The Impact of Digital Literacy on Farmers’ Green Production Behavior: Mediating Effects Based on Ecological Cognition

Abstract

Farmers’ green production behavior is one of the main determinants of the sustainability of the agricultural economy. In this study, Ordered Logit, OLS, and 2SLS models were conducted to evaluate the impact of digital literacy on farmers’ green production behavior. On this basis, the Propensity Score Matching (PSM) method was conducted to deal with the endogeneity bias that may result from the sample self-selection problem. We also adopt the mediation effect model to test the mediating mechanism of ecological cognition between digital literacy and farmers’ green production behavior. The results showed that three different types of digital literacy significantly improved farmers’ green production behavior. We also found that farmers’ green production behavior improved by 19.87%, 15.92%, and 24.16% through digital learning, social, and transaction literacy. Meanwhile, the mediating effect showed that digital literacy improves farmers’ green production behavior by increasing ecological cognition. We demonstrate that three different types of digital literacy significantly improved farmers’ green production behavior. Therefore, policies to increase digital literacy among farmers should be further improved to promote farmers’ green production behavior.

1. Introduction

Vegetable farming is an important part of China’s agricultural economic structure. However, the carbon emissions, ecological pollution, and safety of agricultural products brought about by the vegetable growing process cannot be ignored. According to relevant data from 2016, the weighted average carbon emissions from vegetable cultivation in China amounted to 6244 kg CO₂ equivalent per hectare [1]. Similarly, non-compliant pesticide application practices can lead to ecological degradation and physical health threats [2]. Relevant studies have shown that there is a discrepancy between farmers’ willingness to adopt green production technologies and their behaviors [3]. In 2021, the rate of adoption of crop pest control technologies is 46 percent [4]. It follows that farmers need more intrinsic endowments to motivate them to adopt green production techniques. In China, ecological civilization construction is an important element of the rural revitalization strategy [5]. To further promote the green and low-carbon transformation and high-quality development of the vegetable farming industry, China issued a policy document in the 14th Five-Year Plan for the Modernization of Agriculture and Rural Areas in 2021. The document further standardizes the green production behavior of vegetable cultivation by farmers and also regulates intelligently agricultural products, organic green food, resource utilization, and brand management.

Smart agriculture, digital agriculture, and Precision Agriculture are growing faster and faster. The use of remote sensing satellites and the Global Positioning System (GPS) allows for the visualization of targeted farmland [6,7]. IoT devices with sensor technology can build on the former to monitor and collect farm data in real-time [8]. It saves a lot of labor, time, and material resources, and improves the accuracy of soil temperature, moisture, and data [9]. IoT application technologies in agriculture can help to increase the quantity and quality of green production and improve resource utilization [10]. Drones or robots use artificial intelligence technology to accomplish actions such as weeding and drug spraying during agricultural production. It accelerates the automation of green production and increases productivity [11]. With the improvement in digital village infrastructure, Internet technology drives rapid economic growth. According to the China Internet Information Center, as of December 2022, there were 308 million rural Internet users [12]. Internet technology has become an important information vehicle for farmers to entertain, socialize, learn, grow vegetables, and sell agricultural products [13]. The Internet is embedded in the farmers’ daily lives, promoting the continuous improvement in their digital literacy [14].

Most of the studies related to influencing the green production behavior of farmers have examined the following aspects. Firstly, there is the effect of the external environment on farmers’ green production behavior. With the implementation of regulatory policies for green production in agriculture, most studies focus on the impact of environmental regulations on farmers’ green production behavior. This study shows that guided environmental regulation has a significant effect on farmers’ low-carbon agricultural production behavior. Publicizing the greenhouse effect reduces the cost of information-seeking for farmers, thus promoting farmers’ adoption of low-carbon production technologies [15]. Studies have shown that government technical training for food security promotes low-carbon agricultural management behaviors among farmers, especially for soil testing and formulated fertilizer application techniques [16]. Dual government subsidies for green producers and consumers reduce the costs and help to promote green production behaviors [17,18]. Environmental regulation helps to adjust farmers’ risk perceptions and thus increases their willingness to engage in green production [19].

Secondly, there is the effect of production characteristics on farmers’ green production behavior. Adoption costs discourage small-scale farmers from adopting green production technologies when planting size is below a standard value [20]. The scale effect of higher cultivation scale will reduce the degree of land fragmentation, reduce the cost of technology, and thus increase farmers’ conservation tillage behavior [21]. Relevant scholars introduced whether farmers join cooperatives based on the above research findings. Farmers’ cooperative membership is more helpful in positively moderating the effect of farmers’ perceived value on green production behavior. Membership provides access to productive agricultural services, and technical training provided by cooperatives, thus increasing the probability of adopting Green Prevention and Control technologies [22,23].

Finally, there is the effect of individual characteristics on farmers’ green production behavior. Low-carbon vegetable cultivation requires advanced technology, large capital, and human resources [24]. According to the theory of perceived value proposed by Grönroos [25], farmers will prefer to adopt sustainable agricultural production technologies when they perceive that the economic, social, and environmental values outweigh the costs. Otherwise, they do not adopt [26]. This phenomenon also suggests that large-scale and intensive vegetable cultivation can help reduce the investment costs required for farmers to adopt green production technologies [27]. At the same time, farmers also choose whether or not to adopt environmentally friendly behaviors by perceiving the usefulness and ease of use. Risk perception is an important factor influencing farmers’ green production behavior [28,29]. A study using a sample of farmers growing green vegetables in Southeast Asia found that the higher their perception of pesticide health risks, the more they would reduce pesticide applications [30]. Biological control using predatory insects will reduce reliance on traditional insecticides. Some scholars have also used the Theory of Planned Behavior that incorporates risk perception to conclude that risk perception affects farmers’ safe fertilizer use behavior [31]. Farmers’ knowledge about breeding, pesticide, and fertilizer use can help promote green production. For example, pheromones can also be used to confuse pests about their destructive activities on vegetables. In addition to this are treated seeds, pesticide use intervals, types of prohibited pesticides, etc. [32,33,34]. Subjective norms will help farmers adopt green production techniques [35]. Excessive use of pesticides and fertilizers will lead to increased nitrogen content in the soil, affecting the composition of nitrite in vegetables and ultimately affecting the safety of vegetables [36]. Proper disposal of pesticide and fertilizer packaging helps prevent residual toxins from polluting the water and soil environment [37]. The information gap limits farmers’ access to markets, technology, etc., and production decisions [38,39]. Some scholars have shown that Information and Communication Technology reduces the cost of searching for information and promotes Green Prevention and Control adoption by farmers [40]. The technology mitigates information asymmetry. Farmers have unrestricted access to real-time information related to Green Prevention and Control technology, which promotes green behavior [41,42,43,44]. Digital literacy is the ability of farmers to learn, live, and work efficiently using Internet technologies in the context of the digital era [45]. Digital literacy has been created due to the widespread popularity of the Internet. Increased digital literacy promotes the use of digital information and technologies by farmers to optimize livelihoods and increase farm incomes [46]. Creativity in digital literacy can reduce the learning cost of green production knowledge for farmers [47]. The optimization of production factor inputs improves agricultural green efficiency [48,49]. Digitally literate farmers use communication programs to build a social networking system. Farmers can discuss the technical applications of agricultural green production online [50]. In addition, farmers with high digital literacy can access information and establish market connections on the Internet. With the help of Taobao and Jingdong APP online platforms, they can sell green agricultural products and thus gain revenue [51].

As a result of the above analysis, we found limitations in the existing literature. Firstly, in the context of the rapid development of Internet information technology, research related to the use of information by farmers on the adoption of green behavior remains scarce. Secondly, the ecological cognition required for farmers’ green production behavior comes from many aspects such as knowledge acquisition, technical specifications, value perception, and risk perception, not only health risks, environment, and value perception. Thirdly, aforementioned studies focus on a single dimension of digital literacy. There is a lack of the literature that examines digital learning literacy, digital social literacy, and digital transactional literacy within the same framework. Finally, green production practices are a model for the sustainable development of agricultural technology. It runs through every link of pre-production, production, and post-production, including pre-production soil tilling techniques, chemical fertilizer utilization during production, pest control and prevention, and the resource utilization of post-production [52]. To accurately define green production, every link of pre-production, production, and post-production should be considered.

To fill the gaps in the existing research, the following question is posed in this study. Can increased digital literacy promote green production behavior among farmers?

Given the above findings, we aimed to explore the impact of digital literacy on farmers’ green production behavior. To achieve this research objective, we used a sample of vegetable farmers in Liaoning, China, to estimate digital literacy and farmers’ green production behavior. On this basis, we explore the process where the former acts on the latter. The findings of this study revealed that digital learning literacy, digital social literacy, and digital transaction literacy all have an impact on farmers’ green production behavior. The three types of digital literacy enhance farmers’ green production behavior through ecological cognition.

2. Theoretical Analysis and Research Hypothesis

Internet use provides a channel for the diffusion of green production technologies. Farmers have improved their digital learning literacy related to green production by increasing the frequency of Internet use. Digital learning literacy is the acquisition of a certain level of knowledge and standardized practices in low-carbon agriculture by farmers. The advantages of low information search and payment costs are favored by farmers. Farmers can access knowledge and operating techniques regardless of location and time constraints. Optical, inductive, and ultrasonic sensors sense and transmit signal status within the monitoring system. It helps farmers to monitor crop growth [53]. Farmers use drones to collect scanned images and data from their fields. The digital information fed back can help them make green production decisions [54]. Farm management computer software can be manipulated online to meet plant production needs, such as spraying pesticides, applying moderate amounts of fertilizer, and switching on and off insulation equipment [55,56]. The technical controls for smart production and harvesting need to be set up appropriately by the farmer within the system [57]. Moreover, the knowledge is in a wide range of categories. Green farming experience in all stages of production, including seeding, plowing, and harvesting, can be learned through audio, video, live streaming, and many other ways [58]. For example, biological control techniques, pheromone broadcasting, and placement of traps are alternatives to traditional highly polluting pesticides. Being abreast of news and national policies related to high hazards and pollution can be kept up to date by farmers as well. It can be found that the Internet digital knowledge can be transmitted to farmers on time to break the information occlusion of farmers, thus promoting the endogenous motivation of farmers for green production [59].

Hypothesis 1. 

Digital learning literacy positively influences farmers’ green production behavior.

Internet use provides a channel for social communication among farmers. Farmers have improved their digital social literacy related to green production by increasing the frequency of Internet use. Digital social literacy is the mastery of certain social relationships and interpersonal networks as a sign of social integration. Learning planting experience with agricultural low-carbon planting experts and consulting green production-related issues can be realized. Examples are standardized predatory insect release frequency, release intervals, and use dosages. The most effective insecticidal chemicals have low environmental pollution. Various problems encountered can also be solved through social software. This is manifested in the knowledge management and sharing of social media such as blogs and Facebook [60]. It can also manifest itself in the cross-platform messaging of WhatsApp [61]. WhatsApp caters to farmers to communicate with planting experts on technical operations for green production planting [62], such as variable application rates of pesticides and water. Easy access to information and communication technologies facilitates the provision of green production advice to farmers [63]. Online exchange of green planting and learning experiences with friends and family can also break through geographical restrictions. It is possible to expand online socialization opportunities and receive more support and assistance from online users for green production and planting [64]. It can be found that the social integration of the Internet can broaden farmers’ social networks, increase networking resources, and improve the quality of relationships. The use of the Internet can help farmers change the way they used to communicate offline. It can save more time and promote the standardized operation of farmers’ green production [65,66].

Hypothesis 2. 

Digital social literacy positively influences farmers’ green production behavior.

Internet use provides a channel for farmers to sell their agricultural products. Farmers have improved their digital transaction literacy related to green production by increasing the frequency of Internet use. Digital transaction literacy applies online platforms to buy and sell agricultural products, such as pesticides, fertilizers, and other chemical products, from regular stores online. It benefits from the integrated development of Internet technology, logistics, and information flow for farmers to purchase online products needed for green production. The express transportation service saves transportation costs for farmers and enables them to obtain agricultural products purchased within the agreed time [67]. Similarly, farmers use online sales platforms, facilitated payment methods, and promotion of agricultural products to increase sales. In recent years, the popularity of short live video streaming has been supported by digital media technology [68]. Data suggest that 5G technology has boosted the marketing volume of agricultural products [69]. This online sales model increases the possibility of real-time interaction between farmers and consumers. It brings retailers closer to consumers in a shorter time [70,71]. This can fulfill the consumer’s perception and consumption experience of green agricultural products [72]. In addition, farmers can utilize digital multimedia technology to open online stores to run their farms [73]. As consumer preference for green and healthy food increases, online sales platforms have expanded the new market for green agricultural products. It can be seen that the sales model of the e-commerce platform can help farmers to change their identity of “buying” and “selling”. It can satisfy farmers’ purchasing demand as well as realize farmers’ selling demand [74]. According to the theory of producer behavior, to meet the market’s green consumption preference, farmers will choose green production behavior [75].

Hypothesis 3. 

Digital transaction literacy positively influences farmers’ green production behavior.

According to the principles of information science and the theory of knowledge, information, and action, farmers obtain, process, and handle various types of information on production, marketing, program use, and application techniques. At the same time, they develop new perceptions of environmental alternatives in pest management based on processed knowledge [76]. This will promote farmers’ adoption of green production behaviors. Online planting experts, friends, netizens, and farmers interact with each other on issues related to green production. According to the social learning theory, farmers’ social learning behaviors emerge through observation and communication during interactions. They scientifically master the operational norms of green production technology. The process deepens farmers’ green ecological cognition and promotes behavioral change [77]. According to the perceived value theory, farmers utilize e-commerce platforms for online sales. The process allows farmers to perceive the economic benefits and low operating costs associated with online platform network sales. When the benefits outweigh the costs, farmers will maintain a more positive ecological perception and product sales attitudes and thus will be more likely to adopt green production technologies [78].

Hypothesis 4. 

Digital learning literacy promotes farmers’ green production behavior by enhancing ecological cognition. Digital social literacy promotes farmers’ green production behavior by enhancing ecological cognition. Digital transaction literacy promotes farmers’ green production behavior by enhancing ecological cognition. That is, farmers’ ecological cognition mediates the effect of digital literacy on farmers’ green production behavior.

This paper integrates digital literacy, ecological cognition, and green production behavior into an analytical framework (Figure 1). The definitions of the variables included in the theory model are described in the 3rd table in Section 3.2.

3. Materials and Methods

3.1. Data Source

Vegetable production is a concern in Northeast China, where the cold period is long. Liaoning Province has a temperate continental monsoon climate [79]. The production of greenhouse vegetables in Liaoning Province reached 19.726 million tons. Liaoning is representative of samples in the Northeast [80]. We ensured that the sample was drawn to reflect the whole and to reflect all groups. A stratified random sampling method was used in this study [81]. The data for this study came from a questionnaire survey of farmers in Liaoning Province, China, conducted in 2022. Based on a full consideration of geographic location, economic level, and scale of vegetable cultivation, we used stratified and random sampling methods to select Shenyang City, Liaoyang City, Dandong City, and Panjin City as the prefecture-level cities. Then, we randomly selected 3 counties within each prefecture-level city, 3 townships within each county, 3 villages within each township, and 10–15 vegetable farmers. A total of 950 questionnaires were distributed in this study and finally, 884 valid questionnaires were obtained.

3.2. Variable Selection

The dependent variable is farmers’ green production behavior. We drew on relevant research on digital literacy and ultimately assigned values based on the cumulative extent of the use of green agricultural technologies through every link of pre-production, production, and post-production [82]. The mediating variable was ecological cognition. We referred to the indicator system for ecological cognition developed by Xuezhen Xu [83]. Our revision of the scale is shown in

Table 1. Ecological Cognition Measurement Scale.

Variable	Definition	Value
Hazard Cognition	Failure to treat the seed before sowing increases the risk of microbial damage.	Strongly Disagree = 1, Disagree = 2, Fairly = 3, Agree = 4, Strongly Agree = 5
	Insecticides penetrate the tissue and remain on the vegetables.
	Excessive use of chemical fertilizers can lead to higher nitrate content in vegetables.
	Non-compliant use of pesticides and fertilizers can leave residues on vegetables that are harmful to human health.
	Non-compliant use of pesticides and fertilizers can pollute the ecosystem.
	No use of banned pesticides, or fertilizers.
	Waste chemicals can pollute the ecosystem if not disposed of properly.
Behavioral Cognition	I am willing to choose a regular farm store to buy insecticides.
	I am willing to accept the technical specifications of the agricultural technician.
	I would choose to regulate the application of pesticides and fertilizers according to the instructions.
	I am willing to enforce strict safety intervals for pesticides.
	I am willing to aseptically treat the seeds before planting.
	I am willing to use water-saving irrigation techniques, and pest and disease biological control techniques.
	I am willing to properly dispose of discarded pesticide and fertilizer packaging, etc.
Earnings Cognition	Adoption of green production techniques will reduce ecological pollution.
	Applying chemicals such as pollution-free pesticides can reduce human health hazards.
	Safe use of pesticides and fertilizers is more important than control effectiveness.
	Adoption of green production technologies can help promote low-carbon agriculture.
	Adoption of green production technologies can contribute to carbon neutrality and peak carbon.
	The application of chemicals such as pollution-free pesticides can improve economic incomes.
	The adoption of green production techniques produces safer agricultural products.

.

The independent variables are digital learning literacy, digital social literacy, and digital transaction literacy. Some studies have measured digital literacy in terms of farmers’ online browsing of agricultural production information, and use of apps on smartphones or computers [84]. We drew on the above research findings. Browsing online agricultural information aimed to improve farmers’ digital learning literacy. Using the WeChat program for online communication sought to enhance farmers’ digital social literacy. Using the Taobao program for online promotion aimed to improve farmers’ digital selling literacy. Therefore, this study comprehensively measured digital literacy in terms of digital learning, socializing, and trading. Meanwhile, we referred to Liu’s measurement questions of digital literacy among farmers [85]. We measured digital learning literacy based on the frequency of using the Internet for learning. We measured digital social literacy based on the frequency of using the Internet for socializing. We measured digital transaction literacy based on the frequency of Internet selling. We gave “never”, “rarely”, “sometimes”, “often”, and “very often” with values 1–5.

In this study, the reliability of the Ecological Cognition Measurement Scale was tested using the Internal Reliability (IR) index. The test results are shown in

Table 2. Reliability and validity tests of the Ecological Cognition Assessment Scale.

Variable		Hazard Cognition	Behavioral Cognition	Earnings Cognition
Number of variable items		7	7	7
Cronbach’s alpha coefficient		0.8299	0.8409	0.8829
KMO value		0.836	0.831	0.886
Bartlett’s sphericity test	Chi-square	2225.732	2524.742	3176.121
	Degrees of freedom	21	21	21
	Significance level	0.000	0.000	0.000

, Cronbach’s α coefficients of hazard cognition, behavioral cognition, and earnings cognition are all above 0.8, and the values of the reliability indexes are all greater than the recommended value of 0.7, which indicates the internal consistency, reliability, and stability. The measurement scales of this study are good. On this basis, this study conducted a factor analysis to test the construct validity of the preliminary scales. Most studies were analyzed using KMO values and Bartlett’s spherical test. The results are shown in Table 2; the KMO values of hazard cognition, behavioral cognition, and earnings cognition are all greater than 0.8; and the significance level of Bartlett’s spherical test is 0.000, which rejects Bartlett’s spherical null hypothesis, indicating that the components of the scale constructed in this study are good.

We referred to related studies and added age, gender, education, and membership in a cooperative, and the number of laborers to the control variables of the econometric model. The level of health and education reflects the level of farmers’ human capital. In the TPB model, health is an important factor influencing individual microbehavior [86]. Therefore, this variable was added to this model. The higher the level of education, the higher the environmental awareness of farmers, and the more they have a strong sense of responsibility for green production behavior [87]. Farmers’ digital learning skills increase with age in the first ten years of work. However, after reaching a certain age, digital receptivity decreases with age [88]. Whereas digital literacy affects the green production behavior of farmers, age needs to be included to explore the impact on the green production of farmers. Similarly, the number of laborers has a very important impact on farm performance. The higher the number of laborers, the more it contributes to the development of modern digital agriculture [89]. It has been suggested that women can play a greater role in improving the efficiency of green production in agriculture by utilizing planting techniques. Therefore, in this study, the gender variable was added to explore the impact on green production [90]. Comprehensive farm services influence the tendency of farmers to choose green farm fertilizers [91]. Therefore, this study added the variable of whether farmers use farm machinery services or not. The scale of production affects the cost of adoption of green production technologies in agriculture. In general, large-scale growers are more likely to adopt green production technologies [92]. This variable was added to this study. Farmers joining cooperatives can be trained in green production knowledge and numerical skills, which will promote the adoption of organic fertilizers by farmers [93]. This variable was added to this study. Farmers with political identities are more susceptible to the external political environment, which affects the adoption of green organic fertilizers by farmers [94]. This variable was added to this study. Perceived value and risk are important variables in farmers’ choice to adopt green production technologies [95]. Digitalization is driving high-quality [96] development of the rural economy. The development of the rural economy provides an important development environment for farmers to increase their incomes. The adoption of green production is facilitated by farmers’ perceived benefits from regional development [97]. This variable was added to this study.

Knowledge of green production, access to a market, access to technology for green or digital production, social connections, government subsidies, green farming experience, use of Precision Agriculture technologies, and regulatory controls are important elements that farmers can gain from joining a cooperative [98,99,100,101,102,103,104]. If knowledge of green production techniques, standardized use of green production techniques, regulatory policies, and Internet use are included again, this can lead to duplicate estimates. This is not conducive to accurately estimating the marginal effect of digital literacy on farmers’ green production behavior. In addition, a greater proportion of farmers in our study were involved in cooperatives. Their mastery of digital technology, production knowledge, access to a market, access to technology, government policy subsidies, green farming experience, use of Precision Agriculture technologies, access to information, and use of cell phones and computer equipment all come from cooperatives. Therefore, the above variables were not added separately in this study. Variable selection and sample characteristics are shown in

Table 3. Variable definitions and descriptive statistics.

Variable	Definition	Mean	SD
Green production behavior	Measured based on the cumulative extent of use of green agricultural technologies, aseptic seed treatment, soil-formula fertilization technology, pest and disease biological control technology, water-saving irrigation technology, and waste packaging treatment	2.67	1.08
Digital learning literacy	The frequency of using the Internet for learning. Never = 1, rarely = 2, sometimes = 3, often = 4, and very often = 5	2.08	0.88
Digital social literacy	The frequency of using the Internet for socializing. Never = 1, rarely = 2, sometimes = 3, often = 4, and very often = 5	1.61	0.93
Digital transaction literacy	The frequency of using the Internet for selling. Never = 1, rarely = 2, sometimes = 3, often = 4, and very often = 5	2.72	1.03
Age	Year	55.04	11.27
Gender	Male = 1, female = 0	0.57	0.50
Health	Poor = 1, fair = 2, good = 3, very good = 4, excellent = 5	2.92	1.08
Education	Illiterate = 1, primary school = 2, junior middle school = 3, high school and above = 4	1.95	0.78
Accessibility of villages	Yes = 1, no = 0	0.83	0.37
Political profile	Party members = 1, non-members = 0	0.10	0.30
Number of laborers	Count	4.11	1.76
Village economy	Well below average = 1, below average = 2, average = 3, above average = 4, well above average = 5	2.80	0.99
Use of agricultural machinery	yes = 1, no = 0	0.52	0.50
Perceived risk	Health risk, environmental risk, disaster risk, cultivation risk, and lending risk. Of these risks, how many can be perceived, and what values are assigned	2.65	1.01
Land size	Below 0.33(ha) = 1, 0.33–0.67(ha) = 2, 0.67–1(ha) = 3, 1–1.33(ha) = 4, above 1.33(ha) = 5	2.69	1.31
Membership in cooperatives	Knowledge, normative techniques, social relations, subsidies, and regulation. How many can be acquired, and what values are assigned	3.00	1.18
Ecological cognition	Measured by validated factor analysis	0.00	0.76
Region	Shenyang, yes = 1, no = 0; Liaoyang, yes = 1, no = 0; Panjin, yes = 1, no = 0; Dandong, yes = 1, no = 0

.

3.3. Econometric Model

Previous studies examining the adoption of agricultural innovations use structural equation modeling based on the Theory of Planned Behavior (TPB). In this study, the green production behavior of farmers is ordered continuous data; ordinary least squares (OLS) are more applicable. Therefore, this study drew on relevant studies and used the econometric model of OLS [105]. The regression model in this study is expressed as

$${GPB}_i={\beta }_0+{\beta }_1{digital literacy}_i+{\sum }_{j=1}^n{\beta }_j{Ctrl}_i+{\epsilon }_i$$

(1)

where the outcome variable stands for farmers’ green production behavior, the key explanatory variable indicates digital learning/social/transaction literacy, $Ctrl$ represents a series of control variables, and $\epsilon$ is a random disturbance term. We may be missing some unobservable variables. However, we have considered the impact of unobservable variables on farmers’ green production behavior when designing the econometric model. Our treatment was to put them as disturbance terms in the econometric model [106].

To identify the intermediary process of digital learning/social/transaction literacy in influencing farmers’ green production behavior, we used the mediating effect to test whether digital learning/social/transaction literacy improves farmers’ green production behavior by increasing farmers’ ecological cognition. The mediation model in this study is expressed as

$${GPB}_i={\delta }_1+a_1{digital literacy}_i+{\sum }_{j=1}^n{\gamma }_j{X}_i+{\mu }_1$$

(2)

$${MED}_i={\delta }_2+b_1{digital literacy}_i+{\sum }_{j=1}^n{\gamma }_j{X}_i+{\mu }_2$$

(3)

$${GPB}_i={\delta }_3+c_1{digital literacy}_i+c_2{MED}_i+{\sum }_{j=1}^n{\gamma }_j{X}_i+{\mu }_3$$

(4)

Furthermore, to avoid estimation bias due to reverse causation, we selected the digital literacy of other farmers in the same village as an instrumental variable. We also used a two-stage least squares (2SLS) computational method to determine the IV estimates. On this basis, the Propensity Score Matching (PSM) method was conducted to deal with the endogeneity bias that may result from the sample self-selection problem.

4. Estimation Results and Analysis

4.1. The Baseline Estimation Results

The results of the OLS model are shown in columns (1), (2), and (3) of

Table 4. Results of the baseline estimation and robustness check.

Variable	OLS (1)	OLS (2)	OLS (3)	Ordered Logit (4)	Ordered Logit (5)	Ordered Logit (6)
Digital learning literacy	0.1987 *** (0.0379)			0.4032 *** (0.0923)
Digital social literacy		0.1592 *** (0.0314)			0.3621 *** (0.0715)
Digital transaction literacy			0.2416 *** (0.0485)			0.6262 *** (0.1404)
Education	0.0608 ** (0.0277)	0.0700 ** (0.0281)	0.0732 *** (0.0279)	0.1183 * (0.0631)	0.1240 * (0.0635)	0.1294 ** (0.0632)
Membership in cooperatives	0.1034 *** (0.0376)	0.0846 ** (0.0378)	0.0755 ** (0.0372)	0.2693 *** (0.0872)	0.2390 *** (0.0869)	0.2143 ** (0.0862)
Perceived risk	0.4590 *** (0.0327)	0.4401 *** (0.0332)	0.3682 *** (0.0423)	1.0539 *** (0.0888)	1.0222 *** (0.0874)	0.8103 *** (0.1145)
Village economy	0.1333 *** (0.0277)	0.1389 (0.0280)	0.1238 *** (0.0286)	0.2224 ** (0.1075)	0.2006 * (0.1068)	0.1681 (0.1076)
Age	−0.0024 (0.0029)	−0.0031 (0.0029)	−0.0026 (0.0029)	−0.0065 (0.0065)	−0.0076 (0.0066)	−0.0056 (0.0064)
Gender	−0.0499 (0.0574)	−0.0538 (0.0574)	−0.0661 (0.0567)	−0.1094 (0.1306)	−0.1228 (0.1296)	−0.1521 (0.1285)
Health	0.0269 (0.0324)	0.0277 (0.0323)	−0.0063 (0.0315)	0.0181 (0.0765)	0.0222 (0.0751)	−0.0657 (0.0742)
Political profile	0.0400 (0.0806)	0.0647 (0.0807)	0.0659 (0.0805)	0.0551 (0.1870)	0.1015 (0.1833)	0.1198 (0.1829)
Number of laborers	−0.0208 (0.0167)	−0.0310 * (0.0167)	−0.0253 (0.0166)	−0.0506 (0.0381)	−0.0622 * (0.0374)	−0.0452 (0.0378)
Use of agricultural machinery	0.0619 (0.0574)	0.0596 (0.0577)	0.0783 (0.0572)	0.1897 (0.1311)	0.1873 (0.1303)	0.2289 * (0.1296)
Accessibility of villages	0.0264 (0.0880)	0.0345 (0.0910)	0.0291 (0.0858)	−0.0419 (0.2031)	−0.0582 (0.2046)	−0.0342 (0.1991)
Land size	−0.0115 (0.0421)	−0.0228 (0.0430)	−0.0083 (0.0414)	−0.0441 (0.0986)	−0.0606 (0.1005)	−0.0254 (0.0952)
Constant	0.3368 (0.2695)	0.6219 ** (0.2603)	0.5170 ** (0.2620)	-	-	-
Observations	884	884	884	884	884	884
R2/pseudo R2	0.39	0.39	0.40	0.16	0.16	0.17

Notes: ***/**/* Statistically significant at the 1%/5%/10% level, and values in parentheses are robust standard errors.

. In the OLS model, we added all the control variables and column (1) shows the effect of digital learning literacy on farmers’ green production behavior, with a coefficient of 0.1987, which is significant at the 1% level. Column (2) shows the effect of digital social literacy on farmers’ green production behavior with a coefficient of 0.1592, which is significant at the 1% level. Column (3) shows the effect of digital transaction literacy on farmers’ green production behavior with a coefficient of 0.2416, which is significant at the 1% level.

In addition, it can be found in the three models that perceived risk has a significant positive effect on farmers’ green production behavior, with a coefficient of 45.90%, 44.01%, and 36.82% in each model, respectively. Membership in cooperatives has a significant positive effect on farmers’ green production behavior, which increases by 10.34%, 8.46%, and 7.55%, respectively. The level of education has a significant positive effect on farmers’ green production behavior, which increases by 6.08%, 7%, and 7.32% in each model, respectively. Meanwhile, the results of the Ordered Logit model are shown in columns (4), (5), and (6) of Table 4. In the Ordered Logit model, we added all the control variables, and columns (4), (5), and (6) show the effect of digital learning literacy, digital social literacy, and digital transaction literacy on farmers’ green production behavior, which is significant at the 1% level. The estimation results of the Ordered Logit model are a robustness check of the OLS model’s estimation results. The test proves that the estimates are valid.

4.2. Main Results Based on IV Regression

There may be mutual causality between digital literacy and farmers’ green production behavior. This leads to the problem of endogeneity. In this regard, this study draws on the findings of Yu et al. and selects the digital literacy of other farmers in the same village as an instrumental variable [107]. We use a two-stage least squares approach to study the effect of digital literacy on green adoption behavior after controlling endogeneity. A weak instrumental variable test is conducted for the instrumental variables to ensure the validity of the instrumental variable selection and estimation results. The results of the test are shown in

Table 5. Instrumental variable estimation results.

Variable	Model (1) First Stage	Model (1) Second Stage	Model (2) First Stage	Model (2) Second Stage	Model (3) First Stage	Model (3) Second Stage
Digital learning literacy		0.9678 *** (0.2702)
Digital social literacy				1.0368 *** (0.2397)
Digital transaction literacy						0.7835 *** (0.1688)
Digital learning literacy-IV	0.2114 *** (0.0418)
Digital social literacy-IV			0.0968 *** (0.0169)
Digital transaction literacy-IV					0.3391 *** (0.0393)
Control	YES	YES	YES	YES	YES	YES
Constant	1.8999 *** (0.2491)	−1.0483 * (0.5658)	0.5007 ** (0.2559)	0.3985 (0.3404)	1.1178 *** (0.2369)	0.1107 (0.2953)
F-Statistics	23.0400	43.3641	26.1300	30.9648	54.4185	32.1259
p-Value	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000
Observations	884	884	884	884	884	884

Notes: ***/**/* Statistically significant at the 1%/5%/10% level, and values in parentheses are robust standard errors.

.

The test value of the F-statistic is higher than the recommended value of 10 and the corresponding p-value passes significance at the 1% level, indicating that the selected instrumental variables are not weak instrumental variables. The correlation between the instrumental variables and the endogenous explanatory variables is strong. In the first-stage regression of Model 1 in Table 5, we test the relationship between digital learning literacy and digital learning literacy of other farmers in the same village. The results show that the digital learning literacy of other farmers in the same village increased digital learning literacy by 21.14%. In the second-phase regression of Model 1 in Table 5, we test the relationship between digital learning literacy and farmers’ green production behavior. The results show that digital learning literacy increased farmers’ green production behavior by 96.78%.

In the first-stage regression of Model 2 in Table 5, we test the relationship between digital social literacy and digital social literacy of other farmers in the same village. The results show that the digital social literacy of other farmers in the same village increased digital social literacy by 9.68%. In the second-phase regression of Model 2 in Table 5, we test the relationship between digital social literacy and farmers’ green production behavior. The results show that digital social literacy increased farmers’ green production behavior by 1.04%. In the first-stage regression of Model 3 in Table 5, we test the relationship between digital transaction literacy and digital transaction literacy of other farmers in the same village. The results show that the digital transaction literacy of other farmers in the same village increased digital transaction literacy by 33.91%. In the second-phase regression of Model 3 in Table 5, we test the relationship between digital transaction literacy and farmers’ green production behavior. The results show that digital transaction literacy increased farmers’ green production behavior by 78.35%. A comparison of the baseline regression coefficients shows that the estimated coefficients in the 2SLS model have increased, indicating that the impact of digital learning literacy, digital social literacy, and digital transaction literacy on farmers’ green production behavior is underestimated if endogeneity is not considered.

4.3. Robustness Check

In this study, PSM was used to address the possible self-selection problem of the model and to conduct robustness tests for the effect of digital literacy on farmers’ green production behavior. This study conducted a balancing test for the matching results to ensure that the matching results were valid. The test results are shown in

Table 6. Balance of Matching Results Test for the Impact of Digital learning literacy on farmers’ green production behavior.

Variable	Match Stats	Treatment Group	Control Group	Bias (%)	Reducing Bias (%)	T-Value	p-Value
Age	pre-match	54.883	55.125	−2.2	−209.6	−0.31	0.758
	post-match	54.883	54.133	6.7		0.88	0.378
Gender	pre-match	0.562	0.570	−1.6	−251.0	−0.23	0.819
	post-match	0.562	0.590	−5.6		−0.71	0.475
Education	pre-match	1.994	1.921	9.3	95.7	1.34	0.181
	post-match	1.994	1.997	−0.4		−0.05	0.959
Number of laborers	pre-match	3.741	4.318	−33.6	93.0	−4.76	0.000
	post-match	3.741	3.701	2.3		0.31	0.756
Use of agricultural machinery	pre-match	0.565	0.495	14.1	86.8	2.01	0.044
	post-match	0.565	0.556	1.9		0.24	0.813
Accessibility of villages	pre-match	0.827	0.838	−2.8	40.3	−0.40	0.691
	post-match	0.827	0.821	1.7		0.21	0.837
Political profile	pre-match	0.256	0.161	23.6	90.3	3.47	0.001
	post-match	0.256	0.247	2.3		0.27	0.786

,

Table 7. Balance of Matching Results Test for the Impact of Digital social literacy on farmers’ green production behavior.

Variable	Match Stats	Treatment Group	Control Group	Bias (%)	Reducing Bias (%)	T-Value	p-Value
Age	pre-match	54.678	55.373	−6.2	17.8	−0.92	0.360
	post-match	54.763	55.335	−5.1		−0.72	0.469
Gender	pre-match	0.577	0.557	4.0	−4.9	0.60	0.548
	post-match	0.578	0.557	4.2		0.62	0.535
Education	pre-match	1.981	1.917	8.3	89.1	1.24	0.216
	post-match	1.977	1.984	−0.9		−0.13	0.896
Number of laborers	pre-match	4.220	4.000	12.5	85.1	1.86	0.064
	post-match	4.213	4.180	1.9		0.27	0.785
Use of agricultural machinery	pre-match	0.556	0.487	13.9	66.2	2.06	0.040
	post-match	0.555	0.532	4.7		0.69	0.493
Accessibility of villages	pre-match	0.829	0.838	−2.2	−98.0	−0.33	0.741
	post-match	0.829	0.813	4.4		0.62	0.533
Political profile	pre-match	0.217	0.175	10.5	−0.7	1.57	0.117
	post-match	0.215	0.258	−8.6		−1.45	0.148

and

Table 8. Balance of Matching Results Test for the Impact of Digital transaction literacy on farmers’ green production behavior.

Variable	Match Stats	Treatment Group	Control Group	Bias (%)	Reducing Bias (%)	T-Value	p-Value
Age	pre-match	54.873	55.288	−3.7	2.1	−0.53	0.593
	post-match	54.873	54.467	3.6		0.57	0.571
Gender	pre-match	0.575	0.553	4.5	−17.9	0.65	0.518
	post-match	0.575	0.601	−5.3		−0.87	0.386
Education	pre-match	1.972	1.911	7.9	9.0	1.15	0.251
	post-match	1.972	2.028	−7.2		−1.14	0.253
Number of laborers	pre-match	4.048	4.196	−8.4	95.0	−1.22	0.223
	post-match	4.048	4.041	0.4		0.07	0.944
Use of agricultural machinery	pre-match	0.546	0.481	12.9	79.7	1.87	0.062
	post-match	0.546	0.533	2.6		0.43	0.671
Accessibility of villages	pre-match	0.838	0.827	2.9	−207.5	0.42	0.691
	post-match	0.838	0.804	9.0		1.43	0.152
Political profile	pre-match	0.209	0.176	8.3	48.9	1.20	0.231
	post-match	0.209	0.192	4.3		0.69	0.493

. The test results showed that the standardized error of bias of all covariates in the model after matching was below 10%. The t-value of the covariates after matching varied to less than 1.64 and the p-value corresponding to the t-value was greater than 0.1, indicating no systematic difference between the control group and the control group. The test of equilibrium had been passed. Data matching in the PSM model is divided into pre-matching and post-matching. Before data matching, it is normal for variables to fail the significance test. However, after data matching, the variable p-value must pass the significance test. On this basis, the nearest neighbor matching method was further used to estimate the average treatment effect. The results are shown in

Table 9. ATT Total estimation test.

Variable	Farmers’ Green Production Behavior
	ATT	Standard Errors	t-Value
Digital learning literacy	0.392 ***	0.108	3.615
Digital social literacy	0.375 ***	0.092	4.091
Digital transaction literacy	0.832 ***	0.090	9.240

Notes: *** Statistically significant at the 1% level.

. The average treatment effect value corresponding to the t-value is greater than 2.58, both showing that it passed the test of significance at a 1% level. This indicates that farmers’ green production behavior in the treatment group was significantly better than the control group.

4.4. Main Results Based on Mediating Effect Regression

This study further addresses the mechanism of digital literacy on farmers’ green production behavior. The results are shown in

Table 10. Intermediation results.

Variable	Digital Learning Literacy	Digital Social Literacy	Digital Transaction Literacy
Total effect (a1)	0.1987 *** (0.0340)	0.1592 *** (0.0334)	0.2416 *** (0.0350)
Effect of digital literacy on ecological cognition (b1)	0.3988 *** (0.0668)	0.3658 *** (0.0653)	0.4695 *** (0.0688)
Effect of ecological cognition on farmers’ green production behavior (c2)	0.0831 *** (0.0170)	0.0873 *** (0.0171)	0.0772 *** (0.0170)
Direct effect (c1)	0.1655 *** (0.0343)	0.1273 *** (0.0335)	0.2054 *** (0.0355)
Mediating effect (b1 c2)/bootstrap test value	0.0331 *** (0.0088)	0.0319 *** (0.0085)	0.0362 *** (0.0096)
Mediating effect share (b1 c2/a1) (%)	16.6783	20.0589	14.9977

Notes: *** Statistically significant at the 1% level, and values in parentheses are robust standard errors.

. The total effect estimates of the models in Table 10 passed the significance test at a 1% level, indicating agreement with the above findings. Digital learning literacy, digital social literacy, and digital transaction literacy have a significant positive effect on farmers’ green production behavior. The coefficient of digital learning literacy on ecological cognition, b1, and the coefficient of ecological cognition on farmers’ green production behavior, c2, passed the test of significance at the 1% level, respectively, indicating that the mediating mechanism of ecological cognition in the process of digital learning literacy on farmers’ green production behavior is established. The digital learning literacy promotes the increase in farmers’ green production behavior by influencing the ecological cognition of farmers. The coefficient of digital social literacy on ecological cognition, b1, and the coefficient of ecological cognition on farmers’ green production behavior, c2, passed the test of significance at the 1% level, respectively, indicating that the mediating mechanism of ecological cognition in the process of digital social literacy on farmers’ green production behavior is established. The digital social literacy promotes the increase in farmers’ green production behavior by influencing the ecological cognition of farmers. The coefficient of digital transaction literacy on ecological cognition, b1, and the coefficient of ecological cognition on farmers’ green production behavior, c2, passed the test of significance at the 1% level, respectively, indicating that the mediating mechanism of ecological cognition in the process of digital transaction literacy on farmers’ green production behavior is established. The digital transaction literacy promotes the increase in farmers’ green production behavior by influencing the ecological cognition of farmers. By comparing the proportion of the three types, it can be found that digital social literacy and digital learning literacy influence green production behavior more through ecological cognition.

The relationship between total, direct, and mediating effects is a1 = c1 + b1c2. From the perspective of digital learning literacy on farmers’ green production behavior, the total effect coefficient is 0.1987, the direct effect coefficient is 0.1655, and the mediating effect coefficient is 0.0331. The mediating effect share is 16.68%. From the perspective of digital social literacy on farmers’ green production behavior, the total effect coefficient is 0.1592, the direct effect coefficient is 0.1273, and the mediating effect coefficient is 0.0319. The mediating effect share is 20.06%. From the perspective of digital learning literacy on farmers’ green production behavior, the total effect coefficient is 0.2416, the direct effect coefficient is 0.2054, and the mediating effect coefficient is 0.0362. The mediating effect share is 15%.

5. Discussion

In the related research results, Internet usage time significantly contributed to farmers’ adoption of organic fertilizer behavior. The impact coefficients were 0.065 and 0.487, respectively. The Internet alleviated the information asymmetry in rural areas, thus increasing the rate of organic fertilizer adoption by farmers. Digital learning literacy, digital social literacy, and digital marketing literacy positively promoted the green production behavior of farmers in this study, respectively. The impact coefficients were 0.1987, 0.1592, and 0.2416, respectively [108,109]. Farmers’ ecological cognition mediates the effect of digital literacy on farmers’ green production behavior. The test of mediating mechanisms showed that digital learning, social, and transactional literacy contribute to farmers’ green production behaviors through enhanced ecological cognition. This finding was further validated in a related study. In an apple growers’ fieldwork study, farmers’ cognition also mediates the effect of digital technology on farmers’ fertilizer reduction and efficiency. The adoption of digital technology also had a direct effect on the reduction and efficiency of fertilizer. The study reaffirms that more attention should be paid to farmers’ Internet use, especially digital technology adoption [110]. Data matching in the PSM model is divided into pre-matching and post-matching. Before data matching, it is normal for variables to fail the significance test. However, after data matching, the variable p-value must pass the significance test. From the mechanism results, we can find that the Internet digital knowledge can be transmitted to farmers on time to break the information occlusion of farmers, thus promoting the endogenous motivation of farmers for green production. The use of the Internet can help farmers change the way they used to communicate offline. It can save more time and promote the standardized operation of farmers’ green production. It can also satisfy farmers’ purchasing demand and realize farmers’ selling demand. Farmers will choose green production behavior [111]. Compared with the previous research literature [112], the difference is that the existing literature pays more attention to the influence of individual characteristics of farmers, production characteristics, and external environment on farmers’ green production behavior. The interactive effects of digital literacy and farmers’ green production behavior are neglected. In contrast, this study highlights digital literacy as an important factor influencing green production behavior. This study fundamentally mobilized the endogenous motivation of farmers to participate in green agricultural production. The findings can provide inspiration for future studies on the green development of agriculture in more countries and regions. This is conducive to providing policy recommendations for green agricultural development.

There are limitations in our study. On the one hand, there are the data. Our survey area is limited to Liaoning Province. It would be more convincing to have samples from more regions. On the other hand, there is a lack of relevant exploration of Internet technology options for farmers. In the future, this study will conduct research on technologies related to precision and digital agriculture. It is not enough that our findings apply only to vegetable cultivation. Future research should also be applicable in areas such as cereal cultivation. We will overcome these limitations in our future work.

6. Conclusions

This study analyzes the impact of digital literacy on farmers’ green production behaviors, based on data from 884 farm household surveys in Liaoning Province, China. The conclusions drawn from this study are as follows: Firstly, digital learning literacy, digital social literacy, and digital transaction literacy all have a significant positive impact on farmers’ green production behavior. Secondly, risk perception, rural economy, the degree of farmers’ membership in cooperatives, and education level also have significant effects on farmers’ green production behavior. Finally, farmers’ ecological cognition mediates the effect of digital literacy on farmers’ green production behavior.

Based on the findings of this paper, the following insights can be drawn: On the one hand, local governments enhance the digital literacy of agricultural households in green production in various ways. Local governments develop digital literacy improvement programs. Local governments should contract smart agriculture expert consulting services to guide farmers on standardized production. On the other hand, local governments promote the construction of rural digital infrastructure to provide technological empowerment for low-carbon agriculture. Relying on new technologies such as 5G and artificial intelligence, we can further improve the system functions of online marketing, real-time feedback, online interaction, and live broadcasting with goods. The government has further increased its financial investment in science, technology, and innovation. They aim to provide important support for rural e-commerce development.