Based on bounded rationality theory, farmers will pursue the maximization of interests in the decision-making process, and make rational analysis according to the existing information, so as to determine the optimal scheme within the cognitive range. Pesticide reduction not only affects the quantity and quality of agricultural products, but also directly relates to the production cost and household income of farmers. Therefore, in the decision-making process, farmers will rationally evaluate the effectiveness of pesticide reduction, with the goal of maximizing income and minimizing risk, and form expectations for agricultural income. When farmers expect pesticide reduction to bring higher agricultural income, their willingness to adopt pesticide reduction technology will be significantly enhanced. On the contrary, if the expected income is low or the risk is high, farmers may tend to maintain the status quo.

E-commerce adoption has significantly improved farmers’ agricultural income expectations, and thus achieved pesticide reduction. E-commerce platform provides farmers with key data such as market supply and demand information, price fluctuation trend and consumer preferences, which helps farmers to predict agricultural income more accurately, and directly improves the market competitiveness and premium ability of agricultural products by reducing intermediate links and broadening sales channels, thus enhancing farmers’ optimistic expectations of agricultural income (37), enabling farmers to more clearly foresee the economic benefits brought by pesticide reduction (38). At the same time, the technical training and support provided by the e-commerce platform further reduces the risk of FPRB, making them more inclined to achieve income growth through technological innovation.

Based on the above analysis, this paper puts forward research hypothesis 2:

Awareness and cognition are important factors affecting farmers’ adoption decision (39–41). If farmers do not understand the role of the agricultural technology sufficiently, they may be less likely to adopt it (42). Therefore, food safety perception has a significant impact on FPRB. Food safety perception reflects farmers’ understanding and attention to the safety production standards of agricultural products, which is an important internal driving force for farmers’ technical behavior choice (43). With the increasing demand of consumers for food safety, farmers’ perception of safety production technology directly affects their production decisions. When farmers have a higher perception of food safety, they will be more inclined to adopt technical behaviors that can ensure the safety of agricultural products, such as pesticide reduction technology, to meet market demand and enhance their competitiveness. In addition, food safety perception can also standardize farmers’ production behavior and urge them to actively reduce pesticide use in the production process to meet the safety production standards (44).

E-commerce adoption has significantly improved farmers’ perception of food safety, and thus achieved pesticide reduction. E-commerce platform provides farmers with information on consumers’ demand for food safety, market access standards and related policies and regulations, helping farmers to understand the importance of food safety more comprehensively. At the same time, through the establishment of a transparent supply chain system, the e-commerce platform enables farmers to directly perceive the role of safe production technology in improving the market competitiveness of products, and realize the positive role of pesticide reduction in ensuring food safety and enhancing product value, thus enhancing their willingness to reduce.

Based on the above analysis, this paper puts forward research hypothesis 3:

The socialized service of agricultural machinery plays an important role in farmers’ decision on pesticide reduction and is an important way to promote the sustainable transformation and sustainable development of agriculture (45). First of all, the socialization service of agricultural machinery can accurately control the application amount of pesticides and effectively reduce the abuse of pesticides by introducing specialized agricultural machinery and technologies, such as drone plant protection and precision spraying system (46, 47). Secondly, the socialization service of agricultural machinery can improve agricultural production efficiency by reducing farmers’ production cost and time cost, and then promote farmers to adopt pesticide reduction technology (48). This service model reduces the burden of farmers on equipment purchase, maintenance and labor input, and improves the utilization efficiency of pesticides through specialized operation. Finally, the socialization service of agricultural machinery ensures the scientificity and effectiveness of pesticide reduction measures through standardized service process and professional management, improves the attach rate and utilization rate of pesticides, and reduces the waste and abuse of pesticides.

E-commerce adoption has significantly promoted the adoption of socialized agricultural machinery services by farmers, and thus achieved pesticide reduction. E-commerce platform provides farmers with information about agricultural machinery socialization service, including service content, technical advantages and cost-benefit analysis (46), which promotes farmers’ willingness to adopt agricultural machinery socialization service. At the same time, the e-commerce platform reduces the information asymmetry by establishing the direct connection between farmers and agricultural machinery service providers, and enables farmers to obtain specialized services more conveniently (45). With the support of e-commerce platform, farmers can obtain specialized agricultural machinery services more efficiently, and reduce pesticide consumption through precise operation, thus achieving pesticide reduction.

Based on the above analysis, this paper puts forward research hypothesis 4:

The theoretical analysis framework of this paper is shown in Figure 1.

The dependent variable of this paper is FPRB. The questionnaire inquired in detail about the changes in the average pesticide use per hectare of farmers compared with 5 years ago. Therefore, select the questionnaire “Compared with 5 years ago, does the average pesticide use per hectare decrease?” This problem is used to measure FPRB, and the sample that answers “reduction” is assigned to 1, otherwise it is assigned to 0.

The core independent variable of this paper is ECA. Using the research of Qiu et al. (49) for reference, this paper selects the question “Does your family operate products through online transactions” in the questionnaire to measure the ECA of farmers. If farmers sell products through e-commerce, the value is 1, otherwise the value is 0.

Based on the theoretical analysis above, it can be seen that ECA can promote FPRB by enhancing agricultural income expectations, enhancing food safety perception and adopting agricultural machinery socialization services. This paper selects the questionnaire “How do you think your family’s agricultural income level will change in 2020 compared with that in 2019?” This indicator is used to measure farmers’ agricultural income expectation. If the answer is “increased” or “increased more,” it is assigned to 1, otherwise it is assigned to 0. At the same time, this paper uses the questionnaire “Do you usually care about food safety?” This question is used to measure farmers’ perception of food safety. The answer to “very concerned” or “relatively concerned” is assigned to 1, otherwise it is assigned to 0. In addition, this paper measures the adoption of agricultural machinery socialization service by the question of “mode of pesticide application” in the questionnaire. If you buy machinery service to apply pesticide, it is assigned to 1, otherwise it is assigned to 0.

Referring to the research of Su (50), this paper uses the spherical distance from Hangzhou to cities as a tool variable. The e-commerce represented by Taobao originated in Hangzhou, so Hangzhou is in a leading position in the development of e-commerce. Generally speaking, the closer to Hangzhou geographically, the stronger the e-commerce radiation-driven effect on the city, and the higher the degree of e-commerce development. From the perspective of correlation, Hangzhou, as the core area of e-commerce development, has a significant radiation effect on the ECA in the surrounding areas. There is a strong correlation between the spherical distance from Hangzhou to cities and the development degree of ECA, which can effectively reflect the geographical differences of ECA. From the externality point of view, the spherical distance is a geographical objective index, which is not directly affected by economic activities or policy factors, and has nothing to do with the error term in the model, and meets the exogenous requirements of tool variables. At the same time, the spherical distance only indirectly affects the pesticide reduction by affecting the radiation effect of ECA, but does not directly affect the pesticide reduction, which is in line with exclusive constraints. Therefore, the spherical distance from Hangzhou to cities is a reasonable tool variable, which can effectively solve the possible endogenous problems between ECA and pesticide reduction.

In addition to ECA, individual characteristics, family characteristics and village characteristics of farmers will all affect FPRB to some extent. Referring to the existing research (20, 21), this paper selects control variables from three dimensions: individual characteristics, family characteristics and village characteristics. Among them, individual characteristics include gender, age, education level and marital status; Family characteristics include family agricultural insurance, family political identity and family cultivated land scale; the characteristics of a village include its topography, its distance from the county government and its economic development level.

Table 1 shows the definition, assignment and statistical characteristics of the above variables.

This paper compares the pesticide reduction of e-commerce farmers and non-e-commerce farmers. Figures 3a, b shows the density distribution and cumulative distribution of pesticide reduction, respectively. It can be seen from the figure that the probability of pesticide reduction by e-commerce farmers is higher than that of non-e-commerce farmers. From this, it can be preliminarily judged that ECA plays an important role in promoting pesticide reduction. Of course, to confirm the impact of e-commerce on pesticide reduction, more rigorous quantitative analysis is needed. The following will empirically analyze the impact of ECA on FPRB.

Considering that the dependent variable “FPRB” is a binary classification variable, this paper uses binary Logit model for regression analysis, and the model is set as follows:

In the formula (Equation 1), FPRB represents the farmers’ pesticide reduction behavior; ECA represents e-commerce adoption; X is the control variables, including the individual characteristics, family characteristics and village characteristics of farmers; β₁ and β₂ are the coefficient to be estimated; β₀ is constant term; ε is error term.

Referring to the research (49) of this paper uses the intermediary effect model to test the mechanism of agricultural income expectations, food safety perception and agricultural machinery socialization service adoption. The model formula (Equations 2–4) are as follows:

In the formula, FPRB represents the farmers’ pesticide reduction behavior; ECA represents e-commerce adoption; M represents agricultural income expectations, food safety perception and agricultural machinery socialization service adoption; X is the control variables, including the individual characteristics, family characteristics and village characteristics of farmers; c, a, c′, α1, α2, α3, b represent the coefficient to be estimated; λ1, λ2, λ2 represent constant term; μ1, μ2, μ3 represent error term.

In order to ensure the robustness of the benchmark regression results, this paper replaces the binary Logit model with the binary Probit model, and then carries out the estimation test again. The results are shown in column (1) of Table 3. After replacing the model, ECA still has a significant positive impact on FPRB, and the previous estimation results are robust.

Considering the particularity of farmers over 60 years old in physical fitness and health status, this paper deletes the samples of farmers over 60 years old and regresses them to eliminate the interference of some special factors and make the results more representative. The results are shown in column (2) of Table 3. ECA still has a significant positive impact on FPRB, which further verifies the robustness of the previous results.

In addition, this paper regresses by replacing the dependent variable to verify the robustness of the previous results. In the benchmark regression, “whether the use of pesticides is reduced compared with 5 years ago” is used to measure FPRB, which is replaced by “the cost of self-purchased preventive pesticides by farmers” and re-estimated. The result is shown in column (3) of Table 3. ECA has a significant negative impact on farmers’ self-purchasing cost of disease prevention pesticides, which indicates that ECA promotes FPRB, is conducive to reducing pesticide application and promoting agricultural sustainable production, further confirming the robustness of the previous results.

In order to avoid the influence of extreme values on the regression results, 1% of the data at the left and right ends of continuous variables in the controlled variables are truncated and then regressed. The results are shown in column (4) of Table 3. ECA still has a significant positive impact on FPRB, and the above estimation results are robust.

In order to verify whether the unobserved random factors will affect the regression results, this paper set up an experimental group to carry out placebo test, and the remaining samples were used as the control group. As shown in Figure 4, after repeating the randomization operation for 1000 times, the average value of the regression coefficient of ECA approaches to zero, and presents the characteristics of approximate normal distribution. In addition, this average value is significantly different from the estimated value in the benchmark regression model. Therefore, the potential interference caused by unobserved factors can be ruled out, which indirectly proves that ECA can significantly promote FPRB.

Considering that the behavior decision of ECA is not random, it may be influenced by other factors such as individual, family or village characteristics. Therefore, in order to avoid the estimation bias caused by self-selection, this paper uses the propensity score matching method (PSM) to re-estimate. The average treatment effect of ECA on FPRB is shown in Table 4. ECA has a significant positive impact on FPRB, whether it is estimated by the propensity score matching methods such as nearest neighbor matching, radius matching, kernel matching, mahalanobis matching or local linear matching.

The PSM model specification must meet two preconditions: overlapping assumption and balanced characteristics. For the overlapping hypothesis, it is necessary to test the common support domain. After the matching is completed, if the area of the common support domain is too small, it means that there are too many sample losses and the matching effect is not good. On the contrary, it means that the matching result is good. Figure 5 is a trend score distribution diagram of the test common support domain. The area of common support domain after matching Figure 5a is larger than that before matching Figure 5b, and there is a large common support interval, which shows that the matching effect is good and the common support domain test passes.

At the same time, in order to examine whether the above propensity score matching estimation results balance the data well, it is necessary to carry out a balance test, and the balance test results are shown in Table 5. In this paper, the kernel matching method is taken as an example. The results show that the standard deviation rate of most control variables is reduced after matching, and the standard deviation rate is less than 10%. At the same time, most of the t-test results of control variables do not reject the original assumption that there is no systematic difference between the experimental group and the control group, which effectively balances the difference of covariate distribution between the experimental group and the control group. Therefore, the matching results of propensity score pass the balance test.

There may be endogenous problems caused by reverse causality in benchmark regression, which makes the previous conclusions biased. On the one hand, ECA promotes FPRB through economic incentives and the spread of sustainable production concepts; On the other hand, after pesticide reduction, the economic value of sustainable and high-quality agricultural products in the market is higher, and farmers are more likely to carry out ECA activities to expand the scope of market transactions and increase income.

In order to solve this endogenous problem and improve the robustness of the model estimation, this paper replaces the measurement index of FPRB with the cost of self-purchased preventive drugs by farmers, and uses two-stage lease squares (2SLS) and Extended regression model (ERM) to estimate. Referring to the research of Su and Cao (51), this paper selects the spherical distance from Hangzhou to each city as a tool variable for ECA. From the correlation point of view, this tool variable is closely related to ECA. As the birthplace of e-commerce in China, generally speaking, the closer to Hangzhou, the greater the radiation-driven effect, and the greater the probability of ECA for farmers in this area; from the exogenous point of view, it is difficult for this tool variable to directly affect FPRB. Even if it has an impact on FPRB, it is often through the channel of ECA. Therefore, the tool variables selected in this paper are reasonable.

In this paper, 2SLS model is used to test the weak instrumental variables. The first-stage regression results show that the impact of instrumental variables on farmers’ ECA is significantly negative at the statistical level of 1%, and the f-value of joint significance test is greater than 10, indicating that there is no problem of weak instrumental variables. The test values of Wald and Corr in Table 6 show that there are endogenous problems in the influence of e-commerce on FPRB. After introducing instrumental variables, ECA still has a significant negative impact on farmers’ pesticide application behavior, indicating that ECA can significantly reduce pesticide application and promote pesticide reduction.

Referring to the research of Tian et al. (52), in this paper, farmers over 65 years old are defined as elderly farmers, and those under 65 years old are defined as non-elderly farmers. With junior high school education as the dividing line, those who have received education above junior high school are defined as having a high educational attainment, and those who have received education below junior high school are defined as having a low educational attainment, and investigates the differences in the influence of ECA on FPRB under different ages and education levels. From the regression results in Table 8, it can be seen that the influence of ECA on FPRB is heterogeneous in age and education level. From the perspective of age heterogeneity, the pesticide reduction effect of ECA is more obvious for elderly farmers. The reason is that under the influence of urbanization, a large number of young and middle-aged laborers flow to cities, and most of them are elderly farmers engaged in agricultural production in rural areas. Under the influence of rural e-commerce, this group is more likely to reduce pesticide application. Judging from the heterogeneity of education level, compared with farmers with high education level, the pesticide reduction effect of ECA of farmers with low education level is more obvious. The possible reason is that farmers with high education level have more employment choices, prefer to go out to work rather than engage in agricultural production and management, and farmers with low education level often face greater economic pressure. Increasing income is their main behavioral motivation. Reducing the use of pesticides can not only reduce production costs, but also obtain the premium of sustainable products brought by e-commerce.

This paper explores the heterogeneous influence of ECA on FPRB under different family political identities and family cultivated land scale. According to whether there is party member in the family, the family political identity is regressed in groups, and the family cultivated land scale is regressed in groups with one hectare as the dividing line. From the regression results in Table 9, it can be seen that the influence of ECA on FPRB is heterogeneous in family political identity and family cultivated land scale. Judging from the heterogeneity of family political identity, the pesticide reduction effect of ECA is more obvious for farmers with party member at home. The possible reason is that party member usually has a strong sense of social responsibility and environmental protection, and it is more likely to actively respond to the national policy call for pesticide reduction, and publicize the concept of environmental protection and sustainable production to family members, and guide family members to practice pesticide reduction in agricultural production. Judging from the heterogeneity of family cultivated land scale, compared with farmers with small cultivated land scale, farmers with large cultivated land scale have more obvious pesticide reduction effect. The possible reason is that farmers with large cultivated land often have stronger ability to adopt advanced technologies, such as drone spraying and physical control. This kind of advanced technology can not only directly improve the efficiency of pesticide application, but also fundamentally reduce the probability of pests and diseases by optimizing the production environment, thus reducing the dosage of pesticides.

This paper explores the heterogeneous influence of ECA on FPRB under different agricultural functional divisions and geographical locations. Based on whether the province where the farmers are located is a major grain-producing area, the agricultural functional zoning is grouped and returned, and the geographical location is grouped and returned with the Hu Huanyong Line as the dividing line. According to the regression results in Table 10, the influence of ECA on FPRB is heterogeneous in agricultural function zoning and geographical location. Judging from the heterogeneity of agricultural function zoning, the pesticide reduction effect of farmers participating in ECA in major grain producing areas is more obvious. The possible reasons are that agricultural production activities in major grain producing areas are more concentrated, pesticide reduction technology is more popular, farmers are more likely to obtain premium of high-quality agricultural products from e-commerce, and their willingness to reduce pesticides is higher. From the perspective of geographical heterogeneity, compared with farmers located in the west of Hu Huanyong Line, farmers located in the east of Hu Huanyong Line have more obvious pesticide reduction effect when they participate in ECA. The possible reason is that the economic development level in the area east of Hu Huanyong Line is higher, and consumers’ demand for sustainable agricultural products is stronger. Farmers are more likely to improve the quality of products through scientific application of pesticides and reducing the application amount of pesticides, and then obtain high income.

Influenced by resource endowment, technology application ability and market orientation, the influence of different e-commerce scale on FPRB may be different. Based on the average sales of e-commerce, this paper divides the scale of e-commerce of farmers into scale e-commerce and non-scale e-commerce, and makes regression analysis. According to the regression results in Table 11, compared with non-scale e-commerce, the pesticide reduction effect of scale e-commerce is more obvious. The possible reason is that scale e-commerce operators usually have stronger resource allocation ability and capital advantage, and can adopt more sustainable production technologies and equipment to achieve pesticide reduction. At the same time, scale e-commerce operators have stronger bargaining power and brand influence in the market, and can obtain positive incentives for pesticide reduction through the “high quality and good price” mechanism.

With the booming of e-commerce industry, rural e-commerce presents a diversified development trend. At present, rural e-commerce mainly includes two modes: platform e-commerce and social e-commerce (53). Platform e-commerce mode mainly means that farmers set up online stores by entering JD.COM, Taobao, Pinduoduo and other well-known large-scale e-commerce platforms, and use the traffic and resources of these platforms to push agricultural products to a broader market (54). The social e-commerce mode mainly refers to farmers using the social attributes and communication advantages of social platforms such as WeChat, Xiaohongshu and Tik Tok to directly promote and sell agricultural products to consumers.

Considering the heterogeneity of operating rules and market response mechanism of different e-commerce models, this paper further explores the differences in the influence of different e-commerce models on FPRB. According to the regression results in Table 12, compared with the platform e-commerce model, the pesticide reduction effect of social e-commerce model is more obvious. The possible reason is that compared with platform e-commerce, which only provides an online trading platform for consumers and businesses to trade directly, social e-commerce focuses more on using social networks to promote and sell goods. With the help of social media platform, farmers recommend products to potential consumers through user sharing, likes and comments, and realize word-of-mouth marketing and spontaneous communication of users. In order to ensure product quality and reputation, farmers will pay more attention to the quality of agricultural products, thus reducing the use of pesticides.

At present, pesticide application methods in China are mainly divided into mechanical pesticide application and manual pesticide application. In terms of mechanical pesticide application, new types of plant protection machinery such as agricultural drones and self-propelled boom sprayers are gradually becoming popular. They have high operational efficiency, high pesticide utilization rate, and can effectively reduce pesticide drift and environmental pollution. However, the promotion of these new machines still faces the cost and technical threshold, and it is difficult for some small farmers to widely apply them. In the aspect of manual application of pesticides, small manual sprayers still dominate, but this method has high labor intensity, low efficiency and low effective deposition rate of pesticides, which is easy to cause problems such as excessive pesticide residues and environmental pollution.

The emergence of e-commerce provides a new path for optimizing agricultural pesticide application methods. The information transparency mechanism and convenience of e-commerce platform can help farmers to obtain new plant protection machinery and sustainable pesticide information more directly, reduce procurement costs, and provide accurate guidance for the promotion of pesticide reduction technology through data analysis and user feedback mechanism. Therefore, this paper further discusses the influence of ECA on pesticide application methods. From the regression results in Table 13, it can be seen that ECA has promoted farmers to apply pesticides mechanically. The possible reason is that e-commerce helps farmers to grasp the market demand more accurately. In order to enhance the market competitiveness, farmers tend to adopt more scientific and reasonable mechanical application methods to reduce labor input costs, improve the uniformity and accuracy of application, and thus ensure the stability and high quality of crop output. This result also confirms the pesticide reduction effect of ECA from the side.