2. Related literature

A rapidly growing body of literature empirically analyses the relationship between environmental factors and human migration. Climate has been shown to interact and work on migration behavior through a variety of direct and indirect transmission channels, including through income [Beine and Parsons (Reference Beine and Parsons2015); Coniglio and Pesce (Reference Coniglio and Pesce2015); Cattaneo and Peri (Reference Cattaneo and Peri2016)], crop production and subsequent food security [Gray and Mueller (Reference Gray and Mueller2012b); Cai et al. (Reference Cai, Feng, Oppenheimer and Pytlikova2016); Jacobson et al. (Reference Jacobson, Crevello, Chea and Jarihani2019)], amenities [Marchiori et al. (Reference Marchiori, Maystadt and Schumacher2012)], urbanisation [Marchiori et al. (Reference Marchiori, Maystadt and Schumacher2012); Maurel and Tuccio (Reference Maurel and Tuccio2016)] and violence [Abel et al. (Reference Abel, Brottrager, Cuaresma and Muttarak2019)].Footnote ⁶ Recent empirical analyses have, in particular, been facilitated by the growing availability of data on climatic factors and human mobility needed to investigate the complex nexus between these two phenomena [Berlemann and Steinhardt (Reference Berlemann and Steinhardt2017)]. Despite these efforts, there is, however, no clear consensus on the role played by environmental factors in determining global migration. This is reflected in a series of recent literature reviews reporting a significant diversity in terms of outcomes. A priori, the impact of climate factors on migration is indeed ambiguous: there is, for instance, strong evidence for a negative impact of climate change on income, particularly from agriculture, which increases incentives to migrate but also limits the ability to do so.

The diversity in outcomes is hence not surprising, given the widely varying research methodologies used and the many different contexts in which the climate-migration nexus has been studied. First, while all regions worldwide might be vulnerable to climate change, there is no reason to expect a uniform migration response to environmental stress across regions. A large part of the empirical research has focused on the migration response to precipitation anomalies in sub-Saharan Africa, where many countries rely heavily on agricultural productivity [see e.g., Henry et al. (Reference Henry, Schoumaker and Beauchemin2004); Marchiori et al. (Reference Marchiori, Maystadt and Schumacher2012); Gray and Mueller (Reference Gray and Mueller2012a); Strobl and Valfort (Reference Strobl and Valfort2013); Kubik and Maurel (Reference Kubik and Maurel2016)]. Yet, Asia is the continent experiencing more natural hazards than any other region [IDMC (2016)]; and also in Latin America, the number of people affected by natural disasters, such as flooding, forest fires, and tropical storms, is not to be underestimated [Robalino et al. (Reference Robalino, Jimenez and Chacón2015); IDMC (2016)]. These events might all spur human mobility, both within and across borders, but research on the effects of climate change on migration in Asia and Latin America is much more scarce. The results for the African case can, however, not simply be generalised to other (developing) regions given its quite particular geographical, socioeconomic, and political context. Furthermore, Piguet et al. (Reference Piguet, Kaenzig and Guélat2018), while analysing the geography of research on environmental migration, critically highlight a disproportional presence of research focus on the Global South. This is maybe not surprising as developing countries are also likely to suffer disproportionally [IPCC (2014)]. Yet, climate change is a universal challenge, such that Northern countries should not be overlooked in an analysis of the environment-migration nexus [Piguet et al. (Reference Piguet, Kaenzig and Guélat2018)].

Second, comparison and generalisation of results are further hampered by differences in the chosen research methodology across studies. In general, a distinction can be made between case studies and cross-country studies, both of which come with specific advantages and disadvantages [Piguet (Reference Piguet2010)]. Micro-level case studies mostly analyse the migration response to a specific shock (e.g., a drought, flood, cyclone, landslide, etc.) in one particular country. They typically do better than cross-country studies in accurately identifying the subpopulation that was effectively exposed to climate shock.

Many case studies provide evidence for environmental stress leading to direct, short-term, short-distance domestic migration, sometimes in the form of (planned) evacuations [Berlemann and Steinhardt (Reference Berlemann and Steinhardt2017); Cattaneo et al. (Reference Cattaneo, Beine, Fröhlich, Kniveton, Martinez-Zarzoso, Mastrorillo, Millock, Piguet and Schraven2019)]. This was for example the case when Hurricane Katrina made landfall in the US 100,000 to 150,000 people were evacuated to Houston, Texas [McIntosh (Reference McIntosh2008)]. Furthermore, adverse environmental stress is often found to result in more permanent rural-urban migration movements (i.e., a so-called urbanisation channel). Gröger and Zylberberg (Reference Gröger and Zylberberg2016), for instance, find evidence for labor migration towards urban areas after a typhoon in Vietnam. Robalino et al. (Reference Robalino, Jimenez and Chacón2015) find evidence of such rural-urban movements in response to hydro-meteorological events (such as storms, floods, rainfall, strong winds, etc.) in Costa Rica, but only for less severe hydro-meteorological events, while more severe events resulting in loss of lives decrease migration towards urban areas. Also, Dillon et al. (Reference Dillon, Mueller and Salau2011) find increasing temperatures in Northern Nigeria to affect agricultural income, in turn, driving migration towards urban areas. But movements are not always directed towards urban areas. For example, Dallmann and Millock (Reference Dallmann and Millock2017) analyse the effect of climate variability on actual flows of domestic migration in India and find drought frequency to have the strongest impact on rural–rural interstate migration. A variety of other case-studies confirm environmental stress to negatively affect populations in rural areas, thereby increasing domestic mobility.Footnote ⁷ Also Bohra-Mishra et al. (Reference Bohra-Mishra, Oppenheimer, Cai, Feng and Licker2017) find the impact of temperature and typhoons in the Philippines stronger in rural areas. Zander et al. (Reference Zander, Richerzhagen and Garnett2019), however, emphasise environmental stress (in the form of individual perceptions of heat stress) also having significant impacts on urban populations, thereby stressing the need for further research to not only focus on rural areas but to also consider urban–urban movements. Other case studies, however, find no or only modest effects of environmental stress on migration [see for instance Gray and Mueller (Reference Gray and Mueller2012b); Mueller et al. (Reference Mueller, Gray and Kosec2014), for the case of flooding in rural Bangladesh and Pakistan, respectively]. Yet others find the impact to be negative. Koubi et al. (Reference Koubi, Spilker, Schaffer and Bernauer2016a), by using individual perceptions of different types of environmental stressors, for instance, find droughts in Vietnam to significantly decrease the likelihood of migration [see also Gray and Bilsborrow (Reference Gray and Bilsborrow2013), for the case of rural Ecuador], while floods seem to increase this likelihood. Possible reasons for the lack of a significant positive effect put forward in the literature include the provision of post-disaster aid [Paul (Reference Paul2005); Boustan et al. (Reference Boustan, Kahn and Rhode2012)], an increased demand for labor in reconstruction affected areas and the destruction of infrastructure which leads to impoverishment or increased migration costs [Millock (Reference Millock2015)].

In general, case studies focus primarily on domestic (rural–urban) migration movements, thereby ignoring cross-border migration [Cattaneo et al. (Reference Cattaneo, Beine, Fröhlich, Kniveton, Martinez-Zarzoso, Mastrorillo, Millock, Piguet and Schraven2019)].Footnote ⁸ Furthermore, the particular event that is being studied is usually very local and time-specific which hampers comparison of results across studies and possibilities to draw overall conclusions [Piguet (Reference Piguet2010); Bertoli et al. (Reference Bertoli, Docquier, Rapoport and Ruyssen2019)]. An important advantage of these studies, however, is that they allow to explore individual heterogeneity in migration decisions depending on age, gender, level of education, wealth, etc. [Black et al. (Reference Black, Adger, Arnell, Dercon, Geddes and Thomas2011b)]. Bohra-Mishra et al. (Reference Bohra-Mishra, Oppenheimer, Cai, Feng and Licker2017), for instance, find a stronger migration response to climatic factors in the Philippines for males, more educated and younger individuals. Also, Mueller et al. (Reference Mueller, Gray and Kosec2014) find a slightly higher response for men compared to women in rural Pakistan [see also Thiede et al. (Reference Thiede, Gray and Mueller2016); Baez et al., (Reference Baez, Caruso, Mueller and Niu2017a, Reference Baez, Caruso, Mueller and Niu2017b), for the case of South America].Footnote ⁹ Mastrorillo et al. (Reference Mastrorillo, Licker, Bohra-Mishra, Fagiolo, Estes and Oppenheimer2016) find a stronger impact from environmental stress on domestic migration in South Africa for black and poorer migrants than for white and richer individuals.

Cross-country studies, on the other hand, typically consider a larger group of countries and time span, which allows to estimate the overall impact of climate change on migration for the countries under study (as well as to explore heterogeneous effects across groups of countries). Recent cross-country studies produce conflicting results on the macro-relationship between environmental stress and international migration. The findings vary from finding direct effects of environmental stress on international migration towards the European Union [Missirian and Schlenker (Reference Missirian and Schlenker2017)] and towards the OECD [Drabo and Mbaye (Reference Drabo and Mbaye2015); Coniglio and Pesce (Reference Coniglio and Pesce2015); Wesselbaum and Aburn (Reference Wesselbaum and Aburn2019)], to find little to no direct effect [Naudé (Reference Naudé2010); Ruyssen and Rayp (Reference Ruyssen and Rayp2014); Beine and Parsons (Reference Beine and Parsons2015); Cattaneo and Peri (Reference Cattaneo and Peri2016); Gröschl and Steinwachs (Reference Gröschl and Steinwachs2017)]; or finding evidence for indirect effects working for example through (agricultural) income [Beine and Parsons (Reference Beine and Parsons2015); Coniglio and Pesce (Reference Coniglio and Pesce2015); Cattaneo and Peri (Reference Cattaneo and Peri2016)] or increased urbanisation [Marchiori et al. (Reference Marchiori, Maystadt and Schumacher2012); Maurel and Tuccio (Reference Maurel and Tuccio2016)].

An important advantage of cross-country studies is that they allow to explore heterogeneity in the migration response across countries. The migration response to environmental stress has been shown to depend on countries' development level [Cattaneo and Peri (Reference Cattaneo and Peri2016); Gröschl and Steinwachs (Reference Gröschl and Steinwachs2017); Beine and Parsons (Reference Beine and Parsons2017)] and their agricultural dependence or main type of economic activity [Marchiori et al. (Reference Marchiori, Maystadt and Schumacher2012); Cai et al. (Reference Cai, Feng, Oppenheimer and Pytlikova2016); Falco et al. (Reference Falco, Donzelli and Olper2018)]. Cai et al. (Reference Cai, Feng, Oppenheimer and Pytlikova2016), for instance, contrast the absence of any direct effect studied by Beine and Parsons (Reference Beine and Parsons2015) and do find direct effects between temperature and international migration, but only in the most agriculture-dependent countries. In contrast, Cattaneo and Peri (Reference Cattaneo and Peri2016) find only increased outmigration rates in middle-income countries, together with a reduction in the probability of migration in poor economies (whose economy typically heavily relies on agriculture) [see also Beine and Parsons (Reference Beine and Parsons2017)].

Yet, these cross-country studies rely on the assumption that every resident of a country is affected by environmental factors in the same way, and they typically focus only on international migration, thereby ignoring domestic population movements.Footnote ¹⁰ Reliable comparisons of domestic migration across countries is, however, challenging due to the widespread variation in data collection practices of migration [Bell et al. (Reference Bell, Charles-Edwards, Kupiszewska, Kupiszewski, Stillwell and Zhu2015)]. Moreover, as pointed out by Findlay (Reference Findlay2011), little attention has been paid to where migrants might move to in response to environmental stress. Nonetheless, accounting for both domestic and international movements in the same study and accounting for environmental migrant's destination choice would improve our understanding of the heterogeneity of migration responses [Cattaneo et al. (Reference Cattaneo, Beine, Fröhlich, Kniveton, Martinez-Zarzoso, Mastrorillo, Millock, Piguet and Schraven2019)]. Furthermore, both case studies and cross-country studies mostly look either into the impact of slow or sudden onset hazards. Studies that do account for both types of hazards report conflicting results. Mueller et al. (Reference Mueller, Gray and Kosec2014), for instance, find heat stress in Pakistan (on the basis of both actual and self-reported data) to raise migration, but cannot find evidence for an increased migration response to high rainfall, flooding or moisture. In contrast, Koubi et al. (Reference Koubi, Spilker, Schaffer and Bernauer2016a) find individual perceptions of sudden environmental stress (such as floods and typhoons) to increase the likelihood of migration in Vietnam, while longer-term environmental stress (such as drought or salinity) reduces the likelihood of moving.

In addition, a growing number of studies makes use of self-reported exposure to environmental stress, emphasising the importance of understanding individual perceptions to explain their likely change in behavior in the face of environmental change [Martin et al. (Reference Martin, Billah, Siddiqui, Abrar, Black and Kniveton2014); Koubi et al. (Reference Koubi, Spilker, Schaffer and Bernauer2016a, Reference Koubi, Stoll and Spilker2016b); Parsons (Reference Parsons2019); Zander et al. (Reference Zander, Richerzhagen and Garnett2019)]. In that light, Koubi et al. (Reference Koubi, Stoll and Spilker2016b) argue that “perceptions of risk can act as a mediating factor between environmental stress and migration [Meze-Hausken (Reference Meze-Hausken2008); Black et al. (Reference Black, Kniveton and Schmidt-Verkerk2011a, Reference Black, Adger, Arnell, Dercon, Geddes and Thomas2011b); Hunteretal. (Reference Hunter, Luna and Norton2015)]. The reason is that environmental perception is the means by which individuals seek to understand their environment in order to arrive at the most effective response to environmental hazards given their individual and household level circumstances.” Also, Parsons (Reference Parsons2019) explicitly states that a focus on how the climate is experienced brings meaning to mobility as no two people experience climate change in the same manner due to a variety of objective (i.e., economy, demography, etc.) and subjective realities (norms, emotions, and culture). Understanding individual environmental experience and perceptions can thus help explain migratory movements in response to these changes. Koubi et al. (Reference Koubi, Stoll and Spilker2016b), for instance, examine whether and how individual perceptions of different types of environmental events (i.e., sudden and slow-onset) affect migration decisions in Vietnam and find that migrants perceive slow environmental events, such as droughts, as more extreme compared to non-migrants, while non-migrants perceive sudden and short-term environmental events (floods and hurricanes) as more extreme. Zander et al. (Reference Zander, Richerzhagen and Garnett2019) investigate the influence of self-reported heat stress on migration intentions among urban populations in three South-East Asian countries (Indonesia, Malaysia, and the Philippines). Their results indicate that individuals reporting hot temperatures (whether respondents ever felt stressed by heat) are more likely to intend to migrate, with women, older and richer individuals showing the strongest intentions to move.

In addition, an expanding body of literature has empirically explored the drivers of migration intentions [see among others Jónsson (Reference Jónsson2008); Drinkwater and Ingram (Reference Drinkwater and Ingram2009); Becerra (Reference Becerra2012); Creighton (Reference Creighton2013); Carling and Collins (Reference Carling and Collins2018); Carling and Schewel (Reference Carling and Schewel2018)], some of which in a cross-country framework, relying on the GWP. Dustmann and Okatenko (Reference Dustmann and Okatenko2014) for instance, look into the role of wealth constraints and the quality of local amenities in migration decisions in sub-Saharan Africa, Asia, and Latin America. Docquier et al. (Reference Docquier, Peri and Ruyssen2014), disentangle the role of macroeconomic determinants of migration intentions aggregated at the country level, as well as the probability that these intentions translate into actual migration. Dao et al. (Reference Dao, Docquier, Parsons and Peri2018) also make use of aggregated international migration intentions as well as realisation rates by education level to examine the microeconomic and macroeconomic drivers of the relationship between emigration rates and economic development. Moreover, Docquier et al. (Reference Docquier, Machado and Sekkat2015), Docquier and Machado (Reference Docquier and Machado2016) and Delogu et al. (Reference Delogu, Docquier and Machado2018) use the GWP data to proxy the number of potential migrants who could respond to removal of legal restrictions on migration. Manchin and Orazbayev (Reference Manchin and Orazbayev2018) and Bertoli and Ruyssen (Reference Bertoli and Ruyssen2018) quantify the effect of migrant networks on migration intentions and on prospective migrants' destination choice, respectively. Ruyssen and Salomone (Reference Ruyssen and Salomone2018) track both women's migration desires as well as preparations they have already made to migrate within the next 12 months and disentangle how gender discrimination fosters or impedes female migration across countries. Docquier et al. (Reference Docquier, Tansel and Turati2020) investigate whether intended migrants from MENA countries self-select on cultural traits such as religiosity and gender attitudes. Friebel et al. (Reference Friebel, Manchin, Mendola and Prarolo2018) study the elasticity of migration intentions to illegal moving costs, exploiting the demise of the Gaddafi regime in 2011, and the ensuing opening of the Libyan route to Europe as a quasi-natural experiment. Gubert and Senne (Reference Gubert and Senne2016) consider information on individuals' plans to move within the next 12 months to explore the relative attractiveness of EU-countries as potential destinations. Finally, Bertoli et al. (Reference Bertoli, Docquier, Rapoport and Ruyssen2019) explore the role of weather shocks in six Western African countries at a relatively detailed level of spatial resolution. To the best of our knowledge, our paper is the first to look into the impact of self-reported exposure on environmental stress on intentions to migrate both domestically and internationally, that accounts also for regional variation in migration responses as well as diversity in the preferred destination across individuals.

3. Data and stylised facts

Our analysis rests on individual-level data from 90 countries where at least one Gallup World Poll has been conducted in the year 2010.Footnote ¹¹ The surveys conducted by Gallup typically have a sample of around 1,000 randomly selected respondents per country, and the data are collected either through face-to-face interviews or through phone calls in countries where at least 80% of the population has a telephone land-line.Footnote ¹² The sampling frame represents the entire civilian, non-institutionalised population aged 15 and over covering the entire country including rural areas.Footnote ¹³ Our final sample contains 76,484 individuals at working age (i.e., between 15 and 60 years old) with valid information on all the variables of interest used in the model, interviewed worldwide during the year 2010. In what follows, we explain in detail how the variables of interest have been constructed.

3.1 Migration behavior

The GWP include several related questions on the intention to migrate. Following Manchin and Orazbayev (Reference Manchin and Orazbayev2018), we combine three consecutive questions to categorise respondents' short-run migration intentions: (Q1) “In the next 12 months, are you likely or unlikely to move away from the city or area where you live?”; (Q2) “Ideally, if you had the opportunity, would you like to move permanently to another country, or would you prefer to continue living in this country?”; and (Q3) “Are you planning to move permanently to another country in the next 12 months, or not?”.Footnote ¹⁴

The first question refers to a strong inclination to migrate within the next year regardless of destination. Both the phrase “likely to move” and the relatively short time window of 12 months within which any intended migration response is placed, make it likely that only individuals who have already developed concrete migration plans provide a positive answer to this question [Dustmann and Okatenko (Reference Dustmann and Okatenko2014)]. Also, the migration intentions depicted in the second question are stricter than mere migration considerations typically documented in other surveys [e.g., Creighton (Reference Creighton2013); Dustmann and Okatenko (Reference Dustmann and Okatenko2014)] since they use a stronger formulation which directly asks for the likely response under ideal conditions [Manchin and Orazbayev (Reference Manchin and Orazbayev2018)]. The absence of a time frame, however, does not require any concrete migration plans to answer this question affirmatively, in contrast, to question (Q1), an element which has been proved important to guarantee accurate replies [see e.g., European European Commission (2010); Dustmann and Okatenko (Reference Dustmann and Okatenko2014)]. In order to identify people with strong intentions (concrete plans) to migrate in the short-run, we, therefore, combine questions (Q2) and (Q3).

In order to be able to compare these three questions, some further assumptions need to be made [see also Manchin and Orazbayev (Reference Manchin and Orazbayev2018)]. First, without imposing a constraint on the distance (domestic or international) or the length (temporary or permanent) of the move, question (Q1) elicits firmer intentions than question (Q2). The phrasing of question (Q1) is, nonetheless, much closer to that of question (Q3): both questions consider similar time periods during which the move should take place (“in the next 12 months”) and ask for a relatively firm intention to migrate (there is no reference to ideal conditions or opportunities). A distinction that remains, however, is that question (Q3), just like question (Q2), asks for permanent migration plans only. This implies that for further comparisons, we need to assume that question (Q1) can be interpreted as asking about permanent moves too, which however does not seem implausible given the phrasing “likely to move away”.Footnote ^15, Footnote ¹⁶

Table 1 illustrates all possible combinations of replies to each of these three questions, the resulting migration status and the share of respondents in our sample that belongs to each category. A few things are worth mentioning. First, according to these figures, the share of people intending to move domestically in the next 12 months stands at 14.5% worldwide. Second, on average 2.4% of the respondents questioned worldwide in 2010 intended to migrate permanently abroad in the next 12 months, which is not too far from the overall actual flow of 3.2% documented in the year 2010 by the World Bank [World bank (2010)]. Third, the vast majority of the respondents, 83.1%, indicated not planning to move in the next 12 months.

Table 1. Categorisation of migration behavior (2010)

Notes: Authors' calculations based on the Gallup World Polls. Share denotes the share of respondents in our sample that belongs to each category. (Q1) “In the next 12 months, are you likely or unlikely to move away from the city or area where you live?”; (Q2) “Ideally, if you had the opportunity, would you like to move permanently to another country, or would you prefer to continue living in this country?”; (Q3) “Are you planning to move permanently to another country in the next 12 months, or not?”; and (Q4) “To which country would you like to move?”.

These figures confirm the importance of national borders for migration. Domestic migration is likely to involve shorter travel distances and also cultural differences between the origin and destination location are probably more limited than in the case of international migration, implying lower monetary and psychological costs, which helps to explain the imbalance in migration intentions across domestic and international destinations. Importantly, as put forward by Beine and Parsons (Reference Beine and Parsons2015), international migration does not always entail a move over long distances. They postulate that in regions with porous borders, such as Africa, it might be less costly to cross an international border than to migrate domestically over longer distances. Yet, on average, crossing an international border involves additional costs of obtaining passports and visas, so that in general this conjecture can be assumed to hold. The importance of borders was shown among others by Helliwell (Reference Helliwell1997), who finds that national borders play an even larger role in determining migration flows than they do to explain trade flows (a well-known fact in the trade literature). So even if individuals might want to move internationally (to an OECD country) in the face of environmental hazards, many of them lack the means to actually do so. Hence, if liquidity constraints are binding, individuals hit by environmental hazards may end up moving internally, a possibility which might be particularly relevant in developing countries where people are more likely to be financially constrained [Beine and Parsons (Reference Beine and Parsons2015)].

Furthermore, for those respondents replying positively to question (Q3), there is the follow-up question (Q4) “To which country would you like to move?”. The destination dimension allows us to identify prospective international migrants' preferred destination, which we will use to differentiate between intentions to migrate intraregionally, towards the OECD or elsewhere. We define intraregional migration as migration within the same subcontinent for which we rely on the country classification of the UN DESA Population Division (used among others in the International Migrant Stock database). Migration to the OECD concerns migration towards OECD countries outside the subcontinent in which the respondent resides. Other destinations encompass non-OECD countries that are not part of the subcontinent in which the respondent resides. For a full list of (sub)continental regions and the countries falling within each of them, see Table A1.

A similar logic as above applies when distinguishing between intraregional vs. migration to OECD countries outside the respondent's subcontinent: for most countries (particularly those in developing regions), travel distance to OECD countries is much larger than to countries within the same subcontinental region, and also in terms of culture the latter countries probably are much more similar. Indeed, empirical evidence based on gravity-type studies shows that geographical distance between countries of origin and destination, as well as the presence of a common border, a common language and other shared characteristics of countries, form important determinants to explain bilateral migration flows [see Beine et al. (Reference Beine, Bertoli and Fernández-Huertas Moraga2016), for an overview].

The resulting number of observations in each category and the related shares are presented in Table 1. From the 2.3% of respondents intending to migrate permanently abroad in the next 12 months, about 1.5% has plans to move away towards the OECD and about 0.4% is planning to move intraregionally within the following 12 months. The relatively small share of intended intraregional migrants is the result of the rather small number of countries belonging to each subcontinent according to the UN DESA Population Division country classification. Larger figures would be obtained when regions would be defined more broadly (for instance at the continental level), but this would entail much larger travel distances and smaller cultural similarities within a region.

The share of respondents falling within each category varies considerably across regions and countries. Africa and Latin-America and the Caribbean proved to be the most mobile regions in our sample. Respectively 23.2% and 19.2% of respondents intend to migrate domestically in the next 12 months, compared to only 9.7% in Europe, 11.6% in Asia, and 15.5% and 15.2% in North America and Oceania. Also, within regions, there is a large degree of variation in stated migration intentions, as can be seen in Figure 1 displaying country-specific shares of respondents falling within each category. Within Africa, e.g., we find the highest shares of respondents intending to migrate domestically in Liberia (31.0%), Botswana (31.0%), and Nigeria (29.6%), while they are much lower in Burkina Faso (14.9%), Mali and South-Africa (17.2 and 17.3%) (panel b). As far as international migration intentions are concerned, Africa and Latin America and Caribbean report a higher average share compared to the full sample, 5.5% (3.1%) of African (Latin American) respondents intend to migrate internationally, but only 3.6% (2.4%) is intending to move towards the OECD (panel d), respectively. The highest shares of respondents intending to move towards the OECD can be found in Senegal (with 11.7%), Haiti (8.9%), Liberia (8.1%), Sierra Leone (6.0%), and El Salvador (5.4%). Within Africa, Burkina Faso has the lowest share of domestic intending migrants, but the highest share of respondents intending to migrate intraregionally (panel c) standing at 3.3%, followed by Niger with 2.5%, which is greatly above the African average of 0.9%. Europe, Russia, and Asia seem to be the least mobile regions in our sample. Overall, more than 88.3% of European respondents intend to stay in their country, 93.5% in Russia, and 87.2% in Asia (panel a). Within the latter, only 2.1% of the respondents in Singapore, 4.2% in Kyrgyzstan, 5.4% in Vietnam, and about 6.3% in Japan are intending to move domestically (panel b). These latter figures are in vast contrast with the average of 14.5% of domestic migrants in our total sample.

Figure 1. Share of respondents following their migration intentions by destination.

Notes: The maps report, for each country in our sample, the share of individuals interviewed in 2010 who intend to (a) stay in their current place of residence, (b) migrate domestically, (c) migrate intraregionally and (d) migrate towards the OECD within the next 12 months (based on the categorisation illustrated in Table 1). Darker colors are associated with a higher share of individuals in a country reporting to stay (panel a) or having plans to move (panels b, c and d), while lighter colors denote lower shares. Source: Authors' elaboration on the Gallup World Polls.

3.2 Environmental stress

Information on individual self-reported exposure to environmental stress (Environmental stress) is extracted from the following question (Q5) available in the GWP: “In the past 12 months, have there been any severe environmental problems in your city or area, or not? For example, pollution, floods, droughts, or long periods of extreme heat or cold?”. This question directly asks whether people have experienced extreme environmental stress during the past 12 months, covering a wide variety of both slow and sudden onset environmental threats. It takes the value one if question Q5 is answered affirmatively and zero otherwise.Footnote ^17, Footnote ¹⁸ On average, 37.5% of the respondents in our sample indicate having experienced environmental stress in the last 12 months. Figure 2, plotting country-specific percentages of individuals reporting environmental stress in the past 12 months, indicates significant variety across countries and regions in our sample. The share of individuals having experienced environmental stress in the last year ranges from 4.5% to 6.5% in the Netherlands, Denmark, Japan, and Luxembourg to reaching over 80% in Burkina Faso and Chad (respectively, 82.3% and 81.1%), followed by Kenya (79.3%), Niger and Mongolia (both around 75%).

Figure 2. Share of respondents reporting environmental stress in their city or area in the past 12 months.

Notes: The map reports, for each country, the share of respondents reporting environmental stress in their city or area in the past 12 months. Darker colors are associated with a higher share of individuals in a country reporting environmental stress in the past 12 months in their city or area, while lighter colors denote lower shares. Source: Authors' elaboration on Gallup World Polls.

Bekaert et al. (Reference Bekaert, Ruyssen and Salomone2020) investigate what this measure of exposure to environmental stress exactly captures by computing pairwise correlations with other GWP individual climate indicators as well as external measures of environmental stress commonly used in the literature. Pairwise correlations with other individual indicators of environmental stressFootnote ¹⁹ from the GWP were all found to be positive and highly significant. The strongest correlation is obtained with indicators of a lack of water for growing crops and raising livestock. This seems to suggest that our key variable of interest primarily picks up exposure to water scarcity and its implications, stemming from drought.

In addition, the variable of interest is aggregated at country level (as the share of positive answers by country) and then correlated with objective indicators of the occurrence and intensity of natural disasters taken from the EM-DAT, provided by the Centre for Research on the Epidemiology of Disasters (CRED). The EM-DAT database contains disaster data disaggregated by disaster type (i.e., climatological, geophysical, meteorological, and hydrological disasters) collected from various sources. An event is classified as a disaster when at least one of the following criteria is met: ten or more people reported killed, 100 or more people reported affected, a declaration of a state of emergency or a call for international assistance. The correlations show that our question measuring exposure to environmental stress primarily picks up environmental stress related to drought, riverine floods, and cold waves, and to a lesser extent also fires and tropical cyclones.

Overall, it can be concluded that our GWP individual measure of environmental stress predominately picks up exposure to droughts and water scarcity which are said to “kill more people than any other single weather-related catastrophe” [United Nations Convention to Combat Desertification (2014)]. Correlations with external indicators are not perfect, though, reflecting various objective (related to the economy, demography, as well as individual characteristics) and subjective factors (norms, emotions, and culture) shaping people's self-reported experiences with environmental hazards [Parsons (Reference Parsons2019)]. As argued by Parsons (Reference Parsons2019), no two people experience climate change in the same manner such that a focus on how climate change is experienced brings meaning to mobility. Nonetheless, country-level variation in environmental stress—as would be picked up by objective indicators of the occurrence of environmental hazards—is largely controlled for through the inclusion of our country of origin fixed effects in our empirical specification.

3.3 Individual and household controls

Besides these key variables of interest, we also keep track of the additional individual- and household-level information contained in the GWP. Specifically, we record respondents' age (Age) and gender (Male) at the time of the interview, whether they are highly educated (Higher education) (i.e., have completed secondary education), whether they live in a rural or urban area (Urban) (a rural area covers residence on a farm or in a small town or village while an urban area is defined as a large city or a suburb of a large city), and whether they have a distance-one connection abroad (Network) (i.e., relatives or friends who are living in another country whom they can count on to help them when needed). We also take into account the number of adults (aged 15 and above) in the household (Number of adults), the number of children (below 15 years of age) in the household (Number of children) as well as the log of self-reported household income per capita (HH income pc (ln)). The econometric analysis is conducted on individuals at working age (i.e., between 15 and 60 years old).

Table 2 provides descriptive statistics for the variables used in our empirical analysis. As indicated above, 37.5% of respondents reports having experienced environmental stress in the past 12 months. The share of people answering positively to this question is considerably larger among those expressing an intention to migrate (particularly intraregionally). Furthermore, individuals in our sample are on average 36 years old, but those expressing an intention to migrate are considerably younger. The overall sample contains slightly more females, but men are more likely to state an intention to migrate (both domestically and internationally). Of the total, 68.9% of the individuals in our sample have completed secondary education; 39.4% of respondents live in an urban area. The average number of adults (aged 15 and over) and children (below 15) in the household respectively, stands at 3.3 and 1.4. Respondents in larger households are more likely to express intentions to move abroad, which probably signals the larger responsibilities and pressure on individuals having to support more dependents. Finally, 30% of respondents indicate having a network of family and friends living abroad, but this figure is more than twice as high among those who intend to migrate abroad.

Table 2. Descriptive statistics following the migration categorisation

Notes: Authors' own calculations based on the Gallup World Polls. Standard deviations between brackets. HH income pc (ln) denotes the log of household income per capita.

4. Theoretical foundations

The model that we bring to the data to analyse the migration decision and the prospective migrant's destination choice is a random utility maximisation (RUM) model of migration. Consider an individual i, residing at time t in area r of country j; the choice set D of individual i includes his or her home area r (which we refer to as k = 0 without loss of generality), the rest of country, i.e., R _j/{r} where R _j is the set of other areas in country j (we refer to this second alternative in the choice set as k = 1), the set W/{j} of other countries within the same subcontinental region (k = 2), the set of OECD countries outside the subcontinental region where the respondent resides O/{j, W} (k = 3), and the set of other countries in the world E/{j, W, O} (k = 4). Thus, the choice set D includes five alternatives: staying at origin, moving domestically, migrating to an international destination within the same subcontinental region, migration towards an OECD country outside the subcontinental region, or migrating towards another international destination.

Let U _ikt denote the utility that individual i would derive if opting for alternative k ∈ D at time t. We assume that this alternative-specific utility includes a deterministic component V _ikt and a stochastic component ε _ikt. If the stochastic component follows an independent and identically distributed Extreme Value Type 1 (EVT-1) distribution, then the probability p _ikt that k ∈ D will be the utility-maximising alternative is given by:

(1)$$p_{ikt} = \displaystyle{{e^{V_{ikt}}} \over {\mathop {\mathop \sum \nolimits}\limits_{l\in D} e^{V_{ilt}}}}$$

The relative probability of migrating domestically over staying at origin is given by:

(2)$$\displaystyle{{\,p_{i1t}} \over {\,p_{i0t}}} = e^{V_{i1t}-V_{i0t}}$$

The relative probability of migrating to destination k = 2, 3, 4 over staying at origin is given by:

(3)$$\displaystyle{{\,p_{ikt}} \over {\,p_{i0t}}} = e^{V_{ikt}-V_{i0t}}$$

The relative probability of intending to move (irrespective of the destination) over staying at origin is given by:

(4)$$\displaystyle{{\,p_{i1t} + p_{i2t} + p_{i3t} + p_{i4t}} \over {\,p_{i0t}}} = \displaystyle{{e^{V_{i1t}} + e^{V_{i2t}} + e^{V_{i3t}} + e^{V_{i4t}}} \over {e^{V_{i0t}}}}$$

Relative choice probabilities are solely determined by the difference in the levels of utility associated with each pair of alternatives (and not by the levels themselves). This, in turn, entails that we can normalise the utility associated with the baseline option (staying) to zero. Thus, the estimated coefficient for all the regressors gives us the differential effect of each variable on the attractiveness of moving vs. staying.

We include in the estimation a vector of individual and household-level characteristics x _it. The elements included in the vector are: dummies for different age groups (i.e., 20–29, 30–39, 40–49, 50–60, with 15–19 representing the omitted category) (Aged 20–29; Aged 30–39; Aged 40–49; Aged 50–60), a dummy for male individuals (Male), a dummy for high-educated individuals (i.e., who have obtained a secondary education degree, equivalent to at least 9 years of education) (Higher education), a dummy for individuals living in urban areas (Urban) (i.e., a large city or a suburb of a large city as opposed to residence on a farm or in a small town or village), and a dummy for having a distance-one connection abroad (Network) (i.e., relatives or friends who are living in another country whom an individual can count on to help him or her when needed), the number of adults (aged 15 and above) in the household (Number of adults) and the number of children (below 15 years of age) in the household (Number of children) as well as the log of the self-reported household income per capita (HH income pc (ln)). We also include country of origin fixed effects to account for the fact that the migration behavior of people in the same country might be driven by common unobserved time-invariant factors.

This vector also includes a dummy for whether the individual has experienced any environmental stress in the past 12 months (Environmental stress). If the coefficient $\hat{\beta }$ associated with this dummy is positive, then this means that severe environmental issues make the origin location relatively less attractive than the intended destination. The marginal effect on the probability of intending to move is given by $\hat{\beta }p_{ikt}( {1-p_{ikt}} )$, with k = 1, 2, 3, 4 depending on the choice of the dependent variable, while $\hat{\beta }$ itself represents the partial derivative of the logarithm of the relative choice probability with respect to our variable of interest.

A possible concern in the regression on migration intentions is the following: if an individual considers moving to a neighboring area, then environmental factors at origin could be positively correlated with environmental factors at the destination, and this correlation confounds the effect of the estimated coefficient, possibly biasing it towards zero and reducing its statistical significance.Footnote ²⁰ A further concern related to the data is that individuals might have moved between the occurrence of an extreme environmental shock and the date in which they are interviewed by Gallup. If individuals with the highest propensity to migrate abroad have already moved by the time of the survey, then we would be missing them entirely. If they moved domestically, they might still be included in the sample, but we would be incorrectly matching them to the wrong environmental conditions (the GWP do not provide information on the individual past migration history), i.e., those prevailing in the area to which they moved rather than in their home area.