Establishing institutional platforms for data trading exerts a statistically significant positive effect on agricultural new quality productivity across Chinese cities, according to empirical research published in Frontiers in Sustainable Food Systems. Conducted by researchers Yuting Jin, Daorun Zhang, and Xuebao Fu, the investigation evaluates how market-oriented data allocation mechanisms influence modernization efforts within regional agricultural systems. By exploiting the staggered roll-out of prefecture-level data exchange centers between 2005 and 2023 as a quasi-natural experiment, the authors provide empirical evidence demonstrating that facilitating data circulation directly advances agricultural modernization. The study confirms that treating data as an economic factor of production shifts agricultural growth away from traditional factor-intensive models reliant on land, labor, and chemical inputs, steering primary production toward innovation-driven, intelligent, and sustainable agricultural practices across varied economic environments.

Agricultural new quality productivity represents a systemic qualitative transformation of productive forces characterized by disruptive technological breakthroughs, innovative resource reorganization, and significant enhancements in total factor productivity. While traditional agricultural frameworks emphasize expanding physical inputs such as machinery, fertilizers, and manual labor, this transformed productivity paradigm integrates digital technologies, knowledge resources, and emerging data assets into modern farming operations. However, agricultural data generated across rural jurisdictions often remains isolated, fragmented, and underutilized due to ambiguous property rights and high transaction costs. Data factor marketization introduces standardized institutional architectures—such as data registration, valuation, exchange protocols, and governance frameworks—that transform dispersed information into reusable factors of production. Although these trading facilities frequently operate in urban commercial hubs, data possesses distinctive mobility, replicability, and spillover characteristics that allow digital resources generated in urban sectors to diffuse into surrounding rural economies.

To rigorously identify the causal relationship between data platform establishment and agricultural productivity gains, the authors constructed a comprehensive balanced panel dataset covering prefecture-level administrative units across China from 2005 to 2023. Primary data inputs were compiled from official statistical records, including the Statistical Yearbook of Chinese Cities, the Statistical Yearbook of Chinese Rural Areas, the Statistical Yearbook of Chinese Urban and Rural Construction, local statistical bulletins, and municipal government work reports. Missing observations were supplemented using linear interpolation, while continuous variables were trimmed at the upper and lower 1% percentiles to mitigate the influence of extreme outliers, yielding an analytical sample of 4,958 valid city-year observations. The core econometric design employs a multi-period, staggered difference-in-differences framework where pilot cities establishing data trading venues constitute the treatment group, compared against non-pilot jurisdictions across pre-treatment and post-treatment observation windows.

A major methodological contribution of the investigation is the construction of an objective, text-based measurement index for agricultural new quality productivity using advanced computational linguistics. Rather than relying on rigid keyword frequencies or subjective indicator scorecards, the researchers developed a three-stage machine learning workflow using municipal government work reports. First, an initial seed dictionary of 51 terms was expanded to 104 expressions using a Skip-gram Word2Vec model trained at a context window of 10 and a cosine similarity threshold of 0.70. Second, from 208,475 extracted candidate sentences, 13,000 sentences were sampled and annotated using the GLM-4 large language model, which assigned contextual confidence scores ranging from 0 to 100. Third, 12,666 high-confidence sentences were used to fine-tune a pre-trained Chinese BERT classification architecture, which achieved 86.71% accuracy, 86.12% precision, 84.58% recall, and an F1-score of 85.34% on validation data. The finalized productivity index was calculated as the annual proportion of validated productivity-related sentences relative to total candidate sentences per municipality.

The baseline staggered difference-in-differences estimations demonstrate that introducing data trading platforms significantly increases regional agricultural new quality productivity. In unconditional specifications without fixed effects, the estimated policy coefficient is 0.086 with a clustered standard error of 0.005. When the regression incorporates a comprehensive vector of control variables—including agricultural mechanization, grain production capacity, financial development, chemical input usage, urbanization level, government expenditure, fiscal pressure, industrial structure upgrading, and mobile information penetration—alongside two-way city and year fixed effects, the treatment effect remains statistically significant at 0.057 with a standard error of 0.009. These benchmark regression findings confirm the baseline theoretical hypothesis that establishing regulated data transaction platforms accelerates modern agricultural productivity development across Chinese prefectures, providing substantial quantitative confirmation of the institutional benefits generated by structural data factor market reforms.

To ensure causal credibility, the researchers executed an event-study dynamic analysis confirming that treatment and control jurisdictions exhibited comparable pre-treatment trajectories. Estimated coefficients for the pre-establishment periods fluctuated closely around zero with confidence intervals spanning zero, and a joint F-test across all pre-treatment coefficients yielded a p-value of 0.9504, verifying that parallel trends held prior to policy implementation. Following platform establishment, dynamic treatment coefficients turned strictly positive and remained statistically significant across successive post-adoption years. Furthermore, a non-parametric placebo examination comprising 1,000 random permutations of platform timing and city assignment produced a simulated effect distribution tightly concentrated around zero, with the actual baseline coefficient lying well outside the placebo distribution, thereby confirming that the observed productivity improvements were not driven by random macroeconomic noise or unobserved chance events.

The study further addressed potential endogeneity, selection bias, and econometric specification errors through multiple sensitivity verifications. Employing year-by-year propensity score matching difference-in-differences using nearest-neighbor, kernel, and caliper matching algorithms yielded positive coefficients between 0.034 and 0.035, all statistically significant at the 1% level. To resolve potential reverse causality, the authors applied two-stage least squares instrumental variable estimation interacting each city’s historical 1984 post office tally with lagged municipal internet user counts, yielding a second-stage coefficient of 0.191 under a Kleibergen-Paap rk Wald F-statistic of 23.298. To address potential heterogeneous treatment biases inherent in staggered rollouts, the Callaway and Sant’Anna estimator confirmed an overall average treatment effect on the treated of 0.053. The results remained consistent when controlling for concurrent policies like Broadband China, omitting the COVID-19 pandemic shock, excluding provincial-level municipalities, altering machine learning thresholds, and substituting the dependent metric with an EBM-GML agricultural total factor productivity index.

Empirical heterogeneity evaluations demonstrate that the productivity gains induced by data factor marketization vary substantially according to local economic and institutional endowments. Grouping prefectures around median values reveals that the policy coefficient is statistically significant at 0.064 in regions with advanced broadband internet penetration, but statistically insignificant at 0.007 in areas with weaker digital connectivity. Similarly, prefectures exhibiting higher digital financial inclusion indices record a significant treatment coefficient of 0.072 compared to an insignificant 0.006 in low-inclusion regions. Furthermore, cities characterized by superior market business environments yielded an estimated effect of 0.105 versus an insignificant 0.008 in less favorable settings. Finally, regions with more developed new agricultural entities—such as family farms, cooperatives, and agribusinesses—exhibited a significant response of 0.085 compared to an insignificant 0.016 in regions lacking enterprise density, indicating that local absorption capacities govern technological adoption success.

To open the operational mechanism linking data market expansion to agricultural productivity, the authors estimated two-step mediation models across four pre-specified structural pathways. The empirical results demonstrate that data platform establishment significantly increases regional agricultural patent grants, with an estimated coefficient of 0.516, confirming an innovation enhancement channel driven by improved scientific knowledge diffusion. The policy simultaneously improves primary sector labor productivity with a coefficient of 0.607, driven by precise field management and reduced operational redundancies. Furthermore, higher-education student enrollment ratios expanded by 0.005, supporting human capital accumulation and rural technical talent agglomeration. Finally, text analysis measuring municipal big data deployment, governance, and application yielded a significant positive effect of 0.032, confirming that data marketization actively fosters big data utilization across agricultural value chains to streamline crop management and market forecasting.

In extended distributional and economic resilience analyses, the paper examined broader socioeconomic consequences using Hansen panel threshold regressions and output sensitivity metrics. Modeling agricultural new quality productivity as a threshold variable reveals a single threshold of 0.5714 for farmers’ income, where the positive coefficient increases from 0.130 below the threshold to 0.332 above it, showing an accelerating income-enhancing effect. For the urban-rural income gap, a threshold of 0.4359 shows that the gap-widening coefficient drops from 2.383 in early developmental stages to 1.155 at higher development levels, confirming diminishing marginal disparities as data dividends diffuse. In addition, regressing primary industry economic resilience against the productivity metric yields a positive coefficient of 0.899. While acknowledging constraints related to prefecture-level aggregation, binary platform indicators, and unmodeled data security risks, the authors conclude that data factor marketization provides an institutional framework essential for fostering resilient, high-quality agricultural development.

Subject of Research: The causal effect of data factor marketization on agricultural new quality productivity in China

Article Title: The impact of data factor marketization on agricultural new quality productivity

Article References: Jin, Y., Zhang, D., & Fu, X. (2026). The impact of data factor marketization on agricultural new quality productivity. Frontiers in Sustainable Food Systems, 10, Article 1862789. https://doi.org/10.3389/fsufs.2026.1862789

Image Credits: AI Generated

DOI: 10.3389/fsufs.2026.1862789

Keywords: agricultural economics, data factor marketization, agricultural new quality productivity, staggered difference-in-differences, digital agriculture, natural language processing, rural income, agricultural resilience, impact, data, factor, marketization