Non-Farm Employment Experience, Risk Preferences, and Low-Carbon Agricultural Technology Adoption: Evidence from 1843 Grain Farmers in 14 Provinces in China ^†

Abstract

Guiding and encouraging farmers to adopt low-carbon agricultural technologies is highly significant for reducing greenhouse gas emissions, mitigating climate change, and achieving agricultural production development and food security. This study used survey data from 1843 grain farmers in 14 provinces of China to empirically analyze the impact of non-farm employment experience and risk preferences on grain farmers’ low-carbon agricultural technology (LCAT) adoption. The results show that for grain farmers: (1) non-farm employment experience significantly promoted the adopting of LCAT. The probability of adopting LCAT by those with non-farm experience is 23.5% higher than those without. (2) Non-farm employment experience reinforced their risk preferences and promoted the adoption of LCAT. The adoption probability of LCAT of those with high-risk preferences was 6.1% higher than those with low-risk preferences. (3) The impact of non-farm employment experience on adopting LCAT was more significant for those with high education and training. Non-farm employment experience outside the province and employment experience in the tertiary sector while working outside significantly affect grain farmers’ LCAT adoption.

1. Introduction

Global climate change, primarily characterized by climate warming, has significantly impacted the Earth’s natural systems and human economic and social development [1], especially in developing countries [2]. Addressing climate change has become a global consensus and is included as one of the sustainable development goals of the United Nations (UN) [3]. In response to climate change, countries worldwide have created their emission reduction targets. At the 75th session of the UN General Assembly in September 2020, the Chinese government pledged to achieve peak carbon dioxide emissions by 2030 and carbon neutrality by 2060, and integrated carbon peaking and neutrality into the overall layout of ecological civilization construction.

The agricultural sector is crucial to achieving carbon peaking and neutrality goals because of its dual attributes as a carbon sink and carbon source. Agriculture uses photosynthesis in farmland, forests, grasslands, and other ecosystems for biological carbon sequestration, absorbing approximately 30% of global anthropogenic carbon emissions each year [4]. In contrast, carbon emissions from agricultural production and land use change are a vital source of greenhouse gas emissions, accounting for approximately 25% of the total annual average global carbon emissions [5]. China is the world’s largest carbon emitter, and the share of China’s agricultural greenhouse gas emissions worldwide is approximately 11–12% [6]. In 2018, China’s total agricultural carbon emissions amounted to 801.61 million tons of carbon dioxide equivalent, accounting for 6.85% of China’s total emissions; fertilizer application, energy use, straw burning, and rice cultivation accounted for 55.43% among the sources of carbon emissions [7]. This is closely related to China’s traditional extensive agricultural production modes, such as excessive agricultural chemical input and excessive utilization of agricultural land resources. Therefore, regarding carbon reduction in agriculture, promoting green and low-carbon development in agriculture has a great potential for emission reduction.

Promoting low-carbon agricultural technologies (LCAT) is critical for green and low-carbon agricultural development and achieving greenhouse gas emission reduction [8]. LCAT combines various low energy consumption, emission, and pollution methods and means adopted in agricultural production and operation, including soil testing and formulated fertilization, organic fertilizer, straw returning, conservation tillage technology, and others [9]. Adopting LCAT in crop farming and animal husbandry can reduce greenhouse gas emissions from agrifood systems by 23% in 2060, compared to 2020 levels [10].

Studies have suggested that farmers’ LCAT adoption can effectively reduce pollution emissions and save agricultural production costs, improve agricultural output and farmers’ income, and achieve the compatibility of ecological and economic benefits [11,12,13,14]. However, agricultural production has the characteristics of a long cycle, low added value, and high risk. Additionally, the long-term production concept makes farmers more dependent on traditional agricultural production methods, leading to low enthusiasm and proportion for adopting LCAT [15,16]. These factors have severely restricted food security and the green, low-carbon development of agriculture [17,18].

The literature on farmers’ LCAT adoption behavior primarily analyzes whether they adopt one or more LCAT based on their characteristics, land endowment, subjective cognition, and information technology use [19,20,21]. For example, Yu and Luo [22] used a logistic model and found that gender, age, and agricultural income significantly promoted farmers’ soil testing and fertilizer recommendation adoption, while farming experience inhibited their soil testing and fertilizer recommendations adoption. Qi et al. [23] showed that land scale and fragmentation should be critical factors affecting farmers’ eco-friendly fertilization technology adoption. Hou and Hou [24] found that farmers’ attitudes, subjective norms, perceived behavior control, and order farming participation significantly affect LCAT adoption, such as straw-biogas production, soil testing for formulated fertilizer, and water-saving irrigation. Zheng et al. [25] confirmed that internet usage can improve LCAT adoption. The probability of farmers who use the internet adopting new low-carbon crop varieties, water-saving irrigation technology, and straw-returning technology could be increased by 20%, 15.7%, and 15.5% than those who do not, respectively. For different farmers, adopting LCAT may mean risks, and risk bias also determines LCAT use. Liu and Huang [26] believed that risk aversion should be the critical factor in the slow diffusion of LCAT. Research on farmers’ technology adoption behavior based on scale and type heterogeneity show that risk aversion has an inhibitory effect on farmers’ LCAT adoption [27]. Cracking the restraining effect of farmers’ risk aversion on adopting LCAT is crucial for its promotion. Expected utility theory suggests that farmers’ risk preferences were not fixed [28], and the increase in their family income or assets would affect the risk preferences change [29]. Research shows that guiding farmers to purchase agricultural insurance [27], accumulating social capital [30], participating in agricultural technology training [31], and farmers’ cooperatives [32,33] are effective means.

The non-farm employment experience is an essential characteristic difference between different farmers. With the gradual return of the first generation of migrant workers, the number of farmers with non-farm employment experience has gradually increased. However, few studies examine the impact of non-farm employment experience on adopting LCAT. Their “learning by doing” characteristic is considered a particular human capital investment and training, which exerts a subtle influence on the individual characteristics and psychological status of farmers and may increase farmers’ risk awareness and anti-risk ability, mitigating the inhibition of risk aversion on LCAT adoption [34]. Since China’s reform and opening up, the process of industrialization, urbanization, and marketization in China has accelerated, and rural labor working outside has become common. In recent years, with the development of the county economy and the implementation of the rural revitalization strategy, China’s population flow has gradually broken through the traditional single mode of transferring from rural areas to cities, and the rural-urban backflow has become a crucial form of migration. Currently, the active and selective backflow of migrant workers is substantial, and they are less likely to migrate after returning to their hometowns [35]. Most studies on migrant workers’ return focus on their decision-making and performance of return to their hometown entrepreneurship [36,37,38]. Notably, in the context of the overall promotion of the rural revitalization strategy, many returning migrant workers engaged in agricultural production. Guo and Zhou [39] found that nearly 50% of the older generation of returning labor are engaged in agricultural production, while 31.4% of the new generation migrant workers are willing [40]. The critical difference between farmers with and without non-farm employment experience is reflected in their different attitudes toward risk preferences. Therefore, it is of great practical significance and academic value to analyze the impact of non-farm employment experience and risk preferences on farmers’ LCAT adoption.

This study uses survey data from 1843 grain farmers in 14 provinces of China to explore how the non-farm employment experience and risk preferences affect grain farmers’ LCAT adoption. The main contributions of this study are as follows. First, it analyzes grain farmers’ LCAT adoption from the perspective of non-farm employment experience, analyzes risk preferences into a unified research framework, and expands the boundaries of existing research. Second, it focuses specifically on grain farmers. Grain farmers are the main body of grain production in China. Analyzing their LCAT adoption for green agricultural development and food security is essential. Third, it uses grain farmers’ LCAT adoption types to measure their adoption behavior. LCAT integrates multiple technologies [18]. Hence, it is crucial to focus on whether farmers adopt the technology and also the degree of adoption [41].

2. Theoretical Analysis and Hypotheses

2.1. Non-Farm Employment Experience and Farmers’ LCAT Adoption

The impact of personal experience has been extensively studied in medicine, psychology, and behavioral economics [42,43]. Personal experience, as a kind of wealth and capital, accumulates relevant knowledge, experience, or specific skills and affects production decisions [44]. Non-farm employment experience affects farmers’ adoption of LCAT. Compared to those without, farmers with non-farm employment experience have higher incomes [31], more capital, and savings accumulation [45]. Capital accumulation and income increase directly affect farmers’ agricultural production decisions, investing in agricultural production, and improving their financial ability to adopt LCAT. Furthermore, non-farm employment experience can improve their courage, reaction, information acquisition capability, expand their social network, enhance their technology acceptance ability and agricultural sustainable development awareness [46]. Therefore, non-farm employment experience is conducive to the adoption of LCAT. We propose our first hypothesis as follows.

Hypothesis (H1): 

Farmers with non-farm employment experience are more likely to adopt LCAT than those without non-farm experience.

2.2. Risk Preferences and Adoption of LCAT

Agricultural production integrates natural and market risks. Farmers should consider profit maximization and risk minimization when making agricultural production decisions. Although LCAT is crucial in reducing chemical pesticide use, ensuring agricultural production safety, and promoting sustainable agricultural development [12,13], its adoption also poses a greater risk for farmers. First, there are unknown risks in adopting new technologies. Due to the imperfect agricultural insurance system, most farmers have a weak ability to resist risks. If they blindly adopt new technologies, they may not gain economic benefits and face agricultural losses [47]. Second, there is a risk of net income uncertainty. The agricultural price changes frequently. Regarding asymmetric information, farmers’ bargaining power is generally weak. They often become passive recipients of prices and face difficulties in obtaining higher agricultural income [48]. The high-quality agricultural products produced by adopting LCAT may not reflect the “high quality and good price” in the market due to the “lemon effect” (the “lemon effect” is that in the case of information asymmetry, often the inferior goods will gradually replace the good goods, resulting in the market is full of inferior goods), resulting in uncertainty of net income. Third, there is the risk of improper technology use. The complexity of agricultural technology determines its high requirements for technology adopters. If farmers have not yet mastered the new technology, an improper operation can create an inferior or even an opposite effect than expected [49]. However, farmers with different types of risk preferences hold different attitudes toward LCAT [26,50]. Therefore, we propose our second hypothesis as follows.

Hypothesis (H2): 

Farmers with high-risk preferences are more likely to adopt LCAT than those with low-risk preferences.

2.3. Non-Farm Employment Experience, Risk Preferences, and LCAT Adoption

Non-farm employment experience directly affects farmers’ adoption behavior of LCAT and promotes its adoption by strengthening their risk preferences. First, Non-farm employment experience improves farmers’ awareness of LCAT [34]. The premise of adopting LCAT is to recognize the LCAT. Farmers with non-farm employment experience have a stronger awareness of environmental protection and ecological civilization and are more likely to understand the long-term benefits of LCAT [51]. This may reduce the risk of LCAT adoption. Second, Non-farm employment experience strengthens farmers’ ability to obtain market information [46]. Compared with local farmers, they may moderately perform commercial agriculture production following specific market needs to rectify the information asymmetry of the market price for agricultural products and achieve a “high-quality price” for their products [52]. They also make necessary adjustments to their production and management mode following the changes in market demand to reduce uncertainty risk [48]. Third, non-farm employment experience accumulates human capital [53]. The higher human capital of farmers helps them learn about technical operation specifications through various channels, accurately use various LCATs, and reduce the risk of improper use of technologies [34]. Therefore, we propose our third hypothesis as follows.

Hypothesis(H3): 

Non-farm employment experience can strengthen farmers’ risk preferences and promote their adoption of LCAT.

Based on the theoretical analysis above, the conceptual framework is shown in Figure 1.

3. Materials and Methods

3.1. Model Selection

This study aims to analyze the impact of non-farm employment experience and risk preferences on grain farmers’ LCAT adoption. Previous primary research used binary logit and ordered logit models to analyze one or several technology adoptions [31,41]. As the quantity of grain farmers’ LCAT adoption is a counting variable, the value range is 0 to 6, wherein 0 indicates no LCAT adoption, and 6 indicates the maximum number of LCAT types adopted.

Table 1. Definition and Description of Six LCAT.

LCAT	Definition and Description	Mean	Std.Err.
Straw retuning	1 if grain farmers adopted it,0 otherwise	0.482	0.499
Deep plowing subsoiling	1 if grain farmers adopted it,0 otherwise	0.438	0.496
Soil testing fertilization	1 if grain farmers adopted it,0 otherwise	0.173	0.378
No-tilling direct sowing	1 if grain farmers adopted it,0 otherwise	0.145	0.352
Green field crops	1 if grain farmers adopted it,0 otherwise	0.142	0.348
Soil conditioner	1 if grain farmers adopted it,0 otherwise	0.103	0.304

shows the descriptive analysis of grain farmers’ six LCAT adoptions. Poisson regression model is often used for counting data. If the number of grain farmers’ LCAT adoption $Y_i$ follows the Poisson distribution with parameter ${\lambda }_i$, then its probability distribution function is:

$$P(Y_i=y_i|x_i)=\frac{e^{-{\lambda }_i}{\lambda }_i^j}{y_i!} y_i=0,1,2\cdot \cdot \cdot$$

(1)

where ${\lambda }_i$ represents the “Poisson arrival rate,” representing the average number of grain farmers’ LCAT adoption, determined by the explanatory variable $x_i$. To ensure that ${\lambda }_i$ is non-negative, assume that the conditional expectation function of $Y_i$ is:

$$E(Y_i|x_i)={\lambda }_i=\exp (x_i^{’}\beta )$$

(2)

Taking the logarithm of both sides of the Equation (2), we can achieve:

$$Ln{\lambda }_i=x_i^{’}\beta$$

(3)

where $\beta$ is an estimated parameter, and $x_i$ is a set of vectors affecting grain farmers’ LCAT adoption, including non-farm employment experience, risk preferences, farmers’ individual characteristics, family characteristics, land characteristics, and village characteristics. However, a critical assumption of Poisson regression is that the mean and variance of the explained variable must be equal. If the mean and variance of the explanatory variables of the observed sample data are not equal, or there is “excessive dispersion”, using negative binomial regression model is more advantageous [54]. The expression of negative binomial regression is constructed by introducing an unobservable individual effect into the conditional mean of the Poisson model:

$$Ln{\lambda }_i=x_i^{’}\beta +{\epsilon }_i$$

(4)

In Equation (4), ${\epsilon }_i$ is individual heterogeneity or unobservable part, and other variables are consistent with Equation (3).

3.2. Variables

3.2.1. Explained Variable

LCAT is not a specific technology, but a variety of methods and means used by agricultural producers to reduce carbon emissions before, during, and after agricultural production in the process of agricultural production and management [9]. LCAT emphasizes the reduction of carbon emissions to protect the environment [55]. Based on existing studies [51] and considering the characteristics of grain production, soil conservation low-carbon agricultural technologies were mainly selected, including straw returning, deep plowing subsoiling, soil testing fertilization, no-tillage direct sowing, green field crops, and soil conditioner.

3.2.2. Key Explanatory Variable

This refers to the non-farm employment experience, indicated by “whether the householder has engaged in non-farm work for more than six consecutive months.” If the householder has engaged in non-farm work for more than six consecutive months, the value is 1; otherwise, the value is 0.

3.2.3. Intermediary Variable

Risk preferences refers to the respondents’ attitude toward risk. By referring to the risk preferences measurement methods in financial investment and China Household Finance Survey (CHFS), this paper uses investment preference to measure risk preference [56], and their risk preferences are classified into three types: low (low-risk preferences, the value is 1), medium (medium-risk preferences, the value is 2), and high (high-risk preferences, the value is 3) (“If you have 10,000 RMB to buy wealth management, which of the following three choices do you personally prefer?” A. Best case earn RMB 400 (4%), worst case no loss; B. Best case earn RMB 1700 (17%), worst case lose RMB 1000 (10%); C. Best case earn RMB 9600 (96%), worst case lose RMB 4800 (48%)).

3.2.4. Control Variables

To control the impact of other variables on grain farmers’ LCAT adoption, we introduce four characteristic variables, individual characteristics, including the age and education level of the household head [57,58,59]. Family characteristics, including the labor force number, agricultural income, whether to participate in farmers’ cooperatives and whether to adopt social services [33,60]. Land characteristics, including land area, block number, and quality [61]. Village characteristics, including village terrain and economic level. Considering the regional differences among China’s regions, we used dummy variables to control them [62]. Variable definitions and descriptive statistics are shown in

Table 2. Variable definitions and descriptive statistics.

Variable	Variable Definition	Mean	Std.Err.	References
LCAT	Number of LCAT adoption.	1.482	1.257	[51]
Non-farm employment experience	1 = Yes; 0 = No	0.595	0.491
Risk preferences	1 = low; 2 = medium; 3 = high	1.559	0.714	[56]
Age	householder’ age (years)	52.115	10.839	[57,58,59]
Education level	1 = junior high school and above; 0 = primary school and below	0.602	0.490
Labor force number	The total number of 18–60 year old labor force in the family	3.395	1.683	[33,60]
Agricultural income	Logarithm of household agricultural income (Yuan)	8.523	2.915
Cooperatives	Whether to participate in cooperatives (1 = Yes; 0 = No)	0.144	0.352
Social services	Whether to adopt social services (1 = Yes; 0 = No)	0.551	0.497
Land area	Logarithm of land area (mu) a	2.227	1.203	[61]
Land blocks number	Logarithm of land blocks number	1.631	0.657
Land quality	1 = very poor; 2 = poor; 3 = ordinary; 4 = good; 5 = very good	3.035	0.868
Village terrain	1 = plain; 0 = non plain	0.501	0.500
Village economic level	1 = very poor; 2 = poor; 3 = ordinary; 4 = good; 5 = very good	3.291	0.924

^a 1 mu = 667 m² or 0.067 ha.

.

3.3. Data Source

This study utilized data from a 2019 survey of grain farmers in 14 provinces of China by the National Agricultural and Rural Development Research Institute of China Agricultural University. First, considering China’s major grain-producing areas, we selected 14 provinces, including Hebei, Shandong, Jiangsu, Henan, Anhui, Hubei, Jiangxi, Hunan, Inner Mongolia, Sichuan, Gansu, Heilongjiang, Jilin, and Liaoning, covering eastern, central, western and northeastern China (there are 13 major grain-producing areas in China, namely Hebei, Shandong, Jiangsu, Henan, Anhui, Hubei, Jiangxi, Hunan, Inner Mongolia, Sichuan, Heilongjiang, Jilin and Liaoning) The sample distribution is shown in Figure 2. Second, in the selected provinces, 124 counties and 142 villages were selected by using a typical sampling method, and 12 to 15 grain farmers were randomly selected as research participants in each village. The survey was conducted as one-on-one interviews with the investigators, and a total of 1952 research questionnaires were collected. Finally, excluding questionnaires with missing data and inconsistent information, 1843 valid questionnaires were obtained, for a valid rate of 94.42%. The survey included two levels: family and village. The family level includes farmers’ characteristics, labor employment, land operation scale, grain production, LCAT adoption, and others. The village level considers the village terrain, economic level, population characteristics, land size, and other relevant information.

4. Results

4.1. Baseline Regression

Using Stata17, we analyzed the impact of non-farm employment experience on grain farmers’ LCAT adoption. The likelihood ratio test rejected the original hypothesis at the 5% significance level, rejecting the over-dispersion parameter “alpha = 0.” Therefore, using negative binomial regression was more appropriate in this study. The regression results are shown in

Table 3. Non-farm employment experience impact on grain farmers’ LCAT adoption.

Variable	Model (1)	Model (2)	Model (3)	Model (4)
	Coefficient	IRR	Coefficient	IRR
Non-farm employment experience	0.211 ***	1.235 ***	—	—
	(0.042)	(0.052)	—	—
Risk preferences	—	—	0.059 **	1.061 **
	—	—	(0.026)	(0.028)
Age	−0.001	0.999	0.003	1.003
	(0.002)	(0.002)	(0.002)	(0.002)
Education level	0.017	1.017	-0.002	0.988
	(0.039)	(0.040)	(0.038)	(0.038)
Labor force number	0.042 ***	1.042 ***	0.035 ***	1.036 ***
	(0.010)	(0.011)	(0.011)	(0.011)
Agricultural income	0.036 ***	1.037 ***	0.039 ***	1.040 ***
	(0.010)	(0.010)	(0.010)	(0.011)
Cooperatives	0.170 ***	1.185 ***	0.139 ***	1.149 ***
	(0.047)	(0.055)	(0.047)	(0.054)
Services	0.564 ***	1.758 ***	0.548 ***	1.729 ***
	(0.042)	(0.074)	(0.042)	(0.073)
Land area	0.059 ***	1.061 ***	0.079 ***	1.082 ***
	(0.020)	(0.022)	(0.020)	(0.022)
Number of land plots	−0.099 ***	0.906 ***	−0.101 ***	0.904 ***
	(0.034)	(0.031)	(0.034)	(0.030)
Land quality	0.063 ***	1.066 ***	0.066 ***	1.069 ***
	(0.022)	(0.023)	(0.022)	(0.023)
Village terrain	0.037	1.038	0.026	1.026
	(0.038)	(0.040)	(0.039)	(0.040)
Village economic level	0.156 ***	1.169 ***	0.164 ***	1.178 ***
	(0.022)	(0.025)	(0.022)	(0.026)
Regional dummy variable	controlled	controlled	controlled	controlled
Constant	−0.253	0.777	−0.388 **	0.679 **
	(0.172)	(0.133)	(0.183)	(0.124)
R-squared	0.075	0.075	0.072	0.072

Note: ***, ** mean that the estimated results are significant at 0.01, and 0.05; the robustness standard error is reported in parentheses.

.

The estimated coefficient of non-farm employment experience in model (1) is 0.211, which satisfies the 1% significance level test implying that non-farm employment experience positively impacts grain farmers’ LCAT adoption. The incidence rate ratio (IRR) in model (2) is 1.235, confirming that compared with grain farmers without non-farm employment experience, those with non-farm employment experience have a 23.5% higher probability of adopting LCAT, supporting H1. The estimated coefficient of risk preferences in model (3) is 0.059. Satisfying the 5% significance level test shows that the impact of risk preferences on grain farmers’ LCAT adoption is significantly positive. The IRR in model (4) is 1.061, indicating that, compared with grain farmers with low-risk preferences, those with high-risk preferences have a 6.1% higher probability of adopting LCAT, supporting H2.

Both the labor force number and agricultural income have a significant positive impact on LCAT adoption, indicating that grain farmers with more labor and higher agricultural income are more willing to adopt LCAT, aligning with the results of Mugi Ngenga et al. [63]. Land area and quality significantly positively impact LCAT adoption, indicating that grain farmers with larger land area and higher land quality were more inclined to adopt LCAT. The larger the land area and higher the quality of grain farmers, the higher the utilization efficiency of LCAT, and the cost of adopting LCAT is relatively reduced.

Social services and participation in farmer cooperatives significantly positively affect grain farmers’ LCAT adoption. Purchasing social services can help them obtain more technical information and services [64]. Participation in farmer cooperatives can effectively reduce their input–output and market transaction costs [65], encouraging them to adopt LCAT. The number of land plots significantly negatively impacts the adoption of LCAT; the more land blocks, the more dispersed the land and the less obvious its scale effect, reducing grain farmers’ willingness to adopt LCAT. The village economic level has a significant positive impact on the adoption of LCAT, indicating that the higher the village’s economic development, the more able the grain farmers are to pay for the adoption of LCAT, and the more willing the grain farmers are to adopt LCAT.

4.2. Mechanism Test

The non-farm employment experience may affect grain farmers’ LCAT adoption by changing their risk preferences. To explore the mechanism of non-farm employment experience on grain farmers’ LCAT adoption, the following recursive model is constructed based on the intermediary effect test method [66]:

$$\{\begin{array}{l}LCA{T}_i={\theta }_1+c{\mathrm{NFE}}_i+\gamma Z+{\epsilon }_1 \left(5\right) \\ Ris{k}_i={\theta }_2+a{\mathrm{NFEe}}_i+\eta Z+{\epsilon }_2 \left(6\right)\\ LCA{T}_i={\theta }_3+c^{’}{\mathrm{NFE}}_i+bRis{k}_i+\omega Z+{\epsilon }_3 \left(7\right)\end{array}$$

$LCA{T}_i$ is the explained variable, indicating the number of LCAT adoptions. ${\mathrm{NFE}}_i$ is an explanatory variable. $Ris{k}_i$ is an intermediary variable. ${\theta }_i$ is a constant term. $\gamma$, $\eta$, and $\omega$ are the coefficient vector to be estimated for the control variable. ${\epsilon }_i$ is the error term. In Equation (5), $c$ is the total effect of non-farm employment experience on the adoption of LCAT. In Equation (6), $a$ is the effect of non-farm employment experience on the intermediary variable. In Equation (7), $b$ is the effect of the intermediary variable on the adoption of LCAT after controlling the influence of non-farm employment experience. $c^{’}$ is the indirect effect of non-farm employment experience on the adoption of LCAT after controlling the influence of intermediary variables. In the above recursive model, the mediating effect can be judged by comparing the significance between the estimated coefficients. The estimated results are shown in

Table 4. Mechanism test of non-farm employment experience on grain farmers’ LCAT adoption.

Variable	LCAT	Risk Preferences b	LCAT
	Model (5)	Model (6)	Model (7)
Non-farm employment experience	0.211 ***	0.191 *	0.206 ***
	(0.042)	(0.108)	(0.042)
Risk preferences	—	—	0.050*
	—	—	(0.026)
Control variable	controlled	controlled	controlled
Constant	−0.388 **	—	−0.359 **
	(0.183)	—	(0.181)
R-squared	0.072	0.027	0.075
Wald chi2	362.61	95.64	378.88

Note: ***, **, * mean that the estimated results are significant at 0.01, 0.05, and 0.1; the robustness standard error is reported in parentheses. ^b Since risk preferences is an ordered variable, Ologit model is selected for estimation in model 6.

.

Model (5) shows that without considering risk preferences, the influence of non-farm employment experience on LCAT adoption behavior of grain farmers is significantly positive, indicating it increases the number of LCAT adoptions. In model (6), the impact of non-farm employment experience on grain farmers’ risk preferences is positive. It satisfies the significance test at the 10% level, indicating that compared with grain farmers without non-farm employment experience, those with non-farm employment experience have higher risk preferences. In model (7), non-farm employment experience and risk preferences significantly impact grain farmers’ LCAT adoption. Considering the intermediary effect judgement, three coefficients $a$, $b$ and $c$ are significant. The sign of $a*b$ is the same as that of $c$, indicating that non-farm employment experience has an intermediary effect on grain farmers’ LCAT adoption through strengthening risk preferences. However, $c^{’}$ is also significant, indicating that risk preferences play a partial mediating role in the impact of non-farm employment experience on grain farmers’ LCAT adoption, supporting H3.

4.3. Robustness Test

4.3.1. Replace Explained Variable

We set the LCAT adoption behavior of grain farmers as a 0–1 variable; the value of 0 was assigned to those who did not adopt any LCAT, and the value of 1 was assigned to those who adopted at least one LCAT. Logit model was used for regression, and the estimated results are shown in

Table 5. Robustness test.

Variable	Model (8): Logit	Model (9): Negative Binomial Regression
Non-farm employment experience	0.394 ***	—
	(0.133)	—
Non-farm employment time	—	0.003 ***
	—	(0.001)
Control variable	controlled	controlled
Constant	−0.334	−0.257
	(0.487)	(0.173)
R-squared	0.145	0.072
LR chi2	313.14	349.72

Note: *** mean that the estimated results are significant at 0.01; the robustness standard error is reported in parentheses.

. The result of model (8) shows that the coefficient of non-farm employment experience is 0.394, affecting grain farmers’ LCAT adoption at the 1% significance level, verifying the reliability of the baseline regression results.

4.3.2. Replace Key Explanatory Variable

We use the non-farm employment time as a proxy variable for non-farm employment experience to estimate its effect on the adoption of LCAT by grain farmers. The negative binomial regression model is used and the estimation results are shown in Table 5. As seen in model (9), the direction and significance level of the effect of non-farm employment time on farmers’ LCAT adoption behavior remain consistent with the baseline regression results.

4.4. Heterogeneity Analysis

This study analyzes the heterogeneity impact of non-farm employment experience on grain farmers’ LCAT adoption from four aspects: education level, receiving training, non-farm employment location, and non-farm employment industry (as shown in

Table 6. Heterogeneity test.

Variable	High Education Level	Low Education Level	Trained	Untrained
Non-farm employment experience	0.214 ***	0.164 **	0.141 *	0.077
	(0.050)	(0.075)	(0.076)	(0.079)
Control variable	controlled	controlled	controlled	controlled
Constant	−0.150	−0.316	−0.106	0.158
	(0.204)	(0.289)	(0.347)	(0.300)
R-squared	0.061	0.102	0.060	0.052
Wald chi2	190.98	204.21	84.83	110.02

Note: ***, **, * mean that the estimated results are significant at 0.01, 0.05, and 0.1; the robustness standard error is reported in parentheses.

and

Table 7. Heterogeneity of non-farm employment experience.

Variable	Model (10)	Model (11)	Model (12)	Model (13)
Within the province	0.076 *	—	—	—
	(0.041)	—	—	—
Outside the province	—	0.118 ***	—	—
	—	(0.038)	—	—
Secondary	—	—	0.120 *	—
	—	—	(0.063)	—
Tertiary	—	—	—	0.159 ***
	—	—	—	(0.039)
Control variable	controlled	controlled	controlled	controlled
Constant	−0.269	−0.252	−0.268	−0.252
	(0.174)	(0.172)	(0.173)	(0.172)
R-squared	0.071	0.072	0.071	0.073
Wald chi2	349.57	351.36	347.25	357.64

Note: ***, * mean that the estimated results are significant at 0.01, and 0.1; the robustness standard error is reported in parentheses.

). These are discussed below:

4.4.1. Education Level

Farmers with a primary school and below are classified as having a low education level and those with a junior high school and above are classified as having a high education level [25]. The coefficient of non-farm employment experience of farmers with a high education level is significantly higher than that of farmers with a low education level, indicating that non-farm employment experience has a higher impact on LCAT adoption by farmers with a high education level. This may be because those with a high education level have a better understanding and response ability to new technology information and can realize the long-term economic benefits of LCAT, promoting its adoption [34].

4.4.2. Receive Technical Training

Concerning the technical training characteristics of grain farmers, non-farm employment experience significantly promotes LCAT adoption for those who receive technical training, but has no significant impact for those who do not receive technical training. This may be because trained grain farmers can improve their understanding of LCAT and better understand its risks, production costs, and economic benefits, aligning with the results of Liu’s [31] study, which suggests that technical training encourages farmers to adopt LCAT.

4.4.3. Non-Farm Employment Location

Considering the varied employment places of grain farmers working outside, their non-farm employment locations are divided into non-farm employment within and outside the province. According to the Monitoring and Investigation Report of Migrant Workers 2021, the total number of China’s migrant labor will be 293 million in 2021, including 172 million non-local migrant labor and 121 million local migrant labor (http://www.stats.gov.cn/xxgk/sjfb/zxfb2020/202204/t20220429_1830139.html, accessed on 25 November 2022), see the results of model (10) and model (11) in Table 7. We find that the impact of non-farm employment experience outside the province on LCAT adoption is greater than the impact of non-farm employment experience within the province, indicating that non-farm employment experience outside the province is more conducive to the adoption of LCAT than non-farm employment experience within the province. Non-farm employment experience outside the province can help farmers break the blood and geographical limitations, provide a wider social network, help farmers better understand market information, technical information, and technological achievements, reduce cognitive bias, reduce the uncertainty of adopting LCAT, and be more at profit from LCAT adoption [67].

4.4.4. Non-Farm Employment Industry

The grain farmers’ employment industry is divided into secondary and tertiary (non-farm employment experience in the secondary industry refers to the non-farm employment of farmers in the manufacturing and construction industries; non-farm employment experience in the tertiary industry refers to the non-farm employment of farmers in accommodation, catering, wholesale and retail, transportation, business services and other industries), see the results of model (12) and model (13) in Table 7. We find that the non-farm employment experience in the tertiary sector has a more significant impact on LCAT adoption than that in the secondary sector. There are significant differences in the industries that rural laborers employ in the city. According to the Monitoring and Investigation Report of Migrant Workers 2021, the proportion of rural labor employed in the secondary and tertiary sector is 48.60% and 50.9%, respectively, leading to varied learning abilities and receptivity to new things. Compared with the labor-intensive secondary sector, the tertiary sector has greater labor flow and fierce competition, conducive to the acquisition of labor skills and accumulation of human capital among grain farmers, and thus significantly impacts LCAT adoption [68].

5. Discussion

Compared to previous studies, this study differs in the following aspects. First, previous studies have focused on the direct effect of risk preferences on the adoption of LCAT [69]. As the main body of LCAT adoption in agriculture, the risk preferences of grain farmers are an important reason for the slow diffusion of LCAT [26]. The flow trend of China’s rural labor shows that the first generation of migrant workers will cross 60 years old, and the possibility of their employment in non-farm industries is gradually decreasing. Therefore, they will gradually return to agriculture. Additionally, the Chinese government has continued to increase financial support for agricultural and rural areas, and it has become a new trend for migrant laborers to return to their hometowns to start businesses. The Ministry of Agriculture and Rural Affairs of the People’s Republic of China estimated that, in 2021, approximately 11.2 million migrant laborers returned to agricultural and rural areas for employment, indicating that the number of farmers with non-farm employment experience will increase. This study included non-farm employment experience and risk preferences in the unified theoretical analysis framework, analyzed the impact of non-farm employment experience on grain farmers’ LCAT adoption, and tested the mechanism of influence of non-farm employment experience on strengthening risk preferences and promoting their adoption of LCAT, expanding the boundaries of the existing research. Second, previous studies were mainly conducted from the perspective of whether or not the head of household had non-farm employment experience [70] and focused on the impact of non-farm employment experience on farmers’ return to their hometown to start businesses [71]. For example, Batista et al. [38] confirmed that the presence of family members with non-farm employment experience would increase the probability of family entrepreneurship by 13%. Few studies have explored the possible differential effects of heterogeneity of non-farm employment experience on the adoption of LCAT. The location and industry of non-farm employment experience affect both the social capital accumulation and human capital accumulation of rural-returning labor [53]. Regardless of whether the location of non-farm employment is local or non-local, the non-farm employment industry is secondary or tertiary. Information, technology, and ideas will be transmitted through social networks, influencing the adoption of LCAT. Third, LCAT represents an integration of multiple technologies [9]. Previous studies mainly focused on whether farmers adopted LCAT or not [24]. For example, Liu and Zheng [72] analyzed the impact of social capital on farmers’ willingness to adopt LCAT. Liu et al. [31] analyzed the influence of technical training on the adoption of LCAT. This study focuses on the number of grain farmers’ LCAT adoption. Grain farmers will adopt a variety of LCAT in agricultural production. The whole process of adopting LCAT cannot be analyzed by using a single technology measurement method, and the substitution relationship between technologies cannot be included in the measurement method of a single technology [16]. Research on the adoption behavior of LCAT should focus not only on whether or not farmers adopt LCAT, but pay more attention to the degree of adoption as well [57].

6. Conclusions and Policy Implications

6.1. Conclusions

Based on the survey data of 1843 grain farmers in China’s 14 provinces, this study uses the counting model and the intermediary effect model to explore the impact and mechanism of non-farm employment experience on grain farmers’ LCAT adoption. The conclusions are as follows: non-farm employment experience can significantly promote grain farmers’ LCAT adoption. The adoption probability of LCAT by those with non-farm experience is 23.5% higher than those without non-farm experience. Mechanism analysis shows that non-farm employment experience can promote grain farmers’ LCAT adoption by strengthening the risk preference. Grain farmers with high-risk preferences are 6.1% more likely to adopt LCAT than those with low-risk preferences. The heterogeneity analysis shows that the impact of non-farm employment experience on adopting LCAT is more significant for grain farmers with high education and technical training. When grain farmers have non-farm employment experience outside the province, they are more willing to adopt LCAT. Grain farmers with tertiary sector employment experience are significantly more likely to adopt LCAT than those with employment in the secondary sector.

6.2. Policy Implications

To promote the transformation of the agricultural production mode from traditional extensive to green and low-carbon, vigorously developing LCAT, realizing the coordination between resources and the environment, and ensuring food security are crucial to cope with climate change. The following policy recommendations are proposed based on our study’s results: First, the risks of grain farmers’ LCAT adoption must be reduced. We should improve the early warning system for major agricultural products for production, supply, demand, price, monitoring, and reduce information asymmetry in the market. Vigorously developing contract agriculture, realizing the effective connection between farmers and the market, and reducing the market risk of farmers. Second, the government should form a relatively complete LCAT training system. With the help of Internet information technology, the government should disseminate LCAT training contents to farmers and increase the scale of adoption of LCAT among grain farmers. Third, the government should encourage grain farmers to join professional cooperative organizations, regularly perform technical training in cooperatives, and exchange agricultural production technologies with each other. The promotion of social services, encouraging grain farmers to adopt social services, and reducing the threshold for grain farmers to participate in social services are also needed.

6.3. Limitations

This study has the following limitations. First, we use the questionnaire method to measure farmers’ risk preferences. This is time-efficient and easy to understand but is influenced by farmers’ self-awareness and subjective attitude, which may cause some deviation. Future studies must consider using more accurate experimental economics methods to measure risk preferences. Second, farmers’ LCAT adoption is constantly changing. This study only uses cross-sectional data; hence, future research can consider using dynamic panel data.