The Dynamics of Refugee Return: Syrian Refugees and Their Migration Intentions

Abstract

We study the drivers of refugees' decision making about returning home using observational and experimental data from a survey of 3,003 Syrian refugees in Lebanon. We find that the conditions in refugee-hosting countries play a minor role. In contrast, conditions in a refugee's home country are the main drivers of return intentions. Even in the face of hostility and poor living conditions in host countries, refugees are unlikely to return unless the situation at home improves significantly. These results challenge traditional models of decision making about migration, where refugees weigh living conditions in the host and home countries (“push” and “pull” factors). We offer an alternative theoretical framework: a model of threshold-based decision making whereby only once a basic threshold of safety at home is met do refugees compare other factors in the host and home country. We explore some empirical implications of this new perspective using qualitative interviews and quantitative survey data.

Mass forced displacement has proven to be an enduring challenge in contemporary international politics. Forcibly displaced people face joblessness and food insecurity, lack legal status, and experience hostility and violence in host countries. Meanwhile, the governments of many hosting countries struggle to meet the additional demands that refugees place on public services and infrastructure (Verme et al. 2015). The consequences of forced migration are most acute in developing countries—where a large majority of refugees reside—due to constrained government budgets, weak state capacity, and limited public infrastructure.¹ Making matters worse, as of 2018–19, 78 per cent of all refugees worldwide were in protracted refugee situations, living in exile for more than five consecutive years. The largest protracted refugee populations are Afghans, Syrians, and South Sudanese; the vast majority of these people reside in neighboring countries in the Middle East, North Africa, and South Asia.²

Despite the significant challenges that refugee crises pose to refugees themselves, hosting countries, and international donors, effective responses are lacking. Each year over the last decade, less than 1 per cent of refugees worldwide received citizenship in a hosting country and only 1–2 per cent were resettled.³ Further, governments in hosting countries often restrict refugees' rights and prospects for integration in order to accelerate return (Janmyr 2016). The vast majority of refugees remain in a state of limbo, being unable to either integrate locally or find a new home through resettlement.

How, then, do refugee crises come to an end? To answer this question, we need an understanding of whether, when, and why refugees return home. However, this proves to be a challenging issue to explore empirically. Existing administrative data on refugee return are incomplete: in the past, many returns went unrecorded and the definition of return varied across organizations and across countries, making systematic analysis difficult. Moreover, data collection is especially challenging with mobile populations. The unpredictable timing of return means that it has been difficult to capture household return migration through surveys, especially in contexts of ongoing violence, which compound challenges related to data collection and sample attrition.

We tackle the challenges of studying refugee return with original, cross-sectional survey data from a nationally representative sample of 3,003 Syrian refugee households in Lebanon. This constitutes the first important contribution of our article. By supplying data on return intentions and preparations for refugees, as well as how intentions vary in the short and long term, our study provides novel and hard-to-collect descriptive data on how refugees think about return. We then use these data to examine predictors of return intentions and preparations, to explore the role of information, and to identify differences in the drivers of short- and long-term return intentions. We supplement analysis of observational data with a conjoint experiment to isolate the causal effect of conditions in Syria and Lebanon on return intentions, as well as semi-structured interviews with Syrians living in Lebanon. Finally, we explore the generality of our findings using a second original survey of Syrian refugees in Jordan.

The Syrian refugee crisis in Lebanon provides a useful setting in which to examine the dynamics of refugee return. When we conducted data collection in mid- and late-2019, active conflict in Syria was diminishing and many governmental and humanitarian organizations had begun discussing and even facilitating returns. Conditions across Syria varied widely—many areas remained insecure, and overall prospects for safety, economic recovery, and service provision were uncertain. At the same time, Syrian refugees in Lebanon experienced highly differentiated living conditions, local government policies, and levels of community hostility. In some municipalities, local governments actively targeted refugees for harsh treatment and prominent politicians called for accelerating their return, while in others, refugees were integrating both economically and socially. We leverage this variation in prospects in the country of origin and well-being in the host country to learn about the drivers of return intentions.

Our theoretical starting point is the “push” and “pull” framework, which suggests that refugees decide whether to return to their home country or stay in the host country based on a comparison of conditions in the two locations. This model draws its intellectual origins from the neoclassical economic model of migration (Borjas 1987; Borjas and Bratsberg 1996) and has long informed both research and policy discussions of return migration (see, for example, El Asmar, Shawaf, and Mikdashi 2019; Oxfam 2018; World Bank 2020). Our empirical results call into question the relevance of the push–pull framework for refugee decision making. In particular, we find that, on average, push factors in the hosting country have a small impact on refugees’ aspirations to return. Across a range of potential drivers, conditions in Lebanon do not substantially shape return intentions, even though many Syrians confront extremely challenging living situations.

Before data collection for this project, we registered our research hypotheses (along with statistical models to test them), including our expectations that host-country conditions would shape return intentions.⁴ We expected host-country conditions to matter for a number of reasons. First, existing theoretical models of migration operate under a framework of comparing well-being in both the place of residence and potential destinations (Borjas 1987; Borjas and Bratsberg 1996; Massey et al. 1993). Second, related studies of return among labor migrants and internally displaced populations find that conditions in the place of residence are important drivers of return choices (Arias, Ibáñez, and Querubin 2014; Constant and Massey 2003; Dustmann and Weiss 2007; Stefanovic, Loizides, and Parsons 2015). Further, governments in many refugee-hosting countries implement policies that undercut refugee well-being with the goal of pushing refugees to return (Janmyr 2016). Advocacy organizations and media reports have repeatedly called attention to Lebanon's policy of refugee immiseration to coerce Syrians to return.⁵

Instead, we find that conditions in the home country predict the return intentions of refugees. To explain these results, we propose an alternative framework for understanding refugee return. Specifically, our threshold model suggests that refugees do not trade safety in the country of origin for other goods, such as economic conditions and public service provision, so long as their safety concerns are not met. Only once safety in the country of origin passes a certain threshold do refugees begin to consider other factors. We provide qualitative and survey data that are consistent with this explanation. Overall, we find that despite having been displaced for nearly a decade and having little hope to return in the near future, people who have fled the violence and societal devastation of civil war generally want to return home when those threats dissipate.

This article contributes to an emerging body of work on the drivers of return among displaced populations. Much of the existing work focuses on internally displaced persons (IDPs), who face the question of whether to return to their place of origin after flight within their country (Arias, Ibáñez, and Querubin 2014; Camarena and Hagerdal 2020; Stefanovic, Loizides, and Parsons 2015; Weber and Hartman 2022). A small but growing body of research has begun to shed light on the return choices of refugees, that is, people who flee their home country during conflict and, as a result, face choices and constraints distinct from those that IDPs confront (Beaman, Onder, and Onder 2022; Beber, Roessler, and Scacco 2021; Ghosn et al. 2021). This is important because refugees generally have fewer prospects for long-term settlement in their place of refuge than do IDPs as citizens in their home country, and refugees face distinct legal, political, and economic challenges in a foreign country from those that IDPs face. We also contribute to this work by providing evidence on the role of a comprehensive set of theoretically motivated drivers of return decisions. Building on recent studies like that by Ghosn et al. (2021), which focuses on refugees' wartime experiences and psychological anchoring, we extend the scope and examine a broad range of factors, including material well-being and future prospects. Finally, our detailed measurement of return intentions and preparations allows us to study the interrelated roles of timing and aspirations, building on recent work that focuses on observed migration behavior (see, for example, Beaman, Onder, and Onder 2022; Camarena and Hagerdal 2020). In light of the many constraints that refugees face, evidence on return intentions is important because focusing solely on migration behavior does not allow us to distinguish if someone stays in exile because they were unable to return, preferred to not do so, or both (Schewel 2020).

More broadly, our research advances the literature on host-country politics. Traditionally, research on immigrant–native dynamics focuses on host populations in the receiving country, examining the effect of immigration and refugees on local labor markets (see, for example, Scheve and Slaughter 2001), political attitudes and behavior (see, for example, Hainmueller and Hopkins 2014), and tensions, discrimination, and conflict (see, for example, Adida 2014). This article explores the possibility that these host-country dynamics may, in turn, shape choices about return. Importantly, despite the significant tensions with locals and active efforts to make host societies less hospitable, our results show that refugees may be willing to live with extreme hardship in the absence of a viable opportunity to return to their home country.

When Do Refugees Return?

We approach return migration through the lens of household decision making—an approach that enables us to consider the impact of macro-level factors in a host country or the home country, subnational processes like localized violence and anti-refugee sentiments, and micro-level measures of household experiences, beliefs, and resources. We define return as moving from a host country to one's home country with no immediate plans to depart again. Our focus is on the binary choice of whether to return to the home country, thereby setting aside other migration-related choices that refugees face, such as internal migration within a host country, location choice within their home country after return, and formal or informal migration to a third country.

In identifying the factors that influence return, our starting point is the “push” and “pull” factors framework. The intellectual origins of this framework flow from neoclassical economic theories of migration (Borjas 1987). Although developed to explain patterns of labor migration, these models provide a useful framework for understanding individual decisions to migrate based on the costs and benefits of living in different countries. From this perspective, potential migrants consider their long-term expected well-being in the country in which they live against conditions in possible destination countries, while accounting for the costs of travel, the challenges of adapting to a new labor market and culture, and any nonmonetary costs or benefits of migration.

This framework is what underlies a focus in policy discussions on the relative importance of “push” and “pull” factors in the decisions of refugees about return. In this context, “push” refers to the conditions in the hosting country, while “pull” refers to the situation at home. For example, drawing on comparative experience, a recent World Bank (2020) report describes the “mobility calculus” of refugees in terms of a weighing of different structural conditions—peace, security, and protection; livelihoods and economic opportunities; housing, land, and property; and infrastructure and access to services—in the host and home countries. Recent nongovernmental organization (NGO) reports from the Middle East adopt a similar framing, arguing that refugees actively consider whether life will be better at home than in their host country. They also explicitly state that push factors may encourage refugees to return prematurely (see, for example, Oxfam 2018). However, because conditions are often so difficult in both places, some argue that we need to think of these conditions as push factors in both settings (for example, which is the better of two bad options), rather than push and pull factors (El Asmar, Shawaf, and Mikdashi 2019).

The logic of the push–pull framework has also informed Lebanon's response to the refugee crisis (Janmyr 2016). Under the assumption that making life difficult for refugees will push them to return, the Lebanese government has long imposed significant restrictions on Syrian refugees. For instance, the Lebanese government has increased the difficulty and expense of obtaining residency permits for Syrian refugees. This effectively limits Syrians' access to education and healthcare, and subjects them to detention and forcible return (Amnesty International 2015). A recent report by the Carnegie Endowment calls on the international community not only to stop forced return, but also to reduce “factors in the host countries that push refugees to return home prematurely” (Yahya 2018, 52).

Although one might question the value of rational choice models of migration in contexts of forced displacement, recent research on refugees and IDPs has found evidence that this framework provides important insights even in environments where initial decisions to move were influenced by violence. For example, a study of Lebanese Christians who were internally displaced during the country's civil war in the 1980s finds that in the absence of attractive economic opportunities, people may not return to their home areas even if they have strong aspirations to do so (Camarena and Hagerdal 2020). In the Colombian context, researchers found that, across a range of measures, IDPs were more likely to return home when their prospects were good (including landownership, work opportunities, and social networks); conversely, they were less likely to return home when they could do better in their reception site (Arias, Ibáñez, and Querubin 2014). Building on this framework, we hypothesize that refugees' choices about return are shaped by four key conditions: (1) the situation in the host country; (2) the dynamics in the country of origin; (3) the costs of movement; and (4) the quality of information about the costs and benefits of return.

Conditions in the Host Country

When assessing their prospects in the host country, refugees evaluate their household well-being along multiple dimensions: “Are they employed?”; “Do they have access to humanitarian aid?”; “Are public services available?”; “Are they welcomed by their surrounding communities?”; and “Is there a path to formal legal status?” These critical elements have been shown in previous work to be primary determinants of decisions to return. For example, Constant and Massey (2003) find that a lack of stable full-time employment roughly doubles the odds of return migration for foreign workers in Germany. Arias, Ibáñez, and Querubin (2014) find that IDPs in Colombia are roughly 4 percentage points more likely to intend to return if the household head is unemployed. Moreover, Stefanovic, Loizides, and Parsons (2015) find that integration into a new environment in western Turkey, measured by economic advancement and knowledge of Turkish, decreased return intentions among displaced Kurds from eastern Turkey. All else equal, we expect that improvement on any of these dimensions will increase the perceived value of remaining in the host country, making return less likely. Likewise, a worsening of the situation in the host country will increase the aspiration to return. Historically, host governments have often applied intense pressures for refugees to return en masse (see, for example, Janmyr 2016; Schwartz 2019), gradually ramping up anti-refugee rhetoric and undercutting refugees' legal residency and right to work. Often, the rationale behind such restrictions is that harsher living conditions will incentivize refugees to return home.

We expect conditions in the host country to play an important role in our study context. As we explain in the next section, the Lebanese government has long attempted to make life difficult for Syrian refugees, including through restrictions on the right to work and access to services (Janmyr 2016). Furthermore, there is significant local variation in the levels of hostility toward Syrian refugees. Some Lebanese welcome Syrians because they perceive them to be escaping a brutal dictatorship and a violent civil war. Others view Syrians as a demographic threat to the country's sectarian balance. This variation in hostility is reflected in local government policies, with some municipalities imposing curfews targeting Syrians, as well as in discrimination by the local authorities and the host community.

Conditions in the Home Country

Decisions about return also depend on the environment in the home country. Refugees must consider both the current conditions in their country of origin and their expectations about how the situation will evolve. Economic prospects are one key part of the decision-making calculus, as refugees assess the quality of the post-war economy, access to public services, and the viability of meeting their family's basic needs. For example, in Colombia, Arias, Ibáñez, and Querubin (2014) show that people who own land or have prospects for employment in their place of origin have higher return intentions. Beber, Roessler, and Scacco (2021) find that the South Sudanese residents of North Sudan who were least likely to return were the middle class for whom employment opportunities were scarce in the South. War may also impact people's economic prospects by generating disputes over public policy or property rights (Schwartz 2019). In one example of such dynamics, Weber and Hartman (2022) demonstrate that displaced Iraqis were more likely to return if they had property at the place of origin only when they possessed written documentation of their rights and the property was not disputed or badly damaged.

Refugees also must consider the security situation at home. For example, will violence pick up again, and would it affect the region they are from? Moreover, might they be at risk of targeted persecution or arrest upon return? As households assess their safety if they were to return, they may consider current levels of violence in their hometown, their connections or proximity to existing political divisions, and expectations about continued violence and persecution by the government or armed groups. Overall, we expect that better conditions in the home country will be associated with greater aspirations and preparation to return.

In addition, we expect networks of friends and family to predict return intentions. After over a decade of civil war, many refugees have not seen their relatives and friends in Syria for several years. As the regime regains broad control in Syria and fighting abates, reuniting with relatives and friends could provide a particularly important reason for people to return. Further, when people consider returning, networks in their places of origin can provide critical support by offering shelter and connections to opportunities. This argument is in line with Arias, Ibáñez, and Querubin (2014), who find that social networks in people's places of origin (measured through membership in a peasant organization or collective landownership) increase return intentions.

Mobility Costs

Households considering return migration also weigh the financial costs and physical risks associated with moving (Hunt and Mueller 2004). In particular, long-distance travel for refugees may be expensive and require passage through unsafe territory. Returning refugees may also face the prospect of being stopped at checkpoints run by the state or armed groups that charge tolls or taxes, steal possessions, or detain, interrogate, or abuse travelers. Depending on people's legal status and relationship to the state and other actors, they may need to undertake costly or dangerous informal travel, using smugglers or traveling through remote territory to avoid authorities. Given these concerns, we expect that households facing higher mobility costs will be less likely to return.

Information

Finally, any effort to compare the prospects for household well-being in the host and home countries depends critically on the quality of information about conditions in the country of origin. Yet, refugees fleeing violence may lack access to information about localized conditions in their home country. Social networks are often disrupted, and those who remain behind may face strong disincentives to share information about local conditions, especially in repressive countries. Moreover, the situation on the ground can change often and differ even across small geographies. This makes it challenging for refugees to evaluate their potential well-being in the place of origin. In the absence of good information about conditions in their places of origin, we expect refugees to be less likely to return home.

The Syrian Refugee Crisis in Lebanon

To shed light on the drivers of refugee return intentions, we focus on the Syrian refugee crisis in Lebanon. (We later validate our findings using similar data from Jordan.) Given the diversity of localities in which Syrians have settled and the heterogeneity of conditions in Syria, this is a helpful case for examining the role of push and pull factors, mobility costs, and information in shaping return intentions. Lebanon, in particular, provides a critical test of the importance of push factors given the documented hostility, discrimination, and violence that many Syrians have faced in Lebanon (Lehmann and Masterson 2020). Advocacy groups have repeatedly reported that Lebanon's unlawful evictions, curfews, raids, and arrests make life miserable for Syrian refugees, forcing many to return to Syria despite its dangerous conditions.⁶ In addition, the context provides meaningful variation in prospects in Syria, mobility costs, and access to information. Syrians in Lebanon vary widely in their characteristics and backgrounds, originating from all of Syria's regions and spanning the country's prewar socioeconomic spectrum.

What began in Syria in 2011 with street demonstrations and calls for political reforms collapsed into a devastating civil war, which caused an enormous refugee crisis, with millions of people fleeing to Lebanon, Jordan, Turkey, Iraq, Egypt, and beyond. As of late 2019, when our study was conducted, more than 5 million Syrians had fled to neighboring countries and more than 6 million were displaced inside Syria. Approximately 930,000 Syrians lived in Lebanon, alongside 4.5 million native residents, in a small country with the smallest land area in continental Asia.⁷

One driving assumption behind Lebanon's national policy agenda for Syrian refugees is that exploitation, vulnerability, and material hardship will force Syrians to leave the country (Janmyr 2016). Syrians in Lebanon face widespread hostility, confront significant restrictions on the right to work, and have only limited legal status in the host country. Most Syrians in Lebanon lack reliable access to education, healthcare, stable housing, and safe transportation (see, for example, Lehmann and Masterson 2020; Mourad 2017). They live primarily in urban and peri-urban settings, with 15 per cent in camps that are informally managed by NGOs, as the UN did not establish official refugee camps in the country. The situation of Syrians in Lebanon is similar in many respects to the hardship that refugees face worldwide, notably, many governments restrict refugee rights in order to accelerate return, and less than one third of the world's 26 million refugees live in camps.

As the Syrian government regains control of much of the country, tens of thousands of Syrians have begun returning home, even as violence continues to displace more people. State and nonstate actors in Lebanon have begun taking steps to facilitate and push for the return of refugees, tensions between Lebanese and Syrians remain high, and discussions about the return of refugees are increasingly prominent in journalistic and policy circles. Looking to Syria, the war has devastated the country's infrastructure and public services, including water supply, electricity, schools, and healthcare. Many people fear the persecution and violence that may result from government retribution and collective punishment in the postwar period. Men aged 18–42 are subject to military conscription in Syria, and serving in the Syrian military is likely to put conscripts in violent situations for years to come. Even if the ultimate victor in the war is no longer in question, the specter of future violence remains.

Research Design and Data

Survey Design

We use original survey data from interviews with a nationally representative sample of 3,003 Syrian refugee households living in Lebanon.⁸ The survey measured a wide range of household characteristics, predictors of return, and migration intentions, and included a conjoint experiment to identify drivers of return intentions. The research team contracted a Lebanese survey firm to conduct data collection and participated in all stages of the research, including enumerator training, survey piloting, and oversight of data collection. Data collection for the survey took place from August to October 2019.

To obtain a representative sample of Syrian households in Lebanon, we used stratified random sampling to ensure variation in Syrian and Lebanese demographics in localities and households sampled. A household head (either gender) served as survey respondent. Section 1 in the Online Appendix provides a detailed discussion of sampling protocols.

Measuring Return Intentions

Measuring return intentions is challenging, and survey instruments must account for the different time horizons across which households consider decisions, in addition to the uncertainty that people face. Capturing intentions is also difficult in the absence of concrete behaviors consistent with stated intentions. As a result, we also measure preparations to return as a self-reported but behavioral manifestation of return intentions. We asked respondents about their return intentions in three ways:

• • Return intentions. We asked: “Do you plan to return to Syria in the next twelve months with the goal of staying there?”

• • Return preparations. We asked a battery of binary questions about legal, financial, and logistical steps to prepare for return. Metrics of preparation included saving resources for return, collecting such paperwork as birth certificates or marriage documents, reaching out to Lebanese authorities and the UNHCR about return, and taking scoping trips. We use these questions to calculate a preparations index with polychoric principal component analysis (PCA).

• • Long-term return intentions. We asked: “Do you hope to move back to Syria and live there one day?”

The return intentions measures are binary variables, coded as 1 if the head of household plans to return within the specified time frame and 0 otherwise.⁹ It is worth noting that our key outcomes are stated intentions and self-reported preparations to return, not a retrospective measure of return behavior. Such forward-looking outcomes are an important quantity of interest as people consider whether to return and policymakers design and implement programs to address refugee situations. A foundational principle of return policy is ensuring its voluntary nature, which requires placing people's intentions to return at the center of planning (see, for example, Mixed Migration Centre 2019, 93).

Measuring Drivers of Return

We measure four key concepts that we hypothesize will drive return decisions: (1) well-being in Lebanon; (2) prospective well-being in Syria; (3) information; and (4) mobility costs. To measure the first three concepts, we draw on data from multiple survey questions and use PCA to construct indices to capture aspects of respondents' living situation in Lebanon, prospects in Syria, and access to information. We present the full set of PCA inputs in Sections 2 and 3 in the Online Appendix.¹⁰ In both Syria and Lebanon, we measure economic well-being, using data on assets and earning potential in each country, and current employment, earnings, and aid in Lebanon. We also examine the availability of services, including education, healthcare, water, and electricity, in Lebanon and Syria. We analyze the size of social networks and the number of friends and family in Lebanon and Syria. We measure people's ability to move freely and safely around Lebanon, as well as their integration in the country, using the measures from the Immigration Policy Lab Integration Index (IPL-12) (Harder et al. 2018). To construct an index for the security situation in Syria, we focus on both general factors, such as whether there is still fighting, and personal factors, such as whether a family has any draft-aged men and whether the respondent personally experienced violence. The index on safety also includes an input about safety expectations in one year. The indices on economic conditions and services in Syria also include inputs that measure expectations about the future situation. We also construct an index for regime control, including detailed questions on which parties currently and formerly controlled a respondent's hometown.¹¹ The index for information includes whether the respondent speaks regularly with family or friends in Syria about the situation, as well as measures of people's confidence in the information they have about safety, jobs, services, and conscription in their hometown.

In addition to measuring people's confidence in information directly, we ask questions about the size of refugees' networks in the host and the home country. Family and friends may serve as important sources of information about the conditions in one's hometown. Networks of family and friends may also directly impact people's return choices independently of the information they provide, in the sense that many people want to live in the same place as others in their close network.

We study mobility costs using two metrics: travel distance to one's hometown and household size. We calculate travel distance from each survey respondent's town of residence in Lebanon to their hometown in Syria, via the Beirut–Damascus highway and border crossing, using the Google Maps Application Programming Interface (API). Our fieldwork revealed that this was the only legal border crossing open at the time of research and that a majority of Syrians moving back travel via official routes.

One potential concern with our survey is that affective biases may shape both people's self-reports of their situation and their return intentions, possibly leading to spurious correlations. To mitigate this threat, our metrics of well-being in Lebanon and prospects in Syria aim to measure objective facts rather than people's opinions. For instance, we ask respondents such questions as: “Did you work outside the home for money in the past 30 days?”; and “Do you receive humanitarian aid through the UN cash assistance program?” (rather than about affect, such as: “How would you rate the quality of your life in Lebanon?”).

Conjoint Experiment

We also present a conjoint analysis that experimentally manipulates potential drivers of return intentions.¹² This allows us to isolate the effects of conditions in Lebanon and Syria, individual circumstances, and social networks in shaping respondents' thinking about return. During the survey interviews, enumerators read to respondents a sequence of five separate hypothetical vignettes and, after each one, asked the respondents whether, under these conditions, they would return to Syria. In the vignettes, each of the following numbered attributes was randomly given one of the lettered values, and the order of the attributes was randomized across respondents. The vignettes were presented as follows:

Imagine that one year from now, regarding the security situation in Syria, [INSERT FROM (1) BELOW]. It appears that in [INSERT HOMETOWN], [INSERT FROM (2)]. As for conscription, [INSERT FROM (3)]. In Lebanon, [INSERT FROM (4)]. Finally, regarding your friends and relatives, [INSERT FROM (5)].^[13]

1. (1) Safety in Syria: (a) Your hometown is quite safe; (b) Your hometown remains insecure; (c) All of Syria is quite safe.

2. (2) Economic conditions in Syria: (a) There are many job opportunities; (b) Public services, such as health centers and schools, are relatively easy to attain; (c) There are few job opportunities; (d) Public services, such as health centers and schools, are difficult to attain.

3. (3) Personal safety: (a) Military conscription has stopped; (b) Military conscription is still in place.

4. (4) Conditions in Lebanon: (a) You have a good job in Lebanon; (b) You do not have a good job in Lebanon; (c) Health centers and schools in Lebanon are available and affordable; (d) Health centers and schools in Lebanon are unavailable and unaffordable.

5. (5) Network effects: (a) Most of your friends and relatives are in Lebanon; (b) Most of your friends and relatives are in Syria; (c) Most of your friends and relatives are in Jordan, Turkey, and Iraq.

Results: Observational Data on Return Intentions

We begin by describing our sample. Around 50 per cent of our 3,003 respondents reside in urban areas in Lebanon and 33 per cent live in informal settlements. The median year of arrival for respondents was 2013. The majority (80 per cent) are registered or recorded with the UNHCR. In terms of education levels, 49 per cent had an education level less than completing primary school, 39 per cent completed primary school, and 12 per cent had a secondary education or higher. As for aid, 48 per cent of respondents received cash transfers, 62 per cent received food vouchers, and 32 per cent received both. Discrimination toward refugees in Lebanon is quite high but far from universal. A total of 37 per cent of respondents reported living in towns that had curfews in the past two years (which usually target refugees) and 40 per cent reported facing discrimination when searching for housing. Finally, when it comes to conditions in Syria, 67 per cent of respondents reported that protests occurred in their hometown during the revolution and 96 per cent said that there was heavy fighting in their hometown at some point during the war. By the time the survey was conducted, 66 per cent of respondents said that their hometowns were controlled by the government.

We examine the distribution of return intentions in Figure 1. We find that return intentions are increasing with the time horizon. Only 5 per cent of Syrians plan to return in the next twelve months, that is, before approximately September 2020, and about a quarter of Syrians anticipate returning before September 2021. A total of 63 per cent plan to return at some point in the future.¹⁴ To put these numbers in context, the median year of arrival for respondents was 2013, meaning that the median respondent had been displaced for more than six years at the time of data collection.

Figure 1. Return intentions (short-, medium-, and long-term).

Note: The vertical lines represent 95 per cent confidence intervals.

To study how cross-sectional differences shape return intentions, we examine the predictive power of the range of potential drivers of refugee return described earlier. We estimate the following regression model:

(1)$$Y_i = \alpha + \beta T_i + \gamma X_i + \epsilon _i, \;$$

for each outcome Y and a vector of indices T. Each index is the first principal component from a PCA analysis of the measures detailed in the section titled “Measuring Drivers of Return” earlier. We also adjust for a range of control variables, X, including household-level covariates and locality-level fixed effects. Control variables were defined in the preanalysis plan and are presented in full in Section 2.3 in the Online Appendix. Finally, $\epsilon $ is a mean-zero error term. We also run a series of regression models similar to Equation 1 but where the vector of indices T is replaced with each respective index in one model.¹⁵

Drivers of Return Intentions

We present results for the drivers of return intentions in Figure 2. Each dot represents the point estimate for the relationship between a given index, labeled on the y-axis, and a metric of return, labeled at the top of each panel. Circles represent point estimates drawn from our main model in Equation 1, and triangles represent point estimates drawn from models with each respective index in a separate regression. The independent variables are grouped into four categories: people's prospective situation in Syria; people's living situation in Lebanon; mobility costs to return to Syria; and people's confidence in the information they possess about Syria. The horizontal line around each point estimate shows the 90 per cent and 95 per cent confidence intervals (dark and light, respectively). Standard errors are clustered at the locality level, following from the sampling strategy. Indices are normalized to have mean 0 and standard deviation 1, and the point estimates present the change in the probability of return intentions that corresponds to a one-standard-deviation shift in an index. As shown in Section 6.5 in the Online Appendix, results are robust to using additive indices, rather than PCA indices, and using alternative control sets.

Figure 2. Index results: effects on return intentions and preparations.

Notes: Each dot represents the effect on return intentions (left panel) and return preparations (right panel) presented with its corresponding 95 per cent (transparent lines) and 90 per cent (solid lines) confidence intervals. The triangles represent the regression that includes each index alone, as well as demographic controls. The circles represent the regression that includes all the indices in the same regression, as well as demographic controls.

Figure 2 provides strong evidence for a relationship between conditions in Syria and intentions to return within twelve months (first panel). We see that safety in Syria, economic prospects in Syria, the availability of public services in one's hometown, and respondents' family and friend networks in Syria are positively and significantly associated with return. For each of these indices, we see that a one-standard-deviation shift in the index corresponds with about a 2-percentage-point increase in return intentions. In light of the small fraction of refugees (only 5 per cent) who plan to return in the next year, this constitutes a large increase in return intentions in percentage terms (roughly 40 per cent). Control by the Syrian government correlates negatively with intentions to return, though we cannot rule out a null relationship at either the 90 per cent or 95 per cent level.

The relationship between conditions in Syria and preparations for return (second panel) is less clear but points in the same direction. Point estimates are consistently positive, but only the availability of services and the size of social networks are statistically significant. Security in Syria and economic prospects predict preparations to return, but the results are not statistically significant. Regime control has no detectable relationship with preparations to return.

The results on push factors in Lebanon are quite different. First, looking at the left panel, we do not find a clear correlation between well-being in Lebanon and return intentions. We cannot rule out a zero association for most of the indices. The one index that demonstrates a statistically significant association with return intentions is social well-being. In contrast to the lack of evidence for a role of push factors in shaping return intentions, the second panel reveals evidence for an association between conditions in Lebanon and return preparations. We find that higher levels of economic well-being, networks, and social well-being in Lebanon exhibit a detectable positive correlation with having taken steps to prepare to return to Syria in at least one specification. The direction of the relationship is not what we expected ex ante, based on a theory of preparations being driven by a simple utility comparison between conditions in Lebanon and prospects in Syria. The finding highlights that the theory's focus on migration costs and incentives may have overlooked migration capacities. Indeed, return is a complex and daunting process, and people with more financial and social resources may be better able to undertake a safe voluntary return.

Looking at the next group of drivers, we see in the first panel that the results do not provide evidence of a relationship between mobility costs and return intentions. In the second panel, we find a negative association between mobility costs and preparations for return, significant at the 10 per cent level, when we consider indices separately. Looking at the bottom row of Figure 2, we see that confidence in information about one's hometown is positively associated with both intentions and preparations. Information access may both have a direct effect on return intentions and play a moderating role. We examine this possibility in Section 6.6 in the Online Appendix, where we test whether conditions in Syria have a larger effect on people's intentions when they have high levels of confidence in their information about the situation in Syria. Results provide evidence that the relationship between conditions in Syria and return intentions and preparations is shaped by respondents' confidence in their information sources.

As one additional test of the findings in Figure 2, we fit predictive models based on home-country factors and host-country factors using tenfold cross-validation. We present the results in Section 6.7 in the Online Appendix and find that models based on conditions in Syria consistently demonstrate higher predictive power than models based on conditions in Lebanon.

In light of our theoretical priors, what explains the limited impact of host-country conditions on return intentions? One possible explanation could be that there is insufficient variation in the conditions of refugees in Lebanon. We do not find evidence in the observational data consistent with this possibility. For example, in Section 3.1 in the Online Appendix, the descriptive statistics demonstrate wide variation in the living conditions of Syrians in Lebanon. In Section 6.3 in the Online Appendix, we rerun all models that controlled for Lebanese locality-level fixed effects but without adjusting for locality. Our findings are robust to this alternative specification, suggesting that our null findings for the role of factors in Lebanon are not driven by a lack of variation in living conditions within localities. In Section 6.8 in the Online Appendix, we examine this further by producing a map of Lebanon that shows the variation in respondents' conditions by district using an index of all the components used to measure conditions in Lebanon. The map shows significant variation across districts: the difference in the index between the district with the, on average, worst and the best conditions for respondents is about 2.4 standard deviations. Of course, we cannot rule out the fact that host-country conditions might matter if our sample included a greater diversity of host countries (for example, Western Europe, the United States, Canada, and so on). However, given that most refugees are hosted in the Global South by neighboring countries, our observational analysis suggests that host-country conditions may not be very important determinants of decision making in these contexts. In the next section, we provide evidence based on even wider variation in home- and host-country conditions by leveraging a conjoint survey experiment.

Results: Conjoint Experiment

The analysis of observational data strongly suggests that pull factors are more predictive of return intentions than push factors. Yet, despite our extensive set of control variables, our correlational estimates might be affected by other factors not included in the model. In this section, we present the results of a conjoint experiment designed to provide greater leverage on the causal effects of these drivers on return intentions. We follow a standard approach for analyzing conjoint experiments, using ordinary least squares (OLS) regressions to estimate the average marginal component effect (AMCE) for each attribute (Hainmueller, Hopkins, and Yamamoto 2014). Figure 3 displays the effects on respondents' answers to the question: “Under these conditions, would you be willing to return to Syria?”

Figure 3. Conjoint experiment results.

Notes: Each dot represents the effect on the probability that respondents would return to Syria in a given hypothetical situation, presented with its corresponding 95 per cent (transparent lines) and 90 per cent (solid lines) confidence intervals. The empty circles indicate reference categories. We cluster standard errors at the respondent level.

The main findings from the conjoint experiment are consistent with our analysis of the observational data. On average, conditions in Syria play a more important role in shaping people's return intentions than conditions in Lebanon. Results suggest that safety is the most powerful driver of return, with security in one's hometown increasing return intentions by 35 percentage points and nationwide security increasing return intentions by 42 percentage points. The fact that safety in one's hometown has nearly as large of an effect as nationwide safety suggests that the majority of variation in people's consideration of security is driven by conditions in their hometown, highlighting the local nature of security concerns in postwar environments. An end to military conscription also plays an important role in shaping people's return intentions and increases the likelihood of return by 18 percentage points. In comparison, the availability of jobs and public services in Syria have a more modest effect: both increase return intentions by 8 percentage points.

Both access to a good job and public services in Lebanon play a small, negative role in people's return intentions. Someone with a good job in Lebanon is 2 percentage points less likely to return, and if someone has access to public services, they are 3 percentage points less likely to return. Despite the statistical significance of these estimates, the differences in magnitudes between push and pull factors is substantial.

At the bottom of the figure, we see the effect of networks on people's responses. People were nearly 5 percentage points more likely to say that they would return to Syria if they have family and friends there (compared to having people outside of Syria and Lebanon). In contrast, we see a precisely estimated null effect for having family and friends in Lebanon on people's return intentions. These network results align with our earlier findings about the relative importance of the conditions in the home country compared to the hosting country.

Beyond Lebanon: Return Intentions in Jordan

Syrian refugees migrated to numerous countries, including three primary hosting countries: Lebanon, Jordan, and Turkey. In order to ascertain whether our results are driven by unique circumstances among Syrians in Lebanon, we ran a separate survey with 1,286 Syrian refugees in Jordan. These data offer a test of the external validity of our findings to the broader population of Syrian refugees. Our sampling strategy selected individuals from the four metropolitan areas in Jordan with the largest refugee populations: Amman, Irbid, Mafraq, and Zarqa (including the town of Azraq). In the summer of 2019, enumerators interviewed a random sample of Syrians who received services from the NGO CARE during the study period. The participants were recruited from Syrian refugees living outside of camps, as do more than 80 per cent of Syrians in Jordan (Verme et al. 2015, 40).

The two cases make for a valuable comparison given some key similarities and critical differences. Similar to Lebanon, Jordan hosts a large number of Syrian refugees relative to its population, and public discourse in the country widely frames refugees as having large negative economic and fiscal impacts. Unlike Lebanon, national political discourse in Jordan at the time of the survey was not pushing aggressively for Syrians to return. Further, the baseline rate of return intentions for Syrians in Jordan is very low. When we asked Syrian refugees in Jordan if they plan to ever return to Syria, we find that a large majority of respondents (around 75 per cent) reported that they never want to return to Syria. Therefore, the data enable us to examine whether our results from Lebanon pertain only to a context with major pressure to return and where a large share of people hope to return home someday.

The difference in baseline return intentions between our samples in Lebanon and Jordan is likely driven by different selection into displacement to the countries. Similar to Lichtenheld's (2020) theory of assortative displacement, we can imagine that refugees sort into host countries based on a range of personal characteristics, including their relationship to a war's armed groups in the country of origin, and these views may influence the return decision. First, we see a difference between the two samples in self-reported level of security in respondents' hometowns. As of the summer of 2019, 51 per cent of the sample in Jordan said that their place of origin continues to be very dangerous. In contrast, only 28 per cent of respondents in Lebanon said so when we conducted our survey there a few months later in August–October 2019. Second, our fieldwork suggests that the political attitudes of Syrians living in Jordan tend to be more anti-regime, whereas the Syrian population in Lebanon is more divided in its views toward the Syrian government, which aligns with public opinion surveys on the topic (Corstange 2018).

Using our data from Jordan, we construct indices for dimensions of people's well-being in Jordan and prospective well-being in Syria. We then regress return intentions on the indices, as defined in the “individual indices” specification of Equation 1, to estimate the impact of each factor on people's stated plans to ever return to Syria.¹⁶

Figure 4 presents results from analysis of the Jordan data. Despite the sizable difference in baseline return intentions and the political climate, the drivers of return intentions in Jordan are strikingly similar to Lebanon. First, prospective conditions in Syria play an important role. We see that conditions in respondents' place of origin in Syria—specifically safety, economic prospects, and public services—are positively correlated with return intentions. Also, having family and friend networks in Syria is positively correlated with return intentions.

Figure 4. Index results: effects on plans to ever return in Jordan.

Notes: Each dot represents the effect on return intentions presented with its corresponding 95 per cent (transparent lines) and 90 per cent (solid lines) confidence intervals. We control for gender, age, household size, education, and female-headed households, as well as place of origin in Syria and locality in Jordan. Missing values were imputed using mean imputation.

Second, in line with results from Lebanon, we do not find strong evidence that conditions in Jordan drive return intentions. First, we see in Figure 4 that economic conditions, access to public services, social well-being, and legal conditions are not strongly associated with return intentions. Networks is the one dimension of conditions in Jordan where we find a relationship with return intentions. This contrasts with results from Lebanon, where social well-being is the only push factor that consistently predicts return intentions.

Finally, looking at the impact of information, we do not find evidence of a relationship between information quality and return intentions in Jordan. This contrasts with the evidence we found in Lebanon for the importance of information for Syrians' decision making about return.¹⁷

Toward an Alternative Framework for Understanding Refugee Return

In contrast to our predictions, we find that host-country conditions have little bearing on refugees' decision making about return. At least in this context, the “push” and “pull” framework, informed by traditional models of labor migration, has less explanatory power than anticipated. The empirical results challenge us to reconsider whether a model of migration developed for people weighing how to improve their economic situation applies to the return choices of people who have fled war to neighboring countries in the Global South. While we accept that any binary distinction between different categories of international movers, that is, “labor migrants” and “refugees,” is often too simplistic, even problematic, to stand alone as an explanation (see, for example, Abdelaaty and Hamlin 2022), our intuition is that the conditions that drive people's initial departure from their home country will structure their decision making about return in important ways. We highlight two features of wartime displacement that are particularly relevant for our context and discuss their implications for an alternative theoretical framework for understanding return.

First, when people flee violence and other consequences of war, they may depart based on a sense that they have no other choice in order to find safety and security. In general, these may not be households that would have chosen migration in the absence of conflict and violence at home. Second, the destination choices of refugees are often highly constrained. This is distinct from international movers who choose a destination country based on an opportunity to increase their income or achieve a related economic goal. Many refugees have to flee quickly to a subset of immediately accessible neighboring countries in search of safety.

As a result, we may need an alternative framework to understand refugees' decisions about whether or not to return. A different starting point is that if refugees flee their homes to avoid violence, destruction, and military service, living conditions in the home country may need to exceed a certain threshold before refugees are willing to return. If refugees flee to accessible host countries for the sake of safety, rather than to optimize their income, then challenging conditions in the host country may be unlikely to increase the probability of return before conditions at home have improved. This framework suggests a different set of predictions than those implied in the push–pull model: (1) return only occurs when safety conditions at home improve beyond certain thresholds; and (2) as long as these conditions are not met and refugees remain safe(r) in the host country, the difficulties they face there, as well as non-safety factors at home, will have a minimal impact on return.

One way to conceptualize the non-compensatory decision-making process underlying this alternative framework is to allow that (some) refugees do not trade off safety for other goods (such as income, access to services, and so on) as long as their safety concerns are not met. This conception of hierarchical preferences (see, for example, Marshall 1949; Scott 2002) over safety lies somewhere on the following continuum: on one end, we have the assumption of perfect substitutability of all kinds of goods embedded in standard choice models—a class of models that encompasses neoclassical economic theories of migration, the theoretical foundation for the push–pull model; on the other end, we have strict lexicographic ordering, which in this context implies that refugees maximize safety without regard to other goods. In contrast, this alternative framework assumes that refugees set a target (threshold) for safety that must be reached before other goods and factors are considered. As long as the safety conditions are not met, no amount of other goods will make the refugee indifferent between returning and not doing so.¹⁸ While other factors, such as risk aversion, are likely correlated with the safety threshold that any individual household might apply, for the moment, we treat the threshold as exogenous, though recognize that it may vary across refugees.

Theories of hierarchical preferences and non-compensatory decision making have a rich tradition in psychology and economics, which includes the satisficing theory of Simon (1966), the “elimination by aspect” model of Tversky (1972), and lexicographic trade-off structures like those considered by Luce (1978), among others. Research in psychology has provided ample experimental evidence that individuals indeed use such non-compensatory decision making in a range of choice situations, both with low and high stakes (see, for example, Gigerenzer and Todd 1999; Payne, Bettman, and Johnson 1993).

To explore the plausibility of this alternative framework for understanding refugee return, we look for suggestive evidence with two sources of complementary data: thirty-six qualitative interviews with Syrian refugees living in Lebanon and a reanalysis of the conjoint experiment discussed earlier. The qualitative, semi-structured interviews were conducted with a separate sample of Syrian refugees living in Lebanon between February 2020 and June 2021 by a member of the research team.¹⁹ The semi-structured interviews took place one on one in Syrian Arabic over encrypted WhatsApp calls. The discussions focused on people's migration histories, migration intentions and aspirations, and their process of decision making about the future. We analyze the qualitative data to provide an interpretative understanding of Syrian refugees' lived experiences of return decision making. Section 9 in the Online Appendix offers a detailed presentation of the qualitative data collection, as well as an ethics discussion.

The qualitative data support the proposition that people are waiting for the security and safety situation in Syria to improve before returning. Many respondents emphasized that while they want to return to Syria at some point, they would only do so after certain conditions are met. One respondent told us: “If the war ended in Syria, I'd think about returning, but as long as the war continues and the security situation isn't good, I won't return.”²⁰ Another respondent echoed the same sentiment, saying: “We can only return to our region when there is no longer war there.”²¹ Even when people strongly desired to return in order to reconnect with family or, in one case, to attend a father's funeral, many respondents said they could not return because of how difficult the situation had become.²² Another respondent explained that she never expected to stay in Lebanon as long as she did, but she remained because the war continued: “I said to myself we'd stay in Lebanon until the situation gets better in Syria, then we'd return. I thought that we'd stay for just a short period, then we'd return, but we're still here.”²³

The threshold model also implies that host-country conditions have little effect on return as long as safety concerns in the home country are not addressed. The qualitative data reconcile this prediction with the reality that harsh conditions in Lebanon cause real suffering for the people experiencing them. Respondents explained that while they are very unhappy with their circumstances in Lebanon, they feel like the conditions in Syria simply preclude the option of return. One respondent said that “Lebanon is a prison” but he would return to Syria only “once the crisis ends,” referring to the ongoing civil war.²⁴ Another respondent explained that she cannot afford rent in Lebanon, saying that “there's nothing available” for her or her family in Lebanon, but she still refuses to return to Syria without a significant improvement there, in particular, noting the military draft requirement for her husband.²⁵ Another respondent explained that even extreme hardship in Lebanon will not push them to return to a country at war:

[My husband] can't work very much. His health isn't good right now. He works a little, but it's not enough for a living.… In the beginning, when I came here in 2012 and 2013, [the UN] gave me food aid. They gave me a card and food supplies, but eventually they stopped. I was very dependent on that card. It was about 50 per cent of our food supply, but they stopped all that aid three years ago.… My homeland is at war. So, I can't return. But here [in Lebanon], I can't live.²⁶

We can also use data from our conjoint experiment to assess whether refugees' response behavior is consistent with predictions from the threshold model. As outlined earlier, the threshold model has two (testable) implications: first, if safety conditions are not met, return is unlikely; and, second, other push and pull factors only begin to play a role in refugees' decision making once safety exceeds the minimal threshold. Both the observational and the experimental evidence support the first prediction that, on average, safety concerns—in particular, the security situation in the hometown or country and the ongoing policy of military conscription in Syria—figure prominently in refugees' return decisions.

To explore the second prediction, we reanalyze the conjoint experiment from Figure 3 but subset the sample of the conjoint vignettes to those where conditions at home are described as safe (that is, vignettes in which either the hometown or all of Syria is described as safe and in which military conscription has ended) or unsafe (that is, hometown is described as unsafe and military conscription remains).²⁷ Focusing on the vignettes that describe conditions in Syria as unsafe, the left panel of Figure 5 shows the effects of other push and pull factors. In line with the second prediction, we find precisely estimated null effects for the availability of jobs and public services in Lebanon. We also document null effects for the same conditions in Syria. The only factors that have significant, albeit small, positive effects are social networks in Lebanon and, slightly larger, networks in Syria. The right panel shows the effects for the same factors but subsets to those vignettes that describe conditions as safe. In contrast to the previous estimates, we find that once safety concerns are addressed, the availability of jobs and public services in Lebanon reduces return intentions, while the availability of jobs and public services in Syria increases return. These results are not only consistent with the threshold model, but also suggest an important qualification to our main result based on the entire sample of vignettes: it is not the case that conditions in the host country do not factor into refugees’ decision making about return. Push factors matter too, but only if return is a safe option.

Figure 5. Conjoint experiment results by whether the hypothetical vignette mentioned that respondents' hometowns are unsafe and military conscription remains (left) or that their hometowns/all of Syria are safe and military conscription has ended (right).

Notes: Each dot represents the effect on the probability that respondents would return to Syria in a given hypothetical situation, presented with its corresponding 95 per cent (transparent lines) and 90 per cent (solid lines) confidence intervals. The empty circles indicate the reference categories. We cluster standard errors at the respondent level.

It is important to note that both the conjoint results and qualitative data, while consistent with, and parsimoniously explained by, the threshold model, cannot provide definitive tests against the standard choice model underlying the push–pull framework (Lancsar and Louviere 2006).²⁸ We offer suggestions for how to design future research to more directly test the threshold model in the next section.

Conclusion

We began this article by presenting novel evidence about the return plans of refugees in the short and long term using a representative survey of Syrians in Lebanon. We show that while a minority of refugees want to return in the short term, most refugees hope to return at some point in the future. We also advance understanding of the dynamics of refugee return by examining four major drivers of return at the household level: host-country conditions, home-country conditions, the cost of mobility, and the role of information. These initial hypotheses reflect the dominant model of decision making in the refugee context, which emphasizes how refugees weigh possible outcomes at home versus where they live now. We tested our hypotheses in the context of the Syrian refugee crisis—one of the largest refugee crises in the past century—in Lebanon using observational and experimental survey data from a representative sample of refugees, and explored the external validity of our results with a second survey in Jordan.

Our findings challenge the conventional view that refugees make return decisions by evaluating whether they can do better at home than in their hosting country. In particular, we find strong evidence that conditions at home matter most. By contrast, the dynamics in the hosting country do not have large effects, on average, on the return intentions of Syrian refugees. This finding, which stands in contrast to our registered expectations, is important because governments often restrict refugee rights based on the view that doing so will accelerate return (Janmyr 2016). To explain these results, we propose an alternative framework for understanding refugee return. A threshold model suggests that refugees do not trade safety in the country of origin for other goods, such as economic conditions and public service provision, as long as their safety concerns are not met. We find evidence for this alternative framework in both qualitative and quantitative data.

While these findings challenge the “push” and “pull” models that dominate discussions of return in policy contexts, we cannot evaluate the generality of our findings to contexts outside of a warring country's neighbors in the Global South. Results may differ in countries with robust social welfare systems or with clear pathways to citizenship (though, we note, a study on the drivers of return migration for Syrians in Germany did not find conditions in Germany to drive decisions [Kaya and Orchard 2020]). While this is an important scope condition for our findings, our results nonetheless speak to a significant proportion of the global refugee population. Worldwide, the vast majority of refugees live in developing countries and a similarly large portion live in countries neighboring their home country.²⁹

The article also raises a number of important questions for a growing research agenda on refugee crises and the dynamics of return. First, further work is needed to develop and validate the threshold model proposed in the article. Valuable extensions could explore direct questions and survey experiments to elicit refugees' thresholds and what safety and security mean to them. In particular, more fine-grained levels for the safety attribute will allow researchers to home in on the thresholds refugees apply. Further, exploring variation in thresholds across individuals—and even for a given individual over time—may prove fruitful. These correlates could include refugees' psychological traits (such as risk aversion and time preferences), socio-demographic variables (such as ethnic and religious identity), or political affiliation, as well as past experiences of violence and conflict.

Second, although previous research has explored why refugees seek out information about potential destinations (Holland and Peters 2020), little is understood about how refugees acquire and assess information about the situation at home. It is intuitive that high-quality information will condition migration choices, especially given the potential negative consequences of returning prematurely to a dangerous context. However, theories accounting for risk aversion would predict that the uncertainty that refugees have about the situation at home may lead them to underweight outcomes in the home country relative to those in the host country (see, for example, Kahneman and Tversky 1979). Given the complexity of our findings on information and the absence of a well-identified causal effect, further research is needed on how information quality influences return decisions.

Third, future research should explore the degree to which return intentions predict people's subsequent migration choices. Recent studies of labor migration with direct measurement of both migration intentions and behavior find that intentions are strong predictors of future emigration (see, for example, Docquier, Peri, and Ruyssen 2014; Tjaden, Auer, and Laczko 2019; Van Dalen and Henkens 2013). Future work should explore the conditions under which refugees' return intentions do and do not translate into behavior.

We conclude with two key takeaways for policymakers and humanitarian organizations. First, many refugees intend to return to their home country when threats to their physical, economic, and social well-being have decreased, and when they feel that they possess credible information. We find that more than two thirds of Syrians in Lebanon want to return home, and prospects for a good job and access to public services in Lebanon do not influence people's likelihood of staying. The findings also suggest that, at least while a home country remains unsafe, humanitarian agencies can support refugee well-being without undermining the goal of safe voluntary return. Efforts to deliver humanitarian assistance and provide broader economic opportunities appear unlikely to substantially change the calculus of refugees in the absence of improving conditions at home. Thus, continuing support and the provision of economic opportunities to the displaced can benefit both refugees and host societies, without exacerbating the long-term challenges of hosting governments.