Does Social Learning Promote Farmers’ Cooperative Pest Control?—Evidence from Northwestern China

Abstract

Pest management is pivotal for ensuring secure grain production and constitutes a fundamental strategy in combating pests that detrimentally affect grain supplies. Given the complexity and dynamic nature of pests, it is imperative that farmers implement coordinated prevention and control strategies. Such measures are essential to augment the efficacy of these efforts and to reduce the risks posed by pests to agricultural crops. This research involved a survey of 1205 agricultural households spanning three representative provinces in Northwestern China. By employing an endogenous switching Probit model and addressing sample selection bias, the study investigates the influence of social learning on the adoption of cooperative pest control strategies by farmers. The findings indicate that social learning significantly enhances farmers’ adoption of cooperative pest control measures. In a counterfactual scenario, introducing social learning to farmers previously unexposed to it would result in a 10.3% increase in the likelihood of adopting these practices. Additionally, factors such as the health status of the household head, income level, and size of land under management are critical determinants of farmers’ participation in social learning. The differential access to scientific, accurate, and systematic information, coupled with resource disparities among farmers, can partially account for the varying average treatment effects observed in different learning methods on the propensity to adopt cooperative pest control practices. Furthermore, social learning plays a crucial role in fostering such adoption by establishing trust among farmers, facilitating consensus in decision-making, and enhancing the dissemination of information.

1. Introduction

Multiple ongoing global phenomena have negatively impacted food security. The rise in food prices has forced households to allocate a greater share of their income to food, thereby lowering living standards, heightening the risk of hunger, and hindering progress toward the achievement of the Sustainable Development Goals aimed at eradicating poverty and hunger. Additionally, factors such as extreme weather events, global commodity price volatility, and disruptions in supply chains have further driven up food costs [1]. Global shocks, including the COVID-19 pandemic, climate change, and the Russia–Ukraine conflict, have intensified production interruptions, increased energy costs, and fueled inflation, posing significant threats to the security of both food supply and access [2]. Consequently, the issue of how to safeguard the basic security of food production has become a central focus of attention. Pests not only directly inflict significant reductions in grain yields but may also precipitate a deterioration in grain quality and safety, thereby endangering national food security [3,4,5]. As the foremost grain producer globally, China plays a pivotal role in ensuring the food security of approximately 20% of the world’s population. Recent data from 2023 suggest a marked resurgence in significant crop pests across China, impacting an area of approximately 329.5 million acres—an increase of 29.5% compared to the 2022 average. The estimated yield losses attributable to these outbreaks are projected to surpass 175 billion kilograms (http://www.moa.gov.cn/ztzl/2023cg/jszd_29356/202302/t20230209_6420225.htm, accessed on 9 January 2023). Consequently, it is imperative to implement proactive strategies to curb the proliferation of pests, thus enhancing the protection of national food security more effectively.

Despite substantial mitigation efforts, pests continue to pose a major threat to agricultural producers globally, resulting in estimated annual losses of approximately USD 70 billion [6,7]. The fundamental issue stems from the fact that farmers’ individual pest control strategies often overlook the presence of identical pests in neighboring plots. Additionally, empirical studies suggest that collaborative pest management strategies surpass traditional individual approaches by significantly enhancing the judicious use of chemicals, the efficacy of control measures, and the optimization of production techniques and resource distribution, ultimately leading to increased income for farmers [8,9,10]. A plausible explanation for the inefficacy of uncoordinated pest control lies in its associated negative externalities. When farmers implement pest control measures asynchronously, those plots treated earlier may inadvertently displace pests to neighboring plots. This displacement can lead to an increased pest burden in these adjacent areas, subsequently resulting in diminished crop yields [11]. Collaborative pest management among farmers offers a strategic solution to counteract the negative externalities inherent in individualized approaches [12], thereby enhancing control efficacy and mitigating agricultural production risks [13]. Nonetheless, several impediments—such as deficient cooperation and coordination mechanisms [14,15], limited awareness among farmers [16,17], technical shortcomings [13], and strategic uncertainties including mistrust toward fellow farmers [18]—compromise the effective implementation of these collective measures, resulting in less than optimal outcomes. Consequently, the challenge of fostering effective coordinated pest control practices among farmers and continuously improving these interventions remains an urgent and significant concern.

It merits attention that the efficacy with which farmers manage mobile pests is significantly influenced by the behaviors of their counterparts in neighboring plots [19]. Current literature exploring the determinants of farmers’ cooperative behaviors in pest management frequently neglects the pivotal role of social learning. According to social learning theory, an individual’s behavior is shaped not only by personal experiences but also by observing the actions and consequences experienced by others [20]. The impact of social learning on cooperative pest control among farmers manifests in several key ways: Initially, farmers gain insights and adopt practices by observing the control measures and outcomes implemented by their peers in analogous conditions. Furthermore, through dialog and collaboration with fellow farmers and agricultural technical departments, the sharing of resources and information is facilitated. This exchange fosters increased trust and cooperative intent among farmers, thereby enhancing the likelihood of establishing an effective collaborative network for pest management.

Moreover, subsequent research has corroborated that social learning significantly mitigates the barriers encountered by farmers in the adoption of cooperative pest control strategies, thereby promoting a more integrated approach to management. Social learning plays a crucial role in the diffusion of pest control technologies [21] and other agricultural innovations [22], while also enhancing farmers’ awareness of control measures [23] and improving both the efficiency and effectiveness of these practices [24]. Additionally, social learning fosters cooperative pest management among farmers. For instance, Xu, et al. [13] found that mutual learning within organized groups significantly encourages the collective adoption of pest control practices. Similarly, Kruger [25] demonstrated that social learning can strengthen collaborative efforts in pest management, and Vänninen, et al. [26] observed that knowledge exchange among farming households can enhance their collective resilience against pests. With the advancement of digital technology, utilizing online live streams or videos to seek and acquire agricultural knowledge, as well as sharing and emulating successful online practices, has increasingly become a prevalent learning method [27]. Mwambi, et al. [28] argue that smartphones can play a crucial role in identifying pests, offering guidance on pesticide use, and enhancing overall control measures.

Consequently, drawing from the preceding analysis, this study collectively defines learning from relatives, neighbors, and friends (neighborhood communication); learning through agricultural technical departments (participation in training); and online learning (self-directed learning) under the umbrella term “social learning”. Theoretically, farmers can establish mutual trust through learning and communication [29], which, in turn, facilitates cooperative pest control efforts. Engaging in various forms of learning allows farmers to gain valuable insights from the practical experiences and control strategies of other farmers and experts, thereby fostering a shared understanding [30] that supports coordinated action. Furthermore, learning and communication among farmers promote the exchange of pest control knowledge and skills [31,32], which helps align behaviors through learning and imitation. However, existing research has largely overlooked the impact of different social learning approaches on farmers’ adoption of cooperative pest control practices, as well as the mechanisms driving these effects.

Building on this framework, the present study examines Northwest China as a case study to investigate the mechanism through which social learning influences farmers’ cooperative pest control behavior. Specifically, the research employs an endogenous switching probit model to empirically test whether social learning effectively promotes farmers’ participation in cooperative pest control. Furthermore, the study explores the pathways through which social learning impacts this behavior and utilizes counterfactual analysis to assess the average treatment effect of social learning on farmers’ cooperative pest control practices. In comparison with the existing literature, the marginal contributions of this study are as follows: First, this research integrates network learning into the conceptual framework of social learning to investigate its impact on farmers’ cooperative pest control behavior, thereby enriching the body of research on social learning. Second, this study incorporates trust relationships, decision consensus, and information sharing into the analytical framework of social learning and farmers’ cooperative pest control behavior. By exploring the pathways through which social learning impacts cooperative pest control practices, this research provides valuable insights into the internal dynamics between farmers’ social learning and their cooperative behavior in pest control. This deepens the understanding of how social learning influences farmers’ decision-making and collaboration in pest control initiatives. Additionally, this study considers the distinct characteristics of various learning styles by examining the impact of three different approaches on farmers’ cooperative pest control behavior: learning from relatives, neighbors, and friends (neighborhood communication); learning from agricultural technology departments (training); and online learning (active learning). This analysis provides valuable insights into how different learning styles influence farmers’ participation in cooperative pest control initiatives.

2. Theoretical Analysis and Research Hypotheses

2.1. Direct Influence of Social Learning on Farmers’ Adoption of Cooperative Pest Control Practices

Social learning theory primarily focuses on observational and imitative learning, as well as the role of self-regulation in shaping human behavior. It also underscores the significant interaction between the environment and human behavior, illustrating a complex dynamic that influences behavioral outcomes [21,33]. When a farmer observes other farmers successfully managing pests through collaborative efforts, they are likely to emulate these strategies by engaging in interactions and communications with peers or by soliciting assistance from relevant agronomic departments [21,34]. This transfer of knowledge and sharing of experiences can expedite the adoption of cooperative pest control behaviors among farmers. This acceleration is attributed primarily to two factors: firstly, social learning generates a knowledge spillover effect within a farming community, facilitating the dissemination and diffusion of individual behaviors; secondly, the adoption of behaviors among farmers exhibits a cohort effect, leading to a convergence of practices through social learning [20]. Additionally, when neighboring farmers adopt consistent and proven methods for pest management, individual farmers may feel compelled to adopt similar strategies due to social pressures, aiming to align with group norms and avoid the stigma of non-conformity [10]. Echoing the findings of Xu, et al. [13], Kruger [25], and Vänninen, et al. [26], social learning is likely to encourage the adoption of cooperative pest control behaviors. Based on the aforementioned discussion, this study proposes the following hypothesis:

Hypothesis 1. 

Social learning will significantly enhance farmers’ adoption of cooperative pest control behaviors.

2.2. Indirect Influence of Social Learning on Farmers’ Adoption of Cooperative Pest Control Practices

Firstly, social learning, facilitated through interactions with neighbors, relatives, friends, agricultural departments, and online platforms, enables farmers to gain insights into the behaviors and attitudes of their peers. This diverse array of information sources allows farmers to more thoroughly assess the credibility of other farmers and to establish, based on these assessments, trusting relationships [35,36,37]. Secondly, social learning fosters interaction and communication among farmers, facilitating the development of mutually supportive relationships characterized by increasing trust [37]. Lastly, by observing and emulating the behaviors and attitudes of others, farmers may develop a sense of similarity that enhances regional cohesion and belonging. This emerging consensus and emotional connection further deepen trust relationships, thereby promoting cooperative behaviors among farmers, particularly in the management of pests [38]. Based on this detailed analysis, the study proposes the following hypothesis:

Hypothesis 2. 

Social learning will facilitate the adoption of cooperative pest control behaviors among farmers by fostering trusting relationships within the agricultural community.

Social learning enables farmers to grasp the opinions and perspectives of their peers, thereby enhancing communication and mutual understanding [39]. It allows them to gain diverse experiences and insights, fostering a collective recognition that pest control is a communal challenge requiring cooperative solutions [40]. Furthermore, through social learning, farmers engage in discussions and negotiations to better understand each other’s needs, thus increasing their propensity to collaborate, see Epanchin-NiellWilen [11]. Such dynamic learning processes, encompassing communicative learning, deliberation, and decision-making, facilitate a unified perspective on the issues at hand. Additionally, social learning aids in the dissemination and acquisition of knowledge and skills related to pest control, leading to a comprehensive understanding and a consensus on coordinated action [41]. This concurs with findings by KangCao [42], who noted that social learning fosters a new consensus and unifies group behavior. Based on these observations, this study proposes the following hypothesis:

Hypothesis 3. 

Social learning will facilitate the adoption of cooperative pest control behaviors among farmers by fostering consensus in decision-making processes within the community.

Information sharing among farmers is crucial for enhancing agricultural productivity and improving quality of life [43]. Through various learning methods, farmers acquire insights into the agricultural production experiences and techniques of their peers, facilitating the exchange of knowledge and observations on pest control practices. This exchange is instrumental in equipping farmers with essential information, such as pest counts and their trends across different plots [26]. This aligns with WangLi’s [44] observation that social learning enables access to more valuable information and knowledge. Specifically, observing the successful practices of others motivates farmers to experiment with new technologies and methods, fostering the dissemination of pertinent knowledge and skills. This mirrors Shikuku’s [45] finding that imitation and learning address information constraints in the adoption of agricultural technologies and promote consistency in farmers’ behaviors. Based on the detailed discussion above, this study proposes the following hypothesis:

Hypothesis 4. 

Social learning will facilitate the adoption of cooperative pest control behaviors among farmers by promoting effective information sharing within the community.

In examining the motivations behind different farmers’ behaviors, it is evident that participation in cooperative disease and pest control is influenced by a multitude of factors. Consequently, this study draws upon the research of StallmanJames Jr [10] and Epanchin-NiellWilen [11]. Based on the available data, individual characteristics of farmers (such as gender, age, and health status), family characteristics (including income level, land management area, loan access, and distance from the main road), and production and management characteristics (such as agricultural insurance purchases and soil quality) were selected as control variables. These variables were incorporated to mitigate the potential interference of external factors, thereby ensuring the robustness of the research findings and enhancing the depth of their interpretation. First, male farmers typically assume a leading role in agricultural production, engage more frequently with external communities, and are therefore more likely to participate in cooperative disease and pest control efforts due to their broader access to social resources [46,47]. In contrast, elderly farmers, constrained by physical limitations and traditional notions, may be less inclined to engage in cooperative disease and pest control activities, particularly those that demand significant physical effort [48]. Additionally, farmers with health limitations may rely more heavily on community resources and cooperation [49,50], making them more likely to participate in cooperative pest control efforts. Similarly, farmers with higher income levels may be more motivated to engage in cooperative pest control as a means of safeguarding their agricultural investments [51]. Farmers managing larger land areas may be reluctant to participate in cooperative disease and pest control due to the increased complexity of coordinating with other farmers [52]. In contrast, farmers who rely on borrowing, facing greater financial pressure and risk, may be more inclined to utilize all available resources [53], including engaging in cooperative pest control efforts. Additionally, geographically remote farmers may be less willing or able to participate effectively in cooperative pest control due to limited access to information and resources. Farmers who purchase agricultural insurance may be less inclined to participate in cooperative pest control, as they have already demonstrated concern for risk management. However, since insurance provides risk coverage [54,55], it may reduce the perceived need to engage in collective preventive measures. Conversely, farmers with high-quality soil may be more willing to participate in cooperative pest control to protect their valuable land and ensure optimal crop yields and quality [56,57]. In summary, farmers’ decisions regarding participation in cooperative pest control are influenced by a variety of factors that collectively shape their behavioral choices.

The specific mechanism through which social learning influences farmers’ adoption of cooperative pest control behavior is illustrated in Figure 1 below.

3. Materials and Methods

3.1. Data Sources

Pests represent significant challenges to agricultural production in China [58], particularly under the impact of global climate change, which has increased the frequency of extreme climatic events such as droughts and heat waves [59,60]. These conditions not only widen the range of pests but also intensify their impact. This study’s data were collected through a questionnaire survey administered to farmers by the research team from July to September 2021 across three provinces in Northwest China: Shaanxi, Gansu, and Ningxia. This region is characterized by sparse rainfall and frequent droughts, factors that exacerbate the prevalence of agricultural pests. Based on the analysis and assessment conducted by the plant protection department and experts in Shaanxi Province, it is projected that in 2023, food crop diseases and pests in the province will experience significant development, with an estimated affected area of approximately 14.5 million acres (http://nynct.shaanxi.gov.cn/www/snynctbgswj/20230320/9818355.html, accessed on 17 March 2023). Meanwhile, crop diseases and pests in Gansu and Ningxia provinces are expected to remain relatively severe, with a moderate occurrence trend, affecting a total area of approximately 6.59 million acres (http://nynct.nx.gov.cn/zzb/zwdt_67876/202301/t20230112_3913608.html, accessed on 12 January 2023; http://nync.lanzhou.gov.cn/art/2023/1/28/art_2202_1192518.html, accessed on 28 January 2023). Therefore, selecting the three provinces in Northwest China as the research area provides a representative basis for this study.

The specific research area is shown in Figure 2. In this study, data were collected from a total of 1285 rural households. After excluding samples with missing or contradictory variables, the final dataset comprised relevant information from 1205 rural households. The survey was spearheaded by the National Natural Science Foundation of China (NSFC) under two projects: “The Demand-induced Mechanism of Conservation Tillage Technology Adoption: Organizational Support, Inter-temporal Choice, and Incentive Effects” and “The Research on the Linkage Mechanism between Small-scale Farmers and Modern Agricultural Industry”, the latter being managed by the Shaanxi Provincial Cooperative Innovation and Extension Alliance. Under a sampling framework developed in consultation with relevant government officials and agricultural technology experts, counties (or cities), townships (or towns), and villages were randomly selected from each sampling stratum within Shaanxi, Gansu, and Ningxia provinces. The specific sampling procedure was as follows: First, 1 to 3 counties (or cities) were randomly selected from each of the provinces. Second, 1 to 2 townships (or towns) were randomly selected within the chosen counties. Third, 2 to 5 villages were randomly selected from each of the chosen townships. Finally, 14 to 40 households were randomly selected from these villages, and data were collected through one-on-one household interviews to achieve the required sample size.

3.2. Model Construction

3.2.1. Endogenous Switching Probit Model

Farmers’ decisions to engage in social learning, aimed at acquiring information and enhancing their skills and knowledge, are not arbitrary but characterized by self-selection. This necessitates methodological corrections to ensure the estimation of consistent parameters. The determinants of social learning encompass both observable and unobservable factors, thereby introducing selection bias. To address this issue, our study employs the endogenous switching probit model framework as proposed by LokshinSajaia [61]. The advantages of this model are as follows: on the one hand, it accounts for both observable and unobservable factors when addressing the issues of self-selection and endogeneity in social learning. On the other hand, the use of full information maximum likelihood estimation effectively mitigates the problem of information omission. This framework facilitates the construction of a selection equation for social learning, which is delineated below:

$$S_i^{*}=Z_i\alpha +{\mu }_i, S_i=\left\{\begin{array}{ll}1,& S_i^{*}>0\\ 0,& S_i^{*}\le 0\end{array}\right.$$

(1)

In the specified model, the treatment variable ($S_i)$ signifies whether a farmer opts to participate in social learning, with $S_i$ assigned a value of 1 if the treatment is accepted and 0 otherwise. The latent variable $S_i^{*}$ captures the unobservable determinants influencing a farmer’s decision to engage in social learning. $Z_i$ encompasses the observable factors that affect this decision, functioning as the set of control variables. The parameter $\alpha$, which requires estimation, and ${\mu }_i$, the random error term, are also integral components of the model.

$$y_i=\left\{\begin{array}{cccc}{y}_{1i},& S_i=1,& y_{1i}=I\left(y_{1i}^{*}>0\right),& y_{1i}^{*}=X_{1i}{\beta }_1+{\epsilon }_{1i}\\ y_{0i},& S_i=0,& y_{0i}=I\left(y_{0i}^{*}>0\right),& y_{0i}^{*}=X_{0i}{\beta }_0+{\epsilon }_{0i}\end{array}\right.$$

(2)

Equation (2) delineates the outcome equation for the effect of social learning on farmers’ adoption of cooperative pest control practices across different states. Specifically, $y_{1i}^{*}$ and $y_{0i}^{*}$ represent the latent variables corresponding to the adoption of cooperative pest control behaviors with and without the influence of social learning, respectively. These latent variables determine the observed outcomes $y_{1i}$ and $y_{0i}$. The factors influencing the adoption of these behaviors are denoted by $X_{1i}$ and $X_{0i}$, while ${\beta }_1$ and ${\beta }_0$ are the parameters to be estimated. Lastly, ${\epsilon }_{1i}$ and ${\epsilon }_{0i}$ represent the random error terms.

Assuming that the random error terms ${\mu }_i$, ${\epsilon }_{1i}$, and ${\epsilon }_{0i}$ in Equations (1) and (2) possess mean values of zero and adhere to a joint normal distribution, they are expressed as follows:

$${\mathsf{\Omega }}_M=\left[\begin{array}{ccc}1& {\rho }_1& {\rho }_2\\ & 1& {\rho }_3\\ & & 1\end{array}\right]$$

(3)

In Equation (3), ${\rho }_1$ represents the correlation coefficient between the random error terms ${\mu }_i$ and ${\epsilon }_{0i}$, ${\rho }_2$ denotes the correlation coefficient between ${\mu }_i$ and ${\epsilon }_{1i}$, and ${\rho }_3$ is the correlation coefficient between ${\epsilon }_{0i}$ and ${\epsilon }_{1i}$. Given that $y_{1i}^{*}$ and $y_{0i}^{*}$ cannot be observed simultaneously, ${\rho }_3$ remains unidentifiable; therefore, the assumption that ${\rho }_3=1$ follows the precedent set by LokshinSajaia [61].

Linking Equations (1) and (2) and employing the full information maximum likelihood method of estimation facilitates the determination of the parameter values in the endogenous switching probit model.

3.2.2. Average Treatment Effect on the Treated (ATT)

This paper adopts the research methodology outlined by Liu, et al. [62]. Upon deriving the estimated parameter values within the endogenous switching probit model delineated previously, the average treatment effect of farmers’ social learning on their adoption of cooperative pest control behaviors can be ascertained through counterfactual analysis. This involves comparing the adoption of these behaviors under actual and counterfactual scenarios, wherein the latter scenario does not involve social learning. The treatment effect of social learning on the adoption of cooperative pest control behaviors is formulated as follows:

$$\begin{gathered} TT\left(x\right)=\mathrm{Pr}\left(y_{1i}=1|S_i=1,X_i=x\right)-\mathrm{Pr}\left(y_{0i}=1|S_i=1,X_i=x\right) \\ ={\displaystyle \frac{{\mathrm{\Phi }}_2\left(X_{1i}{\beta }_1,Z_i\alpha ,{\rho }_2\right)-{\mathrm{\Phi }}_2\left(X_{0i}{\beta }_0,Z_i\alpha ,{\rho }_1\right)}{F\left(Z_i\alpha \right)}} \end{gathered}$$

(4)

In this model, $\mathrm{\Phi }(·)$ and $\mathrm{F}(·)$ represent the binary and univariate normal distribution cumulative functions, respectively. The condition $S_i=1$ signifies that farmer $i$ engages in social learning; $X_i=x$ indicates that farmer $i$ possesses the observed characteristics $x$; $y_{1i}=1$ denotes the cooperative pest control behavior observed under social learning conditions; and $y_{0i}=1$ represents the cooperative pest control behavior observed under the counterfactual scenario, which assumes the absence of social learning. The treatment effect, $TT\left(x\right)$, quantifies the impact of social learning on the adoption of cooperative pest control behaviors by farmers who genuinely participate in social learning and exhibit the characteristics $x$, relative to a counterfactual baseline. Utilizing Equation (4), the average treatment effect on the treated $\left(ATT\right)$ is calculated by averaging $TT\left(x\right)$ across the sample of farmers engaged in social learning, as demonstrated in the subsequent equation:

$$ATT={\displaystyle \frac{1}{N_{S_i=1}}}{\sum }_{S_i=1}TT\left(x\right)$$

(5)

3.2.3. Propensity Score Matching (PSM)

This paper also utilizes the research methods of Yan, et al. [63] to construct a propensity score matching (PSM) model. PSM is frequently employed to address the issue of sample “self-selection.” The core idea is to identify or select a non-social learning farmer from the sample of non-social learning farmers, ensuring that the other characteristics of the two groups are comparable, except for the social learning behavior. Consequently, the outcome variables of the two groups can be treated as the estimated results of two different experiments (social learning and non-social learning) on the same farmer. The difference in these estimated outcomes represents the net effect of social learning. In this paper, four matching methods were employed to assess the impact of social learning on farmers’ cooperative pest control behavior: K-nearest neighbor matching (one-to-one matching), K-nearest neighbor matching with a caliper (one-to-four matching with a caliper range of 0.01), radius (caliper) matching (caliper range of 0.05), and kernel matching (using the default kernel function and bandwidth). The difference in outcomes between the two experiments (social learning and non-social learning) for the same farmer represents the average treatment effect on the treated (ATT). The corresponding expression is as follows:

$$ATT=E\left(Y_{1m}|R_m=1\right)-\mathrm{E}\left(Y_{0m}|R_m=1\right)=E\left(Y_{1m}-Y_{0m}|R_m=1\right)$$

(6)

In Equation (6), $Y_1$ represents the cooperative pest control behavior of farmers engaged in social learning, while $Y_0$ represents the cooperative pest control behavior of farmers not engaged in social learning. The value of $E\left(Y_{1m}|R_m=1\right)$ can be directly observed, whereas $\mathrm{E}\left(Y_{0m}|R_m=1\right)$ cannot be observed and is instead a counterfactual estimate.

3.2.4. Mediation Effect Model

Consistent with the hypotheses advanced in our theoretical framework, social learning may influence farmers’ cooperative pest control behaviors by fostering trust relationships, facilitating consensus in decision-making, and enhancing information sharing. This study adopts the mediation effect model proposed by HuangPan [64], which comprises three distinct components: the relationship between the independent and dependent variables, the relationship between the independent variable and the mediator variable, and the comprehensive model incorporating the independent variable, mediator variable, and dependent variable. The formulas for estimating the specific mediation effects are provided in Equations (7)–(9).

$$Y_i=c_0+c_1{Study}_{1i}+{\sum }_{m=2}^M{c}_m{Control}_{im}+{\mu }_{1i}$$

(7)

$${Mediator}_{ji}=a_0+a_1{Study}_{i1}+\sum _{m=2}^M{a}_2{Control}_{im}+{\mu }_{2i}$$

(8)

$$Y_i={\gamma }_0+{\gamma }_1{Study}_{1i}+{\gamma }_2{Mediator}_{ij}+\sum _{m=3}^M{\gamma }_m{Control}_{im}+{\mu }_{3i}$$

(9)

In Equations (7)–(9), $Y_i$ symbolizes the cooperative pest control behavior of the $i$-th farmer, ${Study}_{1i}$ indicates the social learning activities of the $i$-th farmer, and ${Control}_{im}$ encompasses a set of control variables relevant to the $i$-th farmer. The mediator variable, ${Mediator}_{ij}$, is specified such that $M_{1i}$ represents the trust relationship for $j=1$, $M_{2i}$ encapsulates the decision-making consensus for $j=2$, and $M_{3i}$ denotes information sharing for $j=3$. The random perturbation terms, ${\mu }_{1i}$, ${\mu }_{2i}$, and ${\mu }_{3i}$, are presumed to follow a zero-mean normal distribution. Equation (7) specifies $c_1$ as the total effect of social learning on cooperative pest control behavior. Equation (8) defines $a_1$ as the influence of social learning on the mediator variable. Equation (9) calculates ${\gamma }_2$ as the impact of the mediator on cooperative pest control behavior, controlling for the effect of social learning, with ${\gamma }_1$ representing the direct effect of social learning and the product of $a_1{\gamma }_2$ quantifying the mediating effect.

3.3. Variable Selection and Description

(1) Dependent variable: Cooperative Pest Control. This variable is coded as 1 if the farmer, along with neighboring plot farmers, successfully implements pest control through cooperative efforts and unified crop protection directives, and 0 otherwise.

(2) Core independent variable: Social Learning. Drawing from previous analyses and the research by Qiao, et al. [20] and Tafesse, et al. [41], social learning is assessed through three specific inquiries. The first question, “Do you share your pest control experience with other farmers?” evaluates learning through interactions with neighbors, relatives, and friends (neighborhood communication). The second, “Do you seek pest control methods and strategies through production training?” gauges learning from formal instruction provided by agricultural departments (attending training). The third, “Do you seek pest control information and practices through live agricultural instruction or videos?” assesses online learning (active learning). These questions collectively serve as a proxy variable for measuring farmers’ social learning, with a value of 1 assigned if farmers engage in any or all of these methods to share and access agricultural resources, and 0 otherwise.

(3) Mediating variables: Trust relationship, decision consensus, and information sharing are articulated through distinct question items derived from prior studies. The trust relationship is gauged by the question, “Do you trust your relatives, neighbors, friends, etc.?” as employed in the framework by OnyxBullen [65]. Decision consensus is assessed through the query, “Do you believe that the behavior of other farmers influences your behavior?” following the methodology of Mills, et al. [66]. Information sharing is measured by the question, “Did you receive information from others about disaster prevention, etc., before the disaster?”, which is based on the approaches of Garbach, et al. [67] and Maguire-Rajpaul, et al. [68].

(4) Control variables: Drawing on the methodologies of StallmanJames Jr [10] and Epanchin-NiellWilen [11], this study selectively incorporates control variables including individual, family, and production and management characteristics of the surveyed farmers, as informed by the available data. Detailed definitions and descriptive statistics for these variables are presented in

Table 1. Variable definition and descriptive statistics.

Variable Name	Variable Definition	Mean	Standard Deviation
Farmers’ Cooperative Pest Control	Whether pest control is achieved through cooperative supply and unified prevention and control with neighboring farmers. 1 = Yes, 0 = No	0.485	0.500
Social learning	Whether any or multiple methods such as learning from relatives, neighbors, and friends (neighborhood communication), learning from agricultural technical departments (attending training), or learning online (self-directed learning) are used to achieve the sharing and acquisition of agricultural resources. 1 = Yes, 0 = No	0.503	0.500
Learn from relatives, neighbors, and friends	Do you share your pest control experience with other farmers? 1 = Yes, 0 = No	0.630	0.483
Learn from the agricultural technology department	Do you seek pest control methods and strategies through production training? 1 = Yes, 0 = No	0.337	0.473
Online learning	Do you seek pest control information and practices through live agricultural instruction or videos? 1 = Yes, 0 = No	0.447	0.497
Gender	Household head gender. 1 = Male, 0 = Female	0.925	0.263
Age	Household head age (years)	57.616	10.21
Health status	Household head health status. 1 = Unable to take care of oneself, 2 = Can take care of oneself with long-term chronic illness, 3 = Frail with minor illnesses, 4 = Healthy, 5 = Very healthy	3.993	0.879
Income level	Total household income (ten thousand yuan)	6.74	9.054
Scale of land operation	Household land management area (acres)	2.471	37.964
Borrowing situation	1 = Took out a loan, 0 = Did not take out a loan	0.212	0.409
Distance from the main traffic artery	Distance to main traffic artery (kilometers)	1.093	2.957
Agricultural insurance	1 = Purchased agricultural insurance, 0 = Did not purchase agricultural insurance	0.437	0.496
Soil quality	1 = Very poor, 2 = Poor, 3 = Average, 4 = Good, 5 = Excellent	3.784	0.875
Trust relationship	Do you trust your relatives, neighbors, friends, etc.? 1 = Yes, 0 = No	0.847	0.360
Decision consensus	Do you believe that the behavior of other farmers influences your behavior? 1 = Yes, 0 = No	0.503	0.500
Information sharing	Did you receive information from others about disaster prevention, etc., before the disaster? 1 = Yes, 0 = No	0.759	0.428

.

4. Results

4.1. Model Regression Results

In this study, the adoption of cooperative pest control behaviors by farmers was assessed as the outcome variable using full information maximum likelihood estimation for Equations (1) and (2). The results are presented in

Table 2. Regression results for farmers’ adoption of cooperative pest control behavior (Endogenous switching probit model).

Variable Name	Selection Equation	Result Equation
	Whether Social Learning Is Conducted	Social Learning	No Social Learning
Gender	0.117 (0.82)	0.352 ** (1.98)	0.162 (0.89)
Age	0.003 (0.75)	−0.016 *** (−3.50)	−0.008 (−1.55)
Health status	−0.099 ** (−2.29)	−0.022 (−0.45)	0.015 (0.28)
Income level	−0.016 *** (−3.50)	0.010 * (1.95)	0.010 * (1.70)
Scale of land operation	0.019 *** (6.40)	−0.001 (−0.60)	0.012 *** (3.07)
Borrowing situation	0.506 *** (5.47)	0.074 (0.69)	0.233 ** (2.10)
Distance from the main traffic artery	−0.037 *** (−2.87)	−0.019 (−1.20)	0.029 ** (2.03)
Agricultural insurance	0.068 (0.91)	−0.338 *** (−3.69)	−0.489 *** (−5.25)
Soil quality	−0.053 (−1.25)	0.018 (0.38)	−0.203 *** (−3.55)
New Year’s Eve visits to friends and relatives	1.298 *** (2.85)
Constant term	311.9 *** (2.85)	1.609 *** (3.76)	−0.049 (−0.10)
${\rho }_1$		−1.766 *** (−3.46)
${\rho }_0$			−11.290 (−0.04)
Goodness-of-fit test for the model	111.87 ***
Log-likelihood value	−1449.14
Test of equation independence	32.90 ***

Note: Parentheses contain the t-values of the coefficients; ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

below.

The results from the endogenous switching Probit model, as presented in Table 2, reveal several key findings:

Firstly, the residual correlation coefficient ρ₁ from the model is statistically significant at the 1% level, suggesting the presence of sample selection bias [69]. This indicates that there are unobserved factors influencing both the likelihood of a farmer engaging in social learning and their adoption of cooperative pest control behaviors. Consequently, it is crucial to adjust for the endogeneity of farmers’ decisions to participate in social learning to mitigate estimation bias. Additionally, both the model’s goodness-of-fit and the tests for equation independence demonstrate significance at the 1% level, affirming the model’s robustness and appropriateness [70].

Secondly, individual and family business characteristics significantly influence farmers’ propensity to engage in social learning. Analysis of the choice equation reveals that farmers’ health status is negatively associated with their participation in social learning, significant at the 5% level. This suggests that healthier farmers, likely possessing greater resources and capabilities to address agricultural challenges independently, may rely less on social interactions for acquiring information and learning new technologies. Furthermore, the income level of the farm household, significant at the 1% level with a negative coefficient, implies that wealthier farmers are less inclined to engage in social learning. This tendency might be attributed to higher-income farmers’ ability to employ agricultural specialists or consult experts directly—advantages that diminish their reliance on and frequency of engagement in social learning activities. The coefficient for land management area is significantly positive at the 1% level, suggesting that farmers with larger land holdings are more likely to engage in social learning. This propensity may be attributed to the greater risks associated with managing larger areas, where social learning enables a more comprehensive understanding of potential risks and solutions through diverse knowledge sources, thereby enhancing their capacity to manage uncertainties and mitigate risks. Additionally, the likelihood of engaging in social learning is positively correlated at the 1% level with the propensity to borrow, indicating that indebted farmers may seek to acquire specialized knowledge and resources through social interactions to enhance their production capabilities and product quality, ultimately aiming to alleviate financial pressures. Conversely, distance from major transportation routes shows a significant negative impact at the 1% level, suggesting that farmers located farther from these arteries face higher costs in terms of time and access to learning resources, which inhibits their participation in social learning activities.

Building upon the work of Qiao, et al. [20], this study employs the number of New Year’s Eve visits to friends and relatives as an instrumental variable for social learning. These visits, which are characterized by social interactions involving the frequent exchange of information, experiences, and knowledge, serve as a measure of an individual’s engagement in social activities within the village. An active social network, as indicated by numerous New Year’s visits, facilitates significant opportunities for social learning, potentially enhancing farmers’ abilities to access new information and adopt innovative technologies. Importantly, such social interactions are unlikely to directly influence other specific agricultural outcomes, such as cooperative pest management, thereby satisfying the exogeneity requirement for an effective instrumental variable. According to the findings presented in Table 1, the impact of New Year’s Eve visits on social learning is statistically significant at the 1% level.

Thirdly, the regression analysis for the adoption of cooperative pest control behavior exhibits distinct outcomes between the treatment group (S = 1) and the control group (S = 0). In this analysis, male farmers significantly influenced the adoption of cooperative pest control behaviors among those involved in social learning, suggesting that these farmers are more likely to facilitate such practices. Conversely, among farmers who did not engage in social learning, male farmers still played a substantial role in the adoption of these behaviors. This influence may stem from male farmers’ tendency to achieve consensus through regular interactions with other farmers, experts, and institutions, thereby increasing their propensity to implement cooperative pest control measures. Additionally, the analysis revealed that older farmers engaged in social learning were less likely to promote cooperative pest control behaviors, while age did not significantly affect the adoption of these practices among farmers who were not socially learned. Social learning theory posits that individuals acquire skills by observing and imitating others. Within this framework, age can notably influence the adoption of cooperative pest control behaviors. Older farmers, for instance, may prefer to maintain existing practices, exhibiting reluctance toward adopting new knowledge and ideas. Furthermore, factors such as the size of landholdings and borrowing status significantly and positively affect the adoption of cooperative pest control behaviors among farmers who are not engaged in social learning, in contrast to those who are. This discrepancy may arise because farmers who do not participate in social learning—and who manage larger tracts of land and carry greater debt—are more compelled to adopt cooperative pest control practices to mitigate risks and enhance yields and revenues. Such strategies are crucial for maximizing financial returns and ensuring loan repayment capacity. Proximity to major transportation routes significantly influenced the adoption of cooperative pest control behaviors among farmers not engaged in social learning, possibly because those farther from these routes lacked easy access to external resources and support. However, this disadvantage might be mitigated by collaborative engagements in agricultural production with other farmers. Conversely, the soil quality of farms managed by socially learned farmers did not significantly impact the likelihood of adopting cooperative pest control strategies compared to their non-social learning counterparts, although a positive coefficient suggests that superior soil quality correlates with a higher propensity for such practices. This may be attributed to the fact that better soil quality generally enhances crop health and resistance, prompting farmers to implement cooperative pest control measures more readily. This approach not only increases the effectiveness of these measures but also boosts crop yields and the overall economic benefits for farmers confronting pest issues.

4.2. Treatment Effect Estimation

The average treatment effect of social learning on the adoption of cooperative pest control behaviors among farmers is calculated using Equations (3)–(5), as presented in

Table 3. Average treatment effect of social learning on farmers’ adoption of cooperative pest control behavior.

	Endogenous Switching Probit Model			Probit Model
	Social Learning	No Social Learning (Counterfactual)	ATT	Marginal Effect
The probability of farmers adopting cooperative pest control behavior	0.300	0.198	0.102 *** (0.005)	0.07 *** (0.008)

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

. The data show that farmers engaged in social learning have a 30% likelihood of adopting cooperative pest control measures. Conversely, in a counterfactual scenario where these farmers do not participate in social learning, the likelihood decreases to 19.8%. Additionally, the estimated Average Treatment Effect on the Treated (ATT) value, detailed in Table 3, is 0.102, reaching statistical significance at the 1% level. This indicates that farmers who engage in social learning are 10.2 percentage points more likely to adopt cooperative pest control practices compared to their likelihood in the hypothetical scenario where they do not engage in social learning. This confirms the validation of Hypothesis 1.

The final column of Table 3 presents the marginal effects of social learning, estimated using the Probit model. This analysis distinctly reveals that the average treatment effect on the probability that farmers will adopt cooperative pest control behaviors is underestimated by 3.2 percentage points if the endogeneity issue within the model is not addressed. Consequently, employing a robust econometric model to correct for biases arising from endogeneity is essential for accurately estimating the treatment effect. This approach ensures that the influence of social learning on pest control practices is correctly quantified, highlighting the importance of accounting for underlying biases in econometric evaluations.

4.3. Robustness Check

To verify the robustness of the aforementioned estimation results, this study employs a recursive binary probit model to further analyze the impact of social learning on the adoption of cooperative pest control behaviors among farmers. In this model, the variable indicating whether farmers are social learners is included as a factor influencing the adoption of cooperative pest control behaviors, but this does not reciprocally influence the model assessing the characteristics of social learners. This approach differs from the endogenous switching probit model in that the recursive binary probit model relaxes the assumption of equation independence, allowing for more flexible modeling of the relationships between variables [20,71]. As depicted in columns 2 and 3 of

Table 4. Robustness test results for the impact of social learning on farmers’ adoption of cooperative pest control behavior.

Variable Name	Recursive Binary Probit Model		IV Probit Model
	Social Learning	Farmers’ Cooperative Pest Control Behavior	Social Learning	Farmers’ Cooperative Pest Control Behavior
New Year’s Eve visits to friends and relatives	0.087 ** (2.53)		0.039 ** (2.44)
Social learning		1.547 *** (25.68)		3.001 * (1.93)
Constant term	0.128 (0.32)	−0.289 (−0.81)	0.540 *** (3.45)	−0.404 (−0.34)
Control variable	Controlled	Controlled	Controlled	Controlled
Residual correlation coefficient		−1.860 *** (−5.36)
Wald test of exogeneity				4.38 **
AR				6.28 **
Wald				3.73 *

Note: Parentheses contain the t-values of the coefficients. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively.

, the regression outcomes for the core independent variables align with findings from the previous sections, confirming that social learning facilitates the adoption of cooperative pest control behaviors among farmers. Additionally, this study employs an IV Probit model to further examine the impact of social learning on the adoption of these behaviors. This approach contrasts with the endogenous switching Probit model typically applied via the two-stage least squares (2SLS) method. Results presented in columns 4 and 5 of Table 4 reinforce that social learning exerts a positive and significant influence on the likelihood of farmers engaging in cooperative pest control practices. This substantiates the reliability of the preceding conclusion, thereby confirming Hypothesis 1.

This study further employs the propensity score matching (PSM) technique to assess the robustness of the baseline estimation results regarding social learning’s impact on farmers’ adoption of cooperative pest control behaviors. Initially, as illustrated in Figure 3, the propensity score intervals post-matching exhibit enhanced overlap and consistency in probability densities compared to pre-matching, signifying that the matched data satisfy the common support domain criteria and exhibit improved matching quality. Moreover, the robustness of these results is corroborated through the application of four distinct matching methods: k-nearest neighbor matching (one-to-one matching), intra-caliper k-nearest neighbor matching (one-to-four matching with a caliper range set at 0.01), radius (caliper) matching (with a caliper range set at 0.05), and kernel matching (utilizing the default kernel function and bandwidth). These methods and their outcomes are detailed in

Table 5. Average treatment effect of social learning on farmers’ adoption of cooperative pest control behavior.

Matching Method	Treatment Group Mean	Control Group Mean	ATT	Standard Error	T-Value
K-nearest neighbor matching	0.673	0.530	0.143 ***	0.034	4.23
K-nearest neighbor Matching in caliper	0.781	0.635	0.146 ***	0.032	4.58
Radius (caliper) match	0.783	0.642	0.141 ***	0.029	4.89
Kernel matching	0.783	0.642	0.140 ***	0.029	4.85
Mean value	0.755	0.612	0.143 ***

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

. The average treatment effect on the treated (ATT) across the four matching methods is estimated at 0.143, indicating that farmers who engage in social learning are 14.3% more likely to adopt cooperative pest control behaviors compared to those who do not. This significant positive effect underscores the efficacy of social learning, aligning with results from the endogenous switching probit model. Notably, the average treatment effect derived through k-nearest neighbor matching, intra-caliper matching, radius (caliper) matching, and kernel matching exceeds that from the endogenous transformation probit model, which estimated an effect of 10.3%. This discrepancy might stem from the propensity score matching (PSM) model’s limitation in addressing the impact of unobserved variables, potentially leading to biased outcomes. Conversely, the endogenous transformation probit model integrates both observable and unobservable factors, thoroughly accounting for selectivity bias. It incorporates a bias correction term directly into the outcome equation’s estimation, yielding more robust and scientifically valid results.

4.4. Heterogeneity Analysis

The empirical findings presented above demonstrate that social learning significantly influences farmers’ adoption of cooperative pest control behaviors. However, the impact appears to vary according to different learning styles. To further investigate this variability, the current study conducts separate endogenous transformation probit regressions for each learning style. The detailed results of these regressions are displayed in

Table 6. Differences in the effects of various learning methods on farmers’ adoption of cooperative pest control behavior.

Variable	Social Learning	No Social Learning	Average Treatment Effect	Standard Error	T-Value
Learn from relatives, neighbors, and friends	0.331	0.300	0.032 ***	0.005	5.837
Learn from the agricultural technology department	0.201	0.139	0.062 ***	0.004	17.247
Online learning	0.247	0.187	0.059 ***	0.006	10.521

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

. This analysis aims to discern the distinct effects that various styles of learning have on the implementation of cooperative pest control measures among farmers.

The probability of farmers adopting cooperative pest control behaviors after learning from their relatives, neighbors, and friends stands at 33.1%. Conversely, in a counterfactual scenario where these farmers did not receive such social learning, the probability drops to 30%. The Average Treatment Effect on the Treated (ATT) for this group is 3.2%, significant at the 1% statistical level. This indicates that the likelihood of adopting cooperative pest control behaviors is 3.2 percentage points higher among farmers who engaged in social learning from their social circle compared to those who did not. Similarly, the probability increases by 6.2 percentage points for those who learned from agricultural departments and by 5.9 percentage points for those who utilized Internet resources to learn about pest control, compared to their counterparts who did not use these sources of learning. These differences underscore the varied impact of different learning sources on the adoption of cooperative pest control strategies.

The impact of social learning on the adoption of cooperative pest control behaviors among farmers may vary due to different socio-economic conditions and resource constraints associated with varying income levels. Social learning theory posits that farmers acquire agriculture-related information and technology through their social networks; however, the structure and efficacy of these networks can differ significantly based on income, influencing farmers’ decisions regarding cooperative pest control practices. To explore these dynamics, this study segmented farmers into two groups based on income levels—above and below the mean income. An endogenous switching probit model was employed to estimate the effects, with results detailed in

Table 7. Results on group differences in the impact of social learning on farmers’ adoption of cooperative pest control behavior.

Variable	Grouping	Social Learning	No Social Learning	Average Treatment Effect	Standard Error	T-Value
Income level	High-score group	0.295	0.179	0.115 ***	0.008	14.16
	Low-score group	0.304	0.209	0.094 ***	0.007	13.98
Scale of land operation	High-score group	0.444	0.171	0.273 ***	0.010	26.18
	Low-score group	0.248	0.207	0.041 ***	0.005	8.29

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

. The analysis reveals that social learning markedly enhances the adoption of cooperative pest control behaviors across both income groups, with adoption probabilities increasing by 11.5% in the higher-income group and 9.4% in the lower-income group.

As agricultural operations scale up, disparities in expertise, access to technology, and social capital emerge among farmers of varying sizes. Large-scale farmers, in comparison to their small-scale counterparts, typically possess greater expertise and a more nuanced understanding of the benefits and drawbacks associated with joint pest control efforts. Consequently, they are more likely to adopt cooperative pest control behaviors as a strategy to mitigate risks effectively. Drawing on Xu, et al. [72], this study categorizes farmers based on the mean value of the land operation area—those above the mean are classified within the higher grouping, and those below within the lower grouping. Utilizing an endogenous transformed probit model for the analysis, the results, detailed in Table 7, demonstrate that social learning markedly enhances the adoption of cooperative pest control practices across both groups. Specifically, the probability of adopting such behaviors increased by 27.3% among farmers with larger operational areas, compared to a 4.1% increase among those with smaller areas, highlighting a significant disparity in the levels of adoption.

4.5. Mechanism Analysis

The theoretical analysis suggesting the existence of transmission pathways—such as trust relationships, decision consensus, and information sharing—between social learning and farmers’ adoption of cooperative pest control behaviors necessitates empirical validation to ascertain the magnitude of these mediating effects. Accordingly, this study employs the nonparametric percentile Bootstrap method with bias correction to evaluate the mediating influences of trust relationships, decision consensus, and information sharing. The methodology involves setting the random sampling capacity to 1000 and establishing a confidence level of 95%. The significance of the mediation effects is determined by examining whether the confidence intervals exclude zero; an interval that does not contain zero indicates a significant mediation effect, whereas an interval that does include zero suggests the effect is not significant. The results of these tests are presented in

Table 8. Analysis results of the transmission mechanism of social learning on farmers’ adoption of cooperative pest control behavior.

Independent Variable	Mediating Variable	Dependent Variable	Effect	Boot SE	Boot A	Boot B	Mediation Effect
Social learning	Trust relationship	Farmers’ cooperative pest control behavior	0.005 *	0.003	0.001	0.012	Yes
Social learning	Trust relationship	-	0.173 ***	0.027	0.126	0.228	-
Social learning	-	Farmers’ cooperative pest control behavior	0.177 ***	0.027	0.131	0.232	-
Social learning	Decision consensus	Farmers’ cooperative pest control behavior	0.006 *	0.003	0.001	0.015	Yes
Social learning	Decision consensus	-	0.172 ***	0.027	0.118	0.221	-
Social learning	-	Farmers’ cooperative pest control behavior	0.177 ***	0.026	0.125	0.227	-
Social learning	Information sharing	Farmers’ cooperative pest control behavior	0.030 ***	0.009	0.014	0.050	Yes
Social learning	Information sharing	-	0.148 ***	0.025	0.099	0.200	-
Social learning	-	Farmers’ cooperative pest control behavior	0.178 ***	0.027	0.131	0.233	-

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

.

The mediation effect of trust relationships was quantified at 0.005, with a 95% confidence interval ranging from 0.001 to 0.012; for decision consensus, the mediation effect was 0.006, with a confidence interval of 0.001 to 0.015; and for information sharing, it was 0.030, with a confidence interval of 0.014 to 0.050. None of these confidence intervals include zero, indicating that all mediation effects are statistically significant. This confirms that social learning enhances farmers’ adoption of cooperative pest control behaviors through the establishment of trust relationships, achievement of decision-making consensus, and facilitation of information sharing. Thus, the proposed theoretical relationships are empirically supported. Hypotheses 2, 3, and 4, as previously discussed, are validated.

5. Discussion

In the realm of pest control, farmers predominantly adopt independent management strategies. A significant drawback of these strategies is the mobility of pests, which often results in ineffective control. Specifically, confining pest management to individual plots without accounting for adjacent areas can lead to recurrent problems due to pest incursions from neighboring lands. Thus, a collaborative control approach may serve as a viable remedy for the shortcomings of independent management practices. Notably, prior research has predominantly focused on the adoption of green control technologies [28,73,74], frequently neglecting the critical role of regional cooperative pest management. This article extends the discourse by incorporating insights from related studies [20,21,30] and delineating social learning into three distinct categories: learning from relatives, neighbors, and friends (neighborhood communication); from agricultural departments (through training); and via online platforms (self-directed learning). It highlights the influence of social learning on the adoption of cooperative pest management practices, undertakes a heterogeneity analysis based on different social learning modes, income levels, and land management sizes, and introduces concepts such as trust relationships, decision-making consensus, and information sharing to explore potential mechanisms that may impact these outcomes.

The research indicates that social learning exerts a significant positive influence on farmers’ adoption of cooperative pest management practices. This effect is likely attributable to the enhanced willingness among farmers to engage in cooperative efforts, as social learning cultivates mutual trust through the exchange of experiences and shared learning [35,36,37], thereby facilitating these management behaviors. Moreover, social learning enables farmers to comprehend the synergistic effects of pest control, which bolsters their cooperative consciousness [42] and fosters consensus in collective decision-making. Social learning further encourages information sharing among farmers [26], enriching their knowledge about pest management strategies. Consequently, the dissemination of experiences can prompt imitative behaviors, leading to identification imitation [31,36], which in turn augments cooperative pest management activities.

Furthermore, social learning mechanisms, such as learning from relatives, neighbors, and friends (neighborhood communication), agricultural technology departments (via training), and online platforms (self-directed learning), have all been found to significantly enhance farmers’ engagement in cooperative pest management practices. Counterfactual analyses reveal that learning from agricultural technology departments and online platforms is associated with a notably higher likelihood of adopting these practices, compared to learning from personal contacts. This is likely because learning from formalized and digital sources tends to provide more scientific, systematic, and accurate information about pest control. Such education assists farmers in comprehending joint control strategies and the associated benefits. Moreover, these learning avenues supply extensive resources and information, including epidemic prevention strategies and pest monitoring techniques, thereby offering substantial support for pest management and consequently increasing the propensity for cooperative behavior in this context [75,76]. The analysis further examines heterogeneity in income levels and land management sizes, demonstrating that social learning significantly enhances the adoption of cooperative pest management practices among high-income farmers. This may be attributed primarily to the greater accessibility of agricultural knowledge and technologies for these farmers, who often participate in agricultural training and maintain relationships with experts to acquire up-to-date information. Such interactions elevate the cognitive understanding of high-income farmers, fostering awareness of the benefits of joint pest management and promoting their engagement in these practices. In a similar vein, social learning exerts a pronounced effect on farmers with extensive landholdings, who are likely to adopt cooperative management strategies. Farmers managing larger tracts of land generally encounter heightened risks, and cooperative approaches can mitigate these risks and potential losses. Furthermore, as these farmers depend more heavily on agricultural income, they are incentivized to pursue cooperative strategies to optimize pest control and thereby enhance agricultural productivity.

6. Conclusions

Amid escalating resource and environmental constraints, coupled with the inflexible growth in food demand, China faces the urgent challenge of enhancing disaster resilience and agricultural risk management to achieve green and sustainable development in agriculture. In this context, the study employs micro-level survey data from farmers in Northwest China to examine social learning across three dimensions: learning from relatives, neighbors, and friends (neighborhood communication); learning from agricultural technical departments (participation in training); and online learning (self-directed learning). Cooperative pest control behavior is defined by whether farmers collaborate with neighboring plots for integrated pest management. Utilizing an econometric model, the study constructs a “counterfactual” analysis framework to estimate the average treatment effect of social learning on the likelihood of farmers adopting cooperative pest control practices. Additionally, the study evaluates the average treatment effects of different learning methods on this likelihood and delves into the mechanisms through which social learning influences farmers’ adoption of these practices. The study reaches the following conclusions:

First, social learning plays a critical role in enhancing farmers’ adoption of cooperative pest control practices. In a counterfactual scenario, farmers who have not engaged in social learning would see a 10.2% increase in the likelihood of adopting these practices if they were to do so. However, when the issue of endogeneity is overlooked, this estimated increase drops to just 7%, suggesting that the true impact of social learning on farmers’ adoption of cooperative pest control practices has been underestimated. The findings demonstrate robustness even after accounting for endogeneity concerns. Moreover, variables such as the health status of farmers, their income levels, the extent of land they manage, their borrowing status, and proximity to major transportation arteries significantly influence the likelihood of farmers participating in social learning.

Second, given the variety of learning methods and distinct characteristics among farmers, the influence of social learning on the adoption of cooperative pest management strategies is likely to differ. Consequently, this study utilizes an endogenous switching probit model to assess how various learning modalities impact farmers’ decisions to engage in cooperative pest management. The findings, predicated on a counterfactual scenario, suggest that farmers who acquire knowledge from agricultural technology departments or through digital networks are more likely to implement cooperative pest management strategies compared to those who learn from family, neighbors, and friends. Furthermore, the impact of social learning on the adoption of cooperative pest management practices varies significantly across different income and land size categories. Specifically, the likelihood of implementing these practices has risen by 11.5% among high-income farmers and by 9.4% among low-income farmers. Similarly, for farmers managing larger tracts of land, social learning has facilitated a 27.3% increase in adoption rates, whereas for those with smaller land areas, the increase is markedly lower at 4.1%.

Third, social learning plays a pivotal role in promoting the adoption of cooperative pest management practices among farmers. This process is facilitated through the establishment of trust relationships, achievement of consensus in decision-making, and enhancement of information sharing among participants.

Based on the research findings, several implications emerge: First, collaborative action among farmers is effectively promoted through experience sharing, the development of cooperatives, and targeted technical training. To facilitate this learning process, it is essential for village administrative bodies to organize sessions where farmers with proven success can disseminate their agricultural expertise and experiences, thus enhancing peer learning and knowledge transfer. Efforts should be intensified to foster the establishment and growth of farmers’ cooperatives and collective organizations. Simultaneously, it is crucial to encourage farmer participation in these entities. By actively engaging in these groups, farmers are equipped to enhance their competencies in pest management through continuous learning and interaction. Village administrative bodies should regularly schedule agricultural technology training sessions that leverage online platforms to facilitate practical demonstrations by agricultural experts from neighboring towns. This strategy continually enhances the cognitive capabilities of farmers, equipping them with innovative techniques and methods for managing pests. Such collective activities and varied learning methodologies are instrumental in effectively fostering cooperative pest management practices among farmers. Second, establishing mutual aid organizations, information sharing platforms, and enhancing educational outreach are crucial for fostering cooperation and joint management strategies among farmers. The creation of mutual aid organizations promotes communication and collaboration among farmers, simultaneously building trust to collectively tackle challenges such as pests. Within these organizations, it is advantageous to form specialized groups dedicated to pest control, which can facilitate regular discussions and sharing of management strategies. Furthermore, it is imperative to develop information-sharing platforms that provide farmers with up-to-date information on pest control measures, including relevant policies, technologies, and the associated costs and benefits. Enhancing educational initiatives is also essential, as it strengthens farmers’ understanding and acceptance of the significance of collaborative pest control efforts, thus fostering a consensus on cooperative management and promoting the sustainable development of agriculture. Third, governments must advance the specialization of pest control by leveraging digital technologies to augment monitoring and forecasting capabilities, thereby enhancing agricultural risk management. Enhanced governmental support for pest control initiatives is crucial to deepen the professionalization of service organizations, facilitating specialized guidance and collaborative governance. Additionally, it is imperative to further delineate monitoring roles, refine crop pest research strategies, and bolster the infrastructure of field monitoring points and forecasting stations. Implementing digital tools such as drones, remote sensing technologies, Internet of Things (IoT) technology, and data analytics will significantly improve monitoring and forecasting precision. Collectively, these initiatives will progressively fortify crop pest control systems, substantially increasing the agricultural sector’s capacity to mitigate risks.

This study significantly contributes to theoretical development by enhancing the application of social learning and collective action theories within the context of pest management and sustainable agriculture. It enriches the theoretical foundation for these areas and underscores the importance of incorporating social learning into the formulation of rural development policies. Furthermore, the research advances interdisciplinary methodologies, offering a comprehensive analytical framework for addressing agricultural and rural challenges. This approach deepens our understanding of household behavioral dynamics and aids in the scientific formulation of relevant policies. Finally, this study acknowledges certain limitations that warrant further in-depth analysis. The analysis relies solely on binary variables to determine whether farmers engage in cooperative pest control behaviors. Due to data constraints, these variables do not account for the nuances between fully independent and partially joint control scenarios. Future research could explore these distinctions more thoroughly, thereby refining and expanding upon the conclusions drawn in this study. This study focuses on the provinces of Shaanxi, Gansu, and Ningxia, which represent the arid and semi-arid regions of Northwest China, characterized by a high incidence and widespread scale of pest and disease outbreaks. By using pest control practices in Northwest China as a case study, the findings may provide valuable insights for pest management in other regions of China and in developing countries with similar ecological conditions. Nevertheless, due to variations in resource endowments, agricultural practices, and pest types across different regions, future research should aim to expand the scope of analysis. Conducting annual follow-up surveys to generate time series data would enable longitudinal studies that explore the dynamic evolution of social learning and the adoption of cooperative pest control methods. Such research could enhance the generalizability of the findings.