If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Early marriage has multiple drivers including cultural and social norms alongside lack of educational and economic opportunities. This complexity may explain why few programs have demonstrated marriage delays and suggests multisectoral interventions are necessary. This study examined a 2-year multisectoral program designed to delay marriage in a marginalized setting.
Methods
The study used a prospective 80-cluster randomized trial following up 2,147 girls aged 11–14 years from 2015 to 2019. Interventions included community dialogs about inequitable gender norms (violence prevention), a conditional cash transfer (education), weekly group meetings with health and life skills training (health), and financial literacy training (wealth creation). Villages were randomized to one of four study arms: (1) violence prevention only (V-only); (2) violence prevention and education (VE); (3) VE and health (VEH); or (4) all four interventions (VEHW). We used analysis of covariance to estimate intent-to-treat impacts of each study arm with an education component, as well as a pooled study arm combining the VE, VEH, and VEHW arms, in reference to V-only, 2 years after the intervention ended, when girls were 15–18 years old.
Results
There were small but insignificant reductions on primary outcomes in unadjusted analyses that were larger and significant in adjusted analyses. Effects were particularly large for girls not in school at baseline—the pooled study arm reduced marriage by 18.0 and pregnancy by 15.6 percentage points, a relative reduction of 34% and 43%, respectively.
Discussion
The article demonstrates the potential for multisectoral interventions with education components to delay early marriage in an impoverished, socially conservative, pastoral setting.
This article demonstrates the potential of multisectoral interventions including education components to delay early marriage in marginalized, socially conservative settings. The largest effects were for the most at-risk girls, indicating it was possible to shift girls off a path to early marriage and onto one of continued education.
Although early marriage rates are declining worldwide, annually nearly 12 million girls are married before the age of 18 years, including 35% of girls in Sub-Saharan Africa [
Early marriage has multiple drivers including cultural and social norms underlying inequitable gender norms and a lack of educational and economic opportunities, which are exacerbated by poverty, economic instability, and humanitarian crises [
]. In pastoralist communities, social norms promoting early marriage that limit female involvement in the marriage decision and economic incentives associated with marriage—including bride price—are particularly important [
]. The value of livestock given to the bride's family can be substantial and helps ensure that she will remain in the marriage by creating a social link with her economic inheritance [
There is a growing literature examining interventions to delay early marriage in low- and middle-income countries. A 2012 systematic review of such programs concluded that providing economic incentives or encouraging girls empowerment are promising approaches [
]. A 2016 review identified a range of mechanisms, including education, economic support, life skills training for girls, and community engagement, that could successfully delay marriage [
Interventions to prevent child marriage among young people in low-and middle-income countries: A systematic review of the published and gray literature.
The Samata intervention to increase secondary school completion and reduce child marriage among adolescent girls: Results from a cluster-randomised control trial in India.
Programs addressing the economic drivers of early marriage have also had mixed results. In Malawi, short-term conditional cash transfers (CCTs) delayed marriage and fertility in both the short-term and 2 years after the program ended, but only for girls not initially in school. Unconditional transfers for girls in school at baseline also had short-term effects on these outcomes, but the effects did not persist after the transfers ended [
]. In qualitative work in Zambia, participants reported that early marriage declined because cash transfers increased the time girls could focus on schooling and allowed their parents to rely less on bride price [
]. A recent review concluded that interventions providing cash or in-kind support for girls' education had the highest likelihood of delaying marriage [
Given the interconnected drivers of early marriage and the limitations of single-sector interventions, multisectoral approaches that cover multiple levels in an adolescent's ecosystem (i.e., individual, household, community) should be explored [
The case for a multi-sectoral approach to preventing child marriage and early childbearing in Sub-Saharan Africa. Global perspectives on women’s sexual and reproductive health across the lifecourse.
]. Using a quasi-experimental research design, a program in Ethiopia combining mentorship, economic incentives, and informal education was shown to delay marriage [
]. In Bangladesh, a randomized controlled trial tested several multiple component programs with training via community-based girls' groups in all packages, each having one of the three add-on components: tutoring, gender rights awareness, or livelihoods training. The study found that all three programs delayed marriage [
]. There is little rigorous evidence, however, on programs that delay marriage in pastoralist settings in Sub-Saharan Africa.
In addition to intervention content, timing also may be critical. Adolescence is a distinct phase of development during which significant physical and brain maturation takes place and skills related to social engagement and emotional control develop—changes that shape lifelong behaviors [
]. Early adolescence has been posited as a critical window for intervention because many life events—school leaving, marriage, and childbirth—commonly happen in mid-adolescence. Earlier intervention offers the potential to delay such outcomes [
The Adolescent Girls Initiative-Kenya (AGI-K) was a randomized trial designed to test the short-term (at the end of the program) and longer term (after 4 years) effects of 2-year multisectoral and multilevel, “cash plus” programs for adolescent girls aged 11–14 years at the start of the program. It was implemented in two different marginalized areas of Kenya where girls face many of the aforementioned challenges: (1) Kibera, an urban informal settlement in Nairobi; and (2) Wajir, a pastoralist rural county [
]. This article assesses the postprogram (after 4 years) effects in Wajir on primary outcomes marriage, pregnancy, and birth, as well as on a range of secondary outcomes. Short-term results, and results after 4 years in Kibera, are reported elsewhere [
Impacts of two-year multisectoral cash plus programs on young adolescent girls’ education, health and economic outcomes: Adolescent Girls Initiative-Kenya (AGI-K) randomized trial.
Impacts of multisectoral cash plus programs after four years in an urban informal settlement: Adolescent girls initiative-Kenya (AGI-K) randomized trial.
]. Our study adds to the limited rigorous research available on the effectiveness of such interventions to delay early marriage and fertility in highly impoverished, challenging, and socially conservative settings.
Methods
Intervention context
AGI-K was implemented in rural Wajir, a county located on Kenya's northeastern border with Somalia that is difficult to access and challenging to operate in. Wajir is semi-arid with little arable land and characterized by low population density (<20 people per square kilometer), minimal infrastructure with a sparse paved road network, and limited electricity coverage (only 14% of households). The region suffers from insecurity related to local clan disputes and instability in Somalia. Owing to the vast terrain, villages are isolated, and travel between them is difficult. It is the second poorest county in Kenya and has a predominant religious and ethnic group, Muslim Somalis [
]. More than three-quarters of women aged 15–49 years in Wajir have never attended school, 53% of 20- to 24-year-olds were married by age 18, and the total fertility rate was 7.8 in 2014, the highest in the country [
The case for a multi-sectoral approach to preventing child marriage and early childbearing in Sub-Saharan Africa. Global perspectives on women’s sexual and reproductive health across the lifecourse.
Impacts of two-year multisectoral cash plus programs on young adolescent girls’ education, health and economic outcomes: Adolescent Girls Initiative-Kenya (AGI-K) randomized trial.
]. Interventions in the violence prevention and education sectors targeting norms and economic barriers at the household and community levels were combined with individual-level interventions in education, health, and wealth sectors. The individual-level interventions were designed to empower girls, improving their “ability to formulate strategic choices and to control resources and decisions that affect important life outcomes” [
]. Consistent with low rates of premarital sex, median age at first birth (19.4) was more than 1 year higher than the median age at marriage. In this context, we hypothesized that delayed childbearing would result from delayed marriage [
The case for a multi-sectoral approach to preventing child marriage and early childbearing in Sub-Saharan Africa. Global perspectives on women’s sexual and reproductive health across the lifecourse.
The AGI-K intervention packages included nested combinations of four single-sector interventions implemented for 2 years from August 2015 through July 2017. The study examined the effectiveness of three different multisectoral packages, compared with a base single-sector intervention as follows:
Tabled
1
Intervention package/study arm
Abbreviation
Violence prevention
V-only
Violence prevention + education
VE
Violence prevention + education + health
VEH
Violence prevention + education + health + wealth creation
Villages were randomized to one of the four packages. All four included the community-level violence prevention intervention in which local committees discussed and analyzed inequitable gender norms and violence against girls and then formulated and implemented action plans. Three study arms included the education intervention, providing CCTs to households, and providing school supply kits to girls. The health and wealth interventions comprised mentor-led weekly group meetings with girls covering comprehensive health, life skills and financial education curricula, and savings activities. To ensure quality, the trained mentors met monthly to identify and address challenges and undertake additional training. Extensive supervision and monitoring during program operation ensured high program fidelity, quality, and take-up. A process evaluation conducted after 1 year confirmed both fidelity and quality [
Impacts of two-year multisectoral cash plus programs on young adolescent girls’ education, health and economic outcomes: Adolescent Girls Initiative-Kenya (AGI-K) randomized trial.
We carried out cluster-level randomization of the different intervention packages, with clusters defined as villages with one public primary school. Eighty clusters located throughout the county were identified and stratified by district: Wajir West (20), Wajir East (28), and Wajir South (32). Random assignment was performed in each district at public meetings with local leaders and stakeholders between April and June 2015. Each cluster representative blindly drew a slip of paper indicating a study arm from a set with the number of clusters in the district, equally divided across the four study arms.
By design, all community members living in study villages were exposed to the violence prevention intervention. For the other interventions, the primary beneficiaries were girls aged 11–14 years in 2015. A rapid household listing was conducted in each cluster just before the baseline survey. All girls aged 11–14 years residing in the village were eligible for the program except for ∼3% who were enrolled in boarding school at the time. In clusters with fewer than 40 households with an age-eligible girl, all households and girls aged 11–14 years were selected for the baseline interview. In clusters with 40 or more households with an age-eligible girl, survey team leaders used prespecified cluster-specific random number lists to randomly select 40 households and, within each household, one girl for interview. All eligible girls in the cluster were invited to participate in the program in the same study arm regardless of whether they were included in the survey.
The baseline survey was conducted between March and May 2015, before randomization in each district. Longitudinal surveys were conducted 2 years later at the end of the programs and 4 years later between June and August 2019, or endline. All girls interviewed at baseline were targeted for follow-up.
Outcomes
The overall objective of AGI-K was to improve the well-being of girls after 4 years (when 15–18 years old) enabling safe, healthy, and productive transitions into young adulthood and delaying childbearing (Appendix A). The three primary outcomes examined included binary 0/1 variables measured at endline equal to one if the girl had ever (1) been married; (2) been pregnant; or (3) given birth. We also specified multiple secondary outcomes within each of the four domains reflecting the underlying mechanisms in the theory of change, constructing a summary measure for each: violence prevention, education, health (including sexual and reproductive health [SRH] knowledge), and wealth, oriented in such a way that increases represent improvements. Appendix G provides variable definitions.
Sample size and power analysis calculations
A minimum detectable effect approach was used to conduct power analysis based on the number of potential beneficiaries and clusters that could be covered by the program budget and included in the survey, approximately 3,000 girls across the four study arms. Minimum detectable effects comparing each of the VE, VEH, or VEHW study arms to the V-only study arm were estimated for prevalence of first birth at endline. Setting the power at 80% and the significance level at 5%, the power analysis was conducted for two-sample proportions tests using the Optimal Design Plus Software Version 3.0 (William T. Grant Foundation, New York City, NY) for a cluster-randomized trial with individual-level outcomes [
]. Given the resource constraints, the most efficient design was 20 clusters per arm with 32 girls per cluster (40 girls at baseline, assuming 20% loss to follow-up). Based on the 2014 Kenya Demographic and Health Survey [
We first assessed balance on baseline characteristics across the randomized study arms for the sample reinterviewed at endline to explore potential bias from nonrandom attrition. We also analyzed attrition using ordinary least squares to estimate the probability of reinterview at endline and examined whether the correlates of attrition differed by study arm.
We then estimated the intent-to-treat (ITT) effect of each package of interventions relative to the V-only study arm at endline. Following the theory of change, violence prevention was included in all four study arms, and there was no pure control group. In other words, the study was not designed to estimate the impact of the violence prevention intervention alone. With this research design, the estimated ITT effect for each package of interventions captures both the direct effect of the additional interventions as well as any complementarities between the interventions (including with violence prevention). ITT was defined as a girl living at baseline in a cluster assigned to a specific study arm, irrespective of her actual participation in the program or subsequent residential location. Analysis of covariance models were used.
Because three related primary outcome variables were evaluated, we accounted for multiple hypotheses testing by (1) grouping them into a summary measure and (2) recalculating statistical significance. First, for each variable, we calculated a z-score based on the mean and standard deviation (SD) of the V-only study arm at endline. Using those, we constructed an inverse covariance weighted index, restandardizing to be mean 0 and SD 1 [
Multiple inference and gender differences in the effects of early intervention: A reevaluation of the abecedarian, perry preschool, and early training projects.
]. We then estimated the same ITT model on the summary z-score. Second, we recalculated statistical significance using false discovery rates, reporting the adjusted q-values [
]. The same methodologies were applied to the secondary outcome domains.
All regressions included controls for age, stratification, and when available, the baseline value of the outcome measure. In addition, as outlined in the protocol [
], we report adjusted regressions with extended controls including baseline schooling to improve precision and account for potential initial imbalances. Finally, we constructed a binary 0/1 indicator for a pooled study arm combining all three study arms with the education component (VE, VEH, and VEHW) to estimate an average overall effect relative to V-only.
Because education plays a central role in the theory of change and school enrollment was moderately imbalanced across study arms at baseline, we also carried out a post-hoc subgroup analysis for girls enrolled versus not enrolled in school at baseline, as performed for similar interventions [
]. Finally, we examined sensitivity of the results to attrition using inverse probability weights (IPW) derived from a comprehensive model of attrition to estimate attrition-weighted results and calculated various attrition bounds.
All regressions were estimated with robust standard errors accounting for clustering at the village level. Because there were only 20 clusters per study arm, we assessed sensitivity of statistical significance to randomization-based inference. We set statistical significance at 5%. Analysis was conducted using Stata 15.1 (Stata Corp LLC, College Station, TX).
Ethical approval and funding
The Population Council Institutional Review Board (p661) and the Amref Ethics and Scientific Review Committee (p143-2014) approved the study. Written informed consent was obtained from all girls older than 18 years; parental or guardian consent was obtained for girls younger than 18 years, with those girls providing oral assent.
Results
Appendix C presents the sample flow. The target sample was 2,297 girls (across 80 clusters), of which 2,147 (93.5%) were interviewed at baseline. For security reasons, one cluster was not followed up, and in 2019, 2,039 (95.0%) baseline girls were reinterviewed with rates differing across study arms (92.3%–97.6%).
Table 1 presents the baseline 2015 means for girls reinterviewed at endline. The average age of the girls was 11.9 years, and three-fourths of them lived with their parents, few of whom had any formal education. About three-fourths of girls were currently enrolled in school, with most not currently in school never having been enrolled. Average grade attainment was 2.9 years. SRH knowledge was low, and less than 1% had ever been married, been pregnant, or given birth (Appendix B). The endline sample was balanced across study arms on most baseline characteristics, although somewhat different for preprogram school enrollment, which ranged from 68.8% to 81.6% and was highest in V-only.
Table 1Baseline means for endline analytical sample, by study arm
V-only
VE
VEH
VEHW
p value
Background
Age, mean (SD)
11.9 (1.3)
12.0 (1.3)
11.8 (1.2)∗
11.8 (1.3)
.050
Cognitive score (0–16), mean (SD), n = 2,002
5.3 (2.9)
4.9 (3.0)
4.9 (3.2)
5.2 (3.0)
.595
Lives with both parents, % (n = 2,018)
73.9
73.2
77.2
74.5
.621
Mother ever attended school, % (n = 2,017)
1.3
1.4
0.6
0.4
.236
Father ever attended school, % (n = 2,014)
5.6
2.7∗
3.3
3.2†
.206
Violence prevention
Positive gender attitudes score (0–4), mean (SD), n = 2,014
3.1 (1.0)
2.8 (1.1)†
2.9 (1.1)
2.9 (1.1)
.315
Education
Grade attainment, mean (SD)
3.0 (2.2)
2.7 (2.3)
2.7 (2.2)
2.5 (2.1)†
.403
Enrolled in current school year, %
81.6
68.8∗
76.0
71.8∗
.044
Literate in Swahili and English, % (n = 2,002)
39.7
33.2
38.0
33.5
.436
Health
Knows the most fertile period during menstrual cycle, % (n = 2,014)
1.3
0.6
0.8
0.8
.663
General self-efficacy score (0–6), mean (SD), n = 2,014
2.3 (2.0)
2.3 (1.9)
2.2 (1.9)
1.7 (2.0)∗∗
.010
Wealth creation
Financial literacy score (0–10), mean (SD), n = 2,014
4.4 (2.0)
4.1 (2.1)
4.0 (2.0)
4.0 (2.0)
.519
Saved money in the past 6 months, % (n = 2,014)
0.6
0.4
0.4
0.8
.748
Household level
Household expects girl to complete secondary, % (n = 2,010)
86.6
80.1
81.2
81.8
.293
Household wealth quintile (1–5), mean (SD), n = 2,014
3.0 (1.4)
2.7 (1.5)
2.9 (1.5)
2.9 (1.4)
.758
Sample by arm when n = 2,039
467
516
524
532
2,039
N = 2,039 (unless otherwise indicated) reinterviewed at endline. Asterisks in columns 2–4 indicate statistically different from V-only. p values in column 5 are from an F test for joint differences across study arms. All statistical tests are based on regressions that control for district per the stratified design, with robust standard errors calculated accounting for clustering at the village level. Baseline means for the full baseline sample are shown in Appendix B.
SD = standard deviation; VE = violence prevention and education; VEH = VE and health; VEHW = all four interventions.
The probability of reinterview at endline was higher for the youngest, those in school and with higher completed grades, and those living with both parents (Appendix C). There were few statistically significant differences in the relationships between the covariates and attrition in each study arm with a notable exception—baseline school enrollment. This pattern and moderate imbalance across study arms suggest that although large systematic biases due to attrition in the ITT estimates seem unlikely, there remains potential concern related to baseline schooling. In the following paragraphs, we present results controlling for and splitting the sample by baseline school enrollment.
Table 2 presents the ITT analyses on the primary outcomes, and Figure 3, a summary of the main results. At endline, in the V-only study arm, 14.1% of girls had ever married, and 7.5% had given birth. Although all point estimates in columns 2–5 suggest reductions in the primary outcomes and summary fertility measure, none are significant in the full sample. The estimates with extended controls (including baseline school enrollment), however, are modestly larger (columns 6–9), and although not significantly different from the other study arms, the VE and pooled study arms significantly reduced marriage, whereas the study arms that were more complex to implement did not. For example, the pooled estimate indicates a 4.1 percentage point, or nearly one-third, decrease in marriage compared with the V-only study arm. Although only marginally significant, the probabilities of being pregnant or giving birth in VE and the pooled study arm were also reduced by about one-third. The pooled estimate indicates a reduction of 0.11 SD in the fertility summary, with point estimates for individual arms ranging between −0.08 and −0.16 SD. Accounting for potential attrition bias by reweighting strengthens the findings; IPW results with the extended controls indicate that VE and VEHW study arms significantly reduced all three primary indicators and the fertility summary while VEH marginally significantly reduced all but having ever given birth (Appendix E).
Table 2Estimated intent-to-treat effects on primary outcomes at endline
V-only endline, mean
VE, estimate
VEH, estimate
VEHW, estimate
VE-VEH-VEHW pooled, estimate
VE, estimate: extended controls
VEH, estimate: extended controls
VEHW, estimate: extended controls
Pooled, estimate: extended controls
Full sample
Ever married (=1), n = 2,039
0.141
−0.024
−0.009
−0.004
−0.012
−0.062∗∗
−0.028
−0.034
−0.041∗
95% CI
−0.07, 0.03
−0.08, 0.07
−0.06, 0.05
−0.06, 0.04
−0.11, −0.02
−0.08, 0.02
−0.08, 0.01
−0.08, 0.00
FDR-adjusted q-value
0.400
0.822
0.881
0.617
0.023
0.368
0.153
0.077
Ever pregnant (=1), n = 2,033
0.097
−0.020
−0.021
−0.022
−0.021
−0.041†
−0.031
−0.040†
−0.037†
95% CI
−0.06, 0.02
−0.07, 0.03
−0.06, 0.02
−0.06, 0.02
−0.08, 0.00
−0.07, 0.01
−0.08, 0.00
−0.07, 0.00
FDR-adjusted q-value
0.400
0.822
0.803
0.617
0.057
0.368
0.153
0.077
Ever given birth (=1), n = 2,033
0.075
−0.014
−0.006
−0.010
−0.010
−0.031†
−0.015
−0.025
−0.024†
95% CI
−0.05, 0.02
−0.05, 0.04
−0.04, 0.02
−0.04, 0.02
−0.06, 0.00
−0.05, 0.02
−0.06, 0.01
−0.05, 0.00
FDR-adjusted q-value
0.400
0.822
0.803
0.617
0.057
0.368
0.153
0.081
Fertility outcomes summary index z-score (n = 2,039)
0.000
−0.066
−0.029
−0.025
−0.040
−0.160∗
−0.078
−0.102
−0.113∗
95% CI
−0.20, 0.07
−0.21, 0.15
−0.17, 0.12
−0.17, 0.09
−0.28, −0.03
−0.21, 0.05
−0.23, 0.02
−0.22, −0.01
Baseline out-of-school sample (25.6% of full sample)
Ever married (=1), n = 523
0.523
−0.243∗∗
−0.078
−0.168∗
−0.172∗
−0.247∗∗
−0.087
−0.173∗
−0.180∗
95% CI
−0.39, −0.10
−0.28, 0.12
−0.33, 0.00
−0.32, −0.02
−0.40, −0.09
−0.29, 0.11
−0.34, 0.00
−0.33, −0.02
FDR-adjusted q-value
0.005
0.651
0.076
0.060
0.007
0.586
0.075
0.059
p value test diff in-school
.001
.524
.038
.023
.002
.483
.039
.023
Ever pregnant (=1), n = 519
0.360
−0.182∗
−0.111
−0.154†
−0.153∗
−0.185∗
−0.109
−0.157†
−0.156∗
95% CI
−0.33, −0.03
−0.28, 0.06
−0.31, 0.00
−0.30, −0.01
−0.34, −0.03
−0.27, 0.06
−0.31, 0.00
−0.30, −0.01
FDR-adjusted q-value
0.027
0.567
0.076
0.060
0.026
0.567
0.075
0.059
p value test diff in-school
.012
.232
.052
.040
.012
.235
.052
.040
Ever given birth (=1), n = 520
0.244
−0.089†
−0.007
−0.065
−0.058
−0.091†
−0.010
−0.067
−0.062
95% CI
−0.19, 0.01
−0.14, 0.12
−0.17, 0.04
−0.16, 0.04
−0.19, 0.01
−0.14, 0.12
−0.18, 0.04
−0.16, 0.04
FDR-adjusted q-value
0.087
0.911
0.232
0.252
0.084
0.883
0.228
0.232
p value test diff in-school
.105
.875
.314
.351
.108
.890
.309
.333
Fertility outcomes summary index z-score (n = 523)
0.000
−0.471∗∗
−0.218
−0.345∗
−0.358∗
−0.480∗∗
−0.225
−0.355∗
−0.370∗
95% CI
−0.77, −0.17
−0.58, 0.15
−0.67, −0.02
−0.66, −0.06
−0.79, −0.17
−0.59, 0.14
−0.69, −0.02
−0.68, −0.06
p value test diff in-school
.004
.684
.067
.054
.005
.650
.070
.054
Column 1 reports means for the V-only study arm at endline, and columns 2–4, the estimated ITT effects for each study arm relative to V-only. Column 5 pools all three study arms with education into a single treatment variable. Columns 6–9 report estimated ITT effects with extended controls. Regressions estimated with robust standard errors clustered at the village level and included controls for district per the stratified design, age, and outcome variable at baseline. Extended controls regressions additionally included baseline measures of in school, grade attainment, cognitive score, father ever attended school, co-residence with both parents, household wealth quintile, and whether any missing baseline covariates were imputed (with the cluster median). p value test diff in-school reports the p value of the interaction between out-of-school at baseline and study arm using full sample. FDR-adjusted q-values are for the set of three primary outcomes in each sample. Results for the baseline in-school sample are shown in Appendix D.
CI = confidence interval; FDR = false discovery rate; ITT = intent to treat; SD = standard deviation; VE = violence prevention and education; VEH = VE and health; VEHW = all four interventions.
For the baseline in-school subsample, marriage, pregnancy, and birth rates at endline in the V-only study arm were even lower than those for the full sample, and estimates of all ITT effects were small and insignificant across specifications (Appendix D). There were large program impacts for the out-of-school subsample (Table 2), however, in which, at endline, more than half of girls in V-only had married and a quarter had given birth. Estimated effects were largest for the VE study arm, which reduced marriage and pregnancy by nearly half, followed by VEHW by a third. For these outcomes, impacts for both the VE and VEHW study arms were significantly different between the baseline in-school and out-of-school subsamples. As seen for the full sample, the pooled estimates indicate large reductions in both marriage and pregnancy, although smaller than those in VE alone. The summary measure shows a decline of ∼0.4 SD for all but the VEH study arm, which was ∼0.2 SD and insignificant. The extended control regression results (columns 6–9) are nearly identical.
There were no significant effects on either the violence prevention or health summary measures for the full sample (Table 3). The VE, VEH, and VEHW study arms increased educational outcomes between 0.1 and 0.2 SD, although VEH was only marginally significant. VEHW incrementally increased the wealth summary measure. The findings are similar to the pattern of estimated effects at the 2-year follow-up just after the program ended in 2017 [
Impacts of two-year multisectoral cash plus programs on young adolescent girls’ education, health and economic outcomes: Adolescent Girls Initiative-Kenya (AGI-K) randomized trial.
], pointing to persistence in effects on education and financial literacy, although effect sizes are smaller at endline. For the baseline in-school subsample, the only significant effect observed for secondary outcomes was an increase in wealth creation in the VEHW study arm (Appendix D). For the baseline out-of-school subsample, however, there were large effects (0.3 SD or more) on both violence prevention and education for the VE and VEHW study arms, but somewhat smaller and insignificant effects for the VEH study arm (Table 3). For VEH, there was a negative significant effect on wealth creation, driven by low financial literacy scores, which were not part of the VEH intervention. Extended control regression results (columns 6–9) were nearly identical. Results for secondary outcomes were robust to IPW weighting for attrition, with point estimates and significance levels changing only slightly (Appendix E). Appendix F presents results for the individual components of the secondary outcomes.
Table 3Estimated intent-to-treat effects on secondary outcomes summary measures at endline
VE, estimate
VEH, estimate
VEHW, estimate
VE-VEH-VEHW pooled, estimate
VE, estimate: extended controls
VEH, estimate: extended controls
VEHW, estimate: extended controls
Pooled, estimate: extended controls
Full sample
Violence prevention outcomes summary index z-score (n = 2,011)
−0.006
0.103
0.074
0.058
0.010
0.114†
0.095
0.074
95% CI
−0.14, 0.13
−0.04, 0.24
−0.09, 0.23
−0.06, 0.18
−0.13, 0.15
−0.02, 0.25
−0.06, 0.25
−0.05, 0.19
FDR-adjusted q-value
0.975
0.202
0.461
0.682
0.890
0.141
0.306
0.447
Education outcomes summary index z-score (n = 2,039)
0.193∗∗∗
0.099†
0.111∗
0.134∗∗
0.199∗∗∗
0.103†
0.110∗
0.137∗∗
95% CI
0.09, 0.29
−0.02, 0.21
0.01, 0.22
0.05, 0.22
0.10, 0.30
−0.01, 0.22
0.01, 0.21
0.06, 0.22
FDR-adjusted q-value
0.002
0.202
0.075
0.007
0.001
0.141
0.072
0.005
Health outcomes summary index z-score (n = 2,011)
0.003
0.043
0.065
0.037
0.030
0.063
0.091
0.062
95% CI
−0.17, 0.18
−0.15, 0.23
−0.11, 0.24
−0.10, 0.18
−0.15, 0.21
−0.13, 0.25
−0.08, 0.27
−0.08, 0.20
FDR-adjusted q-value
0.975
0.651
0.461
0.799
0.890
0.507
0.306
0.504
Wealth creation outcomes summary index z-score (n = 2,011)
−0.091
−0.169
0.263∗
0.000
−0.027
−0.139
0.319∗∗
0.049
95% CI
−0.28, 0.09
−0.37, 0.04
0.02, 0.51
−0.17, 0.17
−0.21, 0.16
−0.31, 0.03
0.08, 0.56
−0.12, 0.21
FDR-adjusted q-value
0.664
0.202
0.075
0.999
0.890
0.141
0.037
0.554
Baseline out-of-school sample (25.6% of full sample)
Violence prevention outcomes summary index z-score (n = 518)
0.295∗
0.230†
0.378∗∗
0.307∗∗
0.284∗
0.216
0.369∗∗
0.298∗∗
95% CI
0.06, 0.53
−0.03, 0.49
0.13, 0.63
0.08, 0.53
0.05, 0.52
−0.05, 0.49
0.12, 0.61
0.08, 0.52
FDR-adjusted q-value
0.033
0.170
0.007
0.017
0.035
0.193
0.008
0.018
p value test diff in-school
.015
.299
.012
.025
.012
.349
.012
.025
Education outcomes summary index z-score (n = 523)
1.103∗∗∗
0.361
0.748∗∗∗
0.779∗∗∗
1.095∗∗∗
0.345
0.750∗∗∗
0.779∗∗∗
95% CI
0.78, 1.43
−0.11, 0.83
0.41, 1.09
0.46, 1.10
0.77, 1.42
−0.12, 0.81
0.43, 1.07
0.47, 1.09
FDR-adjusted q-value
0.001
0.170
0.001
0.001
0.001
0.193
0.001
0.001
p value test diff in-school
.000
.334
.000
.000
.000
.377
.000
.000
Health outcomes summary index z-score (n = 518)
0.059
0.088
−0.005
0.044
0.040
0.089
−0.019
0.031
95% CI
−0.22, 0.34
−0.31, 0.48
−0.29, 0.28
−0.22, 0.31
−0.22, 0.30
−0.29, 0.47
−0.29, 0.25
−0.22, 0.28
FDR-adjusted q-value
0.678
0.662
0.973
0.742
0.763
0.644
0.891
0.805
p value test diff in-school
.898
.871
.477
.869
.966
.888
.394
.771
Wealth creation outcomes summary index z-score (n = 518)
−0.078
−0.340∗
0.048
−0.101
−0.075
−0.320∗
0.052
−0.089
95% CI
−0.33, 0.18
−0.62, −0.06
−0.24, 0.34
−0.35, 0.15
−0.35, 0.20
−0.61, −0.03
−0.26, 0.36
−0.36, 0.18
FDR-adjusted q-value
0.678
0.078
0.973
0.555
0.763
0.134
0.891
0.679
p value test diff in-school
.519
.025
.101
.122
.530
.041
.123
.163
Columns 1–3 report the estimated ITT effects for each study arm relative to the V-only study arm measured in z-scores (see Appendix G). Column 4 pools all three study arms with education into a single treatment variable. Columns 5–8 report estimated ITT effects with extended controls. Regressions estimated with robust standard errors clustered at the village level and included controls for district per the stratified design, age, and the outcome variable at baseline. Extended controls regressions additionally included baseline measures of in school, grade attainment, cognitive score, father ever attended school, co-residence with both parents, household wealth quintile, and whether any missing baseline covariates were imputed (with the cluster median). p value test diff in-school reports the p value of the interaction between being out of school at baseline and study arm using full sample. FDR-adjusted q-values are for the set of four secondary outcomes in each sample. Results for the baseline in-school sample are shown in Appendix D.
CI = confidence interval; FDR = false discovery rate; ITT = intent to treat; SD = standard deviation; VE = violence prevention and education; VEH = VE and health; VEHW = all four interventions.
Attrition bounds for the estimated effects on the primary outcomes for the full sample all remain negative and most remain at least marginally significant (Appendix E). Alternative approaches to estimating standard errors indicate statistical significance holds with few exceptions and is often stronger under randomization inference (Appendix E).
Discussion
Four years after the start of AGI-K, we find no overall effect in unadjusted analyses; however, for analyses with extended controls including baseline education, the program reduced early marriage by 4 percentage points for girls aged ∼15–18 years. The average ITT effect, however, masks substantial heterogeneity. In particular, there were no effects for girls initially enrolled in school at baseline but large ones for those not initially enrolled for whom education increased, marriage was a remarkable 18.0 percentage points (34%) lower, and pregnancy 15.6 percentage points (43%) lower. These ITT differences did not reflect differential program take-up rates, which were 25 percentage points lower for girls not enrolled at baseline. Because outcomes were assessed 2 years after the end of the program, the results represent persistent, postintervention effects.
Although the research design does not allow identification of which specific mechanisms underlie delayed marriage and fertility, there is suggestive evidence that schooling was an important driver in this context. Apart from wealth creation for VEHW, there were no significant differential effects on potential mechanisms across the various study arms with the education CCT. There were also no program effects on a household-level wealth index [
Impacts of two-year multisectoral cash plus programs on young adolescent girls’ education, health and economic outcomes: Adolescent Girls Initiative-Kenya (AGI-K) randomized trial.
] indicating the cash transfers, which stopped 2 years before endline, did not substantially alter household wealth. A simple mediation analysis re-estimating models in Table 2 while controlling for endline secondary outcome summary measures one at a time indicated that the only endline measure whose inclusion reduced the significant program effects was education. Finally, field reports and implementing partners highlight local perceptions of increased valuation of girls' education and its importance in expanding future opportunities and serving as an alternative to marriage during adolescence.
Discussions with mentors and implementing staff indicate that the limited impact on the measured health domain was probably in part the result of the discomfort mentors felt teaching the SRH curriculum, either because of personal beliefs or concerns that covering the material would disrupt the efficacy of other program components. Because there were short-term effects on other topics covered in the girls groups [
Impacts of two-year multisectoral cash plus programs on young adolescent girls’ education, health and economic outcomes: Adolescent Girls Initiative-Kenya (AGI-K) randomized trial.
], we suggest this was due to social norms rather than weak implementation. We do not conclude, however, that SRH content or girls' empowerment groups should be excluded from future interventions. Similar to most countries, Kenya has joined international agreements such as the 2012 Commission on Population and Development Resolution, calling on governments to provide sex education to adolescents and enshrining access to SRH information as a right [
]. From a program design perspective, it may be that starting with interventions that first address educational and economic deficits and then introducing the SRH components would be more effective. Future research should explore culturally appropriate and effective ways to incorporate SRH content in similar settings.
Taken together, the results in this article demonstrate the potential for multisectoral, multilevel interventions with an education component to delay marriage in a marginalized and socially conservative setting with high rates of early marriage. With effects concentrated among the most at-risk girls, AGI-K demonstrated that with early intervention, it was possible to shift girls off the path to early marriage and onto one of continued education. The results do not suggest only targeting out-of-school girls, however, given the risk of disincentivizing schooling among those already enrolled and recognizing the existing norms and substantial poverty that are barriers to girls' progress.
The study has several limitations. First, we are unable to generalize the findings beyond similarly pastoralist and socially conservative regions. Second, it was not feasible to incorporate a pure control group or implement a full factorial design that would have enabled independent assessment of single-sector interventions on their own. Third, the baseline and final samples were smaller than the planned target sample, possibly decreasing power.
At the same time, the study has significant strengths. It is one of few studies using a longer term, longitudinal design to examine these types of interventions for difficult-to-reach, marginalized, and socially conservative populations. Such follow-up is essential for understanding program impacts among adolescents because early investments may only pay dividends well after the interventions have ended [
]. Fidelity and take-up of the AGI-K program were high, and there was no contamination, allowing an ITT approach in the analysis. Finally, the study used a rigorous randomized design and had minimal attrition.
This study suggests program implementers and policy makers should continue exploring interventions in early adolescence that include educational components to delay negative outcomes such as early marriage and fertility. The case for this is even stronger when one considers that such delays have been shown elsewhere to have longer term individual and intergenerational benefits [
The authors would like to thank the UK Foreign Commonwealth and Development Office for funding this research study (PO6171). They thank Save the Children in Kenya for overseeing quality implementation of the intervention and the members of the AGI-K National External Advisory Committee (EAC) and the Wajir EAC for their invaluable input into the interpretation of the endline results. The authors thank the survey teams in Wajir for their dedication to collecting data and Lizzie Friesen for her careful work cleaning the endline data. They are also grateful for the contributions of Eunice Muthengi to the inception and implementation of the study while she was a staff member at the Population Council. The authors similarly thank Caroline Kabiru and Joyce Mumah, in former roles at the African Population and Health Research Center (APHRC), for contributions in the earlier phases of the study. They also thank those from the Foreign, Commonwealth and Development Office in Kenya and the AGI-K Expert Research Committee who provided feedback on drafts of preliminary results. Finally, the authors thank all the adolescent girls who took the time to participate in the study.
Interventions to prevent child marriage among young people in low-and middle-income countries: A systematic review of the published and gray literature.
The Samata intervention to increase secondary school completion and reduce child marriage among adolescent girls: Results from a cluster-randomised control trial in India.
The case for a multi-sectoral approach to preventing child marriage and early childbearing in Sub-Saharan Africa. Global perspectives on women’s sexual and reproductive health across the lifecourse.
Impacts of two-year multisectoral cash plus programs on young adolescent girls’ education, health and economic outcomes: Adolescent Girls Initiative-Kenya (AGI-K) randomized trial.
Impacts of multisectoral cash plus programs after four years in an urban informal settlement: Adolescent girls initiative-Kenya (AGI-K) randomized trial.
Multiple inference and gender differences in the effects of early intervention: A reevaluation of the abecedarian, perry preschool, and early training projects.
00685-6&id=gr1.jpg){kind=link}
00685-6&id=gr2.jpg){kind=link}
00685-6&id=gr3.jpg){kind=link}