Can the Adoption of Green Pest Control Technologies Reduce Pesticide Use? Evidence from China

Abstract

The widespread use of pesticides has long been a cornerstone of modern agriculture, but their overuse has led to several challenges, including increased production costs, food safety risks, and environmental damage. Green pest control technologies (GPCTs) have emerged as a promising alternative to traditional chemical methods, although their widespread adoption is still in progress, and their environmental and economic impacts require further examination. This study aims to evaluate the adoption of GPCT in apple orchards by employing a rigorous framework to measure pesticide intensity per unit, assess the impact of GPCT on pesticide reduction, and analyze the associated environmental effects in large-scale apple farming systems in China. Based on survey data collected from apple farmers across key production regions in China, we apply an Endogenous Treatment Effect Regression (ETR) model to evaluate the effects of these technologies on pesticide usage and concentration. This model allows for more accurate estimates of the treatment effects by addressing selection bias and accounting for both observable and unobservable factors. Our results show that the adoption of GPCT leads to a significant reduction in pesticide use intensity. Notably, the reductions are more pronounced among specific groups of farmers, particularly those who are less risk-averse and those with larger or more fragmented landholdings. These findings underscore the dual ecological and economic benefits of GPCT, providing strong support for policy initiatives that promote sustainable agricultural practices and encourage land consolidation.

1. Introduction

Chemical pesticides have long played a critical role in addressing global food security challenges [1]. However, despite their inefficiency, their widespread use continues due to factors such as limited access to alternatives, risk aversion, and the entrenched nature of traditional pest management practices. Over time, the prolonged and excessive application of these chemicals has led to numerous adverse consequences, including escalating agricultural production costs, recurring food safety concerns, and the degradation of biodiversity [2]. To address these issues, a shift towards green agricultural practices has become increasingly important for achieving sustainable development in agriculture [3]. Central to this transition is GPCT, which offers effective solutions to reduce pesticide use, safeguard agricultural productivity, improve food quality, protect ecosystems, and promote sustainability [4]. Sustainable agricultural development, in this context, refers to farming practices that balance the need for increased food production with the protection of ecological systems and long-term economic viability [5]. GPCT aligns with these goals by reducing reliance on harmful chemicals and enhancing ecological resilience [6]. These technologies hold significant potential to mitigate pesticide use while minimizing their environmental impact [7].

GPCT has emerged as a promising solution in sustainable agriculture, aiming to reduce reliance on chemical pesticides and mitigate their harmful environmental effects [2]. These technologies include methods such as ecological control, where pest-resistant crop varieties are selected, biological control, which uses natural predators or biological agents, physical control, where pest attraction to light or color is used to trap them, and scientific pesticide use, where low-toxicity pesticides are applied in a targeted and efficient manner [6]. By reducing pesticide use, GPCT contributes not only to environmental protection but also to long-term agricultural sustainability, addressing both ecological and economic challenges associated with conventional pesticide application [7].

Despite progress in understanding the ecological effects of GPCT, there remains a lack of consensus on appropriate measurement indicators for pesticide reduction. Existing studies typically evaluate pesticide reduction from two main approaches: relative and absolute quantities [8]. The relative quantity approach compares pesticide use among farmers and their neighbors over different time periods, or against recommended pesticide application guidelines [5]. In contrast, the absolute quantity approach focuses on metrics such as per-acre pesticide costs [1], application volumes [8], and frequency of pesticide applications per acre. Some studies combine both approaches [2]. However, many studies fail to account for the issue of endogeneity in regression analyses. While some studies have addressed selection bias arising from observable factors, their analyses typically focus on estimating Average Treatment Effect (ATT) values rather than directly evaluating the impact of GPCT on pesticide reduction [4]. ATT refers to the average effect of adopting GPCT on the treated group of farmers, specifically estimating the impact of this adoption on pesticide use. However, this approach does not provide a direct assessment of the overall impact on pesticide reduction, as it only estimates the effect for those who have already adopted the technology.

The existing literature provides valuable insights but also reveals notable limitations, especially in measuring pesticide use. Relative quantity assessments often overlook individual variations between farmers, while absolute quantity assessments face challenges related to inaccurate recall of pesticide types and quantities, inconsistencies in pesticide-to-water ratios, and differences in application frequencies. These factors can introduce significant biases. Since farmers are generally more accurate in recalling input costs and labor efforts, indicators such as total pesticide costs, biopesticide quantities and costs, and pesticide solution volumes may provide more reliable measures for assessing pesticide reduction.

The primary objectives of this study are to evaluate the adoption of GPCT in apple orchards, measure pesticide intensity per unit, assess the impact of GPCT on pesticide reduction, and analyze the environmental effects associated with GPCT implementation in large-scale apple farming systems in China. It incorporates heterogeneity analysis to identify the characteristics of farmers who benefit the most from adopting these technologies. Grounded in the damage-control production function, the study establishes a theoretical framework to assess the effects of adopting GPCT on pesticide use in apple orchards. Drawing on 2022 survey data collected from apple farmers in two major production regions in China—the Jiaodong Peninsula and the Yimeng Mountains in Shandong Province—we analyze three key variables related to pesticide use: pesticide application intensity, chemical pesticide application intensity, and chemical pesticide concentration.

• Pesticide application intensity refers to the expenditure on pesticides per Mu within the orchard, measured in CNY/Mu. This variable captures the total spending on pesticides, which includes both chemical and biopesticides, and serves as an indicator of the overall investment in pest control.
• Chemical pesticide application intensity specifically refers to the expenditure on chemical pesticides per Mu within the orchard, measured in CNY/Mu. This variable isolates the spending on chemical pesticides, excluding biopesticides, and reflects the farmer’s reliance on synthetic chemicals for pest control.
• Chemical pesticide concentration refers to the expenditure on chemical pesticides per half kilogram of diluted solution, following the prescribed dosage, measured in CNY/half kilo. This variable indicates the strength of the pesticide mixture applied to the orchard, highlighting the concentration of chemicals in the pesticide solution and its potential environmental and health impacts.

These three variables are distinct but complementary. While pesticide application intensity reflects the total pesticide expenditure, chemical pesticide application intensity focuses on chemical pesticide spending, and chemical pesticide concentration provides insights into the potency of the pesticide solution used.

This study is innovative in its empirical assessment of GPCT, focusing on both total pesticide input and chemical pesticide input. By examining the impact of these technologies on pesticide application intensity and solution concentration, it provides new insights into the ecological and economic benefits of green pest control. Additionally, the use of the Endogenous Treatment Effect Regression (ETR) model allows for a more accurate evaluation of the direct effects of green pest control adoption, addressing selection bias in a way that previous studies have not. The findings of this study provide valuable theoretical and practical insights to support the promotion of green production technologies and foster agricultural sustainability.

2. Theoretical Analysis and Research Hypotheses

Unlike productive inputs such as improved seeds and fertilizers, pesticides function primarily as protective inputs. Their primary role is to mitigate potential losses caused by pests and diseases, thereby safeguarding crop yields rather than directly enhancing potential yields [9]. Building on prior studies [8,10], this research adopts the damage control production function to develop a theoretical framework for examining the impact of GPCT on farmers’ pesticide use. The proposed framework involves three key steps: (1) constructing the damage control production function, (2) identifying the factors that influence farmers’ optimal pesticide application intensity, and (3) elucidating the mechanisms through which GPCT facilitates farmers in achieving optimal pesticide application intensity.

First, a damage control production function is constructed based on the process through which pesticides influence crop yields. Pesticides represent a unique production input whose effectiveness depends on the presence of pests and diseases. Since the traditional Cobb–Douglas (C–D) production function does not incorporate the occurrence of pest infestations, it often overestimates the marginal productivity of pesticides [11]. The damage control function, therefore, typically employs variables such as output value and input cost rather than output volume and input quantity. This approach is justified by the significant variation in pesticide types and specifications (e.g., pesticides for disease control, insecticides, etc.), as well as the differing units of measurement across pesticide categories. Consequently, farmers often find it challenging to precisely quantify the amount of pesticide used, relying instead on pesticide costs, which are easier to monitor and manage.

The following equations are used to model the relationship between pesticide application intensity, pest population, and crop yield. Equation (1) represents the pest damage control production function, let $B$ is the actual number of pests, and $B_0$ denote the number of pests in the absence of pesticide control, and let $K(T)$ represent a function describing the number of pests under pesticide control as influenced by the pesticide application intensity $T$. The effectiveness of pesticides in reducing pest numbers increases with application intensity $T$, such that $\partial K(T)/\partial T>0$. Consequently, the pest damage control production function can be expressed as:

$$B=B_0[1-K(T)]$$

(1)

From Equation (1), we can infer the following: when $T\to \infty$, $K(T)\to 1$, which indicates that as the pesticide application intensity becomes sufficiently large, the pest population approaches zero. Conversely, when $T=0$, $K(0)=0$ indicating that in the absence of pesticide application, the pest population remains at its initial level $B_0$.

Assuming the actual crop yield is $Q$, and the potential yield is $F(q)$, where $\delta$ represents the proportion of yield loss caused by pests, the role of pesticides is to suppress pests, diseases, and weeds, thereby enabling the realization of the potential yield. Here, $q$ represents other input factors, such as capital and labor. The impact of pests on yield is modeled using a damage function $D(B)$, where $B$ represents the pest population. Consequently, the pest damage control production function can be expressed as:

$$Q=(1-\delta )F(q)+\delta F(q)[1-D(B)]$$

(2)

Typically, the impact of the pest population on crop yield increases with the growth of the pest population. Therefore, we can assume that $\partial D(B)/\partial B>0$. Furthermore, from Equation (2), we know that when the pest population is zero, $D(B)=0$, and thus $Q=F(q)$, indicating that the actual crop yield reaches its potential maximum. Conversely, when the pest population is sufficiently large, $D(B)=1$, and therefore $Q=(1-\delta )F(q)$, meaning the actual crop yield is minimized. By substituting Equation (1) into Equation (2), we derive the pest damage control production function, which incorporates pesticide inputs:

$$Q=(1-\delta )F(q)+\delta F(q)\{1-D[B_0(1-K(T))]\}$$

(3)

Assume that $G(T)$ represents the probability distribution of $D[B_0(1-K(T))]$. Although there is abundant research adopting various functions for damage assessment, variations exist in the assumptions regarding their functional forms. Commonly used distributions include Exponential, Pareto, Weibull, and lognormal distributions. These distributions are selected for their empirical suitability in quantifying damage effects, as demonstrated in both direct and indirect approaches [12].

Second, based on the objective of profit maximization, we analyze the factors influencing farmers’ optimal pesticide application intensity. Let the price of agricultural products be $p$, the price of input factors $q$ (other than pesticides, e.g., labor, capital) be $w$, and the price of pesticide application intensity (i.e., chemical pesticide input) $T$ be $r$. The farmer’s profit function can then be expressed as Equation (4).

It is important to note that farmers may have varying perceptions of the effectiveness of pesticides in controlling pests. Therefore, both $K(T)$ and $\delta$ must account for heterogeneity across farmers. To enhance the model’s realism, we adjust $K(T)$ and $\delta$ to $K_i(T)$ and ${\delta }_i$, where $i$ represents the $i-th$ farmer. Consequently, the farmer’s profit function is adjusted accordingly and expressed as Equation (5).

$$\pi =pQ-wq-rT$$

(4)

$$\begin{array}{l}\pi =p[(1-\delta )F(q)+\delta F(q)G(T)]-wq-rT\\ =p[(1-{\delta }_i)F(q)+{\delta }_{i}F(q)G_i(T)]-wq-rT\end{array}$$

(5)

To determine the optimal pesticide application intensity $T^{*}$ for farmers, this study assumes that the functional form of $G_i(T)$ follows an Exponential model, i.e., $G_i(T)=1-\exp (-m_{i}T)$, where $m_i$ represents the $i-\mathrm{th}$ farmer’s perception of the effectiveness of pesticide use. Substituting this into Equation (5) and differentiating with respect to $G_i(T)$ (i.e., $\partial \pi (T)/\partial T>0$), we can derive the decision condition for the optimal pesticide application intensity, as shown in Equation (6). Taking the natural logarithm of both sides of Equation (6), we obtain the expression for the farmer’s optimal pesticide application intensity, as shown in Equation (7).

$$p\partial {\delta }_{i}F(q)G_i(T)/\partial T=r$$

(6)

$$T^{*}=\{\ln [{\delta }_i{m}_{i}pF(q)]-\ln r\}/m_i$$

(7)

The relationship among $T^{*}$, ${\delta }_i$ and $m_i$ is evident. Specifically, as shown in Equation (7), $T^{*}$ and ${\delta }_i$ exhibit a positive correlation, indicating that as ${\delta }_i$ increases, farmers are more inclined to raise both pesticide application intensity and pesticide usage levels. Furthermore, given the functional form of $G_i(T)$, there is an inverse relationship between $T^{*}$ and $G_i(T)$. Taking the partial derivative of $G_i(T)$, $\partial G_i(T)/\partial m_i=m_{i}T/\exp (m_{i}T)>0$, demonstrates a positive correlation between $G_i(T)$ and $m_i$, which implies an inverse relationship between $T^{*}$ and $m_i$. In other words, as $m_i$ decreases, farmers are more likely to increase their pesticide application intensity and pesticide usage levels.

Notably, assuming $p$ and $r$ are exogenous variables, the primary determinants of farmers’ optimal pesticide application intensity are $m_i$ and ${\delta }_i$. Therefore, factors influencing ${\delta }_i$ and $m_i$ directly affect farmers’ decisions regarding optimal pesticide application intensity.

Third, based on the factors affecting pest-induced yield loss rates and farmers’ recognition of pesticide effectiveness, we clarify the mechanism by which the adoption of GPCT facilitates reductions in pesticide application intensity and concentration. Currently, pesticide application remains the primary method of pest control in crop production. However, this approach has resulted in increased pest resistance, imposing greater demands on control methods, pesticide types, application timing, and techniques. Farmers often rely heavily on experiential knowledge for pest control, which limits the precision of targeted applications and consequently reduces $m_i$. Furthermore, pest control is time-sensitive, and the mobility of pests allows surrounding fields to provide new habitats, leading to increased and ultimately resulting in the overuse of pesticides by farmers.

In contrast to apple growers who have not adopted GPCT, those who have adopted these technologies improve $m_i$ through the use of scientific pesticide application methods and biological control techniques, thereby reducing the intensity of chemical pesticide use [13]. Additionally, through the adoption of non-chemical control methods, these farmers decrease the scope and severity of pest damage [14], thereby lowering ${\delta }_i$. This ensures both the yield and quality of apples while partially substituting for chemical pesticides, further reducing pesticide application intensity. Finally, by adopting ecological control technologies, farmers enhance the disease and pest resistance of apple trees, thereby further reducing ${\delta }_i$ and supporting yield and quality preservation [15].

Through these mechanisms, the adoption of GPCT reduces both the absolute quantity and relative concentration of pesticide application within the ecosystem. Accordingly, this study proposes the following research hypotheses:

H1. 

The adoption of GPCT significantly reduces both the intensity and concentration of pesticide and chemical pesticide applications per unit area in farmers’ orchards.

H2. 

Different types of GPCT exert distinct and significant negative effects on the intensity and concentration of pesticide applications per unit area in farmers’ orchards.

3. Materials and Methods

3.1. Model Construction

This study adopts an empirical approach to test the proposed hypotheses, aiming to validate the logical framework established in the theoretical analysis. By selecting and constructing appropriate models, we seek to verify whether there exists a stable, reliable, and demonstrable causal relationship between the adoption of GPCT and pesticide use intensity. Among the potential econometric methods, the Endogenous Treatment Effect Regression (ETR) model was ultimately chosen as the benchmark model for this study due to its ability to address selection bias and estimate causal effects accurately. The rationale for this choice is detailed below.

There are two primary approaches to quantifying causal effects: parametric and non-parametric methods. Among non-parametric methods, the Inverse Probability Weighted Regression Adjustment (IPWRA) is widely used. However, IPWRA is constrained by its susceptibility to selection bias arising from unobservable factors, which limits its ability to provide accurate estimates in certain settings [16].

In contrast, the parametric approach, represented by the Endogenous Switching Regression (ESR) model, offers broader applicability and is particularly effective in addressing selection bias caused by unobservable factors. However, the ESR model requires strict assumptions about the functional form of the outcome variables, which can sometimes be difficult to justify in practice. Moreover, both IPWRA and ESR belong to the category of indirect evaluation methods and primarily assess effects through ATT (Average Treatment Effect on the Treated) values [17], which are limited in providing a direct effect evaluation. This is a significant limitation because ATT values primarily estimate the effect of the treatment (in this case, the adoption of GPCT) on the treated group but do not provide direct causal inference for the entire population.

To more accurately estimate the relationship between the adoption of GPCT and farmers’ income, this study employs an Endogenous Treatment Effect Regression (ETR) model, which is an advanced parametric approach. The ETR model addresses the selection bias problem by directly estimating the effect of the treatment on the outcome, thereby providing a more direct evaluation compared to ATT-based methods. This capability makes the ETR model a more accurate and reliable tool for causal analysis in this context [18].

The regression process of the ETR model consists of two stages:

• Selection equation regression: The first stage examines the relationship between individual and household variables and the decision to adopt GPCT. It uses a binary choice model (e.g., logit or probit model) to determine the likelihood of adoption based on observable characteristics.
• Outcome equation regression: The second stage estimates the direct effect of adopting GPCT on pesticide application intensity among apple growers while controlling for endogeneity (i.e., the possibility that unobserved factors affect both adoption and pesticide use). This stage uses a continuous outcome model (e.g., linear regression) to quantify the impact of the treatment on the outcome variable.

The Endogenous Treatment Effect Regression (ETR) model, although commonly used in causal inference and econometric analysis, is particularly well suited for this study due to its flexibility in estimating direct treatment effects, even when the assumptions required by other models, such as the Endogenous Switching Regression (ESR) model, are not fully satisfied. While the ESR model is useful for addressing selection bias, it relies on assumptions about outcome distributions that may not always hold, especially when there is heterogeneity in treatment effects. In contrast, the ETR model provides a more robust framework by accounting for both observable and unobservable factors influencing the adoption decision, allowing for more accurate estimates of the impact of GPCT on pesticide use [19,20].

Although the ETR model is widely applied in other contexts, its adaptation to studying GPCT in agricultural settings—particularly in China—represents an innovative approach. This adaptation addresses the unique challenges of assessing the environmental impact of these technologies and contributes to the existing literature by offering new insights into their effects on pesticide use.

Specifically, the first and second stages are represented by the following equations:

$$GPC{T}^{*}={\delta }_i{X}_i+{\gamma }_i{I}_i+{\epsilon }_i,O_i=\left(\begin{array}{l}1,if\left(GPC{T}^{*}>0\right)\\ 0,if\left(GPC{T}^{*}\le 0\right)\end{array}\right.$$

(8)

$$Y^{*}={\alpha }_{i}GPC{T}_i+{\beta }_i{X}_i+{\mu }_i$$

(9)

Equation (8) represents the selection equation, where $GPC{T}_i^{*}$ denotes the decision to adopt GPCT. This is a binary choice variable associated with a random utility model, representing the difference in utility between apple growers who adopt GPCT ($U_{iU}$) and those who do not ($U_{iN}$). Specifically, if $GPC{T}_i^{*}=U_{iU}-U_{iN}>0$, then $GPC{T}_i=1$, indicating that the grower has adopted GPCT. Conversely, if $GPC{T}_i^{*}=U_{iU}-U_{iN}\le 0$, then $GPC{T}_i=0$, indicating non-adoption of GPCT.

Equation (9) represents the outcome equation, where $Y_i$ denotes the pesticide application in the orchard, including both pesticide application intensity and concentration. $X_i$ represents factors influencing the adoption of GPCT and pesticide application within the orchard. ${\delta }_i$, ${\gamma }_i$, ${\alpha }_i$ and ${\beta }_i$ are parameters to be estimated, with ${\alpha }_i$ representing the direct effect of adopting GPCT on pesticide application. ${\epsilon }_i$ and ${\mu }_i$ are random error terms, and $I_i$ is an instrumental variable.

It is important to note that the ETR model utilizes the full-information maximum likelihood estimation method to jointly estimate the correlation coefficient between the error terms ${\epsilon }_i$ and ${\mu }_i$. This approach integrates the selection bias from the first stage into the second, effectively addressing selection bias caused by both observable and unobservable factors, thereby alleviating endogeneity issues. Additionally, the ETR model reports the Wald test statistic for independence, which measures the correlation between the selection and outcome equations.

In addition to analyzing the overall impact of GPCT adoption, this study further explores its heterogeneous effects across different farmer characteristics, such as risk preferences, cultivated area per laborer, and degree of land fragmentation. These heterogeneous effects are particularly important for understanding which groups of farmers benefit most from adopting GPCT, and they provide valuable insights for tailoring policy recommendations. The methodology for this heterogeneity analysis is described in detail in Section 3.4.

3.2. Variable Selection

3.2.1. Outcome Variables

The primary outcome variable in this study is the pesticide application per unit area within farmers’ orchards. As previously noted, the existing literature on farmers’ pesticide application presents diverse conclusions, largely due to variations in the definition and measurement of pesticide use. Building on prior research, this study selects three variables to represent pesticide application within orchards: pesticide application intensity (encompassing both chemical and biological pesticides), chemical pesticide application intensity, and chemical pesticide concentration.

This study assumes that apple growers do not significantly differ in their choice of pesticide types and prices, using “pesticide expenditure” as a proxy for “pesticide application intensity”. Accordingly, the variable for pesticide application intensity (including biological pesticides) refers to the average expenditure on pesticides per Mu (CNY/Mu) during the apple-growing process. Similarly, the variable for chemical pesticide application intensity represents the average expenditure on chemical pesticides per Mu (CNY/Mu).

For the chemical pesticide concentration variable, this study defines it as the concentration of chemical pesticides in the pesticide solution when diluted according to prescribed dosages. This is measured by the “amount of chemical pesticide per half kilo in the diluted solution according to prescribed dosage” (with the expenditure on pesticides used as a proxy for this amount). The derivation process is as follows:

$$P_i=T_i/V_i^{*}={\displaystyle \frac{K_i/r}{V_i/{\alpha }_i}}$$

(10)

In Equation (10), $i$ represents the $i-\mathrm{th}$ apple grower; $T_i$ denotes the amount of chemical pesticide applied; $V_i^{*}$ is the volume of water used to dilute the chemical pesticide; $K_i$ represents the input of chemical pesticides; $r$ is the average price per unit of chemical pesticide; $V_i$ is the actual water volume used for dilution; and ${\alpha }_i$ is the dilution coefficient, which is 100% when the pesticide is diluted according to the prescribed dosage.

As shown, a larger $T_i/V_i^{*}$ results in a lower concentration of chemical pesticide in the solution when prepared according to the recommended dosage. Therefore, a larger $K_i$ corresponds to a weaker chemical pesticide concentration, while a smaller $V_i$ corresponds to a stronger concentration. In this context:

• Total pesticide input and chemical pesticide input represent pesticide application intensity, capturing both the economic cost and the quantity of toxic chemicals applied. These variables reflect both economic and environmental dimensions.
• Chemical pesticide concentration per Mu reflects the level of toxic chemicals introduced into the environment. Higher concentrations are more likely to directly harm the local ecosystem or leave residues in soil and water sources. These residues can potentially accumulate through the food chain, posing risks to the entire food web including humans, and threatening the broader ecosystem. Consequently, this metric primarily captures the environmental implications of adopting GPCT.

3.2.2. Treatment Variable

The treatment variable represents the decision to adopt GPCT and is defined as a binary choice variable. This study defines the variable as follows: if the surveyed apple grower adopted at least one GPCT in apple cultivation in 2020, the variable is assigned a value of 1, categorizing the grower in the treatment group. Conversely, if no GPCT was adopted, the variable is assigned a value of 0, placing the grower in the control group.

3.2.3. Control Variables

Drawing on the work of other researchers [21,22,23], this study identifies the following control variables:

• Individual characteristics of respondent farmers: This category includes five variables: age, education level, village leadership role, agricultural training, and social connections (e.g., with friends or neighbors).
• Household production and management characteristics: This category comprises four variables: years of apple cultivation, per capita apple farming area, percentage of income from non-apple sources, and other relevant factors.

Additionally, to account for regional characteristics and economic conditions, the model incorporates a set of regional dummy variables to control for variations across different areas.

3.2.4. Instrumental Variable

The instrumental variable used in this study is the distance to the nearest demonstration orchard employing physical pest control technologies, such as insect-proof nets, sticky traps, and insecticidal lamps. According to the model specifications, measurement errors or omitted variables may introduce endogeneity issues in the empirical regression process, necessitating the use of an instrumental variable. This study selects the distance to the nearest demonstration orchard utilizing physical control technologies, phrased as “the distance between your household and the nearest demonstration orchard using physical control technologies”, as the instrumental variable. This variable significantly influences the decision to adopt GPCT but is exogenous to pesticide application intensity within the orchard, making it a valid and appropriate instrumental variable.

3.3. Data Source

The micro-level data for this study were collected in 2022 through a questionnaire survey targeting apple farmers in the primary apple-producing regions around the Bohai Sea. The surveyed areas included Muping District, Penglai District, and Qixia City in Yantai, Shandong Province, as well as Mengyin County and Yishui County in Linyi. The survey employed one-on-one questionnaire interviews and utilized stratified sampling to account for regional economic development levels, agricultural production scales, and geographical characteristics.

In each county or district, two to three townships were selected, followed by the selection of two to three administrative villages within each township. Within each village, 10 to 20 apple farmers were randomly chosen for one-on-one interviews and questionnaire surveys. A total of 475 questionnaires were distributed, of which 409 were valid, yielding an effective response rate of 86.11%.

3.4. Methodology for the Impact of Various Sub-Technologies

To assess the impact of various sub-technologies on farmers’ pesticide application levels, this study employs the Ordinary Least Squares (OLS) regression method with robust standard errors. This analytical approach allows for the estimation of linear relationships between pesticide application levels and the influencing factors while accounting for potential heteroskedasticity in the data.

Ordinary Least Squares (OLS) regression is utilized to estimate the relationship between the dependent variable (farmers’ pesticide application levels) and multiple independent variables (sub-technologies and other factors). OLS aims to minimize the sum of the squared residuals, providing the best linear unbiased estimates (BLUE) of the coefficients under the assumption of homoscedasticity and no autocorrelation.

To address potential heteroskedasticity, robust standard errors are calculated. This adjustment ensures that the standard error estimates remain reliable even when the variance of the error terms is not constant across observations. By implementing robust standard errors, the analysis maintains the validity of hypothesis tests and confidence intervals for the estimated coefficients.

The regression model is specified as follows:

$$P{A}_i={\beta }_0+{\displaystyle \sum _n{\beta }_{n}Sub-Te{c}_n}+{\displaystyle \sum _m{\beta }_{m}Contro{l}_m}+{\nu }_i$$

(11)

In Equation (11), $P{A}_i$ represents the pesticide application behavior of the $i-\mathrm{th}$ farmer. The variable $Sub-Te{c}_n$ indicates whether the $n-\mathrm{th}$ sub-technology has been adopted by the farmer, where $Sub-Te{c}_n=1$ signifies the adoption of the specific sub-technology. $Contro{l}_m$ denotes the $m-\mathrm{th}$ control variable included in the model to account for other factors influencing pesticide application behavior. The parameters ${\beta }_0$, ${\beta }_n$, ${\beta }_m$ are the coefficients to be estimated, representing the impact of each respective independent variable on $P{A}_i$. Lastly, ${\nu }_i$ is the random error term, capturing the variability in pesticide application behavior that is not explained by the model.

3.5. Methodology for Heterogeneous Effects Analysis

This study conducts a heterogeneous effects analysis to explore how GPCT adoption impacts different farmer groups. Subgroups were categorized based on three key variables: risk type, average cultivated area per laborer, and degree of land fragmentation. Interaction terms were included in the analysis to capture differential impacts across these groups, and robust standard errors were applied to address potential heteroskedasticity.

The key variables used for subgroup categorization are defined as follows:

• Risk type: This variable reflects household decision-making characteristics and distinguishes between risk-averse and non-risk-averse farmers.
• Average cultivated area per laborer: This variable represents labor endowment characteristics and is calculated as the total cultivated area divided by the number of household laborers.
• Degree of land fragmentation: This variable measures the operational complexity of orchard management and is defined as the average number of orchard plots per mu.

To measure risk type, a game-based approach was incorporated into the survey questionnaire. This method utilized real monetary incentives to elicit respondents’ risk preferences. Specifically, respondents were presented with the following two options:

• Option 1: Receive a guaranteed cash reward of CNY 5.
• Option 2: Participate in a game with uncertain outcomes but an expected reward of CNY 5.

The respondents’ choices determined their classification: those selecting Option 1 were classified as risk-averse, while those selecting Option 2 were classified as non-risk-averse. The cash rewards were implemented in real-time to ensure the authenticity and reliability of responses.

This approach provided a simple yet effective means of quantifying risk preferences, offering a robust measure directly relevant to decision-making in agricultural contexts. These variables were essential for identifying heterogeneous effects and tailoring the analysis to the diverse characteristics of farmer groups.

3.6. Descriptive Statistical Analysis

3.6.1. Variable Definitions and Descriptive Statistics

The variables, definitions, and descriptive statistics for this study are summarized in

Table 1. Variable definitions and descriptive statistics.

Variable Name	Variable Definition	Mean	Standard Deviation
Outcome Variables
Pesticide Application Intensity	Expenditure on pesticides per Mu within the orchard (CNY/Mu)	1054	378
Chemical Pesticide Application Intensity	Expenditure on chemical pesticides per Mu within the orchard (CNY/Mu)	948.1	407.3
Chemical Pesticide Concentration	Expenditure on chemical pesticides per half kilo of diluted solution following prescribed dosage (CNY/half kilo)	3.342	4.749
Treatment Variable
Adoption Decision for Pest Control Technology	Adoption of GPCT in apple cultivation: Yes = 1; No = 0	0.566
Control Variables
Age	Actual age (years)	55.59	10.09
Education Level	Years of education	8.299	2.881
Village Leadership	Household head’s status as village official: Yes = 1; No = 0	0.0695	0.255
Agricultural Training	Number of agricultural training sessions attended by household members in the past three years	2.156	1.611
Social Connections	Interaction with relatives, friends, and neighbors: Never = 1; Rarely = 2; Occasionally = 3; Often = 4; Frequently = 5	3.931	0.931
Years of Cultivation	Years of apple cultivation	21.64	11.87
Operational Area	Apple farming area (mu)	6.289	6.755
Number of Laborers	Number of laborers aged between 18 and 65	2.487	0.9
Proportion of Non-Apple Income	Proportion of total household income from non-apple sources	0.21	0.244
Regional Dummy Variable	Yantai = 1; Linyi = 0	0.417	0.494
Instrumental Variable
Distance to Demonstration Orchard	Distance to the nearest demonstration orchard using technologies like insect-proof nets, sticky traps, and insecticidal lamps (km)	5.749	4.132

Data source: Compiled based on survey data.

. Analysis of the table reveals the following key findings: among the surveyed apple farmers, 269 (approximately 56.6% of the total sample) adopted GPCT, while 206 farmers (about 43.4%) did not. The average pesticide expenditure per Mu among the surveyed farmers is 1054 CNY/Mu (approximately 896.67 USD/acre), with a maximum of 2040 CNY/Mu (approximately 1730.15 USD/acre). For chemical pesticides specifically, the average expenditure per Mu is 948.1 CNY/Mu (approximately 805.18 USD/acre), with a maximum of 2000 CNY/Mu (approximately 1698 USD/acre). Furthermore, the cost of chemical pesticides in the diluted pesticide solution, based on the prescribed dosage, averages CNY 3.342 per half kilo (approximately 0.934 USD per kilo), with a maximum of CNY 57.14 per half kilo (approximately 15.99 USD per kilo). In addition, the average distance to the nearest demonstration orchard employing technologies such as insect-proof nets, sticky traps, and insecticidal lamps is 5.7 km, with a maximum distance of 15 km.

3.6.2. Mean Difference Analysis

Table 2. Difference analysis of the impact of GPCT adoption on pesticide application.

Variable Name	Non-Adoption Group	Adoption Group	Mean Difference
Pesticide Application Intensity (CNY/Mu)	1199	943.3	256.090 ***
Chemical Pesticide Application Intensity (CNY/Mu)	1199	755.6	443.737 ***
Chemical Pesticide Concentration (CNY/half kilo)	3.721	3.052	0.670 *
Age (years)	51.72	58.54	−6.819 ***
Education Level (years)	7.806	8.677	−0.871 ***
Village Leadership	0.073	0.067	0.006
Agricultural Training	2.083	2.212	−0.129
Social Connections	4.034	3.851	0.183 **
Years of Cultivation (years)	19.88	22.99	−3.114 ***
Number of Laborers	2.882	2.157	0.724 ***
Operational Area (mu)	6.627	6.008	0.619
Proportion of Non-apple Income	0.171	0.239	−0.068 ***
Regional Dummy Variable	0.024	0.717	−0.693 ***
Distance to Demonstration Orchard (km)	7.129	4.691	2.438 ***
Sample Size	186	223

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

presents the differences in selected variables between apple farmers who did not adopt GPCT (control group) and those who did (treatment group). Columns 2 and 3 display the mean values for the control and treatment groups, respectively, while Column 4 shows the mean differences. The results reveal significant differences between the two groups of farmers. For instance, pesticide application intensity, chemical pesticide application intensity, and chemical pesticide concentration are all significantly higher in the control group than in the treatment group. Moreover, the treatment group is characterized by older farmers, higher education levels, longer cultivation experience, and a greater proportion of non-apple income.

Notably, apple farmers in the treatment group are located closer to demonstration orchards utilizing insect-proof nets, sticky traps, and insecticidal lamps, indicating a potential relationship between the explanatory variable and the instrumental variable. However, further testing is necessary to validate the instrumental variable.

Overall, while the observed differences between the two groups are statistically significant, they cannot be solely attributed to the adoption of GPCT. Consequently, an empirical analysis is required to assess the impact of GPCT adoption on the intensity and concentration of pesticide application within the ecosystem by apple farmers.

4. Empirical Results

4.1. Impact of GPCT on Farmers’ Pesticide Application Behavior

Table 3. Joint estimation results of the ETR model on the impact of GPCT adoption on farmers’ pesticide application.

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	Adoption Decision	Pesticide Application Intensity	Adoption Decision	Chemical Pesticide Application Intensity	Adoption Decision	Chemical Pesticide Concentration
GPCT Adoption Decision		−357.389 ***		−830.464 ***		−7.719 ***
		(−122.311)		(−108.289)		(−0.587)
Age	0.018 *	0.102	0.018	0.808	0.019	0.003
	(−0.01)	(−2.373)	(−0.01)	(−2.443)	(−0.01)	(−0.033)
Education Level	0.125 ***	0.685	0.128 ***	5.433	0.126 ***	0.321 ***
	(−0.032)	(−7.867)	(−0.032)	(−7.858)	(−0.032)	(−0.098)
Village Leadership	−0.704 *	−88.078	−0.649 *	−113.586	−0.645	−1.847 *
	(−0.364)	(−75.77)	(−0.359)	(−78.142)	(−0.359)	(−1.057)
Agricultural Training	0.079	−4.298	0.067	−3.987	0.067	−0.041
	(−0.066)	(−11.666)	(−0.061)	(−12.058)	(−0.061)	(−0.164)
Social Connections	−0.01	4.17	−0.018	8.719	−0.025	−0.366
	(−0.11)	(−19.726)	(−0.107)	(−20.403)	(−0.107)	(−0.278)
Years of Cultivation	−0.003	0.53	−0.004	0.499	−0.004	−0.047 **
	(−0.006)	(−1.708)	(−0.006)	(−1.764)	(−0.006)	(−0.024)
Number of Laborers	−0.029	−48.092 *	−0.055	−28.712	−0.22
	(−0.125)	(−28.391)	(−0.119)	(−29.347)	(−0.399)
Operational Area	0.001	−3.861	0.005	−5.221 *	0.005	−0.031
	(−0.016)	(−2.854)	(−0.015)	(−2.949)	(−0.015)	(−0.04)
Proportion of Non-Apple Income	0.205	−31.088	0.121	15.047	0.112	−1.572
	(−0.426)	(−86.893)	(−0.406)	(−89.505)	(−0.405)	(−1.207)
Regional Dummy Variable	Controlled	Controlled	Controlled	Controlled	Controlled	Controlled
Distance to Demonstration Orchard	−0.099 ***		−0.097 ***		−0.098 **
	(−0.018)		(−0.016)		(−0.016)
Constant	−2.008 **	1430.086 ***	−1.981 ***	1315.635 ***	−2.140 **	6.278 **
	(−0.831)	(−183.288)	(−0.799)	(−188.996)	(−0.735)	(−2.555)
Tests and Additional Information
${\rho }_{\epsilon \mu }$	0.513 ** (0.231)		0.810 *** (0.235)		1.281 *** (0.097)
Lnsigma	5.890 *** (0.046)		5.924 *** (0.050)		1.628 *** (0.038)
Goodness-of-Fit Test	87.49 ***		149.38 ***		186.58 ***
Log Pseudo-Likelihood	−3110.950		−3106.676		−1315.101
Wald Independence Test	${\chi }^2(1)=3.80$ $prob>{\chi }^2=0.0513$		${\chi }^2(1)=8.71$ $prob>{\chi }^2=0.0032$		${\chi }^2(1)=56.15$ $prob>{\chi }^2=0.0000$
Sample Size	409	409	409	409	409	409

Note: (1) *, **, *** denote significance at the 10%, 5%, and 1% levels, respectively. (2) Values in parentheses are standard errors.

presents the results of tests assessing the impact of GPCT on farmers’ pesticide application intensity, chemical pesticide application intensity, and chemical pesticide concentration. Specifically, the table reports the correlation coefficients ${\rho }_{\epsilon \mu }$ of the random error terms (${\epsilon }_i$ and ${\mu }_i$) from the selection and outcome equations, the Wald independence test results, and the validity test for the instrumental variable.

The correlation coefficients ${\rho }_{\epsilon \mu }$ from the three ETR regression models are statistically significant at the 5% level for 0.513 and at the 1% level for 0.810 and 1.281. These findings indicate that unobservable factors influence both the adoption decision for GPCT and farmers’ pesticide application intensity across the three regression equations. This highlights the presence of selection bias in the outcome equations, which, if unaddressed, would result in an underestimation of the effects.

The Wald independence test values for the three regressions are 3.80 (significant at the 5% level), 8.71, and 56.15 (both significant at the 1% level). These results reject the null hypothesis that the selection and outcome equations are independent in each regression. Consequently, it is both reasonable and necessary to jointly estimate these equations using the ETR model.

To assess the validity of the instrumental variable, a weak instrument test was conducted. The F-statistic is 52.64, which exceeds the critical threshold of 10, effectively ruling out the possibility of a weak instrument. Additionally, the estimated coefficients for the variable “distance to the nearest demonstration orchard using technologies such as insect-proof nets, sticky traps, and insecticidal lamps” in regressions (1), (3), and (5) of Table 3 are −0.099, −0.097 (both significant at the 1% level), and −0.098 (significant at the 5% level), respectively. These results provide robust evidence of a significant association between the distance to the demonstration orchard and the decision to adopt GPCT, thereby confirming the suitability of the instrumental variable.

4.1.1. ETR Model Estimation Results for the Selection Equation

Table 3’s regressions (1), (3), and (5) report the ETR model selection equation estimation results for factors influencing pesticide application intensity, chemical pesticide application intensity, and the general factors affecting farmers’ pesticide application intensity, respectively. In the selection equation, factors influencing farmers’ adoption of GPCT are identified. A positive coefficient that is statistically significant suggests that the factor has a meaningful positive effect on adoption behavior, thereby increasing the likelihood that farmers with higher values for this factor will adopt GPCT. For example, based on adoption decision Equations (1), (3), and (5), education level exhibits coefficients of 0.125, 0.128, and 0.126, respectively, each significant at the 1% level. These results demonstrate that education level significantly and positively influences the decision to adopt GPCT. This indicates that as the education level of apple farmers increases, they become more inclined to adopt at least one GPCT. Farmers with higher education levels tend to have better access to information and a broader perspective, enabling them to acquire relevant knowledge about GPCT and increasing their willingness to adopt these practices.

Conversely, holding a village leadership position has a significant negative effect on the adoption decision. This outcome is reasonable, as such a role often demands significant time for non-farm activities, reducing the household head’s leisure time and limiting their capacity to access information on GPCT, thereby lowering the likelihood of adoption. Additionally, village leaders may receive wage income and greater social resources, which can shift their focus away from apple cultivation. As a result, they may devote less time and effort to adopting new agricultural technologies.

4.1.2. ETR Model Estimation Results for the Outcome Equation

Table 3’s regressions (2), (4), and (6) present the ETR model outcome equation estimation results for factors influencing pesticide application intensity, chemical pesticide application intensity, and chemical pesticide concentration among apple farmers. In the outcome equation, factors influencing farmers’ pesticide application behavior are identified. A negative coefficient that passes the significance test indicates that the factor significantly diminishes farmers’ pesticide application practices.

In regression (2), the estimated coefficient for the adoption of GPCT is −357.389, which is statistically significant at the 1% level. This finding indicates that the adoption of GPCT significantly reduces pesticide application intensity among apple farmers, thereby supporting hypothesis H1.

In regression (4), the estimated coefficient for the adoption decision is −830.464, also significant at the 1% level. This result suggests that adopting GPCT substantially decreases chemical pesticide application intensity, further validating hypothesis H1.

In regression (6), the estimated coefficient for the adoption decision is −7.719, significant at the 1% level. This finding implies that the adoption of GPCT contributes to a reduction in chemical pesticide concentration, thereby supporting hypothesis H1.

Furthermore, this study shows that the adoption of GPCT reduces pesticide application intensity, which indirectly supports the economic benefits by demonstrating how farmers can reduce pesticide application costs per unit area, leading to overall cost savings.

4.2. Impact of Sub-Technologies Within Green Pest Control on Farmers’ Pesticide Application

GPCT encompasses a variety of approaches, including physical control, biological control, ecological control, and scientific pesticide application, each characterized by specific techniques. In this study, these sub-technologies are defined as follows: the use of insect-proof nets, insecticidal lamps, sticky traps, and insect traps represents physical control; the application of biological pesticides represents biological control; the use of disease-resistant crop varieties represents ecological control; and the practice of regular, measured pesticide application represents scientific pesticide application.

To evaluate the impact of these sub-technologies on farmers’ pesticide application levels, this analysis employs the Ordinary Least Squares (OLS) regression method with robust standard errors. The results are shown in

Table 4. Impact of green pest control sub-technologies on farmers’ pesticide application.

	OLS + Robust Standard Errors
Variable	Pesticide Application Intensity		Chemical Pesticide Application Intensity		Chemical Pesticide Concentration
Physical Control	59.018	(68.594)	−11.825	(77.156)	−1.250 **	(0.498)
Biological Control	−134.546 **	(59.396)	−208.642 ***	(66.523)	0.502	(0.521)
Ecological Control	−50.756	(45.888)	−84.924 *	(43.336)	−0.899	(0.734)
Scientific Application	10.226	(61.869)	−166.902 **	(67.200)	−0.372	(0.416)
Age	−0.972	(2.193)	−1.337	(2.005)	−0.020	(0.028)
Education Level	−7.469	(6.084)	−9.853	(6.342)	0.118	(0.078)
Village Leadership	−61.672	(76.795)	−53.447	(74.614)	−1.226 ***	(0.427)
Agricultural Training	−4.838	(10.054)	−2.101	(11.066)	−0.057	(0.158)
Social Connections	4.043	(18.621)	6.683	(19.120)	−0.378	(0.290)
Years of Cultivation	1.423	(1.760)	1.940	(1.688)	−0.038 **	(0.018)
Number of Laborers	−46.237	(28.783)	−26.190	(28.234)	−0.346	(0.424)
Operational Area	−3.080	(2.483)	−4.126	(2.772)	−0.020	(0.022)
Proportion of Non-Apple Income	−73.703	(73.916)	−46.430	(74.755)	−2.302 **	(0.952)
Regional Dummy Variable	−253.214 ***	(65.167)	−80.349	(69.250)	0.356	(0.557)
Constant	1456.116 ***	(163.883)	1383.080 ***	(157.019)	7.606 ***	(2.787)
R-squared	0.1874		0.2898		0.0513

Note: (1) *, **, *** denote significance at the 10%, 5%, and 1% levels, respectively. (2) Values in parentheses are standard errors.

.

In Table 4, coefficients accompanied by * denote that the independent variables have passed the significance tests and exert a statistically significant influence on the dependent variable. Positive values indicate that the variable exacerbates farmers’ pesticide application, whereas negative values imply that the variable helps alleviate or decrease farmers’ pesticide application. The regression results show that:

• Pesticide application intensity: The estimated coefficient for the impact of biological control on pesticide application intensity is −134.546, significant at the 5% level. This finding confirms that biological control, as represented by the use of biological pesticides, reduces pesticide application intensity among farmers. This reduction not only lowers the combined use of chemical and biological pesticides but also yields notable economic and environmental benefits. Economically, the OLS model estimates that the adoption of biological control reduces total pesticide expenditures for farmers by 134.546 CNY/Mu (approximately 11.68 USD/acre), as the decreased need for chemical pesticides offsets the higher initial costs of biological alternatives.
• Chemical pesticide application intensity: The estimated coefficients for the impacts of biological control, ecological control, and scientific pesticide application on chemical pesticide intensity are −208.642, −84.924, and −166.902, significant at the 1%, 10%, and 5% levels, respectively. These findings confirm that all three technologies effectively reduce chemical pesticide application intensity. Consequently, these technologies yield substantial economic benefits by lowering chemical pesticide expenditures for farmers by 208.642 CNY/Mu (approximately 18.13 USD/acre) for biological control, 84.924 CNY/Mu (approximately 7.37 USD/acre) for ecological control, and 166.902 CNY/Mu (approximately 14.50 USD/acre) for scientific pesticide application, as the decreased need for chemical pesticides offsets the higher initial costs of these alternative methods. Additionally, these technologies offer significant environmental advantages by decreasing chemical pesticide usage, thereby mitigating negative ecological impacts.
• Chemical pesticide concentration: The estimated coefficient for the impact of physical control on chemical pesticide concentration is −1.250, significant at the 5% level. This result demonstrates that physical control significantly reduces the concentration of chemical pesticides applied per mu, thereby mitigating the environmental impact of chemical pesticide use. These findings underscore the notable environmental benefits of physical control technologies.

In summary, these results validate hypothesis H2, confirming that different types of GPCT have distinct and significant negative effects on both the intensity and concentration of chemical pesticide applications.

4.3. Heterogeneity Analysis

The preceding analysis has demonstrated that the adoption of GPCT significantly reduces pesticide application intensity, chemical pesticide intensity, and chemical pesticide concentration among farmers. To further explore the differential impact of GPCT adoption on pesticide application across farmers with varying characteristics, this section conducts a group difference analysis using the ETR model. The data for variables such as risk type, cultivated area per laborer, and land fragmentation were collected through the same structured questionnaire administered to the farmers in this study.

Table 5. Group difference analysis results of the impact of green technology adoption on farmers’ pesticide application.

Variable	ETR Model
	Baseline Model	Risk Type	Cultivated Area per Laborer	Land Fragmentation
		Non-Risk-Averse	Above Median	Above Median
Pesticide Application Intensity	−357.389 ***	−760.290 ***	−390.049 **	−468.903 ***
	(122.311)	(107.688)	(138.707)	(157.715)
Chemical Pesticide Application Intensity	−830.464 ***	−1108.901 ***	−870.938 ***	−921.366 ***
	(108.289)	(148.982)	(134.662)	(131.530)
Chemical Pesticide Concentration	−7.719 ***	0.141	−9.224 ***	−4.518 ***
	(0.587)	(1.929)	(1.008)	(0.745)
Control Variables	Controlled	Controlled	Controlled	Controlled
Regional Variables	Controlled	Controlled	Controlled	Controlled
Sample Size	409	220	200	175

Note: (1) **, *** denote significance at the 5%, and 1% levels, respectively. (2) Values in parentheses are standard errors.

presents the regression results, highlighting the heterogeneous effects of GPCT adoption on pesticide application among farmers with different characteristics.

From the regression results in the second and third columns of Table 5, it is evident that the adoption of GPCT has a stronger negative effect on pesticide application intensity and chemical pesticide intensity among non-risk-averse apple growers compared to the overall sample. This finding indicates that GPCT adoption is more effective in reducing pesticide intensity for non-risk-averse farmers. The phenomenon can be explained by the behavioral tendencies of non-risk-averse apple farmers, who are generally more open to adopting new technologies. Unlike their risk-averse counterparts, these farmers are more willing to experiment with innovative methods and quicker to embrace new technologies. Non-risk-averse farmers actively seek information on GPCT—such as cost, effectiveness, purchasing channels, and application methods—through modern tools like the internet and smartphones. This reduces transaction costs and alleviates uncertainties associated with adopting new technologies. Consequently, the negative impact of GPCT adoption on pesticide application intensity and chemical pesticide intensity is more pronounced among this group.

In contrast, risk-averse farmers tend to have stable risk preferences, making them less susceptible to external influences. Their pesticide application practices are often shaped by years of experience and are deeply ingrained, resulting in fixed usage patterns. However, despite these differences, GPCT adoption still significantly impacts pesticide application behaviors across the overall sample, indicating that even risk-averse farmers benefit from the adoption of these technologies.

Furthermore, the results in the second and fourth columns of Table 5 reveal that GPCT adoption has a stronger negative effect on pesticide application intensity and chemical pesticide concentration among apple farmers with a larger cultivated area per laborer, compared to those with a smaller cultivated area per laborer. This suggests that the adoption of GPCT has a greater impact on farmers managing larger per-laborer cultivated areas.

This finding is consistent with the characteristics of apple farmers with larger cultivated areas per laborer, who tend to invest more time, effort, and capital in their agricultural operations. They are more inclined to adopt GPCT to enhance management efficiency and improve pest control outcomes. Moreover, the substitution effect of GPCT for chemical pesticides is further illustrated, as these technologies not only replace chemical pesticides but, in some cases, offer superior effectiveness in pest control.

On the other hand, the regression results in the second and fifth columns of Table 5 indicate that the adoption of GPCT has a stronger negative effect on pesticide application intensity and chemical pesticide concentration among apple farmers with more fragmented land. The phenomenon can be attributed to advancements and improvements in biological pesticides, which have significantly enhanced pest control effectiveness, reduced production costs, and lowered sales prices. These advancements make biological pesticides increasingly viable substitutes for traditional chemical pesticides. Farmers with more fragmented land are often inclined to adopt intensive farming practices to maximize returns on labor. As a result, GPCT proves to be more effective among this group.

4.4. Robustness Test

To verify the robustness of the baseline ETR model’s results, the IPWRA (Inverse Probability-Weighted Regression Adjustment) method is employed. The regression results are presented in

Table 6. Robustness test based on IPWRA method.

Outcome Variable	ATE	Robust Standard Error	z-Value	Sample Size
Pesticide Application Intensity	−170.004 ***	40.285	−4.22	409
Chemical Pesticide Application Intensity	−435.430 ***	38.005	−11.46	409
Chemical Pesticide Concentration	−0.681 **	0.315	−2.16	409

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

, which reports the impact of GPCT adoption on pesticide application intensity, chemical pesticide application intensity, and chemical pesticide concentration among apple growers.

Overall, the IPWRA regression results align with those of the baseline ETR model, confirming that the adoption of GPCT significantly reduces pesticide application intensity, chemical pesticide application intensity, and chemical pesticide concentration among apple farmers. Specifically, the regression results in Table 6 demonstrate that GPCT adoption has a significant negative impact on pesticide application behavior, with ATE (Average Treatment Effect) values of −170.004 and −435.430 at the 1% significance level, and −0.681 at the 5% significance level. These findings indicate that the conclusions remain robust even when an alternative econometric method is used.

5. Discussion

This study aimed to evaluate the adoption of GPCT in apple orchards by measuring pesticide intensity per unit, assessing the impact on pesticide reduction, and analyzing the environmental effects in large-scale apple farming systems in China. The findings indicate that GPCT adoption leads to significant reductions in pesticide use, particularly in pesticide application intensity, chemical pesticide application intensity, and chemical pesticide concentration. These results support the hypothesis that GPCT has the potential to mitigate the environmental damage associated with conventional pesticide use. Furthermore, this study addresses existing gaps in the literature by exploring the heterogeneous impacts of GPCT across different farmer groups and providing empirical evidence to inform policy decisions.

5.1. Impact of GPCT on Pesticide Use

The reductions in pesticide application observed in this study align with previous research highlighting the ecological benefits of GPCT. For instance, Chandler et al. [8] emphasized the potential of GPCT to reduce chemical inputs while maintaining agricultural productivity. However, their work primarily involved qualitative assessments and broad analyses of biopesticide development and regulation, whereas our study provides quantitative evidence from a large-scale farming context in China. This distinction underscores the practical effectiveness of GPCT in diverse agricultural settings and offers valuable insights for policymakers aiming to promote sustainable farming practices.

Our findings confirm that GPCT adoption significantly reduces pesticide intensity and chemical concentration, thereby minimizing immediate environmental risks such as water and soil contamination. Additionally, by reducing reliance on chemical pesticides, GPCT helps mitigate long-term issues like biodiversity loss and soil degradation, which are well-documented concerns in pesticide-intensive agricultural systems [2]. Furthermore, prior research has shown that traditional methods for assessing pesticide risks may fail to capture the extreme risks posed by certain pesticides [11]. This highlights the need for technologies like GPCT that reduce both pesticide use and associated risks.

5.2. Effectiveness of Different Sub-Technologies

This study reveals that the effectiveness of GPCT varies across different sub-technologies, offering nuanced insights into their comparative advantages.

• Biological control methods: The use of natural predators and biocontrol agents significantly reduces pesticide reliance. These findings align with Gurr et al. [24], who emphasized the ecological efficiency of biological pest control in integrated pest management systems.
• Ecological control methods: Practices such as crop rotation and pest-resistant varieties also contribute to pesticide reduction by enhancing pest management systems. Lewis et al. [25] similarly highlighted the long-term sustainability of these methods in agricultural systems.
• Scientific application: Scientific application technologies, such as precision pesticide application and pre-harvest pesticide withdrawal, have been found to effectively reduce chemical pesticide application intensity. Precision pesticide application technologies, including air-assist sprayers and recycling tunnel sprayers, reduce pesticide waste, improve spray uniformity, and decrease environmental pollution by targeting specific areas effectively [26]. Additionally, pre-harvest pesticide withdrawal intervals significantly lower pesticide residue levels in vegetables, aligning with safety thresholds for low-residue production [27]. These findings collectively provide robust evidence that precision agricultural practices and careful pesticide withdrawal management can substantially reduce the intensity and environmental impact of chemical pesticide application.
• Physical control methods: Tools like light traps and insecticidal lamps specifically target pest populations, significantly lowering pesticide concentrations. This finding supports the broader literature on physical pest control methods, such as the work by Ehler [28].

These results emphasize that the success of GPCT depends on adopting a combination of technologies tailored to specific farming challenges, further highlighting the importance of integrated pest management (IPM) approaches.

5.3. Heterogeneous Environmental Effects

The study uniquely contributes to the literature by analyzing the heterogeneous effects of GPCT adoption across farmer groups. The findings demonstrate that the environmental benefits of GPCT adoption vary based on farmers’ characteristics:

• Risk preferences: Non-risk-averse farmers are more likely to adopt GPCT and experience greater environmental benefits. This finding builds on the work of Koundouri et al. [29], who emphasized the role of risk preferences in technology adoption.
• Land size: Farmers managing larger landholdings tend to derive greater benefits from GPCT due to economies of scale.
• Land fragmentation: Farmers with fragmented landholdings also experience significant reductions in pesticide use, as the complexity of land management incentivizes efficient practices [30].

These results underscore the importance of considering farmer heterogeneity when evaluating the effectiveness of GPCT. Previous studies have largely overlooked these individual differences, focusing instead on aggregate outcomes. By identifying these heterogeneous effects, this study provides valuable insights for tailoring policies and interventions to maximize the benefits of GPCT adoption.

5.4. Policy Implications

The findings have important implications for policy and practice. The adoption of GPCT is influenced by several barriers, including fragmented landholdings, financial constraints, risk aversion, and a lack of technical knowledge among farmers. Given the heterogeneous effects of GPCT adoption, policies should be designed to address the specific needs and characteristics of different farmer groups:

• Encourage land consolidation: Farmers with fragmented landholdings benefit significantly from GPCT. Policies that promote land consolidation can help these farmers optimize their use of green technologies by providing larger, more manageable plots.
• Expand agricultural extension services: Targeted education and training programs can address barriers such as risk aversion and lack of technical knowledge, encouraging wider GPCT adoption by equipping farmers with the necessary skills and confidence to implement these technologies.
• Provide financial incentives: Subsidies or financial incentives for purchasing and implementing GPCT can alleviate financial constraints, particularly for smallholders who may lack the necessary capital to invest in green technologies.
• Promote technological integration: Investments in advanced technologies, such as precision pesticide application tools, can amplify GPCT’s environmental benefits while improving cost-efficiency, thereby making adoption more attractive to farmers.

By addressing the barriers identified in this study, policymakers can facilitate the creation of farming systems that are both ecologically resilient and economically viable.

6. Conclusions

This study provides empirical evidence on the environmental effects of adopting GPCT in apple farming. By quantifying GPCT’s impact on pesticide use and exploring their heterogeneous effects across different farmer groups, the findings address key gaps in the existing literature.

• Reduction in pesticide use: GPCT adoption leads to significant reductions in pesticide application intensity, chemical pesticide intensity, and chemical pesticide concentrations. These findings support the hypothesis that GPCT can effectively mitigate the environmental risks associated with conventional pesticide use.
• Effectiveness of different sub-technologies: Among the sub-technologies examined, biological control methods were the most effective in reducing pesticide application intensity, while scientific application technologies contributed significantly to reducing chemical concentrations. These results highlight the importance of combining multiple technologies to maximize GPCT’s benefits.
• Heterogeneous effects: The study finds that the environmental benefits of GPCT adoption vary across farmer groups. Farmers with larger or more fragmented landholdings, as well as those who are less risk-averse, derive greater benefits. This highlights the role of socio-economic and operational factors in shaping the effectiveness of GPCT adoption.

These conclusions, derived directly from the empirical findings, provide a foundation for tailoring GPCT promotion strategies to maximize their environmental and economic benefits. By addressing the specific needs and characteristics of diverse farmer groups, policymakers and agricultural practitioners can enhance the adoption and effectiveness of GPCT, fostering sustainable and resilient farming systems.