Mechanizing Together: Leveraging Local Governments for Rice Mechanization

The study has two interventions. In both interventions farmers will be able to rent drum seeders on a daily basis. In one arm the rental will be managed by the village government with its elected leader responsible for managing the rentals. In the other arm, agricultural extensionists employed by the state government not accountable to the village elected government will manage the rentals. An additional individually randomized intervention will be an unconditional 250 INR payment to large farmers in a village with an additional 1000 INR/ acre of drum seeded rice to be paid once the planting is verified by enumerators.

2022-12-01

2024-01-04

The uptake primary outcome is whether a farmer used a drum seeder on any of their land. This is the overall primary outcome of the study for which I expect to be most powered to detect significant treatment effects.
The primary profit outcome is reduction in cultivation costs.
The primary extensionist effort outcome is whether the farmer ever reported receiving information from the extensionist on drum seeders.
The primary labor market outcome will be wages for workers on the gram panchayats NREGA rolls who worked as transplanters during either the past winter or summer seasons. The working data will be collected multiple times throughout the season and treatment and control gram panchayats will be matched based on nearest neighbor, based on smallest Kolmogorov distance of the distribution of planting dates. I plan on obtaining the full distribution of planting dates using remote sensing.
The primary political economy outcome will be the population level support for the sarpanch with the NREGA and farmer frames re-weighted using dual frame weights to reflect the overall population.

To compare the sarpanch managed and extensionist managed rental arms the total amount of rental income which is a function of the days the drum seeders were rented serves as a secondary uptake outcome. Acreage under drum seeding and profitability are all farmer level secondary outcomes. We will also try to collect cultivation costs and profitability at a plot level. Another secondary outcome is the wages for male and female workers with active NREGA job cards who had not worked as transplanters. The industry of these NREGA workers is a further secondary outcome.

Village leaders, sarpanches, were asked to apply to be considered for the rental program. Among those that applied, one village was selected per cluster. These villages were randomized into control, a sarpanch rental arm, and an agricultural extensionist rental arm. There is also a second level of individual randomization.

At the time of the initial pre-registration, the intervention has begun but only limited administrative monitoring data on self-reported rentals had been collected for the initial winter 2022-2023 season. The survey which will be the primary data source for the primary outcomes had yet to take place. Similarly, the reconciliation of village bank accounts to verify rental income has yet to take place. Due to immense delays in implementation and receiving the necessary data for sampling from the government, the original winter season will not be used as the main season for analysis. Instead, I plan on using the latest season where I am able to collect data as the main season for analysis. Depending on funding and state elections that will either be the summer 2023 season or the winter 2023-2024 season. An alternative specification will be to use the sum of treatment effects across program seasons since the sum of effects across the two main agronomic seasons is also policy-relevant.

We will sample from all farmers who cultivated rice in the past winter and summer season seasons in the gram panchayat using the administrative crop booking database. We will sample 25% of farmers from those with less than half a hectare and the remainder from those with more than half a hectare so that we can focus on farmers whose uptake will meaningfully effect the acreage under drum seeding and thus labor markets, carbon emissions, and water consumption. The half-hectare cut-off was chosen to match how landholding is reported in the agricultural census. We will further sample a particular plot within a farmer's lands in the event some of the land is cultivated by someone other than the owner we will interview the cultivator of the sampled plot. While this is not a fully representative sample of all cultivators it is the best possible sample that can be constructed using administrative data.

For the laborer sample we will take a simple random sample from holders of active NREGA job cards in treatment and control villages.

The number of farmers sampled in each village is not uniform. Some districts and gram panchayats had very low numbers of rentals in the monitoring data. Thus it was not a prudent use of survey resources to extensively sample those strata. The number of rentals recorded in the administrative monitoring data was predicted using administrative data on historical drum seeder usage, a stratification variable, and district, also a stratification variable. Then the randomization strata were divided into quintiles of predicted uptake. In the lowest uptake quintile, "extra" control villages were dropped if a strata had multiple control villages. In these villages, 4 farmers are sampled per village. In the next quintile, no villages were dropped, but still only 4 farmers were sampled. In the 40th to 60th percentile quintile, 8 farmers were sampled and in the 60th to 80th percentile quintile and final quintiles 12 farmers are sampled per village. For the main analysis, I will focus on the third through 5th quintiles, e.g. the strata in the 40th to 100th percentiles of predicted uptake. However, I will also report reweighted results representative of the full sample of GPs.

Randomization was done on a computer using Stata 17. The randomization was stratified by district, historical use of drum seeders, and wages for transplanting labor.

The individual randomization is stratified by any past direct seeding, extent of paddy cultivation, and gram panchayat. We will only randomize in GPs with more than 5 rentals in either the past winter or the current kharif season. Gram Panchayats with more than 20 rentals in either season will also be excluded. GPs whose reported nursery planting time was after the date of the initial randomization were also included.

The administrative name for the unit of randomization is confusingly "cluster." The agriculture department defines the group of 3-5 gram panchayats, village level administrative unit, in which agricultural extensionists work as a "cluster". One gram panchayat was selected from each cluster and then randomization was done at the cluster level to ensure each agricultural extensionist only handled one treatment gram panchayat of either type.

A second level of individual randomization will take place at the farmer level. This randomization will take place among farmers with more than 2 acres of historical paddy cultivation who picked up and fully listened to an IVRS call describing the benefits of drum seeding. This filter is necessary since only a subset of the phone numbers in the government crop booking database are valid. The individual randomization was clustered by village and experience using a drum seeder. Those that had used a drum seeder themselves or seen a drum seeded field in the village were treated as the "experienced" group all others were treated as "inexperienced" for the purpose of randomization.

Yes

361

2,888, 172 individually randomized

100 clusters sarpanch rents, 100 clusters AEO rents, and 161 clusters control. 86 treatment and 86 control individual farmers.

Between the two 100 cluster arms the MDE at 80% power is .18 SD assuming an ICC of .1 with observations per cluster of 8. Between the 200 cluster treatment arms and the 161 cluster control arm the MDE is .16 SD.