To estimate the impacts of 2 interventions, early stimulation (ES) for children aged <3 years and enhanced preschool (EP) for children aged 3+ years, and their interactions.
In Odisha, India, 192 villages were randomly assigned to ES or to no ES. Within each village, about 8 mothers with children initially aged 7 to 16 months were enrolled, receiving ES or no ES accordingly (n = 1449). Subsequently, when children were aged ∼3 years, the villages were rerandomized to either EP at Anganwadi centers or no EP. This yielded 4 groups: (1) ES and EP, (2) only ES, (3) only EP, and (4) no intervention. Trained Anganwadi workers ran the EP. Primary outcomes, measured at baseline and follow-up after ∼1 year, were children’s IQ (summarizing cognition, language, and executive functioning) and school readiness (SR). Secondary outcomes were home environments, caregivers’ child-development knowledge. and preschool quality.
Fifteen months after ES ended, onlyES had a sustained benefit on IQ (0.18 SD, P <.04) and on SR (0.13 SD, P <.08). Only EP improved IQ (0.17 SD, P <.04) and SR (0.24 SD, P <.01). Receiving both interventions improved IQ (0.24 SD, P <.01) and SR (0.21 SD, P <.01). No statistically significant interactions between the 2 interventions were observed.
Both ES and EP increased IQ and SR. Only ES impacts were sustained for 15 months. Only EP resulted in considerable catch-up for children who did not receive only ES. The absence of significant complementarities should be investigated further because of its profound policy implications.
Millions of children are at risk for developmental deficits in low and-middle-income countries (LMICs).1–3 Reviews find that psychosocial interventions for children aged <3 years improve short-run child cognition and language (0.28–0.47 SD).4–6 Similarly, a meta-regression analysis of 54 preschool interventions for children aged ≥3 years found significant improvements in children’s cognitive skills (0.15 SD), executive functioning, social–emotional learning, and behavior (0.12 SD). Only 18 of these interventions were from LMICs, with 2 from India,7,8 which has the world’s largest population of children attending preschool (36 million children enrolled in Integrated Childhood Development Services [ICDS]). Interventions have had benefits in math and language.7,8 However, a survey of 298 Indian preschools found generally poor quality.9,10 Although short-run impacts of some interventions fade,11–15 some rigorous studies with long-term follow-ups found later benefits in educational attainment, reduced crime, and increased income.13,14,16–20
Whereas nutritional interventions are most effective in <2 years,21 best ages for psychosocial interventions are unclear. Is it necessary to initiate stimulation at <1 year or is starting with preschool at ∼3 years soon enough, as in some successful interventions22–25 ? Are effects greater if birth to school enrollment is covered? Is there any complementarity between early parenting interventions and preschool quality?
To address these issues, we build on an early stimulation (ES) 2-year intervention for 7- to 16-months-old children in poor rural areas in Odisha, India. This was evaluated by a cluster randomized-controlled trial in 192 villages, with several significant impacts.26 After this ES trial, the intervention children, like most Indian children aged ≥3 years were eligible (and attend) the ICDS preschool Anganwadi centers (AWCs). At that time, we rerandomized villages in each study group to either (1) enhanced preschool (EP), or (2) no enhanced preschool (NEP), the usual practice.
We investigate 3 primary hypotheses about policy design:
Children receiving only ES have sustained intelligence (or IQ, summarizing cognition, language, executive-functioning) and school readiness (SR) benefits compared with children with no intervention.
Children attending EP benefit compared with those with no intervention.
Children receiving both ES and EP benefit more than children who receive 1 intervention.
These hypotheses focus on intelligence and SR benefits. Three parallel secondary hypotheses focus on home environments, caregivers’ child-development knowledge, and preschool quality.
Methods
We conducted 2 sequential open-label, cluster randomized controlled trials.
Early Trial
A total of 192 Odisha villages (clusters) were stratified by district and randomized into 4 equal groups:
control;
nutritional education (NE);
home-visiting (HV) stimulation based on Reach Up curriculum adapted for local contexts27 plus NE; and
similar stimulation in group sessions (GS) of ≤8 children with primary caregivers plus NE.
Children were identified through mid-2015 household censuses. Singleton children aged 7 to 16 months with no obvious disabilities were eligible, with ≤8 children enrolled per village. After 2 years (December 2017), at age 31 to 40 months, we found no effects of NE and similar significant positive impacts for HV and GS in cognitive and language development. Details are available elsewhere.26
Late Trial
We combined HV and GS into ES. To preserve contact with the families, we extended the GS curriculum to run 8 weekly group sessions, starting 4 months after ES ended, in all 192 villages. After obtaining governmental permission to implement EP (July 2018), including quarterly training sessions for Anganwadi staff, we rerandomized all 192 villages to EP or to usual practice (NEP) with 50/50 for each early trial arm, when the children were aged 36 to 45 months. We identified 216 centers in treatment and 217 in control villages in a 2017 census. This yielded 4 groups: (1) ES and EP, (2) only ES, (3) only EP, (4) no intervention. Many older children already had enrolled in AWCs. Anganwadi workers (AWWs) and helpers (AWHs) in the 216 AWCs in the 96 treatment villages were trained. EP was implemented from July 2018 in all 216 AWCs.
Data were collected by Morsel Research and Development, an independent organization, with training by study investigators. Consent (written or oral if illiterate) was obtained from household heads and primary caregivers. Informed consent for EP participation was collected from all households at late-trial baseline and follow-up surveys. EP and measurements not used previously in India were piloted and adapted as necessary.
Data were collected at the late-trial baseline (coinciding with the early trial endline) and follow-up, 9 months after EP implementation (2019) when the children were aged 46 to 55 months. Although the intervention ended 11 months later in March 2020, it was not possible to collect data because of coronavirus disease 2019. However, further follow-up is in progress.
Institutional review board approvals: Yale University (IRB-1112009492), University College London (IRB-2168/014), University of Pennsylvania (IRB-815027), and Pratham Education Foundation (IRB-PEF/AC-1/2).
Participants
EP Intervention
EP builds on the ICDS AWC curriculum, New Arunima,28 implemented by AWWs with AWHs (Supplemental Information 1). EP was implemented in all treatment-village AWCs. AWWs’ average age was 38 years, professional experience in their current job averaged 13 years, 39.0% had bachelor’s degrees, 57.7% had secondary or higher-secondary education, and 3.3% had not completed secondary education.
Enhanced Curriculum
We reviewed the existing ICDS curriculum using a framework based on recommendations on Developmentally Appropriate Practices29 and the Early Childhood Curriculum Framework.28 We added items from the Reach Up curriculum27 and “Tools of the Mind”30 ; early literacy, numeracy, and self-regulation activities were designed to engage children in higher-order thinking and included dialogic-reading methods.31 We added activities focusing on executive functioning and cognitive concepts, including size, shape, color, and position through play activities, connected to children’s experiences. We aimed to improve teacher–children interactions by encouraging 2-way conversations, using open-ended questions and positive feedback. Scaffolding and building from easy to difficult helped ensure that activities fell within each child’s proximal developmental zone.30 We promoted socioemotional development by encouraging sharing, taking turns, and empathy, often through stories. Activity corners were set up daily. Children chose activities daily, attended at least 1 structured group-learning session and had singing and story time. We promoted conversations among children and adults during meals. All AWC children were included regardless of study membership.
Mentors and Implementers
AWWs and AWHs were the main EP implementers. In addition, Pratham recruited 108 mentors with secondary-level education from the communities. These mentors were responsible for training and in-service mentoring for the AWWs and AWHs in the enhanced curriculum and its implementation. Each visited 2 AWCs twice weekly to help curriculum implementation and to discuss any issues. To avoid interrupting AWC activities, in-service mentoring was the main training approach. The mentors were supervised by 3 to 4 senior staff, “super-mentors,” who were in frequent contact with the research team. The super-mentors received 4-day training quarterly. The mentors received 10-day initial training and 7-day training quarterly. The AWWs and AWHs received 2-day initial training and 2-day refresher training quarterly. The overall approach was designed to improve local acceptance, use existing infrastructure, and engage the ICDS, which can enable scaling up successful interventions.
Parenting Meetings
Starting in intervention month 3, parents and their children were invited to AWCs for 90-minute monthly parent–teacher meetings. Parents were introduced to activities and materials, teaching them the importance of language during daily activities, book use, how to encourage learning, provide good nutrition and health care, and activities for home play. Intervention books were taken home from the lending library and introduced in each AWC to reinforce activities promoted in the sessions. In addition, 3 community melas/fairs were organized for each AWC, where children, their parents, extended family members, and the communities-at-large were invited to engage in games and activities showcasing what children had learned in AWCs.
Materials
The intervention used play materials, including books, puzzles, and age-appropriate toys. If possible, these were made locally, initially by mentors, and then together with AWWs and AWHs using readily available materials to lower costs, improving scalability. Books, designed by Pratham and the Research Team, were printed in Delhi.
Outcomes
Primary outcomes were children’s intelligence (IQ, summarizing cognition, language, executive functioning), and SR. We selected subscales from the Wechsler Preschool and Primary Scale of Intelligence-IV (WPPSI-IV32,33 ), which is widely used internationally: Verbal Comprehension, Visual Spatial, Working Memory, Processing Speed Indices, and Matrix Reasoning subscales. The last 2 scales are only for children aged ≥4 years, whereas the others are for children aged ≥2.5 years. All subscales were combined into the Full-Scale IQ. To assess SR, we adapted and selected age-appropriate items from the Daberon-II Screening for SR test,34 which assesses preacademic knowledge including knowing body parts, color and number concepts, prepositions, following directions, general knowledge, visual perception, gross-motor skills, and categories. We selected items from all subscales except from motor development, ending up with 83 items, and extended number concepts with more difficult items (32 items for identifying, matching, and naming numbers), and we added items to measure early literacy (40 items for identifying, matching, and naming letters).35
Children were tested at home accompanied by their primary caregivers. The testers had tertiary education or equivalent experience working with children and had 4-week training for adapted WPPSI-IV and 2 weeks for adapted Daberon-II. Interobserver reliabilities were assessed before and during the study, and intraclass correlations ranged from 0.82 to 0.99 (n = 298) for adapted-WPPSI-IV and 0.95 to 0.97 (n = 148) for adapted Daberon-II. The correlation between IQ and SR was 0.54.
Secondary outcomes included home environments, caregivers’ child-development knowledge, preschool program quality, and child behaviors. Surveyors had 2-week training and conducted interviews with mothers/primary caregivers. To assess home environment quality, we used:
the United Nations Children’s Fund family care indicators (FCI36 ), including a play-materials scale, which records presence of certain types of toys and books, and a play-activities scale, which includes parental/caregiver involvement with children in certain play activities in the last 3 days; and
selected items from the acceptance, responsivity, physical environment, language stimulation and academic stimulation of the Early Childhood Home Observation for the Measurement of the Environment.37
We assessed mothers’ knowledge using adapted items from the Knowledge of Infant Development Inventory38 and child behaviors by parental reports with the Strength and Difficulties Questionnaire.39
Finally, AWC quality was assessed at baseline and follow-up by a newly designed preschool assessment tool that included subscales and items from several of the most widely used instruments for measuring process quality:
the Classroom Assessment Scoring System,40 which captures teacher–children interaction quality. The adapted and selected subscales focus on positive and negative climate, instructional learning formats, concept development, quality of feedback, and language modeling;
the Early Childhood Education Rating Scale-Revised,41 with selected subscales focused on personal-care activities; and
the Early Childhood Education Quality Assessment Scale (developed by the Ambedkar University Centre for Early Childhood Education and Development) for activity–quality assessment in Indian preschools.10
The selected subscales focus on activities and AWC physical conditions (eg, space use, display of children’s work, teacher–child ratios). The AWC observers had graduate degrees or equivalent experience and 2-week training. Interobserver reliabilities were assessed before the study and intraclass correlations ranged from 0.76 to 0.95 (n = 44). Because of financial constraints, we collected data in 192 AWCs, 1 randomly selected in each of the 192 villages in the study. The observers visited each AWC for 1 day when the mentor was not present. Before the first activity began and during free play, the observers investigated classrooms and recorded all materials children had access to independent of whether children were using these materials. Then, they recorded all activities during the day, assessed interactions between teachers and children, and interviewed the AWWs. Finally, they collected enrollment and attendance data of target children. To measure preschool quality, we created (1) an aggregate raw index for teacher–children interaction quality, and (2) individual indicators of physical conditions.
All instruments measuring outcomes were translated into the local language Odia and extensively piloted and adapted to be culturally appropriate without changing the underlying constructs. Internal reliabilities were satisfactory (Cronbach’s α 0.7–0.9) for adapted WPPSI-IV, adapted Daberon-II, and preschool assessment tool modules; however, they were low for the adapted Strength and Difficulties Questionnaire (0.5–0.6), therefore we dropped this outcome because our translated version may not be valid for this population.
Sample Size
The sample comes from the early trial endline (192 [100%] clusters, 1298 [92%] children). The minimum detectable effect sizes42 for a single main hypothesis test are 0.17 SD for 2-tailed test 5% significance, 80% power, 80% compliance, 5% attrition, and 0.04 intracluster correlation (found in the early trial). Given that we have 2 primary outcomes, the minimum detectable effect sizes’42 upper bound, adjusting for multiple testing,43 is 0.21 SD for each hypothesis.
Randomization and Blinding
We randomized half (96) of the communities in each of the early trial arms to EP. A Stata-14 random number generator was used with reproducible seeds by an independent researcher. Testers and surveyors were blinded to treatments but participants could not be blinded. All tests and interviews were conducted in family homes, minimizing chances of deducing treatment status from observing materials made for EP in AWCs.
Statistical Analysis
Results
Most children enrolled in AWCs (Supplemental Information 2; Supplemental Fig 2); proportions attending private preschools or not in preschools were similar for EP and NEP (Supplemental Table 5). Table 1 indicates no significant differences between characteristics of the 3 treatment groups and controls at early trial baseline, except for Ages and Stages Questionnaires, Third Edition, communication and personal social scores.
Sample Characteristics by Treatment Group
| Difference | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| N | Control | Only ES | Only EP | EP and ES | Only ES | Only EP | EP and ES | Only EP–only ES | EP and ES–only ES | EP and ES–only EP | |
| Panel A: child-level characteristics at baseline early trial | |||||||||||
| Child is male | 1298 | 0.54 | 0.50 | 0.53 | 0.48 | −0.05 | −0.01 | −0.06 | 0.04 | −0.02 | −0.05 |
| (0.5) | (0.5) | (0.5) | (0.5) | (0.04) | (0.04) | (0.04) | (0.04) | (0.04) | (0.04) | ||
| Child age in mo | 1298 | 12.3 | 12.3 | 12.3 | 12.0 | −0.1 | −0.1 | −0.3 | −0.1 | −0.3 | −0.2 |
| (2.7) | (2.5) | (2.6) | (2.7) | (0.2) | (0.2) | (0.2) | (0.2) | (0.) | (0.2) | ||
| Panel B: sample characteristics at baseline early trial | |||||||||||
| ASQ-3 communication score | 1298 | 84.7 | 85.9 | 83.9 | 87.3 | 1.2 | −0.8 | 2.5 | −2.0 | 1.4 | 3.4* |
| (20.1) | (19.0) | (20.5) | (18.7) | (2.2) | (2.2) | (2.0) | (2.2) | (2.0) | (2.0) | ||
| ASQ-3 gross motor score | 1298 | 88.1 | 89.2 | 86.0 | 87.5 | 1.1 | −2.1 | −0.6 | −3.1 | −1.7 | 1.4 |
| (27.0) | (26.8) | (24.8) | (24.0) | (2.2) | (2.3) | (2.3) | (2.2) | (2.3) | (2.4) | ||
| ASQ-3 fine motor score | 1298 | 94.0 | 94.4 | 94.2 | 96.4 | 0.4 | 0.2 | 2.4 | −0.1 | 2.0 | 2.1 |
| (18.8) | (17.4) | (18.0) | (16.0) | (1.6) | (1.8) | (1.7) | (1.5) | (1.4) | (1.6) | ||
| ASQ-3 problem-solving score | 1297 | 94.7 | 95.7 | 94.3 | 95.9 | 0.9 | −0.4 | 1.1 | −1.4 | 0.2 | 1.5 |
| (21.3) | (19.5) | (21.6) | (19.2) | (1.8) | (2.0) | (1.6) | (2.0) | (1.6) | (1.8) | ||
| ASQ-3 personal social score | 1298 | 80.5 | 80.4 | 77.4 | 80.9 | −0.2 | −3.1 | 0.4 | −3.0 | 0.5 | 3.5* |
| (21.0) | (21.1) | (21.0) | (18.8) | (1.9) | (2.0) | (1.9) | (1.9) | (1.7) | (1.8) | ||
| Mother’s Raven’s score | 1298 | 0.1 | 0.0 | 0.0 | 0.0 | 0.1 | 0.1 | 0.1 | 0.0 | 0.0 | 0.0 |
| (0.9) | (0.8) | (0.9) | (0.9) | (0.1) | (0.1) | (0.1) | (0.1) | (0.1) | (0.1) | ||
| Mother’s years of education | 1298 | 7.4 | 7.1 | 7.4 | 7.4 | −0.3 | −0.1 | −0.1 | 0.2 | 0.2 | 0.0 |
| (3.4) | (3.7) | (3.4) | (3.3) | (0.4) | (0.3) | (0.4) | (0.4) | (0.4) | (0.4) | ||
| FCI, activities | 1292 | 2.6 | 2.7 | 2.7 | 2.6 | 0.1 | 0.1 | −0.1 | 0.1 | −0.2 | −0.1 |
| (1.4) | (1.5) | (1.4) | (1.4) | (0.2) | (0.1) | (0.1) | (0.2) | (0.2) | (0.1) | ||
| FCI, materials | 1292 | 2.2 | 2.1 | 2.1 | 2.1 | −0.1 | −0.1 | −0.1 | 0.0 | −0.1 | 0.0 |
| (1.4) | (1.4) | (1.4) | (1.3) | (0.2) | (0.2) | (0.2) | (0.2) | (0.2) | (0.2) | ||
| Panel B: preschool-level characteristics at baseline late trial | |||||||||||
| Teacher–children interaction quality | 188 | 5.5 | 5.3 | 5.5 | 5.6 | −0.2 | 0.0 | 0.1 | 0.2 | 0.3 | 0.1 |
| (1.0) | (0.9) | (1.1) | (1.3) | (0.2) | (0.2) | (0.2) | (0.2) | (0.2) | (0.2) | ||
| Difference | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| N | Control | Only ES | Only EP | EP and ES | Only ES | Only EP | EP and ES | Only EP–only ES | EP and ES–only ES | EP and ES–only EP | |
| Panel A: child-level characteristics at baseline early trial | |||||||||||
| Child is male | 1298 | 0.54 | 0.50 | 0.53 | 0.48 | −0.05 | −0.01 | −0.06 | 0.04 | −0.02 | −0.05 |
| (0.5) | (0.5) | (0.5) | (0.5) | (0.04) | (0.04) | (0.04) | (0.04) | (0.04) | (0.04) | ||
| Child age in mo | 1298 | 12.3 | 12.3 | 12.3 | 12.0 | −0.1 | −0.1 | −0.3 | −0.1 | −0.3 | −0.2 |
| (2.7) | (2.5) | (2.6) | (2.7) | (0.2) | (0.2) | (0.2) | (0.2) | (0.) | (0.2) | ||
| Panel B: sample characteristics at baseline early trial | |||||||||||
| ASQ-3 communication score | 1298 | 84.7 | 85.9 | 83.9 | 87.3 | 1.2 | −0.8 | 2.5 | −2.0 | 1.4 | 3.4* |
| (20.1) | (19.0) | (20.5) | (18.7) | (2.2) | (2.2) | (2.0) | (2.2) | (2.0) | (2.0) | ||
| ASQ-3 gross motor score | 1298 | 88.1 | 89.2 | 86.0 | 87.5 | 1.1 | −2.1 | −0.6 | −3.1 | −1.7 | 1.4 |
| (27.0) | (26.8) | (24.8) | (24.0) | (2.2) | (2.3) | (2.3) | (2.2) | (2.3) | (2.4) | ||
| ASQ-3 fine motor score | 1298 | 94.0 | 94.4 | 94.2 | 96.4 | 0.4 | 0.2 | 2.4 | −0.1 | 2.0 | 2.1 |
| (18.8) | (17.4) | (18.0) | (16.0) | (1.6) | (1.8) | (1.7) | (1.5) | (1.4) | (1.6) | ||
| ASQ-3 problem-solving score | 1297 | 94.7 | 95.7 | 94.3 | 95.9 | 0.9 | −0.4 | 1.1 | −1.4 | 0.2 | 1.5 |
| (21.3) | (19.5) | (21.6) | (19.2) | (1.8) | (2.0) | (1.6) | (2.0) | (1.6) | (1.8) | ||
| ASQ-3 personal social score | 1298 | 80.5 | 80.4 | 77.4 | 80.9 | −0.2 | −3.1 | 0.4 | −3.0 | 0.5 | 3.5* |
| (21.0) | (21.1) | (21.0) | (18.8) | (1.9) | (2.0) | (1.9) | (1.9) | (1.7) | (1.8) | ||
| Mother’s Raven’s score | 1298 | 0.1 | 0.0 | 0.0 | 0.0 | 0.1 | 0.1 | 0.1 | 0.0 | 0.0 | 0.0 |
| (0.9) | (0.8) | (0.9) | (0.9) | (0.1) | (0.1) | (0.1) | (0.1) | (0.1) | (0.1) | ||
| Mother’s years of education | 1298 | 7.4 | 7.1 | 7.4 | 7.4 | −0.3 | −0.1 | −0.1 | 0.2 | 0.2 | 0.0 |
| (3.4) | (3.7) | (3.4) | (3.3) | (0.4) | (0.3) | (0.4) | (0.4) | (0.4) | (0.4) | ||
| FCI, activities | 1292 | 2.6 | 2.7 | 2.7 | 2.6 | 0.1 | 0.1 | −0.1 | 0.1 | −0.2 | −0.1 |
| (1.4) | (1.5) | (1.4) | (1.4) | (0.2) | (0.1) | (0.1) | (0.2) | (0.2) | (0.1) | ||
| FCI, materials | 1292 | 2.2 | 2.1 | 2.1 | 2.1 | −0.1 | −0.1 | −0.1 | 0.0 | −0.1 | 0.0 |
| (1.4) | (1.4) | (1.4) | (1.3) | (0.2) | (0.2) | (0.2) | (0.2) | (0.2) | (0.2) | ||
| Panel B: preschool-level characteristics at baseline late trial | |||||||||||
| Teacher–children interaction quality | 188 | 5.5 | 5.3 | 5.5 | 5.6 | −0.2 | 0.0 | 0.1 | 0.2 | 0.3 | 0.1 |
| (1.0) | (0.9) | (1.1) | (1.3) | (0.2) | (0.2) | (0.2) | (0.2) | (0.2) | (0.2) | ||
SDs (columns 2 to 5) and cluster SEs at the village level (columns 6 to 11) in parentheses. P values calculated on the basis of clustered SEs. N: sample size at baseline early or late trial. ASQ-3, Ages & Stages Questionnaires, Third Edition..
P < .1.
Primary Outcomes
Table 2 gives estimated impacts, with P values adjusted for multiple testing43 for the 6 hypotheses embedded in equation (1), namely 3 impacts each (β1, β2, β3) for IQ and SR. Results include baseline controls registered in the protocol.35 Supplemental Table 6 reports all outcome scores means and SD by treatment arm at follow-up EP.
Treatment Effects on IQ and School Readiness
| IQ | SR | |
|---|---|---|
| Only ES | 0.179 | 0.133 |
| (0.072) [0.035] | (0.076) [0.080] | |
| Only EP | 0.166 | 0.239 |
| (0.068) [0.035] | (0.073) [0.005] | |
| ES and EP | 0.243 | 0.209 |
| (0.077) [0.007] | (0.069) [0.009] | |
| N | 1283 | 1272 |
| IQ | SR | |
|---|---|---|
| Only ES | 0.179 | 0.133 |
| (0.072) [0.035] | (0.076) [0.080] | |
| Only EP | 0.166 | 0.239 |
| (0.068) [0.035] | (0.073) [0.005] | |
| ES and EP | 0.243 | 0.209 |
| (0.077) [0.007] | (0.069) [0.009] | |
| N | 1283 | 1272 |
Estimated coefficients are expressed in SDs of the control group (SDs are 10.38 for IQ and 24.37 for SR). Sample size includes all target children who completed the IQ (adapted WPPSI-IV) at follow-up for column 1 and who completed the SR (adapted Daberon-II) at follow-up for column 2. The SEs (reported in parentheses) and the 2-tailed P values (reported in square brackets) are calculated using the bootstrap (5000 replications), allowing for clustering at the village level and for stratification used in the randomization protocol. The P values are also corrected for multiple (6) hypotheses testing using the Romano-Wolf step-down method.43 Results control for the following covariates: child’s sex; tester fixed effects; baseline age in months; Ages & Stages Questionnaires, Third Edition; mother’s education and Raven’s score; and preschool interaction quality at baseline late trial. IQ refers to intelligence measured as the scaled score by the adapted WPPSI-IV and SR was measured using the adapted Daberon-II.
Only EP improves IQ by 0.17 SD (P <.04) and SR by 0.24 SD (P <.01). Only ES increases IQ by 0.18 SD (P <.04) and SR by 0.13 SD (P <.08). Thus, the ES impacts were sustained for ∼15 months for children without EP. As noted, most of these children were enrolled in AWCs. Finally, receiving both interventions increased IQ by 0.24 SD (P <.01) and SR by 0.21 SD (P <.01). Although the IQ improvement from ES and EP is higher than that obtained from only EP, the difference is not significant and thus we cannot reject the hypothesis that ES did not affect the EP impact. Supplemental Table 7 reports estimated impacts on IQ by subscales. Only EP improves Visual Spatial by 0.22 SD (P <.03). Only ES improves Verbal Comprehension by 0.20 SD (P <.05). Receiving both interventions increased both Verbal Comprehension and Visual Spatial by 0.26 SD (P <.01) and 0.24 SD (P <.01), respectively.
Secondary Outcomes
EP had no effects on primary caregivers’ knowledge, but the teacher–child interaction quality aggregate index improved substantially (0.32 SD) (P <.01) (Table 3).
Treatment Effects on Primary Caregiver’s Knowledge and Preschool Interaction Quality
| Panel A | Parenting Knowledge: Adapted KIDI |
| Only ES | 0.020 |
| (0.073) [0.943] | |
| Only EP | −0.029 |
| (0.064) [0.943] | |
| EP and ES | 0.048 |
| (0.066) [0.805] | |
| N | 1258 |
| Panel B | Teacher–children interaction quality |
| EP | 0.318 |
| (0.113) [0.005] | |
| N | 186 |
| Panel A | Parenting Knowledge: Adapted KIDI |
| Only ES | 0.020 |
| (0.073) [0.943] | |
| Only EP | −0.029 |
| (0.064) [0.943] | |
| EP and ES | 0.048 |
| (0.066) [0.805] | |
| N | 1258 |
| Panel B | Teacher–children interaction quality |
| EP | 0.318 |
| (0.113) [0.005] | |
| N | 186 |
Estimated coefficients are expressed in SDs of the control group. Sample size includes all mothers who completed the adapted Knowledge of Infant Development Inventory questionnaire in Panel A. Sample size includes all AWCs observed using the preschool assessment tool at follow-up (1 per village) in column 1, panel B to assess teacher–children interaction quality. To recover the impact of the teacher–children interaction quality aggregate index and subscales, we compare the treated versus control for the late trial only. The SEs (reported in parentheses) and the 2-tailed P values (reported in square brackets) are calculated using the bootstrap (5000 replications), allowing for clustering at the village level and for stratification used in the randomization protocol. The P values are also corrected for multiple (3) hypotheses testing using the Romano-Wolf step-down method.43 Covariates included: child’s sex; tester fixed effects; baseline age in months; Ages & Stages Questionnaires, Third Edition; mother’s education and Raven’s score; and preschool interaction quality at baseline late trial. KIDI, Knowledge of Infant Development Inventory.
EP significantly improved teacher–child interaction quality (Supplemental Table 8) in instructional learning formats, concept development, quality of feedback, and language modeling. For physical conditions, there was more children’s work displayed and at children’s eye levels (Supplemental Table 9).
Table 4 reports impacts on home environments. FCI activities improved substantially for only ES and EP and ES, and adapted Early Childhood Home Observation for the Measurement of the Environment increased for both only ES and receiving both EP and ES. Supplemental Table 10 reports EP impacts on home environments by subscales, though with no significant impacts except for academic stimulation for both only ES and receiving both EP and ES.
Treatment Effects on Home Environments
| FCI, Activities | FCI, Materials | Adapted EC-HOME | |
|---|---|---|---|
| Only ES | 0.238 | 0.095 | 0.170 |
| (0.064) [0.002] | (0.075) [0.504] | (0.057) [0.016] | |
| Only EP | 0.065 | 0.122 | 0.042 |
| (0.063) [0.504] | (0.070) [0.277] | (0.060) [0.504] | |
| ES and EP | 0.186 | 0.144 | 0.127 |
| (0.055) [0.006] | (0.063) [0.101] | (0.049) [0.053] | |
| N | 1264 | 1264 | 1264 |
| FCI, Activities | FCI, Materials | Adapted EC-HOME | |
|---|---|---|---|
| Only ES | 0.238 | 0.095 | 0.170 |
| (0.064) [0.002] | (0.075) [0.504] | (0.057) [0.016] | |
| Only EP | 0.065 | 0.122 | 0.042 |
| (0.063) [0.504] | (0.070) [0.277] | (0.060) [0.504] | |
| ES and EP | 0.186 | 0.144 | 0.127 |
| (0.055) [0.006] | (0.063) [0.101] | (0.049) [0.053] | |
| N | 1264 | 1264 | 1264 |
Estimated coefficients are expressed in SDs of the control group. Sample size includes all mothers who completed the FCI at follow-up in columns 1 and 2, and the adapted Early Childhood Home Observation for the Measurement of the Environment at follow-up in column 3. The SEs (reported in parentheses) and the 2-tailed P values (reported in square brackets) are calculated using the bootstrap (5000 replications), allowing for clustering at the village level and for stratification used in the randomization protocol. The P values are also corrected for multiple (9) hypotheses testing using the Romano-Wolf step-down method.43 Covariates included: child’s sex; tester fixed effects; baseline age in months; Ages & Stages Questionnaires, Third Edition; mother’s education and Raven’s score; and preschool interaction quality at baseline late trial.
Discussion
Our estimates have 3 important implications. First, children assigned to ES (HV or GS), showed small-to-moderate cognition and language benefits (0.24–0.32 SDs26 ) at early trial end. At follow-up, 15 months later, they showed sustained IQ (0.18 SD) and SR (0.13 SD) benefits, and improvement in FCI play activities with adults (0.24 SD) and home environments (0.17 SD), which suggest benefits might continue. Few LMIC early childhood interventions have been followed up after the intervention ended,8,12,13 and most studies have used small samples. In 2 studies with larger samples, findings were inconsistent.13,44 Most children in our study area attended AWCs (enhanced or not), and this may have contributed to sustained only ES effects.
Second, children attending EP with or without ES showed improvements in IQ and SR. Impact sizes compare favorably with a recent preschool review,7 which found an average cognitive benefit of 0.15 SD. In our study, testing occurred 9 months after program start, but it took several months to establish fully the program in all treatment centers. The AWWs were unavailable for substantial initial training, so we used slow in-service training delivered by mentors twice weekly. Observations of AWCs at 9 months showed that interventions were successfully delivered, and teacher–children interaction quality improved substantially (0.32 SD), particularly in instructional learning formats, concept development, quality of feedback and language modeling.
Third, the point estimates suggest that receiving both ES and EP confers larger cognitive benefits than receiving just 1 intervention. However, the effect sizes are not statistically distinguishable for receiving ES and EP versus receiving only ES or only EP. This result, relating to whether an early start is essential, needs further follow-up with larger samples because it has potentially major implications for early childhood development design. Moreover, only ES but not only EP shows improvement on the FCI play activities with adults and home environment, suggesting that sustainability may differ by group.
The study has limitations. There was some delay between the completion of ES and the initiation of EP waiting for governmental approval, possibly reducing interaction effects of the 2 interventions. Our study allows for minimum detectable effects of 0.2 SD between any 2 arms. More importantly, the pandemic prevented endline data collection, making us rely on midline data when the intervention may have not reached its full impact.
Observations of AWCs showed that they had improved interaction and teaching quality with children, which holds promise for the sustainability of the intervention and its broader impact, beyond the study cohort of children. However, program quality often deteriorates without ongoing support, so continued mentoring at reduced frequencies is probably necessary. Mentors who spend the whole day and participate in the program, just as the study mentors did, are more likely to maintain the AWWs’ motivation and program quality than supervisors checking lists. Furthermore, play materials will need replacing with time. Both the mentors and materials would require additional expenditure, but it is a great opportunity to improve these children’s life chances, mostly using the existing infrastructure.
In summary, the EP effect and the ES sustainability are encouraging. Follow-up data, under collection presently, may illuminate further program impacts.
Acknowledgments
We thank Diana Perez Lopez and Gianina Morales Suazo for their excellent research assistance.
Drs Attanasio, Behrman, Grantham-McGregor, and Meghir conceptualized and designed the study, supported the development of the curriculum/intervention, contributed to the design and adaptation of the data collection instruments, contributed to interpretation of the data analysis, and drafted and critically reviewed the manuscript for important intellectual content; Dr Pal and Ms Gupta supported the development of the curriculum/intervention, and trained and supervised staff for implementation; Mr Phimister contributed to the design and adaptation of the data collection instruments, supported training of staff for implementation and data collection, and conducted analysis and interpretation of the data; Dr Jervis conceptualized and designed the study, supported the development of the curriculum/intervention, contributed to the design and adaptation of the data collection instruments, contributed to interpretation of the data analysis, drafted and critically reviewed the manuscript for important intellectual content, supported training of staff for implementation and data collection, and conducted analysis and interpretation of the data; Dr Day and Ms Makkar supported the development of the curriculum/intervention, trained and supervised staff for implementation, and critically reviewed the manuscript for important intellectual content; Ms Vernekar contributed to the design and adaptation of data collection instruments. and trained and supervised staff for implementation and data collection; and all authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.
This study is registered at ISRTN, #12916148, and AEA RCT Registry, #0003161.
FUNDING: Funded and supported by the European Research Council under the European Union’s Horizon 2020 Research and Innovation Programme (grant #695300, HKADeC, European Research Council-2015-AdG, PI Dr Attanasio); the Early Learning Partnership, World Bank; the Economic and Social Research Council (grant #ES/M010147/1); the National Institutes of Health (grant #R01 HD 72120, Principal Investigator Dr Meghir); the Cowles Foundation at Yale University; the Population Studies Center at the University of Pennsylvania; Dubai Cares; and the Jacobs Foundation (Klaus J. Jacobs Research Prize 2016, Dr Attanasio). Dr Jervis receives financial support from the Institute for Research in Market Imperfections and Public Policy (#ICS13_002 ANID) and the Center for Research in Inclusive Education, Chile (#SCIA ANID CIE160009). The funders reviewed and approved the preliminary study design, but had no further roles in the design or conduct of this study.
CONFLICT OF INTEREST DISCLOSURES: The authors have indicated they have no conflicts of interest relevant to this article to disclose.
- AWC
Anganwadi center
- AWH
Anganwadi helper
- AWW
Anganwadi worker
- ES
early stimulation
- FCI
family care indicators
- GS
group sessions
- HV
home visiting
- ICDS
Indian Integrated Child Development Services
- IRB
institutional review board
- LMIC
low- and-middle-income country
- NE
nutritional education
- NEP
no enhanced preschool
- SR
school readiness
- WPPSI-IV
Wechsler Preschool and Primary Scale of Intelligence-IV
Comments