This is our fourth blog post for our Job Market Paper Series blog for 2024-2025.
Tereza Varejkova is a PhD candidate in Economics at the University of Maryland. Her research interests lie in development and environmental economics. You can find her JMP here.
As climate change makes rainfall patterns more variable, large irrigation projects can serve as an important climate adaptation strategy. However, constructing large dams requires substantial public funding and is often controversial due to the unequal distribution of benefits and costs. While previous research (Duflo and Pande, 2007) has focused on how geographic proximity to dams affects who benefits from irrigation, it is equally important to examine whether the benefits of irrigation are shared equitably between farmers with different scales of operation.
In my job market paper, I study the distributional impacts of irrigation canals on crop yields and land use patterns by type of farmer. I address this question in South Africa, which offers an ideal context for several reasons. First, the agricultural sector in South Africa has a dual structure. On one hand, there is a very well-developed commercial sector with over 40,000 farms. On the other hand, there are approximately 2 million subsistence farmers who grow crops at a small scale for own consumption. These farmers belong to the poorest segment of the population. Second, South Africa has a pattern of spatial exclusion. During apartheid, black population was relocated to special designated areas called the homelands. The land in these areas was mostly of poor quality or quickly degraded due to overgrazing and overcultivation following the quick increase in population density (Butler et al., 1978). In my study sample, subsistence farmers are located almost exclusively in these former homelands, which nowadays still lag behind in terms of different economic indicators (Lochmann, 2022). Finally, South Africa is a dry country, and irrigation has been necessary to sustain livelihoods since the colonial times. For this reason, the system of irrigation dams and canals is well-developed relative to other countries in sub-Saharan Africa.
Identification
To estimate causal effects, I leverage the fact that by gravity water in irrigation canals flows naturally downhill in a spatial regression discontinuity (RD) design with relative elevation to the nearest canal as the running variable, a strategy previously used for India (Asher et. al., 2022). Areas below canals are considered as treated and areas above as control. To make the treated and control areas even more comparable, I restrict the sample to places that are within 10 kilometers of the nearest canal and within 50 meters of relative elevation (Figure 1). There are over 140 canals in my sample.
The RD design is motivated by the fact that at the zero relative elevation threshold, the probability of land being irrigated jumps discontinuously, although not from zero to one. Control areas located above canals have a positive probability of irrigation because water can be pumped uphill. However, as the vertical distance up to which the water must be pumped increases, so does the cost, which in turn reduces the likelihood of irrigation. In contrast, treated areas below the canals eliminate this pumping cost, resulting in a discrete jump in irrigation probability. This discontinuity in the cost of irrigation creates the identifying variation used in my analysis.
Data
This identification strategy requires very granular data. To eliminate measurement error, the unit of analysis should ideally be much smaller than a district or even a village. In this project, I use remote sensing data which allows me to measure outcomes at a level of 30 meters by 30 meters grid cells. I use three main data sources:
- Enhanced Vegetation Index is used to proxy for crop yields. The index measures the greenness of a pixel and therefore captures the size of the above-ground biomass, which is directly related to yields of annual crops.
- South African National Land Cover is used to determine which grid cells are agricultural in order to eliminate noise from non-crop vegetation in the estimation of the yield effects. It also allows me to estimate effects on land use. Crucially, the data set distinguishes between commercial and subsistence land. Subsistence farmers are mostly located in the former homelands and commercial farmers in non-homelands.
- 2011 Census at the level of small areas (approximately 500 people/unit) is used to study the effects on employment.
Commercial farmers increase yields and cultivated area…
For commercial farmers in non-homelands, I find clear productivity gains from irrigation. In areas below canals, wheat yields increase by 4% and maize yields increase by 2%. The effects are intent-to-treat because of the fuzzy RD design. Moreover, between 1990 and 2020, treated farmers converted 30% more land for agricultural purposes than control farmers. The expansion of commercial farms contributed to local economic development through job creation. Using census data, I find a 5-percentage point reduction in unemployment, which represents a 20% drop compared to the control areas.
… while subsistence farmers do not benefit
Subsistence farmers in the former homelands located below canals see no benefit from irrigation canals; in fact, their maize yields decreased by 3%, and there is no evidence of cultivated land expansion. The absence of positive effects for these farmers suggests that either the irrigation infrastructure in the former homelands is poorly maintained and unable to supply water effectively, or that these farmers face constraints—such as insufficient resources to build necessary field-level irrigation infrastructure, limited access to complementary inputs like fertilizer, or labor market frictions (Jones et. al., 2022)—that prevent them from adopting irrigation. The inability to pinpoint the mechanism behind these results is a limitation of my study and an area for future research.
Policy implications
Policymakers are interested in whether large irrigation projects represent a worthwhile investment. To address this question, I conduct a cost-benefit analysis for a hypothetical project, comparing the net present value of benefits with the net present value of costs over the project’s lifetime, with both values discounted at a rate of 3%. For cost estimates, I use data from a feasibility study assessing the cost-effectiveness (measured as cost per cubic meter of water supplied) of proposed solutions to water shortages in the Limpopo province. The benefits are quantified as the increase in value of agricultural production, as implied by my RD estimates. Overall, I find that large irrigation infrastructure is a cost-effective investment despite the unequal distribution of its benefits. I also assess the sensitivity of this analysis to a scenario in which the dam only benefits existing agricultural land without any expansion of cultivated land. This might occur if conservation efforts prevent new land from being converted to cropland. In this scenario, the benefit-cost ratio falls below one and the investment would not recover its costs.
The main takeaway from my paper is that large public infrastructure investments in developing countries can yield unequal benefits across different segments of the population. I find that while large-scale irrigation infrastructure can boost agricultural productivity and is a cost-effective investment, its benefits are limited to commercial farmers; small-scale subsistence farmers do not experience the same gains. As a result, the infrastructure does not help to reduce existing place-based inequalities in South Africa. This finding has important implications for policymakers, who should consider compensating farmers who do not benefit from these projects and ensure that new irrigation initiatives support the expansion of cultivable land to fully justify the investment.
Feature image is the Vanderkloof Dam in South Africa. Source: Facebook post of Storm Report SA from January 25, 2021 here.
