Minimum detectable effect size for main outcomes (accounting for sample
design and clustering)
For administrative data outcomes, part of the driver-level panel of active ours, earnings and trips completed, we find that, on average, we detect a 9-14% change in the outcome relative to the control mean. This is equivalent to approximately 0.16 standard deviations. For the survey outcomes, the minimum detectable effect varies across outcomes. For the driver's overall monthly expenditure (including household expenditures, outstanding loans, school fees, and remittances), we detect an effect of approximately 0.72 standard deviations. For the proportion of time spent on the platform, we find a minimum detectable of less than 13% of the control mean or less than 0.408 units of standard deviation (meaning that we can detect less than a 10% increase in the proportion of time spent on the platform). For the binary outcome of whether the driver self-reports that they can meet their household expenditure, 80% power is achieved at a control mean of 18% (approximately 0.44 standard deviations). For the total hours spent on income-generating activities, we detect a 33\% change relative to the control mean, or 0.45 standard deviations. Finally, for the outcome that the days of the week to work are most important when working on the platform, we find that 80% power is achieved with 61% control of the mean. As expected, the survey outcomes have a higher minimum detectable effect than the administrative outcomes. We lose power when collapsing the data into an ANCOVA specification, thereby reducing variation over time. This is compounded by the fact that survey outcomes tend to exhibit greater noise.