Impacts of internet access and use on grain productivity: evidence from Central China

China must raise agricultural efficiency to meet food security and sustainability goals, given limited arable land and historically heavy input use. Digitalization—especially the rapid expansion of Internet infrastructure and penetration in rural areas—offers potential efficiency gains. Prior literature links Internet development to improved agricultural productivity, but most farm-level evidence focuses on high-value crops and often proxies Internet use by access alone, overlooking purpose-specific usage. Grain farming in China differs from high-value crops due to protective pricing and relatively lower technological demands, possibly reducing incentives to use the Internet for production needs. This study asks whether Internet access and, more critically, Internet use for farming-related purposes, improve grain farmers’ technical efficiency (TE), and through which mechanisms (technology adoption and risk management). Using a representative sample of 855 grain farmers in Henan, Hubei, and Hunan, the paper distinguishes access from purposeful use and quantifies their impacts on TE.

Prior studies show positive associations between the Internet/ICT and agricultural productivity: cross-country evidence finds significant Internet effects on agricultural output; provincial-level analyses in China link Internet development to TFP growth. Farm-level studies in developing countries report that Internet access/use improves efficiency in various crops (e.g., maize, rice, banana, apple). Mechanisms include reduced information search and transaction costs, improved market access, enhanced agricultural extension and technology adoption, and better risk information (weather, pests). However, most existing work uses Internet access as a proxy for use and emphasizes high-value crops. Only limited studies distinguish purpose-specific Internet use, leaving its impacts on grain production largely understudied.

Data: A stratified random sample of 108 villages and 1080 households in Henan, Hubei, and Hunan (surveyed July 2019); analyses focus on 855 grain-farming households. Variables include outputs and inputs (labor hours, expenditures on fertilizers, seedlings, pesticides, and other inputs), Internet access (any of broadband/WiFi/mobile data), Internet use for farming information (any of learning agricultural knowledge online, searching purchase/sales information, or buying agricultural materials online), demographics, household and farm characteristics, village-level infrastructure, and province fixed effects. Technical efficiency (TE) measurement: Technical efficiency of aggregate grain production (rice, wheat, maize) is estimated using a stochastic frontier analysis (SFA) with a translog production function. The production function includes logs of inputs, squared terms, and cross-products, with a composed error term (v for noise; μ≥0 for inefficiency). TE is computed as TE = exp(−μi). The translog specification is preferred to Cobb–Douglas based on a likelihood ratio test (LR=26.02, p<0.01). Econometric strategy: To address endogeneity (self-selection and unobservable heterogeneity) in Internet access/use decisions, endogenous switching regression (ESR) models are estimated in two stages. Stage 1 estimates the binary decision to access or use the Internet as a function of instruments and controls. Stage 2 estimates separate TE equations for users/non-users (or access/non-access), including inverse Mills ratios to correct for selection. Average treatment effects on the treated (ATT) and untreated (ATU) are derived for TE under actual and counterfactual regimes. Instrumental variables (exclusion restrictions): For Internet access, “preference for ICT products” (whether the household owned a smartphone or computer in 2013 or earlier); for Internet use, “years of using the Internet.” Falsification tests confirm instruments affect access/use but not TE; 2SLS Cragg–Donald F-statistics exceed Stock–Yogo critical values, supporting instrument relevance; instrument coefficients are significant at 1% in first-stage ESR. Additional analyses: Unconditional quantile regression (UQR) with a two-stage control function assesses heterogeneity of Internet effects across the TE distribution (25th, 50th, 75th quantiles). Mechanism tests use ESR/ESP to estimate impacts of access/use on technology adoption (count of new technologies) and risk management (purchase of agricultural insurance).

- TE distribution: Mean TE across 855 grain farmers is 0.734 (range 0.255–0.933). Farmers with Internet access have mean TE 0.744 vs. 0.713 without access; difference in means is 0.031 (t-test significant). Density and CDF plots indicate higher concentration around TE≈0.8 for those with access. - ESR average treatment effects: Internet access: ATT = 0.027 (p<0.01); ATU = 0.248 (p<0.01). Access improves TE for both current access users (loss if removed) and especially for those currently without access (gain if provided). Internet access excluding farming-information users: In subsample excluding those who use the Internet for agricultural information, ATT = −0.260 (p<0.01), ATU = 0.287 (p<0.01), indicating that mere access without purposeful agricultural use can impede TE among current accessors. Internet use for agricultural information: ATT = 0.042 (p<0.01), ATU = 0.271 (p<0.01). Purposeful use yields larger gains than access alone. - Determinants: First-stage ESR shows age (−), education (+), village cadre status (+), bus access (mixed), household size (+/NS), and health (+ for use) influence access/use decisions; higher specialization in grains is associated with lower likelihood of using the Internet for farming information. Second-stage ESR indicates TE is positively related to bus access; effects of health, age, household size, and education differ across access/use regimes. - Heterogeneity (UQR): Effects of Internet access/use on TE are concentrated at the lower quartile: Internet access effect at 25th quantile = 0.122 (SE 0.070, p<0.05); Internet use effect at 25th quantile = 0.153 (SE 0.082, p<0.10). No significant effects at median or 75th quantile, implying Internet narrows TE disparities by helping lower-TE farmers. - Mechanisms (ATT/ATU): Technology adoption: Internet access ATT = −0.043 (p<0.01), ATU = 0.040 (p<0.01); Internet use ATT = 0.055 (p<0.01), ATU = 0.197 (p<0.01). Risk management (insurance purchase): Internet access ATT = −0.149 (p<0.01), ATU = 0.309 (p<0.05); Internet use ATT = 0.239 (p<0.01), ATU = 0.369 (p<0.05). Purposeful use promotes technology adoption and insurance uptake; access alone does not and can be detrimental when used for non-farming purposes. - Overall: Purpose-specific Internet use for agricultural information is the key driver of TE improvements; mere access is insufficient and non-farming use may hinder TE gains.

The study disentangles Internet access from purpose-specific Internet use and demonstrates that only using the Internet for agricultural information significantly improves grain farmers’ technical efficiency. Access alone provides potential but does not guarantee productivity gains; when access is used primarily for non-farming purposes, it may crowd out attention or resources and reduce TE. Treatment effect estimates and mechanism analyses show that purposeful use operates through increased adoption of new technologies and better risk management (insurance), consistently aligning with theoretical channels of reduced information costs, enhanced extension, and improved preparedness. Heterogeneity results indicate that these benefits accrue most strongly to lower-TE farmers, implying that Internet use can reduce efficiency disparities within the grain-farming population. These findings directly address the research question by showing that how the Internet is used—rather than access per se—determines its productivity impact in grain systems, which differ from high-value crops in market orientation and technology intensity.

This paper contributes by (1) focusing on grain farming—representing the majority of crop production in China—and (2) distinguishing Internet access from purpose-specific Internet use. Using SFA-derived TE and ESR to address endogeneity on a large Central China sample, the study finds that purposeful Internet use for acquiring agricultural information raises TE, while mere access does not and may even impede TE if used for non-farming activities. Mechanism analyses confirm positive pathways through technology adoption and risk management. Heterogeneity analysis shows stronger effects among lower-TE farmers, suggesting an equalizing role of purposeful Internet use. Policy implications include: guiding and incentivizing farmers to use the Internet for production-related purposes, and enriching supply of high-quality digital agricultural information and services (e.g., specialized apps, extension content). Future research could: refine TE measurement by crop and production stage; examine dynamic effects as digital ecosystems evolve; explore complementary investments (training, digital literacy, extension) that enhance productive use; and assess external validity across regions and countries.

The TE measurement aggregates across rice, wheat, and maize rather than distinguishing by crop, which may mask crop-specific effects; the authors argue this likely does not overturn core findings. Data are cross-sectional (2019) from three provinces, which may limit causal dynamics and national generalizability. Purpose-specific Internet use is captured by self-reported indicators that may not fully reflect intensity or quality of use.