The Logic of Letting Go: Family and Individual Migration from Rural Bangladesh

Running Head: Family and Individual Migration from Rural Bangladesh

Randall S. Kuhn* Josef Korbel School of International Studies University of Denver

October 2010 * The author gratefully acknowledges the support of the International Pre-Dissertation Fellowship of the Social Science Research Council (supported by Ford Foundation and American Council of Learned Societies); the J. William Fulbright Scholarship (funded by United States Information Agency, administered by Institute for International Education); the Population Council Dissertation Fellowship in the Social Sciences; the Mellon Fund for Research in Population in Developing Countries; and the University of Colorado Population Aging Center (funded by National Institute on Aging). Many provided valuable commentary on the paper, including Jane Menken, Douglas Massey, Leah Vanwey, Robert Retherford, Jeroen van Ginneken, Lynn Karoly, Erin Trapp, Julie DaVanzo, Richard Rogers, Fernando Riosmena, Tania Barham, and Linda Mamoun. Most of all, the author wishes to thank the research and technical staff of the Matlab Health and Demographic Surveillance Unit at ICDDR,B: International Centre for Health and Population Research, and the patient and thoughtful citizens of Matlab. Author’s Contact Information Randall Kuhn Josef Korbel School of International Studies 2201 S. Gaylord St. Denver, CO 80208

It is now understood that voluntary and forced migration constitute a continuum, yet this paper is among the first studies to develop a theoretical test of these connections. Processes of family migration that drive the global growth of informal settlements are driven by structural factors such as environmental degradation and considerable individual selectivity. This paper introduces a three-outcome model in which individual and family migration constitute distinct family livelihood strategies: individual migration supplementation rural livelihoods while family migration replaces a rural livelihood with an urban one. I test this model using 14 years of migration data from the rural Matlab area of Bangladesh. Family migration becomes more likely than individual migration for men with very low household land holdings. This effect is exacerbated during the period following a catastrophic flood, when the likelihood of family migration rose, particularly for landless men.

Keywords: Rural-Urban Migration, Family Migration, Bangladesh, Developing Countries


Recent years have witnessed increasing concern about the growth of informal settlements or slums in the megacities and rapidly emerging towns of the developing world. Informal settlements pose many inherent challenges relating to the lack of property rights or tenure, inaccessibility to markets and public services, and overcrowding (UN-Habitat 2003; Davis 2006). Yet the roots of many of these challenges lie in the underlying forces of economic deprivation, social exclusion, and environmental degradation -- so called “push factors” -- that drive rural-urban migration in the first place. Efforts to address the needs and harness the capabilities of rural-urban migrants thus require an indepth understanding of the rural vulnerabilities and selectivity behind the migration process. Social scientists have developed an acute understanding of the heterogeneity in the motivations for migration, particularly in relation to family livelihoods (Lindstrom and Lauster, 2001; Massey and Espinosa, 1997). The New Economics of Labour Migration has moved migration theory beyond the simple idea of the lone migrant in search of personal advancement by accounting for the role of migration in ensuring the collective security and livelihood of a family unit living in two places at once (Stark 1991; VanWey, 2001). Contextual models account for the role of ecological resources and other entitlements in individual- and family-level decisions (Hunter 1998, 2005; Ezra and Kiros 2001; Henry et al. 2004; Massey et al. 2009). Despite progress, few studies have addressed the distinct livelihood functions of different modes of migration and their underlying forces of selectivity. One distinguishing characteristic of many informal settlement dwellers is the movement of an entire family unit to the city, or family migration, as opposed to the movement of only those family members most likely to find employment in the city, or individual migration. Family migration poses numerous challenges for migrants and for planners, including residential overcrowding, the need to provide social and

health services to new population groups, and the risk that children will opt out of school in favour of labour market activity. Few demographic studies have modelled the movement of an entire family, whether in a separate model or as a direct alternative to individual migration. Instead, a separate literature on forced migration has emerged primarily in the fields of political geography, refugee studies, and ecology. While many studies explore welfare outcomes among forced migrants, few address the determinants of such moves beyond the occurrence of precipitating crises such as conflict or disaster (Keely, Reed, and Waldman 2001). Such crisis events constitute neither sufficient nor necessary condition for family migration (Castles 2006). First, many family migration episodes would be traced not to a particular crisis but to a gradual deterioration of social, economic, and ecological position. More importantly, the vast majority of families move in or near their original location except in cases of total catastrophe, while only a small share will make it to a city. As population growth and climate change increase the occurrence of precipitating events, the magnitude and nature of forced migration flows would depend greatly on the positive and negative forces of selectivity that drive families to move to the city. This paper develops a rational choice model of family and individual migration and tests it using data from rural Bangladesh. The typical mover-stayer utility maximization, in which potential urban income is compared to current rural income, is reframed as a three-outcome maximization. Rural income is instead compared to two potential utility streams, a rural-urban utility in which the best outcome involves working in the city but raising family in the rural area (implying individual migration) and a strictly-urban utility in which the best outcome involves moving the entire family to the city (and thus undertaking family migration). The model thereby draws a theoretical connection between the total loss of livelihood underlying forced migration



and the livelihood diversification motives behind the New Economics of Labour Migration. This model is employed to explore the determinants of family and individual migration by married men in the rural Matlab area of Bangladesh between 1982 and 1996. Data come from the Matlab Health and Demographic Surveillance System (HDSS), a unique data archive that is ideally suited to the prospective modelling of migration behaviour in a representative population. The primary finding shows that family migration is far more likely than individual migration among men with minimal land holdings. This effect is exacerbated in the years following a flood, when the majority of excess family migration occurred among men with little land.

Migration, Livelihoods, and Family
Familial Models of Migration Decision-Making
Existing theory highlights the unique role of migration as a labour market, housing, and livelihood decision. Neo-classical economists initially conceived migration as a simple labour market response to an unequal spatial distribution in wages or earnings, suggesting that the likelihood of migration will increase either with a decline in origin-area wages or a rise in the expected wage in the destination (Lewis, 1954; Ranis & Fei, 1961; Sjaastad 1962; Todaro, 1970; Harris & Todaro, 1969; Massey and Espinosa 1997). The typical wage differential model (hence referred to as WD) predicts migration decision (M) for individual i in terms of the utility (U) to be derived moving to the urban area (u) versus remaining in the rural area (r).
P(M i  k)  M ax(U ), k  u, r k


Mincer (1978) incorporated the role of residential decision-making by extending the WD model to the utility of an entire migrant family rather than a single labourer. In order for migration to occur, the benefits accruing to a single working member (usually a husband) would have to be sufficient enough to compensate any potential “tied movers” for the costs of relocation, social dislocation, and other psychic costs. In other words, for a family j, the migration of all members Kuhn 3

(i) depended on the joint utility that all family members could achieve in the urban or rural area:
P(M j ,i*  k)  Max(U ), k  u, r k


This model spawned an extensive literature on family migration, largely focused on developed countries, in which a family unit collectively chose family migration or no migration or, failing to resolve a joint solution, the family dissolved via divorce. In much of the developing world, individual migration offers another option (Rogaly, 2003; Rogaly and Rafique, 2003; Lauby and Stark, 1988). If migration improves an individual’s labour market position, but not his family’s residential position, he may move alone, often for long durations, while other family members remain in the area of origin, before eventually returning to the origin area. In other words, a family j chooses whether to send a member i to the city by maximizing on the income to be derived from remaining in the rural area versus the income to be derived from a multi-sectoral existence (m) in which a member moves to the city:
P(M j ,i  k)  Max(U ), k  r, m k


While the value of the multisectoral utility ( U m ) is difficult to quantify explicitly, the New Economics of Labour Migration (hence referred to as "NELM") provides a systematic framework for understanding the potential effects of individual migration that would cause a family to value U m over U r (Stark, 1982). Most notably, the migration of an individual can substantially improve the welfare of those left behind by migration through the consumption and investment of remittances, or financial transfers, sent by the migrant (Taylor & Wyatt, 1996; de Haan, 1999; Kanaiaupuni & Donato, 1999; de Haan and Rogaly 2002; Kuhn 2005). Remittances may both diversify the sources of family livelihood, thereby offering protection against crisis, and generate liquidity for rural economic activities (Lauby and Stark, 1988; Durand et al., 1996; Ellis, 1998; Winters et al., 2002; Chen et al., 2003; Sana 2005; Massey et al., 1999). Kuhn 4

NELM highlights important non-linearities in the land-migration relationship, most notably that households with small landholdings may benefit more from individual or temporary migration since they would have more to gain from income diversification since families with large land holdings might not need either extra income or diversification (VanWey 2003; Taylor & Wyatt 1996). VanWey (2003) points to the distinct role of temporary migration for those with moderate land holdings, where migration might contribute to a strategy of investment and expansion, and for those with smaller land holdings, who might simply be combining rural and urban wages to achieve a minimum livelihood target. By extension, below a certain land-holding threshold the rural contribution to wages could become so minimal that rural-urban income would drop below strictly-urban income, making family migration more likely. By ignoring the migration of an entire family, equation (3) implicitly assumes that the joint utility resulting from all family members moving to the city, U u , is necessarily lower than
U m or U r , or that an entire family would never migrate. There are strong empirical reasons for

ignoring this third alternative. Legal barriers (as in modern China or Apartheid-era South Africa), high urban living costs, low wages, and limited job-related benefits may all encourage migrants to raise children and retire in rural areas (de Janvry & Garammon, 1977; Gereffi & Korzeniewicz, 1994; Mackenzie 2002; Brueckner & Kim 2001). Yet continued increases in the urban youth dependency ratios point to the importance of family migration flows. While some migrants may have achieved sufficient income to move their families to the city, the world's informal settlements include a great many families living in poverty. At present, such moves are addressed, if at all, through the literature on forced migration, which tends to address structural and political factors rather than individual heterogeneity and agency (Castles 2006).



Distress, and other sources of family migration
Traditionally, forced migration has been understood both as resulting from collective instigators (e.g., armed groups) and as affecting collective classes of victims requiring institutional redress, typically connected to conflict. A gradual relaxing of the collective nature of both cause and affected group has opened up new space for understanding forced migration as part of a common continuum with voluntary migration, and thus as a process that might be understood in terms of within-group heterogeneity (Castles 2006). First, the forced migrant category has grown to include a broader range of political, environmental and even economic causes. This includes development-driven displacement resulting from government public works projects (e.g., roads and dams) and environmental displacement due to population pressure or poor resource management (Scudder 1993; Cernea 1995, 2006). The increasingly apparent displacement risks posed by climate change have driven interest in these processes (Myers 2002). Second, the literature has gradually moved its focus from specific precipitating events such as disasters or conflicts to the long-term, structural precursors of forced migration, particularly in fragile political or ecological systems (Curran 2002; Hunter 2005). Systematic political neglect or abuse may lie at the heart of modern cycles of displacement and discrete events may precipitate moves, but these are often connected via complex mezo chains of causation that blur the distinction between voluntary and involuntary migration and introduce a role for individual agency (Castles 2006). Even a half-century ago, Peterson (1958) distinguished between forced migration and “impelled migration”, in which migrants retain some agency to stay or go. The question remains how to operationalize this continuum into a testable framework. Only recently have a few demographic studies modelled individual migration decisions in response to a crisis (Engel and Ibañez 2007; Ibañez and Velez 2008; Alvarado and Massey Kuhn 6

2009). None has yet accounted for heterogeneity in the decision of whether to move, though Bohra-Mishra and Massey (2010) address migrant destination choice in the context of the Nepali civil war. A recent World Bank report on climate change and displacement highlights the role of rural environmental degradation and environmental crises in precipitating rural-urban migration and urbanization (Raleigh et al. 2008, referring to Pederson 1995; Findley 1994; Ezra 2001). In keeping with the New Economics literature the report emphasizes the role of the “migration of certain family members to urban areas” during periods of crisis and as a means of mitigating future crises, emphasizing that most moves are circular, not permanent (Ezra and Kiros 2001; Caldwell et al. 1986; Paul 1995; Perch-Nielsen 2004). The authors’ exploration of “distress migration” emphasizes that such displacements are usually temporary and quite often local, with most of the distressed travelling fewer than 3km from the origin area, partly in order to protect their assets and interests in the origin area (Hutton and Haque 2004; Zaman 1991). A smaller number of studies draw a more direct association between structural change, family livelihoods, and migration (Bilsborrow 1991, 1992, 2002; Hugo 2008). The mode, permanence, and destination of crisis migration may be determined both by so-called pull factors, such as having the connections and skills to earn incomes in other environments, and push factors, such as the household’s level of attachment to rural livelihoods. Bilsborrow and DeLargy (1990) begin to identify a continuum of labor market responses to diminishing agricultural holdings within a peasant “household survival strategy”, proceeding from changes in crop mix, to local off-farm labor market opportunities; to individual; circular; and seasonal migration; and finally to family migration. This continuum of migration responses points to a clear gap in the mover-stayer model of migration. WD implies a monotonic relationship between land holdings and migration, while



NELM suggests that households with small land holdings might actually draw greater benefit from migration than those with no land holdings. These factors may each carry weight for the decision to migrate in general, but may further explain distinct family and individual migration decisions, particularly at the margins of economic sustainability. If individual migration represents an effort to diversify between rural and urban livelihood resources and family migration represents a departure from rural livelihoods, then family migration should be more strongly responsive to a lack of land holdings than individual migration. Similarly, the period immediately following an environmental catastrophe might push a large number of households, particularly those with constrained livelihood options, below a threshold of rural viability and into family migration. Before testing a formal model of family and individual migration, I describe the context of migration. This exploration is not meant to exceptionalize the Bangladesh experience, but merely to illustrate how a particular livelihood context might set the stage for family migration.

Research context
The interplay between ecological context, livelihoods, and migration can be seen in the unfolding process of urbanization in Bangladesh. Most generally, Bangladesh is one of the most densely populated countries in the world, with 1,127 persons per square mile. A truer reckoning of the ecological and economic fragility of the country would also account for the fact that as of 2005, only 26% of this population lived in cities (UNPD 2010). Furthermore, approximately 30% of Bangladesh’s total land area consists of rivers, swamps, or charlands that shift between habitable and uninhabitable with the changing course of rivers. The southeast monsoon weather system, which continues to feed most of Bangladesh’s rice production, floods a further 20% of land for half of each year, and up to two-thirds of all land in extreme flood years (Mirza 2002). Given these conditions, it is not surprising to find that Bangladesh has been undergoing Kuhn 8

rapid urbanization. Between 1970 and 1995, the proportion of the Bangladeshi population living in cities grew from 7.6 per cent to 21.7 per cent, yet even the latter figure was the lowest urban proportion among the ten largest Asian nations (UNPD 2010). The most prominent rural-urban migration flows involve movement from areas along the Meghna River Basin in southern Bangladesh, such as Chandpur, Comilla, and Barisal Districts, to large cities such as Dhaka (Nabi, 1992). Matlab is situated in the heart of this migrant-sending region. Situated near the highway about halfway between the two largest cities of Dhaka and Chittagong, travel by boat (in the rainy season) or bus (in the dry season) to either destination is about six hours, close enough to carry out regular circular migration but not close enough to commute. As in many places, urbanization in Bangladesh results at least in part from a process of rural livelihood diversification in which migration serves as a response to persistent risks to agricultural production (Kuhn 2003). In Matlab, as elsewhere in Bangladesh, households depending on underwater rice cultivation are exposed to high levels of unmanaged risk from price fluctuations and severe flooding. Monsoon rice production also results in extreme seasonal fluctuations in cash flow, nutrition, employment, and prices (Chen et al., 1979). Small landholders finance production and mitigate the worst seasonal effects of flooding by borrowing rice at high, pre-harvest grain prices, repaying these loans at the lower, post-harvest price, resulting in an effective annualized interest rate of between 30 per cent and 400 per cent, threatening land foreclosure and fragmentation (Kuhn, 1999; Jensen, 1987; Jahangir, 1979; van Schendel, 1981; Momin, 1992; van Schendel & Faraizi, 1984). Remittances from migrants can alleviate each of these risks and contribute to economic progress. Results from the Matlab Health and Socioeconomic Survey (MHSS) in 1996 reveal that households with out-migrant kin received 27 per cent of total household income from such



transfers (Kuhn, 2001). Transfers finance agricultural production and growing-season consumption, thereby reducing the need to incur debt (Kuhn, 1999; Gardner, 1995; Afsar, 1994; Stark, 1991). They also enable economically secure households to extend credit to indebted households (Gardner, 1995). In the process, however, remittance-related liquidity may hasten cumulative processes of economic displacement (Massey et al. 1998). Individual migration greatly facilities the sending of transfers. Solo movers can minimize the extent of consumption at the higher urban cost of living by living in low-cost group housing, living in company housing, or often sleeping on the floors of their workplaces (Kuhn, 2003). Furthermore, many men are able to return home during the rice harvest, mitigating much of the impact on rural labour supply. Matlab’s age-sex distribution in 1996 demonstrates the significance of male individual migration (Mostafa et al., 1998). In 1996, the male-female ratio for the 10-14 and 15-19 age groups was 1.08, compared to 0.95 at age 20-24, 0.78 for 25-29, and 0.79 for 30-34. Matlab’s relative proximity to major urban destinations (six hours travel compared to twenty to twenty-five hours for other areas with high levels of outmigration) should further facilitate temporary and individual migration (Lucas, 2000). Despite these forces, the practice of family migration is widespread, as quantified below. While family migration may have ecological roots, qualitative research points to the underlying role of economic disenfranchisement and social exclusion (Kuhn 2003). Family migrants were not merely landless, but often those who had been landless for quite a long time. Landlessness was often amplified by inadequate access to informal social and patronage entitlements including sharecropping arrangements, off-farm employment on public works projects, and access to ecological resources such as ponds and fallow resources (Kuhn 2003, Das Gupta, 1987). For these households, the dwindling benefits of low rural consumption prices and informal rural



exchange are outweighed by the costs, both personal and financial, of having one household member move back and forth between rural and urban areas. Rather than making a gradual transition to urban economic activity as temporary labourers in support of a rural household, family migrants make an immediate transition to the city as economic producers, consumers, and permanent residents (Unnithan-Kumar, 2003; Roy et al., 1992). The social and economic processes underlying family and individual migration are further amplified by the fragile, though not entirely unique, ecological conditions of deltaic areas of Bangladesh. Extreme floods are the most devastating ecological crisis affecting inland areas such as Matlab, destroying crops, altering the course of rivers, and permanently submerging land. The October 1988 flood in Bangladesh occurred at the end of the annual growing season, resulting not from heavy local rains, but from unusual flows of water from rivers draining the Himalayas. Most of the permanent property loss was sustained by households living on river banks, where buildings and land were inundated. But a broader segment of households suffered lasting indirect effects, including disruptions in subsequent planting seasons, broken communications and transport links, and commodity price distortions. A household’s ability to sustain the short-term effects of a flood may be determined not merely by ecological factors, but also by its own ability to manage risk and cope with crises. Quantitative analysis of family and individual migration will address both the main effects of the flood and interactions with land holdings. Before addressing patterns and determinants of family migration, it is first critical to understand the significance of migration by married men more generally.

The Migration of Married Men in Matlab
The Matlab HDSS provides a dynamic, computerized population event history for the Matlab study site from 1974 to the current era, conferring numerous analytic advantages for the study of vital events in general and migration in particular. During the study period, HDSS data collection Kuhn 11

assistants made monthly visits to each household to record the timing and circumstances of all birth, death, migration, marriage, and divorce events occurring to household members. Periodic censuses in 1974 and 1982 collected additional data on factors such as education, household living conditions, and land holdings. The large sample size, time span, and precision provide what amounts to a quantitative history of a population, with the sole limitation that events occurring entirely outside the HDSS area are not recorded (e.g. marriages of out-migrants, onward migrations).i This analysis focuses on vital events occurring from July 1982 through June 1996, a period covering the significant social and ecological changes described above. A census conducted in July 1982 provides baseline characteristics such as land holdings and education. The 1982 census captured 186,695 individuals living in 149 study villages. The 1982-1996 followup period saw an additional 94,255 new births and 43,684 new in-migrations from outside the HDSS area (largely women who married into the study site and men who had migrated out before the 1982 census and returned after). The master HDSS event file thus tracks the vital events of 324,634 individuals in total. The demographic events of all of these individuals were used to construct dynamic, time-varying counts of household size and coresident children. This study population of 53,901 married males included 30,956 males who were married at the time of the census, 12,710 who were unmarried at the time of the census but subsequently married, and 10,235 men who were newly migrated into the HDSS area after the census. Education and land data were added for individuals who entered the system via inmigration. Because censuses occurred in mid-year, the annual monitoring of events for this study is organized on a mid-year basis (e.g., 1982 refers to events occurring between July 1982 and July 1983, 1995 refers to events occurring from July 1995 to July 1996. Each man contributed an observation for each person-year in which he was present at any time during the year, starting



from the 1982 census, in-migration, or marriage until censoring. Censoring occurred if a man died or migrated outside the household during the observation year. The 53,901 eligible males contributed 473,492 person-years to the analysis, or 8.78 person years per person out of a possible 14 years. Levels of overall migration were substantial as a source of livelihood and a check on population growth. While the area’s population grew from 186,232 in mid-1982 to 209,843 in mid-1996, the 1996 population would have been 40,327 higher in the absence of net outmigration, accounting for about 20% of the population. Although migration was quite common for women as well as men, women’s moves were primarily for marriage (with an equal number of moves in as out) or as tied movers with husbands. Overall, 58% of net out-migration during the study period was accounted for by men. Among adults over age 15, 76% of men’s moves were to urban or overseas destinations compared to only 48% of women’s moves. Among those moves, work was the reported reason for migration among 73% of men’s moves compared to 9% of women’s moves. Because most moves by married women during this period were tied to the moves of men, this analysis focuses on the migration of men. The analysis focuses on moves to urban areas, accounting for about 60% of all moves. Although the results hold for international migration as well, the infrequency of international family migration (except among Hindus moving to India) made models more difficult to interpret. Migrants to overseas and rural destinations were censored but not analyzed. Given the exclusive emphasis on the migration behaviour of married men, it is important to understand the salience of post-marital migration within men’s lifetime migration histories. Fortunately HDSS data make it possible to construct such a history. HDSS field assistants were trained to ensure that migrants returning from outside the study area were always assigned the



same permanent Registration Identification (RID) number upon returning to the study area, thus enabling the classification as return migrants those who migrated after 1974. The long time span of HDSS also makes it possible to go back to an earlier time period to understand the selectivity of the sample prior to our 1982 study baseline. Figure 1 depicts the marriage and migration behaviour of a group of unmarried men age 15+ living in the Matlab HDSS area at the time of the 1974 census. These men are a subset of the larger study population, which also includes men who were already married in 1974, men who are under 15 in 1974, and men who were already living outside the study area in 1974. In this stylized comparison, men begin as unmarried non-migrants (segment 1) and can transition to either being married before migrating (segments 2 and 3, 74% of all men) or migrated before marriage (segments 4-6, about 24% of the total). Only 2% of men neither married nor migrated during the study period. These results involve migration to any destination (e.g. urban, rural, or overseas) to capture the full extent of attrition. Figure 1 illustrates the salience of migration, and post-marital migration in particular, in the typical adult male life course. Although the youngest of these men were still only 36 by the end of followup, a sizeable 37% of them had already migrated to the city at some point (segments 3-6). About half of these migrants (18% of the total cohort) experienced a move after marriage (segments 3 and 4). Although the risk of migration was higher for those men who had also moved prior to marriage, over one-third of all ever-migrants experienced their first move only after marriage. Among the remaining 24% of men who had migration experience prior to marriage (segments 4-6), a little less than half (47%), or 11% of the total cohort, never returned to the HDSS area and thus are excluded from this analysis (segment 6). Although those who never returned represent a source of attrition from the study population, HDSS data on returned



migrants reduces the magnitude of attrition and accounts for their past migration experience.

A Model of Family and Individual Migration
The staggering level of mobility among married men, including those with no prior migration experience, constitutes the starting point for the statistical analysis. Building on the earlier background discussion, I identify a three-outcome migration decision (Mi,j) whereby a family chooses between remaining in the rural origin area (r), sending one member to the city (m), or moving as a family to the urban area (u)
P(M j,i  k)  Max(Uk ), k  r, m, u


This three-outcome model offers ready contrast to the simple mover-stayer model identified in equation (3), which treats family migration as no different from individual. As with any migration model, we cannot fully operationalize the three potential utility streams ( U k ), but we can present divergent predictions for family and individual migration with respect to key correlates of migration such as land holdings. While both individual and family migration should become more likely at lower levels of land holding, family migration should become the dominant form of migration at lower levels of land holdings. Modest land might encourage individual migration, while no land holdings would encourage family migration. Initial statistical results compare a test of equation (4) using a multinomial logistic hazards regression model against a test of equation (3) using a binomial logistic hazards model. For each year t until an event occurred or a respondent was censored, the outcome for man k was Mkt = j, where j= i (individual), f (family) or n (no) migration in the multinomial models, and where j= m (migration) or n (no) migration men in the binomial.ii Only rural-urban moves were modelled; rural-rural and international migrants were censored, with no event recorded.iii Family migration was recorded if the migrant’s wife moved on the same day to the same destination,



individual migration if the man moved alone or with a group that did not include his wife. iv Both equations fit a vector of predictive coefficients X and a measure of time (t) according to the following equation:

Pr( M kt  j, j  i, f , n) 

e s i

  j s X ks   j s X ks
s i n


j i , f , n


Because estimation could result in any number of solutions, no migration was chosen as a reference event for which all values of  n  0 . Table 1 summarizes key dependent and independent variables at the person and personyear levels, and compares men who were in the 1982 census against those migrants who returned after the 1982 census. Family migration is about twice as likely as individual migration. Although neither mode of migration is particularly likely in any given person-year, the cumulative proportion of men who migrated is quite high, with 4.3% individual and 8.4% family. The study size of 473,561 person years for 53,905 men offers sufficient statistical estimation power. Both individual and family migration are about twice as likely among men who had previously migrated. Past migrant destination reports make it possible to account for men who had any previous rural-urban migration experience, 9.2% of all men and those who had moved to other destinations (6.6%). As noted above, the mean agricultural holdings were quite low at 113 decimals, or 1.13 acres, with a median holding of just over 0.5 acres per household. [Insert Table 1 about here]

Table 2 presents a base model of the covariates of migration using the multinomial and binomial specifications. The models also included controls for household headship, total number of



household members (to control for membership in an extended family household unit), whether the respondent’s occupation was fisherman, and whether the respondent’s religion was Hindu (with Muslim as the reference category). Each of these controls was negatively associated with any, family, and individual migration. Models included village-level fixed-effects specified through controls for village of origin to eliminate the impact of community-level factors such as ecology, market development, and community-level variation in migration rates.v The fixed effects controls were jointly significant according to a chi-square test of difference in both the binomial (χ2 = 802.8, d.f. = 143) and multinomial (χ2 = 1284.9, d.f. = 286) specifications. Covariates estimates and significance were robust to the inclusion of fixed effects, with only a minor reduction in the magnitude of the Hindu control coefficient owing to the extreme clustering of half the Hindu populations in just 9 of the 149 study villages, while 60 village had no Hindus at all. This suggests that the any individual- or household-level heterogeneity in migration pattern is not simply the result of clustering, for instance the tendency for people with little land to live in areas with a common ecological predisposition to migration. Subsequent presentation of results focuses on the fixed effects specifications. [Insert Table 2 about here] Finally, all models control for the positive association between prior and subsequent migration experience, with separate controls for past rural-urban migration and other forms of migration. All key findings with respects to land holdings, schooling, age, family composition, and flood effects are robust to the inclusion of exclusion of this control (not shown). All models described below were tested separately for men with and without prior migration experience (not shown). The magnitude of most coefficients was lower for men with prior experience,



unsurprising since barriers to entry into migration would be lower for past migrants. In no case did the coefficient direction differ between those with or without past migration experience. Presentation of results focuses on measures of household land holdings, individual schooling attainment, age, and the effects of a flood. Livelihood measures The principal point of distinction between family and individual migration in the above model lies in the differential role of rural livelihood resources as both a potential deterrent to migration (of any form) and as a platform for livelihood diversification (favouring individual migration). The principal variable of interest is the total land holdings of the man’s household, measured in HDSS in decimals, or hundredths of an acre. Before presenting the results of a simple loggedland specification (shown in columns 4-6 of Table 2), Figure 2 explores a detailed categorical specification that demonstrates the robustness of the logged relationship. A stable categorization was create by joining rare land values with neighbouring categories until all groups contained at least 1% of the The resulting land distribution was roughly power law in form, with low values far more common than higher ones. Landless men were by far the modal category, accounting for 21%. While 79% had some land, the median holding was only 50 decimals and 87% had less than 200 decimals. As a result, the mean land holding of 114 decimals was considerably higher than the median. The resulting 23 land holding categories were entered into fixed effects logistic regression models with the same controls as the models in Table 2. Coefficient estimates are shown in Table A1. For each land holding category, predicted probabilities of family and individual migration were derived from these estimations using the Stata clarify package and plotted in Figure 2. [Insert Figure 2 about here] The categorical specification demonstrates the greater value of land as a predictor of Kuhn 18

family versus individual migration, particularly when delineating between men with no land at all and men with very small land holdings. For example, relative to men with between 20 and 100 decimals of land (the middle 40% of the sample), men with no land at all were only 12% more likely to practice individual migration, a difference that was not statistically significant, but were 45% more likely to practice family migration. A binomial model would split this difference by predicting a 35% increase in the likelihood of migration. Men with just 10 decimals of land were only 2% more likely to practice individual migration than men with 20 to 100 decimals, but 27% more likely to practice family migration. While small variations in land ownership have little effect on individual migration, they have quite large effects on family migration. The differential effects of low-level land holdings on family and individual migration also go a long way towards explaining the relative frequency of family migration in this context (recall that the annual probability of family was 0.89% compared to 0.48% for individual moves). For the top quartile of landowners (with 150 decimals or more of land), the adjusted likelihood of family migration is about 50% higher and the difference is only moderately significant. This gap widens with decreasing land holdings, to about 2-to-1 for men with 20 to 150 decimals, 2.3-to-1 for men with 10 decimals, and 2.6-to-1 for men with no land at all. A decomposition across the actual distribution of land holdings reveals that two-thirds of the difference between the overall family and individual migration probability was accounted for by men holding less than 50 decimals (one-half acre) of land and 40% of the gap was explained by landless men alone. Although the magnitude and significance of the land-migration relationship diverges substantially between family and individual moves, both are well captured by the single variable, logged-land specification employed in the base regression models in Table 2. The single measure



takes the log of land holdings (measured in decimals); landless men are given a value of 0. Since the lowest non-zero value of land holding is 10 (natural log = 2.3), the transformation generates a large linear distance between landless households and those with very little land without the use of an extra landless dummy variable. The categorical results suggest that the likelihood of individual migration begins to rise at the very highest levels of land holdings, supporting earlier findings of an economies of scale effect whereby households with very large land holdings may use migration as part of an accumulation strategy (VanWey 2005). Yet such households represent just over 2% of men in this land distribution. The cubic land specification used by VanWey (2005) to capture curvilinear land holding effects was significant only at the 10% level and could not actually predict the non-monotonic pattern (results not shown). For both family and individual moves the log-land specification is highly significant compared to other single variable specifications, and captures the shape of the land-migration relationship across the distribution, as shown in the comparison of fitted probabilities in Figure 2. Although the single-variable specification explains both family and individual migration, the separation of the two streams is nonetheless justified on the basis of the substantial and significant variations in coefficient magnitude and significance. Coefficient estimates suggest that a doubling of total household land holdings would be associated with a 8.5% decline in the relative risk of family migration but only a 3.2% decline in the risk of an individual move. Under this specification, the relative risks of migration for a man from a landless household compared to a household with the median landholding of 50 decimals (0.5 acres) would be 64% higher for family migration (off a larger base) but only 20% higher for individual moves. To indirectly address some of the environmental aspects of rural livelihoods, the base model also explores the role of internal migration behaviour as a predictor of family and



individual migration. As noted above, family migration outside the district may represent the culmination of a gradual process of economic and ecological displacement that may include rural displacement. Flooding, environmental degradation, and the short-term unintended consequences of the flood control embankment resulted in fairly high levels of internal displacement. The base model includes two measures of internal migration, one for whether the household moved internally (to a separate village) in the current or previous year, and another accounting for internal migration experience in any previous year since 1982. A simple mover-stayer model of migration, would have suggested a reduced risk of out-migration in the year following internal migration, followed by a return to baseline risk. In fact, such a result would mask divergent effects of internal migration on family and individual migration. Internal migration in the current or previous year was associated with a substantial reduction in the likelihood of both family and individual migration, with men about 80% less likely to leave the study site if they had recently moved villages. Net of this effect, however, strikingly different migration patterns emerge for those with less immediate internal migration experience. Internal migrants had a substantially reduced risk of individual migration (50% relative risk reduction, significant at the p<0.001 level), but a significantly increased risk of family migration (44% increased risk, p<0.001). In the case of both land holdings and internal migration, livelihood vulnerability appears to act in the manner of a “push factor” for family migration, but less so for individual migration. Human capital and life course measures The model of family and individual migration offers less clear predictions for the role of individual human capital or life course factors on migration. It would be simple to assume that family migrants are simply more vulnerable in every respect compared to individual migrants. Yet it is also possible that, after adjusting for the effects of rural livelihood vulnerability, that Kuhn 21

family migration would be driven by classic pull factors like education. Furthermore, the Mincer model of tied movers suggests that only those men with high levels of schooling would be better able to finance the high long-term costs of permanent urban residence associated with family migration. The base model includes controls for three levels of respondent's schooling completion: 1-4 years, 5-9 years (lower secondary school), and 10+ years (higher secondary school), each compared to the reference category of no schooling. In the case of education, a two-outcome model of migration versus no migration would generate the statistically correct result that schooling increases the likelihood of migration, but would miss some salient differences between family and individual moves. Each additional level of schooling attainment is associated with an increased likelihood of both family and individual migration, suggesting that forces of positive educational selection operate even in the family migration decision. Yet for each level of schooling, the coefficient for individual migration is significantly larger. An interesting pattern emerges across different levels of schooling whereby each additional level of schooling associated with a relatively large increase in the likelihood of individual migration (coefficients = +.272, +.576, +.821), whereas family migration responds primarily to higher levels of schooling (coefficients = +.150, +.229, and +.672). One interpretation of these effects is that some family migrants move not because their strictly rural income is so low but rather because their strictly urban income is so high that they can actually afford to raise their children in the city. Nevertheless, it should be clear that family migration involves some level of individual agency and responsiveness to individual human capital, not merely livelihood loss. The three-outcome migration model also generates predictions about the effects of age and life course transition on migration. In general, older men and men with more children should



be less likely to migrate at all, yet there are multiple reasons why migration declines over the life course. One factor relates to familial obligations, which merely require family unity, whether all members are living in the rural area or urban area and thus should have less effect on family migration. Another factor relates to the diminishing returns to long-term investments over the life-course; older men may have fewer opportunities to acquire further human capital in the urban area or to see the benefits of investments in long-term livelihood improvements. Both interpretations suggest that the likelihood of individual migration should decline more rapidly with age. At the same time, if individual migration represents a relatively short-term livelihood response, then one might expect that individual moves would become relatively more frequent at older ages as an opportunity to generate a temporary livelihood supplement without long-term separation from family. The age effects in the base model indicate that individual migration experiences a much more rapid age-related decline than family migration. Using an age and agesquared specification, the likelihood of individual migration declines rapidly with age, before levelling off slightly at older ages. By contrast, the likelihood of family migration remains relatively unchanged throughout the 30s before declining more rapidly through the 40s. To further explore the incentives for family and individual migration across the life course, Table 3 introduces controls for the effects of children of different ages. Family migration is not merely more compatible with aging, it is also more compatible with childbearing. The risk of both family and individual migration is substantially reduced if a man has a child under age 1 (54% reduction for individual, 57% reduction for family, not statistically different from one another). But whereas the likelihood of individual migration remains substantially reduced for men with children age 1-4 (56% relative risk reduction), family migration is actually slightly but significantly more likely (14% increased risk) if a man has a child age 1-4. Less strikingly,



having a child age 5-9 results in a moderately decreased risk of individual migration (31% reduction) but no significant impact on family migration. Children age 10-14 had no effect on individual migration and slightly reduced the likelihood of family migration, but these effects were not jointly significant. These results suggest that the presence of small children is a demonstrable impediment to individual migration, but a less clear impediment to family migration. While the presence of an infant may create logistical challenges to moving an entire family, somewhat older children offer no such limitation. One interpretation of this divergence would suggest that potential individual migrants have a choice not to move given their familial obligations, whereas potential family migrants do not have such a choice at their disposal and are forced to migrate irrespective of childrearing responsibilities. The finding of a positive effect of children age 1-4, however, may actually suggest slightly greater agency. Vulnerable families may be forced to consider alternatives to rural life, yet they may also take advantage of windows of opportunity for family resettlement, much as in rich countries (Rogers et al. 19xx). After controlling for the effects of young children, age patterns of individual and family migration become remarkably similar. Variation over time – main effects and interactions One final factor driving family and individual patterns was variation over time, particularly relating to ecological conditions related to the construction of a river embankment and the occurrence of a major flood in 1988. While analysis of the specific interactions between ecological and socioeconomic vulnerability is beyond the scope of this paper, it is important to establish that excess migration risks in the post-1988 period are indeed associated with heavy flooding. To isolate this effect, Figure 3 illustrates spatial variation in the relative risk of family migration in 1988 in relation to the three major rivers in the area, the primary river Meghna touching only the southwest corner of the site, the major tributary Gumti touching only the Kuhn 24

northern tip, and the Dhonnagoda running through the site. The shape and flow of the Dhonnagoda created a tidal flow on the left bank of the river, thereby exacerbating flood risks and justifying the placement of the river embankment on that side. Results are based on the regression models in Table 3, thereby controlling out some of the socioeconomic variations (like high rates of internal migration) that may also vary spatially. Village-year controls were added to capture spatial variation. Map A depicts village-level relative risks of family migration adjusted for baseline variation in risk in all other years. The pattern of spatial variation demonstrates that the spike in family migration rates observed following the flood was indeed associated with flooding, though the pattern is perhaps counterintuitive. After adjusting for baseline risk, the map indicates lower relative risks at the water’s edge, largely because every year is a flood year for those living right on the river. Moving away from the immediate river bank, relative risks are highest along the main canals of the river, particularly on the one major canal on the embankment side. Finding are adjusted for baseline migration risks, and thus control for lower levels of family migration on the right-hand side of the river. Panel B instead maps absolute spatial variation in 1988, thereby incorporating the general tendency towards high levels of family migration in riverine areas. This map illustrates that overall family migration risk in 1988 was universally higher on the embankment side of the river, particularly in areas along the major Meghna and Gumti rivers which always have high family migration risk, and in the major canal areas that saw a temporary spike to comparable migration rates. While excess family migration risk is not simply a matter of proximity to water, the story is clearly related to the flood. The risk of excess family migration was not confined to the flood year, however, but rather carried on through 1991. Table 4, columns 1 and 2 present the coefficients of models with



observation year dummies for each year in the study period (1982 is the reference year). There was a gradual rise in the family migration rate over time. The relative risk of family migration peaked in 1988, but remained significantly higher through 1991 before declining. Individual migration was relatively high in 1988, but actually peaked in 1985 and 1986. Columns 3-6 combine observation year dummies with an interaction between each observation and household land holdings. Land effects on family migration were more negative for the years 1988 to 1991, ranging from -0.149 to -0.159, all significantly different from zero. Joint significance tests further support that the land interaction coefficients for these four years were jointly significantly different from land interaction coefficients for all other years. Main effects on family migration from 1988 to 1991 remain jointly significant as well, such that family migration increased during this period, but especially among households with lower land holdings. As with the main effects, observation year-land interactions for individual migration were not statistically significant. Interactions effects of land holdings and time on the risk of family migration are summarized in Figure 3, which compares the predicted probability of individual and family migration over time for a landless man and a man with the mean landholding of 115 decimals. For men with mean landholdings, the probability of family migration between 1988 and 1991 was about 15% higher than in the two years preceding and following this era (1.32% vs. 1.15%). Landless men experienced a more pronounced four-year peak on top of an already high baseline family migration risk, rising from 1.51% to 1.93%, or a 28% increase.

The preceding theoretical framework and analysis set out to bridge the divide separating the literatures on voluntary and forced migration, particularly with respect to the role of family livelihoods. In this area of rural Bangladesh as in much of the world, rural-urban migration represents a critical livelihood adaptation strategy even after marriage, yet individual and family Kuhn 26

migration represent two distinct approaches to migration given the gravity of the decision to raise children in the urban setting and to forgo many of the social and economic benefits of rural life. A number of specific differences in the determinants of family and individual migration would be missed by a two-outcome model. The relative likelihood of family migration is much higher among men who lack land, the principal economic resource in the area. Family migration is also more likely among those who had recently experienced internal migration, which is itself typically associated with the loss of economic livelihoods or housing security in the former location. Extensive qualitative research carried out in the study site indicate that family migration patterns could be further understood in the context of deeper forms of socioeconomic vulnerability such as a long history of landlessness, limited access to patronage resources including employment and informal resources, and exclusion from social exchange networks. In spite of the ecological and socioeconomic vulnerabilities, family migration requires some amount of agency. This is most evident in the positive selectivity of family migration on schooling, though this selectivity is substantially less than exists for individual migration. While children are less likely to discourage family migration than they are to discourage individual migration, the pattern of child age selectivity nevertheless reveals considerable agency. Family migration is quite unlikely among families with infant children, yet it is somewhat more likely among families with children age 1 to 4, suggesting that families may be strategic in finding opportune stages in the life course to carry out this drastic relocation decision. These findings point to the need for further expansion and integration of existing models of the role of livelihoods and involuntary migration. For the literature on labour migration, including the New Economics of Labor Migration, the spectrum of livelihood-based migration options could be broadened to incorporate the movement of an entire family. The existing model



of livelihood diversification through temporary and individual migration works well enough when modest livelihoods can be supplemented by migration, but more adverse conditions may imply a point to the largely rational response of replacing rural livelihoods entirely with urban ones. Such conditions may be more common in rural Bangladesh than in places with a broader base of rural capital, but they may surely play out in a variety of settings. The adverse conditions underlying family migration also raise the continued need to inject a stratification perspective into the study of migration. For forced migration studies, these results allow us to broaden our conception of involuntary migration to include moves that are non-coercive and uncoordinated, yet occur over long distances and durations. It also adds depth to our existing understanding of the motives behind such migration, moving beyond the specific crisis and a general sense of vulnerability to a specific set of predisposing risks for family migration. Ultimately, the significance of this study and of an integrated model of voluntary and involuntary migration derive from the need for behavioural models of population displacement and welfare in anticipation of the long-term effects of global climate change. Episodes like the 1988 flood illustrates both the significance of family migration under such crisis conditions, the duration of such effects, and the importance of livelihood effects rather than simple ecological effects. The likelihood of family migration rose substantially in the years immediately following a catastrophic flood, far more so than the likelihood of individual migration. The family migration peak lasted not one but four years, with little difference between 1988 and the three subsequent years. The immediate spike in family migration risk came not in the most floodprone areas, which see flooding and high rates of family migration every year, but in areas of moderate flood risk. Most importantly, the relative risk of family migration in the entire four



year post-flood period were higher for men from households with smaller land holdings, on top of their already greater risk of family migration. Post-flood main effects and land interactions were robust to the inclusion of village fixed-effects, suggesting that excess family migration was less the result of the rising waters themselves than of the ensuing social and economic disruptions. Future research should explore the specific interactions between exposure to environmental risk, generalized economic disruption, and household livelihoods in greater depth. Extrapolating from this one small study area to the broader coastal ecosystem, this postflood dislocation of population could have had substantial impacts on rural-urban migration. Over a four-year period, about 1,400 married left the study site with their families, or about 250 more than would have left in a normal year. Roughly 30% of these men were landless, compared to 21% of the total sample. Each of these men travelled with a wife and an average of 2 children under age 15. In a study area population of 200,000, between 5,000 and 6,000 moved during a four-year period under highly adverse conditions, half of them children, about 1,000 of whom would not have moved in the absence of a flood. Simply extrapolating this level of displacement to the quarter of Bangladesh’s population living in deltaic areas, about 30 million at the time, would imply the displacement of approximately 750,000 people in a four-year period, about 150,000 of whom would not have moved had there not been a flood. While these migrants do not constitute the flood of displaced individuals that might be associated with a worst-case scenario climate event, they point a more relevant and equally challenging form of displacement, sustained and highly selective on livelihood vulnerability, with occasional spikes that dip ever more deeply into the pool of vulnerable households. The forces of vulnerability and agency at play in the family migration decision raise questions about the welfare of family migrants and appropriate policy measures. Even in arguing



that family migration is rational and thus perhaps not easily or sensibly discouraged, it may nonetheless be sub-optimal or avoidable under some conditions. Periods of ecological disaster offers periods of potential risk and opportunity. Some displacement is unavoidable, yet continued excess displacement over a period of four years places considerable burdens on urban areas, thereby justifying efforts to promote rural population retention. The duration of risk and its concentration among socioeconomically vulnerable families suggests that medium-term livelihood support (such as employment on reconstruction) would at least mitigate the migration flow. For the subset of families who could not and should not be discouraged from abandoning rural life, urban transition programs targeting the needs of socially excluded and economically vulnerable families could carry further benefits. While forced migration is normally attributed to acts of God, politics, or both, this paper proceeded to understand the role of household vulnerability and agency in driving the process of family migration. More family migration can be expected in the future. Much more can be done to understand it, and to provide families and societies with the tools they need to make it both avoidable and tolerable.



Afsar, Rita. 1994. Internal migration and women: an insight into causes, consequences, and policy implications, Bangladesh Development Studies 2/3: 217-243. Alvarado, Steven E. and Douglas S. Massey. 2010. “In Search of Peace: Structural Adjustment, Violence, and International Migration.” Annals of the American Academy of Political and Social Science, forthcoming. Amin, Sajeda, Ian Diamond, Ruchira T. Naved, and Margaret Newby. Transition to adulthood of female garment-factory workers in Bangladesh, Studies in Family Planning 29: 185-200. Arango, Joaquin. 2000. Explaining migration: a critical view, International Social Science Journal 52(3): 283-296. Bohra-Mishra, Pratikshya and Douglas S. Massey. 2010. Individual Decisions to Migrate During Civil Conflict. Paper presented at "Migration: A World in Motion", University of Maastricht, Maastricht, Netherlands. Brueckner, Jan K. and Hyun-A Kim. 2001. Land Markets in the Harris-Todaro Model: A New Factor Equilibrating Rural-Urban Migration. Journal of Regional Science 41(3): 507-520. Castles, Stephen. Towards a Sociology of Forced Migration and Social Transformation. Sociology 77(1): 13-34. Cernea, Michael M. 1995. Social Integration and Population Displacement:The Contribution of Social Sciences. International Social Science Journal 43 (1): 91–112. Cernea, Michael M. 2006. Development-induced and Conflict-induced IDPs: Bridging the Research Divide. Forced Migration Review 15: 25-27. Chen, Kong-Pin, Shin-Hwan Chiang, and Siu Fai Leung. 2003. Migration, family, and risk diversification, Journal of Labour Economics 21(2): 353-380. Chen, L.C., A.K.M.A. Chowdhury and S.L. Huffman. 1979. Seasonal dimensions of energy protein malnutrition in rural Bangladesh: The role of agriculture, dietary practices, and infection, Ecology of Food and Nutrition 8: 175-187. Connell, John, Biplab Dasgupta, Roy Laishley, and Michael Lipton. 1976. Migration from Rural Areas: The Evidence from Village Studies. Delhi: Oxford University Press. Curran, Sara R. 2002. Migration, Social Capital, and the Environment: Considering Migrant Selectivity and Networks in Relation to Coastal Ecosystems. In: Population and Environment: Methods of Analysis, Population and Development Review. Lutz, W. Prskawetz, A. and Sanderson, W. (eds). A supplement to Volume 28, 89–125. de Haan, Arjan. 1999. Livelihoods and poverty: The role of migration – a critical review of the Kuhn 31

migration literature, Journal of Development Studies 36: 1-47. de Haan, Arajan and Ben Rogaly. 2002. Introduction: migrant workers and their role in rural change, Journal of Development Studies, Special Issue on Labour Mobility and Rural Society, 38(5): 1-14. de Janvry, Alain and Carlos Garramon. 1977. The dynamics of rural poverty in Latin America, Journal of Peasant Studies 5(2): 206-216. Durand, Jorge, Douglas S. Massey, Rene M. Zenteno. 2001. Mexican immigration to the United States: continuities and changes, Latin American Research Review 36(1): 107-127. Durand, Jorge, William Kandel, Emilio A. Parrado, and Douglas S. Massey. 1996. International migration and development in Mexican communities, Demography 33(2): 249-264. Ellis, Frank. 1998. Household livelihood strategies and rural livelihood diversification, Journal of Development Studies, 35: 1-38. Engel, Stefanie, and Ana Maria Ibanez. 2007. “Displacement Due to Violence in Colombia: A Household-Level Analysis.” Economic Development and Cultural Change 55:335–365. Ezra, Markos and Gebre-Egziabher Kiros. 2001. “Rural Out-Migration in the Drought Prone Areas of Ethiopia: A Multilevel Analysis.” International Migration Review. 35(3):749-771. Faist, Thomas. 2000. The Volume and Dynamics of International Migration and Transnational Social Spaces. New York: Clarendon Press. Fauveau, Vincent. 1994. Matlab: Women, Children and Health. Dhaka, Bangladesh: ICDDR,B. Frankenberg, Elizabeth and Randall Kuhn. 2001. Old age support from sons and daughters: Bangladesh and Indonesia in comparison. Paper presented at the Annual Meeting of the Population Association of America, Washington, DC. Gardner, Katy. 1995. Global migrants, local lives: Travel and transformation in rural Bangladesh. Oxford: Clarendon Press. Gereffi, Gary and Miguel Korzeniewicz (ed.). 1994. Commodity Chains and Global Capitalism. Westport, CT: Greenwood Press. Harris, John R. and Michael P. Todaro. 1970. Migration, unemployment and development: a two sector analysis, American Economic Review 60(1): 126-142. Henry, Sabine, Bruno Schoumaker, and Cris Beauchemin. 2004. The Impact of Rainfall on the First Outmigration: A Multi-Level Event-History Analysis in Burkina Faso. Population and Environment 25(5): 423-60.



Hunter, Lori M. 1998. “The Association between Environmental Risk and Internal Migration Flows.” Population and Environment. 19(3):247-277. Hunter, Lori. M. 2005. Migration and environmental hazards. Population & Environment: A Journal of Interdisciplinary Studies 26 (4): 273-302. Ibáñez, Ana Maria, and Carlos Eduardo Vélez. 2008. “Civil Conflict and Forced Migration: The Micro Determinants and Welfare Losses of Displacement in Colombia.” World Development 36(4):659-676. Jahangir, B.K. 1979. Differentiation Polarization and Confrontation in Rural Bangladesh. Dhaka, Bangladesh: Centre for Social Studies, University of Dhaka. Jansen, Eirik G. 1987. Rural Bangladesh: Competition for Scarce Resources. Oslo, Norwegian University Press. Kanaiaupuni, Shawn Malia and Katharine M. Donato. 1999. Migradollars and mortality: The effects of migration on infant survival in Mexico’ Demography 36(3): 339-353. Keely, Charles B., Holly Reed, Ronald J. Waldman, R. 2001. Understanding mortality patterns in complex humanitarian emergencies. In Reed, H. E. & Keely, C. B. (eds) Forced Migration and Mortality. National Academies Press, Washington, DC. Kuhn, Randall. 1999. The Logic of Letting Go: Family and Individual Migration from Rural Bangladesh, Unpublished Doctoral Thesis, University of Pennsylvania, Philadelphia. Kuhn, Randall. 2003. Identities in motion: social exchange networks and rural-urban migration in Bangladesh, Contributions to Indian Sociology, 37(1-2): 311-337. Kuhn, Randall. 2005. A longitudinal analysis of health and mortality in a migrant-sending region of Bangladesh, in S. Jatrana, M. Toyota, and B. Yeoh (eds.), Migration and Health in Asia, London: Routledge Press, pp. 318-357. Lauby, Jennifer and Oded Stark. 1988. Individual migration as a family strategy: young women in the Philippines, Population Studies 42: 473-486. Lewis, W.Arthur. 1954. Economic Development with unlimited supplies of labour, The Manchester School of Economic and Social Studies 22: 139-191. Lindstrom, David P. 1996. Economic opportunity in Mexico and return migration from the United States, Demography 33(3): 357-374. Lindstrom, David P. and Nathanael Lauster. 2001. Local economic opportunity and the competing risks of internal and US migration in Zacatecas, Mexico, International Migration Review 35(4): 1232-1256.



Lipton, Michael. 1980. Migration from rural areas of poor countries: the impact on rural productivity and income distribution, World Development 8: 1-24. Lomnitz, Larissa A. 1977. Networks and Marginality: Life in a Mexican Shantytown, New York: Academic Press. Lucas, Robert E B. and Oded Stark. 1985. Motivations to remit: evidence from Botswana, Journal of Political Economy 93: 901-918. Mackenzie, Peter. 2002. Strangers in the City: The Hukou and Urban Citizenship in China. Journal of International Affairs 56(1): 305-319. Massey, Douglas S. and Emilio A. Parrado. 1994. Migradollars: the remittances and savings of Mexican migrants to the United States, Population Research and Policy Review 13: 3-30. Massey, Douglas S. and Kristen Espinosa. 1997. What’s driving Mexico-US migration: a theoretical, empirical and policy analysis, American Journal of Sociology 102: 939-999. Massey, Douglas S., Joaquin Arango, Graeme Hugo, Ali Kouaouci, Adela Pellegrino, and J. Edward Taylor. 1999. Worlds in Motion: Understanding International Migration at the End of the Millennium, New York: Oxford University Press. Massey, Douglas S., Nathalie Williams, William G. Axinn, and Dirgha J. Ghimire. 2009. Community Services and Out-Migration. International Migration 48(3): 1-41. Mincer, Jacob. 1978. Family migration decisions, Journal of Political Economy 8(2): 749-773. Myers, Norman. 2002. Environmental refugees: a growing phenomenon of the 21st century. Philosophical Transactions of the Royal Society London. 357:609-613, Nabi, A.K.M. Nurun. 1992. Dynamics of internal migration in Bangladesh, Canadian Studies in Population 19(1): 81-98. Perlman, Janice E. 1976. The Myth of Marginality: Urban Poverty and Politics in Rio de Janeiro, Berkeley: University of California Press. Portes, Alejandro and Julia Sensenbrenner. 1993. Embeddedness and immigration: notes on the social determinants of economic action, American Journal of Sociology 98(6): 1320-1350. Raleigh, Clionadh, Lisa Jordan, and Idean Salehyan. 2008. Assessing the Impact of Climate Change on Migration and Conflict. Working paper on Social Dimensions of Climate Change. The Social Development Department, World Bank Group, Washington, DC. Ranis, Gustav and John C.H. Fei. 1961. A theory of economic development, American Economic Review 51: 533-565.



Retherford, Robert D. and Minja Kim Choe. 1993. Statistical Models for Causal Analysis, New York: John Wiley and Sons. Rogaly, Ben. 2003. Who goes? Who stays back? Seasonal migration and staying put among rural manual workers in Eastern India, Journal of International Development 15: 1-10. Rogaly, Ben and Abdur Rafique. 2003. Struggling to Save Cash: Seasonal Migration and Vulnerability in West Bengal, India, Development and Change 34(4): 659-681. Root, Brenda Davis and Gordon F. De Jong. 1991 Family migration in a developing country, Population Studies 45(2): 221-233. Rosenzweig, Mark R. and Oded Stark. 1989. Consumption smoothing, migration, and marriage, Journal of Political Economy 97: 905-926. Roy, Kartik C., Clem Tisdell, and Mohammad Alauddin. 1992. Rural-urban migration and poverty in South Asia, Journal of Contemporary Asia 22(1): 57-72. Sana, Mariano. 2005. Buying Membership in the Transnational Community: Migrant Remittances, Social Status, and Assimilation. Population Research and Policy Review 24: 231-261. Scudder, Theodore. 1993. Development-induced Relocation and Refugee Studies: 37 years of Change and Continuity among Zambia’s Gwenbe Tonga. Journal of Refugee Studies 6(3): 123–152. Shaw, Annapurna 1988. The income security function of the rural sector: the case of Calcutta, India, Economic Development and Cultural Change 36: 303-314. Sjaastad, Larry A. 1962. The costs and returns of human migration, Journal of Political Economy 70: 80-93. Stark, Oded. 1982. Research on rural-to-urban migration in LDCs: the confusion frontier and why we should pause to rethink afresh, World Development 10(1): 63-70. Stark, Oded and Eliakim Katz. 1986. Labour migration and risk aversion in less developed countries, Journal of Labour Economics 4: 134-149. Taylor, J. Edward and T.J. Wyatt. 1996. The shadow value of migrant remittances, income and inequality in a household-farm economy, Journal of Development Studies 32: 899-912. Todaro, Michael P. 1969. A model of labour migration and urban unemployment in lessdeveloped countries, American Economic Review 59: 138-148. United Nations Population Division. 2010. World Population Prospects: The 2006 Revision and World Urbanization Prospects: The 2007 Revision. Department of Economic and Social Kuhn 35

Affairs of the United Nations Secretariat. Accessed at, Saturday, September 11, 2010; 1:52:17 PM. Unnithan-Kumar, Maya. 2003. Spirits of the womb: migration, motherhood and healthcare in Rajasthan, Contributions to Indian Sociology 37(1/2): 163-187. VanWey, Leah K. 2003. Land ownership as a determinant of temporary migration in Nang Rong, Thailand, European Journal of Population 19: 121-145. Winters, Paul, Benjamin Davis, and Leonardo Corral. 2002. Assets, activities and income generation in rural Mexico: factoring in social and public capital, Agricultural Economics 27(2): 139-156. Zaman, M.Q. 1991. “The Displaced Poor and Resettlement Policies in Bangladesh.” Disasters. 15(2):117-125.



Table 1: Means of key dependent and independent Variables Person Level Migration Outcomes Individual Migration Family Migration Past Migration Experience Rural-Urban Other External Internal Any in current year Independent Variables Household Land Years of Schooling 1-4 Years 5-9 Years 10+ Years Age Any child age 0-1 Any child age 1-4 Any child age 5-9 Any child age 10-14 Household head Household Size Fisherman Hindu Cases 112.9 (168.8) 3.1 (3.8) 23% 21% 10% 37.5 (12.4) 18% 39% 34% 28% 51% 7.8 (3.6) 5% 15% 45,382 113.4 (170.3) 3.1 (3.7) 22% 22% 9% 43.1 (12.2) 13% 46% 48% 42% 60% 8.4 (4.1) 5% 15% 473,561 9.2% 6.6% n/a 8.5% 3.6% 7.1% 2.8% 4.3% 8.4% 0.49% 0.96% PersonYear Level

Notes: * - First entry to sample is in 1982 for cases in census, and on first return post-1982 for others. Standard deviations in parenthesis where applicable.



Table 2: Covariates of Migration Using Two-Outcome and Three-Outcome Models No Village Fixed Effects Binomial Multinomial Logit Logit Any Individual Family Migration Migration Migration -0.101*** (0.007) -1.428*** (0.240) 0.194* (0.077) 0.228*** (0.035) 5-9 Years 10+ Years Age Age squared Any Rural-Urban Migration Experience Any Other External Migration Experience Household head Household Size Fisherman Hindu Constant 0.379*** (0.036) 0.724*** (0.044) -0.044*** (0.010) 0.000 (0.000) 0.964*** (0.034) 0.637*** (0.041) -0.537*** (0.035) -0.053*** (0.004) -0.864*** (0.125) -0.757*** (0.056) -1.908*** (0.192) Observations Log Likelihood: DF RSquared 458448 5048.640 14 0.0753 -0.048*** (0.012) -1.489* (0.591) -0.538** (0.186) 0.321*** (0.060) 0.628*** (0.059) 0.843*** (0.074) -0.106*** (0.015) 0.001*** (0.000) 0.601*** (0.058) 0.59*** (0.071) -0.935*** (0.061) -0.084*** (0.007) -1.107*** (0.270) -1.029*** (0.104) -1.502*** (0.303) -0.128*** (0.008) -1.415*** (0.262) 0.415*** (0.085) 0.181*** (0.043) 0.232*** (0.044) 0.664*** (0.054) -0.007 (0.013) -0.000** (0.000) 1.155*** (0.041) 0.782*** (0.049) -0.331*** (0.042) -0.035*** (0.005) -0.806*** (0.141) -0.628*** (0.067) -3.214*** (0.248) Village Fixed Effects Binomial Multinomial Logit Logit Any Individual Family Migration Migration Migration -0.101*** (0.007) -1.427*** (0.240) 0.114 (0.081) 0.193*** (0.036) 0.359*** (0.036) 0.723*** (0.045) -0.044*** (0.010) 0.000 (0.000) 0.913*** (0.035) 0.596*** (0.041) -0.540*** (0.036) -0.051*** (0.004) -0.929*** (0.129) -0.591*** (0.070) -2.140*** (0.238) 458448 5851.483 157 0.0846 -0.048*** (0.012) -1.504* (0.591) -0.684*** (0.192) 0.272*** (0.061) 0.576*** (0.059) 0.821*** (0.075) -0.113*** (0.015) 0.001*** (0.000) 0.504*** (0.059) 0.308*** (0.072) -0.924*** (0.062) -0.081*** (0.007) -1.157*** (0.290) -0.767*** (0.125) -1.421*** (0.381) -0.127*** (0.008) -1.404*** (0.263) 0.366*** (0.090) 0.150*** (0.044) 0.229*** (0.045) 0.672*** (0.056) -0.004 (0.013) -0.000** (0.000) 1.127*** (0.042) 0.747*** (0.049) -0.344*** (0.043) -0.035*** (0.005) -0.869*** (0.144) -0.509*** (0.084) -3.597*** (0.304)

Variables Household Land Holdings (logged) Internal Migrant in Previous Year Any Internal Migration History 1-4 Years+

458448 5195.511 28 0.0717

458448 6480.410 314 0.0850

Notes: Robust standard errors in parentheses; *** p<0.001, ** p<0.01, * p<0.05



Table 3: Covariates of Multinomial Logistic Regression on Three-Outcome Migration Model: Young Child Effects Variables Has child 0-1 Year Old Has child 1-4 Year Old Has child 5-9 Year Old Has child 10-14 Year Old Age Age squared Individual -0.766*** (0.070) -0.827*** (0.048) -0.372*** (0.055) -0.093 (0.068) 0.002 (0.019) -0.001** (0.000) Observations Log Likelihood: DF RSquared 458448 7292.9 328 0.0979 Family -0.835*** (0.053) 0.141*** (0.037) -0.076 (0.040) -0.144** (0.046) 0.009 (0.015) -0.001*** (0.000)

Notes: Robust standard errors in parentheses; *** p<0.001, ** p<0.01, * p<0.05



Table 4: Annual Variation in Individual and Family Migration, Main Effects and Land Interactions Year Effects Only Individual 1982 (Omitted) --0.083 (0.141) 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 0.171 (0.133) 0.550*** (0.123) 0.509*** (0.123) 0.210 (0.130) 0.413*** (0.125) 0.304* (0.127) 0.148 (0.131) 0.251* (0.127) 0.189 (0.128) 0.004 (0.131) -0.074 (0.132) -0.215 (0.135) Observations Log Likelihood DF RSquared Family --0.373*** (0.111) 0.052 (0.100) 0.327*** (0.094) 0.227* (0.095) 0.364*** (0.092) 0.472*** (0.091) 0.404*** (0.092) 0.237* (0.095) 0.358*** (0.092) 0.053 (0.097) 0.291** (0.093) 0.275** (0.093) 0.044 (0.096) Year Effects Individual --0.360 (0.260) -0.072 (0.243) 0.161 (0.225) 0.166 (0.226) -0.054 (0.237) 0.089 (0.228) 0.053 (0.232) -0.343 (0.254) -0.026 (0.230) -0.330 (0.239) -0.223 (0.234) -0.228 (0.230) -0.513* (0.243) Family --0.720*** (0.171) -0.348* (0.157) -0.156 (0.147) -0.179 (0.145) -0.002 (0.138) 0.175 (0.134) 0.117 (0.136) -0.030 (0.141) 0.072 (0.136) -0.446** (0.152) -0.135 (0.141) -0.216 (0.142) -0.371* (0.145) Land Interactions Individual -0.151*** (0.046) 1983 -0.068 (0.048) -0.078 (0.042) -0.037 (0.033) -0.050 (0.035) -0.072 (0.039) -0.056 (0.035) -0.077* (0.038) -0.012 (0.044) -0.070 (0.037) -0.004 (0.038) -0.085* (0.038) -0.108** (0.037) -0.066 (0.042) Family -0.266*** (0.033) -0.130*** (0.038) -0.112*** (0.033) -0.087** (0.028) -0.112*** (0.027) -0.125*** (0.025) -0.149*** (0.024) -0.152*** (0.024) -0.159*** (0.026) -0.153*** (0.024) -0.085** (0.028) -0.107*** (0.025) -0.088*** (0.025) -0.112*** (0.026)

458448 7657.267 366 0.101

458448 7751.8 386 0.101

Notes: Robust standard errors in parentheses; *** p<0.001, ** p<0.01, * p<0.05



Figure 1: First Migration and Marriage Status by Year, 1974 Census Cohort of unmarried men age 18+

80% 70% 60% 50%

Unmarried, never migrated

Married, never migrated

30% 20% 10% 0% 1975 1977 1979 1981

Migrated after marriage Migrated before marriage, never returned
1983 1985

Migrated before marriage, returned






Source: ICDDR,B (1996)



Figure 2: Predicted probability of any, family, or individual migration, by land holdings, categorical and linear logged specifications

Source: ICDDR,B (1996). Categorical specification from Appendix Table 1. Linear specification from Table 2, columns 4-6.



Figure 3: Spatial variation in relative risk of family migration in 1988, with and without adjustment for baseline risk

Adjusted for baseline Risk Source: ICDDR,B (1996). Models not shown

Unadjusted for baseline risk



Figure 4: Annual predicted probabilities of family Migration, for landless and median landholding households (with confidence intervals shaded)

Source: Model shown in Table 4, columns 3-6


Migration in and out of the study site and household was recorded only after a period of six months, excluding

most seasonal or circular migration episodes, business trips, or vacations.

Multinomial models must account for the “Independence of Irrelevant Alternatives” (IIA) assumption that the

relative choice between outcomes 1 and 2 would not be affected by elimination of outcome 3. Hausmann specification tests compare coefficients and standard errors of possible two-outcome models with those of the chosen model. All tests show no significant differences between the two-outcome models and the chosen model (at the p<=0.01 level).




Rural-rural moves, which typically involve individual seasonal migration episodes or nuclear family resettlement,

require a separate analytic model that includes destination-specific resources that are relevant in the rural context (land purchase opportunities, local labor opportunities). International moves are difficult to model because income expectations depend more on labor relations and social connections than on education, and because international family migration is a rare event.

All models were also tested using a definition of family migration as involving a husband, wife, and children (if

any were alive). Cases in which a husband and wife moved without their children were rare, and the models produced statistically similar results. Separate models also relaxed the requirement that wives move on the same day as husbands, instead defining family migration as husbands and wives moving in the same week or month. It was also quite rare for wives to move in the same month if they did not move on the same day, and thus results were unchanged.

While estimating a fixed-effects multinomial logistic regression is problematic because of the categorical nature of

the dependent variable, the inclusion of village-level controls in a standard model results in no bias as long as the number of observations per cell (village) approaches infinity. An average cell contains 1,800 married male personyears and 600 unmarried ones.

These groupings build on already-observed clumping patterns. Between 100 and 200 decimals, values are grouped

to the nearest 20 decimal unit (so 110 decimals becomes 100, 130 becomes 120). Above 200 decimals, values are group to the nearest 50 decimal unit.



Sign up to vote on this title
UsefulNot useful