Box 1.

Study 1—A randomized control trial to evaluate the effect of labor market choices on poverty in rural households in Bangladesh

The goal of the study is to quantify the effect of an intervention to push poor households out of poverty through enabling the poorest women to take on the same work activities as better-off women in their villages (Bandiera et al. 2017). It is an example of causal reasoning in the context of statistical causal inference using an experimental approach and a difference-in-difference regression model. The authors first study household characteristics and the labor markets in villages in a region in Bangladesh. Using regression analysis, they show that there is a correlation between women’s type of labor activities and poverty, where poor women allocate most of their labor to low-return, casual jobs while richer women specialize in high-return livestock rearing. To assess the effect of changes in labor activities on poverty they conduct a randomized control trial (RCT), which builds on a manipulability account of causation. In the RCT, a cause, i.e., occupational choice of the ultra-poor, is manipulated through a treatment, i.e., a one-off provision of livestock assets and skills, to evaluate whether and to what extent the treatment changes the outcomes of interest, i.e., increases in earnings and diversification of the asset base which sets them on a sustained trajectory out of poverty. A positive effect of the treatment on the number of treated women switching to livestock rearing is thought to represent a mechanism to ensure a way out of poverty. Because the intervention provided both money and training it was, however, not possible to separate whether the cause for switching occupation was an increase in savings or skills In designing their study, the authors pay large attention to the process of randomizing the households selected for treatment and control. Randomization is important to control for any household differences that are not a result of the treatment. The quality of this randomization and the absence of spillover between treatment and control (Ferraro et al. 2019) are critical for the reliability of the causal inference, so researchers pay much attention to the design. The authors find evidence for their claim that reallocation of time from casual labor to livestock rearing leads to an increase in net annual earnings relative to control households by 21%. The account of causation and the theoretical commitments associated with its use in RCT direct the causal reasoning of this study (and others) to focus on properties of individuals that can be manipulated, such as access to skills or the means to purchase livestock. It thus neglects other structural influences on people’s (labor market) choices, such as gender and asset inequalities that influence and constrain agency in ways that perpetuate poverty (Akram-Lodhi 2020)