Emigration and employment impacts of a disastrous earthquake: country of birth matters

ABSTRACT Two earthquakes in Christchurch, New Zealand, in 2010–11 caused 185 casualties, demolished much of the inner-city and caused significant job losses. We examine one form of adjustment to these outcomes: (international) emigration. Using a difference-in-difference approach, we analyse employment and emigration responses of Christchurch workers relative to matched workers elsewhere. We also examine heterogeneity in responses across subgroups defined by sex, age and country of birth. Significant emigration occurs following the second (more severe) earthquake mirroring job loss patterns. Effects differ by sex and age, and also by country of birth with the foreign-born much more likely to emigrate than the locally born.


INTRODUCTION
Natural disasters are forecast to rise in frequency and severity through climate change (Berlemann & Steinhardt, 2017; Intergovernmental Panel on Climate Change (IPCC), 2013).Accordingly, increased attention has been paid to population and labour market dynamics following a natural disaster.We focus attention on migration and employment dynamics following the 2010 and 2011 earthquakes in Christchurch, New Zealand (NZ), the second of which caused much loss of life and great physical damage to the city.
Several studies have analysed population effects of disasters either at the aggregate level across multiple countries (or regions) or, as in our study, at the level of the individual for a specific event such as Hurricane Katrina (New Orleans).Significant attention has been paid to employment dynamics and to population flows into and out of the affected region following a disaster.Population flows, in turn, have been decomposed into internal and international flows, but with the latter receiving considerably less attention than the former.
Pre-disaster regional characteristics and differing degrees of resilience, standards of governance and postdisaster policies have led to heterogeneous outcomes following disasters (Borsekova & Nijkamp, 2019).This diversity in circumstances means that a body of post-disaster studies is required to establish commonalities and differences in disaster responses.Diversity in outcomes is especially marked in analyses of emigration 1 flows following a disaster.Some studies show a minimal or even negative effect, while others show a positive emigration response, with responses conditioned partly by employment dynamics and by initial characteristics of the place and its people.
Our focus is on the employment and emigration dynamics following the earthquakes that severely impacted Christchurch.The February 2011 quake killed 185 people, forced many from their homes and resulted in Christchurch's central business district being legally cordoned off for over two years.Both the September 2010 and February 2011 earthquakes caused major damage to land, property and infrastructure. 2 The population of Christchurch City dropped by 4% in the two years to June 2012 following the earthquakes (Statistics New Zealand, 2014a) and firms reported difficulty hiring workers (Ministry of Business, Innovation & Employment (MBIE), 2012).Firms' sales revenues and profitability fell immediately after the second major quake, accompanied by an elevated rate of firm exit, particularly for previously poorly performing businesses and for firms in the most hard-hit areas (Fabling et al., 2019).
Firm-level outcomes varied by industry with a large jump in workers employed in the construction industry (How & Kerr, 2019;Statistics New Zealand, 2014a), and a substantial short-term fall in retail and hospitality sector employment (Statistics New Zealand, 2014b).
We shift the focus to examine the dynamic effects of the disaster on the employment of workers who were in employment in Christchurch at the time of the first earthquake and on their decisions of whether to stay in NZ following the disaster.We concentrate on their inter-rather than intra-country (regional) location choice for two reasons.First, the effects of disasters on migration is subject to considerable uncertainty (see section 2).Second, we are confident that the technical requirements needed to ascertain impacts of the disaster (i.e., parallel trends before the disaster) are met for the analysis of employment and emigration responses whereas we cannot be sure that these requirements are met for regional location choices.Our analysis uses rich unit record data in which the unit of analysis is each private sector employee in Christchurch before the September 2010 earthquake.We follow the monthly employment status and international location (in or out of NZ) for these workers through to 2019.
The worker-level perspective enables exploration of heterogeneity in employment and migration outcomes according to pre-disaster personal characteristics.The analysis uses a series of difference-in-difference type comparisons using linked employer-employee tax data in which we compare monthly outcomes across nine years for Christchurch workers to matched workers in two comparator cities (Wellington and Hamilton).Our data enable us to follow the employment status of workers who were employed at the time of the first earthquake over the nine years and to examine whether these workers subsequently moved overseas (for at least 12 months).For those in employment, we also examine whether the worker remained employed in their pre-earthquake job.
We disaggregate the responses by sex, age and birthplace (NZ-versus foreign-born) to highlight outcomes for particular subgroups.Heterogeneity is apparent across each of these dimensions but is particularly noticeable with respect to country of birth.In this respect, NZ's high proportion of foreign-born people (22.9% in the 2006 census 3 ) provides a new perspective on the source of emigration flows following a natural disaster.
We find that while short-term employment outcomes were adversely affected by the earthquakes, workers were more likely to have jobs three years later (relative to the matched control group).Employment impacts vary by worker characteristics (sex, age and country of birth).Christchurch workers were more likely to have emigrated following the earthquakes than their matched counterparts elsewhere; this emigration response was particularly strong for foreign-born workers.In addition, we find different responses to the moderate September 2010 earthquake compared with the severe February 2011 earthquake, indicating non-linearities in response to disaster destructiveness.
The remainder of the paper is structured as follows.Section 2 places the paper's contribution in the context of prior research on economic and demographic impacts of natural disasters.Section 3 outlines our empirical methods and data.Results for employment and emigration dynamics are presented in section 4, with findings summarized in section 5.

LITERATURE REVIEW
Research on employment and migration effects of disasters has been spurred by challenges arising from climate change (Berlemann & Steinhardt, 2017;Mbaye & Zimmermann, 2016).Studies include impacts of slow-onset events such as rising temperatures (Cattaneo & Peri, 2016) and of sudden-onset events (natural disasters) such as floods (Koubi et al., 2016a(Koubi et al., , 2016b)).Given our focus on impacts of a natural disaster, we restrict attention to studies of sudden-onset events.
Migration responses to natural disasters have been categorized as comprising: (1) internal (intra-country) population flows; and (2) international population flows.Some studies have considered a migration systems perspective in which in-and outflows of migrants occur interactively in response to a disaster (Curtis et al., 2015;Fussell, 2009;Fussell et al., 2014;Ouattara & Strobl, 2014).Given our concentration on international flows, we highlight studies that analyse employment and emigration responses.We particularly highlight studies that use micro-data (as we do) in which the individual is the unit of analysis rather than area-level data pertaining to counties, regions or countries.The latter studies have nevertheless formed an important part of the literature (e.g., Beine & Parsons, 2017;Mahajan & Yang, 2020;Shakya et al., 2022;Spitzer et al., 2021) and we draw on the insights of those studies.
Migration is 'the most extreme adaptation strategy' to a disaster (Berlemann & Steinhardt, 2017).A disaster, having damaged or destroyed productive capital, reduces the ability of residents to maintain income in the affected region either because of job loss or impaired production.One response to this income loss is to send family members elsewhere to earn income and send back remittances (Gröger & Zylberberg, 2016).Findlay (2011) posits that following a disaster, most potential migrants prefer not to move, but if they must do so (for instance, to earn income), most will prefer to move only a short distance or to places in which they have family or other ties.From a regional perspective, temporary migration and long-term migration may have contrary effects on resilience (Béné et al., 2014).People leaving temporarily may enhance resilience as demands on city resources are reduced at a time when social services are already stretched.Permanent departures are more problematic as valuable skills are lost to the city, accompanied by negative side-effects of depopulation.
While depopulation may occur, places are rarely abandoned after a shock (Curtis et al., 2015); emigration is often followed by immigration, either of people who left temporarily or by immigrants from other areas.Davis and Weinstein (2002) provide a taxonomy of potential area-level outcomes for a disaster-affected area: If growth follows a random process, a disaster may have a permanent (level-shift) effect on population and production whereas if spatial relationships are determined by fundamentalsthat are unaffected by the disasterproduction and population should revert to trend.For a place that is already in decline but which was initially above its long-run equilibrium level, a disaster may shift activity permanently downwards towards the equilibrium level.This taxonomy indicates that the longer term impacts of a disaster may reflect prior growth trajectories of those regions.Hence, it is informative to have studies of natural disasters that occur in places that are expanding and in others that are contracting to incorporate differing adjustment paths to a shock.Our study is of Christchurch, a city that was growing before the earthquakes. 4It has a diversified regional economy and is the largest city on the South Island of NZ.This study of the Christchurch responses to a natural disaster therefore contrasts with studies of a city such as New Orleans (e.g., Curtis et al., 2015;Deryugina et al., 2018;Fussell, 2009;Fussell et al., 2014;Groen et al., 2020) which was in decline before Hurricane Katrina.
The experience of two Japanese earthquakes illustrates the importance of alternative prior paths.The population fell by approximately 1% in Hyogo prefecture in the year after the 1995 Hansin-Awaji ('Kobe') earthquake, but rose back to its previous positive growth trend over the following two years (Ishikawa, 2019;Morisugi et al., 2019). 5 This pattern is consistent with the prefecture's economy returning to the growth path determined by its (unaffected) fundamentals.By contrast, the Great East Japan ('Fukushima') earthquake saw the populations of Miyagi, Iwate and Fukushima prefectures decline over the following year with no reversion to previous trends (Ishikawa, 2019).Setting aside the specific situation of Fukushima, where the nuclear disaster compounded the effects of the earthquake, these prefectures were already in decline before the earthquake, so the impact of the disaster is consistent with shifting population and production outcomes towards the equilibrium path.
One determinant of migration choice following a disaster is the employment situation facing an individual or household (Beine & Parsons, 2017).Analyses using micro-data have been particularly useful in highlighting this channel.Deryugina et al. (2018) and Groen et al. (2020) make use of administrative tax data to examine long-run consequences of Hurricane Katrina in New Orleans for individual workers. 6Using linked employeremployee tax data and a difference-in-difference estimation strategy, Groen et al. (2020) find that Katrina had a short-run negative impact on average earnings, driven by increased unemployment, while the (seven-year) long-run effect was to raise earnings, reflecting increased labour demand.On average, women fared worse than men as a result of job sorting and the heterogeneity of impacts by industry.Using annual household tax returns, Deryugina et al. (2018) identified short-run increases in non-employment, unemployment benefit receipt and self-employment income.These labour market outcomes underpinned large outward migration, with 27% of New Orleans residents initially moving elsewhere because of the hurricane. 7 Emigration responses to disasters have mostly been analysed using aggregate (county-, district-or countrylevel) data.One reason for the lack of studies using unit record data is that most data collected after a disaster do not include emigrants (who have already departed).When micro-data (e.g., census data) have been used to analyse emigration responses, the emigration component has often had to be estimated as the residual of those not included in a subsequent census after accounting for other reasons (such as death) for the absence of a person (Berlemann & Steinhardt, 2017).Estimates of emigration in such studies may therefore be extremely noisy.Nevertheless, some micro-level evidence of migration responses following a disaster has been obtained using small-scale panel surveys.Halliday (2006) finds a negative effect of earthquakes on the emigration of rural households in El Salvador to the United States and Canada, with this negative effect being particularly prevalent for females (Halliday, 2012).We are unaware of any study using microdata that analyse the emigration response to a natural disaster for a developed country such as NZ.
Studies that use area-level data to analyse migration responses contribute additional valuable insights.Shakya et al. (2022) use district level data to test whether the 2015 Nepal earthquake affected the number of work permits issued to individuals in Nepal for migration.They estimate a (statistically significant) 49% fall in issued work permits for males in earthquake-hit districts relative to control districts following the earthquake, but find no statistically significant effect for females.Mahajan and Yang (2020) use country-level data to analyse migration flows to the United States from countries hit by hurricanes, finding that hurricanes cause immediate and substantial increases in migration to the United States from the stricken countries.Notably, the effect is strengthened for countries with larger pre-existing stocks of migrants in the United States.Beine and Parsons (2017) adopt a panel estimation approach using country data and find that across middle-income and poorer countries, natural disasters deter emigration in total, but spur emigration to neighbouring countries and (for middle-income countries) to former colonial powers.Beine and Parsons (2017) exclude Organisation for Economic Co-operation and Development (OECD) countries (and hence exclude NZ), but it is reasonable to infer that results for NZ are more likely to reflect those for middle-income than for poorer countries.At the time of the Christchurch earthquakes, almost a quarter of NZ's population had been born overseas; the country with the largest proportion of this migrant stock was NZ's former colonial power, the UK.The closest neighbouring countries (Australia and some Pacific Islands) also had large stocks of migrants living in NZ. 8 Beine and Parsons' analysis indicates that these factors may have enhanced the potential for emigration following the earthquakes.
Most studies of migration responses to a natural disaster are in the context of low-to middle-income countries, so may not generalize to developed countries.One study that analyses emigration from a Western European country following a natural disaster is Spitzer et al. (2021).Noting the ambiguity of prior findings on the effects of a natural disaster on migration flows, they analyse the emigration consequences of the 1908 Messina-Reggio Calabria earthquake in Italy.This catastrophic earthquake occurred at a time when mass migration was prevalent with open borders.Using commune-level data, they find no evidence of a large positive impact of the earthquake on emigration from affected communes, with some evidence of a small emigration decline.However, places in which people had a weak connection to the land (i.e., places with a high prevalence of day labourers) experienced increased emigration in response to the earthquake.
While there is a consensus from prior studies that a major natural disaster has initial negative impacts on employment, the cited evidence on the effects of a disaster on emigration flows is far from unanimous.Studies vary from finding no effects, negative effects or positive effects in aggregate.Heterogeneity in response may reflect personal characteristics.For instance, Himes-Cornell and Hoelting (2015) document that in the face of a major disaster, one personal coping mechanism that may assist some individuals to remain resilient is to leave their current location.A potential complement to this process is the presence of family members who already reside in an alternative location.Conversely, local family responsibilities may inhibit emigration responses.This latter factor may help explain the findings of Halliday (2012) that female emigration was reduced following a natural disaster.Spitzer et al.'s (2021) finding that people with a weak connection to place experienced increased emigration in response to an earthquake may be relevant to NZ given the country's high proportion of the population born overseas.Foreign-born people (and especially those on temporary visas) may have reduced ties to the country relative to others, so may be more prone to emigrate as a personal coping mechanism than would NZ-born people.The strength of ties with the country may also reflect the length of time the person had lived in NZ, so emigration responses may be greater for younger foreign-born migrants than for older migrants. 9Beine and Parsons' (2017) finding that there is increased emigration to former colonial powers and to neighbouring countries following a disaster is relevant to NZ given the high proportion of foreign-born migrants from such countries.
NZ not only has a high ratio of foreign-born residents, but also it has the third highest ratio amongst OECD countries of domestically born people residing overseas, especially in Australia (Carey, 2019).This feature is important in the light of Mahajan and Yang's (2020) finding that emigration is increased when the origin country has a larger stock of migrants already situated in a potential host country.Given the potential response mechanisms highlighted by prior literature, we hypothesize that emigration plays a role in the adjustment to the Christchurch earthquakes of 2010 and 2011.These effects may be impacted by employment outcomes and by country of birth and other demographic factors.
Prior studies also have lessons at a methodological level.Berlemann and Steinhardt (2017) emphasize that controls for other factors affecting migration outcomes should pertain to the pre-disaster period to avoid including 'bad controls' that confuse the fundamental versus mediating mechanisms which affect final outcomes.Accordingly, we match each Christchurch private sector worker who was in employment immediately before the first earthquake with similar workers in our comparator cities based only on pre-earthquake data.The richness of our data (covering over 62,000 Christchurch employees plus a similar number of matched employees in other cities, monthly over nine years) enables us to disaggregate the earthquake's employment and emigration responses with respect to worker characteristics (sex, age and country of birth).Disaggregation by country of birth has not been possible in prior studies.The results of this analysis therefore add valuable insights into the existing literature on heterogeneous employment and emigration responses to a severe disaster within a modern developed country.

ESTIMATION APPROACH AND DATA
Our focus is to examine the effects of the Christchurch earthquakes on emigration of people employed in the private sector 10 before the September 2010 earthquake.We subsequently refer to these employees as 'workers' (each of whom was employed in the base, i.e., pre-earthquake, period).A key transmission channel of the earthquakes to the emigration decision is whether the person continued in employment following the earthquakes.Some workers will have continued employment with their base period firm following the earthquake, which we refer to as 'job retention'. 11Others will have shifted to other firms, either by choice or because their prior firm collapsed or reduced employment (Fabling et al., 2019), and some will have emigrated.To provide context, we therefore examine post-earthquake employment and job retention outcomes for workers as well as emigration choices.
In estimating the employment, job retention and emigration effects, we adopt a difference-in-difference-based approach.First, we take the difference in outcome (employment, job retention or emigration) for individual i at time t relative to the (pre-earthquake) base period.Second, after controlling for individual characteristics, we test whether this difference in outcome across time differs for 'treated' individuals (i.e., those who lived in Christchurch at the time of the first earthquake) relative to 'control' individuals (i.e., those who lived in a city unaffected by the earthquake).To ensure similar samples of individuals, we match each person initially employed in a private sector firm in Christchurch with one or more similar individuals employed in private sector firms in Wellington and Hamilton. 12Wellington is an almost identically sized city to Christchurch (each with population of 361,000 in 2006), while Hamilton is the next largest city in NZ (population of 185,000 in 2006).We choose to match against two cities since Wellington, being the capital city, has a smaller share of its working population employed in the private sector than does Christchurch.From a technical viewpoint, using Wellington/Hamilton as a combined source of control individuals results in parallel trends before the earthquakes for our three variables of interest.
In additional estimates, we adopt Auckland (NZ's largest city; population of 1.3 million in 2006) as an alternative source of control individuals, but in this case we are less confident of prior parallel trends for the emigration and employment variables.The dynamic responses to the earthquakes are nevertheless estimated to be similar whether we use Wellington/Hamilton or Auckland controls.Auckland's results are presented in Appendix A in the supplemental data online for comparison.We also perform a placebo test where we exclude Christchurch workers from the analysis, and estimate the impact of the Christchurch earthquakes on Wellington/Hamilton workers (using Auckland as the control).Consistent with expectations, we find no evidence of an impact of the earthquakes on Wellington/Hamilton workers. 13 This null finding implies that control region labour markets were not impacted by Christchurch events (e.g., through internal migration), strengthening the case that Hamilton/Wellington workers are an appropriate control group.
Each Christchurch worker is matched to individuals in comparator cities using a July 2010 base month, two months before the first earthquake to ensure exogeneity.Matching uses a combination of exact and approximate match characteristics.Exact matching criteria are: sex, whether NZ-or foreign-born, and whether the employing firm has at least 50 (FTE) employees.The approximate matching criteria are: age (within two years), earnings (within 10%), tenure in the current job (within three months if tenure is less than two years, and within six months if tenure is at least two years); and the proportion of the employee's employed months within the base region over the previous five years (within 10 percentage points).This last match criterion controls for prior ties to the region.
While individuals are matched on each of these characteristics, we also control for these characteristics in our regressions and for a large number of other personal (non-match) characteristics. 14The non-match characteristics include controls for receipt of state benefits and ACC payments, 15 nature of employment, and whether a person was in NZ in the base month and in the previous year.Controls for the employee's firm include whether the firm has multiple plants in the region, firm age, return on sales, total employment, the change in employment over the past three years, and the firm's industry (with 457 four-digit industry categories).Additionally, we include proxies for worker skill and for the firm's wage premium, as derived in Maré et al. (2017). 16  Inclusion of this rich set of (match and non-match) characteristics enables us to control for the heterogeneous effects of individual and firm characteristics on emigration, employment and job-retention outcomes.The characteristics are itemized in Appendix A in the supplemental data online together with means for both the treated (Christchurch) and control (Wellington/Hamilton) individuals.For each variable, we indicate whether the treated and control means are significantly different.As expected, the means of the exact and approximate match criteria do not differ significantly.Almost all other variables display differences that (even when significantly different) are materially similar.For instance, average firm FTE employment for Christchurch and Wellington/Hamilton employees is 14.9 and 13.5, respectively; while these figures differ in terms of statistical significance, the difference is minor.We are confident therefore that the matching process results in very similar groups of treated and control employees; furthermore, inclusion of the control variables in our regressions accounts for heterogeneous responses due to individual differences.
Our sample comprises 62,232 initial employees in Christchurch for whom we can find matched employees in Wellington/Hamilton.The number of control individuals who meet these criteria is 57,333; weights are used to match control individuals with each treated (Christchurch) individual.
Data (including border movement data) come from Stats NZ's Integrated Data Infrastructure, exploiting the linking of full-coverage firmand worker-level administrative datasets.Stats NZ's Longitudinal Business Database provides firm characteristics (plant location, profitability and detailed industry). 17These firm-level variables are constructed as in Fabling et al. (2019).The jobs data use the FTE measure of labour input as derived by Fabling and Maré (2015).We include only employees who are employed by a firm that operates solely within the base region so that we can assign the employee with certainty to the relevant region (Christchurch, Wellington or Hamilton) in the reference month.
Our principal outcome variable is the worker's longterm emigration status (Emigration), where emigration is defined as being out of the country for at least 12 consecutive months (matching the official definition of a longterm migrant).We also examine worker's employment (Employment) and job retention at their base period firm (Job_Retention).Each outcome variable is represented by a binary indicator equal to 1 if the outcome holds (i.e., if the worker has emigrated, is in employment or has retained their job, respectively), and 0 otherwise.To allow the control variables to have a time-varying influence on outcomes, we estimate a separate regression for each outcome for each month through to September 2019, nine years after the earthquake (and before any disruptions caused by the Covid-19 pandemic).
The difference-in-difference-based equation that we estimate for individual i in month t for Emigration (Emigration it ) is shown as equation ( 1), where 0 is the base (pre-earthquake) month; Quake i0 = 1 indicates that Emigration and employment impacts of a disastrous earthquake: country of birth matters 2495

REGIONAL STUDIES
individual i lived in Christchurch in the base month (and 0 otherwise); Z i0 is a vector of control variables for the individual measured in the base month; a t , b t , g t are coefficients to be estimated; and 1 it is the residual.
The dependent variable in equation ( 1) is the difference in emigration status across time (between the base period and time t) for individual i, while b t represents the treatment effect at time t for Christchurch individuals.We hypothesize b t = 0 for all periods before the first earthquake (the parallel trends assumption); if this hypothesis is not rejected we can conclude that individuals in Christchurch shared the same pre-earthquake emigration trends as individuals in the control cities. Equation ( 1) is estimated separately for each month by ordinary least squares (OLS), so represents a linear probability model in which the dependent variable is equal to 1 if the individual has emigrated, and 0 if the individual remains in NZ (since Emigration i0 = 0 for all individuals).Individuals may have correlated outcomes based on their initial employer's circumstances (e.g., if the firm closed); hence, in estimating equation (1) we adopt robust standard errors clustered on the base month firm.
For employment and job retention, we estimate corresponding equations ( 2) and (3): They are again estimated by OLS with robust standard errors clustered on the base month firm.For these two equations, the dependent variable is equal to 0 if the individual is still employed (or has retained their employment at their base period firm), and −1 otherwise (since Employment i0 = Job Retention i0 = 1 for all individuals).The hypothesis of parallel trends before the earthquake applies for each equation.We initially present monthly results for the effect of the earthquakes on each outcome variable graphically.The graphs show the dynamic responses of each variable to the quakes, corresponding to the monthly estimates of b, with the sample comprising all (treated and control) workers covered by the study. 18Each graph includes estimates that begin a year before the first earthquake (i.e., each graph begins in September 2009) to provide a test of parallel trends before the September 2010 earthquake. 19We restrict attention to a one year interval before September 2010 to avoid the impact of the Global Financial Crisis on the outcome variables since this shock may have had differing impacts across cities.The initial 12 months' results represent a placebo test in which the placebo earthquake is timed as at September 2009, 12 months before the first earthquake and 17 months before the severe February 2011 earthquake.The vertical lines in the graphs indicate the timing of the two (actual) earthquakes.
After presenting the full sample dynamic results, we explore differential impacts for emigration and job retention by population subgroup.Given the prior literature's identification of local ties as a potential determinant of emigration outcomes, we focus on whether outcomes differ by country of birth.Specifically, we test whether foreignborn people have differing employment, job retention and emigration responses to the earthquakes relative to the NZ-born.We disaggregate responses of the two subgroups further according to sex and (separately) age (young, prime-age, older 20 ).We present these results in tabular form focusing on initial peak or trough effects and long-term effects (as at September 2019).Troughs and peaks are identified from the estimated monthly responses for all workers pooled.The subgroup results (by sex) for Emigration are obtained by estimating equation ( 4), in which Female is a dummy variable for female (¼ 1 if female; 0 otherwise) and NZBorn is a dummy variable for born in NZ (¼ 1 if born in NZ; 0 otherwise).
An analogous equation is estimated for age with three age categories.Corresponding equations are estimated for Employment and Job_Retention as the outcome variable.In each case, we test whether responses for demographic groups differ by country of birth.

RESULTS
We first examine the dynamic employment responses to the earthquakes.Figure 1 graphs estimated employment outcomes; the solid line corresponds to each monthly estimate of b in equation ( 2) and the dotted lines indicate the 95% confidence interval.Employment status of Christchurch workers (relative to Wellington/Hamilton workers) was not significantly different to base period employment status for each month before the first earthquake (i.e., for September 2009-August 2010), consistent with parallel trends before the earthquake.
After the moderate September 2010 earthquake, there is a (significant) lift in employment reaching a local peak (relative to Wellington/Hamilton) in January 2011.This temporary employment strength is consistent with earthquake recovery activities such as clearing rubble and fixing roads.The severe February 2011 earthquake resulted in significant firm closure (Fabling et al., 2019) reflected in Figure 1 through a sharp loss of jobs.Employment reached a trough relative to comparator cities in June 2011 with a (significant) loss in employment of over 1 percentage 2496 Richard Fabling et al.

REGIONAL STUDIES
point (p.p.).Christchurch employment then rebounded as the recovery effort took hold.The result was that employment (of those initially employed in Christchurch) rose relative to comparator cities, significantly so from April 2012 onwards.Long-term employment outcomes for (initially employed) Christchurch workers remained elevated through to the end of the sample.Figure 2 graphs outcomes for job retention (conditional on being in a job).Christchurch workers' job retention did not differ significantly (relative to Wellington/Hamilton workers) in any month before the first earthquake (i.e., for September 2009-August 2020), again indicating parallel trends.There is some decrease in job retention following the first earthquake (albeit not statistically significant, i.e., the confidence interval for each month before the second earthquake includes zero).Following the severe February 2011 earthquake, job retention drops sharply for Christchurch workers relative to those in comparator cities.The trough in job retention relative to comparator cities occurred in August 2012 with a differential of over 5 p.p. Thereafter, attrition in job retention in comparator cities leads to a lessening effect of the earthquakes on attachment with the base period employer.
Figure 3 graphs the dynamic response of long-term emigration from NZ.The first 12 months show no deviation between Christchurch and comparator cities since emigration is defined as being out of the country for at least the past 12 consecutive months.The parallel trends test therefore relates to the 12 months starting in September 2010.Christchurch workers' emigration did not differ significantly (relative to Wellington/Hamilton workers) at any time over these 12 months, again indicating parallel trends before the earthquakes.
Emigration of Christchurch workers (relative to workers in comparator cities) jumps sharply over February-December 2012, that is, 12-22 months after the second, more severe, earthquake.Since emigration is defined as being out of NZ for at least 12 consecutive months, this result indicates that emigration jumped immediately from the time of the second earthquake (February 2011) and kept rising for a further 10 months (relative to comparator cities).The (statistically significant) peak emigration in December 2012 corresponds to a 0.34 p.p. increase in workers who had emigrated at this point (relative to Wellington/Hamilton). Thereafter, emigration converges to that experienced in the comparator cities, implying that these cities' emigration subsequently rose relative to Christchurch.Thus, the February 2011 earthquake appears to have brought forward the emigration plans of Christchurch workers.
The magnitudes and population subgroup impacts of the earthquakes on the outcomes are explored further in Table 1.We present results for the trough in each of employment and job retention and the peak in emigration.We also present end-period results to show long-run effects of the earthquakes.
The top panel of Table 1 shows the overall effect across Christchurch workers in the specified month relative to workers in comparator cities.The middle two panels show average effects for NZ-versus foreign-born by sex, and for NZ-versus foreign-born by age group.Each panel reports p-values for a test of the equivalence of specified coefficients for NZ-versus foreign-born (by sex and   employment reductions than males (for both the NZ-and foreign-born).Foreign-born workers had considerably greater employment loss than NZ-born workers (after controlling for industry and other characteristics), confirmed by the two significance tests in the second panel of Table 1.
The third panel of Table 1 shows that young NZ-born workers suffered no significant reduction in their employment rate, but each other category did experience a reduction at the trough.The significance tests in this panel show that foreign-born young and foreign-born prime-age workers had a significantly greater reduction in their employment rate than did NZ-born workers of the same age group.There was no significant difference in the rate of employment loss for older workers between NZ-and foreign-born.
We cannot ascertain whether the excess loss in foreignborn employment (other than for older workers) was due to greater involuntary job loss and/or greater voluntary job loss for foreign-born workers.Voluntary job loss may have been associated with a choice to leave the country following the disaster, especially for those with weaker ties to the country.NZ's immigration settings favour admission of younger workers; hence, older foreign-born workers are likely to have been settled in the country for a much longer periodand have stronger family and friendship ties in the countrythan young or prime-age workers.
Column 3 of Table 1 shows that the trough in Christchurch workers' job retention conditional on being in employment was 5.5 p.p. lower than for Wellington/Hamilton workers.This figure compares with 26.4% of comparator city workers in employment who no longer remained in their base period job.Hence, the excess reduction in job retention (conditional on employment) in Christchurch represented 21% of the outcome in the comparator cities.
Loss of job with the base period firm was experienced across all 10 subgroups shown in Table 1 with excess reduction in job retention varying from 4.3% (prime-age NZ born workers) to 8.7% (young foreign-born workers); the reduction was statistically significant for each worker category.Recall that each of the regressions includes controls for 457 industry categories plus other control variables so these reductions occur after controlling for industry and other characteristics of the worker and firm.None of the five tests for a difference in job retention outcomes between NZ-and foreign-born workers (according to sex or age) indicates a significant difference (at the 5% level) based on country of birth.
The final two columns of Table 1 report peak (December 2012) and long-run results for emigration.The proportion of initially employed Christchurch workers who had emigrated peaked at 0.34 p.p. above that of Wellington/Hamilton workers.The mean proportion of comparator city workers who had emigrated at that time was 1.8%.Hence, the excess migration rate of Christchurch workers represented 19% of the comparator cities' emigration rate, indicating a material impact.
As with employment, peak emigration showed considerable heterogeneity in response to the earthquakes.Neither female nor male NZ-born workers experienced significant excess emigration relative to Wellington/ Hamilton workers at the peak (column 5).By contrast, in the peak month, female and male foreign-born Christchurch workers recorded excess emigration of 1.6 and 1.2 p.p. relative to Wellington/Hamilton workers, representing 26% and 20% respectively of the mean emigration rate of foreign-born in comparator cities at that time.
For those under the age of 50 (i.e., young and primeage subgroups), there are differences between foreignand NZ-born workers.Both young and prime-age foreign-born workers exhibited excess emigration relative to comparator city workers, especially amongst the young (at 3.8 p.p.).By contrast, prime-age NZ-born workers in Christchurch were no more likely to emigrate than their counterparts in comparator cities, and young NZ-born workers were less likely to emigrate than their counterparts elsewhere.For older people, the pattern was reversed: older NZ-born workers showed excess emigration behaviour while older foreign-born workers were less likely to emigrate than their Wellington/Hamilton counterparts.
The heterogeneity shown in emigration outcomes is consistent with the employment outcomes, especially with respect to age and birth country.Young and primeage foreign-born workers in Christchurch experienced both excess employment loss and excess emigration relative to workers in Wellington/Hamilton.By contrast, older foreign-born Christchurch workers fared no worse than older NZ-born workers in terms of employment loss and were less likely to depart the country following the earthquakes than workers in comparator cities.These differential emigration patterns continue through to September 2019 (Table 1, column 6).As discussed in section 5, these results are consistent with the importance of ties to place in determining emigration responses to a disaster.

CONCLUSIONS
Christchurch's massive earthquakes in 2010/11particularly the severe February 2011 quakeprovide a natural experiment enabling us to assess the importance of emigration as a channel through which adjustment to disaster takes place.Using data for over 60,000 Christchurch workers, matched to similar workers in comparator cities, we find that emigration from Christchurch rose immediately after the February 2011 earthquake, staying elevated for approximately two years relative to emigration from the comparator cities.At its peak, the excess emigration from Christchurch following the February 2011 quake represented approximately one-fifth of the comparator cities' emigration rate.The emigration patterns mirror the impacts on employment of the February 2011 earthquake, with the Christchurch employment rate suffering a sharp fall in March 2011, lasting through to December 2011 after which employment rebounded as recovery took hold.The recovery in employment occurred despite a prolonged reduction in Christchurch workers' attachment to their pre-earthquake employer, consistent with the collapse of many firms following the February earthquake.
The contrast between employment and job retention outcomes indicates that many workers who separated from their initial employer found work with other firms during the recovery.
A key finding of the paper is that employment and emigration outcomes were heterogeneous depending on age, sex and whether the worker was born in NZ or elsewhere.Female employment was initially hit harder than male employment while young and prime-age foreignborn workers experienced a greater reduction in employment than did NZ-born workers.There was little difference between female and male peak emigration rates but there were stark differences in emigration by birthplace for young and prime-age workers.For these age groups, the foreign-born were considerably more likely to emigrate than the NZ-born.The patterns for older workers (aged over 50 years) were quite different to those for younger workers in terms of birth country.Foreign-born older workers were no more likely to lose employment than NZ-born workers and were less likely to emigrate than the NZ-born.
The country of birth results are consistent with some findings from other settings that attachment to place is important as a determinant of population adjustment following a major disaster.Older foreign-born workers are likely to have resided in NZ for a long period (given the country's immigration rules) and so they may have similar attachment to the country as those born in NZ.Both NZborn and older foreign-born residents can therefore be expected to have greater attachment to the country than younger workers not born in NZ; the latter are more likely to choose to emigrate following a disaster.
Another contrast which the two earthquakes allow us to consider are the differential effects of a moderate versus a severe earthquake on employment and emigration outcomes.The employment rate in Christchurch rose relative to that in comparator cities immediately after the moderate September 2010 earthquake and there was no discernible effect of this earthquake on emigration.By contrast, the severe February 2011 earthquake induced both a fall in employment and a rise in emigration.The implication of these contrasting results is that impacts of a disaster on employment and emigration are likely to be non-linear in relation to the destructiveness of a disaster; in particular, an increased emigration response may be apparent only following a severe disaster and may have little or no role in adjustment to a moderate shock.An extension to our work would be to analyse responses across multiple disasters that vary in intensity to test our conjecture about nonlinear emigration responses to disaster severity.While we control for a range of characteristics in our analysis, another useful extension could be to analyse heterogeneous effects across a greater range of personal characteristics (potentially including visa status for the foreign-born, if the data were available) and also according to industry or job characteristics.Our results are based on high quality, comprehensive tax data for a modern developed country.Technically, the analysis is constrained both by a short period of pre-disaster observations (earlier data may be clouded by differential impacts of the Global Financial Crisis across cities) and by a paucity of cities with which to compare Christchurch.The Wellington/Hamilton comparison meets the technical requirements for the analysis, while analysis of responses using Auckland as a control city shows similar patterns.If Christchurch were of a more typical size in the country's city hierarchy, one could extend the technical toolkit to a synthetic control analysis (Abadie, 2021).
The findings related to Christchurch are important for civil defence and other planners considering how a severe disaster might affect their region.A disaster-struck region with a high number of foreign-born residents may experience a strong population outflow relative to that of a region with a mostly domestically-born population.Given that recovery often requires a labour force increase in the years following a disaster, a region that is susceptible to high emigration by virtue of a large foreign-born population (especially if that group is young), may therefore be more hamstrung in attracting the requisite labour and skills than one with established foreign-born workers or domestically born workers.These findings therefore show that, in planning for disaster responses, officials should consider that emigration reactions may be both non-linear in terms of the size of disaster and heterogeneous depending on the composition of their workforce.Richard Fabling et al.

NOTES
1. Throughout, 'emigration', 'immigration' and 'migration' refer to cross-border flows, unless explicitly stated otherwise.2. The New Zealand Treasury (2015) estimated that total investment associated with the rebuild would represent around one-fifth of the country's annual gross domestic product (GDP).3. The March 2006 census was the most recent census before the earthquakes.Over the period 2006-10, annual immigration as a proportion of the population remained stable, varying between 1.9% and 2.1%; the annual emigration proportion was also moderately stable, varying between 1.5% and 1.9% (New Zealand Productivity Commission (NZPC), 2022, figs 2.8-2.9).Hence, the earthquakes followed a stable period for migration.Fabling et al. (2022) estimate that, in 2009, migrants accounted for 28% of aggregate full-timeequivalent labour input, with two-thirds of that share being attributable to non-Australian migrants who had lived in New Zealand for at least five years.The share of immigrants on temporary (student, visitor or work) visas rose from an average of 72% of all immigrants over 2003-06 to 79% over 2007-10 (NZPC, 2022, fig. 2.11), fig. 2.11), potentially reducing recent migrants' attachment to New Zealand in the lead up to the earthquakes.Temporary migrants do not have the same eligibility for social security payments as do permanent migrants so are more exposed to a negative employment shock.4. The Christchurch population grew by 8.8% in the decade to 2006. 5. duPont et al. (2015) ) show that population dynamics varied depending on proximity to the centre of Kobe. duPont and Noy (2015) show that the downturn in Kobe City lasted for at least 10 years.6. Groen and Polivka (2008), Vigdor (2007) and Zissimopoulos and Karoly (2010) use data on small samples of evacuees to examine impacts one year after the disaster.7. Deryugina et al. (2018) do not differentiate between internal and international flows.8.While these countries are referred to as 'neighbouring' countries, New Zealand's closest neighbour is almost 2000 km distant.9. New Zealand immigration rules favour younger migrants, so older foreign-born residents are likely to have lived in New Zealand for a longer period.10.The public sector is excluded to focus on workers in firms at risk of closure or substantial job loss.Public sector jobs that workers subsequently move to post-earthquake are included in the employment outcome measure.11.Job retention is measured as conditional on being in a job.12. 'Christchurch' is defined as being within Christchurch City, Waimakariri District and Selwyn District; 'Wellington' is defined as being within Wellington City, Porirua City, Lower Hutt City and Upper Hutt City; and 'Hamilton' is defined as Hamilton City.13.Detailed placebo test results are available from the authors upon request, but are excluded here for brevity.14.We cannot match on all available characteristics since any individual will inevitably differ from another person according to at least one characteristic; hence, we match on what we consider to be the most important criteria and then control for other characteristics.15.ACC payments are worker compensation payments for injuries, received from the Accident Compensation Corporation.16.Maré et al. employ a two-way fixed-effects model, including controls for year and worker age by gender.The average (FTE-weighted) contribution of observed worker characteristics is included as a firm-level control, as is the average worker's 'skill' level and the firm fixed effect.17.Firms are enterprises with longitudinal identifiers repaired following Fabling (2011).18.Given the large number of coefficients estimated for each month's equation, the results are not reported for the control variables.19.For these estimates, the base period for matching becomes July 2009, two months before the first month's estimates.The difference in base periods between the graphical results and the subsequent tabular results (for which the base period is July 2010) produces fractional changes in estimated timing and magnitude of peaks and troughs in responses.These changes are likely due to the fact that starting the analysis earlier means some treated/ control individuals had migrated internally in the additional year leading up to the earthquakes.20.'Young' is defined as aged less than 26 years, 'primeage' as aged 26-50 years and 'older' as aged more than 50 years in the base period.

Figure 2 .
Figure 2. Job retention (in pre-earthquake job).Note: See Figure 1 for notes; estimated outcomes are conditional on employment.

Figure 3 .
Figure 3. Emigration (for at least the past 12 months).Note: See Figure 1 for notes; by definition, the outcome is zero through to June 2010 since all workers are in New Zealand in the base month (July 2009).

Figure 1 .
Figure 1.Employment (in any job).Note: Difference-in-difference effects (solid line) are estimated separately for each month using ordinary least squares (OLS) including regression controls (listed in Appendix A in the supplemental data online).July 2009 is the base month.The control group is Wellington and Hamilton workers matched on July 2009 characteristics.Dotted lines indicate the 95% confidence interval (calculated using robust standard errors clustered on base month firm).Vertical grey bars indicate the timing of major earthquakes (September 2010 and February 2011).