Experimental Design
The experiment was conducted in urban or semi-urban settings near factories that employ a large number of women in Ethiopia. Interviews were timed such that interviews in half of these communities took place in the ten days leading up to the factory’s pay-day, and interviews in the other half of these communities took place in the ten days just after the factory’s pay-day. This was done to generate potential (non-experimental) background variation in cash-on-hand and other household conditions related to pay-day.
A sampling frame was constructed from the communities surrounding ten factories, via snowball sampling starting with women who are currently employed in factories. Inclusion criteria were being female; currently cohabiting with a male partner, who is currently present in the household or has been away temporarily for less than three months; not being within 2 months before or 6 months after giving birth; and belonging to one of four employment groups: i) those who have never had a formal job or owned a business; ii) those who are currently working in factories, iii) those who have left factory employment and are not working formally or owning a business, and iv) those who have left factory employment and are now engaged in non-factory formal employment or running a business. We did not screen ex ante for experiences of DV. The sample of 400 women was randomly drawn from this sampling frame, stratified on whether her community was just before or after pay-day and on employment group.
To ensure confidentiality, participants were interviewed one-on-one at a location pre-arranged with the respondent where she would feel safe, by a female from a similar background. During the interview itself, respondents were assigned a number and neither their name, the name of their partner nor any contact details were elicited or recorded.
If the respondent consented, she was first asked to respond to a survey covering: personal characteristics; relationship and husband’s/partner’s characteristics; intra-household decision-making and household public goods; household assets and expenditure; time use; and labor supply. The pre-conducted computer randomization then determined whether the respondent was primed about: (1) domestic violence using a well-established survey instrument (based on the DHS and WHO instruments); (2) income and expenditures (similar to Mani et al., 2013); or (3) positive experiences such as fashion and festivals. We then elicited incentivized measures of risk preferences, time preferences and cognitive function, as well as unincentivized measures of beliefs about employment, altruism, and self-efficacy. At the end of all tasks, the respondent drew a ball from a bag to select which of the incentivized tasks would be used to determine her actual monetary payment (as well as the timing of payment, and exertion of real effort, if the time preference task was drawn). Finally, women in the treatment arms, who had not received the violence prime, were also asked about their experiences of DV, such that data on DV was collected from all participants.