Ethno-Caste Influences on Migration Rates and Destinations

Abstract

While studies commonly show differences in out-migration between ethnic groups, ethnicity most often features no more than a side note in the emigration literature, and we have very little insight about why people from different ethnic groups migrate at different rates. Understanding ethnic differences in migration rates and destination choice has important implications for the present-day and future potential for either dampening or exacerbating ethnic discrimination and opportunity structures. Building on existing migration theory, we identify three possible mechanisms through which ethnicity might influence out-migration rates and destination choice: human and economic capital, contemporary discrimination, and historical legacies that are perpetuated through social networks. Our empirical investigation uses longitudinal panel survey data from Nepal and we find that all three of these mechanisms likely influence out-migration and destinations of the five major ethno-caste groups. However, we show that historical legacy and human and economic capital emerge as the key drivers of ethnic differences in out-migration here. We discuss what these results mean for migration studies as well as the potential for the institution of migration to affect patterns of ethno-caste-based disadvantage in Nepal. The theoretical basis and empirical evidence from our study also suggest ways to understand the reasoning for and consequence of ethnic and racial differentials in migration patterns in other areas of the world.

Introduction

In this study, we explore how socially determined categories of ethnicity are related to both the likelihood of out-migration and destination choice. While ethnicity commonly appears in the out-migration literature, it is most often a side note or a control variable. We join Asad and Hwang (2019a) in noting that there is very little research that seeks to understand why there are ethnic differences in out-migration rates throughout the worldⁱ. One reason for this might be the predominance in the literature of economic resources and education as key predictors of out-migration. Given the strong and documented relationship in many countries between ethnicity and economic resources/education (Barron, 2008; Central Department of Sociology/Anthropology, 2014; Sen, 2000) and the likewise strong and documented relationship between economic resources/education and out-migration (Bhandari, 2004; Bhandari & Ghimire, 2016; McKenzie & Rapoport, 2007; Stark & Taylor, 1991; VanWey, 2005; Williams, 2009), it could logically be assumed that any ethnic differences in migration rates and destination choice result from ethnic differences in economic resources and education. Indeed, when Asad and Hwang (Asad & Hwang, 2019a, 2019b) find lower rates of documented migration to the United States for people from communities in Mexico that are characterized as “indigenous”, they attribute this largely to community level social and economic disadvantage. We follow their lead, but examine the question at an individual level—are individuals from different ethnic groups more or less likely to migrate and more or less likely to migrate to specific destinations?

Examining this question at an individual level creates additional theoretical pathways to consider. We build on a long history of social science showing that the social consequences of ethnicity are much broader than just economics and education. In taking a more comprehensive perspective, we address multiple mechanisms that might drive this relationship and how they might sum up to create differences in migration rates and destination choices between ethnic groups. The mechanisms we focus on are educational and economic resources, contemporary discrimination, and historical legacies of migration perpetuated through social networks.

Our study is based in Nepal, home to a variety of ethnic groups that have long been and continue to be treated differently due to a Hindu-based caste system. While rates of out-migration are very high in general, surveys show substantial variation by ethno-caste in out-migration and destination choice, making Nepal an interesting case for the investigation of ethnicity and out-migration. We use detailed longitudinal data from the Chitwan Valley Family Study (CVFS) (Axinn et al., 2019) and multinomial logistic regression models to examine why some ethno-caste groups are more or less likely to migrate and why some are likely to migrate to high or low earning potential destinations.

Understanding the multiple mechanisms that might simultaneously influence ethnic differences in migration behavior is not just an academic exercise; it can have important implications for understanding present-day social hierarchies and the potential for change(Howell, 2017; Karell, 2014). Migration is common in many countries around the world, with an estimated 272 million international migrants in 2019 (International Organization for Migration, 2019) and a substantial majority of these migrants are from poor countries where remittances are a key strategy for sustaining a basic standard of living. Social and political remittances are also important consequences of migration with significant impacts on origin communities. Numerous studies find that returned migrants from disadvantaged ethnic groups have used the money and prestige from migration to renegotiate traditionally degrading occupations, change consumption practices, and actively resist domination and discrimination (Adhikari & Hobley, 2015; Carswell & De Neve, 2014; Gidwani & Sivaramakrishnan, 2003; Ilahiane, 2001; Kurien, 2002; Levitt, 1998; Osella & Osella, 2000; Sunam, 2014). Given the relatively clear case that migration influences the well-being of individuals and their families—aired with the high rates of migration today—it seems likely that migration would impact the status of ethnic groups at the broader level of society. However, in order to understand if that happens, we empirically investigate whether disadvantaged ethnic groups have higher or lower rates of migration, particularly to more profitable destinations. This informs our discussion about the potential for migration to influence patterns of ethno-caste-based stratification in Nepal.

Theoretical Framework

Our primary purpose in this study is to understand the constellation of mechanisms that create ethnic differences in migration likelihood and destination choice. For a thorough analysis and in line with recent scholarship, we take a broad view of migration theory, considering the well-known economic factors as well as the more and less known non-economic factors that could mediate the relationship between ethnicity and migration. Economic factors are generally well known as they almost entirely dominated the earlier period of theorizing about migration (Massey et al., 1993; Sjaastad, 1962; Stark & Bloom, 1985; Stark & Levhari, 1982; Stark & Taylor, 1989; Stark & Taylor, 1991; Taylor, 1986, 1987; Todaro, 1969; Todaro & Maruszko, 1987). Although these economically-based theories explain much about migration around the world (Donato, 1993; Greenwood, 1985, 2016; Liang & White, 1997; Lundquist & Massey, 2005; Massey et al., 1993; Massey et al., 1994; Massey & Espinosa, 1997; Reed, Andrzejewski, & White, 2010; Ricketts, 2014; VanWey, 2005; Wright & Ellis, 2016), recent decades have shown that a variety of non-economic factors also have key impacts on migration. For example, studies have shown that social factors like networks, cumulative causation, and culture (Curran, Garip, Chung, & Tangchonlatip, 2005; Fussell & Massey, 2004; Garip, 2008, 2012; Garip & Asad, 2016; Kandel & Massey, 2002; Massey, 1990a, 1990b; Massey et al., 1998; Osella & Osella, 2000; Williams et al., forthcoming), as well as ideational factors like community satisfaction, intentions, aspirations, and fatalism (Bach & Smith, 1977; Bjarnason & Thorlindsson, 2006; Carling & Collins, 2018; Creighton, 2013; De Jong, 2000; De Jong et al., 1983; De Jong, Richter, & Isarabhakdi, 1996; Deane, 1990; Fischer & Malmberg, 2001; Irwin, Blanchard, Tolbert, Nucci, & Lyson, 2004; Lee, Oropesa, & Kanan, 1994; Marrow & Klekowski von Koppenfels, 2018; Schewel & Fransen, 2018; South & Crowder, 1997; Speare, 1974; Speare, Goldstein, & Frey, 1975; Speare, Kobrin, & Kingkade, 1982; Stinner & Van Loon, 1992; Thissen, Fortuijn, Strijker, & Haartsen, 2010; Thornton, Bhandari, et al., 2019; Thornton, Williams, et al., 2019; Uhlenberg, 1973; van Dalen & Henkens, 2013) have substantial influences on migration and destination choice. Following this contemporary trend in migration studies to think broadly about economic, social, and ideational influences, we identify three key mechanisms that could create ethnic differences in migration and destination choice—human and economic capital, contemporary discrimination, and historical legacy.

Human and Economic Capital

We begin with two studies by Asad and Hwang (Asad & Hwang, 2019a, 2019b) that examine how living in a Mexican community with a high proportion of ethnically indigenous people influences migration to the United States. They find that indigenous community origin does not influence the likelihood of migration, but it does impact whether individuals migrate with documentation or not. Drawing on neo-classical, new economics, and social networks theories of migration, they make a strong case that the social and economic disadvantages of living in an indigenous community impact the ability to obtain documentation for migration to the U.S. for all who live there, regardless of their own ethnicity. While the Asad and Hwang studies theorize and measure indigeneity as a characteristic of a place, it is not a difficult logical leap to apply the same reasoning to the individual. In that case, the expectation would be that Mexican individuals who are ethnically indigenous would be less likely to migrate to the United States because of economic and social disadvantages. Two other studies have examined the ethnicity-migration question on an individual-level. Based in India and Nepal Tsujita and Oda and Gurung (Gurung, 2012; Tsujita & Oda, 2014) find ethno-caste differences in migration rates and, similar to Asad and Hwang, attribute this largely to the mediation of education and economic capital. While they are not able to test this causal assertion, the logic is again compelling.

In many (if not most) countries, minority ethnic groups experience lower educational attainment, less wealth, and lower incomes. We also know that education, wealth, and income are strong predictors of migration outcomes (Bhandari, 2004; Caldwell, 1969; McKenzie & Rapoport, 2007; Stark & Taylor, 1991; VanWey, 2005; Williams, 2009). This suggests that systematic differences in education, wealth and incomeⁱⁱ across ethnic groups will create systematic ethnic differences in the likelihood of migration on an individual level. Further, given that some destinations require more financial input (for travel, resettlement, and/or employment contracts), and others have high educational requirements for legal entry, we can expect human and economic capital to mediate the relationship between ethnicity and destination choice as well.

Discrimination

Also useful for our purposes are studies of the “Great Migration” of African Americans from the southeast to the northeast and midwestern U.S. in the early-mid twentieth century. While economic capital is generally believed to be part of the calculus, many also conclude that discrimination (in the form of lynching and other racialized violence, political disenfranchisement, behavioral restrictions, and inferior educational and employment opportunities) were key factors that pushed African Americans to migrate (Tolnay, 2003).

Classic migration theories also suggest that discrimination might affect migration. The neo-classical economics theory of migration (Sjaastad, 1962; Todaro, 1969; Todaro & Maruszko, 1987) suggests that people are likely to migrate when they can expect higher wages elsewhere. Extending this beyond economics, we can take from this that when people can expect better outcomes (in terms of wages, but also in terms of other social benefits), they will be more likely to migrate. Furthermore, this also suggests that, ceteris paribus, migrants will choose destinations with the best expected economic and social outcomes. Bringing discrimination into this discussion, we can predict that if an individual experiences discrimination in any aspect of life where they live, they might expect better outcomes elsewhere and thus be more likely to migrate away than individuals who are more advantaged at the origin.

Historical Legacy

While contemporary pressures might encourage or discourage migration and choices of particular destinations, the phenomenon of migration also has a distinct historical character due to the strong influence of social networks. A large body of research shows that if a person has more household or community members in their social network who have migrated, then they will be more likely to migrate themselves (Curran & Rivero-Fuentes, 2003; McKenzie & Rapoport, 2007; Palloni, Massey, Ceballos, Espinosa, & Spittel, 2001; Sjaastad, 1962; Todaro, 1969; Todaro & Maruszko, 1987). Further, as social networks perpetuate more migration, the behavior becomes normative, leading to what is now called “cumulative causation” and a “culture of migration” (Durand & Massey, 2004; Kandel & Massey, 2002). Social networks can also influence destination choice, whereby one is more likely to migrate to the same destination as those in their social network (Williams et al., forthcoming). In the case of ethnic groups, if a person’s social network is comprised heavily of others in their same social group (McPherson, Smith-Lovin, & Cook, 2001; Mollica, Gray, & Treviño, 2003) and if members of their own ethnic group have migrated, they might be more likely to choose the same destinations as others in their ethnic group.

Many ethnic groups have different histories of migration and destination choice, stemming from events or processes of differential treatment in the far distant or recent past. As social networks grow within specific historically entrenched migration streams and as they are perpetuated by the process of cumulative causation, this can link distant events or differential treatment to the present day, even if the conditions that caused the original migration are no longer present (Hatlebakk, 2016). In other words, the well-known social networks effect is essentially a narrative of historical legacy, whereby migration of particular people to particular places in the past perpetuates those migration and destination choice patterns into the present-day.

Just as social networks and cumulative causation might increase migration for ethnic groups in which migration is common, they might also function to depress migration for groups where migration is uncommon or place attachment is high (Barcus & Brunn, 2009; Li & McKercher, 2016). Specifically, many ethnic groups claim particular areas as their historic homeland. If an individual lives in their claimed homeland, we can expect they will have higher place attachment and will be less likely to migrate away. Alternately, if one lives outside their claimed historic homeland, we might expect that social networks and normative proscriptions on behavior will make them more likely to migrate to that homeland than other places. Thus, in the same way that social networks make migration normative for some, they can also make it unusual or transgressive for others.

Setting: Migration and Caste in Nepal

The three mechanisms we discuss above are not mutually exclusive or competitive. Instead, it is possible that ethnic differences in migration could be influenced simultaneously by the mediation of human and economic capital, contemporary discrimination, and historical legacy. Thus, we might expect that empirical outcomes will display elements of each of these mechanisms. In seeking to identify whether each mechanism plays a role in ethnic differences, it is essential to examine the particular context of a study in order to determine what would be evidence of discrimination, human and economic capital, or historical legacy effects. We undertake this exercise next, with a description of the setting of our study area in Nepal.

Migration

Nepal has a long history of out-migration and is now a major migrant-sending country. International labor migration formally began in the early nineteenth century with the recruitment of youths from the hill regions (particularly the Gurung and Magar ethnicities) into the British Army Gurkha Brigades (Rathaur, 2001). Since then, migration to India has been regular, with the same hill ethnicities making up a substantial portion of Nepali migrants to India.

In 1985, the Nepali government began encouraging migration to non-Indian destinations by licensing non-governmental institutions to export Nepalese workers abroad. The result was an explosion of international migration outside of India by all ethnic groups. Recent estimates suggest that there may be as many as three million Nepalis, or about 10 percent of the total population, in 131 countries at any time (Government of Nepal, 2014; World Bank, 2009). Of course, not all destinations are equal in terms of possible remuneration, living standards, accessibility, educational requirements, and influences of social networks. Destinations within Nepal and India are relatively accessible, require lower travel costs, and have relatively similar cultural conventions. Travel to India does not require a visa or passport and there are no restrictions on employment. Most jobs require menial labor, and wages are generally low. Data from the Chitwan Valley Family Study (CVFS) show an average monthly wage of 181 USD for Chitwan migrants in India and 192 USD for other places in Nepal.

Middle East destinations, are farther away, require higher initial investment (for travel costs and securing employment contracts through labor recruiting agencies), and are culturally dissimilar to Nepal. Common jobs include unskilled labor in construction, trucking, and service, and working conditions are extremely difficult with long work hours in very high temperatures. However, wages there are much higher, with an average monthly wage of 349 USD.

Countries in Europe, North America, and East and Southeast Asia are the most desirable destinations in terms of higher wages (average 1180 USD per month) and better working and living conditions (Seddon, Adhikari, & Gurung, 2002). However, access to these places is difficult due to the distances involved, travel costs, and restrictive immigration policies.

Ethno-Caste in Nepal: Historic Migration Patterns and Discrimination

Ethnicity and caste affect almost every aspect of life in Nepal. The population is comprised of 126 ethnic groups of Indo-Aryan and Tibeto-Burmese origins (Central Bureau of Statistics (CBS), 2012). In 1854, a new civil code institutionalized caste for every ethnic group (some of which previously had not had caste) and placed each in a hierarchy of inclusion and exclusion based on the Hindu caste system, similar to that in India. Although discriminatory practices based on caste were legally abolished in 1962, caste-based discrimination is still rampant today (Bennet, Dahal, & Govindasamy, 2008).

Brahmin and Chhetri peoples, generally termed ‘high caste Hindu,’ hail from all regions of Nepal and have enjoyed the best access to various economic, social, and political resources (Bennet et al., 2008; Bista, 1991; Dahal, 2003). In addition to being involved in government and politics, they have also long served as ambassadors and in other high-level positions in the foreign service. Thus, while most Nepalis were precluded from leaving the country for centuries, there is a history of some Brahmin and Chhetri people (along with their household staff of the same caste) traveling to many countries around the world, especially wealthier and more politically powerful countries where the Nepali foreign service was located.

The Newar ethnic group is lower in the caste hierarchy than the high caste Hindu people. But given their claimed origin in the Kathmandu Valley—the seat of the Nepali government—Newars have long enjoyed substantial advantages, opportunities, and involvement in business, trade, and politics (Bennet et al., 2008; DFID & World Bank, 2006). Kathmandu Valley not only has influenced the employment advantages of Newars but is also a central figure in their ethnic identity and serves as a homeland to be proud of and a place to ‘return’ to (Gellner, 1986), with important implications for migration.

Hill Janajati people (including Gurung, Magar, Rai, Limbu, and others) historically come from the middle hills and mountainous regions across Nepal. They are considered lower in the caste hierarchy than the Brahmin/Chhetri people and enjoy fewer advantages than those of the Newars. In terms of migration, it is the Hill ethnicities (specifically Gurungs and Magars) that comprised approximately 99 percent of the British Gurkha regiments since the 1800’s (M. Pariyar, 2016) and thus have a long history of migration to India.

Terai Janajati comprise several different ethnic groups (Tharu, Darai, Kumal, Bote, and others) that are indigenous to the lowland terai area of Nepal that borders India. Similar to indigenous groups in other parts of the world, Terai Janajati people are thought to be somewhat ethnically different than the other ethno-caste groups. They are also more residentially segregated and tend to live in villages that are exclusively of their own group. While they are in the middle of the Hindu-based caste system, occupying ranks similar to the Hill Janajati people, they are nonetheless generally heavily disadvantaged in social, economic, health, and political spheres. In the Chitwan study area, most of the Terai Janajati people come from the Tharu ethnic group (Guneratne, 2002). Of all ethnic groups, it is the Tharu who are indigenous to the Chitwan area. Indeed, Chitwan and other parts of the Nepali terai are called by some “Tharuwan” meaning “Tharu homeland”. Just as with the Newars and Kathmandu, Chitwan and the surrounding terai areas serve a central role in Tharu identity formation. Tharus and other terai indigenous groups are also closely attached to their land and farming lifestyles.

Dalit is the term for the caste groups that occupy the lowest rank in the Hindu caste hierarchy, were previously known as “untouchables,” and are heavily disadvantaged in almost all aspects of life (Bennet et al., 2008). Because they are considered impure, religious customs dictate a bevy of restrictions on their interactions with people in higher caste groups. Peoples of higher caste are not supposed to eat or share cooked food or water touched by Dalits. Dalits are often not extended the traditional respectful greeting “Namaste” and are not supposed to enter the homes of higher caste people. Thus, in addition to the tangible benefits of society (access to housing, employment, education, health, etc.), they are also generally excluded from social relations and civic participation. Although Dalits just as skilled as other caste groups, this social exclusion makes it very difficult for those who migrate within Nepal to negotiate employment and housing. However, a published study (B. Pariyar & Lovett, 2016) and our ethnographic experience show that some Dalit migrants to India avoid caste-based discrimination there by changing their family names to hide their caste.

In summary, ethno-caste groups in Nepal differ in terms of geographic origination (claimed homeland), historic migration patterns, economic and educational disadvantage, and discrimination. In terms of disadvantage and discrimination, they can generally be arranged the following order of progressive disadvantage: 1) Brahmin/Chhetri, 2) Newar/Hill Janajati, 3) Terai Janajati, and 4) Dalit. In terms of geography, the ethno-caste groups can generally be arranged by historic/claimed homeland as follows: Newar-Kathmandu, Terai Janajati-Chitwan, Brahmin-Chhetri/Hill Janajati/Dalit-all Nepal. In terms of historic migration patterns, the general patterns are Brahmin-Chhetri-international, especially wealthy geopolitical powers, Hill Janajati-India, Newar/Dalit/Terai Janajati-nowhere.

Data

Our data comes from the Chitwan Valley Family Study (CVFS) (Axinn et al., 2019), collected in the western Chitwan District, which lies in the south-central part of Nepal. Chitwan was historically inhabited by the indigenous Tharu people of the Terai Janajati ethno-caste. However, in the 1950s, the Nepali government cleared the forest and encouraged in-migration of people from the mountains and hills by offering land grants. Present-day Chitwan is now home to the Tharu people as well as many other ethno-caste groups who moved into the region within the last few generations. Outside the city of Narayanghat, which is situated in the northern corner of our study area, most residents live in small villages and sustain themselves with a combination of agriculture and animal husbandry.

The CVFS encompasses a population-based representative sample of all individuals aged 12–59 who were living in the western Chitwan Valley in 2008. Our study period is 2008–2012. A baseline survey was conducted in 2008 with all residents in sample neighborhoods aged 15–59. Those aged 12–14 in 2008 completed a baseline survey when they turned 15. The response rate was 97 percent.

Baseline respondents residing in Nepal (either in Chitwan or elsewhere in Nepal) were re-interviewed thrice yearly through 2012, with 93 percent of this group interviewed in all waves. All respondents who moved outside Nepal between 2008 and 2011 were re-interviewed before and after their moves. Migrants were asked extra questions about their migration experience, including the timing and destination.

The CVFS also conducted thrice yearly household-level interviews that provide continuous demographic information about each respondent. In particular, these household interviews provide a history, with monthly precision, of all moves. Thus, we have data from household interviews for an average of 49 months of demographic events for our study sample (with the exact number of months depending on the exact dates of the interviews). Because very few households moved in their entirety, were lost to follow-up, or refused, we have these reports for 98 percent of our original sample.

With reports of migration from both respondents and household interviews, the CVFS provides two migration reports for almost all respondents. A comparison of the two reports reveals a high degree of consistency, with 98 percent of respondents reporting the same migration outcome for themselves as their household members reported for them. Even when the comparisons are limited to only respondents who migrated during the study period, the individual and household reports provided the same first destination status for 91 percent of the migrants. These comparisons provide substantial confidence in the quality of the migration information reported by household informants. For our analyses, we use the information about migration obtained from the migrants themselves when that information is available (94 percent of the sample). When it is not, we use reports of the household interviews. Thus, we have nearly complete migration information for 97 percent of the original respondents.

The full sample of CVFS respondents eligible for our study is 4936 people. However, the tables we present in this article do not include respondents for whom household resources information (described below) is not available; thus, the final analytic sample that we use is 4406 people. Models that include the full possible sample of 4936 people, with indicators for missing household resources information, produce results that are substantively equivalent to those presented here.

Outcome Measures

We use two measures of migration: (i) any migration outside Chitwan (dichotomous), which is measured as any first time (after 2008) departure from Chitwan lasting six months or more, and (ii) destination-specific migration, which categorizes migrations into different destinations, including: no migration, migration in Nepal (but outside Chitwan), to India, to the Persian Gulf, and to wealthy Westernⁱⁱⁱ and Asian countries, which we call “WWA”.

The main international destinations that we identified as being in the Persian Gulf are Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, and the United Arab Emirates. Seven percent of the migrants in this Persian Gulf category went to other countries: Afghanistan, Iraq, Israel, and Lebanon. The primary destinations in the WWA category were Australia, the United Kingdom, the United States, Japan, Malaysia, South Korea, and Thailand, with these seven countries accounting for 86 percent of all migrants in this category. The remaining 14 percent were divided evenly among Belgium, Belize, Canada, China, Finland, Germany, Iceland, Maldives, Poland, Portugal, and Spain. Although these countries represent a wide range of geographical locations and cultures, they are all quite culturally dissimilar from Nepal and all are classified by the World Bank as upper middle or high-income countries, whereas Nepal is categorized as a low-income country.

Our choice of analyzing the first time a person migrates out of Chitwan during the study period is important. It would be possible to analyze any migrations during the study period (to include second, third, or higher migration spells). However, with subsequent migration spells, the relationship between migration and our predictor variables (like work experience and household resources) becomes increasingly endogenous. It would also be possible to analyze the first migration ever, by excluding from the analysis all respondents who had ever migrated before the study period began. While this would greatly reduce our sample size, it is also problematic for a different reason. Similar to many other places, migration in Nepal over the life course tends to follow a stepwise pattern (Paul, 2011) whereby many people first migrate shorter distances (within Nepal or India) and later to farther destinations as they gain experience. Accordingly, including only people with no migration experience could produce misleading results, particularly for the farther destinations. Thus, our strategy is to examine first migration during the study period as a middle ground that tempers the different limitations of these other strategies.

Explanatory Measure

Our measure of ethno-caste is as reported by respondents and grouped into five categories: (a) Brahmin-Chhetri, (b) Dalit, (c) Hill Janajati, (d) Newar, and (e) Terai Janajati. As the Brahmin-Chhetri group are the most advantaged ethno-caste group, we use it as our reference category in all models.

Human and Economic Capital

One of the main mechanisms that we propose links ethno-caste to migration is human and economic capital. We can directly investigate this with several measures that are strongly suggested by the existing migration literature and that we can measure well with this study. For human capital, we include a time-varying measure of education, categorized to match the primary credentialing points in the Nepali school system (0, 1–5, 6–8, 9–10, and 11+ years), and a time-varying measure of non-family work experience (never worked, wage only, any salary). For economic capital we used a variable that is a composite of land ownership, livestock ownership, housing quality, and income. Each of these indicators provides partial measures of economic status and together provide a more comprehensive record. For each of these four measures, we logged the indicators to correct for skewness, calculated a z-score for each of the logged variables, and then added the z-scores. We also use a variable for relative household resources. This was done by comparing the household resources of the respondent’s household with the resources of the households in their neighborhood. For this measure, respondents were divided into thirds: the lower third of relative household wealth, the middle third, and the upper third. Our economic resources variables are non-time varying and measured in 2006.

Controls

We control many other factors that have been shown in this and other study areas to have important influences on migration. Individual level controls include: gender, age (a time-varying measure categorized into nine groups – 14–19, 20–24, 25–29, 30–34, 35–39, 40 plus), marital status (a time varying measure – ever married and never married), and individual migration experience before the baseline interview (no migration, domestic migration only, any international migration). Household level controls include the total number of household members (a time varying measure) and a time varying measure of household level migration (percent migrants logged). Finally, the community level controls include community level migration experience (time varying, percent migrants logged) and the distance to the main urban area of Narayanghat (in miles).

Analytic Strategy

We first present bivariate results, comparing ethno-caste to migration rates and destinations. This presentation of the gross effects of caste on migration is key to understanding our broader concern in this article of how ethno-caste variation in migration might influence social change in Nepal.

Next, we seek to understand whether the reasons for these differences are due to human and economic capital, contemporary discrimination, or historical legacy. To accomplish this, we estimated multivariate logistic regression equations to model the monthly hazard of out-migration. Table 3 tests the dichotomous outcome of migration to any destination versus no migration. We progressively add controls for education, work experience, and land ownership, enabling us to examine how these human and economic capitals mediate the relationship between ethno-caste and migration. The models in Table 3 use discrete-time event history models with person-months as the unit of analysis.

Table 3.

Logistic regression estimates of the likelihood of migration (to any destination).

Measures	Model 1	Model 2	Model 3	Model 4
Caste/ethnicity
Brahmin-Chhetri	Ref	Ref	Ref	Ref
Newar	1.14 (0.86)	1.19 (1.15)	1.20 (1.22)	1.20 (1.23)
Hill Janajati	1.05 (0.42)	1.15 (1.22)	1.15 (1.23)	1.14 (1.25)
Dalit	0.99 (0.5)	1.09 (0.65)	1.05 (0.38)	1.05 (0.35)
Terai Janajati	0.69** (3.02)	0.78* (1.96)	0.75* (2.23)	0.75* (2.27)
Individual characteristics
Gender (1=female)	0.43*** (10.11)	0.45*** (9.40)	0.46*** (8.62)	0.46*** (8.59)
Age: 14–19	Ref	Ref	Ref	Ref
20–24	1.97*** (6.53)	1.80*** (5.26)	1.74*** (4.93)	1.74*** (4.95)
25–29	0.97 (0.17)	0.91 (0.58)	0.86 (0.88)	0.86 (0.90)
30–34	0.61** (2.76)	0.60** (2.73)	0.56** (3.07)	0.57** (2.96)
35–39	0.37*** (4.84)	0.38*** (4.54)	0.36*** (4.81)	0.36*** (4.79)
40+	0.12*** (11.0)	0.14*** (9.52)	0.13*** (9.76)	0.13*** (9.64)
Ever married	0.72** (2.60)	0.76* (2.14)	0.76* (2.20)	0.75* (2.23)
Personal migration experience: None	Ref	Ref	Ref	Ref
Domestic migration only	1.54*** (3.68)	1.58*** (3.89)	1.57*** (3.80)	1.56*** (3.76)
Any International migration	2.67*** (7.77)	2.71*** (7.77)	2.59*** (7.04)	2.56*** (7.02)
Education: no school		Ref	Ref	Ref
1–5 years of school		1.87** (3.17)	1.83** (3.04)	1.86** (3.11)
6–8 years of school		1.61* (2.43)	1.58* (2.34)	1.60* (2.37)
9–10 years of school		1.91** (3.23)	1.90** (3.19)	1.90** (3.18)
11+ years of school		2.16*** (3.90)	2.14*** (3.82)	2.14*** (3.77)
Work experience: None			Ref	Ref
Wage work only			1.16 (1.54)	1.16 (1.49)
Any salary work			1.23+ (1.79)	1.22+ (1.74)
Household characteristics
Total # of HH members	0.93* (2.42)	0.92** (2.62)	0.92** (2.66)	0.92** (2.66)
Log % HH member migration	1.03 (0.84)	1.03 (0.75)	1.03 (0.86)	1.04 (0.90)
Household Resources				0.99 (0.27)
Relative HH Resources: Lower Third				Ref
Middle Third				1.02 (0.16)
Upper Third				1.04 (0.26)
Neighborhood characteristics
Distance to Narayanghat (miles)	1.01 (1.27)	1.01 (1.34)	1.02 (1.36)	1.02 (1.50)
Log % NBH member migration	1.05 (0.35)	1.03 (0.21)	1.05 (0.33)	1.09 (0.62)
Number of P Periods	193233	193233	193233	193233
Number of Moves	722	722	722	722
Fit Statistics
AIC	8821.16	8809.24	8808.87	8814.32
−2 Log L	8783.16	8763.24	8758.87	8760.32

Odds ratios (OR) shown. Z-statistics in parentheses.

Significance:

⁺p<0.10

^*p<0.05

^**p<0.01

^***p<0.001

One-tailed tests for caste/ethnicity, two- tailed tests for controls.

In Table 4 we test destination outcomes. When we consider destinations, we must acknowledge that migration decision-making could be a one- or two-step process and must be modeled accordingly. A one-step decision process is where people decide to migrate and decide on a destination all at once, e.g. ‘I want to migrate to Australia’. A two-step decision process is where people decide to migrate and then subsequently decide where to migrate, e.g. ‘I want to migrate. Where should I go? I think I’ll choose Australia’. There is not substantial evidence whether migration decisions are largely made through one- or two-steps, and our fieldwork suggests that both happen in Nepal. As such, we analyze both for a thorough understanding of the relationship between ethno-caste and migration. Table 4 presents results from models that include the full analytic sample (as in Table 3), but include a five category outcome (no migration, within Nepal, to India, to the Persian Gulf, to WWA), with no migration as the reference outcome. This essentially models a one-step migration decision making process. We also present Table 5, which is equivalent to Table 4 except that the reference outcome is migration to WWA. Again, this models a one-step decision-making process. The results in Table 5 are substantively equivalent to those in Table 4, but with a different reference outcome (no migration in Table 4 and WWA in Table 5), we can glean slightly different information that is helpful to sorting out the complex relationship between ethno-caste and migration.

Table 4.

Multinomial logistic regression estimates of destination choice. All respondents included (4406 people). Reference outcome is No Migration.

	Model 5				Model 6
Measures	Within Nepal	India	Persian Gulf	WWA	Within Nepal	India	Persian Gulf	WWA
Caste/ethnicity
Brahmin-Chhetri	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Newar	1.38+ (1.57)	1.92+ (1.45)	0.60 (1.04)	0.51+ (1.37)	1.42* (1.68)	1.57 (.99)	0.53 (1.23)	0.58 (1.09)
Hill Janajati	0.99 (.08)	2.94*** (3.91)	1.36 (1.20)	0.53* (1.82)	0.99 (.08)	2.06** (2.45)	1.15 (.51)	0.75 (.79)
Dalit	0.74+ (1.47)	1.73* (1.67)	1.94** (2.54)	0.38* (2.16)	0.72+ (1.47)	1.04 (.10)	1.44 (1.24)	0.71 (.71)
Terai Janajati	0.43*** (4.64)	0.63+ (1.37)	0.48** (2.72)	0.28*** (3.35)	0.41*** (4.61)	0.41** (2.49)	0.34*** (3.64)	0.51* (1.67)
Individual characteristics
Gender (1=female)	1.08 (.57)	0.33*** (4.56)	0.70*** (10.2)	0.17*** (6.62)	1.18 (1.26)	0.30*** (4.56)	0.07*** (9.10)	0.19*** (5.65)
Age: 14–19	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
20–24	0.92 (.52)	0.80 (.82)	3.24*** (3.82)	3.40** (3.03)	0.98 (.13)	1.09 (.29)	3.64*** (3.96)	3.02** (2.63)
25–29	0.44*** (3.39)	0.29** (2.59)	3.62*** (3.33)	1.62 (.96)	0.42*** (3.41)	0.35* (2.12)	3.76** (3.27)	1.40 (.64)
30–34	0.24*** (5.06)	0.10*** (4.01)	1.70 (1.24)	0.75 (.52)	0.22*** (5.20)	0.12*** (3.64)	1.57 (1.00)	0.69 (.65)
35–39	0.12*** (6.13)	0.06*** (4.58)	0.94 (.14)	0.43 (1.37)	0.11*** (6.13)	.08*** (4.15)	0.84 (.36)	0.42 (1.33)
40–63	0.08*** (9.50)	0.01*** (7.57)	0.08*** (5.25)	0.07*** (4.49)	0.07*** (8.38)	0.01*** (6.91)	0.06*** (5.24)	0.09*** (3.71)
Ever married	0.46*** (4.16)	0.82 (.61)	1.26 (.82)	1.63 (1.47)	0.43*** (4.46)	0.61 (1.43)	0.99 (.04)	1.79+ (1.72)
Personal migration experience
No migration	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Domestic only	2.94*** (6.04)	0.89 (.21)	1.55 (1.43)	2.68** (3.18)	2.81*** (5.72)	0.79 (.41)	1.27 (.76)	2.93*** (3.34)
Any International	2.56*** (3.55)	15.0*** (7.28)	4.55*** (5.67)	4.94*** (4.52)	2.04* (2.56)	14.3*** (6.20)	3.23*** (3.95)	6.59*** (4.86)
Education: no school					Ref	Ref	Ref	Ref
1–5 years					1.40 (1.15)	1.77 (.94)	1.34 (.67)	1.59 (.54)
6–8 years					1.23 (.71)	1.74 (.93)	1.34 (.67)	2.61 (1.20)
9–10 years					1.32 (.93)	1.15 (.17)	0.82 (.41)	2.21 (.96)
11+ years					0.96 (.16)	0.49 (1.10)	0.56 (1.24)	3.49 (1.57)
Work experience: None					Ref	Ref	Ref	Ref
Wage work only					1.15 (.97)	1.72* (2.08)	1.71* (2.05)	0.45* (1.98)
Any salary work					1.87** (3.10)	1.06 (.15)	2.08** (2.59)	0.62 (1.42)
Household characteristics
Total # of HH members	0.99 (.14)	1.00 (.02)	1.24** (3.15)	1.06 (.63)	0.98 (.36)	1.01 (.09)	1.25** (3.17)	1.01 (.07)
Log % HH member migration	1.01 (.13)	1.00 (0.0)	0.92 (.97)	1.53** (3.20)	1.00 (.02)	1.02 (.16)	0.94 (.75)	1.41* (2.55)
Household Resources					1.05 (1.32)	0.96 (.61)	1.04 (.73)	1.11 (1.38)
Relative HH Resources:
Lower Third					Ref	Ref	Ref	Ref
Middle Third					0.88 (.75)	1.05 (.16)	1.03 (.10)	0.83 (.48)
Upper Third					0.82 (.91)	1.32 (.72)	0.85 (.48)	1.03 (.06)
Neighborhood characteristics
Distance to Narayanghat (miles)	0.98 (1.30)	1.10** (3.24)	1.06* (2.38)	0.93* (2.35)	0.98 (1.35)	1.09** (2.98)	1.06* (2.38)	0.93* (2.29)
Log % NBH member migration	0.37*** (5.14)	0.33** (3.04)	0.26*** (4.60)	0.29*** (4.29)	0.37*** (5.14)	0.34** (2.82)	0.26*** (4.54)	0.28*** (3.58)
Number of Moves	376	103	155	88	376	103	155	88
Fit Statistics
	AIC: 5677.56				AIC: 5677.56
	−2 Log L: 5669.56				−2 Log L: 5669.56

Odds ratios (OR) shown. Z-statistics in parentheses.

Significance:

⁺p<0.10

^*p<0.05

^**p<0.01

^***p<0.001

One-tailed tests for caste/ethnicity, two- tailed tests for controls.

Table 5.

Multinomial logistic regression estimates of destination choice. All respondents included (4406 people). Reference outcome is WWA.

	Model 5				Model 6
Measures	No Move	Within Nepal	India	Persian Gulf	No Move	Within Nepal	India	Persian Gulf
Caste/ethnicity
Brahmin-Chhetri	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Newar	1.96+ (1.37)	2.71* (1.94)	3.76* (2.06)	1.16 (.22)	1.73 (1.09)	2.47* (1.71)	2.73+ (1.52)	0.92 (.11)
Hill Janajati	1.87* (1.82)	1.85* (1.66)	5.51*** (3.99)	2.56* (2.28)	1.33 (.79)	1.31 (.70)	2.74* (2.23)	1.53 (.98)
Dalit	2.64* (2.16)	1.95+ (1.39)	4.58** (2.82)	5.12*** (3.30)	1.40 (.71)	1.02 (.03)	1.45 (.64)	2.01+ (1.30)
Terai Janajati	3.55*** (3.35)	1.53 (1.04)	2.24+ (1.64)	1.71 (1.21)	1.96* (1.67)	0.80 (.51)	0.80 (.42)	0.67 (.83)
Individual characteristics
Gender (1=female)	5.91*** (6.62)	6.35*** (6.42)	1.96+ (1.90)	0.40* (2.47)	5.23*** (5.65)	6.19*** (5.82)	1.57 (1.16)	0.37** (2.63)
Age: 14–19	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
20–24	0.29** (3.03)	0.27** (3.11)	0.24** (3.05)	0.95 (.10)	0.33** (2.63)	0.32** (2.57)	0.36* (2.02)	1.21 (.36)
25–29	0.62 (.96)	0.27* (2.43)	0.18* (2.56)	2.23 (1.32)	0.72 (.64)	0.30* (2.14)	0.25* (1.98)	2.69 (1.55)
30–34	1.33 (.52)	0.31+ (1.94)	0.14* (2.57)	2.26 (1.22)	1.45 (.65)	0.31+ (1.86)	0.18* (2.16)	2.27 (1.17)
35–39	2.34 (1.37)	0.27+ (1.88)	0.15* (2.28)	2.20 (1.06)	2.36 (1.33)	0.25+ (1.91)	0.18* (1.98)	1.99 (.88)
40–63	14.5*** (4.49)	1.08 (.12)	0.15* (2.28)	1.08 (.10)	10.6*** (3.71)	0.76 (.39)	0.13* (2.33)	0.68 (.48)
Ever married	0.61 (1.47)	0.28*** (3.46)	0.50 (1.54)	0.77 (.64)	0.56+ (1.72)	0.24*** (3.84)	0.34* (2.30)	0.55 (1.41)
Personal migration experience
No migration	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Domestic only	0.37** (3.18)	1.10 (.27)	0.33+ (1.76)	0.58 (1.33)	0.34*** (3.34)	0.96 (.12)	0.27* (2.04)	0.43+ (1.93)
Any International	0.20*** (4.52)	0.52 (1.56)	3.03** (2.63)	0.92 (.20)	0.15*** (4.86)	0.31* (2.57)	2.17 (1.39)	0.49 (1.55)
Education: no school					Ref	Ref	Ref	Ref
1–5 years					0.63 (.54)	0.88 (.15)	1.11 (.10)	0.84 (.18)
6–8 years					0.38 (1.20)	0.47 (.90)	0.67 (.41)	0.51 (.74)
9–10 years					0.45 (.96)	0.60 (.60)	.50 (.67)	0.37 (1.05)
11+ years					0.29 (1.57)	0.27 (1.54)	0.14+ (1.93)	0.16* (2.02)
Work experience: None					Ref	Ref	Ref	Ref
Wage work only					2.24* (1.98)	2.58* (2.24)	3.86** (2.84)	3.82** (2.85)
Any salary work					1.61 (1.42)	3.00** (2.93)	1.71 (1.07)	3.35** (2.90)
Household characteristics
Total # of HH members	0.94 (.63)	0.94 (.64)	0.94 (.46)	1.16 (1.41)	0.99 (.07)	0.98 (.24)	1.00 (.01)	1.24+ (1.89)
Log % HH member migration	0.65** (3.20)	0.66** (2.94)	0.65* (2.50)	0.60*** (3.35)	0.71* (2.55)	0.71* (2.38)	0.72+ (1.88)	0.66** (2.64)
Household Resources					0.90 (1.38)	0.94 (.69)	0.86 (1.48)	0.94 (.70)
Relative HH Resources:
Lower Third					Ref	Ref	Ref	Ref
Middle Third					1.20 (.48)	1.06 (.14)	1.27 (.49)	1.24 (.47)
Upper Third					0.97 (.06)	0.80 (.46)	1.28 (.42)	0.83 (.38)
Neighborhood characteristics
Distance to Narayanghat (miles)	1.08* (2.35)	1.05 (1.55)	1.19*** (4.04)	1.14*** (3.48)	1.08* (2.29)	1.05 (1.50)	1.18*** (3.79)	1.15*** (3.40)
Log % NBH member migration	3.46*** (3.65)	1.28 (.68)	1.15 (.29)	0.90 (.25)	3.56*** (3.58)	1.32 (.74)	1.21 (.38)	0.93 (.16)
Number of Non-Moves or Moves	3684	376	103	155	3684	376	103	155
Fit Statistics
	AIC: 5677.56				AIC: 5677.56
	−2 Log L: 5669.56				−2 Log L: 5669.56

Odds ratios (OR) shown. Z-statistics in parentheses.

Significance:

⁺p<0.10

^*p<0.05

^**p<0.01

^***p<0.001

One-tailed tests for caste/ethnicity, two- tailed tests for controls.

We also model a two-step migration decision-making process. Table 3, which tests migration versus no migration is the first step. Table A1 in our supplementary materials includes migrants only and separates destination outcomes (Nepal, India, Persian Gulf, WWA), and addresses the second step in a two-step process. Notably, the results of the one-step process (Tables 4 and 5) are substantively equivalent to those for the two-step process (Tables 3 and A1).

In both cases (one- or two-step decision-making process), we are interested in understanding the mediating influences of human and economic capital, contemporary discrimination, and historical legacies. We directly test for mediation of human and economic capital by progressively controlling for these factors in Tables 3, 4, and 5. However, we cannot use this strategy to directly test for discrimination and historical legacies. Instead, we do this by carefully examining the results in our models that fully control for human and economic capital (Models 4, 6, and 8). In cases where a particular ethno-caste group is more likely to migrate to a destination where they have a history of migrating, we attribute that to the historical legacy explanation. Specifically, we will interpret as historical legacy cases where we find higher likelihoods of Brahmin-Chhetri people going to WWA destinations, Newars to within Nepal (as Kathmandu is considered their traditional homeland), Hill Janajati to India, and Terai Janajati not migrating (as Chitwan is considered their traditional homeland). Alternately, we will interpret as discrimination any cases where an ethno-caste group, especially the disadvantaged groups of Dalit and Terai Janajati, are more likely to migrate to a destination outside South Asia (destinations where Hindu-based caste discrimination is not practiced). For our purposes, the fact that our hypotheses for discrimination-influenced migration predict different destinations for each ethno-caste group and for historical-influenced migration

Results

Table 1 provides univariate descriptive statistics of each variable appearing in our models. Of the full sample, 16.4 percent undertook a migration to any destination during our study period. 8.6 percent of respondents had a domestic migration, 2.3 percent migrated to India, 3.5 percent to the Persian Gulf, and 2.0 percent to WWA.

Table 1.

Descriptive statistics at the time of baseline interview (N=4424)

Measures	Percent distribution or Mean	SD	Min-Max
Outcome measures – Any Migration
Non-migrants	83.6
Any migrants	16.4
Migration Destinations
Non-migrants	83.6
Migrants outside Chitwan in Nepal	8.6
India	2.3
Persian Gulf	3.5
Wealthy Western and Asian	2.0
Caste/ethnicity
Bhramin-Chhetri	44.5
Newar	6.6
Hill Janajati	15.5
Dalit	11.0
Terai Janajati	22.4
Individual characteristics
Gender: Female	60.8
Male	39.2
Age: 14–19	27.6
20–24	11.6
25–29	10.6
30–34	9.7
35–39	9.5
40+	30.8
Ever married	65.1
Personal migration experience
None	62.8
Domestic migration only	23.2
Any international migration	14.0
Work experience
None	46.0
Wage work only	34.9
Any salary work	19.1
Education
None	25.6
1–5 years of school completed	15.5
6–8 years of school completed	26.6
9–10 years of school completed	14.5
11+ years of school completed	17.8
Household characteristics
Total # of HH members	3.4	1.4	1–10
Household Resources	0.45	2.48	−9.30 – 8.56
Relative Household Resources:
Lower Third	24.4
Middle Third	37.3
Upper Third	38.2
Percent HH member migration	39.3	30.6	0–100
Neighborhood characteristics
Distance to Narayanghat (miles)	8.6	3.93	0.02–17.7
Percent NBH member migration	40.9	10.8	0–85.2

Table 2 shows bivariate distributions for the total sample and by ethno-caste. Looking at the first column, titled “Percent migrated”, we can already see variance in migration rates between ethno-castes. The highest rate was the Hill Janajati and Newar people, 19.2 percent and 19.1 percent (respectively) of whom migrated during the study period, and the lowest rate is found with the Terai Janajati, at 11.5 percent. Recall that it is the Hill Janajati who have the history of migrating to India to serve in the Gurkha regiments, it is the Newar people who have strong place attachments with Kathmandu, and it is the Terai Janajati who claim Chitwan as their ancestral homeland with strong place attachments to our study area. In the middle of the distribution, we find Dalit people (who are the most disadvantaged ethno-caste group) with a migration rate of 18.0 percent and Brahmin-Chhetri people (the most advantaged group) at 16.9 percent. Already we can see hints that some of our suggested mechanisms are appearing in empirical results, but we will examine the multivariate results before making conclusive assessments.

Table 2.

Distribution of migrations, by destination and ethno-caste

		Percent of migrants, by destination
Caste/ethnicity	Percent migrated	within Nepal	India	Persian Gulf	WWA	Total
Brahmin-Chhetri	16.9	57	9	17	17	100
Newar	19.2	70	12	9	9	100
Hill Janajati	19.1	47	24	21	8	100
Dalit	18.0	39	19	35	7	100
Terai Janajati	11.5	46	16	29	9	100
Total number moved	722	376	103	131	155	722

Turning to destinations (the remainder of Table 2), while most migrations occurred within Nepal, destinations outside of Nepal vary considerably by ethno-caste. Newars who migrated were less likely than others to migrate outside of Nepal; 70 percent of Newars who migrated had a destination within Nepal, compared to a low of 39 percent of Dalit migrants whose destination was within Nepal. Hill Janajati and Dalit migrants were more likely than others to go to India, and Dalits and Terai Janajati migrated to countries of the Persian Gulf more often than the other groups. Notably, Brahmin-Chhetris migrants were the most likely to go to WWA countries and the least likely to go to India.

Table 3 provides multivariate estimates of the association of ethno-caste with the rate of migration to any destination. Model 1 indicates that when controlling for basic demographic and social characteristics, the odds ratios, at 0.99, for Dalit people (who experience the most contemporary discrimination) are essentially the same as for Brahmin-Chhetris (who experience the least). It is only with the Terai Janajati people (with a historic homeland in Chitwan and little history of migration) that we find a statistically significant association, which, with an odds ratio of 0.69 is substantial.

In Models 2–4, we show the estimated influences of human and economic capital (education, work experience, and economic resources) on migration out of Chitwan. As shown in Model 2, educational attainment has a strong, positive, and statistically significant effect on migration. Model 3 indicates that any salary work is associated with higher migration likelihoods, with marginal statistical significance, compared to no non-family work experience. As shown in Model 4, neither economic resources nor relative economic resources had statistically significant associations with migration.

Despite the substantial association of human capital with migration, their addition to the equations produces only slight changes in the estimated relationship between ethno-caste and migration. Looking at Terai Janajati results, those that are strongest in Model 1, we find that the odds ratio increases to 0.78 (from 0.69) once education is controlled, but then decreases again to 0.75 when work experience and economic resources are controlled. We find a similar pattern with other ethno-caste groups, where only a small amount of the ethno-caste association is mediated by education, work experience, and land ownership, with the suggestion that the strongest mediating factor is education. These results suggest that contrary to conjectures in the existing literature (Gurung, 2012; Tsujita & Oda, 2014), the ethno-caste differences in migration that we find in Chitwan are not, to any great extent, the result of differentials in education, work experience, and land ownership.

Our results in fully controlled Model 4 still show one substantial caste difference in migration. Specifically, the odds ratio for Terai Janajati is 0.75, meaning that they are much less likely to migrate compared to Brahmin-Chhetri people. At the same time, we find no reportable difference between migration rates of Brahmin-Chhetri and Dalit people. Again, the association of ethno-caste with migration appears to only be present for the Terai Janajati people, implicating that it is likely a historical effect due to living in their historical homeland in Chitwan and very low rates of migration in the past.

Destinations

While we find a direct effect of ethno-caste on any migration for only one group (Terai Janajati people), we find some more interesting differences in our models that separate destinations (some of which are much more profitable than others), shown in Tables 4 and 5.

Model 5 shows results where human and economic capital are not controlled. Here we see that ethno-caste plays a substantial role on destination choice. Dalit people are almost twice as likely (odds ratios of 1.73 and 1.94) as Brahmin-Chhetri people to migrate to India and the Persian Gulf (compared to not migrating at all). We also find a substantial positive effect for Hill Janajati people going to India. With an odds ratio of 2.94 that is statistically significant, Hill Janajati are almost three times more likely to go to India, compared to Brahmin-Chhetri people. At the same time, we find no statistically significant effect of Hill Janajati people going to the Persian Gulf or Terai Janajati people going to any destination.

Notably, we find large ethno-caste differences for migration to WWA countries, with all caste groups substantially less likely to move to this destination compared to Brahmin-Chhetri people. Odds ratios for Newar, Hill Janajati, Dalit, and Terai Janajati are 0.51, 0.53, 0.38, and 0.28, respectively, indicating that these people were about half to one quarter as likely to move to WWA countries as their Brahmin-Chhetri counterparts. The effects are statistically significant in all cases (with marginal significance for Newar people). The arge magnitude of the odds ratios warrants attention and we address this pattern further in our discussion of Model 6.

In Model 6 of Table 4 we introduce into the analysis education, work experience, and land ownership. Comparing coefficients in Model 6 with Model 5, we see that the introduction of these human and economic capital measures has only modest effects on the coefficients for destination choices of Newars compared to Brahmin-Chhetris. By contrast, the coefficient for Hill Janajati to India is reduced from 2.94 to 2.06, suggesting that about half of the Hill Janajati association with migration to India is the result of different socioeconomic attributes. Despite that decrease, Model 6 indicates that Hill Janajati people are still about twice as likely to move to India compared to Brahmin-Chhetris. This could be due to contemporary discrimination or historical legacy. Note that the Hill people experience moderate discrimination (much less than Dalit people) and that they have a long (and proud) history of large migration streams to India for work in the British and Indian Army Gurkha Brigades. Thus, while we cannot be definitive in attributing causation, the context is highly suggestive that the remaining positive result for Hill Janajati moving to India (in Model 6) is due to the historical explanation.

Interestingly, the introduction of the socioeconomic attribute variables substantially reduces the magnitude of the positive coefficient of Dalits to both India and the Persian Gulf. Although the odds ratio for Dalits moving to the Persian Gulf is still notably positive (1.44), it is not statistically significant. This indicates that much of the overall association between Dalit ethnicity and destinations is a product of socioeconomic attributes, similar to what Tsujita and Oda and Gurung (Gurung, 2012; Tsujita & Oda, 2014) suggest.

We noted earlier that all of the ethno-caste groups had lower rates to the WWA countries than did the Brahmin-Chhetris. In Model 6 we see that even with the socioeconomic controls, most of these odds ratios remain strong, between 0.51 and 0.75. Notice, however, that the magnitudes of the odds ratios are generally somewhat weaker in Model 6 than in Model 5 and it is only for Terai Janajati that we find statistical significance.

Migration to WWA destinations is important for this study, due to the fact that these countries offer much higher remuneration and social prestige for migrants. We also hypothesized that Brahmin-Chhetri people would be more likely to migrate there due to historical legacy. We find some evidence in this regard in Table 4. To examine this destination more closely, we turn to Table 5, where our models use WWA as the reference outcome. In Table 5, we see the same general patterns. Most odds ratios are smaller when socio-economic factors are controlled (Model 8) compared to when they are not (Model 7). The fully controlled model in Table 4 (Model 6) shows that Terai Janajati people were less likely to go to any destination as opposed to not migrating and the fully controlled Model 8 in Table 5 also shows that Terai Janajati people were about twice as likely to not migrate as to go to WWA destinations (odds ratio 1.96). Compared Brahmin-Chhetris, Newar people were more than twice as likely to migrate within Nepal (OR 2.47) or to India (OR 2.73) as they were to go to WWA. Hill Janajati were more than twice as likely to go to India (OR 2.74) and Dalit people were more than twice as likely to go to the Persian Gulf (OR 2.01, although this is marginally statistically significant).

Considering Tables 4 and 5 together, we find that when socio-economic factors are controlled one destination stands out for each ethno-caste group. Compared to all others, Brahmin-Chhetri people are more likely to go to WWA destinations than any other place. Newar people are more likely to migrate within Nepal, Hill Janajati to India, Dalits to the Persian Gulf, and Terai Janajati are about equally unlikely to migrate to any destination. These destinations largely match our predictions of where people would migrate if the historical legacies pathway is influential. The lone exception is Dalits, whose migration to the Persian Gulf is indicative of escaping contemporary discrimination.

Conclusion

We investigated the relationship between ethnicity, out-migration, and destination choice using detailed panel data from rural Nepal. Our analyses were guided by previous literature that suggests that human and economic capital might mediate the effect of ethnicity on out-migration. In addition, we drew on existing theories of emigration to hypothesize that the expectation of caste-based discrimination will lead people from the more disadvantaged castes to not migrate within Nepal, and instead be more likely to choose international destinations where they will not experience caste-based discrimination. We also addressed the possibility that ethno-caste differences in migration could be based on an entirely different mechanism, that of historical patterns perpetuated into the present day by social networks.

Our bivariate results, by not controlling for any other factors, show actual patterns of ethno-caste migration differences. From these, we see that the high-caste Brahmin-Chhetri are more likely to migrate to the premier destinations of wealthy Western and Asian countries. But Dalits and the indigenous Terai Janajati people, both disadvantaged castes, are more likely to migrate to India and the Persian Gulf, destinations where remuneration is higher than for migrants who move within Nepal.

With multivariate models intended to better isolate the mechanisms that influence these patterns, our results for migration to any destination show little evidence that contemporary discrimination influences the ethno-caste differences in migration. Instead, we find that human and economic capital is a likely mechanism, but its role is relatively small and notably less influential than the overwhelming influence that is suggested by previous literature. The one important ethno-caste relationship to emerge is the significantly lower likelihood of Terai-Janajati people to migrate and we attribute this to the legacy of very low migration amongst this group of people who still live in their historic homeland.

Alternately, it is with destination choice that we find evidence that supports all three of our proposed mechanisms. Here we find that the mediation of human and economic capital has an important influence on destination choice, especially for Dalits who are the most disadvantaged group. It is also with Dalits that we find results that suggest contemporary discrimination might be influencing destination choice.

For the other ethno-caste groups (Brahmin-Chhetri, Newar, Hill Janajati, and Terai Janajati), our results show that destination choice is partially mediated by socio-economic factors. Notably, we do not find evidence that contemporary discrimination influences destination choice, but this is not entirely surprising because these groups occupy the top and middle of the caste hierarchy in Nepal. The striking results we find for these groups is historical legacy likely has the strongest influence on their destinations. In some cases, ethno-caste groups (like Hill Janajati and Brahmin-Chhetri) are more likely to migrate to the same destinations where they have a long history of migration. In another case, it is the historic homeland and lack of migration history and culture of migration that appears to discourage out-migration to any destination for the Terai Janajati. In other words, the differential treatment and discrimination of ethnic groups in the past remains powerful in present-day migration patterns, perpetuated by social networks and cumulative causation. Indeed, so powerful is the past that it appears to overwhelm much of the possible effect of contemporary discrimination on contemporary out-migration.

This study also allows us to comment on the potential for the institution of migration to influence social change. Specifically, can ethno-caste differences in migration cause change in the dynamics of caste-based discrimination? A relatively strong literature from many countries already shows that there are multiple benefits (economic and social) from migration, for the returned migrant, their households, and their communities in general (Acosta, Calderón, Fajnzylber, & López, 2006; Adams, 2011; Massey & Parrado, 1988). In Nepal alone, studies show that remittances have dramatically reduced poverty (KC., 2003; Lokshin, Bontch-Osmolovski, & Glinskaya, 2007; Sharma & Gurung, 2009; World Bank, 2005).

These benefits of migration to individuals and households only influence broader social change if migration is available to those from caste groups that experience discrimination. We find that Dalits, the most disadvantaged ethno-caste group, migrated at slightly higher rates than the most advantaged group, the Brahmin-Chhetris. Further, they were much more likely to migrate to India and the Persian Gulf, which are better remunerated than within Nepal. Thus, there is potential for migration to reduce economic inequality and dampen existing caste-based hierarchies, creating more opportunities for those who were previously disadvantaged. The obvious contingency to this conclusion is our finding that the highest status group (Brahmin-Chhetri) were more likely than any other group to migrate to the premier destinations in wealthy Western and Asian countries. Thus, migration might be creating space to dampen caste-based hierarchies, but historical legacies that limit Dalits from migrating to premier destinations also might limit the potential for changing the caste hierarchies (see Howell 2017).

Of course, this study represents the period 2008–2012. Since then and looking into the future, we expect that different patterns have and will emerge. Just as Paul (2011) describes for the Philippines, we believe that intentional stepwise migration is common in Nepal, whereby people first migrate to a closer destination, to accumulate social and economic capital, which they use to migrate to a farther preferred destination. If this is indeed the case, then we might expect to find more Dalits moving to wealthy Western and Asian destinations since our study period and in the future, after they or their family members accumulate resources in the Persian Gulf (where they were located during our study period). This could then influence even more renegotiation of caste-based hierarchies in Nepal.

Even with our new evidence that allows us to surmise how migration might impact caste-based discrimination in the future, more investigation is still necessary. Specifically, in addition to being better remunerated, premier migration destinations also provide the possibility of longer-term stays and even permanent relocation, compared to Nepal, India, and the Persian Gulf. Thus, migrants to the wealthier destinations, who are heavily comprised of the advantaged Brahmin-Chhetri people, are probably less likely to return to Nepal, and it is this return that ultimately creates the greatest potential for changes in patterns of caste-based discrimination in Nepal. Drawing on Portes (Portes, 2010), we note that to fully understand the potential for migration to influence social change in Nepal (or any other sending country), it will be necessary to examine ethno-caste selection in length of stay and return migration. The theoretical development and methods we use here should contribute to these endeavors.

HIGHLIGHTS.

1. The most advantaged ethno-caste group are the most likely to migrate to preferred destinations in wealthy Western and Asian countries.

2. Contemporary discrimination likely influences the migration patterns of the most disadvantaged ethnic group—Dalits.

3. Historical patterns perpetuated by social networks likely influence the migration patterns of the more advantaged ethnic groups.

4. Identification with the traditional homeland likely influences the migration patterns of the Terai Indigenous peoples.

5. Ethnic differences in migration and destination choice likely influence caste-based discrimination and opportunity in the future.