Adoption of multiple sustainable agricultural practices among farmers in northwest, China

Abstract

It has become almost a stylized fact that sustainable agricultural practices (SAPs) are central to improving productivity, welfare, and food security. However, SAPs adoption rates have been perceived to be generally low in rural China. Understanding the factors influencing SAPs adoption is critical for sustainable food system development. Drawing on a survey of Chinese farmers, this study examined the determinants associated with adoption decision and intensity of multiple SAPs using a Heckman two-stage model. The findings showed that the adoption of SAPs is a multi-stage sophisticated process resulting from the interplay of several elements, such as individual characteristics, government support, and social networks. Age, education, risk attitude, cognition of SAPs, market access, services from Agricultural Science and Technology Park, policy subsidies, technical training, access to credit, and strong ties were the main determinants encouraging SAPs adoption. Adoption decision also significantly depended on farm size; whereas, adoption intensity relied on off-farm activities, town access, and information channels. This study contributes to the literature by identifying the critical determinants of farmers’ adoption behavior of SAPs in China. Furthermore, our findings are of great value to policymakers as they can target the aforementioned factors to achieve higher SAPs adoption.

Public Interest Statement

Sustainable agricultural practices are considered effective measures to improve human well-being and promote the sustainability of agriculture. However, poor adoption occurs in most parts of the world. This article attempts to investigate the main factors affecting the adoption process of sustainable agricultural practices through a survey of 605 vegetable growers in northwest China, using the Heckman two-stage model. The results indicated that psychological characteristics, government support, and social networks were critical in promoting adoption. These findings have significant policy implications for the development of sustainable agricultural policies.

Keywords:

• Agriculture
• sustainability
• factors
• adoption process
• Heckman model
• China

REVIEWING EDITOR:

• Manuel Tejada, Universidad de Sevilla, Spain

SUBJECTS:

• Agriculture & Environmental Sciences; Conservation - Environment Studies; Sustainable Development; Rural Development

1. Introduction

Agriculture plays a central role in alleviating poverty and promoting development (Ehiakpor et al., 2021; Wainaina et al., 2016). China, a typical resource-constrained country (OECD-FAO, 2023), has been committed to exploring agricultural modernization since the 1980s and has achieved persistent growth in household income and agricultural productivity. Despite remarkable achievements in agricultural economic growth, the pressures of resources and the environment caused by excessive use of resources and chemical inputs, such as land degradation, water scarcity, groundwater pollution, climate change, and soil-borne diseases, have become increasingly prominent (Subedi et al., 2009; Zhang et al., 2013). Thus far, the restriction of agricultural resources and the environment remains one of the main challenges to agriculture and significantly threatens food security and agricultural sustainability in China (Huang et al., 2020; Lu et al., 2015), as well as in other countries (Ehiakpor et al., 2021; Kassie et al., 2013; Manda et al., 2016; Teklewold et al., 2013).

According to statistics, land resources in China have reached a critically low level, with a per capita arable land area of only 0.08 ha. Natural resource constraints have led to the intensive application of purchased inputs in China (OECD-FAO, 2023). China consumes ∼43% of global pesticides on <9% of the global croplands (Zhu & Wang, 2020). Furthermore, available statistics demonstrate that synthetic fertilizer use per hectare of croplands in China is the highest among all regions worldwide, exceeding 300 kg per ha (Hu & Wang, 2022). The excessive application of fertilizers and pesticide use has caused a series of negative environmental externalities, disrupting sustainable agricultural development in China. Official figures reveal that ∼19.6% of the arable land in China has been contaminated (Yan et al., 2016), and high levels of phosphorous and nitrates have also been detected in almost all of the major rivers and lakes in China (Huang et al., 2008).

Sustainable agricultural practices (SAPs) provide the potential to cope with the above-mentioned challenges (Bopp et al., 2019; Manda et al., 2016), and refer to a portfolio of practices with attributes of resources conservation, environment friendliness, technically feasibility, social acceptability, and economic viability (Ehiakpor et al., 2021; FAO, 1995). Specifically, SAPs may be applicable to different agricultural zones and systems, including conservation tillage, soil and water conservation measures, crop rotation, stubble incorporation, improved varieties, organic fertilizers, and integrated pest management (Bopp et al., 2019; Tey et al., 2014). These approaches aim to maximize human well-being while minimizing environmental damage in countries around the world (Ehiakpor et al., 2021; Kassie et al., 2013; Pham et al., 2021; Teklewold et al., 2013; Zeng et al., 2017). Therefore, policymakers have implemented several targeted measures in different regions to popularize SAPs that are adapted to their respective agricultural sustainability issues (Lee, 2010; Pham et al., 2021). The development policy agendas of countries in Sub-Saharan Africa, for instance, have recurrently emphasized the importance of disseminating and adopting SAPs (Kassie et al., 2013; Zeweld et al., 2020). The central government in China has also continuously devoted preferential policies to SAPs promotion.

Despite the potentially extensive benefits of SAPs, enthusiastic attention, and strong encouragement by policymakers to promote SAPs, their adoption has been lower than expected in most countries. For example, recent empirical evidence has observed a generally unsatisfactory adoption rate of SAPs in Sub-Saharan African countries, such as Tanzania (Kassie et al., 2013), Nigeria (Oyetunde-Usman et al., 2021; Wossen et al., 2017), Mali (Mwalupaso et al., 2019), Ethiopia (Zeng et al., 2017), and Morocco (Pilarova et al., 2018). Moreover, other countries, such as Malaysia (Tey et al., 2014), Vietnam (Pham et al., 2021), and China, also appear disappointing adoption rates in SAPs. The investigation by Guo et al. (2021) indicated that the adoption rate of organic fertilizer in China was only 56.8%. Zhang et al. (2021) found that only 33.96% of the sample households adopted balanced fertilization in a survey of 1062 households. In addition, Liu et al. (2019) pointed out that although the Chinese government vigorously promoted SAPs, the adoption rate was generally lower than expected. The low adoption of SAPs by farmers necessitates the determination of the catalysts and inhibitors of their adoption to facilitate their pervasive adoption (Pham et al., 2021).

Recently, considerable efforts have been made to explain SAPs adoption from different viewpoints. The most common approach is to analyze the independent adoption of SAPs, including expounding its determinants in terms of human capital (Pham et al., 2021; Pilarova et al., 2018; Tey et al., 2014; Wossen et al., 2017), physical capital (Kassie et al., 2013; Wainaina et al., 2016), internal perceptions (Arbuckle & Roesch-Mcnally, 2015; Tey et al., 2014) and external elements (Hunecke et al., 2017; Wossen et al., 2017), and evaluating its effect from the perspective of household income, technical efficiency (Mwalupaso et al., 2019), child nutrition (Zeng et al., 2017), and so on. Furthermore, a minority of scholars, considering the limitations of SAPs that are adopted independently, have increasingly shifted their attention to the combined adoption of multiple SAPs and explained it using more sophisticated models (Bopp et al., 2019; Issahaku & Abdulai, 2020; Khonje et al., 2018; Manda et al., 2016). The possible effect of combined adoption has attracted attention in a very small proportion of studies, which revealed a higher potential for SAPs to improve farmers’ welfare (e.g. household income and crop yield) (Khonje et al., 2018; Manda et al., 2016; Teklewold et al., 2013) than single adoption. However, the influence of each factor on adoption typically varies depending on the technical attributes, composite pattern, and specific area, as well as the effect after adoption. Thus, targeted research must be conducted to evaluate SAPs adoption in individual countries (Ehiakpor et al., 2021).

In China, literature focusing on the combined adoption of various SAPs is scanty, but emerging. For example, Li et al. (2017) demonstrated that the adoption intensity of conservation tillage, as well as the binary decision, have significant associations with households’ endowment, social network, and government subsidies, and Zeng et al. (2019) discovered an appreciable effect from external incentives and farmers’ perception on the adoption intensity of pro-environmental agricultural practices. Huang et al. (2020) conducted research on the adoption intensity of soil and water conservation technology in terms of resources endowment and farmers’ cognition of ecology, and Liu et al. (2020) considered society, economy, and nature factors. Evidently, these empirical studies in China contribute to the literature in many ways; however, an overwhelming majority were aimed at food crops, such as rice, maize, and wheat, and few focused on economic crops, particularly vegetables, even though China is the largest vegetable producer and consumer. The study by Yang and Mu (2020) is an exception, whereby it analyzed the intermediary effect of the adoption number of SAPs between information literacy and vegetable growers income; however, the author did not dissect the multidimensional influencing factors of adoption intensity.

In the above studies, some academics shifted their focus to the combined adoption of SAPs in China. However, less attention has been paid to the factors that affect the combined adoption from a multidimensional perspective, particularly in vegetables. This article compensates for the deficiency of the existing literature by providing a diversified perspective on the combined adoption of SAPs based on a survey in northwest China. This study makes two contributions: First, previous studies in China have assessed the specific technology adoption decision for SAPs in vegetables but have not accounted for the combination of different practices. The present study considered technology adoption as a two-stage process and involved a comprehensive and rigorous analysis of the combined adoption of SAPs in China. Second, although there is a well-developed literature on the effect of numerous explanatory variables on technology adoption, this study provides new evidence on plot characteristics, governance indicators, and social networks.

This study aimed to address the following questions: (i) What is the adoption rate of SAPs linked to vegetable production in northwest China? (ii) What factors affect the adoption of SAPs by vegetable growers when adopted independently or in combination? This study makes the following contributions to address these questions. First, we visually compared the adoption rates of various SAPs and different sample counties using ArcGIS software. Second, we regarded farmers’ adoption of SAPs as a two-stage process from adoption decision to adoption intensity and then identified the elements associated with this process from two dimensions: internal and external. The Heckman two-stage model was implemented to eliminate the potential selection bias. The remainder of the article is organized as follows. The conceptual framework is presented in the next section, and the research region, data sources, and econometric methodologies employed in this study are outlined in Section 3. Section 4 presents and discusses the findings, and Section 5 concludes the study and provides policy implications.

2. Theoretical analysis

According to the diffusion of innovation theory, innovation adoption should be conceptualized as a gradual and dynamic decision-making process over a period of time (Doss, 2006; Feder et al., 1985), rather than as an independent and static decision. Nevertheless, limited by the acquisition of panel data, previous empirical studies have typically focused on static descriptions of adoption using cross-sectional data (Dimara & Skuras, 2003). Although the majority of studies simplify adoption to a single-stage dichotomous decision (i.e. adopt or not), some treat adoption as a multistage process occurring in an incremental or stepwise manner with a sequence of sub-decisions ranging from being informed of innovations to actual adoption (Dimara & Skuras, 2003; Noltze et al., 2012; Pannell et al., 2006; Tang et al., 2016). Such multistage studies provide more informative and precise estimates of the adoption process. According to Brown et al. (2017), adoption progress includes a sequence of exposure, non-trial evaluation, trial evaluation, and adoption. An early study by Smale et al. (1995) proposed that adoption is mainly composed of three interrelated choices: adoption, land allocation, and input intensity. Similarly, adoption progress consisting of three phases was outlined in Noltze et al. (2012): whether to adopt (status), how much acreage to use (intensity), and how many components to apply (depth). Further, other empirical studies, including Birhanu et al. (2017), Dimara and Skuras (2003), Feather and Amacher (1994), and Sall et al. (2000), Shiferaw and Holden (1998) consider the adoption process following a two-phase sequential pattern. In our analysis of China, we assumed that individual decisions to adopt SAPs primarily included two stages that occurred simultaneously or successively. The first stage solely serves to explore the adoption decision (AD), defined as a discrete choice of whether to adopt any SAPs. In the next stage, considering that innovations are often adopted in packages (Manda et al., 2016), we interpreted the actual number of SAP components adopted as a measure of adoption intensity (AI).

Theoretically, farmers are the direct adopters and beneficiaries of improved practices. The field theory proposed by Lewin (1951) advocates that individual behavior depends on the interaction between individual characteristics and the external environment. Following this theory, this study believes that farmers’ behavior toward SAPs adoption results from both internal and external factors. Drawing on previous literature, the factors affecting adoption were divided into six categories in this study: individual, psychological, household, geographical, government support, and social networks. The first three characteristics are essential individual characteristics, and the remainder are external environments.

2.1. Influence of individual characteristics on technology adoption

Individual characteristics, including gender, age, education, and respondents’ experience, were valued as pivotal influential factors in determining the adoption decision or intensity elsewhere. As is widely considered, the gender gap must not be overlooked in the face of adopting innovation because potential inequality exists in access to complementary inputs between males and females, with males being in ascendancy (Doss, 2006; Manda et al., 2016; Wainaina et al., 2016). Male managers, therefore, are hypothesized to be more likely to adopt and adopt more SAPs. It is evident that the effect of age on SAPs adoption is remarkable; however, the direction of the influence is controversial. Older people may adopt innovations as a result of ample accumulation in production experience, physical capital, and social capital compared to younger people (Kassie et al., 2013; Manda et al., 2016); however, they may resist adoption because of their excessive adherence to long-term habits. Hence, the effects of age on adoption decision and depth are indeterminate (Kassie et al., 2015). Additionally, the effect of farmers’ experience on the SAPs adoption process is complex in determining a priori. Farmers with extensive experience may be more able to judge whether an innovation is beneficial or suitable for them. They may be reluctant to change and adopt innovations (Donkor et al., 2019) or may tend to adopt and adopt more innovations because of their positive experiences (Embaye et al., 2018). Regarding the education of the respondents, there is a general consensus that well-educated farmers may have more access to technical information (Kassie et al., 2013), a stronger consciousness of sustainable development, and a greater ability to acquire, decode, and implement innovation (Kassie et al., 2015). Education may increase an individual’s ability to acquire, absorb, and implement new technologies. Therefore, we expect farmers with higher education to have a greater possibility and intensity of adoption.

2.2. Influence of psychological characteristics on technology adoption

The considered psychological characteristics were the farmers’ subjective perceptions of risk and SAPs. Conventional wisdom states that the adoption of innovation is often accompanied by risks. Many studies have reported the importance of attitudes toward risk in adopting. They reached an undisputed point that risk aversion may cause slow and low adoption of innovations because of resistance to uncharted inputs and potential risks (Wainaina et al., 2016). Accordingly, we expected risk attitude to affect the adoption decision and intensity of SAPs in the same direction. In terms of the perceived attributes of SAPs, better cognition of the features of SAPs ordinarily signifies that farmers can alleviate the scruples caused by incomplete information with deeper insight into the nature and benefits of innovations and further stimulate enthusiasm for adopting SAPs. Thus, we propose that improving farmers’ awareness of SAPs will promote the possibility and intensity of their adoption. Moreover, perceived usefulness is an essential determinant of adoption, which denotes farmers’ recognition of improvements in innovation performance. As summarized by Gao et al. (2017), when farmers perceive innovations as applicable, their willingness and intensity of adoption are higher. Thus, we assumed that the perceived usefulness of SAPs positively affects adoption decisions and intensity.

2.3. Influence of household characteristics on technology adoption

Household characteristics were captured using four variables: family size, off-farm activity, farm size, and plot fragmentation. The positive effect of family size on the adoption decision and intensity of certain SAPs is well-documented (Darkwah et al., 2019; Kassie et al., 2015; Teklewold et al., 2013). Acknowledging this, we hypothesized that the probability and number of adopted SAPs increase with household size. Regarding off-farm activity, the majority of agricultural households in China engage in off-farm employment. Some such households mainly earn their living from farming; whereas, others regard farming as a sideline, and off-farm work is the principal income source. The 2nd National Agricultural Census in China defined these two household types as farm-dominated and off-farm-dominated households and referred to households specializing in agriculture as full-farm households. Ordinarily, engaging in off-farm activities can enable farmers to obtain additional salaries that can be directed into farming, making innovation inputs more affordable for farm households (Kassie et al., 2013; Ojo & Baiyegunhi, 2020; Thinda et al., 2020). However, farmers with more off-farm work may have limited time and effort for farm operations and less enthusiasm for adoption (Howley, 2013; Kassie et al., 2013). Thus, the effect of off-farm activities on adoption decisions and intensity may be either positive or negative. Farm size, as a pivotal measure of asset ownership, is often considered to be closely associated with adoption (Ojo & Baiyegunhi, 2020). In line with Hunecke et al. (2017), Manda et al. (2016), and Wainaina et al. (2016), we postulate that farm size can increase the likelihood of farm households adopting SAPs, even though farm size has been reported to negatively affect adoption (Marenya et al., 2020). In this study, land fragmentation was defined as the number of vegetable greenhouses owned by farm households. Households with considerable land fragmentation may confront more restrictions when adopting SAPs because of the increase in mechanical operation and transaction costs compared to those with small plot fragmentation (Li et al., 2017; Xie & Huang, 2021). Accordingly, we assumed that plot fragmentation was inversely related to the decision and intensity of SAPs adoption.

2.4. Influence of geographical characteristics on technology adoption

We controlled for geographical characteristics by including plot altitude, market access, and town access. To some extent, plot altitude represents the environmental constraints on vegetable growth, including temperature, solar radiation intensity, and sunshine duration. Previous studies have reported the positive effect of altitude on adoption decisions (Chen et al., 2010). Accordingly, we expected the coefficient sign of the altitude variable to be positive for the adoption decision. Nevertheless, high-altitude plots often encounter difficulties in input purchases, input transportation, and equipment installation. It is reasonable to assume that altitude negatively impacts the intensity of SAPs adoption. The wholesale market is the primary site for vegetable growers in Ningxia to trade agricultural products and exchange crucial information, directly or indirectly. Therefore, we used the computed driving distance to the nearest wholesale market as a proxy for market access because it determines the transaction costs of farmers’ access to information and technologies (Kassie et al., 2015). An increased distance from the wholesale market may mean high transaction costs, poor access to information, and, therefore, lower adoption probability and intensity (Kassie et al., 2013; Oyetunde-Usman et al., 2021). Therefore, market access was expected to discourage SAPs adoption. The effect of town access on adoption is also worth emphasizing, as it was substituted by the shortest driving distance to the town government. Town governments play a prominent role in disseminating agricultural information, projects, and technologies. Hence, town access was expected to negatively correlate with the probability and intensity of SAPs adoption. The distance data from the samples to the corresponding wholesale market and town government were calculated using Amap software.

2.5. Influence of government support on technology adoption

Policymakers have launched various supports to foster efficient SAPs adoption, ranging from policy subsidies to technical training and services from the Agricultural Science and Technology Park (ASTP). Appropriate policy subsidies, whether in monetary form, complementary supplies, or equipment, are often regarded as prerequisites for catalyzing the adoption of a particular innovation (Bopp et al., 2019; Kassie et al., 2013), as they can assist farm households in reducing the expected cost and potential risks of adoption as well as obtaining spiritual incentives. Based on this, we presumed that policy subsidies have a positive impact on farmers’ adoption of SAPs in terms of both decision and intensity. Furthermore, the literature also shows that technical training is a fruitful measure to accelerate farmers’ adoption of innovation by imparting adequate innovation-related information on the nature, benefits, and applications of innovations and enhancing their capability and confidence in adoption (Birhanu et al., 2017; Ehiakpor et al., 2021; Mwalupaso et al., 2019). In addition, we posit that trained farmers are more interested in adopting innovation and most likely adopt it more. In China, the ASTP is a typical form that demonstrates and extends agricultural innovations and practices (Li et al., 2014). The Wuzhong National ASTP, i.e. the first national ASTP established in Ningxia, has been undertaking the task of promoting SAPs for greenhouse vegetable production across Ningxia for many years. The Wuzhong National ASTP can provide farmers with free agricultural technical services, including on-farm demonstrations, online consultations, on-site guidance, and centralized training. Farmers who have received any technical service from the Wuzhong ASTP may understand the innovation more soundly, perceive its benefits more intuitively and thus tend to adopt it more actively. Thus, the variable coefficient was expected to be positive. Credit constraints are often cited as key obstacles to the adoption of agricultural innovations (Balana et al., 2022). Therefore, the relationship between credit constraints and the decision and intensity of SAPs was assumed to be positive in our study.

2.6. Influence of social network on technology adoption

Social networks are frequently mentioned as a crucial consideration in the literature regarding adoption (Birhanu et al., 2017; Hunecke et al., 2017; Kassie et al., 2013; Manda et al., 2016; Noltze et al., 2012; Pham et al., 2021). We drew insights from previous studies and incorporated information channels, interactions with strong ties, and contacts with weak ties as agents of social networks. Access to information cannot be overlooked in the innovations adoption process, as the majority of smallholder decision-makers hold relatively limited information and understanding of innovations (Hunecke et al., 2017; Manda et al., 2016; Pham et al., 2021). Ordinarily, the channels that a farmer relies on to obtain innovation information include peers, extension agents, ASTP, cooperatives, traders, expositions, radio, television, books, and networks. Multiple channels may allow farmers to master sufficient and comprehensive innovation information that can help them eliminate the barriers of credit and resources to a greater extent and make effective decisions (Kassie et al., 2013; Manda et al., 2016). Therefore, we speculated that the possibility and adoption intensity of SAPs increase with the number of channels available for farmers to obtain information. Further, it is well known that the network ties, which were divided into strong and weak ties by Granovetter (2005), are also essential factors affecting adoption. Strong ties are defined as interactions between farmers and their peers (e.g. neighbors, relatives, or friends). Conversely, weak ties are described as connections between farmers and external groups (e.g. extension agents, cooperatives, the ASTP, traders, or exposition) (Granovetter, 2005; Thuo et al., 2014). It is undeniable that both these ties can affect farmers’ information acquisition and innovations adoption, although controversy exists regarding which has the most prominent influence (Conley & Udry, 2010; Granovetter, 2005; Li et al., 2017; Thuo et al., 2014). Strong ties provide farmers with cost-effective and feasible information (Li et al., 2017). For example, as Conley and Udry (2010) confirmed, farmers are likely attempt to adjust their habitual inputs according to the information provided by neighbors with successful experiences. In contrast, weak ties may transmit more novel, non-redundant, and unique information to farmers (Granovetter, 2005), broaden their horizons, and enrich their innovation-related knowledge (Li et al., 2017). Therefore, we envisaged that increased interactions between farmers and either of these ties will improve the likelihood of farmers adopting SAPs and their intensity.

Based on the above discussion, this study proposes the theoretical analysis framework used in the current study, as shown in Figure 1.

Figure 1. Research hypothesis.

3. Data collection and methods

3.1. Data collection

Situated in northwestern China in the upper and middle reaches of the Yellow River (35°14′–39°23′N, 104°17′–107°39′E), Ningxia covers a land area of ∼66,400 km² with an average elevation of above 1000 m and annual precipitation from 167.2 to 618.3 mm. Ningxia has a typical temperate conti­nental climate with a comprehensive daily range of temperatures, abundant sunlight, thermal resources, and superior soil conditions. According to data from the official website of the China Meteorological Administration, the annual sunshine duration in Ningxia is ∼2250–3100 h, total solar radiation is 4950–6100 MJ/m² per year, the annual average temperature is 5.6–10.1 °C, and the daily temperature fluctuates by up to 13 °C. Owing to these exceptional advantages in geographical and climatic conditions, Ningxia has consistently been recognized as an ideal area for greenhouse vegetables in China, with superior quality and high vegetable yields. Since 2007, the Ningxia government has successively implemented numerous preferential measures, including financial support, building production bases, and free training, aimed at developing greenhouse vegetable production into a renewed impetus for increasing farmers’ income. These efforts have resulted in rapid expansion of greenhouse vegetable cultivation areas in Ningxia, followed by problems, such as soil-borne diseases, low soil fertility, and low resource utilization. This necessitates the promotion and adoption of SAPs, along with relevant research.

The study used samples collected through a one-to-one household survey conducted between April and June 2020 among greenhouse vegetable growers in Ningxia. The sample size was calculated to ensure a sampling error of 0.06 using the sampling formula proposed by Scheaffer et al. (1986). The results indicated that the sample size in this study should be at least 278. To collect representative data, a multistage stratified sampling approach was employed to identify six typical counties with high greenhouse vegetable production intensity, purposely selecting three areas from each county and randomly selecting 35–40 households from each area to obtain the final sample. Overall, 640 households were selected across six counties and 18 areas, 605 of which were valid questionnaires collected after data cleaning (accounting for nearly 94.53%), which met the principle of minimum sample size. Figure 2 depicts the locations of the sampled counties sampled and the size of the valid samples in each county. A structured questionnaire containing detailed information on the respondents, households, and plots was administered. Specific items included SAPs adoption, personal characteristics, psychological features, household information, plot characteristics, geographical features, social relations, subsidies, and training. Before the formal survey, a literature review and preliminary survey were conducted to ensure the validity and accuracy of the questionnaire.

Figure 2. Map showing the geographical location of studied counties and the respective areas covered in the data collection.

3.2. Description of SAPs

The SAPs examined in this study refer to a package of common sustainable practices related to greenhouse vegetables in Ningxia, namely straw bioreactors, balanced fertilization, crop diversification, and organic fertilizers. The specific definitions of each SAP are given as straw bioreactors (S) are a sustainable practice that can deal with the problems of high input and environmental pollution with the rationale of rapidly transforming waste straw into necessities for vegetable growth using particular biological strains. This practice provides many benefits to farmers, such as controlling soil-borne diseases, improving soil fertility, and increasing yields. Balanced fertilization (B) emphasizes scientific fertilization based on targeted formulas derived by testing the soil nutrient status. This prevents farmers from inefficient fertilization caused by blind and excessive application, and ensures the sustainability of the soil by reducing nutrient loss. As the core of a sustainable farming system, crop diversification (C) can regulate soil fertility, control pests, and maximize ecological and economic benefits by rationally rotating vegetables with different environmental stress–response characteristics and reaping times on the same land (Kassie et al., 2015; Pham et al., 2021). The organic fertilizers (O) considered here refer to processed and packaged commodity organic fer­tilizers, in particular, which are of great benefit in reducing soil erosion, maintaining soil nutrient balance, and developing sustainable agriculture (Lu et al., 2011).

3.3. Ascertaining explanatory variables affecting SAPs adoption

The selection of explanatory variables in this study relied on a literature review, in which various factors that influence the adoption of improved practices have been emphasized (Darkwah et al., 2019; Doss, 2006; Kassie et al., 2013, 2015; Manda et al., 2016; Ojo & Baiyegunhi, 2020; Teklewold et al., 2013; Thinda et al., 2020; Wainaina et al., 2016). In this study, we classified these expected factors into two broad categories: intrinsic and extrinsic. The considered intrinsic factors were split into three modules: individual, psychological, and household characteristics, and the extrinsic factors included geographical characteristics, government support, and social networks. Table 1 presents the definitions and expected effects of the explanatory variables included in the model. The justification for selecting these variables is as follows.

Table 1. Definitions and excepted effect of explanatory variables included in the model.

Variables	Description	Expected effect
		AD	AI
Dependent variables
Adoption decision (AD)	Whether to adopt any SAPs (1 = Yes, 0 = No)
Adoption intensity (AI)	Adoption intensity [from low intensity (1) to high intensity (4)].
Individual characteristics
Gender	Gender of respondent (1 = male, 0 = female)	+	+
Age	Age of respondent (years) (1 = under 30; 2 = 30–39; 3 = 40–49; 4 = 50–59; 5 = 60 and above)	±	±
Education	Education of respondent (1 = under primary school; 2 = primary school; 3 = middle school; 4 = high school; 5 = university and above)	+	+
Experience	Years of growing greenhouse vegetables (years)	±	±
Psychological characteristics
Risk attitude	Attitude toward the risk of adopting SAPs (1 = risk averter; 2 = risk neuter; 3 = risk taker)	+	+
Cognition of SAPs	Respondent knows SAPs well [5-point scale: from strongly disagree (1) to strongly agree (5)].	+	+
Perceived usefulness	Respondent perceives that SAPs are useful [5-point scale: from strongly disagree (1) to strongly agree (5)].	+	+
Household characteristics
Family size	Family size	±	±
Off-farm activity	1 = non-agricultural households with farm work; 2 = agricultural households with off-farm work; 3 = agricultural households with full-time farm work	±	±
Farm size	Total cultivated are (in mu, 1 mu = 0.067 ha) (1 = under 2; 2 = 2–4; 3 = 4–6; 4 = 6–8; 5 = 8 or above)	+	+
Plot fragmentation	Number of greenhouses	±	±
Geographical characteristics
Plot altitude	Altitude of the plot (in meters)	+	−
Town access	Driving distance to town (in kilometers)	−	−
Market access	Driving distance to the nearest wholesale market (in kilometers)	−	−
Government support
Policy subsidies	Household has obtained subsidies (1 = Yes; 0 = No)	+	+
Technical training	Household has attended free technical training (1 = Yes; 0 = No)	+	+
Service from ASTP	Household has accepted technical services from the Wuzhong ASTP (1 = Yes; 0 = No)	+	+
Access to credit	Difficulty for farmers to obtain credit [5-point scale: from big (1) to small (5)]	+	+
Social networks
Information channels	Number of channels for farmers to obtain information on SAPs	+	+
Strong ties contact	Frequency of strong ties contact [5-point scale: from strongly infrequent (1) to strongly frequent (5)]	+	+
Weak ties contact	Frequency of weak ties contact [5-point scale: from strongly infrequent (1) to strongly frequent (5)]	+	+

3.4. Empirical model

Various models have been developed to assess the adoption of specific innovations. The determinants of a dichotomous choice are generally estimated using Logit or Probit models (Mwalupaso et al., 2019), and those of multiple choice are often assessed using ordinary least squares (OLS), the Tobit model, the multivariate Probit model (MVP), and the Multinomial Logit Selection Model (MNLS) (Khonje et al., 2018). However, it is impractical and biased to estimate the decision and intensity of adoption separately using these methods, principally because these two stages are inseparable. The second stage is conditional on the first. In other words, only when farmers decide whether to adopt SAPs can they choose the number of components to adopt (Birhanu et al., 2017). Therefore, modeling the process in this study required a two-stage estimate that simultaneously accounted for the decision and depth of adoption. Both the Heckman two-stage model and double-hurdle estimation have been demonstrated to have analytical superiority in this regard (Birhanu et al., 2017); however, the former presents a more precise estimation by effectively addressing the endogeneity arising from latent sample selection biases (Heckman, 1979). Accordingly, this study used the Heckman two-stage model (comprising a sample selection and an outcome equation) to identify the factors influencing farmers’ adoption of SAPs.

Following Heckman (1979), we first formulated a binary Probit model following the random utility model as a benchmark for estimating adoption decisions. We then exploited a modified OLS model to estimate the determinants of adoption intensity, where the inverse mills ratio (IMR) was incorporated as an additional independent variable to correct the sample selection bias. Here, referring to Li et al. (2017), we considered the adoption intensity as a continuous interval variable from one to four.

The first stage assumes the existence of an underlying relationship as follows: $$Y_{\mathit{\text{da}}}^{*}=\alpha X_{\mathit{\text{da}}}+{\epsilon }_{\mathit{\text{da}}}$$ where $Y_{\mathit{\text{da}}}^{*}$ is an unobserved variable that measures the conditional probability that an individual decides to adopt SAPs. That which is observed can be expressed as: $$Y_{\mathit{\text{da}}}^{}=\left\{\begin{array}{c}1,\text{if}\hspace{0.17em}{Y}_{\mathit{\text{da}}}^{\ast }>0\hspace{0.17em}\hspace{0.17em}\\ 0,\text{if}\hspace{0.17em}{Y}_{\mathit{\text{da}}}^{\ast }\le 0\hspace{0.17em}\end{array}\right.$$ where Y_da denotes a dichotomous variable that takes the value of one if a vegetable grower adopts any one of the aforementioned SAPs; otherwise it is zero. X_da refers to a vector of independent variables associated with adoption decisions. α indicates a vector of parameter coefficients to be estimated, reflecting the effect of the independent variable on the adoption decision. ε_da is a normally distributed error term.

The algebraic representation of the second stage model is as follows: $$Y_{\mathrm{i}a}^{}=\beta X_{\mathit{\text{ia}}}^{}+\rho \sigma \mathit{\text{IMR}}+{\epsilon }_{\mathit{\text{ia}}}$$ where Y_ia is a continuous interval variable representing the adoption intensity measured by the number of SAPs adopted by the ith vegetable grower. Y_ia is only observed when Y_da equals 1. Y_iagradually increases from one to four, indicating a gradual increase in adoption intensity from low to high. X_ia shows a vector of explanatory variables for the adoption intensity, β is a vector of parameter coefficients to be estimated, and ε_ia is the error term. ρ denotes the correlation between ε_da and ε_ia. σ is the standard deviation of ε_ia. IMR contains information on the unobserved factors affecting the first decision and helps improve the parameter estimates and correct selection bias in the second stage (Birhanu et al., 2017); it was obtained from the first stage, and the formula is as follows: $$\mathit{\text{IMR}}=\frac{\phi (\alpha X_{\mathit{\text{da}}})}{\phi (\alpha X_{\mathit{\text{da}}})}$$ where φ and ϕ are the normal density function and the cumulative density function of a standard normal distributed variable, respectively. Sample selection bias exists if IMR is statistically significant. Hence, the OLS regression may bring about a biased estimate, which can be effectively corrected by the Heckman two-stage model, indicating that it is appropriate for estimation.

4. Results and discussion

4.1. Descriptive statistics

Table 2 provides the descriptive statistics of the explanatory variables included in the model. The results showed that our sample comprised 57.69% males and 42.31% females, with an average age of ∼50 years and a mean of 9.32 years of planting experience, ranging from 1 to 26 years. A relatively large proportion of the sample graduated from elementary and junior high school, accounting for 61.16% of the total, and there were no participants with university degrees or above. Adopters had a higher level of education than non-adopters. This implied that the growers were predominantly middle-aged people with lower educational levels. In addition, men still occupied a dominant position in small-scale agricultural production in China, although previous studies have shown that the off-farm transfer phenomenon of rural male labor is increasingly prominent (Xie & Lu, 2017). The average household size was ∼4, with minimum and maximum sizes of one and eight members, respectively; households with three to five people accounted for the most significant proportion (83.64%). This finding conforms to the demographic characteristics of rural households in China. Approximately 74.05% of the respondents were engaged in off-farm activities, and 42.86% considered off-farm activities as their dominant economic source. It can be seen that participating in part-time activities has become common in Ningxia. Households held three vegetable greenhouses on average (ranging from 1 to 22), of which three greenhouses or fewer accounted for the largest proportion (73.55%). The average number of greenhouses owned by adopters (three) was significantly higher than that owned by non-adopters (two). The samples were primarily vegetable growers with a total cultivated area of <4 mu (also known as small-scale growers) (Zhou et al., 2012), accounting for 71.24%. Only 42 households operated over eight mu (also known as large-scale growers) (Zhou et al., 2012), accounting for 7.44% of the total sample. Furthermore, ∼63.31% of the sample had a high awareness of SAPs and 77.19% perceived that SAPs were helpful. Most participants had concerns regarding the risks of technology adoption and 17.19% were risk-takers.

Table 2. Summary statistics of explanatory variables included in the model.

Variables	Whole sample				Adopter	Non-adopter	Mean diff
	Mean	SD	Min	Max	Mean	Mean
Individual characteristics
Gender	0.577	0.495	0	1	0.585	0.551	0.034
Age	3.474	0.943	1	5	3.488	3.428	0.061
Education	2.344	1.002	1	4	2.415	2.101	0.314***
Experience	9.322	5.388	1	26	9.263	9.522	−0.258
Psychological factors
Risk attitude	1.726	0.737	1	3	1.842	1.333	0.508***
Cognition of SAPs	3.691	1.037	1	5	3.824	3.239	0.585***
Perceived usefulness	3.853	1.263	1	5	3.891	3.725	0.166
Household characteristics
Family size	3.980	1.087	1	8	3.957	4.058	−0.101
off-farm activity	1.942	0.758	1	3	1.991	1.775	0.216***
Farm size	2.205	1.146	1	5	2.300	1.884	0.416***
Plot fragmentation	3.117	2.501	1	22	3.268	2.605	0.663***
Geographical characteristics
Plot altitude	1141	41.616	1088	1233	1140	1144	−4.011
Town access	5.589	3.320	0.380	16.510	5.427	6.138	−0.711**
Market access	10.858	7.183	1.830	32.900	10.332	12.640	−2.308***
Government support
Policy subsidies	0.855	0.353	0	1	0.923	0.623	0.300***
Technical training	0.666	0.472	0	1	0.728	0.457	0.272***
Service from ASTP	0.106	0.308	0	1	0.131	0.022	0.109***
Access to credit	2.704	1.335	1	5	2.979	1.775	1.203***
Social networks
Information channels	2.312	1.090	1	6	2.450	1.848	0.602***
Strong ties contact	3.145	1.145	1	5	3.287	2.667	0.620***
Weak ties contact	2.878	1.177	1	5	3.015	2.370	0.645***

diff = mean (Non-adopter) − mean (Adopter). H0: diff = 0; HA: diff ≠ 0. Significance: ***p < 0.01, **p < 0.05.

Regarding external factors, the average altitude of the sample is 1141 m, ranging from 1088 to 1233 m. Driving distance from the samples to the town government ranged from 0.38 to 16.51 km with an average of 5.59 km, and the average distance to the wholesale market was 10.86 km (0.38–16.51 km). 85.45% of the sample had been subsidized by the local government through agricultural materials or funds, and more adopters received subsidies than non-adopters. Moreover, a significant proportion of the sample have received technical training related to SAPs, and adopters acquired more technical training than non-adopters, which was statistically significant. More than half of the sample believed that it was difficult to obtain credit, indicating that credit access remained a significant obstacle in the rural areas of Ningxia. Furthermore, the data showed that 64 samples received technical services, such as on-site guidance, base demonstration, and centralized training from the Wuzhong ATSP. Regarding social networks, on average, farmers obtained technical information from 1 to 6 sources, and adopters had more access to extension services and information regarding SAPs than non-adopters. In addition, the samples had more frequent interactions with strong ties and less interaction with weak ties (average values of 3.15 and 2.88, respectively). We also observed that social connections differed statistically between adopters and non-adopters; i.e. adopters had stronger and weaker ties than non-adopters.

Overall, of the 605 samples considered in the analysis, ∼77.19% benefited from at least one SAP considered above, and the remaining 22.81% did not adopt any SAPs. Despite the high adoption rate, it is worth notable that all four SAPs were adopted simultaneously in only 32 households among all adopters, accounting for only 5.29% of the total samples. Additionally, the majority of the samples (55.70%) adopted two or three SAPs, and 98 households employed only one SAP; of these, 52.04% adopted organic fertilizers, 35.71% received crop diversification, 11.22% adopted straw bioreactors, and only one sample adopted balanced fertilization. Figure 3 presents the adoption of SAPs disaggregated by county, and shows that there were similarities and differences in the combined adoption of SAPs in the sampled counties. Litong had an overwhelmingly high adoption rate of 89.32%, indicating that 89.32% of the sampled households in Litong had adopted at least one SAP. One plausible explanation for the high adoption rate in Litong is that it is the location of the first National ASTP in Ningxia, making it easier to receive radiation from the ASTP and adopt more SAPs. Conversely, Lingwu had the lowest adoption rate (66.02%) and was the only county that did not adopt all four SAPs concurrently. This may have been due to the lack of straw and information resources required to adopt SAPs in Lingwu. The adoption of SAPs in Qingtongxia, Yongning, Helan, and Shapotou was equivalent, with adoption rates of 79.41, 75.25, 75.76, and 77.32%, respectively. Remarkably, more than half of the farmers in Litong, Lingwu, and Helan concentrated on adopting one or two SAPs concurrently; whereas, Qingtongxia, Shapotou, and Yongning were dominated by households that adopted two or three SAPs simultaneously.

Figure 3. Distribution of adoption.

Coming to the adoption of specific SAPs, Figure 3 shows certain differences in farmers’ adoption of different SAPs. Crop diversification was revealed to be the most universally accepted, with comprehensive adoption rates of 65.12%, and balanced fertilization had the lowest adoption rate of 13.55%. This evidence can be preliminarily interpreted as a higher input of balanced fertilization compared with crop diversification, which makes farmers hesitate to adopt balanced fertilization. In addition, our survey found that vegetable growers in Ningxia generally had a low awareness of balanced fertilization, which may have reduced its adoption rate. The adoption rates of organic fertilizer and straw bioreactors were nearly equivalent at 47.77 and 36.36%, respectively. The results presented in Figure 3 indicate that crop diversification was the most widely adopted SAP in every county, except Lingwu, which had the highest organic fertilizer adoption rate. In addition to the high-quality adoption of crop diversification, higher adoption rates were also observed for organic fertilizers in Litong and Helan, and for straw bioreactors in Qingtongxia and Yongning. Our investigation found that the governments in Qingtongxia and Yongning made significant efforts to promote straw bio-reactors. Therefore, we can preliminarily conclude that government support may be a key driving force in promoting SAPs adoption. In addition, these two counties have sufficient straw resources to provide a material basis for straw bioreactors. Most notably, balanced fertilization was the SAP with the lowest adoption rate in all sampled counties except Lingwu, and the straw bioreactor was the least adopted SAP in Lingwu, with only four households. One possible explanation for the widespread low adoption rate of balanced fertilization is that farmers were generally concerned about the high cost of formula fertilizers. According to our investigation, the limitation of crop straw may have led to this low adoption rate in Lingwu, as straw is the necessary material for straw bioreactors.

4.2. Determinants of adoption

In this study, Stata software (version 13.0) was used to estimate the parameters of the Heckman two-stage model and identify the factors that affected the adoption decision and intensity of SAPs by the sampled households. The results are summarized in Table 3. The model passed the chi-squared test at a significance level of 1%, indicating that it fit reasonably well. We report the coefficients of the explanatory variables, standard deviations, and IMR (lambda). Lambda was found to be statistically significant and positive at the 1% level, suggesting a selection bias in the sample, and that the Heckman two-stage model was suitable for our analysis. Finally, we conducted a variance inflation factor (VIF) test among the independent variables and the IMR to ensure the validity of the results and avoid multicollinearity problems. The maximum and mean VIF were 2.95 and 1.04, respectively, indicating that there was no multicollinearity problem.

Table 3. Estimated results of Heckman two-stage model.

Variables	Adoption decision		Adoption intensity
	Coefficient	Std. Err	Coefficient	Std. Err
Individual characteristics
Gender	−0.1925	0.1564	−0.0.771	0.0643
Age	0.2461**	0.1058	0.1597***	0.0438
Education	0.1845*	0.0992	0.0968**	0.0401
Experience	−0.0333**	0.0154	−0.0260***	0.0067
Psychological factors
Risk attitude	0.2531*	0.1330	0.3191***	0.0469
Cognition of SAPs	0.2391***	0.0743	0.1205***	0.0338
Perceived usefulness	0.0320	0.0571	0.0213	0.0231
Household characteristics
Family size	−0.0632	0.0661	−0.0055	0.0281
Off-farm activity	0.0811	0.1239	0.1113**	0.0483
Farm size	0.2450**	0.1194	0.0351	0.0420
Plot fragmentation	0.0123	0.0495	−0.0078	0.0182
Geographical characteristics
Plot altitude	−0.0007	0.0019	−0.0002	0.0008
Town access	−0.0224	0.0252	−0.0191*	0.0112
Market access	−0.0230*	0.0119	−0.0120**	0.0052
Government support
Policy subsidies	0.8112***	0.1798	0.5864***	0.1339
Technical training	0.3322**	0.1605	0.1908**	0.0760
Service from ASTP	0.7249*	0.4212	0.3870***	0.0972
Access to credit	0.4561***	0.0701	0.1841***	0.0319
Social networks
Information channels	0.1002	0.0860	0.0860**	0.0305
Strong ties contact	0.1738**	0.0669	0.1898***	0.0303
Weak ties contact	0.0461	0.0698	0.0184	0.0289
Lambda	—	—	0.5331**	0.1915
_cons	−3.5502	2.2650	−1.5984*	0.9565
Rho	0.8359
Wald chi^2(24)	195.66
Prob > chi^2	0.0000

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

4.2.1. Factors associated with adoption decision

This section reports the results of the first-stage estimation using the Heckman two-stage method. The estimated correlation coefficients were statistically significant in 12 terms for the adoption decision, two of which were negative and the remainder of which were positive. These terms included age, education, experience, risk attitude, cognition of SAPs, farm size, market access, policy subsidies, technical training, service from ASTP, access to credit, and strong ties contact. Experience and market access negatively affected adoption decisions, whereas the other terms positively affected adoption decisions.

Concerning personal characteristics, age was found to have a positive effect on the decision to adopt SAPs at the 1% significance level, showing that older farmers are more inclined to adopt SAPs than younger farmers. This result corresponds to that of Ehiakpor et al. (2021), which unveiled that younger farmers were less likely to apply animal manure to their maize plots. Education had statistically significant and positive effects on the decision to adopt SAPs, reflecting that highly educated farmers were more willing to adopt SAPs. The probability of adopting SAPs increased with farmers’ education levels because of their greater awareness of the availability and benefits of new agricultural technologies. This finding is in keeping with the previous research in this field, which found that education plays a critical role in encouraging farmers to adopt innovation (Asfaw & Admassie, 2015). This is also supported by Wainaina et al. (2016), who indicate that better-educated farmers have fewer capital constraints and higher time opportunity costs. In addition, farmers’ experience in growing greenhouse vegetables was negatively related to their decision to adopt SAPs, indicating that seasoned farmers showed less disposition to adopt SAPs than inexperienced farmers. This observation is consistent with Embaye et al. (2018), who found that farmers with more farming experience have a lower probability of growing oil seed crops. However, this is contrary to the findings of Donkor et al. (2019), who reported that experience in cassava farming had a positive effect on the adoption of mineral fertilizers. This is most likely because farmers who have accumulated more experience tend to carry out agricultural production according to their ingrained habits and traditional practices rather than introducing innovation, whereas farmers who lack experience prefer to optimize their agricultural production by adopting innovations.

Regarding psychological factors, risk attitude had a positive and significant impact on the decision to adopt SAPs, with a significance level of 10%, suggesting that risk-takers were more likely to adopt SAPs. This finding is supported by Wainaina et al. (2016), who highlight that risk aversion may lead to slower and lower adoption of agricultural technologies, particularly when it comes to inputs that need to be purchased. Seminal studies on this subject also support our findings (Bopp et al., 2019). They indicated that risk perception control fosters sustainable agriculture. A positive and significant effect of farmers’ cognition of SAPs on adoption decisions at the 1% significance level was also revealed, signifying that farmers with better cognition of SAPs had a higher probability of adopting them. This finding is plausible because the enhancement of technology cognition can alleviate farmers’ risk aversion and fear of difficulties, thereby promoting the adoption of new technologies. A similar viewpoint was reported by Liu et al. (2020), who concluded that farmers’ willingness to adopt soil and water conservation increases with improvements in technology cognition. We did not find a significant effect of perceived usefulness on adoption decisions. A possible explanation for this is that there were no significant differences in terms of perceived usefulness between the groups, meaning that farmers generally believed that SAPs were beneficial for agricultural production.

Consistent with our expectations, the results further showed that farm size was positively associated with the adoption decision of SAPs at a 5% significance level, implying that the likelihood of adoption increased as farm size increased. This finding corroborates existing evidence that those possessing greater endowment with farmland are more likely to invest in improved practices (Xie & Huang, 2021), partly because they enjoy lower opportunity costs, more access to economies of scale, and a greater capacity to manage the potential risks brought about by adopting technologies. Rich farmland resources provide farmers with more opportunities to experiment with improved practices. This agrees with earlier studies on the adoption of climate change adaptation strategies in South Africa (Thinda et al., 2020), integrated pest management in Iran (Allahyari et al., 2016), and irrigation technologies in Chile (Hunecke et al., 2017).

Regarding geographical characteristics, only market access was found to have a negative association with the adoption decision of SAPs, with a significance level of 10%, revealing that adoption was more likely to occur on farm plots closer to the wholesale market. Often, proximity to the wholesale market allows farmers to enjoy more opportunities to contact extension agencies, where they can obtain sufficient technical information and guidance (Ojo & Baiyegunhi, 2020). Meanwhile, the increased distance to the wholesale market may discourage the probability of adopting SAPs because of the increasing transaction costs of accessing the input and output markets, which confirms the conclusion of Makate et al. (2019). Their findings identified that a greater distance to the market could discourage the adoption of conservation agriculture. A similar pheno­menon has been observed in countries, such as Ethiopia, where market access is considered one of the most critical factors in adoption decision (Cavatassi et al., 2011).

Government support played an essential role in the decision to adoption SAPs, as all variables in this section passed the significance test. In particular, the positive and significant coefficient of the government subsidy variable revealed that households that received government subsidies had a stronger tendency to adopt SAPs. This finding is consistent with the results of many studies that demonstrate the critical role of government incentives in promoting the adoption of agricultural technologies (Katengeza et al., 2018). In general, government incentives, such as financial incentives, and agricultural material subsidies, can greatly reduce the potential cost of technology adoption and stimulate farmers’ enthusiasm for adoption, thereby enhancing the possibility of technology adoption. Moreover, the positive coefficient of access to credit indicated that credit constraints reduced the likelihood of adopting SAPs. A plausible interpretation of this result is that by lowering the credit threshold, the risk associated with technology adoption is reduced, and farmers are more likely to adopt SAPs in their farming practices. This finding aligns with Ali (2021), who reveals that access to credit significantly positively affects the adoption of improved upland rice. The coefficients of services from the ASTP also showed an expected positive sign at the 1% significance level, indicating that farmers who had received services from the ASTP were more motivated to adopt SAPs. This finding supports the prevailing notion that the ASTP plays a crucial role in promoting SAPs adoption (Mo & Yu, 2020). Technical services offered by the ASTP, such as technical training, experimental demonstration, knowledge dissemination, and field guidance, can enhance farmers’ comprehension of SAPs and provide them with practical experience of their adoption benefits. This result further underscores the importance of technical training in explaining SAPs adoption, showing that technical training was significantly positively correlated with adoption decision at the 5% significance level. This result is identical to our expectation that the probability of SAPs adoption would be high for households that had participated in technical training related to SAPs, and is consistent with studies carried out by Asfaw and Neka (2017) and Mengistu and Assefa (2019), who found that training was positively associated with the adoption of watershed management and soil and water conservation practices. Hence, as an essential avenue for farmers to learn and communicate with technology, technical training can effectively improve farmers’ cognition of SAPs and stimulate their enthusiasm for their adoption.

Regarding social networks, the results confirmed our hypothesis that strong ties positively affected SAPs adoption decision, revealing that farmers who interacted more frequently with their neighbors, relatives, and peers were more likely to adopt SAPs more promptly. This is congruent with the study by Li et al. (2017) on the adoption of conservation tillage technology, which highlighted that frequent connections with strong ties can facilitate adoption behavior by providing emotional support and trust. One explanation for this finding is that strong network ties lead to more effective and efficient work, along with an effective way to cope with risks (Hunecke et al., 2017). Besides that, in the presence of incomplete information, strong ties composed of farmers’ relatives, friends, and neighbors may become the primary source of information because of the generally low educational level of farmers (D’Souza & Mishra, 2018). Frequent homogeneous network connections can facilitate adoption behavior by providing emotional support and trust (Li et al., 2017).

4.2.2. Factors associated with adoption intensity

Regarding the determinants of adoption intensity, the results of the second step reported in Table 3 showed that 14 indicators had significant effects on adoption intensity, of which 11 had positive effects, including age, education, risk attitude, cognition of SAPs, off-farm activity, policy subsidies, technical training, services from ASTP, access to credit, information channels, and strong-tie contact. In contrast, experience, town access, and market access had negative impact. Notably, adoption intensity was not appreciably influenced by household and farm characteristics, such as family size, farm size, and plot fragmentation, which may have been due to the slight differences in these data between the two groups.

The sign for age was positive and significant, meaning that adoption intensity increased with age. This contradicts the findings of Alem et al. (2010), which unveiled that younger farmers may have longer planning horizons and greater enthusiasm for investing in SAPs. One possible explanation for this trend could be that older farmers accumulated more social and physical capital (Kassie et al., 2013). On one hand, farmers’ financial situation continues to improve as they grow older, making it conducive for them to adopt more technologies. On the other hand, older farmers may have a deeper understanding of the value of innovation because of the accumulation of social capital, leading them to adopt more innovation. Education also plays a crucial role in adoption intensity because there is a positive and significant correlation between these two items. As posited by Manda et al. (2016), education enables farmers to interpret technical information and appreciate the importance and benefits of adopting innovation. This finding is also in line with the conclusion reached by Thinda et al. (2020), which indicated that education is a significant determinant of the adoption intensity of climate change adaptation strategies. Further examination showed that experience negatively affected the adoption intensity of SAPs with a 1% significance level, indicating that farmers with less experience may adopt more SAPs. Veteran farmers may tinge on empiricism in the process of agricultural production, preventing them from adopting more SAPs. Consistent with this notion, Thinda et al. (2020) concluded that farming experience is a significant incentive for enhancing farmers’ adaptive capacity by adopting multiple adaptation approaches. Furthermore, inexperienced farmers may adopt more SAPs because they have more energy and enthusiasm to attempt more SAPs.

Risk attitude was positively correlated with adoption intensity at the 1% significance level, suggesting that risk-takers were likely to adopt more SAPs. This positive effect is conceivable because risk takers may be eager to explore innovation and are more likely to obtain technical information through continuous learning. This result broadly agrees with the findings of Ghadim et al. (2005), who prove that risk aversion reduces adoption intensity. Our investigation found that risk-takers ordinarily focused more on the benefits of technology adoption than on the risks involved. In addition, their optimism towards agricultural innovation made them inclined to constantly adopt innovations. Nevertheless, for those wary of risks, the potential uncertainties in technology adoption may lead to the adoption of less or at the latest. Besides that, the cognition of SAPs was also proven to positively affect adoption intensity, with a significant difference of 1%, indicating that farmers who have a greater understanding of SAPs tend to adopt them more. This is consistent with the findings of Li et al. (2017), who found a significant positive relationship between the degree of understanding of conservation tillage and the adoption intensity. This result was also confirmed in Gansu (Huang et al., 2018), where the cognition of irrigation water-saving technologies is the basis of adoption intensity.

Among household characteristics, the influence of off-farm activity on adoption intensity was positive, with a significance level of 5%, which was in line with our expectations. This means that, for adopters, the probability of adopting more SAPs was higher in households without off-farm activities. Contrary to our findings, Issahaku and Abdul-Rahaman (2019) found that participation in off-farm work significantly increased the adoption intensity of organic manure because of the income effect of participation in off-farm activities. This may have been related to the fact that households with off-farm activities mostly undertook farming as a sideline and were not willing to invest excessive cost and energy in technology adoption; whereas, households without off-farm mostly expressed higher enthusiasm to adopt SAPs and tended to adopt more SAPs to improve their economic benefits.

In terms of geographical location variables, the parameters for market access and town access were both negative at significance levels of 1 and 5%, respectively, implying that an excellent location could promote deeper adoption of SAPs to a certain extent. Our investigations revealed that farmers in Ningxia mainly sold greenhouse vegetables through door-to-door acquisition by vegetable vendors, wholesale markets, and collection point acquisition. Although door-to-door acquisition provides farmers with great convenience, farmers located farther away from the wholesale market still incurred higher transport costs to acquire information and inputs; thus, they adopted less SAPs. This observation is in accordance with previous findings undertaken in previous studies (Teklewold et al., 2013), in which it was found that market access negatively influences smallholders’ adoption intensity of SAPs. The negative impact of town access on adoption intensity indicated that households with convenient access to towns were more likely to adopt SAPs more. One plausible explanation is that farmers close to towns had more opportunities to contact agricultural extension agencies, obtain technical information and guidance, and adopt more SAPs. In addition, proximity to towns can minimize transportation costs, which in turn encourages farmers to adopt more.

Government support, as anticipated, was a pivotal factor affecting SAPs adoption intensity because all relevant variables have noticeable effects on adoption intensity. Table 3 shows that adoption intensity is positively related to policy subsidies, which further corroborates the hypothesis that government incentives are a critical element in facilitating SAPs adoption. This is largely because the supply of cash or SAP-related technical materials, such as organic fertilizers and microbial strains, can partially shift the cost burden associated with adopting SAPs for farmers, maximize their relative returns, and stimulate their enthusiasm to adopt more SAPs. This discovery is in conformity with findings reported by Bopp et al. (2019) and Kassie et al. (2013).

Moreover, the parameter for technical services from the ASTP was positive and significant at the 1% level, indicating that adopters receiving technical services from the ASTP were more inclined to adopt in-depth services. Probably because technical services from the ASTP, such as theoretical training, centralized demonstration, and practical guidance, allowed farmers not only a deeper comprehension of the principles of SAPs and more intuitive observation of the benefits of adoption but also to grasp the ability to adopt SAPs and further promote the adoption of innovation. Another possible explanation is that adopters who obtained technical services from the ASTP were mostly professional farmers or local technology leaders. Unlike traditional farmers, these farmers have more solid theoretical knowledge and a stronger ability to accept improved technologies. Receiving technical services from the ASTP can further enhance their cognition of improved technologies and promote their adoption of more SAPs.

As expected, farmers with easy access to credit were more likely to adopt more SAPs, with the regression results showing a significant positive correlation between access to credit and adoption intensity at the 1% significance level. This result highlights the fact that credit constraints significantly restricted farmers’ deep adoption of improved technologies. This finding confirmed the previous research of Ali (2021), which found that access to credit is crucial for the deep adoption of climate adaptation technologies in Togo. For smallholders, lowering the credit threshold can provide sufficient financial security, weaken the risk and uncertainty in their agricultural production process, and increase the possibility of deep adoption. Generally, limited access to agricultural credit is a major impediment to the adoption of agricultural technologies in most countries. Thus, government extensions of agricultural technology should focus more on optimizing the credit environment.

Technical training was significantly and positively correlated with adoption intensity, as expected, symbolizing that farmers who received technical training were more likely to adopt multiple SAPs. Farmers often have limited education and poor cognitive abilities towards new concepts. Technical training can effectively shorten the time required by farmers to search for technical information and learn new technologies, and therefore has a significant incentive effect on improving farmers’ technological literacy, updating their technological knowledge, and enhancing their willingness to adopt more. According to Li et al. (2017), farmers who participate in technical training have a high level of understanding and awareness of their economic benefits of conservation tillage technology, which promotes the adoption of conservation tillage technology.

Among the social network characteristics, information channel was reported to have a significant positive association with adoption intensity at the 1% significance level, implying that diversified channels for obtaining information can help promote the adoption intensity of SAPs. This result aligns with previous research on water-saving irrigation (Qiao et al., 2017), which found that increasing access to information can help improve the intensity of water-saving irrigation adoption. Increased access to information enables farmers to mitigate the cost escalation arising from incomplete information and make more scientifically informed decisions regarding SAPs adoption in an efficient information environment, leading to higher intensity of SAP adoption. Thus, broadening farmers’ information channels is helpful for improving the intensity of technology adoption. Notably, the broadening of information channels did not affect farmers’ adoption decision. One possible explanation is that farmers who had already adopted SAPs had initially realized the benefits of adopting SAPs and had a positive and open attitude towards technology adoption. When information channels were further expanded, and they obtained more technical information, they chose to adopt more improved technologies to improve production.

In addition, the hypothesis that strong-ties contact is significantly correlated to adoption intensity was also supported by the results, which demonstrated the importance of informal networks in promoting in-depth SAPs adoption. This positive effect indicated that the probability of farmers adopting more SAPs may increase as the frequency of communication with peers increases, which can be explained to some extent by farmers’ trust in their peers (e.g. neighbors, relatives, or friends) and their multitude-followed mentality. As pointed out by Hu (2016), strong ties based on trust and normative constraints can facilitate high-intensity sharing of technical information and knowledge, leading to deep adoption behavior.

5. Conclusions and implications

This study analyzed the determinants of the adoption of four interdependent SAPs in northwest China using data obtained from a sample of over 600 farm households. The Heckman two-stage model was used to estimate the main factors affecting the adoption process and to control for self-selection bias.

Three conclusions were drawn from this study, which provide novel contributions to the literature regarding the determinants of SAPs adoption behavior. First, although 77.19% of the sample adopted at least one of the aforementioned SAPs, the combined adoption rate was not ideal, with only 32 households simultaneously adopting four SAPs. Second, SAP adoption is a multi-stage intricate process resulting from interactions between multiple factors, such as individual characteristics, government support, and social networks. The critical factors shaping the adoption decision and intensity of SAPs are somewhat heterogeneous. Third, government influences, such as services from the ASTP, policy subsidies, technical training, and access to credit, were consistent determinants of SAPs adoption decision and intensity. In line with our theory, the risk attitude and cognition of SAPs were important psychological factors affecting this two-stage adoption process. Frequent connections with strong ties significantly increased the likelihood and intensity of adoption. In addition to common factors, adoption decision was also closely related to farm size, and adoption intensity was significantly influenced by off-farm activity, town access, and information channels. These results are relevant in both practice and research. For researchers, a better understanding of the factors influencing SAPs adoption can help build and predict future adoption. For policymakers, this information can serve as a reference for adjusting policy measures to achieve higher SAPs adoption and promote local agricultural sustainability.

Based on the above results, the following recommendations can be proposed to guide agricultural policies.

First, the risk attitude and cognition of SAPs significantly and positively influenced their adoption. These results highlighted the importance of popularizing knowledge of agricultural risk and information on SAPs to promote their adoption. It is, therefore, essential to strengthen the dissemination of potential risks in agricultural production and knowledge related to agricultural insurance so that farmers can maintain correct risk attitudes when making decisions. Policymakers should also perfect the risk diversification mechanism to avoid the potential production risks caused by technology adoption. In addition, multi-channel and multi-form publicity related to SAPs should be implemented to enhance farmers’ awareness and accelerate the promotion of SAPs.

Second, the results of this study suggested that government support, including policy subsidies, technical training, services from the ASTP, and access to credit, had a significant bearing on adoption. Efforts to increase government support may be of practical value, as it is the primary pathway for promoting adoption behavior. Specifically, the ASTP should frequently provide diversified technical services to stimulate farmers’ enthusiasm for adoption. Local governments are expected to increase funding or material subsidies to alleviate economic pressure on farmers and develop enthusiasm for technology adoption. The frequency of technical training should be increased to ensure that all farmers receive the training. Necessary incentive measures can be taken for farmers who are unwilling to participate in technical training. Moreover, the government should broaden access for farmers to obtain credit, provide relatively stable support for farmers to obtain credit, and alleviate constraints on the amount, interest rate, and term of agricultural loans for farmers.

Third, given that social networks are as essential as government support for adoption, we strongly recommend that policymakers should strengthen the construction of social networks. Farmers with powerful social cachets and extensive social contacts can be selected as crucial targets for technology promotion. Inclined support can be provided to cultivate these individuals as demonstrators. Technological information can then be rapidly disseminated by relying on farmers’ extensive influence within social networks. Furthermore, great emphasis should be placed on the role of popular media, such as WeChat groups, short video platforms, and webcasts in the Internet era. Although the impact of weak ties on adoption was not prominent, attention should be paid to the advantages of weak ties in dissemina­ting heterogeneous and high-value technological information.

Our study provides insights into the factors contributing to the behavior of vegetable growers in China. However, two aspects require further research for improvement. First, technology adoption is a dynamic process. This study adopted the Heckman two-stage model to reveal the multi-stage characteristics of technology adoption. However, reflecting the dynamic characteristics of a time series remains challenging. Thus, tracking surveys over a continuous period needs to be conducted in the future to form panel data, laying the foundation for future research on topics, such as dynamic changes in adoption behavior. Second, the selection of investigation areas for this study was limited owing to time constraints. Although the research data fully satisfied the representativeness and typicality requirements, the re­search scope could be expanded in future studies. In addition to northwestern China, Shandong Province in eastern China is a typical and essential vegetable production base. Therefore, sampling surveys should be conducted between different vegetable production bases to elucidate the heterogeneous effects of regional environmental differences.

Disclosure statement

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

Additional information

Funding

This work was supported by the National Natural Science Foundation of China under Grant No. 41771129.

Notes on contributors

Jing Chao

Jing Chao is a lecturer at Shandong Youth University of Political Science. She has a PH.D. from Northwest University in China. Her main research interests include agriculture innovation adoption and rural development. Jing’s doctoral research compared the technology adoption demands, adoption behaviors, and adoption effects of smallholders and new agricultural business entities in China for improved practices.

Tongsheng Li

Tongsheng Li is a professor and doctoral supervisor at Northwest University in China. He has been committed to the research of humanities and economic geography for a long time, focusing on the innovation and diffusion of agricultural science and technology parks, agricultural development, urban-rural integration development, poverty alleviation research and evaluation, and land spatial planning. His current research provides theoretical and practical guidance for the promotion of sustainable agricultural practices and the sustainable development of agriculture in developing countries.