Abstract
This paper provides strong evidence that adult mortality has a negative impact on children educational outcomes, both over the short and the long run, in rural Madagascar. The underlying longitudinal data and the difference-in-differences strategy used overcome most of the previous cross-sectional study limitations, such as failure to control for child and household pre-death characteristics and unobserved heterogeneity. This paper also pays special attention to the heterogeneity, robustness, and long-run persistence of effects. Results show that orphans are on average 10 pp less likely to attend school than their nonorphaned counterparts, this effect being even more pronounced for girls and young children from poorer households. Results on adults further show that those orphaned during childhood eventually completed less education. These findings suggest that not only do households suffering unexpected shocks resort to schooling adjustments as an immediate risk-coping strategy, but also that adversity has long-lasting effects on human capital accumulation.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Explore related subjects
Discover the latest articles, news and stories from top researchers in related subjects.Notes
Many national and international programs were launched by international organizations such as the World Bank or UN agencies, in partnership with community-based associations and NGOs, to reduce school fees and expenses, to supply uniforms, improve access to credit, or promote part-time education for orphans (see, for instance, the UNESCO program to provide education to orphans and vulnerable children, the Orphan Support Africa program or the numerous National Orphan programs in several countries from East and Southern Africa).
For the sake of simplicity, any child experiencing an adult death in the household will be referred to as an “orphan” throughout this paper, even if he (resp. she) is not strictly the son (resp. daughter) of the deceased member. This is not so restrictive in the Malagasy context where most households are nuclear families. Most adult deaths registered in data are then affecting parents and, if not, a very close relative. For further details, see Section 3.
Several studies point out the importance of informal insurance in developing countries where formal mechanisms are failing. This includes risk-sharing systems within extended families or community networks, likely to provide financial, human, or moral assistance and thus working as safety nets (Townsend 1995). Likewise, child fostering is a common practice among orphans. Akresh (2007) show that foster children are more likely to be enrolled in school years after adult death than their nonfoster siblings. However, the same authors put forward the reciprocity involved in those mechanisms, which is hardly sustainable in case of large and permanent shocks such as death (Townsend 1994).
Using previous rounds of this survey, Rakoto-Tiana (2011) investigates children time allocation within the household and shows that there is a trade-off between schooling and domestic and productive child work, depending on the structure of the household. One interesting feature for our issue of interest is that the absence of one of the parents has a negative impact on school enrollment and seems to increase child work. Although this study does not precisely deal with adult death, it provides first evidence that child work is considered by Malagasy households as a relevant informal strategy to cope with risk.
The ROR was implemented as part of the Madagascar-DIAL-INSTAT-Orstom project, established at the Malagasy National Institute of Statistics (INSTAT) and founded through the European Union, the French “Institut de Recherche pour le Développement” (ex-Orstom), and the French Ministry of Cooperation. Surveys are conducted through various rural observatories which are referred to as regions in the paper. Depending on the donors’ interest for such-and-such rural issue, some observatories were created, whereas others were phased out over time. There were a total of 13 observatories in 1999, 17 in 2000 and 2001, and 15 in 2002 and 2003. For more information, see the ROR webpage: http://www.dial.ird.fr/enquetes-statistiques/observatoires-ruraux.
Other reasons were refusal, temporary absence, migration, marriage, divorce, or back in the home household. The rate of missing values is 18 % for this question. However, we are rather confident with the fact that death is well reported by the remaining members of the household, so that we do not underestimate the number of deaths in this setting.
The 2004 survey initially covered 14 regions. Nevertheless, four regions were phased out in 2005 and 2006. We assumed this “technical” attrition as random since it is related to a lack of funding more than to specific village or household characteristics.
Most of them are “half-orphans,” and some lost a grandparent or other adult in the household. However, 76 % of the deaths recorded in the survey affect formerly resident parents of the sampled children, mostly in their working age. The average age at death is indeed 56 years for men and 54 years for women. Following results are robust if we restrict the sample to children who lost a parent. They are available from the author upon request.
This is one limitation of our study since were are not able to evaluate the potential mitigating effects of informal insurance networks through fostering and migration.
Gross enrollment ratios can actually be greater than 100 % as a result of grade repetition and entry at younger or older ages than the typical age at a given level
Among reasons mentioned by respondents in orphans’ households for not sending them to school in 2005, shortage of labor on crops and high school fees stand in the first and second positions, which is in line with our theoretical framework.
\(t\in \{0,1\}\) where 0 refers to the 2004 baseline period and 1 to the subsequent 2005 period, in what follows.
Formally, if we denote \(S^{0}_{it}\) the potential schooling outcome of a child had he not experienced death, this identifying assumption can be written as \(\mathbb {E}[S^{0}_{i1}-S^{0}_{i0}/D=1] = \mathbb {E}[S^{0}_{i1}-S^{0}_{i0}/D=0] =\mathbb {E}[S_{i1}-S_{i0}/D=0]\). It states that there is no selection into treatment, which will be further discussed in details.
Again, these results rely on the same fundamental DID identifying assumption which states that, conditional on observables and/or unobservables, there is no selection into treatment, i.e., death is as good as random. In this regression setting, this assumption can be formally rewritten \(P(D_{iht}=1/\varepsilon _{it})=P(D_{iht}=1)\).
In our setting, \(D_{ih0}=0\) for every child, while \(D_{ih1}\) switches to 1 for a child in household h incurring an adult death between the two periods.
In order to control for the potential inequality of treatment between household heads’ own children and other child relatives with respect to schooling.
We implement the conditional logit estimator for fixed-effects panel data discrete choice model], suggested by Chamberlain (2008).
This figure corresponds to the average marginal change in probability calculated from logit specification (2). The marginal effect at the mean is about 20 pp. Unfortunately, deriving marginal effects from conditional (fixed-effects) logit estimations remains quite tricky. Nevertheless, as the estimated coefficients are quite close in all specifications, we are confident with the fact that marginal changes in probability are of similar magnitude.
This is partly due to the fact that child fixed-effects models are “overidentified,” since adult death shocks are measured at the household level. Thus, household fixed effects probably capture most of the unobservable heterogeneity correlated with death.
This scheme builds on propensity score matching methods proposed by Heckman et al. (1998). The DID/matching combination appears to be robust by comparison to random experiments.
The propensity score is computed from a probit regression where the dependent variable is a dummy equal to one if child i became orphan in 2005 and independent variables include the whole set of baseline child and household characteristics. Results are not reported but available from the author upon request.
However, only 0.9 % of households were not reinterviewed for such a reason.
Rakoto-Tiana (2011) specifically investigates the impact of child fostering on education outcomes in Madagascar and show mixed results depending on the structure of the host household. Young fostered children that are biologically related to the head of host household are more likely to be enrolled after fostering. This is, however, not true for nonrelated ones that might be involved in child (domestic or productive) work.
Formally, we test the missing at random (MAR) assumption (Little and Rubin 1987), \(\mathbb {E}[S_{i0}/X] = \mathbb {E}[S_{i0}/X,A_{i}=1]\), where \(A_{i}\) is a dummy for nonattrition. The MAR assumption is rejected in our data.
Note that we only present corrected LPM, to keep the sample size fixed and also because Heckman’s procedure is better suited to linear models.
Ai and Norton (2003) show that interaction effects in nonlinear models cannot be evaluated simply by looking at the sign, magnitude, or statistical significance of interaction term coefficients, estimated by standard softwares. It requires more sophisticated computation of “true” marginal effects.
Note that the scope of the analysis is here restricted to the impact of parental death, since other adult deaths which occurred in the household during childhood were not recorded.
Secondary education is not rare in the adult sample. Indeed, 15.2 % of adults have engaged in some post-primary schooling, which amounts to 21.6 % of adults who went to school.
Using this subsample of adults for which the exact timing of parental death and the age at end of schooling is recorded, we estimate that 12.6 % of adults who lost a parent had not finished schooling when he died, which amounts to 45.2 % of adults who lost a parent before the age of 18.
In order to account for year-cohort specific trends, we more precisely interact period dummies with date of birth.
We estimate the following model: \(S_{it}=\eta _{i}+\delta _{t}+ \alpha D \centerdot T_{it}+\varepsilon _{it}\), where D is a dummy equal to 1 if individual i lost a parent before age 18, and \(T_{it}\) indicates the number of years since parental death for individual i at period t. This specification is similar to Eq. 8, but postulates a constant effect \(\alpha \) of death on schooling outcomes each year after its occurrence.
References
Abadie A (2005) Semiparametric difference-in-differences estimators. Rev Econ Stud 72(1):1–19
Ai C, Norton E (2003) Interaction terms in logit and probit models. Econ Lett 80(1):123–129
Ainsworth M, Filmer D (2002) Poverty, AIDS, and children’s schooling: a targeting dilemma. World Bank Publications, Washington DC
Ainsworth M, Beegle K, Koda G (2005) The impact of adult mortality and parental deaths on primary schooling in North-Western Tanzania. J Dev Stud 41(3):412–439
Akbulut-Yuksel M, Turan B (2010) Left behind: intergenerational transmission of human capital in the midst of HIV/AIDS. J Popul Econ 26(4):1523–1547
Akresh R (2007) School enrollment impacts of non-traditional household structure. Unpublished manuscript, University of Illinois at Urbana–Champagne
Becker G, Tomes N (1979) An equilibrium theory of the distribution of income and intergenerational mobility. J Polit Econ 87(6):1153–1189
Becketti S, Gould W, Lillard L, Welch F (1988) The panel study of income dynamics after fourteen years: an evaluation. J Labor Econ 6(4):472–492
Beegle K (2005) Labor effects of adult mortality in Tanzanian households. Econ Dev Cult Change 53(3):655–683
Beegle K, De Weerdt J (2008) Adult mortality and consumption growth in the age of HIV/AIDS. Econ Dev Cult Change 56(2):299–326
Beegle K, Dercon S, De Weerdt J (2010) Orphanhood and human capital destruction: is there persistence into adulthood? Demogr 47(1):163–180
Bicego G, Rutstein S, Johnson K (2003) Dimensions of the emerging orphan crisis in sub-Saharan Africa. Soc Sci Med 56(6):1235–1247
Case A, Ardington C (2006) The impact of parental death on school outcomes: longitudinal evidence from South Africa. Demogr 43(3):401–420
Case A, Paxson C, Ableidinger J (2004) Orphans in Africa: parental death, poverty, and school enrollment. Demogr 41(3):483–508
Chamberlain G (2008) Heterogeneity, omitted variable bias, and duration dependence. In: Heckman JJ, Singer B (eds) Longitudinal analysis of labor market data. Cambridge University Press, Cambridge
Cogneau D, Dumont JC, Glick P, Razafindrakoto M, Razafindravonona J, Roubaud F (2003) Madagascar. In: Morrisson C (ed)Dépenses d’éducation, de santé et réduction de la pauvreté en Afrique de l’Est: Madagascar et Tanzanie. OECD Publishing, Paris, pp 120–160
Evans DK, Miguel E (2007) Orphans and schooling in Africa: a longitudinal analysis. Demogr 44(1):35–57
Foster G, Shakespeare R, Chinemana F, Jackson H, Gregson S, Marange C, Mashumba S (1995) Orphan prevalence and extended family care in a peri-urban community in Zimbabwe. AIDS care 7(1):3–18
Gertler P, Levine D, Ames M (2004) Schooling and parental death. Rev Econ Stat 86(1):211–225
Heckman JJ (1979) Sample selection bias as a specification error. Econometrica 47:153–161
Heckman JJ, Ichimura H, Todd P (1998) Matching as an econometric evaluation estimator. Rev Econ Stud 65(2):261–294
Hunter S, Williamson J (2000) Children on the brink: strategies to support children isolated by HIV/AIDS. USAID, Washington DC
Kamali A, Seeley JA, Nunn AJ, Kengeya-Kayondo JF, Ruberantwari A, Mulder DW (1996) The orphan problem: experience of a sub-Saharan Africa rural population in the AIDS epidemic. AIDS care 8(5):509–516
Little RJ, Rubin DB (1987) Statistical analysis with missing data. Wiley, New York
Lloyd CB, Blanc AK (1996) Children’s schooling in sub-Saharan Africa: the role of fathers, mothers, and others. Pop Dev Rev 22(2):265–298
McPherson MF (2001) HIV/AIDS impacts on capacity deepening and economic growth. Unpublished manuscript, Brooking Institutions, Washington DC
Naidu V, Harris G (2005) The impact of HIV/AIDS morbidity and mortality on households—a review of household studies. S Afr J Econ 73(1):533–544
Rakoto-Tiana N (2011) L’école primaire pour tous? Politiques éducatives, demande d’éducation et travails des enfants à Madagascar. PhD thesis, Université Paris 13 - Paris Nord
Ryder RW, Kamenga M, Nkusu M, Batter V, Heyward WL (1994) AIDS orphans in Kinshasa, Zaire: incidence and socioeconomic consequences. Aids 8(5):673
Soares RR (2006) The effect of longevity on schooling and fertility: evidence from the Brazilian demographic and health survey. J Popul Econ 19(1):71–97
Townsend RM (1994) Risk and insurance in village India. Econometrica 62(3):539–591
Townsend RM (1995) Consumption insurance: an evaluation of risk-bearing systems in low-income economies. J Econ Perspect 9(3):83–102
USAID, UNICEF, UN (2002) Children on the brink 2002: a joint report on orphan estimates and program strategies
WorldBank (2001) Education and HIV/AIDS: a window of hope
Yamano T, Jayne TS (2004) Measuring the impacts of working-age adult mortality on small-scale farm households in Kenya. World Dev 32(1):91–119
Yamano T, Jayne TS (2005) Working-age adult mortality and primary school attendance in rural Kenya. Econ Dev Cult Change 53(3):619–653
Author information
Authors and Affiliations
Corresponding author
Additional information
Responsible editor: Junsen Zhang
Rights and permissions
About this article
Cite this article
Senne, JN. Death and schooling decisions over the short and long run in rural Madagascar. J Popul Econ 27, 497–528 (2014). https://doi.org/10.1007/s00148-013-0486-4
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00148-013-0486-4