From Online Markets to Green Fields: Unpacking the Impact of Farmers’ E-Commerce Participation on Green Production Technology Adoption

Abstract

Amid the global push for agricultural green transformation, sustainable agriculture requires not only technological innovation but also market mechanisms that effectively incentivize green practices. Agricultural e-commerce is increasingly viewed as a potential driver of green technology diffusion among farmers. However, the extent and mechanism of e-commerce’s influence on farmers’ green production remain underexplored. Using survey data from 346 rural households in Inner Mongolia, China, this study develops a conceptual framework of “e-commerce participation–green cognition–green adoption” and employs propensity score matching (PSM) combined with mediation analysis to evaluate the impact of e-commerce participation on green technology adoption. The empirical results yield four main findings: (1) E-commerce participation significantly promotes the adoption of green production technologies, with an estimated 29.52% increase in adoption. (2) Participation has a strong positive effect on water-saving irrigation and pest control technologies at the 5% significance level, a moderate effect on straw incorporation at the 10% level, and no statistically significant impact on plastic film recycling or organic fertilizer use. (3) Compared to third-party sales, the direct e-commerce model more effectively promotes green technology adoption, with an increase of 21.64% at the 5% significance level. (4) Green cognition serves as a mediator in the relationship between e-commerce and green adoption behavior. This study makes contributions by introducing e-commerce participation as a novel explanatory pathway for green technology adoption, going beyond traditional policy-driven and resource-based perspectives. It further highlights the role of cognitive mechanisms in shaping adoption behaviors. The study recommends that policymakers subsidize farmers’ participation in e-commerce, invest in green awareness programs, and support differentiated e-commerce models to enhance their positive impact on sustainable agricultural practices.

1. Introduction

As global issues such as climate change, environmental pollution, and biodiversity loss become increasingly severe, agriculture, which depends heavily on natural resources and is highly sensitive to environmental changes, is facing a shared challenge of transformation and development. United Nations projections estimate that the global population will reach 9.2 billion by 2050, with food demand in developing countries expected to double. The combined pressures of population growth and resource scarcity have exacerbated the vulnerability of agricultural ecosystems. For example, global pesticide use rose by 70% between 2000 and 2022. To address these challenges, the United Nations introduced 17 Sustainable Development Goals (SDGs) to guide global efforts. Among them, Goals 2 and 12 explicitly advocate for reducing inputs, increasing outputs, and improving quality in the pursuit of sustainable agriculture [1]. Indicators such as carbon footprint, eco-labels, and green certification are increasingly becoming key benchmarks in global agricultural markets and supply chain governance. The development priorities of many countries have gradually shifted from merely eliminating hunger and poverty to simultaneously enhancing food security while minimizing the environmental impacts of agricultural production and operations. Against this backdrop, green productivity has increasingly emerged as a key competitive advantage in niche markets such as organic and certified agricultural products. Although currently only about one-third of consumers are willing to pay a premium for green products [2,3], this group is predominantly composed of urban residents who are highly sensitive to ecological, health, and quality attributes. Their selective willingness to pay, grounded in actual purchasing behavior, provides market traction for green agricultural production with realistic expectations of profit returns.

Green agriculture emphasizes the integrated consideration of economic, social, and environmental factors within the constraints of limited land resources. Through the adoption of advanced technologies and management practices, it aims to achieve synergistic improvements in agricultural productivity, ecological sustainability, and social well-being [4,5]. For instance, organic and farmyard manures are considered central to soil quality management [6], while practices such as straw incorporation not only enhance soil fertility and crop yields but also contribute to carbon emission reduction [7,8]. In recent years, scholars have paid increasing attention to green production. Governments regulate farmers’ behavior and mitigate negative externalities through legislation, supervision, and penalties [9,10]; they also offer incentives and strengthen farmers’ risk management capabilities through public awareness campaigns, education, technical training, subsidies, and agricultural insurance [11]. As the primary agents of agricultural production, farmers’ behavioral norms and adoption of green technologies are critical to realizing green agriculture [12,13]. Due to present biases, some farmers may undervalue the long-term benefits of green production and thus delay adoption [14]. The expansion of socialized services for green agriculture has lowered the thresholds and costs associated with green technologies, thereby encouraging broader farmer participation [15]. Although green production is widely recognized as essential for agricultural transformation, adoption rates remain relatively low in developing countries [16,17]. One contributing factor is the lack of a systematic set of green technology indicators and operable assessment standards in many developing countries, which leads to significant uncertainty and makes it difficult to align green investments precisely with agricultural practices [18]. Farmers’ adoption of green technologies is influenced not only by external institutional and resource constraints but also by their internal knowledge structures. Knowledge is generally categorized into two types: subjective and objective. Subjective knowledge refers to a farmer’s self-perceived understanding of a particular domain, for example, their belief in how well they understand the advantages and applicability of green production technologies [19]. In contrast, objective knowledge is typically measured through standardized tests to assess an individual’s factual understanding of green technologies [20]. Within the frameworks of the Theory of Planned Behavior and the Knowledge–Attitude–Behavior model, subjective cognition is considered a key psychological precursor to behavioral decision-making, significantly shaping attitudes and intentions toward green practices, which in turn drive actual adoption behavior [21,22]. Incorporating subjective cognition into the analysis of green technology adoption thus offers a more comprehensive understanding of the mechanisms underlying farmers’ green production behaviors.

As a major agricultural country, China has historically relied on production-oriented policies. Despite accounting for only 9% of the world’s arable land, it produces approximately 25% of global grain output [12] and feeds over 18% of the world’s population [23]. Excessive application of agricultural inputs such as pesticides and chemical fertilizers [24] has resulted in serious consequences for China’s agroecosystems, particularly in key regions such as the Loess Plateau, the southern rice-growing areas, and the northeastern maize belt [24]. On average, Chinese farmers apply 305 kg of nitrogen fertilizer per hectare annually. Overapplication of nitrogen can result in crop lodging and increased susceptibility to pests and diseases [25]. Studies show that more than 40% of China’s land area is affected by soil erosion, with the most severe erosion occurring in mountainous regions of central China, especially in the Loess Plateau [26]. Moreover, based on data from 2006 to 2020, high levels of pesticide residues have been detected in the soils of the middle and lower Yangtze River rice-producing areas, the northeastern maize and soybean belt, and the southern hilly zones, with risk levels ranging from moderate to high [27] Despite the highest nitrogen fertilizer application occurring in southern China, nitrogen use efficiency remains extremely low. A significant portion of nitrogen is lost to the soil, water, and atmosphere, contributing to non-point source pollution [28]. These issues highlight the far-reaching challenges of extensive agricultural practices, which, while increasing yields, have compromised agricultural sustainability and deteriorated the living and production environments of farmers, posing a major bottleneck for China’s green agricultural transition [17].

E-commerce platforms have emerged as key channels linking farmers, particularly those in disadvantaged positions, to national and global markets [29]. The global wave of digitalization has led to a surge in Internet users, making online shopping a widespread norm. Digital e-commerce platforms offering efficient logistics and diverse, high-quality products have gained increasing popularity among consumers [30]. Simultaneously, agricultural production and business models are undergoing digital transformation, supported by Internet and information technologies. The digital integration of agricultural supply and demand has become a key driver of economic growth, particularly in developing countries [31]. Data from China’s Ministry of Agriculture and Rural Affairs show that online retail sales of agricultural products totaled RMB 587.03 billion in 2023, marking a 12.5% year-on-year increase and continuing a trend of sustained rapid growth. China has become the world’s largest agricultural e-commerce market. Importantly, e-commerce serves not only as a sales channel but also as an external incentive mechanism shaping farmers’ production behavior. While both conventional and eco farmers engage in e-commerce, the behavioral motivations and outcomes may differ. In China, many traditional farmers engage in agricultural e-commerce through WeChat social networks and third-party platforms such as JD.com and Taobao, primarily aiming to expand sales channels and improve operational efficiency [32], whereas green producers may utilize e-commerce to capture niche markets that reward environmentally responsible practices. Previous research has shown that participation in e-commerce can, to some extent, encourage farmers to adopt green production technologies [33]. On the one hand, e-commerce enhances the efficiency of agricultural supply chains by precisely matching consumer demand with product supply, thereby reducing product loss and resource waste [34]. On the other hand, mandatory regulations often fail to induce lasting behavioral change, whereas rural e-commerce, with its market-driven and benefit-oriented nature, links green production behavior with tangible economic returns. This provides stronger intrinsic incentives for farmers, aligning with behavioral economics theories that posit rational decision-making based on expected returns [35].

Although the importance of green and sustainable development in agriculture is widely recognized, academic research lacks a systematic and in-depth analysis of how e-commerce affects farmers’ adoption of green technologies and the mechanisms driving this process. Existing studies have mostly concentrated on the economic benefits of e-commerce, such as enhancing farmers’ bargaining power, optimizing income structures, improving market access, and increasing transaction efficiency [36]. Other studies highlight its social benefits, including the enhancement of farmers’ social capital, better access to information, and increased household resilience [37]. However, the potential influence of e-commerce on farmers’ green technology choices remains underexplored. Similarly, research on the adoption of green production technologies has largely emphasized external policy incentives and techno-economic variables, while paying limited attention to the cognitive shifts and behavioral restructuring potentially triggered by e-commerce participation. In fact, agricultural e-commerce has increasingly been recognized as a promising platform for promoting green practices [38]. Yet, whether it can serve as an effective mechanism for behavioral transformation remains an open empirical question. To address this gap, this study draws on micro-level survey data from Bayannur City, a representative agricultural region in Inner Mongolia, China, and employs a propensity score matching (PSM) approach to examine whether and how e-commerce participation, as a market mechanism, influences farmers’ adoption of green production technologies. Unlike previous studies that ask whether green farmers are more likely to use e-commerce, our core inquiry investigates the effect of e-commerce participation on subsequent green adoption.

This study contributes to the literature in three key dimensions. First, within the context of agricultural digitalization, it constructs a behavioral pathway model of “e-commerce participation–green cognition–green adoption” by drawing on the Theory of Planned Behavior and Diffusion of Innovations [21,39], introducing cognitive factors as mediators to enrich the explanatory framework of green behavior. Second, on the technological front, green production practices are disaggregated into five core sub-technologies, enabling the identification of differentiated effects of e-commerce participation across various stages of green production. Third, the study incorporates heterogeneity in e-commerce participation structures by comparing direct online sales with third-party agency sales, revealing institutional differences in behavioral motivations and cognitive responses. These contributions respond to the theoretical need for identifying micro-level mechanisms in the agricultural green transition of developing countries and provide a behavior- and cognition-based framework to support strategy design and policy refinement for sustainable agricultural development.

2. Theoretical Analysis and Research Hypothesis

2.1. E-Commerce Participation and Green Production Technology Adoption

In the context of modern agricultural transformation, farmers’ adoption of green production technologies has evolved from a purely technical decision to a rational process shaped by various factors, including information access, expected returns, behavioral cognition, and institutional incentives. This process involves not only the rational assessment of whether green technologies can be used, but also comprehensive evaluations of whether they are worth using or necessary to use [11]. In the domain of macro-level green planning, Ikram et al. 2022 proposed a comprehensive hybrid decision-making framework (SWOT–GAHP–GTOPSIS) tailored to the context of developing economies [40]. This framework serves to identify the key driving factors of agricultural green technologies and to prioritize relevant policy measures. At the national level, it plays a crucial role in strategic planning and the optimization of environmental investment pathways [40]. As an emerging mechanism for agricultural information dissemination and market transactions, e-commerce is now deeply integrated into farmers’ production and sales processes. It reshapes farmers’ decision-making logic and demonstrates significant advantages in promoting the adoption of green production technologies through five key pathways: lowering transaction costs, enhancing income incentives, establishing institutional constraints, improving resource access, and fostering social connectivity.

From the perspectives of information economics and transaction cost theory [41], farmers’ adoption of green production technologies is not solely based on rational profit maximization. Instead, their decisions are also influenced by a range of non-price transaction costs. In traditional agricultural markets, information asymmetry is widespread [42], making it difficult for farmers to accurately assess the applicability, economic viability, and market potential of green technologies. This information gap often leads to conservative behavior or risk aversion. E-commerce platforms, as critical channels for information integration and dissemination, significantly reduce farmers’ search and evaluation costs [43]. Furthermore, by offering standardized certification processes, platform regulations, and pre-sale mechanisms, these platforms help alleviate financial constraints and market uncertainties associated with green investments, thereby lowering institutional transaction costs and enhancing farmers’ willingness to adopt green technologies.

From a behavioral perspective, adopting green production technologies is fundamentally an economic decision driven by the trade-off between marginal benefits and costs. Although green agricultural practices are heavily promoted by national policies, they often require high initial investments and involve delayed returns in traditional market systems. Particularly under China’s predominantly small-scale and fragmented farming structure, influenced by traditional, experience-based reasoning, many farmers are concerned that adopting green practices may negatively affect their income [44,45]. Consequently, they are often reluctant to reduce their dependence on chemical inputs such as fertilizers and remain unenthusiastic about adopting green technologies [46]. In e-commerce systems, however, green production is reframed not as a purely ethical or ecological choice, but as one with clear market value. Through mechanisms such as green labeling, premium pricing, and traffic prioritization [47], e-commerce platforms concretize the advantages of green production. By enabling farmers to directly perceive the price signals associated with green agricultural products, these mechanisms enhance farmers’ subjective expectations of profitability and feasibility, thereby strengthening their motivation to adopt green practices. E-commerce platforms also promote contract farming and cooperative mechanisms that align farmers with governments, transport agents, and third-party sellers, forming green-oriented stakeholder coalitions that further enhance the external benefits of green production. Embedded standards, certification rules, and traffic allocation mechanisms on platforms also act as institutional norms that progressively guide farmers toward green production practices. The convergence of consumer preferences, platform-driven incentives, and the alignment of policy and market forces creates multidimensional behavioral pressures that shift green adoption from optional to normative among farmers.

Capability matching is a critical determinant of farmers’ adoption of green production technologies. E-commerce platforms improve farmers’ access to essential green inputs, including organic fertilizers, biopesticides, and water-saving irrigation systems, by streamlining acquisition processes. For farmers with limited technical foundations, platforms also provide “integrated custodial” green production services that aggregate the necessary elements for green farming, thereby alleviating the practical capacity constraints that determine whether farmers can adopt green practices. Within various e-commerce communities, such as WeChat group purchases or through interactive features on short video platforms, farmers can engage across regions and levels to share experiences related to green cultivation and certification processes, gaining rapid access to relevant knowledge and practical guidelines. E-commerce platforms also amplify the “peer imitation effect” [48]: when farmers observe others adopting green technologies and benefiting visibly, such as selling organic vegetables at higher prices, they become more inclined to follow suit. It is important to emphasize that observation alone does not constitute a sufficient driving force. Imitation intentions can be effectively translated into actual adoption behavior only when supportive conditions such as access to resources, technical capabilities, and institutional guarantees are in place.

The adoption of green technologies is a complex process driven by cognitive motivation, capacity conditions, and institutional context. Existing studies have shown that farmers’ adoption of green production technologies is not solely determined by their participation in e-commerce, but also influenced by a range of individual and structural factors. At the micro level, personal characteristics such as education, age, and health status [49,50], as well as household attributes including cultivated land area, household income, cooperative membership, and land transfer status [51,52,53,54,55], all significantly affect their willingness and behavior to adopt green practices. Accordingly, this study not only identifies the mechanism through which e-commerce influences green adoption at the theoretical level but also incorporates these subjective and objective variables into the empirical analysis as control factors to enhance the accuracy of effect identification and the robustness of causal inference.

In summary, e-commerce integrates green production technologies into a holistic value system that aligns economic returns, social recognition, and ecological sustainability. It encourages farmers to transition from passive compliance with policy mandates to active, market-driven adoption of green production practices. Specifically, e-commerce participation reshapes farmers’ decision-making processes regarding green technology adoption through a structured pathway comprising cost perception, intention formation, behavioral constraints, capability alignment, and social interaction. This process improves both the accessibility and comprehensibility of green technologies while enhancing the feasibility and long-term viability of green practices. As a result, it offers robust practical support for the broader transition to sustainable agriculture. Based on this, the following hypothesis is proposed:

H1. 

Participation in e-commerce has a positive effect on farmers’ adoption of green production technologies.

2.2. An Analysis of the Mediating Effect of Farmers’ Green Cognition Level Between Farmers’ Participation in E-Commerce and Adoption of Green Production Technologies

Studies have shown that farmers’ adoption of green production technologies is jointly influenced by both subjective and objective factors, such as green cognition, market trust, social capital, and information accessibility. These variables play critical mediating or moderating roles in the behavioral pathway from observation to imitation and eventual adoption [56]. Therefore, it is necessary to further investigate, from a cognitive–behavioral perspective, how e-commerce participation may indirectly promote the adoption of green technologies by enhancing farmers’ level of green cognition.

From the perspectives of behavioral economics and cognitive–behavioral theory [57], individual decision-making is inherently uncertain and constrained by internal factors such as cognitive capacity and behavioral awareness. Cognitive levels shape how individuals interpret and evaluate external information and, over time, subtly influence their motivations and behavioral tendencies. Consequently, farmers with higher levels of green cognition are more likely to adopt green production technologies. Specifically, green cognition refers to farmers’ integrated understanding, value assessment, and behavioral alignment with green production technologies, grounded in their economic, ecological, and social interests [58]. It serves as a critical intermediary linking green concepts to practical implementation. Drawing on the Technology Acceptance Model (TAM) and Perceived Value Theory [59], the theoretical foundation of perceived benefit lies in perceived usefulness, which is recognized as a core psychological factor influencing an individual’s behavioral intentions [60]. In agriculture, improved green cognition is primarily reflected in farmers’ heightened perception of the usefulness of green technologies. This involves recognizing the economic benefits of green inputs and their positive externalities for the ecological environment and public welfare, which fosters a stronger moral motivation for adoption [61]. Alongside perceived usefulness, the theoretical foundation of perceived feasibility lies in perceived ease of use. Perceived ease of use and operational simplicity also play crucial roles in translating intention into action [62]. Improved cognitive levels help farmers better understand the operational procedures, certification processes, and management standards associated with green production technologies. This reduces their subjective expectations of behavioral complexity and enhances their perceived behavioral control and attainability. As a result, farmers are more likely to shift from potential to actual adoption. This process also aligns with the theoretical logic in behavioral economics, which emphasizes the pathway from cognitive bias to expectation adjustment and ultimately to the reconstruction of behavioral motivation.

According to the Theory of Planned Behavior, adopting green production technologies involves a process of cognitive restructuring, motivation formation, and behavioral transformation [63]. In this process, green cognition functions both as a mediating variable and as the psychological foundation for farmers’ behavioral rationalization, institutionalization, and socialization [64]. E-commerce platforms such as Pinduoduo and TikTok serve as key hubs for aggregating information and disseminating knowledge. Through formats like short videos and livestreams, they visually demonstrate key techniques in green production, such as the mixing ratio of organic fertilizer and methods for using biopesticides, addressing the common limitations of traditional agricultural extension. The adoption of green production technologies by farmers typically unfolds as a progressive and incremental process. The repeatable and interactive features of digital content enhance farmers’ exposure to and understanding of green concepts, helping resolve the initial question of “Is green worth doing?”, which serves as a key trigger for adoption decisions. Through e-commerce training, demonstrations, and technical toolkits, farmers gradually acquire essential knowledge and skills to manage green production, which helps address concerns about “Can I implement green practices?” [65]. Moreover, platform regulations, public opinion preferences, and consumer feedback transform green practices into market norms, group identity markers, and symbols of social responsibility. Functioning as a channel for technological diffusion, social networks play a key role in helping farmers comprehend the rationale behind institutional arrangements. Through continuous group interaction, green production becomes internalized as a normative and expected behavior, thereby fostering an effective mechanism of information demonstration, imitative learning, and widespread dissemination. Based on this, the following hypothesis is proposed:

H2. 

Farmers’ green cognition plays a mediating role in the relationship between e-commerce participation and the adoption of green production technologies.

In summary, this study constructs a behavioral pathway of “e-commerce participation–green cognition–green adoption,” achieving a theoretical integration of behavioral economics, innovation diffusion theory, and cognitive behavioral theory. From the perspective of innovation diffusion theory [39], the adoption of green technologies by farmers typically progresses through stages of awareness, persuasion, decision, and implementation. E-commerce platforms, through mechanisms such as visualization, demonstration, and social interaction, enhance both the cognitive penetration and adoption rates of green technologies. As illustrated in Figure 1, e-commerce influences farmers’ green adoption behavior through five primary mechanisms: reducing transaction costs, improving information accessibility, reinforcing institutional regulation, enhancing technical empowerment, and strengthening social network connections. These mechanisms collectively enhance farmers’ green cognition, increasing their perceived benefits and feasibility of green technologies, thereby promoting green production behaviors. Green cognition, as a bridge between the awareness and persuasion stages, strengthens farmers’ understanding and trust in green technologies, forming a cognition-to-behavior transformation pathway. This is consistent with the knowledge–attitude–behavior model, which emphasizes the sequential process of knowledge acquisition, attitudinal change, and behavioral transformation [66]. This theoretical framework not only provides a conceptual basis for understanding how e-commerce platforms influence green behavior through cognitive mechanisms but also offers a clear and structured foundation for subsequent empirical analysis.

3. Data and Methods

3.1. Data Sources

Wuyuan County, located in the Hetao Plain of Bayannur, Inner Mongolia Autonomous Region, China, benefits from abundant sunlight, adequate rainfall, and moderate temperatures, creating highly favorable conditions for agriculture. These climatic conditions support concentrated crop growth and stable yields. Specialty agricultural products such as Denglonghong cantaloupe and yellow tomatoes have been granted National Geographic Indication status. In this study, we selected Wuyuan County in Inner Mongolia, China, as the survey site based on three main considerations. First, Wuyuan is designated as a national-level demonstration county for agricultural e-commerce and represents a typical case of “e-commerce-enabled agriculture,” providing a practical basis for identifying the mechanisms through which e-commerce participation affects farmers’ green behaviors. Second, the region’s agriculture emphasizes green practices such as water-saving irrigation and sustainable cultivation. It has been included in several national pilot lists, including the “Water-Saving Society Construction Counties,” “Green Pest Control Demonstration Counties,” and the “National Agricultural Product Quality and Safety Counties,” making it a suitable site for studying the adoption of green technologies. Third, the county features diverse modes of production organization among farmers, including those engaged in direct online sales, those using third-party platforms, and those not participating in e-commerce, which enables comparative analysis across different groups. Although the focus on Wuyuan County introduces a degree of geographic limitation, the county possesses a unique combination of favorable agricultural endowments, a mature rural e-commerce infrastructure, and a supportive policy environment for green development. These features make it broadly representative in terms of institutional settings, ecological context, and policy initiatives, offering policy insights for similar regions. Future research could extend this work through cross-regional comparisons in different agricultural settings to enhance the external validity of the conclusions.

In May 2023, a field survey of farm households was conducted in Wuyuan County, Inner Mongolia Autonomous Region. To enhance the generalizability of the research findings, a stratified sampling method was adopted. With the assistance and support of the Wuyuan County Rural Revitalization Bureau and the local E-commerce Industrial Park, six townships were selected for the survey: Hesheng Township, Longxingchang Town, Shengfeng Town, Taohai Town, Xingongzhong Town, and Rongfeng Subdistrict. Within each township, 3 to 4 villages were randomly chosen, from which 15 to 20 farm households per village were further selected at random for face-to-face interviews. Final data were collected through structured questionnaires. The survey covered a broad geographic scope, and the inclusion criteria required that the respondents be currently engaged in the cultivation of fruits and melons, such as cantaloupes and honeydew melons. These crops are characterized by short shelf lives, high sensitivity to market conditions, and strong dependence on e-commerce channels, making them well-suited for analyzing the behavioral mechanisms linking e-commerce participation to green production practices. The sample included both ordinary farmers and e-commerce-oriented professional growers. Respondents were randomly selected from sampling frames constructed using rosters provided by village committees.

The research design of this study is developed based on a combination of theoretical and empirical foundations, particularly drawing on the Theory of Planned Behavior and the Diffusion of Innovations Theory. The questionnaire covers a wide range of topics, including farmers’ personal characteristics, household information, production and operation details, income and expenditure, e-commerce participation, and subjective perceptions. The list of green technologies referenced in the survey is adapted from previous sustainable research on rural green technology adoption [67,68,69], which also informed the selection of key variables and the overall questionnaire structure. Before the survey, researchers clearly explained the academic purpose and confidentiality policy to the participants and obtained their informed consent prior to formal data collection.

A total of 360 questionnaires were collected during the field survey. After removing responses with missing or abnormal data, 346 valid questionnaires were retained, yielding a valid response rate of 96.1%. Among them, 46 households had not adopted any green production technologies at the time of the survey, while the remaining 300 households had adopted at least one such technology. The number of green technologies adopted ranged from one to five, with the highest proportion of households adopting three types. Notably, only one household adopted all five green production technologies. Of the five green production technologies, plastic film recycling was the most commonly adopted, with 215 households reporting its use.

Table 1. Production and marketing characteristics of sampled farmers.

Variables	Definition	Value	Notes
Major crop types	Types of crops grown, such as melons and tomatoes	100%	All sampled farmers grow melons and tomatoes
Proportion of farmers with certified specialty products	Whether the melons and tomatoes planted are certified under “Green, Organic or GI labels”	60.4%	60.4% of farmers cultivate certified specialty products
Area of certified specialty products	Total planting area of certified specialty products among sampled farmers	893 (ha)	Total planting area of certified products
Proportion of certified area	Certified area/total cultivated area	67.0%	Certified products account for 67% of the total cultivated area
Proportion of farmers using e-commerce channels	Farmers using e-commerce platforms to sell agricultural products	34.6%	34.6% of total sampled farmers
Type of e-commerce channel	Distribution of e-commerce farmers using direct sales or third-party agencies	Direct: 57.6% Third-party agency: 42.4%	Based on the classification of e-commerce sample farmers

Source: Field survey data.

presents the descriptive statistical analysis of farmers’ production and marketing activities. According to the statistics, 60.4% of the surveyed farmers grow specialty agricultural products that are certified as “green,” “organic,” or have geographical indications, with a total certified cultivation area reaching 893 hectares, accounting for 67.0% of the sample’s total planting area. This indicates a strong overall tendency among the sample group toward high-quality and green-oriented production. In terms of marketing channels, 34.6% of the farmers sell their products through e-commerce. Among the 118 farmers using e-commerce, 57.6% choose direct online sales, such as via WeChat or Taobao, while the remaining 42.4% sell through third-party intermediaries. These data provide a solid foundation for further analysis of how farmers’ participation in e-commerce influences their adoption of green production technologies. The dataset is considered accurate and reliable for empirical analysis.

3.2. Method and Model Specification

3.2.1. Propensity Score Matching Method (PSM)

Although e-commerce participation may appear to be an individual decision based on subjective willingness, it is not a random event. Instead, it represents a typical case of self-selection, which may be influenced by observable factors such as personal and household endowments, as well as unobservable factors that may simultaneously affect farmers’ adoption of green production technologies. Given differences in initial conditions between participants and non-participants, conventional regression models cannot adequately address the potential endogeneity between e-commerce participation and green technology adoption, potentially resulting in biased estimates. Furthermore, since each farmer can only make a single decision in reality, either to participate in e-commerce or not, the observed outcome reflects behavior under a single chosen state. Consequently, their outcomes under the unchosen scenario remain unobservable, resulting in missing counterfactual data. To address these issues, this study employs the propensity score matching (PSM) method to estimate the impact of farmers’ e-commerce participation on their adoption of green production technologies. The PSM approach, originally proposed by Rosenbaum and Rubin (2002) [70], mitigates self-selection bias by matching participants with comparable non-participants based on their estimated propensity scores. This method enables observational data to approximate the conditions of a randomized experiment. The core idea of PSM is to construct a counterfactual framework, comparing the potential differences in outcomes under two hypothetical scenarios: what the adoption of green production technologies would have been for e-commerce participants if they had not participated and for non-participants had they participated. The analytical procedure of PSM consists of the following steps:

In the first step, the treatment group (participating e-commerce farmers) and the control group (non-participating e-commerce farmers) of this study were set up to construct the set of covariates.

In the second step, the conditional probability fit value of farmers’ e-commerce participation, also known as propensity score value (${\mathrm{P}}_{\mathrm{i}}$), was estimated using the Logit model, and the calculation is publicized as follows:

$${\mathrm{P}}_{\mathrm{i}}={\mathrm{P}}_{\mathrm{s}}\left({\mathrm{D}}_{\mathrm{i}}=1|{\mathrm{X}}_{\mathrm{i}}\right)=\mathrm{E}({\mathrm{D}}_{\mathrm{i}}=0|{\mathrm{X}}_{\mathrm{i}})$$

(1)

In Equation (1), ${\mathrm{D}}_{\mathrm{i}}=1$ indicates that the farmer participates in e-commerce (treatment group), while ${\mathrm{D}}_{\mathrm{i}}=0$ represents non-participating farmers (control group). ${\mathrm{X}}_{\mathrm{i}}$ denotes observable characteristics of the farmer, including individual attributes, household features, production conditions, and external environmental factors.

In the third step, propensity score matching was performed. Four methods, namely, nearest neighbor matching, radius matching, caliper matching, and kernel matching, were selected to match the samples with close propensity score values in the treatment and control groups. After matching, they were further subjected to the dual test of balance and common support domain.

Different sample matching methods may produce varying estimation results due to methodological differences. However, these methods are not inherently superior or inferior to one another. If the estimated coefficients remain similar in magnitude and consistent in statistical significance across different matching approaches, this indicates the reliability and robustness of the matching results, as well as good model fit and validity. Specifically, nearest-neighbor matching pairs treated and control units based on the smallest propensity score distance (e.g., 1-to-2 or 1-to-4), offering a transparent and straightforward approach that effectively reduces pairing errors. Caliper matching introduces a maximum allowable difference in propensity scores, thereby excluding poorly matched pairs with large score discrepancies. This improves matching quality and better satisfies the comparability assumption, theoretically ensuring the “common support” condition. Radius matching expands the matching pool by allowing one treated unit to be matched with multiple control units within a specified radius, improving sample utilization without sacrificing match quality. Kernel matching uses a weighted average of all control units based on a kernel function, offering higher estimation efficiency and greater robustness, particularly when the data follow a continuous distribution and the sample size is relatively large.

In the fourth step, we calculated the difference in the adoption levels of green production technologies between the treatment and control groups, namely, the average treatment effect on the treated (ATT), to evaluate the impact of farmers’ e-commerce participation on their adoption of green production technologies. The expression is as follows:

$$\mathrm{A}\mathrm{T}\mathrm{T}=\mathrm{E}\left({\mathrm{Y}}_1|{\mathrm{D}}_{\mathrm{i}}=1\right)-\mathrm{E}\left({\mathrm{Y}}_0|{\mathrm{D}}_{\mathrm{i}}=1\right)=\mathrm{E}({\mathrm{Y}}_1-{\mathrm{Y}}_0|{\mathrm{D}}_{\mathrm{i}}=1)$$

(2)

In Equation (2), ATT represents the average treatment effect on the treated, indicating the impact of e-commerce participation on farmers’ adoption of green production technologies. ${\mathrm{Y}}_1$ denotes the adoption level for farmers who participate in e-commerce, while ${\mathrm{Y}}_0$ represents the adoption level for those who do not.

3.2.2. Mediation Model

To investigate the mechanism and transmission pathway through which e-commerce participation affects farmers’ adoption of green production technologies, this study adopts a two-step approach based on the mediation effect testing method proposed by Jiang Ting (2022) [71]. The first step involves selecting an appropriate mediating variable, namely farmers’ level of green production cognition, and evaluating its effect on the dependent variable, which is the adoption of green production technologies. The second step estimates the causal relationship between the core independent variable, e-commerce participation, and the mediating variable, in order to determine whether a mediating effect exists. The model is specified as follows:

$${\mathrm{M}}_{\mathrm{i}}=\mathsf{\alpha }+\mathsf{\beta }{\mathrm{D}}_{\mathrm{i}}+\mathsf{\gamma }{\mathrm{X}}_{\mathrm{i}}+{\mathsf{\epsilon }}_{\mathrm{i}}$$

(3)

In Equation (3), ${\mathrm{M}}_{\mathrm{i}}$ is the mediating variable farmers’ level of green production cognition; $\mathsf{\alpha }$ is a constant term; $\mathsf{\beta }$ and $\mathsf{\gamma }$ are parameters to be estimated; and ${\mathsf{\epsilon }}_{\mathrm{i}}$ is a random interference term. The positive effect of farmers’ level of green production cognition on farmers’ adoption of green production technologies is clear, so if $\mathsf{\beta }$ is statistically significant, it proves that the mediating effect exists.

3.3. Variable Description and Descriptive Statistical Analysis

3.3.1. Dependent Variable

The dependent variable in this study is farmers’ adoption of green production technologies. Adoption of green technologies refers to the selection and application of one or more environmentally sustainable agricultural practices. These technologies span the full agricultural cycle, including pre-production, production, and post-production stages. Pre-production practices include soil testing and conservation tillage [72,73]. In-production practices include pest control, precision fertilization, and water-saving irrigation [65,74,75]. Post-production practices involve straw incorporation and plastic film recycling [76,77]. Based on the existing literature, in this study, we selected five representative sub-technologies to capture green production adoption behavior: organic fertilizer application, water-saving technologies, pest control, straw incorporation, and plastic film recycling. These five components reflect practices across different stages of agricultural production. Organic fertilizer application behavior refers to whether farmers use green organic fertilizers, such as decomposed chicken manure, cow dung, or organic compost. Agricultural water-saving technology behavior refers to the adoption of water-saving practices such as drip irrigation, sprinkler irrigation, plastic film mulching, deep plowing and loosening of soil, and alternate furrow irrigation. Pest control behavior measures whether green pest management techniques are adopted, including agricultural control, physical trapping, and biological pesticide control. Straw incorporation behavior indicates whether farmers return straw to the field through composting and fermentation or via animal digestion and excretion as manure. Plastic film recycling behavior assesses whether farmers collect and recycle plastic mulch films, such as 0.01 mm transparent or black polyethylene film. Each sub-technology is coded as “1” if adopted and “0” otherwise. The total number of adopted sub-technologies is summed to create an index reflecting each farmer’s overall level of green production technology adoption.

3.3.2. Core Independent Variable

The core independent variable in this study is farmers’ e-commerce participation. It is defined as whether a farmer sells agricultural products via online platforms such as Taobao, WeChat, Pinduoduo, etc. Farmers who engage in online sales are assigned a value of 1; those who do not are assigned a value of 0.

3.3.3. Mediating Variable

In this study, we selected farmers’ level of green production cognition as the mediating variable. According to rational behavior theory, behavioral cognition is a prerequisite for action. Only when farmers understand the value, application methods, and long-term benefits of green technologies will they be motivated to adopt them. Participation in e-commerce is regarded as an external behavior, whereas the adoption of green production technologies is an internal decision. Between the two, farmers’ green cognition serves as a psychological conduit linking e-commerce participation and the adoption of green practices.

Following prior research [78], this study measures farmers’ green production cognition across four dimensions: perceived responsibility, technical awareness, awareness of resource waste, and environmental pollution cognition. The concept of “green cognition” in this study primarily refers to farmers’ perceptions and understanding of green production technologies in terms of their social benefits, ecological value, and technical feasibility. It is thus classified as a form of subjective knowledge. To minimize potential bias arising from subjective indicator selection, the entropy weighting method is used to construct a composite index that reflects the relative importance of each dimension. The specific indicator framework and corresponding weights are reported in

Table 2. Weighting results of green production cognition dimensions.

Variables	Dimension	Measurement Item	Assignment Criteria	Weight
Level of green production cognition	Perceived Responsibility	The government and village councils are the main bodies of environmental management; it has little to do with me	Strongly disagree—strongly agree: 1~5	0.441369453
	Technical awareness	Degree of understanding of green production technologies (straw incorporation, toxic waste recycling, plastic film recycling, and livestock and poultry manure resource utilization)	Very uninformed-very informed: 1~5	0.051990866
	Awareness of resource waste	Improper straw disposal, fertilizer application, etc., can lead to resource wastage	Strongly disagree—strongly agree: 1~5	0.258263739
	Environmental pollution awareness	Poor management of arable land leads to degradation	Strongly disagree—strongly agree: 1~5	0.161484797
		Inappropriate use of fertilizers, pesticides, etc., can contaminate agricultural products		0.086891146

.

3.3.4. Control Variables

To mitigate potential estimation bias due to omitted variables, this study includes a range of control variables that may affect both farmers’ e-commerce participation and their adoption of green production technologies. These factors encompass individual characteristics, household attributes, production conditions, and external environmental factors. Specifically, the control variables include age, education level, social capital, number of household laborers, years of farming experience, cultivated land area, land quality, annual net household income, engagement in specialized crop production, and the extent of non-agricultural employment among household members.

Table 3. Definition of subgroup variables and tests for differences in means.

Variable Classification	Variables	Variable Definition	Mean			Mean-Value Difference
			Full Sample	Participation in E-Commerce	Not Involved in E-Commerce	T-Test
Dependent variable	Green production technology adoption behavior	Number of green production technology adoptions: 0~5	1.896 (1.142)	2.415 (0.097)	1.627 (0.073)	0.788 ***
	Organic fertilizer application behavior	Whether green organic fertilizer is applied (the use of decomposed chicken manure, cow dung, and organic compost): Yes = 1, No = 0	0.448 (0.498)	0.466 (0.046)	0.439 (0.033)	0.028
	Agricultural water-saving technology Behavior	Whether drip irrigation, sprinkler irrigation, water conservation by covering with plastic film, deep plowing and loosening of soil, alternate furrow irrigation, and other techniques are used: Yes = 1, No = 0	0.338 (0.474)	0.458 (0.046)	0.276 (0.030)	0.181 ***
	Pest control behavior	Adoption of green pest control techniques (agricultural control, physical trapping, and biological pesticide control): Yes = 1, No = 0	0.0607 (0.239)	0.144 (0.032)	0.018 (0.009)	0.127 ***
	Straw incorporation behavior	Whether the straw is returned to the field (straw composting and fermentation for returning to the field and straw returned via animal digestion and excretion (i.e., livestock manure)): Yes = 1, No = 0	0.457 (0.499)	0.678 (0.043)	0.342 (0.031)	0.336 ***
	Plastic film recycling behavior	Whether mulch is recycled (0.01 mm transparent or black polyethylene (PE) film): Yes = 1, No = 0	0.592 (0.492)	0.669 (0.043)	0.553 (0.033)	0.117 ***
Core independent variable	E-commerce participation	Whether farmers participate in e-commerce sales of agricultural products: Yes = 1, No = 0	0.341 (0.475)	1	0	-
Mediating variable	Level of green production cognition	Calculated from the entropy method	0.493 (0.240)	0.617 (0.021)	0.429 (0.014)	0.189 ***
Control variables	Age	Actual age (years)	56.86 (9.622)	51.915 (0.905)	59.417 (0.560)	−7.501 ***
	Education level	1 = None, 2 = Elementary school, 3 = Middle school, 4 = High school or junior college, or 5 = College and above.	2.734 (0.837)	3.220 (0.072)	2.482 (0.050)	0.738 ***
	Social capital	Expenditures on favors and gifts by farm families in a year (RMB)	1.039 (0.807)	1.296 (0.091)	0.906 (0.044)	0.390 ***
	Number of family laborers	Based on the actual number of persons	1.708 (0.864)	1.992 (0.060)	1.561 (0.061)	0.430 ***
	Years of farming	You are engaged in cultivation/year	31.11 (11.560)	28.568 (1.062)	32.430 (0.754)	−3.862 ***
	Cultivated area	Actual planted area/acre	57.82 (45.300)	66.517 (4.844)	53.325 (4.844)	13.192 ***
	Land quality	How do you think the quality of land in your home compares to others: 1 = Worst, 2 = Worse, 3 = Average, 4 = Better, or 5 = Best	3.136 (0.586)	3.195 (0.058)	3.105 (0.037)	0.090
	Annual net household income	Annual per capita net household income/RMB	3.23 (2.501)	4.658 (0.268)	2.492 (0.125)	2.166 ***
	Specialized planting	Is the produce you sell a local specialty? Yes = 1, No = 0	0.604 (0.490)	0.822 (0.035)	0.491 (0.033)	0.331 ***
	Extent of part-time work	Whether the family member works part-time: Yes = 1, No = 0	0.168 (0.374)	0.203 (0.037)	0.149 (0.024)	0.054

Note: *** indicates significance at the 1% level. Source: Field survey data.

presents variable definitions, descriptive statistics for the full sample and for e-commerce participants versus non-participants, along with the results of independent sample t-tests.

The results show that the actual adoption rates of organic fertilizer application, water-saving practices, pest control, straw incorporation, and plastic film recycling technologies among e-commerce participants are 0.466, 0.458, 0.144, 0.678, and 0.669, respectively, which are all higher than those of non-participants. Except for organic fertilizer application, the differences in all other technologies are statistically significant at the 1% level. Additionally, all household characteristics except land quality and off-farm employment are significantly different between the two groups at the 1% level. Overall, farmers engaged in e-commerce tend to be younger, better educated, more socially connected, and supported by a larger family labor force. They also have fewer years of farming experience but larger farming areas, enjoy higher land quality, and are more likely to engage in specialty agricultural product cultivation.

4. Analysis and Discussion of Results

In this paper, Stata16.0 software was used, and the PSM model was selected to empirically test the e-commerce behavior of farmers and their green production technology adoption.

4.1. Estimation Results of E-Commerce Participation Decision of Farmers Based on Logit Modeling

To achieve sample matching between farmers who participate in e-commerce and those who do not, a propensity score matching (PSM) analysis was preceded by the estimation of farmers’ conditional probability of participating in e-commerce using a Logit regression model. The dependent variable was farmers’ participation in e-commerce, while the independent variables included individual characteristics, household attributes, production conditions, and external environmental factors. The detailed regression results are presented in

Table 4. Analysis of Logit model estimation results based on propensity score.

Variables	Coefficient	Z-Value	S.E.
Age	−0.069 ***	−3.15	0.022
Education level	0.897 ***	3.88	0.231
Social capital	0.093	0.54	0.172
Number of family laborers	0.37 *	1.89	0.196
Years of farming	−0.01	−0.62	0.016
Cultivated area	−0.007 *	−1.77	0.004
Land quality	−0.058	−0.22	0.269
Annual net household income	0.287 ***	3.57	0.081
Specialized planting	1.151 ***	3.45	0.334
Extent of part-time work	−0.084	−0.21	0.393
Constant	−1.012	−0.64	1.572
Log likelihood	−153.78518
LR chi2(10)	136.50
Pseudo R2	0.3074
Observations	346

Note: * and *** indicate significance at the 10% and 1% levels, respectively.

.

As shown in Table 4, education level, annual household net income, and engagement in specialty crop cultivation all have a significantly positive effect on e-commerce participation at the 1% level. This suggests that better-educated farmers are more capable of understanding and utilizing the benefits of e-commerce platforms for selling agricultural products, thus exhibiting a higher propensity to participate. Higher annual household net income, reflecting greater economic capacity, enhances farmers’ tolerance for trial and error and encourages more proactive engagement with emerging marketing channels. As consumer demand for healthy, environmentally friendly, and naturally produced agricultural products continues to grow, specialty products, due to their scarcity and added value, exhibit greater competitive differentiation in the market. In contrast, conventional crops face challenges in establishing brand identity and enhancing value through e-commerce platforms. Consequently, farmers involved in specialty crop cultivation tend to have greater confidence in product quality and market potential, making them more likely to utilize e-commerce to reach targeted consumer segments and maximize product value. In addition, the number of household laborers positively influences e-commerce participation at the 10% significance level, indicating that greater labor availability allows for more flexible allocation between production and e-commerce activities, thus reducing the opportunity cost of participation. Conversely, both age and cultivated land area have significant negative effects on farmers’ e-commerce behavior. One possible explanation is that younger farmers and those managing smaller-scale operations tend to be more adaptable to new technologies and responsive to the e-commerce environment. This indicates that e-commerce platforms are more attractive to younger and small-scale farmers. These findings are consistent with the current trend in China’s rural e-commerce, which features younger and small-scale operators. This further supports the view that digital transformation is reshaping the behavioral dynamics of agricultural actors at the micro level.

4.2. Common Support Domain and Balance Test

4.2.1. Common Support Domain Test

To assess the reliability of the matching process, a common support assumption test was performed. The overlapping region represents the common support domain. If, after matching, the common support region is too small or the overlap interval is too narrow, it indicates that too many samples have been lost, and those outside the overlapping region cannot be effectively matched. Figure 2 displays the kernel density distributions of the propensity scores before and after matching, based on the nearest-neighbor matching method. As shown in Figure 2, prior to matching, the density functions of the treatment and control groups exhibit limited overlap and poor alignment. After matching, however, the kernel density curves of the two groups align more closely, and the common support region expands significantly. This suggests that the propensity score matching procedure effectively adjusted for the differences in score distributions between the two groups. Overall, the matching results are satisfactory, and the model meets the common support assumption.

4.2.2. Balance Test

To further enhance the robustness of the results and ensure that no systematic differences exist between e-commerce participants and non-participants, this study employs a series of matching techniques based on Rubin’s approach [79]. Specifically, nearest-neighbor matching (with k = 2 and k = 4), radius matching, caliper matching, and kernel matching are applied. Several balance diagnostics are used to assess the quality of the matching, including the pseudo R², the likelihood ratio (LR) statistic, mean and median bias, as well as B and R values. The detailed results of the balance test are reported in

Table 5. Balance test results.

Matching Algorithm	$\mathbf{Pseudo}{\mathit{R}}^2$	LR chi2	Mean Bias (%)	Median Bias (%)	B-Value	R-Value
Before matching	0.309	137.03	53.2	50.2	142.1 *	1.2
Nearest neighbor matching (one-to-two)	0.015	4.35	6.8	5.5	28.8 *	1.49
Nearest neighbor matching (one-to-four)	0.005	1.55	4.7	4.4	17.1	1.38
Radius matching	0.006	1.71	3.3	1.7	18	1.52
Caliper matching	0.008	2.27	5.9	5.8	20.8	1.29
Kernel matching	0.006	1.70	3.1	2.3	17.9	1.39

Note: * indicates significance at the 10% level.

. As shown in Table 5, post-matching pseudo R² values decrease significantly from 0.309 before matching to a range of 0.005 to 0.015. Similarly, LR statistics decline from 137.03 before matching to between 1.55 and 4.35 after matching. Both the mean and median standardized biases fall below 10% after matching, well below the commonly accepted threshold of 20% for covariate balance. All covariates become statistically insignificant after matching. Moreover, the B value decreases from 142.1 to between 17.1 and 28.8, and the R values remain within the acceptable range of 1.0 to 2.0. These results indicate that the four PSM algorithms substantially reduce covariate differences between treatment and control groups, resulting in a well-balanced matched sample. Hence, the matching quality is deemed satisfactory, and the balancing diagnostics are successfully passed.

4.3. Estimation of the Impact of Farmers’ E-Commerce Participation on Their Green Production Technology Adoption and Analysis of Group Differences

4.3.1. Analysis of Average Treatment Effects on the Treated

This study employs five matching methods, including nearest neighbor matching with 1-to-2 and 1-to-4 ratios, radius matching, caliper matching, and kernel matching, to estimate the average treatment effect on the treated (ATT) of farmers’ e-commerce participation on their adoption of green production practices. The average treatment effects on the treated (ATT) are presented in

Table 6. Analysis of the estimated average treatment effect of farmers’ e-commerce participation on green production behavior.

Variables	Matching Algorithm	Treatment Group Mean (Participation in E-Commerce)	Control Group Mean (Non-Participation in E-Commerce)	ATT	T-Value
Whether farmers participate in e-commerce	Nearest neighbor matching (one-to-two)	2.286	1.776	0.510 ***	2.88
	Nearest neighbor matching (one-to-four)	2.286	1.774	0.512 ***	3.03
	Radius matching	2.286	1.728	0.558 ***	3.39
	Caliper matching	2.286	1.800	0.486 **	2.78
	Kernel matching	2.286	1.745	0.541 ***	3.27
	Mean	2.286	1.765	0.521 ***	3.07

Note: ** and *** indicate significance at the 5% and 1% levels, respectively.

, which reports the ATT estimates under different matching methods. Despite variations in matching techniques, the ATT values are relatively consistent, all passing significance tests at the 5% or 1% level, indicating strong robustness of the model results. Specifically, for farmers who do not participate in e-commerce, the average number of green production technologies adopted is 1.765. Under the counterfactual scenario where these same farmers are assumed to have participated in e-commerce, the expected adoption level would increase to 2.286. The resulting average treatment effect on the treated (ATT) is 0.521, representing an approximate increase of 29.52%. This finding provides strong evidence that e-commerce participation significantly promotes the adoption of green production technologies among farmers, thereby confirming Hypothesis 1.

From the producers’ perspective, e-commerce platforms not only expand agricultural marketing channels but also enhance price bargaining power and stabilize market expectations. These advantages improve the perceived economic return of green production practices, thereby increasing farmers’ willingness to invest in green technologies. In addition, the interactive features of e-commerce platforms, such as community-based engagement, livestream marketing, and short video sharing, provide farmers with intuitive and experience-based insights into the production practices of their peers. This fosters a behavioral pathway of “mimicry–learning–adoption”, which facilitates the organic diffusion of green technologies within farmer communities.

On the demand side, increasing consumer preferences for green, natural, and high-quality agricultural products have heightened their willingness to invest time and money in finding suitable goods via e-commerce platforms. This demand sends strong market signals. The convenient information feedback mechanisms provided by e-commerce platforms, including user reviews and repeat purchase data, function as subtle yet effective forms of institutional incentives and constraints. These mechanisms prompt farmers to adjust their production decisions in a timely and adaptive manner based on market signals.

In a market environment where green products increasingly yield premium prices, farmers become more attuned to the profit logic of linking green production, brand premiums, and income enhancement. To align with this market trend, farmers are more likely to transition toward green production as a means to enhance their market competitiveness. Moreover, the ease of accessing green inputs, technical assistance, and agricultural services through e-commerce platforms helps alleviate the resource constraints traditionally associated with green transitions. By reducing intermediaries and streamlining supply chains, e-commerce lowers both the financial burden and trial-and-error costs associated with green technology adoption, making the actual adoption cost more manageable.

Overall, the sale of agricultural products through e-commerce platforms, by integrating profit incentives, social diffusion, institutional feedback, resource accessibility, and cost efficiency, promotes a behavioral transition among farmers from passive compliance to proactive adoption of green practices.

4.3.2. Group Difference Analysis

Although the previous analysis identified the overall impact of farmers’ e-commerce participation on their adoption of green production technologies based on the average treatment effect on the treated (ATT), this estimate only reflects the average effect within the treated group and does not reveal the heterogeneous impacts across different dimensions. In other words, measuring average effects may obscure structural differences among individuals in their tendencies to adopt green technologies, their choices of specific technologies, and their modes of e-commerce participation. This limitation restricts a deeper understanding of the mechanisms through which policies exert their effects. To gain a more comprehensive understanding of how e-commerce participation influences farmers’ green production behaviors, this study further conducts a group difference analysis. It examines the variations in the impact of e-commerce participation on the adoption of different green production technologies, such as organic fertilizer substitution, water-saving practices, and pest control. Additionally, it explores the heterogeneity of these impacts under different e-commerce participation modes, including direct online sales and sales through third-party organizations. This analysis aims to uncover the potential mechanisms underlying the heterogeneous effects of e-commerce participation from multiple perspectives, thereby enriching the policy implications and theoretical understanding of related research.

(1)

Analysis of the Impact of E-commerce Participation on the Adoption of Different Green Production Technologies.

Given the ample size of the control group and the good overlap in the distribution of propensity scores, this study adopts 1:4 nearest-neighbor matching under the PSM framework to estimate the effects of e-commerce participation on the adoption of different green production technologies. This approach maintains the precision of treatment effect identification while significantly increasing the number of matched samples, thereby enhancing the statistical power of subsequent heterogeneity analyses across green technologies. The 1:4 matching method is particularly suitable in scenarios where the treated group is relatively limited but the control group is abundant, as it allows for smaller standard errors, better estimation efficiency, and greater methodological stability. It also ensures that the subsample size for each type of green technology remains sufficiently representative to meet the empirical requirements of this study. The estimation results are presented in

Table 7. Estimated effects of farmers’ e-commerce participation on the adoption of different green production technologies (1-to-4 nearest-neighbor matching).

Type of Technology Adoption	Treatment Group	Control Group	ATT
Organic fertilizer application	0.438	0.355	0.083
Water-saving technology	0.429	0.250	0.179 **
Pest control technology	0.133	0.045	0.088 **
Straw incorporation	0.648	0.495	0.152 *
Plastic film recycling	0.648	0.629	0.019

Note: * and ** indicate significance at the 10% and 5% levels, respectively.

. As shown, e-commerce participation significantly promotes farmers’ adoption of water-saving technologies and pest control practices, increasing the adoption probability by 17.9% and 8.8%, respectively, both statistically significant at the 5% level. Water-saving technologies are essential to green agriculture, providing manageable costs and stable returns that balance economic and environmental objectives. Guided by green certification and branding mechanisms on e-commerce platforms, water-saving measures have become a “ticket” to access high-end agricultural markets, encouraging greater farmer investment. Similarly, the adoption of pest control practices is notably influenced by the “green traceability” certification system promoted by e-commerce platforms.

On highly transparent online platforms, consumers’ demand for “low pesticide residues” creates market pressure that incentivizes producers to adopt environmentally friendly pest management practices. The ATT for straw incorporation is also positive and statistically significant at the 10% level, suggesting that e-commerce participation modestly promotes the adoption of straw return practices. However, the effect is relatively weaker than that of water-saving and pest control technologies. This may be attributed to the fact that, unlike the latter two technologies, which are closely tied to market signals, straw incorporation relies more heavily on local policy support, mechanization levels, and the availability of straw collection infrastructure. Since e-commerce platforms primarily target the product end of the supply chain, their influence on upstream agronomic practices tends to be indirect.

In contrast, although the ATT values for organic fertilizer application and plastic film recycling are positive, they are not statistically significant. This suggests that the marginal impact of e-commerce participation on these two technologies is limited. Several contextual factors help explain this outcome. First, the economic incentives associated with organic fertilizer application are relatively weak, while its environmental benefits are largely externalized. Unlike chemical fertilizers, organic fertilizers do not generate significant short-term yield increases; their benefits are primarily long-term, such as soil restoration and ecological improvement. These delayed and intangible returns make it difficult for farmers to perceive benefit signals through the price-feedback mechanisms of e-commerce platforms, thereby dampening their willingness to adopt. Moreover, e-commerce platforms mainly focus on the product sales stage and are less capable of visualizing the ecological benefits of practices such as soil improvement, limiting their influence over such adoption decisions. Although the use of organic fertilizers is generally encouraged by policy, their high price means that adoption decisions depend more on costs and subsidies than on participation in e-commerce.

Plastic film recycling, on the other hand, is a government-led, policy-driven environmental governance measure in China and has little direct connection with e-commerce. Its implementation relies heavily on policy instruments and organizational coordination. Since plastic film recycling does not directly generate economic returns and involves time-consuming labor, it generally lacks individualized market transaction mechanisms. Currently, e-commerce platforms do not offer service systems or supply–demand matching mechanisms related to agricultural waste recycling. Consumers rarely inquire whether plastic film was recycled during the crop production process, making it difficult for platforms to incentivize such non-productive green behaviors via price mechanisms. Instead, this type of activity depends more on collective action led by village organizations or cooperatives, as well as on government subsidies, rather than on the decentralized transactional logic of e-commerce. In summary, due to their low market visibility, weak short-term profitability, and strong externalities, these two green technologies are inherently less compatible with the information dissemination, behavioral guidance, and incentive feedback mechanisms offered by e-commerce platforms. As a result, the marginal effect of e-commerce participation on their adoption is not significant.

(2)

Analysis of the Impact of Participation in Different E-commerce Models on the Adoption of Green Production Technologies.

As demonstrated in the previous analysis, farmers’ participation in e-commerce significantly enhances the adoption of green production technologies. Building on this finding, the study further investigates whether different modes of e-commerce participation generate heterogeneous effects on green technology adoption. Specifically, e-commerce participation is classified into two categories: direct online sales and third-party agency sales. Direct online sales refer to cases where farmers independently market their agricultural products through online platforms, without intermediaries, enabling direct access to consumers. By contrast, under the third-party sales agency model, farmers deliver their agricultural products to external entities such as e-commerce vendors, logistics centers, or other intermediaries, which are responsible for marketing and selling the products on online platforms.

The average treatment effects on the treated (ATT), calculated using the radius caliper matching method, are presented in Figure 3. The results indicate that e-commerce participation significantly improves the adoption of green production technologies among farmers under both direct sales and third-party agency models. However, the impact differs in magnitude between the two participation modes. For farmers who did not initially participate in e-commerce, the number of green technologies adopted increased from 1.657 to 2.061 after engaging in third-party agency sales. The ATT is 0.404, which is statistically significant at the 5% level, corresponding to an increase of approximately 24.38%. In comparison, for those adopting a direct online sales model, the number of technologies adopted rose from 1.655 to 2.463. The ATT in this case is 0.808, which is significant at the 1% level, indicating an increase of about 48.82%. A further comparison between the two e-commerce models, using the group of farmers engaged in third-party agency sales as a reference, shows that switching to direct sales increases the average number of green technologies adopted from 2.103 to 2.558. The corresponding ATT is 0.454, which represents a 21.64% increase and is statistically significant at the 5% level. These results suggest that direct online sales are more effective than third-party platforms in encouraging farmers to adopt green production technologies.

Overall, the impact of e-commerce participation models on the adoption of green production technologies exhibits a clear hierarchical pattern. The direct online sales model shows the greatest advantage in encouraging farmers to adopt green production technologies, followed by the third-party agency model, while farmers who do not participate in e-commerce have significantly lower adoption rates. This disparity may result from the fact that direct e-commerce participation provides farmers with more abundant market information, higher income expectations, and greater autonomy in production and sales decisions, thereby better supporting their transition toward green production practices. In contrast, although third-party agency sales can also promote green technology adoption to some extent, their effectiveness is limited by information asymmetry and the lack of farmers’ dominant role in production decisions.

4.4. Endogeneity Test

To identify the causal effect of farmers’ e-commerce participation on their adoption of green production technologies, this study first applies an instrumental variable approach by constructing a two-stage least squares (2SLS) model to address potential endogeneity. A bidirectional causal relationship may exist between e-commerce participation and green technology adoption. On the one hand, participation in e-commerce can enhance the likelihood of adopting green technologies by improving access to information and market incentives. On the other hand, the adoption of green technologies may increase the competitiveness of agricultural products, thereby motivating farmers to engage more actively in e-commerce platforms.

To address this issue, in this study, we selected the average e-commerce participation rate of other farmers in the same village as the instrumental variable. On the one hand, this variable captures the village-level e-commerce atmosphere and significantly influences individual farmers’ participation decisions through neighborhood effects, thereby satisfying the relevance condition of a valid instrument. On the other hand, the behavior of other farmers in the village is unlikely to directly affect a specific household’s adoption of green technologies. Its impact operates indirectly by influencing the household’s e-commerce participation, thus meeting the exclusion restriction requirement. The 2SLS estimation results are presented in

Table 8. Endogeneity test results.

Variables	2SLS		GMM
	Phase 1	Phase 2	Phase 1	Phase 2
	Farmers’ E-Commerce Participation Behavior	Farmers’ Green Production Technology Adoption Behavior	Farmers’ E-commerce Participation Behavior	Farmers’ Green Production Technology Adoption Behavior
Farmers’ e-commerce participation behavior		0.832 *** (0.160)		0.543 ** (0.225)
E-commerce participation of other sample farmers in the same village	5.212 *** (0.332)
E-commerce training experience			0.384 *** (0.048)
Distance from the household’s farmland to the nearest express delivery point			−0.029 *** (0.002)
Control variables	Controlled		Controlled
Constant	0.057 *** (0.178)	1.612 *** (0.081)	0.719 *** (0.187)	0.934 (0.598)
Phase 1 F-value	47.39		102.32
Adjusted/Uncentered R2		0.0702		0.219
Observations	346		346

Note: ** and *** indicate significance at the 5% and 1% levels, respectively.

. The first-stage regression shows that the instrumental variable is significantly positively associated with e-commerce participation, with an F-statistic of 47.39, which exceeds the conventional threshold of 10, indicating that weak instrument concerns are not present. The second-stage results indicate that after accounting for potential endogeneity, e-commerce participation continues to significantly promote the adoption of green production technologies at the 1% level. These findings are consistent with the previous PSM estimates and reinforce the robustness of the conclusions.

To further enhance the robustness of identification, this study extends the 2SLS framework by introducing a Generalized Method of Moments (GMM) estimation model and incorporating two sets of instrumental variables. The first instrument is whether the farmer has received e-commerce training, and the second is the distance between the household’s farmland and the nearest courier service point. The former affects farmers’ e-commerce capabilities and willingness to participate, while the latter determines the ease of access to logistics and related services. Both variables are significantly associated with e-commerce participation. After controlling for individual characteristics and farming conditions, they are unlikely to directly influence the adoption of green production technologies, thus satisfying the requirements for valid instrumental variable selection.

The GMM estimation results are presented in Table 8. In the first-stage regression, both instrumental variables exhibit significant effects on e-commerce participation. Specifically, having e-commerce training significantly increases the likelihood of participation at the 1% level, with a regression coefficient of 0.384. Greater distance from the nearest courier point significantly reduces the probability of participation, also at the 1% significance level. The first-stage F-statistic is 102.32, indicating the absence of a weak instrument problem. In the second-stage results, after controlling for endogeneity bias, e-commerce participation remains positively associated with the adoption of green technologies at the 5% significance level, with a regression coefficient of 0.543. Compared with the 2SLS results, the direction of the effect remains consistent, and although the estimated coefficients differ slightly, the findings demonstrate strong robustness.

In summary, both estimation methods consistently indicate that farmers’ participation in e-commerce positively promotes the adoption of green production technologies.

4.5. Robustness Tests

4.5.1. Substitution of the Independent Variable

As a robustness check, the study replaces the independent variable “e-commerce participation behavior” with “the degree of e-commerce participation.” The degree of participation is measured by the ratio of per capita e-commerce sales income to total per capita annual household sales income. A higher ratio indicates a greater depth of engagement in e-commerce activities.

The regression results are presented in

Table 9. Robustness test results with replacement of the core independent variable.

Variables	Model (1)	Model (2)
Level of e-commerce participation of farmers	3.572 *** (0.560)	1.509 ** (0.673)
Control variables	NO	YES
Constant	1.689 *** (0.067)	1.103 * (0.608)
Adj R-squared	0.1032	0.182
Observations	346

Note: *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

. As shown, the degree of farmers’ e-commerce participation has a significantly positive effect on their adoption of green production technologies, regardless of whether control variables are included. This finding suggests that farmers with higher levels of e-commerce engagement tend to emphasize product differentiation and market competitiveness. Specifically, as e-commerce platforms increasingly promote green labels, product certifications, and related market mechanisms, such farmers are more likely to regard green production as a crucial pathway to brand development and premium capture. Consequently, they are more inclined to proactively adopt green practices such as water-saving techniques, pest control, and straw incorporation, in order to maintain their online reputation and brand credibility. This also enables them to meet platform requirements for green certification and product traceability, thereby enhancing market visibility and facilitating advancement along the e-commerce value chain.

Moreover, the information-sharing mechanisms and technical service networks integrated within e-commerce platforms help lower both the adoption barriers and perceived uncertainties of green technologies. Driven by policy guidance, platform incentives, and market forces, farmers are increasingly shifting from passive compliance to proactive green awareness. These findings are consistent with the baseline regression results and reinforce the robustness of the model.

4.5.2. Double/Debiased Machine Learning Model

To further verify the robustness of the impact of farmers’ e-commerce participation on their adoption of green production technologies, this study introduces the Double/Debiased Machine Learning (DML) model as a supplementary test based on the main regression analysis. The DML model utilizes K-fold cross-fitting techniques, which repeatedly split the sample randomly and conduct cross-validation to effectively mitigate potential overfitting issues in traditional regression models, thereby improving the accuracy of the estimation results. In this study, the number of folds is set to K-fold = 3, 5, and 8, and two representative machine learning algorithms, Lasso regression and Random Forest, are selected for estimation. This approach identifies the robust impact of e-commerce participation without relying on specific model assumptions and enhances the credibility of causal inference. The regression results are presented in

Table 10. Double/Debiased machine learning test results.

Variables	Model (1)	Model (2)	Model (3)	Model (4)	Model (5)	Model (6)
	Farmers’ Green Production Technology Adoption Behavior
	K-Fold = 3		K-Fold = 5		K-Fold = 8
	Lasso	Random Forest	Lasso	Random Forest	Lasso	Random Forest
Farmers’ e-commerce participation behavior	0.525 *** (0.144)	0.711 *** (0.137)	0.467 *** (0.144)	0.649 *** (0.134)	0.466 *** (0.144)	0.616 *** (0.132)
Constant	−0.018 (0.058)	0.056 (0.053)	−0.003 (0.058)	0.023 (0.051)	0.007 (0.058)	0.028 (0.050)
Control variables	Controlled
Observations	346

Note: *** indicates significance at the 1% level.

.

Table 10 presents the estimation results of the impact of farmers’ e-commerce participation on the adoption of green production technologies under different fold settings and algorithm specifications. Regardless of whether the Lasso or Random Forest algorithm is used, the e-commerce participation variable consistently shows a significant positive effect on farmers’ adoption of green production technologies across all models, with significance levels at 1%. This finding indicates that e-commerce participation indeed increases the likelihood of farmers adopting green technologies. Specifically, the estimated coefficients under the Lasso method range from 0.466 to 0.525, while the coefficients under the Random Forest method are higher, ranging from 0.616 to 0.711. This suggests that the nonlinear model has a stronger ability to capture the heterogeneity of farmers’ characteristics and provides better explanatory power. In addition, the estimation results show minimal variation under different K-fold settings, confirming the stability and robustness of the model. Overall, the DML model effectively enhances the robustness of causal inference and further supports the critical role of e-commerce channels in promoting the green transformation of agriculture.

4.6. Analysis of the Mediating Effect of Green Production Cognition Level

From the perspective of behavioral economics and the Theory of Planned Behavior, individual decision-making is often influenced by cognitive biases, bounded rationality, and psychological expectations, which may lead to suboptimal or irrational choices [80]. Among farmers, the level of green production cognition plays a critical role in influencing their evaluation and adoption of green technologies. Farmers with high levels of cognitive awareness are more likely to recognize the environmental benefits, market competitiveness, and policy support associated with green production technologies. They understand that agricultural production extends beyond individual economic gains and encompasses broader social and environmental responsibilities. As a result, they are more willing to develop positive attitudes toward green production, accept short-term costs, and adopt experimental practices. In contrast, farmers with lower green cognition often lack long-term planning and tend to view non-point source pollution as a natural occurrence unrelated to their own production behavior. They may also regard green technologies as complex, difficult to implement, and lacking sufficient incentives, which contributes to their reluctance to change traditional farming practices. Therefore, the level of green production cognition is positively associated with a farmer’s likelihood of adopting green technologies.

This study employs a mediation model to investigate how farmers’ participation in e-commerce affects their adoption of green production technologies. The detailed results are presented in

Table 11. Mediated effects model regression results for green production cognition level.

Variables	Model (1)
	Level of Green Production Cognition
	Coefficient	S.E.
Farmers’ e-commerce participation	0.140 ***	0.030
Constant	0.292 **	0.128
Control Variables	Controlled
Observations	346

Note: ** and *** indicate significance at the 5% and 1% levels, respectively.

.

As shown, after controlling for other variables, e-commerce participation significantly promotes farmers’ adoption of green technologies at the 1% significance level. Together with earlier findings that green production cognition positively influences technology adoption, the results suggest that e-commerce participation enhances farmers’ green behavior by improving their cognitive awareness. Therefore, Hypothesis 2 is supported. One possible explanation is that e-commerce platforms offer comprehensive information on green cultivation practices, input usage, certification processes, and consumer preferences, while also visualizing market demand via traceability systems and real-time sales feedback. This enhanced transparency helps farmers develop more stable expectations. Through e-commerce platforms, farmers gain direct insight into consumer demand for green and pollution-free agricultural products, along with the potential for price premiums. Traditional farmers, constrained by information asymmetry, often hold cognitive biases such as the belief that green production is costly, risky, and yields low returns. E-commerce platforms, by disseminating specialized knowledge and showcasing successful cases, can effectively correct such misconceptions and improve farmers’ understanding of green production. Thus, e-commerce functions not only as a sales channel but also as a catalyst for farmers’ transition to green and sustainable agricultural practices.

5. Discussion

The empirical results discussed above indicate that farmers’ participation in e-commerce significantly enhances their overall tendency to adopt green production technologies. This suggests that e-commerce participation, as a market-driven behavioral option, offers farmers stronger external incentives and more transparent expectations regarding technology adoption. Unlike traditional adoption pathways primarily driven by government initiatives, e-commerce platforms directly connect green standards with tangible benefits such as price premiums and product differentiation, thereby enhancing farmers’ intrinsic motivation toward green production. However, this influence is not purely direct. Farmers’ green cognition plays a key mediating role in shaping their green production decisions. E-commerce platforms facilitate this process by providing experiential interaction, disseminating information, and resource transparency, all of which collectively improve farmers’ understanding and acceptance of green production concepts. A more detailed classification of green production technologies reveals that, compared to environmentally beneficial but economically less immediate practices such as plastic film recycling and organic fertilizer application, e-commerce participation has a more pronounced effect on the adoption of technologies like water-saving irrigation and pest control. These technologies offer faster/quicker economic returns and are more easily integrated into the standardized systems of e-commerce platforms. Furthermore, the influence of different e-commerce models on farmers’ adoption of green technologies varies significantly. Compared to third-party agency sales, the direct-to-consumer model, which allows for greater autonomy and involves fewer decision-making steps, is more effective at encouraging farmers to invest in green practices. In the broader context of rural green transformation, e-commerce is not merely a sales channel but an enabling mechanism that activates farmers’ potential for green behavioral change. Its influence is multifaceted, encompassing cognitive reinforcement, technological targeting, and adaptability to various participation models.

The findings of this study align with those of research conducted in other countries and regions. For example, Ikram and Sadki (2024) [81] proposed a resilience-oriented strategic evaluation model for green technologies in Morocco and emphasized the strategic importance of green technology selection and institutional arrangement in developing countries. The strategic logic emphasized in this study, which combines green technology selection, strategic prioritization, and institutional arrangement, complements the “cognition, behavior, and incentive” mechanism revealed at the micro level in our research. Morepje et al., 2024 [82], in evaluating the impact of e-commerce platforms on agricultural production among smallholder farmers in sub-Saharan Africa, pointed out that such platforms contribute positively to sustainable agriculture by facilitating information flow and improving market access. Similarly, Kanagavalli et al., 2024 [83], in a study on agricultural development in India, highlights that for a country where agriculture serves as the foundation of the economy, integrating e-commerce with agriculture and leveraging digital technologies to directly connect farmers with consumers can help address the long-standing inefficiencies of traditional market transactions. Hoang and Tran (2023) also found that in Vietnam, smallholders are increasingly using the Internet and digital platforms to manage both agricultural production and sales [84]. Collectively, these studies underscore the critical role of e-commerce platforms in disseminating information and educating farmers. In the developing countries mentioned above, in particular, e-commerce not only expands market opportunities but also increases farmers’ awareness of green agricultural technologies.

However, the mechanisms and pathways through which e-commerce influences green production vary significantly across countries and regions. These variations are shaped by differences in natural conditions, policy environments, infrastructure, farmers’ cognitive capacities, and market structures. While efforts have been made to control for potential confounding factors, several limitations still remain. The primary constraint lies in the representativeness of the sample area. The data were collected from Wuyuan County, characterized by high-value crop production and recognized as a national e-commerce demonstration county. While this region provides a compelling case, its findings may not be directly generalizable to other areas such as the southern hilly regions or the agricultural zones of northwest China, which differ in natural ecological conditions, institutional settings, and the intensity of green agricultural policy support. In future studies, researchers could further conduct cross-regional comparative studies by selecting different agricultural ecological zones, such as hilly and mountainous areas, major rice-producing regions, and economically developed eastern coastal areas. Moreover, as most existing studies rely on cross-sectional data and focus on a limited set of representative technologies, future research should consider employing experimental designs and longitudinal tracking to build panel datasets on farmers’ green production behaviors. In addition, future studies may benefit from cross-country comparative analyses by drawing on national statistics and international organization databases. By doing so, comparative research could uncover both commonalities and divergences in the mechanisms linking e-commerce participation and green technology adoption.

Based on the findings, this study recommends that government policies support e-commerce-driven green technology adoption through three dimensions: market entry, behavioral incentives, and organizational support. Performance-based subsidies and targeted support for high-cost inputs and logistics, particularly in remote areas, can enhance participation. Second, support policies should differentiate by technology type. Given stronger e-commerce-driven adoption of water-saving irrigation and pest control, these should be prioritized via subsidies and low-interest loans, while for technologies like straw incorporation and film recycling, organizational support should be enhanced through a “cooperative + e-commerce” model for standardized practices and centralized recycling. Third, support should be model-specific. For direct sellers, governments should support live-streaming, demonstrations, trademarks, and market access. For third-party sellers, green standards should be embedded in contracts, with subsidies for group purchases of eco-friendly inputs. Fourth, sustained investment in green education is essential. Free livestreams, lectures, and public information materials should disseminate green practices.

6. Conclusions

Based on survey data collected in 2023 from 346 farmers in Wuyuan County in China, this study applies the PSM method to analyze the impact of e-commerce participation on farmers’ adoption of green production technologies. It then further investigates the heterogeneity of these effects across various green production technologies and different e-commerce participation modes. Finally, a mediation model is used to test the mediating role of farmers’ green cognition in the mechanism of influence. The main findings are as follows:

First, participation in e-commerce exerts a significantly positive effect on farmers’ adoption of green production technologies, increasing adoption levels by approximately 29.52%. Regarding the heterogeneity of effects across different green practices, the most pronounced impacts are observed for water-saving technologies, followed by pest control and straw incorporation, whereas the effects on organic fertilizer application and plastic film recycling are statistically insignificant. In addition, compared with third-party agency sales, direct online sales significantly promote adoption at the 5% significance level, resulting in a 21.64% increase. Moreover, green cognition serves as a mediating mechanism linking e-commerce participation to green technology adoption.