The data include three years of retrospective information which I use to estimate a fixed effects
regression that measures the effect of fostering on school enrollment and deals with the potential
biases arising from using cross-sectional data. While cross-sectional results suggest that only 17.6
percent of foster children are enrolled compared to 31.1 percent of host family siblings (Table 1),
controlling for the child’s enrollment status prior to the fostering episode indicates that, when
compared to the host family siblings they live with, foster children are no worse off after moving
away from their biological parents. The fixed effects approach shows that not controlling for omitted
variables in measuring the welfare impacts of child fostering can yield seriously misleading results.
During the data collection, I located the sending and receiving household participating in each
fostering exchange. This research methodology makes these data particularly appropriate for un-
derstanding the impact of fostering, not only on the foster child and the host siblings, but also on
the biological siblings who stayed behind. The results show that after being fostered, foster chil-
dren are, on average, 3.6 percent more likely to be enrolled when compared to their non-fostered
biological siblings, and the impact is larger for young children. However, these results mask sub-
stantial heterogeneity depending on the reason for the fostering and where the sending and receiving
households live.2 Children who, according to their parents, were fostered for schooling reasons are
significantly more likely to experience a positive welfare outcome in terms of school enrollment
compared to children fostered for child labor reasons.
The fixed effects regressions in this analysis control for household level unobservables and provide
evidence that after a household selects which child to send, there is a strong positive impact of
the fostering on that child’s enrollment, relative to both the child’s host and biological siblings.3
2Evidence of this welfare outcome heterogeneity is also seen in rural Mali where children who were requested by
the receiving family had better nutritional outcomes than children sent due to crisis fostering (Castle, 1995).
3A related empirical literature that attempts to understand why households foster children finds that the demand
for child labor, risk-coping in response to exogenous income shocks, human capital investment in the child, parent
death, and high quality social networks are possible motivations for why a household decides to send out a child
(Isiugo-Abanihe, 1985; Page, 1989; Ainsworth, 1996; Zimmerman, 2003; Akresh, 2004; Cichello, 2004). In addition,