Menu
Menu
Menu
Menu
Recent evidence has shown support for the United Nations Development Programme (UNDP) accelerator concept, which highlights the need to identify interventions or programmatic areas that can affect multiple sustainable development goals (SDGs) at once to boost their achievement. These data have also clearly shown enhanced effects when interventions are used in combination, above and beyond the effect of single interventions. However, detailed knowledge is now required on optimum combinations and relative gain in order to derive policy guidance. Which accelerators work for which outcomes, what combinations are optimum, and how many combinations are needed to maximise effect? The current study utilised pooled data from the Young Carers (n = 1402) and Child Community Care (n = 446) studies. Data were collected at baseline (n = 1848) and at a 1 to 1.5- year followup (n = 1740) from children and young adolescents aged 9–13 years, living in South Africa. Measures in common between the two databases were used to generate five accelerators (caregiver praise, caregiver monitoring, food security, living in a safe community, and access to community-based organizations) and to investigate their additive effects on 14 SDG-related outcomes. Predicted probabilities and predicted probability differences were calculated for each SDG outcome under the presence of none to five accelerators to determine optimal combinations. Results show that various accelerator combinations are effective, though different combinations are needed for different outcomes. Some accelerators ramified across multiple outcomes. Overall, the presence of up to three accelerators was associated with marked improvements over multiple outcomes. The benefit of targeting access to additional accelerators, with additional costs, needs to be weighed against the relative gains to be achieved with high quality but focused interventions. In conclusion, the current data show the detailed impact of various protective factors and provides implementation guidance for policy makers in targeting and distributing interventions to maximise effect and expenditure. Future work should investigate multiplicative effects and synergistic interactions between accelerators.
The current study pooled data from two longitudinal studies, which were designed in close collaboration and utilized the same or similar measures of relevant constructs, to generate a sample of 1848 young people (n = 1740 follow-up– 94.2% rate). Of these, 446 were drawn from the Child Community Care (CCC) study, and 1,402 from the Young Carers (YC) study. The CCC study investigated effects of attendance of community-based organisations (CBOs) in three countries (South Africa, Malawi, Zambia–only South African data is used in this paper) on child and adolescent outcomes, with high study enrolment (99.0%) at baseline and retention at a 12–18 month follow-up (86.0%). The YC study focused on the well-being of adolescents from disadvantaged backgrounds living in South Africa. Participants were drawn randomly from four census enumeration areas, with one child chosen at random from all visited households. Enrolment was 97.5% at baseline, and 96.8% were retained at the 1-year follow up. Adolescents in this study received no CBO support at either baseline or follow up. Detailed methods of recruitment for both studies are described elsewhere [23,24]. To pool both databases into one overall sample, only young people living in South Africa aged 9–13 years (overlapping age range between both studies) were selected. All data were obtained by trained data collectors in participant’s language of choice and all participants and their caretakers provided written consent. Ethical approval for the YC study was obtained from the Universities of Oxford (SSD/CUREC2/11-40) and Cape Town (Ref: CSSR 389/2009), and the respective provincial Health and Education Departments. For the CCC study, ethical approval was obtained from University College London (1478/002) and Stellenbosch University (N10/04/112), as well as the funding agencies supporting the participating CBOs. The current analyses build on a previous manuscript [13], in which path analyses were used to investigate the effects of seven hypothesized protective factors on 14 SDG-related outcomes. All protective factors were hypothesized to simultaneously affect various SDG outcomes and thus to act as accelerators. Each factor had to be present at both baseline and follow-up to count as a potential accelerator, given evidence of the importance of sustained provision during child developmental progression [8]. Measures used and coding decisions for all protective factors and SDG outcomes can be found in the original paper [13]. The hypothesized protective factors comprised: 1) food security, coded as present if the child had not gone to bed hungry recently, 2) receipt of at least one of five government-provided cash grants over the past year in the household the child lived in (measured at follow up to cover the preceding year), 3) living in a safe community, indicated by children not witnessing or directly being exposed to community violence, 4) consistent access to healthcare when needed, 5) regular caregiver praise, 6) caregiver monitoring of child activities, and 7) access to CBOs, with the YC sub-sample specifically chosen to not have access to CBOs at any time-point, thus posing a comparison group. Fourteen outcomes that aligned with five SDGs were retrospectively identified (for coding details: see [13]: S1 Table). This includes no symptoms major depression (MDD) and post-traumatic stress disorder (PTSD), no suicidality, as well as overall good mental health (combined score of all three previous measures), no peer problems, high prosocial behaviour (all SDG 3.4); no substance abuse (SDG 3.5); school enrolment, school attendance, being in the right grade for age, being able to concentrate at school (all SDGs 4.1/ 4.4); no sexual debut (given that the target population was relatively young) (SDG 5.6); no delinquent behaviours (due to their common link with aggression/violence grouped under SDG 16.1, “reduce violence everywhere”); and no exposure to physical and emotional abuse by the caregiver (SDG 16.2). Measures at follow-up were used as the main outcomes. We controlled for baseline score where possible to account for potential pre-existing differences between the YC and CCC samples. Exceptions were the peer problems and prosocial behaviour subscales and the sexual debut variable, for which measures were only available at follow-up. We included seven covariates in our analyses that were measured at baseline: child age, child sex, whether the family lived in formal versus informal housing, maternal/paternal death, caregiver HIV-status and child caretaking responsibilities for other children or adults. Descriptive analyses and analyses of those retained versus lost to follow-up were conducted using χ2 tests and two-tailed t-tests as appropriate. As described above, the current analyses build on previous work using path analyses to investigate the effect of seven hypothesized protective factors on 14 SDG-related indicators (standardized approach developed by Rudgard and colleagues 2020; code: https://osf.io/n6jy7/?view_only=17f148085fde4b3fb645106c6c6e418b). After the absence of multi-collinearity between accelerators was established, separate multivariable logistic regressions were conducted, with each outcome being simultaneously regressed on all protective factors and covariates. The aim was to identify “accelerators”, defined as protective factors that were related to three or more SDG outcomes after correcting for multiple testing via Benjamini Hochberg corrections (false discovery rate: 0.1) [25]. Accelerators affecting three or more outcomes identified in these analyses were: caregiver praise, caregiver monitoring, living in a safe community, CBO access and food security. The current paper presents a more in-depth follow-up analysis, focusing on what combinations of these five accelerators affected specific outcomes, and on identifying the quantum of benefit. For this purpose, adjusted probabilities and adjusted probability differences were calculated using the “margins” command in Stata, based on the original model (including covariates). We compared adjusted probabilities for the hypothesized presence of no accelerator, single accelerators and all possible combinations of accelerators to determine optimal combinations. We also calculated probability differences and associated confidence intervals to establish whether there were significant increases in predicted probabilities through the addition of additional accelerators. For this, we focused on the most effective accelerator combinations (i.e., the single most effective accelerator, combinations of the two to four most effective accelerators, all five accelerators). All analyses were conducted in Stata SE v.16.
