Rules versus Discretion in Eerly Childhood Development Programs.

The study experimentally varies beneficiary selection mechanisms while holding program content constant. Randomization occurs at the district level, with each community served by one ATPI.

Baseline census data identified all eligible households prior to implementation.

Eligible districts are randomly assigned to one of three groups:

Group 1: Business-as-Usual Recruitment. Local agents in charge of program sessions---Asistente Técnico de Primera Infancia (ATPI)---recruit participants using standard discretionary practices.
Group 2: Needs-Based Recruitment. ATPIs receive a ranked list of households based on a proxy measure of parental investments and invite families sequentially.
Group 3: Random-Order Recruitment. ATPIs receive a randomly ordered list of eligible households and invite sequentially.

All other program aspects, including content, frequency, and monitoring, remain identical across arms.

2025-05-01

2026-03-31

Program Participation:
Invitation to Círculos de Familias
Attendance intensity
Session completion

Differential Selection into Invitation, Attendance and program completion:

\item Demographic characteristics of caregiver: education, age, sex, employment status, time use.
\item Household characteristics: probability of being in poverty, index of dwelling quality, household size.
\item Baseline child development outcomes.
\item Baseline parenting practices and investments.
\item Engagement with other families.
\item Distance to Circulos de Familia sessions.

Child Development Outcomes: Standardized child development indexes (EDIN II)

Parenting Practices:
\item Early stimulation activities
\item Variety of play materials.
\item Parenting behavior indexes

The objective of the study is to study selection and how selection affects development outcomes.

For selection, the main variables are: Baseline household poverty (measured as an indicator taking the value of 1 if the household's income is below the national poverty line), dwelling characteristics (Inverse covariance weighted index based on five dimensions: access to electricity, sewer, drinking water, trash collection and internet), household size, distance to program's sessions (in km and travel time), frequency of interactions with other families in the district, network centrality and eigenvector centrality (based on self-reported interactions with other families in the district), child development outcomes and parental practices (as described below).


For impacts the main outcomes are:

Administrative records:

Likelihood of being invited to participate in the program, likelihood of attending to at least one session, number of program sessions attended, attendance to the final sessions.

Survey records:
Child development outcomes are measured using the EDIN tool across five domains: gross motor, fine motor, language, cognitive, and socio-affective skills. Raw scores range from 0 to 6 per dimension and are standardized using the control group mean and standard deviation. An aggregate index is constructed using an Anderson Index of the five standardized scores.

Parenting practices are measured across three dimensions: activities with the child, number of toys, and positive attitudes. Principal component analysis is used to construct dimension scores, which are standardized and aggregated using an inverse covariance index.

Class Quality and Conflict:
\item Engagement and participation
\item Friction and anger between participants
\item Interactions with other families in the district.
Recruitment costs:
\item Monetary costs.
\item Time.
\item Number of attempts to reach target children.

Thirty nine districts are randomly assigned to three different recruitment mechanisms. Recruitment continues until class capacity is reached, following arm-specific rules. Program implementation is otherwise identical across experimental arms.

Within districts assigned to groups 2 and 3, individuals were assigned to priority lists given to implementing agents (ATPIs):

Group 2: Within each district, households are grouped in pairs based on a neediness score. WIthin each pair, households are randomized to the short list given to ATPIs.

Group 3: Within each district, households are randomized to the short list given to ATPIs.

Not available

Computer

Districts level for main analysis.
Individual level, within a district for secondary analyses.

Yes

39 districts

6,000 households

13 districts (control)
13 districts (needs-based targeting)
13 districts (random order)