Impact of Early Childhood Comprehensive Care on Children and their Families in the Dominican Republic

Using a randomized control trial, we will estimate the impact of enrollment to daycare centers on both children and their families. We will focus on early childhood outcomes such as language development, fine and gross motor, socio-emotional development and cognitive skills, as well as parental skills, employment, income, women’s agency, IPV exposure and gender norms.

2026-08-23

2028-06-30

Child: Children’s socio-emotional and cognitive development. Household: Employment and income, Time use, Mental well-being (general stress, parental stress), Parenting practices, parent-child interaction, Spousal relation, Women’s empowerment, intra-household decision-making, Social ties, Social norms

Children’s development: To collect information on child development we will use INAIPI’s assessment tool (SIMEDID – Sistema de Medicion de Desarrollo Infantil Dominicano) which measures gross and fine motor skills, socio-emotional development, and language development. This tool will be complemented with instruments that measure children’s cognitive development. These instruments will be used during the interview with the mother and will collect data on all children 0-5 years old.
Intra-household: We will interview the mother to collect data on her time use (or time use of main carer if she is not the mother) on care activities as well as economic activities, income, education, social connections, women’s agency (e.g., involvement in household
decisions), well-being (including mental health and stress) and spousal conflict.
Social norms: We will elicit mothers’ opinions about what is acceptable or appropriate behavior for men and women in domains such as parenting activities (playing with children, reading, etc.), household chores, economic activities outside the house, and autonomous
decision-making (e.g., leaving the house to visit friends or family, financial decisions, etc.).
Intermediate outcomes: An important set of intermediate outcomes relates to the impact of the workshops on parents’ attitudes, skills, knowledge, etc. We will develop a tailor-made set of questions, including which parents participated in which workshop.

Anthropometric measures

Height-for-age (HAZ), weight-for-age (WAZ), weight-for-height (WHZ), and head-circumference-for-age (HCAZ) z-scores computed from WHO Child Growth Standards are pre-specified as secondary confirmatory outcomes within the children’s development domain, contingent on funding being secured to field these measures at endline.

We will use two treatment dimensions assigned at the household level. First, we will randomly vary whether a household has access to the services provided by a CAIPI center (T2) or not (T1). This involves a place for all children 0-5 years’ old in the household. In addition, parents will be invited to monthly parental workshops that cover a variety of topics.
Second, among the households that received a place at a childcare center, we will vary whether parents receive a Whatsapp message on their mobile phone that reminds them of the upcoming monthly workshop. The message includes an infographic and a link to a very short video that encourage them to attend the workshop.

Not available

The randomization is conducted at the household level rather than the child level, to preserve household integrity: all eligible children in a selected household are admitted together. This rule reflects both the way INAIPI operates (families enrol siblings jointly) and the way we
will analyse outcomes (intra-household dynamics are part of the theory of change).
Household-level randomization combined with age-group capacity caps creates a non-trivial allocation problem: a uniform draw at the household level can violate age-group capacity in any single age cell. To respect both capacity and household integrity while pre-serving well-defined ex-ante admission probabilities, we use a scarcity-ordered stratified lottery with household integrity (hereafter, the bucket-lottery protocol). The protocol proceeds within each center as follows:
(i) Eligible age cells are defined at one-year resolution: g0–1, g1–2, g2–3, g3–4, g4–5. Per-cell capacities are set ex ante based on the center’s expected age distribution. Two indicative vectors were used in simulations, both summing to 195 seats per center: a conservative allocation (23, 28, 44, 50, 50) and a benevolent allocation (30, 36, 39, 45, 45) (scenarios S-A and S-C, see attached documents for more information).
(ii) Each household is assigned to a bucket corresponding to the age cell in which it is hardest to accommodate (its scarcest occupied cell). Households with children in multiple cells are bucketed by the scarcest occupied cell; households with children in a single cell are bucketed by that cell.
(iii) Buckets are processed in descending order of demand-to-capacity ratio (scarcest age cell first). Within each bucket, a uniform random ordering of households is drawn. Households are admitted in order until the bucket’s target age cell is filled (the treatment tranche), accounting for any spillover of siblings into other cells from earlier buckets; all remaining households constitute the control tranche. The same ordering is retained for the substitution rule described below.
(iv) A residual fill-up step admits additional households (drawn uniformly from eligible buckets) if any cell remains under-filled after all primary buckets are processed. Ex ante simulations show this step does not activate under realistic scenarios (zero activations in 4,000 replications), but it is pre-specified for completeness.
This protocol delivers (a) household integrity by construction, (b) known ex-ante admission probabilities per bucket, and (c) capacity filled by design. Analysis of treatment effects proceeds with center × bucket fixed effects, which fully absorb the differential admission
probabilities across strata.
Substitution for declined slots. A share of admitted households is expected to decline the offered slot (central assumption: 10%; see the compliance discussion in Section 5.3). Declined slots must be filled for operational reasons. To do so without compromising the design,
the same uniform random ordering of households drawn within each bucket in step (iii) is used to define a substitution order : when an admitted household declines, its slot is offered to the next control-tranche household in that pre-drawn order, subject to the same age-cell
capacity checks as steps (iii)–(iv). The household that accepts the substituted slot is retired from the analysis (it counts neither as treatment nor as control), because its enrolment depended on the realised pattern of declines rather than on the initial assignment. This rule
preserves one-sided noncompliance: only households promoted through the substitution order can transition from the control tranche to enrolment, and those households are excluded from the confirmatory comparison, so no household that remains in the control tranche for
analysis can become enrolled. The ITT/TOT framework therefore applies unchanged. The cost is a modest reduction in the size of the confirmatory control group, in expectation equal to the realised number of declined slots.
Two sources of heterogeneity in our sample motivated this design. First, based on existing survey data, we expect around 45% of households to be single-mother households (no male partner present), and the other 55% to be households with a male partner. This implies that for only around half of the sampled households will we have data on the male partner and on intra-household dynamics. Second, INAIPI attends children from 0 to 5; treatment relevance and outcome instruments differ across this age span, motivating the age-cell structure above. Stratification on mothers’ employment and marital status at baseline is not necessary because the realised variation in these dimensions (around 40% of women are employed; roughly one third are single mothers) is sufficient to support heterogeneity analyses without ex ante balancing.

Household

Yes

We will include 6 new CAIPI centers in this study. The intended CAIPIs for the evaluation will be located in urban and densely populated communities.

390 pupils per center, for a total of 2,340.

T1: 1,170 (195 pupils per center x 6 centers) control households.
T2: 1,170/2=585 attending CAIPI centres and not receiving workshop reminders
T3: 1,170/2=585 attending CAIPI centres and receiving workshop reminders.

This subsection reports minimum detectable effects (MDEs) computed under the realised sample sizes from the simulation scenarios (see Section 5.4 in attached document). All MDEs use a two-sided significance level of 5% and power of 80%, so that (z1−α/2 + z1−β ) = 2.80. Family-wise error across the four primary domains is controlled separately, by the Romano–Wolf step-down procedure (see Section 5.3); the per-test level here is therefore the conventional 5% rather than a stricter threshold, since applying both would double -adjust the same family. Standardised MDEs are reported in standard-deviation units; for the children’s development instruments, the corresponding raw-unit MDEs can be recovered by multiplying by the SIMEDID outcome standard deviations reported in Table 5. Two estimands are distinguished. The ITT MDE is the smallest assignment effect detectable at the stated sample size. The TOT MDE is the smallest treatment-on-the-treated (enrolment) effect detectable, obtained by inflating the ITT MDE by 1/(1 − decline rate); at the central decline rate of 10%, this is an inflation factor of 1.11. All MDEs further incorporate an endline attrition rate of 20%, which reduces the analytic sample in both arms and inflates every MDE by 1/√1 − 0.20 = 1.12. Attrition and slot decline are kept as separate, multiplicative adjustments: attrition is loss-to-follow-up affecting both arms and both estimands, whereas decline is the treated-side non-take-up that distinguishes ITT from TOT. The pooled and age-block child-outcome MDEs incorporate a household design effect of DEFF = 1 + ( ¯m − 1) ρ, with mean children per admitted household ¯m ≈ 1.14 and an assumed within-household outcome correlation ρ = 0.30, giving DEFF ≈ 1.04; because ¯m is close to one (81.6% of households contribute a single child), the MDEs are insensitive to ρ over any plausible range. One-year-cell MDEs assume DEFF = 1, because within a single one-year cell almost every household contributes exactly one child. Table 3 reports the results. Panel A shows that the primary confirmatory levels are adequately powered. The pooled (across-age-cell) analysis attains a standardised ITT MDE of approximately 0.13 SD (TOT ≈ 0.15 SD) across all four scenario combinations. The two age-block analyses attain standardised ITT MDEs of approximately 0.18–0.19 SD for the 0–3 block and 0.18–0.21 SD for the 3–5 block (TOT ≈ 0.21–0.24 SD), the wider range for 3–5 reflecting the reduced older-cell control counts under the stressed demand scenario. For mothers’ employment (a household-level binary outcome with baseline rate 0.43), the pooled ITT MDE is approximately 6.8–7.2 percentage points. All of these figures already incorporate the 20% attrition and 10% decline adjustments.