Experimental Design
We exploit three key data features. First, we combine information on the paternal and maternal surnames of household heads and their spouses, with the Spanish naming convention to build two types of extended family link within the same village — (i) inter-generational links, such as those from the head and spouse to their parents, and to their adult sons and daughters; (ii) intra-generational links, such as those from the head and spouse to their brothers and sisters. Combined with information from household rosters that identify extended family members that co-reside in the household, this provides an almost complete mapping of extended family structures across 506 villages, covering around 22,000 households and over 130,000 individuals.
Second, we exploit the multiple components of PROGRESA, each of which provides cash transfers conditional on a different household behavior. One component provides cash transfers conditional on the attendance of children in primary and secondary school. However, as preprogram initiation primary school enrolment rates are above 90%, transfers provided for this purpose represent a de facto unconditional cash transfer for households with primary school aged children. In contrast pre-program secondary enrolment rates are 65% so that for many households cash transfers will be obtained only if a change in behavior is induced. This is important because the value of transfers only corresponds to between one half to two thirds of the full time child wage in the survey villages [Schultz 2004], and so do not fully compensate for foregone earnings of secondary school aged children employed full time in the labor market. Hence if households are credit constrained, PROGRESA’s effect on secondary enrolment may be a function of the presence of primary school aged children, who receive de facto unconditional transfers. In particular, households can use these transfers to supplement those specifically conditioned on secondary school enrolment, thus fully offsetting the opportunity costs of enrolling children into full time secondary school. This channel affects both connected and isolated households. In addition, if families share resources, and in particular they share the unconditional transfers obtained from the primary school component of PROGRESA, the response of connected households will also depend on the demographic composition of eligible households within their family network. This drives a wedge between the program responses of connected and isolated households in terms of secondary enrolment.
Third, we exploit the randomized research design used to evaluate PROGRESA. Of the 506 sampled villages, 320 were randomly assigned to be a treatment group, namely villages where PROGRESA would be implemented, and 186 villages were controls. Data was collected on a panel of around 22,000 households every six months over the pre and post-program initiation periods. In each village the baseline survey provides a complete census of all eligible and all non-eligible households. Under standard assumptions this research design identifies the average treatment effect of PROGRESA from a comparison of eligibles in treatment and control villages. The core of our analysis identifies whether this treatment effect varies across connected and isolated households.
The 2015 paper combines both insights by studying whether extended family networks pool resources and thus help network members: (i) smooth consumption against idiosyncratic income risk; (ii) relax credit constraints, enabling additional investments to be undertaken. In this paper we first develop a stylized framework to make precise how consumption and investment decisions are interlinked for households subject to idiosyncratic income shocks and imperfect credit markets. We do so for two types of household: those we describe as being connected because they are embedded within an extended family network that pools resources among its members, and those that are isolated and have none of their family members in close geographic proximity and therefore have to self-insure. This simple framework demonstrates how relative to isolated households, family networks can both insure their members against consumption risk and relax credit constraints member households face in investment choices. We are also able to study the long run impacts of positive resource inflows by exploiting a final wave of data from November 2003, some five years after households first experience receipt of transfers. This allows us to provide insights on whether the increased capacity of family networks to insure against income risk and relax credit constraints leads to sustained increases in consumption and investment outcomes for network members relative to isolated self-insuring households.