Evaluating the Addition of Asset Transfers and Training to a Cash for Work Program in Yemen

Comparison between a pure cash-for-work intervention and the same cash-for-work intervention with an add-on livelihoods component that provides assets and training to encourage new self-employment activities in rural areas.

2025-04-06

2026-01-17

• Engagement in productive work (binary)
• Paid for work (binary)
• Self-employment in small businesses/agriculture (binary)
• Any profit from self-employment (binary)
• Profit from self-employment (revenue minus costs, in YER)
• Self-efficacy (PCA score based on 3 questions)
• Days/week spent on economic activities
• Women's decision-making in specific activities (binary)
• Women’s decision-making index (PCA on recoded indicators)

• Revenue from self-employment (continuous, in YER)
• Type of economic activity (agriculture vs. other)
• Time use (hours/week) for leisure and domestic care by key individuals
• Household Dietary Diversity Score (WFP)
• Food Insecurity Experience Scale (FAO)
• Reduced Coping Strategies Index (WFP)
• Subjective well-being/life satisfaction
• Household borrowed money (binary)
• Total loan amount received
• Household has savings (binary)
• Total amount of savings
• Trust levels within and across communities
• Cooperation levels
• Women’s involvement in borrowing decisions (binary)
• Who decides how to use borrowed money (binary)
• Store credit limit (conditional on borrowing)

Randomization at the household level among registered households between the traditional cash for work only modality and the new pilot livelihoods modality.
Households were randomly assigned to one of two groups:
1. Training + Asset Transfer: Eligible for up to $500 in earnings in cash for work program, plus received training valued at $100 (primarily payments for attendance plus share of wages for trainer) and $100 towards the purchase of a productive asset
2. Extra Workdays: Eligible for up to $700 in earnings in cash for work program.

Not available

Randomization was stratified by a community-level poverty index derived from baseline data (value of livestock and fixed assets). Households below the median were classified as asset-poor, and those above as asset-rich. Randomization was implemented within each community and asset group using Stata.
Eligible households that changed their mind and dropped out after randomization will still be included in the study dataset. Households that were not originally in the sample but later chose to participate in the intervention will not be included.

Household

No

22 communities

2,550 households.

1243 Control households (Cash for Work with extra work days)
1307 Treated households (Cash for Work plus livelihoods training and asset)

Our primary variables of interest are the share of primary respondents (the household member likely to participate in LIWP component 2 if available) who are engaged in any productive work (either self-employment or paid work), getting paid in-cash or in-kind, being self-employed in a small business or agricultural activity, self-employment revenue at the household level (YER), profit from self-employment at the household level, and normalized indices of women’s decision-making. The table below shows the baseline summary statistics and minimum detectable effect for each of these outcome variables given the sample size of 2550 with household level randomization at statistical power of 0.8 and significance level of 0.05. Because we sample from all available communities in the population of interest for this analysis and sample a large proportion of available households in each community and randomized assignment is at the household level, there is little justification for clustering standard errors at the community level as this would substantially over-inflate the standard errors. Our primary specification will not include any clustering, but as a robustness check we will use two-stage bootstrapped standard errors following the recommendation in Abadie et al. (2023). Any productive work: baseline value 0.61 (0.49); MDE 0.05 Any self-employment: baseline value 0.42 (0.49); MDE 0.06 Self-employment revenue: baseline value 126,495 (479,141); MDE 532,000 YER Self-employment profit: 23,130 (315,201); MDE 350,000 YER Normalized index of women's decision-making: MD 0.11 std deviations The minimum detectable effect for self-employment revenue and profit is extremely high due to the highly skewed distribution and high numbers of zeroes so we are unlikely to find impacts using the basic OLS approach and plan on using either Lee bounds as recommended in Chen and Roth (2024) or binary value (any profit) to analyze these outcomes.