Experimental Design Details
Following VESA level livelihoods and gender training provided to both women and their male partners, we conduct a randomized control trial (RCT) using participants of the LfR and SPIR programs. We randomly assign the men either to the treatment group or the placebo group. Subjects in the treatment group receive phone-based reinforcement of training messages designed to encourage them to participate in household chores and child care and those in the placebo group receive similar phone calls addressing an unrelated placebo topic. The treatment spans three months, with six phone calls placed on a biweekly schedule. During each call, a well-trained enumerator administers a carefully designed 5 minutes script in which he discusses gender norms with the male respondent. For the control group, an enumerator places calls at the same frequency but instead discusses a placebo topic – the man’s food consumption in the last 24 hours.
After the treatment is over, two weeks following the final reinforcement phone call, information on the outcome variables (i.e., male’s participation in household chores and child care) are collected from the targeted women through phone calls. We ask the women if the following activities were performed in service of the family yesterday: collecting fuelwood, fetching water, cooking, child care giving, processing grains, cleaning the house and washing dishes. We then ask who performed the activities and how many minutes were spent on these activities by the person who performed the activities, including the male spouse. These questions were also asked on the baseline survey. We will evaluate whether men in the treatment group participate in more household chores and child care than men in the control group at the end of the intervention period.
One possible concern in this experimental design is that the treatment (phone calls targeting specific males within a community) could influence the behavior of men in the same community who have been assigned to the control group. Randomly assigning communities to a treatment group (compared to individuals) would minimize this spillover problem, but also requires us to have a much larger sample that amplifies treatment and survey cost. After carefully considering both levels of treatment assignment, we decided to assign treatment at the individual level. In our decision, we considered the nature of the treatment, the outcome variables of interest and the direction of spillover bias. First, the treatment is complex enough that is not easily transferable to households in the control group. We administer six calls and discuss complex gender norms behavior that is very difficult for households in the control group to cope up with just simply through spillovers. For example, as part of the treatment, we ask the respondent to commit to fulfilling an assignment which requires him to do one of the household chores or child caring activity by the end of each call. Provided the respondent did the assignment, we allow him to slowly grow accustomed to the activities. For spillovers to exist, households in the control group would need to continue cope up with such activities for three months which is less likely. Second, our outcome variables consist of public and private household chores. For example, collecting fuelwood is public while bathing babies is private in the sense that it can be conducted without others in the neighborhood watching. We expect spillovers to be minimal on private household chores and we will separately analyze differences in the impact on public and private household chores to trace spillovers, if any. Third, spillovers will not make us overinflate our impact estimates because the spillovers would bias our treatment effect towards zero.
Another possible concern is related to mobile phone ownership. The experiment requires mobile phone ownership, but respondents may or may not own a mobile phone. Sampling only those who own mobile phones would introduce selection bias. For the same reason, the “Her Time” research project distributes free mobile phones to all respondents regardless of treatments status. Although mobile ownership has a potential effect on outcome variables, this effect will occur in all treatment groups. We will interpret our results accordingly, providing appropriate caveat about their external validity, particularly that the results may not be generalizable to poor households that do not participate in livelihoods groups and do not own mobile phones.
This experimental design allows the research team to carefully and credibly measure the effect of using phone calls to reinforce complex behavior change communications.