The impact of a Caribbean home-visiting child development program on cognitive skills

doi:10.1016/j.econedurev.2013.12.003

Economics of Education Review
《教育经济学评论》

Volume 39, April 2014, Pages 22-37
第 39 卷，2014 年 4 月，第 22-37 页

https://doi.org/10.1016/j.econedurev.2013.12.003 Get rights and content 获取权利和内容

Highlights 突出

•
We estimate the impact of a home-based intervention on cognitive outcomes of young children.
我们估计了家庭干预对幼儿认知结局的影响。
•
The program does not affect the cognitive development of the average child or the older birth cohort.
该计划不会影响普通儿童或年龄较大的出生群体的认知发展。
•
The program significantly improves early reading and writing skills of the younger cohort.
该计划显著提高了年轻群体的早期阅读和写作技能。
•
No impact is found on the language development of either cohort.
未发现对两个队列的语言发展产生影响。
•
The findings suggest that home-based programs may have an early window of opportunities.
研究结果表明，以家庭为基础的计划可能有一个早期的机会窗口。

Abstract 抽象

This paper provides a short-term impact evaluation of a home-visiting Early Child Development (ECD) program in the Caribbean aimed at vulnerable children from birth to three years. The analysis is based on a quasi-experimental research design including approximately four hundred children in treatment and comparable control communities. The differences-in-differences methodology estimates intention-to-treat effects. One year after implementation, we find no significant effects on the cognitive development of the average child, but pronounced differences by birth cohort. The program has significantly improved Fine Motor Skills and Visual Reception scores, related to early reading and writing abilities, of the youngest children aged below 18 months at program start. There is no program impact on the older cohort, whose cognitive development appears to be more strongly correlated with center-based ECD services. Language development has not improved for either cohort. The findings suggest that an early window of opportunities may exist for home-based programs.
本文对加勒比地区针对出生至三岁弱势儿童的家访儿童早期发展（ECD）计划进行了短期影响评估。该分析基于准实验研究设计，包括治疗和可比对照社区的大约 400 名儿童。双重差分方法估计意向性治疗效果。实施一年后，我们发现对普通儿童的认知发展没有显着影响，但出生队列之间存在明显差异。该计划显著提高了 18 个月以下最小儿童在计划开始时的精细运动技能和视觉接收分数，与早期阅读和写作能力有关。该计划对老年群体没有影响，他们的认知发展似乎与基于中心的 ECD 服务更密切相关。这两个队列的语言发展都没有改善。研究结果表明，家庭计划可能存在早期机会之窗。

JEL classification JEL 分类

I29

C33

I29

C33

型

Keywords 关键字

Early childhood development

Cognitive skills

Home-based intervention

Impact evaluation

The Caribbean

儿童早期发展

认知技能

家庭干预

影响评估

加勒比地区

1. Introduction 1. 引言

A child's experiences in the first years of life lay the basis for outcomes in adolescence and adulthood (Grantham-McGregor et al., 2007, Heckman, 2006, Shonkoff and Phillips, 2000, Young, 2002). Young children who lack adequate nutrition, early cognitive stimulation, or emotional bonding may suffer developmental delays that will be more difficult to overcome as the years progress. Especially children from disadvantaged households, such as poor or lowly educated families, run the risk of falling behind their more fortunate counterparts at a young age. This might result in an ever-widening gap between poor and non-poor children.
儿童在生命最初几年的经历为青少年期和成年期的结果奠定了基础（Grantham-McGregor et al.， 2007， Heckman， 2006， Shonkoff and Phillips， 2000， Young， 2002）。缺乏足够营养、早期认知刺激或情感纽带的幼儿可能会遭受发育迟缓，随着时间的推移，这些迟缓将更难克服。特别是来自弱势家庭的儿童，如贫困或受教育程度低的家庭，有可能在很小的时候就落后于更幸运的同龄人。这可能会导致贫困儿童和非贫困儿童之间的差距不断扩大。

To prevent the perpetuation of this ‘intergenerational cycle of poverty’, early intervention programs have been set up all over the world to reach the most vulnerable children. However, rigorous evaluations of Early Childhood Development (ECD) programs in developing countries are limited, in particular of home-based interventions.² This paper provides an impact evaluation of a Caribbean home-visiting program aimed at disadvantaged children between birth and three years of age. The paper estimates the short-term program impact on cognitive child development one year after implementation.
为了防止这种“贫困的代际循环”持续存在，世界各地都设立了早期干预计划，以惠及最脆弱的儿童。然而，对发展中国家儿童早期发展（ECD）计划的严格评估是有限的，尤其是基于家庭的干预措施。² 本文对针对出生至 3 岁间弱势儿童的加勒比地区家访计划进行了影响评估。该论文估计了实施一年后短期计划对认知儿童发展的影响。

Quantitative evaluations of home-based interventions in Latin America and the Caribbean have been published for Jamaica (Powell et al., 2004, Walker et al., 2005) and Boliva (Super, Guillermo Herrera, & Mora, 1990). These studies focus specifically on undernourished children that were selected into the sample based on anthropometric indicators of stunting or general malnutrition. They find significant short-term effects of early stimulation on the overall cognitive development of children and on eye–hand coordination, but not on loco-motor development, i.e. children's increased mastery of mobility such as crawling, running, and jumping. Impact on vocabulary and speech is inconclusive. Sweet and Appelbaum (2004) conduct a meta-analysis of 60 impact studies of home-based programs in the United States. They report a significant but small average effect on cognitive development. The authors comment that the small size of the benefits may not outweigh the costs of the interventions, although impact estimates may vary across subgroups of children.
已经针对牙买加（Powell等人，2004年，Walker等人，2005年）和玻利瓦（Super，Guillermo Herrera和Mora，1990年）发布了对家庭干预的定量评估。这些研究特别关注根据发育迟缓或一般营养不良的人体测量指标被选入样本的营养不良儿童。他们发现早期刺激对儿童的整体认知发展和手眼协调有显着的短期影响，但对运动发育没有影响，即儿童对爬行、跑步和跳跃等活动能力的掌握程度提高。对词汇和语音的影响尚无定论。Sweet 和 Appelbaum （2004）对美国 60 项家庭计划的影响研究进行了荟萃分析。他们报告了对认知发展的显著但较小的平均影响。作者评论说，尽管影响估计可能因儿童亚组而异，但益处的小规模可能不会超过干预的成本。

This study evaluates the Roving Caregivers Program (RCP), a program that is widely implemented throughout the Caribbean. To our knowledge, this is one of only two existing impact evaluations of home-based interventions in the region that target the general low-income population instead of undernourished children. Powell (2004) reports on the randomized evaluation of a similar program in Jamaica.
本研究评估了 Roving Caregivers Program （RCP），该计划在整个加勒比地区广泛实施。据我们所知，这是该地区仅有的两项现有的针对一般低收入人群而不是营养不良儿童的家庭干预措施影响评估之一。Powell （2004）报告了牙买加类似计划的随机评估。

The evaluation is set up as a longitudinal, quasi-experimental study that follows almost four hundred children in St Lucia over time from the baseline survey in 2006 to the follow-up survey in 2008. The children were aged between birth and 24 months at the start of the study. Half of them lived in communities where RCP was introduced after the baseline. The other children lived in comparable communities that did not receive the intervention.
该评估是一项纵向的准实验研究，从 2006 年的基线调查到 2008 年的后续调查，对圣卢西亚的近 400 名儿童进行了跟踪调查。研究开始时，这些孩子的年龄在出生到 24 个月之间。他们中的一半生活在基线后引入 RCP 的社区。其他儿童生活在未接受干预的类似社区。

To measure program impact, the econometric analysis uses a differences-in-differences methodology with multivariate child fixed-effect regressions. The evaluation is based on an intention-to-treat approach. It focuses on a comparison of all eligible children in the intervention communities with similar children in the non-intervention communities. The analysis is complemented with a more detailed investigation of child outcomes in relation to home-based versus center-based care.
为了衡量项目影响，计量经济学分析使用具有多变量子固定效应回归的双重差分方法。评估基于意向性治疗方法。它侧重于将干预社区中所有符合条件的儿童与非干预社区中的类似儿童进行比较。该分析还补充了对家庭护理与中心护理相关的儿童结局的更详细调查。

Child development cannot be understood in isolation from its cultural context (Shonkoff & Phillips, 2000). Therefore the next section will start with a description of early childhood experiences and parenting practices in the Caribbean. It also describes the characteristics of RCP. Section 3 discusses the research design with a focus on the comparability of the treatment and the matched control group. It discusses response rates as well as the psychological assessment tool used to measure cognitive development. The econometric strategy is explained in Section 4. Section 5 shows enrollment rates in RCP and discusses targeting effectiveness. Impact results are given in Section 6. The final section concludes.
儿童发展不能脱离其文化背景来理解（Shonkoff & Phillips， 2000）。因此，下一节将首先描述加勒比地区的幼儿经历和育儿实践。它还描述了 RCP 的特征。第 3 节讨论了研究设计，重点是治疗和匹配对照组的可比性。它讨论了反应率以及用于测量认知发展的心理评估工具。计量经济学策略在第 4 节中解释。第 5 节显示了 RCP 的注册率并讨论了定位有效性。影响结果在第 6 节中给出。最后一部分结束。

2. The Roving Caregivers Program in St Lucia
2. 圣卢西亚的巡回看护人计划

2.1. Early childhood and parenting practices in the Caribbean
2.1. 加勒比地区的幼儿和育儿实践

St Lucia is an island in the Eastern Caribbean with a population of 172,000 individuals. One third of the population is aged less than eighteen years, and 8.7% are under five.³ As elsewhere in the region, many St Lucian families can be characterized as matrifocal households: female-headed family units centered on living arrangements between grandmother, mother and children without resident spouse. Males are often physically absent or play a marginal role only (Barrow, 2003). Their role is to be the financial provider, whether living in the household or not, as long as they are the boyfriend of a female household member or the father of a child in the household. However, many unwed, teenaged mothers are left to provide for their children without support from the father (Augustin, 2004).
圣卢西亚是东加勒比地区的一个岛屿，人口为 172,000 人。三分之一的人口年龄在 18 岁以下，8.7% 的人口年龄在 5 岁以下。³ 与该地区的其他地方一样，许多圣卢西亚家庭可以被归类为母系家庭：女性为户主的家庭单位以祖母、母亲和孩子之间的生活安排为中心，没有常住配偶。雄性通常身体上不存在或只扮演边缘角色（Barrow，2003 年）。他们的角色是成为经济提供者，无论是否生活在家庭中，只要他们是女性家庭成员的男朋友或家庭中孩子的父亲。然而，许多未婚的十几岁的母亲在没有父亲支持的情况下供养孩子（Augustin，2004）。

At the start of the study in 2006, 28.8% of the St Lucian population was living below the national poverty line of US$ 158.74 per month.⁴ Young children were disproportionately represented among the poor. Inadequate parenting practices are symptomatic in part of poverty, lack of education and environmental factors (Williams & Brown, 2005). Nonetheless, socio-economic characteristics are not the sole determinants of a child's development. The experiences of young children also depend to a large extent on common conceptualizations of childhood and parenting, i.e. on cultural aspects, which shape the interaction patterns between parent and child.
在 2006 年研究开始时，28.8% 的圣卢西亚人口生活在每月 158.74 美元的全国贫困线以下。⁴ 年幼儿童在穷人中的比例过高。不足的育儿实践是部分贫困、缺乏教育和环境因素的症状（Williams & Brown， 2005）。尽管如此，社会经济特征并不是儿童发展的唯一决定因素。幼儿的经历在很大程度上也取决于童年和养育子女的常见概念化，即文化方面，这些方面塑造了父母和孩子之间的互动模式。

Home-visiting programs such as RCP aim to stimulate parenting practices that enhance children's development. The sociological and anthropological literature emphasizes some salient features of Caribbean childhood that are relevant in this respect (Barrow, 2003, Roopnarine, 2005, Williams and Brown, 2005, Williams et al., 2006). Young Caribbean children receive a lot of affection. However, there is also a strong need among parents to feel in control over their children. Ideally, children should be obedient and compliant. Children's curiosity and creativity are not encouraged. This may stem in part from the fact that many parents do not see play as a learning activity. As a result, homes often lack play materials. In general, little two-way verbal communication takes place between adults and young children. The lack of stimulating interaction patterns between caregivers and children substantially reduces the opportunities for children's reasoning and language development. Books are absent in many households.
RCP 等家访计划旨在刺激促进儿童发展的育儿实践。社会学和人类学文献强调了加勒比童年的一些显著特征，这些特征与此相关（Barrow，2003 年，Roopnarine，2005 年，Williams 和 Brown，2005 年，Williams 等人，2006 年）。年轻的加勒比儿童受到了很多爱。然而，父母也强烈需要对孩子有控制权。理想情况下，孩子应该听话和顺从。孩子们的好奇心和创造力不被鼓励。这可能部分源于许多父母不将游戏视为一种学习活动的事实。因此，家里经常缺乏游戏材料。一般来说，成人和幼儿之间很少进行双向口头交流。照顾者和儿童之间缺乏刺激性的互动模式大大减少了儿童推理和语言发展的机会。许多家庭没有书籍。

Another risk factor commonly found to affect healthy child development in the Caribbean is parental stress caused by poor health, depression, isolation or limited spousal support (Samms-Vaughan, 2004). Parental stress is often related to difficulties in making ends meet. The practice of ‘child-shifting’ between relatives may cause its own problems for children's development (Barrow, 2003, Roopnarine, 2005). Such practices are often due to a parent's migration in search of employment.
在加勒比地区，另一个常见的影响儿童健康发展的风险因素是因健康状况不佳、抑郁、孤立或配偶赡养费有限而引起的父母压力（Samms-Vaughan，2004 年）。父母的压力通常与入不敷出有关。亲属之间 “儿童转移 ”的做法可能会给儿童的发展带来问题（Barrow， 2003， Roopnarine， 2005）。这种做法通常是由于父母为寻找工作而移民。

2.2. Center-based ECD interventions in St Lucia
2.2. 圣卢西亚以中心为基础的 ECD 干预

Early childhood services in the Caribbean mostly take the form of center-based facilities such as daycare centers or preschools. Notwithstanding an increasing number of facilities in St Lucia, access to appropriate ECD services remains very limited. National enrollment rates are a mere 15% among the birth to two years age cohort that is eligible for daycare facilities, and 48% of the children between three and five attend preschool (Charles, 2004). Access is especially problematic for the poorest children. Low-income mothers usually cannot afford to pay for ECD services even when these are provided in their community.⁵ However, children growing up in low-income families are precisely the ones who are most likely to benefit from ECD interventions (Engle et al., 2011, Grantham-McGregor et al., 2007).
加勒比地区的幼儿服务主要采用以中心为基础的设施形式，例如日托中心或学前班。尽管圣卢西亚的设施数量不断增加，但获得适当的 ECD 服务的机会仍然非常有限。在有资格使用日托机构的出生至两岁年龄组中，全国入学率仅为 15%，而 48% 的 3 至 5 岁儿童上学前班（Charles，2004 年）。对于最贫困的儿童来说，获得疫苗尤其成问题。低收入母亲通常负担不起儿童早期发展服务的费用，即使她们的社区提供这些服务。⁵ 然而，在低收入家庭中长大的儿童恰恰是最有可能从儿童早期发展干预中受益的人（Engle et al.， 2011， Grantham-McGregor et al.， 2007）。

Moreover, children do not always enroll in the center that is most appropriate for their age in terms of facilities, activities and training of staff. Of all children in the age group from birth to two years old who attend an ECD facility, only 48% are enrolled in a daycare center. The remaining 52% attend a preschool (Charles, 2004). An important reason thereof is the limited (geographical) access to facilities of each type.
此外，儿童并不总是在设施、活动和工作人员培训方面最适合他们年龄的中心注册。在参加 ECD 设施的从出生到两岁的所有年龄段的儿童中，只有 48% 在日托中心注册。其余 52% 的人在学前班上学（Charles，2004 年）。其中一个重要原因是对每种类型设施的（地理）访问有限。

2.3. Description of the Roving Caregivers Program
2.3. 巡回看护者计划的描述

RCP targets vulnerable families with children in the age group from birth to 36 months of age. It aims to change parenting practices in order to enhance healthy child development. The program has been implemented in Jamaica since 1993. Since 2002, it has been extended to Belize as well as four Eastern Caribbean islands: St Lucia, St Vincent and the Grenadines, Grenada and Dominica. RCP was introduced in the North and East of St Lucia in 2004. The impact study concentrates on the program's expansion to the South and the West of the island in January 2007.
RCP 针对有出生至 36 个月年龄段儿童的弱势家庭。它旨在改变育儿方式，以促进儿童的健康发展。该计划自 1993 年以来一直在牙买加实施。自 2002 年以来，它已扩展到伯利兹以及四个东加勒比岛屿：圣卢西亚、圣文森特和格林纳丁斯、格林纳达和多米尼加。RCP 于 2004 年在圣卢西亚北部和东部推出。影响研究的重点是该计划于 2007 年 1 月扩展到该岛的南部和西部。

The program works with paraprofessional facilitators, so-called “Rovers”, who are living in the communities and who receive an intensive introductory training with regular follow-up. Twice per week for a maximum period of three years, the Rover visits the home of the child for 45 min. In the presence of the child's caregiver the Rover engages in age-appropriate stimulating activities with the child through play, such as singing songs or playing with blocks, shapes and colors. The caregiver is required to be present during the home visits, and expected to observe and join in the activities with the explicit aim of continuing the stimulating interaction on a daily basis also when the Rover is not present. The Rover explains to the caregiver how certain activities enhance cognitive development and discusses related issues such as disciplining practices.
该计划与生活在社区的辅助专业辅导员（即所谓的“漫游者”）合作，他们接受强化的入门培训并定期跟进。每周两次，最长为期三年，漫游者访问孩子的家 45 分钟。在儿童看护人在场的情况下，Rover 通过游戏与儿童进行适合年龄的刺激活动，例如唱歌或玩积木、形状和颜色。看护人在家访期间必须在场，并期望观察并参与活动，其明确目标是在漫游者不在场时每天继续刺激互动。漫游者向看护者解释某些活动如何促进认知发展，并讨论相关问题，例如管教做法。

An additional component of RCP are the monthly parenting meetings in local community centers, which are open to all caregivers of young children in the community. During these interactive group meetings, a variety of topics are introduced and discussed such as nutrition and eating habits, self-made educational toys, or cognitive milestones.
RCP 的另一个组成部分是每月在当地社区中心举行的育儿会议，这些会议向社区中所有幼儿的照顾者开放。在这些互动小组会议期间，会介绍和讨论各种主题，例如营养和饮食习惯、自制益智玩具或认知里程碑。

To reach the most vulnerable children, RCP focuses on communities with above-average poverty rates. Within program communities, all children in the eligible age range can join. The objective of RCP is to fill in the gap caused by a lack of access to other ECD services aimed at young children, not to substitute for them. Access can be limited either because ECD services are absent in the community, or because they are too expensive. Once the child enrolls in daycare or preschool, he or she is supposed to exit the program. At 36 months, the child graduates from RCP.
为了帮助最脆弱的儿童，RCP 专注于贫困率高于平均水平的社区。在计划社区中，符合条件的年龄段的所有儿童都可以加入。RCP 的目标是填补因无法获得针对幼儿的其他 ECD 服务而造成的空白，而不是替代他们。访问可能会受到限制，要么是因为社区中没有 ECD 服务，要么是因为它们太贵了。一旦孩子进入日托或学前班，他或她就应该退出该计划。在 36 个月大时，孩子从 RCP 毕业。

3. Research methodology 3. 研究方法

3.1. Evaluation design 3.1. 评估设计

In the absence of a randomized evaluation design, this short-term impact evaluation is set up as a quasi-experimental study that follows children in a selection of fifteen communities over time. The communities are located in two districts that were designated for expansion of the program: the southern Vieux-Fort district and the western Anse-la-Raye district. Half of the communities were assigned to the program group in which RCP was introduced after the baseline survey in 2006. The remaining seven communities are included in the study as control. The research team together with the implementers explored possibilities of random assignment of children within communities to a treatment and control group. However, given the rural nature of the villages, where all inhabitants know each other, it was expected that such a research design would create tensions within communities and encounter strong resistance. Therefore it was decided to assign treatment at the level of the community instead of the individual child.
在没有随机评估设计的情况下，这种短期影响评估被设置为一项准实验研究，随着时间的推移跟踪 15 个社区中的儿童。这些社区位于两个被指定扩展该计划的地区：南部的 Vieux-Fort 区和西部的 Anse-la-Raye 区。在 2006 年基线调查后，一半的社区被分配到引入 RCP 的项目组。其余 7 个社区作为对照纳入研究。研究小组与实施者一起探索了将社区内的儿童随机分配到治疗组和对照组的可能性。然而，鉴于村庄的乡村性质，所有居民都彼此认识，预计这样的研究设计会在社区内部造成紧张局势并遇到强烈的阻力。因此，决定在社区层面分配治疗，而不是在儿童个体层面进行治疗。

The first stage of the study consisted of the selection of villages into the study sample. Sixteen potential intervention communities were selected from the full list of communities in the two districts based on two main RCP selection criteria: a high poverty rate and a large number of young children. In both districts, the average poverty index was calculated using the official community-level poverty rates.⁶ Precise numbers on children aged between birth and three were not available. As a proxy, the eight largest communities below the poverty line in each district were included in the sample using the population totals from the latest Census 2001. The third official RCP selection criterion – lack of ECD facilities in the community, was not used for the selection process. Enrollment in center-based care was that low in the South and West region, that RCP management did not consider the actual presence of an ECD facility as a good indicator of accessibility or a relevant criterion for the exclusion of communities from their program. The final number of sampled communities was fifteen instead of sixteen because two of the selected communities turned out to be administratively separate but fully merged in practice.
研究的第一阶段包括选择村庄进入研究样本。根据两个主要的 RCP 选择标准，从两个地区的社区完整名单中选择了 16 个潜在的干预社区：高贫困率和大量幼儿。在这两个地区，平均贫困指数都是使用官方社区级贫困率计算的。⁶ 没有出生至三岁儿童的确切数字。作为代理，使用 2001 年最新人口普查的人口总数，将每个地区贫困线以下的八个最大社区包括在样本中。第三个官方 RCP 选择标准 – 社区缺乏 ECD 设施，未用于选择过程。南部和西部地区以中心为基础的护理的入学率非常低，以至于 RCP 管理层没有将 ECD 设施的实际存在视为可访问性的良好指标或将社区排除在其计划之外的相关标准。最终抽样社区的数量是 15 个而不是 16 个，因为选定的社区中有两个在行政上是独立的，但在实践中完全合并。

The second stage concerned the assignment of the selected communities to a treatment and a control group. Communities were assigned to a group such that: (a) in each district there were four program communities and four control communities, and (b) comparability of the program group and the control group was maximized with respect to the RCP selection criteria, weighted for population size. The two groups were subsequently compared on a large number of other community characteristics to verify that they were also similar in other respects: presence of community facilities (e.g. health centers, daycare centers, shops, public transport), the prevalence of social and environmental problems (crime, alcoholism, drugs, prostitution, flooding, lack of sanitation), the number of community organizations (such as women's groups, credit groups, sports clubs) and geographic characteristics (such as distance to main road). Appendix A provides detailed data on the matching criteria and the additional community characteristics.
第二阶段涉及将选定的社区分配到治疗组和对照组。将社区分配到一个组，以便：（a）每个区有四个项目社区和四个控制社区，以及（b）根据 RCP 选择标准（根据人口规模加权），项目组和对照组的可比性最大化。随后，根据大量其他社区特征对这两个群体进行了比较，以验证它们在其他方面也相似：社区设施的存在（如保健中心、日托中心、商店、公共交通）、社会和环境问题的普遍性（犯罪、酗酒、毒品、卖淫、洪水、缺乏卫生设施）、社区组织的数量（如妇女团体、信用组、体育俱乐部）和地理特征（例如到主干道的距离）。附录 A 提供了有关匹配标准和其他社区特征的详细数据。

As a complication to the evaluation design, four of the communities had been promised the program at an earlier consultation stage (Augustin, 2004). These were the two most populous communities in each district. They were included in the treatment group. To avoid confounding effects due to discrepancies in village population size, we included for each district an additional large and equally poor community that was located geographically close but just outside the administrative boundaries. A regression of baseline child outcomes on a dummy variable for prior earmarking as well as a number of control variables indicates that there were no significant differences in child outcomes between the four earmarked communities and the rest of the communities prior to program introduction.
作为评估设计的复杂化，其中四个社区在早期咨询阶段就被承诺提供该计划（Augustin，2004 年）。这是每个地区人口最多的两个社区。他们被纳入治疗组。为了避免由于村庄人口规模的差异而造成的混杂效应，我们为每个地区增加了一个额外的大型且同样贫穷的社区，该社区位于地理上很近，但刚好在行政边界之外。先前指定用途的虚拟变量以及一些控制变量的基线儿童结果回归表明，四个指定用途社区与计划引入前的其他社区之间的儿童结果没有显着差异。

3.2. Data collection and measurement instruments
3.2. 数据收集和测量仪器

The panel dataset is based on two surveys: a baseline survey in the second half of 2006 and a follow-up survey early 2008. Each survey round consisted of two components. The first was a structured interview with the child's primary caregiver conducted by interviewers from the St Lucia Statistical Office. The second component was a child assessment at the local health center, including cognitive and socio-emotional development assessments as well as a measurement of anthropometrics, conducted by trained nurses. Participants received a reimbursement for their transportation costs to the health centers as well as a small cash gift to buy snacks and a food hamper as a token of appreciation.⁷
面板数据集基于两项调查：2006 年下半年的基线调查和 2008 年初的后续调查。每轮调查由两个部分组成。第一个是由圣卢西亚统计局的访谈员对孩子的主要照顾者进行的结构化访谈。第二个组成部分是在当地卫生中心进行的儿童评估，包括认知和社会情感发展评估以及人体测量学测量，由训练有素的护士进行。参与者收到了前往健康中心的交通费用报销，以及用于购买零食和食品篮的小额现金礼物，以示感谢。⁷

To measure children's cognitive ability and loco-motor development, the study used the Mullen Scales of Early Learning (Mullen, 1995). This test is a widely used, individually administered, comprehensive measure of cognitive functioning. It assesses the child's visual, hearing, speech and motor skills, and distinguishes between receptive and expressive processing. It is an age-standardized tool based on a North-American reference population for children from birth through 68 months.⁸
为了测量儿童的认知能力和运动发展，该研究使用了 Mullen 早期学习量表（Mullen， 1995）。该测试是一种广泛使用的、单独管理的、全面的认知功能测量方法。它评估孩子的视觉、听觉、言语和运动技能，并区分接受和表达处理。它是一种基于北美儿童参考人群的年龄标准化工具，适用于从出生到 68 个月的儿童。⁸

The Mullen Scales of Early Learning provide normative scores for five scales as well as a single composite summary score. Each scale has a standardized average score of 50 points and a standard deviation of 10 points. The average of the standardized summary score is 100 points with a standard deviation of 15 points.
Mullen 早期学习量表提供五个量表的标准分数以及一个综合总分。每个量表的标准化平均分为 50 分，标准差为 10 分。标准化总分的平均分是 100 分，标准差为 15 分。

The Visual Reception and Fine Motor scales reflect cognitive abilities that are important for a smooth transition to a school setting. The Visual Reception scale tests a child's performance in processing visual patterns. As they develop, children become able to discriminate and memorize in growing detail the spatial characteristics of drawings, pictures and geometric forms as well as printed letters and words (Mullen, 1995, p. 17). These abilities are at the basis of reading readiness (through letter recognition) and are a prerequisite for Fine Motor Skills and eye–hand coordination.
视觉接收量表和精细运动量表反映了认知能力，这些能力对于顺利过渡到学校环境很重要。视觉接收量表测试儿童在处理视觉模式方面的表现。随着他们的成长，孩子们能够越来越详细地区分和记住图画、图片和几何形式以及印刷字母和文字的空间特征（Mullen，1995 年，第 17 页）。这些能力是阅读准备（通过字母识别）的基础，是精细运动技能和手眼协调的先决条件。

The Fine Motor scale measures visual-motor ability. It reflects the expressive side of visual organization and discrimination. The primary ability areas covered in the assessment are Fine Motor control and writing readiness. Fine Motor control at a young age lays the foundation for early drawing and writing skills from the age of 3 onwards. During this period, children become more skilled at pencil and paper tasks (e.g. copying geometric figures or letter patterns), and at such tasks as using scissors, folding paper, and the independent but coordinated use of fingers (Mullen, 1995, p. 17).
精细运动量表测量视觉运动能力。它反映了视觉组织和辨别的表现力。评估涵盖的主要能力领域是精细运动控制和写作准备。从小就掌握精细运动为从 3 岁开始的早期绘画和书写技能奠定了基础。在此期间，孩子们更擅长用铅笔和纸张（例如复制几何图形或字母图案），以及使用剪刀、折叠纸张以及独立但协调地使用手指等任务（Mullen，1995 年，第 17 页）。

The Gross Motor scale assesses such skills as head control, sitting, and walking, i.e. strength, control and balance in the major muscle groups. Although they are not a direct reflection of cognitive development, motor control and mobility play an important role in the development of cognitive abilities, as they enable the child to experience and experiment with new schemas, objects and people (Mullen, 1995, pp. 10–11). This scale is measured only up to the age of 36 months.
粗大运动量表评估头部控制、坐姿和行走等技能，即主要肌肉群的力量、控制和平衡。虽然它们不是认知发展的直接反映，但运动控制和移动性在认知能力的发展中起着重要作用，因为它们使孩子能够体验和试验新的图式、物体和人（Mullen，1995 年，第 10-11 页）。该量表仅在 36 个月之前测量。

The Receptive Language scale measures a child's ability to process linguistic input, i.e. auditory comprehension and auditory memory. The scale assesses how well the child is able to decode and understand verbal input. The Expressive Language scale measures a child's ability to use language productively, in particular the child's speaking ability and language formation, including the ability to verbalize concepts (Mullen, 1995, p. 11).
接受性语言量表衡量儿童处理语言输入的能力，即听觉理解和听觉记忆。该量表评估孩子解码和理解口头输入的能力。表达性语言量表衡量儿童有效使用语言的能力，特别是儿童的口语能力和语言形成，包括用语言表达概念的能力（Mullen，1995，第 11 页）。

3.3. Survey response rates
3.3. 调查回复率

The sample consisted of a complete census of children in the fifteen study communities who were aged between 0 and 24 months at baseline. Overall, 461 children born between July 1st 2004 and July 1st 2006 were identified; 229 in the program group and 232 in the control group (Table 1). Virtually all identified caregivers participated in the household interview. In addition, 419 children also participated in the baseline child assessment, at an overall participation rate of 90.1%.
该样本包括 15 个研究社区中基线年龄在 0 至 24 个月之间的儿童的完整普查。总体而言，确定了 2004 年 7 月 1 日至 2006 年 7 月 1 日期间出生的 461 名儿童;程序组 229 例，对照组 232 例（表 1）。几乎所有确定的护理人员都参与了家庭访谈。此外，419 名儿童还参与了基线儿童评估，总体参与率为 90.1%。

Table 1. Overview of the children in the sample.
表 1.样本中的子项概述。

Empty Cell	Baseline 2006 基线 2006			Follow-up 2008 2008 年后续行动			Balanced 2006–2008 2006-2008 年平衡
Empty Cell	Total 总	Control 控制	Treat 治疗	Total 总	Control 控制	Treat 治疗	Total 总	Control 控制	Treat 治疗
Total # of children on list 列表中的子项总数 #	461	232	229	410^a ^一个 410	217	193	461	232	229
# Children who participated in cognitive test # 参加认知测试的儿童	419	222	197	391	207	184	389	207	182
Participation rate in cognitive test (%) 认知测试参与率（%）	90.9	95.7	86.0	95.4	95.4	95.3	84.4	89.2	79.5

a: The follow-up child list in 2008 only includes children who participated in the cognitive test at baseline and whose caregiver participated in the caregiver interview at baseline.
2008 年的随访儿童名单仅包括基线时参加认知测试的儿童及其照顾者在基线时参加照顾者访谈的儿童。

Baseline participation rates were higher in the control communities at 95.7% than in the program communities at 86.0%. At baseline, the survey in treatment communities was introduced in conjunction with RCP. Some eligible respondents in treatment communities may have been hesitant to come to the child assessment, for example if they misunderstood that this would also commit them to participating in RCP. In control communities, the link with RCP was mentioned as well but may have featured less prominently in individuals’ decision to participate in the study. It is not clear a priori how this would affect the composition of the sample in treatment communities. Caregivers least willing to enroll in RCP may have been either less aware of the importance of ECD and also invest less in their child's development, or instead may have felt sufficiently equipped (in terms of financial resources, educational background, information) to optimally support their child's development without need for additional services. To filter out time-invariant differences in parenting skills and awareness, and limit potential participation bias, all analyses will include child-fixed effects.
对照社区的基线参与率为 95.7%，高于计划社区的 86.0%。在基线时，治疗社区的调查与 RCP 一起引入。治疗社区中一些符合条件的受访者可能对参加儿童评估犹豫不决，例如，如果他们误解了这也将使他们参与 RCP。在对照社区中，也提到了与 RCP 的联系，但在个体参与研究的决定中可能不太突出。目前尚不清楚这将如何影响治疗社区的样本组成。最不愿意参加 RCP 的照护者可能不太意识到 ECD 的重要性，并且对孩子的发展投入较少，或者可能觉得有足够的能力（在财务资源、教育背景、信息方面）可以最佳地支持孩子的发展，而无需额外的服务。为了过滤掉育儿技能和意识的时间不变差异，并限制潜在的参与偏倚，所有分析都将包括儿童固定效应。

Only the 419 children who were assessed at baseline were revisited in 2008. Of them, 410 were tracked at follow-up (97.9%). Nine children, all living in treatment communities, had migrated and could not be found. Participation rates among those tracked in the follow-up survey were very high in both treatment and control communities at 95.4% on average. The analysis focuses on the 389 children with child assessments and household information in both the baseline and the follow-up survey. They represent 84.4% of the originally identified children.⁹
2008 年仅对基线评估的 419 名儿童进行了复访。其中，410 例在随访中进行了跟踪（97.9%）。9 名儿童都生活在治疗社区，他们已经迁移，无法找到。在随访调查中跟踪的人中，治疗和对照社区的参与率都非常高，平均为 95.4%。该分析侧重于基线和后续调查中 389 名儿童的儿童评估和家庭信息。他们占最初确定的儿童的 84.4%。⁹

3.4. Comparison of the treatment and control group at baseline
3.4. 基线治疗组和对照组的比较

This section investigates the comparability of the two treatment groups. Standard errors are clustered at the community level. Table 2 panel A shows that there were more girls in the treatment group, significant at the 10% error level. There were no baseline differences between the treatment and the control group with respect to the children's average age in months. The percentage of children attending center-based daycare at baseline was also similar in the two groups and much lower at an average of 2.9% than the national average of 15%. Household demographics (household size, number of children, sex of the head of household, age of the caregiver) were very comparable. There was a statistically significant difference in socio-economic characteristics. In particular, caregivers in the control group were more likely to be employed and their wealth index was higher, although the absolute size of the wealth difference was limited.¹⁰ The regressions will control for these variables. Differences across the two groups in terms of caregiver education were not significant.
本节研究了两个治疗组的可比性。标准错误聚集在社区级别。表 2 面板 A 显示治疗组中的女孩更多，在 10% 的误差水平上显着。治疗组和对照组在儿童的平均年龄（以月为单位）方面没有基线差异。两组基线时参加中心日托的儿童百分比也相似，平均为 2.9%，远低于全国平均水平 15%。家庭人口统计数据（家庭规模、孩子数量、户主性别、照顾者的年龄）非常相似。社会经济特征存在统计学上的显著差异。特别是，对照组的护理人员更有可能就业，他们的财富指数更高，尽管财富差异的绝对大小有限。¹⁰ 回归将控制这些变量。两组在照顾者教育方面的差异不显著。

Table 2. Baseline comparison of the treatment group and the control group.
表 2.治疗组与对照组的基线比较。

Empty Cell	Total # obs 总计 # 个	Total mean 总均值	Total min 最小合计	Total max 最大总计	Total s.d. 总计 n.d.	Control mean 控制均值	Treat mean 治疗均值	p-Value^a p 值^a
Panel A. Child and household characteristics at baseline 图 A. 基线时的儿童和家庭特征
Child characteristics 儿童特征
Sex (% female) 性别（女性百分比）	389	54.0	0.0	100.0	49.9	50.2	58.2	.059^* 0.059^*
Age (in months) 年龄（月）	389	15.7	1.1	38.1	8.2	15.6	15.9	.486
Enrolled in center-based care (%) 参加以中心为基础的护理（%）	377	2.9	0.0	100.0	16.9	3.4	2.3	.675
Household characteristics 住户特征
Household size 家庭人数	382	5.9	2.0	16.0	2.5	6.0	5.8	.269
# of children 0–17 # 0-17 岁儿童	382	3.1	1.0	9.0	1.7	3.1	3.2	.586
Female head of household (%) 女性户主（%）	377	42.4	0.0	100.0	49.5	44.6	40.0	.415
Age of caregiver 照护者的年龄	372	27.8	6.0	62.0	7.5	28.2	27.2	.129
Years of education of caregiver	365	10.9	0.0	25.0	4.0	11.1	10.6	.375
Employment of caregiver (%)	377	34.2	0.0	100.0	47.5	37.9	29.9	.024^**
Wealth index	380	0.0	−2.2	3.5	1.0	0.1	−0.2	.077^*
Panel B. Cognitive child outcomes at baseline
Summary score	389	106.2	49.0	155.0	16.1	105.9	106.5	.813
Gross Motor	377	59.1	21.0	80.0	11.0	58.9	59.3	.684
Visual Reception	389	54.3	20.0	80.0	10.3	54.7	53.9	.487
Fine Motor	389	51.3	20.0	80.0	11.6	50.7	52.0	.466
Receptive Language	389	52.8	20.0	80.0	10.9	52.8	52.8	.978
Expressive Language	389	53.7	20.0	80.0	10.6	53.2	54.2	.514

Empty Cell	Without controls (1)		With controls^b (2)
Empty Cell	Coef.	s.e.	Coef.	s.e.
Panel C. Regression analysis of baseline differences in cognitive child outcomes^a
Summary score	0.65	(2.71)	−0.54	(2.83)
Gross Motor	0.40	(0.97)	−0.06	(1.28)
Visual Reception	−0.81	(1.13)	−1.54	(1.37)
Fine Motor	1.29	(1.72)	−0.33	(1.90)
Receptive Language	−0.05	(1.59)	−0.14	(1.57)
Expressive Language	1.03	(1.54)	0.86	(1.49)

a: Standard errors are clustered at the community level.
b: Included control variables are child sex, child age in months, number of children aged 0–17, gender of household head, education and employment of caregiver, household wealth index, district code, tester id-codes.
*: p-Value < .100.
**: p-Value < .050.

Panel B compares cognitive child outcomes at baseline. As the descriptive statistics show, baseline child development status was highly comparable across the treatment and the control group, despite the imbalance in socio-economic indicators. The average scores were high compared to the reference population due to the fact that some items of the tool were adjusted during the nurse training in order to better reflect local circumstances. Moreover, young Caribbean children generally outperform North-American children in terms of Gross Motor Skills (e.g. Samms-Vaughan, 2004). In fact, 21 children (5.6%) reached a ceiling score on the Gross Motor scale at baseline, reducing the interpretability of this particular scale for the St Lucian context (Paxson & Schady, 2007). Ceiling scores on the other scales were much less common with frequencies between 1 (0.3%) and 9 (2.3%) out of 389 observations per round.

To further analyze the comparability of the treatment and the control group, the child development outcomes at baseline are regressed on a dummy variable indicating whether the child lived in a treated or a non-treated community (Table 2 panel C), without control variables in column (1), equivalent to panel B, and with controls in column (2). The coefficient on the RCP community variable is not significantly different from zero in either specification and for any of the scales, providing further evidence of the comparability of the two groups.

4. Econometric methodology

To estimate the impact of RCP on child outcomes, we use a differences-in-differences estimator. This estimator compares changes in outcomes over time in the treatment group with the changes over that same period of time in the control group. It includes child fixed-effects to eliminate any unobserved time-invariant heterogeneity between the two groups. This method yields consistent impact estimates of RCP as long as unobserved individual differences between the two groups (such as innate ability or parenting skills) did not differentially change over time independent of RCP, as long as external factors affected both groups to the same extent (such as price rises, fluctuating employment rates, hurricanes, fertility trends), and new policies and interventions aimed at young children were not systematically favoring the treatment group over the control group or vice versa. It is impossible to completely rule out the presence of unobserved time-varying differences. However, the geographical proximity of the communities, and their similarities in key community characteristics provide support for the assumption of parallel trends. The comparability of child outcomes at baseline is also reassuring, because the outcomes reflect the cumulative effect of both observed and unobserved environmental factors prior to baseline that may have been of influence on child development.

To estimate program impact we estimate the following panel regression equation:

Y_{i t} = α + β P_{i} + γ P_{i} \times T_{t} + δ T_{t} + θ X_{i t} + η_{i t}

where Y_it measures child outcome Y for child i at time t (with t = 0 for the baseline in 2006 and t = 1 for the impact round in 2008). P_i is a dummy variable equal to 1 if child i lives in a community assigned to the program and 0 if the child lives in a community assigned to the control group. The coefficient β captures any baseline differences between the program and the control communities. Given the comparability of treatment and control communities, we expect this variable not to be significantly different from zero.¹¹ T_t is a dummy variable equal to 0 for the baseline survey and equal to 1 for the follow-up survey in 2008. The coefficient δ captures any time trends in child outcomes and environmental factors that are common to the treatment and the control group, e.g. general economic trends in St Lucia, the aging of the children, the changing team of child testers.

The term P_i × T_t captures program introduction after the baseline survey. It is equal to 0 for all children in 2006 as well as for control children in 2008; and equal to 1 for children living in treatment communities in 2008. In other words, the coefficient γ measures the impact of RCP on children living in a community where RCP was implemented after the baseline. It compares changes in child outcomes from baseline to follow-up for children in RCP communities with the changes in outcomes for children in the control group. To analyze treatment heterogeneity, impact will not only be estimated for the full sample but also by gender and by birth cohort.

The coefficient γ is an “intention-to-treat” (ITT) impact estimator. It measures the effect of being offered RCP services irrespective of whether the child actually participated in RCP or not. It represents a lower bound on the average treatment effect on the treated (ATET) on children who actually enrolled in the program. Given the voluntary nature of participation in RCP and in the absence of instrumental variables, the impact of the program on participants cannot directly be assessed. A comparison of participants in treatment communities with either the population in control communities or with non-participants in treatment communities would yield biased impact estimates since the subsamples are likely to differ systematically in unobserved characteristics that directly affect children's developmental paths over time. Unfortunately, there are no adequate instruments available to plausibly correct for selection bias.

X_it is a vector of explanatory variables that measure individual child and household characteristics for child i at time t. It includes child age in months, employment status of the primary caregiver, and a measure of household wealth.¹² The analysis also includes categorical variables identifying the nurses who tested the children. Despite intensive training and regular consistency checks across testers, some of them gave systematically higher or lower scores than the others. Since testers were randomly assigned to children within districts, it is unlikely that these deviations were caused by real differences in child developmental status. Rather, they must have been related to tester differences, for example in the interpersonal contact with the child. Interviewer effects are common, as has been shown for other types of large-scale surveys,¹³ but they are often ignored in empirical analyses and rarely explicitly discussed. Each tester was purposively assigned to an approximately equal number of treatment and control children to avoid multicollinearity with the treatment indicator. This allows filtering out systematic level effects across testers through the inclusion of tester identifiers in the regressions, in line with e.g. Powell et al. (2004).

Finally,

η_{i t}

is an unobserved component for child i at time t. The analysis assumes that

η_{i t} = μ_{i} + ε_{i t}

, i.e. that it consists of a fixed component

μ_{i}

that captures child-specific effects that do not vary over time (such as innate ability, or time-invariant household characteristics and parenting abilities) and a random term

ε_{i t}

. Standard errors are clustered at the community level to account for community heterogeneity. This is equivalent to a Hierarchical Linear Model with nests at the community level to capture community random effects.

5. Enrollment in RCP after baseline

The RCP program was introduced in the treatment communities in January 2007. At that time, children in the study were between 6 and 30 months of age. The balanced sample in treatment communities consists of 182 children. The majority of them (142, or 78%) participated in RCP (Table 3 panel A). Twenty two percent of caregivers in the RCP communities did not enroll their child in the program. In total 68 children (37.4%) were still enrolled at the time of the follow-up survey in 2008.¹⁴ Twenty-seven children had been enrolled in RCP but had reached the age limit and graduated from the program. Approximately one quarter had dropped out of RCP before reaching 36 months of age.

Table 3. Enrollment rates in RCP among eligible children in RCP communities.

Ever enrolled in RCP?	#	%	Age at enrollment			Length of enrollment
Empty Cell	Empty Cell	Empty Cell	Mean	Min.	Max.	Mean	Min.	Max.
Panel A. Total population
Yes	142	78.0	17.6	5.8	33.4	13.2	0.0	19.4
Yes, currently enrolled	68	37.4	15.8	6.1	33.4	15.6	4.1	19.4
Yes, but graduated	27	14.8	24.9	7.6	32.9	11.0	2.4	17.4
Yes, but dropped out	47	25.8	16.2	5.8	27.0	10.5	0.0	17.2
No	40	22.0
Total	182
Panel B. Youngest birth cohort only
Yes	83	86.5	12.6	5.8	24.6	13.8	0.0	19.4
Yes, currently enrolled	48	50.0	12.6	6.1	24.6	15.5	4.1	19.4
Yes, but graduated	4	4.2	13.0	7.6	15.7	10.3	10.1	10.5
Yes, but dropped out	31	32.3	12.6	5.8	17.9	11.3	0.0	17.2
No	13	13.5
Total	96
Panel C. Oldest birth cohort only
Yes	59	68.6	24.8	17.3	33.4	12.2	2.4	19.4
Yes, currently enrolled	20	23.3	23.6	17.4	33.4	15.7	12.1	19.4
Yes, but graduated	23	26.7	27.0	19.6	32.9	11.1	2.4	17.4
Yes, but dropped out	16	18.6	23.1	17.3	27.0	8.8	4.0	13.9
No	27	31.4
Total	86

The average age at enrollment for children in the study was 17.6 months, while the average length of enrollment at the time of follow-up was slightly more than a year (13.2 months). Children who graduated from RCP because they reached the age limit of 36 months, had been participating for 11 months on average, while children who dropped out before reaching 36 months had been enrolled for 10.5 months.

Column (1) in Table 4 takes a closer look at the probability that eligible children in treatment communities enrolled in the program. Standard errors are not clustered in this specification to allow for the simultaneous inclusion of all control variables. Clustering and entering variables in subsets yields similar results. Girls were equally likely to enroll in RCP as boys. Children who were younger at the start of the study were significantly more likely to enroll. Each additional month of age at baseline decreased the probability of participation with 0.9%. There was no significant relationship between the likelihood of participation in RCP and household size, number of children in the household or the gender of the household head.

Table 4. Probability of enrollment in ECD programs in 2008.

Dependent variable	Likelihood of participation in RCP	Likelihood of enrollment in center-based care
Sample	RCP communities only (1)	Total sample (2)
Child lives in an RCP community		−0.062
		(0.059)
Sex (male = 0, female = 1)	0.039	0.038
	(0.068)	(0.058)
Age in months (2006)	−0.009^*	0.023^***
	(0.005)	(0.004)
# of children in the household (2006)	0.003	−0.032
	(0.037)	(0.032)
Household size (2006)	0.005	−0.004
	(0.023)	(0.021)
Female head of household (2006)	−0.036	0.158^***
	(0.076)	(0.059)
Age of the caregiver (2006)	0.007	−0.001
	(0.004)	(0.004)
Years of education of caregiver (2006)	0.003	−0.001
	(0.009)	(0.008)
Employment of caregiver (2006)	−0.164^**	0.133^**
	(0.082)	(0.062)
Wealth indicator (2006)	−0.058	0.091^**
	(0.044)	(0.037)
Vieux-Fort district	0.162^**	−0.314^***
	(0.075)	(0.068)

Gross Motor Skills (2006)	0.000	−0.002
	(0.004)	(0.003)
Visual Reception (2006)	0.006	0.001
	(0.004)	(0.004)
Fine Motor Skills (2006)	0.004	−0.002
	(0.004)	(0.003)
Receptive Language (2006)	−0.006^*	0.005^*
	(0.004)	(0.003)
Expressive Language (2006)	0.000	0.001
	(0.004)	(0.003)

Joint test of child outcomes: Chi²-stat (p-value)	7.17 (.209)	5.08 (.406)

The reported coefficients are marginal effects (i.e. the percentage change in probability related to a one point change at the mean of each explanatory variable). Robust standard errors are given in parentheses.

*: p-Value < .100.
**: p-Value < .050.
^***p-Value < .010.

Caregivers who were employed were significantly less likely to enroll their child in RCP. This is in line with the program requirement that a caregiver should be present during the Rover visit. Age, education level of the caregiver, and household wealth were not significantly correlated with RCP participation. The lack of a significant correlation with education or wealth is a reflection of program implementation. RCP selects poor communities but it does not specifically target the children most at risk within the selected communities.

The regression also includes cognitive development status at baseline in 2006. This allows examining whether RCP attracted especially those children who started off with relatively high or instead low scores. As the results show, Visual Reception was positively correlated with subsequent enrollment in RCP, suggesting that participants may have been slightly better off than non-participants. None of the other baseline cognitive skills were statistically significant, either individually or jointly.

6. Impact results

6.1. Intention-to-treat effects

The impact results of the child fixed-effects regressions are given in Table 5. The table only shows the coefficients for the treatment variable. Panel (a) shows the main impact estimates for the full sample, respectively without controls¹⁵ and with controls. The impact estimates are small compared to the standardized average scores of 50 for the individual scales and 100 for the composite summary scale. The estimates range from minus 2.20 points on the Expressive Language scale to plus 1.27 points on the Receptive Language scale. Except for one, none of the impact estimates is statistically significant. The coefficient on the Expressive Language score becomes significant at the 10% level once controls are included.

Table 5. Impact of RCP on cognitive child development.

Changes in child development outcomes	Summary score	Gross Motor	Visual Reception	Fine Motor	Receptive Language	Expressive Language
(a) Total population
RCP community (without controls)	−1.04	−1.26	0.16	−0.22	0.00	−1.98
	(1.91)	(0.95)	(1.11)	(1.61)	(1.31)	(1.15)
RCP community (with controls)	−0.39	−0.79	0.31	−0.03	1.27	−2.20^*
	(2.07)	(0.77)	(1.14)	(1.68)	(1.42)	(1.17)
(b) Girls only
RCP community (without controls)	−1.22	−0.32	0.99	1.03	−0.87	−3.33^*
	(2.52)	(1.85)	(1.35)	(2.23)	(1.27)	(1.71)
RCP community (with controls)	−0.76	1.50	1.09	1.02	0.43	−3.70^**
	(2.53)	(1.86)	(1.39)	(2.31)	(1.51)	(1.51)
(c) Boys only
RCP community (without controls)	−1.31	−1.56	−1.39	−2.61	1.42	−0.26
	(2.17)	(1.78)	(1.53)	(1.60)	(1.96)	(1.55)
RCP community (with controls)	−0.29	−2.01	−1.20	−1.92	2.65	−0.27
	(2.17)	(1.48)	(1.64)	(1.59)	(1.91)	(1.69)
(d) Youngest birth cohort only
RCP community (without controls)	1.50	−0.44	3.14	2.52	−0.14	−2.39^*
	(2.62)	(0.84)	(2.16)	(1.75)	(1.46)	(1.32)
RCP community (with controls)	2.66	0.26	3.60^*	3.82^**	1.06	−2.91^**
	(2.05)	(0.67)	(1.90)	(1.42)	(1.57)	(1.23)
(e) Oldest birth cohort only
RCP community (without controls)	−0.56		−1.17	−0.72	0.84	0.01
	(3.07)		(1.60)	(2.38)	(2.21)	(1.97)
RCP community (with controls)	0.58		−0.86	−0.26	2.10	0.40
	(3.00)		(1.60)	(2.39)	(2.33)	(1.91)

*: p-Value < .100.
**: p-Value < .050.

The inclusion of control variables has a modest but discernible effect on the sizes of the impact coefficients, although most unadjusted and adjusted estimates are of the same order of magnitude. This indicates that some socio-economic differences between the treatment and control group may exist that should be controlled for, despite the sample being well-balanced at baseline in terms of child outcomes, and the majority of household characteristics.

6.2. Treatment heterogeneity by child sex and age

The subsequent panels in Table 5 examine potential treatment heterogeneity by sex and by age of the child, without and with control variables. They show estimates of the fixed-effects model for four different subsamples: (b) girls, (c) boys, (d) youngest birth cohort, and (e) oldest birth cohort. The younger cohort is defined as the group of children who were between 0 and 12 months old on July 1st 2006. They were between 6 and 18 months old at program start in January 2007. The older cohort includes the children who were between 18 and 30 months at program start. The estimations compare children of the respective gender/birth cohorts in RCP communities with children of similar gender/birth cohort in control communities. The impact on Gross Motor Skills is not calculated for the older age group because the scale ends at 36 months.

Table 5 panels (b) and (c) show no evidence of significant heterogeneous effects of RCP by sex, although girls in RCP communities score significantly lower on Expressive Language than girls in control communities. The findings are robust to the in- or exclusion of controls. It is difficult to conceive that RCP has actually harmed language development. Further analysis suggests that the negative coefficient concentrates on girls in the youngest birth cohort in program villages who did not participate in RCP and who were enrolled in center-based daycare at follow-up. However, if there were any detrimental effects of daycare on infants, one would expect a similar developmental lag among the youngest in control villages, instead of a significant difference. We hence have no adequate explanation for this finding.

The results in panels (d) and (e) suggest that the program has been partially effective in enhancing cognitive development among the youngest age cohort but not the older age cohort. The youngest birth cohort in the treatment group experienced a significantly different change in scores between baseline and follow-up compared to the control group, both on the Visual Reception scale and on the Fine Motor scale. Their improvement on Visual Reception is 3.60 points higher compared to children of similar age in non-RCP communities, and 3.82 points higher on the Fine Motor scale. These are sizeable differences, equal to more than a third of a standard deviation. The estimates are lower, less precise and not statistically significant at conventional levels in the unadjusted specifications without control variables. None of the impact estimates for the older birth cohort are statistically significant, and except for the Receptive Language scale they are all very close to zero.

6.3. Participation, age at enrollment and length of enrollment

To analyze program effects in more detail, this section investigates how participation, age at enrollment and length of enrollment are related to differential child outcomes.¹⁶ It should be kept in mind that these participation indicators are choice variables. Therefore, results in this section should be interpreted in terms of correlation instead of causality. They are reported for the total sample as well as by birth cohort. Age at enrollment was on average 12.6 months for the youngest birth cohort and 24.8 months for the oldest birth cohort (Table 3 panels B and C). Average length of enrollment at follow-up was slightly larger for the youngest birth cohort at 13.8 months compared to 12.2 months for the oldest cohort.

Table 6 panel A shows the coefficients on the RCP community variable when non-participants in program villages are excluded from the sample. If RCP had an impact on its participants, one would expect the coefficients in these estimations to be larger than in the concomitant intention-to-treat estimates in Table 5. Excluding non-participants does not affect the results for the full sample. The coefficients for the youngest cohort substantially increase in size and significance, providing additional support for a positive impact on the youngest children. The estimates for the oldest cohort however become negative and significant for the Visual Reception scale, indicating that the oldest children who were enrolled in RCP perform worse compared to the older birth cohort in the control area. The next section will investigate to what extent this is due to a selection effect.

Table 6. Participation, age at enrollment and length of enrollment.

Changes in child development outcomes	Summary score	Gross Motor	Visual Reception	Fine Motor	Receptive Language	Expressive Language
Panel A. Excluding non-participants from the sample
(a) Total population
RCP community	−1.60	0.15	−0.31	−1.64	1.60	−2.65^**
	(2.21)	(0.85)	(1.45)	(1.97)	(1.22)	(1.10)
(b) Youngest cohort
RCP community	3.77^*	1.45	4.29^**	4.38^**	1.84	−2.56^**
	(2.02)	(0.84)	(1.64)	(1.61)	(1.71)	(1.03)
(c) Oldest cohort	−2.10		−2.75^*	−3.23	1.51	0.36
RCP community	(3.21)		(1.42)	(2.55)	(2.03)	(1.98)
Panel B. Program participation
(a) Total population
RCP community	2.83	−5.80^**	1.79	4.33	0.29	−0.83
	(3.26)	(2.54)	(1.88)	(3.23)	(3.01)	(2.41)
RCP community × participation	−4.15	5.76^*	−1.91	−5.60	1.26	−1.75
	(3.01)	(2.84)	(2.13)	(3.79)	(2.59)	(2.19)
Test of joint significance	.396	.097^*	.630	.358	.387	.077^*
(b) Youngest cohort
RCP community	−4.85	−6.26^**	−0.35	−1.30	−3.62^*	−5.35^*
	(3.51)	(2.67)	(3.95)	(2.93)	(2.03)	(2.93)
RCP community × participation	8.72^**	7.44^**	4.59	5.95^*	5.44^**	2.83
	(3.43)	(3.08)	(3.16)	(3.34)	(2.38)	(2.38)
Test of joint significance	.043^**	.088^*	.021^**	.029^**	.106	.093^*
(c) Oldest cohort
RCP community	6.08		2.83	5.95	3.38	0.53
	(4.75)		(3.61)	(4.77)	(3.62)	(2.16)
RCP community × participation	−8.01		−5.36	−9.05	−1.86	−0.19
	(4.60)		(3.57)	(5.20)	(2.61)	(1.69)
Test of joint significance	.252		.134	.223	.651	.970
Panel C. Age at enrollment
(a) Total population
RCP community	1.13	1.18	1.51	1.43	2.94	−3.27^*
	(2.82)	(1.73)	(1.65)	(2.26)	(2.53)	(1.66)
RCP com. × age at enrollment	−0.11	−0.17	−0.09	−0.11	−0.12	0.08
	(0.13)	(0.13)	(0.07)	(0.10)	(0.12)	(0.08)
Test of joint significance	.663	.300	.496	.592	.520	.154
(b) Youngest cohort
RCP community	1.12	−0.84	2.36	2.01	−0.51	−1.88
	(2.63)	(2.35)	(3.59)	(2.23)	(1.15)	(2.36)
RCP com. × age at enrollment	0.14	0.10	0.11	0.16	0.14	−0.09
	(0.20)	(0.20)	(0.20)	(0.18)	(0.11)	(0.15)
Test of joint significance	.428	.826	.066^*	.051^*	.481	.053^*
(c) Oldest cohort
RCP community	5.91		1.67	5.12	4.77	1.10
	(4.93)		(3.56)	(4.40)	(3.87)	(2.01)
RCP com. × age at enrollment	−0.33		−0.16	−0.33^*	−0.16	−0.04
	(0.21)		(0.14)	(0.18)	(0.14)	(0.08)
Test of joint significance	.325		.262	.185	.475	.817
Panel D. Length of enrollment
(a) Total population
RCP community	2.08	−2.10	1.00	3.62	0.97	−1.24
	(2.67)	(2.81)	(1.55)	(2.48)	(2.24)	(1.95)
RCP com. × length of enrollment	−0.25	0.11	−0.07	−0.37^*	0.03	−0.10
	(0.19)	(0.25)	(0.13)	(0.18)	(0.14)	(0.13)
Test of joint significance	.429	.454	.807	.162	.621	.097^*
(b) Youngest cohort
RCP community	−4.15	−2.26	0.01	−1.28	−1.24	−6.21^**
	(3.92)	(2.95)	(3.61)	(2.91)	(1.41)	(2.65)
RCP com. × length of enrollment	0.58^**	0.21	0.30	0.43^*	0.19	0.28
	(0.24)	(0.25)	(0.18)	(0.21)	(0.11)	(0.16)
Test of joint significance	.041^**	.659	.005^***	.019^**	.240	.079^*
(c) Oldest cohort
RCP community	6.53		2.58	6.67^*	3.72	1.04
	(4.01)		(2.76)	(3.78)	(3.01)	(1.70)
RCP com. × length of enrollment	−0.78^**		−0.45^*	−0.91^**	−0.21	−0.08
	(0.33)		(0.22)	(0.33)	(0.17)	(0.14)
Test of joint significance	.097^*		.059^*	.037^**	.413	.718

Robust standard errors, clustered at the community level, in parentheses. All regressions control for round and tester. Additional control variables are age of child (in months), caregiver employment status, and household wealth. The last rows in panels B–D show the p-value of a test of joint significance of the RCP community variable and its interaction with the participation variable.

*: p-Value < .100.
**: p-Value < .050.
^***p-Value < .010.

The next panel takes a different approach to assess the impact on participants. It is based on the entire sample of participants and non-participants. The specification includes both the RCP community variable and its interaction with (prior) participation in RCP. Since the interaction term captures the effect of participation, the RCP community variables picks up outcomes for children who live in an RCP community but who did not participate in the program. The estimates as well as the tests of joint significance of the RCP community variable and its interaction with participation reiterate the positive impacts on the younger but not the older birth cohort, as found in panel A.

Panel C adds an interaction term between the RCP community variable and age at enrollment. None of the coefficients is significant at the 5% error level, either individually or jointly. Hence, age at enrollment does not seem strongly correlated with program impact. Note however that the interaction coefficients are consistently positive for the youngest birth cohort while they are consistently negative for the older birth cohort. These counterbalancing patterns may explain the lack of significant findings for the full sample, despite strong indications of treatment heterogeneity by birth cohort.

In panel D, the RCP community variable is interacted with length of enrollment. For the full sample, again no clear pattern can be discerned. Differences between the two birth cohorts are pronounced, showing opposite trends. Whereas cognitive development scores of the youngest birth cohort steadily increase with increasing length of enrollment, outcomes significantly decrease for children in the oldest birth cohort the longer they are enrolled.

The most important reason for children to prematurely drop out of RCP is to enroll in daycare or preschool. Indeed, there is a strong negative correlation between length of enrollment in RCP and being enrolled in a center-based ECD facility at follow-up. The correlation coefficient is −0.613 (p-value .000) for the oldest birth cohort and −0.204 (p-value .051) for the youngest birth cohort in RCP communities. The next section will therefore investigate the relation between child development outcomes and center-based as opposed to home-based ECD services.

6.4. Daycare and preschool enrollment

As all real-world programs, RCP is not implemented in a vacuum but within a broader context of ECD-related services and economic development. After 2006, there was a steep and unexpected rise in job opportunities in many of the study communities. This has led to a general increase in both the demand for center-based care as well as the financial means to pay for it. In 2006, only fifteen out of every hundred children aged birth to five years old in our study households (i.e. study children plus their siblings) were enrolled in a center-based ECD facility. By 2008, enrollment in center-based care had more than doubled to 36.9% of all children aged 0 to 5 years.

Table 4 column (2) analyzes which child and household characteristics at baseline were correlated with enrollment in daycare (for the majority of the younger cohort) or preschool (for the majority of the older cohort) in 2008. The first row indicates that there were no significant differences in center-based enrollment between children living in treatment communities and children living in control communities. Thus, there are no indications that RCP deterred or instead stimulated children to attend an ECD center. Children in the younger age group from male-headed, poorer, unemployed households were least likely to attend a center-based facility. There were no differences in enrollment rates between low and high educated families. Finally, the findings indicate that enrollment in daycare was unrelated to child developmental outcomes at baseline. These results suggest that the decision to enroll a child was mostly financially motivated.

Table 7 further explores the relationship between RCP, center-based ECD programs and child outcomes. Panel A shows the estimates using the same basic specification as the impact regressions, but adding a variable indicating whether the child was enrolled in a center-based ECD facility at follow-up. The estimates should be interpreted with caution. They do not reflect causal effects because of self-selection into ECD centers. For the full cohort, the RCP estimates remain insignificant, while the coefficients on the daycare/preschool enrollment variable are large and statistically significant for the summary score as well as the Visual Reception and the Fine Motor scales. Thus, the average child who was enrolled in an ECD center at the time of follow-up showed substantially higher scores on two of the individual child development scales. This effect is entirely driven by the oldest birth cohort. Daycare attendance of the youngest children is not correlated with cognitive outcomes, while the RCP coefficients remain statistically significant. These findings provide a potential explanation for the significant negative coefficient on length of enrollment for older participants in Table 6.

Table 7. Home-based RCP versus center-based ECD facilities.

Changes in child development outcomes	Summary score	Gross Motor	Visual Reception	Fine Motor	Receptive Language	Expressive Language
Panel A. RCP and center-based care
(a) Total population
RCP community	0.36	−0.63	0.97	0.83	1.02	−1.97
	(2.26)	(0.80)	(1.16)	(1.89)	(1.44)	(1.35)
Center-based care	6.75^***	−1.84	4.39^**	9.52^***	−2.14	1.96
	(2.27)	(2.09)	(1.67)	(2.08)	(1.46)	(1.16)
Test of joint significance	.032^**	.436	.033^**	.002^***	.296	.042^**
(b) Youngest cohort
RCP community	3.21	0.48	4.29^*	3.97^**	1.24	−2.74^*
	(2.13)	(0.78)	(2.15)	(1.37)	(1.65)	(1.3)
Center-based care	1.15	−2.06	1.87	0.26	−1.25	1.63
	(3.72)	(2.39)	(2.55)	(2.62)	(2.83)	(1.93)
Test of joint significance	.324	.637	.113	.024^**	.743	.041^**
(c) Oldest cohort
RCP community	1.29		−0.18	1.37	1.60	−0.02
	(3.22)		(1.40)	(2.64)	(2.54)	(2.04)
Center-based care	6.68^***		4.97^***	11.79^***	−1.82	−1.11
	(1.85)		(1.56)	(2.51)	(2.04)	(0.86)
Test of joint significance	.009^***		.020^**	.000^***	.383	.448
Panel B. Interaction between RCP participation and center-based care
(a) Total population
RCP community	1.61	−5.40^*	1.01	2.58	0.60	−1.16
	(3.63)	(2.65)	(1.75)	(3.26)	(2.84)	(2.46)
RCP community × participation	−0.63	4.70	0.45	0.02	0.43	−1.50
	(3.29)	(3.10)	(2.28)	(3.38)	(2.71)	(2.30)
Center-based care	7.50^***	−3.18	4.84^**	11.37^***	−2.20	1.46
	(2.21)	(2.80)	(2.01)	(1.73)	(1.60)	(1.16)
RCP community × participation × center-based care	−3.49	4.17	−1.72	−7.94^*	0.44	1.50
	(3.67)	(3.37)	(2.23)	(4.24)	(2.61)	(1.98)
Test of joint significance	.052^*	0.145	.093^*	.000^***	0.495	0.114
(b) Youngest cohort
RCP community	−5.54	−6.07^**	−0.92	−1.84	−3.54	−5.85^*
	(3.93)	(2.71)	(4.27)	(3.50)	(2.20)	(3.03)
RCP community × participation	11.51^***	6.63^*	6.44^*	8.59^**	6.22^**	3.74
	(3.73)	(3.31)	(3.52)	(3.59)	(2.47)	(3.10)
Center-based care	4.04	−3.42	2.91	3.74	0.20	2.09
	(4.31)	(2.84)	(3.28)	(3.29)	(3.48)	(2.39)
RCP community × participation × center-based care	−5.68	4.40	−1.53	−7.99^*	−2.76	−0.47
	(5.60)	(3.10)	(4.02)	(3.95)	(5.15)	(4.00)
Test of joint significance	.038^**	.058^*	.022^**	.012^**	0.115	0.159
(c) Oldest cohort
RCP community	5.68		2.55	5.18	3.47	0.63
	(5.23)		(3.53)	(4.98)	(3.52)	(2.11)
RCP community × participation	−6.47		−3.91	−4.56	−3.11	−1.72
	(5.72)		(3.83)	(5.87)	(3.42)	(2.22)
Center-based care	6.04^**		4.64^**	11.85^***	−2.30	−1.65
	(2.39)		(2.06)	(2.44)	(2.53)	(1.05)
RCP community × participation × center-based care	−0.25		−0.47	−3.13	0.88	2.06^*
	(3.75)		(3.64)	(4.45)	(3.36)	(1.14)
Test of joint significance	.012^**		.015^**	.002^***	0.705	.074^*

Robust standard errors, clustered at the community level, in parentheses. All regressions control for round and tester. Additional control variables are age of child (in months), caregiver employment status, and household wealth. The last row in each panel shows the p-value of a test of joint significance of the reported estimates.

*: p-Value < .100.
**: p-Value < .050.
***: p-Value < .010.

Table 7 panel B investigates whether RCP reinforces any beneficial effects of center-based ECD services for children previously enrolled in RCP, or whether RCP and center-based care are instead substitutes. For this purpose, the estimations include the RCP community variable, an indicator equal to 1 if a child in an RCP community ever participated in the program, an indicator equal to 1 if the child is currently enrolled in a center-based ECD facility and an interaction term of (prior) participation in RCP and current center-based enrollment. In line with previous results, the estimates for RCP participation are positive and significant only for the youngest cohort whereas the estimates for center-based enrollment are positive and significant only for the oldest cohort. Results are not suggestive of synergistic effects. The majority of interaction terms is statistically insignificant. The negative interaction coefficient in the estimation of Fine Motor Skills suggests that current daycare enrollment among the youngest cohort diminishes the positive results from RCP. Only for Expressive Language in the oldest birth cohort is the interaction term positive and significant at the 10% error level.

Although these estimates cannot be interpreted as impact effects of enrollment in daycare or preschool, the results provide suggestive indications that RCP is only beneficial for children who enroll at a young age whereas older children may benefit more from center-based ECD services, regardless of prior participation in RCP. Additional research is required to allow for causal attributions.

7. Discussion and conclusion

This evaluation of the Roving Caregivers Program in St Lucia is one of the few existing impact evaluations of home-based interventions in the Caribbean. One year after implementation, we do not find a program impact on the average child in treatment communities. However, the estimates for the full sample mask pronounced treatment heterogeneity by age. The findings show significant positive program effects for the younger birth cohort, which was on average twelve months old at program introduction, on the two Mullen Scales of Early Learning that are related to eye–hand coordination and early reading readiness. The impact estimates indicate that the outcome changes since baseline are more than a third of a standard deviation larger in the treatment group than changes over that same period of time in the control group.

The effects on the older birth cohort, which was on average twenty-four months of age when the program started, are not significant. The lack of impact does not seem attributable to a ceiling effect. Except for Gross Motor Skills, at most 3.6% of the scores of the older cohort were at the maximum on any scale of the assessment tool, and 94.1–99.0% of the cohort scored below +2 standard deviations from the mean on any scale. Nor does it seem related to vanishing gains over time for the oldest birth cohort since the average length of enrollment as well as the time elapsed since drop-out for those who left the program, were relatively comparable among the two cohorts. A core feature of the program relates to the involvement of parents in the Rover interactions with children, enabling them to adopt these practices on a daily basis. Opportunities for caregiver practice with their children with feedback have been identified as a key success factor for home-based programs (Engle et al., 2011). However, especially in the early stages of RCP implementation, caregivers were not always fully involved during Rover visits. This may provide a partial explanation of the low impact, although it is not clear why this would undermine impact for the older but not the younger cohort. Unfortunately, data limitations prevent us from analyzing this in more detail.

There is no evidence of a positive impact of RCP on language development of either birth cohort, despite this being an explicit aim of the program. A non-published randomized controlled evaluation of RCP in Jamaica finds similar positive effects on eye–hand coordination and no effect on hearing and speech (Powell, 2004). Personal communications suggest that the lack of impact in this respect may be due to relatively little focus of the Rovers on language stimulation.¹⁷ Intensified Rover training in the language domain seems warranted. This recommendation is supported by a recent review that suggests home-based programs are most effective when based on a structured, evidence-based curriculum coupled with systematic training methods for the home visitors (Engle et al., 2011). Noteworthy is that a meta-analysis of 60 home-visiting interventions in the United States finds no differences in performance of paraprofessionals such as the Rovers compared to professional home visitors (Sweet & Appelbaum, 2004).¹⁸

This evaluation measures only the impact on cognitive development in the short run. There may be beneficial effects of RCP on other child outcomes such as non-cognitive development or on parent outcomes such as maternal stress and knowledge, which are not captured in the evaluation (Cunha and Heckman, 2008, Engle et al., 2011, Sweet and Appelbaum, 2004). Moreover, benefits may not materialize until a later age (Heckman & Kautz, 2013).

Medium- and long-term evaluations of the program are also necessary to shed light on the sustainability of the gains among the youngest cohort. Findings from other studies are promising in this respect (e.g. Claessens et al., 2009, Deming, 2009, Gertler et al., 2013, Hazarika and Viren, 2013, Walker et al., 2005). They find that the impact of early learning interventions on school performance and adult labor productivity can be substantial, with improvements in indicators of up to 40%.

Overall, the findings suggest that there are windows of opportunity for improving cognitive child development through home-based interventions. Programs may therefore opt to enroll families as soon as possible after birth of the child. From a programmatic point of view, cooperation with maternal and child health centers seems a promising way forward as it allows reaching caregivers at an early, even pre-natal, stage. Research points out to a number of health and nutrition concerns in St Lucia, especially for children living in the poorest and least educated households (Groot Bruinderink & Janssens, 2010). Being a program that works with parents and specifically targets young vulnerable children, RCP seems particularly well suited to address health problems and malnutrition in this population. This would directly benefit the healthy development of the most disadvantaged children, and may indirectly translate into better cognitive outcomes, although evidence in this respect is mixed (Alderman et al., 2001, Maluccio et al., 2009, Walker et al., 2005).

The favorable economic climate after baseline allowed an unexpectedly high proportion of children to enroll in daycare or preschool. Whereas daycare attendance among the younger birth cohort appears unrelated with child outcomes, preschool attendance among the older birth cohort is strongly correlated with higher cognitive scores, regardless of children's prior enrollment in RCP. These results cannot be interpreted in a causal sense because of selectivity. But they are in line with ample evidence documenting the potential benefits of center-based ECD programs. Longitudinal studies find that the impact of center-based interventions can be long-lasting and reach far into the future (Currie, 2001, Engle et al., 2011, Garces et al., 2002, Schweinhart et al., 2005), although the effects of early childhood education are sometimes found to be detrimental, especially with respect to behavioral skills (e.g. DeCicca and Smith, 2013, Loen et al., 2007). Studies that evaluate center-based interventions in Latin-America and the Caribbean have found beneficial cognitive effects in for example Argentina (Berlinksi, Galiani, & Gertler, 2009) and Bolivia (Behrman, Cheng, & Todd, 2004). After the follow-up survey in 2008 however, the world financial crisis hit St Lucia particularly hard due to its economic dependence on tourism. If caregivers of young children lost their employment again, this may have fueled the need for alternative and low cost ECD services such as the home-based RCP.

A key requirement for enrollment in RCP is that both the child and the child's caregiver must be present during the Rover visits. As a result, most mothers who participate in RCP are unemployed. This may introduce a bias against the working poor. The program addresses this as much as possible by allowing other persons who take care of the child during the day to participate in the program instead. Nevertheless, a significant negative correlation exists between enrollment and caregiver employment. Further efforts could be made to accommodate working caregivers’ schedules. Although conducting the visits in the evening is impractical for young families as well as the Rovers, alternative options can be thought of, such as offering services in the weekends. Concerns have also been raised that the program, if highly valued, may create unintended side-effects by providing incentives to remain unemployed. Given the poverty levels in RCP communities, it seems however unlikely that parents would forego employment in order to remain eligible for the program.

Ideally, the program would have been evaluated through a randomized controlled trial. However, as most programs that are being scaled up, RCP was not randomized at implementation. The validity of the estimated intention-to-treat effects rests on the absence of unobserved time-varying differences between the treated and untreated communities, which appears to be a plausible assumption given the similarities in key indicators between the two groups. From a policy perspective it is crucial to extend the scope of evaluations beyond randomized designs and include quasi-experimental studies of existing child development programs in order to increase understanding of potential impacts, and add to the, as of yet limited, knowledge base on home interventions in low-income countries.

Acknowledgements

We gratefully acknowledge the cooperation with Jeroen van Spijk at the Amsterdam Institute for International Development and the excellent research assistance of Marije Groot Bruinderink. We thank Jacques van der Gaag, Orazio Attanasio, Sally Grantham-McGregor, Jan Willem Gunning, Robert Sparrow, and two anonymous referees, as well as participants at the Institute of Fiscal Studies seminar in London, the 3IE conference 2009 in Cairo, the 2009 PEGNet conference in The Hague, and the Development Economics group at the VU University for useful comments. We very much appreciate the support of and discussions with the Caribbean Child Support Initiative, the Roving Caregivers Program in St Lucia and the Bernard van Leer Foundation, in particular with Susan Branker, Ruth Fevrier, and Huub Schreurs. We thank the Bernard van Leer Foundation for financial support with grant number CAR-2005-116.

Appendix A. Comparison of baseline community characteristics between the treatment and the control group

Variable	Average	St. dev.	Min	Max	Total (weighted average)	Control (weighted average)	Treat (weighted average)
Empty Cell	(i)	(ii)	(iii)	(iv)	(v)	(vi)	(vii)
Panel A. Poverty
Poverty index	7.80	.68	6.82	9.49	8.08	8.19	7.97
% Unemployment	22	7	12	41	20	19	22
Panel B. Demographics
Total number of children 0–3 in community	81	53	11	183	109	119	98
Total community population size	889	642	224	2403	1322	1488	1159
Panel C. ECD services in the community (0 = no; 1 = yes)
Presence of at least one ECD facility in community	.80	.41	0	1	.93	.91	.95
Public daycare	.33	.49	0	1	.28	.05	.50
Private daycare	.33	.49	0	1	.51	.48	.53
Preschool	.53	.52	0	1	.77	.73	.81
Panel D. Presence of other facilities (0 = no; 1 = yes)
Primary school	.60	.51	0	1	.83	.85	.81
Health center	.47	.52	0	1	.70	.70	.69
Playground	.67	.49	0	1	.86	.85	.86
Sport facilities	.53	.52	0	1	.76	.70	.81
Church	.73	.46	0	1	.86	.85	.87
Public transport	.93	.26	0	1	.94	1.00	.87
Police station	.13	.35	0	1	.28	.36	.19
Post office	.53	.52	0	1	.74	.73	.74
Bank	.20	.41	0	1	.43	.36	.48
General store	.67	.49	0	1	.81	.89	.74
NELP Centers	.33	.49	0	1	.40	.48	.33
Panel E. Security and social issues
Is … a problem in the community? (0 = no; 1 = yes)
Flooding/natural hazards	.67	.49	0	1	.61	.72	.51
Bad hygiene and lack of sanitation	.40	.51	0	1	.55	.40	.70
Drugs	.80	.41	0	1	.90	.94	.86
Alcohol	.80	.41	0	1	90	.95	.86
Prostitution	.13	.35	0	1	.06	.03	.09
Crime and violence	.93	.26	0	1	.82	.64	1.00
How severe is the problem? (1 = no problem; 2 = it happens but is not a problem; 3 = minor problem; 4 = fairly big problem; 5 = big problem)
Flooding/natural hazards	3.20	1.57	1	5	3.13	3.18	3.08
Bad hygiene and lack of sanitation	2.13	1.51	1	5	2.34	1.86	2.81
Drugs	3.40	1.40	1	5	3.64	3.43	3.85
Alcohol	3.53	0.99	2	5	3.80	3.86	3.73
Prostitution	1.80	1.08	1	5	1.71	1.62	1.80
Crime and violence	3.87	0.99	2	5	3.59	3.16	4.02
Panel F. Community organizations (0 = not present in community; 1 = present in community)
Total number of community groups	17	12	0	36	22	22	22
Mother's and Father's group	.67	.49	0	1	.53	.57	.49
Women's groups	.40	.51	0	1	.46	.36	.55
Parent–teacher association	.60	.51	0	1	.79	.85	.73
Youth groups	.87	.35	0	1	.94	.89	1.00
Sports group	.67	.49	0	1	.66	.37	.95
Cultural group (e.g. music, theater, dance)	.67	.49	0	1	.72	.66	.78
Religious/charity groups	.73	.46	0	1	.86	.77	.95
Elderly group	.20	.41	0	1	.20	.12	.28
Farmers or agricultural groups	.33	.49	0	1	.44	.49	.40
Business or professional groups	.00	.00	0	0	.00	.00	.00
Savings and credit group	.20	.41	0	1	.43	.36	.48
Development committee	.60	.51	0	1	.65	.89	.42
Self-help group	.47	.52	0	1	.42	.26	.56
Disaster preparedness group	.53	.52	0	1	.74	.89	.59
Other groups	.27	.46	0	1	.26	.12	.41
Panel G. Accessibility and mobility
Community along main road (0 = no; 1 = yes)	.33	.49	0	1	.43	.55	.32
Community along coast (0 = no; 1 = yes)	.20	.41	0	1	.34	.36	.32
Road quality (1= bad, 2 = fair, 3 = good)	2.40	.63	1	3	2.33	2.20	2.46
Geographic spread of houses (0 = spread out over large area; 1 = close to each other)	.53	.52	0	1	.65	.55	.75

Source of panels A and B: Statistical Office St Lucia, based on the Census 2001.

Source of panels C–G: Own data collection (2006).

References

Alderman et al., 2001
H. Alderman, J.R. Behrman, V. Lavy, R. Menon
Child health and school enrollment: A longitudinal analysis
Journal of Human Resources, 36 (1) (2001), pp. 185-205
Crossref Google Scholar
Augustin, 2004
M. Augustin
Parenting support and early childhood intervention: A replication plan for Saint Lucia
Caribbean Child Support Initiative (CCSI), Barbados (2004)
Google Scholar
Barrow, 2003
C. Barrow
Child rearing and socialisation approaches in the Caribbean—Phase I: Trinidad; Phase II: Dominica
Caribbean Child Support Initiative (CCSI), Barbados (2003)
Google Scholar
Behrman et al., 2004
J.R. Behrman, Y. Cheng, P.E. Todd
Evaluating preschool programs when length of exposure to the program varies: A nonparametric approach
Review of Economics and Statistics, 86 (1) (2004), pp. 108-132
View in Scopus Google Scholar
Berlinksi et al., 2009
S. Berlinksi, S. Galiani, P. Gertler
The effect of pre-primary education on primary school performance
Journal of Public Economics, 93 (2009), pp. 219-234
Google Scholar
Charles, 2004
L. Charles
Early childhood policy
Government of Saint Lucia, Saint Lucia (2004)
Google Scholar
Claessens et al., 2009
A. Claessens, G. Duncan, M. Engel
Kindergarten skills and fifth-grade achievement: Evidence from the ECLS-K
Economics of Education Review, 28 (2009), pp. 415-427
View in Scopus Google Scholar
Cunha and Heckman, 2008
F. Cunha, J. Heckman
Formulating, identifying and estimating the technology of cognitive and non-cognitive skill formation
Journal of Human Resources, 43 (2008), pp. 738-782
View in Scopus Google Scholar
Currie, 2001
J. Currie
Early childhood education programs
Journal of Economic Perspectives, 15 (2) (2001), pp. 213-238
Crossref View in Scopus Google Scholar
Davis et al., 2009
R.E. Davis, M.P. Couper, N.K. Janz, C.H. Caldwell, K. Resnicow
Interviewer effects in public health surveys
Health Education Research, 25 (1) (2009), pp. 14-26
View in Scopus Google Scholar
DeCicca and Smith, 2013
P. DeCicca, J. Smith
The long-run impacts of early childhood education: Evidence from a failed policy-experiment
Economics of Education Review, 36 (2013), pp. 41-59
View in Scopus Google Scholar
Deming, 2009
D. Deming
Early childhood intervention and life-cycle skill development: Evidence from Head Start
American Economic Journal: Applied Economics, 1 (3) (2009), pp. 111-134
Crossref Google Scholar
Durrant et al., 2010
G.B. Durrant, R.M. Groves, L. Staetsky, F. Steele
Effects of interviewer attitudes and behaviors on refusal in household surveys
Public Opinion Quarterly, 74 (1) (2010), pp. 1-36
Crossref View in Scopus Google Scholar
Engle et al., 2011
P. Engle, L. Fernald, H. Alderman, J. Behrman, C. O’Gara, A. Yousafzai, et al.
Strategies for reducing inequalities and improving developmental outcomes for young children in low-income and middle-income countries
The Lancet, 378 (2011), pp. 1339-1353
View in Scopus Google Scholar
Garces et al., 2002
E. Garces, D. Thomas, J. Currie
Longer-term effects of head start
American Economic Review, 92 (4) (2002), pp. 999-1012
View in Scopus Google Scholar
Gertler et al., 2013
P. Gertler, J. Heckman, R. Pinto, A. Zanolini, C. Vermeerch, S. Walker, et al.
Labor market returns to early childhood stimulation
World Bank Policy Research Paper 6529
World Bank, Washington, DC (2013)
Google Scholar
Grantham-McGregor et al., 2007
S.M. Grantham-McGregor, Y.B. Cheung, S. Cueto, P. Glewwe, L. Richter, B. Strupp, et al.
Developmental potential in the first 5 years for children in developing countries
The Lancet, 369 (2007), pp. 60-70
View in Scopus Google Scholar
Groot Bruinderink and Janssens, 2010
Groot Bruinderink, M., & Janssens, W. (2010). Child Health and Nutrition in St Lucia. AIID Research Paper, Amsterdam Institute for International Development.
Google Scholar
Hazarika and Viren, 2013
G. Hazarika, V. Viren
The effect of early childhood developmental program attendance on future school enrollment in rural North India
Economics of Education Review, 34 (2013), pp. 146-161
View in Scopus Google Scholar
Heckman, 2006
J. Heckman
Skill formation and the economics of investing in disadvantaged children
Science, 312 (5782) (2006), pp. 1900-1902
Crossref View in Scopus Google Scholar
Heckman and Kautz, 2013
J. Heckman, T. Kautz
Fostering and measuring skills: Interventions that improve character and cognition
NBER Working Paper 19656
National Bureau of Economic Research, Cambridge, MA (2013)
Google Scholar
Loen et al., 2007
S. Loen, M. Bridges, D. Bassok, B. Fuller, R.W. Rumberger
How much is too much? The influence of preschool centers on children's social and cognitive development
Economics of Education Review, 26 (2007), pp. 52-66
Google Scholar
Maluccio et al., 2009
J.A. Maluccio, J. Hoddinott, J.R. Behrman, R. Martorell, A.R. Quisumbing, A.D. Stein
The impact of improving nutrition during early childhood on education among Guatemalan adults
The Economic Journal, 119 (2009), pp. 734-763
Crossref View in Scopus Google Scholar
Mullen, 1995
E.M. Mullen
Mullen scales of early learning, AGS edition: Manual
American Guidance Service (AGS), Inc., Circle Pines, MN (1995)
Google Scholar
Nores and Barnett, 2010
M. Nores, W.S. Barnett
Benefits of early childhood interventions across the world: (Under) Investing in the very young
Economics of Education Review, 19 (2010), pp. 271-282
View in Scopus Google Scholar
Paxson and Schady, 2007
C. Paxson, N. Schady
Cognitive development among young children in Ecuador: The roles of wealth, health and parenting
Journal of Human Resources, 42 (1) (2007), pp. 49-84
View in Scopus Google Scholar
Powell, 2004
C.A. Powell
An evaluation of the roving caregivers programme of the rural family support organisation
University of the West Indies, Tropical Medicine Research Institute, Kingston, Jamaica (2004)
Google Scholar
Powell et al., 2004
C.A. Powell, H. Baker-Henningham, S. Walker, J. Gernay, S. Grantham-McGregor
Feasibility of integrating early stimulation into primary care for undernourished Jamaican children: Cluster randomized controlled trial
British Medical Journal, 329 (89) (2004)
Google Scholar
Roopnarine, 2005
J.L. Roopnarine
The Roving Caregivers Program in Jamaica: Its theoretical and research foundations and efficacy
Syracuse University, Syracuse, NY (2005)
Google Scholar
Samms-Vaughan, 2004
M. Samms-Vaughan
Profiles—The Jamaican pre-school child: The status of early childhood development in Jamaica
Planning Institute of Jamaica, Kingston, Jamaica (2004)
Google Scholar
Schady, 2006
N. Schady
Early childhood development in Latin America and the Caribbean
Economía, 6 (2) (2006), pp. 185-225
Crossref Google Scholar
Schweinhart et al., 2005
L. Schweinhart, J. Montie, Z. Xiang, W.S. Barnett, C. Belfield, M. Nores
Lifetime effects: The high/scope Perry preschool study through age 40. Monographs of the High/Scope Educational Research Foundation 14
(2005)
Google Scholar
Shonkoff and Phillips, 2000
J.P. Shonkoff, D.A. Phillips (Eds.), From neurons to neighborhoods: The science of early childhood development, National Academy Press, Washington, DC (2000)
Super et al., 1990
C.M. Super, M. Guillermo Herrera, J.O. Mora
Long-term effects of food supplementation and psychosocial intervention on the physical growth of Colombian infants at risk of malnutrition
Child Development, 61 (1) (1990), pp. 29-49
View in Scopus Google Scholar
Sweet and Appelbaum, 2004
M.A. Sweet, M.I. Appelbaum
Is home visiting an effective strategy? A meta-analytic review of home visiting programs for families with young children
Child Development, 75 (5) (2004), pp. 1435-1456
View in Scopus Google Scholar
Walker et al., 2005
S.P. Walker, S.M. Chang, C.A. Powell, S.M. Grantham-McGregor
Effects of early childhood psychosocial stimulation and nutritional supplementation on cognition and education in growth-retarded Jamaican children: Prospective cohort study
The Lancet, 366 (9499) (2005), pp. 1804-1807
View in Scopus Google Scholar
Williams and Brown, 2005
S. Williams, J. Brown
Research approaches and findings on childrearing and socialisation of young children (birth to eight years of age) in the Caribbean
Caribbean Child Support Initiative (CCSI), Bridgetown, Barbados (2005)
Google Scholar
Williams et al., 2006
S. Williams, J. Brown, J. Roopnarine
Child rearing in the Caribbean: A literature review
Digital Solution, Kingston. Jamaica (2006)
Google Scholar
Young, 2002
M.E. Young (Ed.), From early child development to human development, World Bank, Washington, DC (2002)

Cited by (7)

Identifying the Common Elements of Early Childhood Interventions Supporting Cognitive Development in Low- and Middle-Income Countries
2024, Adversity and Resilience Science
Validation of a developmental checklist for young (0-5) Surinamese children
2023, Pedagogische Studien
Teaching home-visitors to support responsive caregiving: A cluster randomized controlled trial of an online professional development program in Brazil
2022, Journal of Global Health
Indian Academy of Pediatrics Position Paper on Nurturing Care for Early Childhood Development
2021, Indian Pediatrics
Systematic Review and Meta-Analyses: Motor Skill Interventions to Improve Fine Motor Development in Children Aged Birth to 6 Years
2020, Journal of Developmental and Behavioral Pediatrics
Effects of home-based play-assisted stimulation on developmental performances of children living in extreme poverty: A randomized single-blind controlled trial
2018, BMC Pediatrics

View all citing articles on Scopus

¹: Present address: 41 Fairbridge Road, London N19 3EW, United Kingdom. Tel.: +44 075 924863.

²: Engle et al. (2011) and Nores and Barnett (2010) review impact evaluations of ECD programs in developing countries. Schady (2006) provides a detailed overview of the Latin American and Caribbean evidence.

³: “At a Glance: Saint Lucia Statistics", Unicef, 2009. http://www.unicef.org/infobycountry/stlucia_statistics.html.

⁴: “The Assessment of Poverty in St Lucia, Volume I”, Caribbean Development Bank, 2006.

⁵: Average monthly fees range from US$ 55 to US$ 200 (Government of Saint Lucia, “Poverty Assessment Report St. Lucia—Executive Summary", 1995).

⁶: “Core Welfare Indicators Questionnaire Survey (CWIQ): A Pilot Study in St. Lucia", Statistical Office, Government of Saint Lucia, 2004.

⁷: Participants in the treatment communities did not receive the food hamper at baseline. Since the hamper was not distributed until after conclusion of the assessment and its value was modest (***US$15), this is not likely to have affected outcomes of the child assessments at baseline or the follow-up two years later.

⁸: The scales behave as expected for the St Lucian sample in relation to relevant child and household characteristics such as sex, age, maternal education and wealth.

⁹: Comparisons of children who did and who did not participate in the baseline assessment (non-response), as well as of children who did and who did not participate at follow-up (attrition) suggest that bias in the sample is limited. Non-respondents are a bit older on average, and children from poorer families are less likely to (continue to) participate, significant at the 10% level, but absolute numbers are low. Results are available upon request.

¹⁰: The wealth indicator was calculated based on the first loading of a factor analysis of dwelling characteristics (such as quality of the roof, walls and floor, number of rooms, presence of inside kitchen/bathroom/toilet facilities) and the possession of a large number of assets. This indicator was normalized to have mean zero and standard deviation equal to one.

¹¹: This is confirmed in a random effects panel regression of the child outcomes on baseline treatment assignment and the full set of other regressors.

¹²: Because of limited degrees of freedom the estimations focus on the variables with most explanatory power in the majority of specifications. Characteristics that are time-invariant, such as child sex, caregiver education and districts indicators, drop out of the fixed-effects estimation.

¹³: See for example Davis et al. (2009) or Durrant et al. (2010).

¹⁴: The total number of children participating in RCP at follow-up was much larger due to the enrollment of new-born children not included in the study.

¹⁵: All regressions correct for round and tester effects.

¹⁶: Data on attendance in the monthly parenting meetings are not available. Therefore, this component of the program cannot be included in the analysis.

¹⁷: Based on discussions with program coordinators.

¹⁸: The evidence with respect to most other program design features such as length of enrollment, target population, or single versus multiple site interventions were inconclusive, except for the intensity of the program (more visits, more hours of visits) and a focus on low-birth weight children, which both tended to increase effect sizes (Sweet & Appelbaum, 2004).

View Abstract

[1] Alderman et al., 2001
H. Alderman, J.R. Behrman, V. Lavy, R. Menon
Child health and school enrollment: A longitudinal analysis
Journal of Human Resources, 36 (1) (2001), pp. 185-205
Crossref Google Scholar

[2] Augustin, 2004
M. Augustin
Parenting support and early childhood intervention: A replication plan for Saint Lucia
Caribbean Child Support Initiative (CCSI), Barbados (2004)
Google Scholar

[3] Barrow, 2003
C. Barrow
Child rearing and socialisation approaches in the Caribbean—Phase I: Trinidad; Phase II: Dominica
Caribbean Child Support Initiative (CCSI), Barbados (2003)
Google Scholar

[4] Behrman et al., 2004
J.R. Behrman, Y. Cheng, P.E. Todd
Evaluating preschool programs when length of exposure to the program varies: A nonparametric approach
Review of Economics and Statistics, 86 (1) (2004), pp. 108-132
View in Scopus Google Scholar

[5] Berlinksi et al., 2009
S. Berlinksi, S. Galiani, P. Gertler
The effect of pre-primary education on primary school performance
Journal of Public Economics, 93 (2009), pp. 219-234
Google Scholar

[6] Charles, 2004
L. Charles
Early childhood policy
Government of Saint Lucia, Saint Lucia (2004)
Google Scholar

[7] Claessens et al., 2009
A. Claessens, G. Duncan, M. Engel
Kindergarten skills and fifth-grade achievement: Evidence from the ECLS-K
Economics of Education Review, 28 (2009), pp. 415-427
View in Scopus Google Scholar

[8] Cunha and Heckman, 2008
F. Cunha, J. Heckman
Formulating, identifying and estimating the technology of cognitive and non-cognitive skill formation
Journal of Human Resources, 43 (2008), pp. 738-782
View in Scopus Google Scholar

[9] Currie, 2001
J. Currie
Early childhood education programs
Journal of Economic Perspectives, 15 (2) (2001), pp. 213-238
Crossref View in Scopus Google Scholar

[10] Davis et al., 2009
R.E. Davis, M.P. Couper, N.K. Janz, C.H. Caldwell, K. Resnicow
Interviewer effects in public health surveys
Health Education Research, 25 (1) (2009), pp. 14-26
View in Scopus Google Scholar

[11] DeCicca and Smith, 2013
P. DeCicca, J. Smith
The long-run impacts of early childhood education: Evidence from a failed policy-experiment
Economics of Education Review, 36 (2013), pp. 41-59
View in Scopus Google Scholar

[12] Deming, 2009
D. Deming
Early childhood intervention and life-cycle skill development: Evidence from Head Start
American Economic Journal: Applied Economics, 1 (3) (2009), pp. 111-134
Crossref Google Scholar

[13] Durrant et al., 2010
G.B. Durrant, R.M. Groves, L. Staetsky, F. Steele
Effects of interviewer attitudes and behaviors on refusal in household surveys
Public Opinion Quarterly, 74 (1) (2010), pp. 1-36
Crossref View in Scopus Google Scholar

[14] Engle et al., 2011
P. Engle, L. Fernald, H. Alderman, J. Behrman, C. O’Gara, A. Yousafzai, et al.
Strategies for reducing inequalities and improving developmental outcomes for young children in low-income and middle-income countries
The Lancet, 378 (2011), pp. 1339-1353
View in Scopus Google Scholar

[15] Garces et al., 2002
E. Garces, D. Thomas, J. Currie
Longer-term effects of head start
American Economic Review, 92 (4) (2002), pp. 999-1012
View in Scopus Google Scholar

[16] Gertler et al., 2013
P. Gertler, J. Heckman, R. Pinto, A. Zanolini, C. Vermeerch, S. Walker, et al.
Labor market returns to early childhood stimulation
World Bank Policy Research Paper 6529
World Bank, Washington, DC (2013)
Google Scholar

[17] Grantham-McGregor et al., 2007
S.M. Grantham-McGregor, Y.B. Cheung, S. Cueto, P. Glewwe, L. Richter, B. Strupp, et al.
Developmental potential in the first 5 years for children in developing countries
The Lancet, 369 (2007), pp. 60-70
View in Scopus Google Scholar

[18] Groot Bruinderink and Janssens, 2010
Groot Bruinderink, M., & Janssens, W. (2010). Child Health and Nutrition in St Lucia. AIID Research Paper, Amsterdam Institute for International Development.
Google Scholar

[19] Hazarika and Viren, 2013
G. Hazarika, V. Viren
The effect of early childhood developmental program attendance on future school enrollment in rural North India
Economics of Education Review, 34 (2013), pp. 146-161
View in Scopus Google Scholar

[20] Heckman, 2006
J. Heckman
Skill formation and the economics of investing in disadvantaged children
Science, 312 (5782) (2006), pp. 1900-1902
Crossref View in Scopus Google Scholar

[21] Heckman and Kautz, 2013
J. Heckman, T. Kautz
Fostering and measuring skills: Interventions that improve character and cognition
NBER Working Paper 19656
National Bureau of Economic Research, Cambridge, MA (2013)
Google Scholar

[22] Loen et al., 2007
S. Loen, M. Bridges, D. Bassok, B. Fuller, R.W. Rumberger
How much is too much? The influence of preschool centers on children's social and cognitive development
Economics of Education Review, 26 (2007), pp. 52-66
Google Scholar

[23] Maluccio et al., 2009
J.A. Maluccio, J. Hoddinott, J.R. Behrman, R. Martorell, A.R. Quisumbing, A.D. Stein
The impact of improving nutrition during early childhood on education among Guatemalan adults
The Economic Journal, 119 (2009), pp. 734-763
Crossref View in Scopus Google Scholar

[24] Mullen, 1995
E.M. Mullen
Mullen scales of early learning, AGS edition: Manual
American Guidance Service (AGS), Inc., Circle Pines, MN (1995)
Google Scholar

[25] Nores and Barnett, 2010
M. Nores, W.S. Barnett
Benefits of early childhood interventions across the world: (Under) Investing in the very young
Economics of Education Review, 19 (2010), pp. 271-282
View in Scopus Google Scholar

[26] Paxson and Schady, 2007
C. Paxson, N. Schady
Cognitive development among young children in Ecuador: The roles of wealth, health and parenting
Journal of Human Resources, 42 (1) (2007), pp. 49-84
View in Scopus Google Scholar

[27] Powell, 2004
C.A. Powell
An evaluation of the roving caregivers programme of the rural family support organisation
University of the West Indies, Tropical Medicine Research Institute, Kingston, Jamaica (2004)
Google Scholar

[28] Powell et al., 2004
C.A. Powell, H. Baker-Henningham, S. Walker, J. Gernay, S. Grantham-McGregor
Feasibility of integrating early stimulation into primary care for undernourished Jamaican children: Cluster randomized controlled trial
British Medical Journal, 329 (89) (2004)
Google Scholar

[29] Roopnarine, 2005
J.L. Roopnarine
The Roving Caregivers Program in Jamaica: Its theoretical and research foundations and efficacy
Syracuse University, Syracuse, NY (2005)
Google Scholar

[30] Samms-Vaughan, 2004
M. Samms-Vaughan
Profiles—The Jamaican pre-school child: The status of early childhood development in Jamaica
Planning Institute of Jamaica, Kingston, Jamaica (2004)
Google Scholar

[31] Schady, 2006
N. Schady
Early childhood development in Latin America and the Caribbean
Economía, 6 (2) (2006), pp. 185-225
Crossref Google Scholar

[32] Schweinhart et al., 2005
L. Schweinhart, J. Montie, Z. Xiang, W.S. Barnett, C. Belfield, M. Nores
Lifetime effects: The high/scope Perry preschool study through age 40. Monographs of the High/Scope Educational Research Foundation 14
(2005)
Google Scholar

[33] Shonkoff and Phillips, 2000
J.P. Shonkoff, D.A. Phillips (Eds.), From neurons to neighborhoods: The science of early childhood development, National Academy Press, Washington, DC (2000)

[34] Super et al., 1990
C.M. Super, M. Guillermo Herrera, J.O. Mora
Long-term effects of food supplementation and psychosocial intervention on the physical growth of Colombian infants at risk of malnutrition
Child Development, 61 (1) (1990), pp. 29-49
View in Scopus Google Scholar

[35] Sweet and Appelbaum, 2004
M.A. Sweet, M.I. Appelbaum
Is home visiting an effective strategy? A meta-analytic review of home visiting programs for families with young children
Child Development, 75 (5) (2004), pp. 1435-1456
View in Scopus Google Scholar

[36] Walker et al., 2005
S.P. Walker, S.M. Chang, C.A. Powell, S.M. Grantham-McGregor
Effects of early childhood psychosocial stimulation and nutritional supplementation on cognition and education in growth-retarded Jamaican children: Prospective cohort study
The Lancet, 366 (9499) (2005), pp. 1804-1807
View in Scopus Google Scholar

[37] Williams and Brown, 2005
S. Williams, J. Brown
Research approaches and findings on childrearing and socialisation of young children (birth to eight years of age) in the Caribbean
Caribbean Child Support Initiative (CCSI), Bridgetown, Barbados (2005)
Google Scholar

[38] Williams et al., 2006
S. Williams, J. Brown, J. Roopnarine
Child rearing in the Caribbean: A literature review
Digital Solution, Kingston. Jamaica (2006)
Google Scholar

[39] Young, 2002
M.E. Young (Ed.), From early child development to human development, World Bank, Washington, DC (2002)

Outline 大纲

Cited by (7)

Tables (8)

Economics of Education Review
《教育经济学评论》

Highlights 突出

Abstract 抽象

JEL classification JEL 分类

Keywords 关键字

1. Introduction 1. 引言

2. The Roving Caregivers Program in St Lucia
2. 圣卢西亚的巡回看护人计划

2.1. Early childhood and parenting practices in the Caribbean
2.1. 加勒比地区的幼儿和育儿实践

2.2. Center-based ECD interventions in St Lucia
2.2. 圣卢西亚以中心为基础的 ECD 干预

2.3. Description of the Roving Caregivers Program
2.3. 巡回看护者计划的描述

3. Research methodology 3. 研究方法

3.1. Evaluation design 3.1. 评估设计

3.2. Data collection and measurement instruments
3.2. 数据收集和测量仪器

3.3. Survey response rates
3.3. 调查回复率

3.4. Comparison of the treatment and control group at baseline
3.4. 基线治疗组和对照组的比较

4. Econometric methodology

5. Enrollment in RCP after baseline

6. Impact results

6.1. Intention-to-treat effects

6.2. Treatment heterogeneity by child sex and age

6.3. Participation, age at enrollment and length of enrollment

6.4. Daycare and preschool enrollment

7. Discussion and conclusion

Acknowledgements

Appendix A. Comparison of baseline community characteristics between the treatment and the control group

References

Cited by (7)

Identifying the Common Elements of Early Childhood Interventions Supporting Cognitive Development in Low- and Middle-Income Countries

Validation of a developmental checklist for young (0-5) Surinamese children

Teaching home-visitors to support responsive caregiving: A cluster randomized controlled trial of an online professional development program in Brazil

Indian Academy of Pediatrics Position Paper on Nurturing Care for Early Childhood Development

Systematic Review and Meta-Analyses: Motor Skill Interventions to Improve Fine Motor Development in Children Aged Birth to 6 Years

Effects of home-based play-assisted stimulation on developmental performances of children living in extreme poverty: A randomized single-blind controlled trial

The impact of a Caribbean home-visiting child development program on cognitive skills加勒比地区家访儿童发展项目对认知技能的影响

Highlights 突出

Abstract 抽象

JEL classification JEL 分类

Keywords 关键字

1. Introduction 1. 引言

2. The Roving Caregivers Program in St Lucia2. 圣卢西亚的巡回看护人计划

2.1. Early childhood and parenting practices in the Caribbean2.1. 加勒比地区的幼儿和育儿实践

2.2. Center-based ECD interventions in St Lucia2.2. 圣卢西亚以中心为基础的 ECD 干预

2.3. Description of the Roving Caregivers Program2.3. 巡回看护者计划的描述

3. Research methodology 3. 研究方法

3.1. Evaluation design 3.1. 评估设计

3.2. Data collection and measurement instruments3.2. 数据收集和测量仪器

3.3. Survey response rates3.3. 调查回复率

3.4. Comparison of the treatment and control group at baseline3.4. 基线治疗组和对照组的比较

4. Econometric methodology

5. Enrollment in RCP after baseline

6. Impact results

6.1. Intention-to-treat effects

6.2. Treatment heterogeneity by child sex and age

6.3. Participation, age at enrollment and length of enrollment

6.4. Daycare and preschool enrollment

7. Discussion and conclusion

Acknowledgements

Appendix A. Comparison of baseline community characteristics between the treatment and the control group

References

The impact of a Caribbean home-visiting child development program on cognitive skills
加勒比地区家访儿童发展项目对认知技能的影响

2. The Roving Caregivers Program in St Lucia
2. 圣卢西亚的巡回看护人计划

2.1. Early childhood and parenting practices in the Caribbean
2.1. 加勒比地区的幼儿和育儿实践

2.2. Center-based ECD interventions in St Lucia
2.2. 圣卢西亚以中心为基础的 ECD 干预

2.3. Description of the Roving Caregivers Program
2.3. 巡回看护者计划的描述

3.2. Data collection and measurement instruments
3.2. 数据收集和测量仪器

3.3. Survey response rates
3.3. 调查回复率

3.4. Comparison of the treatment and control group at baseline
3.4. 基线治疗组和对照组的比较