**Background:** Approximately 75% of sub-Saharan Africa's population depends on agriculture. The diffusion of modern seed varieties has been associated with large economic and health benefits, but its environmental impact—particularly on deforestation—is theoretically ambiguous. The Borlaug hypothesis suggests yield-enhancing technologies reduce pressure on forests by intensifying production on existing farmland, while Jevons' paradox predicts that increased profitability of agriculture may instead drive land expansion. This question is critical for the Congo Basin, the world's second-largest tropical forest, where 90% of deforestation is linked to smallholder agriculture. Prior evidence has lacked credible causal estimates, especially for the Congo Basin.
**Methods:** The study used a randomized controlled trial (RCT) embedded within the PARRSA program in the former Equateur province of DRC. Ninety-two villages were stratified by population size, remoteness, and extension exposure; 60 were randomly assigned to receive a seed subsidy intervention and 32 served as controls. Within treatment villages, households were randomly selected via lottery to receive vouchers offering 30%, 60%, 90%, or 100% subsidy for improved seeds of rice, maize, groundnuts, or cassava. In 35 of the 60 treatment villages, seeds were delivered by truck to overcome transportation barriers. The proportion of households receiving vouchers also varied across villages (20%, 45%, or 70%). Deforestation was measured using: (1) remote sensing data from Hansen et al. (30m resolution, 90% canopy cover threshold) for village-level estimates; (2) household survey data (n=904 households) providing plot-level classification of primary vs. secondary forest conversion; and (3) household labor allocation data. Primary forest was defined as forest with no observable traces of previous cultivation; secondary forest as regrowth on previously cultivated land. Intent-to-treat (ITT) effects were estimated using OLS regressions with village-clustered standard errors and stratification fixed effects.
**Key Results:** In control villages, only half of cultivated area in 2014 was under cultivation the previous year; of new area, 12% came from primary forest, 72% from secondary forest, and 16% from savanna/abandoned plantations. Village-level remote sensing estimates showed no statistically significant overall deforestation effect: for cumulative deforestation 2013-2016, the coefficient for lottery without truck was -0.006 (SE=0.182, p=0.972) and for lottery with truck was -0.150 (SE=0.166, p=0.368); joint F-test p=0.60. For 2014 alone, the coefficient for lottery without truck was -0.092 (SE=0.094, p=0.337) and for lottery with truck was -0.144 (SE=0.086, p=0.099); joint F-test p=0.13. However, household-level data revealed significant shifts in forest type. Deforestation of primary forest increased by 0.108 ha per household in villages without truck delivery (SE=0.038, p=0.006) and by 0.06 ha in villages with truck delivery (SE=0.031, p=0.052); joint F-test p=0.00. This represents a more than 100% increase relative to the control mean of 0.1 ha. High-subsidy households in non-truck villages deforested an extra 0.126 ha of primary forest (SE=0.049, p=0.012), while low-subsidy households cleared an extra 0.086 ha (SE=0.039, p=0.028). These increases were partly offset by reductions in secondary forest conversion (e.g., -0.110 ha for high-subsidy non-truck households, SE=0.071, p=0.124). Total deforestation remained unchanged (joint F-test p=0.54). Labor data showed increased household members working on-farm (joint F-test p=0.00) and increased person-days in labor-sharing arrangements for land preparation (joint F-test p=0.06), consistent with reallocation of labor toward primary forest clearing.
**Clinical Implications:** This study provides the first rigorous causal evidence on the link between modern seed variety promotion and deforestation in the Congo Basin. The findings demonstrate that agricultural intensification without complementary soil fertility management can accelerate primary forest loss—a critical outcome for biodiversity, as primary forests support at least two-thirds of the world's biodiversity and provide irreplaceable habitat. The distinction between primary and secondary forest is essential: while overall deforestation was unchanged, the shift toward primary forest represents irreversible biodiversity loss. The results suggest that promoting legume seeds (groundnuts) is associated with less deforestation than promoting cereals, likely due to lower nitrogen requirements. For policymakers, these findings imply that seed subsidy programs in remote, fertilizer-constrained environments should either be combined with soil fertility management practices (organic fertilizer, conservation agriculture) or focus on crops less demanding of nitrogen-rich soils to avoid unintended environmental harm. The study also highlights that labor constraints may limit overall deforestation responses, but do not prevent ecologically damaging shifts in the type of forest cleared.