The impact of daycare programmes on child health, nutrition and development in developing countries: a systematic review

Abstract

The authors conducted a systematic review of impact evaluations of daycare interventions on the health, nutrition and development of children under five in low-income and middle-income countries. The review was limited to intervention studies controlling for potential self-selection of beneficiaries. Papers were excluded based on study scope, type and quality. Only six studies (conducted in Latin America) met the inclusion criteria. Large positive effects on child development were found. Because of the limited evidence, no conclusions can be drawn regarding the impact on child health and nutrition. More rigorously conducted studies on the impact of daycare programmes in developing countries are needed.

Keywords:

• daycare
• nutrition
• health
• development
• review

Introduction and background

Many women in developing countries face the difficult challenge of competing demands on their time. Women are often in charge of activities related to childrearing and other time-intensive domestic activities such as collecting fuel-wood and fetching water. Additionally, they engage in non-domestic production activities, such as formal and informal labour market activities and agricultural work, resulting in high workloads (Engle et al. 1997). The situation of mothers has been coined as a zero-sum game: a new activity can only be added if she drops another activity or becomes more efficient in it (McGuire and Popkin 1989).

Many hypothesise that the higher unemployment rates and fewer working hours for females, as compared with their male counterparts, are – at least partly – due to those conflicting demands on women's time (Quisumbing et al. 2007). Urbanisation and increases in female labour market participation have increased the demand for non-parental childcare in many low-income and middle-income countries. Especially in formal sector jobs, work and caring for one's child cannot be combined. Government daycare programmes currently underway in several developing countries seek to promote labour force participation through relieving one of the most pressing constraints faced by working parents, especially mothers, which is access to reliable and affordable childcare.

The effects of maternal labour market participation and its effect on the child have been studied widely. Most studies focus on the trade-off between the potential positive effects of increased income earned by the mother and the negative effect of the decrease in time available for childcare. This model, however, ignores several important issues. First, there is the self-selection of women that decide (not) to work; second, contrasting working with non-working is a simplification, since working conditions may differ considerably relative to providing childcare; finally, the availability and quality of alternative childcare is an important variable as well (Engle and Pedersen 1989).

Zoritch et al. (2000) conducted a systematic review of randomised and quasi-randomised controlled daycare trials. The review was limited to studies conducted in the United States. Based on eight published trials, the authors concluded that daycare had positive effects on child development and mothers' interaction with their children. Few studies looked at health outcomes. The authors emphasised the urgent need for higher quality studies in this area. No systematic review of the evidence from studies in low-income and middle-income countries has been conducted. The aim of this review was to fill in this gap in the literature.

We conducted a systematic review of the evidence regarding the impact of daycare on the health, nutrition and development of children under five years of age in low-income and middle-income countries. We used a programme impact theory approach to identify the pathways through which daycare may improve child outcomes.

Methods

The definition of daycare

Daycare is defined as any type of institutional out-of-home care for children younger than five years of age, independent of who provided the daycare (government, private or a combination of both). Daycare can range from a service just providing a safe place where children are taken care of while their mothers work, to preschool services that offer health, nutrition and education components.

It is important to note that the focus of our review is different from that of a number of recent reviews on early childhood development (ECD) interventions (Engle et al, 2007, Baker-Henningham and Lopez-Boo 2010, Nores and Barnett 2010). The objective of ECD interventions is to promote child development through a variety of activities, such as improving nutritional status and health, reducing the social, environmental and infectious risks children are exposed to, and providing cognitive and socio-emotional stimulation. ECD interventions can be offered in different settings, including home-based, centre-based, formal preschool, and parent/community-based arrangements. The interventions we reviewed may or may not have specific ECD components in place. Common to all of them, however, is that they provide formal out-of-home care for children under the age of five that allows parents to participate in the labour market.

Inclusion and exclusion criteria

The studies selected for our review included intervention studies (defined as studies evaluating the impact of an exogenous change in daycare provision or utilisation on child outcomes) using quasi-experimental methods (for example, propensity score matching), as well as regression-based methods to control for potential self-selection (endogeneity) of programme participation.

We limited the review to papers published after 1980. No language restrictions were imposed. The selection of articles for full review from the pool obtained during the database searches was conducted in three phases, with increasing specificity. Papers were excluded based on study scope, type and quality (Table 1). We first scanned the title to exclude studies that were out of the scope of the review (see below). We then read the abstract and excluded papers not meeting the criteria for study scope, type and quality. To avoid incorrectly excluding studies that met the inclusion criteria, we erred on the side of caution: when in doubt we always decided to review the full text. The searches were conducted in parallel by two researchers; that is, two authors independently reviewed all identified titles and abstracts. The full text review was also conducted independently by both authors. Any disputes were resolved by a third author.

Table 1. Overview of exclusion criteria

Exclusion criterion	Study excluded if …
Study scope	(a) Study did not evaluate the impact of a daycare intervention on child health, nutrition or development, comparing (groups of) children receiving this intervention with (groups of) children not receiving it.
	(b) Study's main focus was on disabled children, orphans, children living with HIV/AIDS, child abuse, malnourished children, and so forth.
	(c) Study only described the situation in daycare centres or the quality of services provided.
Study type	(a) Editorial, commentary or book review.
	(b) Policy document.
	(c) Survey solely reporting the prevalence or incidence of the use of daycare centres or the prevalence or incidence of diseases, malnutrition, dietary patterns of children attending daycare centres.
	(d) Non-evaluated intervention.
	(e) Theoretical or methodological study.
	(f) Single-case study.
	(g) Study that only evaluated the process of an intervention.
Study quality	(a) Study that did not provide sufficient details about:
	• the intervention (including setting, beneficiary population, benefits);
	• sample characteristics (age, sample size);
	• study type and analytic model; statistical tests on the studied associations; coefficients and level of significance reported numerically or graphically; and
	• detailed description of the methodology used to control for confounding and selection bias, that is providing details on the instruments used in case of instrumental variables estimation or the variables used for matching in case of (propensity) matching.
	(b) Study only used univariate methods for data analysis.

Data sources and search strategies

Our primary searches were conducted in PubMed and EconLit, using a wide range of terms intended to maximise the number of potentially relevant articles identified. In addition, we searched a number of databases including grey literature (IDEAS, POPLINE, the British Library for Development Studies and Literatura Latinoamericana y del Caribe en Ciencias de la Salud) and the websites of the World Bank (including the JOLIS catalogue) and the International Food Policy Research Institute.

The search strategy combined a group of ‘exposure’ terms (representing the daycare programmes), with ‘outcome’ terms (child health, nutrition and development), ‘subject’ terms (children) and ‘context’ terms (developing countries) (see Table 2). In the final search syntax, exposure terms were combined with the outcome, subject and context terms by the Boolean operator ‘AND’ (that is, ‘exposure terms’ AND ‘outcome terms’ AND ‘context terms’). Search terms within these four domains were connected by the Boolean operator ‘OR’. Where spellings differed between British and American English, we searched for both. Full details on the search procedures and the final search strings used for each database have been published elsewhere (Leroy et al. 2011).

Table 2. Overview of search terms

Domain	Description	Search terms
Exposure	Intervention	Child daycare centers, child care, infant care, child daycare, nursery*, nursery school*, pre-school*, kindergarten
Outcome	Health	Health, infection*, disease*, morbidity, diarrhea, diarrhoea
	Nutrition	Diet, nutrition, growth, height, weight, anemia, stunting, wasting, deficiency* nutrient*
	Development	Intelligence, infant development, child development, psychomotor, motor, sensory, language, social, emotional, cognitive, cognition
Subjects	Children	Child, children, preschool child, preschool children, infant*
Context	Low-income and middle-income countries	All names of low and middle income countries and the following terms: developing country*, third world country*, low income country*, middle income country*, developing nation*
*Refers to the same word in plural.

A number of complementary strategies were used to identify additional studies. We conducted citation tracking (back and forward referencing of the selected studies) in Google Scholar and Web of Science; we hand-searched the journals (and websites) where the selected studies were published. Since the oldest selected study was published in 2004, these hand searches were limited to papers published after 1999. A regular Google search (limited to the first 500 hits) was conducted as well. We finally added papers suggested by colleagues and international experts in the field, as well as papers identified through reviewing the reference list of non-systematic reviews; we also contacted the authors of the selected papers regarding published or unpublished work relevant for this review.

To obtain complementary information on the characteristics of the evaluated programmes and their beneficiaries and to identify any other information relevant for the pathway analysis, we searched for ‘sister studies’ (such as process evaluations, qualitative studies, and so forth) of the selected impact evaluations. To identify these studies, we contacted the authors of the selected papers and conducted a regular Google search (limited to the first 50 hits sorted by relevance) using the name of the programme as the search term.

Data summary and analysis

The studies that met the inclusion criteria were reviewed in depth. The characteristics of the articles were described and their results were tabulated. Data were summarised into the following categories: country, intervention (including the eligibility criteria), sample characteristics (data sources and years, age of the children, sample size), evaluation design and analytic method (definition of treatment, outcomes, analyses), study quality assessment and results by outcome (health, nutrition and development).

After reviewing the evidence on the impact of daycare on child nutrition, health and development, we used a ‘programme theory’ approach to identify the pathways through which daycare may improve child outcomes. Programme theory refers to the pathways by which a programme is intended to achieve its impacts (Rossi et al. 2004). Identifying these pathways allows for the best understanding of whether, why, and how a programme achieved, or did not achieve, its intended impacts. The programme theory presented in Figure 1 is based on Ruel's framework (Ruel 2008). While the framework refers explicitly to the mother as the primary caregiver, it is applicable to cases where the primary caregiver is another household member. Maternal employment may increase household income, which might increase household food security and allow households to purchase better (more nutrient dense) and more foods. This, in turn, could lead to improved child dietary intake. Maternal employment may also increase women's control over income and their decision-making power with respect to child feeding and health. The meals provided to children while attending daycare may directly – either positively or negatively – affect the adequacy of child dietary intake. The hygiene and sanitation levels of the daycare attended might lead to changes in childhood illness, especially infectious diseases. Additionally, daycare may increase the risk of exposure to communicable diseases through contact between children. Daycare safety determines the odds of accidents and consequently child health. The potential impacts on child development depend – in part – on the quality of the educational activities provided. Social interaction may have a positive impact on child development as well, but may also lead to more exposure to communicable diseases. Maternal employment may have a negative effect on women's time, which, in turn, might limit her ability to provide adequate care. Finally, the services provided at daycare might affect the level of care provided at home. For instance, the food offered to children in daycare might cause parents to offer less (or lower quality) foods to children at home (so-called substitution). Daycare might have similar effects on the level of hygiene, sanitation and child stimulation provided at home. For instance, parents may consider the amount of stimulation their children receive in daycare to be sufficient and hence reduce or even stop engaging in these types of activities at home. Alternatively, parents might learn from the services provided at daycare and improve the care environment and practices at home.

Figure 1. Mechanisms by which daycare programmes might affect child nutritional status, health and development.

Results

Study selection

A total of 13,190 papers were identified through our searches (Figure 2). After reviewing titles and abstracts, 27 articles were selected for a full text in-depth review. Of those, 21 studies were excluded because they did not meet the inclusion criteria. From the back and forward citation tracking of the selected studies, six additional studies were identified for review, but none met the inclusion criteria. As such, only six studies were included in the review.

Figure 2. Study selection process.

Brief description of the evaluated programmes

All six studies evaluated the impact of programmes in Latin American countries. Three of the interventions (Colombia, Bolivia and Guatemala) used a community-based approach in which women from the community provided daycare in their homes. The programmes in Argentina and Uruguay used an institutional approach through the provision of preschool education. The studies are briefly described below. More details have been reported elsewhere (Leroy et al. 2011).

Community-based interventions

Attanasio et al. (2010) assessed the impact of the Hogares Comunitarios de Bienestar (HCB) programme in Colombia. HCB targets children under the age of six and provides, for a small fee, daycare and food (70% of nutrient requirements) at the house of a madre comunitaria. Bernal et al. (2009) assessed the impact of the same HCB programme. The impact of the Bolivian Programa Integral de Desarrollo Infantil (PIDI) was estimated by Behrman et al. (2004). PIDI provides fully subsidised daycare to children between six and 72 months. Services are provided in the homes of women in low-income areas and include food (70% of nutrient requirements) as well as health and nutrition monitoring and educational activities. The third community-based intervention was Guatemala's Hogares Comunitarios (HC) programme evaluated by Ruel et al. (2006). HC services include full-time daycare provided by women in the community, food (80% of nutrient requirements), psycho-pedagogical activities and general care and hygiene.

Institutional daycare

Berlinski et al. (2009) evaluated the impact of the expansion of public pre-primary provision (through preschool classroom construction) in Argentina from 1993 to 1999. The second study by Berlinski et al. (2008) evaluated the impact of the expansion of public preschool services (through classroom construction) in Uruguay.

A wide variety of estimation strategies were used. Statistical methods included instrumental variable estimation and different types of (propensity score) matching. The impact was measured by comparing children in the programme with different types of comparison groups: (eligible) children who self-selected out of the programme, (eligible) children living in areas where the programme was not offered, children who (very) recently joined the programme, and siblings who did not participate in the programme. Complete details on the estimation approach used in each study have been published elsewhere (Leroy et al. 2011).

Effectiveness results

Impact on child health

The only study (Table 3) including child health outcomes was the second evaluation of the Colombia programme. Significant reductions in the prevalence of diarrhoea (up to –6.9 percentage points) and acute respiratory infections (up to –3.4 percentage points) were found for children with longer exposures to the programme. It must be noted that the diarrhoea and acute respiratory infections effects can be explained in two different ways. They could represent a true positive health effect of the programme. Alternatively, the children who just entered the programme (that is, with less than one month of exposure) and who served as the comparison group for these analyses might have suffered from a steep increase in diarrheal and respiratory infections right after entering daycare. As children stayed in the programme longer, this negative effect might have slowly disappeared, which would explain the positive programme effect in the older age groups. The programme had an unexpected negative impact of up to –3.9 percentage points on the proportion of children who were completely immunised (Bernal et al. 2009, Universidad de los Andes and Profamilia 2009).

Table 3. Summary of included studies: health outcomes.^a

Country Reference	Intervention	Comparison	Study quality assessment	Outcome measure	Descriptive statistics	Effect size (95% CI)^b
ColombiaBernal et al. (2009)	Hogares Comunitarios de Bienestar (HCB). Madres comunitarias (MC) provide childcare in their homes for a small monthly fee (3.5 USD/mo in 2008) to groups from 12 to 14 children 0 to 6 y. Care includes lunch and snacks (50% to 70% of caloric and nutritional needs). There are two modalities full time or part time. Eligibility criteria: households with children 0 to 6 y belonging to SISBEN levels 1 to 2.	Treatment: 2–4, 5–15 and 16+ mo of attendance (vs. <1 mo)^c Analyses: propensity score matching (kernel) Medium term impact of the programme on test scores in 5^th grade, using instrumental variables.	• Matched treatment- control estimates suffer from selection bias; • Not clear how subsample was selected; • Relatively low prevalence of malnutrition, so potential to benefit is low; • Positive effect on diarrhoea and ARI might be due to an increase in morbidity in children just after joining daycare	ARI, complete vaccination scheme.	Diarrhoea: 15.65% ARI: 5.33% Complete vaccination scheme: 20.05%	Diarrhoea: 0–24 mo: –6.9pp (–12.8, –1.0) if exposure 16+ mo; 25–36 mo: –4.2pp (–8.1, –0.3) if exposure 5–15mo and –6.8pp (–10.7, –2.9) if exposure 16+ mo; 49+ mo: –4.2pp (–4.6, –3.8) if exposure 16+ mo; all other age/exposure combinations: NS. ARI: 0–24 mo: –3.4pp (–3.6, –3.2) if exposure 16+ mo; 25–36 mo: –3pp (–5,–1) if exposure 16+ mo; all other age/exposure combinations: NS. Complete vaccination: Age 0–24 mo: –3.9pp (–7.9, 0.0) if exposure 16+ mo. Age 25–36 mo: +2.9pp (0.9, 4.9) if exposure 16+ mo.
^aAdditional study details are provided in Leroy et al. (2011). ARI, acute respiratory infections; NS, not significant; pp, percentage points; SISBEN, Identification System for Potential Beneficiaries of Social Programs (Sistema de Identificación de Potenciales Beneficiarios de Programas Sociales). SISBEN is an indicator of economic well-being; indicator ranges from 1 (poorest) to 6.
^b95% confidence intervals (CI) were calculated as follows: estimated impact ± 1.96 × reported standard error.
^cProgramme impact was estimated using three different comparisons: treatment versus control, children exposed for <1, 2–4, 5–15 and over 16 months versus control and children exposed ≤1 month versus 2–4, 5–15 and over 16 months. The results strongly indicate that the impact estimates using the control group as the counterfactual suffer from selection bias. This was confirmed by the author (personal communication with Raquel Bernal, December 2010). We only report and discuss the results obtained from the estimations using the <1 versus >1 exposure comparison.

Impact on child nutrition

Three studies – Colombia (Universidad de los Andes and Profamilia 2009, Attanasio et al. 2010) and Bolivia (Behrman et al. 2004) – estimated the impact on child anthropometry (Table 4). Current attendance at daycare in Colombia assessed by Attanasio et al. (2010) had an estimated impact of 0.448 standard deviations (SD) on the height-for-age z-score (HAZ) of children between the age of zero and six years in rural areas and 0.826 SD in urban areas. Full exposure (defined as a –hypothetical – child that spent the first 72 months of its life in the HCB) was estimated to have an effect on the child HAZ of 0.945 SD in rural areas and 1.227 SD in urban areas. When estimating the effects by child height quintiles, the authors found larger effects for shorter children. The impact estimates in this study are biologically implausible. They are considerably larger than the effects of interventions specifically aimed at improving child growth. Additionally, the average age of children in the study was 49 and 36 months for the rural and urban samples, respectively. It is well known that the growth effect of nutrition interventions is mostly limited to children younger than 24 months (see, for instance, Schroeder et al. 1995). A second limitation of this study is the diffculty to interpret the exposure variable (defined as the proportion of the child's life spent in daycare). It is unlikely that the effect of daycare on child growth is linear with respect to the proportion of the child's life spent in daycare, rather than with the absolute amount of time (Attanasio et al. 2010). The second evaluation of the same programme found no consistent nutrition effect. The only significant effect on stunting (a 6.3 percentage point increase) was found in children who were zero to 24 months old with two to four months of exposure to the programme. It is very unlikely that the programme had such an impact on the height of children within a matter of months. The prevalence of being underweight dropped by 2.6 percentage points among children between 25 and 36 months with five to 15 months of exposure to the programme and by 2.1 percentage points for children between 37 and 48 months with two to four months of exposure. The prevalence of wasting did not change as a consequence of the programme (Bernal et al. 2009, Universidad de los Andes and Profamilia 2009).

Table 4. Summary of included studies: nutrition outcomes.^a

CountryReference	Intervention	Comparison	Study quality assessment	Outcome measure	Descriptive statistics	Effect size (95% CI)^b
ColombiaAttanasio et al. (2010)	Hogares Comunitarios de Bienestar (HCB). Madres comunitarias provide childcare in their homes for a small fee to groups of up to 15 children 0 to 6 y. Care includes lunch, 2 snacks and a nutritional beverage (70% of caloric needs) Eligibility criteria: households with children 0 to 6 y belonging to SISBEN levels 1 to 2 (SISBEN is an indicator of economic well-being; indicator ranges from 1 (poorest) to 6).	Treatment: current attendance (vs. not attending) and exposure to HCB (fraction of life spent in HCB) Analyses: instrumental variable regression. Overall impact and impact by child height quintiles.	• The FeA sample only includes SISBEN 1 households; • Not clear how many children were excluded from the analyses; • Effect sizes are biologically implausible; • Unlikely that programme effect is linear with respect to exposure (defined as the proportion of the child's life spent in daycare); • Not clear how 3 rounds of panel data were modelled.	HAZ	HAZ: Rural: –1.25 SDUrban: –0.77 SDStunting: Rural: 23.7% Urban: 11.2%	Overall: Current attendance: Rural: +0.448SD (0.076, 0.820) Urban: +0.826SD (0.450, 1.202) Exposure (=1 if exposed entire life): Rural: +0.945SD (0.228, 1.662) Urban:+1.227SD (0.512, 1.942) Height quintile results: Current attendance: Rural: quintile 1 +0.65SD (0.081, 1.217) to NS in quintile 5 Urban: quintile 1 +1.33SD (0.767, 1.895) to NS in quintile 5 Exposure (=1 if exposed entire life): Rural: +1.737 SD (0.044, 3.43) in quintile 1 to NS in quintile 5 Urban: +3.419SD (1.312, 5.526) in quintile 1 to NS in quintile 5.
Colombia Bernal et al. (2009)	Hogares Comunitarios de Bienestar (HCB). Madres comunitarias (MC) provide childcare in their homes for a small monthly fee (3.5 USD/mo in 2008) to groups of 12 to 14 children 0 to 6 y. Care includes lunch and snacks (50% to 70% of caloric and nutritional needs). There are two modalities: full time or part time. Eligibility criteria: households with children 0 to 6 y belonging to SISBEN levels 1 to 2.	Treatment: 2–4, 5–15 and 16+ mo of attendance (vs. <1 mo)^c Analyses: propensity score matching (kernel) Medium term impact of the programme on test scores in 5^th grade, using instrumental variables.	• Matched treatment- control estimates suffer from selection bias; • Not clear how subsample was selected; • Relatively low prevalence of malnutrition, so potential to benefit is low; • Positive effect on diarrhoea and ARI might be due to an increase in morbidity in children just after joining daycare	Stunting (HAZ<–2SD), underweight (WAZ<–2SD) and wasting (WHZ<–SD);	Stunting: 12.96% Underweight: 7.02% Wasting: 1.59%	Stunting: 0–24 mo: +6.3pp (2.4, 10.2) if exposure 2–4 mo; all other age/exposure combinations: NS; Underweight: 25–36 mo: –2.6pp (–4.6, –0.6) if exposure 5–15 mo; 37–48 mo:–2.1pp (–4.1, –0.1) if exposure 5–15 mo; all other age/duration combinations: NS; Wasting: NS
Bolivia Behrman et al. (2004)	Programa Integral de Desarrollo Infantil (PIDI) Childcare in homes of women in low-income areas; up to 15 children per PIDI, 1 staff/5 children. Food provided (70% of nutritional needs), health and nutrition monitoring and educational activities. Full time, no fees chargedEligibility criteria^d: Children 6–72 mo, poor urban areas, household eligibility evaluated using an index based on housing characteristics and working status	Treatment: attendance (participants (a) vs. eligible non participants living close (b) or in areas not served by programme (c)), cumulative impact (1–6, 7–12, 13–18, 19–24 and 25+ mo; comparisons as for attendance), ≤1 mo vs. ≥2 moAnalyses: generalised matching estimator	Not clear how many children were excluded from the analyses	HA%, WA%	Reported in graphs. HA: Overall mean around 20^th percentile. PIDI children 12mo shorter than (b) and (c) comparison children. WA: Overall mean around 35^th percentile. No apparent differences between groups.	(a) vs (b) attendance & cumulative: NS(a) vs (c) attendance: HA%: NSWA%: all ages negative (–5.0 (–9.9, –0.1) to –7.5pp (–12.8, –2.2)), except for for 42–58mo (NS) (probably due to residual selection bias) cumulative: HA%: NSWA%^e: –6.8 to –8.9 pp for short duration (<12 mo) of exposure (probably due to residual selection bias) ≤1 mo vs ≥2 mo attendance & cumulative: NS
GuatemalaRuel et al. (2006)	Hogares Comunitarios (HC) Daycare provided by women in the community to up to 10 children 0 to 7 y of ageMonday to Friday from 6 a.m. to 6 p.m. Care includes breakfast, lunch, and 2 snacks (80% of nutrient requirements); psycho-pedagogical activities, general care and hygiene. Eligibility criteria: Children 2 to 7 y, mothers working outside the home, 1 urban zone in Guatemala City.	Treatment: attending daycare (vs. not attending) Analyses: matching by design (at the time of subject selection), propensity score matching (PSM, NN and kernel)	Not clear to what extent limiting the PSM to the common support affected the internal validity of the results	Percentage of child daily energy and nutrient requirements (protein, Ca, Fe, Vit C, Vit A, Zn)	Results presented in graphs	Match by design: Results presented in graphs (not tables); similar to PSM results. PSM (kernel)^f Weekdays, during stay in place of care: Energy: +12.4pp (7.9, 16.2); Protein: +25.7pp (17.1,34.4); Ca: NS; Fe: +21.6pp (7.8, 30.9); Vit C: NS; Vit A: +85.2pp (48.6, 126.4); Zn: NSWeekends: Energy: 23.6pp (5.5, 41.3); Protein: +57.5 (13.5, 96.3)pp; Ca: NS; Fe: NS; Vit C: NS; Vit A: NS; Zn: NSWeekdays, before and after stay in place of care: Energy: +5.3pp (–0.1,9.4); Protein: NS; Ca: NS; Fe: +13.4pp (3.6, 20.2); Vit C: 14.7pp (3.2, 29.5); Vit A: NS; Zn: NSNo evidence of substitution home diet
^aAdditional study details are provided in Leroy et al. (2011). NS, not significant; pp, percentage points; PSM, propensity score matching; NN, nearest neighbour; SD, standard deviation; SISBEN, Identification System for Potential Beneficiaries of Social Programs (Sistema de Identificación de Potenciales Beneficiarios de Programas Sociales, SISBEN is an indicator of economic well-being; indicator ranges from 1 (poorest) to 6.); vit, vitamin; HAZ height-for-age Z-score; WAZ weight-for-age Z-score; WHZ weight-for-height Z-score; HA%, height-for-age percentile; WA%, weight-for-age percentile.
^b95% confidence intervals (CI) were calculated as follows: estimated impact ± 1.96 × reported standard error.
^cProgramme impact was estimated using three different comparisons: treatment versus control, children exposed for <1, 2–4, 5–15 and over 16 months versus control and children exposed ≤1 month versus 2–4, 5–15 and over 16 months. The results strongly indicate that the impact estimates using the control group as the counterfactual suffer from selection bias. This was confirmed by the author (personal communication with Raquel Bernal, December 2010). We only report and discuss the results obtained from the estimations using the <1 versus >1 exposure comparison.
^dEligibility criteria changed. The latest criteria are reported here. The original criteria included child malnutrition (no objective measure was used to verify it).
^e95% CI could not be calculated as the authors do not report standard errors for the cumulative effects.
^fNearest neighbour estimates had a substantially lower sample size.

No programme impact on growth was found in Bolivia (Behrman et al. 2004). When comparing participating children with eligible children in areas not served by the programme, a negative impact (seven to nine percentage points) on weight-for-age was found for short programme exposure (less than 12 months). The authors believe this negative effect was due to residual selection bias. The programme's initial eligibility requirements included child malnutrition (subjectively assessed) as a criterion. As a consequence, the programme might have initially enrolled many malnourished children.

The Guatemala study looked at the impact of daycare on child dietary intake on weekdays while attending daycare, on weekdays before and after the children stayed in daycare and during weekends (Ruel et al. 2006). The programme had a positive effect on the percentage of daily requirements consumed while at daycare: energy increased by 12 percentage points, protein by 26 percentage points, iron by 22 percentage points and vitamin A by 85 percentage points. Additionally, the programme improved the diet while not at daycare. Positive effects were found for the times before and after the children's stay in daycare on weekdays: energy increased by five percentage points, iron by 13 percentage points and vitamin C by 15 percentage points. These results show that the positive effect on child dietary intake while at daycare was not attenuated by substitution at home. During the weekends, the percentage of daily requirements consumed increased by 24 and 57 percentage points for energy and protein, respectively. Potentially worrisome is the finding that close to one-half of the effect on vitamin A intake could be attributed to the consumption of sugar (which is fortified with vitamin A in Guatemala) in sweetened homemade fruit juices in the daycare centres.

Impact on child development

Three studies (Table 5) evaluated the impact on child development – Colombia (Bernal et al. 2009), Bolivia (Behrman et al, 2004) and Argentina (Berlinski et al. 2009) – and one estimated the effect on schooling– Uruguay (Berlinski et al. 2008). The programme in Colombia had a generally positive effect on child development (Bernal et al. 2009, Universidad de los Andes and Profamilia 2009). Psychosocial development was measured using the Early Development Instrument (EDI, section on social and emotional development) and the Penn Interactive Peer Play Scale (PIPPS). No effect was found on the EDI. The impact on the PIPPS (scale range: one to four) was mixed: aggressive behaviour increased by 0.085 points (equivalent to a 5% increase using the overall reported mean) in children between the age of 36 and 48 months who had attended the programme for more than five months. Social isolation, however, dropped by 0.128 points (9%) and adequate interaction increased by 0.240 points (8%) in the same children. The authors explained the results as follows: children in a HCB were exposed to more interactions with other children and while they were learning to ‘negotiate’ (competition over toys, turns, and so forth), it is ‘normal’ to observe more aggressive behaviour in children. Cognitive development was measured using the EDI instrument (section on language and cognitive skills; range zero to one), the Peabody Picture Vocabulary Test and the Woodcock Johnson–Muñoz test. Positive impacts were found on the EDI but only among children older than 49 months; the impacts were larger with longer exposure to the programme (0.027 [10%], 0.029 points [10%] and 0.096 points [34%] for an exposure of two to four months, five to 15 months and more than 16 months, respectively). The positive impact on the Peabody Picture Vocabulary Test (scale range: 55–145; population level mean 100 [standard deviation 15]) was limited to children who had been in the programme for over 16 months (2.18 points [2.4%] for children aged 36–48 months and 4.8 points [5%] for children aged over 49 months). Finally, the Woodcock Johnson–Muñoz test (population level mean 100 [standard deviation 15]) showed that the programme had no impact on intellectual ability, a positive impact (3.48 points or 4%) on verbal ability among children 49 months or older who had been in the programme for over 16 months and similar positive impacts on mathematical reasoning and general knowledge among children over 36 months with more than 16 months of exposure. Finally, the authors also estimated medium-term impacts on test scores at fifth grade of primary school. They found an increase of 14.5 percentage points on total test scores.

Table 5. Summary of included studies: development outcomes.^a

CountryReference	Intervention	Comparison	Study quality assessment	Outcome measure	Descriptive statistics	Effect size(95% CI)^b
Colombia Bernal et al. (2009)	Hogares Comunitarios de Bienestar (HCB). Madres comunitarias (MC) provide childcare in their homes for a small monthly fee (3.5 USD/mo in 2008) to groups of 12 to 14 children 0 to 6 y. Care includes lunch and snacks (50% to 70% of caloric and nutritional needs). There are two modalities: full time or part time. Eligibility criteria: households with children 0 to 6 y belonging to SISBEN levels 1 to 2.	Treatment: 2–4, 5–15 and 16+ mo of attendance (vs. <1 mo)^c Analyses: propensity score matching (kernel) Medium term impact of the programme on test scores in 5^th grade, using instrumental variables.	• Matched treatment- control estimates suffer from selection bias; • Not clear how subsample was selected; • Relatively low prevalence of malnutrition, so potential to benefit is low; • Positive effect on diarrhoea and ARI might be due to an increase in morbidity in children just after joining daycare	Early Development Instrument (EDI), Penn Interactive Peer Play Scale (PIPPS): Peabody Picture Vocabulary Test (Peabody), Woodcock Johnson-Muñoz (WJM); and test scores in 5^th grade.^f	EDI social:1.55 PIPPS: aggressive behaviour: 1.83; social isolation: 1.40; adequate interaction: 3.05 EDI cognitive: 0.28 Peabody: 90.53 WJM: intellectual ability: 86.45; verbal ability: 80.89 mathematical reasoning: 82.00; general knowledge: 78.42 Total test score in 5^th grade: 36.36%	EDI social: NSPIPPS: Aggressive behaviour: 36–48 mo: +0.085 (0.026, 0.144) in the index if exposure 5–15 mo and +0.085 (0.007, 0.163) if 16+ mo. Social isolation: 36–48 mo: –0.128 (–0.167, –0.089) if exposure 16+ mo. Adequate interaction: 36–48 mo: +0.240 (0.162, 0.318) if exposure 16+ mo. EDI cognitive: 49+ mo: +0.027 (0.007, 0.047) to +0.029 (0.009, 0.049) if exposure 2–15mo; +0.096 (0.076, 0.116) if exposure 16+ mo. Peabody: 36–48 mo:+2.18 (0.39, 3.96) if exposure 16+ mo; 49+ mo: +4.80 (2.21, 7.38) if exposure 16+ mo. WJM: Intellectual ability: NS. Verbal ability: 49+ mo: +3.48 (0.63, 6.32) if exposure 16+ mo. Mathematical reasoning: 36+ mo: +3.80 (1.62, 5.98) to +4.26 (1.46, 7.07) if exposure 16+ mo. General knowledge: 36+ mo: +2.17 (0.19, 4.15) to +3.36 (1.11, 5.62) if exposure 16+ mo. Total test score in 5^th grade: +14.5pp (2.9, 26.1)
Bolivia Behrman et al. (2004)	Programa Integral de Desarrollo Infantil (PIDI) Childcare in homes of women in low-income areas; up to 15 children per PIDI, 1 staff/5 children. Food provided (70% of nutritional needs), health and nutrition monitoring and educational activities. Full time, no fees chargedEligibility criteria^e: Children 6–72 mo, poor urban areas, household eligibility evaluated using an index based on housing characteristics and working status	Treatment: attendance (participants (a) vs. eligible non participants living close (b) or in areas not served by programme (c)), cumulative impact (1–6, 7–12, 13–18, 19–24 and 25+ mo; comparisons as for attendance), ≤1 mo vs. ≥2 moAnalyses: generalised matching estimator	Not clear how many children were excluded from the analyses	Bulk motor, fine motor, language and auditory, and psycho-social skills.	Test scores: Reported in graphs. No apparent differences between groups.	(a) vs. (b) attendance: age 37–58 mo: all scores +2 to 6%; other ages: mostly NScumulative^f: age 25–58 mo: mostly positive (2 to 11%) if duration 13 mo(a) vs. (c) attendance: age 42 to 59+ mo: all scores +3 to 8%; other ages: mostly NScumulative^f : all ages: mostly positive (2 to 11%) if duration 7 mo≤1 mo vs. ≥2 mo attendance: age 42 to 59+ mo: all scores +2 to 9%; younger ages: mostly NScumulative^g: all ages: mostly positive (2 to 10%) if duration 13 mo; lower duration, some positive
Argentina Berlinski et al. (2009)	Expansion of public pre-primary school provision (through construction of classrooms) from 93 to 99. Pre-primary for children 3 to 5, 3.5 hrs/day, two shifts (am and pm), 5 days a week, 9 mo school year. Class size: 25 children. Eligibility criteria: Children 3 to 5y. Last year of pre-primary school compulsory since 1993.	Treatment: supply of pre-primary schools, attending pre-primary school. Analyses: Retrospective. Used exogenous increase in pre-primary school availability (construction of 3724 classrooms from 93 to 99) as treatment variable; interacted with child gender and municipality specific poverty level to estimate effect modification.	Not clear how many children were excluded from the analyses	Spanish and mathematics test scores; teachers' perception of student behaviour (attention, effort, class participation, discipline)	Mathematics (sd): 61.14 (19.70) Spanish (sd): 62.79 (20.41) Behaviour: Half or more of my students… Pay a lot of attention: 89.9% Put a lot of effort: 87.1% Are well disciplined: 75.6% Participate regularly: 83.5%	Mathematics: Overall: +4.69 (0.90, 8.48) (+8%) Gender interaction: NSPoverty interaction: NSSpanish: Overall: +4.76 (0.69, 8.83) (+8%) Gender interaction: NSpoverty interaction: higher if poorerBehaviour: Half or more of my students… Pay a lot of attention: +12.2pp (0.0, 24.4) Put a lot of effort: +21.1pp (3.7, 38.5) Are well disciplined: NSParticipate regularly: +16.5pp (2.2, 30.8)
Uruguay Berlinski et al. (2008)	Expansion of public pre-primary school classrooms 1995–2002. Pre-primary services for 3 to 5 y olds; 4 hrs/day, two shifts (am and pm), 5 days a week during 180 day school term. The programme aim was to promote child's early socialisation and alphabetisation. Eligibility criteria: Children 3–5 y. Pre-primary school is compulsory for children 4 to 5 y.	Treatment: attended <1 y vs. ≥1 y, ≥2 vs. 1 and 3 vs. 2 y of preschoolAnalyses: within household estimator, using variability between siblings and controlling for birth order, sex, mother´s age and education, living in Montevideo or not, household fixed effects. Additionally, preschool attendance instrumented with average locality and cohort-specific attendance rates.	• Not clear how many children were excluded from the analyses; • No gender specific attendance and grade progression descriptive statistics provided, so impossible to evaluate to what extent the difference in effects by maternal education and location are due to pre-existing differences;	School attendance, years of schooling.	School attendance: 0.97 Years of schooling (primary & secondary): 4.59	School attendance: OLS: <1 y vs. ≥1: effect grows with age from +4.3 (0.4, 8.2) at 7y to + 27.4pp (16.4, 38.4) at 15y; larger for low educated moms +26.9 (14.0, 39.8) vs. NS at 15y, outside Montevideo (+34.2 (21.7, 46.7) vs. +20.3pp (4.0, 36.6) at 15y); results by sex NSIntensive margin: Returns to additional years of preschool (above and beyond one year) NSIV: +39.8pp (0.178, 0.618) at 15y (other ages NS, no clear increase with age) Years of schooling: OLS: <1 y vs. ≥1: effect grows with age from −0.341y (−0.637, −0.045) at 7y to +0.788y (0.48, 1.096) at 15y; larger for low educated moms (0.741y (0.318, 1.164) vs. NS at 15y), outside Montevideo (0.923 (0.317, 0.869) vs. 0.593y (0.402, 1.444) at 15y); results by sex NS
			• Not clear whether households with between sibling variability are different than average beneficiary households.			Intensive margin: Returns to additional years of preschool (above and beyond one year) are insignificantIV: +0.871y (0.109, 1.633) at 14 y (other ages NS, increase with age)
^aAdditional study details are provided in Leroy et al. (2011). IV, instrumental variables; NS, not significant; OLS, ordinary least squares; pp, percentage points; PSM, propensity score matching; SD, standard deviation; SISBEN, Identification System for Potential Beneficiaries of Social Programs (Sistema de Identificación de Potenciales Beneficiarios de Programas Sociales). SISBEN is an indicator of economic well-being; indicator ranges from 1 (poorest) to 6.
^b95% confidence intervals (CI) were calculated as follows: estimated impact ± 1.96 × reported standard error.
^cProgramme impact was estimated using three different comparisons: treatment versus control, children exposed for <1, 2–4, 5–15 and over 16 months versus control and children exposed ≤ 1 month versus 2–4, 5–15 and over 16 months. The results strongly indicate that the impact estimates using the control group as the counterfactual suffer from selection bias. This was confirmed by the author (personal communication with Raquel Bernal, December 2010). We only report and discuss the results obtained from the estimations using the <1 versus >1 exposure comparison.
^dEDI: caregiver's (the mother in this case) perception of the psychosocial development of the child (range: 1–3; lower score indicates fewer behaviour problems). PIPPS: maternal report in the control group and madres comunitarias' report of children in the treatment group, about child social conduct during play time (range: 1–4). EDI cognitive: maternal report of the cognitive development of her child (range: 0–1; lower score is better). Peabody Picture Vocabulary Test: scale of verbal skills directly applied to children (scale range 55–145, larger is better; population level mean [SD] = 100 [15]). Woodcock Johnson–Muñoz: test of intellectual ability, verbal ability, mathematical reasoning and general knowledge applied directly to children (standardised scores for age, larger numbers are better; population level mean [SD] = 100 [15]).
^eEligibility criteria changed. The latest criteria are reported here. The original criteria included child malnutrition (no objective measure was used to verify it).
^fBehrman et al.'s cut-off value for statistical significance was set at 10% for the cumulative results.

The Bolivia daycare programme had a positive effect on bulk (gross) and fine motor, language and psycho-social skills (Behrman et al. 2004). When matching children in the programme to non-participating eligible children living close to PIDI centres, the programme was found to increase all of the skills by two to six per cent in children between the age of 37 and 58 months. The cumulative impact estimates showed that the effects were positive (2–11% increase) for children who had been in the programme for over 13 months and who were 25–58 months of age. Larger effects were found when the programme children were matched to eligible children living in poor urban areas not served by PIDI. Attending daycare had a significantly positive effect on all skills in children older than 42 months (3–8%). The cumulative analyses showed that the programme had a positive impact (2–11%) for children who had been in the programme for at least seven months. In the final set of analyses, the authors only considered children in the programme and evaluated the impact of having been in the programme for one month or less as compared with at least two months. The programme significantly increased all skills (2–9%) for children older than 42 months. The cumulative estimates found that the programme had a positive impact (2–10%) for children who had been in the programme for at least 13 months. Given the residual (negative) selection bias found for child growth (see above), the development effects might actually be underestimated.

In the case of the Argentina evaluation, Berlinski et al. (2009) first estimated the association between the construction of new preschool places and preschool enrolment. They showed that every newly created preschool place was taken up. As a consequence, the estimated effects of each additional preschool place per child can also be interpreted as the effect of attending preschool. The programme was found to have a positive effect on the mathematics and Spanish test scores of third-graders. An increase of one preschool place per child increased mathematics scores by 4.69 points and Spanish scores by 4.76 points (equivalent to 8%). The estimated effect of one year of preschool for all students in a class of third-graders had a positive effect on the teachers' perception of students' attention (likelihood that at least one-half of them pay a lot of attention in class increased by 12 percentage points, effort by 21 percentage points and regular participation by 16.5 percentage points). No effect was found on teachers' perception of discipline.

Having attended at least one year of preschool in Uruguay had a positive effect on later school attendance (Berlinski et al. 2008). Using ordinary least squares regression estimation, the authors found that the effect increased from 4.3 percentage points at the age of seven years to 27.4 percentage points for children who were 15 years old. The effect on school attendance was larger for children of mothers with lower levels of education and in children outside the capital city of Montevideo. No differences in effect were found for boys or girls. The instrumental variables estimate for children who were 15 years old was a 39.8 percentage point increase in school attendance. The instrumental variables estimates did not show the monotonic increase in effect size with age. The programme also led to an increase in years of schooling. The effect estimated with ordinary least squares steadily increased with age, from (a surprisingly negative effect of) –0.341 years at seven years of age to 0.788 years at the age of 15. As for school attendance, the effects were larger for children of mothers with low levels of education. It was also higher for children outside Montevideo. The effect was not different for boys and girls. The instrumental variables estimate for children at 14 years of age was 0.871 years, but the effect for 15 year olds did not reach statistical significance. It is important to note that the authors did not provide gender-specific attendance and grade progression descriptive statistics. It is thus impossible to evaluate to what extent the difference in effects by maternal education and location are due to pre-existing differences in attendance rates and years of schooling. The results by years of preschool attended (presented as graphs) show that returns to additional years of preschool (above and beyond one year) are insignificant. A limitation of the study is that the identification of the programme effect hinges on the existence of families with sibling variability in preschool attendance. It is not clear to what extent these households might be different than the ‘average’ beneficiary households and thus whether and how this estimation strategy might have biased the results.

Pathways

The reviewed studies provide no or only very limited information on the possible mechanisms by which daycare programmes affect child outcomes. The one exception is the Guatemala study. This study showed that the programme not only improved child dietary intake while at daycare but also when at home (Ruel et al. 2006). This indicates that the effects these programmes might have on child nutritional status (which was not assessed in this study) operated directly through the benefits children receive from the programme and indirectly through improvements at home. It is not clear whether the improved diet at home was a consequence of higher income, better maternal knowledge or a combination of both. The Guatemala study also included a strong operational evaluation that identified a number of factors that were reducing the potential impact of the programme. These included problems such as the need for additional training of the caretakers on using menus and substituting foods, the inadequacy of the amount of cash the hogares received earmarked for food, the large amount of time spent away from the daycare centre to collect the programme cash, delays in receiving the food donations, leaving the children unattended to buy food, and the little time spent on psycho-social activities. Finally, a companion document provided useful information on the care poor children received. These were children who did not attend daycare and whose mothers were working in Guatemala City. The limited availability of childcare alternatives (both formal and informal) leaves mothers with no alternative but to work in the informal sector. In a random sample of mothers working in the slums of Guatemala City, more than 40 per cent reported caring for their children while working (International Food Policy Research Institute 2003). This indicates that notwithstanding the often inadequate quality of the hogares, they most probably provided better care to children than what their working mothers could offer.

Berlinski et al. (2009) suggested that the positive impact on student behaviour in Argentina might be part of the pathway by which pre-primary school participation exerts its impact on school test performance. The authors did not, however, test this pathway quantitatively.

As in Guatemala, the Colombia programme suffered from a number of problems limiting its potential impact. A first limitation is the limited knowledge of the madres comunitarias with respect to child development, safety and effective care practices. On average, madres comunitarias knowledge of infant development was low and not very different from the beneficiary mothers' knowledge (a score of 58% versus 54% on the Knowledge Infant Development Inventory). The quality of care was low as well: using the Family Day Care Rating Scale the mean score was 2.67 (out of seven), suggesting that only the minimum conditions for an adequate early childhood care environment were met. Children were mostly engaged in free play (Bernal and Helo 2010).

Discussion

We reviewed the literature on the impact of daycare programmes on child health, nutritional status, and development in low-income and middle-income countries. We identified studies of programmes from around the world. Studies meeting our inclusion criteria, however, were rare and were all conducted in Latin America. Drawing conclusions from the available evidence is challenging, as the outcomes studied were mostly unique to each paper.

Only one study evaluated the impact on child health. Significant reductions in the prevalence of diarrhoea and acute respiratory infections in children with longer exposures to the programme were found in the second Colombia evaluation. Those effects could have, however, been due to an increase in infections in children immediately after joining a daycare centre, since these children – not those not attending daycare – constituted the comparison group.

The impact on child nutrition outcomes was not clear. The Bolivia study found no impact on child growth. The large reported effect sizes in the first Colombia evaluation are biologically implausible. The second study of the Colombia programme found inconsistent effects. A clear positive impact on child diet was found in Guatemala.

In general, the few studies found large positive effects on measures of child development. The programme in Colombia increased aggressive behaviour but reduced social isolation and increased adequate social interaction in the short term; cognitive development (language, mathematical and general knowledge) improved as well. Additionally, positive long-term effects were found. The study in Bolivia found short-term effects on child motor, language, auditory and psycho-social skills. Longer term effects on test scores, child behaviour, school attendance and years of schooling were found in Argentina and Uruguay.

A key limitation of the reviewed studies is the lack of information on the ‘net’ treatment the daycare interventions bring about. We define this net treatment as the difference in the (quality of) care between the daycare programme and the alternative forms of care the child receives in the absence of the programme. This difference is what ultimately determines the potential impact a programme might have. If daycare interventions provide a high-quality alternative to working mothers who would otherwise take their young children to work, a positive impact on child outcomes can be expected. On the other hand, a programme that pulls children from adequate family care into low-quality daycare may have negative effects on children. The reviewed studies provide detailed descriptions of the programmes themselves. Only the Guatemala study, however, described the care situation of children in the absence of the programme: a large proportion of mothers working in Guatemala City's slums reported caring for their child while working. A good description of the situation of children in the absence of the programme is important as it constitutes the counterfactual against which programme impact is measured. A much better understanding of the difference between what the programme offers and the situation without the programme is indispensable to make useful policy recommendations.

A related shortcoming, which was also identified in a recent review of conditional cash transfer programmes (Leroy et al. 2009), is the very limited information on the pathways of impact. Apart from the information on the improved home diet in Guatemala, none of the studies provided information on how the daycare programmes exerted their impact. The limited quality of services described in the Guatemala and Colombia studies indicates that the programmes do not reach their full potential impact. The importance of understanding the pathways of impact is particularly important because it is not unconceivable that daycare interventions have positive effects on some intermediary outcomes and negative effects on others, resulting in a limited or even negative impact on child outcomes. A beneficiary mother who finds a job as a consequence of the programme might be able to earn a higher income (allowing her to buy better foods for her children) but may also lack the time to care for her children. Social interaction between children may improve child development but may also increase morbidity. Ideally, programmes should be designed after carefully measuring and analysing all relevant intermediary variables and pathways through formative research. Rigorous impact evaluations should consider all of these intermediary variables as well. In sum, without a much better understanding of the care children receive in the absence of the programme and the pathways of impact, the assumption that daycare will benefit children may not be warranted.

An important challenge in the evaluation of daycare interventions is the families' self-selection into the programme, which makes the identification of a credible counterfactual difficult. Daycare programmes do not just cater to parents who already work, but also lead to women entering the labour market. Other women may be increasing the number of hours they work. Many of the parents who do not participate in daycare programmes do so by choice. Randomised controlled designs are difficult to implement in the context of daycare programmes. Individually randomising households to receiving or not receiving daycare (using a wait list, for instance) may not be practical. First, the programme to be evaluated is rarely the only daycare option available, so wait-listed parents may explore and use other daycare options. Second, the existence of wait-lists may actually spur community members to start providing daycare. Finally, as treatment children graduate from the programme or drop out and daycare spots open up, it would be difficult to prevent wait-listed (control) children from enrolling in the programme. A cluster randomised design is easier to implement, but does not come without its own challenges. Since only a potentially small proportion of the population participates in these types of programmes, intent-to-treat analyses might not be able to detect a sizeable impact. A valid treatment on the treated approach would require a strategy to match beneficiary households to those in the control group who would benefit from the programme if it was available.

The studies in this review use different non-experimental approaches to solve the selection bias problem: matching (either by design or analytically), instrumental variables or intra-family estimators and dose–response analyses (comparing children who just joined daycare with children who have been there longer). Even though the authors are confident that their approach gets around the selection bias problem, there are reasons to believe this might not be the case. The barely credible effect sizes in the first Colombia study puts into question the validity of the instruments used by the authors. The first author of the second Colombia study confirmed that the estimates based on the treatment–control matching suffered from selection bias. The identification strategy in Uruguay eliminated households with only one beneficiary child and all households in which all siblings received the same treatment. This inevitably leads to the question as to how representative the analysed households were and whether solving one type of selection bias problem came at the expense of a different type of selection bias. Finally, the comparison of children with shorter and longer programme exposure can produce invalid estimates for outcomes that change in the very short term such as child morbidity.

Future studies should consider using alternative evaluation designs and evaluate the extent to which they minimise the selection bias problem. These designs include randomised promotion of the daycare programme and randomly changing the price of daycare. In these designs, having received the promotion or (lower) price is orthogonal to household characteristics. As a consequence, they might serve as valid instruments to identify programme impact.

Conclusions

The reviewed studies suggest that daycare programmes have a positive impact on child development. The evidence on child nutrition and health outcomes is less clear. More rigorously conducted studies on the impact of daycare programmes in low-income and middle-income countries are needed. These should be conducted in a variety of settings and provide a clear description of the counterfactual care scenarios. There is a particular need for studies documenting the impact on health outcomes. These studies should use children not exposed to the programme as the control to avoid the problem described for the Colombia study. Intermediary variables that help explain the pathways of impact need to be measured and taken into account in the analyses. Studying how child age changes the (relative) importance of each pathway (for example, the trade-off between income and care) would provide important insights as well. Additionally, the extent to which the characteristics of the care provided (for example, number of children, hours per week spent in daycare) modify its effect should be studied. Finally, future studies should consider alternative evaluation designs that minimise the risk of selection bias.

Acknowledgements

The present review was supported by the International Initiative for Impact Evaluation (3ie). The authors are grateful to Hugh Waddington and two anonymous reviewers who provided invaluable comments and suggestions. The authors are responsible for all remaining errors.

Notes