The research team conducted field interviews with rice farmers in Sichuan Province, China, from July to September 2023. The selection of research subjects was primarily based on the following considerations: First, rice is one of the staple foods in many countries across Asia, Africa, and Latin America, occupying a vital position in the global food supply chain. Rice pests and diseases are diverse, with over 20 types of pests causing economic losses, leading to an average yield reduction of approximately 20% (Pathak and Khan, 1994). As a result, the use of traditional chemical pesticides in rice cultivation is significant, making the promotion of GCT an urgent priority. Second, China ranks second in rice cultivation area and first in rice production globally, Sichuan Province accounts for 6.27% of China’s rice cultivation area and 7.02% of the annual production (National Bureau of Statistics of China (NBSC), 2023). Therefore, Sichuan is one of China’s major rice-producing provinces, and its rice cultivation practices are representative.

The study first employed a random sampling method to select five sample cities: Chengdu, Mianyang, Guangan, Luzhou, and Yibin. Subsequently, using administrative stratification and stratified random sampling techniques, 1–3 sample counties were randomly selected within each sample city. Similarly, 1–3 sample towns were randomly selected within each sample county, followed by 1–3 sample villages in each sample town. Finally, in each sample village, 15–25 rice-growing households were randomly selected as respondents for insightful interviews. The content included member characteristics, agricultural production, technology adoption, and consumption expenditures. After excluding invalid responses, the study obtained a total of 608 valid questionnaires. The sample distribution is illustrated in Figure 4.

The dependent variable is the adoption behavior of rice farmers toward GCT. In practical application, based on the characteristics of rice production and field survey results, the study focuses on three types of GCTs: biopesticides, natural enemy control, and physical control. These technologies differ in terms of acquisition and operational complexity, implementation costs, and information requirements. For example, natural enemy-based control often demands higher levels of technical knowledge from farmers; biopesticide application requires accurate identification of pest species and appropriate chemical matching; while physical control, though more intuitive, still necessitates understanding of pest life cycles and trapping timing. Regardless of the technology chosen, farmers must access relevant pest information and technical guidance, highlighting their reliance on the information environment. Given that each of these technologies can effectively serve the purpose of pest control, this study emphasizes whether a farmer adopts any of the GCT. Accordingly, a binary indicator is constructed: if a farmer adopts at least one of the three, the variable is coded as 1; otherwise, it is coded as 0.

The use of DT serves as the key independent variable. While smartphones, computers, and other smart devices have become common in rural areas, the effectiveness of DT relies not only on device connectivity but also on farmers’ ability to use these devices to search, access, and analyze agricultural data resources. Building on the approaches of Kai et al. (2023), this study represents the key independent variable as “whether farmers use DT, such as mobile phones or computers, to access agricultural information.” Additionally, to comprehensively measure DT use, the study integrates various aspects of farmers’ production and daily life, including social interaction, information acquisition, input purchase, and product sales. During the survey, farmers are asked to rate the extent of their DT use in these aspects on a five Likert scale. Subsequently, the entropy weight method was applied to assign weights to the indicators based on the amount of information each conveys, and an aggregated index was constructed as an alternative measure of DT use for robustness checks. Compared with simple averaging, the entropy method captures the heterogeneity across indicators more effectively by assigning greater weight to those with higher information variability, thereby enhancing the representativeness and objectivity of the composite index. Therefore, this study adopts the entropy weighting approach for indicator aggregation.

There may be endogeneity issues between GCT adoption and DT use, potentially due to reverse causality or omitted variables. To address this, this study follows the approach of Luo and Liu (2022), selecting household internet access cost as the instrumental variable for DT use. This variable reflects the cost incurred by farmers in accessing digital information and is theoretically expected to be significantly associated with their DT use. However, we also acknowledge that higher household internet expenditure may correlate with stronger economic capacity, which in itself could facilitate the adoption of GCT. To mitigate this potential confounding pathway, the model controls for key variables indicative of household economic capacity, including the household head’s education and occupation, household size, number of laborers, and cultivated land area. Given that these covariates account for the majority of economic differences across households, the effect of internet expenditure on GCT adoption—after controlling for these factors—can be reasonably attributed to its influence through DT use. This ensures the validity of the exclusion restriction in the context of our study. In summary, “household expenditure on communication and internet services” satisfies both the relevance and exclusion criteria, and thus serves as a valid instrumental variable for the core explanatory variable in this research framework.

Technical perception serves as the mediator. Based on theoretical analysis, technical perception is divided into three aspects: PU, PEU, and PR. Drawing on Huang X. et al. (2022), the study further classifies perceived benefits into economic, social, and ecological dimensions. Economic benefits perception is measured through questions such as: “To what extent do you believe green production technologies increase income?,” “enhance yields?,” “reduce material input costs?,” and “save labor?.” Social benefits perception is assessed using statements like: “Adopting green production technologies improves food safety,” “benefits other members of society,” and “contributes to social development.” Ecological benefits are gauged via: “How effective do you perceive green production technologies to be in environmental protection?,” “in conserving natural resources?,” and “in reducing environmental pollution?” Perceived ease of use is measured by the respondent’s level of agreement with the statement: “GCT are easy to master.” PR is assessed by asking whether they perceive GCTs as carrying minimal technical risk (e.g., yield reduction) and market risk (e.g., failure to achieve premium pricing). All items are rated using five-point Likert scales. For multi-item constructs, the entropy weight method is applied to compute composite indices, which assigns greater weights to variables with higher informational variance, thus improving the indicators’ representativeness and objectivity.

Farmers’ decisions to adopt technology are influenced by multiple factors. Therefore, individual, household, and environmental characteristics are incorporated as control variables (Creissen et al., 2021; Wu et al., 2018). Additionally, to account for the potential influence of location, terrain, and unobserved regional factors, the study includes variables such as distance to the nearest county, topography, and locational dummy variables. Variable descriptions are provided in Table 1.

The t-test results further visually illustrate the characteristic differences between rice farmers who use DT and those who do not. Specifically, 72.3% of DT-using rice farmers have adopted GCT, while only 24.8% of non-DT-using farmers have done so, with the difference being significantly positive at the 1% level. It can be intuitively inferred that DT users generally perform better in implementing GCT. Additionally, the internet access cost, as the cost of DT use, is understandably higher among DT users compared to non-users. In terms of technical perception, DT users exhibit stronger PU, PEU, and PR regarding the technology than non-users, indicating greater confidence in the benefits and operability of GCT, along with a lower estimate of its risks. Furthermore, DT users are usually to be younger, more educated, risk-tolerant, engaged in larger-scale farming, located closer to the county, and residing in flatter terrains.

Firstly, the study aims to clarify the direct and overall associations between DT use and the adoption of GCT. Given that farmers’ decisions regarding the adoption of GCT are binary, a Probit model is employed for the foundational analysis, structured as follows:

In Equation 1, GCT i represents the adoption decision of GCT by rice farmer i, and DT i denotes the DT use of rice farmer i, encompassing both the decision to use DT and the capacity to utilize it; Con i includes the individual, household, and environmental control variables of rice farmers; β 0 is the constant term; β 1 and β 2 are the coefficients to be estimated, and ε is the random error term.

Despite incorporating numerous control variables in Equation 1, potential omissions remain inevitable, leading to endogeneity. To address it, the study continues to employ the CF approach for estimation. This approach is advantageous due to its flexibility and ability to intuitively identify endogeneity within the results (Wooldridge, 2015). The process involves two main steps: first, DT use is treated as the dependent variable, with the instrumental variable incorporated as key independent variable to estimate the generalized residual, as shown in Equation 2; second, the generalized residual is then included in Equation 1 for re-estimation, as demonstrated in Equation 3. If the residual is significant, it indicates that DT is an endogenous explanatory variable, and including the residual corrects estimation bias. If the residual is not significant, it suggests that DT is not an endogenous explanatory variable, making the regression results without the residual more valid and unbiased. Based on this approach, the following model is constructed:

In Equations 2, 3, IAC i represents household internet access cost for farmer i, σ i is the generalized residual obtained from Equation 2, δ 0 and ρ 0 are constants, δ 1 ~ δ 2 , and ρ 1 ~ ρ 3 , are the coefficients to be estimated, ϵ and θ are random disturbance terms.

Equation 1 may also be subject to endogeneity arising from “self-selection,” as both the adoption of DT and GCT by farmers may be influenced by some same factors (Gao et al., 2020; Zheng et al., 2022). To address this, the study applies the PSM method within the “counterfactual” framework to verify the relationship between DT use and GCT adoption. The main modeling ideas are as follows: First, a regression model estimates each farmer’s probability of DT use, yielding a propensity score for each sample. Second, all sample farmers are categorized into a treatment group (users) and a control group (non-users) based on DT use, and different matching methods are employed to pair samples with similar propensity scores across both groups. Third, the difference in GCT adoption rates between the two groups is analyzed to determine the average treatment effect on the treated (ATT). The final expression is as follows:

In Equation 4, D i indicates whether farmer i belongs to the treatment group, D i = 1 for using DT, D i = 0 for not adopting DT; P ( Xi ) represents the propensity score, and Y i and Y 0 denote the outcomes for the treatment and control groups, respectively.

After clarifying the direct effects, the study also investigates the mechanisms based on Equation 1. The testing method follows the stepwise regression equation method proposed by Baron and Kenny (1986), constructing the mediation model as follows:

In Equations 5–7, TP i is the mediator, which indicates rice farmers’ perception of GCT, encompassing PU, PRU, and PR. φ 0 and γ 0 are constants, φ 1 , φ 2 , γ 1 , γ 2 , γ 3 are coefficients to be estimated, π and τ represent random disturbance terms.

Table 2 presents the results of the basic regression analysis. Model (1)–Model (4) employs stepwise regression by gradually adding control variables. It is clear that, there is still a positive correlation on the 1% statistical criterion between the adoption behavior of GCT and the use behavior of DT among rice farmers, even after including more restrictive factors.

Moreover, to ensure robustness, Model (5) conducts a re-test by replacing the key independent variable. The robustness check results are consistent with the stepwise regression findings, indicating that the positive association between DT use and GCT adoption remains highly significant.

This study further employs the CF approach and the PSM method for correction. Model (1), the first stage of the CF approach, results indicate a positive correlation between household internet access cost and the use of DT, demonstrating that internet access cost meet the relevance criterion as instrumental variable. Subsequently, in Model (2), the results of the following stage, the residual is significantly positive, confirming the endogeneity of DT use. This suggests that estimating with a standard Probit model alone would lead to an overestimation of DT’s effect, whereas including the residual corrects for estimation bias. Specifically, compared to the baseline regression, the coefficient of DT use derived from the CF approach is lower. However, the association between DT use and the adoption of GCT remains positive and statistically significant at the 1% level. The average marginal effect further clarifies the economic implications of the data: when rice farmers choose to DT, the probability of them adopting GCT will go up approximately 16.1% on average. The specific analysis results are shown in Table 3.

Based on different matching methods, PSM method may yield varying results. To enhance the reliability of the conclusions, this study conducts tests using nearest neighbor matching (1:1), radius matching, and kernel matching. As shown in Table 4, the ATT calculated by all the methods are very similar, and all of them are very positive and significant (due to space constraints, the estimation results of the propensity scores are omitted here). Specifically, the likelihood of rice farmers adopting GCT increased to approximately 68% among those using DT, which is significantly higher by about 26.3% compared to those not using DT.

In summary, after addressing endogeneity, DT use has been confirmed to have value in promoting technological adoption. This finding differs from the studies by Zhou et al. (2023), as well as Yu and Zhang (2009), both of which did not identify a correlation between internet use and IPM application. In the former, the key independent variable was measured directly as “whether or not the internet was used,” which failed to eliminate interference from non-productive uses, potentially obscuring the correlation. The differences with the latter study stem from the historical context. This study is conducted against the backdrop of rapid DT development in rural China. The conclusions are indirectly supported by Cai et al. (2022), as well as Gao et al. (2020), who similarly affirm the necessity of using DT to facilitate the penetration of GCT.

Murage et al. (2015) have pointed out that successfully transmitting information to target recipients is a prerequisite for technology adoption. DT has further expanded farmers’ access to information beyond traditional agricultural technology dissemination models (Barnett et al., 2019). In terms of information scope, digital information technology reduces information search costs and fosters the creation and broad sharing of technical information (Shen et al., 2022; Wu and Zhang, 2020). Meanwhile, given the multi-period characteristics of GCT, farmers face a heightened need for intensive and sustained access to information (Lou et al., 2021). Unlike occasional promotional training, DT continuously accompanies farmers, subtly shaping their green production mindset (Pan et al., 2017).

Based on the extended TAM, this study employs the B-K stepwise regression equation to examine the mediation effects of PU, PEU, and PR in the relationship between DT use and GCT adoption. This analysis is complemented by the Bootstrap and Sobel tests (Wen and Ye, 2014). The specific analysis results are shown in Table 5.

The B-K stepwise regression equation results show that the use of DT significantly enhances rice farmers’ PU, PEU, and PR associated with GCT. After controlling for the direct effects of DT use, these three perceptions further exhibit a clear positive impact on the GCT adoption. The 95% confidence interval of the mediation effects obtained through the Bootstrap test also did not include zero. These data indicate that DT use raises farmers’ expectations regarding the benefits and convenience of GCT while simultaneously reducing their risk assessments, ultimately promoting GCT adoption. H2 and H3 are confirmed, while H4 is rejected.

This finding aligns with the rational smallholder hypothesis and supports the theory put forward by Nowak (1987), which emphasizes that making the target group aware of the necessity of the technology and providing adaptation assistance are prerequisites for technology diffusion. Generally, studies suggest that the most valuable, low-risk innovations are often the quickest to be adopted (Fliegel and Kivlin, 1966). In practice, mastering the essentials of GCT can enhance farmers’ welfare. As a robust information channel, DT can improve farmers’ operational capabilities and promote the realization of technical benefits, providing intrinsic motivation for technology adoption (Samiee et al., 2009; Verma and Sinha, 2018). Risk, meanwhile, implies uncertainty and threat (Trujillo-Barrera et al., 2016). Although theoretically, digital media might expose cases of DT failure, potentially leading farmers to overestimate risks, the opposite has proven true: DT effectively lowers PR and fosters technology adoption. This can be attributed to two factors. First, DT helps to bridge the information asymmetry in the market and enhance individual development capabilities (Zheng and Lu, 2021), thus reducing uncertainty in the technology adoption process (Luh et al., 2014). Second, DT’s visibility and traceability render GCT’s operational steps more transparent and simplified, instilling confidence in farmers about their capacity to use GCT effectively, thereby lowering their PR of adoption. Rather than leaving farmers in a technological “black box,” transparency of technical information helps mitigate the fear associated with unknown risks. Moreover, distinguishing between different sources of digital information may yield varying conclusions. For instance, guidance from research institutions or government extension platforms may reduce farmers’ risk perception, whereas peer exchanges on social media might amplify it by emphasizing negative experiences. Unfortunately, this study does not differentiate farmers’ digital information sources in detail, which represents a promising direction for future research.

In comparing the mediation effects of the three types of cognition, the study also reveals that PU, PEU, and PR account for 45.05, 35.13, and 17.09% of the mediation effect, respectively, showing a decreasing trend. This finding reflects the underlying logic that benefit drives technology dissemination, suggesting that amid growing support and technological maturity, digital media in China places significant emphasis on promoting the benefits of GCT, laying a strong foundation for its adoption.

The breadth and depth of information acquisition using DT vary among farmers, leading to different behavioral responses, due to the gap of farmers’ resources endowments. To capture this variability, the study selects age and area of rice cultivation, and the distance from the housing to the county, corresponding to individual, household, and environmental characteristics, respectively, for group discussions. Notably, to alleviate endogeneity interference, the study continues to use the CF approach for estimation. The research results are shown in Figures 5–7.

The comparative results among different age groups indicate that the use of DT is significantly associated with the adoption of GCT among younger farmers, whereas no significant association is observed among older farmers. Two potential reasons for this finding are: first, due to differences in education levels and technological acceptance, older farmers may not be as proficient in using digital tools like the internet as their younger counterparts, which limits the effectiveness of DT (Chen H. et al., 2023); second, older farmers typically rely on their past planting experiences and traditional methods, leading them to be skeptical of the information obtained through DT, they may be reluctant to alter their established planting habits in the short term (Murage et al., 2015).

The comparative results across groups with varying rice cultivation areas indicate that the use of DT positively influences the adoption of GCT across all scales of rice farming. However, this association is more pronounced among farmers with smaller landholdings. Gao et al. (2020) reached similar conclusions, attributing this effect to the greater capacity increment benefits brought by the adoption of DT for smallholders. However, the study suggests two additional reasons. From a technological demand perspective, smaller-area farmers primarily grow rice for self-consumption, making them more likely to accept GCT to ensure food safety (Yu and Zhang, 2009). Although large-scale farmers possess capital and scale advantages conducive to adopting GCT, their rice production is primarily market-oriented. Under China’s underdeveloped green agricultural product market, it is difficult for green products to command price premiums. Therefore, aiming to maximize benefits and minimize risks, large-scale farmers are more likely to rely on chemically stable pesticides. Regarding behavioral response speed, the decision-making processes of larger-area farmers are often more complex, whereas smaller-area farmers, have shorter decision chains, allowing for more flexibility in adjusting their production methods (Timprasert et al., 2014).

A comparison between closer-distance and farther-distance farmers indicates that the advantage of DT in enhancing the adoption of GCT is significant only among closer-distance farmers, while it is not significant among farther-distance farmers. This result may be due to differences in infrastructure arising from geographical location. Areas near the county usually have more agricultural extension agencies, technical service stations, and tool sales points, providing farmers with face-to-face technical guidance and necessary equipment support. In contrast, remote areas tend to have fewer relevant institutions, limiting farmers’ access to modern agricultural production technologies or equipment (Bello-Bravo et al., 2022).