The Causal Effects of Income Instability

We will experimentally manipulate income instability by generating fluctuations in the labor demand for casual work in rural Ghana.

We will experimentally manipulate income instability by generating fluctuations in the labor demand for casual work in rural Ghana. Participants will be randomly assigned to one of four study arms. A control arm will not receive any intervention; they will only be surveyed. The three treated arms will differ in whether their workload and pay vary over time and whether their workload and pay vary predictably or unpredictably.

The three treated arms will be offered employment for 12 weeks. These 12 weeks will be divided into 2-week blocks—we will refer to these blocks as periods.

- The workload and pay of the Stable Income arm will not vary over time. They will have the same workload and the same pay in every period.
- The workload and pay of the Predictable Volatility arm will vary over time in a predictable manner. In half of the periods, they will have a lower workload and earn less. In the other periods, they will have a higher workload and earn more. Crucially, these participants will know at the beginning of the study how much they will work and how much they will be paid in each period.
- The workload and pay of the Risky Income arm will vary unpredictably over time. In each period, they will have a 50% chance of having a lower workload and earning less and a 50% chance of having a higher workload and earning more.

The design varies the second moment while holding the first moment constant. In particular, Stable Income and Predictable Volatility participants will have on average the same workload per period and earn the same amount on average per period. Similarly, Risky Income participants will have in expectation the same workload per period as the Stable Income and Predictable Volatility participants and will earn in expectation the same amount per period as the Stable Income and Predictable Volatility participants.

In addition to comparing study arms, we will compare (i) Risky Income arm participants who by chance end up with exactly 3 periods with high workload to other study arms; (ii) participants within the Risky Income arm who by chance end up with different income realizations; and (iii) periods with high workload to periods with low workload.

2025-05-22

2025-08-20

Consumption & expenditures; food insecurity and dietary diversity; and mental health.

Savings & Assets; Loans & Debt; Transfers & Gifts; Non-Study Employment & Non-Study Income; Subjective Beliefs about Earnings; and Time Use.

Subjective beliefs about earnings will only be elicited from the Risky Income arm.

We will recruit multiple participants per community (one per household). We will randomize at the individual level, stratifying by community. Participants within a community will be randomly assigned to one of the four study arms.

Randomization to arm is done in office by a computer. For the participants in the Risky Income arm, their workload in a given period is determined by scratch cards in the field. For participants in the Predictable Volatility arm, there are 20 different possible permutations of when the 3 high-workload periods and the 3 low-workload periods may happen. Participants in this arm will be randomly assigned to one of them in advance by a computer.

The unit of randomization is an individual.

No

N/A

2,267 study participants.

400 control, 400 stable income, 400 predictable volatility, and 1,067 risky income.

We consider two-tailed tests with a 0.05 significance level and 80% power. The minimum detectable effect (MDE) is 0.16 standard deviations (SD) for comparisons between the Risky Income arm (approximately 1,067 participants) and any one of the other three arms (each with approximately 400 participants). For comparisons between any two of the smaller arms—Control, Stable Income, and Predictable Volatility (each with about 400 participants)—the MDE is 0.2 SD. We expect that 31.25% of participants in the Risky Income arm (i.e., 333 participants) will, by chance, experience exactly 3 out of 6 periods with a high workload. For comparisons between this subset and any one of the other three arms, the MDE is 0.21 SD. These are conservative estimates, as they do not account for the use of baseline outcome measures, which should reduce MDEs.