Fathers' whereabouts and children's welfare in Malawi

Abstract

Children's welfare may be affected by the absence of a parent – be it due to migration, divorce or death. These reasons for absence have largely been addressed separately in the literature; we present a unified framework. Using Demographic and Health Survey data from Malawi, we compare welfare indicators for four categories of children: those who live with both parents, and those who live with their mother but whose father is absent due to migration, divorce, or death. We find a clear pattern of welfare differences: children whose father is either present or a migrant are better off, and children whose father is deceased or whose parents are divorced are worse off. Our findings indicate that concern about the welfare of migrants' children might be exaggerated. By contrast, vulnerable children of divorcees are at risk of being overlooked in a policy environment that focuses on orphans.

Keywords:

• children
• divorce
• orphans
• migration
• Malawi

JEL codes:

• J12
• J13
• F22

1. Introduction

Two largely separate bodies of literature address the impact of parental absence on children's welfare in developing countries. The first examines how orphanhood resulting from adult HIV/AIDS mortality affects welfare-related outcomes in children. The second investigates the consequences of parental migration for children left behind. Both approaches are primarily topic driven, conducted within the frames of research on HIV/AIDS and migration, respectively. A third group of children live without a parent as a result of divorce. This is a widespread but under-researched phenomenon in developing countries, not reflected in any body of academic literature paralleling those on orphanhood and migration.

In this article we explore the consequences of paternal absence on child welfare in Malawi, distinguishing between orphanhood, migration and divorce as the causes of absence. For methodological and empirical reasons, we restrict ourselves to studying paternal absence among children who live with their mother. We find that children whose father is either present or a migrant are consistently better off than children whose father is deceased or whose parents are divorced. Although the cross-sectional design does not allow for firm conclusions about causality, the analysis makes a significant contribution to the study of living arrangements and vulnerability.

We use data from the 2010 Demographic and Health Survey (DHS) for Malawi. This survey is not designed for analysis of parental migration, but we are able to combine existing variables to identify children whose father is most probably a migrant. This represents a methodological contribution to the use of DHS data.

2. Background

In the following, we summarise relevant research on how orphanhood, parental migration and divorce may affect child welfare, with an emphasis on sub-Saharan Africa. Towards the end of each section, we discuss what is known about the situation in Malawi in particular.

2.1 Orphanhood and child welfare

The HIV/AIDS epidemic in sub-Saharan Africa has raised mortality among young adults and bereaved many children of their parents. The plight of the so-called ‘AIDS orphans’ has received considerable attention from the media, policy-makers, and researchers. In a comprehensive review article, Beegle & De Weerdt (2008:S93) refer to ‘an extensive descriptive and speculative literature on orphans, probably largely motivated by strong previous assumptions that the “orphan” effect would be negative and large’. The empirical conclusions, however, are mixed.

Monasch & Boerma's (2004) analysis of national household surveys from 40 African countries found that orphans are less likely than non-orphans to attend school, but little impact was found on nutritional status or child work. Akwara et al. (2010) found by analysing household surveys from 36 countries with high AIDS prevalence that, for ‘wasting’, early sexual debut and, to a lesser extent, school attendance, there were few significant differences between orphans and non-orphans.

Even when orphan disadvantages can be observed, Beegle & De Weerdt (2008) point out, this may be partly due to selection effects: if orphans come from vulnerable families with higher risk of parental death, they would possibly have been experiencing many of the same disadvantages if both parents were still alive. When donors and policy-makers still focus on orphanhood, the effects could be perverse. Meintjes & Giese (2006) argue that this focus creates economic incentives for orphans who are well integrated in the extended family to nevertheless self-identify as vulnerable motherless or fatherless children.

Malawi has, perhaps more than any other country, been associated with AIDS-related orphanhood. In 1997 a USAID report estimated that 1.2 million or 27% of Malawian children would be orphaned by the year 2000 (Hunter & Williamson, 1997). Even though the real figures did not reach the predicted levels, Malawi saw a clear increase in the number of orphans after the onset of the epidemic. The Demographic and Health Survey 2000 showed that, among children aged 0 to 15, 11% were orphans and 2% were double orphans – that is, they had lost both parents. These proportions peaked in the DHS survey four years later, at 13% and 3%, respectively. By 2010, however, the rates of orphanhood had fallen to their 2000 levels, coinciding with a slight decline in HIV prevalence.

As in sub-Saharan Africa more generally, research on orphan disadvantage in Malawi has produced mixed results. Some qualitative studies report lack of food, property grabbing by relatives, school dropout and lack of social support (Muula et al., 2003; Funkquist et al., 2007). Other, quantitative, approaches find that Malawian orphans are disadvantaged in terms of school attendance, immunisation, weight, fever, and bed-net use, but not when it comes to coughing, diarrhoea, medical treatment or child labour (Ueyama, 2007; Pullum, 2008). For schooling, Pullum (2008) found contradictory effects: a lower likelihood of attendance, but faster progress than non-orphans. Living with parents suffering from AIDS can, in fact, be a greater burden on children than orphanhood. Kidman et al. (2010) observed higher morbidity among Malawian children living with AIDS-sick parents, but found no clear pattern of disadvantage for maternal, paternal or double orphans.

Extended families are taking the paramount responsibility for Malawi's orphans (Chirwa, 2002; Beard, 2005; Floyd et al., 2007; Kidman & Heymann, 2008). Although some warn that this family-based system may become over-stretched, Peters et al. (2008) found that there is no pervasive family ‘breakdown’, that very few orphans are abandoned, and that, when necessary, households members regroup within the extended family.

2.2 Migration and child welfare

A growing body of literature addresses the impact of parental migration on children left behind. This comes partly as a counter-reaction to the optimistic view on the development benefits of migration that has swept the international development community over the past decade. UNICEF has played a central role in raising the issue of children left behind. In a 2007 policy brief, the organisation summarises its country studies on the topic, concluding that ‘children and adolescents left behind may be at greater risk to drug abuse, teenage pregnancy, psychosocial problems and violent behaviour’ (UNICEF, 2007:1). Alongside the somewhat politicised standoff in policy circles and civil society, a number of academic studies have attempted to determine the effect of parental migration on different aspects of child welfare, including child health and schooling (Kandel & Kao, 2001; Frank & Hummer, 2002; Jampaklay, 2006; Lu & Treiman, 2007; Amuedo-Dorantes & Pozo, 2010; Amuedo-Dorantes et al., 2010; Antón, 2010; Mazzucato & Schans, 2011).

There are two separate sets of effects of parental migration on child welfare, caused by the absence of a parent and the inflow of remittances, respectively. The absence of one or both parents can – as in the case of orphanhood or divorce – have negative direct impacts: children could be given less affection and attention in daily life, be required to perform household or non-household work that was previously carried out by the absent adult, or be given inferior treatment to other children in the extended household. The ‘absence’ of a migrant parent is, however, a matter of degree: intervals between return visits range from a few months to several years, and many migrant parents remain closely involved with childcare from a distance (Carling et al., 2012). The remittances that typically accompany migration can benefit child welfare by strengthening household finances, reducing reliance on child labour, and help cover the costs of education, healthcare and other welfare-related expenses.

Migration within, and from, Malawi has been extensive since colonial times. Men in particular have migrated internationally to work in mines and agriculture in South Africa, Zimbabwe, Zambia and the Democratic Republic of Congo, often leaving families behind. Opportunities for such work have diminished since the 1980s, but South Africa remains an important destination. Within Malawi, migration has primarily been directed to the capital, Lilongwe, and the agricultural areas of the Central region. More than one-third of households in Malawi receive financial help from other households in the form of remittances, most often from family members working elsewhere within Malawi (Pugliese & Esipova, 2011).

2.3 Union dissolution and child welfare

Studies of the consequences of divorce on child welfare in sub-Saharan Africa are remarkably scarce (exceptions include Bojuwoye & Akpan, 2009; Oya & Sender, 2009; Clark & Hamplová, 2013; Gnoumou-Thiombiano et al., 2013). We can assume, however, that the impact is modified by intermediate factors such as the nature of the dissolution process, the economic consequences for the child's household, the role of the extended family and relations between parents after separation.

As with orphanhood and parental migration, selection effects can influence the observed differences between children with divorced and cohabiting parents. Factors precipitating divorce in sub-Saharan Africa include medical conditions such as obstetric fistula and HIV infection, which have an independent negative effect on household vulnerability that could persist after the divorce (Porter et al., 2004; Ahmed & Holtz, 2007; Bwirire et al., 2008; Yeakey et al., 2009).

There can, however, also be positive selection mechanisms. In rural Mozambique, Oya & Sender (2009) found that women are more likely to be divorced or separated if they are highly educated and have a strong position on the labour market. Perhaps as a result, divorced or separated women ‘clearly achieve better results in educating their children than other women’ (Oya & Sender 2009:15).

In Malawi, divorce is relatively widespread. Reniers (2003) concludes that marriage in Malawi is a fragile institution, and that probabilities of divorce are among the highest on the continent. Every 10th head of household is divorced or separated (Malawi National Statistical Office, 2008). In a preliminary study of data from the Malawi Longitudinal Study of Families and Health, Chae (2013) found that parental divorce is, on the whole, associated with poorer schooling outcomes, especially for older children.

2.4 An integrated perspective on parental absence

The preceding discussion has differentiated between three reasons for parental absence: death, migration, and divorce. The socio-demographic determinants of each are different, but outcomes may operate through similar processes. In each of the three situations, it is pertinent to consider three separate sets of effects. First, child welfare might be affected by the factors that precipitated death, divorce, and migration, respectively. The transition to orphanhood, for instance, could mark the end of a period of parental illness that was psychologically and physically taxing on a child. Second, the death, divorce, or migration of a parent can affect child welfare via the household economy. Death and divorce can be assumed to have a negative economic impact; migration is often engaged in as an income-generating (or risk-reducing) strategy and can have a positive or stabilising effect on the household economy. The third set of effects is comprised of the non-economic impacts of parental absence. These relate to physical care and affection, non-market household labour needs, and perhaps social stigma.

3. Data and methods

We use data from the 2010 DHS for Malawi. This is a nationally representative household survey, with a sample covering all children residing in households. The data are most detailed regarding reproductive health, but also cover several aspects of household-level welfare and provides information on health and education for each child in the household. Since 1984, more than 300 DHS surveys have been carried out in over 90 countries. Developing new ways of using these data beyond their primary foci is therefore particularly valuable.

The sample used in the DHS for Malawi 2010 covers a total of 58,241 children below 15 years. Among these, 803 (1.4%) were not normally living in the respondent household and an additional 282 (0.5%) had missing information about parents. These children are removed from our sample. All of the descriptive statistics incorporate weights that account for regional variation in sampling ratios.

Our research design is as follows. First we use available data to categorise children by the status of their father. This categorisation constitutes our independent variable. Second, we develop dependent variables in the form of welfare indicators. Third, we identify appropriate control variables that can be assumed to have independent effects of children's welfare. The dependent variables are all binary, and we consequently use logistic regression analysis.

3.1 Classification of children

Information on whether or not parents are alive and reside with the child can be used to construct a cross-table with nine possible combinations (Table 1). Only 60% of children live with both parents. An additional 22% live only with their mother and 2% live only with their father. The four categories in the lower right part of the table (Categories 5, 6, 8 and 9) represent the 16% of children who are not living with any of their parents.

Table 1: Distribution of children by parental co-residence and orphanhood (n = 57 156), Malawi 2010

	Mother's status
Father's status	Mother in household	Mother alive, not in household	Mother deceased	Total
Father in household	Category 1, 59.6%	Category 2, 1.5%	Category 3, 0.5%	61.6%
Father alive, not in household	Category 4, 17.3%	Category 5, 10.8%	Category 6, 1.8%	29.9%
Father deceased	Category 7, 4.6%	Category 8, 1.9%	Category 9, 2.0%	8.5%
Total	81.6%	14.2%	4.2%	100.0%

Note: Percentages are weighted to account for regional variation in sampling ratios.

To study the impact of fathers' co-residence status we limit our analysis to the 46,626 children (82%) who are residing with their mother (Categories 1, 4 and 7 in Table 1). This is a large subset that provides sufficient empirical variation to assess the separate impacts of orphanhood, divorce and migration. Moreover, the survey design is such that there is more information on children whose mothers were interviewed.

Category 4 – children living with their mother and whose father is alive, but not living in the household – is a complex category. It includes children with divorced parents, children with migrant fathers, children whose father has never been around, children where the mother has remarried and children in polygamous marriages where the father lives with another wife. From this category we have picked out two groups: children whose parents have divorced (and whose mother has not remarried), and children whose father is most probably a migrant. The latter group have mothers who are in a conjugal relationship, who report that that the husband/partner is ‘staying elsewhere’, and who are not one of several wives in a polygamous marriage. We cannot rule out that some of these mothers have divorced and remarried, and that their current migrant spouse is not the biological father of the index child. This implies that some of the migrant fathers may be foster fathers. It is also possible that the migrant fathers include persons in the institutionalised population.

The regional variation in paternal migration, according to our estimates, matches what is known about migration patterns in Malawi: Children of migrant fathers constitute 3% of all children in the central region (including the capital Lilongwe), versus 6% and 5% of children in the northern and southern regions.

The categorisation of children that we use as our independent variable is presented in Table 2. The table also includes selected descriptive statistics. Children of divorced parents live in poorer households than the three other groups. Children of migrants stand out as having the most highly educated mothers – perhaps because the educational attainment of spouses is positively correlated and migrants can be expected to have higher levels of human capital.

Table 2: Classification of children by paternal co-residence status (n=42 987), Malawi 2010, descriptive statistics

	Size of group		Average values
	n	%	Age of child (years)	Household wealth (1 to 5)^a	Mother's educational attainment (years)	Number of children (age <15) in household	Proportion rural (%)
Father in household	34 178	79.3	6.0	3.0	4.3	3.6	85
Father migrant	2 281	5.4	4.9	3.1	5.3	3.5	88
Parents divorced	3 861	9.1	6.0	2.3	4.2	3.5	90
Father deceased	2 667	6.2	8.8	2.5	3.8	3.4	88
Total^b	42 987	100.0

Notes: ^aHousehold wealth is measured on a scale where 1 = poorest quintile and 5 = wealthiest quintile. Wealth is estimated by means of a household assets index. Percentages are weighted to account for regional variation in sampling ratios. ^b Excluding 14 169 children in the total sample that do not belong to any of the four categories listed.

3.2 Welfare measures

There are many possible approaches to measuring child welfare with the DHS data. Since we are interested in the effect of father's current co-residence status, we use variables that reflect current welfare, rather than accumulated disadvantage. To estimate welfare outcomes in an age-appropriate way, we conduct separate analyses for younger children (age 0 to 4) and for older children (age 5 to 14). Descriptive statistics for the selected indicators (by paternal co-residence status and for the full sample) are presented in Table 3.

Table 3: Exposure to welfare disadvantages, by paternal co-residence status, Malawi 2010, descriptive statistics (n = 42 987)

	Paternal co-residence status (%)
Welfare disadvantages	Father present	Father migrant	Parents divorced	Father deceased	Total (%)	n
Ages 0 to 4
Non-use of mosquito net	48.6	60.6	64.0	63.7	51.2	17 341
Too thin (–1SD)	31.4	23.4	46.3	44.0	32.0	2 134
Diarrhoea	17.9	17.2	18.7	20.2	18.0	16 486
Coughing	29.4	28.2	27.8	25.2	29.1	16 436
Fever	34.7	36.7	38.6	33.6	35.2	16 490
n	14 159	1 186	1 594	402	17 341
Ages 5 to 14
Non-use of mosquito net	70.0	68.3	75.4	81.3	71.4	25 646
Non-attendance at school^a	11.0	9.0	13.6	12.0	11.2	22 547
Fewer than two meals yesterday	6.8	7.9	13.4	12.5	7.9	25 646
No pair of shoes	35.4	31.0	54.5	58.5	39.0	25 281
Fewer than two sets of clothes	7.8	8.3	15.4	16.9	9.3	25 273
n	20 019	1 095	2 267	2 265	25 646

Notes: Percentages are weighted to account for regional variation in sampling ratios. SD, standard deviation. ^aAges 6 to 14.

One measure, the use of mosquito nets, is examined among both age groups. Heads of household were asked whether each of the household's children slept under a mosquito net the night before the interview. Far from all households in Malawi have bed nets, and particularly not for each bed. Because malaria is especially dangerous to children, use of mosquito nets could be a good indicator of caretakers' capacity and commitment to provide for each child.³

For younger children we also measure recent illness and weight for height. All mothers were asked whether each of her younger children had suffered from diarrhoea, fever or coughing during the two weeks preceding the survey. These symptoms are possible indicators of diseases that account for a large proportion of child mortality in sub-Saharan Africa. They may reflect the quality of care through the presence or absence of risk factors such as poor food hygiene and unhealthy cooking practices. Weight for height is the basic biometric measure that best captures acute or recent malnutrition, as opposed to cumulated chronic malnutrition. We identify children as having low weight for height if their value is more than one standard deviation below the median of the standard reference population. This is a broader definition of at-risk children than ‘wasting’ in the clinical sense, which is diagnosed by weight for height being more than two standard deviations below the median. Weight and height were measured by DHS interviewers for a subsample of the children.

For older children we examine school attendance, meals received yesterday and whether they have shoes and more than one set of clothes. Ensuring that children of appropriate ages attend school is seen as a core aspect of guardianship. The opportunity costs of foregoing child labour and the direct costs for uniforms and transportation, for instance, make school attendance an economic burden for the household, even if schooling itself is nominally free. Since children in Malawi often start school at age six or seven, the five year olds are excluded from the school attendance analysis. As direct measures of material disadvantage, we identify older children who ate fewer than two meals the day before the interview, who do not have any shoes, and who have fewer than two sets of clothes.

The welfare measures are all binary and consistently coded as indicators of disadvantage, for instance as non-attendance at school and non-use of a mosquito net.

We examined correlations between all of the different welfare measures to see whether the same children tend to be exposed to multiple disadvantages. In that case, the variables could be interpreted as overlapping measures of the same underlying, unobserved welfare dimension. However, we found only four correlation coefficients exceeding ±0.2 (for correlations between symptoms of illness, and between ownership of clothes and shoes). The lack of further co-variation shows that child welfare is multidimensional. The complementary measures not only gauge different aspects of welfare, but they differ in the ways in which they may be affected by paternal co-residence status. We therefore repeat the analysis with each welfare measure before assessing the overall relationship between fathers' whereabouts and children's welfare.

3.3 Control variables

To estimate the isolated effect of paternal co-residence status we control for factors that are likely to have an independent effect on welfare, and may be correlated with (but not caused by) the father's whereabouts. These factors are age and sex of the child, the mother's educational attainment, the total number of children in the household, urban versus rural residence, and region. The regional variable distinguishes between the northern, central and southern regions of Malawi.

Current household wealth can clearly be expected to affect child welfare. However, it is also entangled in a two-way relationship with paternal co-residence status. First, current household wealth is presumably correlated with past household wealth, which may have affected paternal migration, separation or death. Second, as discussed in the Introduction, important links between paternal co-residence status and child welfare operate through household wealth. For instance, any welfare disadvantage children experience as a result of paternal orphanhood could be caused partly by a decline in household wealth and partly by other consequences of the father's death. Household wealth should therefore be interpreted as both a background variable and an intermediate variable.

Our approach to the challenge of household wealth is to run the analyses with and without controlling for wealth. If paternal co-residence status affects child welfare partly through household wealth, as we assume, we expect that the effects are reduced, but do not disappear, when household wealth is introduced as a control variable.

The household wealth variable is developed by the DHS, using information on household assets (e.g. paraffin lamp, bicycle, motorcycle) and dwelling characteristics (e.g. water supply and floor material). By means of principal component analysis, each asset is assigned a weight and each household is given a total score. The household scores are subsequently converted into quintiles. Collecting and processing household wealth information in this way is generally recognised as an efficient and reliable approach in developing countries. However, it may not accurately reflect changes in household wealth. If a father migrates, for instance, his remittances may have an immediate impact on the mother's ability to cover expenses that are important for child welfare, but will take a longer time to affect the stock of assets in the household. Conversely, if the father suddenly dies, household income could drop abruptly while assets take a long time to depreciate.

4. Results

In the following we present results on younger (0 to 4 years) and older (5 to 14 years) children separately, and display the effects of paternal co-residence status with and without controlling for household wealth. Paternal co-residence status is a categorical variable of the kind that would normally be presented as a comparison between a single reference category and the other possible categories. In our case, however, we are interested in differences between all four categories – father in household, father migrant, parents divorced, and father deceased – and not in how one specific category compares with the other three. We therefore display the effect of paternal co-residence status in four separate sections in the tables, each one using a different reference category. Since all of the welfare measures are defined and coded as disadvantages, higher values in the odds ratios imply lower welfare for children.

Table 4 shows the estimated effect of paternal co-residence status on the five welfare measures for younger children; Table 5 shows the same effects controlled for wealth. We find that younger children who live with their father have a considerably lower risk of sleeping without a mosquito net. Children whose parents are divorced or whose father is deceased have 1.8 times higher odds of sleeping without a bed net. When we control for household wealth, the effects of paternal co-residence status on bed net use are weakened, but remain significant. Weight for height, diarrhoea and coughing exhibit the same pattern: children whose father is present or a migrant are better off than children whose parents are divorced or whose father is deceased. Prevalence of coughing, by contrast, appears to be relatively unaffected by paternal co-residence status.

Table 4: Odds ratio estimates for selected disadvantages, not controlling for household wealth, Malawi 2010: younger children (ages 0 to 4)

	Non-use of mosquito net	Too thin (–1 SD)	Fever	Diarrhoea	Coughing
Control variables
Age	1.12***	1.13***	0.90***	0.65***	0.91***
Sex (female)	0.97	0.89	0.93**	0.96	1.00
Residence (rural)	1.38***	1.59***	1.29***	1.00	1.18***
Northern region^a	1.11**	1.05	0.72***	0.79***	1.48***
Central region^a	1.08**	1.00	1.32***	1.31***	1.36***
Mother's education	0.93***	0.97**	1.00	0.99*	1.02***
Children in household	1.06***	1.14***	0.96***	0.94***	1.00
Paternal co-residence status
Father present^b
Father migrant	1.63***	0.70	1.12*	0.92	0.94
Parents divorced	1.89***	1.77***	1.21***	1.14*	1.10
Father deceased	1.83***	1.71	1.23*	1.40**	0.92
Father migrant^b
Father present	0.62***	1.44	0.90*	1.08	1.06
Parents divorced	1.16*	2.54***	1.09	1.23*	1.17*
Father deceased	1.13	2.45**	1.11	1.51***	0.98
Parents divorced^b
Father present	0.53***	0.57***	0.83***	0.88*	0.91
Father migrant	0.86*	0.39***	0.92	0.81*	0.86*
Father deceased	0.97	0.96	1.02	1.23	0.84
Father deceased^b
Father present	0.55***	0.57	0.81*	0.72**	1.09
Father migrant	0.89	0.41**	0.91	0.66***	1.03
Parents divorced	1.03	1.04	0.98	0.81	1.20
n	16 666	2 084	16 490	16 486	16 436

Notes: All regressions include a constant term. SD, standard deviation. ^aSouthern region is the reference category. ^bReference category. *Significant at the 0.1 level. **Significant at the 0.05 level. ***Significant at the 0.01 level.

Table 5: Odds ratio estimates for selected disadvantages, controlling for household wealth, Malawi 2010: younger children (ages 0 to 4)

	Non-use of mosquito net	Too thin (–1 SD)	Fever	Diarrhoea	Coughing
Control variables
Age	1.13***	1.13***	0.90***	0.65***	0.91***
Sex (female)	0.97	0.90	0.93**	0.96	1.00
Residence (rural)	1.07	1.38*	1.26***	0.98	1.18**
Northern region^a	1.14***	1.06	0.72***	0.79***	1.48***
Central region^a	1.06	1.00	1.32***	1.30***	1.36***
Mother's education	0.95***	0.98	1.00	0.99	1.02***
Children in household	1.09***	1.15***	0.96***	0.94***	1.00
Household wealth	0.82***	0.89***	0.99	0.99	1.00
Paternal co-residence status
Father present^b
Father migrant	1.64***	0.71	1.12*	0.92	0.94
Parents divorced	1.70***	1.73***	1.20***	1.13	1.1
Father deceased	1.75***	1.69	1.23*	1.39**	0.92
Father migrant^b
Father present	0.61***	1.41	0.90*	1.08	1.06
Parents divorced	1.04	2.45**	1.08	1.22*	1.17*
Father deceased	1.06	2.38**	1.10	1.51**	0.97
Parents divorced^b
Father present	0.59***	0.58***	0.83***	0.89	0.91
Father migrant	0.97	0.41**	0.93	0.82*	0.86*
Father deceased	1.03	0.97	1.02	1.24	0.84
Father deceased^b
Father present	0.57***	0.59	0.81*	0.72**	1.09
Father migrant	0.94	0.42**	0.91	0.66**	1.03
Parents divorced	0.98	1.03	0.98	0.81	1.20
n	16 666	2 084	16 490	16 486	16 436

Notes: All regressions include a constant term. SD, standard deviation. ^aSouthern region is the reference category. ^bReference category. *Significant at the 0.1 level. **Significant at the 0.05 level. ***Significant at the 0.01 level.

Table 6 shows the estimated effect of paternal co-residence status on the five welfare measures for older children; Table 7 shows the same effects controlled for wealth. The pattern of bed-net use confirms what we found in younger children: children whose father is present in the household are most likely to be protected by mosquito net, followed by migrant children, children with divorced parents and paternal orphans. The latter group have 1.7 times higher odds of sleeping without a bed net than those with both parents present. School attendance is significantly higher for children with their father in the household than for children with divorced parents or a deceased father. Controlling for wealth, however, the significance disappears. For meals, shoes and clothes, most of the differences between the groups are significant, even after controlling for wealth: children with a migrant father or both parents present more often eat at least two meals, have shoes and have two or more sets of clothes.

Table 6: Odds ratio estimates for various disadvantages, not controlling for household wealth, Malawi 2010: older children (ages 5 to 14)

	Non-use of mosquito net	Non-attendance at school^a	One or no meals yesterday	No pair of shoes	Fewer than two sets of clothes
Control variables
Age	1.10***	0.71***	0.98**	0.96***	0.96***
Sex (female)	0.97	0.77***	0.94	0.86***	0.84***
Residence (rural)	1.39***	1.40***	1.43***	2.87***	2.26***
Northern region^b	1.27***	0.74***	0.49***	0.85***	0.92
Central region^b	1.24***	1.16***	0.78***	0.89***	1.01
Mother's education	0.90***	0.80***	0.91***	0.84***	0.84***
Children in household	1.04***	1.06***	1.02	1.17***	1.05***
Paternal co-residence status
Father present^c
Father migrant	1.20***	1.16	1.26**	0.95	1.35***
Parents divorced	1.57***	1.41***	2.15***	2.41***	2.47***
Father deceased	1.89***	1.29***	1.81***	2.91***	2.63***
Father migrant^c
Father present	0.83***	0.87	0.79**	1.06	0.74***
Parents divorced	1.31***	1.21	1.71***	2.54***	1.83***
Father deceased	1.57***	1.12	1.44***	3.07***	1.95***
Parents divorced^c
Father present	0.64***	0.71***	0.47***	0.42***	0.41***
Father migrant	0.76***	0.82	0.59***	0.39***	0.55***
Father deceased	1.20**	0.92	0.84*	1.21***	1.07
Father deceased^c
Father present	0.53***	0.78***	0.55***	0.34***	0.38***
Father migrant	0.64***	0.90	0.70***	0.33***	0.51***
Parents divorced	0.83**	1.09	1.19*	0.83***	0.94
n	22 987	20 052	22 987	22 663	22 656

Notes: All regressions include a constant term (n is slightly lower here than in Table 3 because missing information on maternal education is removed).

^aAges 6 to 14. ^bSouthern region is the reference category. ^cReference category. *Significant at the 0.1 level. **Significant at the 0.05 level. ***Significant at the 0.01 level.

Table 7: Odds ratio estimates for various disadvantages, controlling for household wealth, Malawi 2010: older children (ages 5 to 14)

	Non-use of mosquito net	Non-attendance at school^a	One or no meals yesterday	No pair of shoes	Fewer than two sets of clothes
Control variables
Age	1.11***	0.72***	0.99	0.97***	0.97***
Sex (female)	0.97	0.77***	0.94	0.84***	0.83***
Residence (rural)	1.06	0.86	0.87	1.33***	1.07
Northern region^b	1.33***	0.79**	0.53***	0.96	1.05
Central region^b	1.22***	1.12**	0.75***	0.82***	0.94
Mother's education	0.93***	0.82***	0.95***	0.89***	0.89***
Children in household	1.05***	1.08***	1.03*	1.21***	1.07***
Household wealth	0.81***	0.72***	0.71***	0.56***	0.59***
Paternal co-residence status
Father present^c
Father migrant	1.18**	1.10	1.21	0.88*	1.27**
Parents divorced	1.35***	1.12	1.69***	1.67***	1.74***
Father deceased	1.71***	1.09	1.53***	2.36***	2.07***
Father migrant^c
Father present	0.85**	0.91	0.83	1.14*	0.79**
Parents divorced	1.14	1.02	1.40***	1.90***	1.37***
Father deceased	1.45***	0.99	1.27*	2.69***	1.64***
Parents divorced^c
Father present	0.74***	0.89	0.59***	0.60***	0.58***
Father migrant	0.88	0.98	0.71***	0.53***	0.73***
Father deceased	1.27***	0.97	0.90	1.42***	1.19**
Father deceased^c
Father present	0.59***	0.92	0.66***	0.42***	0.48***
Father migrant	0.69***	1.01	0.79*	0.37***	0.61***
Parents divorced	0.79***	1.03	1.11	0.71***	0.84**
n	22 987	20 052	22 987	22 663	22 656

Notes: All regressions include a constant term (n is slightly lower here than in Table 3 because missing information on maternal education is removed).

^aAges 6 to 14. ^bSouthern region is the reference category. ^cReference category. *Significant at the 0.1 level. **Significant at the 0.05 level. ***Significant at the 0.01 level.

For many of our indicators, girls seem to be less vulnerable than boys (as shown in the second line of Tables 4 to 7). To investigate whether fathers' whereabouts have different effects on male and female children, we conducted separate analyses for girls and boys (data not shown). Some effects that were significant in the analyses including both girls and boys became less significant or non-significant in the stratified analyses. This is not surprising, as the sample was halved. In the joint analyses, we found 39 significant differences between the four groups (34 when controlling for wealth). But whereas the analyses on girls only showed 22 significant differences between the groups (21 when controlling for wealth), the boys' analyses still showed 37 significant differences (32 when controlling for wealth). This suggests that boys' welfare may be somewhat more dependent on the father's whereabouts than girls' welfare.

5. Discussion

The results presented in Tables 4 to 7 cover 10 different welfare measures, each of which was examined with and without controlling for household wealth and subjected to six comparisons on the paternal co-residence status variable. We initially recognised that the measures complement each other in describing different aspects of welfare, and possibly reflect different impacts of paternal co-residence status. In order to summarise the findings we present a graphical comparison of the four categories of paternal co-residence status for children of all ages (Figure 1). The two panels present results with and without controlling for household wealth. Each arrow represents one or more comparisons that are statistically significant at the 0.1 level, pointing in the direction of the category of children that is more likely to be exposed to the welfare disadvantage. The thickness of arrows reflects the number measures for which the difference is statistically significant, the highest possible number being 10 (i.e. the total number of measures for younger and older children). The maximum number of significant differences we observe is between children whose father is living in the household and children with divorced parents: nine of the 10 welfare measures displayed a significant difference, all in favour of children residing with their fathers.

Figure 1: Comparison of welfare disadvantages by paternal co-residence status, Malawi 2010, children aged 0 to 14

Notes: For control variables and details on individual comparisons, see Tables 4 to 7. For a summary of welfare disadvantages, see Table 3.

Despite the relatively weak correlations between the welfare measures as such, there are clear patterns in how they are affected by paternal co-residence status. The four categories of children appear to fall into two groups in terms of welfare disadvantage: the high-welfare group, relatively speaking, is composed of children who either live with their father or whose father is a migrant; and the low-welfare group consists of paternal orphans and children whose parents are divorced. Before controlling for household wealth, nine of the 10 welfare measures showed one or more significant differences across the divide between these two groups; after controlling for wealth, eight of the measures did so. All of the statistically significant differences across this divide point in the same direction.

Within each of these two groups, the differences are more ambiguous. Although children residing with their father in general seem to be the least disadvantaged, there are fewer significant differences between them and the children of migrants. For one variable (owning shoes), the children of migrants were the least disadvantaged. Similarly, comparisons between the two most disadvantaged groups – paternal orphans and children whose parents are divorced – do not show consistent results in one direction, even though the majority of significant differences are in disfavour of those with a deceased father.

The overall pattern of relative disadvantage displayed in Figure 1 and summarised in the preceding paragraphs holds true for younger and older children alike. However, the effects of paternal co-residence status are more pronounced in the older group. In order to complement the summation of disadvantages presented in Figure 1 and to consider the relative size of effects, we constructed an additive index of disadvantage. This variable counts the number of disadvantages experienced by each child, excluding weight for height, which was only measured for a small subsample. The index thus has the range zero to four for younger children and zero to five for older children. The older children are restricted to the 6 to 14 age range in order to include school attendance in the index. We use an ordinary least-squares regression to estimate the effect of paternal co-residence status on child welfare, as measured by the additive index of disadvantage. Control variables are the same as in Tables 4 to 7. The results are presented visually in Figure 2, using children living with both parents as the reference category. This figure allows us to also consider the uncertainty of statistical estimation, represented by the 95% confidence intervals of each estimate. The results presented in Figure 2 can be summarised in six points.

Figure 2: Magnitude of effects of paternal co-residence status on additive index of disadvantage, ordinary least-squares regression, Malawi 2010

First, the effect of paternal co-residence status is much stronger for older children than for younger children. The scales are not directly comparable – since the maximum score is four for younger children and five for older children – but this does not affect the conclusion. What is more important for the interpretation is that the measures are different: the welfare indicators we have used for older children are, for the most part, more directly linked to parental resources and priorities than the ones for younger children. In fact, the measure that is included for both age groups – non-use of mosquito net – is affected by paternal co-residence status in almost the exact same way among younger and older children (Tables 4 to 7). We can therefore not conclude that older children are more vulnerable to paternal absence, only that the relevant and available measures of older children's welfare are more affected by paternal co-residence status.

Second, among the younger children, the greatest gap in welfare is between those who have their father present and those who do not – be it because of migration, divorce or death. Children of migrant fathers appear to be less disadvantaged than the two latter groups. The confidence intervals overlap, however, and this difference is therefore not significant at the 95% level.

Third, whether the father is divorced from the mother or deceased appears to make no difference for the welfare of younger children; the point estimates are virtually identical. There is, however, a particularly large confidence interval for the effect of being a paternal orphan, since there are relatively few young children in this category. This specific finding is nevertheless a powerful corrective to the policy-driven focus on orphans.

Fourth, among the older children, there is a large divide between the two advantaged groups – children living with their father or whose father is a migrant—and the two disadvantaged groups – paternal orphans and children of divorcees.

Fifth, being the child of a migrant father has somewhat surprising effects: there is no significant disadvantage for older children, but there is for younger children. As shown in Tables 4 and 5, the disadvantage for younger children is primarily linked to the greater likelihood of sleeping without a bed net. A possible interpretation is that younger children are more likely to sleep in the same bed as the mother, and that bed net use in parents' bed is higher when both parents are physically present.

Sixth, the effects of fathers' whereabouts do not operate primarily through household wealth. For younger children in particular, controlling for household makes virtually no difference. Among the older children, the disadvantage of paternal death or divorce is reduced when we control for income, but remains large. What we see, however, is that controlling for wealth introduces a significant difference between the two disadvantaged groups. This means that at a given level of household wealth, paternal orphans are somewhat more disadvantaged than children of divorcees. However, children of divorcees tend to live in poorer households and therefore experience the same level of disadvantage as orphans when we do not control for wealth.

The results presented in this paper have limitations that must be taken into account. A case in point is that we investigate only paternal absence. There were both substantive and methodological reasons for doing so, but it means leaving out the children who live without their mother – about one-fifth of all children in Malawi. While maternal absence is less common than paternal absence, the consequences for children are possibly more severe.

The data at hand cannot fully account for how economic mechanisms mediate between paternal absence and child welfare. As noted earlier, the measure of household wealth indicates long-term socio-economic status but may not reflect recent fluctuations or shocks. By controlling for household wealth, we can therefore dismiss the hypothesis that children who live without their father are disadvantaged only because they come from lower socio-economic strata – a conclusion that is essential to our research objective. However, we do not know how much of the remaining disadvantage is caused by economic hardship that results from the father's absence but is not (yet) reflected in household assets.

The available data only allow us to make cross-sectional comparisons between groups. The effects we seek to study, however, unfold over time after a father's death, migration or divorce. We do not know the time that has elapsed since these events. Being able to consider the duration of the father's absence would have been an advantage. The cross-sectional design also limits what we can say about causal mechanisms: we cannot firmly conclude that children are disadvantaged because of their father's absence, and that the disadvantages would not have been there if the father was present.

6. Conclusions

What we can conclude, which is important, is that paternal orphans and children of divorcees are significantly disadvantaged compared with otherwise similar children who live together with their father. Moreover, we can conclude that among older children the net welfare effect of having a migrant father is minimal.

The policy relevance of these findings is primarily linked to the identification of vulnerable children. As Kidman et al. (2010) note, orphanhood is often the sole criterion used by international non-governmental organisations for eligibility into child-centred programmes. Even the Malawi DHS, in categorising ‘orphaned and vulnerable children’, only includes children who have a deceased or seriously ill parent, or live in a household in which an adult has been seriously ill or died in the past 12 months. What we show, however, is that the vulnerability of children of divorcees is similar to that of paternal orphans.

The prominence of orphans in child-centred policies may have several explanations. First, donor funding is often related to the HIV/AIDS epidemic, to which orphanhood is directly linked. There are, as noted above, also connections between the epidemic and union dissolution, but these are more subtle. Second, it is plausible that the image of the orphan elicits greater compassion than that of a child of divorcees.

Current demographic trends make it even more important to ensure that children of divorcees are not neglected. While the number of orphans in Malawi seems to have stagnated or even declined during the past decade, a growing proportion of children have parents who are no longer in a union. Their share of the below-15 population rose from 5% in Malawi DHS 2004 to 7% in the 2010 DHS.

Our study makes the most of a widely used survey format – the DHS – to differentiate paternal co-residence status beyond what was envisaged in the survey design. We find among children whose fathers are alive but not residing in the household that there are two very different categories, depending on whether the parents are divorced or the father is a migrant. While the DHS data have limitations – for instance, in measuring recent changes in household finances – the innovative use of this widespread survey format is in itself a contribution to the literature.

This study was inspired by observing that policy discourses and academic research reinforce specific ways of categorising social reality. There are disparate literatures on orphanhood and on the impacts of migration on children left behind, and a striking paucity of research on children and divorce in sub-Saharan Africa. Some recent studies take an integrated approach to widowhood and divorce as causes of single motherhood, but leave out the possible effects of migration (Clark & Hamplová, 2013; Gnoumou-Thiombiano et al., 2013). We argue that in an African context, where parental migration, child fostering, union dissolution and orphanhood are all prevalent, there is a case for systematic comparisons across the categories.

Acknowledgements

The authors wish to thank Øystein Kravdal and two anonymous reviewers for valuable comments to earlier versions of this paper.

Notes

³We have used this indicator for all children, whether or not they live in households with bed nets. Stratified analyses including only children from households with bed nets showed similar results to those including all children.