issues of power, agency, vulnerability and protection (Cernea, 2000). In contrast, the literature linking residential mobility and disasters—often in the remit of demography and human geography—focuses on the decision-making process, the socio-economic factors that influence residential moves following disasters and natural hazards and ensuing population-level changes in the level and spatial patterns of population movement (Curtis et al., 2015; Fussell et al., 2014a, 2014b). In this paper, we draw on the latter literature, which aligns more closely with our key aims of establishing the level and determinants of residential mobility following home damage caused by an extreme weather event. However, where relevant, we also engage with the displacement literature. The terminology used within the remainder of the paper reflects that used in these literatures.
An important distinction across these bodies of work is the temporal and spatial scales at which mobility and displacement operate. Disaster-induced displacement is typically short-lived. Most people return after environmental stressors ease (Findlay, 2011), as documented in the 2010 Haiti earthquake, where most affected residents returned within 5 months. In contrast, migration is understood as a more permanent relocation (Piguet 2011). Yet there are also well-documented instances of long-term displacement following disasters. For instance, just over half of the individuals affected by Hurricane Katrina had returned after 14 months (Fussell et al., 2010; Sastry & Gregory, 2014). Similarly, approximately 10% of evacuees from Hurricane Andrew, which hit Florida in 1992, were still displaced 4 years on (Smith & McCarty, 1996). There are even historical examples of complete settlement abandonment following repeated extreme climate events, such as successive floods and volcanic eruptions (Greenberg et al., 2007; McLeman, 2011). Collectively, existing studies show significant variation in relocation duration across population subgroups, types of natural hazards and broader societal contexts in which a disaster occurs, which highlights the need to better understand the temporal dynamics of disaster-induced residential mobility (Beyer et al., 2023).
Relocations caused by rapid-onset events are not only short-lived, but also tend to occur over short distances—typically within the same city/county or across neighbouring cities/counties—because of the greater social and monetary costs of moving to more distant locations (Findlay, 2011). Referred to as the ‘distance decay’, this well-known pattern characterises all types of population movement regardless of their motivating factors, duration or position on the voluntary-involuntary spectrum (Stillwell et al., 2016). For example, in the case of the 2017 North California wildfires, most affected residents moved within the same city and very few changed their county of residence (Sharygin, 2021). Yet there are also documented instances of long-distance relocation including the aftermath of Hurricane Katrina. With over 40% of the state of Louisiana’s population affected (Gabe et al., 2005), the limited availability of local housing triggered many long-distance moves (Graif, 2016).
The fact that most post-disaster moves are local and short-term is further supported by evidence of persistent aspirations to remain in place after wildfires (Berlin Rubin and Wong-Parodi 2022) and floods (Holley et al., 2022; Koubi, Freihardt, and Rudolph 2022). Ostensibly, this occurs because of place attachment, reliance on social networks and an orientation towards rebuilding for most post-disaster remedial policies (Amaratunga & Haigh, 2011). There are, however, instances of policies that support long-distance relocation. This was the case for Cyclone Tracy, which hit Darwin in Australia’s Northern Territory in 1974. Because of the scale of destruction and the remoteness of the city, the government ordered the mass evacuation of 80% of the population (Haynes et al., 2011). Most of the affected residents initially relocated to the state of New South Wales some 4000 km away from Darwin (West, 2000), yet 70% of them had returned within 7 years (Britton, 1981; Carson et al., 2021).
Taken together, these diverse examples show that residential mobility responses to disasters occur on a continuum of agency, duration and distance (Black et al., 2013; McLeman & Smit, 2006). While many studies recognise that the level of disaster-induced mobility varies across spatial and temporal scales, challenges in accessing appropriate data to measure population movement at a range of spatial scales remain a challenge. Mobility outcomes are diverse because they are highly context-specific and influenced by multiple factors that operate differentially across time and space (Hunter et al., 2015). In the next section, we review the individual-level, household-level, meso-level and societal-level determinants of disaster-induced mobility.
The determinants of disaster-induced residential mobility
Originally formulated in the Fifth Assessment Report by the Intergovernmental Panel on Climate Change (IPCC, 2014), the hazard-exposure-vulnerability framework conceptualises the determinants of post-disaster population movement by recognising that environmental and climatic risks are jointly shaped by (a) the nature of the hazards, (b) the exposure of people, resources and systems to such hazards and (c) their vulnerability. The framework recognises that some conditions, such as poverty, increase the likelihood of being exposed and adversely affected by climate-induced hazards (IPCC, 2014). To explain variations in mobility responses to disasters, we couple this framework with Tierney’s (2020) critical sociological perspective, according to which the level of risk and vulnerability is rooted in social structures and entrenched inequalities.
Across the world, there are numerous examples of disadvantaged groups living in proximity to environmental risks in order to achieve affordable housing, whereas more advantaged individuals often leave such areas because of perceived risk (Hunter, 2005). In the United States, such spatio-structural inequalities in the risk of exposure to climate hazards are the result of historical land-development and residential-segregation patterns that manifest in low-income and minority groups’ concentration in low-lying suburbs that are more impacted by disasters (Fussell et al., 2010). Heightened exposure to environmental hazards for these groups is often coupled with less robust housing and more limited resources to cope with structural damage and loss of utilities (Smith, McCarty, and Durham 2006). These factors cumulatively contribute to increased odds of moving after a disaster among more disadvantaged groups (Elliott & Howell, 2017).
Similar patterns have been observed in Australia. For instance, historical records from the 1893 floods Brisbane and Ipswich revealed that, since the beginning of European settlement, ‘elevation segregated the classes with the heights overlooking the river occupied by the elite, the middle classes on the slopes and the lower classes valleys and flats’ (Cook, 2023: 20). This pattern of socio-spatial segregation continues today, with flood-exposed populations in the 2016/2017 Lismore floods in Northern New South Wales exhibiting comparatively higher levels of socio-economic vulnerability (Rolfe et al., 2020). Similarly, disadvantaged communities were comparatively overexposed to the 2019/2020 Australian bushfire that burnt over 19 million hectares across the eastern seaboard (Akter and Grafton 2021), although the ensuing impact on residential mobility is yet to be quantified.
Another source of heterogeneity in residential mobility responses to extreme climate events is well-established individual- and household-level determinants of voluntary population movements, such as age, housing tenure and duration of residence, with young adults, renters and recently arrived residents being more likely to move (Thomas et al., 2016; Bernard, Bell, and Charles‐Edwards 2014; Morrison & Clark, 2016). This pattern of selectivity holds for disaster-induced mobility behaviour and intentions. For instance, following the Calgary floods of 2013 in Canada, older residents and homeowners were comparatively more likely to stay in flood-affected areas (Haney, 2019).
Disaster-induced mobility is also shaped by a series of interlocked meso-level factors that act as constraints or facilitators (Black et al., 2011). These include community support networks, local government policies (e.g. evacuation orders and rebuilding regulations), access to services and infrastructure and economic opportunities (including those offered through reconstruction efforts). Of particular relevance is access to social networks, which provide essential support post-disaster and tie people to places (Thiede & Brown, 2013). Yet large-scale disasters that affect entire communities, such as Hurricane Katrina, severely reduce reliance on friends or family, which contributed to large-scale relocation (Graif, 2016), particularly among the most disadvantaged groups who are generally more reliant on kinship support (Clarke & Wallsten, 2003). Mobility responses triggered by home damage occur alongside indirect mobility responses to infrastructure damage, which can hinder access to essential services and changes to risk perceptions (Zander & Garnett, 2020).
Recovery policies are another important meso-level determinant of disaster-induced mobility. Following Hurricane Harvey, which hit southern Texas in 2017, flood-insured residents received higher payouts than uninsured residents. The latter were minimally compensated through the federally run ‘Individuals and Households Program’ and consequently suffered higher indebtment than insured households—whose financial recovery was quicker (Rhodes and Besbris 2022a). This policy—coupled with the absence of large-scale buy-back schemes allowing homeowners to voluntarily leave their properties—encouraged residents to stay in neighbourhoods prone to natural hazards instead of relocating (Rhodes and Besbris 2022b). Consequently, disadvantaged social groups may become ‘trapped in place’, being too poor to meet the financial costs of moving when a new natural disaster occurs, particularly in locations where insurance markets are underdeveloped or inaccessible (Cattaneo et al. 2019). This situation has led to growing scholarly attention on the concept of ‘trapped populations’, recognising that ‘immobility may be of equal or greater concern than migration in the context of climate change’ (Skeldon, 2024: 4). To date, most examples of trapped populations are found in the Global South (Nawrotzki & DeWaard, 2018), but discussions of the impact of under-insurance and un-insurability on residential mobility are growing (Blake et al., 2022).
Altogether, the evidence introduced within this section shows that individual, household, community and societal factors interact in complex ways to create differential capabilities to respond to disasters, with such responses being ultimately rooted in broader social structures (Tierney, 2020). With these theoretical tenets in mind, we explore mobility responses to home damage due to extreme weather events in Australia. In the next section, we discuss relevant aspects characterising the Australian context.
The Australian context
Our focus on Australia is motivated by the unequal distribution of empirical studies on climate and disaster-induced population moment, which remains a barrier to the production of comprehensive and balanced global assessments of the links between climate change and population movement (Blicharska et al., 2017). Existing research is typically conducted by researchers from the Global North based on empirical evidence from the Global South, which may stem from wealthy countries historically perceiving themselves as being ‘immune’ to climate-induced displacement (Mullingan et al. 2014). Within scholarship on the Global North, there is also a marked geographic imbalance. In a recent review of 463 empirical studies, Piguet et al. (2018) identified only one contribution from Australia and a handful from Europe, compared to 62 contributions from the United States.
Yet Australia constitutes a meaningful case study in its own right. Particularly, as noted earlier and compared to other countries of the Global North, the country faces more recurrent extreme weather events. Among these, floods and wildfires are the most common natural disasters, followed by cyclones (BOM and CSIRO 2022). Because of an increase in dry weather—in both length and intensity —wildfires are becoming larger and more frequent, particularly in Southern Australia. The Australian Black Summer Fires of 2019/2020, which caused an estimated direct and indirect 450 deaths (Johnston et al., 2021), epitomise this trend. At the same time, heavy rainfall events have also become more frequent and intense. This is demonstrated by the large-scale flooding of Southeast Queensland and coastal New South Wales at the beginning of 2022, which resulted in over AUS$3.35 billion in insurance claims (ISA, 2022). These rapid-onset climate events have occurred in concert with slow-onset climate change, particularly drought and heatwaves, which have increased in intensity, frequency and duration over the last 70 years (Trancoso et al., 2020). At the same time, with 15% of its population changing place of residence each year, Australia is one of the most mobile countries in the world (Bell et al., 2015). As such, one would expect mobility responses to climatic and environmental hazards to occur on a larger scale in Australia than in countries that record lower levels of residential mobility.
Climate-driven mobility research in Australia has been focused on the impact of droughts because of the scale of the problem (Hugo 2012; Vidyattama et al., 2016). Research on mobility responses to rapid-onset climate events remains limited, but there is evidence that residing in a region recently affected by floods, cyclones or bushfires is associated with an increase in mobility intentions (King et al., 2014), although environmental factors remain secondary to economic and lifestyle considerations (Zander et al., 2020). Because of worsening extreme weather events, there is growing concern that home insurance is becoming unaffordable or unavailable in some parts of Australia, which may lead to some groups being trapped in place (Hutley et al., 2022). This situation has led to the emergence of buy-back schemes, whereby the government buys homes from those most severely impacted by recent floods and who are at the greatest risk of future flooding. These schemes should theoretically facilitate out-migration (Durand-Delacre et al. 2023; Piggott-McKellar & Vella, 2023). Indeed, evidence from the United States suggests that, in the absence of buy-back schemes, under-insured individuals are less mobile than their insured peers (Rhodes and Besbris 2022a). Nonetheless, research on insurance coverage and disaster-induced mobility remains lacking in Australia. More generally, Australian research on climate-induced mobility remains descriptive and based on bespoke cross-sectional surveys in selected regions and does not account for the fact that exposure to extreme weather events is not random. As climate change intensifies, nationally representative and robust estimates of climate-induced mobility for new countries, including Australia, are urgently needed. The present study offers a novel contribution by providing the first estimate of the level of residential mobility caused by disaster-induced home damage in Australia and by identifying the socio-economic determinants of this type of population movement.
Data and methods
The Household Income and Labour Dynamics in Australia (HILDA) Survey
We harness 14 years of panel data from the HILDA Survey, a high-quality, multipurpose study collecting longitudinal information from Australian households since 2001 (Watson & Wooden, 2012). Based on a complex probabilistic sampling design, the HILDA Survey is representative of the Australian population aged 15 and older. Further, the survey boasts a high wave-on-wave retention rate, of 90–95% (Summerfield et al., 2021). As a result, panel attrition has had a limited impact on internal migration estimates using the HILDA Survey (Sander & Bell, 2014), which are comparable to those from the national population census (Kalemba et al., 2022; Watson, 2020). Indeed, the HILDA Survey has been used widely in internal migration and residential mobility research for a broad range of topics, including trends in and determinants of population movement (Campbell, 2019; Crown et al., 2020; Perales & Bernard, 2023), its social and economic impacts (Clark & Lisowski, 2019; Korpi & Clark, 2017) and its associations with life-course transitions (Bernard et al., 2016; Clark & Lisowski, 2018; Sander & Bell, 2014; Vidal et al., 2017).
Since 2009, HILDA Survey respondents have been asked the following question on an annual basis: “In the last 12 months, has a weather-related disaster (e.g. flood, bushfire, cyclone) damaged or destroyed your home?” We use responses to this survey item to derive a variable capturing home damage due to extreme weather events (i.e. floods, cyclones and bushfires) (‘yes’/ ‘no’). This survey item has been previously used to establish the impact of extreme weather–related home damage on health and well-being (Gunby & Coupé, 2023; Li et al., 2023). To our knowledge, it has never been used to assess the impact of home damage caused by extreme weather events on residential mobility. This represents the key aim of the present study. It is important to acknowledge that, in using this survey item, we only focus on the direct impact of disasters caused by home damage and do not consider the indirect impacts of exposure to disasters on residential mobility—such as damage to infrastructure (e.g. roads and schools), changes in risk perceptions, shifts in locational preferences and job loss. In addition, this survey item does not allow us to distinguish between the type of natural hazard or to ascertain the degree of damage a home sustained during a weather event. We return to these points in the discussion section.
Climate-induced home damage is one of multiple life-course events for which information is collected annually within the HILDA Survey, along with others such as marriage, pregnancy/childbirth, separation, employment loss and retirement. To benchmark the impact of climate-induced home damage on residential mobility, we compare its estimated effect to those of these well-established determinants of relocation (Bernard, Bell, and Charles‐Edwards 2014; Rindfuss, 1991; Mulder, 1993).
For our purposes, we define residential mobility as a change of address between two consecutive survey waves within an unbalanced panel structure. This information is obtained from an annual binary question: “did you change address of residence in the last 12 months” (‘yes’/ ‘no’). Between 2009 and 2022, an annual average of 12.8% of the Australian population aged 15 + years changed address. The annual average for those who experienced climate-induced home damage was 19.5%, suggesting that this event may be a trigger of residential mobility. However, careful modelling is required to draw firm conclusions, as we discuss in the next section.
Estimation strategy
The analysis proceeds in three sequential steps. First, we calculate the annual share of the Australian population exposed to extreme weather–related home damage. We do so for an unbalanced sample of 192,790 person-year observations from 23,522 individuals. By applying population weights derived by the HILDA Survey team (Summerfield et al., 2021), we report results that are nationally representative.
Second, we use a logistic regression model to identify the determinants of extreme weather–related home damage in Australia. The explanatory variables include an encompassing range of economic, demographic, locational and social characteristics used in previous studies (Korpi & Clark, 2017; Pelikh & Kulu, 2018; Perales & Bernard, 2023). These include respondents’ age, gender, immigrant status, marital and parental statuses, household income, educational attainment, duration of residence, state of residence, survey year and metropolitan status. We also consider housing tenure, distinguishing between renters, insured homeowners and uninsured homeowners.1 Appendix A presents descriptive statistics on all analytic variables. To accommodate repeated observations from the same individuals and account for the nesting of individuals within households, the model’s standard errors are clustered on both individuals and households.
Finally, we estimate the impact of extreme weather–related home damage on changes of address within the year. Given the possible endogeneity of experiencing home damage caused by extreme weather events, we use a matching approach to reduce selection bias. There is no consensus on the best method, but matching techniques aim to maximise the balance between the treated group (home damage) and the control group (no home damage) on the pre-treatment variables and minimise the number of observations removed from the dataset (King et al., 2011). To select the most suitable approach for the data at hand, we assess four well-established methods by comparing (i) the number of observations kept post-matching and (ii) the balance between the treated and control groups, which we measure as the mean absolute standardised difference across all covariates (see King & Nielsen, 2019). Following best practice in quasi-experimental designs, we include a large set of covariates that are associated with the exposure and outcome variables—namely, all variables used as predictors in the logistic regression model described before. The results in Table 1 show that all matching methods improve the balance between the treated and untreated groups. While coarsened exact matching (CEM) performs the best in terms of balance, it reduces the sample size by over 25%. This is a known problem that has led some authors to caution against the use of CEM despite its desirable statistical properties (Lacus, King, and Porro 2012), particularly when using datasets with rich covariate information (Ripollone et al. 2020; Wang 2021) as is the case here. In contrast, propensity score matching (PSM) offers the second-best balance after CEM without reducing the sample size. Although PSM has been criticised for reliance on model specifications and the risk of biased estimates when the model is misspecified (King & Nielsen, 2019), we found that it yields the optimal bias-variance trade-off for our data and therefore opt for this matching technique for our analyses. The results are reported as average treatment effects (ATEs), which denote the difference in residential mobility between the treated (home damage) and control (no home damage) groups after balancing the covariates between the two groups, expressed in percentage points, making the comparison closer to a randomised experiment.
Table 1
Balance and sample size for the raw sample and the data used by selected matching techniques
Balance
Sample size
Average of the absolute standardised mean difference across all covariates (MASD)
Number of observations matched
Raw sample
0.087
3111
Mahalanobis distance nearest neighbour matching
0.074
3111
Mahalanobis distance kernel matching
0.071
3018
One-to-one propensity score matching
0.023
3111
Coarsened exact matching
0.000
2309
Notes: \(\text{MASD}=\frac{1}{K}{\sum }_{k=1}^{K}|{\text{SMD}}_{k}|,\) where K is the number of covariates. \(\text{SMD}= {~}^{{\overline{X} }_{T}-{\overline{X} }_{C}}\!\left/ {\sqrt{{{(S}_{T}^{2}+{S}_{C}^{2})}\left/ \!{~}_{2}\right.}}\right.\), where \({\overline{X} }_{T}\) and \({\overline{X} }_{c}\) are the means of the covariates for the treated and control groups, respectively, and \({S}_{T}\) and \({S}_{C}\) are the standard deviations. Lower MASD values denote better balance between the treated and control groups. Higher sample sizes denote fewer observations being lost in the matching
To detect heterogeneity in mobility responses to extreme weather events and better understand the temporal dynamics of disaster-induced residential mobility, we replicate the PSM analysis at different spatial and temporal scales. We do so by measuring residential mobility using different distance thresholds. For each change of address, the distance moved is calculated based on the great circle formula (Thomas et al., 2019), which accounts for the curvature of the Earth (Small, 2012). Given the vast scale of Australia, this approach is more accurate than a basic Euclidean distance calculation, which gives a straight-line distance and ignores the spheric shape of Earth. From this variable, we construct 20 overlapping distance-based measures of population movement, ranging from ‘up to 1 km’ to ‘up to 350 km’. Given that most moves occur over short distances (Lomax, Norman, and Darlington‐Pollock 2021; Thomas et al., 2019), we define shorter distance-based measures of mobility in 5-km increments from 5 to 65 km since 65 km is the threshold at which employment reasons begin to outweigh housing reasons for moving in Australia (Thomas et al., 2019). We then use 10-km increments from 70 to 100 km, followed by moves up to 200 km and 350 km.
We then measure residential mobility at four different observation intervals: the year disaster-induced home damage occurred and up to 3 years later in 1-year increments. We derive these measures by comparing the census collection district of residence the year before home damage occurred to the census collection district of residence 1, 2 and 3 years later. The census collection district is the finest geographic scale at which the place of residence is made available in the HILDA survey. The 38,700 census collection districts that cover Australia comprise an average of 255 dwellings and are analogous to census tracks in the United States.
Finally, we replicate the PSM analysis of our key residential mobility measure (i.e. change of address in the last 12 months) for different sub-population groups. Specifically, we distinguish between homeowners (with and without insurance) and renters, as well as between low- and high-income households. We assess differences between these groups by comparing the direction of the estimated impact of climate-induced home damage (i.e. whether it increases or decreases residential mobility) and by considering the overlap (or lack thereof) in confidence intervals. The focus on housing tenure is motivated by (i) our event of interest (i.e. disaster-induced home damage) being tightly connected to house ownership, (ii) well-established findings in both the voluntary and forced mobility literatures that renters are more mobile than homeowners (Clark & Lisowski, 2019) and (iii) recent research indicating that flood insurance payouts ground people in place (Rhodes and Besbris 2022a). We also explore variations by socio-economic status, using both self-reported and income-based measures, because of the greater vulnerability of disadvantaged groups to extreme weather events as discussed in the “Literature review” section.
Empirical evidence
Level and determinants of exposure to climate-induced home damage
We begin our empirical analysis by ascertaining the annual share of individuals affected by home damage due to extreme weather events. From our weighted sample, we find that between 2009 and 2022, an average of 1.6% of the Australian population aged 15 + years was affected by extreme weather–related home damage every year. At the population level, this corresponds to an annual average of 308,133 individuals affected during the observation period.
The multivariate logistic regression model in Table 2 identifies the individual and household-level factors associated with exposure to extreme weather–related home damage. To facilitate interpretation, the model parameters are presented as both odds ratios (ORs) and marginal changes in predicted probabilities (PPs), with covariates held at the sample mean. The results show a socio-economic gradient in exposure to climate-induced home damage. All else being equal, sampled individuals in the top (OR = 0.89, p > 0.10, PP = 1.53%) and second-top income quartiles (OR = 0.91, p > 0.10, PP = 1.57%) are significantly less likely to be affected than individuals in the bottom income quartile (PP = 1.72%). However, these income-quartile differences were not statistically significant and thus not extrapolatable to the population. To further explore this socio-economic gradient, we replaced income quartiles with a self-reported measure of economic prosperity based on a five-point scale: ‘very poor’, ‘poor’, ‘just getting along’, ‘reasonably comfortable’ and ‘very comfortable’.2 The results in Appendix C confirm an increased risk of climate-induced home damage among the most disadvantaged groups, with the differences being statistically significant. Ceteris paribus, the probability of experiencing climate-induced home damage is 3.6% and 3.1% for individuals self-classifying as ‘very poor’ and ‘poor’, respectively, compared to 1.5% and 1.3% for those self-classifying as ‘reasonably comfortable’ and ‘very comfortable’.
Table 2
Logistic regression model of extreme weather–related home damage, main results
Odds ratio
Standard errors
Predicted probability
95% confidence interval
Socio-demographic characteristics
Age (ref. cat. 15–24)
1.67
1.45
1.91
25–44 years
1.01
[0.87,1.17]
1.69
1.56
1.83
45–64 years
1.01
[0.85,1.20]
1.69
1.56
1.82
65 + years
0.79*
[0.63,0.96]
1.13
1.14
1.48
Sex (ref. cat. Male)
1.65
1.55
1.75
Female
0.96
[0.89,1.04]
1.59
1.49
1.67
Marital status (ref. cat. married or partnered)
1.64
1.54
1.73
Divorced or separated
0.99
[0.86,1.15]
1.63
1.42
1.83
Never married
0.94
[0.81,1.10]
1.54
1.35
1.73
Dependent children (ref. cat. no)
1.45
1.231
1.58
Yes
1.18*
[1.04,1.34]
1.70
1.59
1.80
Foreign-born (ref. cat. no)
1.64
1.55
1.72
Yes
0.93
[0.82,1.05]
1.52
1.36
1.69
Housing tenure (ref. cat. insured homeowner)
1.63
1.54
1.73
Uninsured homeowner
1.57***
[1.15,2.14]
2.37
1.67
3.06
Renter
1.07
[0.96,1.19]
1.53
1.40
1.66
Socio-economic status
Income quartile (ref. cat. lowest quartile)
1.52
1.51
1.87
Second quartile
0.98
[0.85,1.11]
1.51
1.51
1.79
Third quartile
0.92
[0.79,1.07]
1.42
1.42
1.71
Highest quartile
0.92
[0.79,1.09]
1.41
1.41
1.71
Educational attainment (ref. cat. no tertiary education)
1.67
1.59
1.76
Tertiary education
0.85***
[0.76,0.96]
1.43
1.39
1.57
Duration of residence (ref. cat. less than 1 year)
1.77
1.59
1.96
1 to 4 years
0.96
[0.85,1.08]
1.70
1.57
1.82
5 to 9 years
0.87
[0.75,1.01]
1.55
1.39
1.71
10 + years
0.85*
[0.74,0.98]
1.51
1.38
1.63
Insufficient information
0.74
[0.43,1.26]
1.32
0.06
1.99
Locational characteristics
Urban status (ref. cat. metropolitan)
1.30
1.21
1.38
Non-metropolitan
1.63****
[1.47,1.80]
2.08
1.94
2.22
State/territory (ref. cat. Victoria)
1.11
0.98
1.25
New South Wales
1.77***
[1.53,2.05]
1.96
1.80
2.11
Queensland
2.16***
[1.87,2.50]
2.38
2.19
2.56
South Australia
0.59***
[0.46,0.75]
0.66
0.52
0.80
Western Australia
1.00
[0.81,1.24]
1.12
0.92
1.31
Tasmania
0.79
[0.57,1.08]
0.89
0.63
1.11
Northern Territory
1.83***
[1.19,2.82]
2.02
1.21
2.83
Australian Capital Territory
1.45
[0.99,2.10]
1.61
1.05
2.16
Number of observations
192,780
Number of individuals
23,522
Log likelihood
− 15,131
Notes: HILDA Survey, 2009–2022. The model controls also for survey year (parameters not shown)
***p < 0.001; **p < 0.01; *p < 0.05
In contrast, tertiary-educated individuals3 (OR = 0.85, p < 0.01, PP = 1.43%) are significantly less likely to be exposed to home damage than non-tertiary-educated individuals (PP = 1.67%). Concerningly, insurance coverage also appears to be a strong independent determinant of exposure (OR = 1.57, p < 0.001). Ceteris paribus, uninsured homeowners have a 2.4% chance of experiencing extreme weather–related home damage compared to 1.6% among insured homeowners. This may be due to increasing premiums on climate-unsafe areas coupled with the emergence of uninsurable ‘danger’ zones (CCA, 2022).
Age and life-course stage also seem to exert significant influences on the propensity to sustain climate-induced home damage. For example, parents with dependent children (OR = 1.18, p < 0.05, PP = 1.70%) are more likely to be affected than individuals without children (PP = 1.45%), whereas individuals aged over 65 years (OR = 0.79, p < 0.05, PP = 1.13%) are significantly less likely. In addition, longer durations of residence are significantly associated with a lower likelihood of sustaining climate-induced home damage. However, we observe no significant differences between renters and insured homeowners or between native and foreign-born individuals. These variations are overlaid by broader spatial inequalities. Specifically, individuals in non-metropolitan Australia and those in Queensland, New South Wales and the Northern Territory are significantly more likely to be exposed to home damage due to extreme weather events, all else being equal.
Overall, these results point to systematic differences in the propensity for different population groups to sustain home damage from extreme climate events, with some vulnerable groups being comparatively overexposed. This pattern of results underscores the usefulness of applying PSM techniques to obtain robust estimates of climate-induced residential mobility in Australia.
Climate-induced home damage and residential mobility
Having examined the extent and determinants of climate-induced home damage in Australia, we now analyse its impacts on residential mobility using a PSM methodology. We express the results using average treatment effects (ATEs), which represent the average difference in residential mobility between the treated (i.e. individuals who experienced home damage caused by extreme weather events) and the untreated (i.e. individuals with no such experience) after balancing the covariates between the two groups, expressed in percentage points. The estimated ATE for climate-induced home damage amounts to 0.073, indicating that—on average—having sustained extreme weather–related home damage increases the odds of moving by 7.3 percentage points within a year (95% confidence interval [CI] 5.22–9.34 percentage points), which corresponds to a 56% increase. Considering the exposure rate to climate-induced damage, we infer that—at a population level—an average of 22,261 Australians aged 15 + were displaced every year between 2002 and 2009 (95% CI 15,952–28,561). This corresponds to 0.9% of all changes of address recorded in Australia during this period.
To further contextualise the magnitude of the reported effect, we applied an analogous PSM approach to identify the impacts of other life-course events on the propensity to migrate. These additional analyses, shown in Fig. 1, reveal that the estimated effect of climate-induced home damage on residential mobility (+ 7.28 percentage points; 95% CI 5.22–9.34 percentage points) is comparable to those of other life-course events that have received greater scholarly attention. This is the case for marital separation (+ 10.24 percentage points; 95% CI 7.77–12.70 percentage points) and childbirth (+ 5.78 percentage points; 95% CI 3.51–8.05 percentage points), which exhibit overlapping confidence intervals with climate-induced home damage. Importantly though, the estimated effect of climate-induced home damage is greater than those of job loss (+ 1.82 percentage points; 95% CI 0.55–3.09 percentage points), retirement (not statistically different from zero) and marriage (not statistically different from zero). This finding serves to highlight the importance of considering climate events in residential-mobility research.
Fig. 1
Average treatment effect with 95% confidence interval by life-course event. Notes: HILDA Survey, 2009–2022. The average treatment effect is the average difference in percentage points in residential mobility between the treated (i.e. individuals who experienced home damaged caused by extreme weather events) and the untreated (i.e. individuals with no such experience) after balancing the covariates between the two groups. Statistical significance: ***p < 0.001; **p < 0.01; *p < 0.05
Spatio-temporal dynamics
To assess how far people move, we replicate the PSM analysis for climate-induced home damage using different mobility-distance thresholds. Results in Fig. 2a reveal a strong distance decay, with most moves motivated by climate-induced home damage occurring over short distances. This finding is consistent with well-established patterns in residential mobility (Stillwell et al., 2016) and environmental mobility (Piguet et al., 2011) literatures. More specifically, the odds of moving following climate-induced home damage decrease rapidly from 1 km (ATE = 0.069) to 25 km (ATE = 0.016), after which the ATEs stabilise. Because 65 km is the distance at which mobility begins to affect social networks in Australia (Lomax, Norman, and Darlington‐Pollock 2021), we conclude that most climate-induced moves are short-distance and do not disrupt the social networks of disaster-affected residents.
Fig. 2
Average treatment effect with 95% confidence interval by distance moved and time since disaster. Notes: HILDA Survey, 2009–2022. The average treatment effect is the average difference in percentage points in residential mobility between the treated (i.e. individuals who experienced home damaged caused by extreme weather events) and the untreated (i.e. individuals with no such experience) after balancing the covariates between the two groups. All ATEs are statistically significant (p < 0.05) for a. Statistical significance for b: ***p < 0.001; **p < 0.01; *p < 0.05
Next, we analyse temporal dynamics by tracking respondents’ census collection district of residence for up to 3 years after home damage due to an extreme weather event was recorded. Results in Fig. 2b show a rapid decrease in the odds of living in a new location over time. However, the impact of home damage remains statistically significant, which indicates that up to 3 years after the event, some residents are still displaced. Importantly, the level of residential mobility stabilises after 1 year, with ATEs ranging from 0.023 to 0.032. This pattern of results indicates that approximately 40% of individuals who moved following climate-induced home damage were still living in a different census collection district 3 years on.4 It also suggests that the majority of impacted residents returned within 12 months, yet results remain tentative due to large and overlapping confidence intervals.
Socio-economic gradient in residential mobility responses to climate-induced home damage
To capture possible heterogeneity between sub-population groups in residential mobility responses to climate-induced home damage, we replicate some of the analyses presented above by key population groups. For parsimony, these subgroup analyses are based on the most granular change-of-address measure of mobility recorded in the year of a disaster. Since the risk of sustaining climate-induced home damage was shown to vary by housing tenure, insurance coverage and socio-economic status (see Table 2), the analyses in this section focus on those variables.
Consistent with expectations, the estimated effect of climate-induced home damage on residential mobility is larger for renters (ATE = 0.17) than homeowners (Fig. 3a). This is consistent with the well-established finding in the broader residential-mobility literature that home ownership constrains mobility by increasing the costs of moving (Jia et al., 2023). There are, however, notable differences between homeowners depending on their insurance coverage. Of particular concern is the negative ATE for uninsured homeowners (ATE = − 0.031), for whom experiencing climate-induced home damage decreases the odds of moving. In comparison, the ATE for insured homeowners is positive (ATE = 0.020). This pattern of effects signals a possible risk of “entrapment” for uninsured individuals (Rhodes and Besbris 2022a), who may neither have the means to move nor to rebuild their homes.
Fig. 3
a–c Average treatment effect with 95% confidence interval by housing tenure and socio-economic status. Notes: HILDA Survey, 2009–2022. The average treatment effect is the average difference in percentage points in residential mobility between the treated (i.e. individuals who experienced home damaged caused by extreme weather events) and the untreated (i.e. individuals with no such experience) after balancing the covariates between the two groups. Statistical significance: ***p < 0.001; **p < 0.01; *p < 0.05
Analogous analyses by income quartiles in Fig. 3b reveal that low-income earners from the bottom two quartiles (ATEs = 0.075 and 0.139, respectively) are more likely to be displaced when experiencing climate-induced home damage than high-income earners (ATEs = 0.026 for the top two income quartiles). The ATEs are not statistically significant for individuals within the top two income quartiles, which means that, on average, high-income earners do not move when experiencing home damage. We also replicated the analyses using the self-reported measure of economic prosperity based on a five-point scale, with even more patterned results (see Fig. 3c) confirming significant socio-economic inequalities in the risk of displacement. Specifically, the estimates reveal a linear trend, from an ATE of 0.215 for individuals who self-identify as ‘very poor’ to an ATE of 0.029 for individuals who self-identify as being ‘very comfortable’. For the latter group, the ATE is not statistically significant, indicating that their propensity to move is not affected by climate-induced home damage.
The uneven burden of climate-induced home damage
To obtain a full picture of how climate-induced displacement affects individuals from different socio-economic strata, Fig. 4 juxtaposes the results in the previous section against the results for the risk of exposure presented earlier. The X-axis shows the predicted probability of exposure to climate-induced home damage by self-reported socio-economic status, whereas the Y-axis shows the average treatment effect of climate-induced home damage for each group. Both axes intersect at the sample mean, with the bubble size representing the share of each group in the population.
Fig. 4
Risk of exposure against risk of displacement by self-reported socio-economic status. Notes: HILDA Survey, 2009–2022. The predicted probability of experiencing climate-induced home damage is obtained from the regression model in Appendix E. The ATE is obtained from Fig. 2b. The bubbles represent the size of each group in the population, as reported in Appendix A
The relative location of the different socio-economic groups clearly demonstrates that climate events hit the most vulnerable hardest. All else being equal, socio-economically disadvantaged groups are both at a greater risk of sustaining climate-induced home damage and at a greater risk of being displaced (net of the effect of differences in socio-demographic attributes). For those self-classifying as being ‘very poor’, the predicted probability of experiencing climate-induced home damage is 3.6% and the likelihood of moving, as a result, is 21.5 percentage points higher than in the absence of home damage. Similarly, 3.1% of those self-classifying as ‘poor’ experience climate-induced home damage, and the odds of being moving as a result is 17.2 percentage points higher than in the absence of home damage. In contrast, people self-classifying as ‘very comfortable’ are nearly three times less likely to be exposed (1.3%) and 7 times less likely to be displaced (2.9 percentage point increase in residential mobility). The most disadvantaged Australians thus face a “double vulnerability”: They are both more likely to sustain home damage from extreme weather events and more likely to be displaced as a result.
Using the information in Fig. 4, it is possible to compare the share of the overall and displaced populations by their self-reported socio-economic status. Results in Appendix D show that the bottom 3% of Australians account for 14% of the population displaced due to climate-induced home damage, while the top 19% of the population account for just 5% of those displaced. Results by income quartile confirm the existence of substantial socio-economic inequalities. The bottom two income quartiles represent 50% of the total population but 80% of the displaced population, whereas the top two income quartiles represent 50% of the total population but just 20% of the displaced.
Discussion
In response to calls to expand the geographic coverage of research on population movement caused by natural hazards (Piguet et al., 2018), this paper has provided the first casual estimate of the level of residential mobility triggered by home damage caused by sudden-onset climate events in Australia. The choice of case-study country was motivated by the scale of extreme weather events in Australia, where floods, wildfires and cyclones are increasing in intensity (BOM and CSIRO 2022) and by access to high-quality, nationally representative panel data. Our findings confirm four well-established patterns while providing new insights for a high-income country context, with ensuing implications for policy and future research.
First, our findings confirm that most housing-damage-induced moves take place over short distances and, thus, do not significantly disrupt social networks. We found that experiencing climate-induced home damage increases the odds of moving up to 1 km within a year by 6.9 percentage points, compared to just 1.6 percentage points for moves of up to 25 km. This finding has important implications for future research. Most datasets, including censuses, do not provide distance-based measures of population movement and define internal migration as a change of administrative unit, often using first-tier units, such as states (Thiede et al., 2016), and second-tier units, such as counties (Fussell et al., 2014a, 2014b). As our results show, most climate-induced moves will be missed in studies relying on coarse geographies which are commonly used in climate-induced mobility research (Hoffmann et al., 2023). In the absence of a distance-based measure of population movement, future studies should endeavour to draw on lower-level spatial units that ideally correspond to the neighbourhood level or below to capture the full scale of climate-induced population movement. The fact that most moves occur over short distances raises questions about whether displaced residents will be exposed to similar natural hazards in the future. This observation deserves further attention, given the recurrency of floods and cyclones within certain areas of Australia. For example, the city of Lismore in Northern New South Wales recoded 138 floods in the past 152 years, including 26 major floods (Callaghan & Power, 2014; Morrison, 2023).
Second, the impact of sudden-onset climate events on home damage is highly inequitable and follows a clear socio-economic gradient. Tertiary-educated individuals are significantly less likely to experience disaster-induced home damage, which may be the result of greater access to, or use of, information about at-risk locations or greater knowledge about home-resilient design features and funding opportunities. Duration of residence also plays a role, with individuals residing in the same area for a decade or more reporting a lower likelihood of sustaining climate-induced home damage. This may be due to a lack of knowledge about low-lying areas among newcomers, to housing enhancements by long-term residents (e.g. home elevation against floods or bushfire preparedness) or to a progressively tighter housing market that increasingly places a cost premium on safe areas. Importantly, low-income earners are more likely to experience climate-induced home damage and more likely to be displaced than high-income earners. This may occur due to a concentration of low-income households in low-lying or wildfire-prone areas, an increased likelihood to sustain home damage due to poor quality housing and/or fewer resources for recovery in place—as observed in New Orleans following Hurricane Katrina (Fussell et al., 2010). Australian evidence is limited, but what is available does suggest that disadvantaged communities are more likely to be exposed to both floods (Rolfe et al., 2020) and bushfires (Akter and Grafton 2021). Thus, more work is required to understand the root causes of vulnerability in the Australian context and to identify the social and economic factors that contribute to the creation and exacerbation of risk to natural hazards among the most disadvantaged groups (Tierney, 2020; Rhodes and Besbris 2022b).
The higher risk of disaster-induced mobility among disadvantaged groups runs counter to evidence for less developed countries, where the poorest are often trapped in place because of the lack of resources to move (Ayeb‐Karlsson, Baldwin, and Kniveton 2022). Instead, we found that, in Australia, the poorest 3% account for 14% of the displaced population and the bottom half of the income distribution includes 80% of the population displaced by sudden-onset events. This contrast underscores the importance of considering heterogeneity in the drivers and consequences of climate-induced mobility in different macro-level contexts. In addition, our findings for Australia have important implications for climate-event responses, including the relevance of mean-tested recovery funding. This approach to disaster recovery is currently not the norm within Australia, despite growing recognition that internally displaced Australians need more tailored support (Mortimer et al., 2023).
Third, our findings for Australia revealed one factor that increases the risk of entrapment after climate-inducted home damage, namely, a lack of home insurance among homeowners. Indeed, uninsured homeowners were the only group of those considered that exhibited a decrease in the likelihood of moving after climate-inducted home damage. This is a concerning finding, particularly given forecasts of 1 in 25 dwellings in Australia being uninsurable by 2030 (CCA 2023). Our results therefore suggest that a growing share of the current and future Australian population may be trapped in place because of uninsurance and unable to move after sustaining climate-induced home damage. Indeed, evidence from the United States has shown that disasters increase inequalities, as wealthier individuals and communities are better equipped to recover and rebuild their homes than lower-income populations who face greater challenges and longer recovery times (Rhodes and Besbris 2022b). Our findings call for a deeper understanding of the impact of insurance coverage and government-funded post-disaster schemes on mobility in the Australian context.
Fourth, residential mobility is currently not a common household strategy in response to climate-inducted home damage in Australia. Indeed, our results reveal that most people who sustain such damage remain immobile. Based on a robust and rigorous PSM methodology, we estimate that approximately 22,000 people aged 15 and over were displaced each year from 2009 to 2022—a period characterised by unusually catastrophic and recurrent flooding and wildfires. Importantly, this estimate is substantially lower than what would be concluded by simply glancing at the descriptive statistics. Based on the latter, we would have concluded that 20% of individuals who sustained climate-induced home damage moved houses by the following year. This serves as a reminder of the importance of stringent methodological choices—in this case, PSM over traditional descriptive and regression methods—to derive reliable and realistic estimates of climate-induced residential mobility. Given the ongoing debate about the selection of matching techniques and increasing reliance on coarsened exact matching (King & Nielsen, 2019), future studies on disaster-induced mobility could explore the sensitivity of results to the choice of matching technique—something that has been rarely done in disaster-mobility literature.
While our estimate is nationally representative, only a few communities are affected by extreme weather events every year, and as a result, the impact of displacement is likely to be felt more strongly in some localities. More importantly, climate-induced home damage exerts a significant and long-lasting impact on the subjective well-being of affected individuals (Gunby & Coupé, 2023). Yet it remains unclear whether those displaced due to climate hazards have better outcomes than those trapped in place. These are issues that warrant further investigation.
Our estimates of climate-induced mobility represent just a small fraction of overall population movement in Australia, yet they revealed that the average impact of climate-induced home damage on the odds of moving is comparable in magnitude to the impact of more-widely-researched life-course events such as marital separation and childbirth, but significantly greater than loss of employment, marriage and retirement. However, it is important to remember that our focus is only on the direct impact of housing sudden-onset climate events operating through home damage. The full impact of climate change on residential mobility will be larger when considering the indirect impacts caused by other factors, such as infrastructure damage, increased risk perceptions and changes in livelihood (Black et al., 2011). These are empirically captured in most models by determining whether an individual lived in a neighbourhood, city or county that was affected by a specific extreme weather event. The data at hand did not enable us to assess exposure to extreme weather events without ensuing home damage. Future work could combine the housing-damage question used in this paper with external data on exposure to extreme-weather events. This would help quantify and disentangle the direct and indirect impacts of extreme weather events on residential mobility.
In reflecting on the implications of our findings, it is important to also acknowledge the limitations of this study. First, analyses of the HILDA Survey were restricted by small cell sizes, which prevented us from calculating year-, state- or region-specific estimates. An alternative data source for future research is the Personal Level Integrated Data Asset (PLIDA), a novel administrative longitudinal micro-dataset that provides geographically detailed information on place of residence based on the triangulation of multiple administrative datasets (Bernard et al., 2024; ABS 2024). These new data could potentially be deployed to identify place-based attributes that interact with climate-induced mobility once the reliability of its geo-spatial attributes has been fully assessed. Such a dataset could also help us better understand the temporal dynamics of disaster-induced mobility. Our results tentatively suggest that as many as 40% of residents who moved because of climate-induced home damage were still living elsewhere three years on. By leveraging the full population count offered by PLIDA, future research ought to explore these dynamics in more depth, including spatio-temporal variations in exposure to extreme weather events at smaller scales.
Second, the HILDA Survey does not provide information on the hazard type experienced by respondents as the question on home damage was asked jointly for floods, cyclones and bushfires. This impairs our ability to recognise whether and how climate risks may be shaped by the nature of specific climatic hazards, as posited by the exposure-hazard-vulnerability framework (McLeman et al., 2021). Also missing from the HILDA Survey is information on the scale of climate-induced home damage, which would be required to identify possible thresholds in the impact of sudden-onset climate events on population movement (McLeman, 2018). The available measure only enabled us to estimate the impact of the average home damage on residential mobility. As new data become available, these aspects represent opportunities for future refinement of the findings presented here.
Conclusion
The present study has provided the first causal estimate of climate-induced residential mobility in Australia, where research on the disaster-mobility nexus remains in its infancy. Our findings confirmed the importance of considering how wider societal contexts help drive climate mobility and the need to broaden the evidence base to high-income countries such as Australia, which are increasingly vulnerable to extreme weather events. Given that climate change is expected to intensify in tandem with rapid national population growth (OECD, 2024), more research is urgently needed to fully understand the impacts of climate-induced mobility on both individuals and communities. Among others, policy interventions should attempt to limit exposure among those most impacted, such as low-income residents and uninsured homeowners. In light of our findings, failing to do so will likely result in a growing divide between the rich and the poor, contributing to the exacerbation of existing socio-economic inequalities within the country (Hérault et al., 2024; Sila & Dugain, 2019).
Acknowledgements
This paper uses data from the Household, Income and Labour Dynamics in Australia (HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of Social Services (DSS) and is managed by the Melbourne Institute of Applied Economic and Social Research (Melbourne Institute). The findings and views reported in this paper, however, are those of the authors and should not be attributed to either DSS or the Melbourne Institute.
Declarations
Conflict of interest
The authors declare no competing interests.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
As a caveat, the insurance question in the HILDA Survey combines vehicle, home and content insurance. Therefore, we define uninsured homeowners as individuals with zero expenditure on these three insurance types combined.
Descriptive statistics in Appendix A show that 52.3% of respondents self-classify as being ‘reasonably comfortable’, followed by 25.5% as ‘just getting along’ and 18.9% as ‘very comfortable’. Meanwhile, the most vulnerable groups, ‘poor’ and ‘very poor’, account for 2.5% and 0.7% of the sample.
We obtain this estimate by dividing the ATE the year of housing damage (0.068) by the ATE 3 years later (0.028). Note that the ATE recorded the year of the housing damage (0.068) is lower than the ATE recorded in the “Climate-induced home damage and residential mobility” Sect. (0.073) because here we measure change of census collection district and not change of address but is on par with the estimate for changes of address up to 1 km (0.069).
