How does internet use promote joint adoption of sustainable agricultural practices? Evidence from rice farmers in China

ABSTRACT

Climate change and public health emergencies have severely threatened world food security. In response, the Chinese government has actively promoted sustainable agricultural practices (SAPs) and emphasized the importance of integrated packages to enhance the development of high-quality agriculture. However, due to information failure, farmers’ adoption rate of SAPs is still very low. Meanwhile, internet use can effectively reduce the cost of information barriers and catalyze farmers to promote the joint adoption of SAPs. This paper aims to theoretically elucidate the logic of how internet use impacts farmers’ joint adoption of SAPs. Empirical analysis based on data from 844 rice farmers in southern China reveals that internet use can significantly promote the joint adoption of SAPs, and adoption behaviours are complementary across multiple technologies. Furthermore, internet use in the group of rice farmers with more farmlands or those receiving subsidies is more inclined to adopt SAPs jointly. These findings suggest that developing countries should prioritize public investment in rural internet infrastructure to facilitate joint adoption of SAPs through farmers’ internet use. The moderate scale of farmland usage and targeted agricultural subsidy policies can further enhance the effectiveness of ICT-based joint adoption of SAPs.

KEYWORDS:

• Sustainable agricultural practices
• joint adoption of technology
• internet use
• farmers

Highlights

• Internet use significantly promotes the joint adoption of sustainable agricultural practices (SAPs) among farmers in southern China.

• Complementary adoption across three technologies indicates that new seeds or seedlings technology adoption can be driven if the farmer adopts the other two SAPs.

• Larger farmland scale or receiving subsidies contribute to a higher inclination towards farmers’ joint adoption of SAPs through internet use.

• Developing countries should prioritize public investment in rural internet infrastructure to overcome limited access to information and foster the adoption of SAPs.

• Targeted policies, including agricultural subsidies and encouraging appropriate management scale of farmland, can further enhance the effectiveness of ICT-based joint adoption of SAPs, ensuring sustainable agricultural development.

1. Introduction

Global food security suffers severe shocks due to climate change and public health emergencies. The number of people affected by hunger globally rose to as many as 828 million in 2021, an increase of about 46 million since 2020 (FAO et al., 2022). The 2030 Agenda for Sustainable Development is already off-track, reframing sustainable food systems is urgent. Sustainable Agricultural Practices (SAPs) are strategies that can enhance productivity sustainably by addressing the degradation of ecosystem services and improving the ability of smallholder farmers to adapt to climate variability and change (Teklewold et al., 2013a). To address the issue of food security, the Chinese government emphasized that joint adoptions of SAPs should be promoted to facilitate sustainable agricultural development in an integrated and complementary manner.¹ The rationale of those policies is clear. First, agricultural production is a continuous process, and joint adoption is inevitable in agricultural production activities (Dorfman, 1996). Second, SAPs appear to be independent, but it is still inherently systematic, as SAPs can interrelate with different technologies (Oyetunde-Usman et al., 2021); adopting multiple technologies is more effective than adopting a single technology (Ehiakpor et al., 2021). Third, Farmers’ technology adoption decisions are determined by the characteristics of agricultural technology innovation (Morgan et al., 2023). So, considering synergistic effects, farmers may change the internal components of SAPs integration, deepen the structure of SAPs integration, generate innovation decisions (Arthur, 2009) and allow them to achieve agricultural productivity gains with the joint adoption of SAPs. However, Chinese farmers’ willingness to engage in joint adoption was low (Zeng et al., 2019). For example, 42.77% of rice farmers adopted only one kind of Pro-environmental Agricultural Practice (PEA), 31.03% chose to adopt two PEAs, and only 6.5% chose to adopt three or more PEAs (Zeng et al., 2020).

So far, a few studies have researched the impacts that influence farmers’ joint adoption of SAPs. Due to constraints in resources, ability, and information, farmers face limitations in their behavioural capacity, making it challenging to meet the conditions for the joint adoption of SAPs (Bakker et al., 2021; Gao et al., 2020). Also, since farmers are risk-averse when making decisions (Ellis, 1993), their attitudes regarding new technologies can lack motivation. The rapid development of information and communication technology (ICT) has effectively reduced the cost of information access and dissemination, which can increase the willingness of farmers to adopt SAPs jointly (Ma & Wang, 2020). Research has also demonstrated that many farmers worldwide already use internet search results to make agricultural production decisions and improve production efficiency (Deichmann et al., 2016; Michels et al., 2020; Zhai et al., 2020). Many farmers have internet access, especially in China, the largest developing country (Li et al., 2022). Until June 2022, the internet penetration rate in rural areas of the country was 58.8%. With the rapid development of ICT, agricultural sectors, including farming, aquaculture, fisheries, and forestry, are actively embracing ICT technologies such as IoT and AI to enhance production efficiency and product quality. This growing trend has prompted research into the relationship between internet use and joint adoption of SAPs (Ali et al., 2019; Kuhlmann et al., 2023; Lim et al., 2022; Ma, 2020; Steinke et al., 2021; Tham et al., 2022; Van et al., 2022). Sørensen et al. (2019) argue that farmers can directly engage in the joint adoption of ICT with other technologies. Moreover, farmers can utilize ICT to provide agricultural production information, indirectly assisting them in making joint adoption decisions (Hasanuzzaman, 2019; Khan et al., 2022a; Sørensen et al., 2019). Sheng Tey et al. (2018) argued that ICT could reduce uncertainty in technology adoption among SAPs. Others have argued that ICT enhances the willingness of farmers to adopt SAPs jointly (He et al., 2021). Moreover, ICT can contribute to agricultural productivity (Khan et al., 2022b) and achieve sustainable and high-quality agricultural development (Chandio et al., 2022; Symeonaki et al., 2019). However, literature on theoretical analyses of how internet use influences farmers’ SAPs adoption is still inadequate, and only Ma and Wang (2020) have made a relatively comprehensive and detailed analysis at the farm level. Furthermore, the latest research notices the digital divide among different groups of farmers in technology adoption (Ma, 2020). According to the findings, smaller farming households can suffer from ‘technophobia’ and be reluctant to adopt new technological devices such as smartphones (Marescotti et al., 2021), thus less likely to adopt SAPs; larger farmers were more likely to experiment with new conservation farming technologies on parts of their land (Arslan et al., 2014). Moreover, the government tends to use subsidies policy to promote the adoption of new agricultural technologies. Consequently, the effectiveness of internet use in promoting the joint adoption of SAPs may be constrained by internal and external factors, including endowment characteristics and policy conditions. However, due to the lack of empirical evidence to support them, these perspectives require further exploration (Lankoski & Lankoski, 2023).

The objective of this paper is to cover the gaps mentioned above. First, this paper provides a rationale for internet use as ‘ICT information’ input which drives agricultural production and theoretically analyses how it affects farmers’ joint adoption of SAPs. Then, by utilizing survey data from rice farmers, it empirically analyzes the impact of internet use on the joint adoption of SAPs in intensity perspective, verifies the complementary relationship between the three SAPs’ adoption from a decision correlation perspective, and finds out larger-scale farmers and farmers receiving agricultural subsidies are more likely to adopt SAPs jointly when using the internet. Finally, based on the research findings, it proposes policy recommendations for promoting farmers’ joint adoption of SAPs. This study provides a theoretical and practical reference for improving the ICT-based joint adoption of SAPs in developing contexts.

This paper is organized as follows. After this introduction, section 2 is the theoretical framework and research hypotheses. Section 3 outlines the study sites, data collection and empirical models. Section 4 presents the main results and heterogeneity analysis. Section 5 includes a synthesis of the results, policy and practice recommendations, and research limitations.

2. Theoretical analysis and research hypotheses

Yoav and Shchori-Bachrach (1973) proposed the Diffusion Theory of Technological Innovation for business firms, and Feder and Slade (1984) argued that it was also applicable in the agricultural sector. Their study of technology adoption by Indian farmers underscored the role of information accumulation in agricultural production while distinguishing between active information gathering and passive information acquisition. On this basis, Zheng et al. (2022) included ‘ICT information’ as an input factor in farmers’ agricultural production functions. They found that internet use could promote farmers’ technology adoption in agricultural production.

This paper incorporates ‘ICT information’ as a driver of agricultural production and theoretically analyses farmers’ joint adoption of SAPs. It is assumed that farmers obtain new technology to improve their production utility through passive and active learning, while the most common active learning is from ICT information, and the acquisition of ICT information involves low costs such as time or money. This means that if the farmer uses internet, the more information of SAP farmers can get through the internet, the more likely they would adopt a new SAP. Since the technology adoption of SAPs varies across farmers, farmers with more ICT information of SAPs accumulation will continue to adopt SAPs to maximize utility if the marginal productivity of ICT information is positive. The first hypothesis of this paper is:

H1: Internet use promotes farmers’ joint adoption of SAPs.

Feder (1982) has argued that agricultural innovations usually show higher perceived risk than traditional technologies. Joint adoption by farmers is usually an information-based rational choice (Ryan & Subrahmanyam, 1975): although farmers maximize expected profits by joint adoption, they tend to adopt technologies in some order over time (Atanu et al., 1994) until full joint adoption is achieved, thus maximizing expected utility (Leathers & Smale, 1991). However, Feder (1982) has also argued that the selection patterns of innovations are interrelated. Assume that two innovations are introduced simultaneously: one innovation is scaling neutral, and the cost of the other decreases with increasing farm scale. Using certain technology-to-technology combinations may be more favourable for some farmers, while other farmers actively avoid using them. There are clear complementarities and alternatives between different agricultural technologies (Teklewold et al., 2013b).

In the process of ICT information dissemination, farmers’ information decision-making behaviours on different SAPs may also be correlated with each other. High-volume, low-cost, and rapid internet information may induce farmers to adopt combined innovations for increasing production, income, and efficiency. Moreover, they may integrate and combine other SAPs to maximize the profit of the agricultural output. This leads to the second hypothesis of this paper:

H2: In the internet use promoting the joint adoption of farmers’ SAPs, SAPs complement each other.

Heterogeneity also exists in technology adoption (Foster & Rosenzweig, 2010). The effectiveness of ICT information for the joint adoption of SAPs may be limited by both exogenous and endogenous constraints on farmers’ decision-making (Feder, 1982). On the one hand, farm scale is the main factor of agriculture production and may impact farmers’ joint adoption of SAPs. Smaller farmers have less ability to search for technology information and are usually more risk-averse and psychologically resistant to new technologies. In comparison, larger farmers with higher levels of information accumulation can hold more positive attitudes toward the joint adoption of new technologies. So, these farmers are more likely to use ICT for the joint adoption of SAPs with the help of the internet compared to smaller farmers.

On the other hand, official institutions usually extend agricultural innovations, and subsidy policies can reduce the investment cost of SAPs for farmers and enhance their willingness for technology adoption. Farmers who receive subsidies are more likely to adopt SAPs jointly through the internet. As such, our third hypothesis is:

H3-1: Compared to small-scale farmers, large-scale farmers are more inclined to adopt SAPs jointly when they use the internet.

H3-2: Compared to the farmers who have no subsidies, farmers who received subsidies are more inclined to joint adoption of SAPs when they use the internet.

3. Materials and methods

3.1. Data

The sample of this study is rice farm households in Xinfeng County, Shaoguan City, Guangdong Province, Southern China, as it shown in Figure 1. Xinfeng County has a mild subtropical monsoon climate, high rainfall, abundant light, long frost-free periods, and four distinct seasons. Hills and basins dominate the topography, making it suitable for growing traditional crops such as rice Meanwhile, telecommunication optical ports and mobile signal base stations were constructed to cover all administrative villages in the county in 2019.²

Figure 1. Study area in Chinese territory.

Note: This figure is based on a standard map downloaded from the standard map service system of the Ministry of Natural Resources of the People's Republic of China with approval number GS(2019)1822. The base map has not been modified.

The research group conducted the survey in 2019. Considering Xinfeng county had 142 administrative villages in 7 towns, we adopted the following sampling principles. First, 60 villages were randomly selected from the entire administrative villages; then, 2 natural villages were randomly selected from each administrative village, and 10 farmer households were selected from each natural village. As a result, a total of 1200 farmer households were surveyed, and after excluding non-rice farmer households, the final sample size in this paper is 844.

3.2. Variables

3.2.1. Dependent variables

Joint adoption of SAPs is the dependent variable; it refers to the adoption of the number of SAPs by rice farmers for agricultural production activities (Dorfman, 1996). And each SAP adoption is measured as follows: (1) Whether rice farmer adopts new seeds or seedlings (SAP_S). (2) Whether rice farmers adopt green and environmentally friendly agrochemical technology for production (SAP_PF). (3) Whether rice farmers adopt modern machinery for farmland preparation, planting, or harvesting (SAP_M) (Akram et al., 2020). So, we can construct an ordinal variable to reflect SAPs’ joint adoption. Based on Ehiakpor et al. (2021), we assign a value of 0 to rice farmers who do not adopt SAP, 1 to rice farmers who adopt one SAP, 2 to rice farmers who adopt two SAPs, and 3 to rice farmers who adopt three SAPs.

3.2.2. Independent variable

The independent variable is whether rice farmers use the internet via WIFI or a mobile network, which is assigned a value of 1 if the condition is satisfied and 0 otherwise.

3.2.3. Control variables

To control for external influences on the joint adoption of SAPs by rice farmers (Conley & Udry, 2010; Teklewold et al., 2013b; Wainaina et al., 2016; Zhang et al., 2020), we included control variables as follows: householder characteristics include age, gender, education, training, and risk; household characteristics include family size, farm labourers, and agricultural income; agriculture management characteristics include area, plots, and farming type; policy characteristics include subsidies; and social characteristics include clan, migrant, trust, and associations.

3.3. Descriptive statistics

Table 1 shows the statistical description of the variables used in this paper.

Table 1. Variables definition and description.

Variables	Definition	Mean	S.D.	Min	Max
Dependent variables
SAP_S	=1 if the rice farmer adopts the new seeds and seedlings, 0 otherwise	0.686	0.464	0	1
SAP_PF	=1 if the rice farmer adopts the green and environmentally friendly pesticide and fertilizer technologies, 0 otherwise	0.273	0.446	0	1
SAP_M	=1 if the rice farmer uses machinery for land preparation, seeding or harvesting, 0 otherwise	0.744	0.437	0	1
Joint adoption	adoption number of SAPs (none = 0; one = 1; two = 2; three = 3)	1.703	0.908	0	3
Independent variable
Internet_use	=1 if the rice farmer uses the internet, 0 otherwise	0.705	0.456	0	1
Control variables
Age	Age of the rice farmer(years)	57.564	9.713	21	91
Gender	=1 if the rice farmer is male, 0 otherwise	0.897	0.304	0	1
Education	Education of the rice farmer(below primary school = 1; primary school = 2; junior high school =3; senior high school = 4; above high school = 5)	2.737	0.759	1	5
Training	=1 if the rice farmer has been trained in agricultural technology, 0 otherwise	0.168	0.374	1	0
Risk	The risk appetite of the rice farmer (1–5, risk averse = 1, risk preference = 5)	1.633	1.107	1	5
Family size	Number of family members	5.072	1.982	1	15
Farm_labourers	Number of family farming labourers	1.977	1.251	0	11
Agri_ income	Total household farm income (in natural logarithm)	2.444	3.878	0	12.543
Area	The area of arable land on which the rice farmer cultivates rice (mu)	3.703	4.727	0.4	110
Plots	Number of rice plots	8.515	7.532	1	55
Farming_type	=1 if the rice farmer is growing double-season rice, 0 otherwise	0.765	0.424	0	1
Subsidies	=1 if the rice farmer receives agricultural subsidies, 0 otherwise	0.823	0.382	0	1
Clan	=1 if the rice farmer is a member of a clan, 0 otherwise	0.722	0.448	0	1
Migrant	=1 if the rice farmer is working outside the village, 0 otherwise	0.791	0.406	0	1
Trust	Level of trust of rice farmers in village collectives (very high = 5; very low = 1)	3.874	1.070	1	5
Associations	=1 if the rice farmer is a member of an agriculture technology association, 0 otherwise	0.123	0.329	0	1

A description of technology adoption by rice farmers is shown in Figures 2–4. The adoption of each SAPs by rice farmers is shown in Figure 2, in which 68.60% of rice farmers adopted SAP_S, 27.25% adopted SAP_PF, and 74.41% adopted SAP_M. Figure 3 shows the joint adoption by rice farmers. A significant proportion of rice farmers opted for joint adoption (60.19%).

Figure 2. Adoption rate of each SAP by rice farmers.

Figure 3. Adoption intensity of different types of SAPs by rice farmers.

Figure 4. Percentage of internet use by rice farmers.

Figure 4 shows the internet use of rice farmers. Most rice farmers (70.50%) had chosen to use internet.

3.4. Estimation strategy

The dependent variable is farmers’ joint adoption of SAPs. In our study scenario, there are four possible adoption outcomes: no adoption, adoption of one technology, adoption of two technologies, and adoption of three technologies. Here, the joint adoption behaviour of farmers towards SAPs is coded as an ordinal variable. We use the Ologit model as the benchmark model, and it's shown in the following equation (Greene, 2012): (1) $J\mathrm{\_}{\mathrm{a}\mathrm{d}\mathrm{o}\mathrm{p}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_i^{\ast }=\text{β}{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{n}\mathrm{e}\mathrm{t}}_i^{\mathrm{\prime }}+{ϵ}_i$(1) (2) $J\mathrm{\_}adoptio{n}_i=\{\begin{array}{ccc}0& \mathrm{i}\mathrm{f}& J\mathrm{\_}{\mathrm{a}\mathrm{d}\mathrm{o}\mathrm{p}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_{\mathrm{i}}^{\ast }\le r_0\\ 1& \mathrm{i}\mathrm{f}& r_0<J\mathrm{\_}{\mathrm{a}\mathrm{d}\mathrm{o}\mathrm{p}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_i^{\ast }\le r_1\\ 2& \mathrm{i}\mathrm{f}& r_1<J\mathrm{\_}{\mathrm{a}\mathrm{d}\mathrm{o}\mathrm{p}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_i^{\ast }\le r_2\\ 3& \mathrm{i}\mathrm{f}& J\mathrm{\_}{\mathrm{a}\mathrm{d}\mathrm{o}\mathrm{p}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_i^{\ast }>r_2\end{array}$(2) In Equations (1) and (2), $J\mathrm{\_}{\mathrm{a}\mathrm{d}\mathrm{o}\mathrm{p}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}}_i^{\ast }$ is an unobservable latent variable, indicating the intensity of joint adoption of farmer $i$. ${\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{n}\mathrm{e}\mathrm{t}}_i^{\mathrm{\prime }}$ indicates the $i$th farmer’ internet use affecting the joint adoption of SAPs. $\text{β}$ is a parameter to be estimated. ${ϵ}_i$is the random error term which follows a standard normal distribution. The farmer’ s joint adoption of SAPs is categorized into four outcomes: (1) none, (2) one, (3) two, and (4) three. In Equation (2), 0,1, 2 and 3 are the different intensity of joint adoption(none, one, two, three), $r_0<r_1<r_2$ are the cut points to be predicted for any joint adoption intensity.

Further, referring to Wainaina et al. (2016) and Koppmair et al. (2017), the MVP model is constructed to analyze the effect of internet use on the interrelationship of SAPs for testing H2. the MVP approach simultaneously models the influence of the set of explanatory variables on each of the different practices, while allowing for the potential correlation between unobserved disturbances, as well as the relationship between the adoption of different practices (Belderbos et al., 2004). Failure to capture unobserved factors and interrelationships among adoption decisions regarding different practices will lead to bias and inefficient estimates (Greene, 2012).

The observed outcome of SAP adoption can be modelled following a random utility formulation. The general model can be written as Equations (3) and (4): (3) $y_{ik}^{\ast }={\text{β}}_k\mathrm{I}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{n}\mathrm{e}{\mathrm{t}}_i+{\sigma }_{ik}\left(k=S,PF,M\right).$(3) (4) $y_{ik}=\{\begin{array}{cc}1,& \mathrm{i}\mathrm{f}{y}_{ik}^{\ast }>0\\ 0,& \mathrm{o}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{w}\mathrm{i}\mathrm{s}\mathrm{e}\end{array}\left(k=S,PF,M\right).$(4) In Equation (3), $y_{ik}^{\ast }$ is an unobservable latent variable that captures the unobserved preferences or demand associated with the $k$th choice of SAP, k denotes choice of new seeds and seedlings ($S$), green and environmentally friendly pesticide and fertilizer technologies ($PF$), and modern agricultural machinery ($M$). The independent variable is $Interne{t}_i$, a dummy variable indicating whether $i\mathrm{t}\mathrm{h}$ farmers use the internet. ${\text{β}}_k$ is the vector of parameters to be estimated. Unobserved characteristics captured by the stochastic error term ${\sigma }_{ik}$. Using the indicator function, the unobserved preferences in Equation (1) translate into the observed binary outcome equation for each choice as in Equation (4).

In the multivariate model, where the adoption of several SAPs is possible, the error terms in the model jointly follow a multivariate normal distribution with zero conditional mean and variance normalized to unity. The model generates a variance-covariance matrix$\mathrm{\Omega }$ is given by: (5) $\left[\begin{array}{ccc}1& {\rho }_{\mathrm{S}\mathrm{P}\mathrm{F}}& {\rho }_{\mathrm{S}\mathrm{M}}\\ {\rho }_{\mathrm{P}\mathrm{F}\mathrm{S}}& 1& {\rho }_{\mathrm{P}\mathrm{F}\mathrm{M}}\\ {\rho }_{\mathrm{M}\mathrm{S}}& {\rho }_{\mathrm{M}\mathrm{P}\mathrm{F}}& 1\end{array}\right]$(5) Of particular interest are the off-diagonal element in the covariance matrix represents the unobserved correlation between the stochastic component of SAPs. The results of the MVP model in Equation (5) enable the representation of the unobserved characteristics that influence the SAPs to make decisions, thus reflecting the adoption of correlation between the decision error terms of the three specific technologies (Kassie et al., 2013).

4. Results

4.1. Joint adoption of SAPs: intensity perspective

4.1.1. Benchmark model

The Ordered Logit model benchmark regression revealed that the internet use positively affected the joint adoption of SAPs and was statistically significant at the 5% level (Table 2, Model 1). Furthermore, the Logit model regression showed that the coefficients of internet use for each of the three different SAPs are all positive (Table 2, Model 2). Rice farmers using the internet were more likely to adopt SAP_PF and SAP_M, which were statistically significant at the 5% level; however, internet use did not significantly affect SAP_S (Coefficient = 0.020, P-value = 0.924).

Table 2. Joint adoption regression: Benchmark model estimation.

Variables	Model 1: Ologit model		Model 2: Logit model
			SAP_S		SAP_PF		SAP_M
	Parameter	P-values	Parameter	P-values	Parameter	P-values	Parameter	P-values
Internet_use	0.422**	0.031	0.020	0.924	0.647**	0.012	0.433**	0.030
	(0.195)		(0.211)		(0.256)		(0.199)
Age	0.007	0.311	−0.001	0.861	0.016*	0.071	0.003	0.697
	(0.007)		(0.008)		(0.009)		(0.009)
Gender	0.186	0.425	0.148	0.512	−0.140	0.585	0.558**	0.023
	(0.233)		(0.225)		(0.257)		(0.246)
Education	0.077	0.453	0.064	0.629	0.219**	0.040	−0.142	0.171
	(0.102)		(0.133)		(0.107)		(0.104)
Training	0.181	0.354	0.043	0.805	0.395*	0.074	−0.051	0.829
	(0.196)		(0.174)		(0.221)		(0.235)
Risk	−0.014	0.782	−0.025	0.714	0.022	0.709	−0.004	0.959
	(0.051)		(0.068)		(0.060)		(0.078)
Family size	0.054	0.175	0.029	0.493	0.036	0.430	0.044	0.295
	(0.040)		(0.042)		(0.046)		(0.043)
Farm_labourers	0.073	0.139	0.131**	0.039	0.074	0.239	−0.023	0.690
	(0.049)		(0.063)		(0.062)		(0.058)
Agri_income	0.024	0.227	0.026	0.265	0.004	0.886	0.034	0.194
	(0.020)		(0.023)		(0.026)		(0.027)
Area	0.002	0.884	−0.006	0.730	0.000	0.987	0.016	0.436
	(0.014)		(0.019)		(0.015)		(0.021)
Plots	−0.032**	0.016	−0.017	0.187	−0.021	0.138	−0.039***	0.004
	(0.013)		(0.013)		(0.014)		(0.014)
Farming_type	−0.235	0.126	−0.360**	0.037	−0.114	0.618	−0.092	0.692
	(0.154)		(0.173)		(0.228)		(0.232)
Subsidies	0.071	0.700	0.169	0.449	−0.431*	0.054	0.474**	0.021
	(0.184)		(0.223)		(0.224)		(0.205)
Clan	0.296*	0.062	0.372**	0.050	0.095	0.692	0.192	0.324
	(0.159)		(0.190)		(0.239)		(0.194)
Migrant	0.316*	0.078	0.441**	0.023	0.170	0.531	0.157	0.424
	(0.179)		(0.194)		(0.271)		(0.196)
Trust	0.132**	0.030	0.069	0.346	0.171**	0.046	0.088	0.289
	(0.061)		(0.073)		(0.086)		(0.083)
Associations	0.265	0.265	0.136	0.620	0.379	0.217	−0.020	0.933
	(0.238)		(0.274)		(0.307)		(0.235)
Constant			−0.460	0.578	−3.622***	0.000	−0.393	0.655
			(0.826)		(0.855)		(0.880)
Observation	844		844		844		844
R^2/Pseudo R^2	0.021		0.023		0.042		0.033

Note: * p < 0.1, ** p < 0.05, *** p < 0.01. Robust standard errors adjusted for clustering at village level in brackets.

In addition, we reported the marginal effects of the joint adoption of SAPs in Table 3 (Wollni et al., 2010; Teklewold et al., 2013b). We observed that the probability of joint adoption of two technologies increased by 3.4% while the likelihood of joint adoption of three technologies increased by 6.7% if rice farmers used the internet. The result indicates an increasing tendency for the intensity of joint adoption of SAPs among rice farmers through using the internet (Figure 5).

Figure 5. Marginal effects of joint adoption of SAPs.

Table 3. Joint adoption marginal effects: Benchmark model estimation.

Variables	Y = 0	Y = 1	Y = 2	Y = 3
Internet_use	−0.037**	−0.064**	0.034**	0.067**
	(0.016)	(0.030)	(0.016)	(0.032)
Control variables	Yes	Yes	Yes	Yes
Observation	87	249	336	172

Note: * p < 0.1, ** p < 0.05, *** p < 0.01. Robust standard errors adjusted for clustering at village level in brackets. The control variables of all these models are the same as in Table 2.

4.1.2. Robustness check

The underlying assumption of the Poisson regression model was that events occurred independently over time (Cameron & Trivedi, 1986). Therefore, we tested for potential overdispersion in the count-related variables (Cameron & Trivedi, 1990). The results showed that the response variance of the model was less than the mean, and there was no potential overdispersion in the count-related variables (Mean = 1.702, Response Variance = 0.824). Then, we dropped the negative binomial regression model in favour of a Poisson model. Considering there might have been more zeros in the dependent variable than expected (Thinda et al., 2020), we also evaluated whether a zero-inflated Poisson regression produced a better fit. As the evaluation result was not the case (Vuong test, P-value = 0.500), we used a standard Poisson model for robustness checking (Table 4).

Table 4. Joint adoption robustness check: Poisson model estimation.

Variables	Poisson		Marginal effects
	Parameter	P-values
Internet_use	0.119**	0.032	0.196**
	(0.055)		(0.090)
Control variables	Yes		Yes
Constant	−0.091	0.646
	(0.197)
Observation	844		844

Note: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors adjusted for clustering at village level in brackets. The control variables of all these models are the same as in Table 2.

We found that internet use promoted increased intensity of joint adoption of SAPs among rice farmers, which was statistically significant at the 5% level. The marginal effects of the Poisson model indicate that internet use was positively correlated with the joint adoption of SAPs among rice farmers. Rice farmers using the internet were 19.6% more likely to adopt SAPs jointly than those not.

4.1.3. Endogeneity analysis

In the technology adoption process by rice farmers, there was a possibility of endogeneity problems between the independent and dependent variables due to reverse causality. Specifically, for the joint adoption of new SAPs, rice farmers would seek adequate information through various means, which could effectively contribute to developing information dissemination channels and increase rice farmers’ internet usage (Bi et al., 2022). Therefore, the endogeneity problems would lead to biased results of the benchmark model.

We adopted an instrumental variable approach to deal with potential endogeneity problems for more rigorous results. We chose the logarithm of rice farmers’ communication expenses in 2018 as an instrumental variable for internet use (He et al., 2021). So, the rice farmers’ decision was made given that: (1) higher communication expenses of rice farmers were correlated with lower digital divide problems faced by households (Ma et al., 2020) and (2) higher communication expenses of rice farmers were essential household expenses that do not directly affect their joint adoption of SAPs. After weak-instrument tests, We found that communication expenses were not a weak instrumental variable, meeting the requirements for the exogenous variable (F-value = 36.911).

We used a conditional mixture process (CMP) to deal with potential endogeneity problems (Roodman, 2011), the results were shown in Table 5. The endogeneity test parameter Atanhrho_12 was significant at the 1% statistical level. It indicated the endogeneity problem in the benchmark model. In the first stage of regression, there was a positive correlation between instrumental variables and endogenous explanatory variables, which was statistically significant at the 1% level. In the second stage regression, internet use was positively correlated with the joint adoption of SAPs and was statistically significant at the 1% level. The CMP approach was mainly consistent with the benchmark regression results, supporting H1.

Table 5. CMP method: IV-Oprobit model estimation.

Variables	First stage		Second stage
	Parameter	P-values	Parameter	P-values
Internet_use			1.330***	0.000
			(0.225)
Communication expenses (ln)	0.215***	0.000
	(0.040)
Control variables	Yes		Yes
Atanhrho_12			−0.847***	0.000
			(0.180)
Constant			−0.214	0.578
			(0.384)
Observation	844		844

Note: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors adjusted for clustering at village level in brackets. The control variables of all these models are the same as in Table 2.

4.2. Joint adoption of SAPs: decision correlation perspective

We chose to use the MVP model considering the decision complexity when rice farmers’ Internet use affects the joint adoption of sustainable agricultural production technologies. It allows further analysis of the correlation between unobserved factors among adoption decisions and error terms in the adoption equation (Koppmair et al., 2017). Table 6 shows the regression results of the MVP model. The results show that the coefficient of rice farmers’ internet use on SAPs is positive. Internet use on the adoption of SAP_PF technology by rice farmers is statistically significant at a 1% level, which means using the internet increased the odds of SAP_PF adoption by 39.7%. Internet use on the adoption of SAP_M technology by rice farmers was statistically significant at the 5% level, which means using the internet increased the odds of SAP_M adoption by 27.0%.

Table 6. Correlation test: MVP model estimation.

Variables	SAP_S		SAP_PF		SAP_M
	Parameter	P-values	Parameter	P-values	Parameter	P-values
Internet_use	0.005	0.967	0.397***	0.007	0.270**	0.023
	(0.129)		(0.147)		(0.118)
Control variables	Yes		Yes		Yes
Constant	−0.359	0.465	−2.063***	0.000	−0.224	0.669
	(0.492)		(0.505)		(0.524)
Observation	844		844		844

Note: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors adjusted for clustering at village level in brackets. The control variables of all these models are the same as in Table 2.

SAP adoption is not mutually exclusive, the adoption of one technology does not mean that other technologies cannot be adopted. There may be complementary and alternative relationships between different technologies such that it may make more sense for rice farmers to adopt certain combinations of technologies according to local conditions. This can lead to synergistic relationships that can lead to a ‘1 + 1 > 2’ effect (Koppmair et al., 2017).

Although the internet was not significant for SAP_S in the benchmark regression, there were complementary relationships between SAP_S & SAP_PF, SAP_S & SAP_M and SAP_PF & SAP_M in the correlation matrix, which were statistically significant at the 1%, 1%, 10% level (Table 7), supporting H2. As the ‘chips’ of agriculture, seeds play a critical role in rice production, the results indicated that the adoption of SAP_PF or SAP_M promoted the adoption of SAP_S.

Table 7. SAPs correlation matrix: MVP model estimation.

Variables	SAP_S		SAP_PF		SAP_M
	Parameter	P-values	Parameter	P-values	Parameter	P-values
SAP_S	1		0.557***	0.000	0.212***	0.000
			(0.066)		(0.056)
SAP_PF			1		0.123*	0.062
					(0.066)
SAP_M					1

Note: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors adjusted for clustering at village level in brackets. The control variables of all these models are the same as in Table 2.

4.3. Heterogeneity analysis

However, the above results are tested for the whole sample level, ignoring differences in SAPs technology adoption that might be caused by different endowment and institutional characteristics. Here, we analyze the heterogeneous effects of joint adoption of SAPs via internet use based on rice farmers’ farmland scale and subsidies. On the one hand, the sample of rice farmers was divided into two groups, ‘≥3’ and ‘<3’, using the median farm size as the cutoff (Zhao et al., 2022); On the other hand, the whole sample was divided into two groups based on whether they received subsidies. When the rice farmer received subsidies, he or she was in the ‘yes’ group; otherwise in the ‘no’ group.

As the results shown in Table 8, we found that rice farmers with larger farmland scales that use the internet for joint adoption of SAPs were more favourable. In addition, internet use by rice farmers who received subsidies significantly promoted the joint adoption of SAPs technology relative to rice farmers who did not receive subsidies.

Table 8. Heterogeneous effects tests: Ologit model estimation.

Variables	Farmland scale				Subsidies
	≥3		<3		Yes		No
	Parameter	P-values	Parameter	P-values	Parameter	P-values	Parameter	P-values
Internet_use	0.519**	0.018	0.377	0.157	0.462**	0.020	0.135	0.718
	(0.219)		(0.267)		(0.199)		(0.373)
Control variables	Yes		Yes		Yes		Yes
Observation	432		412		695		149
Pseudo R^2	0.033		0.027		0.022		0.048

Note: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors adjusted for clustering at village level in brackets. The control variables of all these models are the same as in Table 2.

Adopting all three technologies positively benefited rice farmers with more farmland or received subsidies (Table 9). With the result of correlation matrix in Table 10, we found that the complementary relationships between SAP_S & SAP_PF and SAP_S & SAP_M existed in both large-scale rice farmers and small-scale rice farmers (Table 10). It seems inconsistent with literature, suggesting that small-scale rice farmers are more likely to integrate combinations of SAPs according to their local conditions to maximize the profitability of agricultural production (Cacho et al., 2020; Wale & Yalew, 2007). Some research discussed that they did not jointly adopt specified technologies regarding local governments, agricultural cooperatives, or agricultural extension organizations to achieve economies of scale, as large-scale rice farmers did (Zheng et al., 2022). With the results of Table 11, we found a complementary relationship between joint SAP_S and SAP_PF adoption by rice farmers who receive subsidies but not among those who do not. This situation may be due to risk aversion, as non-subsidized rice farmers were more likely to use farmyard manure in rice cultivation instead of the more expensive SAP_PF, as demonstrated in a study of Tanzanian farmers (Kassie et al., 2013).

Table 9. Heterogeneous effects tests: MVP model estimation.

Variables		SAP_S		SAP_PF		SAP_M
		Parameter	P-values	Parameter	P-values	Parameter	P-values
Farmland scale	≥3	0.100	0.514	0.474**	0.013	0.207	0.208
		(0.153)		(0.190)		(0.164)
	<3	−0.092	0.609	0.275	0.120	0.356**	0.026
		(0.179)		(0.177)		(0.160)
Subsidies	Yes	0.016	0.896	0.518***	0.001	0.209	0.102
		(0.126)		(0.160)		(0.128)
	No	−0.073	0.800	0.014	0.947	0.424*	0.083
		(0.290)		(0.214)		(0.245)
Control variables		Yes		Yes		Yes

Note: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors adjusted for clustering at village level in brackets. The control variables of all these models are the same as in Table 2.

Table 10. Heterogeneous effects of farmland scale: SAPs correlation matrix.

<3  ≥3	SAP_S		SAP_PF		SAP_M
	Parameter	P-values	Parameter	P-values	Parameter	P-values
SAP_S	1		0.511***	0.000	0.188**	0.040
			(0.098)		(0.092)
SAP_PF	0.748***	0.000	1		0.045	0.569
	(0.115)				(0.080)
SAP_M	0.189**	0.027	0.088	0.346	1
	(0.085)		(0.093)

Note: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors adjusted for clustering at village level in brackets. The control variables of all these models are the same as in Table 2.

Table 11. Heterogeneous effects of subsidies: SAPs correlation matrix.

No Yes	SAP_S		SAP_PF		SAP_M
	Parameter	P-values	Parameter	P-values	Parameter	P-values
SAP_S	1		0.595***	0.000	0.207***	0.001
			(0.073)		(0.061)
SAP_PF	−0.009	0.949	1		0.123*	0.077
	(0.135)				(0.070)
SAP_M	0.685***	0.000	0.365***	0.001	1
	(0.194)		(0.108)

Note: *p < 0.1, **p < 0.05, ***p < 0.01. Robust standard errors adjusted for clustering at village level in brackets. The control variables of all these models are the same as in Table 2.

In summary, we found that large-scale rice farmers were more likely to adopt multiple SAPs jointly, while small-scale rice farmers were more likely to adopt different types of SAPs in combination. This may be because larger rice cultivation areas create a cost advantage, and large-scale agricultural production may give rice farmers more time and money to pursue new technologies (Hu et al., 2022). We also found that rice farmers who received subsidies were more inclined to joint adoption of SAPs if they used the internet. The subsidy incentive appears to mitigate the perceived risk of joint adoption of SAPs by rice farmers. These results support hypothesis H3: using the internet to promote joint adoption of SAPs among rice farmers is more likely in large-scale rice farmers or rice farmers who receive subsidies.

5. Conclusion

This study offers valuable insights into the role of internet use in promoting rice farmers’ joint adoption of SAPs. The study theoretically revealed that the internet serves as a new type of farm implement for farmers to acquire adequate information in a fast, low-cost, timely way and facilitate the joint adoption of SAPs. This study draws results from Guangdong province, employing field survey data collected from 844 rice farmers. The empirical analysis using the Ologit model testified internet use facilitation of joint adoption intensity and using MVP verified complementary relationships among the adoption of three types of SAPs. The heterogeneity analysis revealed that ICT-based joint adoption of SAPs is more likely among rice farmers with a larger farmland scale or those receiving subsidies, indicating the influence of resource endowment conditions and institutional incentives on the adoption behaviour.

The findings of this study have three important implications for future practice. Firstly, governments in developing countries should prioritize internet infrastructure in rural areas. Secondly, governments should promote the integration of internet use and the adoption of SAPs in a way that aligns with local conditions. Thirdly, bridging the digital divide in the joint adoption of SAPs is crucial. Governments should promote the moderate-scale operation of agriculture. Finally, incentive policies for SAPs and establish a subsidy policy system to overcome potential market failures.

There are two main limitations of this study. First, the sample was recruited in 2019. Although the data was collected in 2019, the technology adoption in Chinese agriculture has not undergone rapid and drastic changes in recent years. Several studies have indicated that factors such as farm scale, farmers’ technological capabilities, and the overall state of agricultural technology adoption have remained relatively stable over the years (Li et al., 2023; Lin et al., 2022; Liu et al., 2023; Zhang et al., 2023). Meanwhile, in 2019, the internet had rapid development, and the influence of the internet was prevalent during the data collection period (Ding et al., 2022). Therefore, the insights and findings obtained from this survey still hold relevance and significance in understanding the current state in China. Second, the mechanisms of how farmers’ internet use influences their joint adoption of SAPs can be explored in future research.

Acknowledgements

The authors would like to express sincere appreciation to the funding agencies that have supported this research.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research is supported by the Project of the National Natural Science Foundation of China (Project No. 72173047), the Social Science Foundation in Guangdong Province of China (Project No. GD21CYJ16), and the Natural Science Foundation of Guangdong Province of China (Project No. 2022A1515012082).

Notes

1 Here we referred to two important official policy issued by the Chinese government. First, in 2016, the Ministry of Agriculture and seven other departments promulgated ‘Plan for the construction of national Experimental Demonstration zones for sustainable agricultural Development’. It pointed to promoting sustainable agricultural development and a need for the innovation of integrated technologies. For more information, please visit the official website of the Central People's Government of the People's Republic of China: http://www.gov.cn/xinwen/2016-08/30/content_5103492.htm;Second, in 2022, the General Office of the Ministry of Agriculture and Rural Affairs promulgated the ‘Guidance on Promoting the Construction of Ecological Farms’. It pointed out that local governments should strengthen the integration of ecological agriculture technology, promote the integration and support the maturation of individual technologies, products, and equipment.For more information, please visit the official website of the Ministry of Agriculture and Rural Affairs of the People's Republic of China: http://www.moa.gov.cn/nybgb/2022/202203/202204/t20220401_6395140.htm.

2 This data came from the Xinfeng County Statistical Yearbook. It was a comprehensive, systematic, and continuous record of the development of the local economy and society in all aspects. Interested readers can view more detailed information online at this website:https://www.xinfeng.gov.cn/attachment/0/78/78676/1770546.pdf.