Demography

, Volume 51, Issue 1, pp 205–228

The Impact of Family Transitions on Child Fostering in Rural Malawi

Authors

    • Department of Sociology, Center for Demography and EcologyUniversity of Wisconsin–Madison
  • Sara Yeatman
    • Department of Health and Behavioral SciencesUniversity of Colorado Denver
Article

DOI: 10.1007/s13524-013-0239-8

Cite this article as:
Grant, M.J. & Yeatman, S. Demography (2014) 51: 205. doi:10.1007/s13524-013-0239-8

Abstract

Despite the frequency of divorce and remarriage across much of sub-Saharan Africa, little is known about what these events mean for the living arrangements of children. We use longitudinal data from rural Malawi to examine the effects of family transitions on the prevalence and incidence of child fostering, or children residing apart from their living parents. We find that between 7 % and 15 % of children aged 3–14 are out-fostered over the two-year intersurvey period. Although divorce appears to be a significant driver of child fostering in the cross-sectional analysis, it is not significantly associated with the incidence of out-fostering. In contrast, maternal remarriage has both a lagged and an immediate effect on the incidence of out-fostering. Furthermore, the likelihood of out-fostering is even higher among children whose mother remarried and had a new child during the intersurvey period. Using longitudinal data collected from living mothers rather than from children’s current foster homes offers new insights into the reasons children are sent to live with others besides their parents.

Keywords

DivorceRemarriageChild fosteringSub-Saharan Africa

Introduction

The out-fostering of children is a common practice across sub-Saharan Africa (Bicego et al. 2003; Hosegood et al. 2007; Monasch and Boerma 2004). Child fostering is the situation in which children with living parents are sent to live in a different household, most often with other relatives. These arrangements are informally negotiated between family members and are usually temporary. Fostering has long been recognized by demographers as a risk-coping mechanism used by households to offset economic or demographic hardships, to take advantage of resources available through extended kin networks, and to redistribute the costs and benefits of childbearing across the extended family (Akresh 2009; Isiugo-Abanihe 1985; Lloyd and Desai 1992; Madhavan 2004; Mason 1997; McDaniel and Zulu 1996; Notermans 2004).

Although most research (largely anthropological) on the practice of child fostering has focused on West Africa, the practice regularly occurs throughout sub-Saharan Africa, including in the AIDS-burdened areas of eastern and southern Africa. Across the region, only 63 % of children live with both parents, and 10 % of children with two living parents live with neither of those parents (Monasch and Boerma 2004). The prevalence of out-fostering is highest in southern Africa, where as many as a quarter of non-orphaned children in Botswana, Namibia, and South Africa do not live with either parent (Monasch and Boerma 2004). In Malawi, 14 % of children under age 15 are not coresident with their living mother, and 31 % are not coresident with their living father (National Statistics Office and ICF Macro 2011).

The literature on the effects of living apart from one’s biological parents in sub-Saharan Africa has been inconclusive (Verhoef and Morelli 2007). This should not be surprising: the reasons children live apart from their parents are varied and carry diametric implications (Serra 2009; Verhoef and Morelli 2007). Most demographic research on the causes of child fostering uses data on children after they have been fostered and focuses on the characteristics of the household into which the children have moved. Although these approaches can be informative, they rely on retrospective accounts of the situation before the child was fostered. In contrast, anthropologic studies provide great depth into the practice of fostering within a village or certain ethnic group but are less informative for answering questions about the breadth of fostering and the most common reasons children are out-fostered.

In this article, we take a unique approach to understanding the causes of child out-fostering in rural Malawi. We use data from living parents about their children’s living arrangements and prospectively follow children over time to examine not just the household characteristics associated with prevalent out-fostering but also the family transitions associated with the incidence of fostering over a two-year period. We focus on divorce and remarriage, which are commonly cited in the literature, but also include other aspects of family change. Our approach enables us to examine both the risk of out-fostering to children living in families that have experienced recent transitions, as well as the relative contribution of these transitions to the incidence of fostering.

Family Change as a Potential Driver of Child Out-Fostering

An estimated 30 % to 50 % of marriages in sub-Saharan Africa end in divorce (Lesthaeghe 1989; Locoh and Thiriat 1995; Reniers 2003; Tilson and Larson 2000), and recent research suggests that divorce rates in AIDS-burdened regions may be increasing as partners use divorce as a strategy to mitigate their HIV risk (Anglewicz 2012; Porter et al. 2004; Reniers 2008; Schatz 2005; Smith and Watkins 2005). Little is known, however, about the implications of divorce for children in this context. Studies from the West have shown that divorce decreases economic resources and social capital available to children and is associated with worse behavioral, schooling, and family outcomes for children (Amato 2000; Amato and DeBoer 2001; McLanahan and Percheski 2008; Seltzer 1994; Sigle-Rushton and McLanahan 2004). Additionally, instability in family structure has been associated with negative outcomes independent of family structure itself (Fomby and Cherlin 2007; Osborne and McLanahan 2007). The consequences of divorce may be more severe in low-income countries, where recent evidence suggests it could raise the risk of child mortality (Clark and Hamplova 2013). One mechanism through which divorce can affect child well-being is through the disruption of household living arrangements (Brown 2004; Magnuson and Berger 2009; Peterson and Zill 1986; Thomson et al. 1994), such as the out-fostering of children.

Anthropologists have long identified divorce as a contributing factor for child fostering (e.g., Alber 2004; Bledsoe and Isiugo-Abanihe 1989; Carsten 1991; Castle 1996; Goody 1973; Oppong 1973; Sinclair 1972; Verhoef and Morelli 2007). Parents may out-foster their children to other relatives after divorce if neither parent independently is able to care for the child or to buffer children from other disruptions during the change in family structure. Child fostering may also precede divorce, such as in cases where women in patrilineal communities organize fostering arrangements in order to secure future contact with children (Alber 2004).

Even if divorce itself does not lead to child fostering, parental remarriage could. In patrilineal ethnic groups, children belong to their father’s clan and inherit resources through their father’s lineage. Although such children may remain with their father’s family following a divorce, they may be out-fostered to other family members when their father remarries or if their father is polygynous, for fear their stepmother might neglect them (Bledsoe 1993; Isiugo-Abanihe 1985). In some cases, children from patrilineal ethnic groups are allowed to remain with their mother following a divorce, particularly if the children are young. However, these children may be out-fostered when their mother remarries because they do not belong to the lineage of their mother’s new husband (Silk 1987). In matrilineal ethnic groups, children belong to their mother’s clan and are expected to remain with their mother following a divorce. Although remarriage in matrilineal ethnic groups does not create a problem for children’s family membership, it may still lead to out-fostering because of concerns that the presence of children from previous unions will create problems for a new marriage (Goody 1982; Mkandawire-Valhmu et al. 2013).

The birth of a new child may also trigger out-fostering, particularly for children whose mother has remarried. In patrilineal ethnic groups, new births create a formal link between a woman and her husband’s lineage, which may further marginalize children from previous unions (Bledsoe 1990). New births may also create pressures on the resources within the household. Previous studies have cited child fostering as a postnatal means of controlling family size in high-fertility societies (Mason 1997; Oppong and Bleek 1982). A positive association between child fostering and a mother’s number of surviving children has been identified across a range of African contexts (Desai 1992; Isiugo-Abanihe 1985). Other studies, however, have noted the importance of the age distribution of surviving children: a cross-sectional study of out-fostering in Senegal, for example, found that the presence of more surviving children aged 0–5 was positively associated with out-fostering for this age group, whereas the presence of more surviving girls aged 6–14 decreased the fostering of younger children (Vandermeersch 2002).

In this article, we examine the diverse predictors of child out-fostering in rural Malawi. Our study is unique in two key dimensions. First, our data contain information on children’s living arrangements as reported by their biological parents rather than by their current head of household. Most of what is known at present about children’s living arrangements in sub-Saharan Africa stems from reported information about the current living situation of children (e.g., Demographic and Health Survey data on fostering follow this model). Although informative for certain questions about fostering, this approach can tell us little about the home the child came from and the reason that a child lives or does not live with their biological parents. In other words, it says little about why children are fostered out.

Second, unlike previous research in the area, we use a panel data set that prospectively follows children’s living arrangements over a period of two years. Most existing demographic studies of the reasons for child fostering in sub-Saharan Africa have relied on cross-sectional survey data. These studies compare children who are fostered with children who are not fostered using cross sections of data and retrospective reports on what a child’s birth household and family characteristics were like prior to fostering. Such studies make inferences about the rationale for fostering based on retrospective reporting, which is vulnerable to recall bias, ex post facto rationalizations, and endogeneity—current child and household characteristics used in models may have been influenced by the act of fostering itself (see Evans and Miguel (2007) for a similar discussion regarding the effects of orphanhood). In contrast, our prospective approach allows us to circumvent these pitfalls and to begin to disentangle the order in which key events occur.

We hypothesize that family transitions are closely associated with changes in children’s living arrangements in rural Malawi. In particular, we are interested in the relationship between marital transitions and child out-fostering and what can be learned by examining the predictors of both prevalent and incident child out-fostering. A number of specific hypotheses guide this work. First, we believe that divorce will alter the living arrangements of children, predicting not only that children will live apart from one parent but that they will live apart from both. Second, we hypothesize that maternal remarriage will lead to further disruptions in children’s living arrangements and increase child fostering. Third, we believe that the way in which these marital changes are measured matters. Therefore, we explore not only the association between child fostering and mother’s current marital status and recent family transitions but also the association between child fostering and the parents’ marital outcome, a more detailed variable that describes whether a child was born outside a marriage or whether a child’s parents’ marriage ended in divorce or widowhood. In addition to these marriage transitions, we also hypothesize that other household transitions, such as new births to a child’s mother, will increase the odds of out-fostering.

Data and Methods

Our analysis uses data from the Malawi Longitudinal Survey of Families and Health (MLSFH),1 a longitudinal survey that followed women and their spouses from 1998 to 2010, to examine the consequences of parental divorce and remarriage on children’s living arrangements. The sample was refreshed in 2004, with the addition of adolescents aged 15–24; at the 2008 survey round, these respondents accounted for 20 % of completed interviews. We use data from the fourth and fifth survey rounds of the MLSFH, collected in 2006 and 2008. Beginning with the 2004 survey round, the MLSFH focused on the consequences of HIV and collected rich data on household composition, socioeconomic status, and marriage and sexual partner histories, as well as providing household-based HIV testing (MLSFH 2009). The 2006 survey round differed from previous years: in addition to collecting data on all household members, respondents were instructed to list any biological children who lived elsewhere. The 2008 MLFSH collected additional information about nonresident biological children; for children aged 5 and older who were not coresident, the survey asked the respondent to identify with whom the child lived.

Study Context

MLSFH data were collected in three rural districts representing the three main regions of Malawi. In Rumphi in the north, most respondents are Tumbuka, an ethnic group that practices patrilineal descent and for whom residence is predominantly patrilocal. In Balaka in the south, most respondents are Yao and follow a matrilineal descent system and matrilocal residence; the Yao are also predominantly Muslim (Mitchell 1956). Finally, Mchinji is located in the central part of the country, and most Mchinji respondents are Chewa, an ethnic group that was historically matrilineal and matrilocal but which has begun adopting more patrilineal practices following the spread of cash cropping in the region (Phiri 1983; Takane 2008).

Marriage in Malawi is early and universal. It also frequently ends in divorce, quickly followed by remarriage. Within this general pattern, however, practices vary by ethnic groups. Among the matrilineal Yao, marriage is particularly fluid (Kaler 2001; Peters 1997; Tawfik 2003). It is also relatively unstable, with close to 60 % of first marriages ending in divorce within 20 years; of these divorced women, approximately 50 % remarry within 2 years of divorce, and 90 % remarry within 10 years (Reniers 2003). In contrast, among the patrilineal Tumbuka, marriage is more stable but divorce is not uncommon: approximately 35 % of first marriages end in divorce within 20 years, with 30 % of women remarrying within 2 years of divorce and 78 % remarrying within 10 years (Reniers 2003). Reniers also found that between 38 % and 65 % of women in Malawi who remarry enter a polygynous union, with the prevalence of polygyny higher in the patrilineal regions.

In Malawi, children from patrilineal ethnic groups belong to their father’s lineage, but it is customary for children to remain under their mother’s care until they come of age (Wanda 1988). In contrast, among matrilineal tribes in southern Malawi, children belong to their mother’s clan and, although children may maintain a relationship with their father following divorce, they would most likely live with their mother or her extended family (Peters 1997).

Analytic Sample

Although the MLSFH followed adult female respondents longitudinally, the biological children identified by respondents in the household rosters were manually linked. We used three criteria to match children: name, age, and gender. A longitudinal link was based on successfully matching any two of the three criteria across the rosters of the same respondent. Age was considered flexibly, in tandem with birth order, given the known patterns of age misreporting in the MLSFH sample (Bignami-Van Assche et al. 2003). Approximately 75 % of child matches satisfied all three criteria, and less than 5 % of matches were made on the basis of age and gender but not name.

Our analysis begins by examining the child and family characteristics associated with child fostering among the 3,217 children aged 3–14 listed by their mothers in the 2006 survey round with no missing data. We limit our sample to children aged 3–14 in 2006 because questions about children’s living arrangements in 2008 were limited to children aged 5 and older. When available, data from men are used to verify the living arrangements of children who did not live with their mother; however, only half of the spouses of the female respondents were interviewed in both survey rounds, and many of these husbands were not the biological fathers of the focal children. Next, we examine the factors associated with the incidence of child fostering over the 2006–2008 survey period. We restrict the second analysis to the 2,248 children aged 3–14 in 2006 who were reported to live with a parent at the time of the survey and for whom longitudinal data were available.

Table 1 compares the sample characteristics of the cross-sectional and longitudinal samples. Many of the differences can be attributed to the selection of the longitudinal sample on parental coresidence status in 2006, but sample attrition may also play a role. Longitudinal information was available for only 73 % of the children who were coresident with a parent in 2006. There are two broad types of explanation for the “missing” children. First, mothers (respondents) were lost to follow-up between the two rounds, which accounts for 74 % of all “missing” children. Second, children were listed by their mother on the 2006 roster but could not be matched on the 2008 roster due to death, reporting differences, or interviewer error. Therefore, we estimate an “upper bound” of incident child out-fostering by assuming that all children in the latter category were out-fostered, which adds 218 children to the final sample. This specification accounts for possible measurement error if children who are considered permanent residents of another household are more likely to be omitted by their mothers.
Table 1

Distribution of variables, cross-sectional and longitudinal samples: Malawi Longitudinal Survey of Families and Households, 2006–2008

 

Cross-sectional

Longitudinal

“Upper Bound” Longitudinal

N

3,217

2,248

2,466

Fostered, 2006

4.72

––

––

Out-Fostered, 2006–2008

––

6.99

15.21

Family Transition Variables

 Mother’s current marital status, 2006

   Still married to child’s father (ref.)

74.02

80.27

79.16

   Not married

11.93

7.71

8.19

   Remarried

14.05

12.02

12.65

 Parents’ marital outcome, 2006

   Still married (ref.)

74.02

80.27

79.16

   Nonmarital birth

11.00

5.79

6.61

   Divorced/separated

10.85

9.80

9.89

   Widowed

4.10

4.14

4.34

 Mother divorced, 2006–2008

––

6.51

6.53

 Mother remarried, 2006–2008

––

3.34

3.86

 New birth 2006–2008

––

32.10

30.94

Child Covariates

 Female child

50.09

49.73

49.96

 Age of child, 2006

   3–4

18.18

18.79

18.61

   5–9 (ref.)

44.00

44.34

44.41

   10–14

37.82

36.87

36.98

 Child in worse relative health, 2006

4.10

4.23

4.26

 Ethnic group

   Yao (ref.)

27.03

25.02

25.87

   Chewa

28.78

28.58

28.39

   Tumbuka

29.74

31.66

30.70

   Other ethnic group

14.45

14.74

15.04

 Number of living siblings, age 0–14 (mean)

 2.31

2.69

2.67

Maternal Covariates

 Mother’s age, 2006

   18–29

20.54

19.23

19.26

   30–34 (ref.)

22.78

23.24

22.91

   35–39

21.16

22.44

21.86

   40–44

15.54

16.21

15.94

   45+

19.98

18.88

20.03

 Mother is in polygynous marriage

27.97

27.87

27.90

 Mother is HIV positive

6.18

4.90

4.95

 Mother is in bad health

37.13

35.98

37.47

Household Covariates

 Family death, 2006–2008

––

15.72

16.67

 Quintile of household asset ownership, 2006

   

   Low (ref.)

16.72

14.38

15.21

   Lower-middle

18.96

19.41

19.30

   Middle

23.37

23.64

23.48

   Higher-middle

20.32

20.97

20.72

   High

20.63

21.59

21.29

Variables of Interest

Child Fostering

Our key dependent variable is child fostering, which we define as children who do not live with a biological parent. Children who live in the same compound but not the same household as their parent are assumed to have regular contact and to share resources with their parents, distinguishing them from children who live elsewhere. In the cross-sectional analysis, we use mothers’ reports of whether their biological children were coresident at the time of the 2006 survey. When fathers were interviewed, we use that information to confirm that children who were not resident with mothers were also not resident with fathers. However, no information about coresidence with fathers was available for 43 children; all of these children are assumed to be fostered in 2006 and account for 27 % of children who were fostered in 2006. Eleven children were identified as not living with their mothers but were listed as coresident with their father; these children are coded as being coresident with a parent in 2006.

The second part of the analysis examines the incidence of out-fostering over the 2006–2008 survey interval. Children are considered to be out-fostered during the study period if they lived with a biological parent at the time of the 2006 survey but no longer lived with either of their biological parents in 2008. In order to identify out-fostered children, we first examine whether a child was coresident with his or her mother in 2008. If a child in the longitudinal sample was not coresident with the mother in 2008, we then examine with whom the child was reported to live. If the child was recorded as living with someone other than the father, we consider that child to have been out-fostered between the two survey rounds. Of children who did not reside with their mother in 2008, less than 9 % were living with their father. In our final regression model, we estimate an “upper bound” for incident out-fostering and assume that all unmatched children who were coresident with a parent in 2006 and whose mother was reinterviewed in 2008 were out-fostered. This specification most likely overestimates the incidence of out-fostering by including children who may have died or been omitted because of other types of misreporting.

Family Transition Variables

Our key independent variables are measures of marital status and changes in marital status over the study period, constructed from the marital histories and marital status variables collected in 2006 and 2008. Because almost all children in our sample lived with their mother and not their father in the event of separation or divorce, we focus on mother’s marital status and marital outcome with the child’s father. First, we construct two child-specific sets of variables for the marital status of a child’s parents in 2006 by comparing a child’s age and estimated year of birth to the respondent’s marital history. We use the year that each union began and ended to define the child’s mother’s marital status in 2006: currently married to the child’s father, not married, and married to someone other than the child’s father (i.e., remarried). The category not married includes mothers who were never married, were divorced, or were widowed at the time of the 2006 survey. The categories not married and remarried are included in the regressions as two binary variables. The second variable identifies the marital outcome of a child’s parents in 2006: still married, divorced/separated, widowed, and nonmarital birth.2 When used in conjunction with the binary variable for whether a child’s mother was remarried by 2006, the parents’ marital outcome variable combines current status information with retrospective data on the fate of a child’s parents’ union. This combination allows us to examine whether past family transitions have consequences for children’s current living arrangements and incident out-fostering. Then, we create two binary variables to identify women who divorced or remarried during the intersurvey period.

In addition to the marital status and change variables, we also include a variable for the birth of a new child to the mother. We examine the association between out-fostering and both the birth of a new sibling and the interaction between a new birth and maternal remarriage.

Control Variables

Our analysis includes child-level, mother-level, and household-level control variables (see Table 1). Unless otherwise noted, all values are taken from the 2006 data. Child-level variables include the child’s gender, age, health status, and number of living siblings aged 0–14 years, as reported by their mother. The child’s health status is coded as a binary variable for whether the child was in worse health relative to other children of the same age living in the respondent’s village. A child’s ethnicity is assumed to be the same as that of the mother. Ethnicity is categorized as Yao, Chewa, Tumbuka, and other ethnic groups that are present in numbers too small to be analyzed separately (including the Lomwe, Ngoni, Sena, Tonga, and Senga).

We also examine several dimensions of the mother’s sociodemographic characteristics and health. We include a categorical variable for the mother’s age and a binary variable for whether the mother was in a polygynous marriage. We include two indicators of the mother’s own health because poor health may be associated with marital instability and children’s living arrangements. The first is an indicator of whether the respondent knew that she was HIV positive. After the completion of the interviewer-administered survey in 2004 and 2006, respondents were offered a home-based HIV test by a trained counselor. By 2006, 99 % of the analytic sample had learned their HIV status through the survey’s testing procedures. The second indicator is a measure of the respondent’s self-reported health status in 2006. This indicator is coded as a binary variable to represent reports of “fair” or “bad” health relative to “good” or “excellent” health.

We also include a measure of the household’s asset ownership. Household asset ownership is based on household ownership of a set of 17 assets. Asset ownership is then run through a principal components analysis and divided into quintiles. For the 86 respondents with a missing asset index, we take the average of the reported asset index from 2004 and 2008.

Finally, we include a control for whether a family member died between the 2006 and 2008 survey rounds. The death of a family member includes all deaths during the intersurvey period to individuals aged 15 and older listed in the family roster, excluding the respondent’s spouse; most of these deaths were the parents or adult children of the respondents. The survey did not capture whether these individuals were coresident at the time of their death.

Model Specification

To test our hypotheses, we use probit regressions to examine the predictors of prevalent child fostering in 2006 and incident out-fostering between the 2006 and 2008 survey rounds. Because the analysis of incident out-fostering is limited to children for whom longitudinal data are available, the selectivity of this group may bias our findings. To account for this selectivity, we use a bivariate probit regression with selection,3 which jointly estimates the probabilities of out-fostering and selection into the longitudinal sample, accounting for a correlation (ρ) between the error terms (Dubin and Rivers 1989). In the event that ρ is equal to zero and the equations are independent, the models are the same as two univariate probit models. The bivariate probit regression with selection recognizes that differences in the independent variables change not only the outcome of interest but also the likelihood that the observation is in the sample.

In the analyses of out-fostering in the main longitudinal sample, the selection variable codes all children who were coresident with a parent in 2006 but not matched in 2008 as censored observations. The selection variable is recoded, however, for our analyses of the “upper bound” sample, such that children who were not matched in 2008 but whose mother was successfully interviewed were moved from the censored to the uncensored category. The selection equation includes all the child, mother, and household variables measured in 2006 and, as the exclusion criteria, an indicator of whether the respondent was successfully interviewed in 2004.4 The out-fostering equation includes the same child, mother, and household variables, plus our measures of family change that occurred between 2006 and 2008. Although the selection and out-fostering regressions are estimated jointly, we present only the coefficients from the out-fostering equation in the Results section; results for the selection equation are available in Online Resource 1.5 The standard errors in all regression models are adjusted to correct for the clustering of multiple children within families.

Results

Descriptive Analysis

Table 1 presents the descriptive characteristics of the two analytic samples. In 2006, 4.7 % of children aged 3–14 were fostered. Two years later, 7.0 % of the children who were coresident with at least one biological parent in 2006 were no longer living with either of their parents. Our alternate specification, or “upper bound,” estimates that 15.3 % of children were out-fostered.

At the first survey round, 74 % of children from the original cross-sectional sample lived in intact biological families. For 12 % of children, their mother was not married in 2006. The remaining 14 % of children had a mother who had remarried or was married to someone other than the child’s biological father. In the main longitudinal sample used to estimate the factors associated with the incidence of out-fostering between 2006 and 2008, 80 % of children had mothers who were still married to the child’s father. Almost 8 % had a mother who was not married in 2006, and 12 % had mothers who had remarried.

The second marital status variable identifies whether a child’s parents were still married to each other and, if not, the fate of that union. In the original sample, 11 % of the children were born outside of a union, and an additional 11 % had parents whose marriage ended in a divorce or separation. The remaining 4 % of children had a deceased father. The distribution of this variable is similar across the cross-sectional and longitudinal samples with the notable exception of a lower prevalence of children who were nonmarital births in the longitudinal samples.

Other family variables of interest capture changes in a child’s family structure over the intersurvey period and are available only for the longitudinal samples. In the main longitudinal sample, the mothers of 6.5 % of children divorced between the 2006 and 2008 survey rounds, and 3.3 % of the children’s mothers remarried over the period. Furthermore, the mothers of one-third of the children gave birth to another child during the intersurvey period. Overall, there were no substantive differences between the main and “upper-bound” longitudinal samples.

Our key interest is how the experience of marital dissolution and remarriage affects the likelihood of child fostering. Table 2 shows the prevalence of child fostering in 2006 and the incidence of out-fostering between 2006 and 2008 by these characteristics. Based on the baseline marital characteristics, less than 3 % of children from intact families were fostered in 2006, as compared with 7 % of children whose mother was not married and 13 % of children whose mother had remarried. Only 4.6 % of children whose mother was widowed were fostered in 2006, compared with almost 11 % of children born outside of marriage and 12.6 % of children whose parents were divorced or separated. These variables were also associated with higher rates of incident out-fostering, such that 17.4 % of children with divorced parents in 2006 and almost 14 % of children with nonmarital births were out-fostered over the intersurvey period.
Table 2

Prevalence and incidence of out-fostering, children aged 3–14, 2006–2008

 

Prevalence 2006

Incidence 2006–2008

“Upper-Bound” Incidence 2006–2008

Mother’s Marital Status, 2006

***

***

***

 Still married to child’s father

2.69

4.99

12.14

 Not married

7.31

12.72

25.24

 Remarried

13.27

16.67

27.88

Parents’ Marital Outcome, 2006

***

***

***

 Still married

2.69

4.99

12.14

 Nonmarital birth

10.73

13.85

31.29

 Divorced/separated

12.64

17.35

25.82

 Widowed

4.54

11.70

22.43

Divorced, 2006–2008

 No

 

6.86

15.05

 Yes

 

8.90

17.39

Remarried, 2006–2008

 

***

***

 No

 

6.59

14.31

 Yes

 

18.67

35.79

New Birth to Mother

 

*

*

 No

 

7.73

16.50

 Yes

 

5.37

12.32

N (children)

3,217

2,248

2,466

Notes: We use chi-squared tests to examine the significance of categorical variables. We use t tests of means to examine the significance of binary variables.

*p < .05; ***p < .001

There was no bivariate association between maternal divorce over the intersurvey period and the incidence of child fostering. In contrast, 18.7 % of children whose mother remarried between 2006 and 2008 were out-fostered, compared with only 6.6 % of children whose mother did not enter a new union. Children whose mother had a new baby had a lower incidence of child fostering than children whose mother did not have a new child.

Prevalent Child Fostering, 2006

The regressions presented in Table 3 examine the child- and mother-level characteristics associated with child fostering in the 2006 sample. The first model uses the mother’s current marital status. Children whose mother was not married at the time of the survey were significantly more likely to not be coresident with their mother relative to children whose mother and father were still married to each other. The probability of being fostered was even higher for children whose mother was married to someone other than the child’s father. These model estimates indicate that the predicted probability that a 10- to 14-year-old girl whose mother and father were still married to each other was fostered was only .02, compared with .06 for girls whose mother was not married and .11 for those whose mother had remarried.6 The second model replaces mother’s current marital status with parents’ marital outcome and a binary variable for whether the mother was remarried. Children who were born to unmarried parents or whose parents were separated or divorced were significantly more likely to be fostered relative to children in intact families. There was no association between having a widowed mother and fostering in the cross-sectional analysis. Furthermore, maternal remarriage was nonsignificant when controlling for parents’ marital outcome.
Table 3

Probit regression results, child fostered in 2006, among children aged 3–14

 

Model 1

Model 2

Family Transition Variables

 Mother not married, 2006

0.41**

 

 Mother remarried, 2006

0.73***

0.25

 Parents’ marital outcome, 2006 (ref. = still married)

  Nonmarital birth

 

0.44**

  Separated/divorced

 

0.57***

  Widowed

 

–0.01

Child Covariates

 Female child

0.04

0.05

 Age of child (ref. = 5–9)

  3–4

–0.38**

–0.40**

  10–14

0.28**

0.29**

 Child in worse relative health

–0.30

–0.28

 Ethnic group (ref. = Yao)

  Chewa

0.19

0.19

  Tumbuka

0.40**

0.42**

  Other ethnic group

0.35*

0.38*

 Number of living siblings, aged 0–14

–0.08*

–0.08*

Maternal Covariates

 Mother’s age (ref. = 30–34)

  18–29

0.19

0.20

  35–39

–0.09

–0.09

  40–44

–0.20

–0.18

  45+

–0.05

–0.03

 Mother is in polygynous marriage

–0.01

–0.02

 Mother is HIV positive

0.18

0.21

 Mother is in bad health

0.15

0.15

Household Covariates

 Quintile of household asset ownership (ref. = low)

  Lower-middle

–0.03

–0.06

  Middle

–0.04

–0.07

  Higher-middle

0.00

–0.03

  High

–0.06

–0.08

Constant

–2.11***

–2.11***

Number of Children

3,217

3,217

Number of Mothers

1,317

1,317

Wald Chi-Square

104.97***

111.38***

Log-Likelihood

–545.19

–541.33

*p < .05; **p < .01; ***p < .001

In addition to these family variables, the probability of child fostering increased with a child’s age. There was also a significant negative association between fostering and a child’s number of living siblings younger than 15 years old. Children who were Tumbuka or were coded in the other ethnic group category had a higher probability of being fostered relative to children from the Yao ethnic group. In contrast, these cross-sectional analyses found no association between child fostering and household socioeconomic status, mother’s age, and mother’s health status.

Incident Out-Fostering, 2006–2008

The second set of regressions, presented in Table 4, examine the association between incident child fostering and family transitions experienced between the 2006 and 2008 survey rounds. The first model replicates the mother’s marital status variables used in Model 1 of the cross-sectional analyses and finds that mother’s marital status in 2006 predicts not only prevalent but also incident child fostering. Children whose mother was remarried in 2006 had a significantly higher probability of being out-fostered between 2006 and 2008 than children whose parents were still married to each other in 2006. The second model adds marital transitions during the intersurvey period to the 2006 marital status variables. After controlling for recent divorces and new marriages, we found no change in the association between maternal remarriage and child fostering. Although there was no significant association between a recent divorce and out-fostering, children whose mother entered a new union during the intersurvey period were more likely to be out-fostered. The predicted probability that a 10- to 14-year-old girl was out-fostered over the two-year period was .03 if her parents were continuously married to each other, compared with .09 if her mother was remarried in 2006 and .11 if her mother was not married in 2006 but had remarried by the 2008 survey round. These results suggest that maternal remarriage has both a lagged and an immediate effect on out-fostering.
Table 4

Bivariate probit regression with selection, child out-fostered 2006–2008, children aged 3–14 in 2006

 

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6a

Family Transition Variables

 Mother not married, 2006

0.35

0.08

  

0.05

–0.11

 Mother remarried, 2006

0.55***

0.55***

0.12

0.39

  

 Parents’ marital outcome, 2006 (ref. = still married)

  Nonmarital birth

  

0.33

0.06

  

  Separated/divorced

  

0.55*

0.28

  

  Widowed

  

0.18

–0.05

  

 Mother divorced, 2006–2008

 

0.05

 

0.04

0.19

0.06

 Mother remarried, 2006–2008

 

0.57

 

0.55

  

 Mother remarried, 2008

    

0.45***

0.40***

 New birth, 2006–2008

–0.13

–0.14

–0.12

–0.13

–0.26*

–0.24*

 Mother remarried, 2008 × new birth, 2006–2008

    

0.43

0.47*

Child Covariates

 Female child

0.15

0.15

0.15

0.15

0.14

0.09

 Age of child, 2006 (ref. = 5–9)

  3–4

–0.30

–0.30

–0.30

–0.31

–0.28

–0.06

  10–14

0.59***

0.60***

0.61***

0.61***

0.60***

0.24***

 Child in worse relative health

0.42*

0.40*

0.43*

0.41*

0.40*

0.18

 Ethnic group (ref. = Yao)

  Chewa

0.21

0.25

0.22

0.26

0.22

–0.01

  Tumbuka

0.35*

0.39**

0.36*

0.40**

0.37**

0.00

  Other ethnic group

0.28

0.34*

0.30

0.35*

0.30

0.19

 Number of living siblings, aged 0–14

–0.10**

–0.10*

–0.10*

–0.09*

–0.09*

–0.04

Maternal Covariates

 Mother’s age (ref. = 30–34)

  18–29

0.33

0.36*

0.34*

0.37*

0.37*

0.14

  35–39

0.26

0.29*

0.25

0.28

0.30*

0.13

  40–44

0.22

0.24

0.23

0.25

0.24

0.11

  45+

–0.04

0.00

–0.02

0.01

0.02

0.25

 Mother is in polygynous marriage

0.17

0.16

0.16

0.16

0.17

0.09

 Mother is HIV positive

0.20

0.21

0.23

0.24

0.23

0.00

 Mother is in bad health

–0.11

–0.10

–0.10

–0.10

–0.10

0.15

Household covariates

 Family death, 2006–2008

0.33**

0.33**

0.34**

0.34**

0.32**

0.19*

 Quintile of household asset ownership (ref. = low)

  Lower-middle

–0.15

–0.14

–0.17

–0.16

–0.13

–0.19

  Middle

–0.15

–0.15

–0.18

–0.18

–0.17

–0.25

  Higher-middle

–0.04

–0.01

–0.06

–0.04

0.00

–0.12

  High

–0.01

0.00

–0.02

–0.01

–0.01

–0.07

Constant

–2.31***

–2.40***

–2.32***

–2.40***

–2.39***

–1.47***

ρ

0.23

0.29

0.21

0.26

0.32

0.71

Wald Test of Independent Equations, ρ = 0

0.58

0.78

0.60

0.83

0.84

1.16

Number of Children

3,051

3,051

3,051

3,051

3,051

3,051

Censored Observations

803

803

803

803

803

585

Uncensored Observations

2,248

2,248

2,248

2,248

2,248

2,466

Number of Mothers

1,282

1,282

1,282

1,282

1,282

1,282

Wald Chi-Square

143.60***

139.92***

154.55***

152.35***

134.04***

99.24***

Log-Likelihood

–2,129.20

–2,126.76

–2,083.86

–2,081.62

–2,123.91

–2,341.93

Note: The results of the selection equation are available in Table S1 (Online Resource 1).

aThe dependent variable in Model 6 is the “upper bound” measure of out-fostering.

p < .10; *p < .05; **p < .01; ***p < .001

The third model substitutes parents’ marital outcome in 2006 for mother’s marital status (similar to Table 3, Model 2). In contrast to the cross-sectional regression, children who were born outside of a union were not more likely to be out-fostered. However, children whose parents were divorced by 2006 had significantly higher probabilities of being out-fostered relative to children from intact families. The fourth model adds the two binary variables for whether the child’s mother divorced or remarried between the 2006 and 2008 survey rounds. After these marital changes were controlled, having divorced parents was no longer significantly associated with out-fostering. Children whose mother entered a new union during the intersurvey period had a higher probability of being out-fostered, echoing the results observed in Model 2.

Models 1–4 show no significant association between out-fostering and the birth of a new baby to a child’s mother during the intersurvey period. Model 5 tests the interaction between the birth of a new baby and a mother’s marital status. The two remarriage variables were collapsed into one variable that indicated whether a child’s mother was remarried at the time of the 2008 survey. This variable was then interacted with the birth of a new baby. Children whose mother was remarried but did not have a new birth had a higher probability of being out-fostered relative to children whose mother was not remarried, translating to predicted probabilities of .08 and .03, respectively. Children whose mother had not remarried but had a new birth were significantly less likely to be out-fostered relative to children who did not have a new sibling, yielding a predicted probability of .02. However, the interaction term between remarriage and new birth is positive, significant, and relatively large; this finding indicates that when remarried mothers had a new birth, their children were particularly likely to be out-fostered, with a predicted probability of .11. Our final model replicates Model 5 using our “upper-bound” measure of incident child fostering. The estimated coefficients for maternal remarriage, the birth of a new baby, and the interaction between remarriage and a new birth are all remarkably similar to those estimated in Model 5, but with higher levels of significance.

All regressions in Table 4 also found that children who experienced a family death in the intersurvey period were more likely to be out-fostered. Having a mother in bad health or who was HIV positive were not associated with out-fostering.

As in the cross-sectional regressions, the analysis of incident out-fostering found a positive and significant association with a child’s age: the older children are, the more likely they are to be out-fostered. There was also an association with ethnic group, such that Yao children were less likely to be out-fostered than all other ethnic groups. Children with more living siblings younger than 15 years old were also less likely to be out-fostered. In contrast to the cross-sectional analyses, several other variables became significant in the longitudinal analysis. Children with worse relative health were more likely to be out-fostered. Children with younger mothers also had higher probabilities of out-fostering. Many of these sociodemographic variables, however, were no longer significant in Model 6, which expanded the definition of out-fostered children to include nonmatched children for whom longitudinal data were otherwise available.

Discussion

Despite the frequency of divorce and remarriage across much of sub-Saharan Africa, little is known about what these events mean for the living arrangements of children. Child fostering is common throughout the region, but its relationship to family transitions remains poorly understood. In this article, we used longitudinal data from Malawi to examine the relationship between these and other family transitions and the prevalence and incidence of child fostering. Using the perspective of children’s living mothers rather than their current living situations, we can offer new insight into the reasons that children are out-fostered.

Our analysis found that both the prevalence and the incidence of child fostering were high in rural Malawi; between 7 % and 15 % of all children were out-fostered over a two-year period. By examining the predictors of both prevalent and incident child fostering, we can begin to carefully untangle the complex relationships between family transitions and fostering. For example, in the cross-sectional models, divorce seems to be the driver of child fostering. However, when we combine the findings from the cross-sectional and longitudinal analyses, the picture becomes more nuanced. Here, we see that it is really maternal remarriage and not divorce that is most closely associated with out-fostering. Remarriage has both a lagged and a more immediate influence on children’s living arrangements.

Our finding that mothers’ new and ongoing marriages to men who are not the fathers of their children raise the risk of out-fostering suggests that a substantial proportion of child fostering does not happen at the time of mother’s remarriage but may be precipitated by circumstances that emerge at some time after remarriage. This supposition is supported by our finding of a positive interaction between maternal remarriage and a new birth during the intersurvey period. The birth of a child may solidify a new union, making the position of children from previous unions more tenuous within the household. If mothers become aware that their children from previous unions do not receive sufficient resources, children may be sent to live with other relatives months or years after the initial family transition. It is also possible that conflicts between children and their stepfathers may take time to develop, creating a lag between the time of a mother’s new marriage and the out-fostering of her child. Unfortunately, data on intrahousehold conflicts and the division of resources were not collected in the survey, so we are unable to test these hypotheses. An examination of the family dynamics that lead to child out-fostering would be a fruitful area for future research. However, it is also important to note that the majority of children whose mother remarried continue to live with their mother.

The relationship between marital change and fostering may also vary by descent system or across ethnic groups. In this study, we found that both prevalent and incident out-fostering were least common among the matrilineal Yao. However, in alternate model specifications (results not shown), we did not find significant interactions between ethnicity and family transitions. It is unclear whether this was due to the absence of a true relationship or whether our study lacked the statistical power to detect such an interaction. Future demographic analyses of child fostering should be sensitive to how descent systems and ethnicity may condition the response to shocks and family change, potentially increasing the risk of vulnerability for some children more than others.

A key strength of our study is our ability to examine the incidence of child fostering over a two-year period and its association with family transitions over this period. Nonetheless, we are unable to identify the order of family transitions and child fostering between the survey rounds, so we are unable to know the extent to which children are out-fostered in anticipation of or as a consequence of these changes. Knowing the order of events would be informative, but it does not seem too problematic given the lagged association we also identify between child fostering and maternal remarriage. Although it is beyond the scope of this study, longitudinal data also enable the examination of fostering reversals, where children return to live with a parent after spending some period of time living with others. Among children for whom longitudinal data are available, 45 % of children who were fostered in 2006 coresided with a parent in 2008. Future research is needed in order to understand the determinants of temporary fostering relative to longer-term arrangements and how these patterns may be related to subsequent family transitions.

Our inclusion of an “upper bound” of out-fostering in the analysis also allows us to comment on the quality of mothers’ reporting of their children’s living arrangements. When we rely entirely on the sample of children who are listed in their mother’s household roster in both 2006 and 2008, we estimate that almost 7 % of children were out-fostered during the intersurvey period. However, we are also able to identify a second group of children who were listed by their mother in 2006 but who were not mentioned by their mother in the 2008 household roster. If we assume that all of these children were out-fostered and erroneously omitted, our estimate of the incidence of out-fostering doubles. We do not know whether those children were omitted because their mothers now considered them to be full members of their new households, because interviewers failed to follow survey instructions to probe for nonresident biological children, or because other issues were at play. Nonetheless, our findings demonstrate that estimates of child fostering that are based on cross-sectional survey data have limited means of identifying omitted children and evaluating the quality of reporting. This may be less of an issue for estimates derived from surveys such as the Demographic and Health Surveys, which are based on the presence of fostered children in receiving households rather than on mother’s reports of out-fostering. Our MLSFH “upper-bound” estimate of 14.6 % of prevalent child fostering among children aged 0–14 in 20087 is close to the DHS estimate that 15.0 % of children of the same age with living mothers who reside in rural Balaka, Mchinji, or Rumphi were fostered (ICF Macro and NSO 2010).

Although longitudinal data improve our ability to examine the process of child fostering, we are unable to examine the living arrangements of children whose mothers have left the sample. Women were followed only if they migrated within the study area, so we are unable to evaluate whether maternal migration raises the risk of out-fostering. Given that divorce raises a woman’s risk of migration out of the sample area (Anglewicz 2012), this is problematic but unavoidable.

Understanding the consequences of family transitions for the residential patterns of children is key to understanding the places of greatest strain due to an increase in AIDS-related orphanhood. Families use child fostering to mitigate shocks, to adjust to changing situations, and possibly to gain advantage. In the context of rising orphanhood, it may no longer be easy to find a household able to take in additional children or to find a household with the resources to appropriately provide for them (Grant and Yeatman 2012). Over the past decade, the incidence of divorce has increased in Malawi because it has become more acceptable to divorce a spouse suspected of bringing AIDS into a marriage (Reniers 2008). As divorce and remarriage become more common, the fostering out of children during these periods of family transition may become more difficult to arrange.

Everywhere, divorce and remarriage involve an adjustment in a child’s living situation. As one parent moves out or children come to split their time between two residences, the time children spend with their parents and the resources available to them will change. In Malawi, divorce may mean that a child is less likely to live with his or her father, but the mother’s remarriage carries the additional risk that the child will be out-fostered from both of the parents entirely. Although there may be some continuity of care if children are “left behind” when their mothers leave the extended-family household in order to enter a new union, the attention and support of their parents would still be reduced. Given the incidence of remarriage and the existing pressure on households in places where orphanhood is on the rise, such patterns may carry serious implications for children as high-quality fostering opportunities become more rare.

The HIV epidemic has focused research on the effects of parental death on child well-being (e.g., Case et al. 2004; Nyamukapa and Gregson 2005; Urassa et al. 1997). In this context, the implications of parental divorce and remarriage have gone unexamined. Future research on child fostering should examine not just the quantity of child fostering but also the quality, since there is good reason to believe that this has been eroded and will carry negative consequences for children’s well-being (Beegle et al. 2010; Grant and Yeatman 2012; Nyambedha et al. 2003).

Footnotes
1

The survey rounds collected from 1998 to 2006 are also known as the Malawi Diffusion and Ideational Change Project (MDICP).

 
2

The data do not enable us to identify whether women who had a nonmarital birth later married the father of that child.

 
3
The bivariate probit regression with selection is estimated using the heckprob command in Stata (StataCorp 2009). Predicted probabilities of out-fostering are obtained from the post-estimation command predict var, pmargin, which estimates the univariate predicted probability of success for the main dependent variable. This regression is based on the assumption of two latent variables, https://static-content.springer.com/image/art%3A10.1007%2Fs13524-013-0239-8/MediaObjects/13524_2013_239_Figa_HTML.gif and https://static-content.springer.com/image/art%3A10.1007%2Fs13524-013-0239-8/MediaObjects/13524_2013_239_Figb_HTML.gif, which represent the probability of being out-fostered and of selection into the longitudinal sample, respectively. From these latent variables, three probabilities can be estimated. First, is the probability that an observation is not censored and is in the longitudinal sample:
https://static-content.springer.com/image/art%3A10.1007%2Fs13524-013-0239-8/MediaObjects/13524_2013_239_Equa_HTML.gif
Second, is the probability of an uncensored success, or the probability that an observation is observed and fostered:
https://static-content.springer.com/image/art%3A10.1007%2Fs13524-013-0239-8/MediaObjects/13524_2013_239_Equb_HTML.gif
And, third, is the probability of an uncensored failure, or the probability that an observation is observed and not fostered:
https://static-content.springer.com/image/art%3A10.1007%2Fs13524-013-0239-8/MediaObjects/13524_2013_239_Equc_HTML.gif
The joint estimation leads to the log likelihood function:
https://static-content.springer.com/image/art%3A10.1007%2Fs13524-013-0239-8/MediaObjects/13524_2013_239_Equd_HTML.gif
 
4

The exclusion criterion, whether the respondent was successfully interviewed in 2004, is significantly correlated with attrition from the 2008 sample (R = .1818) but is uncorrelated with out-fostering in the analytic sample (R = –.0019).

 
5

In the selection equation, we found that children whose mothers were not married in 2006 were significantly more likely to attrit from the sample, consistent with prior research by Anglewicz (2012) on migration out of the study area. Children who were Tumbuka or in the other ethnic group category were significantly more likely than Yao children to be included in the longitudinal sample, as were children whose mothers were successfully interviewed in 2004.

 
6

Unless otherwise stated, all predicted probabilities presented in this article hold all variables constant at either the reference category or the mean value.

 
7

We estimate our “upper bound” of prevalent child fostering in 2008 as the percentage of children aged 0–14 who were not coresident with their mother, plus those children of the same age who were listed by their mother in the 2006 household roster but who were not matched in their mother’s 2008 roster.

 

Acknowledgements

The data used in this article were collected with support from the NIH (R01-HD37276, R01-HD050142, and R01-HD/MH-41713-0) and the Rockefeller Foundation (RF-99009 #199). Monica Grant was also supported by an NIH core grant to the Center for Demography and Ecology at the University of Wisconsin–Madison (R24 HD047873). We thank Philip Anglewicz, Ruben Castro, Sophia Chae, Hans-Peter Kohler, and Serena Tang for assistance linking the children’s data across survey rounds. We also thank Marcia Carlson, Shelley Clark, Rachel Goldberg, and the anonymous reviewers at Demography for comments on previous versions of this article, and Russell Dimond for statistical computing assistance.

Supplementary material

13524_2013_239_MOESM1_ESM.docx (87 kb)
ESM 1(DOCX 86 kb)

Copyright information

© Population Association of America 2013