Understanding the linkages between social safety nets and childhood violence: a review of the evidence from low- and middle-income countries

Abstract

As many as one billion children experience violence every year, and household- and community-level poverty are among the risk factors for child protection violations. Social safety nets (SSNs) are a main policy tool to address poverty and vulnerability, and there is substantial evidence demonstrating positive effects on children’s health and human capital. This paper reviews evidence and develops a framework to understand linkages between non-contributory SSNs and the experience of childhood emotional, physical and sexual violence in low- and middle-income countries. We catalogue 14 rigorous impact evaluations, 11 of which are completed, analysing 57 unique impacts on diverse violence indicators. Among these impacts, approximately one in five represent statistically significant protective effects on childhood violence. Promising evidence relates to sexual violence among female adolescents in Africa, while there is less clear evidence of significant impacts in other parts of the developing world, and on young child measures, including violent discipline. Further, few studies are set up to meaningfully unpack mechanisms between SSNs and childhood violence; however, those most commonly hypothesized operate at the household level (through increases in economic security and reductions in poverty-related stress), the interpersonal level (improved parental behaviours, caregiving practices, improved psychosocial well-being) and at the child-level (protective education and decreases in problem or risky behaviours). It is important to emphasize that traditional SSNs are never designed with violence prevention as primary objectives, and thus should not be considered as standalone interventions to reduce risks for childhood violence. However, SSNs, particularly within integrated protection systems, appear to have potential to reduce violence risk. Linkages between SSNs and childhood violence are understudied, and investments should be made to close this evidence gap.

Cash transfers, child protection, social protection, violence against children

Key Messages

• We review evidence and develop a framework to understand linkages between social safety nets (SSNs) and childhood emotional, physical and sexual violence in low- and middle-income countries.

• We catalogue 14 impact evaluations, 11 with completed results analysing 57 unique impacts on diverse violence indicators, among which approximately one in five represent statistically significant protective effects on childhood violence.

• Promising evidence relates to sexual violence among female adolescents in Africa, while there is less clear evidence of significant impacts for in other parts of the developing world, and on young child measures, including violent discipline.

• Linkages are understudied, however, SSNs may reduce violence risk at the margin, and integrated systems and programming hold promise to maximize violence prevention.

Introduction

Physical, emotional or sexual violence is experienced by as many as one billion children every year (Hillis et al. 2016; UNICEF 2014a), and has detrimental impacts on children’s development, their ability to learn, and their right to healthy and productive lives (Paolucci et al. 2001; Gershoff 2002; Arseneault et al. 2010; Abramsky et al. 2011; Gini and Pozzoli 2013; Devries et al. 2014; Jennings et al. 2015; Ogando Portela and Pells 2015; Ports et al. 2016). Violence experienced in childhood is also of concern because of its intergenerational nature, whereby violent behaviours are typically normalized in childhood, putting children who experience or witness violence at a higher risk of experience and perpetration in adulthood (Abramsky et al. 2011; Fulu et al. 2013; Fleming et al. 2015; Jennings et al. 2015). The United Nations Convention on the Rights of the Child (UNCRC), recognizes a child’s right to protection from all forms of violence. Adopted in 1989, it is the most widely ratified human rights treaty (UN Committee on the Rights of the Child (CRC) 2011; UNICEF 2014b).

Despite these recognized rights, childhood violence rates remain high globally. In 62 low- and middle-income countries (LMICs), 4 out of 5 children aged 2–14 experienced physical punishment or aggression from a parent or caregiver in the past month (UNICEF 2014a), and a global meta-analysis with prevalence rates from 331 samples and nearly ten million participants finds prevalence rates for child sexual abuse at 11.8% (18.0% for girls and 7.6% for boys) (Stoltenborgh et al. 2011). While risk and protective factors for childhood violence have been studied, effective prevention strategies, particularly in LMIC settings, remain elusive. Global policy attention to the prevention of childhood violence was renewed with the adoption of the UN Sustainable Development Goals in 2015, with the commitment to ending abuse, exploitation, trafficking and all forms of violence against and torture of children (Goal 16.2).

Household- and community-level poverty are among the risk factors for child protection violations (Berger 2004; Butchart et al. 2006; Elgar et al. 2009; Gilbert et al. 2009; Akmatov 2011; OECD 2011; Shook Slack et al. 2011; Meinck et al. 2015; Pelton 2015). There is however, little rigorous evidence demonstrating whether this relationship is causal, with some research suggesting that certain child protection issues, including sexual exploitation, unnecessary family separation, child labour and early marriage have a more direct link to poverty, whereas other types of childhood violence, such as child sexual abuse and violent discipline, might be more indirectly related to poverty (Barrientos et al. 2014; Markus and Page 2014; Sheahan 2011). Regardless, economic pressures have been broadly shown to leave children at increased risk of violence (Butchart et al. 2006; OECD 2011; Butchart and Hillis 2016).

In recent years, social safety nets (SSNs) have emerged as a primary policy tool to address poverty and vulnerability. SSNs (also referred to as social assistance or transfers) are non-contributory programmes, designed to provide regular and predictable support to poor and vulnerable populations, and are key components of larger social protection systems (Honorati et al. 2015). There is strong evidence that cash transfers have resulted in considerable reductions in poverty globally (Miller and Samson 2012). Further, a growing evidence base around the world is documenting the role that SSNs play in improving child well-being, regardless of whether these programmes are explicitly child-focused. Child well-being outcomes commonly studied in relation to cash transfers include nutrition, illness, schooling, mental health and stress (Lagarde et al. 2007; Owusu-Addo and Cross 2014). Relatedly, the relationship between SSNs and intimate partner violence (IPV) among adults has increasingly been studied (Bobonis et al. 2013; Hidrobo et al. 2016).

Despite the hypothesized potential of SSNs to reduce childhood violence, little attention has been paid to these linkages by researchers, and hence few studies empirically or theoretically document the pathways through which SSNs affect childhood violence, whether positively or negatively. Most available evidence concerns broader child protection issues rather than childhood violence specifically. For example, systematic reviews have examined links between social protection and child protection outcomes, including birth registration, child labour, family separation and early marriage (Sheahan 2011; Barrientos et al. 2014); and impacts of small- and medium-scale economic strengthening interventions on child labour, child marriage, sexual violence, physical violence, gender-based violence (GBV) and inadequate care in crisis settings (Chaffin 2011; Markus and Page 2014; Chaffin and Mortenson Ellis 2015). Despite their stated focus, neither the Markus and Page (2014) nor the Barrientos et al. (2014) reviews found examples of empirical evidence linking social transfer programmes to reductions in sexual or physical violence. The Chaffin and Mortenson Ellis (2015) review included only one study with a measure on GBV, namely ‘having sex unwillingly’ asked of adolescent girls. This review also considered adverse effects of economic strengthening interventions on child well-being outcomes, and found that while most programmes reported one or more positive effects on child well-being, about one in five also reported at least one adverse effect. No clear patterns emerged regarding these adverse impacts, but they often included increased participation by children in work activities. Finally, a review paper examining the role of cash transfers in the protection of children in emergencies highlighted the lack of evidence around programme impacts on psychosocial distress, sexual exploitation and physical violence (Thompson 2014). Across existing reviews, authors conclude that while these programmes may have had the potential to affect violence, this potential has not been evaluated.

This paper aims to address this research gap by reviewing the evidence base in LMICs linking SSNs and experiences of childhood violence. In particular it aims to assess to what extent and through which pathways non-contributory SSNs can help protect children from childhood physical, emotional and sexual violence drawing on definitions proposed by UNICEF and Together for Girls (UNICEF 2010; UNICEF 2014a; Together for Girls 2016). We focus on the following questions: (1) What are the key pathways through which SSNs have the potential to affect childhood violence? (2) What rigorous evidence exists on the impact of SSNs on childhood violence? (3) Through which mechanisms are impacts realized, or in the cases where no impacts were found, what hypotheses exist as to mechanisms which could strengthen impacts? (4) Where SSNs have actively tried to address issues related to childhood violence, what programme modifications or strategies have been pursued? and (5) What are some of the key research questions and gaps looking forward?

This paper does not constitute a systematic review of the evidence; rather, it is a comprehensive synthesis of a largely emerging and fragmented evidence base, with a focus on understanding existing evidence and gaps that need to be filled in order to support policy makers seeking to utilize SSNs in their effort to prevent childhood violence. Section 2 presents a framework, Section 3 describes the methodology and Section 4 discusses the findings from this review. Section 5 concludes with a discussion of the limitations of the review, implications for programme design and highlights ongoing research efforts and key gaps on the evidence base linking SSNs and childhood violence.

SSNs and childhood violence: a framework

In this section, we present a framework of hypothesized ways in which SSNs can influence childhood violence. First, we define some key concepts.

Consistent with the UNCRC, a child is defined as a person under the age of 18. Following the standard UN-wide categorization of adolescents and young people, an adolescent is defined as a person aged 10–19 years and young people as those aged 10–24 years (UNICEF 2012).

Our focus on SSNs includes five of the six main types of programming identified in the World Bank’s state of Social Safety Nets 2015.¹ These include (1) conditional cash transfers (CCTs), (2) unconditional cash transfers (UCTs), (3) unconditional in-kind transfers, (4) public works (PW) or cash for work (CfW) and (5) vouchers or fee waivers. SSN programming bundled with other services or intervention components, for example, additional livelihoods training, or community information sessions are considered and referred to in this review as ‘plus’ (e.g. ‘CCT plus’ or ‘PW plus’).²

Although we recognize that all forms of child maltreatment are important, we focus on childhood physical, emotional and sexual violence experienced by children under the age of 18 years, drawing on definitions proposed by UNICEF and Together for Girls (UNICEF 2010, 2014a; Together for Girls 2016). We therefore do not consider outcomes such as child marriage, neglect or negligent treatment, female genital mutilation/cutting (FGM/C), child labour or witnessing IPV among adult household members.³ We consider violence perpetrated by peers, intimate partners or adults, in the context of the home, school or community. In addition, while recognizing that children are themselves common perpetrators of violence, we consider only victimization, and not perpetration of violence.⁴ Physical violence, both fatal and non-fatal, is understood to include homicide, violent physical discipline (also known as corporal punishment) and all other forms of torture, cruel, inhuman or degrading treatment or punishment as well as physical bullying and hazing. Emotional violence is understood to include violent psychological discipline and all other non-physical forms of hostile and degrading behaviour, as well as psychological bulling and hazing. Sexual violence comprises sexual abuse (i.e. abusive sexual contact, attempted non-consensual sex acts, coerced/forced sex acts), sexual exploitation (the exploitative use of a child in commercial sexual activities or other unlawful sexual practices in which cash, goods or favours are exchanged for sex acts) and non-contact sexual violence (e.g. verbal sexual harassment, use of children in pornographic performances and materials). Although the discrete categorization of the different forms of childhood violence ignores the fact that children are often exposed to simultaneous and overlapping forms of violence, it has been organized as such for analytical purposes. In addition, while we are guided by these broad definitions, each study included operationalizes its own unique definitions of violence indicator(s), falling into the broad categories as defined above.

Our framework (Figure 1) articulates hypothesized direct and indirect pathways by which SSNs may positively or negatively affect childhood violence (Jaffe et al. 1990; Repetti et al. 2002). We draw on frameworks by Fein and Lee (2003), Markus and Page (2014) and Barrientos et al. (2014), who examine social protection and child well-being; however, either differ in their definition of programming, and/or do not describe in detail the specific pathways and impacts on childhood physical, emotional or sexual violence, as we aim to do here. As highlighted in Fein and Lee (2003), the number and nature of different pathways of potential influence are such that we cannot predict the net impact of SSNs on experiences of childhood violence precisely. However, we can hypothesize the direction of various relationships and the impacts that SSNs can have on factors along the causal pathway. Hence, pathways in our framework are hypothetical, and as yet, not necessarily supported by evidence from the studies reviewed. Rather the framework serves as a starting point for understanding empirical findings and research gaps, and aims to inform future evaluation studies.

Figure 1.

Theoretical framework linking social safety nets and childhood violence

Given that SSNs generally target households, our framework is focused on potential mechanisms at the micro (household- and individual-) level. It considers these mechanisms in the wider context of the meso- and macro-level context (structural, institutional and community). For SSNs to affect the risk of violence, the impacts are generally hypothesized to work first through household-level mechanisms, and subsequently through caregiver/interpersonal-level mechanisms or directly through child-level mechanisms (or both). In Figure 1, drawing on existing literature, we illustrate hypothesized pathways between mechanisms where positive relationships are denoted by large dash arrows, negative relationships by small dash arrows and ambiguous relationships by solid arrows.

Household-level mechanisms

At the household-level, SSNs may affect economic security, labour force participation, intra-household power dynamics (e.g. women’s bargaining power), and overall stress levels. Income transfers have been shown to improve poverty-related outcomes such as food insecurity, consumption, asset ownership and housing conditions (Adato and Bassett 2009; Fiszbein et al. 2009; Kenya OVC-CT Evaluation Team 2012; FAO 2015; Davis et al. 2016). Further, these programmes may alter labour participation of household members (including children), which has further implications for income and time use (de Hoop and Rosati 2014; Handa et al. 2016). Related, household economic survival strategies and poverty can play an important role in decisions around early marriage (Walker 2012). Early marriage in turn is a risk factor for childhood violence. The evidence to date around cash transfers and early marriage is mixed (Baird et al. 2011; Nanda et al. 2014; Handa et al. 2015), but does suggest that cash transfers may help to delay marriage, particularly in the context of sub-Saharan Africa (SSA).

Economic insecurity including food insecurity are major sources of daily stress. By alleviating this insecurity, transfers may reduce both acute and chronic stress. Studies of two cash transfer programmes (one governmental in Mexico and one non-governmental programme in Kenya) showed mixed results on cortisol levels, a biomarker of chronic stress, among adults and children living in beneficiary households (Fernald and Gunnar 2009; Haushofer and Shapiro 2016). Evidence linking cash transfers and self-reported perceived stress, is also mixed, with some studies reporting reductions in stress (Ozer et al. 2011; Haushofer and Shapiro 2016) and at least one other finding no impacts (Paxson and Schady 2010).

Transfers (and increased access to cash, information, social networks, and services that sometimes complement cash benefits), as well as increased female labour force participation through PW/CfW, may alter intra-household power dynamics and women’s bargaining power. A review of programming aimed at strengthening household economic security (including CCTs and UCTs) on women’s empowerment and nutrition found mixed evidence from quantitative impact evaluations, with positive impacts on women’s empowerment generally found only in qualitative evaluations of CCTs (van den Bold et al. 2013). Mixed and non-significant impacts of SSN on quantitative measures of women’s empowerment and decision-making may be a function of poor or inconsistent measurement of the concept, as well as the diversity of gendered contexts which affect conclusions (Peterman et al. 2015).

In turn, the aforementioned changes at the household-level may affect the risk of childhood violence. Economic insecurity, including food insecurity, income poverty and inadequate housing are risk factors for childhood violence (Akmatov 2011; Berger 2004; Butchart et al. 2006; Gilbert et al. 2009; Cancian et al. 2013; Jacob et al. 2013; Meinck et al. 2015; Pelton 2015). Economic insecurity may be a driver of engagement in transactional sex, and parents may directly or indirectly encourage such relationships to obtain food and other goods (Heise et al. 2013; Stoebenau et al. 2016). Unemployment also increases the risk of childhood violence (Pelton 2015; Stith et al. 2009); however, the extent to which this effect is mediated through material hardship, especially among the poor, is not fully understood (Pelton 2015).

Caregiver- (interpersonal-) level mechanisms

Transfer-induced improvements in economic security, changes to labour force participation and time use, reductions in stress, and more equitable intra-household power dynamics may in turn influence factors at the caregiver and interpersonal level, including substance misuse, psychosocial well-being, caregiving behaviours, intra-household conflict and interpersonal violence (including IPV), and caregiver’s supervision of children.

Transfers or other income support may improve psychosocial well-being—including aspects such as personal stress, anxiety, depression, self-esteem, and psychopathology—among caregivers, either directly, or indirectly through positive effects of increased participation in productive activities on self-esteem. Conversely, cash transfer programmes may adversely affect psychosocial well-being through stress related to fulfilling programme conditions. Poverty and poor mental health are mutually reinforcing (Lorant et al. 2003; Lund et al. 2011). Poverty is a risk factor for mental health disorders, through malnutrition, stress, substance abuse, social exclusion and exposure to trauma and violence. Likewise, poor mental health increases the risk of poverty, through increased health expenditures, reduced productivity, stigma and loss of employment and earnings (Lund et al. 2011). Furthermore, poverty-induced stress may cause sadness and anger, thereby increasing short-sighted and risk-averse decision-making which reinforce the cyclical nature of poverty (Haushofer and Fehr 2014). A study in Kenya demonstrated that cash transfers improved happiness and life satisfaction and lowered depression (Haushofer and Shapiro 2016).

Caregiver psychosocial well-being undermine one’s ability to cope with poverty and its stressors and pose a significant threat to child safety and well-being (Belsky 1993; Black et al. 2001; Gilbert et al. 2009; Stith et al. 2009; Meinck et al. 2015; Pelton 2015). Research suggests that economic resources may allow parents to be more responsible, warm and consistent (Wachs et al. 2009; Fernald and Hidrobo 2011), and that these parenting characteristics are associated with a reduced risk of childhood violence. The impacts of transfer programmes on caregiving behaviours is understudied (de Groot et al. 2017); however, transfer-induced changes in such behaviours have the potential to decrease the risk of childhood violence (including the frequency and severity of violent discipline), through decreases in household-level stress and improved caregiver psychosocial well-being. Furthermore, positive caregiving behaviours and children’s problem behaviours are mutually reinforcing, so that improvements in one of these outcomes are likely to lead to improvements in the other (Butchart et al. 2006; Pinheiro 2006; Epps and Huston 2007).

SSNs also have potential to reduce childhood violence through the intra-household conflict pathway. Intra-household conflict may reduce children’s psychosocial well-being, increase problem and risk behaviours, or increase the time children spent in high-risk settings (including residential care, the street or in gangs). Further, there is a growing body of evidence documenting the potential for cash transfers to reduce the risk of IPV among adults (Perova 2010; Bobonis et al. 2013; Hidrobo and Fernald 2013; Hidrobo et al. 2016), and given that maternal experience of IPV is a risk factor for childhood violence (Meinck et al. 2015), these documented reductions are promising for reducing childhood violence.

Caregiver supervision of children may also have implications for the risk of childhood violence. On the one hand, transfers may allow caregivers to spend more time with their children, while on the other, PWs/CfW programmes and stringent programme conditions may reduce this time (Bloom et al. 2000; Beecroft et al. 2002; Gennetian and Miller 2002; Fein and Lee 2003). Increased employment or engagement in productive activities, as well as time consuming conditions linked to transfers, change time-use patterns of caregivers and may leave children without adequate supervision, increasing their susceptibility to violence and abuse, particularly in settings where quality childcare is limited. Among older children and adolescents, inadequate caregiver supervision may be associated with their engagement in problem and risk behaviours.

In theory, SSNs, and income transfers in particular, can influence substance misuse, but the direction of hypothesized impacts is ambiguous: improvements in psychosocial well-being and economic security may decrease motivations to engage in substance misuse, or purchase of these so-called ‘luxury’ goods may increase through an income effect. A systematic review from LMICs of the effects of cash transfers on the use of temptation goods (mostly alcohol and tobacco) showed largely non-significant or negative impacts on expenditures on such goods from 19 studies in Africa, LAC and Asia (Evans and Popova 2014). Substance abuse is a common factor in incidents of child abuse (Milner and Chilamkurti 1991; Famularo et al. 1992; Gilbert et al. 2009; Walsh et al. 2003; Meinck et al. 2015).

Child-level mechanisms

At the child-level, transfers—either directly or through the aforementioned pathways—can affect time spent in school, psychosocial well-being, time in high-risk settings, child marriage, and problem and risk, which in turn all affect childhood violence risk.

Decisions around time use are influenced by transfers and increased economic security. These decisions simultaneously affect the amount of time that children spend in school, home, productive activities (i.e. labour), and high-risk settings (such as unsafe work environments) (de Hoop and Rosati 2014). While cash transfers have been shown to have large, positive impacts on school enrolment (Baird et al. 2013b; de Hoop and Rosati 2014), the impacts of school enrolment on childhood violence risk are ambiguous. Increased time spent in school may lead to decreases in exposure to (hazardous) work environments, thus decreasing the risk of physical and sexual violence. However, children may increase both hours spent in school and work simultaneously as a result of SSNs (de Hoop et al. 2015), and thus both protective and adverse consequences are possible resulting from programme-induced changes in time use patterns and resulting exposure to school and work environments. Conversely, time spent in school may increase childhood violence risk, as teachers and peers may be perpetrators of various types of violence, and children, especially girls, may be at risk of sexual violence while travelling to school (Lalor 2004; Dunne et al. 2006; African Child Policy Forum (ACPF) 2014; Ogando Portela and Pells 2015). On the other hand, research from Ethiopia, India, and Vietnam suggests that out-of-school children are more likely to be physically bullied than their in-school counterparts (Jones and Pells 2016).

Psychosocial well-being, as well as problem and risk behaviours may be influenced by transfers through the economic security, caregiver psychosocial well-being, intra-household conflict, caregiving behaviour and caregiver supervision pathways. Evidence from Kenya and Malawi has shown that cash transfer programmes have potential to improve adolescent mental health, including depression and distress (Baird et al. 2013a; Kilburn et al. 2016). Problem behaviours include externalizing (e.g. aggression, disobedience, bullying) and delinquent behaviours, and interact with time spent in school as well as time spent in high-risk settings. Studies in LMICs have found reductions in problem behaviours (Fernald et al. 2009; Ozer et al. 2009; Paxson and Schady 2010; Fernald and Hidrobo 2011; Macours et al. 2012; Figueroa 2014) and delinquent behaviours (Chioda et al. 2015; DSD, SASSA & UNICEF 2012) resulting from CCTs and housing voucher programmes. Problem behaviours are significantly correlated with the risk of childhood violence (Epps and Huston 2007; Stith et al. 2009). Relatedly, in adolescence, sexual and health risk behaviours may be a function of economic insecurity and other aforementioned pathways, and these in turn may increase the risk of childhood violence, particularly sexual violence. Poverty (both absolute and relative) may, for example, drive adolescents, especially girls, into commercial sex work, or more commonly, transactional or age-disparate sexual relationships (termed ‘intergenerational’ if the age gap is large), all with important limits as to their sexual agency (Lorant et al. 2003; Hallman 2005; Markus and Page 2014; Austrian et al. 2016; Stoebenau et al. 2016). The likelihood of these pathways resulting in negative outcomes may increase in emergencies due to increased vulnerability of separated and unaccompanied children, and vary by the poverty levels of their caregivers and perceptions around protecting children’s ‘honour’ (Thompson 2014).

Contextual factors and vulnerability characteristics

The afore-mentioned pathways describe how SSNs can work through household-, caregiver- and child-levels to influence the risk of childhood violence. The strength of these relationships may vary depending on child, caregiver and household vulnerability characteristics that influence susceptibility to childhood violence, and also by contextual factors that influence associations at the community-, institutional- or structural-level, as recognized by ecological frameworks for childhood violence risk (Cicchetti and Lynch 1993; Lynch and Cicchetti 1998; Scannapieco and Connell-Carrick 2005). Examples of contextual factors include policy and institutional frameworks; legal frameworks; economic and human development; migration patterns; power relations, class structures and levels of inequality; socio-cultural (gender and authoritarianism) norms, beliefs, and practices; and generalized levels and types of violence (may be a function of conflict settings and other forms of community violence). At the household level, vulnerability characteristics may include household structure/composition, social isolation, parental loss or separation and social support networks, and discrimination based on HIV, chronic illness or disability (Belsky 1993; Pinheiro 2006; Stith et al. 2009; Meinck et al. 2015). At the caregiver-level, vulnerability characteristics may include gender and social support, as well as caregiver age, education levels, biological relationship to the child and personal histories of childhood violence (Belsky 1993; Pinheiro 2006; Fang and Corso 2007; Gilbert et al. 2009; Stith et al. 2009; Meinck et al. 2015). At the child-level, vulnerability characteristics include age, gender, sexual orientation, and HIV, disability and orphan status (Butchart et al. 2006; Pinheiro 2006; Gilbert et al. 2009; Markus and Page 2014; Meinck et al. 2015).

Finally, a variety of factors related to programme design characteristics, including targeting, programme conditions, duration, regularity of payments, payment size and recipients’ gender may affect the direction and strength of relationships and/or the plausible pathways outlined above. In the case of PW/CfW, such characteristics as type of work, wage, hours and seasonalities of work, as well as potential child care and supervision arrangements which may be part of programming are important for potential impact pathways and determine whether these are positive, negative or ambiguous.

Methodology

The review required a methodology appropriate to navigate an emerging and fragmented evidence base. We aimed to review published or publicly available (grey literature) studies, as well as ongoing work (presentations, study protocols), from January 2000 to April 2016 which link SSNs and childhood violence outcomes of interest. As previously defined, a child is considered anyone under the age of 18 years; however, as to not exclude relevant evidence, we include studies which span a larger age range, as long as they include individuals under 18 years.⁵

We considered quantitative and/or mixed methods approaches that utilized an experimental or quasi-experimental design. In addition, we draw from key qualitative studies to help explain and discuss mechanisms, complementing findings from quantitative studies. We limit our core review to studies conducted in LMICs; however, we discuss evidence from high-income countries as a contrast to reported findings. Studies were identified through searches in electronic databases, relevant journals, and on institutional websites.⁶ In addition, experts in the field were consulted, and forward and backward citation searches were performed as studies were identified.

As we expected the evidence base to be scattered and childhood violence to be rarely the primary focus of interest, we deliberately did not seek to conduct a meta-analysis or systematic review. We adopted broad search terms most likely to identify evaluations fitting our criteria.⁷ Abstracts or executive summaries were screened against inclusion and exclusion criteria. When these did not provide sufficient information to determine relevance, the full article was retrieved for further examination. While articles in Spanish were included, we only conducted formal database searches in English. A total of 117 articles were scanned as meeting some criteria or marginally relevant, but ultimately not included. It is worth noting that the majority of studies included here do not necessarily focus on childhood violence as the objective of the evaluation, and in several cases analyse violence only as a robustness check rather than a main outcome (Rasella et al. 2013; Rodríguez 2015). Further, measures considered as part of sexual abuse and exploitation (transactional sex and age-disparate sex) can have multiple interpretations. Particularly in SSA, some argue these indicators represent complex social interactions with multiple interpretations (Poulin 2007; Stoebenau et al. 2016; Fielding-Miller et al. 2016). In addition, a number of studies aggregate indicators into a composite measure or a scale [e.g. Home Observation for Measurement of Environment (HOME scale)], and in such instances, it is unclear if the impact (or lack thereof) is a function of variation and change in specific violence measures, or is driven by non-violence factors in the scale.⁸

From each of the core studies identified, we abstracted information about the programme (location, programme name, implementer), the intervention (modality, conditionality, size and regularity of benefits), the target population (age, sex), the study (design, data, sample size) and childhood violence impact results (if available, including baseline prevalence, impact estimates and hypothesized mechanisms). Summary tables organized by region, and alphabetically by country within region were constructed and used to summarize evidence and link back to unpack mechanisms hypothesized in the framework.

Results

Summary of programmes reviewed

Table 1 summarizes the programme components from the identified core papers, organized by region SSA, LAC, Asia and Middle East and Northern Africa (MENA), and country (in alphabetical order) within region. In total, we identified 14 studies meeting our inclusion criteria: three studies were ongoing, 11 were completed. Of the completed studies (with results), 10 were published in peer-reviewed papers, one was a technical report. Approximately 50% (seven studies) were from SSA, 36% (five studies) from LAC, 7% (one study) from Asia and 7% from MENA (one study). In total, 10 countries were represented, with multiple studies in each of Kenya, South Africa and Ecuador. Only two cases, the Palestinian National Cash Transfer Programme (PNCTP) and the Wajir setting in North Eastern province of the Kenyan Adolescent Girls Initiative (AGI-K), qualify as humanitarian or emergency settings.

Table 1.

Review of core papers (programme components)

No	Authors	Country	Programme name	Implementer	Modality	Target population	Recipient	Programme details	Benefits
									Size	Regularity
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
Sub-Saharan Africa (SSA)
1	Handa et al. (2014)a	Kenya	Cash Transfer for Orphans and Vulnerable Children	Ministry of Gender, Children and Social Development (Govt)	UCT	HHs identified as extremely poor and caring for at least one OVC aged <18 c	Primary caregivers	HHs told at enrollment the cash is for the care and development of OVCs; however, no monitoring or conditions imposed. HHs no longer eligible when child turns 18	3000 KES (∼USD 40). Approximately 20% of monthly total HH expenditure.	Every 2 months
2	Rosenberg et al. (2014)a
3	Austrian et al. (2015b, 2016a)	Kenya (Kibera and Wajir)	Adolescent Girls Initiative-Kenya	Plan International (Kibera) and Save the Children (Wajir)	CCT Plus	Female adolescents aged 11–15 years	HH head (or designated HH member)	Four treatment arms: Violence prevention (VP) (community conversations); VP + education (cash and in-kind transfers); VP + education + health (health and life skills education delivered in weekly girls groups by female mentor); VP + education + health + wealth creation (savings and financial education + start-up savings). Education transfer has four components (conditional on ≥ 80% school attendance and continued enrollment): HH cash transfer; Schools fees (paid to the school); School incentive (paid to the school); Schooling kits (every new term).	Kibera: 2250 KES per term (∼USD 23) Wajir: 3000 KES per term (∼25 USD) Approximately 10% of average HH expenditure for 4 month period.	Every 2 months (for 6 school terms)
4	Baird et al. (2012)a	Malawi (Zomba)	Zomba cash transfer program	Invest in Knowledge Initiative (2008) and Wadonda Consultants (2009)	CCT and UCT	Never married young women aged 13–22 years	Parents/guardians and young women	Programme stratified on enrollment status of young women at baseline. All baseline dropouts assigned to CCT. Baseline schoolgirls randomly assigned to UCT or CCT (conditional ≥80% school attendance).	Young women: USD 1, 2, 3, 4 or 5 (lottery) Parents: USD 4, 6, 8, or 10 (RS by EA) Average transfer to HHs: USD 10 or approximately 10% of HH consumption expenditure.	Ten monthly installments per year
Sub-Saharan Africa (SSA)
5	Cluver et al. (2013)a	South Africa	Child Support Grantd	Social Security Agency (Govt)	UCT	HHs with children aged <19 based on income means test	Primary caregivers	Recipients required to participate in “development programmes'' and have children immunized at programme introduction. Conditions lifted in 2001 when studies showed the most vulnerable children were excluded as a result.	ZAR 280 (∼USD 35) for each qualifying child (caps of ZAR 33 600/year for single caregivers and ZAR 67 000/year for married couples)	Monthly
6	Pettifor et al. (2016)a	South Africa (Mpumalanga)	HIV Prevention Trials Network 068	HIV Prevention Trails Network 068 study team	CCT	Never married or pregnant young women aged 13–20 in Grades 8–11	Parents/guardians and young women	Funds deposited into separate bank accounts, conditional on ≥ 80% school attendance.	Young women: ZAR 100 (∼USD 10) Parents: ZAR 200 (∼USD 20) Approximately 15.7% of BL HH expenditure	Monthly
7	Palermo et al. (2017)b	Tanzania	Tanzania’s Productive Social Safety Net	Tanzania Social Action Fund (Govt)	CCT, UCT and PW, Plus	HHs below the food poverty line	Adult women (mothers)	Treatment is comprised of 3 components: Variable UCT/CCT for all HHs; PW for one able-bodied adult per eligible HH aged ≥18 in the lean season; Livelihoods strengthening. UCT to all eligible HHs with an additional transfer to HHs with children <18. CCT involves grants to (i) HHs with pregnant women or children <5 conditional on health check-ups (for pregnant woman and child); or (ii) HHs with school age children, conditional on ≥ 80% school attendance.	UCT: TZS 10 000 (∼USD 6)/TZS 14 000 (∼USD 8.5) if children <18 CCT: max. TZS 38 000 (∼USD 23) per HH PW: max. TZS 138 000 (∼USD 83) annuallye	CCT/UCT: monthly PW: daily for 4 months a year
Latin America and the Caribbean (LAC)
8	Rasella et al. (2013)a	Brazil	Bolsa Familia Program	Ministry of Social Development with Ministry of Health and Education (Govt)	CCT Plus	Extremely poor HHs and HHs deemed poor with children <17 years or lactating women	Adult women (mothers)	Education conditionalities include ≥85% attendance for children 9–15 and ≥75% attendance for children 16–17. Health conditions include vaccination and compliance with health and growth check-ups for children < 7 years; and pre- and post-natal visits and health and nutritional educational activities for lactating women.	USD 18 for each pregnant women or child aged <17 years (cap by category). Additional USD 35 to extremely poor HHs. Total benefit ranges between USD 18–175 per HH.	Monthly
Latin America and the Caribbean (LAC)
9	Rodríguez (2015)b	Colombia	Familias en Acción	Acción Social (Govt)	CCT	Extremely poor HH based on PMT	Adult women (mothers)	Education conditionalities include ≥80% attendance for children 7–18. Health conditionalities include regular medical check-ups for children < 7 years. Benefits paid through pre-programmed ATMs starting in 2007.	Health: USD 25–30 for HH with any children 0–6; Education: USD 6 per primary school child + up to USD 24 per secondary school child in 2014 Approximately 25% of legal minimum wage (HH with 2 children aged 6 and 12).	Monthly
10	Paxson and Schady (2010)a	Ecuador	Bono de Desarrollo Humano	Ministry of Social and Economic Inclusion (Govt)	UCT	Low-income mothers with children aged 0–16 based on PMT	Adult women (mothers)	Originally conceived as a CCT, conditional on preventive health check-ups and minimum school attendance for school-aged children; however, lack of verification of compliance led to implementation of a UCT. Programme advertised as a child benefit.	USD 15. Approximately 6–10% of baseline HH expenditures.	Monthly
11	Fernald and Hidrobo (2011)a
12	Macours et al. (2012)a	Nicaragua	Atención a Crisis	Ministry of the Family (Govt)	CCT Plus	Poor HHs (based on PMT)	Primary caregivers	Three treatment arms: CCT conditional on regular preventive health check-ups (not monitored) for HHs with children aged 0–5. HHs with children aged 7–15 in primary school eligible for an additional educational transfer conditional on school enrollment and regular attendance (monitored); CCT + training package (vocational and business-skills training for up to HH member); CCT + lump-sum payment package (to start a small non-agricultural activity conditional on the HH developing a business plan). All beneficiaries exposed to repeated information and communication efforts during enrollment and paydays. Beneficiaries also expected to attend regular meetings with local programme promoters to discuss programme objectives and conditionalities.	CCT: USD 20 if no children or children <7. HHs with children aged 7–15 additional USD 8 per HH and USD 2 per child. Approximately 15% of per capita expenditures. Lump-sum package: USD 175 in May 2006 and USD 25 in Sept 2006. Approximately 11% of per capita expenditures.	Every 2 months
Asia
13	Ahmed et al. (2016)b	Bangladesh	Transfer Modality Research Initiative	Eco-Social Development Organization (Govt), operational and technical support from World Food Programme	UCT and in-kind, Plus	Ultra-poor HHs with at least one child 0–24 months and not receiving benefits from other SSN interventions	Mothers	Five treatment arms: UCT; Food transfer; 50% UCT, 50% food transfer; UCT + BCC (north only); Food transfer + BCC (south only). BCC involved weekly group meetings and focus on nutrition. Community nutrition workers also visited beneficiaries in their homes to observe level practice and encourage the adoption of positive behaviours.	UCT (full): Taka 1500 (∼USD 18). Approximately 25% HH consumption expenditures. Food transfer (full): 30 kg of rice, 2 kg lentils and 2 l fortified cooking oil.	Monthly
Middle East and North Africa (MENA)
14	Abu-Hamad et al. (2014)a	Occupied Palestinian Territories	Palestinian National Cash Transfer Program	Ministry of Social Affairs (Govt)	UCT and in-kind	Extremely poor HHs (consumption-based PMT) with specific consideration for vulnerable groupsf	HH representative	Combines two cash transfers: EU-funded Social Hardship Case program; Social Safety Net Reform Project. HHs also entitled to other state-provided assistance, including health insurance, school fee waivers, and cash grants to help with one-off emergency needs. All beneficiaries receive in-kind food assistance through non-state provisions.	750–1800 NIS (∼USD 195–468) per HH (to bridge the 50% poverty gap)	Quarterly

Notes on abbreviations: BBC, behaviour change communication; BL, baseline; CCT, conditional cash transfer; EA, enumeration area; HH, household; KES, Kenyan Shillings; NIS, New Israeli Shekel; OVC, orphans or vulnerable children; PMT, proxy means test; PW, public works; RS, randomly selected; SRH, sexual and reproductive health; SSN, social safety net; TZS, Tanzanian shilling; UCT, unconditional cash transfer; USD, United States dollar; VP, violence prevention; ZAR, South African rand.

a

Peer-reviewed journal article.

b

Working paper or technical report.

c

Ultra-poor defined as belonging to the lowest expenditure quintile. OVC defined as having one or more deceased parent, or whose parent or main caregiver is chronically ill.

d

Indicator also included beneficiaries of the Foster Grant, targeted to HHs with children <19 years (or 21 years if attending school) who are orphaned, abandoned, at risk, abused or neglected. However, only 23 (0.7%) of 3401 participants received a Foster Grant, suggesting that preventative effects were probably driven by the CSG.

e

The CCT offers (i) a grant (TZS 4000 per month) to HHs with pregnant women or children under 5 who are in compliance with pre and post-natal exams and regular child health check-ups; (ii) a grant (TZS 2000) to households with children with demonstrating an 80% primary school attendance rate; (iii) an individual grant (TZS 4000) for children demonstrating an 80% lower secondary school attendance rate; and (iv) an individual grant (TZS 6000) for children demonstrating an 80% upper secondary school attendance rate where such services are available. PW component provides TZS 2300 per day (∼1 USD).

f

Female-headed HH, ≥1 HH member aged 65+, or members who are chronically ill, disabled or orphaned.

All SSN programme implementers were national or local government bodies with the exception of three programmes in SSA (the HIV Prevention Trails Network 068 [HPTN 068], the Malawi Zomba cash transfer programme and the AGI-K) and one in Asia (Bangladesh Transfer Modality Research Initiative), which were run by research study teams, NGOs or UN organizations (Column 4). While government-run programmes are often limiting in terms of research and programme design, this implies that results may have a higher degree of external validity and generalizability. Across programme typologies (Column 5), there is a clear majority of CCT and UCT programming (∼41%, or seven programmes each), followed by in-kind programming (12% and two programmes) and a minority of PW/CfW programming (6% or one program).⁹ Further, there are clear differences by region, where cash transfers in SSA tend to have a mix of UCT/CCTs, while in LAC nearly all programmes are CCTs. Importantly, nearly 29% of programmes have some type of ‘plus’ component, for example, linkages to services, information or training.

Virtually all programmes implemented some type of means-based targeting to identify extremely poor households as beneficiaries. In SSA, typically a vulnerability criteria was operationalized, including OVC or labour-constrained categories and in several cases adolescent girls were specifically targeted in HIV-risk motivated programming. LAC programming has a greater focus on early childhood development (ECD) (health and primary education) alongside income-based targeting. In nearly all cases (Column 6), with the exception of three studies in SSA (the Malawi Zomba cash transfer program, the AGI-K and the South African HPTN 068), benefits were given to an adult household member, often designated as primary caregivers (generally females). In the three SSA programmes where benefits were given directly to adolescent girls, additional benefits were often given to an adult household member.

Programme operational details vary widely. On one hand, very basic models, such as the Kenyan Cash Transfer for Orphans and Vulnerable Children (Kenyan CT-OVC) or South African Child Support Grant (CSG) give cash payments with simple messaging about use with no additional conditions or components attached. On the other hand, the majority of the LAC implementation includes more complex conditions around human capital development, such as school attendance and enrolment requirements and health components, including health check-ups, nutrition trainings or child vaccinations. One notable design variation is the extent that programmes explicitly incorporate violence-specific components. We find only one programme that has a specific violence prevention component, namely community conversations to address norms around the value of adolescent girls in the AGI-K.

Finally, benefit levels range from ∼6 to 25% of baseline household income or expenditures with the large majority of benefits in the 10–15% range (Column 9).¹⁰ Approximately half of the programmes involved a flat benefit rate per beneficiary or household, while the remaining implemented a variable benefit rate, generally based on household composition (including number of children within specific age ranges or attending different school levels). Unless tied to specific cycles (e.g. education terms), or given as a lump sum transfer, most programmes delivered benefits on a monthly basis (Column 10).

Summary of evaluations and research findings

Table 2 summarizes the evaluation or research findings of the 14 studies summarized in Table 1 (region of origin and by country alphabetically). Nearly, all study designs involved some sort of experimental design, primarily either cluster, household or individual-level randomized controlled trials (RCTs) (Column 2). The remaining employed quasi-experimental techniques, including propensity score matching (PSM) or other matching techniques or natural experiments. The Palestine paper (Abu-Hamad et al. 2014) was the only truly mixed methods evaluation reviewed.

Table 2.

Review of core papers (evaluation components)

No	Authors	Study design	Data (years)	Sample (size)	Violence outcome	Measure(s)	Baseline Mean	Measure of effect(s)	Hypothesized mechanism
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Sub-Saharan Africa (SSA)
1	Handa et al. (2014)a	cRCT (exp.)	Primary data collection in 28 locations (14 T/14 C) in 7 districts: Kisumu, Homa Bay, Migori, Suba, Nairobi, Garissa, and Kwale (2007–11d)	551 youth aged 15–25 (≤3 youth per HH)	Sexual exploitation	Ever received or given gifts	14.86c	aOR: 0.843 (95% CI 0.461–1.539)	Suggests two pathways: (i) increase in schooling may reduce sexual risk behaviours, (ii) increased economic well-being may reduce engagement in transactional sex. Authors hypothesize that especially for young women, increased economic well-being may reduce dependence on male partners (unwanted sexual relationships).
						Ever received or given gifts (females)	n/r	aOR: 0.979 (95% CI 0.439–2.186)
						Ever received or given gifts (males)	n/r	aOR: 0.711 (95% CI 0.295–1.713)
2	Rosenberg et al. (2014)a	cRCT (exp.)	Primary data collection in 28 locations (14 T/14 C) in 7 districts: Kisumu, Homa Bay, Migori, Suba, Nairobi, Garissa, and Kwale (2007–11d)	684 youth aged 15–25 who reported sexual partner in last 24 months (≤3 youth per HH)	Sexual exploitation (female)	Transactional sex	19.3c	aOR: 0.79 (95% CI 0.40–1.58)	Suggests cash put recipients in contact with safer sex partners through two mechanisms: (i) keeping youth in school, where they are more likely to find partners of their own age, and (ii) offset the economic motive to engage in transactional sex.
						Transactional sex (≤21 years)	25.4c	aOR: 0.65 (95% CI 0.30–1.42)
						Transactional sex (in school)	27.8c	aOR: 0.38 (95% CI 0.13–1.11)
						Transactional sex (out of school)	14.6c	aOR: 1.27 (95% CI 0.47–3.43)
						Transactional sex (< median BL expenditures)	21.1c	aOR: 0.86 (95% CI 0.32–2.31)
						Transactional sex (> median BL expenditures)	18.0c	aOR: 0.65 (95% CI 0.23–1.80)
					Sexual exploitation (male)	Transactional sex	4.1c	aOR: 1.57 (95% CI 0.60–4.07)
						Transactional sex (≤21 years)	5.5c	aOR: 0.96 (95% CI 0.27–3.40)
						Transactional sex (in school)	3.5c	aOR: 1.29 (95% CI 0.24–6.93)
						Transactional sex (out of school)	4.7c	aOR: 1.57 (95% CI 0.60–4.07)
						Transactional sex (< median BL expenditures)	7.0c	aOR: 1.81 (95% CI 0.56–5.88)
						Transactional sex (> median BL expenditures)	2.3c	aOR: 0.70 (95% CI 0.20–2.51)
Sub-Saharan Africa (SSA)
3	Austrian et al. (2015b, 2016a)	Individual and cRCT (exp.)	Primary data collection in urban settlements in Kibera (T), Huruma (C) and rural Wajir County (T) (2015–17–19)	4544 girls aged 11–14 at BL	Emotional violence	Emotional violence (lifetime; Kibera; ages 11–12)	0.21	Ongoingf	Suggests the combination of interventions is key: (i) CCT expected to reduce risks (sex/marriage) driven by a lack of economic resources and to increase educational attainment; (ii) community dialogues expected to prevent violence through changes in attitudes, perceptions and norms, reducing non-consensual sex and increasing educational attainment; (iii) safe spaces expected to improve girls' individual self-esteem and knowledge.
						Emotional violence (lifetime; Kibera; ages 13–14)	0.23
						Emotional violence (lifetime; Wajir; ages 11–12)	0.02
						Emotional violence (lifetime; Wajir; ages 13–14)	0.03
					Physical violence	Physical violence (lifetime; Kibera; ages 11–12)	0.24
						Physical violence (lifetime; Kibera; ages 13–14)	0.23
						Physical violence (lifetime; Wajir; ages 11–12)	0.02
						Physical violence (lifetime; Wajir; ages 13–14)	0.02
					Sexual violence	Sexual violence (lifetime; Kibera; ages 11–12)	0.05
						Sexual violence (lifetime; Kibera; ages 13–14)	0.08
					Sexual abuse	Indecent touching from boys/men (6 months; Kibera; ages 11–12)	0.05
						Indecent touching from boys/men (6 months; Kibera; ages 13–14)	0.07
						Indecent touching from boys/men (6 months; Wajir; ages 11–12)	0.14
						Indecent touching from boys/men (6 months; Wajir; ages 13–14)	0.11
					Emotional, physical and sexual violence	Any emotional, physical, sexual violence (lifetime; Kibera; ages 11–12)	0.33
						Any emotional, physical, sexual violence (lifetime; Kibera; ages 13–14)	0.35
						Any emotional, physical, sexual violence (lifetime; Wajir; ages 11–12)	0.04
						Any emotional, physical, sexual violence (lifetime; Wajir; ages 13–14)	0.03
Sub-Saharan Africa (SSA)
4	Baird et al. (2012)a	cRCT (exp.)	Primary data collection in 176 EAs (88 T/88 C) from urban and rural Zomba district (2007–08g)	3796 never married girls aged 13–22 at BL	Sexual abuse	Sexual partner aged ≥25 years (BL schoolgirls)	0.03c	aOR: 0.21 (95% CI 0.07–0.62)	Results suggest an income effect, as well as a potential link between education and reduced risky sexual behaviours (however, Wald tests fail to distinguish between aORs of CCT vs UCT arms, suggesting income effect dominates).
						Sexual partner aged ≥25 years (BL dropouts)	0.10c	aOR: 0.79 (95% CI 0.42–1.50)
						Sexual partner aged ≥25 years (CCT vs C arm)	0.03c	aOR: 0.08 (95% CI 0.01–0.60)
						Sexual partner aged ≥25 years (UCT vs C arm)	0.03c	aOR: 0.36 (95% CI 0.11–1.19)
5	Cluver et al. (2013)a	Individual PSM (quasi-exp.)	Primary data collection in two rural and two urban health districts in Mpumalanga and Western Cape (September 2009 to December 2011)	2668 adolescents aged 12–18 years (one RS per HH)	Sexual exploitation (female)	Transactional sex (incidence)	0.05c	aOR: 0.42 (95% CI 0.22–0.79)	Suggests cash targets sex and risk-specific risk behaviours and may interrupt those risks driven by economic necessity.
						Transactional sex (prevalence)	NR	aOR: 0.43 (95% CI 0.23–0.78)
					Sexual abuse (female)	Age-disparate sex (incidence)	0.04c	aOR: 0.28 (95% CI 0.13–0.62)
						Age-disparate sex (prevalence)	NR	aOR: 0.36 (95% CI 0.18–0.70)
					Sexual exploitation (male)	Transactional sex (incidence)	0.03c	aOR: 0.93 (95% CI 0.41–2.11)
						Transactional sex (prevalence)	NR	aOR: 1.00 (95% CI 0.45–2.20)
					Sexual abuse (male)	Age-disparate sex (incidence)	0.03c	aOR: 0.69 (95% CI 0.30–1.59)
						Age-disparate sex (prevalence)	NR	aOR: 0.68 (95% CI 0.31–1.52)
6	Pettifor et al. (2016)a	Individual RCT (exp.)	Primary data collection in 28 villages in Ehlanzeni district (2012–15h)	2533 young women aged 13–20 at BL	Physical violence	Physical violence from a partner (12 months)	0.28c	RR: 0.66 (95% CI 0.59–0.74)	Suggests cash may enable young women to leave or not engage in violent relationships.
					Sexual abuse	Partner age difference ≥5 years	0.20	RR: 0.90 (95% CI 0.72–1.12)
					Sexual exploitation	Transactional sex	0.14	RR: 0.95 (95% CI 0.78–1.15)
7	Palermo et al. (2017)b	cRCT (exp.)	Primary data collection in 84 villages (48 T/36 C) in 8 districts (Misungwi, Kahama, Kilola, Kisarawe, Handeni, Mbogwe, Itilima and Uyui) (2015–17)	1357 youth aged 14–28 at BL	Emotional violence	Emotional violence (12 months)	0.55	Ongoing	Suggests cash transfers have potential to effect risk of violence through (i) increased economic security, (ii) increased school attendance, (iii) reduced household stress, (iv) changes in parental time use and supervision and (v) decreased labor force participation among children and adolescents.
					Physical violence	Physical violence (12 months)	0.29
					Sexual exploitation	Transactional sex (12 months)	0.13
					Sexual violence	Sexual violence (12 months)	0.22
						Sexual violence (lifetime)	0.19
Latin America and the Caribbean
8	Rasella et al. (2013)a	Mixed ecological design (quasi-exp.)	Longitudinal data from several databases at the municipality leveli (2004–09)	2853 municipalities	Homicide (under-5 mortality rate/1000 live births)	External causesj (intermediate vs low coveragek)	1.23	RR: 1.03 (95% CI 0.95–1.13)	Mortality attributed to external causes was originally included as a robustness check and no programme impacts were expected.
						External causes (high vs low coverage)	1.23	RR: 0.92 (95% CI 0.79–1.06)
						External causes (consolidated vs low coverage)	1.23	RR: 0.92 (95% CI 0.72–1.16)
9	Rodríguez (2015)b	Variation in payment date over time (quasi-exp.)	Administrative data from Institute of Legal Medicine (2007–10) alongside programme administrative monthly payment data	1020 municipalities	Physical violence	Violence against minors (per 10 000 people)	0.54	OLS: −0.00637 (0.0181)	Although no significant impact found on violence against minors (similar to other community violence examined), there was a significant impact on IPV among adult women (6% reduction). Authors hypothesize mechanisms of women's empowerment and poverty-related stress, however, are unable to directly test pathways.
10	Paxson and Schady (2010)a	cRCT (exp.)	Primary data collection in 118 parishes in 6 provinces (79 T/39 Cl) (2003/04–05/06)	5547 children aged 3–7 years (follow-up)m	Violent discipline	HOME scoren(first quartile, poorest)	NR	OLS: −0.318*	Suggests potential of transfers to improve maternal mental health and the quality of parenting received by children.
						HOME score (top three quartiles, wealthiest)	NR	OLS: −0.031 (NS)
11	Fernald and Hidrobo (2011)a	cRCT (exp.)	Primary data collection in 118 parishes in 6 provinces (79 T/39 Cl) (2003/04–05/06)	1196 children aged 12–35 months (follow-up)	Violent discipline	Harsh parentingo (HOME score)	2.4	OLS: 0.21 (NS)	Suggests income may improve psychological well-being through reductions in subjective feelings of financial strain and deprivation. As poverty affects the ways parents monitor, provide stimulation, and respond to children's needs, increased access to economic resources may allow parents to be more responsive, warm and consistent.
						Harsh parenting (HOME score, rural)	NR	OLS: −0.49 (NS)
						Harsh parenting (HOME score, urban)	NR	OLS: 1.06 (P< 0.10)
Latin America and the Caribbean
12	Macours et al. (2012)a	cRCT (exp.)	Primary data collection in 106 communities (56 T, 50 C) in 6 rural municipalities (2005–06–08/09)	4021 HHs with children aged 0–6 years (3002 T/1019 C)	Violent discipline	HOME scorep (2006, full intervention)	4.018	OLS: −0.265 (0.291)	No hypothesized mechanism, however, suggests parenting behaviour/home environment is a determinant (potential risk factor) of child development broadly.
						HOME score (2006; with HH expenditure controls)	4.018	OLS: −0.088 (0.284)
						HOME score (2006, CCT only vs C)	4.018	OLS: −0.204 (0.308)
						HOME score (2006, CCT vs lump-sum payment)	3.76	OLS: 0.103 (0.182)
						HOME score (2008, full intervention)	4.072	OLS: −0.081 (0.120)
						HOME score (2008; with HH expenditure controls)	4.072	OLS: −0.078 (0.119)
						HOME score (2008, CCT only vs C)	4.072	OLS: −0.128 (0.135)
						HOME score (2008, CCT vs lump-sum payment)	3.83	OLS: 0.151 (0.142)
Asia
13	Ahmed et al. (2016)b	cRCT (exp.)	Primary data collection in 250 villages in northwest (Kurigram and Rangpur) 250 villages in south (Bhola, Patuakhali, Pirojpur, Bagerhat and Khulna) (2012–13–14)	5000 HH with child aged 0–24 months at baseline	Violent discipline	Parent hit child last week	Ongoing	Ongoing	NR
						Parent hits back or other physical punishment if child hit parents
Middle East and North Africa
14	Abu-Hamad et al. (2014)a	Difference in mean (quasi-exp.)	Primary data collection in the Gaza Strip (2013)	Quant: 641 HHs (315 T/326 C) and 200 adolescents aged 11–17 (RS at every third HH) Qual: beneficiary children and adultsq	Violent discipline	Caregivers did not allow any child to leave the house (past month)	NA	Mean: 43.5 (T), 46.0 (C) (NS)	In qualitative discussions caregivers explained their behaviour (violent discipline) in terms of excessive stress and frustration. In addition to poverty-induced stressors, a significant number of respondents highlighted the chronic psychological ill-health they and their household members faced owing to the ongoing conflict. In qualitative discussions, a number of adolescents, while generally positive, emphasized poverty stood in the way of positive peer relations.
						Caregivers shook any child (past month)		Mean: 47.6 (T), 52.0 (C) (NS)
						Caregivers yelled/shouted at any child (past month)		Mean: 60.6 (T), 65.8 (C) (NS)
						Caregivers slapped any child with bare hand/object (past month)		Mean: 41.0 (T), 48.9 (C) (NS)
						Caregivers called any child dumb/lazy (past month)		Mean: 33.4 (T), 38.8 (C) (NS)
					Bullying	Children reported classmates treat them with respect		Mean: 89.8 (T), 95.7 (C) (NS)
						Children reported classmates tease them at school		Mean: 19.8 (T), 13.2 (C) (NS)
					Violent discipline	Children reported teachers treat them worse than other children		Mean: 10.4 (T), 12.0 (C) (NS)

Notes on abbreviations: Study design: cRCT, cluster randomized control trial; exp, experimental; PSM, propensity score matching; quasi-exp, quasi-experimental; RCT, randomized control trial. Measure of effect(s): aOR, adjusted odds ratios; CI, confidence interval; ME, marginal effects; NS, non-significant; OLS, ordinary least squares; OR, odds ratio; RR, risk ratio. Other: BFP, Bolsa Familia Program; BL, baseline; C, control; CCT, conditional cash transfer; HH, household; HOME, Home Observation Measurement of the Environment; IPV, Intimate partner violence; NA, not applicable; NR, not reported; RS, randomly selected; T, treatment; UCT, unconditional cash transfer.

Bold indicates statistical significance at the p < 0.10 level or higher

a

Peer-reviewed journal article.

b

Working paper or technical report.

c

Mean of control group at endline.

d

Module on sexual behaviour with up to three HH members aged 15–25 implemented at follow-up in 2011.

e

Qualitative research including semi-structured interviews of beneficiaries and key adult stakeholders in 2016 and 2018 will supplement and complement the quantitative research findings.

f

Data were also collected on age-disparate sex, transactional sex and forced sexual debut in Kibera and Wajir, and on lifetime sexual violence in Wajir, however, not reported due to small sample size.

g

Data on partner aged 25 years or older collected at 12-month follow-up only.

h

Young women were interviewed at baseline and annually for up to three follow-up visits until the study completion date (March 2015), or at their planning high school completion, whichever came first.

i

Information systems from the Ministry of Health (health-related variables), the Ministry of Social Development (BFP coverage), and the Brazilian Institute of Geography and Statistics (socioeconomic variables); as well as the national census databases (covariates).

j

Mortality attributed to ′external causes′ includes transport accidents, homicides, and accidental injuries. The percentage classified as homicide is approximately 7%.

k

BFP municipality population coverage: Low (0.0–17.1%), Intermediate (17.2–32.0%), High (>32.0%), and Consolidated (>32.0% and target population coverage ≥100 for at least 4 years).

l

Smallest administrative unit in Ecuador. In total, 51 rural and 28 urban treatment parishes, and 26 rural and 13 urban control parishes were included in the sample.

m

Analyses focuses on rural HHs only (51 T/26 C parishes); no evidence of impacts found on any outcome considered for urban areas. The later rollout of the programme in urban areas meant HHs had less exposure to the programme at follow-up.

n

HOME score is based on the Home Observation for Measurement of Environment (HOME) scale, measures maternal punitiveness and lack of warmth, with lower values corresponding with better outcomes. The score includes 8 out of 11 interviewer assessed items. HOME score is standardized to z-score. Direct programme impact NR, only variations by income quartile.

o

HOME score uses two sub-scales of the HOME scale, measures harsh parenting, with lower values corresponding with better outcomes. The score includes 11 observational items (ranges 0–11) divided into responsivity (said kind words or phrases at least twice; responded verbally to any child’s vocalization at least once; told any child the name of an object at least once; praised any child at least twice; conveyed positive feelings toward or about any child; caressed or kissed any child at least once) and punitiveness (yelled at any child; expressed annoyance or hostility towards any child; hit any child; scolded or criticized any child; forbade any child from doing something more than three times). Item 3 on the punitiveness sub-scale is considered violent physical discipline, and Items 1, 2 and 4 on the same sub-scale are considered violent psychological discipline.

p

HOME score (shortened version of the HOME scale) includes 11 observational items (ranges 0–11) and measures positive parenting behaviours (responsivity—see items in table note o) and negative parenting behaviours (punitiveness—see items in table note o). Lower values correspond with better outcomes. One item is considered violent physical discipline (hit any child) and three items are considered violent psychological discipline (expressed annoyance or hostility towards any child; yells at any child; scolded or criticized any child). Each item received an answer of ‘yes’ or ‘no’. The HOME score is equal to the number of answers of ‘no’ to the positive items, plus the numbers of answers ‘yes’ to the negative items.

q

Qualitative data collection took place in June/July 2013 in two cites in the north of Gaza and Gaza city. Participants were purposefully selected from quantitative data to represent specific conditions such as disability or violence in the HH. In total, 71 children took part in small group discussions, 10 children took part in in-depth interviews, 2 households were observed, 6 in-depth interviews took place with caregivers, 14 adults took part in focus group discussions and 11 key informant interviews were undertaken with key stakeholders.

Significance levels are:

*

P <0.10;

**

P < 0.05;

***

P < 0.001.

Nearly all research involved primary data collection at several points before and after programme implementation (Column 3). The exceptions were two LAC studies, which used government and programme administrative data either alone or combined with primary data (Rasella et al. 2013; Rodríguez 2015). Within the primary data collections, a variety of tools were used, including both parental/caregiver reports, caregiver/child interactions or observational data and survey administration to children or adolescents themselves (Column 4). Quantitative sample sizes range from 551 adolescents (Handa et al. 2014) in Kenya to 5547 young children (Paxson and Schady 2010) in Ecuador.

The studies examined various childhood violence outcome indicators (Column 5). Overall, 57 indicators were found across the 11 completed studies. In total, five measures of physical violence only [three indicators of homicide, two indicators of other physical violence (dating or partner violence and violence against minors from administrative data)], 23 measures of physical and/or emotional violence (21 violent discipline indicators and two indicators of bullying) and 29 measures of sexual violence (20 indicators of sexual exploitation including transactional sex, and nine indicators of sexual abuse including age-disparate sex). In most cases, studies analysed results for the same indicator in a variety of ways (e.g. varying subgroups of the target population) or examining both prevalence and incidence. For completeness, we include all variations, regardless of statistical significance—as to not bias summary figures. This means that some studies have only one qualifying violence outcome (e.g. Rodríguez 2015); however, the number presented ranges up to 12 (e.g. Rosenberg et al. 2014). There is significant regional variation in the evidence across childhood violence indicators. For example, the vast majority of sexual abuse and exploitation measures come from SSA and are collected among adolescents (age 13 and above, particularly female adolescents), where there has been more attention to dynamics around youth HIV risk behaviours. Likewise, the bulk of violent discipline evidence, both physical and psychological, comes from LAC where greater emphasis has been placed on ECD and is collected among children under five or primary school aged children. Indicators which are relatively rare, such as homicide, are only collected in settings were administrative or large-scale programme data are available—relatively higher income LAC countries such as Brazil. Finally, we found no evidence for non-contact sexual violence.

Columns 6, 7 and 8 present the specific indicators operationalized in the data, the baseline mean (or control group at endline if not available) and the effect size with accompanying significance level or confidence interval. For the mean and effect size, we have maintained the same number of significant digits or reporting as in reviewed papers. Across all 57 indicators, approximately 11 (or 19%) are statistically significant at the P < 0.10 level or higher(denoted by bolded text). This percentage varies between category of violence examined: In total, 20% of physical violence only indicators [0% for homicide, 50% for other physical violence (100% dating or partner violence and 0% violence against minors)], 9% of physical and/or emotional violence indicators (10% for violent discipline and 0% for bullying) and 40% of sexual violence indicators (20% for sexual exploitation and 44% for sexual abuse) were statistically significant. In all cases when statistically significant findings are found, SSNs have a protective effect (reduce violence) among the treatment group in comparison to the control group.

In interpreting these findings, it is worth noting that for several studies with multiple variations of the same indicator, there was varying significance levels across sub-groups—indicating that even within settings or studies, heterogeneities may exist. In addition, although all studies have provided motivation and validation for their specific measures operationalized, it is worth noting that many of the baseline mean figures are low (<5%), and this limits the power of evaluations to detect statistically significant programme impacts. Further, there is variation in indicator definition between studies, which may drive some variation in results.

We do not consider evidence from high-income countries as part of the core review, as programmes tend to be complex bundled interventions and rely more heavily on social services and policy levers including tax breaks and benefits, with low comparability to programmes in LMICs. However, to enrich the discussion, we include similar summaries of evidence from nine studies in high-income countries (namely the USA and Canada) in the Supplementary Appendix (Bloom et al. 2000, 2002; Beecroft et al. 2002; Gennetian and Miller 2002; Fein and Lee 2003; Huston et al. 2008; Miller and Samson 2012; Cancian et al. 2013; Jacob et al. 2013). Across the nine studies, 44 measures of childhood violence are collected, 12 (27%) of which show significant protective effects, while one (2%) shows a significant increased risk effect due to the program. The majority of indicators analysed were specific to young children, including indicators of violent discipline or abuse. It should be cautioned that many of these effects are not full programme impacts, but rather test a variation in programme design, for example placing conditions on CfW programmes or allowing a tax pass through on child support benefits. In addition, these studies tend to be older when compared with the core papers reviewed here, with data being collected from mid-1990s to 2009 and often are limited to one state in the USA, thus potentially with lower generalizability.

Despite these evidence gaps, the three ongoing studies identified hold some promise to add to the evidence base. In particular, Austrian et al. (2016) is collecting a wide range of emotional, physical and sexual violence indicators within the context of a NGO-implemented adolescent girl’s bundled intervention in Kenya, including community conversations addressing violence. In addition, Palermo et al. (2017) have also collected emotional, physical and sexual violence among female youth aged 14–28 years at baseline in the context of the government of Tanzania’s CCT, UCT and PW, plus program. Finally, Ahmed et al. (2016) have completed an evaluation of a UCT and in-kind transfer programme with behaviour change and communication (BCC) components and plan to examine violent discipline among young children in Bangladesh—making this the only study in Asia catalogued to date.

Summary of mechanism proposed and evaluated

The hypothesized or investigated mechanism(s) outlined among completed studies in Column 9 generally cover all of the pathways described in our theoretical framework, with the exception of substance misuse, caregiver supervision of children and time spent in high-risk settings. Apart from economic security, which was implicit in nearly all the studies, the most commonly hypothesized mechanisms for changes in childhood violence experience (all highlighted by three studies each) were schooling (Baird et al. 2012; Handa et al. 2014; Rosenberg et al. 2014), changes in caregiving behaviours (Paxson and Schady 2010; Fernald and Hidrobo 2011; Macours et al. 2012), and parents’ improved mental health and psychosocial well-being (Paxson and Schady 2010; Macours et al. 2012; Abu-Hamad et al. 2014). The stress pathway was mentioned by two studies (Abu-Hamad et al. 2014; Rodríguez 2015), as was children’s problem or risk behaviours (Baird et al. 2012; Cluver et al. 2013). Finally, one study each mentioned the following pathways: changes in intra-household conflict through women’s empowerment (Rodríguez 2015) and empowerment which allows adolescent girls to leave abusive relationships (Pettifor et al. 2016). These pathways are largely not empirically tested in the papers reviewed, but rather hypothesized in the discussion after impacts were detected, or elicited from qualitative interviews.

Discussion and conclusion

Global initiatives seeking to promote evidence-informed practice around violence prevention and response have identified carefully designed economic empowerment programmes, including SSNs, as key interventions to prevent violence and reduce risk behaviours associated with it. These include THRIVES, developed by the US Centers for Disease Control and Prevention (CDC) (Hillis et al. 2015), and INSPIRE, a multi-national partnership led by the World Health Organization (WHO).¹¹ The most recent, INSPIRE, launched in July 2016, promotes three types of income and economic strengthening as ‘effective’ for reducing childhood violence: (1) cash transfers, (2) group saving and loans combined with gender equity training and (3) microfinance combined with gender norm training (Butchart and Hillis 2016). The latter two approaches have been under criticism by a number of recent reviews for their general lack of transformative effects for households and women specifically, with some evidence of negative effects, which calls into question the choice of these types of instruments for sustaining meaningful economic strengthening, as well as their potential to reduce childhood violence (Stewart et al. 2010, 2012; Duvendack et al. 2011; Vaessen et al. 2014; Banerjee et al. 2015).

We only considered the first instrument (i.e. cash transfers) in this review, and our conclusions from a comprehensive view of the evidence point to large variations in potential for impacts and substantial gaps in the literature on the intersection between SSNs and childhood violence. The lack of evidence is particularly striking when compared with the bodies of evidence on SSNs and other outcomes—including topics such as education, gender and empowerment or child nutrition—all of which have been topics of multiple evidence and systematic reviews as well as meta-analyses (Lagarde et al. 2007; Leroy et al. 2009; Baird et al. 2012; van den Bold et al. 2013; Bastagli et al. 2016; de Groot et al. 2016). Despite these findings, SSNs may play an important role in affecting the frequency and severity of childhood violence, aspects which are not necessarily captured in the current evidence base. In addition, it is also possible that SSNs have impacts on childhood violence indicators, particularly witnessing violence in the home and community, which were excluded from this review. For example, there is increasing evidence that cash transfers in particular can lead to reductions in IPV among adult women, hypothesized to work through many of the same mechanisms as proposed in this review (Bobonis et al. 2013; Hidrobo and Fernald 2013; Rodríguez 2015; Buller et al. 2016; Hidrobo et al. 2016). In addition, we do not consider outcomes of community violence or exposure to such violence. However, a recent review examining the potential of PW/CfW and cash transfer programming to promote economic growth and mitigate risks of violent conflict, largely through channels of productivity gains and economic growth, concludes that the body of evidence is also weak (Beazley et al. 2016).

Our review also points to significant regional variation. We found major evidence gaps by region, for example, there were no violent discipline specific indicators for children in SSA. Likewise we found no measures of sexual violence (including transactional and age-disparate sex) or violence perpetrated against adolescents by intimate partners in LAC. MENA had only one study with particular evidence and Asia had no completed studies—making it difficult to draw conclusions. We also find regional variation in hypothesized mechanisms, which may reflect programme typologies and objectives (as previously touched upon). For example, in LAC, where programmes are largely CCTs, the mechanisms tend to focus on schooling, caregiver stress and the parent–child relationship. Conversely, in SSA, where many programmes are implemented in the context of generalized HIV epidemics, aim to help households with OVCs and are unconditional (thus potentially having the ability to influence broader outcomes), the mechanisms tend to focus on schooling, adolescent risk behaviours, caregiver stress, exposure to high-risk environments, community norms and girls’ empowerment. The only study from MENA, and the only mixed methods study, hypothesized pathways focusing on stress and positive peer relationships driven by increases in household economic security (Abu-Hamad et al. 2014). Countries vary with respect to the generalized levels of poverty, the capacities of implementing agencies to monitor compliance with conditions, coordinate across sectors and offer appropriate referrals and support services, and the maturity of their social protection and child protection systems, and thus their ability to integrate or link the two to maximize violence prevention and response opportunities.

Future research on SSNs and childhood violence should not only focus on filling regional gaps, but also seek to unpack differential effects of various programme design and implementation variations. As a priority, all programmes must be designed to minimize potential harm. Despite their potential to reduce childhood violence, SSNs may also bring about unintended adverse effects, for example in terms of children’s participation in work activities (de Hoop and Rosati 2014; Chaffin and Mortenson Ellis 2015) or exposure to peer-violence, most notably bullying due to program-related stigma or discrimination (Abu-Hamad et al. 2014; Jones and Samuels 2015; Zhang 2016). In addition, while acknowledging differences in the sophistication of social protection and child protection systems across countries, there is potential for: (1) light-touch complementary interventions—for example, BCC strategies delivered within existing programme structures (e.g. programme registration, pay points and monitoring activities), utilizing the wide coverage and unique ability of SSNs to reach people at the level of the household, as done for example within the Kenya CT-OVC (Handa et al. 2014) or the Bangladesh Transfer Modality Research Initiative (Ahmed et al. 2016); (2) intensive or specialized complementary interventions—when dedicated programming is layered onto or integrated within the conditions of existing SSN structures, such as an adapted version of the Sinovuyo parenting programme from South Africa introduced as part of family development sessions conditional for beneficiaries of the Philippines’ Pantawid Pamilyan Pilipino Program (Madrid 2016) and (3) integration and system linkages—focused on case management and referral systems using single registries and programme monitoring as key entry points, with the strongest examples in countries with more evolved and institutionalized social protection systems, such as Chile’s Solidario (Ministerio de Desarrollo Social 2016). While the latter, followed by intensive complementary services, hold the greatest potential to prevent and respond to childhood violence, it should be emphasized that we found no completed rigorous evaluations of programmes utilizing a systems approach or integrating violence-specific components in LMICs. Further, SSNs bundled with light-touch interventions have the potential to reach large numbers of people and achieve impacts at the margin, without overburdening emerging or fragmented systems with extensive or specialized add-ons. It is important to note that these additional services should not necessarily be seen as programme ‘conditions’, which may inadvertently exclude the most vulnerable beneficiaries from accessing benefits.

In conclusion, we demonstrate that the linkages between non-contributory SSNs and childhood violence are understudied, but there exists some emerging evidence to suggest the potential of such programming to reduce aspects of multiple violence typologies across age, gender and regions. These effects may be significant and relevant from a policy perspective and help inform the global violence prevention debate as to the role of economic empowerment programmes. Yet, to fully understand their potential, it is essential for future research to collect validated measures of childhood violence, where ethical and other objectives align, within rigorous impact evaluation (UNICEF 2006; Child Protection Monitoring and Evaluation Reference Group 2014). Further, to move beyond impacts to generalize to other settings and programme typologies, it is essential to include quantitative analysis testing pathways of impact, and where they are found, explore the mechanisms behind them using qualitative approaches. Finally, it is important to emphasize that SSNs, as stated, are never designed, nor should be designed, with violence prevention as the primary objective, and thus should not be looked at alone to reduce the broad-based and interrelated risks and vulnerabilities linked to childhood violence. With this caveat in mind, integrated social and child protection systems and SSN programming appear to be a promising way forward to reduce risk associated with childhood violence.

Ethical approval

No ethical approval was required, as no primary data from human subjects was utilized as part of this research.

Supplementary Data

Supplementary data are available at HEAPOL online

Footnotes

1

As the focus of the review is on the level of the household, school feeding programmes were excluded. Insurance schemes were also excluded unless they were part of a bundled programme—as the mechanisms are likely to be quite different than for traditional SSNs.

2

We refer to ‘plus’ as any additional programming which may be bundled with basic SSNs, while recognizing that this design variation is infinitely diverse, ranging from light add programming such as messaging and behavioural change communication, to large multi-component interventions which may even be larger than the economic component of the SSN itself.

3

While exposure to IPV among adult household members is often considered a form of emotional violence, we do not include it in our definition of emotional violence, as evaluation studies examining the link between SSNs and IPV rarely assess if children witnessed (were exposed) to IPV or not, and impacts of SSNs on IPV experienced by adults are being reviewed elsewhere.

4

Note that perpetration also includes self-harm (fatal and non-fatal). Despite search criteria that allowed us to assess potential impacts of SSNs on suicide, we did not find any studies which measured impact on this outcome.

5

In total, six of the 14 studies reviewed include measures in a sample of individuals age 18 or over (ranging up to age 28).

6

Databases including google scholar, PubMed and ScienceDirect were searched. In addition, journals searched include: Child Abuse and Neglect, Child Maltreatment, Journal of Interpersonal Violence Children, Youth Services Review, The Lancet and Journal of Adolescent Health. Institutional websites searched include: World Bank, UNICEF, Overseas Development Institute (ODI), Sexual Violence Research Initiative (SVRI), Child Protection in Crisis (CPC) Learning Network, Socialprotection.org and the Transfer Project.

7

These terms included for example: child* OR adolesc* OR girl* OR boy* OR youth OR “young people” AND “social protection” OR “cash transfer*” OR “food transfer*” OR “in-kind transfer*” OR “public work*” OR voucher* OR “fee waiver*” OR “social assistance” OR “social safety net*” OR “food stamps.”

8

For example, Rasella et al. (2013) examine Brazil’s Bolsa Familia programme and meet the inclusion criteria for this review by examining under five death due to ‘external causes’ which includes homicide. However, in further discussion with authors, they note that only 7% of the total external causes are homicide cases, whereas the majority is accidents, exemplifying this limitation (personal communication with authors 12 July 2016). In other instances, including Macours et al. (2012) who operationalize the HOME score, the measure combines indicators of physical and emotional violence with those related to more general caregiving behaviours (see Table 2, notes for specific definitions by study). Therefore, one could argue that it is better to refer to these composite measures as general measures of a negative home environment, rather than violence per se (personal communication with lead author, 10 November 2016).

9

Note that some programmes could represent several typologies in one, if they combine multiple components (for example, a CCT with a PW/CfW component), thus totals do not sum to the number of distinct programming types. In addition, at times, multiple studies evaluated the same programme, in which case, it is only counted once.

10

Not all evaluations provided information comparing benefit levels to baseline income or expenditure (four studies did not). In particular, this information was missing for programmes with multi-tiered variable benefits.

11

The INSPIRE package was led by the WHO in collaboration with the CDC, End Violence Against Children: The Global Partnership, the Pan American Health Organization (PAHO), the President’s Emergency Program for AIDS Relief (PEPFAR), Together for Girls, the United Nations Children’s Fund (UNICEF), United Nations Office on Drugs and Crime (UNODC), United States Agency for International Development (USAID) and the World Bank. The seven evidence-based strategies include: (1) implementation and enforcement of laws, (2) norms and values, (3) safe environments, (4) parent and caregiver support, (5) income and economic strengthening, (6) response and support services and (7) education and life skills.

Acknowledgements

We would like to thank the following individuals for their expert consultations and review of this paper: AK Shiva Kumar, Alina Potts, Audrey Pereira, Audrey Pettifor, Bernadette Madrid, Heidi Loening-Voysey, Jacob de Hoop, Jasmina Byrne, Joshua Chaffin, Kathryn Maguire Jack, Leah Prencipe, Lorraine Sherr, Mary Catherine Maternowska, Matthew Morton, Mayke Huijbregts, Meghna Ranganathan, Paola Pereznieto, Rachael Sabates-Wheeler, Ramya Subrahmanian, Richard de Groot, Simone Cecchini, Sudhanshu Handa, Tomoo Okubo and Vivien Stern, as well as authors of the reviewed papers. We are grateful for support on grant administration from Laura Meucci, Michelle Godwin, Prerna Banati and Cinzia Bruschi of the UNICEF Office of Research—Innocenti.

Funding

This work was supported by the Oak Foundation via the Network of European Foundations, facilitated by the Know Violence in Childhood Initiative. Salary support for Amber Peterman and Tia Palermo provided to UNICEF Office of Research—Innocenti by the UK Department for International Development and the Swedish International Development Cooperation.

Conflict of interest statement. None declared.