Wage transparency and farm employment—Experimental evidence from Nigeria

Farmers and workers in villages assigned to receive information treatment (treatment group) will receive information on the average maximum wage farmers in their area are willing to offer. Information will be assigned to experiment participants in the third week of October 2024. To design the information treatment, we will use wage information collected in the survey round just before the intervention, which is in September 2024. The content of the information treatment will summarize wage-specific information from the sub-sample of farms in villages close to each other (known as catchment areas in the study). For instance:

1. “Thank you for participating in the survey. We wish to share wage information with you that you might find interesting. Two weeks ago, we asked N tomato farmers in your village and surrounding villages about the highest wage they would be willing to pay a casual worker. They said X Naira per day.”

The information treatment will be administered through phone calls to the experiment participants and followed up with a text message (short message service). We will also convey the information treatment in local languages (e.g., English, Pidgin-English, and Yoruba) to ensure the farmers and workers, some of whom are less educated, fully understand it.

2024-10-13

2024-10-19

1. Wages paid by farmers
2. Maximum wage farmers would be willing to offer
3. Wages received by workers
4. Reservation wages (workers)
5. Days worked (workers)
6. Unfilled labor needs (farmers)

6. Unfilled labor needs (calculated by subtracting the total number of casual workers hired over the past seven days from the total number of casual workers needed on the farm in the past seven days) measured as a count of workers

At the farm level, the total number of workers needed on the farm in the past seven days and the perceived community-level wage (Naira/day) over the past seven days.

At the worker level, number of days worked in self-employment and on one's own or household farm during the past seven days, and the perceived community-level wage (Naira/day).

The planned study is a cluster randomized control experiment with one information treatment. Participants will be randomly assigned to the treatment arm at the village level instead of opting for randomization at the individual (farmer or worker) level. Although randomizing at the individual level would increase statistical power, it was ruled out due to concerns about treatment spillover—i.e., externalities on untreated participants. For instance, if multiple farms operate in the same community, treating only one may lead to treatment spillover among untreated participants. This could result in biased treatment effect estimates when comparing different experiment arms. While randomizing at the village level reduces the risk of treatment spillover to some extent, it still poses a treatment spillover risk since some villages in our study are in close proximity. Therefore, the random assignment procedures for the allocation of villages to treatment arms will follow a two-step randomized design. In the first step, each of the 5 LGAs is randomly assigned a proportion p (0%, 50%, or 100%) of villages to receive information treatment. In the second step, within each LGA, a fraction P of all villages will be randomly selected into treatment arms. The two-step randomized design also allows us to account for possible spillovers by comparing LGAS where some villages are treated (contaminated group) to those LGAs with no treated villages (pure control groups) or those in villages where everyone is treated (pure treatment group).

Randomization done in office by a computer using STATA

Village level

Yes

88 villages

247 Farmers and 280 Casual (daily) Workers

45 villages in the control group (127 Farmers and 142 Workers)
43 villages in the treatment group (120 Farmers and 138 Workers)

Sample size calculations for a cluster randomized control trial (RCT) with group sizes equivalent to 3 experiment participants per village and 41 villages per treatment arm (one control and one treatment group), as well as an intra-cluster correlation of 0.1, indicate that we should be able to detect a difference in the natural logarithm of wages between pooled treatment and control groups of around 27.8% with a power of around 80% at the 5% significance level. Similarly, based on sample size calculations, it was determined that a sample size of 3 farmers per village and 41 villages per treatment arm (one control and one treatment group), as well as an intra-cluster correlation of 0.1 and power of around 80% at the 5% significance level, would provide sufficient statistical power to detect a change in labor shortages of 0.37 units.