Fragmenting the Family? The Complexity of Household Migration Strategies in Post-apartheid South Africa

Abstract The disruption of family life is one of the important legacies of South Africa’s colonial and apartheid history. Families were undermined by deliberate strategies implemented through the pass laws, forced removals, urban housing policy, and the creation of homelands. Despite the removal of legal restrictions on permanent urban settlement and family co-residence for Africans, patterns of internal and oscillating labor migration have endured, dual or stretched households continue to link urban and rural nodes, children have remained less urbanized than adults, and many grow up without coresident parents. Although children are clearly affected by adult labor migration, they have tended to be ignored in the migration discourse. In this study, we add to the literature by showing how a child lens advances our understanding of the complexities of household arrangements and migration processes for families. In a mixed-methods study, we use nationally representative panel data to describe persistence, and also change, in migration patterns in South Africa when viewed from the perspective of children. We then draw on a detailed case study to explore what factors constrain or permit families to migrate together, or children to join adults at migration destination areas.


Introduction
The disruption of family life is one of the important legacies of South Africa's colonial and apartheid history. The marginalization of Africans in "homelands," where there were few employment opportunities, forced Africans to migrate to "White" urban areas to find employment, but a range of restrictions prevented family migration or permanent settlement at the urban destination. The migrant labor system meant that it was mainly men who worked in urban areas or on the mines, while the rural homelands became places for "surplus" people whose labor contributions were not needed (Platzky and Walker, 1985). Children formed a substantial part of that surplus population, along with women and the elderly.
Despite the removal of legal impediments to permanent urbanization and family co-residence for Africans in the late 1980s, patterns of internal and oscillating labor migration have endured, dual or stretched households continue to link urban and rural nodes, and children have remained less urbanized than adults (Hall and Posel, forthcoming;Posel, 2010). Along with neighboring countries that historically provided migrant labor, South Africa has uniquely high rates of parental absence from children's lives (Hall and Posel, 2012;Martin and Zulaika, 2016;Posel and Devey, 2006). Importantly for children, internal labor migration rates increased in the first decade after apartheid, driven mainly by a rise in the share of migration by prime-age women (Collinson et al., 2007;Posel, 2010;Posel and Casale, 2003). It is only since the lifting of apartheid laws that women have migrated for work in substantial numbers (Williams et al., 2011). At the same time, rates of marriage and union formation, which were already low, continued to fall, remittances declined, and unemployment rates remained persistently high (Hunter, 2010;Posel, 2010;Posel and Rudwick, 2013). Households, and women especially, may have to make difficult choices about how to manage the competing demands of childcare and income generation.
Internationally, and despite efforts to improve data on migration, quantitative surveys have tended to "pay minimal attention to children" (Castaldo et al., 2009: 5), and this is also true in South Africa. Commentators have pointed out that children often migrate "as a consequence of many of the same processes that stimulate adult migration, and in response to living arrangements that emerge due to adult migration" (Hosegood and Ford, 2003: 1). But children do not necessarily migrate together with adults, and it cannot be assumed that their migration patterns follow the timing or directions of parents. Rather, children "participate in migration, both independently, as well as with their parents and caregivers as households relocate" (Richter et al., 2006: 197) and children are also left behind when parents migrate.
In this paper, we address two main research questions. The first question is descriptive and investigates how the migration of children has changed in post-apartheid South Africa, given the absence of legal constraints on family migration and urban settlement. The descriptive analysis, which is undertaken using national quantitative data, identifies both persistence and change in migration patterns from the perspective of children. Migration continues to separate children from their mothers mostly, because mothers migrate away from their children, and also because children migrate away from their mothers. However, with the democratic transition, a sizeable share of children has also migrated with their mother, or to join their mother, and most of this migration has been urban bound.
The second research question then considers what factors constrain or permit the co-migration of mothers and children, or the migration of children to join mothers at destination areas. We explore this question through the lens of a rich qualitative study, which captures decisions about mobility and immobility among a three-generation family.
In the next section, we discuss migration in the South African context and we elaborate upon the conceptual approach which helps to frame the mixed-methods study. In section 3, we present the nationally representative microdata that are analyzed in the study. In Section 4 we describe child-parent coresidential arrangements and how these have changed over the post-apartheid period and in Section 5, we investigate child migration in relation to maternal migration. Section 6 considers the complexity of single and family migration decisions through the analysis of our case study, and in Section 7, we conclude the paper with a discussion of our findings and how a focus on the agency (or constrained agency) of mothers can advance our understanding of family migration dynamics in South Africa and other developing contexts.
2 Context and concepts: migration, family fragmentation, and the fluidity of households Family fragmentation (and concern about it) is not new. Historical studies of kinship have described the extended and complex structure of families in southern Africa, where family members were not always present, and children did not always live with their biological parents (Murray, 1981).
The practice of distributing (and moving) children across households is well documented, and as Russell (2003b: 25) writes, "almost certainly predate[s] the economic and political upheavals of colonialism and industrialism." African children move between households to provide help with errands and companionship to other kin, and according to the household's ability to support children (Russell, 2003b).
Labor migration to (and within) South Africa also predates apartheid; the extended separation of migrants from their home environments was common in the region as far back as the late nineteenth century (Murray, 1981;Russell, 2003a;Walker, 1990 , 2007;Budlender and Lund, 2011;Reynolds and Burman, 1986). The effects on household form would certainly have influenced living arrangements for children, and it was observed that the impact on African household structure "will probably be slow to work itself out … the effects of decades of this system could therefore be expected to survive … the demise of the system itself" (Simkins, 1986: 18).
South Africa and its neighboring countries remain unique in the extent of parental absence from children's lives. A recent international study of childcare and co-residence arrangements found that, of the 77 countries studied, the three with the lowest rates of parental co-residence were South Africa (where 35% of children lived with neither of their biological parents), Namibia (27%), and Swaziland (23%). Lesotho and Zimbabwe also featured in the bottom ten countries (Martin and Zulaika, 2016). What these countries have in common is that they supply labor to South Africa.
Important variations in migration patterns have emerged in the post-apartheid period (Todes et al., 2010), and these may have relevance for children's living arrangements. First, and perhaps because the formal system of migration and labor recruitment no longer exists and mines are no longer major employers, temporary migrants are often involved in insecure and poorly paid work in the informal sector, in domestic employment or security services (Cox et al., 2004). Second, a rise in "temporary" labor migration in South Africa between 1993 and 1999 is largely attributed to the feminization of the labor force and the increasing prevalence of migration among women (Hunter, 2010;Posel and Casale, 2003). In 1993, women made up 29% of all African adults who were nonresident members of a rural household for reasons of employment (or temporary labor migrants); by 2008, this had increased to 37% (Posel, 2010).
When migration is defined more broadly to include any move across a municipal boundary, then Schiel and Leibbrandt (2015) estimate that by 2012, rates of male and female migration had equalized.
In the context of low and falling marriage rates in South Africa, women typically bear both economic and caregiving responsibilities for children (Hatch and Posel, 2018). A persistence or increase in temporary female labor migration therefore suggests that children continue to be "left behind" when mothers migrate, indicating that the constraints on family migration are not just in the form of legal restrictions on urban settlement (Posel, 2010;Potts, 2011).
An aspirations-capabilities migration framework (De Haas, 2014) is useful for framing the kinds of considerations that are likely to affect the migration decisions around family or individual migration. In this framework, migration is driven not just by push-pull factors but also by the capability to be able to respond to those forces and realize aspirations. In these terms, migration is seen as "a function of aspirations and capabilities to migrate within a given set of opportunity structures" and, drawing on Sen's capabilities approach, human mobility is defined as "people's capability (freedom) to choose where to live, including the option to stay" (De Haas, 2014: 2). Where people lack the capability or agency to migrate, De Haas describes this as involuntary immobility, or displacement in place. This concept is particularly relevant to a study of child mobility in the context of adult labor migration, where decisions about child co-migration and co-residence may be informed (or constrained) by factors that are different from those which influence adult capabilities to migrate.
Given women's responsibility for childcare, a key capability for the co-migration of mothers and children is likely to be the availability of support networks in the destination area.
Women's responsibility for childcare is often shared within female networks that span generations. Although urban destinations may offer opportunities for income generation, they can further marginalize women and their children by removing them from established chains of care, and childcare can be unaffordable in the absence of kinship networks (Bray, 2008).
Places of residence at destination areas may also be unfit or unsafe for children. Informal settlements (or squatter camps) are important transitional spaces for urban migration. Informal housing is often used as an initial point of access to the city for migrants who cannot obtain their own property through formal processes (Gilbert and Crankshaw, 1999;Lemanski, 2009), and who are able to access only informal or precarious forms of employment. The main kinds of informality in urban areas are settlements on demarcated or invaded land, and backyard shacks on existing properties. There are known risks associated with informality -particularly for young children, who are vulnerable to a range of threats associated with crowded conditions, poor or absent water, and sanitation and refuse removal services, the use of paraffin stoves and associated risk of fire or poisoning, perpetual problems with drainage and flooding, poor security and policing, and long distances to reach health facilities and schools. Given the risks, mothers may decide that it is in the best interests of their children to live elsewhere even while they themselves must endure these conditions (Meth, 2013;Posel and van der Stoep, 2008).
However, the migration of mothers without their children is also a decision that is only possible with the availability of substitute caregivers, particularly grandmothers, in the sending area (Ardington et al., 2009;Madhavan et al., 2012;. The presence of children at the home of origin, in turn, may serve to sustain ties between urban and rural nodes. Although the terms "fragmentation" or "dissolution" are commonly used to describe household dispersion as a negative consequence of migration patterns, or even abandonment, the practice of leaving children with family members in households of origin could also be regarded as the opposite -a strategy to retain an unfragmented (albeit spatially dispersed) household.
In the remainder of the study, we investigate child migration in relation to adult, and specifically maternal, migration. Categories or ordering of children's migration in relation to that of their mother includes simultaneous or co-migration (where children move together with their mother), chain or staged migration (where children are initially left behind, with the expectation that they will later join their mother at the destination household), circular or reverse migration (where children are sent away from a migrant mother's home to her place of origin), and autonomous or child-led migration (where the child moves independently, or in advance of their mother) (Massey et al., 1993;Orellana et al., 2001).
We use quantitative data first to describe the patterns of parent-child co-residence and to motivate our focus on children's mothers specifically. We then chart the nature of, and the relationship between, child and maternal migration. We complement these national statistics with a qualitative study which probes the factors that affect the capabilities and aspirations of mothers to migrate with, or without, their children.

Quantitative data sources
Given the growing interest in understanding the patterns of population mobility and migration in the region, detailed studies of migration patterns are surprisingly scarce. In particular, little is known about family migration and the dynamics of child mobility and care in relation to adult migration. The scarcity of this research is partly due to the limitations of available data (Posel and Casale, 2003;Todes et al., 2010), including the constraints of cross-sectional or region-specific data, with narrowly defined households and poorly defined intra-household relationships.
To explore child migration in relation to maternal migration, we use data collected in the Despite the low attrition rate for children (less than 10% over the four waves of the panel) there is still the possibility of nonrandom attrition from the sample. Logit regressions were used to estimate the likelihood of child attrition between waves 1 and 4, with a focus on the vital and co-residence status of mothers, as this is directly relevant to child mobility and care in relation to maternal migration. The likelihood of child attrition after wave 1 was negatively and significantly associated with the child's age, suggesting that if attrition was related to movement of the child, then this was more likely to take place in the early years of a child's life. However, the regressions indicate that the variables of interest (related to the mother's vital and co-residence status) do not affect child attrition significantly (see Hall (2017) for further details). 1 We define migration as movement across a municipal district. A migrant child is aged 0-8 years at baseline, who then moved across municipal districts in any of the successive waves. Before exploring patterns of child migration, we describe the extent of parent-child co-residence at the cross-section. To present a longer trend, we also use two cross-sectional national household surveys -the 1993 Project for Statistics on Living Standards and Development (PSLSD) and the 2014 General Household Survey (GHS), which collected information on whether a child's mother or father was a deceased or resident member of the household. Table 1 shows the patterns of reported parental co-residence with African children under 15 years. In 1993, 35% of children had two parents living at home. Co-residence rates then 1 We augmented this analysis of attrition by comparing NIDS wave 4 (2014-15) with a nationally representative crosssectional survey conducted in 2014 (the General Household Survey 2014). The comparison revealed no significant differences in the African child population in terms of provincial distribution, type of area, children's age and sex, or their mothers' vital status and co-residence status (if alive). This confirms that even after attrition the balanced child panel in NIDS offers a reliable representation of the national child population in terms of the relevant variables.

Children's households and parental co-residence arrangements
declined further, reaching as low as 27% in 2008 (when HIV-related orphaning reached its peak) and increased again to 30% in 2017. Table 1 illustrates the strong role of mothers who, in all years, lived together with nearly half of children in the absence of their fathers, whereas only 2-4% of children lived with their fathers but not their mothers.
Orphaning is not the main reason for parental absence, but it did increase as a contributing factor after 1993. Table 2 shows that, from 1993 to 2017, death as a reason for maternal absence rose threefold. By 2017, approximately 3.4 million children were living without a coresident mother, of whom 18% had a mother who was deceased. Far more children lived without a coresident father (9.4 million in total), of whom 15% were paternal orphans.
Parents who are alive but absent from the child's household may still retain contact with their children and support them financially. Table 3 shows the baseline wave of NIDS to describe parental contact and financial support to children, distinguishing between parents who are nonresident household members and those who live elsewhere and are not defined as nonresident members. Children with nonresident parents are more likely to see their parents at least sometimes, compared to those with parents who live elsewhere and are not considered as part of the household. Similarly, children with nonresident parents are considerably more likely to receive financial support from their parents. This supports the idea that nonresidency status, which is often related to the labor migration of parents, signals an attachment with the household of origin and the people in it.
From a child's perspective, many absent parents remain in contact even if they are not part of the household. However, children are far less likely to have contact with, and receive  support from, absent fathers than absent mothers. Only 8% of children whose mothers are not household members "never" see their mother, and well over half see their mothers at least once a month.
Low rates of co-residence between children and fathers also make it difficult to link children to their fathers in the data. In the remainder of the quantitative analysis, we therefore focus on the residential and mobility arrangements between children and their mothers.
The descriptive analysis presented in this section therefore supports the existing literature: many African families in South Africa are dispersed geographically, most children do not have a coresident father, and although children are more likely to live with their mothers than their fathers, maternal co-residence rates have also declined. Rising orphaning rates explain only a small part of maternal absence and, given the increase in female labor migration, it is likely that migration is an important contributing factor: children may still be left behind when their mothers migrate, or those born to migrant mothers may be sent away to be cared for elsewhere. We explore this further in the next section.

Child migration streams
It is not possible to tell whether child migration rates have increased over the long term in South Africa, as there is no reliable baseline for comparison. NIDS provides the first national panel data from which migration can be measured directly rather than through retrospective reporting, which is less reliable. An analysis of the NIDS data found that 2.5 million children (35%) had moved place before the age of 15 years and that nearly one million children (14% of a cohort of seven million) migrated across municipalities from 2008 to 2014 (Hall and Posel, forthcoming). A transition matrix of sending and receiving geotypes (Table 4) suggests multiple migration streams for children between geography types, including both urban-rural and rural-urban migration. There are some differences in the geographic patterns of internal migration of children compared to adults. Consistent with their analysis of adult migration (Schiel and Leibbrandt 2015), we find that the most common directions of child migration are to areas similar to the sending area (i.e., urban-to-urban, or rural-to-rural). In the adult analysis, however, only 26% of migrants from the former homelands had moved to urban areas (compared to 47% of children), while over 70% had remained in rural areas under traditional authority (or the former homelands). However, adults who migrate from an urban area are far more likely than children to remain in an urban area. Schiel and Leibbrandt (2015) found that 85% of adults whose sending areas were urban remained within an urban area after migration, while urban-to-rural migration comprised only 15%. By contrast, Table 4 shows that 36% of children who migrated from an urban area ended up in a rural area.
The multidirectional movement of children may be related to the independent movement of children in the context of adult labor migration, where, for example, children are sent home from urban to rural households, or brought from rural households to join migrant parents in urban areas (see also Bennett et al., 2015). We investigate this further by exploring the relationship between child and maternal migration from 2008 to 2014.

Child and maternal migration
To identify maternal migration status, we used the same definition as for children: mothers were classified as migrants if they had crossed municipal boundaries at least once over the four panel waves. It was not possible to complete this exercise for all the children in the sample as they could not all be matched to mothers, even when it was established that their mothers were alive.
Mothers were mapped to children using data on the residential status of mothers and their unique person identifiers. Of the 3,750 children in the balanced sample, 488 were defined as having migrant mothers (12% when weighted), 2,433 had nonmigrant mothers, 326 were already maternally orphaned in wave 1 or their mother died between waves 1 and 4, and 503 had mothers whose migrant status could not be defined because they were not part of the panel. Child migration is strongly associated with maternal co-residence arrangements. Among all children aged 0-8 years in wave 1, whose mother was still alive in wave 4, 22% were not coresident with their mother at wave 1. These children were significantly more likely to migrate over the course of the panel than children who were coresident with their mother at baseline (22% migrated compared to 12% of coresident children), and just over 60% of these child migrations served to unite mother and child.
Most children (0-8 years in wave 1) with living mothers were coresident with their mother at baseline (78%). Among these children, migration is clearly correlated with maternal migration -specifically with maternal work-seeking behavior. This is shown with a logit regression (Table 5), which estimates the likelihood of child migration in relation to mother's migration status and baseline employment status. The analysis is restricted to children with coresident mothers in wave 1 whose mothers were still alive in wave 4.
Children whose mothers migrated during the course of the panel were 43 times more likely to migrate than those whose mothers did not migrate, when controlling for baseline maternal employment status, and the child's age and geographic area type. Those whose mothers were actively seeking work were two-and-a-half times more likely to migrate than those whose mothers were not economically active. The strong association between child and maternal migration is corroborated by analyses from other more localized surveys (Bennett et al., 2015;Madhavan et al., 2012).
However, a relationship between child and maternal migration does not imply that these migration events occur at the same time or in the same direction. More than half of all children who migrated from 2008 to 2014 did so independently of their mother in terms of timing.
Moreover, migration events (defined as an event where either the child or mother (or both) Log pseudo-likelihood = −1,143,443.9 Notes: The sample consists of African children who were aged 0-8 years in wave 1 and initially coresident with their mothers, and whose mothers were still alive at wave 4. The data are weighted to be nationally representative. Omitted categories: not economically active and traditional authority rural areas. ***Significant at the 1% level. Source: NIDS waves 1-4. NIDS, National Income Dynamics Study.
migrated over the four waves) do not always unite children with their mothers. Slightly more than half of all migration events separated mothers from their children or maintained their separation. Nonetheless, this means also that almost 50% of child-mother migration events resulted in children living with their mothers, because children migrated either with, or to, their mothers.
The spatial patterns of changing co-residence arrangements in the context of migration are explored by comparing the location of the child's receiving household (urban versus rural) across different categories of mother-child co-residence (Figure 1). The analysis is based only on children in the panel who migrated across municipal boundaries between waves 1 and 4 and distinguishes between children who had coresident, nonresident, or absent mothers in these waves.
Migrant children who were always coresident with their mothers were more likely to end up (or remain) in urban areas than rural areas following migration, although this difference is not statistically significant at the 5% level. Migrant children who were always separate from their mothers were more likely to live in rural areas in wave 4, but this difference is also not significant. The other distributions suggest a clearer link between the geography of moves and maternal co-residence arrangements. Migrant children whose mothers changed from being coresident to nonresident or absent were more likely to be living in rural areas than urban areas in wave 4. These categories can be broadly clustered as migrant children who moved apart from their mothers. Conversely, an urban destination was most likely for children who had lived separately from their mothers in wave 1 (i.e., in households where their mothers were nonresident or absent) but were coresident with their mothers in wave 4. These can be broadly defined as migrant children who united with their mothers.
The most striking and statistically significant contrast in the destination area is among child migrants who joined their absent mother: 78% of children whose mother's residence status had changed from absent to coresident ended up in urban areas. The distributions suggest

Figure 1
Urban/rural location of receiving households for child migrants in wave 4, by change in maternal co-residence status over waves 1-4 Notes: The sample is based on African children aged 0-8 years in wave 1 who experienced a migration event over the panel and whose mothers were alive in wave 4. Panel weights are used. Source: NIDS waves 1 and 4 (mother and child co-residence status); waves 1-4 (child migrant status). NIDS, National Income Dynamics Study. that migration to (or between) urban areas can serve to facilitate co-residence with mothers.
An example of this would be when a child who previously stayed with a grandmother or other relative at the home of origin is sent to live with her migrant mother in the city. In contrast, migration to (or between) rural areas may serve to separate children from mothers. An example is when migrant mothers are unable to manage both work (or work-seeking) activities and care responsibilities for young children and send their children to be cared for at their home of origin.
The analysis of the panel data has shown that children are highly mobile, that their mobility is associated with maternal migration, and that there are overlaps, and also differences, in the geographic patterns of child migration relative to those of adults. The migration of children may separate children from their mothers, or children may be left behind when their mother migrates, suggesting a persistence in patterns of solo adult migration enforced during the years of apartheid. However, the quantitative data also indicate changes in child migration patterns, with the migration of children almost as likely to result in co-residence with (as opposed to separation from) their mothers. In the next section, we present a case study which sheds light on some key factors that may constrain or enhance the capacity of mothers and children to co-migrate, or for children to join their mother at the destination area.

Family migration to the city: a case study
The case study centers on the life history and migration experiences of a single migrant mother, Lindiwe Jali, 2 her two children, and her extended family, spanning three generations. The study was deliberately chosen as an example of a well-trodden internal migration route between the rural Eastern Cape and the Cape Town metropole. It provides a detailed account of life at the rural and urban ends of a migration path, and of multiple migrations within a single family.
The effect of these movements is to link family members, through sequential migration, at the urban end, while retaining an occupied home of origin at the rural end. Officially named Imizamo Yethu, Mandela Park is the name used by most residents, including Lindiwe. exception is very young children: the pyramid suggests that many children move away after infancy, a form of child migration that fits with one of the migration streams described in the previous section. Mandela Park is a relatively new residential area, only settled in the mid-1990s, so there is no intergenerational history and few elderly people live here. Given that it is mainly informal, suffers regular shack fires, and is severely under-serviced, with about 20 households sharing four toilets and one tap for drinking water, it is not a place where people would want to retire.
As with quantitative studies, qualitative research can gather retrospective migration histories or participants can be followed over time. The approach used in the case study was a combination of the two, although the main migration events emerged through retrospective reporting. Lindiwe's initial account and its implied causal flow fitted neatly with the idea that if alternative care were available at the home of origin, a migrant mother would be inclined to delay bringing her children to live with her until she had some security of employment and housing and found affordable childcare. However, over the nearly 2 years of research with the same family (starting in 2015), the details and timing of migration events altered many timesa clear demonstration of the recall problems that quantitative surveys are likely to face.
Lindiwe's childhood household was what would be described from survey data as a complex, three-generation household. To Lindiwe, it was not complex; it was fairly normal compared to other households she knew as a child. Childcare arrangements were shared around the extended family according to the needs of the children and of adults -an example of household fluidity that is not easily captured by surveys. At one stage, Lindiwe's youngest brother was sent to live with an aunt in a neighboring district to provide company and support because the aunt did not have children of her own.
Important childhood events occurred that set the course of Lindiwe's life: a rape that resulted in her first pregnancy at the age of 15 years and the birth of her daughter, Asanda; the emotional and financial shock when her father was murdered over a stolen horse; the fact that Asanda was brought up as the child of Lindiwe's mother; and that Lindiwe was forced to drop out of school soon after the birth of her second child, Sipho, when she was sent away to work so that she could remit money to the family.
Lindiwe's father had been the pioneer migrant of the family, working in Cape Town until his retirement. After his death, her mother, Noluthando, sent Lindiwe's beloved oldest brother, Bongani, to work in Cape Town so he could send money back to the household. The household still did not have enough money to survive, so Noluthando also started working as a bus operator, taking tickets and looking after the passengers on the bus line from the Eastern Cape to Cape Town. Lindiwe, now 16 years, was the eldest child at home and was left in charge of her younger siblings and her baby daughter, who was also regarded as a sibling. Their aunt, who lived at a neighboring homestead, kept an eye on the children and made sure they had food.
When asked how she made the decision to do work that required travel, leaving her children behind, Noluthando explained that there was no other option: When you see the situation, you act, you get up and close it because life doesn't stand still.
For us people life changes … and so when life changes you decide. I can't even tell my neighbour [the plan]. I just call them to tell them, 'Please look after those children', you see that? You can feel the hardship, but you also have to do whatever. What I mean to say is that when they are left alone it's not that they are not loved, that is the truth, but it's because of the situation.
This theme, the absence of choice or agency, recurred throughout our conversations with both Lindiwe and her mother when we were discussing migration and childcare decisions. In fact, the question of where children would live in relation to their caregivers was hardly ever described as a choice between options -just a necessary fact, without alternatives.
When Lindiwe was 20 years and had recently given birth to her second child, Noluthando decided Lindiwe should go to the city and work, partly to supplement the family income and partly to care for her brother Bongani, who had become sick in Cape Town and had stopped working. Although Lindiwe had not yet finished high school, her mother, desperately working to support all her dependants, could not think about the needs of an individual child -her focus was on the survival of the family. Lindiwe herself wanted to complete her schooling and stay with her children, but she was not in a position to argue -she was the oldest child and had a responsibility to the family. Ultimately, it was Noluthando's decision. Lindiwe was very clear: "[The decision] was made by her and not by me." Lindiwe and her brother stayed with her aunt's cousin in her shack in Mandela Park, and the cousin helped Lindiwe get a job at a local shebeen (tavern) while she looked for domestic work. The shebeen work was insecure, erratic, and dangerous, keeping her out at night.
Although there were no formal restrictions on family migration and urban settlement, there was no question of Lindiwe taking her own children with her to Cape Town. At first, she thought it was a temporary trip. As time went on, it was clear that this was no place for the children: three adults living in a small, crowded, one-roomed shack, one of them bedridden most of the time and the two women working (or seeking work), with nobody at home who could look after the children. Rather, her daughter Asanda stayed with Noluthando, who had stopped working on the buses altogether, and her son Sipho stayed with his paternal grandmother in another village close to her rural home. Asanda turned 6 years of age and Sipho turned 2 years of age.
They were in a state of what De Haas (2014) would term involuntary immobility in relation to their mother. Lindiwe recounts the following: "Oh it was difficult, more especially as I had left Sipho very young …. All along I knew that he was safe where he was because he was with his grandmother, but it was painful -I mean it was not easy." The following year, Bongani's health deteriorated so badly that they knew it was terminal and he was taken back to the Eastern Cape, where he died. Lindiwe stayed on in Cape Town, doing night shifts at the shebeen and looking for domestic work by day.
Back in the Eastern Cape, Asanda was 8 years old and was going to a local preschool, but it was time for her to start formal schooling. Lindiwe and her mother agreed that Asanda would move to join Lindiwe in Cape Town so she could start school there. "The reason why is because when I got here to Cape Town I found that it's better to go to school here. As I went to school in the Eastern Cape, I didn't learn English, and then I found that to learn in Xhosa didn't help me. So I wanted my children to have a better education than me." Noluthando moved to Cape Town so that she could look after Asanda, but little Sipho remained in the Eastern Cape with his paternal grandmother. The family in Cape Town stayed in a shack belonging to a cousin until they could find a place of their own.
The informal settlement of Mandela Park is regularly gutted by shack fires that sweep through the area: during the early part of the qualitative research for this study, the media reported six fires over a 6-month period, destroying hundreds of shacks. The cousin's shack, in which they were all living, burned down in the first year that they lived together.
Lindiwe, together with her daughter, mother, and younger sister moved into a new shack together, which burned down the following year. It is impossible to get insurance against shack fires, and even the stokvels (group savings schemes) do not cover this kind of risk.
Each time there was a fire, the family had to start building and furnishing a new home from scratch.
After the second fire, Lindiwe and her mother built separate shacks, and Sipho (now aged 6 years) came to live in Cape Town. There were two main reasons for Sipho's move. First, he was due to start Grade 1 the following year and Lindiwe wanted him to benefit from the better schooling available in Cape Town too. Second, Lindiwe had finally managed to get a job as a domestic worker and, although she only worked 2 days a week and was earning very little, at least her income was stable. She had always intended to have both her children with her in Cape Town once she found employment.
Sipho's migration and the subsequent co-residence of the two children with their mother and grandmother were therefore enabled by a range of factors and considerations including a housing arrangement which, although still risky, poorly serviced, and inadequately small, was their own and could accommodate the children; a small but relatively stable income that could provide for their basic needs; the availability of the grandmother to care for the children while Lindiwe was at work; and an educational aspiration for the children. The women in the family took a series of deliberate and strategic steps toward the outcomes they aspired to, with a long view to a better life for the next generation. This entailed making sacrifices, working within a constrained set of options (many of which, such as housing backlogs and unemployment, are structural), coping with a range of shocks and setbacks, and responding to opportunities as they arose.
The siblings Sipho and Asanda were now living together for the first time they could remember, but they stayed with their grandmother Noluthando in her shack, while Lindiwe stayed in a separate shack. Lindiwe described this diversification of the family dwellings as an insurance strategy in case of further shack fires. Having a second shack at a slight distance meant there was a lockable space to store the contents of one shack if it was under threat of fire and, provided one of the shacks survived, there would still be shelter for the family. The distance between the shacks was therefore an important consideration: they needed to be close enough for family life to continue but far enough to have a chance of one shack surviving a fire if the other was destroyed. Although they operated as a single household, sharing resources and eating their meals together, this configuration would be impossible to construct from survey data. In a survey, they would appear as two distinct households: a single-adult household and a skip-generation household where two children do not have a coresident mother.
Lindiwe continued to support the children and her mother, buying their food and other necessities. In 2006, she found a second domestic job and was working full time. She supported the whole family, including her younger brother who was finishing high school, and her sister who was unemployed and living with her boyfriend. In 2008, Noluthando's shack burned down, but in 2010, partly through her involvement in local politics and her contacts, Noluthando fast-tracked her way up a waiting list and secured a state-provided formal house.
However, she only stayed in the house with her two grandchildren for 1 year.
Her whole family had now moved away from the Eastern Cape, and house there was locked up and empty. Their house had been broken into numerous times, and their possessions stolen. Noluthando moved back to the Eastern Cape partly because of concerns about their rural home, and also because she felt she had achieved her purpose in Cape Town, which was to ensure that her children could sustain themselves and the next generation. Her grandchildren were also older, and childcare was no longer an issue.
When Noluthando left, Lindiwe moved from her shack into the formal house and lived with her children. Lindiwe was 32 years old, Asanda was 17, and Sipho was 12. It was the first time the three of them had lived together as a mother with two children since Sipho had been born. However, even this was not a "nuclear" arrangement as Lindiwe's younger brother, the children's uncle, was also living with them.
We did a return trip to Lindiwe's rural home at the end of 2015 and another in December 2016. Lindiwe regards her rural home as her main home, the place where she is rooted and with which she maintains a connection. She remits money regularly to her mother and has also invested in building a separate room at the rural homestead, where she can enjoy some privacy when she returns. She distinguished between being a resident and being a citizen, using those English words. In Cape Town, she was a resident. She had a house there, a job, and a daily life with her children. But in the Eastern Cape, she was a citizen. This place was her true home, although the homestead was her mother's. And so, like many other South Africans, she oscillated between urban and rural nodes that were firmly connected through historical migration patterns, being a member of both households but not feeling fully content in either.
Lindiwe expects to be buried in her home village, although she is not sure whether she would retire there. One of the challenges over the course of our interviews was to try to determine whether she would be classified as a "permanent" or "temporary" migrant. Through the research, it became apparent that these categories are not easily definable and that adult migrants may themselves vacillate between a sense of temporary and permanent status in the city. But she does not imagine that her children would ever want to live permanently at the rural home, suggesting a generational shift in attachment to place, and therefore also in circular migration -the strength of the ties depends on where one grew up.

Discussion and conclusion
Surveys have only limited ability to capture the extent of mobility between households, and the nature of fluidity in and out of households; and they therefore offer at most partial insights into family and social networks. But national household surveys are indispensable for describing broad national patterns and trends that cannot be captured through smaller surveys or more nuanced qualitative work. The quantitative data presented in this study describe high rates of parental absence from children's households in South Africa, which is mostly not accounted for by parental death. These patterns suggest that parents and children in South Africa continue to be separated through processes of migration. A child lens on migration reveals that this is not only because of adult migration but also because of mobility among children.
In this study, we investigated child migration in relation to maternal migration. An analysis of national longitudinal data in the NIDS panel showed that children whose mothers had migrated over the course of the panel were far more likely to move than those whose mothers had not migrated. However, the analysis showed further that children not only migrated away from their mother but also toward their mother. Among all children, just under 50% of all migration events (where a child, mother, or both migrated) ended in the co-residence (rather than separation) of children and their mother, either because they migrated together or because children migrated to their mother -as was illustrated in the qualitative study.
The qualitative study provides further insight into these different patterns of child migration (or non-migration) in relation to maternal migration. First, the case study affirms and provides context for the patterns observed in the national data: that children mostly live with mothers, but a substantial share do not; that fathers rarely feature, but that grandmothers and extended family play a crucial role in providing childcare; that women carry an enormous financial burden, and that to support to their family, mothers may be forced to migrate without their children. In this sense, migration is not about individual agency or aspiration, but a livelihood balancing act to ensure the survival and well-being of the broader family, even at the cost of a mother living without her biological children.
Second, the focus on a single migrant and her family made it possible to trace the connectedness of the rural and urban homes, which are part of the same "single social field" (Trager, 1991: vii). The multiple and multidirectional moves between rural and urban areas illustrate the stepwise migration of children in relation to their mother, as well as other forms of mobility. There are large moves across provinces, and also small moves as the family rebuilds destroyed shacks, reconfigures itself into separate households, and then reunites in a formal house. Not all these moves would be captured as geographic migration or even mobility across place. Including the extended family highlights processes of cumulative causation and chain migration, successive members provide the next in-migrants with accommodation when they arrive in the city.
Third, the case study demonstrates that children's migration is deliberate and strategic, but that it is also strongly influenced by the capabilities of the rural and urban household, including the availability of childcare, living conditions, and opportunity structures. Child migration may happen at different times for siblings depending on the available care arrangements and on their life stage and school readiness; it may be delayed far longer than planned, or happen prematurely, before the hoped-for living conditions are in place. Qualitative research can capture, in ways that surveys do not, the plans and aspirations of families and how these are modified over time as capabilities and circumstances change or fail to change in anticipated ways. For example, despite her desire to be with her children, the central character, a migrant mother, fails to have them living with her for much of their childhood, even after they migrate to the same place.