Leveraging Early Adolescence for Development: Longitudinal and Experimental Evidence from Ghana (LEAD)

As part of this study, we developed and pilot-tested the PFP, a family-based program aiming at supporting parents and young adolescents. The program consisted of three in-person meetings with parents and adolescents, and biweekly SMS-messages to the parents delivered over 9 months. Booster calls and SMS following the main intervention for 9 months aim to foster longer-term behavioural change.

As part of this study, we leverage randomised variation from a previous RCT, the “Quality Preschool for Ghana” study (Wolf et al., 2019). This research initially assessed the effectiveness of a scalable model for kindergarten teacher training and parental awareness programs, implemented by district education officials through a school-randomized trial across six districts in the Greater Accra Region in the 2015-16 academic year. Schools were randomized into three groups: teacher training (TT), teacher training plus parental awareness (TTPA), and a control group following business-as-usual practices. Children were aged 4-5 years in 2015. This sample was followed for over eight survey rounds before LEAD began.

2025-03-14

2025-12-31

We pre-register outcomes at the first follow-up (2025). Later modifications of this pre-analysis plan will register outcomes at the second follow-up (2026). Primary outcomes for the 2025 follow-up include caregiver and adolescent reported: (i) Stress; (ii) Self-regulation; and (iii) parenting practices.

Following our theory of change, primary outcomes include skills and outcomes that PFP directly targeted as part of the in-person sessions for parents and adolescents, and in the SMS to the parents, while secondary outcomes are more distal that we expect to change as a result of improvements in the primary outcomes.
Primary outcomes include parent and child self-reported: stress, self-regulation, and parenting practices. These are measured through, respectively: the Perceived Stress Scale (Cohen et al., 1983); Emotion Regulation Questionnaire (Gross, 2003); and a subset of scales from the Alabama Parenting Questionnaire on discipline and positive parenting (Essau et al., 2006), as reported by both the child and parent. For a target sub-sample of around 575 parent-child dyads measured in 2025, we will directly assess stress through hair cortisol data collection to get a biological marker of long-term stress in addition to the self-report perceived stress measure. We focus on adolescent stress and self-regulation because this developmental period is marked by significant brain and physiological changes that can affect one's ability to respond to their environment. For parents, we focus on the same outcomes, as we are interested in learning how parenting programs can decrease parenting stress in a low-resource environment, and provide parents with skills and knowledge to cope with stress respond to their emotions (e.g. taking deep breaths before punishing the child, use assertive communication tools, etc).

Generally, we aimed to measure both primary and secondary outcomes through scales that have been previously validated in Ghana with similarly-aged participants, or that worked well from a measurement perspective in other projects in Ghana by the study PIs. As this was not always possible, we opted for scales that either: have been validated across other low-resource contexts; demonstrated face validity with local experts if other validated scales have not worked well with samples from Ghana; or have been previously used in adolescent parenting intervention evaluation studies. New measurement tools for this sample were adapted to the local setting through an in-depth local language translation and field-testing process.
As part of our investigation, once data are collected, we will assess the reliability and validity of study scales, including assessing if items missing responses, sufficient item-level variability, concurrent and discriminant validity (e.g. Ausburg et al., 2022; Macours et al., 2023). If these checks highlight that the measures employed are not valid, we will consider dropping some items or the measure entirely if its use would bring into question the validity of the interpretation of the scale as a representation of the underlying construct it aimed at measuring. During this phase of measure assessment, we will also evaluate whether data should be reported as raw scores or factor scores, similar to Augsburg et al. (2022). Outcomes will be standardised based on control means and SD in each round.

Adolescent outcomes: (i) Positive well-being and reduced vulnerability to mental health problems (ii) Risky health behaviors; (iii) School enrollment and attainment.
Caregiver outcomes: (i) Self-efficacy; (ii) Mental health and well-being.
Family outcomes: Strengthened family relationships and well-being.

Following our theory of change, we measure our secondary outcomes at the parent-, adolescent-, and family-level. We consider these more ‘distal’ outcomes from the intervention but connected to the primary outcomes “downstream”. In other words, we expect that any change that might occur in the secondary outcomes stems from changes in the primary outcomes. Secondary outcomes will also be measured using established scales or adaptations of items from other large-scale studies following recent guidelines (Macours et al., 2023).

For adolescents, we will measure: (i) externalising and internalising symptoms with the Youth Externalizing Problems Screener (Renshaw & Cook, 2018) and Kessler-10 Psychological Distress (Kessler et al., 2003), respectively, and well-being with Mental Health Continuum-Short Form (Lamers et al., 2011); (ii) risky behaviour (Young Lives; Seager & De Wet, 2003); (iii) school enrolment and attainment.
For parents, we will measure: (i) Brief Parental Self-efficacy Scale (Woolgar et al., 2023), and (ii) Kessler-10 Psychological Distress (Kessler et al., 2003).
For the family-level, we will assess effects on multiple dimensions of family functioning, as reported by both parents and children through the McMaster Family Assessment Device (Epstein et al., 1983).
Finally, we also explore potential mechanisms underlying program effectiveness: (i) parental and child knowledge of stress management techniques and disciplinary techniques (measured among parents only); (ii) implementation data (for treatment households, both self-reported and based on intervention attendance, SMS delivery rates, monitoring calls); (iii) spousal agreement; (iv) beliefs, preference and costs for child outcomes and time spent with child (as measured by a series of pre-registered lab-in-the-field experiments, see AEARCTR-0015304); time use; and risk/time preferences.
If the adolescent has a young sibling, we will also explore spillover parenting effects on the younger sibling via parent-reported items from the Alabama Parenting Questionnaire (Essau et al., 2006).

The study population for LEAD consists of ~2,400 children and their parents that were recruited in 2023-24 (LEAD baseline, prior to the start of the intervention), who were randomised individually to: (a) treatment (PFP) or (b) control, with equal probability.

This is a subset and a re-randomisation of participants from a previous impact evaluation study, the “Quality Preschool for Ghana (QP4G)” project, where children were recruited and randomized in the fall of 2015 from 240 preschools across six districts of the Greater Accra region. Preschools were randomised into: (a) teacher training (TT); (b) teacher training plus parental-awareness meetings (TTPA); or (c) controls. This sample was followed for over eight survey rounds before the start of the current project.

This crossover design will allow us to investigate four main research questions:
1. What are the main effects of PFP on young adolescents, their caregivers, and their families? This will be analysed by comparing the effects of the program among children, parents, and families that have been randomly offered the PFP in a standard ITT framework.
2. Are there interactive effects between early-life randomised exposure to QP4G and the PFP in adolescence? We will estimate the effects of being randomly assigned to the either PFP, the QP4G, or both programs at different stages of childhood.
3. Is there heterogeneity in the effects of PFP by child and parent characteristics? For question #1, we assess heterogeneity by child and caregiver gender and parental SES. We will explore additional axes of heterogeneity using econometric techniques.
4. What are the mechanisms underlying PFP effectiveness?

Not available

The randomization method was implemented using STATA with a do-file to ensure reproducibility.

Caregiver-adolescent dyad (~2,400 clusters)

No

We did not cluster the treatment for LEAD. However, for QP4G, children were clustered across 240 schools in 2015. We will explore clustering the SEs at QP4G schools in robustness checks.

Approximately 2,400 caregiver-child dyads

1,200 caregiver-child dyads

The key assumptions for our power calculations are: - ICC: 0.2 - Standardised outcomes - Power: 0.8, alpha: 0.05, two-tailed test While randomisation in LEAD was not clustered, QP4G was a clustered-RCT. Thus, we also have run our power calculations with or without considering the initial clustering of the QP4G sample, with 240 clusters: 240 (10 units per cluster). For research question #1 (N=2,400, 1,200 individuals per arm): MDES: 0.11 without clustering MDES: 0.14 with clustering For research question #2 (N=2,400, 400 individuals per arm) MDES: 0.20 without clustering MDES: 0.23 with clustering We note these power calculations are conservative, as they do not take into account the rich panel data that we have, which can explain part of the variance in our outcome variables.