Experimental Design
Randomization was carried out at the level of the municipalitiy (distrito) and using pairwise matching. From a list of all municipalities and villages (centros poblados) in Peru that were eligible for SAF, we eliminated those villages that did not have a minimum number of children 0-24 months old, based on the 2007 Census and the SISFOH household roster used for targeting the conditional cash transfer program. This left a roster of 531 municipalities that were ranked based on their poverty level (based on INEI’s 2009 poverty map). Based on this ranking, municipalities were organized in groups of three and 60 of these trios were randomly selected. Within each group of three municipalities, each one municipality was randomly allocated into T1, T2 and C. Within each municipality, the two villages with the largest number of children 0-24 months of age were selected. During the field work and prior to the survey administration, all households in these villages were registered in a village-level roster. From those with children younger than 24 months of age, 15 households per village were selected at random to be in the baseline survey.
Hence, the final evaluation sample consists of 360 villages, 120 randomly assigned to C and 240 randomly assigned to T (T1+T2).
At baseline, the total sample size was 5,869 children in 5,620 households from these 360 villages in 180 municipalities in 12 provinces of Peru. All children were 1-24 months of age.
Of these children, a subsample was assessed at follow-up using the Bayley-III. Given practical constraints, it was unfeasible to administer the Bayley-III in all municipalities in the sample. Hence, we selected the subsample of municipalities in which to administer the Bayley-III as follows. First, we dropped the 40 villages initially assigned to T, where the Program decided not to intervene at a later stage; as well as 1 village initially assigned to C which ended up receiving the intervention. To preserve the experimental design, we also dropped the matching villages in the trios to which these 41 villages belonged. Second, and given the difficulties and inability of the research team to have the Bayley-III translated and adapted into any indigenous languages, we excluded all villages where more than 30% of the baseline surveys were administered in a language other than Spanish. As before, in order to preserve the experimental design, all villages in the matching trio were dropped. Lastly, we excluded 6 villages (1 in Amazonas, 2 in Cuzco and 3 in Loreto) that were very disperse and hard to reach, together with the matching villages in their corresponding trios.
Moreover, within these municipalities, we identified the subsample of children estimated to be younger than 42 months of age at follow-up, given that the test is designed to assess children up this age.
Hence, the Bayley-III subsample includes 1,492 children who will be younger than 42 months old at follow-up in the 158 villages satisfying the criteria above (109 in T and 49 in C).
In all treatment (T) villages of the Bayley-III subsample, we also filmed a sample of home visits carried out by the home visitors who had been assigned to the children in T. We scored these videos using the HOVRS A+V2.1 and a short checklist we developed to measure the content and other quality features of the home visits, to all home visitors who work with the children in the Bayley-III sample (this is, 109 villages in 71 municipalities). For each home visitor (of a child in the Bayley-III subsample), we collected these measures during a home visit for each child in the Bayley-III subsample and any other child the home visitor regularly serves, on average 3 per home visitor.