Study design and data
Data were drawn from a three-year cluster-randomized trial in rural
Rajasthan, which is described in detail elsewhere \cite{Nandi2016}. Briefly, we recruited 3,177 mothers with at least one child between
the ages of one and six from 160 village hamlets (hamlets are small independent communities near a larger village center), all of which met the following conditions: 1) no ready access to a government-run daycare; 2) a sufficient number of children age-eligible for daycare; 3) an identified structure to operate as a daycare; 4) a local women to manage the daycare; and 5) demand from the community via a formal application for a daycare. Following a census of households and collection of baseline
data, village hamlets were randomly assigned to treatment or control
conditions: the 80 “treated” hamlets received an offer to set up a daycare (called a balwadi) during the observation period, while control hamlets were not to receive any daycare at least until the study’s completion in
September 2017. Additional data were collected via midline and endline
surveys, roughly 11 months and 22 months after the baseline surveys. The primary outcomes of interest for the evaluation included maternal economic
outcomes, maternal mental health, children’s nutritional status, and
children’s educational attainment.
To construct a subsample where older children could potentially take on
some of the childcare/domestic responsibilities, we first restricted to
households with 2 or more children (n=X, Y%), at least one of whom was older than
6 years old (and therefore school-aged) at midline. Ages were derived
based on children’s birthdates and the recorded interview dates; missing birthdates (n=X [Y%] an issue that we revisit in the Discussion section) were randomly
imputed from a uniform distribution. We further restricted to children
within this subset who were not the youngest child in the household, as
the youngest child would not have a sibling under 6 during the
intervention period and would likely be exempt from, or at least less
impacted by, caregiving responsibilities. Baseline data were available
for 9922 of the 11190 children in our dataset; 65 of these children were
lost to follow-up by the time of the midline survey. An additional 1,491
children were excluded in accordance with our aforementioned exclusion
criteria.
Our final analytical subsample for descriptive purposes included xxxx children in xxxx households (Figure 1 ): 51% of these children
resided in households randomized to treated hamlets, and 49% resided
in control households. Given our focus on identifying effects among
older siblings, analyses were restricted to children over 6 years old at
midline. This general sample was further restricted depending upon the
analysis: for example, while our analysis of the impact of exposure on
school enrollment included all children over 6 for whom data were
available, our assessment of the impact of exposure on non-enrollment
due to domestic responsibilities was restricted to children who were
listed as unenrolled. We provide additional details for each analysis in
the Results section.