Factor Market Failures and the Adoption of Irrigation in Rwanda

The study investigates why productive technologies, specifically irrigation, are under-adopted in sub-Saharan African agriculture despite high potential returns. The central research question is whether failures in factor markets—especially land and labor—constrain the adoption of irrigation in Rwanda. The context is three government-built hillside irrigation schemes where irrigation can create an additional productive dry season and enable high-value horticulture. The authors outline three steps: (1) demonstrate that irrigation is productive but adoption is partial; (2) show this partial adoption is inefficient; and (3) identify labor market failures as a binding constraint. Using a spatial discontinuity in access to gravity-fed canals, they quantify returns to irrigation and examine how access on one plot affects production choices on other plots, interpreting cross-plot substitution as evidence of separation failures due to imperfect factor markets. The importance of labor market frictions is posited given the labor-intensive nature of dry-season horticulture and the critical role of the household’s shadow wage in profitability.

The paper situates its contribution within literatures on technology adoption constraints, misallocation, and agricultural household models. Prior work shows risk, credit, and information frictions limit adoption in developing country agriculture (e.g., De Janvry et al., 2017). Factor markets for land and labor are known to be imperfect, inducing misallocation and productivity losses (Udry, 1997; Fafchamps, 1993; LaFave & Thomas, 2016; Adamopoulos & Restuccia, 2014, 2018; Foster & Rosenzweig, 2017). Classical agricultural household models (Singh et al., 1986; Benjamin, 1992) predict separation failures when markets are incomplete, linking household characteristics to production choices. Evidence on irrigation returns largely comes from South Asia leveraging natural experiments (Duflo & Pande, 2007; Sekhri, 2014; Jacoby, 2017), with limited rigorous evidence from Africa (Dillon, 2011; Dillon & Fishman, 2019). Complementary literatures examine how property rights and land titling affect investment (Besley, 1995; Goldstein & Udry, 2008; Deininger & Feder, 2009; Ali et al., 2014; Goldstein et al., 2018). The authors contribute by linking land and labor market frictions to inefficient technology adoption on technologically identical plots, enriching empirical tests of separation and misallocation that infer inefficiencies from input and labor allocations (e.g., Jacoby, 1993; Skoufias, 1994; Restuccia & Santaeulalia-Llopis, 2017).

- Setting and intervention: Three hillside, gravity-fed irrigation schemes in Karongi and Nyanza districts (Rwanda) constructed 2009–2014; fully operational by 2015 Rainy 1. Irrigation water is delivered via main canals and secondary pipes/valves across terraced hillsides. The command area comprises plots below the canal within ~100m of valves. - Data: 969 cultivators and 1,753 plots in the discontinuity sample (overall 1,695 cultivators, 3,332 plots). Four years of panel plot-season data spanning 2014 Dry through 2018 (except 2015 Dry and 2016 Rainy 1). Detailed plot-level inputs, outputs, crop choices, irrigation use; GPS-mapped plots. Two focal plots per household tracked: the randomly sampled “sample plot” (near boundary) and the “most important plot” (MIP). Low attrition; tracked plots across transactions; some CA plots rented to commercial farmers (bias likely downward for impacts). - Identification: Spatial regression discontinuity (RD) at the command area boundary within 50 meters, exploiting the exogenous canal alignment determined by engineering slope constraints. Compare plots just inside (treated) to just outside (control). Account for potential biases from hillside position and plot size using: (i) distance-to-boundary slopes on each side; (ii) log plot area; (iii) spatial fixed effects (demeaning with nearest-neighbor sets) to remove spatially correlated unobservables and selection of non-agricultural terrain. - Estimation: OLS with site-by-season fixed effects; RD with distance interactions; spatial FE with Conley standard errors. Treatment-on-the-treated (ToT) scaling to interpret magnitudes for irrigated plots. Outcomes assessed separately for dry vs. rainy seasons. - Stylized facts informing interpretation: Irrigation is used almost exclusively for dry-season horticulture; horticulture is highly input- and labor-intensive; horticulture yields much higher cash profits; household labor dominates costs, so profitability hinges on the shadow wage. - Testing separation failures: Develop and apply an agricultural household model with uncertainty, labor, input, and insurance constraints. Model predicts under complete markets, irrigation access on plot 1 does not affect allocations on plot 2 (separation). With binding constraints, irrigation access on plot 1 increases input and labor demand there and induces substitution away from other plots (reduced irrigation, horticulture, labor, and inputs on the MIP). Empirical test: regress MIP outcomes on the sample plot’s CA status (and interactions by whether the MIP is in CA), controlling for distance to boundary (of sample plot), plot areas, and MIP CA status. - Heterogeneity to identify binding constraints: Interact the sample plot shock with household size (labor endowment) and asset wealth (credit/consumption smoothing). Model predicts: with insurance or input constraints, larger and wealthier households should be less responsive (less substitution). With labor market frictions and elasticities of on-farm labor supply varying with size/wealth, larger households substitute less while wealthier households substitute more. - Experiments: Three RCTs among CA farmers: (1) empower local monitors to support O&M and report maintenance; (2) randomized subsidies of land taxes intended for scheme O&M; (3) randomized minikit provision (seeds, fertilizer, insecticide) to test financial/information constraints and spillovers via village-level saturation (20/60/100%). Estimate ITT effects on minikit uptake and horticulture adoption with clustering at water user group level.

- Adoption patterns: CA access raises dry-season irrigation by about 16–20 percentage points (pp) and rainy-season irrigation by 4–6 pp. Dry-season shift into horticulture increases by roughly 14–18 pp and substitutes away from bananas by ~13–17 pp; no net change in overall cultivation rates in dry season. - Inputs and labor: In dry seasons, ToT estimates imply irrigation increases household labor by ~340–450 person-days/ha; hired labor expenditures by ~19,000–28,000 RwF/ha; and input expenditures by ~25,000–39,000 RwF/ha. Valuing household labor at a market wage of ~800 RwF/day implies labor costs dwarf purchased inputs and hired labor. - Yields, revenues, and cash profits: Dry-season ToT increases yields by ~300,000–450,000 RwF/ha (about 49–72% of annual agricultural production) and sales rise nearly proportionally (0.76–0.89 RwF per 1 RwF of yield increase). Yields net of purchased inputs and hired labor rise by ~250,000–390,000 RwF/ha (44–71%). When household labor is valued at market wage, net gains are substantially reduced, underscoring dependence on the shadow wage. - Partial and potentially unsustainable adoption: Only ~30% of plots are irrigated four years after canals became operational; low adoption threatens cost recovery for routine maintenance given current fee compliance. - Separation failures (inefficiency): Access to irrigation on the sample plot reduces irrigation, horticulture adoption, labor, and input use on the household’s MIP. For MIPs in the CA, a 16–20 pp increase in sample-plot irrigation coincides with an 8–10 pp decrease in MIP irrigation. MIP household labor and input use fall (e.g., −11 to −33 person-days/ha and −2,100 to −6,700 RwF/ha on average; larger magnitudes when MIP is in CA). Much of the MIP response reflects both intensive and extensive margins. - Magnitude of inefficiency: Back-of-the-envelope indicates that absent within-household substitution, overall irrigation adoption would be at least 21–24% higher. - Heterogeneity consistent with labor market frictions: Larger households substitute substantially less; two additional members reduce responsiveness by roughly 62–86% for labor/irrigation/horticulture. Wealthier households substitute more; a one standard deviation asset increase raises responsiveness by roughly 40–80%. This pattern is consistent with binding labor market constraints and not with pure credit/insurance constraints. - Experiments: O&M monitor empowerment and fee subsidies had no detectable effects on cultivation; compliance with fees was extremely low (~4%). Minikits had strong takeup among assignees (≈40 pp increase in use) but no impact on horticulture adoption; positive selection into usage by farmers already likely to grow horticulture. Together, RCTs suggest financial and informational frictions are not primary constraints in this setting.

The findings directly address the research question: despite high private returns, irrigation adoption is inefficiently low due to factor market failures, specifically labor market frictions compounded by imperfect land markets. The spatial RD demonstrates large dry-season productivity gains from irrigation via a shift into horticulture, but cross-plot substitution implies that constrained households reallocate labor and inputs away from other plots, reducing multi-plot adoption and overall utilization of the system. The heterogeneity analysis (larger households substituting less; wealthier substituting more) pinpoints labor market failures as the primary mechanism, inconsistent with credit or insurance frictions alone. Experimental null results for O&M and fee policies and for minikits (financial/information) reinforce this interpretation. These dynamics matter for scheme sustainability: with low adoption rates and substitution limiting multi-plot irrigation, maintenance cost recovery is jeopardized. The results contribute granular evidence linking misallocation mechanisms to technology adoption, showing that separation failures can differentially suppress adoption on technologically identical plots.

The paper shows that hillside irrigation in Rwanda is highly productive, enabling dry-season horticulture and increasing cash profits by roughly 44–71% net of purchased inputs and hired labor. Yet adoption remains limited due to labor market frictions and land market imperfections that induce cross-plot substitution, leading to inefficiently low take-up; eliminating substitution could raise adoption by at least 21%. Heterogeneous responses by household size and wealth, along with experimental null results targeting O&M and financial/information constraints, indicate labor market failures as the binding constraint. The study’s contributions include a credible natural experiment-based estimate of irrigation returns in sub-Saharan Africa, a novel test of separation failures via cross-plot responses, and integration of modeling, RD, and RCT evidence. Future research could explore interventions that relax labor constraints (e.g., facilitating seasonal labor markets, labor-saving irrigation technologies), investigate long-run market adjustments (wages, prices, land reallocation), and test institutional designs that improve multi-plot irrigation without inducing harmful substitution.

- External validity: Evidence from three sites in Rwanda may not generalize to other agro-ecologies or irrigation modalities. - Timing and dynamics: Analysis covers roughly four years post-implementation; longer-run adjustments in labor markets, land transactions, or crop markets may differ. - Measurement and design constraints: The baseline was not fully pre-treatment (some 2014 Dry access); although balance checks are favorable. Some CA plots were rented to commercial farmers, likely biasing impact estimates downward. General equilibrium effects (e.g., horticulture prices) are not identified within the RD; some suggestive price declines occur in one site. The valuation of household labor is uncertain (shadow vs. market wage). Experiments faced low compliance with fees and selection into minikit uptake. The separation test is joint with land market frictions and does not isolate all possible concurrent constraints, though heterogeneity patterns support labor frictions as primary.