Lambon-Quayefio et al. (2021). Government Unconditional Transfers and Safe Transitions into Adulthood Among Youth in Malawi. IUSSP virtual conference: https://ipc2021.popconf.org/sessions/13
2024: The FAR, Federal Acquisition Regulations - Part 24
Government Unconditional Transfers and Safe Transitions into Adulthood - Lambon-Quayefio at IUSSP
1. Government
Unconditional Transfers
and Safe Transitions into
Adulthood Among Youth
In Malawi
Monica Lambon-Quayefio on behalf of the
Malawi Social Cash Transfer Evaluation Team
IUSSP VIRTUAL CONFERENCE
December 2021
2. Background and Motivation
• Africa youth are recognized as a key demographic critical to harness the
continent’s growth potential and achieve the SDGs
• The use of social safety nets (cash transfers) have become popular tools to
reach poor households and address SDG1
• There is limited research and evidence on the ability of cash transfers to
facilitate adolescent and youth safe transitions:
• Existing evidence is concentrated in Latin America
• Inter-generational impacts mostly focus on a few domains (education and labor force
participation) (Molina Millan et al. 2019)
• Evidence on broader domains comes mainly from smaller programs implemented by NGOs
or from the HIV sector (e.g. Pettifor et al. 2016)
3. Research Questions
Recent review argues social protection can contribute to a “pipeline of investments”
for at-risk youth (Cirillo et al. 2021). For future investment in cash transfers, it is
important to investigate:
• If national “at-scale” cash transfers have implications for youth that reside in participant
households
• Implications for inter-generational transmission of multi-dimensional poverty
Objectives of current study:
• Examines the effects of Malawi’s unconditional cash transfer program on safe
transitions to adulthood among youth aged 13 – 19 at baseline in ultra-poor
households over nearly three years.
• Uses a broad range of indicators across six domains and examines impacts
separately for male and female youth
4. Malawi Social Cash Transfer Program
• The Malawi SCT is implemented by the Ministry of Gender, Children, Disability and
Social Welfare and is currently operational in all 28 districts (started in 2006)
• Targeted population: Ultra-poor and labour-constrained households
• Transfer operational details:
• Bimonthly payments in cash through a local pay-point manager to caregiver
(primarily women – 84%)
• Unconditional
• Transfer value:
• Varying amounts (based on household size and the number of primary and
secondary school-aged children in the home)
• Equivalent to US$1.25 a month (US$60 per household annually)
• Payments represent 17 to 23% of median pre-program household consumption
5. Data and evaluation design
• Cluster randomized control trial (RCT)
• Districts: Salima and Mangochi
• Three waves: 2013 – 2014 – 2015
• Household survey administered to the primary
caregiver in all three waves
• Additional youth surveys: aged 13 to 19 years old
at baseline (by same sex enumerators)
• Analytical data (1,679 unique youth):
• Panel data on youth interviewed at baseline
and again at least one follow up wave
• Data collected on multiple domains
representing youth safe transitions
• Demographics and background characteristics
from household survey Map of study districts and village clusters in
Malawi: Mangochi [bottom] and Salima [top]
6. Modelling Approach
ANCOVA estimation (with DID robustness checks) pooling follow-up waves
• Overall impact: 𝑌𝑖𝑗1= 𝛽0 +𝛽1𝑇𝑟𝑒𝑎𝑡𝑗 +𝛽2𝑌𝑖𝑗0+𝛽3𝑋𝑖𝑗0+𝛾 + 𝜀𝑖𝑗1
• Heterogeneous impact (by gender of youth):
𝑌𝑖𝑗1= 𝛽0 +𝛽1𝑇𝑟𝑒𝑎𝑡𝑗 ∗ 𝐹𝑒𝑚𝑎𝑙𝑒𝑖𝑗 + 𝛽2𝑇𝑟𝑒𝑎𝑡𝑗 + 𝛽3𝐹𝑒𝑚𝑎𝑙𝑒𝑖𝑗 + 𝛽4𝑌𝑖𝑗0+𝛽5𝑋𝑖𝑗0+𝛾 + 𝜀𝑖𝑗1
• Control variables: Age of youth (year splines), gender of youth, female headed
household, household head age, household head ever attended school, household size,
district traditional authority fixed effects
• Baseline balance and attrition analysis show good internal validity – with the exception
of baseline balance of sexual and reproductive health driven by female youth (treatment
group is worse off at baseline)
7. Main Outcomes: Six Domains
Domain Indicators
1 Physical health (a) reported health; (b) morbidity; (c) use of mosquito net; (d) activities of
daily living
2 Emotional wellbeing
and mental health
(a) depressive symptoms; (b) aspirations; (c) social support
3 Education (a) enrollment; (b) highest grade completed; (c) attendance; (d)
expenditure on education
4 Time use and labor (a) time spent on economic activities; (b) hazardous labour; (c) time spent
on domestic activities
5 sexual and
reproductive health
(a) ever pregnant (females only); (b) currently partnered; (c) sexual debut;
(d) any sexual experience; (e) ever experienced forced sex
6 HIV risk (a) number of sexual partners; (b) age disparate sex; (c) transactional sex;
(d) concurrent partnerships; (e) number of sexual acts; (f) condom use
* Six Domains with 25 indicators
**All indicators are coded so more favorable transition outcomes is higher—transformed into a z-score
and aggregated into an index, standardized according to the control group by wave
8. Results: Overall impacts on domains by gender
• Significant impacts on 4 out of
6 domains—all except time
use and HIV risk
• Impacts are largest for
domains of mental health and
education
• Impacts overall are similar for
male and female youth –
however females have
significantly higher impacts on
physical health
* Figure shows 90% confidence intervals
9. Results: Impacts on Domains and Indicators (1)
* Figure shows 90% confidence intervals
10. Results: Impacts on Domains and Indicators (2)
* Figure shows 90% confidence intervals
11. Conclusion and Policy Implications
• Results indicate that adolescents living in ultra-poor households
improved outcomes after one and two years of household-receipt of
transfers
• Improvements go beyond domains directly related to receipts of cash
(e.g., education, physical health) to outcomes including mental health
and emotional wellbeing, as well as sexual and reproductive health
• No impacts on some domains and outcomes—in particular, time use and
labor. The lack of impacts may be due to increased household income
generation and productive activities (which youth may take part in)
• More can be done to intentionally design for, reach and evaluate impacts
for adolescents and youth populations, especially during times of shocks.
12. Thank you!
Co-Authors: Sudhanshu Handa, Amber Peterman, Adria
Molotsky, Frank Otchere, Peter Mvula, Maxton Tsoka, Jacob de
Hoop, Gustavo Angeles, Kelly Kilburn, Annamaria Milazzo
Acknowledgements: We thank the Government of Malawi for their
supportive engagement with the evaluation team, and for their time and
intellectual contributions, specifically Dr. Mary Shawa, Dr. Esmie Kainja,
Mr. Laurent Kansinjiro, Mr. Charles Chabuka and Mr. Gideon Kachingwe of
the Ministry of Gender, Children, Disabilities and Social Welfare, Mr. Harry
Mwamlima of the Ministry of Finance, Economic Planning and
Development, as well as the District Commissioner's Offices of Salima and
Mangochi. We would also like to acknowledge Chantal Elmont of Ayala
Consulting, and the research team CSR for their exceptional work. We
thank the European Union, the German Government through KfW, Irish
Aid, FAO, the International Initiative for Impact Evaluation (3ie), FCDO and
UNICEF Malawi for their financial contributions and support for the study.
In Africa, the population of adolescents and youth aged 15 to 24 is projected to grow faster than any other region, with an estimated near doubling between 2015 and 2050 (from 258 million to 456 million) (UN 2019). For this reason, adolescents and youth have been recognized as a key demographic for the region, for harnessing the economic and social benefits of the demographic dividend, and critical for achievement of the sustainable development goals (SDGs).
In the African region, on average countries have 15 SSN programs and spend approximately 4.6 percent of all government spending on SSN (Beegle et al. 2018). The rise of social protection has also been unprecedented during COVID-19, across the globe, the World Bank has calculated that at total of 3,333 social protection and labor market measures have been implemented or planned to respond to COVID-19 in 222 countries – 1/4 of which have been cash transfers.
https://documents1.worldbank.org/curated/en/281531621024684216/pdf/Social-Protection-and-Jobs-Responses-to-COVID-19-A-Real-Time-Review-of-Country-Measures-May-14-2021.pdf
The program began as a pilot in the Mchinji district in 2006, was expanded to an additional eight districts by 2009, and further expanded starting in 2014.
For this evaluation, we leverage a cluster RCT taking place within the expansion of the government’s program in Salima and Mangochi districts. There were 28 village clusters, and half were randomly assigned to Treatment via a public lottery. Data collection was led by the Center for Social Research at the University of Malawi in Zomba – interviews were conducted by same sex enumerators in Chichewa and Chiyao with tablets. IRB was obtained by the University of Malawi and at UNC.
ANCOVA is preferred to difference-in-difference (DD) models due to power gains when autocorrelations between baseline and follow-up dependent variables are low. Guidelines proposed in McKenzie (2012) suggest “low” autocorrelation is between 0.20 to 0.40, while “high” autocorrelation is between 0.6 to 0.8. As shown in appendix Table A3, across 17 individual outcomes, only one (highest grade of schooling completed, corr: 0.774) can be considered “high” and all but one domain index (education) can be considered “low.”
The point estimates are from the pooled ANCOVA model, with 90% confidence intervals in the bars (representing 10% significance levels) – positive point estimates represent positive impacts (favorable transitions) on the domain, while negative point estimates represent adverse transitions, however only those with CIs not crossing the zero line are significant. We can see that for average impacts (that is pooled between male and female youth – the blue estimates) there are significant positive impacts in 4 out of the 6 domains – all except HIV risk, which is almost significant, and time use (which is the only aggregate that is not positive). Among these, two are large and significant – the program has an average impact on mental health and wellbeing of 0.26 SD in comparison to the control group and an average impact on education of 0.3 SDs in comparison to the control group. Overall the impacts are largely similar between male and female youth (represented by the red and green point estimates), however in one case (physical health) the impacts for females are significantly larger.
Here we have the same figures, including both the aggregate index (as shown before), as well as the indicators that make up the index. All the indicators are standardized to z-scores, so they are in the same units. From the domain on physical health, we can see that impacts are driven by sleeping under a mosquito net, and for females particularly, freedom from illness and injury. If we were to convert the impacts to real terms, the impacts for bednet use translate into a 28% increase compared to the endline control group. For the domain on emotional wellbeing and mental health, we see impacts are driven by the CES-D scale, which is a scale of depressive symptoms, as well as social support. For example, in real terms, the impacts on CES-D translate to a change of 9% on the depression scale.
Here we have the impacts on education and on SRH, which are the two other domains which were significantly improved on aggregate. For education, we see that the impacts are coming from impacts on enrollment and expenditure on education. For enrollment, in real terms these impacts translate to a 26% increase as compared to the endline control group—and for spending they translate into a 79% increase in Malawi kwatcha spent in comparison to the endline control group. For SRH, we see that although most impacts are positive, the only one that is significant is for forced sex. Here, we see that in real terms, the impacts are equivalent to a 6% increase in freedom from being forced, tricked or pressured into sex.
In the paper, we also address some robustness checks – including sensitivity to using difference-in-difference models, evolution of impacts over time and correction of standard errors for small number of clusters.