The Gendered Impacts of Employment

Our experiment implements three interventions. Relative to a control arm, in which individuals receive a nominal fee for weekly survey participation, we design an employment arm, in which we offer gainful employment in the form of surveying tasks for an average of four days per week for six weeks; a cash arm, in which we offer no work but the equivalent sum of money in the form of weekly cash transfers; and a volunteer arm, in which we offer an equivalent but unpaid work opportunity and a nominal fee for weekly survey participation. A comparison of the work arm to the control arm yields the value of employment in its entirety. A comparison of the work arm to the cash arm yields a causal estimate of the non-pecuniary value of employment. A comparison of the work arm to the volunteer arm enables a distinction between the psychosocial value of fulfilling the role of the ‘breadwinner’ of a family (working and earning for work) to that of the experience of working without earning. We enroll 2520 married couples into the experiment. While we offer a randomized treatment to only one member (male or female) of each couple, we survey both members, allowing us to document spillovers of each intervention on the spouses of treated individuals.

2023-02-01

2023-06-01

We collect measures of psychosocial wellbeing, which our PAP describes in more detail. We organize the underlying channels navigating this relationship into three broad buckets: how work shapes the perception of self, household interactions, and relationships with the outside world. Our PAP outlines how each of these outcomes is measured.

Our sample households spanned 280 sub-blocks, with each sub-block randomly assigned to one of the three experimental arms (80 blocks each) and a control group (40 blocks). Each household we selected was randomized into having the husband or the wife receive the respective intervention. Table \ref{baltable} reports summary statistics and balance across the arms. All households were informed that the study would last six weeks, with surveyors returning weekly to a pre-assigned meeting point to conduct five-minute surveys and provide compensation. In the 40 control sub-blocks, the selected wife or husband of the couple was informed that s(he) would receive 50 taka (USD \$0.50) as compensation for the weekly surveys. In the 80 cash sub-blocks, this individual was informed that s(he) would receive 1200 taka (USD \$12) as compensation for the weekly surveys. In the 80 work sub-blocks, the individual was offered an opportunity for gainful employment in which s(he) could work for an average of four days per week (either three or five) over the six weeks and be compensated 300 taka (USD \$3) per day of work, resulting in an average of 1200 taka (USD \$12) weekly; these individuals would also be asked to complete the weekly surveys. Lastly, in the 80 volunteer sub-blocks, the individual was offered an opportunity to engage in the same activity as those in the work arm, but for no pay, beyond the 50 taka (USD \$0.50) received as compensation for completion of the weekly surveys. The randomized nature of treatment allocation was made public to all participants, with surveyors displaying the participant's randomized treatment status on their tablet screen.

randomization done in office by a computer

We randomize at the sub-block level within the refugee camp

Yes

The research team enlisted 2520 households across 10 camps (2E, 2W, 3, 4, 5, 7, 10, 17, 18, 19), with each camp apportioned into 4 to 7 blocks, and 14 to 42 sub-blocks, with the latter as our unit of randomization. Our sample households spanned 280 sub-blocks, with each sub-block randomly assigned to one of the three experimental arms (80 blocks each) and a control group (40 blocks). We selected nine households per sub-block. These households were recruited first through a random walk, in which the field team began near the center of a given sub-block, moved in a randomly selected direction, and proceeded to knock on doors and inquire about interest in participating in our study. Each household we selected was randomized into having the husband or the wife receive the respective intervention.

2520 households

The research team enlisted 2520 households across 10 camps (2E, 2W, 3, 4, 5, 7, 10, 17, 18, 19), with each camp apportioned into 4 to 7 blocks, and 14 to 42 sub-blocks, with the latter as our unit of randomization. Our sample households spanned 280 sub-blocks, with each sub-block randomly assigned to one of the three experimental arms (80 blocks each) and a control group (40 blocks). We selected nine households per sub-block. These households were recruited first through a random walk, in which the field team began near the center of a given sub-block, moved in a randomly selected direction, and proceeded to knock on doors and inquire about interest in participating in our study. Each household we selected was randomized into having the husband or the wife receive the respective intervention.

For the full main treatment arm comparisons, the experiment is well powered to detect 0.15 standard deviation effect sizes at low ICC. At higher ICC levels, minimum detectable effect sizes range from 0.2-0.25 standard deviations. Observed ICCs at baseline tend towards the low end of the examined range while standardized effect sizes in Hussam et al. (2022) are typically between 0.2-0.25, suggesting that our primary specifications are well-powered.