Study on the Effect of Digital Technology Adoption and Farmers’ Cognition on Fertilizer Reduction and Efficiency Improvement Behavior

Abstract

Based on the field survey data for 1216 apple growers in three provinces from China’s Loess Plateau Region, this paper adopted Multivariate Probit (MVP) model and intermediary effect model to analyze the influence mechanisms of digital technology adoption (DTA) on farmers’ choices of Fertilizer-reduction and Efficiency-improving Technologies (FETs) from the perspective of farmers’ cognition (FC). The correlation effects of three typical FETs of soil testing and formula fertilization, integrated water-fertilizer fertilization and slow/controlled release fertilizer, were tested empirically. The results showed that DTA could significantly affect the adoption of FETs by apple growers, and FC played a partial mediating role, and there were complementary effects among the three typical FETs. The results were further confirmed by the propensity score matching (PSM) robustness test and the instrumental variable (IV) endogeneity test conducted in the conditional mixed-process (CMP). The results of our heterogeneity analysis showed that the promotion effect of DTA on FETs in the group with junior high school education or below was more significant than that in the group with higher education, and the promotion effect was more significant in the group with large operation scale. Therefore, the improvement in Internet penetration should not be taken as the ultimate goal, but more attention should be paid to farmers’ mastery and effective use of DTA, promote the improvement in farmers’ cognitive level, and implement the “bundled” publicity and guidance strategy of FETs adoption, so as to help in the green transformation of agriculture.

1. Introduction

As China enters a new phase of building a socialist modern country, modernizing agriculture has become a priority. Agriculture is the foundation of the country and the key to strengthening it. However, in recent years, the excessive use of chemical fertilizers has endangered the quality and safety of agricultural products, human health, and the quality of arable land. It has hindered the increase in farmers’ income levels and created a vicious cycle of excessive reliance on chemical fertilizers due to low returns [1,2]. Therefore, reducing chemical fertilizer use and improving its use efficiency are both crucial for achieving agricultural modernization and improving farmers’ livelihoods. Agriculture needs to transform and upgrade, and agricultural production and management entities also need to transform and progress and keep pace with the times. The 2022 Central No. 1 Document emphasizes the importance of promoting digital rural construction, improving the digital literacy of farmers, and providing them with training in information skills. In 2022, the Ministry of Agriculture and Rural Affairs released the Action Plan for Reducing Fertilizer Use by 2025, which highlights the significance of leveraging the Internet and other media platforms to promote and facilitate the adoption of Fertilizer-reduction and Efficiency-improving Technologies (FETs). Farmers are the ultimate decision-makers in agricultural production, and their internal factors ultimately determine the effects of external elements. The role of farmers’ cognition (FC) in decision-making cannot be overemphasized, given its profound influence on behavior and policy choices. Environmental factors deeply influence FC, which continually evolves through internal assimilation [3,4,5]. In the current era of widespread Internet access and rapid evolution of information infrastructure, it is of paramount reference and practical significance to explore whether digital technology adoption (DTA) can enhance farmer’s choice of FETs by improving their cognition. It is crucial for promoting green transformations in agriculture and fostering the enhancement of farmers’ literacy.

FETs involves techniques such as soil testing and formula fertilization to increase fertilizer utilization without affecting crop growth. It also includes optimizing fertilization structure, increasing the use of organic and new types of fertilizers, and improving fertilization methods to increase the utilization rate of water and fertilizer integration. These efforts promote the improvement in crop quality and yield, as well as the ecological protection of the agricultural environment. Microscopic research on reducing fertilizer usage and increasing efficiency has mainly focused on field crops such as rice and land resource conservation [6,7]. Regarding the study of fertilizer-reduction behavior, some scholars have made judgments by asking respondents questions such as “whether they have reduced fertilizer usage” [8], while others have measured it by the average amount of fertilizer used per mu and the cost involved [9,10]. However, these indicators cannot fully encompass the connotation of reducing fertilizer usage and increasing its use efficiency. Research on the factors influencing fertilizer application mainly focuses on individual farmer characteristics (such as age, labor availability, agricultural income, education level [11], knowledge of scientific fertilization, and risk perception [12]), land resources (such as cultivated land area and degree of fragmentation [13]), social and technological factors (such as technical training [14] and subsidies for organic fertilizers [2,15]), and market conditions (such as the degree of marketability of agricultural products and price compensation mechanisms [16]). There have been limited studies on how the Internet affects the use of fertilizers. Li et al. (2022) conducted research on digital promotion by examining the use of smartphones and found that it significantly promotes the adoption of conservation agricultural practice by farmers [17]. Some scholars have discovered that when farmers are presented with a vast amount of information, they tend to internalize only the information relevant to their needs and personal subjective judgment. And in turn, the information would affect their production and lifestyle decisions [18]. DTA has a dual effect of providing necessary information while also creating confusion. Different individuals interpret information differently. Therefore, how will FC be affected when they use the Internet to obtain information? And how will this impact their choices of FETs?

After reviewing the literature mentioned above, it has been determined that there is still room for further discussion in the following areas. Firstly, current studies on fertilizer application focus more on farmers’ implementation of fertilizer reduction, rather than the efficiency of such reduction. However, it is only when fertilizer reduction does not compromise efficiency that farmers can achieve long-term transformation in their fertilizer application methods. Secondly, there is limited empirical evidence on the impact of DTA on agricultural production by enhancing FC. FC encompasses multiple dimensions, including risk perception, and therefore, it is not comprehensive to generalize FC based on a single dimension. Therefore, this paper aims to explore practical and feasible solutions to fully mobilize the enthusiasm of apple growers for green production by evaluating their choices of three typical FETs, including soil testing and formula fertilization, integrated water-fertilizer fertilization, and slow/controlled fertilization, based on the interpretation of FETs from the Action Plan for Fertilizer Reduction by 2025 and the actual production situation of apple growers. Furthermore, this paper theoretically analyzes the relationship between DTA, FC, and farmers’ choices of FETs. Considering the possible interaction between different technology adoption, the Multivariate Probit model (MVP) is adopted to empirically clarify the realistic logic of the influence among the three. The paper also addresses how to make the external positive factors, such as the popularization of the Internet, play a full role through the internal causes of apple growers, such as the improvement in their cognition (FC).

2. Theoretical Framework and Hypothesis

2.1. Impact Mechanism of Digital Technology Adoption (DTA) on Farmer’s Choices of Fertilizer-Reduction and Efficiency-Improving Technologies (FETs)

According to Shannon’s information theory, information has the ability to reduce randomness and uncertainty, making it a crucial and essential component of effective management [19]. This concept is particularly relevant in enterprise management, where the success of management is closely tied to the collection, integration, and application of information [20,21]. Similarly, farmers’ agricultural production is essentially a process of managing limited resources to maximize individual benefits. Therefore, the importance of information in agricultural production should not be underestimated. However, rural households in rural society often have a low level of education, which puts them at a disadvantage when it comes to accessing information. Additionally, their economic capacity is often weak, and their ability to manage risks is poor. As a result, when rural households encounter unfamiliar behaviors, they may struggle to develop positive psychological expectations, which can lead to resistance towards those behaviors. The development of digital rural construction has significantly increased the Internet penetration rate in rural areas. As of December 2023, the Internet penetration rate in rural areas has reached 66.5%, with 326 million rural Internet users (53rd China Rural Internet Survey Report, China Internet Network Information Center (CNNIC) https://www.cnnic.net.cn/NMediaFile/2024/0325/MAIN1711355296414FIQ9XKZV63.pdf, accessed on 22 March 2024). The rapid expansion of the Internet has greatly impacted farmers’ consumption, entertainment, production, and daily lives. In terms of agricultural production, digital technology can provide farmers with more information channels, effectively addressing the issue of limited information access that rural residents have faced for a long time, who rely solely on traditional channels such as brigade broadcasts and television news. Additionally, it can reduce the problem of farmers’ low level of green production behavior due to asymmetric information. Based on the above analysis, this paper proposes Hypothesis 1:

H1: 

The adoption of digital technology (DTA) can positively influence farmers’ choices of FETs.

2.2. Impact Mechanism of Farmers’ Cognition (FC) on their Choices of FETs

Cognitive behavioral theory is an extension of both cognitive theory and behavioral theory. It suggests that individuals form specific perceptions after gathering and filtering information [22,23,24]. These perceptions guide their decision-making and ultimately determine their behavior. This theory has been further proved by some scholars, e.g., Shikuku (2019) conducted research to test the effect of information exchange links on farmers’ cognition and technology adoption and found that information exchange links had a positive effect on both [25]. Farmers’ cognition includes ecology cognition, market cognition, risk cognition, and technology cognition, with each aspect of cognition further affecting their behavior. In this paper, the mechanism of FC affecting FETs is mainly carried out from the following four aspects. Firstly, ecological economics indicates that the maximization of the benefits that individuals pursue as rational people is not simply the maximization of economic benefits, but the coordination and unification of economic interests, ecological benefits, and social interests, and maximizes intragenerational and intergenerational benefits, that is, the coordination and unification of individuals’ current interests and those of future generations [26]. Therefore, according to the “ecological person” hypothesis of ecological economics, apple growers, as members of the ecological economy system, will make optimal production decisions based on comprehensive consideration of the ecological environment and the reality of resources from the perspective of long-term interests and group interests, on the premise of not harming the welfare of surrounding individuals and future generations. Secondly, Schultz’s theory of rational peasant behavior suggests that apple growers are more likely to adjust their production factor allocation and change traditional production methods when they perceive a higher demand for environmentally friendly agricultural products in the apple market and have a positive outlook on the market price of such products [27]. By adopting green production practices to meet market demand, growers can offset production costs and achieve higher profits [28]. In other words, farmers who have a higher awareness of markets are more likely to adopt FETs. Thirdly, risk perception has been defined as an individual’s ability to perceive and control situational uncertainty [29]. Considering the current emphasis on food quality and safety, there is a growing preference in the market for environmentally friendly agricultural products. This trend is likely to push apple growers towards implementing green production practices. Therefore, the greater the awareness of market demand upgrading and associated risks, the more likely growers are to adopt green production strategies. Fourthly, the active gathering of information on green production by apple growers is a significant step in developing their technology cognition [30], which will then impact their overall attitude towards FETs and ultimately determine whether or not they decide to adopt it [23,25]. Based on the above analysis, this paper proposes Hypothesis 2:

H2: 

Farmers’ cognition (FC) can positively influence their choices of FETs.

2.3. Analysis of the Mediating Effect of Farmers’ Cognition (FC)

Heckman (2012) emphasized that a farmer’s cognitive ability is a form of human capital. Apart from being shaped by their innate abilities, a person’s early experiences and external surroundings also play a crucial role in determining their cognitive ability [31]. The extensive adoption of digital technology has significantly expanded the channels through which farmers can access information [32,33,34]. As farmers acquire more information and enhance their ability to accumulate and internalize it, their knowledge level and cognitive perception will be updated, upgraded, and optimally expanded [17,20,35]. This upgrading of farmers’ cognitive systems will assist them in making informed decisions about adopting new practices [17,18,25,36]. In the current policy environment, the government is promoting green agricultural production and providing certain subsidies. At the same time, the market is gradually recognizing the value of green or organic products, leading to higher prices. Additionally, with the widespread use of the Internet, farmers are becoming more aware of the importance of reducing fertilizer usage and improving its use efficiency. As a result, their perceptions and decisions regarding Fertilizer-reduction and Efficiency-improving Technologies are likely to change in response to these environmental factors. Based on the above analysis, the theoretical analysis framework is shown in Figure 1, and this paper proposes Hypothesis 3:

H3: 

Farmers’ cognition (FC) plays a crucial role in mediating the influence of digital technology adoption (DTA) on their choices in FETs.

Figure 1. Theoretical analysis framework.

Figure 1. Theoretical analysis framework.

3. Data and Methods

3.1. Data Source

The data for this study were collected through field research conducted by our research group in Shaanxi, Gansu, and Shanxi provinces from July to August in 2020 and 2021. These provinces are located in the Loess Plateau, which is the ecological fragile region in China. Meanwhile, it is also a major apple-producing region. As of 2022, this region accounts for 58.44% of the country’s apple planting area, with an apple output as high as 21.6 million tons, which is 54.91% of the national total. In addition to the advantage of quantity, the main apple-producing areas in the Loess Plateau actively promote the improvement in apple planting quality, encourage the adoption of new apple cultivation methods and technological innovations, vigorously promote green production technologies such as fertilizer and water film integration, establish apple science and technology demonstration zones, and build high-standard planting bases to effectively promote the upgrading and high-quality development of the apple industry. Therefore, conducting research in the dominant producing areas of the Loess Plateau, where apple production has absolute comparative advantages, is a typical and appropriate choice.

3.2. Data Processing

In this research process, we used a combination of stratified step-by-step sampling and simple random sampling to ensure accuracy and reliability. Firstly, we selected eight main apple-producing cities in three provinces (Xianyang, Baoji, Yan’an, Weinan, Yulin, Yuncheng, Pingliang, and Tianshui) based on the apple planting situation and green production promotion and development situation. Secondly, we selected 2–3 sample towns from each of the three typical counties (districts) based on production conditions and economic development level. Finally, we randomly selected 1–3 villages from each sample town and interviewed 15–20 farmers from each village in a one-on-one face-to-face questioning format. This approach ensures the authenticity and reliability of the information obtained through the questionnaire. The questionnaire covers various aspects such as personal characteristics, family background, input of production factors, orchard output, farmers’ awareness, and green production. A total of 1227 questionnaires were distributed, and after eliminating 11 questionnaires with missing information, the final questionnaire efficiency reached 99.10%. According to the sample data statistics shown in

Table 1. Distribution of frequency of DTA and choices in FETs among sample farmers.

	DTA/FETs	Sample	Relative Frequency
Are you able to use the Internet to obtain agricultural information?	I can use the Internet to obtain information related to agriculture	436	35.86%
How is your family’s adoption of fertilizer reduction and efficiency-enhancing technology?	Not adopted	455	37.42%
	Adopt only 1 type	528	43.42%
	Adopt 2 or more types	233	19.17%

, only 35.86% of farmers in the surveyed area actively use the Internet to obtain agricultural information, which is still far behind the official Internet penetration rate. Therefore, there is a significant potential for improving farmers’ ability to use the Internet to obtain agricultural information. Additionally, the adoption rate of Fertilizer-reduction and Efficiency-improving Technologies among farmers reached 62.58%. However, only 19.17% of farmers adopted more than one sub-technology, indicating a considerable dependence on traditional chemical fertilizers.

3.3. Variables Design

Explained variable. Based on the current situation of technology promotion and adoption among apple growers in the production process, this paper focuses on three resource-saving and efficient fertilization techniques to measure FETs. These techniques include soil testing and formulated fertilization, water-fertilizer integration, and slow/control fertilization.

Core explanatory variable. Currently, the Internet is the primary way farmers engage with digital technology [20]. Given the widespread use of smartphones among farm households during our research, it is clear that not all of these households are using the Internet to effectively access agricultural information. While some may use the Internet for leisure and entertainment activities, such as watching short videos, others may lack the necessary technological literacy to access agricultural information online. Therefore, this paper aims to establish a benchmark for evaluating DTA by assessing whether farm households are equipped with the technological literacy needed to access agricultural information online.

Mediating variable. The intermediary of this paper is the comprehensive score of FC measured by factor analysis. In accordance with existing research [21,22,23], a evaluation system for FC has been established. The KMO value is 0.774, and the Bartlett sphere test is statistically significant at a level of 0.000, indicating that it is suitable for conducting factor analysis, as depicted in

Table 2. Variable description and factor analysis results.

Category	Definition	Mean	Std. Dev.	Factor Analysis Result
				Factor1	Factor2	Factor3	Factor4
Ecology cognition	I am very familiar with the national ecological restoration policy. 1 = Strongly disagree; 2 = Disagree; 3 = General; 4 = Familiar; 5 = Very familiar	2.370	1.174	0.784	0.148	0.141	0.068
	I can identify environmental issues in the surrounding ecosystem. 1 = Strongly disagree; 2 = Disagree; 3 = General; 4 = Consistent; 5 = Very Consistent	2.310	1.231	0.855	0.098	0.095	0.032
	I have mastered the relevant technologies to deal with ecological and environmental issues. 1 = Not at all; 2 = Not mastered; 3 = General; 4 = Mastered; 5 = Fully mastered	1.640	0.952	0.748	0.200	0.178	0.029
Market cognition	I can grasp the trends of the apple market. 1 = Strongly disagree; 2 = Disagree; 3 = General; 4 = Consistent; 5 = Very consistent	2.410	1.211	0.156	0.843	0.035	0.106
	I think the price fluctuations in the apple market are relatively stable. 1 = Strongly disagree; 2 = Disagree; 3 = General; 4 = Consistent; 5 = Very consistent	2.500	1.257	0.181	0.864	0.180	−0.001
	I can grasp the timing of Apple’s sales 1 = Strongly disagree; 2 = Disagree; 3 = General; 4 = Consistent; 5 = Very consistent	2.790	1.242	0.148	0.699	0.402	0.003
Risk cognition	I can control the production cost of green production. 1 = Strongly disagree; 2 = Disagree; 3 = General; 4 = Consistent; 5 = Very consistent	2.720	1.306	0.043	0.267	0.764	−0.021
	I can control the quality of green production fruits. 1 = Strongly disagree; 2 = Disagree; 3 = General; 4 = Consistent; 5 = Very consistent	3.150	1.279	0.150	0.126	0.762	0.134
	When facing a decline in the price of regular apples, I believe that adopting green production can reduce risks. 1 = Strongly disagree; 2 = Disagree; 3 = Neutral; 4 = Agree; 5 = Strongly agree	2.879	0.655	0.287	0.052	0.625	0.238
Technology cognition	I understand green production technologies. 1 = Very unfamiliar; 2 = Unfamiliar; 3 = General; 4 = Familiar; 5 = Very familiar	3.840	1.109	0.057	0.002	0.086	0.896
	I believe that implementing green production for apples will improve efficiency. 1 = Strongly disagree; 2 = Disagree; 3 = Neutral; 4 = Agree; 5 = Strongly agree	3.490	1.305	−0.005	0.052	0.123	0.708
	In order to attract consumers and wholesalers, I am willing to adopt new technologies to improve the quality of fruits. 1 = Strongly unwilling; 2 = Unwilling; 3 = Neutral; 4 = Willing; 5 = Strongly willing	3.800	1.158	0.088	0.045	0.053	0.920

. After factor analysis, four common factors were extracted. Combined with existing research [26,28,29,30], the four common factors were named ecological cognition, market cognition, risk cognition, and technology cognition.

Control variables. Previous studies provide a valuable reference for selecting control variables [24,36,37,38]. Building upon this foundation, the present paper aims to utilize individual, family, and external environmental characteristics of farmers as control variables.

Table 3. Variable design and descriptive statistical analysis.

Variable Category		Definition	Mean	Std. Dev.
Explained variable	FETs	Do you use soil testing and formulated fertilization? Yes = 1; No = 0	0.121	0.326
		Do you use water-fertilizer integration technology? Yes = 1; No = 0	0.571	0.495
		Do you use slow/controlled fertilization? Yes = 1; No = 0	0.162	0.369
Core explanatory variable	DTA	Is it possible to access agricultural information through the Internet? Yes = 1; No = 0	0.359	0.481
Mediating variable	FC	Comprehensive score after factor analysis of each dimension of FC	0.386	0.318
		Common factor 1 score for ecological cognition	0	1
		Common factor 2 score for market cognition	0	1
		Common factor 3 score for risk cognition	0	1
		Common factor 4 score for technical cognition	0	1
Control variables	Individual characteristics	Gender: Male = 1; Female = 0	0.728	0.444
		Age: Age of the head of household	54.663	9.690
		Educational level: Number of years of education received by the household head	7.294	3.390
	Family characteristics	Orchard area: Apple planting area	15.155	52.119
		Non-agricultural employment: Do any family members have non-agricultural employment? Yes = 1; No = 0	0.590	0.492
	Environmental characteristics	Degree of orchard fragmentation: Number of orchard plots owned	3.989	3.218
		Level of government green production promotion: Is there relevant training? Yes = 1; No = 0	0.750	0.435

provides specific variable definitions and a descriptive statistical analysis. As the mediating variable, FC is obtained through a factor analysis of the evaluation index system. This method standardizes each variable in the index system during operation, resulting in a mean of 0 and a standard deviation of 1 for the four common factors obtained through this method.

3.4. Research Methods

If the adoption of various FETs sub-technologies by farmers is independent of each other, then the random error term μ in the adoption decision model of a single sub-technology should be independently and equally distributed, that is, the adoption of information of one sub-technology will not affect the adoption of other sub-technologies, that is, $\mu ~MVN(0,\psi )$ and the covariance matrix $\psi$ can be expressed as

$$\psi =\left[\begin{array}{ccc}1& {\rho }_{12}& {\rho }_{13}\\ {\rho }_{21}& 1& {\rho }_{23}\\ {\rho }_{31}& {\rho }_{32}& 1\end{array}\right]$$

The non-diagonal elements represent unobservable links among the random components of multiple Fertilizer-reduction and Efficiency-improving Technologies, and the non-zero values indicate that there is a correlation between the error terms of each latent variable, to determine whether there is a complementary or substitute relationship between different technology choices. Therefore, the Multivariate Probit model (MVP) can be used to jointly estimate multiple fertilizer-reduction technologies and then analyze whether there are differences in the influencing factors of different technology choices. At the same time, it can identify unobtrusive factors affecting the adoption of multiple fertilizer-reduction technologies. The model settings are as follows.

$${FET}_i={\alpha }_{1i}+{\beta }_{1i}DTA+{\gamma }_{1i}X+{ϵ}_{1i}$$

(1)

${FET}_i$ represents adoption of three Fertilizer-reduction and Efficiency-improving Technologies (FETs), where $\mathrm{i}$ = 1, 2, 3; $DTA$ indicates the adoption of digital technology by apple growers.

In order to test the mediating role of FC in the influence of DTA on FETs, this paper adopts the stepwise regression method to test the mediating effect. The specific setting of the model is as follows.

$$FC={\alpha }_2+{\beta }_{2}DTA+{\gamma }_{2}X+{ϵ}_2$$

(2)

$${FET}_i={\alpha }_{3i}+{\beta }_{3i}DTA+{\delta }_{i}FC+{\gamma }_{3i}X+{ϵ}_{3i}$$

(3)

$FC$ represents farmer’s cognition, $X$ represents control variables.

4. Results and Analysis

4.1. Main Observations

To assess the impact of DTA and FC on apple growers’ choices of FETs, Equations (1)–(3) were used for regression analysis. The results are presented in

Table 4. Regression analysis on Internet use and farmers’ cognition affecting their choices of FETs.

Variables	Multivariateprobit1			Multivariateprobit2			OLS3
	STFF	WFIT	S/CF	STFF	WFIT	S/CF	FC
DTA	0.302 *** (−0.103)	0.364 *** (−0.08)	0.289 *** (−0.098)	0.190 * (−0.107)	0.319 *** (0.081)	0.208 ** (−0.101)	0.134 *** (−0.028)
FC				0.765 *** (−0.116)	0.418 *** (−0.085)	0.538 *** (−0.106)
Gender	0.151 (−0.21)	−0.207 * (−0.122)	0.095 (−0.215)	0.088 (−0.213)	−0.280 ** (−0.123)	0.086 (−0.215)	0.188 *** (0.043)
Age	0.004 (−0.005)	−0.005 (−0.004)	−0.008 (−0.005)	0.007 (−0.006)	−0.005 (−0.004)	−0.007 (−0.005)	−0.002 (0.001)
Education level	−0.033 ** (−0.015)	−0.017 (−0.012)	−0.003 (−0.015)	−0.028 * (−0.015)	−0.016 (−0.012)	0.000 (−0.015)	−0.003 (0.004)
Orchard area	0.299 *** (−0.062)	0.014 (−0.05)	0.112 * (−0.063)	0.223 *** (−0.064)	−0.036 (−0.051)	0.049 (−0.064)	0.121 *** (0.017)
Non-agricultural employment	−0.009 (−0.103)	−0.022 (−0.077)	−0.001 (−0.097)	−0.048 (−0.107)	−0.031 (−0.078)	−0.022 (−0.099)	0.021 (0.027)
Orchard fragmentation	−0.182 ** (−0.089)	0.068 (−0.064)	−0.106 (−0.088)	−0.168 * (−0.091)	0.068 (−0.064)	−0.09 (−0.089)	−0.002 (0.022)
Government promotion	0.459 *** (−0.142)	0.066 (−0.087)	0.519 *** (−0.132)	0.296 ** (−0.149)	−0.028 (−0.089)	0.421 *** (−0.136)	0.234 *** (0.031)
Whether Shaanxi Province	0.542 ** (−0.239)	0.610 *** (−0.151)	0.887 *** (−0.267)	0.461 * (−0.253)	0.620 *** (−0.152)	0.868 *** (−0.273)	0.000 (0.053)
Whether Shanxi Province	0.661 *** (−0.235)	0.555 *** (−0.146)	1.932 *** (−0.247)	0.693 *** (−0.233)	0.590 *** (−0.147)	1.934 *** (−0.244)	−0.093 * (0.052)
Time-fixed effect	−0.286 (−0.272)	−0.338 * (−0.179)	−0.402 (−0.275)	−0.191 (−0.283)	−0.270 (−0.180)	−0.277 (−0.278)	−0.146 ** (0.062)
Constant	−2.605 *** (−0.515)	0.197 (−0.343)	−2.043 *** (−0.502)	−2.467 *** (−0.535)	0.364 (−0.348)	−1.958 *** (−0.510)	−0.427 *** −0.119
Atrho21	0.238 *** (−0.057)			0.194 *** (−0.058)
Atrho31	0.180 *** (−0.063)			0.118 * (−0.065)
Atrho32	0.308 *** (−0.061)			0.276 *** (−0.061)
Wald chi2(F)	293.44			364.79			26.62
Prob > chi2(F)	0.000			0.000			0.000

Note: STFF, Soil testing and formulated fertilization; WFIT, Water-fertilizer integration technology; S/CF, Slow/controlled fertilization. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively; the values in “()” are the standard error of the regression coefficient.

. Regression 1 examines the relationship between DTA and apple growers’ adoption of FETs. The coefficients of DTA, the key explanatory variable, are all greater than 0 and significant at the 1% level, indicating that after controlling for individual, family, and external environmental factors, DTA has a positive effect on apple growers’ choices in FETs. Therefore, H1 is confirmed. Regression 2 examines the impact of the comprehensive score of FC on apple growers’ choices of FETs. The coefficients of FC are all greater than 0 and significant at the 1% level, indicating that when other variables are held constant, FC has a positive effect on apple growers’ choices of FETs. Therefore, H2 is confirmed.

The analysis of individual characteristics shows that educational attainment has a significant impact on the return of the variable. However, the marginal effect indicates a negative trend, suggesting that farmers with higher education levels are less likely to adopt FETs. This could be due to the fact that they have more livelihood options, with agriculture being only a small part of their daily life, and therefore have little concern for new technologies. On the other hand, the analysis of family characteristics shows that larger orchard sizes are indicative of a closer approach to scale farming practices, leading farmers to opt for FETs. Environmental context features such as the degree of fragmentation within orchards and the extent of government promotion of green production also show a significant impact. The marginal effect of fragmentation is negative, indicating that as the number of blocks within an orchard increases, centralized management becomes increasingly challenging, making it less likely that apple growers would embrace FETs. Conversely, the marginal effect of government promotion of green production is positive, highlighting that when the government places attention on green production, the likelihood of apple growers adopting FETs greatly increases.

The correlation effect of the three sub-technologies can be expressed through the regression coefficients of Atrho in each regression model. Except the regression coefficient of Atrho31 in regression 2, which is statistically significant at a 10% level, the regression coefficients of other Atrho in the regression results are statistically significant at a 1% level, and all coefficients are positive. These results indicate a significant complementary effect between the three typical FETs: soil testing and formulated fertilization, water-fertilizer integration technology, and slow/controlled fertilization. For instance, when apple farmers adopt soil testing and formula fertilization, they are also more likely to adopt integrated water-fertilizer fertilization and slow/controlled fertilization. In other words, the adoption of one kind of FETs by farmers promotes the adoption of the other two sub-technologies. It could be because after adopting one of the FETs, apple growers can compare the costs and benefits with those of traditional fertilization methods. After receiving positive feedback from adopting FETs, their willingness to adopt other FETs significantly increases, and the possibility of adoption also increases. Therefore, the three typical FETs show complementary effects.

4.2. Mediating Effect Examination

The regression coefficient of DTA on FC is 0.134, which is significant at the statistical level of 1%, indicating that the higher the level of DTA, the higher the level of apple farmers’ FC, that is, the positive impact of DTA on apple farmers’ FC, H3, is confirmed. Moreover, it can be seen from Table 4 that the results of regression 2 show that DTA and FC can significantly and positively affect the choices of FETs by apple growers, and the regression coefficient of DTA is not 0. Therefore, combined with the results of regression 1 and 3, it can be seen that DTA can further promote apple growers’ adoption of FETs by improving the level of FC.

4.3. Robustness Tests

4.3.1. Replacement of Mediating Effect Test

To ensure the accuracy of the regression results, this paper employs the Sobel test method instead of the stepwise regression method to examine the mediating effect of FC on the process of DTA that affects apple growers’ FETs. The Sobel test equation is used for this purpose.

$$Z_{\mathrm{i}}={\beta }_4{\delta }_{\mathrm{i}}/S_{{\beta }_4{\delta }_{\mathrm{i}}}$$

(4)

$$S_{{\beta }_4{\delta }_{\mathrm{i}}}=\sqrt{{{\beta }_4}^2{S_{{\delta }_{\mathrm{i}}}}^2+{{\delta }_{\mathrm{i}}}^2{S_{{\beta }_4}}^2}$$

(5)

By applying the Sobel test equation to Equations (4) and (5), we can observe that the Z values for soil formula fertilization, water and fertilizer integrated fertilization, and slow/controlled release fertilizer are 3.87, 3.43, and 3.48, respectively. Additionally, the P values for all three are 0.000. This further confirms that FC plays a mediating role in the impact of DTA on the choices of three types of FETs. This conclusion is consistent with the results obtained from the stepwise regression method, thus ensuring the reliability of our findings.

4.3.2. Counterfactual Estimation of Propensity Score Matching

Whether apple growers use the Internet is a self-selected behavior. To eliminate any bias that self-selection may cause on the regression results, this paper uses the propensity score matching method to construct a counterfactual framework and alleviate the problem of selection bias. This paper will conduct a one-to-one match between apple growers who use DTA and those who do not, ensuring that the two groups are balanced and comparable, and then compare their FETs adoption status. To ensure accurate results, this paper adopts five matching methods, including nearest neighbor matching, radius matching, kernel matching, loacal linear regression matching and Mahalanobis matching to verify the findings. The results are presented in

Table 5. Estimation results of PSM model.

Matching Method	ATT
	STFF	WFIT	S/CF
Nearest neighbor matching	0.172 (0.028)	0.670 *** (0.041)	0.214 (0.033)
Radius matching	0.173 ** (0.022)	0.671 *** (0.031)	0.214 ** (0.024)
Kernel matching	0.173 ** (0.028)	0.671 *** (0.031)	0.215 ** (0.033)
Local linear regression matching	0.173 ** (0.051)	0.673 *** (0.044)	0.215 ** (0.035)
Mahalanobis matching	0.175 ** (0.028)	0.671 ** (0.037)	0.214 (0.029)

Note: STFF, Soil testing and formulated fertilization; WFIT, Water-fertilizer integration technology; S/CF, Slow/controlled fertilization. **, and *** denote statistical significance at the 5%, and 1% levels, respectively; the values in “()” are the standard error of the regression coefficient.

. With the exception of one matching method for soil testing and formula fertilization technology and two matching methods for slow-release fertilizers that failed the significance test, the average treatment effects of the remaining treatment groups (apple growers using DTA) were significantly greater than 0. This indicates that the baseline regression model did not account for the self-selection problem of apple growers and underestimated the role of DTA in promoting their choices of FETs. It also confirms that DTA can positively affect apple growers’ adoption of FETs, making the conclusion robust.

4.4. Endogenous Tests

To avoid endogeneity problems that may arise from omitting unobservable variables or bidirectional causation, which could bias DTA represented by “whether agricultural information can be obtained through the Internet” on the adoption of FETs, this article refers to the study by Yu et al. [39], which used “whether you have online shopping experience” as an instrumental variable. From a correlation perspective, the ability and level of DTA among apple growers are affected by their ability to shop online. From an exogeneity perspective, online shopping experiences have no direct connection to apple growers’ adoption decisions of FETs. Therefore, this instrumental variable meets the correlation and exogeneity assumptions. The instrumental variable in this article is a discrete variable, and IV-Probit can only solve a situation where the instrumental variable is continuous. Therefore, this article uses the conditional mixed-process (CMP) estimation method to conduct a two-stage estimation of the model [40]. This method uses simultaneous likelihood estimation to estimate the DTA model and the adoption decision model of apple growers’ FETs. The specific estimation results are shown in

Table 6. Regression results of endogeneity test of CMP.

Variables	STFF		WFIT		S/CF
	(1)	(2)	(1)	(2)	(1)	(2)
DTA		1.0001 *** (0.379)		1.270 *** (0.250)		1.047 *** (0.350)
Online shopping experience	0.221 *** (0.027)		0.221 *** (0.027)		0.221 *** (0.027)
Control variables	Controlled	Controlled	Controlled	Controlled	Controlled	Controlled
lnsig_2	−0.794 *** (0.020)		−0.794 *** (0.020)		−0.794 *** (0.020)
atanhrho_12	−0.414 * (0.213)		−0.506 *** (0.157)		−0.436 ** (0.200)

Note: STFF, Soil testing and formulated fertilization; WFIT, Water-fertilizer integration technology; S/CF, Slow/controlled fertilization. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively; the values in “()” are the standard error of the regression coefficient.

.

In the estimation results, all atanhrho_12 values are significant at the 1% statistical level, indicating that DTA is an endogenous explanatory variable. Therefore, it is reasonable to use instrumental variables to solve endogenous problems. In the first-stage estimation results, the impact of online shopping experiences on DTA is significant, as evidenced by the significant regression coefficients. This verifies the correlation of instrumental variables. In the second-stage estimation results, the regression coefficients of DTA are all significant at the 1% statistical level. These results are consistent with the regression coefficients of DTA in the baseline regression. Although the direction of the coefficients is the same, the value of the regression coefficient has significantly increased. This indicates that potential endogeneity issues make the benchmark regression model underestimate the impact of DTA on farmers’ adoption of FETs. It is important to note the robustness of the above benchmark regression results.

4.5. Further Discussion

4.5.1. Heterogeneity Analysis Based on Educational Level

This paper categorizes household heads’ education level into three groups: those with primary school education or below, those with junior high school education, and those with high school education or above. The paper then removes education level from the baseline regression model and re-regresses it. The results of the regression analysis are presented in

Table 7. Results of heterogeneity analysis based on education level.

Variables	Multivariateprobit3-Primary School			Multivariateprobit4-Junior High School			Multivariateprobit5-High School
	STFF	WFIT	S/CF	STFF	WFIT	S/CF	STFF	WFIT	S/CF
DTA	0.351 * (−0.183)	0.212 (−0.134)	0.205 (−0.178)	0.097 (−0.161)	0.437 *** (−0.122)	0.294 * (−0.152)	0.117 (0.277)	0.237 (0.197)	−0.019 (0.243)
FC	0.871 *** (−0.194)	0.458 *** (−0.143)	0.602 *** (−0.184)	0.750 *** (−0.176)	0.436 *** (−0.129)	0.495 *** (−0.161)	0.707 ** (0.310)	0.416 ** (0.205)	0.453 * (0.260)
Constant	−1.143 (−0.865)	0.513 (−0.577)	−2.106 ** (−0.834)	−2.151 * (−1.298)	0.667 (−0.845)	5.221 (−2.174)	−4.098 ** (1.622)	−0.289 (0.943)	−0.861 (0.654)
Control variables	controlled	controlled	controlled	controlled	controlled	controlled	controlled	controlled	controlled
Atrho21	0.083 (−0.100)			0.092 (−0.086)			0.285 ** (0.144)
Atrho31	0.032 (−0.132)			0.142 (−0.112)			0.418 *** −0.16
Atrho32	0.269 ** (−0.109)			0.285 *** (−0.091)			0.418 *** (0.160)
Wald chi2	146.70			183.62			59.49
Prob > chi2	0.000			0.000			0.004
N	441			561			214

Note: Primary school, Primary school education and below; Junior high school, Junior high school education; High school, High school education or above. STFF, Soil testing and formulated fertilization; WFIT, Water-fertilizer integration technology; S/CF, Slow/controlled fertilization. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively; the values in “()” are the standard error of the regression coefficient.

. The findings indicate that DTA has a significant positive effect on the adoption behavior of apple growers in the primary school education group or below and the junior high school education group. However, the promotion effect on the high school education group or above is not significant. This may be due to the fact that apple growers with a degree or above have a higher level of knowledge, are more agile and open-minded, and have a better understanding and awareness of FETs. Therefore, DTA does not significantly impact their behavior. On the other hand, for groups with lower education levels, DTA can improve their information asymmetry problem regarding FETs, thereby significantly affecting their choice behavior.

4.5.2. Heterogeneity Analysis Based on Planting Scale

Due to the high standard deviation in the planting area of apple growers, indicating a significant difference in sample sizes, this paper categorizes apple growers into small and large operation groups based on a median of 8.00 mu (in line with the national average arable land area of less than 8 mu). The apple planting area was excluded from the benchmark regression model, and a re-regression was conducted, with the results shown in

Table 8. Results of heterogeneity analysis based on management scale of apple orchards.

Variables	Multivariateprobit6-Small Scale			Multivariateprobit7-Large Scale
	STFF	WFIT	S/CF	STFF	WFIT	S/CF
DTA	0.166 (0.181)	0.211 * (0.128)	0.315 ** (0.160)	0.228 * (0.136)	0.390 *** (0.106)	0.171 (0.134)
FC	0.942 *** (0.197)	0.423 *** (0.127)	0.743 *** (0.169)	0.750 *** (0.151)	0.429 *** (0.120)	0.387 *** (0.144)
Constant	−2.980 ** (0.985)	−0.517 (0.593)	−2.397 ** (0.929)	−2.416 *** (0.690)	1.112 ** (0.481)	−2.033 *** (0.677)
Control variables	controlled	controlled	controlled	controlled	controlled	controlled
Atrho21	0.074 (0.094)			0.242 *** (0.078)
Atrho31	0.138 (0.114)			0.073 (0.086)
Atrho32	0.184 * (0.094)			0.305 *** (0.082)
Wald chi2	186.23			222.21
Prob > chi2	0.000			0.000
N	534			682

Note: STFF, Soil testing and formulated fertilization; WFIT, Water-fertilizer integration technology; S/CF, Slow/controlled fertilization. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively; the values in “()” are the standard error of the regression coefficient.

. The findings indicate that DTA has a more significant impact on the top two sub-technologies adoption behaviors of the large operation group, and the technology correlation effect is more pronounced in this group. This could be due to the higher human capital of the large operation group, which enables them to access and utilize agriculture-related information more effectively. Additionally, the large operation group tends to have more material resources and fewer decision-making barriers than the small operation group, making the promotion of DTA more apparent.

5. Conclusions and Implications

The emergence of the digital era has brought about a considerable impact on agriculture, rural areas, and farmers. Certain scholars have noted that digital technology will initiate the “Fourth Agricultural Revolution” [41]. Accordingly, this study scrutinizes the correlation between DTA, FC, and their choice of FETs. The main conclusions are as follows. (1) DTA has a significant positive impact on apple growers’ choice of FETs, indicating that increasing information access through digital technology will increase farmers’ adoption of FETs. This finding is in line with prior research and underscores the crucial role that information plays in farmers’ decision-making [17,18,32,33]. (2) This paper further confirmed that DTA can not only directly facilitate apple growers’ adoption of FETs, but also indirectly enhance FETs by increasing their FC. This finding shows the importance of FC on their behavior [1,12,23]. It also reflects the positive impact of DTA on FC, which is consistent with the research conclusion that Li et al. (2022) found: the application of digital technology can improve farmers’ information acquisition, further enhance their cognition, and regulate their behavior [17]. (3) There is a complementary phenomenon in FETs. (4) The impact of DTA on FETs is more significant among groups with higher secondary education and above, and the impact on FETs is more significant among smaller business-scale groups compared to larger business-scale groups, which have been explained in the former section. This finding is related to the differences in the ability to use digital technology among farmers with different characteristics [42].

Based on the aforementioned theoretical analyses and empirical findings, this paper derives four possible policy recommendations. (1) Efforts should be made to improve the Internet infrastructure in rural areas. However, the goal should not only be to increase Internet penetration or smartphone ownership in rural areas but also to enhance farmers’ ability to access relevant information effectively through the Internet. Furthermore, the government can promote the establishment of agricultural information platforms to consolidate and disseminate specialized knowledge, technical guidance, market information, and other relevant resources about apple cultivation. Such a platform would offer farmers convenient access to information, amplifying the benefits of utilizing the Internet as a tool for obtaining agricultural insights. (2) It is crucial to value FC in making decisions. In addition to using the Internet to gain effective information, it is important for the government to help apple growers to learn more about green technology and relevant national policies. This approach aims to improve apple growers’ understanding of eco-friendly technology and ultimately influence their production practices. (3) Considering the correlation effect among apple growers’ adoption of FETs, it is necessary to implement a “bundling” combination strategy to accelerate the process of fertilizer reduction and efficiency improvement. (4) To fully leverage the “long tail” impact of DTA on farmers’ decision-making, it is crucial to consider the specific effects on the adoption of FETs among groups with different educational backgrounds and business scales. For individuals with junior and lower education, ample training and guidance should be provided, while entities with a large business scale require more precise support and incentives.

While this study may provide scholars with a reference for studying farmers’ adoption of FETs in the digital age, it has certain limitations. Firstly, this study is based solely on apple growers in China and cannot be generalized to other crops or countries. Further research is needed to explore the relationship between DTA, FC, and the adoption of FETs. Secondly, the data used in this study are cross-sectional. Longitudinal panel data would provide more persuasive research findings if the surveyed farmers are tracked over time. This is also a potential direction for future research advancements.