Can Digital Technology Improve the Resilience of Grain Production: Causal Inference Based on Dual Machine Learning

Leveraging digital technology to enhance grain production resilience is pivotal for ensuring food security and constructing a modernized agricultural support system. The purposes of this study are to: (1) Define the connotation of grain production resilience and construct an evaluation index system across three dimensions: risk resistance, regulatory adaptability, and transformative innovation; (2) Identify the impact and mechanisms of digital technology as a new production factor on grain production resilience; (3) Reveal how digital technology influences resilience through the reorganization of "People-Land-Service" factors; (4) Analyze the heterogeneity of these effects across diverse geographical environments and economic endowments.

Based on panel data from 30 Chinese provinces spanning 2011 to 2023, a rigorous empirical analysis was conducted. First, the entropy method was employed to measure the levels of digital technology and grain production resilience. Second, a dual machine learning model, utilizing support vector machine as base learners with 5-fold cross-validation and double residualization, was constructed for causal inference to address the "curse of dimensionality" and specification biases inherent in traditional linear models. To ensure robustness, an instrumental variable approach was applied, using the product of the number of fixed-line telephones in 1984 and the national internet users of the previous year. Third, a mediation effect model was introduced to test three pathways: non-farm labor transfer, large-scale land management, and agricultural productive services.

Mechanism analysis indicated: (1) In the "People" dimension, digital technology promoted non-farm labor transfer by reducing information search costs, thereby inducing capital reflux to alleviate financial constraints; (2) In the "Land" dimension, digital platforms lowered transaction costs for land transfer, facilitating contiguous large-scale farming and improving the marginal efficiency of digital equipment; (3) In the "Service" dimension, digital technology empowered agricultural service organizations, enhancing resilience through technology diffusion and professional intervention. Heterogeneity analysis showed that the empowerment effect was more pronounced in grain production-marketing balanced zones, high-relief terrains, and economically underdeveloped regions. This highlighted the "gap-filling" nature of digital technology, where precision tools like drones and remote sensing compensated for the limitations of traditional machinery in complex terrains. Notably, while digital technology significantly drove overall resilience, it primarily bolstered regulatory adaptability and transformative innovation while exerting a significant negative impact on initial risk resistance.

Digital technology is not a mere superposition of factors; rather, it catalyzes a fundamental transformation in the dynamics of grain production through the systemic restructuring of elements. To this end, differentiated strategies should be implemented: Major grain-producing areas should prioritize full-chain data integration, whereas production-marketing balanced zones and ecologically disadvantaged regions should focus on overcoming digital infrastructure bottlenecks. Furthermore, a horizontal "digital feedback" compensation mechanism should be innovated, encouraging major consumption areas to shift from "blood transfusion" (passive aid) to "blood-making" (capacity building) through technology licensing and targeted talent support. Finally, a new system of digitalized agricultural social services should be cultivated, leveraging the scale effects of digital platforms to effectively bridge the gap between smallholder farmers and modern agriculture.