TY - JOUR
AU - Goldbach,, Carina
AB - Abstract Projections of climatic and environmental changes have generated a growing effort to assess their implications for human migration. Because migration is always a multicausal phenomenon, this study aims to disentangle the impact of environmental factors from other migration-inducing factors to shed some light on the complex relationship between the environment and migration. Thus, we conducted quantitative microlevel studies in low-lying communities in two high-mobility countries—Ghana and Indonesia—that are particularly exposed to coastal hazards like erosion, land subsidence, storm surges and an increasing sea level, and are prone to flooding on a regular basis. Different measures of environmental threats were collected, ranging from individual perceptions over the household’s distance to the coast to expert opinions. We analyzed the relationships using logistic regressions and controlled for contextual factors on multiple levels. No statistically significant direct impacts of slow-onset environmental events on migration decisions could be detected. Perceptions of storms, a clearly sudden-onset event, however, were found to be significantly linked to out-migration decisions in Ghana. These findings support the hypothesis that environmental factors are generally not a primary cause of migration, and their effects are rather context specific—especially for slow-onset changes. (JEL codes: R23, O15, Q54.) 1. Introduction Scientists expect that increases in global temperatures will lead to sea-level rise and greater weather variability which could result in droughts, increased rainfalls, and intensified extreme coastal events (IPCC 2007). These projections have generated a growing effort to assess the impact of climate change, including its implications for human migration. However, ‘the science of climate change is complex enough, even before considering its impact on societies’ (Brown 2008, p. 8). Consequently, global estimates of future migration caused by climate and environmental change vary extremely. One of the most famous estimations resulted in around 200 million ‘environmental refugees’ until the year 2050, 162 million of them due to sea-level rise in Asia and Africa (Myers 2002). Other scholars have produced even higher estimates. The Christian aid report, for example, predicted that 1 billion people will be forced to leave their home in the same period (Christian Aid 2007). These estimates, however, have been heavily criticized by other scholars who reject this deterministic view of climate change being the primary and direct cause of out-migration. Instead, migration is recognized as a complex phenomenon which is ‘always the result of a multi-causal relationship between environmental, political, economic, social, and cultural dimensions’ (Piguet 2010, p. 517). As these dimensions are closely intertwined, it makes not much sense to consider any of the migration determining factors in isolation. Despite this high level of interest, quantitative micro-level studies accounting for this multi-causal relationship are still underrepresented (Gray and Bilsborrow 2013). Therefore, this study aims to contribute to the existing literature by using a quantitative multilevel model where several migration-determining factors are simultaneously considered to isolate the net effect of the environment. Since low-lying regions in Asia and Africa are considered especially vulnerable, data have been collected in two coastal regions in Indonesia and Ghana which experience floods and erosion on a regular basis. Economic, social, and cultural conditions vary considerably between both regions which helps to understand in how far migratory responses to environmental threats are context specific. Empirically, this study improves on existing works in several ways. First of all, a micro-level survey was conducted in two different developing countries, which enables the analysis of the relationship between the environment and out-migration in a quantitative and comparative way. Second, very different environmental variables ranging from perceptions over expert opinions to more objective measures have been collected. While most quantitative studies have so far focused on sudden changes in the environment, this study especially looks at rather gradual changes and slow-onset events. Third, both people who stayed and people who moved were included which is important for a better understanding of migration decisions. Additionally, instead of relying on information from a proxy respondent who answers on behalf of the migrant, migrants were personally interviewed. This procedure enables the inclusion of individual perceptions and preferences. While this study is able to explain migration decisions well with the help of contextual factors on community, household, and individual level, no direct link between the main environmental factors—floods and erosion—and out-migration could be detected. However, there is evidence of a positive effect of storm perceptions on individual migration decisions in Ghana which lead to the hypothesis that long-term, gradual changes do not tend to increase the likelihood of out-migration directly, while sudden-onset events might do. Additionally, there is weak evidence that the effect of environmental degradation on migration depends on other migration-facilitating factors such as networks in Indonesia and the number of children in Ghana. 2. Migration and the Environment Despite substantial and ongoing progress in migration theories, the environment was not incorporated explicitly for most of the time (Black et al. 2011).1 The idea that human migration patterns may respond to the environment, however, is not new. A large body of literature indicates that prehistoric settlements were strongly linked to changes in the climate (McLeman and Smit 2006). Examples of climatic influences on human settlement patterns can also be found in the more recent past. The Great Plains in Oklahoma, as for example, experienced a high level of in-migration during the 1920s which coincided with a period of favorable agricultural conditions. During the 1930s, however, droughts and dust storms made hundred thousands of Americans leave this ‘Dust Bowl’ (McLeman 2006). Later on, Hurricane Mitch struck Central America leading to wide-scale migration movements in 1998 and Hurricane Katrina resulted in around 2 million displaced people in 2005 (McLeman and Hunter 2010). Yet only 30 years ago, the term ‘environmental refugee’ came into regular use (El-Hinnawi 1985) and in the early 1990s the international community slowly began to recognize the potential implications of environmental change on human migration. The Intergovernmental Panel on Climate Change (IPCC) warned that the greatest single impact of climate change could be on human migration (IPCC 1990). From that moment the subject became increasingly polarized and Suhrke (1993, 1994) identified two main groups: ‘Maximalists’ claimed that the environment is the primary and direct cause of forced migration, whereas ‘minimalists’ suggested that the relationship is more complex and, thus, the environment only a contextual factor. In the course of this debate, Myers (1993, 1997, 2002) became the most prominent and widely cited author within the maximalist school. In his various studies he estimated the current number of environmental refugees at 25 million and predicted the number to rise up to 200 million until the year 2050.2Black (2001), supported by Castles (2002), disagreed with such scenarios and questioned the assumption of the environment being the primary cause of migration. It would not only be problematic to assume that people move directly because of the environment—as this assumption neglects the essential role of humans in dealing with those changes—but also that everyone who leaves a region experiencing environmental change is actually leaving due to that environment (Black 2001). Today, the field has generally moved beyond these polarized debates and most social scientists emphasize multilevel migration drivers and the importance of contexts (Morrissey 2009). Yet, environment–migration theories still do not lead to a consistent prediction in which direction environmental changes will impact on out-migration and how large this impact might be relative to other migration-inducing factors. Adverse environmental factors are mostly seen as stressors, push factors, or local disamenities which decrease individuals’ happiness and incomes and which encourage them to move to places with better environments (Hunter 2005; Gray 2009). Other theories emphasize that migration decisions are rarely an individual choice but rather made by households (Stark 1991). Thus, it is assumed that when environmental changes lead to the loss of assets or income, households may decide to send out migrants to receive remittances and replace lost assets. While this would be a direct response to the effects of environmental events, sending out migrants may also serve as an ex-ante strategy whenever environmental effects are expected in the future. Since migrants’ earnings are mostly uncorrelated with environmental threats in the home community, the out-migration of one or more household members can act as an insurance against future environmental damage by diversifying the household’s income sources (Stark and Bloom 1985). In case of severe events, however, migration of whole households might be necessary and unavoidable to seek shelter in a different area (Gray and Bilsborrow 2013). Nevertheless, there is agreement that migration is generally costly and thus cannot be considered a default response to environmental stressors. As long as the costs of dealing with environmental events are lower than the costs of migrating, individuals are expected to stay in their home community—coping with and adapting to the adverse environmental changes. Similarly, environmental shocks and degradation might make out-migration even less likely because it requires forms of capital, whereas affected populations often experience a decrease in the very capital required for a migratory move (Foresight 2011). Since environment–migration theories do not lead to clear predictions regarding the direction and relative importance of environmental factors on out-migration decisions, it was called for more detailed case studies and analytical attempts to further assess the impact of environmental degradation. Consequently, in the past two decades the literature on environmental change and migration grew extensively; most of the studies suggesting that the environment ‘can’ lead to migration (Henry et al. 2004; Laczko and Aghazarm 2009; Kniveton et al. 2011). Highly vulnerable households of dry areas in Ethiopia were found to send migrants to urban areas during times of famine (Ezra and Kiros 2000). Results from another study indicated that in dry periods male migration is increasing, while female migration decreases (Gray and Müller 2012). In drought-prone areas of Burkina Faso individuals engaged in rural–rural migration to areas with better agricultural outcomes and reduced their international migration (Henry et al. 2003). Gray and Bilsborrow (2013), on the other hand, found that dry periods in Ecuador decreased internal moves but increased international migration. Halliday (2006) found increased international mobility with loss of harvest for wealthy households but overall decreased migration following an earthquake in El Salvador. These and other papers illustrate how diverse findings regarding the environment–migration nexus can be. Despite the already extensive literature, there is still room for further research addressing the shortcomings of existing work. First, existing research mostly focuses on one specific environmental threat and cannot make statements about differences across different environmental factors in the same region. The most investigated environmental threat is drought or rainfall, while other environmental events and their impacts get rather neglected (Jónsson 2010). This study adds by focusing on different coastal events. Second, the majority of micro-level field studies have used qualitative methods. Although these studies offer valuable insights into people’s migratory responses toward environmental change, they cannot successfully isolate environmental aspects from other migration-inducing factors (Piguet 2010). Therefore, quantitative micro-level research methods are used which are still largely underrepresented (see Moriniere 2009).3 Furthermore, since sudden and extreme environmental events have received much more attention (Gray 2011), this study will especially focus on rather gradual and long-term changes. While both types of environmental stressors have the potential to impact on out-migration decisions, long-term changes are often not regarded as severe enough to cause the relocation of whole households. Households experiencing long-term changes are rather expected to send out individual household members to diversify and supplement their income sources and to adapt to those changes over time, since they are generally easier to anticipate than sudden-onset events (Koubi et al. 2016). Thus, long-term gradual environmental changes are expected to have an overall smaller impact on individual out-migration decisions than sudden-onset events. This study tests this hypothesis and adds to the literature by interviewing both migrants and non-migrants in two coastal regions. 3. Methods As mentioned above, quantitative research methods were used to isolate environmental degradation from other migration-inducing factors. To do so, micro-level4 data for both migrants and non-migrants who originally came from the same region were required. This data requirement ensures that the context is similar for every respondent. Thus, a retrospective survey was conducted in two developing countries. 3.1 Research sites Even though many countries in the world face environmental problems, low-lying coastal communities are considered, especially, vulnerable because they are particularly exposed to environmental hazards like coastal erosion, tidal waves, storm surges and an increasing sea level, and thus at the risk of experiencing floods. However, coastal regions are not only particularly exposed to environmental hazards; they are also associated with a large and rapidly growing human population. Currently, low-lying coastal regions are home to 10% of the world population, while nearly half of the world population lives within 150 km from the coast (Foresight 2011). Two coastal regions, characterized by ongoing erosion and regularly occurring floods, and a longstanding tradition in regional migration, were selected: Keta district in southeastern Ghana and Semarang in Indonesia. A comparative case study approach was chosen, since several studies have shown that this approach is especially useful to pinpoint the impact of site-specific factors on the respective outcomes. 3.1.1 Keta, Ghana Keta municipality is located in rural southeastern Ghana, and has a population of about 100,000 inhabitants. It was chosen, since it has been the site of acute coastal erosion since about 1907 (Akyeampong 2001). By independence, more than half of Keta was robbed by the sea. This ongoing erosion process is caused by increased storm intensity, soft geology, and low-lying topography but is also influenced by anthropogenic activities like illegal sand mining or the building of the Akosombo dam on the Volta River in 1964 which decreased the sediment flow to the coast (Boateng 2012). In the end of the twentieth century, annual recession rates ranged from 2 m/year in the northeast to 8 m/year in the southwest (Nairn 2001). Land became extremely scarce and the distance between the sea and the Keta Lagoon is often not exceeding 3 miles. At various sections, especially affected by the environmental change, the lagoon and the sea are within 15–30 m of each other, only separated by a thin tongue of sand. These erosion processes and the concomitant retreating shoreline have direct effects on some coastal households which have to deal with tidal inundation and the threat of losing their house. While erosion and tidal floods are generally seen as rather slow and foreseeable processes, storm surges hit the coastline with destructive and unpredictable power. 3.1.2 Semarang, Indonesia Semarang, the provincial capital of Central Java, has around 1.6 million inhabitants5 and is, thus, essentially bigger than Keta. It is a coastal urban area at the Northern coast of Java, located between Jakarta and Surabaya, the two major cities of Indonesia. During colonial times, Semarang has emerged as a successful and important port, and is still seen as an important regional center and port today (Knaap 2015). Very similar to Keta, coastal communities of Semarang are exposed to massive coastal changes which threaten the development of the area. Substantial land subsidence due to excessive groundwater extractions and extensive construction works causes coastal communities to sink with a rate of 2 up to 10 cm per year. This subsidence in combination with high tides is often resulting in tidal flood inundation which poses a major threat to infrastructure and settlements of urban coastal communities (Marfai and King 2008). Not only is the majority of industry located in these communities but also has a large part of the population of Semarang settled there. Consequently, many people have been experiencing the threat of tidal inundation with different depth of seawater flooding (Marfai et al. 2008). Even though communities are sinking at an alarming rate, subsidence and erosion are rather slow-onset changes, and inundation is regularly experienced by households at risk. Therefore, the great majority of affected households responds to the threats by elevating their houses and raising the floors every 5–10 years if they can afford it. So far, there is no prospect of an end of these environmentally adverse conditions (Harwitasari and van Ast 2009). 3.2 Sampling The two coastal regions—Keta in Ghana and Semarang in Indonesia—were purposefully chosen because of their changing coast. In each of the regions, several communities were also purposefully selected due to their exposure to coastal changes. This non-random selection of communities ensures that the sample contains both affected and non-affected communities and, thus, that there is sufficient variation in the variables of interest.6 Once each community was chosen, households got randomly selected to avoid sampling bias. In Semarang, high-resolution satellite pictures and randomly generated GPS points were used to select households. In Keta, which is substantially smaller and less densely populated, households got carefully chosen at regular intervals. Once the households were selected, a household survey was the main method to gather data. After household characteristics were obtained by interviewing the household head, the enumerator randomly selected and interviewed a household member above the age of 18 years. Since this study does not want to focus solely on migration intentions or the individual’s willingness to migrate but on actual migration, one randomly chosen migrant of the household was additionally interviewed—in case there was any. This sampling strategy has the disadvantage that by definition it does not include households which have moved as a whole. If households which move as a whole are systematically different from households which stay and only send out a migrant, this sampling strategy is likely to produce a sampling bias. However, this study focuses especially on rather gradual and long-term changes which are often not regarded as severe enough to induce the inevitable relocation of whole households. Only very few houses become completely uninhabitable due to coastal changes in Keta and Semarang. Households experiencing long-term changes are therefore expected to be left by individual household members (Koubi et al. 2016). Those migrants, however, are part of this study’s sample. Furthermore, the migrant sample is found to be quite diverse: only 42% of sampled migrants from Keta and 35.4% of migrants from Semarang had actually moved unaccompanied. The rest moved mainly together with their spouses and children but also with parents and siblings—leaving behind only parts of the generally quite large and intergenerational households which have been especially found in the Ghanaian study area. Even though the error caused by the omission of whole households is thus expected to be rather small, one should keep the sampling strategy in mind when interpreting the results. The great advantage of this sampling approach, however, is that environmental migration can be investigated even when high-quality census data are not available. Therefore, this generally applicable approach is very helpful for studying migration decisions in many different contexts. 4. Data and Analysis In line with the Foresight Report (2011), migration is understood as a movement from one place to another for a period of 3 months or more. This study does not focus on international migration only but also considers everyone a migrant who moves within her country—to another region, district, or community. Not only did first informal interviews reveal that only very few people from Semarang actually leave Indonesia, studies have also shown that the majority of environmental migrants move internally (Obokata et al. 2014). While it is common in the literature to get information about the migrant from a proxy respondent like the household head, migrants in this study have been contacted and interviewed by phone. Thereby, it was able to avoid proxy errors and to include very individual perception, preferences, and opinion questions. Since no panel data were available, migrants were asked to provide information about certain characteristics like age, education, and perceptions for the time when they left to avoid reverse causality problems. This also enables the comparison of non-migrants and migrants before their out-migration. Additionally, migrants were only included when they had left within the past 10 years to reduce recall bias. Ultimately, in Indonesia, 240 households got interviewed out of which 105 households (43.75%) listed at least one migrant in the past 10 years. In Ghana, 190 households participated in the survey out of which 101 (53.16%) had at least one migrant. As Table 1 shows, the great majority of migrants in the sample moved internally: only 4% of Indonesian migrants and only 7% of Ghanaian migrants actually left the country. In Indonesia, nearly a third of movements happened within Semarang, while in Ghana only 7% of migrants moved within Keta. Table 1. Destination of migrants . % of migrants, Indonesia . % of migrants, Ghana . Within Semarang/Keta 30.68 6.98 Within region7 38.95 20.93 Within country to capital 26.32 65.12 11.46 36.78 International 4.05 6.98 . % of migrants, Indonesia . % of migrants, Ghana . Within Semarang/Keta 30.68 6.98 Within region7 38.95 20.93 Within country to capital 26.32 65.12 11.46 36.78 International 4.05 6.98 Table 1. Destination of migrants . % of migrants, Indonesia . % of migrants, Ghana . Within Semarang/Keta 30.68 6.98 Within region7 38.95 20.93 Within country to capital 26.32 65.12 11.46 36.78 International 4.05 6.98 . % of migrants, Indonesia . % of migrants, Ghana . Within Semarang/Keta 30.68 6.98 Within region7 38.95 20.93 Within country to capital 26.32 65.12 11.46 36.78 International 4.05 6.98 This finding is not very surprising, since Semarang is essentially bigger and economically stronger than Keta. That might also be the reason why most Indonesian migrants stay within the region (Central Java), whereas Ghanaian migrants tend to leave the region (Volta region), mostly to move to the Greater Accra region. Flow maps (see Figures 1 and 2) further illustrate the range of destinations of migrants. Figure 1. Open in new tabDownload slide Flow map, migrants from Semarang, Indonesia. Source: Author’s illustration. Figure 1. Open in new tabDownload slide Flow map, migrants from Semarang, Indonesia. Source: Author’s illustration. Figure 2. Open in new tabDownload slide Flow map, migrants from Keta, Ghana. Source: Author’s illustration. Figure 2. Open in new tabDownload slide Flow map, migrants from Keta, Ghana. Source: Author’s illustration. The other variable of main interest, environmental threats, was measured in three different ways. First, in line with other papers about the environment–migration link, this research uses perceptions because they are considered central for how people respond to environmental threats and because they can differ substantially between individuals from the same household (Mortreux and Barnett 2009, Koubi et al. 2016). Respondents were asked about their perceptions of flood and erosion in both study areas. Since subsidence poses an additional environmental threat in Indonesia, it was included in the Indonesian questionnaire. The same applies to storms for the Ghanaian case which is the only clearly sudden-onset environmental event. Respondents were asked to indicate on a scale from 1 to 10 how much they have been affected by those environmental threats within the past 5 years. Nevertheless, individuals’ perceptions might be biased or incomplete, and it is frequently argued that studies should focus more on objective measures (Laczko and Aghazarm 2009). Therefore, the household’s distance to the coast is used as a proxy for its exposure to coastal changes. On top of these measures on individual and household level, the sampled communities have been categorized according to their recent exposure to floods (Indonesia) and shoreline erosion (Ghana). This classification of communities into different hazard categories is based on the knowledge of experts. A first look at the environmental variables shows that perceptions are in line and highly correlated with the more objective measures (see Table 2). As expected, individuals from households living further away from the coast perceived to be less affected by erosion, floods, and subsidence. Additionally, those who perceive to be highly affected by environmental threats are significantly more likely to live in high-hazard communities. Table 2. Correlation of environmental variables Indonesia/ Ghana . Individual level: perceptions . Household level . Community level . Flood . Erosion . Subsidence/storm . Distance . Hazard . Flood 1.00 Erosion 0.1764***/ 0.7070*** 1.00 Subsidence/Storm 0.4323***/ 0.5917*** 0.3466***/ 0.4890*** 1.00 Distance −0.3909***/ −0.1578*** −0.0428***/ −0.1935*** −0.3909***/ −0.0724 1.00 Hazard 0.0278/ 0.3094*** 0.1172**/ 0.4089*** 0.2426***/ 0.1972*** 0.2595***/ 0.4319*** 1.00 Indonesia/ Ghana . Individual level: perceptions . Household level . Community level . Flood . Erosion . Subsidence/storm . Distance . Hazard . Flood 1.00 Erosion 0.1764***/ 0.7070*** 1.00 Subsidence/Storm 0.4323***/ 0.5917*** 0.3466***/ 0.4890*** 1.00 Distance −0.3909***/ −0.1578*** −0.0428***/ −0.1935*** −0.3909***/ −0.0724 1.00 Hazard 0.0278/ 0.3094*** 0.1172**/ 0.4089*** 0.2426***/ 0.1972*** 0.2595***/ 0.4319*** 1.00 Note: *p < 0.1, **p < 0.05, ***p < 0.01. Table 2. Correlation of environmental variables Indonesia/ Ghana . Individual level: perceptions . Household level . Community level . Flood . Erosion . Subsidence/storm . Distance . Hazard . Flood 1.00 Erosion 0.1764***/ 0.7070*** 1.00 Subsidence/Storm 0.4323***/ 0.5917*** 0.3466***/ 0.4890*** 1.00 Distance −0.3909***/ −0.1578*** −0.0428***/ −0.1935*** −0.3909***/ −0.0724 1.00 Hazard 0.0278/ 0.3094*** 0.1172**/ 0.4089*** 0.2426***/ 0.1972*** 0.2595***/ 0.4319*** 1.00 Indonesia/ Ghana . Individual level: perceptions . Household level . Community level . Flood . Erosion . Subsidence/storm . Distance . Hazard . Flood 1.00 Erosion 0.1764***/ 0.7070*** 1.00 Subsidence/Storm 0.4323***/ 0.5917*** 0.3466***/ 0.4890*** 1.00 Distance −0.3909***/ −0.1578*** −0.0428***/ −0.1935*** −0.3909***/ −0.0724 1.00 Hazard 0.0278/ 0.3094*** 0.1172**/ 0.4089*** 0.2426***/ 0.1972*** 0.2595***/ 0.4319*** 1.00 Note: *p < 0.1, **p < 0.05, ***p < 0.01. To get a first impression of the general reasons for moving, at the beginning of the interview and thus before mentioning the focus on environmental factors, migrants were asked why they had moved away from their community. Migrants could openly name up to three main reasons, which have been sorted into five different categories. Ultimately, 9% of the Indonesian and 3% of the Ghanaian migrants mentioned floods or other environmental threats as a reason (see Figure 3). Figure 3. Open in new tabDownload slide Reasons for migration, multiple answers per migrant. Source: Author’s illustration. Figure 3. Open in new tabDownload slide Reasons for migration, multiple answers per migrant. Source: Author’s illustration. Even though self-reported reasons give a first impression of the relevance of the environment in individual migration decisions, Van der Geest (2011, p. 3) acknowledges that, ‘the underlying causes of migration […] will not be mentioned by respondents who are asked about their personal motivation to migrate’. Often people use rather standardized answers in explaining migration (Jónsson 2010) or reinterpret the reality after their migration experience (Henry 2006). Therefore, additional quantitative regression analyses are used to answer this question. Since the dependent variable is a dummy reflecting the individual’s decision to migrate or to stay, binary logistic regressions were used to analyze the data. Thus, the model for testing the relationship between environmental threats and the decision to migrate is log  (πi1−πi)=α+β environmenti+γXi+ ui,(1) where πi is the probability of out-migration of individual i. The factor ‘environment’ corresponds to one of the variables which measure the exposure of individual i to environmental change. X stands for a set of control variables which represent important factors mentioned in economic migration theories. U stands for the residuals. The set of control variables includes variables on community, household, and individual level, since it is assumed that migration responses are the result of a complex combination of multiple factors that shape the decision of individuals. The selection of these variables was guided by previous studies and common theories regarding migration decisions. Thus, at the community level, control variables include the community’s population density as well as its percentage of employed inhabitants. In addition, the percentage of the population without an own toilet is included as a wealth proxy. At the household level, control variables include a dummy for female-headed households and household size which are expected to impact positively on the propensity to migrate, as well as the number of children in the household and the ownership status, which are expected to be negatively correlated with migration decisions.8 Another important variable for testing the New Economics of Labor Migration theory is the relative household income, since Stark and Bloom (1985) emphasize that households engage in income comparisons and may migrate due to their relative deprivation within the community. Households with low relative incomes can be expected to have a stronger incentive to migrate or to send out a migrant. To acknowledge the network theory, a control variable indicating an individual’s network was added. According to the theory, most common individual characteristics like sex, age, marital and employment status, education, and previous migration experience have been added. Younger, single males who have a good education but do not have a job at the current place and who have already migrated once or more before are expected to have a higher likelihood to migrate. Since it is assumed that individual preferences matter, individual’s risk aversion and impatience get included as well. Both types of preferences are expected to be negatively correlated with migration decisions which bear uncertainties and risks and can be seen as an investment which just brings benefits in the future. Together these controls account for the most important migration drivers found in previous studies. A further definition and summary statistics of the variables used in this article are provided in Tables 3 and 4. Table 3. Definition of variables Community level  Logarithm population density Logarithm of the community’s population density per square kilometer  Employment ratio Percentage of population employed  Toilet Percentage of population having no toilet or using public toilet  Hazard Indonesia: Community’s flood risk based on flood data of past 5 years, data from Indonesia National Agency for Disaster Management (BNPB), 2015 Ghana: Community’s erosion/flood risk based on average shoreline erosion within past 10 years, data for five communities from Center for Tropical Marine Research Bremen (ZMT), 2015 With 1=low, 2=low-medium, 3=medium, 4=medium-high, 5=high Household (HH) level  Female-headed HHa =1 if household head is female, =0 otherwise  Number of children in HHa Number of children of the age less than or equal to 16 years living in household  Household sizea Total number of household members  Relative HH incomea Household income as reported by HH relative to average community income  Distance to coasta Household’s linear distance to coast in kilometers Individual level  Migrant status =1 if migrant, =0 otherwise  Ownershipa =1 if house is owned by respondent or spouse, =0 otherwise  Networka Index between 0 and 5, based on how many of the following questions could be answered with ‘yes’: 1(2): Do you have immediate family members (other family members or friends) living abroad? 3(4): Do you have immediate family members (other family members or friends) living in another province of Indonesia? 5: Do you have family members or friends living in another community in Semarang (Keta)?  Unemployeda =1 if respondent reported to be unemployed when asked about occupation, =0 otherwise  Sexa =1 if female, =0 otherwise  Agea Age of respondent in years  Age2a Age squared  Marital statusa =1 if married, =0 otherwise  Educationa Years of education  Migration experiencea =1 if has lived somewhere else between age 18 years and now, =0 otherwise  Risk aversion ‘In general, I am very willing to take risks’, Likert scale from 1= Agree strongly to 5= Disagree strongly  Impatience ‘I am a patient person’. Likert scale from 1= Agree strongly to 5= Disagree strongly  Flooda ‘In your opinion, how much have environmental events affected you within the last 5 years? Please indicate your opinion regarding the following events on a scale from 0 to 10, where 0 stands for ‘not affected at all’ and 10 for ‘extremely affected’.  Subsidencea/storma  Erosiona  Aggregated Environmental change indexa Sum of individual’s perception on flood, subsidence/storm, erosion Community level  Logarithm population density Logarithm of the community’s population density per square kilometer  Employment ratio Percentage of population employed  Toilet Percentage of population having no toilet or using public toilet  Hazard Indonesia: Community’s flood risk based on flood data of past 5 years, data from Indonesia National Agency for Disaster Management (BNPB), 2015 Ghana: Community’s erosion/flood risk based on average shoreline erosion within past 10 years, data for five communities from Center for Tropical Marine Research Bremen (ZMT), 2015 With 1=low, 2=low-medium, 3=medium, 4=medium-high, 5=high Household (HH) level  Female-headed HHa =1 if household head is female, =0 otherwise  Number of children in HHa Number of children of the age less than or equal to 16 years living in household  Household sizea Total number of household members  Relative HH incomea Household income as reported by HH relative to average community income  Distance to coasta Household’s linear distance to coast in kilometers Individual level  Migrant status =1 if migrant, =0 otherwise  Ownershipa =1 if house is owned by respondent or spouse, =0 otherwise  Networka Index between 0 and 5, based on how many of the following questions could be answered with ‘yes’: 1(2): Do you have immediate family members (other family members or friends) living abroad? 3(4): Do you have immediate family members (other family members or friends) living in another province of Indonesia? 5: Do you have family members or friends living in another community in Semarang (Keta)?  Unemployeda =1 if respondent reported to be unemployed when asked about occupation, =0 otherwise  Sexa =1 if female, =0 otherwise  Agea Age of respondent in years  Age2a Age squared  Marital statusa =1 if married, =0 otherwise  Educationa Years of education  Migration experiencea =1 if has lived somewhere else between age 18 years and now, =0 otherwise  Risk aversion ‘In general, I am very willing to take risks’, Likert scale from 1= Agree strongly to 5= Disagree strongly  Impatience ‘I am a patient person’. Likert scale from 1= Agree strongly to 5= Disagree strongly  Flooda ‘In your opinion, how much have environmental events affected you within the last 5 years? Please indicate your opinion regarding the following events on a scale from 0 to 10, where 0 stands for ‘not affected at all’ and 10 for ‘extremely affected’.  Subsidencea/storma  Erosiona  Aggregated Environmental change indexa Sum of individual’s perception on flood, subsidence/storm, erosion a If migrant: at the time of migration, if non-migrant: at the time of interview. Table 3. Definition of variables Community level  Logarithm population density Logarithm of the community’s population density per square kilometer  Employment ratio Percentage of population employed  Toilet Percentage of population having no toilet or using public toilet  Hazard Indonesia: Community’s flood risk based on flood data of past 5 years, data from Indonesia National Agency for Disaster Management (BNPB), 2015 Ghana: Community’s erosion/flood risk based on average shoreline erosion within past 10 years, data for five communities from Center for Tropical Marine Research Bremen (ZMT), 2015 With 1=low, 2=low-medium, 3=medium, 4=medium-high, 5=high Household (HH) level  Female-headed HHa =1 if household head is female, =0 otherwise  Number of children in HHa Number of children of the age less than or equal to 16 years living in household  Household sizea Total number of household members  Relative HH incomea Household income as reported by HH relative to average community income  Distance to coasta Household’s linear distance to coast in kilometers Individual level  Migrant status =1 if migrant, =0 otherwise  Ownershipa =1 if house is owned by respondent or spouse, =0 otherwise  Networka Index between 0 and 5, based on how many of the following questions could be answered with ‘yes’: 1(2): Do you have immediate family members (other family members or friends) living abroad? 3(4): Do you have immediate family members (other family members or friends) living in another province of Indonesia? 5: Do you have family members or friends living in another community in Semarang (Keta)?  Unemployeda =1 if respondent reported to be unemployed when asked about occupation, =0 otherwise  Sexa =1 if female, =0 otherwise  Agea Age of respondent in years  Age2a Age squared  Marital statusa =1 if married, =0 otherwise  Educationa Years of education  Migration experiencea =1 if has lived somewhere else between age 18 years and now, =0 otherwise  Risk aversion ‘In general, I am very willing to take risks’, Likert scale from 1= Agree strongly to 5= Disagree strongly  Impatience ‘I am a patient person’. Likert scale from 1= Agree strongly to 5= Disagree strongly  Flooda ‘In your opinion, how much have environmental events affected you within the last 5 years? Please indicate your opinion regarding the following events on a scale from 0 to 10, where 0 stands for ‘not affected at all’ and 10 for ‘extremely affected’.  Subsidencea/storma  Erosiona  Aggregated Environmental change indexa Sum of individual’s perception on flood, subsidence/storm, erosion Community level  Logarithm population density Logarithm of the community’s population density per square kilometer  Employment ratio Percentage of population employed  Toilet Percentage of population having no toilet or using public toilet  Hazard Indonesia: Community’s flood risk based on flood data of past 5 years, data from Indonesia National Agency for Disaster Management (BNPB), 2015 Ghana: Community’s erosion/flood risk based on average shoreline erosion within past 10 years, data for five communities from Center for Tropical Marine Research Bremen (ZMT), 2015 With 1=low, 2=low-medium, 3=medium, 4=medium-high, 5=high Household (HH) level  Female-headed HHa =1 if household head is female, =0 otherwise  Number of children in HHa Number of children of the age less than or equal to 16 years living in household  Household sizea Total number of household members  Relative HH incomea Household income as reported by HH relative to average community income  Distance to coasta Household’s linear distance to coast in kilometers Individual level  Migrant status =1 if migrant, =0 otherwise  Ownershipa =1 if house is owned by respondent or spouse, =0 otherwise  Networka Index between 0 and 5, based on how many of the following questions could be answered with ‘yes’: 1(2): Do you have immediate family members (other family members or friends) living abroad? 3(4): Do you have immediate family members (other family members or friends) living in another province of Indonesia? 5: Do you have family members or friends living in another community in Semarang (Keta)?  Unemployeda =1 if respondent reported to be unemployed when asked about occupation, =0 otherwise  Sexa =1 if female, =0 otherwise  Agea Age of respondent in years  Age2a Age squared  Marital statusa =1 if married, =0 otherwise  Educationa Years of education  Migration experiencea =1 if has lived somewhere else between age 18 years and now, =0 otherwise  Risk aversion ‘In general, I am very willing to take risks’, Likert scale from 1= Agree strongly to 5= Disagree strongly  Impatience ‘I am a patient person’. Likert scale from 1= Agree strongly to 5= Disagree strongly  Flooda ‘In your opinion, how much have environmental events affected you within the last 5 years? Please indicate your opinion regarding the following events on a scale from 0 to 10, where 0 stands for ‘not affected at all’ and 10 for ‘extremely affected’.  Subsidencea/storma  Erosiona  Aggregated Environmental change indexa Sum of individual’s perception on flood, subsidence/storm, erosion a If migrant: at the time of migration, if non-migrant: at the time of interview. Table 4. Summary statistics for Indonesian sample (Ghanaian sample in brackets) Variable . Number of Observations . Mean . Standard deviation . Minimum . Maximum . Community level  Logarithm of the population density 309 8.76 1.09 5.30 9.87  Employment ratio 309 0.55 0.038 0.50 0.61  Toilet 309 0.12 0.08 0.01 0.31  Hazard 309 (174) 2.94 (2.65) 0.80 (1.15) 2 (1) 4 Household (HH) level  Female-headed HH 309 (277) 0.15 (0.44) 0.36 (0.49) 0 1  Number of children in HH 309 (277) 0.73 (0.98) 1.03 (1.33) 0 6 (8)  Household size 309 (277) 4.60 (5.82) 1.99 (2.81) 1 13 (15)  Relative HH income 309 (277) 1 (1) 0.67 (0.91) 0.06 (0.03) 6.94 (5.06)  Distance to coast 309 (277) 2.67 (2.10) 1.49 (2.37) 0 8.64 (7.38) Individual level  Migrant status 309 (277) 0.31 (0.31) 0.46 (0.46) 0 1  Ownership 309 (277) 0.53 (0.10) 0.49 (0.30) 0 1  Network 309 (277) 1.95 (3.93) 1.17 (0.96) 0 5  Unemployed 309 (277) 0.35 (0.37) 0.48 (0.48) 0 1  Sex 309 (277) 0.57 (0.53) 0.49 (0.49) 0 1  Age 309 (277) 36.63 (9.94) 14.45 (4.81) 18 86 (88)  Age2 309 (277) 1550.37 (1709.83) 1261.41 (1527.56) 324 7396 (7744)  Marital status 309 (277) 0.69 (0.42) 0.46 (0.49) 0 1  Education 309 (277) 11.13 (9.94) 3.64 (4.81) 0 18 (20)  Migration experience 308 (277) 0.30 (0.69) 0.45 (0.46) 0 1  Risk aversion 308 (277) 2.29 (2.77) 0.77 (1.33) 1 5  Impatience 308 (277) 2.35 (1.87) 0.83 (0.98) 1 5  Perception, flood 309 (277) 5.69 (5.24) 3.67 (3.67) 1 10  Perception, erosion 301 (277) 1.39 (4.20) 1.40 (3.56) 1 10  Perception, subsidence 302 3.09 3.33 1 10  Perception, storms (277) (4.00) (3.54) 1 10  Aggregated Environmental change index 300 (277) 10.03 (13.23) 6.47 (9.22) 3 30 Variable . Number of Observations . Mean . Standard deviation . Minimum . Maximum . Community level  Logarithm of the population density 309 8.76 1.09 5.30 9.87  Employment ratio 309 0.55 0.038 0.50 0.61  Toilet 309 0.12 0.08 0.01 0.31  Hazard 309 (174) 2.94 (2.65) 0.80 (1.15) 2 (1) 4 Household (HH) level  Female-headed HH 309 (277) 0.15 (0.44) 0.36 (0.49) 0 1  Number of children in HH 309 (277) 0.73 (0.98) 1.03 (1.33) 0 6 (8)  Household size 309 (277) 4.60 (5.82) 1.99 (2.81) 1 13 (15)  Relative HH income 309 (277) 1 (1) 0.67 (0.91) 0.06 (0.03) 6.94 (5.06)  Distance to coast 309 (277) 2.67 (2.10) 1.49 (2.37) 0 8.64 (7.38) Individual level  Migrant status 309 (277) 0.31 (0.31) 0.46 (0.46) 0 1  Ownership 309 (277) 0.53 (0.10) 0.49 (0.30) 0 1  Network 309 (277) 1.95 (3.93) 1.17 (0.96) 0 5  Unemployed 309 (277) 0.35 (0.37) 0.48 (0.48) 0 1  Sex 309 (277) 0.57 (0.53) 0.49 (0.49) 0 1  Age 309 (277) 36.63 (9.94) 14.45 (4.81) 18 86 (88)  Age2 309 (277) 1550.37 (1709.83) 1261.41 (1527.56) 324 7396 (7744)  Marital status 309 (277) 0.69 (0.42) 0.46 (0.49) 0 1  Education 309 (277) 11.13 (9.94) 3.64 (4.81) 0 18 (20)  Migration experience 308 (277) 0.30 (0.69) 0.45 (0.46) 0 1  Risk aversion 308 (277) 2.29 (2.77) 0.77 (1.33) 1 5  Impatience 308 (277) 2.35 (1.87) 0.83 (0.98) 1 5  Perception, flood 309 (277) 5.69 (5.24) 3.67 (3.67) 1 10  Perception, erosion 301 (277) 1.39 (4.20) 1.40 (3.56) 1 10  Perception, subsidence 302 3.09 3.33 1 10  Perception, storms (277) (4.00) (3.54) 1 10  Aggregated Environmental change index 300 (277) 10.03 (13.23) 6.47 (9.22) 3 30 Table 4. Summary statistics for Indonesian sample (Ghanaian sample in brackets) Variable . Number of Observations . Mean . Standard deviation . Minimum . Maximum . Community level  Logarithm of the population density 309 8.76 1.09 5.30 9.87  Employment ratio 309 0.55 0.038 0.50 0.61  Toilet 309 0.12 0.08 0.01 0.31  Hazard 309 (174) 2.94 (2.65) 0.80 (1.15) 2 (1) 4 Household (HH) level  Female-headed HH 309 (277) 0.15 (0.44) 0.36 (0.49) 0 1  Number of children in HH 309 (277) 0.73 (0.98) 1.03 (1.33) 0 6 (8)  Household size 309 (277) 4.60 (5.82) 1.99 (2.81) 1 13 (15)  Relative HH income 309 (277) 1 (1) 0.67 (0.91) 0.06 (0.03) 6.94 (5.06)  Distance to coast 309 (277) 2.67 (2.10) 1.49 (2.37) 0 8.64 (7.38) Individual level  Migrant status 309 (277) 0.31 (0.31) 0.46 (0.46) 0 1  Ownership 309 (277) 0.53 (0.10) 0.49 (0.30) 0 1  Network 309 (277) 1.95 (3.93) 1.17 (0.96) 0 5  Unemployed 309 (277) 0.35 (0.37) 0.48 (0.48) 0 1  Sex 309 (277) 0.57 (0.53) 0.49 (0.49) 0 1  Age 309 (277) 36.63 (9.94) 14.45 (4.81) 18 86 (88)  Age2 309 (277) 1550.37 (1709.83) 1261.41 (1527.56) 324 7396 (7744)  Marital status 309 (277) 0.69 (0.42) 0.46 (0.49) 0 1  Education 309 (277) 11.13 (9.94) 3.64 (4.81) 0 18 (20)  Migration experience 308 (277) 0.30 (0.69) 0.45 (0.46) 0 1  Risk aversion 308 (277) 2.29 (2.77) 0.77 (1.33) 1 5  Impatience 308 (277) 2.35 (1.87) 0.83 (0.98) 1 5  Perception, flood 309 (277) 5.69 (5.24) 3.67 (3.67) 1 10  Perception, erosion 301 (277) 1.39 (4.20) 1.40 (3.56) 1 10  Perception, subsidence 302 3.09 3.33 1 10  Perception, storms (277) (4.00) (3.54) 1 10  Aggregated Environmental change index 300 (277) 10.03 (13.23) 6.47 (9.22) 3 30 Variable . Number of Observations . Mean . Standard deviation . Minimum . Maximum . Community level  Logarithm of the population density 309 8.76 1.09 5.30 9.87  Employment ratio 309 0.55 0.038 0.50 0.61  Toilet 309 0.12 0.08 0.01 0.31  Hazard 309 (174) 2.94 (2.65) 0.80 (1.15) 2 (1) 4 Household (HH) level  Female-headed HH 309 (277) 0.15 (0.44) 0.36 (0.49) 0 1  Number of children in HH 309 (277) 0.73 (0.98) 1.03 (1.33) 0 6 (8)  Household size 309 (277) 4.60 (5.82) 1.99 (2.81) 1 13 (15)  Relative HH income 309 (277) 1 (1) 0.67 (0.91) 0.06 (0.03) 6.94 (5.06)  Distance to coast 309 (277) 2.67 (2.10) 1.49 (2.37) 0 8.64 (7.38) Individual level  Migrant status 309 (277) 0.31 (0.31) 0.46 (0.46) 0 1  Ownership 309 (277) 0.53 (0.10) 0.49 (0.30) 0 1  Network 309 (277) 1.95 (3.93) 1.17 (0.96) 0 5  Unemployed 309 (277) 0.35 (0.37) 0.48 (0.48) 0 1  Sex 309 (277) 0.57 (0.53) 0.49 (0.49) 0 1  Age 309 (277) 36.63 (9.94) 14.45 (4.81) 18 86 (88)  Age2 309 (277) 1550.37 (1709.83) 1261.41 (1527.56) 324 7396 (7744)  Marital status 309 (277) 0.69 (0.42) 0.46 (0.49) 0 1  Education 309 (277) 11.13 (9.94) 3.64 (4.81) 0 18 (20)  Migration experience 308 (277) 0.30 (0.69) 0.45 (0.46) 0 1  Risk aversion 308 (277) 2.29 (2.77) 0.77 (1.33) 1 5  Impatience 308 (277) 2.35 (1.87) 0.83 (0.98) 1 5  Perception, flood 309 (277) 5.69 (5.24) 3.67 (3.67) 1 10  Perception, erosion 301 (277) 1.39 (4.20) 1.40 (3.56) 1 10  Perception, subsidence 302 3.09 3.33 1 10  Perception, storms (277) (4.00) (3.54) 1 10  Aggregated Environmental change index 300 (277) 10.03 (13.23) 6.47 (9.22) 3 30 Since independence of observations cannot be assumed and individuals from one household are expected to be more similar, all models get adjusted for clustering at the household level. Furthermore, it is accounted for the fact that there is only information for one migrant (non-migrant) per household, regardless of the total number of migrants (non-migrants) by weighting the observations based on the inverse of the probability of selection. The model is then estimated separately for the two study regions. 4.1 Indonesia The results for Indonesia are presented in Table 5. Overall, the results of the control variables are consistent with previous studies. At the community level, it is found that the community’s percentage of people with employment has a strong and significant impact on the individual migration decision. The better the employment situation in a community, the less likely are people to leave this community. Table 5. Regression results Indonesia . (1) . (2) . (3) . (4) . (5) . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Logarithm population density −0.094 (0.20) −0.143 (0.23) −0.086 (0.19) 0.091 (0.21) 0.045 (0.23) Employment ratio −12.008**(5.01) −11.922**(5.13) −12.200**(5.06) −11.736**(5.06) −14.721**(5.83) Toilet type −0.077 (2.46) −0.577 (2.29) −0.756 (2.28) −3.060 (2.73) 2.007 (3.02) Female-headed Household (HH) (=1) 0.133 (0.45) 0.062 (0.48) 0.134 (0.48) 0.146 (0.49) 0.149 (0.47) Number of children in HH −0.435**(0.21) −0.580**(0.24) −0.525**(0.23) −0.411**(0.21) −0.430**(0.21) Household size 0.119 (0.10) 0.113 (0.11) 0.136 (0.11) 0.137 (0.11) 0.127 (0.10) Ownership −0.559 (0.40) −0.532 (0.42) −0.529 (0.42) −0.564 (0.42) −0.614 (0.41) Relative HH income 0.143 (0.30) 0.261 (0.31) 0.184 (0.30) 0.118 (0.32) 0.159 (0.31) Network 0.485***(0.17) 0.500***(0.18) 0.488***(0.17) 0.485***(0.17) 0.502***(0.17) Unemployed (=1) −1.396***(0.43) −1.229***(0.43) −1.311***(0.44) −1.576***(0.44) −1.511***(0.44) Sex (female =1) −0.198 (0.40) −0.257 (0.41) −0.172 (0.41) −0.098 (0.41) −0.124 (0.41) Age 1.008***(0.25) 1.014***(0.26) 0.969***(0.26) 1.021***(0.26) 1.020***(0.26) Age2 −0.020***(0.00) −0.020***(0.00) −0.020***(0.00) −0.021***(0.00) −0.020***(0.00) Married (=1) 0.531**(0.25) 0.504**(0.25) 0.542**(0.25) 0.578**(0.25) 0.585**(0.26) Education −0.035 (0.07) −0.055 (0.07) −0.028 (0.07) −0.033 (0.07) −0.036 (0.07) Migration experience (=1) −0.010 (0.43) −0.007 (0.45) −0.093 (0.47) 0.090 (0.42) −0.026 (0.45) Risk aversion −0.526**(0.26) −0.482*(0.28) −0.562*(0.28) −0.586**(0.28) −0.549**(0.27) Impatience −0.740***(0.24) −0.629**(0.30) −0.670**(0.28) −0.749***(0.24) −0.751***(0.24) Flood −0.034 (0.05) Subsidence −0.083 (0.08) Erosion −0.143 (0.16) Distance to coast −0.262*(0.13) Hazard −0.449 (0.41) Constant −0.254 (4.92) −0.105 (5.61) 0.059 (5.19) −1.119 (4.76) −3.065 (5.31) Community fixed effects No No No No No BIC 550.211 533.145 536.327 544.351 549.187 AIC 475.674 459.069 462.319 469.814 474.650 Pseudo R2 0.534 0.543 0.537 0.541 0.535 Percent correctly classified10 83.71% 83.61% 84.00% 85.34% 85.02% Percent reduction in error 48.10% 47.78% 49.03% 53.29% 52.28% N 307 300 299 307 307 . (1) . (2) . (3) . (4) . (5) . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Logarithm population density −0.094 (0.20) −0.143 (0.23) −0.086 (0.19) 0.091 (0.21) 0.045 (0.23) Employment ratio −12.008**(5.01) −11.922**(5.13) −12.200**(5.06) −11.736**(5.06) −14.721**(5.83) Toilet type −0.077 (2.46) −0.577 (2.29) −0.756 (2.28) −3.060 (2.73) 2.007 (3.02) Female-headed Household (HH) (=1) 0.133 (0.45) 0.062 (0.48) 0.134 (0.48) 0.146 (0.49) 0.149 (0.47) Number of children in HH −0.435**(0.21) −0.580**(0.24) −0.525**(0.23) −0.411**(0.21) −0.430**(0.21) Household size 0.119 (0.10) 0.113 (0.11) 0.136 (0.11) 0.137 (0.11) 0.127 (0.10) Ownership −0.559 (0.40) −0.532 (0.42) −0.529 (0.42) −0.564 (0.42) −0.614 (0.41) Relative HH income 0.143 (0.30) 0.261 (0.31) 0.184 (0.30) 0.118 (0.32) 0.159 (0.31) Network 0.485***(0.17) 0.500***(0.18) 0.488***(0.17) 0.485***(0.17) 0.502***(0.17) Unemployed (=1) −1.396***(0.43) −1.229***(0.43) −1.311***(0.44) −1.576***(0.44) −1.511***(0.44) Sex (female =1) −0.198 (0.40) −0.257 (0.41) −0.172 (0.41) −0.098 (0.41) −0.124 (0.41) Age 1.008***(0.25) 1.014***(0.26) 0.969***(0.26) 1.021***(0.26) 1.020***(0.26) Age2 −0.020***(0.00) −0.020***(0.00) −0.020***(0.00) −0.021***(0.00) −0.020***(0.00) Married (=1) 0.531**(0.25) 0.504**(0.25) 0.542**(0.25) 0.578**(0.25) 0.585**(0.26) Education −0.035 (0.07) −0.055 (0.07) −0.028 (0.07) −0.033 (0.07) −0.036 (0.07) Migration experience (=1) −0.010 (0.43) −0.007 (0.45) −0.093 (0.47) 0.090 (0.42) −0.026 (0.45) Risk aversion −0.526**(0.26) −0.482*(0.28) −0.562*(0.28) −0.586**(0.28) −0.549**(0.27) Impatience −0.740***(0.24) −0.629**(0.30) −0.670**(0.28) −0.749***(0.24) −0.751***(0.24) Flood −0.034 (0.05) Subsidence −0.083 (0.08) Erosion −0.143 (0.16) Distance to coast −0.262*(0.13) Hazard −0.449 (0.41) Constant −0.254 (4.92) −0.105 (5.61) 0.059 (5.19) −1.119 (4.76) −3.065 (5.31) Community fixed effects No No No No No BIC 550.211 533.145 536.327 544.351 549.187 AIC 475.674 459.069 462.319 469.814 474.650 Pseudo R2 0.534 0.543 0.537 0.541 0.535 Percent correctly classified10 83.71% 83.61% 84.00% 85.34% 85.02% Percent reduction in error 48.10% 47.78% 49.03% 53.29% 52.28% N 307 300 299 307 307 Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. Table 5. Regression results Indonesia . (1) . (2) . (3) . (4) . (5) . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Logarithm population density −0.094 (0.20) −0.143 (0.23) −0.086 (0.19) 0.091 (0.21) 0.045 (0.23) Employment ratio −12.008**(5.01) −11.922**(5.13) −12.200**(5.06) −11.736**(5.06) −14.721**(5.83) Toilet type −0.077 (2.46) −0.577 (2.29) −0.756 (2.28) −3.060 (2.73) 2.007 (3.02) Female-headed Household (HH) (=1) 0.133 (0.45) 0.062 (0.48) 0.134 (0.48) 0.146 (0.49) 0.149 (0.47) Number of children in HH −0.435**(0.21) −0.580**(0.24) −0.525**(0.23) −0.411**(0.21) −0.430**(0.21) Household size 0.119 (0.10) 0.113 (0.11) 0.136 (0.11) 0.137 (0.11) 0.127 (0.10) Ownership −0.559 (0.40) −0.532 (0.42) −0.529 (0.42) −0.564 (0.42) −0.614 (0.41) Relative HH income 0.143 (0.30) 0.261 (0.31) 0.184 (0.30) 0.118 (0.32) 0.159 (0.31) Network 0.485***(0.17) 0.500***(0.18) 0.488***(0.17) 0.485***(0.17) 0.502***(0.17) Unemployed (=1) −1.396***(0.43) −1.229***(0.43) −1.311***(0.44) −1.576***(0.44) −1.511***(0.44) Sex (female =1) −0.198 (0.40) −0.257 (0.41) −0.172 (0.41) −0.098 (0.41) −0.124 (0.41) Age 1.008***(0.25) 1.014***(0.26) 0.969***(0.26) 1.021***(0.26) 1.020***(0.26) Age2 −0.020***(0.00) −0.020***(0.00) −0.020***(0.00) −0.021***(0.00) −0.020***(0.00) Married (=1) 0.531**(0.25) 0.504**(0.25) 0.542**(0.25) 0.578**(0.25) 0.585**(0.26) Education −0.035 (0.07) −0.055 (0.07) −0.028 (0.07) −0.033 (0.07) −0.036 (0.07) Migration experience (=1) −0.010 (0.43) −0.007 (0.45) −0.093 (0.47) 0.090 (0.42) −0.026 (0.45) Risk aversion −0.526**(0.26) −0.482*(0.28) −0.562*(0.28) −0.586**(0.28) −0.549**(0.27) Impatience −0.740***(0.24) −0.629**(0.30) −0.670**(0.28) −0.749***(0.24) −0.751***(0.24) Flood −0.034 (0.05) Subsidence −0.083 (0.08) Erosion −0.143 (0.16) Distance to coast −0.262*(0.13) Hazard −0.449 (0.41) Constant −0.254 (4.92) −0.105 (5.61) 0.059 (5.19) −1.119 (4.76) −3.065 (5.31) Community fixed effects No No No No No BIC 550.211 533.145 536.327 544.351 549.187 AIC 475.674 459.069 462.319 469.814 474.650 Pseudo R2 0.534 0.543 0.537 0.541 0.535 Percent correctly classified10 83.71% 83.61% 84.00% 85.34% 85.02% Percent reduction in error 48.10% 47.78% 49.03% 53.29% 52.28% N 307 300 299 307 307 . (1) . (2) . (3) . (4) . (5) . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Logarithm population density −0.094 (0.20) −0.143 (0.23) −0.086 (0.19) 0.091 (0.21) 0.045 (0.23) Employment ratio −12.008**(5.01) −11.922**(5.13) −12.200**(5.06) −11.736**(5.06) −14.721**(5.83) Toilet type −0.077 (2.46) −0.577 (2.29) −0.756 (2.28) −3.060 (2.73) 2.007 (3.02) Female-headed Household (HH) (=1) 0.133 (0.45) 0.062 (0.48) 0.134 (0.48) 0.146 (0.49) 0.149 (0.47) Number of children in HH −0.435**(0.21) −0.580**(0.24) −0.525**(0.23) −0.411**(0.21) −0.430**(0.21) Household size 0.119 (0.10) 0.113 (0.11) 0.136 (0.11) 0.137 (0.11) 0.127 (0.10) Ownership −0.559 (0.40) −0.532 (0.42) −0.529 (0.42) −0.564 (0.42) −0.614 (0.41) Relative HH income 0.143 (0.30) 0.261 (0.31) 0.184 (0.30) 0.118 (0.32) 0.159 (0.31) Network 0.485***(0.17) 0.500***(0.18) 0.488***(0.17) 0.485***(0.17) 0.502***(0.17) Unemployed (=1) −1.396***(0.43) −1.229***(0.43) −1.311***(0.44) −1.576***(0.44) −1.511***(0.44) Sex (female =1) −0.198 (0.40) −0.257 (0.41) −0.172 (0.41) −0.098 (0.41) −0.124 (0.41) Age 1.008***(0.25) 1.014***(0.26) 0.969***(0.26) 1.021***(0.26) 1.020***(0.26) Age2 −0.020***(0.00) −0.020***(0.00) −0.020***(0.00) −0.021***(0.00) −0.020***(0.00) Married (=1) 0.531**(0.25) 0.504**(0.25) 0.542**(0.25) 0.578**(0.25) 0.585**(0.26) Education −0.035 (0.07) −0.055 (0.07) −0.028 (0.07) −0.033 (0.07) −0.036 (0.07) Migration experience (=1) −0.010 (0.43) −0.007 (0.45) −0.093 (0.47) 0.090 (0.42) −0.026 (0.45) Risk aversion −0.526**(0.26) −0.482*(0.28) −0.562*(0.28) −0.586**(0.28) −0.549**(0.27) Impatience −0.740***(0.24) −0.629**(0.30) −0.670**(0.28) −0.749***(0.24) −0.751***(0.24) Flood −0.034 (0.05) Subsidence −0.083 (0.08) Erosion −0.143 (0.16) Distance to coast −0.262*(0.13) Hazard −0.449 (0.41) Constant −0.254 (4.92) −0.105 (5.61) 0.059 (5.19) −1.119 (4.76) −3.065 (5.31) Community fixed effects No No No No No BIC 550.211 533.145 536.327 544.351 549.187 AIC 475.674 459.069 462.319 469.814 474.650 Pseudo R2 0.534 0.543 0.537 0.541 0.535 Percent correctly classified10 83.71% 83.61% 84.00% 85.34% 85.02% Percent reduction in error 48.10% 47.78% 49.03% 53.29% 52.28% N 307 300 299 307 307 Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. On the household level, only one variable, namely, the number of children younger than 17 years of age, turns out to be significant. As expected, a higher number of younger children reduces the probability to migrate.9 However, there seems to be a very important role of networks. Individuals, who have more friends and family members living in another district or abroad and thus have a better network which helps facilitating migration, are more likely to move. At the individual level, there is a negative and highly significant correlation between being unemployed and out-migration. Moreover, age is found to have a very significant effect on the migration propensity, with a peak at the age of 25 years. Married individuals and individuals who are rather risk-loving or patient are also more likely to migrate. While non-married people would be expected to be more likely to move, this finding is easy to explain in the Indonesian context. As seen before, marriage was named as the main reason for moving. In Semarang it is common to wait until after the wedding with moving together. This results in many just married individuals who leave their community to move to their spouse. The effect of the stated preferences, however, is exactly as anticipated: rather risk-loving as well as rather patient people are more likely to migrate. Taking a look at the variables of main interest, there is no significant correlation between the individual’s perceptions of flood, subsidence, and erosion, and her out-migration decision. Also the hazard categorization of communities cannot be linked to individual migration decisions. The household’s distance to the coast, however, which serves as a more objective measure of exposure to coastal events has a significant impact on migration behavior. Individuals living in households closer to the coast have a higher probability to leave the community than people living further away. This finding serves as a first indicator that the environment actually has a direct effect on migration. However, the coefficient is only significant at the 10% level and does not survive further robustness tests (see Section 5.3). To evaluate the accuracy and the goodness-of-fit of the models, the AIC and BIC as well as McFadden pseudo R2 are included which do very well.11 Additionally, the percentage of correctly classified cases, which is a commonly used goodness-of-fit measure and assesses how well the predictions fit the observed outcome, and the percent reduction in error are included.12 Even though, the environmental variables do not add much to the goodness of fit, the models are able to predict around 85% of the outcome correctly. 4.2 Ghana The results for Ghana are presented in Table 6. Unfortunately, no official data on community level were available which reduces the set of control variables to household- and individual-level factors. Nevertheless, to account for differences in communities, community fixed effects are included. Overall, the models support common findings. Like in Indonesia, there is a weak link between the number of children in a household and out-migration. Additionally, migrants are significantly more likely to move away from bigger households as well as from relatively deprived households—just as expected. This is supporting the hypothesis that households with a lower relative income send out a migrant to diversify risks and to generate income somewhere else. Networks which turned out to be imperative in Indonesia are not found to explain migration decisions in Ghana which is probably due to the fact that nearly everyone in Ghana has friends or family in other parts or even outside of the country—making it less crucial for own migration decisions. At the individual level, younger persons13 are more likely to move, whereas the marital status does not seem to play a role in the Ghanaian context. Furthermore, there is weak evidence in three of the five specifications that men are more likely to leave than women. Individual unemployment, as well as the level of education, is positively correlated with the decision to migrate, as expected, and significant at the 1% level. Risk aversion and time preference do also play a role again. These findings can be strongly supported by open-ended interviews conducted in those communities. Keta is an economically rather weak and rural region not offering many jobs, especially for well-educated people. Thus, individuals who are not able to find work and better educated people who have better chances to find a job somewhere else often take their chance in urban areas like Accra, Tema, or Lomé. Table 6. Regression results Ghana . (1) . (2) . (3) . (4) . (5) . . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Female-headed Household (HH) (=1) 0.263 (0.33) 0.320 (0.31) 0.231 (0.33) 0.311 (0.32) 0.770*(0.41) Number of children in HH −0.210*(0.12) −0.212*(0.12) −0.177 (0.12) −0.212*(0.12) −0.329*(0.19) Household size 0.314***(0.09) 0.317***(0.09) 0.276***(0.08) 0.314***(0.09) 0.364***(0.12) Ownership −1.429 (0.96) −1.282 (0.97) −1.385 (0.89) −1.353 (0.97) (Dropped)14 Relative HH income −0.329*(0.18) −0.347**(0.18) −0.339*(0.17) −0.339*(0.17) −0.118*(0.06) Network 0.137 (0.20) 0.116 (0.20) 0.155 (0.21) 0.119 (0.20) 0.496*(0.28) Unemployed (=1) 0.984***(0.36) 0.964***(0.36) 0.985***(0.37) 0.983***(0.36) 1.342***_(0.47) Sex (female = 1) −0.583 (0.35) −0.648*(0.35) −0.475 (0.37) −0.615*(0.35) −1.196***(0.46) Age −0.056***(0.02) −0.058***(0.02) −0.055***(0.02) −0.057***(0.02) −0.022 (0.02) Married (=1) −0.236 (0.21) −0.235 (0.21) −0.267 (0.21) −0.232 (0.20) −0.373 (0.23) Education 0.399***(0.15) 0.391***(0.15) 0.405***(0.15) 0.393***(0.14) 0.341**(0.17) Migration experience (=1) 0.934**(0.41) 0.931**(0.41) 0.920**(0.41) 0.935***(0.41) 0.923*(0.50) Risk aversion −0.439***(0.16) −0.434***(0.16) −0.468***(0.17) −0.437***(0.16) −0.570***(0.20) Impatience −0.318*(0.17) −0.318*(0.17) −0.289**(0.17) −0.323*(0.17) −0.535**(0.21) Flood 0.027 (0.05) Erosion −0.027 (0.06) Storm 0.130**(0.06) Distance to coast −0.007 (0.48) Hazard −0.037 (0.25) Constant −0.840 (1.64) −0.437 (1.63) −1.139 (1.64) −0.531 (3.25) −1.215 (1.86) Community fixed effects Yes Yes Yes Yes Yes BIC 646.543 646.619 639.499 647.136 399.020 AIC 566.815 566.891 559.770 567.407 346.956 Pseudo R2 0.362 0.362 0.371 0.361 0.339 Percent correctly classified15 82.31% 81.59% 83.05% 81.59% 78.48% Percent reduction in error 43.75% 41.46% 46.10% 41.46% 37.28% N 277 277 277 277 17416 . (1) . (2) . (3) . (4) . (5) . . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Female-headed Household (HH) (=1) 0.263 (0.33) 0.320 (0.31) 0.231 (0.33) 0.311 (0.32) 0.770*(0.41) Number of children in HH −0.210*(0.12) −0.212*(0.12) −0.177 (0.12) −0.212*(0.12) −0.329*(0.19) Household size 0.314***(0.09) 0.317***(0.09) 0.276***(0.08) 0.314***(0.09) 0.364***(0.12) Ownership −1.429 (0.96) −1.282 (0.97) −1.385 (0.89) −1.353 (0.97) (Dropped)14 Relative HH income −0.329*(0.18) −0.347**(0.18) −0.339*(0.17) −0.339*(0.17) −0.118*(0.06) Network 0.137 (0.20) 0.116 (0.20) 0.155 (0.21) 0.119 (0.20) 0.496*(0.28) Unemployed (=1) 0.984***(0.36) 0.964***(0.36) 0.985***(0.37) 0.983***(0.36) 1.342***_(0.47) Sex (female = 1) −0.583 (0.35) −0.648*(0.35) −0.475 (0.37) −0.615*(0.35) −1.196***(0.46) Age −0.056***(0.02) −0.058***(0.02) −0.055***(0.02) −0.057***(0.02) −0.022 (0.02) Married (=1) −0.236 (0.21) −0.235 (0.21) −0.267 (0.21) −0.232 (0.20) −0.373 (0.23) Education 0.399***(0.15) 0.391***(0.15) 0.405***(0.15) 0.393***(0.14) 0.341**(0.17) Migration experience (=1) 0.934**(0.41) 0.931**(0.41) 0.920**(0.41) 0.935***(0.41) 0.923*(0.50) Risk aversion −0.439***(0.16) −0.434***(0.16) −0.468***(0.17) −0.437***(0.16) −0.570***(0.20) Impatience −0.318*(0.17) −0.318*(0.17) −0.289**(0.17) −0.323*(0.17) −0.535**(0.21) Flood 0.027 (0.05) Erosion −0.027 (0.06) Storm 0.130**(0.06) Distance to coast −0.007 (0.48) Hazard −0.037 (0.25) Constant −0.840 (1.64) −0.437 (1.63) −1.139 (1.64) −0.531 (3.25) −1.215 (1.86) Community fixed effects Yes Yes Yes Yes Yes BIC 646.543 646.619 639.499 647.136 399.020 AIC 566.815 566.891 559.770 567.407 346.956 Pseudo R2 0.362 0.362 0.371 0.361 0.339 Percent correctly classified15 82.31% 81.59% 83.05% 81.59% 78.48% Percent reduction in error 43.75% 41.46% 46.10% 41.46% 37.28% N 277 277 277 277 17416 Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. Table 6. Regression results Ghana . (1) . (2) . (3) . (4) . (5) . . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Female-headed Household (HH) (=1) 0.263 (0.33) 0.320 (0.31) 0.231 (0.33) 0.311 (0.32) 0.770*(0.41) Number of children in HH −0.210*(0.12) −0.212*(0.12) −0.177 (0.12) −0.212*(0.12) −0.329*(0.19) Household size 0.314***(0.09) 0.317***(0.09) 0.276***(0.08) 0.314***(0.09) 0.364***(0.12) Ownership −1.429 (0.96) −1.282 (0.97) −1.385 (0.89) −1.353 (0.97) (Dropped)14 Relative HH income −0.329*(0.18) −0.347**(0.18) −0.339*(0.17) −0.339*(0.17) −0.118*(0.06) Network 0.137 (0.20) 0.116 (0.20) 0.155 (0.21) 0.119 (0.20) 0.496*(0.28) Unemployed (=1) 0.984***(0.36) 0.964***(0.36) 0.985***(0.37) 0.983***(0.36) 1.342***_(0.47) Sex (female = 1) −0.583 (0.35) −0.648*(0.35) −0.475 (0.37) −0.615*(0.35) −1.196***(0.46) Age −0.056***(0.02) −0.058***(0.02) −0.055***(0.02) −0.057***(0.02) −0.022 (0.02) Married (=1) −0.236 (0.21) −0.235 (0.21) −0.267 (0.21) −0.232 (0.20) −0.373 (0.23) Education 0.399***(0.15) 0.391***(0.15) 0.405***(0.15) 0.393***(0.14) 0.341**(0.17) Migration experience (=1) 0.934**(0.41) 0.931**(0.41) 0.920**(0.41) 0.935***(0.41) 0.923*(0.50) Risk aversion −0.439***(0.16) −0.434***(0.16) −0.468***(0.17) −0.437***(0.16) −0.570***(0.20) Impatience −0.318*(0.17) −0.318*(0.17) −0.289**(0.17) −0.323*(0.17) −0.535**(0.21) Flood 0.027 (0.05) Erosion −0.027 (0.06) Storm 0.130**(0.06) Distance to coast −0.007 (0.48) Hazard −0.037 (0.25) Constant −0.840 (1.64) −0.437 (1.63) −1.139 (1.64) −0.531 (3.25) −1.215 (1.86) Community fixed effects Yes Yes Yes Yes Yes BIC 646.543 646.619 639.499 647.136 399.020 AIC 566.815 566.891 559.770 567.407 346.956 Pseudo R2 0.362 0.362 0.371 0.361 0.339 Percent correctly classified15 82.31% 81.59% 83.05% 81.59% 78.48% Percent reduction in error 43.75% 41.46% 46.10% 41.46% 37.28% N 277 277 277 277 17416 . (1) . (2) . (3) . (4) . (5) . . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Female-headed Household (HH) (=1) 0.263 (0.33) 0.320 (0.31) 0.231 (0.33) 0.311 (0.32) 0.770*(0.41) Number of children in HH −0.210*(0.12) −0.212*(0.12) −0.177 (0.12) −0.212*(0.12) −0.329*(0.19) Household size 0.314***(0.09) 0.317***(0.09) 0.276***(0.08) 0.314***(0.09) 0.364***(0.12) Ownership −1.429 (0.96) −1.282 (0.97) −1.385 (0.89) −1.353 (0.97) (Dropped)14 Relative HH income −0.329*(0.18) −0.347**(0.18) −0.339*(0.17) −0.339*(0.17) −0.118*(0.06) Network 0.137 (0.20) 0.116 (0.20) 0.155 (0.21) 0.119 (0.20) 0.496*(0.28) Unemployed (=1) 0.984***(0.36) 0.964***(0.36) 0.985***(0.37) 0.983***(0.36) 1.342***_(0.47) Sex (female = 1) −0.583 (0.35) −0.648*(0.35) −0.475 (0.37) −0.615*(0.35) −1.196***(0.46) Age −0.056***(0.02) −0.058***(0.02) −0.055***(0.02) −0.057***(0.02) −0.022 (0.02) Married (=1) −0.236 (0.21) −0.235 (0.21) −0.267 (0.21) −0.232 (0.20) −0.373 (0.23) Education 0.399***(0.15) 0.391***(0.15) 0.405***(0.15) 0.393***(0.14) 0.341**(0.17) Migration experience (=1) 0.934**(0.41) 0.931**(0.41) 0.920**(0.41) 0.935***(0.41) 0.923*(0.50) Risk aversion −0.439***(0.16) −0.434***(0.16) −0.468***(0.17) −0.437***(0.16) −0.570***(0.20) Impatience −0.318*(0.17) −0.318*(0.17) −0.289**(0.17) −0.323*(0.17) −0.535**(0.21) Flood 0.027 (0.05) Erosion −0.027 (0.06) Storm 0.130**(0.06) Distance to coast −0.007 (0.48) Hazard −0.037 (0.25) Constant −0.840 (1.64) −0.437 (1.63) −1.139 (1.64) −0.531 (3.25) −1.215 (1.86) Community fixed effects Yes Yes Yes Yes Yes BIC 646.543 646.619 639.499 647.136 399.020 AIC 566.815 566.891 559.770 567.407 346.956 Pseudo R2 0.362 0.362 0.371 0.361 0.339 Percent correctly classified15 82.31% 81.59% 83.05% 81.59% 78.48% Percent reduction in error 43.75% 41.46% 46.10% 41.46% 37.28% N 277 277 277 277 17416 Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. Coming again to the variables of interest, no correlation between individual perceptions of flood or erosion, the household’s distance to coast or the community’s level of hazard, and migration decisions can be found. However, the coefficient of storm perceptions is found to be significant at the 5% level and meaningful in size. When looking at predicted probabilities, individuals, who perceived to be highly affected by storms, are 11% more likely to leave the community than their less affected neighbors. 4.3 Robustness checks Environmental events like flooding, erosion, subsidence, or storms are clearly exogenous to the individual’s decision to migrate which avoids the problem of reverse causality. Additionally, also the control variables are exogenous to migration, since they refer to the time just before the migration and are therefore not influenced by the migration itself. Only the control variables risk aversion and patience were measured after the migration took place.17 Thus, they might have changed due to positive or negative feedback of the migration experience.18 Even though the focus is on the effects of environmental factors, which are clearly not influenced by individual migration decisions and therefore exogenous, it has to be ensured that the estimates for the environmental factors are not biased. Additionally, multicollinearity of covariates could also reduce the efficiency of the estimates (see also correlation matrix (Table 7)). Thus, several robustness checks get performed (see Tables 8 and 9). In different specifications it is tested whether the inclusion or exclusion of community fixed effects, the omission of clusters at the household level, or a reduced set of control variables impact on the effect of environmental factors on migration. Since Clarke (2005) argues that including control variables at all may already increase the bias, an additional specification without any controls is also estimated. It is found that results are robust to these changes, and only the anyhow weak coefficient of the Indonesian households’ distance to the coast loses its significance in the majority of robustness tests. An additional robustness test checks whether an analysis at household level provides new insights, since members within one household are expected to be exposed to the same environmental changes.19 However, perceptions of the household head, the household’s distance to the coast, and the community’s hazard categorization do not help to explain whether a household has a migrant. Only the household head’s perception of storms is highly significant, supporting the findings on individual level. Another robustness test addresses the question whether individuals are only found to migrate when experiencing more than one adverse environmental condition simultaneously. Therefore, we sum up the different perception measures to have one aggregated index. However, this index does not turn out to be significant. Table 7. Correlation matrix of covariates Indonesia/ Ghana . Logarithmic population . Employment status . Toilet . Female- headed Household (HH) . Number of children in HH . Household size . Ownership . Relative HH income . Networks . Sex . Age . Marital status . Education . Unemployed . Migration experience . Risk aversion . Patience . Logarithm population 1.00 Employment −0.324*** 1.00 Toilet 0.436*** 0.241*** 1.00 Female headed 0.077 −0.099* 0.030 1.00 Number of children −0.129** 0.073 −0.018 −0.028/ 0.025** 1.00 Household size −0.054 0.041 0.096* −0.122**/ −0.016 0.322***/ 0.699*** 1.00 Ownership −0.109* −0.017 −0.087 0.007/ −0.18*** 0.062/ 0.048 −0.20***/ −0.031 1.00 Relative HH income −0.000 0.000 0.000 −0.103**/ −0.18*** −0.06/ −0.011 0.113**/ 0.111* −0.001/ 0.044 1.00 Networks 0.064 −0.246*** −0.033 −0.059/ 0.074 −0.18***/ 0.015 −0.038/ −0.096 0.008/ 0.088 0.133**/ 0.032 1.00 Sex 0.066 0.044 0.121** 0.133**/ 0.274*** 0.002/ 0.003 0.002/ 0.037 −0.001/ 0.003 −0.086/ −0.170*** −0.044/ −0.026 1.00 Age 0.019 −0.188*** −0.142** 0.049/ 0.000 0.008/ −0.032 −0.008/ −0.110* 0.434***/ 0.339*** −0.101*/ −0.157*** 0.052/ −0.003 0.016/ 0.088 1.00 Marital status −0.069 0.012 −0.052 0.237***/ −0.267*** 0.352***/ 0.111** 0.352***/ −0.003 0.201***/ 0.191*** 0.024/ −0.174*** 0.052/ −0.014 −0.004/ 0.071 0.186***/ 0.225*** 1.00 Education 0.066 −0.142** −0.110* −0.062/ −0.109* −0.144**/ −0.072 −0.144**/ −0.034 −0.116**/ −0.146** 0.288***/ 0.235*** 0.186***/ 0.112* −0.082/ −0.285*** −0.27***/ −0.314*** 0.076/ −0.094 1.00 Unemployed 0.077 0.012 0.028 0.020/ 0.167*** 0.019/ −0.094 0.019/ −0.024 0.119**/ −0.010 −0.069/ 0.018 −0.046/ −0.105* 0.087/ −0.010 0.166***/ −0.153** −0.081/ −0.325*** −0.099*/ −0.18*** 1.00 Migration experience −0.001 −0.181*** −0.138** −0.010/ −0.05 0.072/ 0.063 0.072/ 0.012 0.045/ 0.065 0.11*/ 0.101* 0.16***/ 0.122* −0.121**/ −0.022 0.166***/ 0.184*** 0.056/ 0.087 0.22***/ 0.063 −0.014/ −0.094 1.00 Risk aversion 0.055 −0.065 −0.047 0.076/ −0.003 −0.056/ 0.089 −0.056/ 0.095 −0.017/ 0.130** −0.041/ −0.119** −0.061/ −0.146** 0.176***/ 0.139** 0.137**/ 0.219*** −0.015/ 0.092 −0.102*/ −0.19*** 0.126**/ −0.032 −0.081/ −0.078 1.00 Patience −0.019 0.041 −0.020 0.004/ 0.209*** −0.011/ −0.175*** −0.011/ −0.111* 0.000/ −0.042 0.001/ −0.058 −0.042/ 0.013 0.041/ 0.077 −0.112*/ −0.146** 0.051/ −0.076 0.088/ 0.003 −0.015/ 0.098 0.022/ 0.003 0.035/ −0.086 1.00 Indonesia/ Ghana . Logarithmic population . Employment status . Toilet . Female- headed Household (HH) . Number of children in HH . Household size . Ownership . Relative HH income . Networks . Sex . Age . Marital status . Education . Unemployed . Migration experience . Risk aversion . Patience . Logarithm population 1.00 Employment −0.324*** 1.00 Toilet 0.436*** 0.241*** 1.00 Female headed 0.077 −0.099* 0.030 1.00 Number of children −0.129** 0.073 −0.018 −0.028/ 0.025** 1.00 Household size −0.054 0.041 0.096* −0.122**/ −0.016 0.322***/ 0.699*** 1.00 Ownership −0.109* −0.017 −0.087 0.007/ −0.18*** 0.062/ 0.048 −0.20***/ −0.031 1.00 Relative HH income −0.000 0.000 0.000 −0.103**/ −0.18*** −0.06/ −0.011 0.113**/ 0.111* −0.001/ 0.044 1.00 Networks 0.064 −0.246*** −0.033 −0.059/ 0.074 −0.18***/ 0.015 −0.038/ −0.096 0.008/ 0.088 0.133**/ 0.032 1.00 Sex 0.066 0.044 0.121** 0.133**/ 0.274*** 0.002/ 0.003 0.002/ 0.037 −0.001/ 0.003 −0.086/ −0.170*** −0.044/ −0.026 1.00 Age 0.019 −0.188*** −0.142** 0.049/ 0.000 0.008/ −0.032 −0.008/ −0.110* 0.434***/ 0.339*** −0.101*/ −0.157*** 0.052/ −0.003 0.016/ 0.088 1.00 Marital status −0.069 0.012 −0.052 0.237***/ −0.267*** 0.352***/ 0.111** 0.352***/ −0.003 0.201***/ 0.191*** 0.024/ −0.174*** 0.052/ −0.014 −0.004/ 0.071 0.186***/ 0.225*** 1.00 Education 0.066 −0.142** −0.110* −0.062/ −0.109* −0.144**/ −0.072 −0.144**/ −0.034 −0.116**/ −0.146** 0.288***/ 0.235*** 0.186***/ 0.112* −0.082/ −0.285*** −0.27***/ −0.314*** 0.076/ −0.094 1.00 Unemployed 0.077 0.012 0.028 0.020/ 0.167*** 0.019/ −0.094 0.019/ −0.024 0.119**/ −0.010 −0.069/ 0.018 −0.046/ −0.105* 0.087/ −0.010 0.166***/ −0.153** −0.081/ −0.325*** −0.099*/ −0.18*** 1.00 Migration experience −0.001 −0.181*** −0.138** −0.010/ −0.05 0.072/ 0.063 0.072/ 0.012 0.045/ 0.065 0.11*/ 0.101* 0.16***/ 0.122* −0.121**/ −0.022 0.166***/ 0.184*** 0.056/ 0.087 0.22***/ 0.063 −0.014/ −0.094 1.00 Risk aversion 0.055 −0.065 −0.047 0.076/ −0.003 −0.056/ 0.089 −0.056/ 0.095 −0.017/ 0.130** −0.041/ −0.119** −0.061/ −0.146** 0.176***/ 0.139** 0.137**/ 0.219*** −0.015/ 0.092 −0.102*/ −0.19*** 0.126**/ −0.032 −0.081/ −0.078 1.00 Patience −0.019 0.041 −0.020 0.004/ 0.209*** −0.011/ −0.175*** −0.011/ −0.111* 0.000/ −0.042 0.001/ −0.058 −0.042/ 0.013 0.041/ 0.077 −0.112*/ −0.146** 0.051/ −0.076 0.088/ 0.003 −0.015/ 0.098 0.022/ 0.003 0.035/ −0.086 1.00 Note: *p < 0.1; **p < 0.05; ***p < 0.01. Table 7. Correlation matrix of covariates Indonesia/ Ghana . Logarithmic population . Employment status . Toilet . Female- headed Household (HH) . Number of children in HH . Household size . Ownership . Relative HH income . Networks . Sex . Age . Marital status . Education . Unemployed . Migration experience . Risk aversion . Patience . Logarithm population 1.00 Employment −0.324*** 1.00 Toilet 0.436*** 0.241*** 1.00 Female headed 0.077 −0.099* 0.030 1.00 Number of children −0.129** 0.073 −0.018 −0.028/ 0.025** 1.00 Household size −0.054 0.041 0.096* −0.122**/ −0.016 0.322***/ 0.699*** 1.00 Ownership −0.109* −0.017 −0.087 0.007/ −0.18*** 0.062/ 0.048 −0.20***/ −0.031 1.00 Relative HH income −0.000 0.000 0.000 −0.103**/ −0.18*** −0.06/ −0.011 0.113**/ 0.111* −0.001/ 0.044 1.00 Networks 0.064 −0.246*** −0.033 −0.059/ 0.074 −0.18***/ 0.015 −0.038/ −0.096 0.008/ 0.088 0.133**/ 0.032 1.00 Sex 0.066 0.044 0.121** 0.133**/ 0.274*** 0.002/ 0.003 0.002/ 0.037 −0.001/ 0.003 −0.086/ −0.170*** −0.044/ −0.026 1.00 Age 0.019 −0.188*** −0.142** 0.049/ 0.000 0.008/ −0.032 −0.008/ −0.110* 0.434***/ 0.339*** −0.101*/ −0.157*** 0.052/ −0.003 0.016/ 0.088 1.00 Marital status −0.069 0.012 −0.052 0.237***/ −0.267*** 0.352***/ 0.111** 0.352***/ −0.003 0.201***/ 0.191*** 0.024/ −0.174*** 0.052/ −0.014 −0.004/ 0.071 0.186***/ 0.225*** 1.00 Education 0.066 −0.142** −0.110* −0.062/ −0.109* −0.144**/ −0.072 −0.144**/ −0.034 −0.116**/ −0.146** 0.288***/ 0.235*** 0.186***/ 0.112* −0.082/ −0.285*** −0.27***/ −0.314*** 0.076/ −0.094 1.00 Unemployed 0.077 0.012 0.028 0.020/ 0.167*** 0.019/ −0.094 0.019/ −0.024 0.119**/ −0.010 −0.069/ 0.018 −0.046/ −0.105* 0.087/ −0.010 0.166***/ −0.153** −0.081/ −0.325*** −0.099*/ −0.18*** 1.00 Migration experience −0.001 −0.181*** −0.138** −0.010/ −0.05 0.072/ 0.063 0.072/ 0.012 0.045/ 0.065 0.11*/ 0.101* 0.16***/ 0.122* −0.121**/ −0.022 0.166***/ 0.184*** 0.056/ 0.087 0.22***/ 0.063 −0.014/ −0.094 1.00 Risk aversion 0.055 −0.065 −0.047 0.076/ −0.003 −0.056/ 0.089 −0.056/ 0.095 −0.017/ 0.130** −0.041/ −0.119** −0.061/ −0.146** 0.176***/ 0.139** 0.137**/ 0.219*** −0.015/ 0.092 −0.102*/ −0.19*** 0.126**/ −0.032 −0.081/ −0.078 1.00 Patience −0.019 0.041 −0.020 0.004/ 0.209*** −0.011/ −0.175*** −0.011/ −0.111* 0.000/ −0.042 0.001/ −0.058 −0.042/ 0.013 0.041/ 0.077 −0.112*/ −0.146** 0.051/ −0.076 0.088/ 0.003 −0.015/ 0.098 0.022/ 0.003 0.035/ −0.086 1.00 Indonesia/ Ghana . Logarithmic population . Employment status . Toilet . Female- headed Household (HH) . Number of children in HH . Household size . Ownership . Relative HH income . Networks . Sex . Age . Marital status . Education . Unemployed . Migration experience . Risk aversion . Patience . Logarithm population 1.00 Employment −0.324*** 1.00 Toilet 0.436*** 0.241*** 1.00 Female headed 0.077 −0.099* 0.030 1.00 Number of children −0.129** 0.073 −0.018 −0.028/ 0.025** 1.00 Household size −0.054 0.041 0.096* −0.122**/ −0.016 0.322***/ 0.699*** 1.00 Ownership −0.109* −0.017 −0.087 0.007/ −0.18*** 0.062/ 0.048 −0.20***/ −0.031 1.00 Relative HH income −0.000 0.000 0.000 −0.103**/ −0.18*** −0.06/ −0.011 0.113**/ 0.111* −0.001/ 0.044 1.00 Networks 0.064 −0.246*** −0.033 −0.059/ 0.074 −0.18***/ 0.015 −0.038/ −0.096 0.008/ 0.088 0.133**/ 0.032 1.00 Sex 0.066 0.044 0.121** 0.133**/ 0.274*** 0.002/ 0.003 0.002/ 0.037 −0.001/ 0.003 −0.086/ −0.170*** −0.044/ −0.026 1.00 Age 0.019 −0.188*** −0.142** 0.049/ 0.000 0.008/ −0.032 −0.008/ −0.110* 0.434***/ 0.339*** −0.101*/ −0.157*** 0.052/ −0.003 0.016/ 0.088 1.00 Marital status −0.069 0.012 −0.052 0.237***/ −0.267*** 0.352***/ 0.111** 0.352***/ −0.003 0.201***/ 0.191*** 0.024/ −0.174*** 0.052/ −0.014 −0.004/ 0.071 0.186***/ 0.225*** 1.00 Education 0.066 −0.142** −0.110* −0.062/ −0.109* −0.144**/ −0.072 −0.144**/ −0.034 −0.116**/ −0.146** 0.288***/ 0.235*** 0.186***/ 0.112* −0.082/ −0.285*** −0.27***/ −0.314*** 0.076/ −0.094 1.00 Unemployed 0.077 0.012 0.028 0.020/ 0.167*** 0.019/ −0.094 0.019/ −0.024 0.119**/ −0.010 −0.069/ 0.018 −0.046/ −0.105* 0.087/ −0.010 0.166***/ −0.153** −0.081/ −0.325*** −0.099*/ −0.18*** 1.00 Migration experience −0.001 −0.181*** −0.138** −0.010/ −0.05 0.072/ 0.063 0.072/ 0.012 0.045/ 0.065 0.11*/ 0.101* 0.16***/ 0.122* −0.121**/ −0.022 0.166***/ 0.184*** 0.056/ 0.087 0.22***/ 0.063 −0.014/ −0.094 1.00 Risk aversion 0.055 −0.065 −0.047 0.076/ −0.003 −0.056/ 0.089 −0.056/ 0.095 −0.017/ 0.130** −0.041/ −0.119** −0.061/ −0.146** 0.176***/ 0.139** 0.137**/ 0.219*** −0.015/ 0.092 −0.102*/ −0.19*** 0.126**/ −0.032 −0.081/ −0.078 1.00 Patience −0.019 0.041 −0.020 0.004/ 0.209*** −0.011/ −0.175*** −0.011/ −0.111* 0.000/ −0.042 0.001/ −0.058 −0.042/ 0.013 0.041/ 0.077 −0.112*/ −0.146** 0.051/ −0.076 0.088/ 0.003 −0.015/ 0.098 0.022/ 0.003 0.035/ −0.086 1.00 Note: *p < 0.1; **p < 0.05; ***p < 0.01. Table 8. Robustness checks, Indonesia . (1) . (2) . (3) . (5) . (6) . (7) . Community fixed effects . No clustering . Reduced set of controls . Exclusion of controls . Household levela . Aggregated index . Alternative specification 1  Flood −0.020 (0.06) −0.034 (0.05) −0.035 (0.04) 0.022 (0.03) 0.010 (0.07) − Alternative specification 2  Erosion −0.104 (0.20) −0.143 (0.16) 0.016 (0.13) 0.058 (0.09) −0.533 (0.45) − Alternative specification 3  Subsidence −0.129 (0.10) −0.083 (0.08) −0.080 (0.06) −0.049 (0.05) −0.095 (0.07) − Alternative specification 4  Distance to coast −0.333 (0.38) −0.262*(0.15) −0.014 (0.10) 0.052 (0.06) 0.007 (0.11) − Alternative specification 5  Hazard −0.542 (0.43) −0.449 (0.45) 0.089 (0.18) 0.105 (0.12) −0.144 (0.37) − Alternative specification 6 Aggregated Environmental change index − − − − − −0.042 (0.04) Control variables Yes Yes Age, age2, unemployed, sex No Yes Yes Community dummies Yes No No No No No . (1) . (2) . (3) . (5) . (6) . (7) . Community fixed effects . No clustering . Reduced set of controls . Exclusion of controls . Household levela . Aggregated index . Alternative specification 1  Flood −0.020 (0.06) −0.034 (0.05) −0.035 (0.04) 0.022 (0.03) 0.010 (0.07) − Alternative specification 2  Erosion −0.104 (0.20) −0.143 (0.16) 0.016 (0.13) 0.058 (0.09) −0.533 (0.45) − Alternative specification 3  Subsidence −0.129 (0.10) −0.083 (0.08) −0.080 (0.06) −0.049 (0.05) −0.095 (0.07) − Alternative specification 4  Distance to coast −0.333 (0.38) −0.262*(0.15) −0.014 (0.10) 0.052 (0.06) 0.007 (0.11) − Alternative specification 5  Hazard −0.542 (0.43) −0.449 (0.45) 0.089 (0.18) 0.105 (0.12) −0.144 (0.37) − Alternative specification 6 Aggregated Environmental change index − − − − − −0.042 (0.04) Control variables Yes Yes Age, age2, unemployed, sex No Yes Yes Community dummies Yes No No No No No Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. aThe dependent variable is migrant household (=1 if migrant in household, =0 if otherwise), individual characteristics and perceptions of environmental events included for household head. Table 8. Robustness checks, Indonesia . (1) . (2) . (3) . (5) . (6) . (7) . Community fixed effects . No clustering . Reduced set of controls . Exclusion of controls . Household levela . Aggregated index . Alternative specification 1  Flood −0.020 (0.06) −0.034 (0.05) −0.035 (0.04) 0.022 (0.03) 0.010 (0.07) − Alternative specification 2  Erosion −0.104 (0.20) −0.143 (0.16) 0.016 (0.13) 0.058 (0.09) −0.533 (0.45) − Alternative specification 3  Subsidence −0.129 (0.10) −0.083 (0.08) −0.080 (0.06) −0.049 (0.05) −0.095 (0.07) − Alternative specification 4  Distance to coast −0.333 (0.38) −0.262*(0.15) −0.014 (0.10) 0.052 (0.06) 0.007 (0.11) − Alternative specification 5  Hazard −0.542 (0.43) −0.449 (0.45) 0.089 (0.18) 0.105 (0.12) −0.144 (0.37) − Alternative specification 6 Aggregated Environmental change index − − − − − −0.042 (0.04) Control variables Yes Yes Age, age2, unemployed, sex No Yes Yes Community dummies Yes No No No No No . (1) . (2) . (3) . (5) . (6) . (7) . Community fixed effects . No clustering . Reduced set of controls . Exclusion of controls . Household levela . Aggregated index . Alternative specification 1  Flood −0.020 (0.06) −0.034 (0.05) −0.035 (0.04) 0.022 (0.03) 0.010 (0.07) − Alternative specification 2  Erosion −0.104 (0.20) −0.143 (0.16) 0.016 (0.13) 0.058 (0.09) −0.533 (0.45) − Alternative specification 3  Subsidence −0.129 (0.10) −0.083 (0.08) −0.080 (0.06) −0.049 (0.05) −0.095 (0.07) − Alternative specification 4  Distance to coast −0.333 (0.38) −0.262*(0.15) −0.014 (0.10) 0.052 (0.06) 0.007 (0.11) − Alternative specification 5  Hazard −0.542 (0.43) −0.449 (0.45) 0.089 (0.18) 0.105 (0.12) −0.144 (0.37) − Alternative specification 6 Aggregated Environmental change index − − − − − −0.042 (0.04) Control variables Yes Yes Age, age2, unemployed, sex No Yes Yes Community dummies Yes No No No No No Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. aThe dependent variable is migrant household (=1 if migrant in household, =0 if otherwise), individual characteristics and perceptions of environmental events included for household head. Table 9. Robustness checks, Ghana . (1) . (2) . (3) . (5) . (6) . (7) . No community fixed effects . No clustering . Reduced set of controls . Exclusion of controls . Household levela . Aggregated index . Alternative specification 1  Flood 0.049 (0.05) 0.027 (0.05) −0.015 (0.04) −0.10 (0.03) 0.039 (0.07) − Alternative specification 2  Erosion 0.009 (0.05) −0.027 (0.06) −0.034 (0.04) −0.032 (0.03) 0.076 (0.07) − Alternative specification 3  Storms 0.142**(0.06) 0.130*(0.07) 0.089*(0.05) 0.066**(0.03) 0.168***(0.06) − Alternative specification 4  Distance to coast −0.015 (0.06) −0.007 (0.61) 0.009 (0.04) 0.036 (0.03) −0.435 (0.46) − Alternative specification 5  Hazard −0.189 (0.15) −0.449 (0.45) −0.080 (0.12) −0.059 (0.08) −0.424 (0.28) − Alternative specification 6 Aggregated Environmental change index − − − − − −0.025 (0.02) Control variables Yes Yes Age, unemployed, sex No Yes Yes Community dummies No Yes No No Yes Yes . (1) . (2) . (3) . (5) . (6) . (7) . No community fixed effects . No clustering . Reduced set of controls . Exclusion of controls . Household levela . Aggregated index . Alternative specification 1  Flood 0.049 (0.05) 0.027 (0.05) −0.015 (0.04) −0.10 (0.03) 0.039 (0.07) − Alternative specification 2  Erosion 0.009 (0.05) −0.027 (0.06) −0.034 (0.04) −0.032 (0.03) 0.076 (0.07) − Alternative specification 3  Storms 0.142**(0.06) 0.130*(0.07) 0.089*(0.05) 0.066**(0.03) 0.168***(0.06) − Alternative specification 4  Distance to coast −0.015 (0.06) −0.007 (0.61) 0.009 (0.04) 0.036 (0.03) −0.435 (0.46) − Alternative specification 5  Hazard −0.189 (0.15) −0.449 (0.45) −0.080 (0.12) −0.059 (0.08) −0.424 (0.28) − Alternative specification 6 Aggregated Environmental change index − − − − − −0.025 (0.02) Control variables Yes Yes Age, unemployed, sex No Yes Yes Community dummies No Yes No No Yes Yes Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. aThe dependent variable is migrant household =1 if migrant in household, =0 if otherwise), individual characteristics and perceptions of environmental events included for household head. Table 9. Robustness checks, Ghana . (1) . (2) . (3) . (5) . (6) . (7) . No community fixed effects . No clustering . Reduced set of controls . Exclusion of controls . Household levela . Aggregated index . Alternative specification 1  Flood 0.049 (0.05) 0.027 (0.05) −0.015 (0.04) −0.10 (0.03) 0.039 (0.07) − Alternative specification 2  Erosion 0.009 (0.05) −0.027 (0.06) −0.034 (0.04) −0.032 (0.03) 0.076 (0.07) − Alternative specification 3  Storms 0.142**(0.06) 0.130*(0.07) 0.089*(0.05) 0.066**(0.03) 0.168***(0.06) − Alternative specification 4  Distance to coast −0.015 (0.06) −0.007 (0.61) 0.009 (0.04) 0.036 (0.03) −0.435 (0.46) − Alternative specification 5  Hazard −0.189 (0.15) −0.449 (0.45) −0.080 (0.12) −0.059 (0.08) −0.424 (0.28) − Alternative specification 6 Aggregated Environmental change index − − − − − −0.025 (0.02) Control variables Yes Yes Age, unemployed, sex No Yes Yes Community dummies No Yes No No Yes Yes . (1) . (2) . (3) . (5) . (6) . (7) . No community fixed effects . No clustering . Reduced set of controls . Exclusion of controls . Household levela . Aggregated index . Alternative specification 1  Flood 0.049 (0.05) 0.027 (0.05) −0.015 (0.04) −0.10 (0.03) 0.039 (0.07) − Alternative specification 2  Erosion 0.009 (0.05) −0.027 (0.06) −0.034 (0.04) −0.032 (0.03) 0.076 (0.07) − Alternative specification 3  Storms 0.142**(0.06) 0.130*(0.07) 0.089*(0.05) 0.066**(0.03) 0.168***(0.06) − Alternative specification 4  Distance to coast −0.015 (0.06) −0.007 (0.61) 0.009 (0.04) 0.036 (0.03) −0.435 (0.46) − Alternative specification 5  Hazard −0.189 (0.15) −0.449 (0.45) −0.080 (0.12) −0.059 (0.08) −0.424 (0.28) − Alternative specification 6 Aggregated Environmental change index − − − − − −0.025 (0.02) Control variables Yes Yes Age, unemployed, sex No Yes Yes Community dummies No Yes No No Yes Yes Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. aThe dependent variable is migrant household =1 if migrant in household, =0 if otherwise), individual characteristics and perceptions of environmental events included for household head. Furthermore, it is tested whether the effect of environmental factors is the same for different distances. Thus, a multinomial logit regression is used to estimate the impact of environmental factors on the probability of moves within the region or moves out of the region (relative to staying). Still, no robust significant effect can be detected in the Indonesian case study. However, it reveals that the perception of storms is only significant for rather long-distance internal moves, while it has no effect on moves within the region of origin (see Tables 10 and 11). Table 10. Multinomial logit: different distances, Indonesia . Migration distance . (1) . (2) . (3) . (4) . (5) . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Flood In region −0.047 (0.05) Out of region −0.06 (0.05) Erosion In region −0.022 (0.19) Out of region −0.058 (0.17) Subsidence In region −0.085 (0.08) Out of region −0.016 (0.05) Distance to coast In region −0.233 (0.20) Out of region −0.155 (0.14) Hazard In region −0.455 (0.44) Out of region 0.244 (0.26) Full set of control variables Yes Yes Yes Yes Yes BIC 1042.398 1022.666 1029.510 1029.673 1044.676 AIC 893.324 874.648 881.359 880.599 895.602 Pseudo R2 0.415 0.417 0.414 0.424 0.414 N 307 299 300 309 309 . Migration distance . (1) . (2) . (3) . (4) . (5) . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Flood In region −0.047 (0.05) Out of region −0.06 (0.05) Erosion In region −0.022 (0.19) Out of region −0.058 (0.17) Subsidence In region −0.085 (0.08) Out of region −0.016 (0.05) Distance to coast In region −0.233 (0.20) Out of region −0.155 (0.14) Hazard In region −0.455 (0.44) Out of region 0.244 (0.26) Full set of control variables Yes Yes Yes Yes Yes BIC 1042.398 1022.666 1029.510 1029.673 1044.676 AIC 893.324 874.648 881.359 880.599 895.602 Pseudo R2 0.415 0.417 0.414 0.424 0.414 N 307 299 300 309 309 Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. Table 10. Multinomial logit: different distances, Indonesia . Migration distance . (1) . (2) . (3) . (4) . (5) . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Flood In region −0.047 (0.05) Out of region −0.06 (0.05) Erosion In region −0.022 (0.19) Out of region −0.058 (0.17) Subsidence In region −0.085 (0.08) Out of region −0.016 (0.05) Distance to coast In region −0.233 (0.20) Out of region −0.155 (0.14) Hazard In region −0.455 (0.44) Out of region 0.244 (0.26) Full set of control variables Yes Yes Yes Yes Yes BIC 1042.398 1022.666 1029.510 1029.673 1044.676 AIC 893.324 874.648 881.359 880.599 895.602 Pseudo R2 0.415 0.417 0.414 0.424 0.414 N 307 299 300 309 309 . Migration distance . (1) . (2) . (3) . (4) . (5) . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Flood In region −0.047 (0.05) Out of region −0.06 (0.05) Erosion In region −0.022 (0.19) Out of region −0.058 (0.17) Subsidence In region −0.085 (0.08) Out of region −0.016 (0.05) Distance to coast In region −0.233 (0.20) Out of region −0.155 (0.14) Hazard In region −0.455 (0.44) Out of region 0.244 (0.26) Full set of control variables Yes Yes Yes Yes Yes BIC 1042.398 1022.666 1029.510 1029.673 1044.676 AIC 893.324 874.648 881.359 880.599 895.602 Pseudo R2 0.415 0.417 0.414 0.424 0.414 N 307 299 300 309 309 Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. Table 11. Multinomial logit: different distances, Ghana . . (1) . (2) . (3) . (4) . (5) . . Migration distance . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Flood In region −0.046 (0.05) Out of region −0.428 (0.32) Erosion In region −0.093 (0.06) Out of region −0.044 (0.05) Storm In region 0.025 (0.07) Out of region 0.093**(0.04) Distance to coast In region −0.085 (0.10) Out of region 0.051 (0.05) Hazard In region −0.060 (0.20) Out of region −0.098 (0.13) Full set of control variables Yes Yes Yes Yes Yes BIC 1069.779 1061.697 1064.327 1066.106 696.739 AIC 910.322 902.240 904.870 906.649 583.013 Pseudo R2 0.326 0.333 0.331 0.329 0.345 N 277 277 277 277 174 . . (1) . (2) . (3) . (4) . (5) . . Migration distance . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Flood In region −0.046 (0.05) Out of region −0.428 (0.32) Erosion In region −0.093 (0.06) Out of region −0.044 (0.05) Storm In region 0.025 (0.07) Out of region 0.093**(0.04) Distance to coast In region −0.085 (0.10) Out of region 0.051 (0.05) Hazard In region −0.060 (0.20) Out of region −0.098 (0.13) Full set of control variables Yes Yes Yes Yes Yes BIC 1069.779 1061.697 1064.327 1066.106 696.739 AIC 910.322 902.240 904.870 906.649 583.013 Pseudo R2 0.326 0.333 0.331 0.329 0.345 N 277 277 277 277 174 Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. Table 11. Multinomial logit: different distances, Ghana . . (1) . (2) . (3) . (4) . (5) . . Migration distance . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Flood In region −0.046 (0.05) Out of region −0.428 (0.32) Erosion In region −0.093 (0.06) Out of region −0.044 (0.05) Storm In region 0.025 (0.07) Out of region 0.093**(0.04) Distance to coast In region −0.085 (0.10) Out of region 0.051 (0.05) Hazard In region −0.060 (0.20) Out of region −0.098 (0.13) Full set of control variables Yes Yes Yes Yes Yes BIC 1069.779 1061.697 1064.327 1066.106 696.739 AIC 910.322 902.240 904.870 906.649 583.013 Pseudo R2 0.326 0.333 0.331 0.329 0.345 N 277 277 277 277 174 . . (1) . (2) . (3) . (4) . (5) . . Migration distance . Migrant status . Migrant status . Migrant status . Migrant status . Migrant status . Flood In region −0.046 (0.05) Out of region −0.428 (0.32) Erosion In region −0.093 (0.06) Out of region −0.044 (0.05) Storm In region 0.025 (0.07) Out of region 0.093**(0.04) Distance to coast In region −0.085 (0.10) Out of region 0.051 (0.05) Hazard In region −0.060 (0.20) Out of region −0.098 (0.13) Full set of control variables Yes Yes Yes Yes Yes BIC 1069.779 1061.697 1064.327 1066.106 696.739 AIC 910.322 902.240 904.870 906.649 583.013 Pseudo R2 0.326 0.333 0.331 0.329 0.345 N 277 277 277 277 174 Note: The dependent variable is migrant status. Robust standard errors in parenthesis. * p < 0.1; ** p < 0.05; *** p < 0.01. 4.4 Interaction effects Since environmental factors were found to play a limited direct role, an additional set of exploratory models tested whether the effect of the environment depends on other factors and, thus, whether the environment might only lead to migration in certain contexts. However, only few potential interactions could be detected. In the Indonesian case two conditional effects could be found. First, the community’s level of hazard has a negative impact on out-migration for male respondents, while no effect can be found for female respondents (see Figure 4).20 Second, a highly significant interaction between distance to coast and networks was found. The marginal effect of the distance to the coast decreases with the improvement of networks, i.e. living closer to the coast increases the probability of out-migration for individuals with better networks, while it has no effect on individuals with no or small networks (see Figure 5). Figure 4. Open in new tabDownload slide Marginal effect of hazard on probability of migrating by sex of respondent, Indonesia. Figure 4. Open in new tabDownload slide Marginal effect of hazard on probability of migrating by sex of respondent, Indonesia. Figure 5. Open in new tabDownload slide Marginal effect of distance to coast on probability of migrating by network of respondent, Indonesia. Figure 5. Open in new tabDownload slide Marginal effect of distance to coast on probability of migrating by network of respondent, Indonesia. In Ghana it is found that the effect of individual storm perceptions does also depend on the number of children in the household. The marginal effect of storms on out-migration increases with every additional child in the household. While it is 0 for someone from a childless household and, thus, has no effect on the individual’s out-migration decision, it increases to 0.078 for someone from a household with eight children. Same relation can be found for the number of household members, since this indicator is highly correlated with the number of children in the household (see Figures 6 and 7). These results indicate that there might be a conditional effect of the environment on migration. However, these few interactive effects seem to be very context-specific and could only be found for one of the environmental variables in one of the study regions. Nevertheless, these results could also be seen as an indicator for future in-depth research. Figure 6. Open in new tabDownload slide Marginal effect of storm perception on probability of migrating by number of children, Ghana. Figure 6. Open in new tabDownload slide Marginal effect of storm perception on probability of migrating by number of children, Ghana. Figure 7. Open in new tabDownload slide Marginal effect of storm perception on probability of migrating by number of household members, Ghana. Figure 7. Open in new tabDownload slide Marginal effect of storm perception on probability of migrating by number of household members, Ghana. 4.5 Environmentally induced economic migration? Even though no robust, generalizable impact of environmental characteristics on migration decisions could be detected, the environment could still impact indirectly. Afifi (2011) therefore introduces the term ‘environmentally induced economic migration’ and argues that ‘the economic factor can act as the mechanism through which environmental degradation leads to migration’ (p. 100). It is assumed that environmental conditions impact on economic outcomes like food security, income, and employment—especially in rural developing countries. Using Sobel–Goodman tests, it is tested whether economic conditions like personal income or unemployment status act as a mediator of the relationship between environmental change and out-migration. These tests estimate the effect of the environmental variables on migration and on the economic factors, unemployment, and income, as well as their effects on migration and test whether there is a potential indirect link. In the Indonesian case, these tests cannot detect any significant indirect effect. Only one significant link between an environmental variable and an economic outcome (path a in Figure 8) was found: respondents affected by erosion are earning significantly more than those less affected by erosion processes. While this link might not be causal, it does also not translate into out-migration (path b). However, this absence of indirect effects is not very surprising, since Semarang is a thriving industrialized city with many factories, a big harbor, and good work opportunities, especially in the coastal areas. Figure 8. Open in new tabDownload slide Indirect effect of environmental change on migration through economic factors. Source: Author’s representation, based on Preacher and Hayes (2008). Figure 8. Open in new tabDownload slide Indirect effect of environmental change on migration through economic factors. Source: Author’s representation, based on Preacher and Hayes (2008). Nevertheless, also in the more rural and economically weaker region in Ghana, the Sobel–Goodman test cannot detect any indirect effect of coastal changes on out-migration.21 Again only one significant link between environmental change and economic factors was found: respondents living further away from the coast earn more than those closer to the coast. One explanation for this might be that richer households can afford property which is further inland and thus less threatened by coastal changes. But just as in the Indonesia case, this link does not translate into a significant correlation to out-migration and no indirect effects can be found. This might be due to the very special case of gradual coastal changes. In contrast to agricultural output depending on rainfall or soil quality, or fish catch depending on the health of ecosystems, economic activities of respondents in this study do not depend much on experienced coastal degradation. Thus, it is not very unexpected that no indirect effect of the environment can be found in these case studies. 5. Discussion Various scholars as well as public institutions have suggested that environmental change could result in the migration of millions of people, especially in coastal areas. Migration on such a massive scale would be a challenge for both sending and receiving regions—as current debates about migration flows in Europe show. However, empirical evidence for such environmentally induced migration is mixed. This study sought to contribute to the discussion by using a multilevel model where individual-, household-, and community-level factors are all simultaneously considered to isolate the net effect of slow- and sudden-onset environmental factors. Due to the lack of available high-quality census data, both migrants and non-migrants of households located in the two coastal study regions have been interviewed directly. When interpreting the following results, it should be kept in mind that this approach does not include households which have moved as a whole. In summary, it can be stated that the data explain individual migration behavior of the sampled respondents quite well. In both regions, age and the employment status of the respondent were found to be especially important which emphasizes the influence of individual characteristics. But also other variables like risk aversion, patience, and the number of children in the household in general; networks, the community’s employment situation, and marital status in Indonesia; and migration experience, education, household size, and relative household income in Ghana help in explaining who migrates and who stays, and support different economic migration theories. Taken together, no generalizable direct link between the main coastal events and out-migration could be detected. Only one region-specific environmental event—perceptions of storms in Ghana—turned out to have a robust and direct impact on the respondents’ decision to migrate. This result indicates that the more an individual perceives herself as affected by storms, the higher is the probability of out-migration. Further tests showed that this finding holds only for moves out of the region. This finding is not very surprising, since storms, unlike erosion, do not only affect particular sections of the shoreline. Looking at the nature of the included environmental events, it has to be noted again that most of them can be considered as rather long-term in nature with a limited geographical scope. Both regions experience severe erosion processes; nevertheless, erosion is still a rather gradual and predictable process. The same can be said about subsidence in Semarang which can be clearly anticipated by affected households, since rates of yearly subsidence do not change much.22 Also the floods experienced in both regions are mostly tidal floods and not severe sudden-onset flooding. Respondents perceive them as less severe, since streets and houses are regularly inundated for a shorter period of time without threatening health or lives. Inhabitants of both regions are used to floods and see them as part of daily life.23 Erosion and subsidence have also been experienced for several decades already with the result that their impacts are not new for respondents in those regions. Furthermore, the great majority of people in Semarang adapts to constant subsidence and the concomitant inundation threats by lifting houses, floors and valuables, building drainages, and similar. Even though less adaptation strategies at the household level have been observed in Keta, mostly because the eroding coastline cannot efficiently be stopped by single households, households whose house got eaten by the sea tend to rebuild their houses in the neighborhood—sometimes knowing that the newly built house will also only last for several years before the coastline has also reached the new houses. Altogether, the costs associated with coping with and adapting to those slow-onset environmental changes might be lower than those associated with migration, which include, for example transport, psychic and social costs, and uncertainties about economic success of the migration. Looking at the only statistically significant environmental event, storms24, however, it can be stated that they hit the communities unexpectedly and with greater power, destroying buildings, roofs, and boats, making it difficult for fishermen to fish. Storms are only a problem in Ghana and, thus, there is no comparable sudden-onset measure for the Indonesian case study. Nevertheless, it is very likely that the effect of environmental events strongly depends on the nature of these events. Long-term, gradual changes like sea-level rise, erosion, and land subsidence were not found to increase the likelihood of out-migration, while sudden environmental events could be more likely to induce migration. Overall, however, no convincing evidence for a general direct impact of environmental change on migration decisions could be found. Regardless of the nature of the environmental event, its effect on out-migration might still be either moderated or mediated. In Indonesia the effects of two of the environmental events get moderated by the gender and network of respondents. There is no effect of the community’s level of hazard on out-migration of women, while a higher hazard is found to lead to less out-migration of men. This finding indicates that the likelihood of male out-migration is smaller in communities with high hazards than in those with lower. When further looking at high-hazard communities, it is found that a significantly higher proportion of the population is employed than in low-hazard communities. Thus, the reduced out-migration of men who are the main breadwinner in Indonesian households might not be due to the hazard itself, but to the increased employment opportunities. This effect might not be found for the female subpopulation because of different gender roles. Additionally, people living closer to the coast are more likely to leave if they have good networks, which emphasizes the crucial effect of networks, especially for the more vulnerable coastal population. In Ghana, the number of children seems to act as a moderating factor on the effect of storm perceptions on migration decisions. While someone’s migration decision is not affected by storm perceptions if she is from a childless household, someone from a child-rich household will be more likely to move when affected by storms. This could be attributed to the fact that storms might seem less dangerous for households which are not responsible for children. Childless households might be better able to cope with the immediate consequences of storms. While the effects of environmental change on out-migration might be moderated and context-dependent in some cases, no mediating economic factors have been found which is not much surprising, since economic activities of the respondents do not depend much on the considered coastal changes and, thus, cannot be compared to environmental changes like droughts in agricultural regions. 6. Conclusion The results of this study have relevant implications for environment–migration theories, for future research in this field and for policies in the two study regions. Regarding environment–migration theories, the study’s findings indicate that there is no generalizable direct effect of environmental change on out-migration—especially when looking at slow-onset coastal changes. This finding highlights the importance of contexts in environment–migration relationships and suggests that, if the past can be used to predict future scenarios, then predictions of large-scale displacements are most likely exaggerated. With regard to research methods, this study improves common methods by directly interviewing people who left regions affected by coastal hazards to get personal views on perception and preferences. It also extends the number of studies which have used quantitative multilevel methods to estimate the influence of environmental change on individual migration (Henry et al. 2004; Gray 2011). This approach aims at providing a generalizable methodology in a field ‘where sophisticated empirical applications have lagged significantly behind the high level of interest by academics and policy makers’ (Gray 2011). However, more differentiated analyses are needed to test whether there are differences in the impact of sudden-onset and more gradual environmental events. A high-quality longitudinal data set, preferably collected through quantitative panel studies, which include also households which move as a whole, would additionally help to get a clearer picture of the environment–migration nexus. Furthermore, greater comparability in measurement of the variables of interest would be desirable to compare case studies from a wider range of contexts and, thus, make stronger generalizations. McLeman (2013) even calls for the integration of environmental data and common standards into official censuses, since most environmentally induced migration is likely to occur internally. Such an implementation could improve further research on the environment–migration nexus and allow for further theoretical, methodological, and empirical improvements. Finally, these findings have implications for policies in both study countries. Some evidence suggests that networks in Semarang moderate the effect of environmental change, leading to the conclusion that improving information and institutional support in affected areas might help those who are willing to leave but do not have helpful network ties. Improving economic situations or offering alternative livelihoods to those affected might benefit vulnerable child-rich households without resulting in rural depopulation. However, evidence for these moderating or mediating factors is weak to non-existent, and overall findings indicate that most of the people prefer not to migrate when facing longer-term gradual environmental problems, but to use other forms of adaptation and rather migrate due to more individual or economic reasons. Therefore, it is critical that policies get implemented which do not only take into account that migration ‘might’ occur as consequence of coastal changes but which promote adaptation to environmental change and increase the resilience of coastal populations. Thus, aid could be more targeted to areas affected by environmental changes to promote adaptation on individual and community level. Funding This work was supported by Volkswagen Foundation. Footnotes 1 " At the same time, also research on environmental change has completely ignored the role of migration (Black et al. 2011). 2 " Foresight (2011) shows that also most other estimates of the number of environmental migrants base on Norman Myers’ (1993, 2002) methodology or estimates (so for example the numbers of Christian Aid (2007), Stern (2007) and Friends of The Earth (2007). Piguet (2010) calls these numbers ‘nothing but the rule of the thumb’ (p.517) and the IPCC (2007) labels them as ‘at best, guesswork’ (p. 365). 3 " Moriniere (2009) reviewed the environment–migration literature, consisting of 321 publications and found only two articles in which the researchers used quantitative multivariate methods to examine the effect of environmental factors on out-migration. Laczko and Aghazarm (2009) argue that in fact there have been few more quantitative papers. Nevertheless, they criticize that most of those few studies use rainfall data to investigate the link. The few studies which have focused on other environmental factors are criticized for having clear measurement problems by only using very subjective environmental variables (Ezra and Kiros 2001). 4 " The International Organization for Migration emphasizes in its report ‘Migration, Environment and Climate Change: Assessing the evidence’ that now ‘collecting data on households in rural areas is fundamentally important, since households are the major decision makers about […] migration’ (Laczko and Aghazarm 2009, p. 175). 5 " Dispendukcapil.semarangkota.go.id (2016). Available at: http://dispendukcapil.semarangkota.go.id/statistik/jumlah-penduduk-kota-semarang/2016-04-17 [accessed 31 May 2016]. 6 " Even though communities share the same regional coastline, not every community is affected by coastal changes. In Ghana, few communities are protected by a sea defense and therefore even experiencing accretion instead of erosion. In Indonesia, some communities are located well above the sea level not facing the previously mentioned threats. 7 " To be more precise: Volta Region in Ghana, and Central Java in Indonesia. 8 " The number of children in the household is expected to be negatively correlated with the decision to migrate, since parents and other household members are needed to help with raising the children. The household size, however, is seen as an indicator for household-level labor abundance and expected to be positively correlated with migration propensities (Ackah and Medvedev 2010). 9 " Also other household-level variables like individual or household income in absolute terms do never turn out significant when included. It is not included here due to its correlation with relative household income. 10 " Classification of models calculated without p-weights. 11 " The pseudo R2 cannot be interpreted as the common ordinary Least Squares R2, but nevertheless, higher values of R2 indicate a better model fit. 12 " In specification (1), 83.71% of cases get correctly predicted. This may seem impressive; however, it does not tell anything about the proportion of correctly classified cases beyond the number that would be correctly guessed by choosing the most frequent outcome. Since 212 of 308 respondents are non-migrants, just by chance 68.61% of outcomes would be predicted correctly. Thus, White (2013) recommends using this information and calculating the proportional reduction in error, which is reported in Table 5 as well, and which shows that specification (1) reduces the error by 48.10%. 13 " Please note that age is not included as a quadratic term in the Ghanaian case because no non-linearity could be detected. Tests show that the inclusion of age2 would not improve the goodness-of-fit of the model. 14 " Dropped since sample does not contain migrants who owned a house. 15 " Classification of models calculated without p-weights. 16 " Less observations since hazard data are not available for every community. 17 " While it is easy for a migrant to recall how many children or what kind of occupation she had at the time of migration, it is difficult or confusing to recall how willing she was to take risks or how patient she was at that time. 18 " Nevertheless, studies suggest that preferences are rather stable over time and not affected by major life events like migration (Andersen et al. 2008; Conroy 2009) 19 " However, perceptions of respondents from the same household might still differ due to different personal experiences, different coping mentalities, and/ or different recalling of scope and intensity of the event. 20 " Marginal effect of hazard for male respondents is − 0.115, while the one for female is around 0.02. 21 " P-value of indirect effect of the environment through unemployment status: 0.096. We are aware that these p-values are only accurate estimates of the true p-values if the sample size is big enough (and thus the term a*b is normally distributed). Since it is difficult to test whether the sample is big enough, we re-estimate the indirect effects by using bootstrapping which does not rely on distributional assumptions (Preacher and Hayes 2008). Ultimately, the p-value decreases to 0.104 which leads to the conclusion that there is no indirect effect. 22 " The observed erosion in Ghana and subsidence in Indonesia are occurring at very fast rates (several centimeters per year) compared to erosion or subsidence somewhere else. Nevertheless, compared to all potential environmental events, these are very gradual and well-known processes which can be anticipated by inhabitants long before. 23 " ‘We are fishermen, we are used to water. I just walk through’ (Respondent in community Kedzi (Keta, Ghana), when asked about flood problems, 5 October 2015) 24 " Please note that by ‘storm’ Ghanaians understand strong winds in combination with heavy tidal waves, whereas flood can be understood as any form of inundation, regardless of the cause (often rain or tidal floods inundate the streets). Acknowledgments The author greatly appreciates the constant support provided by Achim Schlüter and would also like to thank the anonymous reviewers for their valuable comments and suggestions. Special thanks also to Micaela Kulesz for her support in the econometric analysis of the data and to all members of the project ‘New Regional Formations: rapid environmental change and migration in coastal areas in Ghana and Indonesia’ for valuable support and discussion. Likewise, the author wants to thank Mariama Awumbila, Joseph Teye, and Gertrude Dzifa Torvikey from the University of Ghana, Accra, and Muh Aris Marfai from the University Gadjah Mada, Yogyakarta, for their assistance in the field. The author is also very grateful to the team of students for their excellent assistance in conducting the survey. Most importantly, the author thanks all the Indonesian and Ghanaian respondents who patiently participated in the interviews because without their help this research would not have been possible. References Ackah C. , Medvedev D.. ( 2010 ), Internal Migration in Ghana: Determinants and Welfare Impacts , The World Bank , Washington, DC . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Afifi T. ( 2011 ), “ Economic or Environmental Migration? The Push Factors in Niger ”, International Migration 49 , 95 – 124 . Google Scholar Crossref Search ADS PubMed WorldCat Akyeampong E. K. ( 2001 ), Between the Sea and the Lagoon: An Eco-social History of the Anlo of Southeastern Ghana; c. 1850 to Recent Times , Ohio University Press , Athens, OH . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Andersen S. , Harrison G. W., Lau M., Rutström E. E.. ( 2008 ), “ Lost in State Space: Are Preferences Stable? ”, International Economic Review 49 , 1091 – 112 . Google Scholar Crossref Search ADS WorldCat Black R. ( 2001 ), “Environmental Refugees: Myth or Reality?”, New Issues in Refugee Research, Working Paper No.34, United Nations High Commissioner for Refugees, Geneva. Black R , et al. ( 2011 ), “ The Effect of Environmental Change on Human Migration ”, Global Environmental Change 21 , 3 – 11 . Google Scholar Crossref Search ADS WorldCat Boateng I. ( 2012 ), “ An Assessment of the Physical Impacts of Sea-level Rise and Coastal Adaptation: A Case Study of the Eastern Coast of Ghana”, Climate Change 114 , 273 – 93 . Google Scholar Crossref Search ADS WorldCat Brown O. ( 2008 ), “ The Numbers Game ”, Forced Migration Review 31 , 8 – 9 . OpenURL Placeholder Text WorldCat Castles S. ( 2002 ), “Environmental Change and Forced Migration: Making Sense of the Debate”, Working Paper No. 70, UNHCR Refugee Studies Centre, Geneva. Christian Aid ( 2007 ), Human Tide: The Real Migration Crisis [Online], http://www.christianaid.org.uk/Images/human-tide.pdf. Clarke K. ( 2005 ), “ The Phantom Menace: Omitted Variable Bias in Econometric Research ”, Conflict Management and Peace Science 22 , 341 – 52 . Google Scholar Crossref Search ADS WorldCat Conroy H. ( 2009 ), Risk Aversion, Income Variability, and Migration in Rural Mexico [Online], www.webmeets.com/files/papers/LACEA-LAMES/2009/393/ HectorVConroyLACEA09.pdf El-Hinnawi E. ( 1985 ), Environmental Refugees , United Nations Environmental Programme , Nairobi, Kenya . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Ezra M. , Kiros G. E.. ( 2000 ), “ Household Vulnerability to Food Crisis and Mortality in the Drought-Prone Areas of Northern Ethiopia ”, Journal of Biosocial Science 32 , 395 – 409 . Google Scholar Crossref Search ADS PubMed WorldCat Ezra M. , Kiros G.-E.. ( 2001 ), “ Rural Out-migration in the Drought Prone Areas of Ethiopia: A Multilevel Analysis ”, International Migration Review 35 , 749 – 771 . Google Scholar Crossref Search ADS PubMed WorldCat Foresight ( 2011 ), “Migration and Global Environmental Change”, Final Project Report, The Government Office for Science, London. Friends of The Earth ( 2007 ), A Citizen’s Guide to Climate Refugees [Online], http://www.safecom.org.au/pdfs/FOE_climate_citizens-guide.pdf. Gray C. ( 2011 ), “ Soil Quality and Human Migration in Kenya and Uganda ”, Global Environmental Change 21 , 421 – 30 . Google Scholar Crossref Search ADS PubMed WorldCat Gray C. ( 2009 ), “ Environment, Land, and Rural Out-migration in the Southern Ecuadorian Andes ”, World Development 37 , 457 – 68 . Google Scholar Crossref Search ADS WorldCat Gray C. , Bilsborrow R.. ( 2013 ), “ Environmental Influences on Human Migration in Rural Ecuador ”, Demography 50 , 1217 – 41 . Google Scholar Crossref Search ADS PubMed WorldCat Gray C. , Müller V.. ( 2012 ), “ Drought and Population Mobility in Rural Ethiopia ”, World Development 40 , 134 – 45 . Google Scholar Crossref Search ADS PubMed WorldCat Halliday T. ( 2006 ), “ Migration, Risk, and Liquidity Constraints in El Salvador ”, Economic Development and Cultural Change 54 , 893 – 925 . Google Scholar Crossref Search ADS WorldCat Harwitasari D. , van Ast J. A.. ( 2009 ), Adaptation Responses to Tidal Flooding in Semarang, Indonesia , Erasmus University , Rotterdam . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Henry S. , Boyle P., Lambin E. F.. ( 2003 ), “ Modelling Inter-provincial Migration in Burkina Faso, West Africa: The Role of Socio-demographic and Environmental Factors ”, Applied Geography 23 , 115 – 36 . Google Scholar Crossref Search ADS WorldCat Henry S. , Schoumaker B., Beauchemin C.. ( 2004 ), “ The Impact of Rainfall on the First Out-migration: A Multi-level Event-history Analysis in Burkina Faso ”, Population and Environment 25 , 423 – 60 . Google Scholar Crossref Search ADS WorldCat Henry, S. ( 2006 ), Some questions on the migration-environment relationship, Panel Contribution to the Population-Environment Research Network Cyberseminar on Rural Household Micro-Demographics, Livelihood and the Environment [Online], http://www.ciesin.columbia.edu/repository/pern/papers/Henry_statement.pdf (last accessed 15 March 2015). Hugo G. ( 2008 ), “International Migration in Indonesia and its Impacts on Regional Development”, in: Van Naersen T., Spaan E., Zoomers A., eds, Global Migration and Development , Routledge , New York, NY , pp. 43 – 65 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Hunter L. ( 2005 ), “ Migration and Environmental Hazards ”, Population and Environment 26 , 273 – 302 . Google Scholar Crossref Search ADS PubMed WorldCat IPCC ( 1990 ), Climate Change. The IPCC Scientific Assessment, https://www.ipcc.ch/ipccreports/far/wg_I/ipcc_far_wg_I_full_report.pdf. IPCC ( 2007 ), Climate change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change , in Parry L., Canziani O. F., Palutikof J. P., van der Linden P. J., Hanson C. E., eds, Cambridge University Press , Cambridge; New York . Jónsson G. ( 2010 ), “The Environmental Factor in Migration Dynamics: A Review of African Case Studies”, Working Paper Series – Paper 21, International Migration Institute, University of Oxford, Oxford. Knaap G. ( 2015 ), “Semarang, a Colonial Provincial Capital and PortCity in Java, c.1775” in Bosma U., Webster A., eds, Commodities, Ports and Asian Maritime Trade Since 1750 , Palgrave Macmillan, UK , pp. 79 – 95 . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Kniveton D. , Smith C., Wood S.. ( 2011 ), “ Agent-based Model Simulations of Future Changes in Migration Flows for Burkina Faso ”, Global Environmental Change 21 , 34 – 40 . Google Scholar Crossref Search ADS WorldCat Koubi V. , Stoll S., Spilker G.. ( 2016 ), “ Perceptions of Environmental Change and Migration Decisions ”, Climatic Change 138 , 439 – 51 . Google Scholar Crossref Search ADS WorldCat Laczko F. , Aghazarm C., eds ( 2009 ), Migration, Environment and Climate Change: Assessing the Evidence , International Organization for Migration , Geneva . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Marfai M. A. , King L.. ( 2008 ), “ Tidal Inundation Mapping Under Enhanced Land Subsidence in Semarang, Central Java Indonesia ”, Nat Hazards 44 , 93 – 109 . Google Scholar Crossref Search ADS WorldCat Marfai M. A. , Lorenz K., Junun S., Sudrajat S., Rahayu B. S., Fajar Y.. ( 2008 ), “ The Impact of Tidal Flooding on a Coastal Community in Semarang, Indonesia ”, Environmentalist 28 , 237 – 48 . Google Scholar Crossref Search ADS WorldCat McLeman R. ( 2006 ), “ Migration Out of 1930s Rural Eastern Oklahoma: Insights for Climate Change Research ”, Great Plains Quarterly , 26 . OpenURL Placeholder Text WorldCat McLeman R. ( 2013 ), “ Developments in Modelling of Climate Change-related Migration ”, Climatic Change 117 , 599 – 611 . Google Scholar Crossref Search ADS WorldCat McLeman R. , Smit B.. ( 2006 ), “ Migration as Adaptation to Climate Change ”, Climatic Change 76 , 31 – 53 . Google Scholar Crossref Search ADS WorldCat McLeman R. , Hunter L. M.. ( 2010 ), “ Migration in the Context of Vulnerability and Adaptation to Climate Change: Insights from Analogues ”, Wiley Interdisciplinary Reviews: Climate Change 1 , 450 – 61 . Google Scholar Crossref Search ADS PubMed WorldCat Morinière L. ( 2009 ), “Tracing the Footprint of ‘Environmental Migrants’ through 50 Years of Literature”, in Linking Environmental Change, Migration and Social Vulnerability . pp. 22 – 30 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Mortreux C. , Barnett J.. ( 2009 ), “ Climate Change, Migration and Adaptation in Funafuti, Tuvalu ”, Global Environmental Change 19 , 105 – 12 . Google Scholar Crossref Search ADS WorldCat Morrissey J. ( 2009 ), “Environmental Change and Forced Migration: A State of the Art Review”, Background Paper, Refugee Study Centre, University of Oxford, Oxford. Myers N. ( 1993 ), “ Environmental Refugees in a Globally Warmed World ”, Bioscience 43 , 752 – 61 . Google Scholar Crossref Search ADS WorldCat Myers N. ( 1997 ), “ Environmental Refugees ”, Population and Environment 19 , 167 – 82 . Google Scholar Crossref Search ADS WorldCat Myers N. ( 2002 ), “ Environmental Refugees: A Growing Phenomenon of the 21st Century ”, Philosophical Transactions of the Royal Society of London, Series B: Biological Sciences 357 , 609 – 13 . Google Scholar Crossref Search ADS WorldCat Nairn R. B. ( 2001 ), “ Coastal Erosion at Keta Lagoon, Ghana – Large Scale Solution to a Large Scale Problem ”, Coastal Engineering Proceedings 1 , 3192 – 3205 . OpenURL Placeholder Text WorldCat Obokata R. , Veronis L., McLeman R.. ( 2014 ), “ Empirical Research on International Migration: A Systematic Review ”, Population and Environment 36 , 111 – 35 . Google Scholar Crossref Search ADS PubMed WorldCat Piguet E. ( 2010 ), “ Linking Climate Change, Environmental Degradation, and Migration: A Methodological Overview ”, Climate Change 1 , 517 – 24 . OpenURL Placeholder Text WorldCat Preacher K. J. , Hayes A. F.. ( 2008 ), “ Asymptotic and Resampling Strategies for Assessing and Comparing Indirect Effects in Multiple Mediator Models ”, Behavior Research Methods 40 , 879 – 91 . Google Scholar Crossref Search ADS PubMed WorldCat Stark O. ( 1991 ), The Migration of Labor , Basil Blackwell , Cambridge . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Stark O. , Bloom D. E.. ( 1985 ), “ The New Economics of Labor Migration ”, American Economic Review 75 , 173 – 8 . OpenURL Placeholder Text WorldCat Stern N. H. ( 2007 ), The Stern Review of the Economics of Climate Change , Cambridge University Press , Cambridge . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Suhrke A. ( 1993 ), “Pressure Points: Environmental Degradation, Migration and Conflict”, Workshop on Environmental Change, Population Displacement, and Acute Conflict, University of Toronto and the American Academy of Arts and Science, Toronto. Suhrke A. ( 1994 ), “ Environmental Degradation and Population Flows ”, Journal of International Affairs 47 , 473 – 96 . OpenURL Placeholder Text WorldCat Van der Geest K. ( 2011 ), “ North-South Migration in Ghana: What Role for the Environment? ”, International Migration 49 , 70 – 94 . Google Scholar Crossref Search ADS WorldCat White J. ( 2013 ), “ Logistic regression model effectiveness: Proportional chance criteria and proportional reduction in error ”, Journal of Contemporary Research in Education 2 , 4 – 10 . OpenURL Placeholder Text WorldCat © The Author 2017. Published by Oxford University Press on behalf of Ifo Institute, Munich. All rights reserved. For permissions, please email: journals.permissions@oup.com
TI - Out-migration from Coastal Areas in Ghana and Indonesia—the Role of Environmental Factors
JF - CESifo Economic Studies
DO - 10.1093/cesifo/ifx007
DA - 2017-12-01
UR - https://www.deepdyve.com/lp/oxford-university-press/out-migration-from-coastal-areas-in-ghana-and-indonesia-the-role-of-trf1fFCIVj
SP - 529
VL - 63
IS - 4
DP - DeepDyve
ER -