Migration and natural disaster: Ex-ante preparedness and contribution to ex-post community recovery

Migration and natural disaster: Ex-ante preparedness and contribution to ex-post community recovery Abstract Economics literature suggests that migration and especially remittances serve as insurance to migrant households in the ex-post recovery. However, the evidence of migration or remittances on ex-ante preparedness has not been focused much in the literature. Additionally, little attention has been devoted to analysing the behavioural difference between migrant and non-migrant households towards ex-post community recovery after an external shock in their country of origin. In this paper, we study the differences in migration characteristics of a household and their behaviour towards the ex-ante preparedness for future disaster. Furthermore, we analyse the differences in their behaviour towards the ex-post recovery of neighbourhood and community in the home country. The earthquake in Nepal in 2015 is considered for the analysis. As migration tends to be self-selected, we use a unique random selection policy of migration to the Republic of Korea as our identification strategy to eliminate self-selection bias. Our empirical results do not find any relationship between migration characteristics of a household and the likelihood of ex-ante preparedness for future disasters. However, we find a substantial difference in their behaviour towards the ex-post recovery of neighbourhood and community. On average, migrant and return migrant households are 18 to 23 percentage points more likely to participate in the community helping and contribute 257 to 279 percentage points more in absolute terms in comparison with non-migrant households. Our findings suggest that international migration increases the relative participation and absolute amount of community helping after a disaster in the home country. 1. Introduction Insurance against natural disasters is relatively poor or non-existent in developing countries. However, the rate of destruction and fatalities due to natural disasters are higher in low-income countries in comparison with high-income countries (see Fig. 1). As a disaster risk reduction strategy, ex-ante and ex-post actions taken by the households in disaster-prone regions, especially in developing countries can substantially reduce the loss of human life and vulnerability.1 Economics literature suggests that migration and especially remittances serve as insurance to migrant households in the ex-post recovery (Yang and Choi 2007).2 There is also a large literature in the field of climate change asserting that migration or mobility can act as an important coping strategy for the households following the events (Black et al. 2011; Mueller et al. 2014). However, the evidence of migration or the impact of remittances on ex-ante preparedness is thin (Mohapatra et al. 2012). Additionally, little attention has been devoted to analysing the behavioural difference between migrant and non-migrant households towards ex-post community recovery after an external shock in their country of origin. Empirical evidence suggests that migrants help others after a disaster in the home country (Fagan 2006; Brown et al. 2014). However, to the best of our knowledge, we have not come across any study that tries to compare the help provided by migrant and non-migrant (households). Taking these scenarios into consideration, we examine the following two questions in this paper: (1) Do overseas migrants and their households from developing countries act differently from non-migrant households towards the ex-ante preparedness for future disaster? (2) Do they also behave and act differently towards helping the community or neighbourhood in the ex-post recovery? We consider the earthquake in Nepal in 2015 for our analysis. Figure 1. View largeDownload slide Differential burden of natural disasters. Source: Linnerooth-Bayer, J., Mechler, R. and Hochrainer, S. (2011) ‘Insurance against Losses from Natural Disasters in Developing Countries. Evidence, gaps and the way forward’, IDRiM Journal, 1/1: 59–81. Data source: Munich Re, 2005. Figure 1. View largeDownload slide Differential burden of natural disasters. Source: Linnerooth-Bayer, J., Mechler, R. and Hochrainer, S. (2011) ‘Insurance against Losses from Natural Disasters in Developing Countries. Evidence, gaps and the way forward’, IDRiM Journal, 1/1: 59–81. Data source: Munich Re, 2005. Forming an unbiased control group is a severe challenge to researchers attempting to analyse the impact of migration, as migrants tend to be self-selected (Gibson et al. 2011; McKenzie and Yang 2012; McKenzie 2015). A host of observable and unobservable differences exist between people who choose to migrate and those who stay behind. Therefore, the estimation results might be misleading, as it is quite difficult to isolate the effect of migration by controlling for all other differences. A better comparison is possible if the control and treatment groups have same characteristics or similar in all dimensions. In other words, migration of an individual or household is random or by chance. To address the challenges of self-selection biases, we exploit the Employment Permit System (EPS) policy. EPS is a programme under which workers are randomly selected to work in the Republic of Korea over a period of two years after passing a Korean language examination. The Republic of Korea is also one of the most desired destinations to work among Nepalese people due to high wages, good working conditions and no corruption in hiring. In this paper, we analyse three kinds of households based on their migration characteristics: non-migrant households, migrant households and return migrant households. First, we examine the differences in migrant, non-migrant and return migrant households’ behaviour towards ex-ante disaster preparedness. There are various ways to measure households’ ex-ante disaster preparedness. For example, households may engage in housing improvement as a disaster preparedness strategy. In this paper, we consider the rate of destruction of houses in the Nepal earthquake as our measure of ex-ante preparedness considering housing improvement as a disaster risk reduction strategy. We consider full destruction, partial destruction, and no destruction of houses to measure the intensity of ex-ante preparedness of a household. Our empirical results do not find any relationship between migration characteristics of a household and the likelihood of ex-ante preparedness. The results are similar in the case of both migrant and return migrant households. Our paper provides the first evidence of the reality of ex-ante disaster preparedness among non-migrant, migrant and return migrant households. A considerable amount of empirical research has focused on the impact of international migration, especially remittances on the ex-post disaster recovery of their own households in the home country (Deshingkar and Aheeyar 2006; Yang and Choi 2007; Ratha and Sirkeci 2010).3 However, there is a possibility that the migrants and their households also help others, especially their neighbours and community after a disaster, as the income of overseas migrant workers is not affected by an external shock in their home country.4 To evaluate the impact of migration on households’ behaviour towards the community after the disaster, we analyse the differences in the migration status of a household and their behaviour towards ex-post disaster recovery of their community. We use Hurdle model (Cragg 1971) for our analysis. The first hurdle examines the likelihood of participation of a household in community help, whereas the second hurdle examines their absolute amount of help if the household chose to participate. Our empirical findings show that migrant and return migrant households have a higher likelihood of choosing to participate in the community helping in comparison with non-migrant households. Furthermore, we also find that migrant and return migrant households help substantially more in absolute terms as compared with non-migrant households. On average, migrant and return migrant households are 18 to 23 percentage points more likely to participate in the community helping and contribute 256 to 279 percentage points more in absolute terms in comparison with non-migrant households. We do not find any statistically significant difference between migrant and return migrant households in their relative and absolute amount of help. Our findings suggest that international migration increases the relative participation and absolute amount of community helping after a disaster in the home country. This paper contributes to three strands of the literature. First, we contribute to the limited literature that studies the impact of migration through a random experiment to eliminate the selection bias (McKenzie et al. 2010; Gibson et al. 2011). We did it by using a unique random selection policy of migration to the Republic of Korea as our identification strategy. Second, we are the first to analyse the reality of the impact of international migration on ex-ante preparedness towards natural disasters as our survey was conducted after the earthquake in Nepal in 2015. Earlier studies in this field (Mohapatra et al. 2012; Manandhar 2016) have considered the construction of houses (mud and brick or concrete) or the quality of construction as a measure of disaster preparedness. However, we considered the intensity of destruction of houses (i.e. full destruction, partial destruction and no destruction) as our measure to analyse the ex-ante preparedness. Third, we are the first to offer evidence on behavioural differences among migrant, return migrant and non-migrant households towards ex-post community recovery after a natural disaster in the home country. Our findings also shed light on the positive impact of migration on disaster recovery for those households who could not be able to engage in migration. The rest of the paper is organized as follows. Section 2 reviews the related literature; Section 3 gives a preliminary idea about the EPS policy in Nepal; Section 4 describes our survey methods; Section 5 describes the data and test the randomization of our control and treatment groups; Section 6 describes the empirical model; Section 7 presents our results; Section 8 presents robustness check to our main findings; and Section 9 summarizes and concludes our findings. 2. Related research The impact of migration on disaster recovery has been extensively studied in the field of economics. Most research concentrates on the flow of remittances. For example, Wu (2006) and Deshingkar and Aheeyar (2006) studied the flow of remittances aftermath of the tsunami in Ache (Indonesia) and Sri Lanka respectively and found that remittances are very fast in reaching affected migrant households and sometimes even more quickly than international aid and government assistance. Le De et al. (2015) studied Samoan households that were affected by the tsunami in 2009 and further hit by Cyclone Evan in 2012. Their study found that remittances are very fast in reaching those affected by the disaster and even remain high in the long run to cope with and recover from the disaster. They also found a complementary relationship between remittances and external aid. Furthermore, Yang and Choi (2007) studied rainfall shock in the Philippines and concluded that remittances serve as a self-insurance to the recipient households. Ex-ante actions taken by households in disaster-prone regions and especially in developing countries can substantially reduce the loss of human life and vulnerability. As a disaster risk reduction strategy migrant households can invest in housing improvements. However, the literature on the impact of migration on ex-ante preparedness for future disasters is thin. Mohapatra et al. (2012) studied the impact of remittances on ex-ante preparedness in Burkina Faso and Ghana. They found that remittance-receiving households, especially from developed countries, tend to have housing built of concrete rather than mud and have greater access to communication. Therefore, they concluded that remittance-receiving households are better prepared against natural disasters as concrete houses are more disaster resilient than mud and brick houses. In contrast, Manandhar (2016) studied the remittances from Qatar and the Republic of Korea and its impact on earthquake preparedness in Nepal and found that remittance-dependent households have a higher likelihood of possessing concrete houses and invest 18 to 22% of their total remittances for the construction of houses. However, he additionally found that the use of remittances for better quality and stronger houses, using engineer and building code awareness for safe construction, tend to decrease. Therefore, he concluded that remittances are fuelling unsafe construction practices in Nepal and increasing risks caused by an earthquake. Mohapatra et al. (2012) considered the construction of concrete houses and Manandhar (2016) examined the quality of construction of houses as a measure of ex-ante preparedness towards a future disaster. However, we have not come across any study that tries to evaluate the reality of the situation, especially after a disaster. Our paper provides the first evidence of the reality of ex-ante disaster preparedness considering the level of destruction of houses after a disaster. We consider full destruction, partial destruction, and no destruction of houses as the measures of ex-ante preparedness. There is a higher possibility that migrants and their households help others in ex-post disaster recovery, as the income of overseas migrant workers is not affected by an external shock in their home country. For example, Brown et al. (2014) studied migration and wider community sharing norm after Cyclone Pat in 2010 in the Cook Islands. They found that Cook Islanders living in regional locations (Riverina) have higher propensity to remit to others (community remittances) in comparison to migrants living in the metropolitan region (Sydney). They concluded that there exists a ‘donor fatigue’ where the social pressure is stronger in the metropolitan region. Furthermore, anecdotal evidence by Fagan (2006) also shows that in-kind transfer from friends and relatives abroad played an important role in relieving the immediate distress caused by Cyclone Jeane in Haiti in 2004. These evidences show that migration not only helps in recovery of their own households, but they also help others after a disaster in their countries of origin. However, little attention has been devoted to analysing the behavioural difference between migrant and non-migrant households towards ex-post community recovery after an external shock in their country of origin. We investigate the difference in migration characteristics of households and their behaviour towards the likelihood of participation and absolute amount help towards the community. Furthermore, we consider total help by a migrant household (including help by household members living in the home country instead of only members living in the destination country) for the following reasons. First, either migrant(s) or their household members living in the country of origin or both might be participating in community help. Second, the total income of the household includes the income of the migrant so the participation decision could be jointly determined. 3. EPS policy in Nepal In 2007, the Government of Nepal and the Republic of Korea signed an MOU under the Employment Permit System (EPS) that allows Nepali workers to work in the Republic of Korea. It is a government-to-government agreement and currently accessible to 15 Asian countries. In Nepal, it is managed by the EPS Korea Section, under the Department of Foreign Employment (DOFE). The Republic of Korea has emerged as a favourable destination country among Nepalese because of the government-to-government initiative that enables higher wages and better living and working conditions. Also, unlike other foreign employment programmes where the recruiting agencies may charge a higher amount of money over and above the actual amount, there is no corruption involved in this system. As of 2014/2015 33,960 individuals (31,771 men and 2189 women) have participated in this programme since 2008.5 EPS workers work in small enterprises (fewer than 300 employees) and perform mostly low-skilled jobs. Nepalese workers under the EPS system generally work in manufacturing and agriculture and livestock sectors. The average wage EPS workers receive in the Republic of Korea is quite high in comparison to the average wage in Nepal. Individuals between the age of 18 and 39 with no criminal records are eligible to apply for this programme. Furthermore, all the applicants have to pass a Korean Language Test (Paper Based Test or PBT) to be eligible to work.6 The minimum score to pass the examination is 80 out of 200. However, due to a limited number of quotas every year, a designated number of candidates pass in order of their highest score on the test. The test score is valid for two years from the date of announcement of the result. After passing the test, workers fill out their job application form at the EPS Korea Section in Nepal. Then, each one is randomly drawn and introduced to an employer in the Republic of Korea.7 So a worker’s timing of going to the Republic of Korea for work is randomly decided over two years. Furthermore, the Korean language test is conducted almost every year.8 As the randomization process of introducing workers goes on for two years, workers from two different test taking years are selected simultaneously. So, the workers are treated similarly after winning the lottery even if they have appeared the examination and were selected in different years. These unique policies act as our identification strategy to eliminate the selection biases. Under the EPS policy, it is mandatory for workers to return to their home countries after finishing their contract. The general contract period is three years and can further be extended up to four years and 10 months. Foreign workers who have worked for four years and ten months without changing their workplace are considered as ‘committed workers’. Workers with these characteristics can come back to the Republic of Korea with a new contract after three months of departure and work for an additional four years and ten months in the same organization. Workers who have worked three years or more under the EPS system but without the characteristics mentioned above, that belong to the age group between 18 and 39 years, and have no illegal records are eligible to go to the Republic of Korea through a special Korean language test (Computer-Based Test or CBT). This test is conducted four times per year.9 Workers who pass this test can easily and quickly re-enter the Republic of Korea six months after their departure. As the average wage in the Republic of Korea is very high in comparison to Nepal, most workers under these categories opt to go back again.10 The Special Korean language examination (CBT) started in 2013 in Nepal, and as of 2015 September, 11 CBT examinations have already been conducted. We analyse three kinds of households based on their migration characteristics: non-migrant, migrant and return migrant. We term ‘non-migrant households’ as those with a member who has passed the PBT, won the EPS lottery and were in the process of going to the Republic of Korea to work for the first time. We consider this group as households with no migration experience. We term ‘migrant households’ as those with a member who is presently working in the Republic of Korea under his/her first contract.11 We consider this group as households with short-term migration experience. We term ‘return migrant households’ as those with a member who has completed his first contract period and returned from the Republic of Korea or has gone back again with a new contract. We consider this group as households with medium to long-term migration experience. We consider non-migrant households as our control group to analyse the impact of migration. 4. Survey methods A customized survey was conducted between September and October 2015.12 All the survey questions were written in simple Nepali language to avoid any difficulty in understanding. Three different groups of households were selected for the survey. The first group is ‘non-migrant households’.13 Contacting household members with these characteristics were difficult due to unavailability of personal information. We used different approaches to survey household with these characteristics. Each and every prospective worker has to go through a week of preliminary training for 45 hours in an approved public institution in the sending country after a person is selected randomly through the lottery process and concluded labour contracts with their respective employers.14 There is only one institution in Nepal that provides training for these workers. We contacted the institution and with their permission conducted the survey of those who were taking training during that time. Additionally, we also surveyed individuals who have already completed their training and were about to fly to the Republic of Korea. People come to the EPS Korea Section office in Nepal to collect their final documents (like passport, dress, and batch to be worn during the departure period) before their departure. We requested the ministry to conduct a survey at their campus. With their permission, we carried out the survey for those who were about to depart to the Republic of Korea. In this way, we avoid selection bias in our survey as we surveyed those who have been selected through the random selection process. Our second group is ‘migrant households’. It is also difficult to obtain personal information of these households because of the restriction by the ministry to disclose their personal data. We used a novel approach to contact households with these characteristics. Facebook has many community pages dedicated to Nepalese workers living in the Republic of Korea through EPS. The number of subscribers in these community pages ranges from 2,000 to 30,000 members. The community pages with most subscribers are EPS NEPAL FAMILY (around 30,000 subscribers), EPS Korea Nepal (more than 28,000 subscribers) and Eps Nepalese in Korea (more than 10,000 subscribers). People share valuable information and interact with other members of the community. They also post information regarding various events held by migrant community/groups in the Republic of Korea. We created a survey page using Google forms. We contacted the administrator of the respective community sites and posted the Google forms link containing customized survey with their permission. In this way, we avoided the spatial bias of snowballing effect and conducted the survey on a wider scale throughout the Republic of Korea. Although the mail-out survey is the preferred method for surveying migrants and diaspora, it produces a severe sample selection bias as they tend to be highly skewed (Crush et al. 2012). In this context, web-based research methodology could be of significant advantage. There is a specific chapter devoted to research methods in World Wide Web (WWW) and Social Networking Sites (SNSs) in particular, in the Handbook of Research Methods in Migration. However, one limitation could be that our sample might exclude those who do not have access to these community sites or Internet. The Republic of Korea has the fastest speed Internet in the world, and moreover, the Internet user rate was 89.65% in 2015.15,16 From interaction with those who have already returned from the Republic of Korea, we came to know that Facebook is indeed the best way of communication and that it is widely used by Nepalese workers to interact with friends in the Republic of Korea and Nepal. As most Nepalese workers work in very small organizations, browsing Facebook is one of the best ways to pass their leisure time and keep in contact with their friends and family members. Besides, the number of members in these community pages is considerably large. This evidence additionally reflects that the majority of the migrant workers have access to the Internet and these community sites. Another drawback of our sample survey method through Facebook pages could be that we may fail to recognize whether our respondents only include our target groups, as various kinds of people could be a member of these community pages. To avoid these problems, we specifically put some questions which only the target groups could answer, so we can identify respondents who belong to our target group and those who do not. Additionally, we also exclude repeated respondents from our sample. Our third group is ‘return migrant households’. As the wage in the Republic of Korea is much higher than in Nepal, most workers under these categories choose to go back again to work.17 These groups have about 3 to 7 years of migration experience. We used the Facebook mode of survey for those who have not appeared for any examinations (Committed workers) and for those who have already gone back again with a new contract during our survey period. Additionally, with the permission of the ministry in Nepal, we surveyed people who have appeared for CBT examination in September 2015 and have been qualified to go back to the Republic of Korea again.18 5. Descriptive statistics and the test of randomization We present the descriptive statistics in Table 1. In columns (1) and (2), we show the descriptive statistics of our control group households. Furthermore, we show the descriptive statistics of our treatment group households in columns (3) to (6). In columns (7) and (8) we present the difference between control and treatment group households and their statistical significance on average. The first four variables are our outcome variables, and the remaining variables are our control variables. Table 1. Descriptive statistics and test of randomization Non-migrant household Migrant household Return migrant household Difference Difference (1) (2) (3) (4) (5) (6) (7) (8) Variables Mean Std Dev. Mean Std Dev. Mean Std Dev. (3)–(1) (5)–(1) Help (Yes) 0.53 0.501 0.729 0.447 0.766 0.428 0.199*** 0.236*** Log (Amount of help+1) 4.233 4.321 6.725 4.666 6.635 4.35 2.492*** 2.402*** Full destruction of house 0.202 0.403 0.119 0.326 0.113 0.32 −0.083 −0.089 No. destruction of house 0.518 0.501 0.393 0.491 0.585 0.497 −0.125* 0.067 Affected intensity 1.989 1.868 1.871 1.696 2.077 1.69 −0.118 0.088 No. of family members 6.316 2.733 6.775 3.46 6.418 3.705 0.459 0.102 Family member(s) abroad 0.452 0.499 0.398 0.492 0.42 0.499 −0.054 -0.032 District 34.706 17.72 35.847 17.038 34.712 16.804 1.141 0.006 Education in years 12.929 2.21 13.356 2.122 13.273 2.475 0.427 0.344 Age 24.937 5.498 26.831 4.513 31.058 2.6 1.894*** 6.121*** Household living in Kathmandu Valley 0.051 0.22 0.035 0.186 0.135 0.345 −0.016 0.084** Number of observations 184 90 55 Non-migrant household Migrant household Return migrant household Difference Difference (1) (2) (3) (4) (5) (6) (7) (8) Variables Mean Std Dev. Mean Std Dev. Mean Std Dev. (3)–(1) (5)–(1) Help (Yes) 0.53 0.501 0.729 0.447 0.766 0.428 0.199*** 0.236*** Log (Amount of help+1) 4.233 4.321 6.725 4.666 6.635 4.35 2.492*** 2.402*** Full destruction of house 0.202 0.403 0.119 0.326 0.113 0.32 −0.083 −0.089 No. destruction of house 0.518 0.501 0.393 0.491 0.585 0.497 −0.125* 0.067 Affected intensity 1.989 1.868 1.871 1.696 2.077 1.69 −0.118 0.088 No. of family members 6.316 2.733 6.775 3.46 6.418 3.705 0.459 0.102 Family member(s) abroad 0.452 0.499 0.398 0.492 0.42 0.499 −0.054 -0.032 District 34.706 17.72 35.847 17.038 34.712 16.804 1.141 0.006 Education in years 12.929 2.21 13.356 2.122 13.273 2.475 0.427 0.344 Age 24.937 5.498 26.831 4.513 31.058 2.6 1.894*** 6.121*** Household living in Kathmandu Valley 0.051 0.22 0.035 0.186 0.135 0.345 −0.016 0.084** Number of observations 184 90 55 Table 1. Descriptive statistics and test of randomization Non-migrant household Migrant household Return migrant household Difference Difference (1) (2) (3) (4) (5) (6) (7) (8) Variables Mean Std Dev. Mean Std Dev. Mean Std Dev. (3)–(1) (5)–(1) Help (Yes) 0.53 0.501 0.729 0.447 0.766 0.428 0.199*** 0.236*** Log (Amount of help+1) 4.233 4.321 6.725 4.666 6.635 4.35 2.492*** 2.402*** Full destruction of house 0.202 0.403 0.119 0.326 0.113 0.32 −0.083 −0.089 No. destruction of house 0.518 0.501 0.393 0.491 0.585 0.497 −0.125* 0.067 Affected intensity 1.989 1.868 1.871 1.696 2.077 1.69 −0.118 0.088 No. of family members 6.316 2.733 6.775 3.46 6.418 3.705 0.459 0.102 Family member(s) abroad 0.452 0.499 0.398 0.492 0.42 0.499 −0.054 -0.032 District 34.706 17.72 35.847 17.038 34.712 16.804 1.141 0.006 Education in years 12.929 2.21 13.356 2.122 13.273 2.475 0.427 0.344 Age 24.937 5.498 26.831 4.513 31.058 2.6 1.894*** 6.121*** Household living in Kathmandu Valley 0.051 0.22 0.035 0.186 0.135 0.345 −0.016 0.084** Number of observations 184 90 55 Non-migrant household Migrant household Return migrant household Difference Difference (1) (2) (3) (4) (5) (6) (7) (8) Variables Mean Std Dev. Mean Std Dev. Mean Std Dev. (3)–(1) (5)–(1) Help (Yes) 0.53 0.501 0.729 0.447 0.766 0.428 0.199*** 0.236*** Log (Amount of help+1) 4.233 4.321 6.725 4.666 6.635 4.35 2.492*** 2.402*** Full destruction of house 0.202 0.403 0.119 0.326 0.113 0.32 −0.083 −0.089 No. destruction of house 0.518 0.501 0.393 0.491 0.585 0.497 −0.125* 0.067 Affected intensity 1.989 1.868 1.871 1.696 2.077 1.69 −0.118 0.088 No. of family members 6.316 2.733 6.775 3.46 6.418 3.705 0.459 0.102 Family member(s) abroad 0.452 0.499 0.398 0.492 0.42 0.499 −0.054 -0.032 District 34.706 17.72 35.847 17.038 34.712 16.804 1.141 0.006 Education in years 12.929 2.21 13.356 2.122 13.273 2.475 0.427 0.344 Age 24.937 5.498 26.831 4.513 31.058 2.6 1.894*** 6.121*** Household living in Kathmandu Valley 0.051 0.22 0.035 0.186 0.135 0.345 −0.016 0.084** Number of observations 184 90 55 The distribution of the absolute amount of help in our sample is highly skewed towards left. We show the skewness in Figure 2. Econometric models are consistent under the assumption that the normal distribution of the dependent variable is satisfied. Following Cragg (1971) and Amemiya and Boskin (1974) we use the natural logarithm of the amount of help to fix the problem of highly skewed data. Several studies have been undertaken following this approach, such as Newman et al. (2003), Carroll et al. (2005) and Zhang et al. (2008). As can be seen from Figure 3, the log transformation follows the normal distribution rather well. Figure 2. View largeDownload slide Distribution graph of amount of help with and without zero. Figure 2. View largeDownload slide Distribution graph of amount of help with and without zero. Figure 3. View largeDownload slide Distribution graph of Log (Amount of help + 1) with and without zero. Figure 3. View largeDownload slide Distribution graph of Log (Amount of help + 1) with and without zero. We constructed the disaster intensity variable in the following way. The government of Nepal has scaled the earthquake-affected districts into five categories (see Fig. 4). We assigned value 0 to those districts that were not affected at all (white parts) and assigned value 1 to 5 by the affected intensity of the respective district. We use district as our cluster variable to estimate the standard error.19 Our sample represents households from 63 districts out of a total of 75 districts in Nepal. Figure 4. View largeDownload slide Affected intensity in 2015 earthquake in Nepal. Source: National Planning Commission, (2015), Post-disaster needs assessment, Vol. A: Key findings, Government of Nepal. Figure 4. View largeDownload slide Affected intensity in 2015 earthquake in Nepal. Source: National Planning Commission, (2015), Post-disaster needs assessment, Vol. A: Key findings, Government of Nepal. Economic status of the migrant and return migrant households before their migration is a better control variable for comparison with non-migrant households as households’ economic status is highly correlated with their migration status. It is difficult to correctly evaluate the economic status of migrant and return migrant households before their migration. Therefore, we consider education (of the migrant or would be migrant worker), number of family members, household member(s) abroad, household living in Kathmandu valley as proxy variables for household economic status.20 Household member(s) abroad variable does not include the respondent who is now working in the Republic of Korea. It is important for us to check whether our control and treatment groups are uniformly distributed based on their other characteristics. We test the randomization of our treatment and control groups in columns (7) and (8). As we can observe, our control and treatment groups are uniformly distributed except the last two characteristics. Return migrant households are more likely to live in the Kathmandu Valley. This might be due to the information gap in earlier years. The EPS programme started in Nepal for the first time in 2008 and households from the Kathmandu Valley might have had a higher chance of obtaining information earlier as compared with other parts of Nepal. Another interpretation could be that migrant households could have moved to Kathmandu Valley after a period, as their income got much higher in comparison with non-migrant household’s income. We also see a significant difference in age based on our sample groups. Workers with higher international migration experience are older in comparison with workers with no migration experience. This difference is expected as there is a restriction on migration based on age imposed by the EPS. As we can see from Table 1, the percentage of households participating in community helping is higher for migrant and return migrant households. Additionally, they have helped more in absolute terms. These differences are statistically significant from zero. On average, migrant and return migrant households are 19.9 percentage points and 23.6 percentage points more likely to participate in the community helping and contribute 249.2 percentage points and 240.2 percentage points more in absolute terms respectively. However, there is no statistically significant difference between the migration characteristics of a household and the intensity of destruction of houses. No destruction of houses in the case of migrant households is only significant at the margin. 6. Model specifications We use Probit model to analyse ex-ante preparedness of households in Nepal. We estimate the extent of destruction suffered by households with different migration characteristics. We conditioned partial destruction of the house as the base and compare no destruction of a house and full destruction of a house to analyse the level of ex-ante preparedness. The model for ex-ante preparedness is as follows: Prob(y>0 |X)=1[Xγ+u>0]= Φ(Xγ) (1) In the above model, y takes value one if the destruction of a house is full (no destruction) and zero otherwise. X is the vector of independent observable characteristics and γ is the vector of coefficients. If migrant and returned-migrant households are prepared for future disasters, our results should show a negative and statistically significant coefficient towards the full destruction of a house and positive and statistically significant coefficient towards no destruction of a house. The absolute amount of community helping variable in our sample follows a lognormal distribution. So we employ lognormal hurdle model (Cragg 1971) to analyse participation decisions and absolute amount of help. There are two hurdles in the model; in the first hurdle, households decide whether or not to participate in helping. In the second hurdle, households decide how much to help if they choose to participate in helping. The equations are: y=s.w*=1Xγ+v>0 exp Xβ+u (2) Where u and v are independent, of X, and w* has a lognormal distribution, and w*=exp(Xβ+u) u|X∼Normal(0,σ²) In Equation (2)y takes the value zero or the amount of donation depending on the household’s decision to participate. X is the vector of observable independent household characteristics, and β and γ are the vectors of coefficients. The estimation is preceded in two steps (Wooldridge 2009). The coefficient γ is estimated in Probit model of s on the vector of independent variables X. Then the coefficient β is estimated by OLS regression of log(y) on X for observations with y > 0 (type II Tobit model). The expected value of donation conditional on y > 0 is Ey|X,y>0= Ew*|X,s=1=Ew*X=expXβ+σ22 (3) And the unconditional expected value of donation is Ey|X=ΦXγexpXβ+σ22 (4) We use a Lognormal hurdle model in our analysis as suggested by Brown et al. (2014). Brown et al. (2014) analysed the remittances data using both the Heckman selection model and Cragg’s double-hurdle model and concluded that Cragg’s double hurdle model is more appropriate for analysing remittances data. 7. Estimation results We use modified zero-order regression to deal with the missing variables as instructed in Greene (2010). We present all regression results with and without controlling for other characteristics to examine the robustness of our findings. We also check the robustness of our findings without considering the missing variables. 7.1 Ex-ante preparedness The regression results on household behaviour towards ex-ante preparedness are shown in Tables 2 and 3. We estimate the coefficients based on Probit model stated in Equation (1). Standard errors are clustered at the district level. We analyse full destruction of houses in Table 2 and no destruction of houses in Table 3 as our dependent variable. We condition non-migrant households as the base and compare it with migrant and return migrant households. Column (1) of Tables 2 and 3 show the estimates without any control. In column (2) we show the estimates with all controls presented in descriptive statistics. In column (3) we further include interaction variable between affected intensity and the migration characteristics to see any differences in the destruction rate. Table 2. Full destruction of house (1) (2) (3) Return migrant household –0.376 –0.571 –0.374 (0.241) (0.496) (2.667) Migrant household –0.347 –0.615* –2.937 (0.270) (0.373) (12.52) Affected intensity * Migrant household 0.628 (2.752) Affected intensity * Return migrant household –0.077 (0.693) Other controls No Yes Yes Chi2 4.301 242.3 94.03 Log likelihood –133.999 –78.460 –75.182 R² 0.015 0.413 0.437 N 305 292 292 (1) (2) (3) Return migrant household –0.376 –0.571 –0.374 (0.241) (0.496) (2.667) Migrant household –0.347 –0.615* –2.937 (0.270) (0.373) (12.52) Affected intensity * Migrant household 0.628 (2.752) Affected intensity * Return migrant household –0.077 (0.693) Other controls No Yes Yes Chi2 4.301 242.3 94.03 Log likelihood –133.999 –78.460 –75.182 R² 0.015 0.413 0.437 N 305 292 292 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table 2. Full destruction of house (1) (2) (3) Return migrant household –0.376 –0.571 –0.374 (0.241) (0.496) (2.667) Migrant household –0.347 –0.615* –2.937 (0.270) (0.373) (12.52) Affected intensity * Migrant household 0.628 (2.752) Affected intensity * Return migrant household –0.077 (0.693) Other controls No Yes Yes Chi2 4.301 242.3 94.03 Log likelihood –133.999 –78.460 –75.182 R² 0.015 0.413 0.437 N 305 292 292 (1) (2) (3) Return migrant household –0.376 –0.571 –0.374 (0.241) (0.496) (2.667) Migrant household –0.347 –0.615* –2.937 (0.270) (0.373) (12.52) Affected intensity * Migrant household 0.628 (2.752) Affected intensity * Return migrant household –0.077 (0.693) Other controls No Yes Yes Chi2 4.301 242.3 94.03 Log likelihood –133.999 –78.460 –75.182 R² 0.015 0.413 0.437 N 305 292 292 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table 3. No destruction of house (1) (2) (3) Return migrant household 0.170 0.197 0.175 (0.214) (0.288) (0.408) Migrant household –0.317 –0.362 –0.065 (0.216) (0.232) (0.299) Affected intensity * Migrant household –0.198 (0.159) Affected intensity * Return migrant household 0.0108 (0.153) Other controls No Yes Yes Chi2 2.472 72.75 75.42 Log likelihood –208.591 –165.972 –164.709 R² 0.013 0.215 0.221 N 305 305 305 (1) (2) (3) Return migrant household 0.170 0.197 0.175 (0.214) (0.288) (0.408) Migrant household –0.317 –0.362 –0.065 (0.216) (0.232) (0.299) Affected intensity * Migrant household –0.198 (0.159) Affected intensity * Return migrant household 0.0108 (0.153) Other controls No Yes Yes Chi2 2.472 72.75 75.42 Log likelihood –208.591 –165.972 –164.709 R² 0.013 0.215 0.221 N 305 305 305 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table 3. No destruction of house (1) (2) (3) Return migrant household 0.170 0.197 0.175 (0.214) (0.288) (0.408) Migrant household –0.317 –0.362 –0.065 (0.216) (0.232) (0.299) Affected intensity * Migrant household –0.198 (0.159) Affected intensity * Return migrant household 0.0108 (0.153) Other controls No Yes Yes Chi2 2.472 72.75 75.42 Log likelihood –208.591 –165.972 –164.709 R² 0.013 0.215 0.221 N 305 305 305 (1) (2) (3) Return migrant household 0.170 0.197 0.175 (0.214) (0.288) (0.408) Migrant household –0.317 –0.362 –0.065 (0.216) (0.232) (0.299) Affected intensity * Migrant household –0.198 (0.159) Affected intensity * Return migrant household 0.0108 (0.153) Other controls No Yes Yes Chi2 2.472 72.75 75.42 Log likelihood –208.591 –165.972 –164.709 R² 0.013 0.215 0.221 N 305 305 305 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Construction of strong houses is an important measure for ex-ante preparedness for future disasters, in particular in an earthquake-prone country like Nepal. Our analysis in Tables 2 and 3 do not find any statistically significant relation between migration status of households and ex-ante preparedness except in column (2) of Table 2. The same is similar both in the case of migrant and return migrant households. In column (2) of Table 2, we only find a marginally negative relationship between the full destruction of houses and households with migrant workers currently working in the Republic of Korea. In column (3) of Tables 2 and 3, the interaction coefficients are also not statistically significant. 7.2 Ex-post community help In Table 4 we analyse households’ behaviour towards ex-post community helping based on their migration characteristics.21 Columns (1), (3) and (5) estimate the likelihood of a households’ participation in helping, and columns (2), (4) and (6) estimate the absolute amount of help conditional on households who choose to participate. We keep non-migrant households as our control group for the analysis and compare them with migrant and return migrant households. Table 4. Relative participation and absolute donation (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.651** 3.722*** 0.693*** 3.791*** 0.561** 3.472** (0.286) (1.368) (0.266) (1.428) (0.284) (1.682) Migrant household 0.536*** 3.714*** 0.534*** 3.517*** 0.530*** 3.561*** (0.176) (0.942) (0.163) (0.934) (0.167) (1.027) Age No No No No Yes Yes Other controls No No Yes Yes Yes Yes Sigma 6.947*** 6.838*** 6.779*** (0.447) (0.435) (0.426) Chi2 9.801 16.49 37.85 40.61 47.27 59.31 R² 0.038 0.013 0.053 0.018 0.074 0.022 Log likelihood –179.596 –604.281 –176.67 –601.044 –172.816 –598.724 Return migrant household = Migrant household 0.630 0.995 0.483 0.830 0.900 0.945 N 283 258 283 258 283 258 (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.651** 3.722*** 0.693*** 3.791*** 0.561** 3.472** (0.286) (1.368) (0.266) (1.428) (0.284) (1.682) Migrant household 0.536*** 3.714*** 0.534*** 3.517*** 0.530*** 3.561*** (0.176) (0.942) (0.163) (0.934) (0.167) (1.027) Age No No No No Yes Yes Other controls No No Yes Yes Yes Yes Sigma 6.947*** 6.838*** 6.779*** (0.447) (0.435) (0.426) Chi2 9.801 16.49 37.85 40.61 47.27 59.31 R² 0.038 0.013 0.053 0.018 0.074 0.022 Log likelihood –179.596 –604.281 –176.67 –601.044 –172.816 –598.724 Return migrant household = Migrant household 0.630 0.995 0.483 0.830 0.900 0.945 N 283 258 283 258 283 258 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. Table 4. Relative participation and absolute donation (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.651** 3.722*** 0.693*** 3.791*** 0.561** 3.472** (0.286) (1.368) (0.266) (1.428) (0.284) (1.682) Migrant household 0.536*** 3.714*** 0.534*** 3.517*** 0.530*** 3.561*** (0.176) (0.942) (0.163) (0.934) (0.167) (1.027) Age No No No No Yes Yes Other controls No No Yes Yes Yes Yes Sigma 6.947*** 6.838*** 6.779*** (0.447) (0.435) (0.426) Chi2 9.801 16.49 37.85 40.61 47.27 59.31 R² 0.038 0.013 0.053 0.018 0.074 0.022 Log likelihood –179.596 –604.281 –176.67 –601.044 –172.816 –598.724 Return migrant household = Migrant household 0.630 0.995 0.483 0.830 0.900 0.945 N 283 258 283 258 283 258 (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.651** 3.722*** 0.693*** 3.791*** 0.561** 3.472** (0.286) (1.368) (0.266) (1.428) (0.284) (1.682) Migrant household 0.536*** 3.714*** 0.534*** 3.517*** 0.530*** 3.561*** (0.176) (0.942) (0.163) (0.934) (0.167) (1.027) Age No No No No Yes Yes Other controls No No Yes Yes Yes Yes Sigma 6.947*** 6.838*** 6.779*** (0.447) (0.435) (0.426) Chi2 9.801 16.49 37.85 40.61 47.27 59.31 R² 0.038 0.013 0.053 0.018 0.074 0.022 Log likelihood –179.596 –604.281 –176.67 –601.044 –172.816 –598.724 Return migrant household = Migrant household 0.630 0.995 0.483 0.830 0.900 0.945 N 283 258 283 258 283 258 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. The estimates in Table 4 show a positive and statistically significant relationship between migration and households’ participation in helping the community. The results in column (1) show a higher likelihood of migrant as well as return migrant households participating in helping after the disaster. After controlling for other household characteristics in columns (3) and (5), the coefficient estimates of migrant and return migrant households are still statistically significant. On average, migrant and return migrant households are 18.8 to 19 and 19 to 23 percentage points more likely to participate in helping in comparison with non-migrant households respectively. The estimation results in column (2) of Table 4 show a positive and statistically significant relationship between migration characteristics of a household and their absolute amount of community help. Both migrant and return migrant households help more in comparison to non-migrant households. After controlling for other household characteristics in columns (4) and (6), the coefficient estimates are still statistically significant. Conditional on the participation decision, on average, migrant and return migrant households help 352 to 371 and 372 to 379 percentage points more in comparison with non-migrant households respectively. Overall, migrant and return migrant households donate 256 to 270 and 261 to 279 percentage points more on average as compared with non-migrant households respectively.22 We are also interested in estimating whether there is any difference between migrant and return migrant households towards participation decisions and absolute amount of help. We present the probabilistic statistics in each column (H0: Return migrant household = migrant household). Our results do not find any statistically significant difference between migrant and return migrant households on their likelihood of participation and absolute amount of community help. We are also interested in analysing the households’ behaviour within migration status. Social pressure to help and altruism might be related to the likelihood of participation and absolute amount of help. For example, expectation or social pressure to help and altruism would be higher among migrant households living in highly affected districts, as the earthquake did not affect a majority part of their income. To see the effect we additionally control for interaction terms of affected intensity with migration characteristics of households. We show the results in Table 5. The interaction coefficient estimates are not statistically significant. So the results in Table 5 do not support the altruistic or community sharing norm behaviour of migrant and return migrant households. However, the hypothesis of donor fatigue based on the place of residence at the destination country cannot be rejected. Table 5. Behaviour within households (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.837* 4.265*** (0.429) (1.652) Migrant household 0.400** 2.599* (0.192) (1.343) Affected Intensity –0.016 0.066 (0.060) (0.351) Affected Intensity * Migrant household 0.082 0.531 (0.092) (0.524) Affected Intensity * Return migrant household –0.135 –0.399 (0.185) (0.995) Other controls Yes Yes sigma 6.757*** (0.423) chi² 45.52 81.33 Log Likelihood –171.756 –598.012 R² 0.080 0.023 N 283 258 (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.837* 4.265*** (0.429) (1.652) Migrant household 0.400** 2.599* (0.192) (1.343) Affected Intensity –0.016 0.066 (0.060) (0.351) Affected Intensity * Migrant household 0.082 0.531 (0.092) (0.524) Affected Intensity * Return migrant household –0.135 –0.399 (0.185) (0.995) Other controls Yes Yes sigma 6.757*** (0.423) chi² 45.52 81.33 Log Likelihood –171.756 –598.012 R² 0.080 0.023 N 283 258 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. Table 5. Behaviour within households (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.837* 4.265*** (0.429) (1.652) Migrant household 0.400** 2.599* (0.192) (1.343) Affected Intensity –0.016 0.066 (0.060) (0.351) Affected Intensity * Migrant household 0.082 0.531 (0.092) (0.524) Affected Intensity * Return migrant household –0.135 –0.399 (0.185) (0.995) Other controls Yes Yes sigma 6.757*** (0.423) chi² 45.52 81.33 Log Likelihood –171.756 –598.012 R² 0.080 0.023 N 283 258 (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.837* 4.265*** (0.429) (1.652) Migrant household 0.400** 2.599* (0.192) (1.343) Affected Intensity –0.016 0.066 (0.060) (0.351) Affected Intensity * Migrant household 0.082 0.531 (0.092) (0.524) Affected Intensity * Return migrant household –0.135 –0.399 (0.185) (0.995) Other controls Yes Yes sigma 6.757*** (0.423) chi² 45.52 81.33 Log Likelihood –171.756 –598.012 R² 0.080 0.023 N 283 258 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. 8. Robustness check In this paper, we employed modified zero-order regression to deal with the missing values as instructed in Greene (2010) for our primary results. To test the robustness of our main findings, we present the results without considering the missing values. The regression results on household’s behaviour towards ex-ante preparedness are shown in Tables A1 and A2 in the Appendix. We analyse full destruction of houses in Table A1 and no destruction of houses in Table A2 as our dependent variable. The results in Tables A1 and A2 are quite similar to the results in Tables 2 and 3 respectively. In Table A3 we analyse households’ behaviour towards ex-post community helping based on their migration characteristics. Columns (1), (3) and (5) estimate the likelihood of households’ participation decision whereas columns (2), (4) and (6) estimate the absolute amount of help conditional on households who choose to participate. The results in Table A3 are quite similar to the results in Table 4. In Table A4 we analyse the households’ behaviour within migration status similar to Table 5 and find quite similar results as well. 9. Discussion and conclusion In this paper, we studied the difference in households’ migration characteristics and their behaviour towards ex-ante preparedness for a future disaster. Additionally, we also analysed the differences in their behaviour towards ex-post helping towards community after a natural disaster. We considered the recent earthquake in Nepal in 2015 as an exogenous shock for our analysis. Furthermore, as migration tends to be self-selected, we use a unique random selection policy of migration to the Republic of Korea as our identification strategy to eliminate self-selection bias. In contrast to earlier literature, we did not find any statistically significant results in relation to migration and ex-ante preparedness for a future disaster. We interpret our results in two different ways. First, we support the previous literature by Manandhar (2016). Although migrant and return migrant households have a higher likelihood of possessing concrete houses and invest a significant amount of remittances on the construction of houses, lower awareness of building code for safe construction makes them vulnerable similar to non-migrant households against a future disaster. The second interpretation of our results has relied on the severity of the natural disaster. The earthquake in Nepal in 2015 was scaled at 7.9 on the Richter scale, which is severe according to international standards. It is also one of the worst disasters in the history of Nepal. Even if migrant and return migrant households have followed some sorts of ex-ante preparedness as Mohapatra et al. (2012) found, it may not be sufficient to stand with such a severe natural catastrophe. Simple concrete or brick houses might be disaster proof towards usual events like flood or cyclone but may not be resistant to very unexpected events like the earthquake. However, we found positive and statistically significant evidence in support of migration on the likelihood of participation in the community helping after a disaster in their home country. Furthermore, migrant and return migrant households helped significantly higher amount towards ex-post community recovery. On average, migrant and return migrant households were 18 to 23 percentage points more likely to participate in helping in comparison with non-migrant households. Furthermore, conditional on the participation decision, on average, migrant and return migrant households help 352 to 379 percentage points more in comparison with non-migrant households. Overall, migrant and return migrant households donate 256 to 279 percentage points more on average as compared with non-migrant households respectively. Our general interpretation of the varying households’ behaviour towards the participation decision and absolute amount of help is that the annual income of migrant and returned-migrant households is much higher than the annual income of non-migrant households. It is true as the real wages in developed countries are much higher in comparison to developing countries, even for equivalent workers (Ashenfelter 2012). Previous literature in this field found that migration and remittances serve as self-insurance to recipient households after an exogenous shock in the country of origin. Our results additionally suggest that migration also helps in ex-post disaster recovery of neighbourhood and community. Migration in general and international migration, in particular, is self-selected. Therefore, it is possible that member of households with a certain amount of wealth or some adaptive capacity migrate to cope with the recovery while other households could be particularly vulnerable (Black et al. 2011; Mueller et al. 2014). Our findings shed light on the importance of migration on disaster recovery for those households who could not be able to engage in migration. There could be a concern that migrant and return-migrant households are different from non-migrant households based on some unobservable characteristics that we could not capture in our estimation. For example, return migrant households may be more risk-averse (risk-loving), in comparison with migrant and non-migrant households, as they are the first movers. In our analysis, it is difficult to measure these characteristics, as we do not have further information. Considering variables presented in the descriptive statistics the observable characteristics are the same on average, except the age of the migrants.23 The age variable is systematically different in accordance with the eligibility criteria. On average, non-migrants are the youngest and return-migrants are the oldest in our sample.24 Therefore, it is likely that the unobservable characteristics of the household are similar on average. However, it would be interesting to see if there is any significant difference among households and how it systematically impacts the outcome in future research. We only analysed the impact of migration on short-term ex-post community recovery after a natural disaster. Our dataset did not allow us to measure the magnitude of its impact on ex-post community recovery. Furthermore, we could not be able to track the kind of help delivered as well. Therefore, the insurance hypothesis might not be applicable. However, their valuable contributions in the time of crisis should not be underestimated. Future research in this field could address these limitations. Acknowledgements For valuable comments and suggestions, we are grateful to Toyo Ashida, Yasuyuki Sawada, Fumio Ohtake and the participants of presentations at the East Asian Economic Association Conference 2016, SU-ADBI Workshop 2016. We are also grateful to the EPS staffs in Nepal, Joobong Kim and Jelena Rkman for their valuable help. Funding This work was supported by Osaka University and Asian Development Bank Institute (ADBI). Conflict of interest statement. We have no conflicts of interest to disclose. Footnotes 1. Disaster Risk Reduction is the concept and practice of reducing disaster risks through systematic efforts to analyse and manage the causal factors of disasters, including through reduced exposure to hazards, lessened vulnerability of people and property, wise management of land and the environment, and improved preparedness for adverse events (UNISDR definition). 2. In this paper, we use the term ‘remittances’ as ‘migration induced remittances’. 3. For a literature review in this field, see Le De et al. (2013) 4. We could not differentiate between neighbourhood and community in our survey. For simplicity, we use the term community instead of neighbourhood and community from now onwards. 5. Ministry of Labor and Employment (2015). 6. <https://www.eps.go.kr/ph/index.html> accessed 1 Sept 2016. 7. <https://www.eps.go.kr/ph/index.html> accessed 1 Sept 2016. 8. The 2009 test was abandoned, possibly due to the financial crisis. The 2012 test was not conducted as the Korean government selected around 15,678 workers in 2011 which was more than three times of workers selected in 2010. No one knew beforehand that the 2009 and 2012 tests were going to be abandoned. 9. The test is also conducted to check the Korean language ability. As it is meant for returned migrant workers who have worked in the Republic of Korea, it is conducted on a smaller scale in comparison to the general test (PBT) where anyone with 18–39 years of age can participate. 10. <http://www.korea.net/Government/Briefing-Room/Press-Releases/view?articleId=1553> accessed 1 Dec 2016. 11. This group only includes individuals who have passed the paper-based test (PBT) and work in the Republic of Korea. 12. We did a pilot survey in early September and did our main survey between the third week of September to the first week of October. There was a clash between India and Nepal during the last week of September which intensified in October. As the surveying person was Indian, we had to stop the survey as the clash between India and Nepal might influence the survey. 13. Most of the people in this group leave for the Republic of Korea within one month. 14. <https://www.eps.go.kr/ph/index.html> accessed 1 Sept 2016. 15. Speiser (2015). 16. <https://data.worldbank.org/indicator/IT.NET.USER.ZS?locations=KR> accessed 13 Mar 2018. 17. The minimum hourly wage in the Republic of Korea is almost 8 to 10 times higher compared with the average hourly wage in Nepal (authors’ calculation). 18. We only surveyed those who have passed the examination, which constitutes more than half of the applicants. We exclude those who did not pass from our sample, as we could not get their personal information. 19. There are few observations in each district. Therefore, using 62 district dummy variables may reduce the precision of our estimation. Instead, we clubbed the districts based on the intensity of the earthquake and included it in our estimation. 20. Kathmandu valley constitute Kathmandu, Lalitpur and Bhaktapur districts. 21. We asked the following question to measure the participation in community helping and absolute amount of help. ‘Did your household directly help the earthquake victims or participate in community help after the earthquake? If yes approximately how much in Nepalese Rupees?’ (authors’ translation from the Nepali language). 22. Note that the sample size of participation in helping is higher than the ‘amount of help’ in our analysis. In our survey, some of the respondents responded with qualitative answers such as ‘people stayed at our place for some days’, ‘we helped with 20 kilos of rice’, and so forth. It is difficult to quantify these responses into Nepalese rupees correctly. Therefore, we could not include these responses into the amount of help sample. In our sample analysis on the ‘amount of help’, we only include the respondents who responded in Nepalese rupees. However, we have not noticed any systematic difference in response on the basis of the migration characteristics of the household. 23. Living in Kathmandu Valley is significant only for return migrant households compared with our control group. We discuss the possible reason for this difference in Section 5. 24. The age and age squared control variables are not statistically significant in any of our estimations. References Amemiya T. , Boskin M. ( 1974 ) ‘Regression Analysis When the Dependent Variable is Truncated Lognormal, with an Application to the Determinants of the Duration of Welfare Dependency’ , International Economic Review , 15 / 2 : 485 – 96 . Google Scholar CrossRef Search ADS Ashenfelter O. ( 2012 ) ‘Comparing Real Wage Rates’ , American Economic Review , 102 / 2 : 617 – 42 . Google Scholar CrossRef Search ADS Black R. et al. ( 2011 ) ‘Climate Change: Migration as Adaptation’ , Nature , 478 / 7370 : 447 – 9 . Google Scholar CrossRef Search ADS PubMed Brown R. P. , Leeves G. , Prayaga P. ( 2014 ) ‘Sharing Norm Pressures and Community Remittances: Evidence from a Natural Disaster in the Pacific Islands’ , Journal of Development Studies , 50 / 3 : 383 – 98 . Google Scholar CrossRef Search ADS Carroll J. , McCarthy S. , Newman C. ( 2005 ) ‘An Econometric Analysis of Charitable Donations in the Republic of Ireland’ , Economic and Social Review , 36 / 3 : 229 . Cragg J. G. ( 1971 ) ‘Some Statistical Models for Limited Dependent Variables with Application to the Demand for Durable Goods’ , Econometrica , 39 / 5 : 829 – 44 . Google Scholar CrossRef Search ADS Crush J. et al. ( 2012 ) ‘16 Diasporas on the Web: New Networks, New Methodologies’, in Vargas-Silva C. (ed.) Handbook of Research Methods in Migration . Edward Elgar Publishing, UK . Deshingkar P. , Aheeyar M. M. M. ( 2006 ) ‘Remittances in Crisis: Sri Lanka after the Tsunami.’ Humanitarian Policy Group Background Paper. Fagan P. ( 2006 ) Remittances in Crisis: A Haiti Case Study . London : Overseas Development Institute . Gibson J. , McKenzie D. , Stillman S. ( 2011 ) ‘The Impacts of International Migration on Remaining Household Members: Omnibus Results from a Migration Lottery Program’ , Review of Economics and Statistics , 93 / 4 : 1297 – 318 . Google Scholar CrossRef Search ADS Greene W. H. ( 2010 ) Econometric Analysis , 7th edn. Pearson Education Limited, UK . Korea.net ( 2012 ) Re-entry system for committed workers receives positive response from firms and foreign workers <http://www.korea.net/Government/Briefing-Room/Press-Releases/view?articleId=1553> accessed 31 Dec 2016. Le De L. et al. ( 2015 ) ‘Remittances in the Face of Disasters: A Case Study of Rural Samoa’ , Environment, Development and Sustainability , 17 / 3 : 653 – 72 . Google Scholar CrossRef Search ADS Le De L. , Gaillard J. C. , Friesen W. ( 2013 ) ‘Remittances and Disaster: A review’ , International Journal of Disaster Risk Reduction 4 : 34 – 43 . Google Scholar CrossRef Search ADS Linnerooth-Bayer J. , Mechler R. , Hochrainer-Stigler S. ( 2011 ) ‘Insurance Against Losses from Natural Disasters in Developing Countries: Evidence, Gaps and the Way Forward’ , Journal of Integrated Disaster Risk Management , 1 / 1 : 59 – 81 . Google Scholar CrossRef Search ADS Manandhar B. ( 2016 ) ‘Remittance and Earthquake Preparedness’ , International Journal of Disaster Risk Reduction , 15 : 52 – 60 . Google Scholar CrossRef Search ADS McKenzie D. ( 2015 ) ‘Learning about Migration through Experiments’, in Dustman C. (ed.) Migration: Economic Change, Social Challenge , pp. 68 – 82 . Oxford University Press , USA. Google Scholar CrossRef Search ADS McKenzie D. , Yang D. ( 2012 ) ‘Experimental Approaches in Migration Studies’, in Vargas-Silva C. (ed.) Handbook of Research Methods in Migration . Edward Elgar Publishing, UK . McKenzie D. , Stillman S. , Gibson J. ( 2010 ) ‘How Important is Selection? Experimental vs. Non-Experimental Measures of the Income Gains from Migration’ , Journal of the European Economic Association , 8 / 4 : 913 – 45 . Google Scholar CrossRef Search ADS Ministry of Labor and Employment ( 2015 ) Labor Migration for Employment a Status Report for Nepal: 2014/2015 . Government of Nepal, Nepal . Mohapatra S. , Joseph G. , Ratha D. ( 2012 ) ‘Remittances and Natural Disasters: Ex-post Response and Contribution to Ex-ante Preparedness’ , Environment, Development and Sustainability , 14 / 3 : 365 – 87 . Google Scholar CrossRef Search ADS Mueller V. , Gray C. , Kosec K. ( 2014 ) ‘Heat Stress Increases Long-Term Human Migration in Rural Pakistan’ , Nature Climate Change , 4 / 3 : 182 – 5 . Google Scholar CrossRef Search ADS PubMed National Planning Commission ( 2015 ) Post Disaster Needs Assessment, Vol. A. Key findings, Government of Nepal, Nepal. Newman C. , Henchion M. , Matthews A. ( 2003 ) ‘A Double-Hurdle Model of Irish Household Expenditure on Prepared Meals’ , Applied Economics , 35 / 9 : 1053 – 61 . Google Scholar CrossRef Search ADS Ratha D. , Sirkeci I. ( 2010 ) ‘Remittances and the Global Financial Crisis’ , Migration Letters , 7 / 2 : 125 – 31 . Speiser M. ( 2015 ) The 10 Countries with World’s Fastest Internet Speeds <https://www.weforum.org/agenda/2015/05/the-10-countries-with-the-worlds-fastest-internet-speeds/> accessed 1 Sept 2016. Wooldridge J. M. ( 2009 ) Hurdle and “selection” models, Michigan State UniversityBGSE/IZA Course in Micro-econometrics. Wu T. ( 2006 ) ‘The Role of Remittances in Crisis. An Aceh Research Study’, Overseas Development Institute Background working paper, London, UK. Yang D , Choi H. ( 2007 ) ‘Are Remittances Insurance? Evidence from Rainfall Shocks in the Philippines’ , The World Bank Economic Review , 21 / 2 : 219 – 48 . Google Scholar CrossRef Search ADS Zhang F. et al. ( 2008 ) ‘Modeling Fresh Organic Produce Consumption with Scanner Data: A Generalized Double Hurdle Model Approach’ , Agribusiness , 24 / 4 : 510 – 22 . Google Scholar CrossRef Search ADS Appendix Table A1: Full Destruction of House. (1) (2) (3) Return migrant household –0.365 –0.420 –0.035 (0.250) (0.561) (2.131) Migrant household –0.332 –0.564 –2.886 (0.264) (0.379) (11.36) Affected intensity * Migrant household 0.625 (2.528) Affected intensity * Return migrant household –0.138 (0.543) Other controls No Yes Yes Chi2 3.960 63.95 39.58 Log likelihood –129.319 –71.025 –67.672 R² 0.014 0.415 0.442 N 287 260 260 (1) (2) (3) Return migrant household –0.365 –0.420 –0.035 (0.250) (0.561) (2.131) Migrant household –0.332 –0.564 –2.886 (0.264) (0.379) (11.36) Affected intensity * Migrant household 0.625 (2.528) Affected intensity * Return migrant household –0.138 (0.543) Other controls No Yes Yes Chi2 3.960 63.95 39.58 Log likelihood –129.319 –71.025 –67.672 R² 0.014 0.415 0.442 N 287 260 260 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table A1: Full Destruction of House. (1) (2) (3) Return migrant household –0.365 –0.420 –0.035 (0.250) (0.561) (2.131) Migrant household –0.332 –0.564 –2.886 (0.264) (0.379) (11.36) Affected intensity * Migrant household 0.625 (2.528) Affected intensity * Return migrant household –0.138 (0.543) Other controls No Yes Yes Chi2 3.960 63.95 39.58 Log likelihood –129.319 –71.025 –67.672 R² 0.014 0.415 0.442 N 287 260 260 (1) (2) (3) Return migrant household –0.365 –0.420 –0.035 (0.250) (0.561) (2.131) Migrant household –0.332 –0.564 –2.886 (0.264) (0.379) (11.36) Affected intensity * Migrant household 0.625 (2.528) Affected intensity * Return migrant household –0.138 (0.543) Other controls No Yes Yes Chi2 3.960 63.95 39.58 Log likelihood –129.319 –71.025 –67.672 R² 0.014 0.415 0.442 N 287 260 260 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table A2: No destruction of house. (1) (2) (3) Return migrant household 0.186 0.149 0.014 Migrant household (0.222) (0.328) (0.592) –0.256 –0.295 0.144 (0.226) (0.258) (0.407) Affected intensity * Migrant household –0.273 (0.246) Affected intensity * Return migrant household 0.064 (0.202) Other controls No Yes Yes Chi2 1.635 53.25 45.11 Log likelihood –196.845 –140.224 –137.924 R² 0.010 0.221 0.234 N 287 260 260 (1) (2) (3) Return migrant household 0.186 0.149 0.014 Migrant household (0.222) (0.328) (0.592) –0.256 –0.295 0.144 (0.226) (0.258) (0.407) Affected intensity * Migrant household –0.273 (0.246) Affected intensity * Return migrant household 0.064 (0.202) Other controls No Yes Yes Chi2 1.635 53.25 45.11 Log likelihood –196.845 –140.224 –137.924 R² 0.010 0.221 0.234 N 287 260 260 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table A2: No destruction of house. (1) (2) (3) Return migrant household 0.186 0.149 0.014 Migrant household (0.222) (0.328) (0.592) –0.256 –0.295 0.144 (0.226) (0.258) (0.407) Affected intensity * Migrant household –0.273 (0.246) Affected intensity * Return migrant household 0.064 (0.202) Other controls No Yes Yes Chi2 1.635 53.25 45.11 Log likelihood –196.845 –140.224 –137.924 R² 0.010 0.221 0.234 N 287 260 260 (1) (2) (3) Return migrant household 0.186 0.149 0.014 Migrant household (0.222) (0.328) (0.592) –0.256 –0.295 0.144 (0.226) (0.258) (0.407) Affected intensity * Migrant household –0.273 (0.246) Affected intensity * Return migrant household 0.064 (0.202) Other controls No Yes Yes Chi2 1.635 53.25 45.11 Log likelihood –196.845 –140.224 –137.924 R² 0.010 0.221 0.234 N 287 260 260 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table A3: Relative participation and absolute donation. (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.578** 3.458** 0.569** 3.431** 0.511* 3.777** (0.281) (1.468) (0.275) (1.536) (0.310) (1.912) Migrant household 0.501*** 3.555*** 0.572*** 3.745*** 0.617*** 4.124*** (0.180) (1.031) (0.184) (1.010) (0.197) (1.134) Age No No No No Yes Yes Other Controls No No Yes Yes Yes Yes Sigma 6.929*** 6.774*** 6.730*** (0.466) (0.461) (0.450) Chi2 8.261 12.42 22.77 27.58 31.52 37.66 R² 0.0316 0.0117 0.0474 0.0175 0.0745 0.0239 Log likelihood –170.49 –575.771 –155.29 –535.50 –147.31 –518.05 Return migrant household = Migrant household 0.7535 0.9408 0.9922 0.8263 0.6894 0.8179 N 267 245 248 228 242 222 (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.578** 3.458** 0.569** 3.431** 0.511* 3.777** (0.281) (1.468) (0.275) (1.536) (0.310) (1.912) Migrant household 0.501*** 3.555*** 0.572*** 3.745*** 0.617*** 4.124*** (0.180) (1.031) (0.184) (1.010) (0.197) (1.134) Age No No No No Yes Yes Other Controls No No Yes Yes Yes Yes Sigma 6.929*** 6.774*** 6.730*** (0.466) (0.461) (0.450) Chi2 8.261 12.42 22.77 27.58 31.52 37.66 R² 0.0316 0.0117 0.0474 0.0175 0.0745 0.0239 Log likelihood –170.49 –575.771 –155.29 –535.50 –147.31 –518.05 Return migrant household = Migrant household 0.7535 0.9408 0.9922 0.8263 0.6894 0.8179 N 267 245 248 228 242 222 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. Table A3: Relative participation and absolute donation. (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.578** 3.458** 0.569** 3.431** 0.511* 3.777** (0.281) (1.468) (0.275) (1.536) (0.310) (1.912) Migrant household 0.501*** 3.555*** 0.572*** 3.745*** 0.617*** 4.124*** (0.180) (1.031) (0.184) (1.010) (0.197) (1.134) Age No No No No Yes Yes Other Controls No No Yes Yes Yes Yes Sigma 6.929*** 6.774*** 6.730*** (0.466) (0.461) (0.450) Chi2 8.261 12.42 22.77 27.58 31.52 37.66 R² 0.0316 0.0117 0.0474 0.0175 0.0745 0.0239 Log likelihood –170.49 –575.771 –155.29 –535.50 –147.31 –518.05 Return migrant household = Migrant household 0.7535 0.9408 0.9922 0.8263 0.6894 0.8179 N 267 245 248 228 242 222 (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.578** 3.458** 0.569** 3.431** 0.511* 3.777** (0.281) (1.468) (0.275) (1.536) (0.310) (1.912) Migrant household 0.501*** 3.555*** 0.572*** 3.745*** 0.617*** 4.124*** (0.180) (1.031) (0.184) (1.010) (0.197) (1.134) Age No No No No Yes Yes Other Controls No No Yes Yes Yes Yes Sigma 6.929*** 6.774*** 6.730*** (0.466) (0.461) (0.450) Chi2 8.261 12.42 22.77 27.58 31.52 37.66 R² 0.0316 0.0117 0.0474 0.0175 0.0745 0.0239 Log likelihood –170.49 –575.771 –155.29 –535.50 –147.31 –518.05 Return migrant household = Migrant household 0.7535 0.9408 0.9922 0.8263 0.6894 0.8179 N 267 245 248 228 242 222 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. Table A4: Behaviour within households. (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.595 4.011* (0.486) (2.305) Migrant household 0.430 3.030* (0.269) (1.639) Affected intensity −0.006 0.196 (0.065) (0.423) Affected intensity * Migrant household 0.105 0.555 (0.102) (0.551) Affected intensity * Return migrant household −0.043 −0.131 (0.186) (1.045) Other controls Yes Yes Sigma 6.711*** (0.447) Chi2 33.19 44.00 R² 0.078 0.025 Log likelihood −146.756 −517.575 N 242 222 (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.595 4.011* (0.486) (2.305) Migrant household 0.430 3.030* (0.269) (1.639) Affected intensity −0.006 0.196 (0.065) (0.423) Affected intensity * Migrant household 0.105 0.555 (0.102) (0.551) Affected intensity * Return migrant household −0.043 −0.131 (0.186) (1.045) Other controls Yes Yes Sigma 6.711*** (0.447) Chi2 33.19 44.00 R² 0.078 0.025 Log likelihood −146.756 −517.575 N 242 222 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. Table A4: Behaviour within households. (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.595 4.011* (0.486) (2.305) Migrant household 0.430 3.030* (0.269) (1.639) Affected intensity −0.006 0.196 (0.065) (0.423) Affected intensity * Migrant household 0.105 0.555 (0.102) (0.551) Affected intensity * Return migrant household −0.043 −0.131 (0.186) (1.045) Other controls Yes Yes Sigma 6.711*** (0.447) Chi2 33.19 44.00 R² 0.078 0.025 Log likelihood −146.756 −517.575 N 242 222 (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.595 4.011* (0.486) (2.305) Migrant household 0.430 3.030* (0.269) (1.639) Affected intensity −0.006 0.196 (0.065) (0.423) Affected intensity * Migrant household 0.105 0.555 (0.102) (0.551) Affected intensity * Return migrant household −0.043 −0.131 (0.186) (1.045) Other controls Yes Yes Sigma 6.711*** (0.447) Chi2 33.19 44.00 R² 0.078 0.025 Log likelihood −146.756 −517.575 N 242 222 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Migration Studies Oxford University Press

Migration and natural disaster: Ex-ante preparedness and contribution to ex-post community recovery

Loading next page...
 
/lp/ou_press/migration-and-natural-disaster-ex-ante-preparedness-and-contribution-lMUOP7wdaU
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com
ISSN
2049-5838
eISSN
2049-5846
D.O.I.
10.1093/migration/mny006
Publisher site
See Article on Publisher Site

Abstract

Abstract Economics literature suggests that migration and especially remittances serve as insurance to migrant households in the ex-post recovery. However, the evidence of migration or remittances on ex-ante preparedness has not been focused much in the literature. Additionally, little attention has been devoted to analysing the behavioural difference between migrant and non-migrant households towards ex-post community recovery after an external shock in their country of origin. In this paper, we study the differences in migration characteristics of a household and their behaviour towards the ex-ante preparedness for future disaster. Furthermore, we analyse the differences in their behaviour towards the ex-post recovery of neighbourhood and community in the home country. The earthquake in Nepal in 2015 is considered for the analysis. As migration tends to be self-selected, we use a unique random selection policy of migration to the Republic of Korea as our identification strategy to eliminate self-selection bias. Our empirical results do not find any relationship between migration characteristics of a household and the likelihood of ex-ante preparedness for future disasters. However, we find a substantial difference in their behaviour towards the ex-post recovery of neighbourhood and community. On average, migrant and return migrant households are 18 to 23 percentage points more likely to participate in the community helping and contribute 257 to 279 percentage points more in absolute terms in comparison with non-migrant households. Our findings suggest that international migration increases the relative participation and absolute amount of community helping after a disaster in the home country. 1. Introduction Insurance against natural disasters is relatively poor or non-existent in developing countries. However, the rate of destruction and fatalities due to natural disasters are higher in low-income countries in comparison with high-income countries (see Fig. 1). As a disaster risk reduction strategy, ex-ante and ex-post actions taken by the households in disaster-prone regions, especially in developing countries can substantially reduce the loss of human life and vulnerability.1 Economics literature suggests that migration and especially remittances serve as insurance to migrant households in the ex-post recovery (Yang and Choi 2007).2 There is also a large literature in the field of climate change asserting that migration or mobility can act as an important coping strategy for the households following the events (Black et al. 2011; Mueller et al. 2014). However, the evidence of migration or the impact of remittances on ex-ante preparedness is thin (Mohapatra et al. 2012). Additionally, little attention has been devoted to analysing the behavioural difference between migrant and non-migrant households towards ex-post community recovery after an external shock in their country of origin. Empirical evidence suggests that migrants help others after a disaster in the home country (Fagan 2006; Brown et al. 2014). However, to the best of our knowledge, we have not come across any study that tries to compare the help provided by migrant and non-migrant (households). Taking these scenarios into consideration, we examine the following two questions in this paper: (1) Do overseas migrants and their households from developing countries act differently from non-migrant households towards the ex-ante preparedness for future disaster? (2) Do they also behave and act differently towards helping the community or neighbourhood in the ex-post recovery? We consider the earthquake in Nepal in 2015 for our analysis. Figure 1. View largeDownload slide Differential burden of natural disasters. Source: Linnerooth-Bayer, J., Mechler, R. and Hochrainer, S. (2011) ‘Insurance against Losses from Natural Disasters in Developing Countries. Evidence, gaps and the way forward’, IDRiM Journal, 1/1: 59–81. Data source: Munich Re, 2005. Figure 1. View largeDownload slide Differential burden of natural disasters. Source: Linnerooth-Bayer, J., Mechler, R. and Hochrainer, S. (2011) ‘Insurance against Losses from Natural Disasters in Developing Countries. Evidence, gaps and the way forward’, IDRiM Journal, 1/1: 59–81. Data source: Munich Re, 2005. Forming an unbiased control group is a severe challenge to researchers attempting to analyse the impact of migration, as migrants tend to be self-selected (Gibson et al. 2011; McKenzie and Yang 2012; McKenzie 2015). A host of observable and unobservable differences exist between people who choose to migrate and those who stay behind. Therefore, the estimation results might be misleading, as it is quite difficult to isolate the effect of migration by controlling for all other differences. A better comparison is possible if the control and treatment groups have same characteristics or similar in all dimensions. In other words, migration of an individual or household is random or by chance. To address the challenges of self-selection biases, we exploit the Employment Permit System (EPS) policy. EPS is a programme under which workers are randomly selected to work in the Republic of Korea over a period of two years after passing a Korean language examination. The Republic of Korea is also one of the most desired destinations to work among Nepalese people due to high wages, good working conditions and no corruption in hiring. In this paper, we analyse three kinds of households based on their migration characteristics: non-migrant households, migrant households and return migrant households. First, we examine the differences in migrant, non-migrant and return migrant households’ behaviour towards ex-ante disaster preparedness. There are various ways to measure households’ ex-ante disaster preparedness. For example, households may engage in housing improvement as a disaster preparedness strategy. In this paper, we consider the rate of destruction of houses in the Nepal earthquake as our measure of ex-ante preparedness considering housing improvement as a disaster risk reduction strategy. We consider full destruction, partial destruction, and no destruction of houses to measure the intensity of ex-ante preparedness of a household. Our empirical results do not find any relationship between migration characteristics of a household and the likelihood of ex-ante preparedness. The results are similar in the case of both migrant and return migrant households. Our paper provides the first evidence of the reality of ex-ante disaster preparedness among non-migrant, migrant and return migrant households. A considerable amount of empirical research has focused on the impact of international migration, especially remittances on the ex-post disaster recovery of their own households in the home country (Deshingkar and Aheeyar 2006; Yang and Choi 2007; Ratha and Sirkeci 2010).3 However, there is a possibility that the migrants and their households also help others, especially their neighbours and community after a disaster, as the income of overseas migrant workers is not affected by an external shock in their home country.4 To evaluate the impact of migration on households’ behaviour towards the community after the disaster, we analyse the differences in the migration status of a household and their behaviour towards ex-post disaster recovery of their community. We use Hurdle model (Cragg 1971) for our analysis. The first hurdle examines the likelihood of participation of a household in community help, whereas the second hurdle examines their absolute amount of help if the household chose to participate. Our empirical findings show that migrant and return migrant households have a higher likelihood of choosing to participate in the community helping in comparison with non-migrant households. Furthermore, we also find that migrant and return migrant households help substantially more in absolute terms as compared with non-migrant households. On average, migrant and return migrant households are 18 to 23 percentage points more likely to participate in the community helping and contribute 256 to 279 percentage points more in absolute terms in comparison with non-migrant households. We do not find any statistically significant difference between migrant and return migrant households in their relative and absolute amount of help. Our findings suggest that international migration increases the relative participation and absolute amount of community helping after a disaster in the home country. This paper contributes to three strands of the literature. First, we contribute to the limited literature that studies the impact of migration through a random experiment to eliminate the selection bias (McKenzie et al. 2010; Gibson et al. 2011). We did it by using a unique random selection policy of migration to the Republic of Korea as our identification strategy. Second, we are the first to analyse the reality of the impact of international migration on ex-ante preparedness towards natural disasters as our survey was conducted after the earthquake in Nepal in 2015. Earlier studies in this field (Mohapatra et al. 2012; Manandhar 2016) have considered the construction of houses (mud and brick or concrete) or the quality of construction as a measure of disaster preparedness. However, we considered the intensity of destruction of houses (i.e. full destruction, partial destruction and no destruction) as our measure to analyse the ex-ante preparedness. Third, we are the first to offer evidence on behavioural differences among migrant, return migrant and non-migrant households towards ex-post community recovery after a natural disaster in the home country. Our findings also shed light on the positive impact of migration on disaster recovery for those households who could not be able to engage in migration. The rest of the paper is organized as follows. Section 2 reviews the related literature; Section 3 gives a preliminary idea about the EPS policy in Nepal; Section 4 describes our survey methods; Section 5 describes the data and test the randomization of our control and treatment groups; Section 6 describes the empirical model; Section 7 presents our results; Section 8 presents robustness check to our main findings; and Section 9 summarizes and concludes our findings. 2. Related research The impact of migration on disaster recovery has been extensively studied in the field of economics. Most research concentrates on the flow of remittances. For example, Wu (2006) and Deshingkar and Aheeyar (2006) studied the flow of remittances aftermath of the tsunami in Ache (Indonesia) and Sri Lanka respectively and found that remittances are very fast in reaching affected migrant households and sometimes even more quickly than international aid and government assistance. Le De et al. (2015) studied Samoan households that were affected by the tsunami in 2009 and further hit by Cyclone Evan in 2012. Their study found that remittances are very fast in reaching those affected by the disaster and even remain high in the long run to cope with and recover from the disaster. They also found a complementary relationship between remittances and external aid. Furthermore, Yang and Choi (2007) studied rainfall shock in the Philippines and concluded that remittances serve as a self-insurance to the recipient households. Ex-ante actions taken by households in disaster-prone regions and especially in developing countries can substantially reduce the loss of human life and vulnerability. As a disaster risk reduction strategy migrant households can invest in housing improvements. However, the literature on the impact of migration on ex-ante preparedness for future disasters is thin. Mohapatra et al. (2012) studied the impact of remittances on ex-ante preparedness in Burkina Faso and Ghana. They found that remittance-receiving households, especially from developed countries, tend to have housing built of concrete rather than mud and have greater access to communication. Therefore, they concluded that remittance-receiving households are better prepared against natural disasters as concrete houses are more disaster resilient than mud and brick houses. In contrast, Manandhar (2016) studied the remittances from Qatar and the Republic of Korea and its impact on earthquake preparedness in Nepal and found that remittance-dependent households have a higher likelihood of possessing concrete houses and invest 18 to 22% of their total remittances for the construction of houses. However, he additionally found that the use of remittances for better quality and stronger houses, using engineer and building code awareness for safe construction, tend to decrease. Therefore, he concluded that remittances are fuelling unsafe construction practices in Nepal and increasing risks caused by an earthquake. Mohapatra et al. (2012) considered the construction of concrete houses and Manandhar (2016) examined the quality of construction of houses as a measure of ex-ante preparedness towards a future disaster. However, we have not come across any study that tries to evaluate the reality of the situation, especially after a disaster. Our paper provides the first evidence of the reality of ex-ante disaster preparedness considering the level of destruction of houses after a disaster. We consider full destruction, partial destruction, and no destruction of houses as the measures of ex-ante preparedness. There is a higher possibility that migrants and their households help others in ex-post disaster recovery, as the income of overseas migrant workers is not affected by an external shock in their home country. For example, Brown et al. (2014) studied migration and wider community sharing norm after Cyclone Pat in 2010 in the Cook Islands. They found that Cook Islanders living in regional locations (Riverina) have higher propensity to remit to others (community remittances) in comparison to migrants living in the metropolitan region (Sydney). They concluded that there exists a ‘donor fatigue’ where the social pressure is stronger in the metropolitan region. Furthermore, anecdotal evidence by Fagan (2006) also shows that in-kind transfer from friends and relatives abroad played an important role in relieving the immediate distress caused by Cyclone Jeane in Haiti in 2004. These evidences show that migration not only helps in recovery of their own households, but they also help others after a disaster in their countries of origin. However, little attention has been devoted to analysing the behavioural difference between migrant and non-migrant households towards ex-post community recovery after an external shock in their country of origin. We investigate the difference in migration characteristics of households and their behaviour towards the likelihood of participation and absolute amount help towards the community. Furthermore, we consider total help by a migrant household (including help by household members living in the home country instead of only members living in the destination country) for the following reasons. First, either migrant(s) or their household members living in the country of origin or both might be participating in community help. Second, the total income of the household includes the income of the migrant so the participation decision could be jointly determined. 3. EPS policy in Nepal In 2007, the Government of Nepal and the Republic of Korea signed an MOU under the Employment Permit System (EPS) that allows Nepali workers to work in the Republic of Korea. It is a government-to-government agreement and currently accessible to 15 Asian countries. In Nepal, it is managed by the EPS Korea Section, under the Department of Foreign Employment (DOFE). The Republic of Korea has emerged as a favourable destination country among Nepalese because of the government-to-government initiative that enables higher wages and better living and working conditions. Also, unlike other foreign employment programmes where the recruiting agencies may charge a higher amount of money over and above the actual amount, there is no corruption involved in this system. As of 2014/2015 33,960 individuals (31,771 men and 2189 women) have participated in this programme since 2008.5 EPS workers work in small enterprises (fewer than 300 employees) and perform mostly low-skilled jobs. Nepalese workers under the EPS system generally work in manufacturing and agriculture and livestock sectors. The average wage EPS workers receive in the Republic of Korea is quite high in comparison to the average wage in Nepal. Individuals between the age of 18 and 39 with no criminal records are eligible to apply for this programme. Furthermore, all the applicants have to pass a Korean Language Test (Paper Based Test or PBT) to be eligible to work.6 The minimum score to pass the examination is 80 out of 200. However, due to a limited number of quotas every year, a designated number of candidates pass in order of their highest score on the test. The test score is valid for two years from the date of announcement of the result. After passing the test, workers fill out their job application form at the EPS Korea Section in Nepal. Then, each one is randomly drawn and introduced to an employer in the Republic of Korea.7 So a worker’s timing of going to the Republic of Korea for work is randomly decided over two years. Furthermore, the Korean language test is conducted almost every year.8 As the randomization process of introducing workers goes on for two years, workers from two different test taking years are selected simultaneously. So, the workers are treated similarly after winning the lottery even if they have appeared the examination and were selected in different years. These unique policies act as our identification strategy to eliminate the selection biases. Under the EPS policy, it is mandatory for workers to return to their home countries after finishing their contract. The general contract period is three years and can further be extended up to four years and 10 months. Foreign workers who have worked for four years and ten months without changing their workplace are considered as ‘committed workers’. Workers with these characteristics can come back to the Republic of Korea with a new contract after three months of departure and work for an additional four years and ten months in the same organization. Workers who have worked three years or more under the EPS system but without the characteristics mentioned above, that belong to the age group between 18 and 39 years, and have no illegal records are eligible to go to the Republic of Korea through a special Korean language test (Computer-Based Test or CBT). This test is conducted four times per year.9 Workers who pass this test can easily and quickly re-enter the Republic of Korea six months after their departure. As the average wage in the Republic of Korea is very high in comparison to Nepal, most workers under these categories opt to go back again.10 The Special Korean language examination (CBT) started in 2013 in Nepal, and as of 2015 September, 11 CBT examinations have already been conducted. We analyse three kinds of households based on their migration characteristics: non-migrant, migrant and return migrant. We term ‘non-migrant households’ as those with a member who has passed the PBT, won the EPS lottery and were in the process of going to the Republic of Korea to work for the first time. We consider this group as households with no migration experience. We term ‘migrant households’ as those with a member who is presently working in the Republic of Korea under his/her first contract.11 We consider this group as households with short-term migration experience. We term ‘return migrant households’ as those with a member who has completed his first contract period and returned from the Republic of Korea or has gone back again with a new contract. We consider this group as households with medium to long-term migration experience. We consider non-migrant households as our control group to analyse the impact of migration. 4. Survey methods A customized survey was conducted between September and October 2015.12 All the survey questions were written in simple Nepali language to avoid any difficulty in understanding. Three different groups of households were selected for the survey. The first group is ‘non-migrant households’.13 Contacting household members with these characteristics were difficult due to unavailability of personal information. We used different approaches to survey household with these characteristics. Each and every prospective worker has to go through a week of preliminary training for 45 hours in an approved public institution in the sending country after a person is selected randomly through the lottery process and concluded labour contracts with their respective employers.14 There is only one institution in Nepal that provides training for these workers. We contacted the institution and with their permission conducted the survey of those who were taking training during that time. Additionally, we also surveyed individuals who have already completed their training and were about to fly to the Republic of Korea. People come to the EPS Korea Section office in Nepal to collect their final documents (like passport, dress, and batch to be worn during the departure period) before their departure. We requested the ministry to conduct a survey at their campus. With their permission, we carried out the survey for those who were about to depart to the Republic of Korea. In this way, we avoid selection bias in our survey as we surveyed those who have been selected through the random selection process. Our second group is ‘migrant households’. It is also difficult to obtain personal information of these households because of the restriction by the ministry to disclose their personal data. We used a novel approach to contact households with these characteristics. Facebook has many community pages dedicated to Nepalese workers living in the Republic of Korea through EPS. The number of subscribers in these community pages ranges from 2,000 to 30,000 members. The community pages with most subscribers are EPS NEPAL FAMILY (around 30,000 subscribers), EPS Korea Nepal (more than 28,000 subscribers) and Eps Nepalese in Korea (more than 10,000 subscribers). People share valuable information and interact with other members of the community. They also post information regarding various events held by migrant community/groups in the Republic of Korea. We created a survey page using Google forms. We contacted the administrator of the respective community sites and posted the Google forms link containing customized survey with their permission. In this way, we avoided the spatial bias of snowballing effect and conducted the survey on a wider scale throughout the Republic of Korea. Although the mail-out survey is the preferred method for surveying migrants and diaspora, it produces a severe sample selection bias as they tend to be highly skewed (Crush et al. 2012). In this context, web-based research methodology could be of significant advantage. There is a specific chapter devoted to research methods in World Wide Web (WWW) and Social Networking Sites (SNSs) in particular, in the Handbook of Research Methods in Migration. However, one limitation could be that our sample might exclude those who do not have access to these community sites or Internet. The Republic of Korea has the fastest speed Internet in the world, and moreover, the Internet user rate was 89.65% in 2015.15,16 From interaction with those who have already returned from the Republic of Korea, we came to know that Facebook is indeed the best way of communication and that it is widely used by Nepalese workers to interact with friends in the Republic of Korea and Nepal. As most Nepalese workers work in very small organizations, browsing Facebook is one of the best ways to pass their leisure time and keep in contact with their friends and family members. Besides, the number of members in these community pages is considerably large. This evidence additionally reflects that the majority of the migrant workers have access to the Internet and these community sites. Another drawback of our sample survey method through Facebook pages could be that we may fail to recognize whether our respondents only include our target groups, as various kinds of people could be a member of these community pages. To avoid these problems, we specifically put some questions which only the target groups could answer, so we can identify respondents who belong to our target group and those who do not. Additionally, we also exclude repeated respondents from our sample. Our third group is ‘return migrant households’. As the wage in the Republic of Korea is much higher than in Nepal, most workers under these categories choose to go back again to work.17 These groups have about 3 to 7 years of migration experience. We used the Facebook mode of survey for those who have not appeared for any examinations (Committed workers) and for those who have already gone back again with a new contract during our survey period. Additionally, with the permission of the ministry in Nepal, we surveyed people who have appeared for CBT examination in September 2015 and have been qualified to go back to the Republic of Korea again.18 5. Descriptive statistics and the test of randomization We present the descriptive statistics in Table 1. In columns (1) and (2), we show the descriptive statistics of our control group households. Furthermore, we show the descriptive statistics of our treatment group households in columns (3) to (6). In columns (7) and (8) we present the difference between control and treatment group households and their statistical significance on average. The first four variables are our outcome variables, and the remaining variables are our control variables. Table 1. Descriptive statistics and test of randomization Non-migrant household Migrant household Return migrant household Difference Difference (1) (2) (3) (4) (5) (6) (7) (8) Variables Mean Std Dev. Mean Std Dev. Mean Std Dev. (3)–(1) (5)–(1) Help (Yes) 0.53 0.501 0.729 0.447 0.766 0.428 0.199*** 0.236*** Log (Amount of help+1) 4.233 4.321 6.725 4.666 6.635 4.35 2.492*** 2.402*** Full destruction of house 0.202 0.403 0.119 0.326 0.113 0.32 −0.083 −0.089 No. destruction of house 0.518 0.501 0.393 0.491 0.585 0.497 −0.125* 0.067 Affected intensity 1.989 1.868 1.871 1.696 2.077 1.69 −0.118 0.088 No. of family members 6.316 2.733 6.775 3.46 6.418 3.705 0.459 0.102 Family member(s) abroad 0.452 0.499 0.398 0.492 0.42 0.499 −0.054 -0.032 District 34.706 17.72 35.847 17.038 34.712 16.804 1.141 0.006 Education in years 12.929 2.21 13.356 2.122 13.273 2.475 0.427 0.344 Age 24.937 5.498 26.831 4.513 31.058 2.6 1.894*** 6.121*** Household living in Kathmandu Valley 0.051 0.22 0.035 0.186 0.135 0.345 −0.016 0.084** Number of observations 184 90 55 Non-migrant household Migrant household Return migrant household Difference Difference (1) (2) (3) (4) (5) (6) (7) (8) Variables Mean Std Dev. Mean Std Dev. Mean Std Dev. (3)–(1) (5)–(1) Help (Yes) 0.53 0.501 0.729 0.447 0.766 0.428 0.199*** 0.236*** Log (Amount of help+1) 4.233 4.321 6.725 4.666 6.635 4.35 2.492*** 2.402*** Full destruction of house 0.202 0.403 0.119 0.326 0.113 0.32 −0.083 −0.089 No. destruction of house 0.518 0.501 0.393 0.491 0.585 0.497 −0.125* 0.067 Affected intensity 1.989 1.868 1.871 1.696 2.077 1.69 −0.118 0.088 No. of family members 6.316 2.733 6.775 3.46 6.418 3.705 0.459 0.102 Family member(s) abroad 0.452 0.499 0.398 0.492 0.42 0.499 −0.054 -0.032 District 34.706 17.72 35.847 17.038 34.712 16.804 1.141 0.006 Education in years 12.929 2.21 13.356 2.122 13.273 2.475 0.427 0.344 Age 24.937 5.498 26.831 4.513 31.058 2.6 1.894*** 6.121*** Household living in Kathmandu Valley 0.051 0.22 0.035 0.186 0.135 0.345 −0.016 0.084** Number of observations 184 90 55 Table 1. Descriptive statistics and test of randomization Non-migrant household Migrant household Return migrant household Difference Difference (1) (2) (3) (4) (5) (6) (7) (8) Variables Mean Std Dev. Mean Std Dev. Mean Std Dev. (3)–(1) (5)–(1) Help (Yes) 0.53 0.501 0.729 0.447 0.766 0.428 0.199*** 0.236*** Log (Amount of help+1) 4.233 4.321 6.725 4.666 6.635 4.35 2.492*** 2.402*** Full destruction of house 0.202 0.403 0.119 0.326 0.113 0.32 −0.083 −0.089 No. destruction of house 0.518 0.501 0.393 0.491 0.585 0.497 −0.125* 0.067 Affected intensity 1.989 1.868 1.871 1.696 2.077 1.69 −0.118 0.088 No. of family members 6.316 2.733 6.775 3.46 6.418 3.705 0.459 0.102 Family member(s) abroad 0.452 0.499 0.398 0.492 0.42 0.499 −0.054 -0.032 District 34.706 17.72 35.847 17.038 34.712 16.804 1.141 0.006 Education in years 12.929 2.21 13.356 2.122 13.273 2.475 0.427 0.344 Age 24.937 5.498 26.831 4.513 31.058 2.6 1.894*** 6.121*** Household living in Kathmandu Valley 0.051 0.22 0.035 0.186 0.135 0.345 −0.016 0.084** Number of observations 184 90 55 Non-migrant household Migrant household Return migrant household Difference Difference (1) (2) (3) (4) (5) (6) (7) (8) Variables Mean Std Dev. Mean Std Dev. Mean Std Dev. (3)–(1) (5)–(1) Help (Yes) 0.53 0.501 0.729 0.447 0.766 0.428 0.199*** 0.236*** Log (Amount of help+1) 4.233 4.321 6.725 4.666 6.635 4.35 2.492*** 2.402*** Full destruction of house 0.202 0.403 0.119 0.326 0.113 0.32 −0.083 −0.089 No. destruction of house 0.518 0.501 0.393 0.491 0.585 0.497 −0.125* 0.067 Affected intensity 1.989 1.868 1.871 1.696 2.077 1.69 −0.118 0.088 No. of family members 6.316 2.733 6.775 3.46 6.418 3.705 0.459 0.102 Family member(s) abroad 0.452 0.499 0.398 0.492 0.42 0.499 −0.054 -0.032 District 34.706 17.72 35.847 17.038 34.712 16.804 1.141 0.006 Education in years 12.929 2.21 13.356 2.122 13.273 2.475 0.427 0.344 Age 24.937 5.498 26.831 4.513 31.058 2.6 1.894*** 6.121*** Household living in Kathmandu Valley 0.051 0.22 0.035 0.186 0.135 0.345 −0.016 0.084** Number of observations 184 90 55 The distribution of the absolute amount of help in our sample is highly skewed towards left. We show the skewness in Figure 2. Econometric models are consistent under the assumption that the normal distribution of the dependent variable is satisfied. Following Cragg (1971) and Amemiya and Boskin (1974) we use the natural logarithm of the amount of help to fix the problem of highly skewed data. Several studies have been undertaken following this approach, such as Newman et al. (2003), Carroll et al. (2005) and Zhang et al. (2008). As can be seen from Figure 3, the log transformation follows the normal distribution rather well. Figure 2. View largeDownload slide Distribution graph of amount of help with and without zero. Figure 2. View largeDownload slide Distribution graph of amount of help with and without zero. Figure 3. View largeDownload slide Distribution graph of Log (Amount of help + 1) with and without zero. Figure 3. View largeDownload slide Distribution graph of Log (Amount of help + 1) with and without zero. We constructed the disaster intensity variable in the following way. The government of Nepal has scaled the earthquake-affected districts into five categories (see Fig. 4). We assigned value 0 to those districts that were not affected at all (white parts) and assigned value 1 to 5 by the affected intensity of the respective district. We use district as our cluster variable to estimate the standard error.19 Our sample represents households from 63 districts out of a total of 75 districts in Nepal. Figure 4. View largeDownload slide Affected intensity in 2015 earthquake in Nepal. Source: National Planning Commission, (2015), Post-disaster needs assessment, Vol. A: Key findings, Government of Nepal. Figure 4. View largeDownload slide Affected intensity in 2015 earthquake in Nepal. Source: National Planning Commission, (2015), Post-disaster needs assessment, Vol. A: Key findings, Government of Nepal. Economic status of the migrant and return migrant households before their migration is a better control variable for comparison with non-migrant households as households’ economic status is highly correlated with their migration status. It is difficult to correctly evaluate the economic status of migrant and return migrant households before their migration. Therefore, we consider education (of the migrant or would be migrant worker), number of family members, household member(s) abroad, household living in Kathmandu valley as proxy variables for household economic status.20 Household member(s) abroad variable does not include the respondent who is now working in the Republic of Korea. It is important for us to check whether our control and treatment groups are uniformly distributed based on their other characteristics. We test the randomization of our treatment and control groups in columns (7) and (8). As we can observe, our control and treatment groups are uniformly distributed except the last two characteristics. Return migrant households are more likely to live in the Kathmandu Valley. This might be due to the information gap in earlier years. The EPS programme started in Nepal for the first time in 2008 and households from the Kathmandu Valley might have had a higher chance of obtaining information earlier as compared with other parts of Nepal. Another interpretation could be that migrant households could have moved to Kathmandu Valley after a period, as their income got much higher in comparison with non-migrant household’s income. We also see a significant difference in age based on our sample groups. Workers with higher international migration experience are older in comparison with workers with no migration experience. This difference is expected as there is a restriction on migration based on age imposed by the EPS. As we can see from Table 1, the percentage of households participating in community helping is higher for migrant and return migrant households. Additionally, they have helped more in absolute terms. These differences are statistically significant from zero. On average, migrant and return migrant households are 19.9 percentage points and 23.6 percentage points more likely to participate in the community helping and contribute 249.2 percentage points and 240.2 percentage points more in absolute terms respectively. However, there is no statistically significant difference between the migration characteristics of a household and the intensity of destruction of houses. No destruction of houses in the case of migrant households is only significant at the margin. 6. Model specifications We use Probit model to analyse ex-ante preparedness of households in Nepal. We estimate the extent of destruction suffered by households with different migration characteristics. We conditioned partial destruction of the house as the base and compare no destruction of a house and full destruction of a house to analyse the level of ex-ante preparedness. The model for ex-ante preparedness is as follows: Prob(y>0 |X)=1[Xγ+u>0]= Φ(Xγ) (1) In the above model, y takes value one if the destruction of a house is full (no destruction) and zero otherwise. X is the vector of independent observable characteristics and γ is the vector of coefficients. If migrant and returned-migrant households are prepared for future disasters, our results should show a negative and statistically significant coefficient towards the full destruction of a house and positive and statistically significant coefficient towards no destruction of a house. The absolute amount of community helping variable in our sample follows a lognormal distribution. So we employ lognormal hurdle model (Cragg 1971) to analyse participation decisions and absolute amount of help. There are two hurdles in the model; in the first hurdle, households decide whether or not to participate in helping. In the second hurdle, households decide how much to help if they choose to participate in helping. The equations are: y=s.w*=1Xγ+v>0 exp Xβ+u (2) Where u and v are independent, of X, and w* has a lognormal distribution, and w*=exp(Xβ+u) u|X∼Normal(0,σ²) In Equation (2)y takes the value zero or the amount of donation depending on the household’s decision to participate. X is the vector of observable independent household characteristics, and β and γ are the vectors of coefficients. The estimation is preceded in two steps (Wooldridge 2009). The coefficient γ is estimated in Probit model of s on the vector of independent variables X. Then the coefficient β is estimated by OLS regression of log(y) on X for observations with y > 0 (type II Tobit model). The expected value of donation conditional on y > 0 is Ey|X,y>0= Ew*|X,s=1=Ew*X=expXβ+σ22 (3) And the unconditional expected value of donation is Ey|X=ΦXγexpXβ+σ22 (4) We use a Lognormal hurdle model in our analysis as suggested by Brown et al. (2014). Brown et al. (2014) analysed the remittances data using both the Heckman selection model and Cragg’s double-hurdle model and concluded that Cragg’s double hurdle model is more appropriate for analysing remittances data. 7. Estimation results We use modified zero-order regression to deal with the missing variables as instructed in Greene (2010). We present all regression results with and without controlling for other characteristics to examine the robustness of our findings. We also check the robustness of our findings without considering the missing variables. 7.1 Ex-ante preparedness The regression results on household behaviour towards ex-ante preparedness are shown in Tables 2 and 3. We estimate the coefficients based on Probit model stated in Equation (1). Standard errors are clustered at the district level. We analyse full destruction of houses in Table 2 and no destruction of houses in Table 3 as our dependent variable. We condition non-migrant households as the base and compare it with migrant and return migrant households. Column (1) of Tables 2 and 3 show the estimates without any control. In column (2) we show the estimates with all controls presented in descriptive statistics. In column (3) we further include interaction variable between affected intensity and the migration characteristics to see any differences in the destruction rate. Table 2. Full destruction of house (1) (2) (3) Return migrant household –0.376 –0.571 –0.374 (0.241) (0.496) (2.667) Migrant household –0.347 –0.615* –2.937 (0.270) (0.373) (12.52) Affected intensity * Migrant household 0.628 (2.752) Affected intensity * Return migrant household –0.077 (0.693) Other controls No Yes Yes Chi2 4.301 242.3 94.03 Log likelihood –133.999 –78.460 –75.182 R² 0.015 0.413 0.437 N 305 292 292 (1) (2) (3) Return migrant household –0.376 –0.571 –0.374 (0.241) (0.496) (2.667) Migrant household –0.347 –0.615* –2.937 (0.270) (0.373) (12.52) Affected intensity * Migrant household 0.628 (2.752) Affected intensity * Return migrant household –0.077 (0.693) Other controls No Yes Yes Chi2 4.301 242.3 94.03 Log likelihood –133.999 –78.460 –75.182 R² 0.015 0.413 0.437 N 305 292 292 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table 2. Full destruction of house (1) (2) (3) Return migrant household –0.376 –0.571 –0.374 (0.241) (0.496) (2.667) Migrant household –0.347 –0.615* –2.937 (0.270) (0.373) (12.52) Affected intensity * Migrant household 0.628 (2.752) Affected intensity * Return migrant household –0.077 (0.693) Other controls No Yes Yes Chi2 4.301 242.3 94.03 Log likelihood –133.999 –78.460 –75.182 R² 0.015 0.413 0.437 N 305 292 292 (1) (2) (3) Return migrant household –0.376 –0.571 –0.374 (0.241) (0.496) (2.667) Migrant household –0.347 –0.615* –2.937 (0.270) (0.373) (12.52) Affected intensity * Migrant household 0.628 (2.752) Affected intensity * Return migrant household –0.077 (0.693) Other controls No Yes Yes Chi2 4.301 242.3 94.03 Log likelihood –133.999 –78.460 –75.182 R² 0.015 0.413 0.437 N 305 292 292 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table 3. No destruction of house (1) (2) (3) Return migrant household 0.170 0.197 0.175 (0.214) (0.288) (0.408) Migrant household –0.317 –0.362 –0.065 (0.216) (0.232) (0.299) Affected intensity * Migrant household –0.198 (0.159) Affected intensity * Return migrant household 0.0108 (0.153) Other controls No Yes Yes Chi2 2.472 72.75 75.42 Log likelihood –208.591 –165.972 –164.709 R² 0.013 0.215 0.221 N 305 305 305 (1) (2) (3) Return migrant household 0.170 0.197 0.175 (0.214) (0.288) (0.408) Migrant household –0.317 –0.362 –0.065 (0.216) (0.232) (0.299) Affected intensity * Migrant household –0.198 (0.159) Affected intensity * Return migrant household 0.0108 (0.153) Other controls No Yes Yes Chi2 2.472 72.75 75.42 Log likelihood –208.591 –165.972 –164.709 R² 0.013 0.215 0.221 N 305 305 305 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table 3. No destruction of house (1) (2) (3) Return migrant household 0.170 0.197 0.175 (0.214) (0.288) (0.408) Migrant household –0.317 –0.362 –0.065 (0.216) (0.232) (0.299) Affected intensity * Migrant household –0.198 (0.159) Affected intensity * Return migrant household 0.0108 (0.153) Other controls No Yes Yes Chi2 2.472 72.75 75.42 Log likelihood –208.591 –165.972 –164.709 R² 0.013 0.215 0.221 N 305 305 305 (1) (2) (3) Return migrant household 0.170 0.197 0.175 (0.214) (0.288) (0.408) Migrant household –0.317 –0.362 –0.065 (0.216) (0.232) (0.299) Affected intensity * Migrant household –0.198 (0.159) Affected intensity * Return migrant household 0.0108 (0.153) Other controls No Yes Yes Chi2 2.472 72.75 75.42 Log likelihood –208.591 –165.972 –164.709 R² 0.013 0.215 0.221 N 305 305 305 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Construction of strong houses is an important measure for ex-ante preparedness for future disasters, in particular in an earthquake-prone country like Nepal. Our analysis in Tables 2 and 3 do not find any statistically significant relation between migration status of households and ex-ante preparedness except in column (2) of Table 2. The same is similar both in the case of migrant and return migrant households. In column (2) of Table 2, we only find a marginally negative relationship between the full destruction of houses and households with migrant workers currently working in the Republic of Korea. In column (3) of Tables 2 and 3, the interaction coefficients are also not statistically significant. 7.2 Ex-post community help In Table 4 we analyse households’ behaviour towards ex-post community helping based on their migration characteristics.21 Columns (1), (3) and (5) estimate the likelihood of a households’ participation in helping, and columns (2), (4) and (6) estimate the absolute amount of help conditional on households who choose to participate. We keep non-migrant households as our control group for the analysis and compare them with migrant and return migrant households. Table 4. Relative participation and absolute donation (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.651** 3.722*** 0.693*** 3.791*** 0.561** 3.472** (0.286) (1.368) (0.266) (1.428) (0.284) (1.682) Migrant household 0.536*** 3.714*** 0.534*** 3.517*** 0.530*** 3.561*** (0.176) (0.942) (0.163) (0.934) (0.167) (1.027) Age No No No No Yes Yes Other controls No No Yes Yes Yes Yes Sigma 6.947*** 6.838*** 6.779*** (0.447) (0.435) (0.426) Chi2 9.801 16.49 37.85 40.61 47.27 59.31 R² 0.038 0.013 0.053 0.018 0.074 0.022 Log likelihood –179.596 –604.281 –176.67 –601.044 –172.816 –598.724 Return migrant household = Migrant household 0.630 0.995 0.483 0.830 0.900 0.945 N 283 258 283 258 283 258 (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.651** 3.722*** 0.693*** 3.791*** 0.561** 3.472** (0.286) (1.368) (0.266) (1.428) (0.284) (1.682) Migrant household 0.536*** 3.714*** 0.534*** 3.517*** 0.530*** 3.561*** (0.176) (0.942) (0.163) (0.934) (0.167) (1.027) Age No No No No Yes Yes Other controls No No Yes Yes Yes Yes Sigma 6.947*** 6.838*** 6.779*** (0.447) (0.435) (0.426) Chi2 9.801 16.49 37.85 40.61 47.27 59.31 R² 0.038 0.013 0.053 0.018 0.074 0.022 Log likelihood –179.596 –604.281 –176.67 –601.044 –172.816 –598.724 Return migrant household = Migrant household 0.630 0.995 0.483 0.830 0.900 0.945 N 283 258 283 258 283 258 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. Table 4. Relative participation and absolute donation (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.651** 3.722*** 0.693*** 3.791*** 0.561** 3.472** (0.286) (1.368) (0.266) (1.428) (0.284) (1.682) Migrant household 0.536*** 3.714*** 0.534*** 3.517*** 0.530*** 3.561*** (0.176) (0.942) (0.163) (0.934) (0.167) (1.027) Age No No No No Yes Yes Other controls No No Yes Yes Yes Yes Sigma 6.947*** 6.838*** 6.779*** (0.447) (0.435) (0.426) Chi2 9.801 16.49 37.85 40.61 47.27 59.31 R² 0.038 0.013 0.053 0.018 0.074 0.022 Log likelihood –179.596 –604.281 –176.67 –601.044 –172.816 –598.724 Return migrant household = Migrant household 0.630 0.995 0.483 0.830 0.900 0.945 N 283 258 283 258 283 258 (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.651** 3.722*** 0.693*** 3.791*** 0.561** 3.472** (0.286) (1.368) (0.266) (1.428) (0.284) (1.682) Migrant household 0.536*** 3.714*** 0.534*** 3.517*** 0.530*** 3.561*** (0.176) (0.942) (0.163) (0.934) (0.167) (1.027) Age No No No No Yes Yes Other controls No No Yes Yes Yes Yes Sigma 6.947*** 6.838*** 6.779*** (0.447) (0.435) (0.426) Chi2 9.801 16.49 37.85 40.61 47.27 59.31 R² 0.038 0.013 0.053 0.018 0.074 0.022 Log likelihood –179.596 –604.281 –176.67 –601.044 –172.816 –598.724 Return migrant household = Migrant household 0.630 0.995 0.483 0.830 0.900 0.945 N 283 258 283 258 283 258 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. The estimates in Table 4 show a positive and statistically significant relationship between migration and households’ participation in helping the community. The results in column (1) show a higher likelihood of migrant as well as return migrant households participating in helping after the disaster. After controlling for other household characteristics in columns (3) and (5), the coefficient estimates of migrant and return migrant households are still statistically significant. On average, migrant and return migrant households are 18.8 to 19 and 19 to 23 percentage points more likely to participate in helping in comparison with non-migrant households respectively. The estimation results in column (2) of Table 4 show a positive and statistically significant relationship between migration characteristics of a household and their absolute amount of community help. Both migrant and return migrant households help more in comparison to non-migrant households. After controlling for other household characteristics in columns (4) and (6), the coefficient estimates are still statistically significant. Conditional on the participation decision, on average, migrant and return migrant households help 352 to 371 and 372 to 379 percentage points more in comparison with non-migrant households respectively. Overall, migrant and return migrant households donate 256 to 270 and 261 to 279 percentage points more on average as compared with non-migrant households respectively.22 We are also interested in estimating whether there is any difference between migrant and return migrant households towards participation decisions and absolute amount of help. We present the probabilistic statistics in each column (H0: Return migrant household = migrant household). Our results do not find any statistically significant difference between migrant and return migrant households on their likelihood of participation and absolute amount of community help. We are also interested in analysing the households’ behaviour within migration status. Social pressure to help and altruism might be related to the likelihood of participation and absolute amount of help. For example, expectation or social pressure to help and altruism would be higher among migrant households living in highly affected districts, as the earthquake did not affect a majority part of their income. To see the effect we additionally control for interaction terms of affected intensity with migration characteristics of households. We show the results in Table 5. The interaction coefficient estimates are not statistically significant. So the results in Table 5 do not support the altruistic or community sharing norm behaviour of migrant and return migrant households. However, the hypothesis of donor fatigue based on the place of residence at the destination country cannot be rejected. Table 5. Behaviour within households (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.837* 4.265*** (0.429) (1.652) Migrant household 0.400** 2.599* (0.192) (1.343) Affected Intensity –0.016 0.066 (0.060) (0.351) Affected Intensity * Migrant household 0.082 0.531 (0.092) (0.524) Affected Intensity * Return migrant household –0.135 –0.399 (0.185) (0.995) Other controls Yes Yes sigma 6.757*** (0.423) chi² 45.52 81.33 Log Likelihood –171.756 –598.012 R² 0.080 0.023 N 283 258 (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.837* 4.265*** (0.429) (1.652) Migrant household 0.400** 2.599* (0.192) (1.343) Affected Intensity –0.016 0.066 (0.060) (0.351) Affected Intensity * Migrant household 0.082 0.531 (0.092) (0.524) Affected Intensity * Return migrant household –0.135 –0.399 (0.185) (0.995) Other controls Yes Yes sigma 6.757*** (0.423) chi² 45.52 81.33 Log Likelihood –171.756 –598.012 R² 0.080 0.023 N 283 258 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. Table 5. Behaviour within households (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.837* 4.265*** (0.429) (1.652) Migrant household 0.400** 2.599* (0.192) (1.343) Affected Intensity –0.016 0.066 (0.060) (0.351) Affected Intensity * Migrant household 0.082 0.531 (0.092) (0.524) Affected Intensity * Return migrant household –0.135 –0.399 (0.185) (0.995) Other controls Yes Yes sigma 6.757*** (0.423) chi² 45.52 81.33 Log Likelihood –171.756 –598.012 R² 0.080 0.023 N 283 258 (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.837* 4.265*** (0.429) (1.652) Migrant household 0.400** 2.599* (0.192) (1.343) Affected Intensity –0.016 0.066 (0.060) (0.351) Affected Intensity * Migrant household 0.082 0.531 (0.092) (0.524) Affected Intensity * Return migrant household –0.135 –0.399 (0.185) (0.995) Other controls Yes Yes sigma 6.757*** (0.423) chi² 45.52 81.33 Log Likelihood –171.756 –598.012 R² 0.080 0.023 N 283 258 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. 8. Robustness check In this paper, we employed modified zero-order regression to deal with the missing values as instructed in Greene (2010) for our primary results. To test the robustness of our main findings, we present the results without considering the missing values. The regression results on household’s behaviour towards ex-ante preparedness are shown in Tables A1 and A2 in the Appendix. We analyse full destruction of houses in Table A1 and no destruction of houses in Table A2 as our dependent variable. The results in Tables A1 and A2 are quite similar to the results in Tables 2 and 3 respectively. In Table A3 we analyse households’ behaviour towards ex-post community helping based on their migration characteristics. Columns (1), (3) and (5) estimate the likelihood of households’ participation decision whereas columns (2), (4) and (6) estimate the absolute amount of help conditional on households who choose to participate. The results in Table A3 are quite similar to the results in Table 4. In Table A4 we analyse the households’ behaviour within migration status similar to Table 5 and find quite similar results as well. 9. Discussion and conclusion In this paper, we studied the difference in households’ migration characteristics and their behaviour towards ex-ante preparedness for a future disaster. Additionally, we also analysed the differences in their behaviour towards ex-post helping towards community after a natural disaster. We considered the recent earthquake in Nepal in 2015 as an exogenous shock for our analysis. Furthermore, as migration tends to be self-selected, we use a unique random selection policy of migration to the Republic of Korea as our identification strategy to eliminate self-selection bias. In contrast to earlier literature, we did not find any statistically significant results in relation to migration and ex-ante preparedness for a future disaster. We interpret our results in two different ways. First, we support the previous literature by Manandhar (2016). Although migrant and return migrant households have a higher likelihood of possessing concrete houses and invest a significant amount of remittances on the construction of houses, lower awareness of building code for safe construction makes them vulnerable similar to non-migrant households against a future disaster. The second interpretation of our results has relied on the severity of the natural disaster. The earthquake in Nepal in 2015 was scaled at 7.9 on the Richter scale, which is severe according to international standards. It is also one of the worst disasters in the history of Nepal. Even if migrant and return migrant households have followed some sorts of ex-ante preparedness as Mohapatra et al. (2012) found, it may not be sufficient to stand with such a severe natural catastrophe. Simple concrete or brick houses might be disaster proof towards usual events like flood or cyclone but may not be resistant to very unexpected events like the earthquake. However, we found positive and statistically significant evidence in support of migration on the likelihood of participation in the community helping after a disaster in their home country. Furthermore, migrant and return migrant households helped significantly higher amount towards ex-post community recovery. On average, migrant and return migrant households were 18 to 23 percentage points more likely to participate in helping in comparison with non-migrant households. Furthermore, conditional on the participation decision, on average, migrant and return migrant households help 352 to 379 percentage points more in comparison with non-migrant households. Overall, migrant and return migrant households donate 256 to 279 percentage points more on average as compared with non-migrant households respectively. Our general interpretation of the varying households’ behaviour towards the participation decision and absolute amount of help is that the annual income of migrant and returned-migrant households is much higher than the annual income of non-migrant households. It is true as the real wages in developed countries are much higher in comparison to developing countries, even for equivalent workers (Ashenfelter 2012). Previous literature in this field found that migration and remittances serve as self-insurance to recipient households after an exogenous shock in the country of origin. Our results additionally suggest that migration also helps in ex-post disaster recovery of neighbourhood and community. Migration in general and international migration, in particular, is self-selected. Therefore, it is possible that member of households with a certain amount of wealth or some adaptive capacity migrate to cope with the recovery while other households could be particularly vulnerable (Black et al. 2011; Mueller et al. 2014). Our findings shed light on the importance of migration on disaster recovery for those households who could not be able to engage in migration. There could be a concern that migrant and return-migrant households are different from non-migrant households based on some unobservable characteristics that we could not capture in our estimation. For example, return migrant households may be more risk-averse (risk-loving), in comparison with migrant and non-migrant households, as they are the first movers. In our analysis, it is difficult to measure these characteristics, as we do not have further information. Considering variables presented in the descriptive statistics the observable characteristics are the same on average, except the age of the migrants.23 The age variable is systematically different in accordance with the eligibility criteria. On average, non-migrants are the youngest and return-migrants are the oldest in our sample.24 Therefore, it is likely that the unobservable characteristics of the household are similar on average. However, it would be interesting to see if there is any significant difference among households and how it systematically impacts the outcome in future research. We only analysed the impact of migration on short-term ex-post community recovery after a natural disaster. Our dataset did not allow us to measure the magnitude of its impact on ex-post community recovery. Furthermore, we could not be able to track the kind of help delivered as well. Therefore, the insurance hypothesis might not be applicable. However, their valuable contributions in the time of crisis should not be underestimated. Future research in this field could address these limitations. Acknowledgements For valuable comments and suggestions, we are grateful to Toyo Ashida, Yasuyuki Sawada, Fumio Ohtake and the participants of presentations at the East Asian Economic Association Conference 2016, SU-ADBI Workshop 2016. We are also grateful to the EPS staffs in Nepal, Joobong Kim and Jelena Rkman for their valuable help. Funding This work was supported by Osaka University and Asian Development Bank Institute (ADBI). Conflict of interest statement. We have no conflicts of interest to disclose. Footnotes 1. Disaster Risk Reduction is the concept and practice of reducing disaster risks through systematic efforts to analyse and manage the causal factors of disasters, including through reduced exposure to hazards, lessened vulnerability of people and property, wise management of land and the environment, and improved preparedness for adverse events (UNISDR definition). 2. In this paper, we use the term ‘remittances’ as ‘migration induced remittances’. 3. For a literature review in this field, see Le De et al. (2013) 4. We could not differentiate between neighbourhood and community in our survey. For simplicity, we use the term community instead of neighbourhood and community from now onwards. 5. Ministry of Labor and Employment (2015). 6. <https://www.eps.go.kr/ph/index.html> accessed 1 Sept 2016. 7. <https://www.eps.go.kr/ph/index.html> accessed 1 Sept 2016. 8. The 2009 test was abandoned, possibly due to the financial crisis. The 2012 test was not conducted as the Korean government selected around 15,678 workers in 2011 which was more than three times of workers selected in 2010. No one knew beforehand that the 2009 and 2012 tests were going to be abandoned. 9. The test is also conducted to check the Korean language ability. As it is meant for returned migrant workers who have worked in the Republic of Korea, it is conducted on a smaller scale in comparison to the general test (PBT) where anyone with 18–39 years of age can participate. 10. <http://www.korea.net/Government/Briefing-Room/Press-Releases/view?articleId=1553> accessed 1 Dec 2016. 11. This group only includes individuals who have passed the paper-based test (PBT) and work in the Republic of Korea. 12. We did a pilot survey in early September and did our main survey between the third week of September to the first week of October. There was a clash between India and Nepal during the last week of September which intensified in October. As the surveying person was Indian, we had to stop the survey as the clash between India and Nepal might influence the survey. 13. Most of the people in this group leave for the Republic of Korea within one month. 14. <https://www.eps.go.kr/ph/index.html> accessed 1 Sept 2016. 15. Speiser (2015). 16. <https://data.worldbank.org/indicator/IT.NET.USER.ZS?locations=KR> accessed 13 Mar 2018. 17. The minimum hourly wage in the Republic of Korea is almost 8 to 10 times higher compared with the average hourly wage in Nepal (authors’ calculation). 18. We only surveyed those who have passed the examination, which constitutes more than half of the applicants. We exclude those who did not pass from our sample, as we could not get their personal information. 19. There are few observations in each district. Therefore, using 62 district dummy variables may reduce the precision of our estimation. Instead, we clubbed the districts based on the intensity of the earthquake and included it in our estimation. 20. Kathmandu valley constitute Kathmandu, Lalitpur and Bhaktapur districts. 21. We asked the following question to measure the participation in community helping and absolute amount of help. ‘Did your household directly help the earthquake victims or participate in community help after the earthquake? If yes approximately how much in Nepalese Rupees?’ (authors’ translation from the Nepali language). 22. Note that the sample size of participation in helping is higher than the ‘amount of help’ in our analysis. In our survey, some of the respondents responded with qualitative answers such as ‘people stayed at our place for some days’, ‘we helped with 20 kilos of rice’, and so forth. It is difficult to quantify these responses into Nepalese rupees correctly. Therefore, we could not include these responses into the amount of help sample. In our sample analysis on the ‘amount of help’, we only include the respondents who responded in Nepalese rupees. However, we have not noticed any systematic difference in response on the basis of the migration characteristics of the household. 23. Living in Kathmandu Valley is significant only for return migrant households compared with our control group. We discuss the possible reason for this difference in Section 5. 24. The age and age squared control variables are not statistically significant in any of our estimations. References Amemiya T. , Boskin M. ( 1974 ) ‘Regression Analysis When the Dependent Variable is Truncated Lognormal, with an Application to the Determinants of the Duration of Welfare Dependency’ , International Economic Review , 15 / 2 : 485 – 96 . Google Scholar CrossRef Search ADS Ashenfelter O. ( 2012 ) ‘Comparing Real Wage Rates’ , American Economic Review , 102 / 2 : 617 – 42 . Google Scholar CrossRef Search ADS Black R. et al. ( 2011 ) ‘Climate Change: Migration as Adaptation’ , Nature , 478 / 7370 : 447 – 9 . Google Scholar CrossRef Search ADS PubMed Brown R. P. , Leeves G. , Prayaga P. ( 2014 ) ‘Sharing Norm Pressures and Community Remittances: Evidence from a Natural Disaster in the Pacific Islands’ , Journal of Development Studies , 50 / 3 : 383 – 98 . Google Scholar CrossRef Search ADS Carroll J. , McCarthy S. , Newman C. ( 2005 ) ‘An Econometric Analysis of Charitable Donations in the Republic of Ireland’ , Economic and Social Review , 36 / 3 : 229 . Cragg J. G. ( 1971 ) ‘Some Statistical Models for Limited Dependent Variables with Application to the Demand for Durable Goods’ , Econometrica , 39 / 5 : 829 – 44 . Google Scholar CrossRef Search ADS Crush J. et al. ( 2012 ) ‘16 Diasporas on the Web: New Networks, New Methodologies’, in Vargas-Silva C. (ed.) Handbook of Research Methods in Migration . Edward Elgar Publishing, UK . Deshingkar P. , Aheeyar M. M. M. ( 2006 ) ‘Remittances in Crisis: Sri Lanka after the Tsunami.’ Humanitarian Policy Group Background Paper. Fagan P. ( 2006 ) Remittances in Crisis: A Haiti Case Study . London : Overseas Development Institute . Gibson J. , McKenzie D. , Stillman S. ( 2011 ) ‘The Impacts of International Migration on Remaining Household Members: Omnibus Results from a Migration Lottery Program’ , Review of Economics and Statistics , 93 / 4 : 1297 – 318 . Google Scholar CrossRef Search ADS Greene W. H. ( 2010 ) Econometric Analysis , 7th edn. Pearson Education Limited, UK . Korea.net ( 2012 ) Re-entry system for committed workers receives positive response from firms and foreign workers <http://www.korea.net/Government/Briefing-Room/Press-Releases/view?articleId=1553> accessed 31 Dec 2016. Le De L. et al. ( 2015 ) ‘Remittances in the Face of Disasters: A Case Study of Rural Samoa’ , Environment, Development and Sustainability , 17 / 3 : 653 – 72 . Google Scholar CrossRef Search ADS Le De L. , Gaillard J. C. , Friesen W. ( 2013 ) ‘Remittances and Disaster: A review’ , International Journal of Disaster Risk Reduction 4 : 34 – 43 . Google Scholar CrossRef Search ADS Linnerooth-Bayer J. , Mechler R. , Hochrainer-Stigler S. ( 2011 ) ‘Insurance Against Losses from Natural Disasters in Developing Countries: Evidence, Gaps and the Way Forward’ , Journal of Integrated Disaster Risk Management , 1 / 1 : 59 – 81 . Google Scholar CrossRef Search ADS Manandhar B. ( 2016 ) ‘Remittance and Earthquake Preparedness’ , International Journal of Disaster Risk Reduction , 15 : 52 – 60 . Google Scholar CrossRef Search ADS McKenzie D. ( 2015 ) ‘Learning about Migration through Experiments’, in Dustman C. (ed.) Migration: Economic Change, Social Challenge , pp. 68 – 82 . Oxford University Press , USA. Google Scholar CrossRef Search ADS McKenzie D. , Yang D. ( 2012 ) ‘Experimental Approaches in Migration Studies’, in Vargas-Silva C. (ed.) Handbook of Research Methods in Migration . Edward Elgar Publishing, UK . McKenzie D. , Stillman S. , Gibson J. ( 2010 ) ‘How Important is Selection? Experimental vs. Non-Experimental Measures of the Income Gains from Migration’ , Journal of the European Economic Association , 8 / 4 : 913 – 45 . Google Scholar CrossRef Search ADS Ministry of Labor and Employment ( 2015 ) Labor Migration for Employment a Status Report for Nepal: 2014/2015 . Government of Nepal, Nepal . Mohapatra S. , Joseph G. , Ratha D. ( 2012 ) ‘Remittances and Natural Disasters: Ex-post Response and Contribution to Ex-ante Preparedness’ , Environment, Development and Sustainability , 14 / 3 : 365 – 87 . Google Scholar CrossRef Search ADS Mueller V. , Gray C. , Kosec K. ( 2014 ) ‘Heat Stress Increases Long-Term Human Migration in Rural Pakistan’ , Nature Climate Change , 4 / 3 : 182 – 5 . Google Scholar CrossRef Search ADS PubMed National Planning Commission ( 2015 ) Post Disaster Needs Assessment, Vol. A. Key findings, Government of Nepal, Nepal. Newman C. , Henchion M. , Matthews A. ( 2003 ) ‘A Double-Hurdle Model of Irish Household Expenditure on Prepared Meals’ , Applied Economics , 35 / 9 : 1053 – 61 . Google Scholar CrossRef Search ADS Ratha D. , Sirkeci I. ( 2010 ) ‘Remittances and the Global Financial Crisis’ , Migration Letters , 7 / 2 : 125 – 31 . Speiser M. ( 2015 ) The 10 Countries with World’s Fastest Internet Speeds <https://www.weforum.org/agenda/2015/05/the-10-countries-with-the-worlds-fastest-internet-speeds/> accessed 1 Sept 2016. Wooldridge J. M. ( 2009 ) Hurdle and “selection” models, Michigan State UniversityBGSE/IZA Course in Micro-econometrics. Wu T. ( 2006 ) ‘The Role of Remittances in Crisis. An Aceh Research Study’, Overseas Development Institute Background working paper, London, UK. Yang D , Choi H. ( 2007 ) ‘Are Remittances Insurance? Evidence from Rainfall Shocks in the Philippines’ , The World Bank Economic Review , 21 / 2 : 219 – 48 . Google Scholar CrossRef Search ADS Zhang F. et al. ( 2008 ) ‘Modeling Fresh Organic Produce Consumption with Scanner Data: A Generalized Double Hurdle Model Approach’ , Agribusiness , 24 / 4 : 510 – 22 . Google Scholar CrossRef Search ADS Appendix Table A1: Full Destruction of House. (1) (2) (3) Return migrant household –0.365 –0.420 –0.035 (0.250) (0.561) (2.131) Migrant household –0.332 –0.564 –2.886 (0.264) (0.379) (11.36) Affected intensity * Migrant household 0.625 (2.528) Affected intensity * Return migrant household –0.138 (0.543) Other controls No Yes Yes Chi2 3.960 63.95 39.58 Log likelihood –129.319 –71.025 –67.672 R² 0.014 0.415 0.442 N 287 260 260 (1) (2) (3) Return migrant household –0.365 –0.420 –0.035 (0.250) (0.561) (2.131) Migrant household –0.332 –0.564 –2.886 (0.264) (0.379) (11.36) Affected intensity * Migrant household 0.625 (2.528) Affected intensity * Return migrant household –0.138 (0.543) Other controls No Yes Yes Chi2 3.960 63.95 39.58 Log likelihood –129.319 –71.025 –67.672 R² 0.014 0.415 0.442 N 287 260 260 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table A1: Full Destruction of House. (1) (2) (3) Return migrant household –0.365 –0.420 –0.035 (0.250) (0.561) (2.131) Migrant household –0.332 –0.564 –2.886 (0.264) (0.379) (11.36) Affected intensity * Migrant household 0.625 (2.528) Affected intensity * Return migrant household –0.138 (0.543) Other controls No Yes Yes Chi2 3.960 63.95 39.58 Log likelihood –129.319 –71.025 –67.672 R² 0.014 0.415 0.442 N 287 260 260 (1) (2) (3) Return migrant household –0.365 –0.420 –0.035 (0.250) (0.561) (2.131) Migrant household –0.332 –0.564 –2.886 (0.264) (0.379) (11.36) Affected intensity * Migrant household 0.625 (2.528) Affected intensity * Return migrant household –0.138 (0.543) Other controls No Yes Yes Chi2 3.960 63.95 39.58 Log likelihood –129.319 –71.025 –67.672 R² 0.014 0.415 0.442 N 287 260 260 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table A2: No destruction of house. (1) (2) (3) Return migrant household 0.186 0.149 0.014 Migrant household (0.222) (0.328) (0.592) –0.256 –0.295 0.144 (0.226) (0.258) (0.407) Affected intensity * Migrant household –0.273 (0.246) Affected intensity * Return migrant household 0.064 (0.202) Other controls No Yes Yes Chi2 1.635 53.25 45.11 Log likelihood –196.845 –140.224 –137.924 R² 0.010 0.221 0.234 N 287 260 260 (1) (2) (3) Return migrant household 0.186 0.149 0.014 Migrant household (0.222) (0.328) (0.592) –0.256 –0.295 0.144 (0.226) (0.258) (0.407) Affected intensity * Migrant household –0.273 (0.246) Affected intensity * Return migrant household 0.064 (0.202) Other controls No Yes Yes Chi2 1.635 53.25 45.11 Log likelihood –196.845 –140.224 –137.924 R² 0.010 0.221 0.234 N 287 260 260 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table A2: No destruction of house. (1) (2) (3) Return migrant household 0.186 0.149 0.014 Migrant household (0.222) (0.328) (0.592) –0.256 –0.295 0.144 (0.226) (0.258) (0.407) Affected intensity * Migrant household –0.273 (0.246) Affected intensity * Return migrant household 0.064 (0.202) Other controls No Yes Yes Chi2 1.635 53.25 45.11 Log likelihood –196.845 –140.224 –137.924 R² 0.010 0.221 0.234 N 287 260 260 (1) (2) (3) Return migrant household 0.186 0.149 0.014 Migrant household (0.222) (0.328) (0.592) –0.256 –0.295 0.144 (0.226) (0.258) (0.407) Affected intensity * Migrant household –0.273 (0.246) Affected intensity * Return migrant household 0.064 (0.202) Other controls No Yes Yes Chi2 1.635 53.25 45.11 Log likelihood –196.845 –140.224 –137.924 R² 0.010 0.221 0.234 N 287 260 260 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses *p < 0.1, **p < 0.05, ***p < 0.01. Standard errors are clustered at district level. Table A3: Relative participation and absolute donation. (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.578** 3.458** 0.569** 3.431** 0.511* 3.777** (0.281) (1.468) (0.275) (1.536) (0.310) (1.912) Migrant household 0.501*** 3.555*** 0.572*** 3.745*** 0.617*** 4.124*** (0.180) (1.031) (0.184) (1.010) (0.197) (1.134) Age No No No No Yes Yes Other Controls No No Yes Yes Yes Yes Sigma 6.929*** 6.774*** 6.730*** (0.466) (0.461) (0.450) Chi2 8.261 12.42 22.77 27.58 31.52 37.66 R² 0.0316 0.0117 0.0474 0.0175 0.0745 0.0239 Log likelihood –170.49 –575.771 –155.29 –535.50 –147.31 –518.05 Return migrant household = Migrant household 0.7535 0.9408 0.9922 0.8263 0.6894 0.8179 N 267 245 248 228 242 222 (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.578** 3.458** 0.569** 3.431** 0.511* 3.777** (0.281) (1.468) (0.275) (1.536) (0.310) (1.912) Migrant household 0.501*** 3.555*** 0.572*** 3.745*** 0.617*** 4.124*** (0.180) (1.031) (0.184) (1.010) (0.197) (1.134) Age No No No No Yes Yes Other Controls No No Yes Yes Yes Yes Sigma 6.929*** 6.774*** 6.730*** (0.466) (0.461) (0.450) Chi2 8.261 12.42 22.77 27.58 31.52 37.66 R² 0.0316 0.0117 0.0474 0.0175 0.0745 0.0239 Log likelihood –170.49 –575.771 –155.29 –535.50 –147.31 –518.05 Return migrant household = Migrant household 0.7535 0.9408 0.9922 0.8263 0.6894 0.8179 N 267 245 248 228 242 222 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. Table A3: Relative participation and absolute donation. (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.578** 3.458** 0.569** 3.431** 0.511* 3.777** (0.281) (1.468) (0.275) (1.536) (0.310) (1.912) Migrant household 0.501*** 3.555*** 0.572*** 3.745*** 0.617*** 4.124*** (0.180) (1.031) (0.184) (1.010) (0.197) (1.134) Age No No No No Yes Yes Other Controls No No Yes Yes Yes Yes Sigma 6.929*** 6.774*** 6.730*** (0.466) (0.461) (0.450) Chi2 8.261 12.42 22.77 27.58 31.52 37.66 R² 0.0316 0.0117 0.0474 0.0175 0.0745 0.0239 Log likelihood –170.49 –575.771 –155.29 –535.50 –147.31 –518.05 Return migrant household = Migrant household 0.7535 0.9408 0.9922 0.8263 0.6894 0.8179 N 267 245 248 228 242 222 (1) (2) (3) (4) (5) (6) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Help (Yes) Log (Amount + 1) Return migrant household 0.578** 3.458** 0.569** 3.431** 0.511* 3.777** (0.281) (1.468) (0.275) (1.536) (0.310) (1.912) Migrant household 0.501*** 3.555*** 0.572*** 3.745*** 0.617*** 4.124*** (0.180) (1.031) (0.184) (1.010) (0.197) (1.134) Age No No No No Yes Yes Other Controls No No Yes Yes Yes Yes Sigma 6.929*** 6.774*** 6.730*** (0.466) (0.461) (0.450) Chi2 8.261 12.42 22.77 27.58 31.52 37.66 R² 0.0316 0.0117 0.0474 0.0175 0.0745 0.0239 Log likelihood –170.49 –575.771 –155.29 –535.50 –147.31 –518.05 Return migrant household = Migrant household 0.7535 0.9408 0.9922 0.8263 0.6894 0.8179 N 267 245 248 228 242 222 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. Table A4: Behaviour within households. (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.595 4.011* (0.486) (2.305) Migrant household 0.430 3.030* (0.269) (1.639) Affected intensity −0.006 0.196 (0.065) (0.423) Affected intensity * Migrant household 0.105 0.555 (0.102) (0.551) Affected intensity * Return migrant household −0.043 −0.131 (0.186) (1.045) Other controls Yes Yes Sigma 6.711*** (0.447) Chi2 33.19 44.00 R² 0.078 0.025 Log likelihood −146.756 −517.575 N 242 222 (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.595 4.011* (0.486) (2.305) Migrant household 0.430 3.030* (0.269) (1.639) Affected intensity −0.006 0.196 (0.065) (0.423) Affected intensity * Migrant household 0.105 0.555 (0.102) (0.551) Affected intensity * Return migrant household −0.043 −0.131 (0.186) (1.045) Other controls Yes Yes Sigma 6.711*** (0.447) Chi2 33.19 44.00 R² 0.078 0.025 Log likelihood −146.756 −517.575 N 242 222 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. Table A4: Behaviour within households. (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.595 4.011* (0.486) (2.305) Migrant household 0.430 3.030* (0.269) (1.639) Affected intensity −0.006 0.196 (0.065) (0.423) Affected intensity * Migrant household 0.105 0.555 (0.102) (0.551) Affected intensity * Return migrant household −0.043 −0.131 (0.186) (1.045) Other controls Yes Yes Sigma 6.711*** (0.447) Chi2 33.19 44.00 R² 0.078 0.025 Log likelihood −146.756 −517.575 N 242 222 (1) (2) Help (Yes) Log (Amount + 1) Return migrant household 0.595 4.011* (0.486) (2.305) Migrant household 0.430 3.030* (0.269) (1.639) Affected intensity −0.006 0.196 (0.065) (0.423) Affected intensity * Migrant household 0.105 0.555 (0.102) (0.551) Affected intensity * Return migrant household −0.043 −0.131 (0.186) (1.045) Other controls Yes Yes Sigma 6.711*** (0.447) Chi2 33.19 44.00 R² 0.078 0.025 Log likelihood −146.756 −517.575 N 242 222 Other controls include age, age squared, affected intensity, number of family members, family members abroad, education in years and household living in Kathmandu valley. Standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01. Standard errors are clustered at district level. © The Author(s) 2018. Published by Oxford University Press. All rights reserved. For permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Migration StudiesOxford University Press

Published: Mar 27, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off