containing multiple households.6 Within each compound, an enumerator individually interviewed
the head of every household and then separately interviewed all of his wives, if applicable.7
The tracking phase of the survey consisted of finding the 316 paired households that had ex-
changed a foster child and interviewing the head of each household along with all of his wives
using the same survey instrument as the initial phase. I restricted the tracking to those households
that had exchanged a foster child between 1998 and 2000 and where the child’s age at the time of
fostering was between five and fifteen inclusive.
Children under age five were excluded from the tracking for three reasons. First, these children
cannot be enrolled in school. Second, researchers studying child fostering in Africa have argued
that young children are fostered for different reasons than older children (Vandermeersch, 2002).
In particular, young children are not routinely performing domestic chores and are essentially
just consumers. Around age five, children are expected to become economic contributors to the
family, undertaking tasks in the household, fields, and marketplace. At this time, a household
would become concerned with human capital investment and possibly with offsetting demographic
imbalances in the number of its children of a given age and gender. Third, results from this
survey confirm that fostering of young children is much less common than older children, showing
a significant jump in fostering rates at age six. Approximately one percent of children under age
five were fostered between 1998 and 2000, compared to ten percent of children aged five to fifteen.
Children aged sixteen and older were also excluded from the tracking because, at that age,
6To increase the number of households in the sample that had fostered children, I adopted a two part sampling
frame that included a random sample and a choice-based sample both drawn from a village level census that included
information about the fostering status of every household (for more details, see Akresh, 2004b). The choice-based
sample consisted of compounds that had fostered a child between 1998 and 2000. All results in this paper use the
entire sample, but results are qualitatively similar when I restrict the observations to just the random sample. Using
the population fostering weights from the village level census to adjust the choice-based sample does not significantly
alter the results. A total of 383 compounds containing 606 households were selected with approximately sixty percent
of the compounds in the random sample.
7The particular household definition (described in Akresh, 2004b) that assigned every individual living in the
compound to a specific household was implemented to ensure that individuals in the compound who might have been
involved in making a fostering decision would be interviewed.