capital investment, Cichello is not able to address the endogeneity of fostering.
Related research attempts to measure the school enrollment impact when a child’s parent dies.
As parent death is one of several reasons why children are fostered, it is informative to understand
the empirical estimation strategies employed in that literature. Several papers use cross-sectional
data to estimate this impact of orphanage on enrollment (Ainsworth and Filmer, 2002; Case,
Paxson, and Ableidinger, 2004; Gertler, Levine, and Ames, 2004), but the results are subject to
potential biases due to omitted variables being correlated with both orphanage and enrollment.
There are two papers that address the endogeneity problem by using the time dimension in a panel
dataset to estimate a child fixed effects regression (Evans and Miguel, 2004; Yamano and Jayne,
2004). With this estimation strategy they are able to control for time-invariant factors, such as
wealth and network quality, that might be correlated with both orphanage and school enrollment.
These papers studying orphans have the advantage that parent death might be unexpected
and measuring the schooling impact due to this potentially exogenous event seems straightforward.
However, these papers focus on only one of the reasons why a child lives away from his biological
parents, and their data do not allow for comparisons with the biological siblings left behind. This
paper is able to address the broader question of the impact on children of fostering for potentially
endogenous and exogenous reasons. This is possible because the fieldwork design collected data not
just on a foster child and his host siblings, but also his left behind biological siblings. The biological
siblings are a good comparison group if the fostering endogeneity operates purely at the household
level, and thus is differenced out when comparing a foster child with his biological siblings.
3.2 Identification Strategy
In this paper, I employ two main estimation strategies, household and child fixed effects regressions,
to address the endogeneity problems regarding the fostering decision discussed in the previous
10