The Effect of Social Network on Controlled-Release Fertilizer Use: Evidence from Rice Large-Scale Farmers in Jiangsu Province, China

Abstract

The reduction and efficiency of fertilizer use has been a recent focus of governments and scholars. As a new agricultural technology, controlled-release fertilizer can not only increase yield and save labor, but also improve efficiency and reduce the use of fertilizer, thus promoting sustainable agricultural development. Drawing on a sample of 231 farmers of Jiangsu Province, China, this paper applies a probit model to assess the adoption behavior of controlled-release fertilizer by large-scale households in terms of three dimensions of social network, i.e., communication intensity, trust level, and network size, specifically exploring how science popularization influences their adoption intention, and comparing the heterogeneity of impact that social network has on the adoption intention of farmers when the information is obtained adequately or not. The empirical results demonstrate that: (1) At the early stage of technology diffusion, the size of social network has a positive effect on farmers’ cognition of controlled-release fertilizer, and the communication intensity with neighboring farmers has a positive effect on the adoption behavior of controlled-release fertilizer; (2) Farmers’ adoption intention of controlled-release fertilizer is significantly influenced by their original knowledge of new technology and science popularization; (3) When the information is sufficient, the social network of large-scale households has no significant effect on their willingness to adopt. Therefore, in promoting controlled-release fertilizer, the government should highlight the synergistic effect of farmers’ cognition and science popularization activities, fully consider the characteristics of farmers’ social network, facilitate the infrastructure of rural informatization, and regulate the agricultural promotion networks so that farmers can obtain sufficient and effective information.

1. Introduction

The overuse of chemical fertilizers in Chinese agricultural production is very serious [1]. FAO statistics show that in China, 397.07 kg of chemical fertilizer was used in arable land per hectare in 2019, which is about 2.91 times the world average (136.29 kg) [2]. The massive use of chemical fertilizers increases grain production significantly [3], but the traditional quick-release one has also exacerbated the issues of greenhouse gas emissions, water pollution, and soil contamination [4]. To fix conventional fertilizer issues, the use of controlled-released fertilizers (Hereinafter referred to as CRFs) is recommended [5]. The nutrients in CRFs can be released at a set rate through various regulating mechanisms, at a rate consistent with the nutrient uptake requirements of the crop, thus reducing the harmful environmental impact of unconverted fertilizers. As the most important food crop in China, there is a large demand for rice. Therefore, it is especially important to actively explore the feasible green path for the transformation of rice production methods, so that the utilization efficiency of fertilizer can be improved, and the pollution rate can be reduced, which can not only increase food production and quality, but also ensure the “safety on the tongue” of the residents while maintaining the sustainable development of the ecological environment [6,7,8,9,10,11,12].

As a green agricultural technology, CRFs can control the release of nutrients, which could effectively improve nutrient utilization, meet the nutrient demand of crops during growth, and help increase the quality and safety of agricultural products while reducing pollution. As a “rational man”, farmers will conduct a cost–benefit analysis and decide on whether to adopt CRFs. Studies have shown that CRFs can effectively reduce the quantity of fertilizer use and change the traditional habit of multiple use, yet still promote crop yield, reduce production costs, and increase planting profit [13,14]. Although the production of CRFs is more expensive than that of conventional fertilizers, with large-scale production these costs may drop [15]. Owing to rising energy costs and labor shortages in agriculture, CRFs will be more attractive, and farmers will have the motivation to adopt CRFs from the cost–benefit perspective. However, at the early stage of technology diffusion, farmers’ perceptions of CRFs are characterized by information asymmetry. When the income increase per unit area is certain, the larger the total arable land area is, the higher the total return is. Therefore, large-scale agricultural operators tend to have stronger incentives to obtain information about new technology and adopt it.

Existing studies mostly focus on the effect of CRFs itself, and there are few studies on farmers’ adoption behavior. The application of a technology starts from research and development to small-scale pilot, and then to large-scale promotion, which is a gradual process and cannot be leapfrogged. At present, CRFs are still being piloted on a small-scale, in which people tend to decide whether to adopt the new technology through initial private signals or observation of neighbors’ behavior in social networks [16]. Unlike the period promoting straw return and formula fertilization, the Internet is developing rapidly today, people have more diversified access to information, and the scope of social networks has been further expanded, from acquaintance circles to geographic interest community. In China, the mobile network is one of the important sources for farmers to obtain new knowledge, so we choose mobile phone bills as a proxy variable for farmers’ Internet use. It is of strategic importance to further promote CRFs by exploring the influence of social networks on farmers’ CRF adoption behavior.

Therefore, from the perspective of information asymmetry, this paper investigates the impact of different social network dimensions on the adoption of CRFs by large-scale farmers at different stages respectively. The research questions to be answered are: How do the social network dimensions affect farmers’ perceptions and adoption behavior in the early stage of technology diffusion? Are farmers not adopting CRFs because they lack proper knowledge of the technology? Do social networks influence the willingness of large-scale farmers to adopt when information is available? What is the most appropriate approach in terms of extension efficiency?

The marginal contribution of this study is: taking the mobile phone bill as the proxy variable for farmers’ use of social network, measuring the capability of obtaining relevant information from their unfamiliar net friends, and rethinking their cognition to the short-term use behavior and long-term adoption intention. At the theoretical level, through the application of econometric models such as the probit model, this study lays a solid micro-theoretical foundation for the study of deeper agricultural social service issues. At the practical level, this study is conducive to promoting a detailed and oriented training service system for farmers and puts forward relevant countermeasure suggestions with a view to providing reference for the policy decisions of relevant departments, which is of great importance for the correct adjustment of farmers’ fertilizer application decisions, the reduction of agricultural ecological environment pollution, and the sustainable development of agriculture in the future.

2. Background of CRF Extension

Fertilizer is an indispensable factor of production in agricultural production. However, the long-term excessive and inefficient application of chemical fertilizers has brought about agricultural non-point source pollution and environmental quality decline, which is a real contradiction limiting the sustainable development and green transformation of Chinese agriculture. Since the Ministry of Agriculture proposed the “Action Plan for Zero Growth in Chemical Fertilizer Use by 2020” in 2015, China has implemented a number of measures to promote fertilizer reduction and efficiency; after that, the government has actively promoted the rational application of fertilizers, such as the General Office of the CPC Central Committee, the General Office of the State Council, and the Ministry of Agriculture and Rural Development promulgated the “Opinions on Innovative Institutional Mechanisms to Promote Green Agricultural Development” in 2017 and the “Technical Guidelines for Green Agricultural Development (2018–2030)” in 2018, aiming to build a green production technology system featuring “safe and harmless agricultural inputs and environmentally friendly production processes”, to fully activate the endogenous power of green agricultural development, and to vigorously increase the supply of green and high-quality agricultural products and turn green into efficiency. Furthermore, the Central Document No. 1 was released in China in 2019, which emphasizes the need to “intensify efforts to combat agricultural non-point source pollution, carry out fertilizer and biocide conservation actions, and achieve negative growth in fertilizer and pesticide use”. China’s policy on fertilizer application and control has gradually evolved from “increase the amount and yield” to “reduce the amount and increase the efficiency”.

The efficiency of traditional fertilizers is generally low, and the unconverted ones can be harmful to the environment. Therefore, improving the efficiency of fertilizers is an important way to achieve the coordinated development of China’s agricultural industry and environmental resources [17]. “Controlled-release” refers to the release mode in which nutrients are released at a set rate through various regulatory mechanisms, and the release rate is consistent with the nutrient absorption requirements of crops [18]. Field trials in different regions have shown that CRFs have a positive effect on rice tillering and yield. It can improve the stress resistance and increase the effective panicle, seed-setting rate, and thousand grain weight of rice, thus achieving the goal of enhancing the fertilizer efficiency. Studies have shown that the use of CRFs have significant economic and ecological benefits, specifically on reducing fertilizer application, saving labor costs, and improving fertilizer efficiency [19,20,21,22].

With the low agricultural mechanization level and the labor shortage in China, disposable fertilizer application technology provides an effective way to achieve sustainable grain production, and CRFs are one of the technologies. Whether existing agricultural machinery is suitable for CRFs determines the diffusion effectiveness [23]. In the past, bagged CRFs relied on manual trenching and fertilizing, which required a large amount of labor. Due to the lack of suitable fertilization technology, CRFs were limited in large-scale promotion and could not achieve the purpose of improving overall fertilizer efficiency and agricultural ecological environment. At present, the combination of synchronized side-deep fertilization technology and CRFs has achieved good results. By using the side-deep fertilization technology, CRFs can achieve one-time operation during rice transplanting, thus reducing the intensity of field operation and improving the application effect. In addition, by using the technology, CRFs can also be applied to a fixed location to ensure positioning, quantification, and uniform fertilization, saving cost and labor while increasing rice production stably.

3. Literature Review

The adoption of green agricultural technology has always been a great concern of scholars. Existing studies mostly focus on the technology of straw return, formula fertilization, and conservation tillage [24,25,26]. There are relatively few articles on CRFs, a new green agricultural technology. As an agricultural technology of fertilizer use reduction, CRFs should have similar influencing factors, transmission modes and diffusion pathways as other green agricultural technologies. Therefore, this part will review the relevant literature from two aspects: firstly, the adoption of fertilizer use reduction technology by farmers; secondly, the diffusion and promotion of agricultural technology.

Most of the existing studies are based on the hypothesis of “economic man” or “social man”. Studies with the hypothesis of “economic man” believe that the expected benefits brought by technology are an important factor affecting the adoption behavior of farmers [27]. Some scholars also believe that social motivation also plays an important role in the adoption decision of farmers. The “semi-acquaintance society” characteristic of rural social environment makes the “social circle” of acquaintance association become a common way of interaction and communication among farmers [28,29]. Previous studies have shown that the role of “social man” of farmers has a particularly important impact on the adoption of new technologies. In the rural environment, neighborhood demonstration and communication can reduce the risk expectation of farmers to a certain extent, thus increasing their adoption enthusiasm [30].

Research on factors influencing technology adoption behavior can be roughly divided into two aspects. Firstly, as to the adopting intention and influence factors, academic literature recognizes that individual characteristics, family resources endowment, social and economic networks, non-agricultural and part-time employment, policy system, technology application environment, and natural conditions have significant effects on technology adoption [27,30,31,32,33]. Secondly, concerning the impact of a specific factor on new technology adoption, it is believed that the adoption cost, risk and technical difficulty, policy tools, and cognitive conflict have impact on the adoption behavior of farmers to varying extents [31,34,35].

It is generally believed that the diffusion of agricultural technology has an important impact on the adoption behavior of farmers. Empirical studies on the diffusion of agricultural technology show that the diffusion of new technology among farmers is a gradual process. Adoption of new methods usually starts through the actions of a few innovators and then spreads rapidly as more people reach out to previous adopters in their social networks [36]. Therefore, the adoption of agricultural technology is usually closely related to non-economic factors such as social interaction among farmers [17], and the number of technology adopters often shows an “S-shaped” growth trend [37]. As micro subjects and technology adopters of agricultural production, farmers’ green production behavior is fundamental to solve the quality and safety problems of agricultural products and plays an important role in transforming agricultural production modes [38]. Scholars have explained the factors affecting the promotion of new agricultural technologies in rural areas from different perspectives. From the perspective of individual characteristics of farmers, factors such as gender, age, education level, planting experience, and political identity are directly related to the adoption of agricultural technologies [39,40,41,42,43]. At the household level, household income, income structure, number of household labor force, Internet installation, and quantity of land also affect the adoption behavior of farmers [44,45,46,47]. In addition, scholars also believe that farmers’ insufficient knowledge of agricultural technology is the main factor hindering their adoption of technology [48,49]; as such, limited information dissemination channels may become a major obstacle to the early diffusion of new technologies [50,51].

To sum up, this paper took CRFs, a new green agricultural technology, as the research object, and mobile phone bill as a proxy variable for the extent of farmers’ Internet use, which is an important component of farmers’ social network, to explore the influence of social network on the behavior of large-scale operators to use CRFs. On this basis, some families were randomly selected to receive education on the knowledge of CRFs, so as to explore the influence of science popularization on the adoption intention of scale operators and compare the heterogeneity of impact that social network has on their expected adoption intention when the information is sufficient or not, hoping to provide reference for the promotion of CRFs.

4. Theoretical Analysis and Hypotheses

In the process of agricultural production, farmers always make “optimal” production decisions. They can make rapid and accurate adjustments according to market price changes and improve productivity by reconfiguring production factors [52]. Economic theory holds that, as “rational “, farmers’ willingness to adopt technology is an economic behavior, which means after making a cost–benefit analysis and combining with their own resource endowment conditions, they would choose appropriate ones under the goal of profit maximization [27]. Taking CRFs as an example, previous studies have shown that the use of CRFs can significantly increase production, save fertilizer and labor, and improve efficiency. However, the total cost of adopting a certain agricultural technology also includes the time and effort that farmers contribute to it in order to grasp it. Therefore, based on the above analysis, this paper puts forward three research hypotheses.

Social network refers to the formal and informal relationships built by the flow of information, material, and other resources in the process of interaction between subjects [53], which is formed by individual decisions. According to social network theory, due to the lack of information access channels, farmers are often in a state of incomplete information [54], which can be alleviated by obtaining information and learning knowledge in social networks [55]. However, access to information and learning comes at a cost. People often need to weigh the costs against the potential returns when building and maintaining social networks of connections. Different from the restriction of spatial distance and communication mode in the past, today in the early stage of technology transmission, farmers can obtain the revision or supplement of new technology-related knowledge through mobile phones in social networks such as WeChat and TikTok, and their cognitive status may change during this process. In addition, for a technology that has not been rolled out locally on a large scale, even farmers have obtained some knowledge from remote social interaction; as agricultural production is greatly influenced by regional environment, farmers may still pay special attention to their neighboring farmers’ behavior, communicating and imitating to a certain degree when making decisions to use CRFs. Based on the above analysis, we pose the following hypothesis:

Hypothesis 1 (H1).

In the early stage of technology dissemination, farmers’ cognitive level of CRFs will be affected by the scale of social networks, and their use behavior will be affected by the intensity of communication.

From the perspective of economics, it is usually assumed that farmers make rational production decisions with the goal of maximizing profits [52]. That is, farmers usually decide whether to adopt a technology or to what extent based on economic benefits. The heterogeneity of farmers’ willingness to adopt new technologies is mainly caused by their own endowments and external stimuli. Previous studies have shown that farmers’ insufficient knowledge of agricultural technologies is the main factor hindering their willingness to adopt them [48], and limited information dissemination channels may become the main obstacle to the early diffusion of new technologies [50]. When deciding whether to adopt new technology, farmers will make a binary decision relying on their private signals and past choice. Science popularization can change a farmer’s private signal, and play an important role in influencing farmers’ willingness to adopt CRFs as external stimulus, i.e., science popularization increases opportunities for farmers to understand and learn about CRF. As a result, effective information is brought to the farmers, thus improving their willingness to accept the new technology of CRF. Therefore, we posited the following:

Hypothesis 2 (H2).

In the early stage of technology dissemination, both farmers’ original cognition of new technology and science popularization will have a positive influence on farmers’ intention to adopt CRF.

Traditional small-scale farmers are individuals whose primary goal is to satisfy their own survival need [56]. Due to the small operation area, even if the yield per unit area increases, the total revenue increase is not obvious. Therefore, in most cases, small-scale farmers will choose to maintain the status quo, rather than to spend time and energy acquiring and updating new technology, so the use of new technologies should be closely related to social networks. However, large-scale operators are different from small farmers. With a larger land, even the per unit area yield keeping the same increase, their total income and net profit will be higher. Thus, the large-scale operators will have a stronger motivation to gather new technology information, higher willingness to pay more time and energy to obtain new technology, thus more likely to be an independent decision maker. In addition, due to the large demand for agricultural materials, such as fertilizer, the large-scale households can directly purchase from the upstream agricultural stores or manufacturers, and their purchase behavior is not restricted by the product category of the agricultural stores which is close to where they live, allowing them to obtain a more favorable price. Therefore, when the information is fully obtained, the large-scale households will make an independent decision on whether to use CRFs after comprehensive study and judgment with the knowledge they have acquired. Thus, we hypothesized:

Hypothesis 3 (H3).

When information is obtained sufficiently, the social network of large-scale households has no significant influence on their adoption intention.

Based on the above analysis, the theoretical framework that this paper hopes to verify is proposed as shown in Figure 1.

5. Materials and Methods

5.1. Data Source

The data in this paper came from a survey in Taizhou and Yancheng of Jiangsu Province, China in May 2021, totally containing 39 villages. Jiangsu is one of the main rice producing areas in China, and conventional rice cultivation is mainly concentrated in northern Jiangsu and central Jiangsu, accounting for 94% of the total cultivated area of the province in 2020. Therefore, Taizhou, the representative city of central Jiangsu, and Yancheng, the representative city of northern Jiangsu, were selected as our preliminary survey sites. Then, stratified sampling was conducted, and a total of 39 villages in the two places were finally selected as survey sites. The distribution is shown in Figure 2.

The quantity of the initial sample was 328 households. As the research object of this paper is the large-scale agricultural households, we firstly excluded the small-scale farmers under 3.33 ha (equal to 50 mu, according to the regulations of the Chinese government). To ensure the rigor and effectiveness of data analysis, the data was cleaned. After removing the samples containing missing values and singular values, 231 households with operation scale over 50 mu were retained. The relevant data were used in statistical analysis and empirical study, and the sample efficiency was 70.43%.

5.2. Variable Selection and Descriptive Statistics

In this paper, the study on adoption behavior was divided into three stages: initial cognition of farmers, short-term use behavior, and long-term adoption intention. There are five categories of independent variables. Specifically, social network, characteristics of family management decision maker, and family characteristics are common among the three studies, and the characteristics of family management decision maker and family characteristics exist as control variables. In the study of adoption intention, apart from the above three categories, initial cognition of farmers and science popularization are added as independent variables.

Table 1. Definition of variables.

Name	Assign a Value Standard
Intention	Farmers are willing to adopt CRFs when planting rice in the future (1= yes; 0 = no)
Use behavior	Farmers have adopted CRFs in rice production (1 = yes; 0 = no)
Initial cognition	Farmers have heard of CRFs (1 = yes; 0 = no)
Promotion	Farmers watch promotional videos (1 = yes; 0 = no)
Social Networks 1
Trust level	Factor analysis of extracted common factors 1
Communication	Factor analysis of extracted common factors 2
Network size	Factor analysis of extracted common factors 3
Characteristics of family decision makers
Gender	Female = 0; Male = 1
Age	Respondents’ age at the time of the survey (years)
Education	Literacy level (1 = illiterate; 2 = primary; 3 = middle; 4 = high; 5 = university and above)
Health	Current health status (1 = better; 2 = equal; 3 = worse)
Risk	0 = risk aversion; 1 = risk appetite
Visit	Whether or not you have visited a rice growing demonstration area (1 = yes; 0 = no)
Training	The actual number of agricultural technology training participation events last year (times)
Family Features
Land size	Actual rice planting area last year (mu 2)
Family income	Last year’s actual household income (million yuan)
Revenue Structure	Non-farm income as a share of total income (%)

¹ Standardization was used in the paper for the social network factor analysis, so the mean value of each dimension of the social network is 0. ² one Mu is equal to 0.165 ac.

is the definition of variables and

Table 2. Descriptive statistics of variables ¹.

Variable Names	Mean	Std. Err	Minimum	Maximum
Intention	0.51	0.50	0	1
Use behavior	0.32	0.47	0	1
Initial cognition	0.58	0.49	0	1
Promotion	0.51	0.50	0	1
Trust level 2	0	1	−2.89	1.90
Communication 2	0	1	−3.22	5.52
Network size 2	0	1	−1.33	7.91
Gender	0.94	0.23	0	1
Age	52.93	8.55	27	76
Education	3.22	0.86	1	5
Health	1.17	0.39	1	3
Risk	0.24	0.43	0	1
Visit	0.65	0.48	0	1
Training	2.16	1.86	0	12
Land size	364.88	484.17	50	5000
Family income	61.80	105.42	1.86	1046.53
Revenue structure	0.21	0.24	0	0.98

¹ The data source is compiled from fieldwork. ² The social network factor analysis was carried out using a standardization process, so that the mean value of each dimension of the social network is zero.

is the descriptive statistics of the variables involved in this paper. Nearly 60% of the households in the surveyed area have heard of CRFs, but only 30% of the farmers have used them. Nearly half of the interviewed farmers were randomly selected, and after further science popularization, the adoption intention of farmers increased to 0.51.

5.2.1. Dependent Variables

In this paper, we addressed the influence of social network on the use of CRFs by large-scale households. We asked farmers about their initial knowledge, short-term use behavior, and long-term adoption intention. Specifically, the initial knowledge was measured by the question “Have you heard of special CRF for rice?”; the use behavior was measured by the question “Have you ever used CRF?”; and the adoption intention was measured by the question “After the experiment, are you willing to use CRF in 2021?”

5.2.2. Core Explanatory Variables

Social network. For the description of social network variable, the following nine indicators were classified into three dimensions of trust level, communication, and network size. Referring to existing studies [57,58,59,60], we used factor analysis for comprehensive measurement. The nine selected indicators were measured using a five-level Likert scale, except for the “number of mobile phone contacts” and “average monthly mobile phone bills”, which were actual values. The details are shown in

Table 3. Social network rotation component matrix ¹.

Dimension	Metrics	Mean	Std. Err	Results of Factor Analysis
				Factor 1	Factor 2	Factor 3
Communication	Regular communication with village officials	3.69	1.09	0.0255	0.3641	0.0650
	Regular communication with large grain growers	4.15	0.91	−0.1042	0.4539	0.0848
	Willing to participate in village collective activities	3.84	0.86	−0.0081	0.4158	−0.1719
	Annual expenses for favors	1.46	1.73	−0.1589	0.3995	−0.0693
Trust level	Relatives	4.29	0.63	0.4048	−0.1228	−0.0361
	Neighbors	4.16	0.68	0.4386	−0.1644	−0.0371
	Village officials	4.05	0.82	0.3015	0.0104	0.1011
Network size	Number of mobile phone contacts	219.52	156.01	0.1110	−0.1563	0.6188
	Average monthly mobile phone bills	143.98	162.82	−0.1375	0.0966	0.5890

¹ Extraction method: principal component analysis.

.

In this paper, the above variables were firstly standardized and then factor analyzed by principal component analysis. In order to improve the correlation and accuracy of the extracted factors with the original factor time, the maximum variance method was used to rotate the factors, and after rotation, the correlation improved and the aggregation of the factors became higher, and finally three common factors (communication, trust level, network size) were extracted with a cumulative variance contribution of 58.26%.

Dissemination of CRF. Science popularization will affect the adoption intention of farmers. However, the literature has not considered whether popularization of the technology is “effective” for farmers, i.e., there may be someone to publicize the technology, but due to some reasons, such as inappropriate publicity and little information, farmers may not have a comprehensive understanding of the technology. Thus, there may be a situation that they have heard of but have little knowledge of it, thus weakening the positive impact on their intention to adopt the technology. The use of publicity videos can reduce the misunderstanding caused by different expressions in the process. To this end, farmers were first asked whether they had heard of CRFs, and then they were randomly divided into two groups of accepting and not accepting popularization before the study; the popularization behavior was measured by the question “Did the farmer watch the publicity video on CRF for rice?”.

5.2.3. Control Variables

To avoid the influence of individual and household differences on the research results, we took the characteristics of family decision makers as control variables in the empirical model. Individual characteristics included gender, age, education level, health status, risk attitude, whether they have visited demonstration areas, and the number of trainings; household characteristics included land area, household income, and income structure.

Regarding the risk attitude measurement of the business decision maker, this paper used a coin toss game to make a judgment, in which farmers have two choices, to toss a coin or not to toss a coin. If a farmer chooses to toss a coin, he or she faces the situation of “getting $40 for heads toss and $60 for tails toss”, which is considered as risk-tolerant; if a farmer chooses not to toss a coin and receives $50 as a definite reward, he or she is considered as risk-averse. To avoid inconsistency between the farmers’ true risk preference and their stated preferences, a cash incentive was used for further confirmation in this study, i.e., after they made their choices, a coin was tossed on the spot and a work allowance was given according to the results.

5.3. Model Specification

5.3.1. Probit Model

Initial cognition, use behavior, and adoption intention are all typical discrete binary choice variables. Therefore, a binary probit model was selected for empirical analysis in this study, and the model was initially constructed as follows.

Prob_r (Y_j = 1│X) = y = βx = β₀ + β₁x₁ + β₂x₂ + ⋯ + β_kx_k

(1)

where the dependent variable y is the binary choice variable; β₀ is the model intercept term; β_i (i = 1, 2, …, k) is the coefficient of each influencing factor; and x_i (i = 1, 2, …, k) is each type of factor that affects technology adoption behavior.

When studying farmers’ cognition, the dependent variable y represents farmers’ cognition. When studying farmers’ technology use behavior, the dependent variable y represents farmers’ use of CRFs, and if the large-scale household has not heard of rice-specific CRFs, it is assumed that he does not use it. x is the factors affecting farmers’ cognition, including three aspects, such as farmers’ social network, characteristics of household decision makers, and farmers’ household characteristics.

In the study of farmers’ adoption intention, the dependent variable y represents whether farmers are willing to adopt CRFs in the future. x is the factor that influences farmers’ willingness to adopt; in addition to the three aspects of farmers’ social network, household decision maker characteristics, and farmers’ household characteristics, it includes farmers’ existing knowledge of CRFs and science popularization.

5.3.2. Heckman Two-Stage Model

Adoption to CRFs, a new agricultural technology, involves a two-stage process: first, heard of CRFs and, second, deciding whether to adopt it. This leads to a sample selectivity problem, since only those who have heard of CRFs would adopt, whereas we need to make an inference about adoption in general, which implies the use of Heckman’s sample selectivity model.

In the first stage, a selection equation was developed to examine the determinants of whether a large-scale farmer chooses to adopt CRFs, i.e., whether the large-scale farmer has heard of CRFs. Specifically, the event “heard of CRFs” is indicated by z = 1, and z can be expressed as its potential variable z*, which is expressed as follows.

$$z_i^{*}={\omega }_i^{\prime }\alpha +e_i$$

(2)

where i refers to the ith surveyed farmer; $\omega$ is the set of covariates that may influence whether they have heard of CRFs; the independent error term $e$ follows a normal distribution, which has a mean of 0 and a variance of ${\sigma }_e^2$; and $z_i$ and $z_i^{*}$ are related to each other, such that if $z_i^{*}\le 0$ then $z_i=0$, and if $z_i^{*}>0$ then $z_i=1$.

In the second stage, a primary equation was established. The inverse Mills ratio obtained in the first stage of estimation would be regressed as a control variable along with other variables to examine the factors influencing the adoption of CRFs. Specifically, by expressing farmer’s choice of adoption as y, then y can be expressed by its potential continuous variable y*, and the expression of y* is as follows.

$$y_i^{*}=X_i^{\prime }\beta +u_i$$

(3)

where X is the set of covariates that may affect the farmer’s adoption behavior, and it does not need to be mutually disjointed from $\omega$; and the independent error term $u$ obeys a normal distribution with mean 0 and variance ${\sigma }_u^2$. Estimating z = 1, the vector $X_i$ determines the conditional expectation of y.

$$E(y_i|z=1,X_i)=X_i^{\prime }\beta +\rho {\sigma }_e{\sigma }_u{\lambda }_i$$

(4)

where ${\lambda }_i$ is the inverse Mills ratio estimated for the full sample using the probit model in the first stage estimation, and $\rho$ is the correlation coefficient of the two variances.

6. Results and Discussion

6.1. Baseline Regression Results

A binary probit model was used to investigate the influence of social network dimensions on the adoption behavior of large-scale agricultural operators in using CRFs by introducing control variables, further explore the influence of science popularization behavior on their willingness to adopt, and compare the heterogeneity of the influence that a social network has on farmers’ adoption behavior when information access is more adequate or not.

When studying farmers’ cognition (model 1), the dependent variable is farmers’ cognition. When studying farmers’ technology use behavior (model 2), the dependent variable is farmers’ use behavior of CRFs, and if the large-scale household has not heard of CRFs, it is assumed that he does not use it in this part. In the study of farmers’ adoption intention (model 3), the dependent variable represents whether farmers are willing to adopt CRFs in the future. Model 4 and 5 are further decompositions of model 3, exploring the impact of the three social network dimensions on the adoption of CRFs by large-scale farmers in the presence or absence of popularization behavior and comparing the differences.

From model (1) and (2) in

Table 4. Estimated results of factors influencing the adoption behavior of CRFs by large-scale households.

Variable Name	Before Intervention		After Intervention
	(1) Cognition	(2) Use	(3) Intention	(4) Popularization	(5) Non-Popularization
Popularization			0.7284 ***
			(0.1819)
Initial cognition			0.6223 ***	0.6246 **	0.6523 **
			(0.1940)	(0.2837)	(0.2970)
Trust	−0.0947	−0.0699	−0.0490	−0.0971	−0.0192
	(0.0931)	(0.0895)	(0.0941)	(0.1461)	(0.1399)
Communication	−0.0594	0.1517 *	0.0148	−0.0098	0.1165
	(0.0921)	(0.0899)	(0.0929)	(0.1428)	(0.1429)
Network size	0.2476 **	−0.0062	−0.1624	−0.0909	−0.3716 *
	(0.1222)	(0.0903)	(0.0999)	(0.1301)	(0.1912)
Gender	−0.7522 *	0.0104	−0.7992 *	0.0000	−0.3182
	(0.4283)	(0.3889)	(0.4519)	(.)	(0.5801)
Age	−0.0003	0.0043	0.0043	−0.0045	0.0096
	(0.0118)	(0.0119)	(0.0116)	(0.0175)	(0.0167)
Education	0.2774 **	0.1506	0.0475	−0.3492 *	0.3953 **
	(0.1259)	(0.1212)	(0.1231)	(0.1928)	(0.2002)
Health	0.4390 *	0.3179	0.0536	−0.0067	0.1162
	(0.2406)	(0.2267)	(0.2389)	(0.4079)	(0.3419)
Risk	0.3235	0.1348	−0.4001 *	−0.2887	−0.2691
	(0.2184)	(0.2101)	(0.2163)	(0.3194)	(0.3496)
Visit	0.4034 **	0.2528	0.1166	−0.2403	0.2966
	(0.2010)	(0.2053)	(0.2022)	(0.2874)	(0.3456)
Training	0.1074 **	0.0319	0.0301	−0.0149	0.1056
	(0.0508)	(0.0499)	(0.0513)	(0.0629)	(0.0974)
Land size	0.0003	0.0002	−0.0001	0.0004	−0.0006
	(0.0002)	(0.0002)	(0.0002)	(0.0004)	(0.0005)
Family income	−0.0004	0.0002	0.0006	−0.0010	0.0006
	(0.0009)	(0.0008)	(0.0008)	(0.0026)	(0.0010)
Revenue structure	−0.5164	0.1240	0.1322	0.4895	0.0054
	(0.3960)	(0.3924)	(0.3915)	(0.5774)	(0.6264)
Constant	−0.9633	−1.9735 **	−0.4721	1.5056	−2.5048 *
	(0.9414)	(0.9195)	(0.9132)	(1.3266)	(1.3711)
Observations	231	231	231	111	114

Note: ***, **, and * indicate significant at 1%, 5%, and 10% significance levels, respectively; numbers in parentheses are standard errors of variables.

, the coefficient of the network size on farmers’ initial cognition was 0.2476, which was significant at 5% statistical level, confirming that the network size had a positive influence on farmers’ cognition of CRFs at the early stage of technology dissemination. The coefficient of the communication intensity on farmers’ use behavior was 0.1517, which was significant at 10% statistical level, confirming that at the early stage of technology dissemination, the communication intensity with neighboring farmers had a positive influence on farmers’ use behavior of CRF. The results implied acceptance of Hypothesis 1.

From model (3), the coefficient of science popularization was 0.7284 and passed the significance test at 1% level, which indicated that the popularization behavior had a significant positive influence on farmers’ intention to adopt CRF. The coefficient of farmers’ initial cognition was 0.6223 and passed the significance test at 1% level, which indicated that farmers’ initial cognition had a significant contribution to farmers’ intention to adopt. These caused us to accept Hypothesis 2.

To further compare the heterogeneity of the social networks’ influence on their expected intention to adopt when information access is adequate or not, this study divided farmers into two groups by whether they watched the popularization video or not. As shown in model (4) and (5), there was a significant increase in the willingness to adopt for large-scale households who watched the video compared to those who did not. The three social network variables of trust, communication intensity, and network size were not significant among farmers who received science popularization, which implied acceptance of Hypothesis 3.

In addition, among farmers who did not receive science popularization, the coefficient of the influence of social network size on farmers’ intention to adopt was −0.3716 and passed the 10% significance test, indicating that the larger the social network size of farmers, the lower their willingness to adopt CRF. Probably it was because the larger the network size of farmers was, the more complex information they would receive, which interfered with their judgment of risk. Furthermore, during the initial stage of technology diffusion, the information that farmers obtained from their social activities may not be completely accurate, and most of the information they received may even be negative, thus reducing their intention to adopt.

6.2. Robustness Tests

Although in analyzing the adoption behavior of CRFs by large-scale households, this paper considered some factors that may affect the adoption behavior, there is a still possibility of estimation bias in the measurement results. To ensure the robustness of the results, different strategies were tried to conduct robustness tests for different questions, and the test results are shown in

Table 5. Robustness test results.

Variable Name	Heckman Two-Stage Model		Possible Missing Variables
	(6) Cognition	(7) Use	(8) Intention	(9) Popularization	(10) Non-Popularization
Popularization			0.7444 ***
			(0.1826)
Initial cognition			0.6062 ***	0.5552 *	0.6489 **
			(0.1946)	(0.2878)	(0.2990)
Trust	−0.0947	−0.0162	−0.0523	−0.1078	−0.0187
	(0.0931)	(0.0441)	(0.0945)	(0.1482)	(0.1400)
Communication	−0.0594	0.1119 **	0.0001	−0.0668	0.1165
	(0.0921)	(0.0475)	(0.0955)	(0.1537)	(0.1430)
Network size	0.2476 **		−0.1667 *	−0.1081	−0.3738 *
	(0.1222)		(0.1010)	(0.1323)	(0.1933)
Control variables	Yes	Yes	Yes	Yes	Yes
observations	231	135	231	111	114

Note: (a) The network size variable had a direct influence on the cognition of CRFs by large-scale households, and affected their CRF adoption behavior only through farmers’ cognition, which qualified as an exclusionary constraint variable of the model. (b) ***, **, * denoted significant at 1%, 5%, and 10% significance levels, respectively. (c) The numbers in parentheses were the standard errors of the variables.

.

For the initial cognition and use behavior of farmers before the intervention, this section used a replacement model to test the robustness of the above results. There were two situations where farmers did not choose to use CRFs: one was that farmers had not heard of CRFs and then had no use behavior; another case was that farmers had heard of CRFs but did not choose to use it. Farmers’ choice of whether to use CRFs was not random. It was decided on a combination of potential profitability, their own conditions, and family situation. Thus, the statistical results found only from the sample of farmers who had heard of CRFs would be biased because the sample selection was not exogenous at random. Therefore, this section used the Heckman two-stage model to conduct regression analysis again. The first stage was farmers’ cognition, and the second stage was farmers’ use behavior. The regression results were more consistent with the results of the above probit model analysis, and the IMR was not significant, indicating that there was no self-selection problem.

For the adoption intention of farmers after the intervention, from the perspective of possible omitted variables, this section chose the method of adding control variables to test the robustness of the empirical results. In China, agricultural extension workers are an important source of farmers’ exposure to new agricultural technologies, so this important information might be missing upwards. Compared with ordinary farmers, village cadres had more access to media, such as agricultural promotion agencies, agricultural dealers, and village radio, so it might be necessary to include them as control variables, which have been included in past literature [61].

7. Conclusions

This paper investigated the effect of social networks on the adoption behavior of CRFs by large-scale households from the perspective of information asymmetry, specifically in three stages: initial cognition of CRFs, use behavior, and willingness to adopt after the dissemination of science and technology. The dichotomous choice model found and verified that the social network of the large-scale households influenced their adoption behavior in different dimensions at the early stage of CRF technology diffusion, but the decisions made by the large-scale households were not influenced by their social network when they had access to effective information.

The main findings of this paper were threefold. First, at the early stage of technology diffusion, the social network of large-scale households had a significant influence on farmers’ cognition and use behavior. Specifically, the network size had a positive contribution to the farmers’ cognitive level of CRFs, and the communication intensity with neighboring farmers had a positive effect on the use of CRFs. Second, at the early stage of technology diffusion, both the initial cognition of large-scale households about the new technology and the popularization behavior had a positive influence on their willingness to adopt CRFs. Third, when information was sufficient, there was no significant effect of the social network of the large-scale households on their willingness to adopt.

Based on this, the following policy recommendations are made. In the initial dissemination of the new technology, we should give full play to the advantages of today’s information technology and make use of social network to disseminate relevant information so that more farmers have at least a superficial understanding of the technology. After that, farmers’ common learning among acquaintances should be encouraged. This will strengthen farmers’ common understanding of the new agricultural technology and accelerate the spread of the technology in this area. In addition, when promoting new agricultural technologies such as CRF, it is especially important for the government to break down information barriers so that farmers can obtain accurate and effective technology-related information. Furthermore, the government should carry out relevant technology popularization according to local conditions in an easy-to-understand way, such as categorizing and publishing technology-related knowledge on websites, official accounts, and other channels that are easy for farmers to find and retrieve. This will facilitate farmers to obtain relevant information at a lower cost, and then to order relevant agricultural products on the Internet, thus promoting independent decision-making by this group of large-scale households and further influencing other farmers around them through their use behavior.