Do Green Production Technologies Improve Household Income? Evidence from Rice Farmers in China

Abstract

Revealing the behavioral decision-making mechanism and behavioral effects of farmers adopting green production technologies will help motivate farmers to actively adopt green production technologies, thereby promoting the development of green agriculture. A stratified random sampling technique was used to select 607 Chinese rice farmers, while the endogenous switching regression model was used to analyze the influencing factors and behavioral effects of farmers adopting green production technologies. The results show that the adoption of green production technologies by farmers can significantly increase household income. Among the green production technologies, soil testing and formula fertilization technology has the greatest impact on farmers’ income, followed by straw returning technology and planting green manure. The main influencing factors of farmers adopting green production technologies are education level, cultivated land area, cultivated land fragmentation, soil fertility, and the distance between home and agricultural technology stations. The results of this study provide an empirical basis for relevant government departments to carry out agricultural technology extension work and formulate relevant policies.

1. Introduction

In order to better address global hunger and malnutrition, the green revolution, mainly featuring agricultural technological innovation, is already in full swing around the world [1,2,3]. The agricultural technology developed during the green revolution is to improve grain production, with a specific focus on the use of fertilizer to improve soil fertility, using pesticides to protect crops from pests, cultivate high-yield and drought resistant crop varieties and use agricultural machinery to improve agricultural management efficiency and reduce production costs [4,5,6]. The development of these agricultural techniques has proved to play a crucial role in increasing food production and agricultural productivity, promoting economic growth and improving the quality of life in rural communities, and still profoundly affects agricultural production today [7,8,9].

Since the green revolution, traditional agricultural technology has been gradually been replaced by modern agricultural technology, and the overuse of pesticides and fertilizers has become a common and long-standing phenomenon in many countries, especially in developing countries [10,11]. However, inputs such as pesticides and fertilizers being used intensively and soil being cultivated with high intensity have caused a series of ecological and environmental problems such as soil pollution, cultivated land degradation, non-point source pollution and greenhouse gas emissions [12,13,14,15]. To deal with this unsustainable trend, many countries have formulated a series of policies and programs to promote green production technologies (GPTs), ranging from economic incentives to long-term technological packages [12,16].

However, the implementation of GPTs is inseparable from farmers. As the actual users of cultivated land, farmers are also the most direct participants and stakeholders in the protection and quality improvement of cultivated land, playing a key role in the sustainable agricultural practice of cultivated land [12,17,18]. Although a large amount of human, material and financial resources have been invested to promote GPTs, the actual promotion results did not meet the expectations. Some reports pointed out that the farmers’ enthusiasm for adopting GPTs was not high, and the actual proportion of adoption was low [19,20], even staying at the cognitive level of “hearing or not” [21,22].

So, what is the reason for the low adoption rate of GPTs by farmers? After reviewing the existing research, we found that we can carry out research from two aspects. On the one hand, the transition from traditional farming to green agriculture is a radical transformation for farmers [23,24], which is also a complex decision-making process involving not only economic factors, but also social and psychological factors [12,23,25]. However, the existing studies cannot give consistent conclusions on the influencing factors, and there are contrary results. It is necessary to study the factors that affect farmers’ adoption of GPTs. On the other hand, farmers may not fully recognize the benefits of GPTs or doubt them [26]. To solve the “last kilometer” problem of laboratory research and farmer practice, more empirical results of field investigation are needed to show the adoption behavior benefits of GPTs, especially the economic benefits that farmers are concerned about. In view of this, we tried to answer the following questions: (1) What are the barriers to farmers’ adoption of GPTs? (2) What are the behavioral and economic effects of farmers in adopting GPTs? (3) What aspects should the subsequent promotion policy of GPTs focus on?

There are two potential contributions of this article. Firstly, an endogenous switching regression model that can address the issues of missing variables, and sample self-selection and heterogeneity was employed to evaluate the income processing effect of farmers adopting GPTs. Specifically, the use of ordinary least squares (OLS) method cannot solve the problem of sample self-selection, the propensity score matching (PSM) method cannot solve the problem of endogenous missing variables caused by unobservable factors, and the instrumental variable method does not consider the heterogeneity of treatment effects [27,28]. Secondly, existing studies have explored processing effects more from a single GPT or a single dimension (whether adopted or not). However, in the agricultural production process, farmers may adopt a package of GPTs, and only considering whether farmers adopt a single GPTs often leads to biased estimates. Therefore, this article selects three types of GPTs commonly adopted by farmers from the three stages of rice production (before, during and post-harvest), and evaluates the treatment effect from two dimensions (whether adopted and degree of adoption), which can more scientifically and reasonably reveal the behavioral effects of farmers.

In order to answer the research questions mentioned above, Section 2 introduces the research design and research methods, such as describing the study area, overviews the research design and data collection method and selects the econometric model method. Section 3 describes the differences in rice harvest among farmers, the regression model results and treatment effects of farmers adopting GPTs. Section 4 discusses the empirical results, while Section 5 concludes the study with policy implications.

2. Materials and Methods

2.1. Study Area

The Poyang Lake Plain lies between 27°32′–30°06′ N and 115°01′–117°34′ E and is located in the north of Jiangxi Province and the southwest border of Anhui Province, China. It is an alluvial plain formed by the five major river systems and the Yangtze River into the Poyang Lake, and is an important part of the middle and lower reaches of the Yangtze River. The area has gentle terrain, dense river network, diverse geomorphic forms, alternating hills, low mountains, plains and lakes. The Poyang Lake Plain belongs to the middle latitude region of east Asia, with obvious subtropical humid monsoon climate characteristics, showing the characteristics of more precipitation, abundant light and heat, a mild and humid climate and long frost-free period. The annual average temperature of the Poyang Lake Plain is 17.3 °C, the annual average precipitation can reach 1612 mm, the annual frost-free period is 260 days and the sunshine duration is 1803 h [29].

The total area of the Poyang Lake Plain is about 38,760.6 km², and the land area accounts for 23.2% of Jiangxi Province, while the arable land is 876.8 thousand hectares, accounting for 37.3% of Jiangxi Province. Abundant rainfall and superior natural environment have created high-quality cultivated land resources in the Poyang Lake Plain. Convenient irrigation conditions and farming environment have provided unique conditions for the development of agriculture in the Poyang Lake Plain and benefited the agricultural population in this area. The Poyang Lake Plain has always been the main grain producing area in Jiangxi Province and one of the important rice producing areas in China [30]. The area of rice planting in this area exceeds half of the total sown area of crops in Jiangxi Province, and 86% of the sown area of grain crops in Jiangxi Province is concentrated in the Poyang Lake Plain.

2.2. Research Design and Data Collection

The research design consisted of four stages. First was design of the initial questionnaire. After extensive reading of relevant literature, preliminary research topics were identified. Four experts in agricultural economics and land resource management were invited to discuss the feasibility of the research topic, the research topic was finalized and an initial questionnaire was designed. Second, we determined the questionnaire. The initial questionnaire was pretested on 50 farmers in Poyang County in October 2020. Three local government officials and eight local farmers were invited to participate in the focus group discussions (FGDs). An informal, semi-structured discussion format lasting about 2 h each time was employed for the FGDs. The questionnaire was modified based on the results of the FGDs and pre-surveys. Third, we conducted a formal survey. A field survey was conducted in the main grain producing areas of the Poyang Lake Plain in December 2020. According to the evaluation results of China’s advanced grain production counties and the geographical distribution characteristics (upper, middle and lower reaches of the Poyang Lake basin), six sample counties (Yongxiu County, Xinjian District, Jinxian County, Yugan County, Poyang County and Duchang County) were selected as sample counties. Four, we reviewed and analyzed data. After logical relationship checks, the questionnaires with incomplete information and self-contradiction were eliminated, and the valid questionnaire data were used for study and analysis.

The final questionnaire consisted of three main parts. The first part confirmed the agricultural production situation of farmers. Some farmers were asked questions about land resources, fertilizer application, information access channels and technical support, which could help farmers better understand the content and process of the interviews. In the second part, we focused on farmers’ adoption and preference of GPTs. For example, respondents were asked about their cognition, willingness and adoption behavior relating to green agriculture and GPTs. The third part confirmed the personal and family information of farmers. The gender, age, education, farming years, risk preference and other personal information of farmers were asked, and family income, housing, agricultural machinery and equipment, social network and other issues were discussed.

A field survey was conducted to obtain the research data for this article. The simple random sampling and stratified random sampling techniques were used to select sample points and sample farmers in the main grain producing areas of the Poyang Lake Plain in December 2020. According to the township size, population and geographical location of the sample counties (districts), each county (district) selected 2 sample townships, each township randomly selected 3 sample villages and we randomly invited 15–20 farmers in each village to conduct the questionnaire survey. The head of household or the agricultural decision-maker of the family was invited to accept the interview. In order to better improve the data quality, training for the research team was carried out in the early stage. The members of the research group used face-to-face interviews to obtain relevant information from the interviewed farmers. Finally, 630 questionnaires were distributed, 607 valid questionnaires were obtained and the effective recovery rate was 96.35%.

2.3. Data Analysis

As a rational economic person, a farmer makes the decision to adopt GPTs based on maximization of utility, which is reflected in the difference between the cost and benefit of farmers adopting the technology, that is, the net utility. Specifically, when the benefits of adopting the technology are greater than the cost, farmers will tend to adopt the technology; otherwise, farmers will tend not to adopt the technology. In this study, whether GPTs could increase the output value of rice or reduce the amount of chemical fertilizer application were the most important issues for farmers. Specifically, assuming that the behavioral decision of farmers to adopt GPTs is $A_i$, the potential net utility that farmers $i$ can obtain by adopting GPTs is $U_{\mathit{iy}}$, and the potential net utility of not adopting GPTs is $U_{\mathit{in}}$. Then, the precondition for farmers to adopt GPTs is that the net utility obtained by adopting the technology is greater than the net utility of not adopting the technology; that is, when $U_{\mathit{iy}}-U_{\mathit{in}}\hspace{0.17em}>\hspace{0.17em}0$. Since the net effect of whether farmers adopt GPTs is a latent variable that cannot be directly observed, it needs to be represented by a series of functional expressions of observable exogenous variables. Then, the decision model of whether farmers adopt GPTs can be expressed as:

$$A_i=\{\begin{array}{l}0,U_{\mathit{iy}}-U_{\mathit{in}}\hspace{0.17em}\le 0\\ 1,U_{\mathit{iy}}-U_{\mathit{in}}\hspace{0.17em}>\hspace{0.17em}0\end{array}\}$$

(1)

In the formula, $A_i=0$ indicates that farmers choose not to adopt GPTs and $A_i=1$ indicates that farmers choose to adopt GPTs.

As mentioned earlier, this study used an endogenous switching regression model to explore the income increase effect test of farmers adopting GPTs. The endogenous switching regression model estimates three equations: behavioral choice equation (whether farmers adopt GPTs), result equation I (control group; that is, the behavioral effect level equation of farmers not adopting GPTs) and result equation II (treatment group, that is, adopting farmers’ behavior effect level equation). Its equation expressions are Equations (2)–(4):

$$A_i=\phi Z_i+\delta W_i+{\mu }_i$$

(2)

$$Y_{in}={\beta }_n{X}_{in}+{\epsilon }_{in}$$

(3)

$$Y_{iy}={\beta }_y{X}_{iy}+{\epsilon }_{iy}$$

(4)

In the formula, $i$ means the $i$-th farmer, $n$ means that the farmers have not adopted GPTs, and $y$ means that they have adopted them; $A_i$ represents the binary selection variable of whether farmers adopt GPTs; $Z_i$ represents various factors affecting whether farmers adopt GPTs; $W_i$ represents the identification variable vector, which is used to ensure the identifiability of the endogenous switching regression model; $\varphi$ and $\delta$ represent estimated coefficients of $Z_i$ and $W_i$, respectively; ${\mu }_i$ is the random error term. $Y_{in}$ and $Y_{iy}$ indicate the level of behavioral effects of the two sample groups of non-adopting and adopting farmers, respectively; $X_{in}$ and $X_{iy}$ indicate a series of factors affecting the behavioral effects of farmers, ${\beta }_n$ and ${\beta }_y$ represent the estimated coefficients of the $X_{in}$ and $X_{iy}$, respectively; ${\epsilon }_{in}$ and ${\epsilon }_{iy}$ are the stochastic error terms of the respective equations.

Through the estimation coefficient of endogenous switching regression model and the counterfactual analysis framework, the average processing effect of GPTs’ adoption decisions on household income was estimated between the real situation and the counterfactual hypothesis scenario. The inverse Mills ratios ${\lambda }_{iy}$, ${\lambda }_{in}$ and their covariance ${\sigma }_{\mu y}=cov\hspace{0.17em}({\mu }_i,{\epsilon }_{iy})$ and ${\sigma }_{\mu n}=cov\hspace{0.17em}({\mu }_i,{\epsilon }_{in})$ are introduced, and the complete information maximum likelihood method is applied to jointly estimate Equations (2)–(4), resulting in Equations (5)–(8) as follows.

$$E\left[Y_{in}|A_i=0\right]={\beta }_n\hspace{0.17em}{X}_{in}+{\sigma }_{\mu n}\hspace{0.17em}{\lambda }_{in}$$

(5)

$$E\left[Y_{\mathit{iy}}|A_i=1\right]={\beta }_y\hspace{0.17em}{X}_{\mathit{iy}}+{\sigma }_{\mu y}\hspace{0.17em}{\lambda }_{\mathit{iy}}$$

(6)

$$E\left[Y_{\mathit{iy}}|A_i=0\right]={\beta }_y\hspace{0.17em}{X}_{\mathit{in}}+{\sigma }_{\mu y}\hspace{0.17em}{\lambda }_{\mathit{in}}$$

(7)

$$E\left[Y_{\mathit{in}}|A_i=1\right]={\beta }_n\hspace{0.17em}{X}_{\mathit{iy}}+{\sigma }_{\begin{array}{l}\mu n\\ \end{array}}\hspace{0.17em}{\lambda }_{\mathit{iy}}$$

(8)

Therefore, the average treatment effect of farmers who did not adopt GPTs, namely the average treatment effect of the control group (ATU), can be expressed as the difference between Equations (5) and (7):

$$\mathrm{A}\mathrm{T}{\mathrm{U}}_i=E\left[Y_{in}|A_i=0\right]-E\left[Y_{iy}|A_i=0\right]=\left({\beta }_n-{\beta }_y\right)X_{in}+\left({\sigma }_{\mu n}-{\sigma }_{\mu y}\right){\lambda }_{in}$$

(9)

Accordingly, the average treatment effect of farmers who have adopted GPTs, namely the average treatment effect of the treatment group (ATT), can be expressed as the difference between Equations (6) and (8):

$$\mathrm{A}\mathrm{T}{\mathrm{T}}_i=E\left[Y_{iy}|A_i=1\right]-E\left[Y_{in}|A_i=1\right]=\left({\beta }_y-{\beta }_n\right)X_{iy}+\left({\sigma }_{\mu y}-{\sigma }_{\mu n}\right){\lambda }_{iy}$$

(10)

2.4. Variable Setting and Descriptive Statistics

According to the setting of the endogenous transformation regression model, this study set the explained variable, key explanatory variables, other explanatory variables, and identifying variables. The specific variables were set and described as follows.

Explained variable. Whether to obtain higher income is the most important motivation for farmers to make a decision to adopt GPTs. Compared with the growth of rice yield, farmers attach greater importance to rice output value, because low prices for grain hurt farmers and will not only hit the confidence of farmers continuing to engage in agriculture, but also cover the economic effect of GPTs. Therefore, the rice output value was selected as the dependent variable.

Key explanatory variables. Given that GPTs is a general term containing a variety of specific agricultural technologies, for this paper we selected planting green manure (PGM), soil testing and formula fertilization technology (STFFT) and straw returning technology (SRT) as three agricultural technologies commonly used by farmers from the three links of rice production (before, during and post-harvest) to characterize GPTs. Therefore, the behavioral decision-making of farmers using three types of GPTs was considered as a key explanatory variable.

Control variables. Referring to existing research [18,25,31,32], control variables were selected from three aspects: individual characteristics of farmers, family characteristics, and cultivated land resource characteristics. In terms of individual characteristics, three indicators were selected to measure: age, gender and education level. In terms of family characteristics, three indicators were selected to measure total household income, household population, and government officials. In terms of cultivated land resource characteristics, four indicators were selected to measure cultivated area, cultivated land fragmentation, soil fertility, and distance from cultivated land.

Identification variables. The distance between farmers’ home and agricultural technology station was selected as the identification variable. The reason for choosing this variable was that closer proximity to agricultural technology station can help farmers better access agricultural information and services such as agricultural technology and subsidies, which will directly affect farmers’ behavioral decisions to adopt GPTs. In addition, the distance between farmers and agricultural technology station is objective and will not have a direct impact on the output value of rice planting by farmers. The definitions and descriptions of the above variables were statistically analyzed as shown in

Table 1. Definition and descriptive statistics of selected variables.

Variable	Definition	Mean	S.D.
PGM	1 If farmer adopts PGM, 0 otherwise	0.37	0.48
STFFT	1 If farmer adopts STFFT, 0 otherwise	0.32	0.47
SRT	1 If farmer adopts SRT, 0 otherwise	0.72	0.45
Rice output value	The output value of rice per mu, yuan/mu	1260.71	216.11
Age	Age of interviewed farmers (years)	60.00	9.90
Gender	1 = man, 0 = woman	0.68	0.47
Education	1 = illiteracy, 2 = primary; 3 = junior high school; 4 = senior high school; 5 = university and above	2.20	0.81
Population	Number of family members of the interviewed farmers	6.03	2.65
Income	The total household income of the interviewed farmers (ten thousand yuan)	14.23	20.11
Official	1 If a family member is a government official, 0 otherwise	0.27	0.44
Cultivated area	Area of cultivated land planted by farmers (mu)	46.23	114.03
Fragmentation	Fragmentation of cultivated land, planting area/number of plots	1.11	0.92
Soil fertility	Soil fertility, 1 = very poor, 2 = relatively poor, 3 = average, 4 = relatively good, 5 = very good	3.60	1.01
Cultivation distance	The average distance from farmer’s house to the cultivated land (km)	2.44	2.30
Distance to service station	Distance from farmer’s house to the nearest agricultural technology extension station (km)	6.65	11.13

Note: yuan is Chinese currency unit ($1 = 6.87 yuan), 1 mu = 1/15 hectare.

.

3. Results

3.1. Socioeconomic Characteristics of the Sample

The main socioeconomic characteristics of the surveyed farmers are reported in

Table 2. Socioeconomic characteristics of samples.

Variable	Variable Description and Assignment	Mean	S.D.
Age	Age of the surveyed farmers	60.00	9.90
Gender	Gender of the surveyed farmers. Female = 0, Male = 1	0.68	0.47
Education	Uneducated = 1; primary school = 2; junior middle school = 3; senior middle school = 4; junior college and above = 5	2.20	0.81
Years of farming	Years engaged in farming production	39.14	12.48
Population	Household size of the surveyed farmers	6.03	2.65
Farm area	The area of farmland contracted by households (mu)	8.21	5.62
Income	Total annual net household income (yuan)	142,304	20.11

. Overall, the average age of the interviewees was about 60 years and ranged from 30 to 84 years. The average educational level of the interviewees ranged from primary school to junior high school and was generally low. The survey results show that the interviewees generally engaged in agricultural production a long time, with an average of 39.14 years. These findings are consistent with the actual situation in Jiangxi Province, where farmers who are engaged in agriculture generally have the characteristics of old age, low educational level and many years of agricultural production. The results indicate that the average family size of the surveyed farmers was about six people, with 8.21 mu of contracted cultivated land. The average total household income of the surveyed farmers was about 142,304 yuan. Overall, the socioeconomic characteristics of the farmers surveyed were basically in line with the actual situation of farmers in Jiangxi Province [33].

3.2. Difference Test of Rice Output Value

The harvest of farmers planting rice is shown in Figure 1. The rice yield for most farmers was 500~700 kg/mu, and very few farmers had rice yields over 700 kg/mu. It cannot be ignored that there were many farmers whose rice yields were less than 500 kg/mu. Similarly, the rice output value of most farmers was 1100~1500 yuan/mu, and for a few farmers, it was less than 1100 yuan or higher than 1500 yuan.

The mean differences in the behavior of farmers adopting GPTs is presented in

Table 3. Rice harvests under different GPT adoption scenarios.

Category	PGM		STFFT		SRT
	Adopted	Unadopted	Adopted	Unadopted	Adopted	Unadopted
Rice yield (kg/mu)	593.79 *** (4.87)	528.71 *** (4.14)	626.69 *** (3.80)	518.50 *** (3.60)	578.21 *** (3.29)	487.22 *** (6.51)
Rice output value (yuan/mu)	1353.11 *** (12.48)	1206.67 *** (10.94)	1426.39 *** (10.46)	1184.06 *** (9.82)	1318.05 *** (8.75)	1113.31 *** (17.28)

Note: Standard errors in parentheses. *** p < 0.01.

. As is evident from the table, the rice yield and rice output value for adopted farmers were higher than those of unadopted farmers, and statistically significant in mean difference. These descriptive comparisons seem to show that GPTs play a significant role in increasing rice yield and rice output value of adopted farmers compared with non-adopting farmers. However, from the results in Table 3 we cannot make an inference that GPTs have an impact on rice yield and rice output value. It is also necessary to explore the effects of some observable variables (e.g., the characteristics of farmers at the individual, family and farm levels) and unobservable factors (e.g., farmers’ personal skills and motivation for technology selection).

3.3. Estimated Results of Farmers Adopting GPTs

Table 4, Table 5 and Table 6 present the estimates of factors influencing farmers’ decision to adopt GPTs and the impact of GPTs’ adoption on rice output value. As described previously, the selection and outcome equations were jointly estimated by the FIML approach. Specifically, the second column of Table 4, Table 5 and Table 6 present the selection equation for the determinants of GPTs’ adoption. The third and fourth columns of Table 4, Table 5 and Table 6 give the outcome equations of the effect of GPTs on the output value of non-adoption and adoption, respectively. The estimation results of the endogenous switching regression model show that the estimation coefficients of the identification variable passed the significance test, which indicates that the identification variables were effective [34]. At least one of the structural variables (ρ_μy and ρ_μn) in the same model has passed the significance test, indicating that the sample has a self selection problem. Whether farmers choose to adopt GPTs is not randomly generated, but rather a “self selection” [27,28]. In other words, both observable and unobservable factors influenced the decision and outcome of farmers choosing to adopt GPTs. Therefore, ignoring this problem may lead to biased estimation results.

3.3.1. Estimated Results of Farmers Adopting PGM

The estimates of the factors that influenced a farmer’s decision to adopt PGM, and the impact of PGM adoption on rice output value, are presented in Table 4. In selection equations, the age variable was tested for significance, indicating that older farmers were less likely to adopt PGM. The gender variable passed the significance test, indicating that male farmers were more likely to adopt PGM than female farmers. The education variable had a significant positive correlation, suggesting that farmers with higher education levels were more likely to adopt PGM. The variables of cultivated land area and cultivated land fragmentation had significant negative effects, indicating that farmers with larger cultivated land area and cultivated land fragmentation were less likely to adopt PGM. Moreover, the identification variable of the distance from home to agricultural technology station passed the significance test, indicating that farmers who were more likely to access agricultural technology services were more likely to adopt PGM.

According to the estimation results in Table 4, we can explain the impact of PGM adoption on rice output value. The official variable indicates a significant and positive impact on rice output value for both unadopted and adopting farmers, suggesting that farmers whose family members hold positions in government departments were more likely to obtain higher rice output value. Similarly, the soil fertility variable showed a significant positive impact on the rice output value for both unadopted and adopted farmers, indicating that the farmers with better soil fertility obtained higher rice output value. Conversely, the variable of home-to-land distance had a significant negative impact on the rice output value for both unadopted and adopted farmers, indicating that the closer the distance between farmers’ homes and cultivated land, the more beneficial the farmers’ agricultural production and the improvement of rice output value. Furthermore, the estimation results of age, family population, and cultivated land fragmentation variables indicate that older, smaller households, and lower cultivated land fragmentation among unadopted farmers were more likely to obtain higher output value.

Table 4. Determinants of farmers adopting PGM and its impact on rice output value.

Variables	Selection		Adopted		Unadopted
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
Age	−0.017 **	0.007	−0.001	0.001	0.002 **	0.001
Gender	0.399 **	0.157	−0.020	0.027	−0.011	0.019
Education	0.861 ***	0.101	−0.015	0.020	0.002	0.017
Population	−0.029	0.025	−0.004	0.004	−0.006 **	0.003
Income	0.004	0.009	0.001	0.001	0.001	0.001
Official	−0.076	0.144	0.063 ***	0.018	0.090 ***	0.021
Cultivated area	−0.005 ***	0.002	0.000	0.000	0.000	0.000
Fragmentation	−0.383 ***	0.108	−0.008	0.022	−0.040 ***	0.010
Soil fertility	0.172 **	0.077	0.036 ***	0.012	0.025 ***	0.009
Cultivation distance	0.002	0.028	−0.014 ***	0.004	−0.011 ***	0.003
Distance to service station	−0.087 ***	0.022
Constant	−1.096 *	0.605	7.216 ***	0.100	6.912 ***	0.079
Lnσμy			−2.090 ***	0.098
Lnσμn					−1.805 ***	0.049
ρμy			−0.527 **	0.259
ρμn					−0.595 ***	0.121
Log likelihood	77.518753
LR test of indep. eqns.	7.68 **

Note: * p < 0.1, ** p < 0.05, *** p < 0.01.

Table 4. Determinants of farmers adopting PGM and its impact on rice output value.

Variables	Selection		Adopted		Unadopted
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
Age	−0.017 **	0.007	−0.001	0.001	0.002 **	0.001
Gender	0.399 **	0.157	−0.020	0.027	−0.011	0.019
Education	0.861 ***	0.101	−0.015	0.020	0.002	0.017
Population	−0.029	0.025	−0.004	0.004	−0.006 **	0.003
Income	0.004	0.009	0.001	0.001	0.001	0.001
Official	−0.076	0.144	0.063 ***	0.018	0.090 ***	0.021
Cultivated area	−0.005 ***	0.002	0.000	0.000	0.000	0.000
Fragmentation	−0.383 ***	0.108	−0.008	0.022	−0.040 ***	0.010
Soil fertility	0.172 **	0.077	0.036 ***	0.012	0.025 ***	0.009
Cultivation distance	0.002	0.028	−0.014 ***	0.004	−0.011 ***	0.003
Distance to service station	−0.087 ***	0.022
Constant	−1.096 *	0.605	7.216 ***	0.100	6.912 ***	0.079
Lnσμy			−2.090 ***	0.098
Lnσμn					−1.805 ***	0.049
ρμy			−0.527 **	0.259
ρμn					−0.595 ***	0.121
Log likelihood	77.518753
LR test of indep. eqns.	7.68 **

Note: * p < 0.1, ** p < 0.05, *** p < 0.01.

3.3.2. Estimated Results of Farmers Adopting STFFT

The estimates of the factors that influence a farmer’s decision to adopt STFFT, and the impact of STFFT adoption on rice output value are presented in Table 5. In selection equations, the income variable had a significant positive correlation, suggesting that farmers with higher income were more likely to adopt STFFT. The variables of cultivated land area and cultivated land fragmentation had significant positive and negative effects, respectively, indicating that farmers with large cultivated land area and low degree of cultivated land fragmentation were more likely to use STFFT. Moreover, the identification variable of the distance from home to agricultural technology station passed the significance test, indicating that farmers with easier access to agricultural technology services were more likely to adopt STFFT.

According to the estimation results in Table 5, we can explain the impact of STFFT adoption on rice output value. For both unadopted and adopted farmers, the variables of officials, soil fertility and home-to-land distance were the main influencing factors of rice output value. Specifically, farmers whose family members held positions in government departments or whose soil fertility was better or whose home was closer to the cultivated land were more likely to obtain high rice output value. For adopted farmers, the education variable indicated a significant and positive impact on rice output value, suggesting that farmers with better education were more likely to obtain higher rice output value. For farmers, the variables of family population, cultivated land area and cultivated land fragmentation had significant negative impact on rice output value, indicating that farmers with smaller household size, smaller cultivated land area, or more concentrated cultivated land were more likely to obtain higher rice production value.

Table 5. Determinants of farmers adopting STFFT and its impact on rice output value.

Variables	Selection		Adopted		Unadopted
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
Age	−0.018	0.011	0.001	0.001	0.001	0.001
Gender	0.336	0.248	−0.001	0.015	0.005	0.019
Education	0.311 **	0.138	0.030 ***	0.009	0.012	0.012
Population	−0.014	0.045	−0.004	0.003	−0.008 **	0.004
Income	0.068 *	0.035	0.001	0.001	0.001	0.003
Official	−0.136	0.209	0.044 ***	0.013	0.090 ***	0.020
Cultivated area	0.071 ***	0.009	−0.000	0.000	−0.004 ***	0.001
Fragmentation	−1.771 ***	0.469	−0.053	0.038	−0.035 ***	0.010
Soil fertility	0.104	0.129	0.032 ***	0.009	0.030 ***	0.009
Cultivation distance	0.025	0.040	−0.010 ***	0.003	−0.015 ***	0.003
Distance to service station	−0.131 ***	0.042
Constant	−0.936	1.076	7.055 ***	0.066	6.967 ***	0.074
Lnσμy			−2.520 ***	0.051
Lnσμn					−1.829 ***	0.037
ρμy			−0.304 *	0.174
ρμn					−0.851 ***	0.067
Log likelihood	315.58592
LR test of indep. eqns.	14.96 ***

Note: * p < 0.1, ** p < 0.05, *** p < 0.01.

Table 5. Determinants of farmers adopting STFFT and its impact on rice output value.

Variables	Selection		Adopted		Unadopted
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
Age	−0.018	0.011	0.001	0.001	0.001	0.001
Gender	0.336	0.248	−0.001	0.015	0.005	0.019
Education	0.311 **	0.138	0.030 ***	0.009	0.012	0.012
Population	−0.014	0.045	−0.004	0.003	−0.008 **	0.004
Income	0.068 *	0.035	0.001	0.001	0.001	0.003
Official	−0.136	0.209	0.044 ***	0.013	0.090 ***	0.020
Cultivated area	0.071 ***	0.009	−0.000	0.000	−0.004 ***	0.001
Fragmentation	−1.771 ***	0.469	−0.053	0.038	−0.035 ***	0.010
Soil fertility	0.104	0.129	0.032 ***	0.009	0.030 ***	0.009
Cultivation distance	0.025	0.040	−0.010 ***	0.003	−0.015 ***	0.003
Distance to service station	−0.131 ***	0.042
Constant	−0.936	1.076	7.055 ***	0.066	6.967 ***	0.074
Lnσμy			−2.520 ***	0.051
Lnσμn					−1.829 ***	0.037
ρμy			−0.304 *	0.174
ρμn					−0.851 ***	0.067
Log likelihood	315.58592
LR test of indep. eqns.	14.96 ***

Note: * p < 0.1, ** p < 0.05, *** p < 0.01.

3.3.3. Estimated Results of Farmers Adopting SRT

The estimates of the factors that influence a farmer’s decision to adopt SRT, and the impact of SRT adoption on rice output value are presented in Table 6. In selection equations, the family population variable had a significant negative effect, indicating that farmers with a smaller family were more likely to adopt SRT. The income variable had a significant positive correlation, suggesting that farmers with higher income were more likely to adopt SRT. The official variable had a significant negative correlation, suggesting that farmers without family members holding positions in government departments were more likely to adopt SRT. The cultivated land area variable had significant positive effects, indicating that farmers with large cultivated land area were more likely to use SRT. The variables of soil fertility and the distance from cultivated land had significant positive and negative effects, respectively, indicating that farmers with better soil fertility and closer distance to cultivated land were more likely to use SRT. Moreover, the identification variable of the distance from home to agricultural technology station passed the significance test, indicating that farmers who were more likely to receive agricultural technology support and services were more likely to adopt SRT.

According to the estimation results in Table 6, we can explain the impact of SRT adoption on rice output value. The official variable indicates a significant and positive impact on rice output value for both unadopted and adopted farmers, suggesting that farmers whose family members held positions in government departments were more likely to obtain higher rice output value. For adopted farmers, the variables of education and soil fertility indicated significant and positive impacts on rice output value, suggesting that farmers with better education or better soil fertility were more likely to obtain higher rice output value. The variables of family size, cultivated land fragmentation and home-to-land distance had significant negative impact on rice output value, indicating that farmers with smaller family size, smaller cultivated land fragmentation and smaller home-to-land distance were more likely to obtain higher rice output value.

Table 6. Determinants of farmers adopting SRT and its impact on rice output value.

Variables	Selection		Adopted		Unadopted
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
Age	0.011	0.009	−0.000	0.001	0.001	0.002
Gender	−0.092	0.196	−0.001	0.014	0.043	0.038
Education	−0.131	0.118	0.030 ***	0.008	0.021	0.024
Population	−0.173 ***	0.039	−0.005 **	0.002	0.001	0.007
Income	0.103 ***	0.037	0.001	0.001	−0.005	0.007
Official	−0.448 **	0.218	0.065 ***	0.013	0.146 ***	0.043
Cultivated area	0.086 ***	0.020	−0.000	0.000	−0.001	0.005
Fragmentation	−0.166	0.109	−0.051 ***	0.013	−0.014	0.015
Soil fertility	0.376 ***	0.091	0.022 ***	0.008	0.016	0.018
Cultivation distance	−0.098 **	0.039	−0.015 ***	0.002	−0.002	0.007
Distance to service station	−0.200 ***	0.031
Constant	0.188	0.772	7.113 ***	0.056	6.800 ***	0.139
Lnσμy			−2.179 ***	0.034
Lnσμn					−1.619 ***	0.066
ρμy			−0.135	0.173
ρμn					−0.523 ***	0.170
Log likelihood	215.5087
LR test of indep. eqns.	6.38 **

Note: ** p < 0.05, *** p < 0.01.

Table 6. Determinants of farmers adopting SRT and its impact on rice output value.

Variables	Selection		Adopted		Unadopted
	Coef.	Std. Err.	Coef.	Std. Err.	Coef.	Std. Err.
Age	0.011	0.009	−0.000	0.001	0.001	0.002
Gender	−0.092	0.196	−0.001	0.014	0.043	0.038
Education	−0.131	0.118	0.030 ***	0.008	0.021	0.024
Population	−0.173 ***	0.039	−0.005 **	0.002	0.001	0.007
Income	0.103 ***	0.037	0.001	0.001	−0.005	0.007
Official	−0.448 **	0.218	0.065 ***	0.013	0.146 ***	0.043
Cultivated area	0.086 ***	0.020	−0.000	0.000	−0.001	0.005
Fragmentation	−0.166	0.109	−0.051 ***	0.013	−0.014	0.015
Soil fertility	0.376 ***	0.091	0.022 ***	0.008	0.016	0.018
Cultivation distance	−0.098 **	0.039	−0.015 ***	0.002	−0.002	0.007
Distance to service station	−0.200 ***	0.031
Constant	0.188	0.772	7.113 ***	0.056	6.800 ***	0.139
Lnσμy			−2.179 ***	0.034
Lnσμn					−1.619 ***	0.066
ρμy			−0.135	0.173
ρμn					−0.523 ***	0.170
Log likelihood	215.5087
LR test of indep. eqns.	6.38 **

Note: ** p < 0.05, *** p < 0.01.

3.4. Estimating Treatment Effects of Farmers Adopting GPTs

Table 7. Average treatment effect of different GPTs adopted by farmers.

Category	Adoption Scenarios	Decision Stage		ATT	ATU	Change/%
		To Adopt	Not to Adopt
PGM	Adopted	7.200	7.022	0.178 ***	-	2.472
	Unadopted	7.240	7.079	-	0.161 ***	2.274
STFFT	Adopted	7.258	6.595	0.663 ***	-	9.135
	Unadopted	7.177	7.062	-	0.115 ***	1.628
SRT	Adopted	7.174	6.891	0.283 ***	-	3.945
	Unadopted	7.061	6.993	-	0.068 ***	0.972

Note: Since the dependent variable entered into the endogenous switching regression outcome equation is the logarithm of the rice output value, the predicted results are also given in logarithmic form. As the arithmetic mean and geometric mean are not equal, converting the predicted mean back to a meaningful value will result in inaccuracy. *** p < 0.01.

presents the estimated results of the impact of green production technologies on rice output value. In general, the adoption of the three GPTs had a significant positive effect on farmers’ rice output value, which indicates that farmers’ adoption of GPTs can much improve farmers’ income. Specifically, the estimation results show that if the farmers who actually adopted PGM did not adopt PGM, their rice output value would have decreased by 2.47%. Similarly, for those farmers who actually adopted STFFT and SRT, if they did not adopt the corresponding technologies, their rice output value would have decreased by 9.14% and 3.95%, respectively. Moreover, for farmers who did not adopt technology, if PGM, STFFT and SRT were adopted, their rice output value would have increased by 2.27%, 1.63% and 0.97%, respectively.

In order to further reveal the behavioral economic effects of farmers adopting GPTs, we again estimated the treatment effect from different degrees of adoption perspectives. Overall, farmers who adopted more types of GPTs were more likely to achieve greater rice output value (

Table 8. Average treatment effect of different degrees GPTs adopted by farmers.

Category	Adoption Scenarios	Decision Stage		ATT	ATU	Change/%
		To Adopt	Not to Adopt
One kind	Adopted	7.084	6.909	0.175 ***	-	2.470
	Unadopted	7.005	6.984	-	0.021 ***	0.300
Two kinds	Adopted	7.195	6.866	0.329 ***	-	4.573
	Unadopted	7.174	7.037	-	0.137 ***	1.947
Three kinds	Adopted	7.260	6.956	0.304 ***	-	4.187
	Unadopted	7.293	7.093	-	0.200 ***	2.820

Note: “One kind” indicates that farmers have adopted one of the three GPTs, “two kinds” indicates that farmers have adopted two of the three GPTs, “three kinds” indicates that farmers have adopted the three GPTs. *** p < 0.01.

). Specifically, the estimation results show that if farmers who actually adopted one kind, two kinds, and three kinds did not adopt the corresponding number of varieties in the future, their rice output value would decrease by 2.47%, 4.57% and 4.19%, respectively. Similarly, for farmers who have not actually adopted one kind, two kinds, and three kinds, if they did adopt the corresponding number of varieties in the future, their rice output value would increase by 0.30%, 1.95% and 2.82%, respectively.

4. Discussion

In the process of achieving green agricultural development and cultivated land protection, GPTs can not only protect the agricultural ecological environment and realize the sustainable utilization of cultivated land [35,36], but also help to realize the yield and income increase of farmers [32,37]. However, GPTs have not been adopted by enough farmers, especially in developing countries [17,20,38]. This is unfortunate news and has seriously hindered the green development of agriculture. In this context, this study tried to explore the obstacles to farmers adopting GPTs, and show the economic benefits of GPTs. Taking China as a typical case study, it is hoped to provide a reference basis for the country in promoting green agricultural development, achieving ecological environmental protection and increasing farmers’ production and income.

It was found that the cultivated land area did not always have a significant positive or even negative effect on the adoption of GPTs. Previous studies have shown that expanding the area of cultivated land is conducive to the internalization of cultivated land construction costs, enabling farmers to achieve economies of scale, encouraging farmers to invest in medium- and long-term investments to improve the quality of cultivated land and motivating them to engage in environmentally friendly behaviors [39]. However, some studies have shown that expanding the scale of farmland has a negative scale effect and no significant impact on the pro-environmental behaviors of farmers [31,40]. Our results indicate that cultivated land area has a significant negative impact on farmers’ adoption of PGM. It is worth noting that, according to our field survey, planting green manure in paddy fields will cause agricultural machinery to stop working and even damage agricultural machinery. However, farmers with large-scale farming often use agricultural machinery to replace labor at a high frequency, to reduce input costs and improve production efficiency [20,41], thereby inhibiting large-scale farmers from adopting PGM.

The results showed that the negative impact of cultivated land fragmentation on GPTs adoption is consistent with that shown in existing studies [18,39], which suggests that the greater the area of cultivated land, the higher the degree of land fragmentation, the higher the cost of GPT adoption. Furthermore, the farmers with smaller distance from their homes to agricultural technology stations were more likely to adopt GPTs. This is in line with the existing literature that obtaining correct or sufficient agricultural information and technical guidance can effectively promote the adoption of GPTs by farmers in production and management [42,43].

The T-test results of the independent samples show that the farmers with the adoption of the three GPTs have a significantly higher rice output value and rice yield than the unadopted farmers. More interestingly, this behavioral economic effect of farmers is also verified by the endogenous switching regression model. The positive effect of GPTs in helping farmers to increase rice yield and income is consistent with other studies. Specifically, green manure can meet the needs of crops by enhancing the soil water storage capacity, regulating the soil nutrients and improving the organic matter content of the soil, so as to realize the improvement of crop yield and quality [44,45]. STFFT improves the soil fertility through precise fertilization to meet the needs of crops for various nutrients, so as to increase the yield and income of crops [46,47]. SRT can make full use of the nutrients in the straw (such as nitrogen, phosphorus, potassium and a variety of trace elements) through the straw returning treatment, improve the soil nutrient content, improve soil fertility and meet the needs of crops for nutrients, so as to increase crop yield and income [48,49].

Finally, one point that cannot be overlooked is that, although the article reveals some important findings, the study also has some limitations. It is worth pointing out that the estimated behavioral effect of farmers adopting GPTs may not be so high. Combined with our field survey, farmers showed two extreme phenomena in the adoption of GPTs, namely, either adopting a variety of GPTs at the same time, or adopting none of them at all. Therefore, this paper also contributes from two aspects in order to obtain more scientific and accurate estimates. On one hand, as many technologies as possible were selected from the three production links of rice, and the three GPTs are also widely adopted by farmers. On the other hand, the two indicators of rice yield and rice output value were selected to estimate and verify the behavioral effect of farmers adopting GPTs from the two dimensions of “adoption or not” and “degree of adoption”. Overall, how to more comprehensively and scientifically evaluate the behavior effect of farmers adopting GPTs also needs to be the subject of follow-up research. This paper is an attempt to enrich the research ideas and research basis.

5. Conclusions and Policy Implications

In order to better answer whether GPTs can improve family income, the 607 farmer survey data from six grain producing counties in Poyang Lake Plain of China were used to explore the influencing factors and behavioral effects of rice farmers’ adoption of GPTs. Given that GPTs is a general term containing a variety of specific agricultural technologies, this paper selected three agricultural technologies (PGM, STFFT and SRT) commonly used by farmers from the three links of rice production (before, during and post-harvest) to characterize GPTs. A simple comparison of rice yield and output value between technology adopters and non-adopters found some significant differences. Considering that these comparisons are only descriptive, without considering the confounding factors affecting the differences, we also employed the endogenous switching regression model, which considered the observed and unobserved factors to solve the problem of selection bias. The results do suggest that the sample had a self-selection problem. Specifically, the choice of farmers to adopt GPTs was not randomly generated, but the choice made by farmers based on changes in their own utility before and after adopting the technology. If this problem is not corrected, the estimated results will be biased.

The empirical results show that the adoption of GPTs can significantly and positively improve the level of household income. Specifically, if farmers who have actually adopted PGM, STFFT, and SRT do not continue to adopt corresponding technologies, their rice output value will decrease by 2.47%, 9.14% and 3.95%, respectively. If farmers who have not adopted PGM, STFFT and SRT choose to adopt corresponding technologies, the rice output value will increase by 2.27%, 1.63% and 0.97%, respectively. The estimation results indicate that farmers with higher education levels, larger cultivated land areas, lower fragmentation of cultivated land, better soil fertility and closer proximity to agricultural technology stations are more likely to adopt GPT. Furthermore, there are endowment differences in the welfare effect of GPT adoption. Farmers with higher education, family relatives and friends serving as government officials and better soil fertility will have better welfare effects of GPT adoption. Farmers with higher education, family relatives and friends serving as government officials and better soil fertility will have better welfare effects of GPT adoption. However, the larger the family, greater the cultivated land fragmentation and greater the distance from home to cultivated land, the worse the welfare effect of the adoption of GPTs.

Based on the above conclusions, the following policy implications can be drawn. First, reduce the fragmentation of cultivated land and realize moderate scale operation. It is suggested to reduce the fragmentation of cultivated land by means of land leveling and land adjustment, cultivate the farmland circulation market and encourage farmers to carry out moderate scale operations. Second, strengthen the publicity and promotion of technology and build a technology training system. It is recommended that the agricultural technology extension department take the lead in organizing publicity and education activities for GPTs, improve farmers’ cognition of GPTs, integrate the strength of multiple parties (agricultural technology research institutions, universities, agricultural enterprises, etc.), and build a multi-channel, multi-form and multi-level technical training system and training network. Third, show the benefits of technology and give full play to the demonstration and driving effect of technology. By screening some farmers or new agricultural business entities to carry out technology promotion and demonstration, farmers can more truly feel the benefits of GPTs and improve farmers’ enthusiasm to adopt GPTs.