This dataset provides de-identified panel data from a long-term randomized controlled trial examining the causal impact of access to factory employment on labour market outcomes, wellbeing, and fertility among women in Ethiopia (IZA/FCDO G²LM|LIC Project GA-6-775).
The study follows 1,464 married women who applied for jobs in 27 garment and shoe factories across five industrialising regions. Applicants were randomly assigned to either receive a job offer (treatment group) or not (control group), enabling causal identification of the effects of formal employment.
Data collection spans multiple survey waves over approximately nine years and includes detailed information on labour market participation, earnings, income and savings, physical and mental health, life satisfaction, empowerment, and complete fertility histories.
Short- to medium-term findings show that access to factory jobs increased formal employment and earnings. However, long-run results reveal a more complex dynamic: while fertility did not differ in the early years, women who received job offers ultimately had higher fertility, including lower rates of childlessness. This pattern suggests that women postponed childbearing while working under time constraints but later increased fertility once income and savings had accumulated.
The dataset enables analysis of both short- and long-run treatment effects and provides rich information to study mechanisms linking employment and fertility, including income effects, time constraints, childcare arrangements, and intra-household decision-making.
The data contribute to key debates in development economics and demographic research, particularly regarding the relationship between female employment and fertility in low-income settings. The findings challenge the conventional assumption that increased female labour force participation necessarily leads to fertility decline.
All personal identifiers have been removed. Users must not attempt re-identification.
Sampling and randomization
- Universe. Married/partnered women who applied for factory jobs and passed eligibility screening at one of 27 study factories. Recruitment was done in batches across multiple hiring rounds (2015–2016).
- Sample size. 1,464 randomly assigned women at baseline. Wave-7 endline interviewed 1,130 women (77% retention).
- Randomization. Within each hiring batch (a “block”), women were randomly assigned to treatment (job offer) or control (no offer). The randomization-block ID is in the variable block (opaque integer; do not interpret block IDs).
- Pre-registration (long-run follow-up). AEA RCT registry: AEARCTR-0014209.
Dataset overview Study type: Randomized controlled trial (individual-level randomisation)
Setting: 27 garment and shoe factories across five industrialising regions in Ethiopia
Intervention: Job offers for low-skilled factory employment
Randomisation: Treatment (job offer) vs control (no job offer) among applicants
Timeline: Baseline (2016–2018), multiple follow-up waves up to ~9 years
Sample: 1,464 married female job applicants
Key outcomes:
- Labor market participation and earnings
- Wellbeing (incl. physical & mental health)
- Fertility outcomes (births, timing, childlessness)
- Household income and savings
Data components:
- Baseline survey
- Multiple follow-up waves (panel data)
- Fertility histories
- Household and partner-level variables
Anonymization The released file is built from the source endline file by removing every variable that is not strictly required for analysis or that could re-identify a respondent. The following are dropped: respondent name and identifiers, GPS coordinates, region/kebele/factory names, all interview and submission timestamps, household roster (member-level), free-text comment fields, free-text asset descriptions, exact day and month of birth events (year is retained), and all baseline (Wave-1) variables. The respondent identifier N_Id is replaced with a random permutation id_anon; the crosswalk is held privately by the project team. A k-anonymity audit (k = 5) is run on quasi-identifier combinations. Analysis variables are kept at native precision; the only coarsening applied is collapsing very small sector categories into “Other”.
