Cash, Income Generating Activities, and Parental Training for Vulnerable Families in Malawi

Households were randomly assigned to one of four arms in a 2×2 design, crossing two forms of economic support with a parental training program.

**Income-generating activity (IGA) package.** SOS's standard livelihood intervention. Caregivers first received general financial and business management training and chose a trade. Between January and April 2026 they received trade-specific skills training and start-up assets tailored to their choice — for example, goats or pigs with livestock-management training; tailoring, welding, or salon training with equipment; irrigation training and inputs; or start-up stock for small trading businesses. The package is valued at roughly MWK 1,500,000 (approximately USD 900) per household.

**Unconditional cash transfer.** Households received MWK 1,500,000 (approximately USD 900), matching the value of the IGA package, in two equal payments of MWK 750,000: the first in October–November 2025 and the second in March–April 2026. No conditions were attached to the use of the funds.

**Parental training.** Half of the households in each economic-support arm were additionally invited to SOS's positive parenting program, delivered from May 2026 in community-based group sessions using an eleven-module curriculum: parenting styles, fathers' involvement in childcare, feeding children for healthy growth, caregiver mental health, domestic violence, alternatives to corporal punishment, parents' role in education, teen parenting and school dropout, harmful cultural practices, the child protection referral system, and health-seeking behaviours.

2025-09-01

2026-10-31

Study A:

1. Attends school or vocational training regularly and not repeating a grade, ages 6–17, binary. Endline survey.
2. Child development and learning index: early development direct assessment adapted from IDELA and similar tools for ages 3–5 and early-grade reading and mathematics assessment subtasks (EGRA/EGMA, Chichewa and English) for ages 6 and older, standardized within age band. Endline survey.
3. Child nutrition index: meals per day and dietary diversity in the past 7 days and in the most recent hungry season, ages 3 and older, standardized index. Implementation survey and endline survey.
4. Not malnourished: BMI-for-age at or above −2 SD, enumerator-measured, ages 3–17, binary. Endline survey.
5. Household cash income, past year: enumerator-assisted reconstruction by source over the past 12 months, excluding remittances, transfers, and own-produced food, household level, continuous. Endline survey.
6. Net household wealth: savings + durable assets + productive assets + livestock + net land transactions − debt, household level, continuous. Endline survey.
7. Household consumption per capita: food consumed in the past 7 days (purchased, own-produced, or received) and non-food expenditure, valued, per household member per day, household level, continuous. Endline survey.

Study B: TBC before endline data collection. Endline only.

**1. Attends school or vocational training regularly and not repeating a grade.** Ages 6–17; child-reported when the child is present, otherwise caregiver-reported. "Do you go to school regularly or have you received other types of education or vocational training during the last year?" No = 0; Yes, vocational training = 1; Yes, school: 1 if not repeating the current class, 0 if repeating.

**2. Child development and learning index.** Ages 3 and older; administered directly to the child. Within each instrument, an inverse-covariance-weighted index (Anderson 2008), standardized to the cash arm within age band (3–5, 6–12, 13–17; mean 0, SD 1).

**3. Child nutrition index.** Ages 3 and older; caregiver-reported for ages 3–5; for ages 6 and older, child-reported when the child is present, otherwise caregiver-reported. Inverse-covariance-weighted index (Anderson 2008) of four components, each oriented so that higher is better: meals per day in the past 7 days; meals per day in the most recent hungry season; dietary diversity (number of 10 food groups consumed in the past 7 days); dietary diversity in the most recent hungry season. Standardized to the cash arm within survey round.

**4. Not malnourished.** Ages 3–17; enumerator-measured height and weight. BMI-for-age from the WHO sex-specific simplified field tables: 1 if at or above −2 SD, 0 if below.

**5. Household cash income, past year.** Household level; caregiver-reported, with enumerator-assisted reconstruction by source over the past 12 months (months or units worked × typical amount per unit, per source). Sources: wages/salary, ganyu (casual labor), own business or trading, sale of crops or livestock, and other income; excludes remittances and government/NGO transfers (which include the study cash transfers) and the value of own-produced food (see primary outcome 7).

**6. Net household wealth.** Household level. Sum of (i) total savings (cash at home, bank, mobile money, savings groups); (ii) estimated resale value of durable assets; (iii) estimated resale value of productive/business assets; (iv) livestock, as counts by type valued at district-median unit resale prices from the same survey; and (v) net land transactions since randomization (purchases minus sales); minus (vi) total debt outstanding (banks/MFIs, savings groups, moneylenders, shop credit, family/friends). The total can be negative.

**7. Household consumption per capita.** Household level; caregiver-reported. Food: quantity of each item consumed by the household in the past 7 days, with purchases valued at the amount paid and own production and gifts at district-median unit prices from the same survey. Non-food: household items, fuel, airtime, transport, and clothing in the past 30 days; school and health spending and housing repairs in the past 12 months. Durables and ceremonies excluded. Food and annual items rescaled to 30 days; total monthly consumption divided by household size and by 30.

Study A:

1. Adequate care composite: composite of five indicators (school, healthcare access, nutritional status, violence at home, mental health), ages 3–17, binary. Endline survey.
2. Access to healthcare: no difficulty accessing healthcare for the child in the past 12 months, ages 3–17, binary. Implementation survey and endline survey.
3. Not exposed to violence at home: no physical punishment or withholding of meals, and no physical fights between adults seen at home in the past year, child-reported, ages 6–17, binary. Implementation survey and endline survey.
4. Mental health in normal range: Pediatric Symptom Checklist (PSC-10) score at or below 8, ages 6–17, binary. Implementation survey and endline survey.
5. Health inputs index: health passport, vaccination, vitamin A, deworming, and bed net use, ages 0–5, standardized index. Implementation survey and endline survey.
6. Access to menstrual products: always has access to sanitary pads or other adequate materials when needed, girls 11–17, binary. Endline survey.
7. Child labor: days worked in the past month on the family farm (ages 6+) or for pay (ages 14+), counts, implementation survey; months worked in the past year on the family farm, looking after livestock, in a household business, or for pay, ages 6+, count (0–12), endline survey.
8. No school days missed because of work in the past year, ages 6+ enrolled at some point in the past year, binary. Endline survey.
9. Consumption over the poverty line: household consumption per capita at or above the World Bank extreme poverty line, household level, binary. Endline survey.

Study B: TBC before endline data collection.

**1. Adequate care composite.** Ages 3–17; scored five child indicators as measured in this study (primary outcomes 1 and 4 and secondary outcomes 2–4). Ages 3–5: 1 if both healthcare access and not malnourished = 1, 0 if either = 0. Ages 6–17: 1 if at least four of the five indicators = 1 and none = 0; 0 if any = 0; missing otherwise. Endline only.

**2. Access to healthcare.** Ages 3–17; caregiver-reported (child-reported if 13 or older and living with a spouse/partner). Difficulty accessing healthcare for the child in the past 12 months: Yes = 0; No = 1; Didn't need = 1; Don't know = missing.

**3. Not exposed to violence at home.** Ages 6–17; child-reported, asked only if the child is present, assents, and privacy is ensured. Two questions: (a) no physical punishment nor withholding of meals; (b) no physical fights between adults in the home seen in the past year. 1 only if both = 1; 0 if either = 0; "Don't know / will not say" scored 0.

**4. Mental health in normal range.** Ages 6–17; caregiver-reported PSC-10 (child self-report if 13 or older and living with a spouse/partner). Ten items scored never or rarely = 0 / sometimes = 1 / very often = 2, with "Don't know" scored 1; raw score 0–20; 1 if at or below 8 (normal range), 0 if above.

**5. Health inputs index.** Ages 0–5 (at endline, including children born since baseline); caregiver-reported. Five binary inputs, higher is better: has a health passport (enumerator-verified where possible), fully vaccinated for age, vitamin A in the past 6 months, deworming in the past 12 months, and slept under a mosquito net last night. Inverse-covariance-weighted index (Anderson 2008), standardized to the cash arm within survey round.

**6. Access to menstrual products.** Girls 11–17; child-reported in private, asked only if the girl is present and assents. 1 if she always has access to sanitary pads or other adequate materials when needed, 0 if not; "Not yet menstruating" and "Will not say" treated as missing.

**7. Child labor.** Child-reported when the child is present, otherwise caregiver-reported; higher is worse. Implementation survey: number of days in the past month worked on the family farm (ages 6 and older) and number of days worked for pay (ages 14 and older). Endline survey: number of months (0–12) in the past year with any work on the family farm, looking after livestock, helping in a household business, or for pay, asked of all children aged 6 and older; 0 if none.

**8. No school days missed because of work.** Ages 6 and older; child-reported when the child is present, otherwise caregiver-reported. 1 if the child missed no school days because of work in the past 12 months (including children who did no work), 0 if any days were missed. Defined only for children enrolled at some point in the past year.

**9. Consumption over the poverty line.** Household level; from primary outcome 7. Daily consumption per capita compared with the World Bank international extreme poverty line: 1 if at or above the line, 0 if below.

The trial is a household-level randomized controlled trial with a 2×2 factorial design. In August 2025, 1,200 households enrolled in the SOS Children's Villages family-strengthening program across four districts in Malawi were assigned with equal probability to one of four arms: (1) the income-generating activity (IGA) package, (2) IGA plus parental training, (3) an unconditional cash transfer of equivalent value, and (4) cash plus parental training, giving 300 households per arm. Randomization was stratified by district and by whether the household had any primary-school-aged child out of school at baseline (eight strata), using a pre-specified seed and a stratified randomization routine.

The design supports two comparisons. Study A compares the IGA package to cash (600 versus 600 households), pooling across the parental training arms. Study B evaluates parental training (600 versus 600 households), pooling across the economic-support arms, and examines whether its effects depend on the form of economic support.

Baseline data were collected in 2024 by SOS before randomization and cover all children under 18 in each household. The household sample is all 1,200 randomized households. The child sample is children in these households aged 17 or younger at endline (16 or younger at the implementation survey), with eligibility determined by the age recorded at each survey. The baseline masterlist identifies approximately 3,285 such children; this is approximate because masterlist ages are noisy. Implementation-survey child records whose recorded code is missing, ambiguous, or inconsistent with the masterlist (sex or age) will be matched to the masterlist before analysis, using other survey information (e.g. sex, age, enumerator, and survey timing). The endline survey will record the current age and status (interviewed, absent, moved, aged 18 or older, deceased, refused) of every child in the sample, to measure attrition. Follow-up data are collected in two rounds: an implementation survey in April 2026, after delivery of the cash and IGA support and before parental training began, and an endline survey planned for spring 2027. Each round includes a caregiver survey and an individual survey or assessment for each child, with children aged six and older interviewed directly where present and assenting. Longer-run surveys will be planned conditional on funding.

We will estimate OLS regressions of each outcome on an IGA indicator (the cash arm is the reference category), separately by survey round.

Household-level controls: stratification bin, parental-training assignment, baseline annual income (with a missing indicator), baseline number of earners, and the number of boys and of girls in the child sample according to the baseline masterlist. Child controls: age, age band (as in the heterogeneity analysis), gender, and baseline (2024) measures of the outcome, where available, and of primary outcomes 1 and 4, each with a missing indicator. Heterogeneity by child gender and age bands (0–5, 6–12, 13–17). Child age and gender, for covariates and heterogeneity, are as recorded at the implementation survey, or at endline for children without an implementation-survey record.

Standard errors clustered by household, and randomization inference p-values (10,000 replications, re-randomizing with the original assignment mechanism).

We will conduct further exploratory analyses, including to examine implementation fidelity and mechanisms, to test for robustness, and to provide greater descriptive detail.

Not available

Randomization done in office by a computer, using Stata's randtreat command with seed 20250812, stratified on 8 district × baseline-enrollment strata, with misfits assigned using randtreat's misfits(wglobal) option.

Household

Yes

1,200 households (randomization unit).

1,200 households (household-level outcomes) and approximately 3,285 children aged 17 or younger at endline (child-level outcomes). The child count is based on baseline masterlist ages, which are noisy; eligibility is determined by the age recorded at each survey. At endline, children born into sample households since baseline will be added for some outcomes and analyses.

IGA: 300 households, approx. 823 children
IGA + parental training: 300 households, approx. 814 children
Cash: 300 households, approx. 825 children
Cash + parental training: 300 households, approx. 823 children

Study A (IGA vs. cash, pooled over parental training): 600 vs. 600 households; approx. 1,637 vs. 1,648 children.
Study B (parental training vs. none, pooled over economic support): 600 vs. 600 households; approx. 1,637 vs. 1,648 children.