Orphanhood and schooling outcomes in Malawi.

As in many sub-Saharan countries, the issue of orphan care has risen to the top of the social protection agenda in Malawi, where the prevalence of orphaned children has dramatically increased because of early deaths of parents infected by the HIV/AIDS virus. According to the Malawi Poverty Reduction Strategy Paper (MPRSP) prepared by the Government of Malawi in 2002, HIV infection rates in the 15-49 age group was at around 15% nationally (GOM 2002). The paper reported that about 70,000 children become orphans every year, adding to the already large number of orphans, estimated at about 850,000.

Orphans are a vulnerable group in any socioeconomic. However, Subbarao and Coury 2004, p. 140) state that "erosion of human capital is probably the biggest risk orphans and vulnerable children face in much of Africa." This is a serious concern, as underinvestment in health and education not only leads to serious depravation and hardship for the child, but it also depresses their future lifetime incomes.

Using longitudinal data covering the years 2000-2004, this paper determines the effect of orphanhood on school attendance in rural Malawi. This is done by estimating probit equations that relate school attendance status of children in 2004 to children and household characteristics and previous grade-level achievement.

Relating Schooling Outcomes to Orphanhood

There are two main reasons to expect schooling outcomes of orphans to fall short of schooling outcomes of non-orphan children. First, death of a parent, especially the more significant income-earning parent, is likely to affect labor allocation within the household. Specifically, because education brings financial returns only in the distant future, increased poverty caused by the loss of current income may mean that future returns to schooling are discounted more heavily and, in turn, children are expected to work, both at home and outside the home, at an earlier age to meet current consumption needs.

Second, it is often hypothesized that orphans may be victims of discriminatory practices by the relatives to whom they are entrusted for care. Such relatives or other caregivers are not only less likely to be altruistically motivated to care for their orphan wards, they may also not "invest" in the child's education because of the expectation that future financial returns, unlike in the case of their own children, will not necessarily accrue to them.

Third, it is quite likely that the physical and psychological trauma associated with the death of a parent may affect performance at school, and this way, affect the decision to continue education. This kind of trauma may be especially severe for orphans who witnessed the physical and mental agony of their HIV/AIDS-infected parent/s. There is also the chance that such children face discriminatory practice at school, by teachers as well as by fellow students, and that this makes them more likely to drop out of school than the rest.

There is some empirical evidence to support the above hypotheses. For example, Case, Paxon, and Abledinger (2004) used Demography and Health Survey (DHS) data to examine school enrollment of children 14 years or younger in several sub-Saharan countries, including Malawi. Using DHS 2000 data in Malawi, they found that orphaned children were more likely not to be enrolled in school compared to non-orphans. However, the problem with their estimate is that it is based on a single cross-section, and therefore does not convey very accurate information on whether orphanhood itself affects schooling outcomes. This is because it could well be that many of the orphans had stopped going to school before the death of their parents. In fact, the likelihood of this happening would be greater for HIV/AIDS-infected parents who might have pulled their children out of school while they were still alive either because of reduced income or because of greater need to finance increased medical expenses.

For this reason, tracking education achievement over time is likely to provide a better understanding of orphans' schooling outcomes. It would be of interest, for example, to compare current school enrollment status of orphans with that of non-orphans who had similar levels of schooling achievement in the past. This is what we do here. Specifically, using panel data, we compare school enrollment of orphans versus non-orphans in 2004 controlling for their educational level in 2000.

Data Source

Our analysis of schooling outcomes is based on longitudinal information on school-age children from 534 rural households in Malawi. These households were surveyed in the Complementary Panel Survey (CPS) conducted by the International Food Policy Research Institute in collaboration with the Center for Social Research, University of Malawi. The first round of the survey was completed in 2000 and the fifth round in 2004. Selection of households was done so as to maximize representativeness at the national level. In fact, the CPS household sample is a sub-sample of the much larger sample of households drawn for the 1997-8 Malawi Integrated Household Survey (IHS), which was nationally representative (National Economic Council 2000). A complete description of the sampling process can be found in Sharma et al. (2002).

Results

School enrollment of children in the age group 5-15 years was considered. There were a total of 966 children in this category in 2000, out of which 15.9% had lost at least one parent. About 7.56% were paternal orphans, 3.73% were maternal orphans, and 4.66% were "double" orphans, meaning both parents were deceased. Because of the small sample, and the corresponding smaller number of observations on different types of orphans, all types of orphans are pooled together in the econometric analysis. Out of the 966 children recorded in 2000, 99.98% were again traced in 2004.

Figure 1 shows the age distribution of orphans and non-orphans. While the age distribution of nonorphans in the sample peaks at just over five years of age, the distribution for non-orphans peaks at almost fifteen years, indicating that orphanhood increases with age.

[FIGURE 1 OMITTED]

Consistent schooling information is available for only 729 children in the survey. Hence, analysis of schooling outcomes is based on this subgroup. A simple tabulation of school enrollment in 2004 shows that 86% of nonorphans attended school while about only 81% of orphans did so. However, this 5 % point shortfall does not necessarily indicate that orphans are less likely to go to school. For example, if the likelihood of dropping out of school increases with age, and if the likelihood of orphanhood also increases with age (as was shown in figure 1), such a result would still hold even if there was no relationship between orphanhood and school attendance. There could also be other confounding factors arising out of household wealth and location of households. If orphanhood is correlated with these variables, or if the effect of orphanhood on school enrollment is itself modified by these variables (for example, if orphans from poorer households are more likely to drop out of school compared to orphans from richer households), then not incorporating these attributes in the analysis would lead to an erroneous conclusion.

For this reason, a multivariate probit model that relates school attendance to a range of child- and household-specific variables is estimated (table 1) to extract an unbiased estimate of the effect of orphanhood on school attendance. In order to evaluate the robustness of results, five alternative specifications are used. These are discussed below.

Under Specification 1, school enrollment in 2004 is specified as a function of only child level characteristics, namely the child's age and sex, his/her education level (grade level) in 2000, relationship with the household head, and orphanhood status. The relationship variable is a binary variable that equals one if the child is living in households not headed by either parents or grandparents and zero otherwise. Results (column [1]) show that while orphanhood has a negative effect on school enrollment, this effect is not statistically significant. As for the other variables, boys are more likely to be attending school than girls; the likelihood of dropping from school increases with age; and those at higher grade levels in 2000 are more likely to continue attending school. The effect of the relationship variable is not statistically significant.

Specification 2 is similar, but introduces interaction terms that allow the interaction of age, sex, and previous educational status with orphanhood status (column [2]). Thus, in this model, effects of orphanhood are specified to be conditional on the age, sex, as well as previous educational status. Results are similar to the first specification, except that the interaction term between orphanhood status and education level is negative and statistically significant at the 10% level. This result implies that there is greater likelihood of orphans dropping out of school compared to non-orphans as education level increases.

In Specification 3, the probit equation estimated is augmented by household-level variables (column [3]). Specifically, three household variables are introduced:

* Per capita land cultivated,

* Magnitude of negative agricultural shock experienced by the household as a result of 2001 droughts, and

* Household size.

Per capita land, the most important asset in rural Malawi, is included to control for general wealth level of the household. In 2001, Malawi was affected by one of the most serious and widespread droughts in recent years. The household-specific shock variable included controls for the effects of this shock (1). Finally household size controls for scale effects within the household. It should also be noted that Specification 3 contains interaction terms between orphanhood status and both land size and the 2001 shock. The coefficients of these terms allow us to test whether the effect of orphanhood on school enrollment is conditional on the level of wealth and/or the magnitude of shock experienced. Further, in order to test whether living arrangements modify the effects of orphanhood, Specification 3 also contains a term that interacts the relationship variable with orphanhood status.

Results for Specification 3 are reported in column (3) in table 1. As in Specification 2, the interaction term between orphanhood status and education level turns out to be statistically significant. Not only is the size of the coefficient bigger, the level of statistical significance is also higher. However, none of the household-level variables has a statistically significant relationship with school enrollment, either on their own or when interacted with orphanhood. The coefficient of the interaction term between the relationship variable and orphanhood status is positive and significant at the 10% level, implying that among orphans, those that are staying with non-parent or non-grandparent relatives, have a higher likelihood of attending school. As will be discussed later, this result may well be due to the impoverishing effect of HIV/AIDS-related deaths, and also the fact that orphans are taken under the care of relatives only when wealthier relatives are available.

Given multiple observations within households, one could have taken advantage of within household variations between orphan and non-orphan children to obtain a more efficient estimate of the effect of orphanhood. However, because fixed-effects probit estimators are not well defined, fixed-effect estimation is not pursued. Instead, information on the possible correlation in outcomes between children living in the same household is used in computing standard errors. This is done by specifying the cluster option in STATA at the household level when estimating the probit equations. Hence Specifications 5-6 are counterparts of Specifications 3-4 that recognize within-household correlation. Columns (5) and (6) show that all cases of statistical significance reported in the earlier equations remain, indicating general robustness of results.

Summing up, the estimated probit equations indicate that while an independent orphan effect on school attendance is absent, the likelihood of dropping out of school is higher for orphans than for non-orphans as grade level increases. Also, contrary to the hypothesis suggested earlier in the paper, the results indicate that orphans residing with non-parent or non-grandparent caregivers do not have a higher probability of dropping out. In fact, quite the opposite is true. Orphans under the care of a single parent or grandparent/s are less likely to attend school compared to those living with other relatives. This is most probably due to the fact that income losses associated with the death of a parent are significantly high and that the resulting increase in household poverty has a large negative impact on school attendance. Such a scenario is likely in Malawi where the high prevalence of HIV/AIDS has meant that death rates among income earning young adults are especially high. It is also likely due to the fact that orphans are adopted by relatively wealthier relatives who are less constrained financially.

Summary and Conclusions

The number of orphans in Malawi is growing rapidly, due primarily to the death of young parents to HIV/AIDS. This clearly poses new challenges for Malawi's policymakers. Apart from the psychological trauma associated with the loss of parents at a young age, there is clearly the danger that orphan children may grow up in a deprived environment, unable to benefit from basic investments in health and education. Indeed, our analysis indicates slippage in school enrollment of orphans in Malawi as grade level rises. This is bound to have strong bearings on their social and personal development and limit their lifetime earning potential. Consequently, there exists the danger that orphaned children can get quickly trapped in poverty for the rest of their lives, and this may lead to the emergence of a new generation of underclass citizens in the not-to-distant future. Because education has a strong bearing on both formation of social capital as well as future earnings, policies that ensure that education of orphaned children does not fall behind the rest will have high payoffs. Free or subsidized education for orphan children is therefore worthwhile to be seriously considered.

It is however important that any policy aimed at upholding education of orphaned children be "incentive-compatible" with individuals newly charged to care for the orphans. That is, policies need to have built-in rules such that caregivers (specially surviving, but now poorer parents) have sufficient incentives to actually convey benefits to the orphans in their charge.

References

Benson, T. 2002. Malawi: An Atlas of Social Statistics. Washington DC: International Food Policy Research Institute.

Case, A., C. Paxson, and J. Abledinger. 2004. "Orphans in Africa: Parental Death, Poverty, and School Enrollment." Demography 41(3):483-508.

Glewwe, P., H. Jacoby, and E. King. 2001. "The Impact of Early Childhood Nutritional Status on Cognitive Development: Does the Timing of Malnutrition Matter?" World Bank Economic Review 15:81-114.

Government of Malawi (GOM). 2002. "Malawi Poverty Reduction Strategy Paper." Final Draft. Lilongwe.

National Economic Council (NEC). 2000. "Profile of Poverty in Malawi, 1998--Poverty Analysis of the Malawi Integrated Household Survey, 1997-98." Poverty Monitoring System, Government of Malawi, Lilongwe.

Sharma, M., M. Tsoka, E. Payongayong, and T. Benson. 2002. "Analysis of Poverty Dynamics in Malawi." Washington DC: International Food Policy Research Institute.

Subbarao, K., and D. Coury. 2004. Reaching Out to Africa's Orphans: A Framework for Public Action. Washington DC: The World Bank.

(1) In rural Malawi, it is a common practice to estimate the size of maize harvest by the number of months the harvested maize can support household consumption, given normal consumption patterns. In the survey, each household was asked to provide this estimate for the 2001 and 2002 maize harvest (drought years) and compare this to the harvest level (again in months of consumption support) had the same amounts of land been cultivated in a "normal" year. Agriculture shock, the variable used as a measure of the crop shock received by the household is then defined as the ratio normal year harvest/specified year harvest. An additional advantage of using this measure of shock is that it accounts for the severity of the shock as well, since the larger this ratio, the larger the negative shock.

Manohar P. Sharma is a research fellow, International Food Policy Research Institute, 2033 K Street NW, Washington, DC 20006, USA.

This article was presented in a principal paper session at the AAEA annual meeting (Long Beach, CA, July 2006). The articles in these sessions are not subjected to the journal's standard refereeing process. Table 1. Probit Estimates of School Enrollment and Orphan Status Explanatory Variable (1) (2) (3) Sex of child 0.063 ** 0.062 ** 0.053 **

(2.59) (2.40) (1.97) Age of child in years (2000) 0.038 0.035 0.040

(1.52) (1.43) (1.58) Square of age -0.004 ** -.004 ** -.004 **

(3.15) (3.06) (3.14) Grade level (2000) 0.019 ** 0.024 ** 0.028 **

(2.45) (2.80) (3.18) Relation' 0.037 0.050 -0.119

(0.62) (0.84) (0.90) Orphan indicator (a) -0.012 -0.033 -0.027

(0.33) (0.23) (0.15) Orphan indicator x sex of child (a) -0.022 -0.014

(0.29) (0.17) Orphan indicator x age of child 0.011 0.014

(0.78) (0.90) Orphan indicator x grade level of -0.037 * -.055 **

child (1.85) (2.44) Household size (2000) 0.002

(0.41) 2001 shock 0.001

(0.13) Per capital landholding 0.010

(0.19) Orphan indicator x landholding 0.045

(0.37) Orphan indicator x shock -0.014

(0.99) Orphan indicator x relation (a) 0.105 *

(1.83) Observations 726 726 682 Explanatory Variable (4) (5) Sex of child 0.062 ** 0.053 *

(2.32) (1.95) Age of child in years (2000) 0.035 0.040 **

(1.52) (1.75) Square of age -.004 ** -0.004

(3.30) (3.55) ** Grade level (2000) 0.024 ** 0.028 **

(2.09) (2.28) Relation' 0.050 -0.119

(0.71) (0.83) Orphan indicator (a) -0.033 -0.027

(0.23) (0.19) Orphan indicator x sex of child (a) -0.022 -0.014

(0.35) (0.21) Orphan indicator x age of child 0.011 0.014

(0.83) (1.13) Orphan indicator x grade level of -0.037 * -.055 **

child 1.86 (2.65) Household size (2000) 0.002

(0.33) 2001 shock 0.001

(0.13) Per capital landholding 0.010

(0.19) Orphan indicator x landholding 0.045

(0.44) Orphan indicator x shock -0.014

(1.02) Orphan indicator x relation (a) 0.105 **

(2.02) Observations 726 682 (a) These are dummy variables. Note: Dependent variable equals one if child enrolled in school in 2004. Coefficients are presented in terms of the marginal effects of the regressors and absolute value of Z statistics are in parentheses. Single (*) and double asterisk (**) denote variables significant at 5% and 10%, respectively. District dummies are included but not reported.