Social Science & Medicine 262 (2020) 113194

Contents lists available at ScienceDirect

Social Science & Medicine

journal homepage: www.elsevier.com/locate/socscimed

Parenting, mental health and economic pathways to prevention of violence T

against children in South Africa
L. Cluvera,b,∗, Y. Shenderovicha,c, F. Meincki,j, M.N. Berezina,d, J. Doubta, C.L. Warde,
J. Parra-Cardonaf, C. Lombardg, J.M. Lachmana,h, C. Wittesaelea, I. Wesselsa,e, F. Gardnera,
J.I. Steinerta,k
a
Centre for Evidence-Based Intervention, Department of Social Policy & Intervention, University of Oxford, United Kingdom
b
Department of Psychiatry and Mental Health, University of Cape Town, South Africa
c
Institute of Criminology, University of Cambridge, United Kingdom
d
Department of Applied Psychology, New York University, New York, USA
e
Department of Psychology, University of Cape Town, South Africa
f
Steve Hicks School of Social Work, The University of Texas at Austin, Texas, USA
g
Biostatistics Unit, South African Medical Research Council, South Africa
h
MRC/CSO Social and Public Health Sciences Unit, University of Glasglow, United Kingdom
i
School of Social and Political Science, University of Edinburgh, United Kingdom
j
Optentia, Faculty of Health Sciences, North-West University, South Africa
k
TUM School of Governance, Technical University of Munich, Germany

ARTICLE INFO ABSTRACT

Keywords: Background: Parenting programs based on social learning theory have increasing empirical evidence for reducing
Violence violence against children. Trials are primarily from high-income countries and with young children. Globally, we
Parenting know little about how parenting programs work to reduce violence, with no known studies in low or middle-income
Depression countries (LMICs). This study examines mechanisms of change of a non-commercialized parenting program, Parenting
Adolescence
for Lifelong Health for Teens, designed with the World Health Organization and UNICEF. A cluster randomized trial
Alcohol
Poverty
showed main effects on parenting and other secondary outcomes. We conducted secondary analysis of trial data to
investigate five potential mediators of reduced violence against children: improved parenting, adolescent behaviour,
caregiver mental health, alcohol/drug avoidance, and family economic strengthening.
Methods: The trial was implemented in rural South Africa with 40 sites, n = 552 family dyads (including
adolescents aged 10–18 and primary caregivers). Intervention sites (n = 20) received the 14-session parenting
program delivered by local community members, including modules on family budgeting and savings. Control
sites (n = 20) received a brief informational workshop. Emotional and physical violence against children/
adolescents and each potential mediator were reported by adolescents and caregivers at baseline and 9–13
months post-randomisation. Structural equation modelling was used to test simultaneous hypothesized pathways
to violence reduction.
Results: Improvements in four pathways mediated reduced violence against children: 1) improved parenting
practices, 2) improved caregiver mental health (reduced depression), 3) increased caregiver alcohol/drug
avoidance and 4) improved family economic welfare. Improved child behaviour was not a mediator, although it
was associated with less violence.
Conclusions: Simultaneously bolstering a set of family processes can reduce violence. Supporting self-care and
positive coping for caregivers may be essential in challenging contexts. In countries with minimal or no eco­
nomic safety nets, linking social learning parenting programs with economic strengthening skills may bring us
closer to ending violence against children.

Corresponding author. Department of Social Policy and Intervention, Barnett House, 32 Wellington Square, Oxford, OX1 2ER, United Kingdom.
∗

E-mail addresses: [email protected] (L. Cluver), [email protected] (Y. Shenderovich), [email protected] (F. Meinck),
[email protected] (M.N. Berezin), [email protected] (J. Doubt), [email protected] (C.L. Ward), [email protected] (J. Parra-Cardona),
[email protected] (C. Lombard), [email protected] (J.M. Lachman), [email protected] (C. Wittesaele),
[email protected] (I. Wessels), [email protected] (F. Gardner), [email protected] (J.I. Steinert).

https://doi.org/10.1016/j.socscimed.2020.113194
Received in revised form 3 July 2020; Accepted 5 July 2020
Available online 21 July 2020
0277-9536/ © 2020 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
L. Cluver, et al. Social Science & Medicine 262 (2020) 113194

1. Introduction both? This study tests these five hypothesized mediation pathways,
within a cluster RCT of a parenting program with economic strength­
Violence against children (VAC) – both physical and emotional – has ening components, for families of adolescents in South Africa.
severe detrimental impacts on children and adolescents (WHO, 2020).
Long-term sequelae include increased mortality and morbidity, and 2. Methods
impaired neurological function, mental health, education, and em­
ployment (Moffitt & Klaus-Grawe Think Tank, 2013; Teicher et al., 2.1. Intervention development
2016). Prevalence studies find elevated rates in low- and middle-in­
come countries (LMIC), with Africa as the most affected region (Hillis Parenting for Lifelong Health is an initiative co-developed by aca­
et al., 2016). demics, the World Health Organisation (WHO), and UNICEF. Its goal is
Substantial evidence, primarily from high-income settings, suggests to develop and test parenting programs to prevent violence against
that parenting programs based on social learning theory may reduce children in LMIC. In South Africa, Parenting for Lifelong Health for
violence against children (Barlow et al., 2006; Chen and Chan, 2016; Adolescents (locally named Sinovuyo Teen) was developed over five
Vlahovicova et al., 2017). Within LMIC there is a very new but emer­ years in collaboration with a local NGO, Clowns Without Borders South
ging evidence-base, that also finds positive impacts of social learning- Africa, and the National Department of Social Development. Stages of
based parenting programs on parenting and reducing violence (Knerr development and testing included qualitative piloting, adolescent ad­
et al., 2013; Parra-Cardona et al., 2018; Pedersen et al., 2019). visory group participation, input from 50 academic and programming
In developing effective parenting programs for LMIC, we need to experts, and two pre-post trials with sequential refinement of the
understand mechanisms or mediators of violence reduction. To date, manual, followed by a full RCT (Cluver et al., 2016a, 2016b, 2016c). In
there are no known studies in either high- or low-income countries that brief, the RCT found significant reductions in physical and emotional
quantitively investigate mediators from parenting programs to reduced abuse in caregiver report, whilst adolescents in both the intervention
violence against children. Two studies in Panama and Liberia with and control groups reported reductions in abuse. The intervention
younger children (Giusto et al., 2017; Mejia et al., 2016) used quali­ group reported improved parental supervision/monitoring and im­
tative methods to identify pathways of improved parenting behaviours, proved parental involved parenting, but not improved adolescent be­
increased parental self-efficacy, and improved child behaviour. There is haviour (adolescent and caregiver report). Caregivers in the interven­
some evidence of mediators from parenting programs to improved child tion group reported improved caregiver mental health, reduced
behaviour (rather than parental violence) in high-income countries caregiver substance use and improved household economic welfare
(Forehand et al., 2014), and a very few in LMICs (Puffer et al., 2017). (adolescents did not report on these). All program materials are open
These find pathways of improved positive, involved and supervisory access for non-profit use on the WHO and UNICEF websites
parenting, and reduced caregiver depression. (Shelleby et al., 2018). http://www.who.int/violence_injury_prevention/violence/child/
Reviews of observational studies of child maltreatment suggest as­ PLH-manuals/en/.
sociated factors that may be modifiable by parenting programs: low
parenting skills, caregiver depression or low self-esteem, caregiver 2.2. Study design and recruitment
stress, alcohol/drug use, and child behaviour problems (Meinck et al.,
2014; Stith et al., 2009). A pragmatic cluster RCT including 552 families (adolescents aged
In addition, economic deprivation creates stressful environments 10–18 and primary caregivers) in 40 communities (32 rural villages and
that can reduce parenting capacity (Gershoff et al., 2007; Parra- 8 peri-urban township clusters). The Eastern Cape – along with
Cardona et al., 2018; Meinck et al., 2014). Poverty, unemployment, and Limpopo – is South Africa's poorest province (Statistics SA, 2016), with
decreases in state benefit levels have all been associated with increased a remaining legacy of apartheid resulting in limited infrastructure and
child maltreatment (e.g. Steinberg et al., 1981; Yang, 2015). A sys­ poor service delivery, and the area in which the study took place was
tematic review in 2014 found no parenting programs in LMIC that in­ identified by the national government and UNICEF South Africa as a
corporated economic strengthening components (Marcus and Page, priority for child abuse prevention. The research took place in colla­
2014). Since then, a small number of programs have combined par­ boration with local traditional leaders and government. To reflect real-
enting programmes with economic strengthening, all showing effec­ world service delivery, recruitment in April–August 2015 (prior to site
tiveness in reducing parental violence towards children (Annan et al., randomisation) was informed by local chieftains, schools, community-
2013; Cluver et al., 2018 ; Ismayilova; Karimli, 2018). selected representatives, and door-to-door visits, and was presented as
No research has yet examined mechanisms of violence reduction for general support for families in raising adolescent children. In order to
programs that combine parenting and economic strengthening. prevent any risk of coercion, all referral sources were combined and
Although this trial only has two timepoints which test the outcome and many participants self-referred into the study after announcements in
possible mediators (baseline, followed by randomisation and then community meetings. Adolescents with physical or learning difficulties
follow-up), our trial protocol specified that we would explore mediating were included, unless these disabilities were so severe that they were
pathways from other studies of child abuse prevention and the pro­ unable to give informed consent or participate in any programme
gramme's theory of change. Reviews identified parenting, caregiver (n = 3). No other exclusion criteria applied, and we note that through
mental health and child behaviour as possible mediators of violence this pragmatic RCT recruitment approach, the sample included those
reduction. During in-depth qualitative research led by UNICEF that experiencing many other co-existing problems including severe abuse
took place alongside this trial (Doubt et al., 2018), program participants and other conditions such as mental health problems, HIV/AIDS, in­
identified two additional pathways of change. The first was reduced timate partner violence, and substance use. Adolescents identified their
stress on caregivers, leading to less use of alcohol and drugs as coping primary caregiver as ‘the person who looks after you most’. Inclusion
mechanisms. The second was reduced family conflict around money, criteria were that caregivers and adolescents had to live in the same
due both to shared budget planning and higher collaboration around dwelling for at least four nights per week, with no requirement for a
spending, and also due to improved income smoothing leading to re­ biological relationship.
duced end-of-month severe hardship, and consequent lower stress. Original power calculations for the trial did not include mediation
This paper thus asks: Is reduced violence primarily driven by social models, and therefore power calculations for these were not conducted
learning mechanisms (i.e., improved parenting through better child post-hoc (Dziak et al., 2018; Gelman, 2019). The trial was powered to
behaviour, caregiver mental health, and alcohol/drug avoidance) or by 80% and 95% confidence with 40 clusters for a minimum detectable
alleviation of family distress linked to poverty, or a combination of effect size of 0.36 for the main outcome of violence against children.

2
L. Cluver, et al. Social Science & Medicine 262 (2020) 113194

Power calculations were based on a two-level multi-level model. The 2.4. Measures
0.80 refers to the desired level of power and alpha 0.05 to the level of
statistical significance for the ability to detect the overall treatment All outcomes and potential mediators were measured for the past
effect. We present all relevant parameters for a power calculation in month, and are available online https://www.unicef-irc.org/research-
supplementary material for future studies that may wish to plan similar family-and-parenting/. Measures were reported independently by both
analyses (MacKinnon et al., 2007). adolescents and primary caregivers at baseline and 9–13 month follow-
Randomisation was stratified by rural/urban location and con­ up. Child and caregiver reports were combined (summed into a total
ducted after baseline using a random number generator by an in­ aggregate score) in order to explore family-level processes, incorporate
dependent, blinded statistician (CL). Complete randomisation within both adult and adolescent perspectives, and to reduce the impact of
strata used a ratio of 1:1 intervention: control. The sample included 270 social desirability bias in reporting physical abuse, emotional abuse,
families in the intervention arm and 282 families in the control arm parental supervision, involved parenting, and child problem behaviour.
(mean 14 families per cluster, SD 1.9). Blinding of participants and For caregiver depression, caregiver alcohol/drug use, and household
program providers is not feasible for parenting programs. economic welfare only caregiver report was measured: in rural contexts
Ethical approval was given by the University of Oxford (SSD/ it is often considered inappropriate to inform children about household
CUREC2/11–40), University of Cape Town (PSY2014-001), and gov­ finances, and two thirds of adolescents in this sample reported no in­
ernment Departments of Social Development and Education. The pro­ sight into the family's financial management.
tocol was published (Cluver et al., 2016) and the trial registered on the
Pan-African Clinical Trials Registry PACTR201507001119966 on 27/4/ 2.5. Outcome
2015, but did not include the exploratory mediation analyses conducted
in this paper. An independent trial steering committee oversaw trial Physical and emotional violence at baseline and follow-up were
conduct. Written informed consent was given by all adults and ado­ measured using adolescent and caregiver-report on 14 items of the re­
lescents. No monetary incentives were given for participation, but all levant subscales of the International Society for Prevention of Child
families were given snacks at baseline and small food parcels at follow- Abuse and Neglect Screening Tool for Trials (ICAST-Trial) (Meinck
up as thanks. Confidentiality was maintained unless participants were et al., 2018). Potential mediators: Mediators were measured at baseline
at risk of significant harm or asked for support. In cases of severe abuse, and follow-up, and analyses controlled for baseline values of all med­
rape, suicide attempts or other significant harm, 33 immediate referrals iator and outcome variables. Parenting comprised child and caregiver-
and follow-up assistance were made to social and health services. report on 20 items of the parental supervision and positive involved
parenting subscales of the Alabama Parenting Questionnaire (Frick,
1991). Caregiver depression used seven items from the Centre for
2.3. Procedures Epidemiologic Studies Depression Scale (Radloff, 1977). Caregiver al­
cohol and drug use used an adapted version of the WHO Alcohol Use
Self-report tablet-based questionnaires were completed by primary Disorders Identification Test (3 items) and was coded ordinally as no,
caregivers and adolescents at baseline and 9–13 months (mean 12 regular, and very frequent alcohol or drug consumption (Saunders
months) post-randomisation. A very brief questionnaire including only et al., 1993). Child and caregiver-report of child problem behaviour
primary outcomes was completed at one-month post-program, and used the 35-item Child Behavior Checklist rule-breaking and aggression
therefore was not used in this analysis. Data collectors supported the sub-scales (Achenbach, 2000). For each of the above scales, individual
process and audio-computer assisted self-interviewing was modified items were aggregated into a total sum score, similar to the approach
after pre-piloting for low literacy levels. Participants chose their lan­ used in previous psychological literature. Household economic welfare
guage (isiXhosa or English) and privacy was ensured. There was a de­ used a four-item scale measuring monthly consistent access to neces­
viation from the protocol due to extended political violence related to sities including food, electricity, communication (airtime), and trans­
the 2016 elections, with riots, road blockages and petrol-bombing that port (Morduch, 1995; Townsend, 1995), which was aggregated into a
substantially affected delivery and data collection. This resulted in a principal-component weighted scale centered around 0, thus following
shift of the planned final data collection stage from 19 months post- conventions in the poverty measurement literature (Filmer and
program to 9–13 months post randomisation. Pritchett, 2001; Sahn and Stifel, 2003).
Intervention clusters received the Parenting for Lifelong Health for
Teens program comprising 14 weekly sessions, attended by adolescents 2.6. Covariates
and their primary caregivers. The program was designed for low-re­
source settings with no technology (such as video) or literacy require­ All analyses controlled for baseline values of violence and of all
ments, and used non-didactic, collaborative learning processes in­ hypothesized mediators to account for potential differences between
cluding role-plays, activities, discussions, and songs. Sessions lasted treatment and control group at the study's outset. Additional covariates
1.5–2 h and included praise, managing anger and stress, collaborative included child and caregiver gender, child and caregiver age, rural/
problem-solving, planning together to protect adolescents from com­ urban location, whether the caregiver was a biological parent of the
munity violence, and two sessions focused on family budgeting, saving adolescent, and household living standards using an asset index of the
and financial planning for the future. The program was delivered by eight most highly-voted necessities for households with children,
local community members, who were trained by a local NGO, Clowns identified through the South African National Social Attitudes Survey
Without Borders South Africa, and supported through weekly super­ (Wright, 2008) and scored based on principal-component-weighting.
vision. Sessions were delivered in locally available spaces, such as
community halls. Control clusters received a community level one-day 2.7. Analyses
hygiene promotion workshop (5 h), also delivered by Clowns Without
Borders South Africa, which focused on handwashing skills for children, Analyses used intention-to-treat (ITT) for all clusters and families
delivered through performance and activities. All children received a irrespective of intervention uptake, and included families who were no
soap which – when used – had a small toy inside. The choice of active longer living together at follow-up (n = 53). The full sample was used
placebo was made by the local communities, who were concerned in order to allow testing of intervention impact on hypothesized med­
about child sanitation after a period of drought. iators. Potential mediators were first tested separately following

3
L. Cluver, et al. Social Science & Medicine 262 (2020) 113194

methods to estimate average causal mediation effects (ACME) (Imai Table 2 shows each mediator tested separately. In column (1), we
et al., 2010; Imai et al., 2011). Using nonparametric bootstrapping show parametric estimates of the program's effect on each respective
procedure with 1000 resamples, this analysis splits the average treat­ mediator. In columns (2)–(4), we show nonparametric (based on
ment effect (ATE) of the Sinovuyo Teen programme both into an indirect bootstrapping with 1000 simulations) estimates of the ACME, direct,
effect (the ACME) that runs through an observed intermediate variable and total effect. The ACME is the statistic of interest and indicates a)
and also into a direct effect that runs through unobserved channels. whether a mediation effect exists and b) in which direction it runs.
Mediation analyses controlled for baseline values, and standard errors Three of the four tested pathways were significant: improved parenting
were clustered at the village level, i.e., the unit of randomisation. In a with an average causal mediation effect (ACME) of −2.08 (95% CI
final step, using the full sample, all hypothesized mediating variables [-3.08, −1.29]), improved caregiver mental health ACME -1.10 (95%
(Hayes, 2009) were entered simultaneously into a linear structural CI [ −2.00, −0.47]), improved caregiver alcohol/drug avoidance
equation model (parametric estimation), and analyses controlled for ACME – 0.44 (95% CI [ −0.90, −0.10]), and improved economic
baseline measures of all mediators and outcome. Interactions were not welfare ACME -0.39 (95% CI [ −0.75, −0.14]). Improved adolescent
tested for as we did not have time separation between the mediators behaviour was not a mediator.
and outcome. Goodness of fit for the final model (without clustering) In a subsequent step, all four mediators were entered simulta­
was assessed using the Comparative Fit Index (CFI), the Root Mean neously into a structural equation model. The final structural model
Standard Error of Approximation (RMSEA), and the Standardized Root (see Fig. 1) showed moderate to good fit with CFI = 0.981,
Mean Square Residual (SRMR). We also report χ2 fit statistics but ac­ RMSEA = 0.080, SRMR = 0.011 and χ2 = 20.491***. Effects on
knowledge that the test is inflated by sample size. Following Brown violence reduction ran through four indirect pathways: improvements
(2015), the model was refined for improved goodness of fit by taking in parenting skills, caregiver mental health, alcohol/drug avoidance,
modification indices into account and correlating respective item re­ and family economic welfare. At follow-up, intervention clusters had
siduals. improved parenting ( = 0.33, SE = 0.033, p < 0.001), improved
mental health (less depression) ( = 0.22, SE = 0.042, p < 0.001),
3. Results improved alcohol and drug avoidance ( = 0.14, SE 0.040,
p < 0.001), and increased likelihood of reaching the end of the month
Basic sociodemographic characteristics of the study sample are re­ with enough food, electricity, and transport (β = 0.15, SE = 0.042,
ported in Cluver et al. (2018). Baseline and follow-up values of the p < 0.001). There was no pathway from the intervention to adolescent
outcome and mediating variables are reported in Table 1. Internal behaviours (β = 0.02, SE = 0.039, p = 0.616). All mediating variables
consistency of all aggregated variables was high and ranged between were directly associated with reductions in violence against children:
alpha values of 0.71 and 0.89. Attrition was low, with 11% of the improved parenting (β = −0.14, SE = 0.050, p = 0.005), improved
sample missing items or one of the caregiver/adolescent responses, and mental health (β = −0.14, SE = 0.041, p < 0.001), improved al­
missing status was similar across arms and not associated with the in­ cohol/drug avoidance (β = −0.09, SE = 043, p = 0.029) and eco­
tervention. Given these, complete case analysis was conducted within nomic welfare (β = −0.08, SE = 0.041, p = 0.048). Improved ado­
complete dyads. lescent behaviour (whilst not a significant mediator), was associated
with reduced violence (β = 0.41, SE = 0.046, p < 0.001).
3.1. Mediation
4. Discussion
Primary analyses are reported in Cluver et al., 2018. In summary,
caregivers and adolescents in the intervention group reported reduced This study is, to our knowledge, the first in a low or middle income
physical and emotional violence, improved parental monitoring, and country to quantitatively examine mediators of a parenting program on
involved parenting (all based on combined child- and caregiver-report). reduction of violence against children (WHO, 2016). Results indicate
Caregivers in the intervention group reported improved mental health, four pathways: improved parenting skills, caregiver mental health, al­
alcohol/drug avoidance, and household economic welfare (based on cohol/drug avoidance, and household-level economic welfare. Findings
caregiver report only). Control group adolescents also reported reduced suggest the continuing importance of supporting caregivers with stra­
violence, and there were no differences for child behaviour problems. tegies for protective parenting. They also suggest that improving

Table 1
Sample characteristics.
Baseline Follow-Up

Mean (SD) Treatment Control Treatment Control

Outcome Variable
(1) Emotional and Physical Violence against Children* 15.69 (16.11) 14.14 (14.55) 7.28 (10.85) 8.27 (10.05)
(Cronbach's α = 0.81)
Mediating Variables
(2) Monitored and Involved Parenting* 9.61 (18.59) 11.83 (18.47) 29.00 (18.08) 15.64 (18.99)
(Cronbach's α = 0.79)
(3) Child Problem Behaviour* 33.18 (16.56) 31.98 (15.66) 24.44 (14.44) 24.15 (14.75)
(Cronbach's α = 0.89)
(4) Caregiver Depression 23.13 (11.78) 24.90 (12.09) 11.30 (9.78) 16.82 (11.13)
(Cronbach's α = 0.71)
(5) Caregiver Alcohol/Drug Use 0.25 (0.50) 0.30 (0.54) 0.16 (0.40) 0.33 (0.62)
(6) Household Economic Welfare 0.04 (1.68) −0.04 (1.64) 0.29 (1.60) −0.28 (1.49)
(Cronbach's α = 0.72)
N 270 282 264 278

Notes: * based on combined (summed) caregiver and child reports. Variables (1)–(4) are continuous scale scores, Variable (5) here reported as binary, coded 1 if any
reported drug use or more than three drinks per one day in the past month. Variable (6) is a continuous scale score based on principal component weighting of access
to four necessities (e.g., transport, food, electricity, communication), centered around 0. Cronbach's alpha reported for the baseline scale.

4
L. Cluver, et al. Social Science & Medicine 262 (2020) 113194

Table 2
Mediation Analysis for the Outcome of violence against children (ICAST).
Potential Mediator Effect of Program Participation on Average Causal Mediation Direct Total Sample Size
Mediator (1) Effect (2) Effect (3) Effect (4)
(5)

Improved Monitoring and Involved Parenting 12.87*** [9.87, 15.88] −2.08 [-3.08, −1.29] 1.21 [-0,62, −0.89 [-2.89, 489
3.20] 1.09]
Improved Adolescent Behaviour −0.44 0.16 [-0.75, 1.06] −1.01 [-2.41, −0.86 [-2.66, 489
[-2.94, 2.06] 0.51] 0.97]
Improved Caregiver Mental Health (Reduced 5.16*** [2.96, 7.36] −1.10 [ −2.00, −0.47] 0.27 [-1.39, −0.84 [-2.62, 489
Depression) 2.08] 0.99]
Improved Caregiver Alcohol/Drug Avoidance 0.15*** −0.44[-0.90, −0.10] −0.52 [-2.08, −0.91 [-2.62, 489
[0.06, 0.24] 1.16] −0.90]
Improved Household Economic Welfare 0.49*** [0.21, 0.76] −0.39[ −0.75, −0.14] −0.52 [-2.23, −0.93 [-2.85, 489
1.33] 1.04]

Notes: *p < 0.05, **p < 0.01, ***p < 0.001, 95% CIs in square brackets. Coefficients are unstandardized and standard errors were clustered at the village level.
Column (1) represents the parametric inference of the program's effect on the respective mediator. Estimates shown in columns (2)–(4) are based on a nonparametric
bootstrap procedure with 1000 simulations (consequently significance levels are not provided and CI range is used). Each mediator was tested individually. All
analyses control for caregiver and adolescent age and sex, rural/urban location, biological/non-biological relationship, and household living standards. Sample size
varies for tested mediators as combined adolescent and caregiver scores could only be calculated if both pair members were interviewed at follow-up. As each
mediation model was run separately and the direct and indirect effect varies by mediator, the total effect on violence differs by model – varying from −0.84 to
−0.93.

Fig. 1. Structural Equation Model of Program Mediators, Notes: ***p < 0.001, **p < 0.01, *p < 0.05. All coefficients shown are standardized.

caregiver wellbeing and coping strategies leads to improvements for follow-up (mean 12 months after randomisation). Thus, hypothesized
children. Finally, they suggest that combining parenting and economic mediators and the violence reduction outcome were measured at the
strengthening programs – particularly in poor communities – may boost same time, although both were able to measure change by controlling
effectiveness against violence. These implications are supported by for baseline scores within the model. Although mediation analyses with
qualitative work from other low-resource settings, for example, care­ two time points are common in the literature (Forehand et al., 2014;
givers in Latin America and Africa describing overwhelming parenting Gardner et al., 2010), future studies could valuably use multiple follow-
stress related to poverty (Doubt et al., 2018; Lachman et al., 2016; up time points. Second, there are always limitations of mediation
Parra-Cardona et al., 2009). analyses in establishing causality of discrete intervention components:
This study has a number of limitations. First, our measures were improvements in economic welfare mediated reductions in violence,
limited to baseline, randomisation into trial arms (after baseline) and but these improvements may have been due to other aspects of the

5
L. Cluver, et al. Social Science & Medicine 262 (2020) 113194

intervention in addition to the economic sessions – for example, re­ reflects a broader conceptual shift led by the Sustainable Development
duced caregiver alcohol/drug use may have allowed more money to be Goals, which conceptualise goals such as poverty reduction, mental
allocated to food. Approaches such as factorial experiments could be health and violence prevention as interlinked and inter-dependent.
used in future studies to isolate core components.
Third, the current analysis weighted child and caregiver responses 5. Conclusion
equally, due to limited statistical power and linked to this, a limited
number of degrees of freedom. Future analyses could additionally These findings can inform the scale-up of evidence-based parenting
model variables such as parenting and violence as latent factors. Fourth, programs for low-resource settings (Puffer et al., 2017). The Parenting
due to the real-world sample, the violence outcome was skewed. for Lifelong Health for Teens program is now used by a range of NGOs,
However, these complex mediation models were impossible to compute governments and donors, including USAID-PEPFAR, UNICEF, Catholic
with a non-linear link and transformation does not solve high numbers Relief Services and Pact, and included in the WHO ‘INSPIRE’ policy
of zero scores. Therefore, we used MPlus which has the capacity to guidelines. It is currently implemented in 15 low- and middle-income
conduct SEM on skewed distributions and checked goodness of fit in­ countries, reaching an estimated 600,000 families by 2021.
dicators which were good. Fifth, the trial was conducted by the pro­ Further important implementation science questions remain. These
gram developers, and additional studies should be conducted in­ include how to most cost-effectively combine strategies within pro­
dependently. Sixth, due to sustained election violence, the study was grams, avoiding excessively long or complex delivery processes,
not able to conduct follow-up beyond 9 months. A recent review of training and supporting local staff in sustainable program im­
parenting in conflict zones finds that impacts on children (such as be­ plementation (Ward et al., 2015). Nonetheless, there is growing ad­
haviour changes) may be delayed and therefore longer follow-up times vocacy in both high- and low-income settings for combining parenting
are needed (Murphy et al., 2017). Seventh, the trial was only powered support, mental health and economic strengthening (Richter and
to detect substantially larger than average effects for parenting pro­ Naiker, 2015; WHO, 2020). This study provides empirical support for
grams on violence, thus likely under-estimated program impacts on doing so. It indicates a next step in the field of parenting programs:
both mediators and outcome. Eighth, parenting was measured through capitalizing on the strong evidence-base for social learning methods,
adolescent and caregiver report. Whilst home observations have strong and building on this with approaches from other fields such as devel­
external validity for younger children, it is difficult to get reliable ob­ opment economics. Through such linkages, we can reduce violence
servations of adolescent-caregiver interactions with an interviewer against children in the world's highest-risk, lowest-resource contexts.
present, and in rural areas where large families live in one room
households. Globally, studies find discrepancies between adolescent
Funding
and caregiver report (or perception) of both parenting and adolescent
behaviours, and so we combined and summed both reports to gain a
The intervention and study were supported by the European
family-level average.
Research Council (ERC) under the European Union’s Seventh
Strengths of the study include the pragmatic randomized trial
Framework Program (FP7/2007–2013/ERC Grant Agreement No.
method, which uses real-world recruitment and service delivery plat­
313421), UNICEF Innocenti Office of Research and UNICEF South
forms to provide high external validity. In particular, sampling purpo­
Africa. Continuing support was provided by the ERC under the
sefully reflected real-world service delivery in African contexts, with
European Union’s Horizon 2020 research and innovation program
participants referred by a range of community members, state services,
(Grant Agreement No. 737476). Additional funding is provided by
traditional leaders and self-referrals, and no exclusion criteria. In con­
UKRI GCRF through the Accelerating Achievement for Africa's
texts of very weak social and police services (as across Africa) com­
Adolescents Hub (Grant Ref: ES/S008101/1), Research England, the
munity approaches are the most feasible to reach families experiencing
John Fell Fund, the Leverhulme Trust (PLP-2014-095), the Cambridge
risk factors for violence. However, we note that our study area included
Trust and the University of Oxford’s ESRC Impact Acceleration Account
rural areas and peri-urban townships, but no major inner-city areas –
(1311-KEA-004 and 1602-KEA-189). The South African National
which may be characterised by differing family challenges. It is im­
Department of Social Development provided in-kind support through
portant to recognise the additional needs of families with severe dis­
posting social auxiliary workers to be trained as programme facilitators.
abilities, and the research team are currently working with UNICEF's
Various international agencies and governments are supporting on­
disability team and the Special Olympics to adapt and deliver parenting
going scale-up of the programme. JML was supported by the
support. We also note that the study setting in South Africa has im­
Complexity in Health Improvement Programme of the Medical
plications for the economic aspect of the intervention: most families
Research Council MRC UK (MC_UU_12017/14). FM was supported by
had access to small, regular state poverty alleviation grants of around
the ESRC under a Future Research Leader Award (ES/N017447/1).
$30/month for the household. In contexts where families may have no
Funders of this study had no role in study design, data collection, data
income at all, budgeting and savings plans may require adaptation.
analysis or writing of the report.
The trial shows that violence reduction through simultaneous be­
havioural and economic strengthening pathways is possible during
political unrest and in a very low-resource area. We used intention-to- Declaration of conflicting interests
treat analyses and standardized outcome measures, including actual
violence, rather than commonly-used proxies such as parenting stress or LC, JD, CW, JML and FG are co-developers of the PLH for Young
views about violence, and combined caregiver and child report wher­ Children and PLH for Adolescents programs, which are licensed under a
ever feasible. Creative Commons 4.0 Non-commercial No Derivatives license. CW, LC
These findings may also reflect some of those seen in other fields of and JML are co-founders of the Parenting for Lifelong Health Initiative.
violence prevention. Studies of intimate partner violence in South JML is also the Executive Director of Clowns Without Borders South
Africa, Cote D'Ivoire and Tanzania have shown positive impacts of Africa, a non-profit institution responsible for the dissemination of the
programs combining economic strengthening (such as microfinance or program. JML also receives occasional fees for providing training and
village savings and loans groups) with gender norms training (Gupta supervision to facilitators and coaches. JML, FG, LC and CW have
et al., 2013; Harvey et al., 2018; Pronyk et al., 2006). Recent global participated (and are participating) in a number of research studies
initiatives to prevent violence against children highlight the importance involving the program, as investigators, and the Universities of Oxford,
of both parenting programs and economic strengthening (Know Glasgow and Cape Town receive research funding for these. Conflict is
Violence in Childhood, 2017; World Health Organisation, 2016). This avoided by declaring this potential conflict of interests; and by

6
L. Cluver, et al. Social Science & Medicine 262 (2020) 113194

conducting and disseminating rigorous, transparent and impartial material hardship into models of income associations with parenting and child de­
evaluation research on both this and other similar parenting programs. velopment. Child Dev. 78 (1), 70–95.
Giusto, A., Friis, E., Sim, A., Chase, R., Zayzay, J., Green, E., 2017. A qualitative study of
mechanisms underlying effects of a parenting intervention in rural Liberia. Eur. J.
Acknowledgments Dev. Res. 29 (5), 964–982.
Gupta, J., Falb, K.L., Lehmann, H., Kpebo, D., Xuan, Z., Hossain, M., ... Annan, J., 2013.
Gender norms and economic empowerment intervention to reduce intimate partner
We would like to thank the families who participated, our adoles­ violence against women in rural Cote d'Ivoire: a randomized controlled pilot study.
cent advisory groups in South Africa, the National Department of Social BMC Int. Health Hum. Right 13, 46. https://doi.org/10.1186/1472-698X-13-46.
Development, especially Deputy Director-General Conny Nxumalo and Harvey, S., Lees, S., Mshana, G., Pilger, D., Hansen, C., Kapiga, S., Watts, C., 2018. A
cluster randomized controlled trial to assess the impact on intimate partner violence
Thabani Buthelezi, Lulu Ncgobo, Sbongile Tsoanyane, Nompumelelo of a 10-session participatory gender training curriculum delivered to women taking
Danisa, Sinah Moruane, Heidi Loening, the field team and volunteers part in a group-based microfinance loan scheme in Tanzania (MAISHA CRT01): study
for this study. protocol. BMC Wom. Health 18 (1), 55. https://doi.org/10.1186/s12905-018-
0546-8.
Hayes, A., 2009. Beyond Baron and Kenny: statistical mediation analysis in the new
Appendix A. Supplementary data millennium. Commun. Monogr. 76 (4), 408–420.
Hillis, S., Mercy, J., Amobi, A., Kress, H., 2016. Global prevalence of past-year violence
Supplementary data to this article can be found online at https:// against children: a systematic review and minimum estimates. Pediatrics 137 (3),
e20154079. https://doi.org/10.1542/peds.2015-4079.
doi.org/10.1016/j.socscimed.2020.113194. Imai, K., Keele, L., Tingley, D., 2010. A general approach to causal mediation analysis.
Psychol. Methods 15 (4), 309–334.
Author statement Imai, K., Keele, L., Tingley, D., Yamamoto, T., 2011. Unpacking the black box of causality:
learning about causal mechanisms from Experimental and Observational Studies. Am.
Polit. Sci. Rev. 105 (4), 765–789.
This work reflects analysis of a large randomised controlled trial Ismayilova, L., Karimli, L., 2018. Harsh parenting and violence against children: a trial
with a large group of key contributors to the trial and mediation ana­ with ultrapoor families in francophone west Africa. J. Clin. Child Adolesc. Psychol.
1–18. https://doi.org/10.1080/15374416.2018.1485103.
lysis. The conceptualisation of how economic and social learning theory Knerr, W, Gardner, F, Cluver, L, 2013. Improving positive parenting skills and reducing
pathways may work together required a broader interdisciplinary ap­ harsh and abusive parenting in low- and middle-income countries: a systematic re­
proach. view. Prevent. Sci. 14 (4), 352–363. https://doi.org/10.1007/s11121-012-0314-1.
Know Violence in Childhood, 2017. Ending Violence in Childhood. Global Report 2017.
Retrieved from New Delhi, India.
References Lachman, J M, Sherr, L, Cluver, L, Ward, C L, Hutchings, J, Gardner, F, 2016. Integrating
evidence with local contextual factors to develop a parenting program in low-income
communities in South Africa. J. Child Family Stud. 25 (7), 2337–2352.
Achenbach, T., 2000. Child behaviour checklists (CBCL/2-3 and CBCL/4-18), teacher
MacKinnon, D., Fairchild, A., Fritz, M., 2007. Mediation analysis. Annu. Rev. Psychol. 58,
report form (TRF) and youth self-report (YSR). In: The American Psychiatric
593–614.
Association Taskforce for the Handbook of Psychiatric Measures. The Handbook of
Marcus, R., Page, E., 2014. Economic Strengthening Activities in Child Protection
Psychiatric Measures: The American Psychiatric Association.
Interventions: an Adapted Systematic Review. Retrieved from, Geneva.
Annan, J., Bunderviet, T., Seban, J., Costigan, J., 2013. Urwaruka Rushasha: A
Meinck, F, Boyes, M E, Cluver, L, Ward, C L, Schmidt, P, DeStone, S, Dunne, M P, 2018.
Randomised Impact Evaluation of Village Savings and Loans Associations and
Adaptation and psychometric properties of the ISPCAN Child Abuse Screening Tool
Family-Based Interventions in Burundi. Retrieved from Washington, DC. https://
for use in trials (ICAST-Trial) among South African adolescents and their primary
resourcecentre.savethechildren.net/sites/default/files/documents/new_generation_
caregivers. Child Abuse Neglect 82, 45–58. https://doi.org/10.1016/j.chiabu.2018.
final_report_05312013.pdf.
05.022.
Barlow, J., Johnston, I., Kendrick, D., Polnay, L., Stewart-Brown, S., 2006. Individual and
Meinck, F, Cluver, L, Boyes, M E, Mhlongo, E L, 2014. Risk and protective factors for
group-based parenting programmes for the treatment of physical child abuse and
physical and sexual abuse of children and adolescents in Africa: A review and im­
neglect. Cochrane Database Syst. Rev. 3, 1–20 (Art. No: CD005463).
plications for practice. Trauma Violence Abuse. https://doi.org/10.1177/
Brown, T., 2015. Confirmatory Factor Analysis for Applied Research. Guilford
1524838014523336.
Publications.
Mejia, A., Ulph, F., Calam, R., 2016. Exploration of mechanisms behind changes after
Chen, M., Chan, K.L., 2016. Effects of parenting programs on child maltreatment pre­
participation in a parenting intervention: a qualitative study in a low-resource set­
vention: a meta-analysis. Trauma Violence Abuse 17 (1), 88–104. https://doi.org/10.
ting. Am. J. Community Psychol. 57 (1–2), 181–189.
1177/1524838014566718.
Moffitt, T.E., Klaus-Grawe Think Tank, 2013. Childhood exposure to violence and lifelong
Cluver, L, Lachman, JM, Gardner, F, Hutchings, J, Tsoanyane, S, et al., 2016.
health: clinical intervention science and stress-biology research join forces. Dev.
Development of a parenting support programme to prevent abuse of adolescents in
Psychopathol. 25 (4 Pt 2), 1619–1634. https://doi.org/10.1017/
South Africa: findings from a pilot pre-post study. Res. Soc. Work Practice 1-9.
S0954579413000801.
Cluver, L, Meinck, F, Shenderovich, Y, Ward, C L, Romero, R H, Redfern, A, Lechowics, M,
Morduch, J., 1995. Income smoothing and consumption smoothing. J. Econ. Perspect. 9
et al., 2016. A parenting programme to prevent abuse of adolescents in South Africa:
(Summer), 103–114.
study protocol for a randomised controlled trial. Trials 17 (1), 328. https://doi.org/
Murphy, K., Rodrigues, K., Costigan, J., Annan, J., 2017. Raising children in conflict: an
10.1186/s13063-016-1452-8.
integrative model of parenting in war. J. Peace Psychol. 23 (1), 46–47. Retrieved
Cluver, L, Meinck, F, Yakubovich, A, Doubt, J, Redfern, A, Ward, C L, Gardner, F, et al.,
from. http://psycnet.apa.org/?&fa=main.doiLanding&doi=10.1037/pac0000195.
2016. Reducing child abuse amongst adolescents in low- and middle- income coun­
Parra-Cardona, R, Holtrop, K, Cordova, D, Escobar-Chew, A, Horsford, S, Tams, L,
tries: A pre-post trial in South Africa. BMJ Public Health 16 (567).
Fitzgerald, H, et al., 2009. Latino Immigrants’ call to integrate cultural adaptation
Cluver, L, Steinert, J, Shenderovich, Y, Doubt, J, Romero, R, et al., 2018. Parenting for
with best practice knowledge in a parenting intervention. Family Process 48 (2),
Lifelong Health: A pragmatic cluster randomised controlled trial of a non-commer­
211–231.
cialised parenting programme for adolescents and their families in South Afric. BMJ
Parra-Cardona, R, Leijten, P, Lachman, J M, Meija, A, Baumann, A A, Amador Buenabad,
Global Health 3, e000539.
N G, Domenech Rodriguez, M M, et al., 2018. Latino Immigrants’ call to integrate
Doubt, J, Bray, R, Loening-Voysey, H, Cluver, L, Byrne, J, Nzima, D, Medley, S, et al.,
cultural adaptation iith best practice knowledge in a parenting Intervention. Prev.
2018. ‘It has changed’ Understanding change in a parenting program in South Africa.
Sci. https://doi.org/10.1007/s11121-018-0935-0.
Ann. Glob. Health 83 (6), 767–776.
Parra-Cardona, R, López-Zerón, G, Gisela Leija, S, Maas, M, Villa, M, Zamudio, E,
Dziak, J., Dierker, L., Abar, B., 2018. The interpretation of statistical power after the data
Domenech Rodriguez, M, 2018. A culturally adapted intervention for Mexican-origin
have been gathered. Curr. Psychol. 1–9.
parents of adolescents: the need to overtly address culture and discrimination in
Filmer, D, Pritchett, L H, 2001. Estimating wealth effects without expenditure sata - or
evidence-based practice. Family Process.
tears: an application to educational enrollments in States of India. Demography 38
Pedersen, G., Smallegange, E., Coetzee, A., Hartog, K., Turner, J., Brown, F., Jordans, M.,
(1), 115–132.
2019. A systematic review of the evidence for family and parenting interventions in
Forehand, R., Lafko, N., Parent, J., Burt, K.B., 2014. Is parenting the mediator of change
low-and middle-income countries: child and youth mental health outcomes. J. Child
in behavioral parent training for externalizing problems of youth? Clin. Psychol. Rev.
Fam. Stud. 1–20.
34 (8), 608–619. https://doi.org/10.1016/j.cpr.2014.10.001.
Pronyk, P.M., Hargreaves, J.R., Kim, J.C., Morison, L.A., Phetla, G., Watts, C., Porter, J.D.,
Frick, P., 1991. The Alabama Parenting Questionnaire. University of Alabama.
2006. Effect of a structural intervention for the prevention of intimate-partner vio­
Gardner, F, Hutchings, J, Bywater, T, Whitaker, C, 2010. Who benefits and how does it
lence and HIV in rural South Africa: a cluster randomised trial. Lancet 368 (9551),
work? Moderators and mediators of outcome in an effectiveness trial of a parenting
1973–1983. https://doi.org/10.1016/S0140-6736(06)69744-4.
intervention. J. Clin. Child Adoles. Psychol. 39 (4), 568–580. https://doi.org/10.
Puffer, E.S., Annan, J., Sim, A.L., Salhi, C., Betancourt, T.S., 2017. The impact of a family
1080/15374416.2010.486315.
skills training intervention among Burmese migrant families in Thailand: a rando­
Gelman, A., 2019. Don't calculate post-hoc power using observed estimate of effect size.
mized controlled trial. PloS One 12 (3), e0172611. https://doi.org/10.1371/journal.
Ann. Surg. 269 (1), e9–e10. https://doi.org/10.1097/SLA.0000000000002908.
pone.0172611.
Gershoff, E., Aber, J., Raver, C., Lennon, M., 2007. Income is not enough: incorporating
Radloff, L., 1977. The CES-D scale: a self-report depression scale for research in the

7
L. Cluver, et al. Social Science & Medicine 262 (2020) 113194

general population. Appl. Psychol. Meas. 1, 385–401. 14 (1).

Richter, L., Naiker, S., 2015. A Review of Published Literature on Supporting and Teicher, M.H., Samson, J.A., Anderson, C.M., Ohashi, K., 2016. The effects of childhood
Strengthening Child- Caregiver Relationships (Parenting). Retrieved from Arlington, maltreatment on brain structure, function and connectivity. Nat. Rev. Neurosci. 17
vol. A. (10), 652–666. https://doi.org/10.1038/nrn.2016.111.
Sahn, D E, Stifel, D, 2003. Exploring alternative measures of welfare in the absence of Townsend, R., 1995. Consumption insurance: an evaluation of risk-bearing systems in
expenditure data. Rev. Income Wealth 49 (4), 463–489. low-income economies. J. Econ. Perspect. 9 (Summer), 83–102.
Saunders, J.B., Aasland, O.G., Babor, T.F., de la Fuente, J.R., Grant, M., 1993. Vlahovicova, K, Melendez-Torres, G J, Lejten, P, Knerr, W, Gardner, F, 2017. Parenting
Development of the alcohol use Disorders identification test (AUDIT): WHO colla­ programs for the prevention of child physical abuse recurrence: A systematic review
borative project on early detection of persons with harmful alcohol consumption–II. and meta-analysis. Clin. Child Fam. Psychol. Rev. 20 (3), 351–365. https://doi.org/
Addiction 88 (6), 791–804. Retrieved from. https://www.ncbi.nlm.nih.gov/ 10.1007/s10567-017-0232-7.
pubmed/8329970. Ward, C, Mikton, C, Cluver, L, Cooper, C, Gardner, F, Hutchings, J, Wessels, I, et al., 2015.
Shelleby, E C, Shaw, D S, Dishion, T J, Wilson, M N, Gardner, F, 2018. Effects of the family Parenting for lifelong health: From South Africa to other low-and middle-income
check-up on reducing growth in conduct problems from toddlerhood through school countries. Early Childhood Matters 122, 49–54.
age: An analysis of moderated mediation. J. Consult. Clin. Psychol. 86 (10), 856–867. WHO, 2016. INSPIRE: seven strategies for ending violence against children. WHO,
https://doi.org/10.1037/ccp0000337. Luxembourg.
Statistics, S.A., 2016. Men, Women and Children: Findings of the Living Conditions WHO, 2020. Global Status Report on Preventing Violence against Children. WHO,
Survey 2014/15. Retrieved from Gauteng, South Africa. Geneva.
Steinberg, L.D., Catalano, R., Dooley, D., 1981. Economic antecedents of child abuse and Wright, G, 2008. Findings from the indicators of poverty and social exclusion project: A
neglect. Child Dev. 52 (3), 975–985. Retrieved from. https://www.ncbi.nlm.nih. profile of poverty using the socially perceived necessities approach. Key Report 7.
gov/pubmed/7285664. Yang, M.Y., 2015. The effect of material hardship on child protective service involvement.
Stith, S., Liu, T., Davies, C., Boykin, E., Alder, M., Harris, J., Dees, J., 2009. Risk factors in Child Abuse Negl. 41, 113–125. https://doi.org/10.1016/j.chiabu.2014.05.009.
child maltreatment: a meta-analytic review of the literature. Aggress. Violent Behav.