Evaluating the Effectiveness of Different Demonstration Models on Agricultural Climate-Smart Technology Adoption: Evidence from China’s Cotton Farmers

Abstract

Amid escalating global climate challenges and the accelerating transition toward low-carbon agriculture, the effective diffusion of climate-smart technologies (CSTs) has become a critical pathway to achieving sustainable agricultural development. This study focuses on cotton farmers in Xinjiang and draws on micro-level survey data from 504 respondents to construct an analytical framework of “demonstration–cognition–adoption”. It systematically examines the impact pathways and mediating mechanisms of different demonstration models. The findings reveal that technology demonstration indirectly influences farmers’ adoption of CSTs by shaping their cognitive perceptions, with perceived operational utility emerging as the most critical mediating mechanism in the entire technology dissemination chain. Among current extension models, government-led demonstrations play a central role, while the effectiveness of enterprise-led demonstrations hinges on brand credibility and service quality. Moreover, the ease of operation of a technology outweighs its economic returns in determining adoption outcomes, and farmers exhibit significant heterogeneity in their responses to different demonstration types. Based on these insights, the study recommends the development of a stratified and differentiated dissemination strategy, the strengthening of government-led demonstration functions, the promotion of standardized enterprise participation, and the enhancement of both farmers’ cognitive understanding and technology fit to enable broader and higher-quality adoption of climate-smart technologies.

1. Introduction

The global climate system is undergoing irreversible and tipping-point transformations. According to the IPCC Sixth Assessment Report, the global average temperature has already risen by more than 1.1 °C, triggering cascading effects across critical climate thresholds and substantially increasing the likelihood of future climate crises [1]. Agriculture occupies a central position in global climate action, functioning both as a significant source of greenhouse gas (GHG) emissions and as a major carbon sink [2]. The FAO’s report at COP26 revealed that over the past three decades, GHG emissions from global agricultural and food production have increased by 17%, making agri-food systems one of the primary contributors to emissions. However, agriculture also holds considerable mitigation potential. Enhancing soil organic carbon and increasing vegetative cover can significantly boost carbon sequestration, thereby reducing atmospheric CO₂ concentrations and contributing to climate stabilization [3]. Studies have shown that global croplands could sequester 26–53% of the annual GHG target set by the “4 per 1000” initiative for soil carbon enhancement to improve food security and climate resilience [4]. These findings underscore the pivotal role of agriculture in addressing global climate change and highlight agricultural carbon neutrality as a key component in tackling climate challenges.

How, then, can agriculture contribute effectively to addressing climate challenges? A growing body of research highlights the innovation and diffusion of climate-smart technologies (CSTs) as a critical lever for enabling the transition toward low-carbon agriculture [5,6]. CSTs are derived from the conceptual framework of Climate-Smart Agriculture proposed by the FAO. These technologies aim to reduce GHG emissions from agricultural production by enhancing resource-use efficiency, minimizing agricultural inputs, and improving ecosystem resilience [7]. Recent scientific advances have led to substantial innovation in CSTs, and a growing body of empirical evidence confirms their effectiveness in mitigating emissions during various stages of agricultural production [8,9,10,11]. However, despite their proven mitigation potential in theory and experimental trials, the adoption rate of CSTs in real-world farming practices remains relatively low, with widespread implementation still facing multiple practical constraints.

On the demand side, several factors significantly constrain farmers’ willingness to adopt CSTs, including limited cognitive understanding, high upfront investment costs, uncertain expected returns, and technical complexity in application [12]. Particularly within economically driven agricultural systems, farmers tend to favor low-risk and predictable traditional practices. On the supply side, many CSTs have yet to reach mature commercialization, and the supporting service systems and market safeguards remain underdeveloped, resulting in insufficient supply to support large-scale adoption [12]. This dual constraint on both the demand and supply sides severely hindered the widespread diffusion of CSTs.

However, not all CSTs are economically unviable. Some, such as precision agriculture, conservation tillage, and straw management, have already been widely applied in agricultural production and demonstrated tangible economic as well as environmental benefits. Nevertheless, due to limited technical knowledge and information asymmetries, farmers often misperceive certain economically viable CSTs as lacking profitability. Therefore, compared with institutional or service-oriented reforms on the supply side, a more feasible and policy-oriented approach in the short term is to focus on the demand side, address farmers’ cognitive bias towards CSTs, and release farmers’ demand for CSTs. This challenge is particularly pronounced in the cotton sector. As one of the world’s most important cash crops, cotton production entails high energy use and extensive application of fertilizers and pesticides, making it a high-emission agricultural activity [13]. Studies indicate that producing one ton of cotton emits between 0.3 and 1.4 tons of CO₂-equivalent GHG [14], significantly exceeding emissions from staple crops such as rice, wheat, and maize [15]. With climate-related risks intensifying, cotton farmers are increasingly aware of the adverse impacts of extreme weather on yields and are paying more attention to strategies that integrate economic returns with ecological benefits.

In recent years, researchers have developed a variety of CSTs tailored to cotton production, such as optimized water and fertilizer management, biochar application, and nitrification inhibitors. Most of them have demonstrated promising mitigation potential in experimental settings [16,17,18,19]. Some CSTs continue to face obstacles—such as high costs, technical risks, and incomplete commercialization—that limit their ability to meet farmers’ expectations for cost reduction and efficiency improvement, but others, particularly those that enhance yields and reduce input costs, have already been adopted by commercial farming entities [20,21]. However, many cotton farmers remain unaware of the potential economic value these CSTs can deliver. This indicates that the adoption gap cannot be attributed solely to a lack of economic benefits but also to information asymmetries, cognitive biases, and risk perceptions that influence farmers’ decision-making.

This issue is especially salient in China’s Xinjiang region. As the world’s largest cotton producer, China is increasingly pushing its cotton industry toward a low-carbon, circular, and ecologically sustainable model in response to climate change [22]. Xinjiang, as the country’s most important cotton-growing area, accounted for 86.25% of China’s total cotton acreage and 92.25% of total output in 2024. It not only dominates the national cotton supply but also bears the strategic responsibility for advancing green transformation in the sector. However, cognitive biases regarding CSTs persist among Xinjiang’s cotton farmers. Most farmers perceive the relationship between ecological and economic benefits as a “zero-sum game,” neglecting the long-term environmental and economic gains that climate-smart technologies can deliver, which results in a generally low willingness to adopt such technologies. In fact, some cotton farmers have already achieved cost reductions and increased efficiency through the adoption of CSTs, such as soil testing and fertilization, while also reducing GHG emissions and soil pollution in cotton production.

To address this issue, the Chinese government has considered Xinjiang’s natural endowments and production realities in recent years to actively promote trials and demonstrations of CSTs—such as precision fertilization, precision irrigation, and dry seeding with wet emergence—that offer both yield-enhancing and carbon-reducing benefits. A series of demonstration and extension programs focused on Xinjiang have been implemented through policy incentives and project support, encouraging research institutes, technical service organizations, and grassroots extension personnel to conduct field-based research and technology dissemination. These efforts aim to enhance farmers’ awareness and adoption willingness through a “visible, learnable, and applicable” demonstration mechanism. Numerous agricultural enterprises have actively participated in promoting CSTs, establishing high-standard demonstration fields and experimental sites across Xinjiang to improve both the accessibility and the perceived economic benefits of these technologies through enterprise-led demonstrations. Some emerging agricultural entities, such as large-scale farmers, cooperatives, and agricultural service organizations, have taken the lead in piloting CSTs. These early adopters have achieved notable emission-reduction outcomes and economic returns, gradually exerting a “pioneering demonstration–technology diffusion” effect. As a result, several CSTs have been widely adopted in cotton production practices.

Nonetheless, there remains a lack of systematic evaluation of these demonstration efforts. The following key questions remain unanswered: Have the demonstrations truly influenced farmers’ technology cognition and decision-making behavior? Do different types of technology demonstration mechanisms vary in their effectiveness at promoting CST adoption? How do farmers’ cognitive and behavioral responses evolve when exposed to diverse actors and demonstration approaches? These critical issues call for in-depth empirical research based on micro-level farmer data.

The adoption of CSTs has emerged as a key topic in global social science research and has attracted increasing attention from the international academic community in recent years. Guided by classical theories such as the TAM, the TRA, and the TPB, numerous studies have demonstrated that farmers’ adoption decisions regarding CSTs are shaped by both traditional and emerging factors. On the one hand, traditional factors—including farm size, education level, policy incentives, and regulatory frameworks—consistently serve as significant drivers across regions [23,24,25]. On the other hand, emerging factors—such as climatic conditions and disaster shocks, the opportunity costs of farmland abandonment for reforestation, the development and accessibility of green finance, individual environmental awareness, and social pressure—have also been shown to significantly influence adoption behavior [26,27,28,29]. In addition, scholars have examined cross-country and regional variations in CST adoption to explore differentiated strategies for promoting CSTs in diverse economic contexts and to identify effective approaches for contributing to global climate governance [12,29,30].

Meanwhile, among various approaches to promoting CSTs adoption, technology demonstration has been widely recognized as an effective tool [31,32,33]. Particularly in the dissemination of CSTs, demonstrations play a vital role in enhancing cognitive awareness and guiding behavioral change [34,35,36,37]. Through visible and interactive learning experiences, technology demonstrations can improve farmers’ understanding and confidence in applying new practices [38,39]. Furthermore, some studies have also begun to explore the heterogeneous effects of demonstration models across different farmer groups, suggesting that individual characteristics of demonstration recipients may lead to varied adoption outcomes [33,35,37].

However, systematic investigations into the heterogeneous impacts of different technology demonstration models remain relatively limited. While a few studies have compared the effectiveness of distinct demonstration types on adoption behaviors [40], most fall short of unpacking the underlying mediating mechanisms and lack a complete analytical framework linking demonstration, cognition, and adoption. This research gap is particularly relevant in practice. In real-world agricultural extension systems, multiple demonstration models often coexist. These models vary considerably in terms of promoters’ objectives, resource configurations, demonstration strategies, and emphasis, leading to differentiated pathways for information delivery, trust formation, and cognitive stimulation. As a result, they may exert distinct influences on farmers’ technology cognition and subsequent adoption decisions. Relying on a single demonstration channel or homogeneous policy design is insufficient to address the behavioral diversity and cognitive barriers that characterize rural farming populations.

Therefore, it is essential to analyze the strategic orientation, operational mechanisms, and actual effectiveness of different technology demonstration models. Understanding how these models influence technology adoption through cognitive pathways holds both theoretical and practical significance for improving CSTs’ dissemination strategies, enhancing policy effectiveness, and advancing the broader goal of low-carbon agricultural transformation. Building on this premise, the present study focuses on cotton farmers in Xinjiang and investigates the impacts of diverse CST demonstration models. By constructing an analytical framework of “demonstration–cognition–adoption”, the study systematically examines the mediating mechanisms and behavioral effects of different demonstration approaches. The findings aim to offer theoretical guidance and policy insights for tailoring promotion strategies to local conditions and improving the adoption rate of CSTs.

2. Materials and Methods

2.1. Conceptual Definition

Agricultural CSTs refer to technologies that reduce GHG emissions during agricultural production and enhance soil carbon sequestration. Their primary aim is to improve agricultural productivity, mitigate the negative environmental impacts of farming, and promote the harmonious coexistence of agriculture and ecosystems [7]. The prevailing academic consensus holds that CSTs achieve dual goals of emission reduction and carbon sequestration by optimizing agronomic management and resource allocation, reducing the excessive use of high-carbon inputs such as fertilizers and pesticides, and improving soil physicochemical properties and water retention, thereby mitigating CH₄ and N₂O emissions from agricultural sources and increasing soil organic carbon storage [41].

Considering the specific conditions of cotton production in Xinjiang, the selection of CSTs must balance economic and ecological benefits. On one hand, farmers tend to prioritize yield and economic returns, which necessitates that adopted CSTs provide strong economic incentives. On the other hand, given the scarcity and uneven spatiotemporal distribution of water resources in Xinjiang’s arid regions, CST promotion must integrate water-saving technologies to improve water-use efficiency per unit of resource. Accordingly, this study focuses on five key CSTs in cotton production: precision fertilization, precision irrigation, conservation tillage, straw incorporation, and dry seeding with wet emergence (

Table 1. Key Studies Confirming the Positive Contributions of CSTs to Addressing Climate Challenges.

CSTs	Definition	References
precision fertilization	the optimal ratios and application rates of nutrients such as nitrogen, phosphorus, and potassium based on soil nutrient status and crop requirements	[42,43,44]
precision irrigation	patch-type drip irrigation	[44,45,46]
conservation tillage	minimize soil disturbance, maintain surface cover, and preserve soil aggregate structure	[47,48,49]
straw incorporation	crushing and returning cotton straw to the field	[46,50]
dry seeding with wet emergence	direct sowing without winter or spring irrigation, followed by a single post-sowing watering to promote seedling emergence	[51]

).

Precision fertilization specifically refers to soil testing and formula-based fertilization, which determines the optimal ratios and application rates of nutrients such as nitrogen, phosphorus, and potassium based on soil nutrient status and crop requirements. Its primary function is to reduce excessive nitrogen fertilizer use, decrease nitrogen losses and N₂O emissions, and improve both fertilizer efficiency and crop yield. In Xinjiang cotton production, this practice effectively reduces fertilizer waste caused by blind application, achieving both cost-saving and GHG reduction benefits.

Precision irrigation refers to the use of patch-type drip irrigation, an improved version of traditional drip systems. This technique offers better adaptability and clogging resistance, delivering water directly to plant roots. It significantly conserves water and reduces soil denitrification caused by excessive irrigation, thereby lowering N₂O emissions. Moreover, integrated water–fertilizer management improves the rhizosphere environment, enhances nutrient uptake efficiency, and achieves both resource conservation and yield gains.

Conservation tillage refers to reduced or no-tillage practices that minimize soil disturbance, maintain surface cover, and preserve soil aggregate structure. This approach effectively decreases soil organic carbon mineralization and CO₂ emissions while enhancing long-term soil carbon sequestration. In Xinjiang’s arid cotton-growing regions, conservation tillage also improves soil moisture retention, reduces evaporation losses, and enhances water-use efficiency and drought resistance.

Straw incorporation involves crushing and returning cotton straw to the field, where organic matter decomposition increases soil organic matter content and carbon sequestration capacity. Long-term straw incorporation improves soil physicochemical properties, enhances microbial activity, and contributes to soil carbon pool accumulation. Its nutrient-supplying function partially substitutes for chemical fertilizers, indirectly reducing N₂O emissions.

Dry seeding with wet emergence refers to direct sowing without winter or spring irrigation, followed by a single post-sowing watering to promote seedling emergence. This method significantly reduces irrigation water use, lowers energy consumption and carbon emissions from pumping, and shortens periods of soil saturation, thereby suppressing denitrification-induced N₂O emissions. It provides both water-saving and emission-reduction benefits.

2.2. Theoretical Framework

The rational smallholder theory posits that farmers, under constraints of limited resources, asymmetric information, and bounded rationality, tend to make relatively rational decisions based on the principle of “maximizing returns and minimizing risks”. Due to a lack of technical knowledge and limited opportunities for trial and error, farmers often exhibit caution or even conservatism when confronted with new technologies. They tend to rely on external sources of information, observing the experiences of others or the outcomes of technology demonstrations to assess feasibility and expected benefits, thereby reducing the likelihood of economic losses or operational risks. Therefore, technology demonstration, as a substitute trial-and-error mechanism, provides farmers with an affordable pathway for acquiring information and forming cognitive evaluations. Through visual field demonstrations, yield comparisons, and face-to-face communication, demonstration activities can significantly reduce uncertainty and enhance farmers’ expectations regarding the economic benefits, operability, and reliability of a given technology. Within the rational smallholder decision-making framework, technology demonstration can thus significantly enhance the willingness of farmers to adopt CSTs, making it a critical institutional tool for promoting technology diffusion. Based on this, Hypothesis 1 is proposed as follows:

H1. 

All else being equal, stronger technology demonstration is positively associated with a higher likelihood of CSTs adoption among cotton farmers.

At the same time, cognitive behavioral theory suggests that individual behavior results from cognitive processing of external stimuli. As an external intervention tool, technology demonstration can significantly improve farmers’ understanding, trust, and confidence in new technologies through visual display and interpersonal interaction, thereby enhancing their adoption intentions. According to the TAM, perceived usefulness and perceived ease of use are two key determinants of technology adoption decisions. Technology demonstration can directly reinforce these two cognitive dimensions, thereby indirectly promoting CST adoption [39,40]. Accordingly, Hypothesis 2 is proposed as follows:

H2. 

Technology cognition mediates the relationship between technology demonstration intensity and cotton farmers’ adoption of CSTs.

However, given the diversity of demonstration actors, demonstration models differ in objectives, delivery methods, and areas of emphasis, which in turn affect both the mechanism and effectiveness of farmer adoption. This study categorizes demonstration models into three types based on the identity and purpose of the demonstration actor: government-led, enterprise-led, and peer-driven demonstrations. Drawing on TAM theory, farmers’ cognition of CSTs is further categorized into three dimensions: economic utility, ecological utility, and operational utility.

Government-led demonstrations are typically initiated by universities, research institutes, or agricultural extension agencies. Their main goals are to develop, transform, and promote scientific research outcomes, with a strong public-good orientation and long-term sustainability goals. Supported by government policies, these demonstrations are usually conducted on a large-scale basis and presented through centralized observation sessions, expert field lectures, and official reporting mechanisms. While the credibility of multi-level government and expert endorsements significantly enhances farmers’ trust in the ecological utility of CSTs, the limited participatory experience and weak perception of direct benefits result in a relatively modest effect on actual adoption behavior. Based on this, Hypothesis 3a is proposed as follows:

H3a. 

Government-led demonstrations positively influence farmers’ adoption of CSTs by enhancing ecological utility cognition, but the effect size is relatively modest.

Enterprise-led demonstrations are organized by large agricultural input firms or local agro-dealers with the aim of promoting their own products and securing government subsidies. These demonstrations are often integrated with product marketing and conducted on proprietary plots or those managed in cooperation with local farmers. By showcasing economic performance and cost–benefit comparisons, such demonstrations directly address farmers’ economic interests. Accompanied by after-sales services and subsidy packages, they tend to significantly enhance economic utility cognition. However, the effectiveness of this model depends heavily on corporate reputation; firms with strong local credibility are more likely to earn farmer trust and improve demonstration outcomes. Accordingly, Hypothesis 3b is proposed as follows:

H3b. 

Enterprise-led demonstrations enhance cotton farmers’ economic utility cognition and have a significant positive effect on CSTs’ adoption—stronger than that of government-led demonstrations.

Peer-driven demonstrations are carried out by advanced growers, cooperative leaders, or other new agricultural business entities. Through mechanisms such as farmer-to-farmer learning or neighbor-based sharing, these actors conduct on-farm demonstrations within their own production plots. While primarily driven by their own cost-reduction or efficiency-improvement goals, peer demonstrations benefit from long-standing local trust networks, highly relatable operating procedures, and ease of replication. These factors lower perceived cognitive and technical barriers and enable target farmers to better perceive the economic value of the new technology. As such, this model most effectively stimulates both operational and economic utility cognition, substantially boosting CSTs’ adoption willingness. Based on this, Hypothesis 3c is proposed as follows:

H3c. 

Peer-driven demonstrations significantly promote CSTs adoption by enhancing both operational and economic utility cognition, and exhibit the strongest positive effect among all demonstration models.

2.3. Variable Selection

Dependent Variable. The dependent variable in this study is the level of technology adoption, which measures the actual adoption level of CSTs by cotton farmers. Following established methods in the literature and focusing on five key CSTs, the number of CSTs adopted by each farmer is used as an indicator of adoption level. A greater variety of adopted technologies indicates a higher level of CST adoption, whereas fewer adopted technologies reflect a lower level of adoption. According to the number of CSTs adopted by cotton farmers, their degree of technology adoption is assigned a value of 1 to 5 (there is no zero adoption in the sample), where “1” represents the lowest degree of adoption and “5” represents the highest degree of adoption. The increasing values from 1 to 5 reflect the improvement of the degree of technology adoption.

Independent Variable. The independent variable is technology demonstration intensity, which measures the breadth of technology promotion received by farmers. As shown in

Table 2. Technology Demonstration Intensity Measurement Method. The cotton farmers interviewed will be asked five questions about climate-smart technology demonstration in turn. The answer “yes” is assigned a value of “1” and “no” is assigned a value of “0”, and the five answers are then added together to form a total.

CSTs	Is Demonstration	Government-Led	Enterprise-Led	Peer-Driven
precision fertilization	1	0	1	0
precision irrigation	1	1	0	0
conservation tillage	1	1	0	0
straw incorporation	1	0	1	1
dry seeding with wet emergence	0	0	0	0
intensity	4	2	2	1

, this variable is constructed by summing the number of CSTs demonstrations a farmer has attended. During the survey, farmers were asked five binary questions: “Have you ever attended a demonstration of soil testing and formula fertilization?”, “...of patch-type drip irrigation?”, “...of conservation tillage?”, “...of straw incorporation?”, and “...of dry seeding with wet emergence?”. Each affirmative response was coded as 1, and negative as 0. The total sum yields an ordinal variable representing demonstration intensity. For example, if a farmer had attended four of the five demonstrations, their demonstration intensity was coded as “4”.

To explore the heterogeneous effects of different demonstration modes, this variable is further disaggregated into three sub-variables: government-led demonstration intensity, enterprise-led demonstration intensity, and peer-driven demonstration intensity. The measurement method mirrors that of overall demonstration intensity. Farmers were additionally asked whether they had attended each technology demonstration under these three modes. For instance, if a farmer had attended two government-led, one enterprise-led, and one jointly enterprise- and peer-driven demonstration (for four technologies in total), their demonstration intensities for the three modes would be 2, 2, and 1, respectively.

Mediating variable. To examine the mechanism through which demonstration influences adoption, farmers’ technology cognition is introduced as a mediating variable. Similarly to demonstration intensity, cognition level is measured by summing responses to five yes/no questions (

Table 3. Technology Awareness Level Measurement Method. The cotton farmers interviewed will be asked five questions about climate-smart technology demonstration in turn, and the answers will be summed up according to the respondents.

CSTs	Is Cognition	Economic Utility	Ecological Utility	Operational Utility
precision fertilization	1	5	4	3
precision irrigation	1	4	4	5
conservation tillage	0	0	0	0
straw incorporation	1	3	5	4
dry seeding with wet emergence	0	0	0	0
level	3	12	13	12

): “Do you know about precision fertilization?”, “...precision irrigation?”, “...conservation tillage?”, “...straw incorporation?”, and “...dry seeding with wet emergence?”. Each affirmative response was coded as 1, and negative as 0. For example, if a farmer reported knowledge of three technologies, their technology cognition level was coded as “3”.

To further analyze the mechanisms of different demonstration modes, the mediating variable was decomposed into three dimensions: economic utility cognition, ecological utility cognition, and operational utility cognition. A five-point Likert scale was used to evaluate farmers’ perceptions of each technology in three aspects: perceived economic benefits (1 = very low, 5 = very high), perceived ecological benefits (1 = very low, 5 = very high), and perceived adoption difficulty (reverse-coded: 1 = very difficult, 5 = very easy). Scores across all technologies were summed to generate the three cognition indices. For example, if a farmer knew three technologies and rated them as (5, 4, 3), (4, 4, 5), and (3, 5, 4) across the three dimensions, their economic utility, ecological utility, and operational utility cognition scores would be 12, 13, and 12, respectively.

Control Variables. To account for other potential influences on technology adoption and to improve model robustness, this study includes a range of socio-economic characteristics at the household level as control variables, following related literature [33,37]. These include gender, ethnicity, age, education level, health status, household size, off-farm employment, years of cotton farming, annual household income, and cooperative membership. These variables have been shown to correlate significantly with agricultural technology adoption and help minimize omitted variable bias. The selection, definition and statistics of control variables are shown in

Table 4. The Selection, Definition and Statistics of Control Variables.

Variable	Definition	Mean	SD
Dependent Variable
Technology adoption level	Number of CSTs adopted by the farmer	3.14	0.86
Independent Variables
Technology demonstration intensity	Number of CSTs demonstrated to the farmer	2.80	1.54
Government-led demonstration intensity	Number of CSTs demonstrated by government actors	1.34	1.60
Enterprise-led demonstration intensity	Number of CSTs demonstrated by enterprises	1.39	1.52
Peer-driven demonstration intensity	Number of CSTs demonstrated by peers	1.87	1.45
Mediating Variables
Technology cognition	Number of CSTs the farmer is familiar with	4.10	0.86
Economic utility	Sum of perceived economic utility scores (5 technologies)	14.60	4.27
Ecological utility	Sum of perceived ecological utility scores (5 technologies)	13.34	4.91
Operational utility	Sum of perceived operational utility scores (5 technologies)	7.20	4.67
Control Variables
Gender	1 = Male, 0 = Female	0.86	0.35
Ethnicity	1 = Han, 2 = Uyghur, 3 = Other	1.23	0.50
Age	Age of household head (years)	49.55	8.16
Education level	1 = Primary or below, 5 = Above Bachelor	2.73	0.81
Health status	1 = Very poor, 5 = Very good	4.70	0.79
Household size	Number of household members	3.86	1.34
Off-farm employment	1 = Yes, 0 = No	0.30	0.46
Years of cotton farming	Years engaged in cotton cultivation	18.22	9.81
Household income	1 ≤ 50 K RMB, 2 = 50–100 K RMB, 3 = 100–150 K RMB, 4 = 150–200 K RMB, 5 > 200 K RMB	2.51	1.17
Cooperative participation	1 = Member, 0 = Non-member	0.28	0.45

.

Men often serve as primary decision-makers in agricultural production, and gender differences may influence adoption willingness. Male farmers were coded as 1 and females as 0. Given Xinjiang’s multi-ethnic context, cultural and language differences across ethnic groups may affect cognition and adoption. Han ethnicity was coded as 1, Uyghur (the major ethnicity in Xinjiang) as 2, and other minorities as 3.

Age and education level influence farmers’ capacity to adopt new practices and their risk preferences; younger and better-educated farmers tend to have superior information access, technical understanding, and cost–benefit analysis ability. Health status directly affects labor supply and management capacity. Age was recorded in years, while education and health status were measured using a five-point Likert scale (5 = highest, 1 = lowest).

Household size affects resource endowment and risk-sharing capacity. Larger households with sufficient labor are better positioned to adopt new technologies, but households dominated by non-labor members may face economic constraints suppressing adoption. Household income reflects financial and risk-bearing capacity; higher-income households are more likely to invest in new technologies. Off-farm employment and cooperative membership reflect farmers’ agricultural investment level, access to information, and subsidy channels.

2.4. Econometric Models

The ordered logit model is one of the most widely used classical econometric models in studies of technology adoption behavior. It is particularly suitable for situations where the dependent variable is ordinal with multiple categories and non-equidistant intervals. This model effectively characterizes the behavioral process in which a latent continuous propensity is transformed into discrete ordered outcomes. The model aligns well with the decision-making logic underlying farmers’ gradual transformation from latent willingness to actual adoption behavior and has been extensively validated in agricultural technology adoption research [52,53]. In this study, the dependent variable is an ordered categorical variable, which satisfies the basic assumption of the ordered logit model regarding the ordinal nature of the dependent variable. Furthermore, the proportional odds assumption—a key requirement of the ordered logit model—was tested using the Wald test, a mainstream method for this purpose. The results showed that p-values exceeded 0.05, indicating that the null hypothesis could not be rejected. Thus, the proportional odds assumption holds, confirming the appropriateness of the ordered logit model for this study. Therefore, the ordered logit model was employed for hypothesis testing in this study.

The mediation effect model, in contrast, is an econometric method designed to examine how independent variables indirectly influence dependent variables through mediating variables. Its core function lies in revealing the underlying causal mechanisms, and it has been widely applied in economics and behavioral decision-making research. To test the mediating role of technology cognition in the relationship between demonstration and adoption, this study employed a mediation effect model to analyze the underlying mechanisms. The specific models are specified as follows:

$${\mathrm{C}}_{\mathrm{i}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\mathrm{D}}_{\mathrm{i}}+{\mathsf{\alpha }}_2{\mathrm{X}}_{\mathrm{i}}+{\mathsf{\epsilon }}_{\mathrm{i}}.$$

(1)

Equation (1) examines the effect of technology demonstration intensity on farmers’ level of technology cognition, where ${\mathrm{C}}_{\mathrm{i}}$ represents the cotton farmer’s cognition level regarding CSTs, ${\mathrm{D}}_{\mathrm{i}}$ denotes the degree of exposure to CST demonstrations, and ${\mathsf{\alpha }}_1$ captures the marginal effect of demonstration intensity on technology cognition. A significantly positive ${\mathsf{\alpha }}_1$ indicates that higher demonstration intensity enhances farmers’ cognition of CSTs. Conversely, a negative or insignificant ${\mathsf{\alpha }}_1$ may suggest that overly intensive demonstrations could trigger skepticism or resistance among farmers, thereby reducing their cognition levels.

To further assess both the direct effect of technology demonstration intensity on technology adoption and the mediating role of technology cognition, the following model is constructed:

$$A_i={\beta }_0+{\beta }_1{D}_i+{\beta }_2{C}_i+{\beta }_3{X}_i+{\epsilon }_i.$$

(2)

$$\begin{array}{c}{A}_i\left\{\begin{array}{c}1, A_i^{\ast }\le r_1\\ 2, r_1\le A_i^{\ast }\le r_2\\ 3, r_2\le A_i^{\ast }\le r_3\\ 4, r_3\le A_i^{\ast }\le r_4\\ 5, r_4{\le A}_i^{\ast }\end{array}\right.\end{array}$$

(3)

$A_i$ denotes the degree of CSTs adoption by cotton farmers, while $A_i^{\ast }$ denotes the farmer’s latent (unobservable) true adoption level. The relationship between $A_i$ and $A_i^{\ast }$ is specified in Equation (3), where $r_1$ $r_2$, $r_3$, and $r_4$ are threshold cut-points. Here, ${\beta }_1$ captures the direct effect of demonstration intensity on CST adoption. A significantly positive ${\beta }_1$ indicates that higher demonstration intensity directly promotes farmers’ adoption of CSTs. ${\beta }_2$ measures the effect of technology cognition on CST adoption. A significantly positive ${\beta }_2$ suggests that higher levels of CST cognition are associated with greater adoption, indicating that cognition plays a positive role in promoting adoption behavior. The mediating effect is tested by the product ${\alpha }_1\times {\beta }_2$, which shows whether cognition acts as a mediating mechanism between demonstration and adoption.

However, in real-world agricultural practice, cotton farmers often receive overlapping information and influences from multiple demonstration actors. That is, government-led, enterprise-led, and peer-driven demonstrations commonly coexist. This situation of multi-model coexistence raises two empirical challenges. First, it is difficult to identify which demonstration model plays the dominant role in influencing adoption behavior. Second, it remains unclear whether the joint effect of multiple demonstrations is synergistic, neutralizing, or even mutually interfering.

To more accurately assess the interaction mechanisms among different demonstration models, this study extends empirical strategy by introducing interaction terms into the main regression. To test for possible interaction effects among the three types of demonstration models, the following extended model is specified:

$$\begin{gathered} A_i={\beta }_0+{\beta }_1{GD}_i+{\beta }_2{ED}_i+{\beta }_3{PD}_i+{\beta }_4{(GD}_i\times {ED}_i)+{\beta }_4{(GD}_i\times {PD}_i)+{\beta }_5{(ED}_i\times {PD}_i)+ \\ {\beta }_6{({GD}_i\times ED}_i\times {PD}_i)+{\beta }_7{C}_i+{\beta }_8{X}_i+{\epsilon }_i. \end{gathered}$$

(4)

Here, ${GD}_i$, ${ED}_i$ and ${PD}_i$ denote the intensities of government-led, enterprise-led, and peer-driven demonstrations, respectively. The interaction terms represent combinations of demonstration types that a farmer is simultaneously exposed to. A significantly positive coefficient of the interaction term indicates that combinations of different demonstration modes generate a synergistic effect, jointly facilitating farmers’ adoption of CSTs. A significantly negative coefficient suggests the presence of an interference effect, implying potential signal conflicts or mutual interference among demonstration modes, which may offset their overall impact. A statistically insignificant coefficient indicates no interaction effect among different demonstration modes. Finally, the original five-level technology adoption variable was recategorized into three levels—scores 1–2 as “low adoption” (coded as 1), score 3 as “medium adoption” (coded as 2), and scores 4–5 as “high adoption” (coded as 3)—to examine whether the estimation results remained stable after reclassification.

To ensure the robustness of the results, a series of robustness checks were conducted using alternative models, sample trimming, and variable transformation. First, the ordered logit model was replaced with an ordered probit model to test consistency under different distributional assumptions. Second, observations with extreme values in household income were removed to reduce potential outlier bias. Third, the five-level adoption variable was reclassified into three categories (Scores 1–2 as “low adoption”, Score 3 as “medium adoption”, Scores 4–5 as “high adoption”), which allowed for validation of whether results are sensitive to the scaling of the dependent variable.

2.5. Data Source

Xinjiang is China’s largest high-quality commercial cotton production base and has ranked first in national cotton output for 31 consecutive years. The region holds strategic importance in the national agricultural economy. Geographically, the cotton-growing area is divided by the Tianshan Mountains into southern and northern cotton zones, with minor cultivation in the eastern regions of Turpan and Hami (Figure 1). Xinjiang’s unique natural conditions, such as abundant sunshine and large diurnal temperature variation, combined with drip irrigation and mechanized farming, have pushed cotton yield to 2322.8 kg per hectare, far exceeding the national average. As one of the world’s key cotton production regions, Xinjiang has developed a full industrial chain encompassing breeding, cultivation, processing, and textile manufacturing. In recent years, the Chinese government has launched a series of CST demonstration projects in Xinjiang to promote climate-smart agricultural transformation. This makes Xinjiang an ideal case for analyzing CST demonstration and adoption mechanisms.

The data used in this study were obtained through a structured field survey conducted in Xinjiang from November 2024 to June 2025. Drawing on established local research foundations and extensive experience in CST promotion, the research team conducted in-depth surveys and interviews targeting cotton growers in both southern and northern Xinjiang, including actors from the Xinjiang Uyghur Autonomous Region and the Xinjiang Production and Construction Corps. A total of 504 valid questionnaires were collected.

The data collection process was subject to multiple unique and significant logistical and social challenges, which underscore both the rarity and value of the dataset. Xinjiang’s vast land area and the wide distribution of cotton-growing zones—often located in remote rural regions—posed substantial difficulties for transportation and access. Despite seemingly large populations in some villages, the actual number of households engaged in cotton cultivation was limited, making respondent identification particularly challenging. Additionally, cotton farming in Xinjiang is highly seasonal, with many farmers being migrant workers who return to other provinces during the off-season. Missing the appropriate time window made it nearly impossible to reach these individuals. Linguistic barriers further complicated the process. Xinjiang is home to multiple ethnic groups, and some minority cotton farmers have limited fluency in Mandarin. This raised concerns over questionnaire comprehension and translation bias, requiring careful coordination during survey implementation.

To overcome these obstacles, the research team partnered with local agricultural departments and extension systems, leveraging the assistance of grassroots extension workers, cooperative leaders, and other key informants. A three-phase strategy of early identification–targeted coordination–phased rollout was adopted to ensure accurate farmer selection and flexible scheduling. Multilingual survey assistants were deployed to ensure smooth communication and high-quality responses. In addition, survey activities were carefully timed to avoid critical farming seasons, such as spring sowing and autumn harvest, ensuring farmers were available for interviews.

Ultimately, the survey was completed with high quality. First-hand micro-level data cover not only farmers’ demographic characteristics and production practices but also detailed information on their CST cognition, demonstration exposure, and adoption behavior. The dataset provides rare empirical support into the constraints and experiences of cotton production in China’s western frontier and serves as an important basis for policy formulation in low-carbon agricultural development. To better understand the sample structure, a detailed descriptive analysis of farmer characteristics is provided in

Table 5. Demographic Characteristics of Cotton Farmer Sample.

Characteristics		Count	Proportion
Gender	Male	433	85.91%
	Female	71	14.09%
Ethnicity	Han	408	80.95%
	Uyghur	78	15.48%
	Other	18	3.57%
Age	≤40	77	15.28%
	41–60	400	79.37%
	>60	27	5.36%
Education	Primary or below	53	10.52%
	Junior secondary	264	52.38%
	High school/Technical school	133	26.39%
	University or College	54	10.71%
Health status	Very poor	10	1.98%
	Poor	12	2.38%
	Average	8	1.59%
	Fair	61	12.10%
	Good	413	81.94%
Household size	1–2	59	11.71%
	3–5	312	61.90%
	>5	133	26.39%
Off-farm employment	Yes	151	29.96%
	No	353	70.04%
Years of cotton farming	<10	148	29.37%
	10–20	186	36.90%
	21–30	130	25.79%
	>30	30	5.95%
Household income	<50,000 RMB	109	21.63%
	50–100 K RMB	160	31.75%
	100–150 K RMB	144	28.57%
	150–200 K RMB	49	9.72%
	>200 K RMB	42	8.33%
Cooperative membership	Yes	139	27.58%
	No	365	72.42%
Total		504	100%

and

Table 6. CST Cognition, Demonstration, and Adoption Among Sample Farmers Indicator. The number and proportion of interviewed cotton farmers with different levels of technology awareness, technology demonstration intensity and technology adoption degree.

Characteristics		Count	Proportion
CSTs Cognition Level	0	0	0.00%
	1	0	0.00%
	2	1	0.20%
	3	162	32.14%
	4	127	25.20%
	5	214	42.46%
Demonstration Types Received	None	66	13.10%
	Government-led	253	50.20%
	Enterprise-led	288	57.14%
	Peer-driven	376	74.60%
CSTs Adoption Level	0	0	0.00%
	1	5	0.99%
	2	144	28.57%
	3	217	43.06%
	4	140	27.78%
	5	28	5.56%
Total		504	100%

.

In terms of demographic characteristics, the sample is predominantly male, accounting for 85.91% of respondents. Han ethnicity represents 80.95%, while Uyghur farmers account for 15.48%, reflecting the ethnic composition and regional specificity of Xinjiang. The age structure is skewed toward older farmers, and the overall education level is relatively low. Despite this, the surveyed farmers generally report good health status, and most households consist of 3 to 5 members. Nearly 60% of respondents are full-time farmers, and over 70% have more than ten years of cotton farming experience, suggesting substantial practical knowledge and technical experience. Household income is mostly in the middle range, with 60.32% earning between 50,000 and 150,000 RMB annually. Approximately 27.58% of respondents participate in agricultural cooperatives. In terms of technology cognition and adoption, overall awareness of CSTs is relatively high. A total of 42.46% of respondents rated their CST cognition at the highest level, with another 25.20% scoring 4. However, adoption levels are more concentrated in the mid-range. 43.06% of farmers reported adopting three CSTs, while only 33.34% adopted four or more. This indicates a degree of disconnect between cognition and actual adoption behavior. Regarding technology demonstration exposure, the largest number of cotton farmers, 376, received demonstrations from Peer-driven. Among the remaining respondents, exposure to only one or two types of demonstration was more fragmented. In summary, the sample displays a high degree of diversity and representativeness, providing a robust empirical foundation for analyzing the mechanisms of technology cognition and the pathways of demonstration-based dissemination.

3. Results

3.1. Effects of Technology Demonstration on CSTs Adoption

Table 7. Effects of Technology Demonstration Intensity on CST Adoption.

Variables	CST Adoption Levels
	(1)		(2)
	Coef.	S.E.	Coef.	S.E.
Technology demonstration intensity	0.416 ***	0.085	0.434 ***	0.080
Gender			0.089	0.301
Ethnicity			−0.475 **	0.232
Age			0.023	0.015
Education			−0.139	0.139
Health status			0.099	0.126
Household size			0.066	0.063
Off-farm employment			0.552 ***	0.222
Years of cotton farming			−0.007	0.014
Household income			0.227 ***	0.078
Cooperative membership			1.033 ***	0.257
Fixed effects	Controlled		Controlled
R2	0.219		0.248
Observations	504		504

“***” indicates significance at the 1% level, and “**” indicates significance at the 5% level.

presents the regression results examining the effect of technology demonstration intensity on cotton farmers’ adoption of CSTs, estimated using ordered logit models. In Model (1), without control variables, the coefficient on demonstration intensity is 0.544 and statistically significant at the 1% level, suggesting a strong positive association between demonstration exposure and adoption. After adding control variables in Model (2), the coefficient slightly decreases to 0.484, but remains highly significant, indicating that the direct effect of demonstration intensity is robust and substantial. These findings provide strong support for Hypothesis 1, confirming that higher demonstration intensity significantly increases the likelihood of CST adoption among cotton farmers. Among the control variables, education level, years of cotton farming, household income, and cooperative participation are significantly associated with adoption behavior. The positive effect of education suggests that knowledge capital helps farmers better evaluate and adopt new technologies. More experienced farmers are more likely to adopt, indicating higher technical judgment and openness to innovation. Higher income increases the probability of adoption, reflecting the importance of economic capacity. Participation in cooperatives enhances adoption, suggesting that organizational embeddedness plays a dual role in platform provision and trust-building in technology promotion.

To test the mediating role of technology cognition in the relationship between demonstration and adoption,

Table 8. Mediation Effects of Technology Cognition.

Variables	CST Adoption Levels
	(1)		(2)
	Coef.	S.E.	Coef.	S.E.
Technology demonstration intensity	0.267 ***	0.094	0.304 ***	0.088
Technology cognition	0.871 ***	0.142	1.121 ***	0.167
Mediation effect (α1 × β2)	0.474 ***	0.109	0.543 ***	0.125
Control variables	Uncontrolled		Controlled
Fixed effects	Controlled		Controlled
R2	0.250		0.292
Observations	504		504

“***” indicates significance at the 1% level.

reports the results of the mediation analysis. In Model (1), without controls, technology cognition is found to partially mediate the relationship between demonstration intensity and CST adoption. In Model (2), once controls are added, the direct effect of demonstration increases slightly to 0.304, and the effect of technology cognition becomes even stronger, indicating that the mediating mechanism is more pronounced after controlling for heterogeneity in farmer characteristics. The estimated mediation effect is 0.543 and remains significantly positive, suggesting a substantial indirect pathway. Therefore, Hypothesis 2 is supported. These findings confirm that technology demonstration does not directly lead to behavior change; rather, it works by enhancing farmers’ cognitive understanding, subjective evaluation, and perceived value of CSTs, which in turn influence adoption decisions.

Table 9. Mediation Effects of Technology Cognition. (1) represents the result of the ordered logit model was replaced with an ordered probit model; (2) represents the result of observations with extreme values in household income were removed to reduce potential outlier bias; (3) represents the result of the five-level adoption variable was reclassified into three categories.

	Variables	CST Adoption Levels
		(1)	(2)	(3)
Hypotheses 1	Technology demonstration intensity	0.263 *** (0.042)	0.454 *** (0.084)	0.367 *** (0.080)
	Control variables	Controlled	Controlled	Controlled
	Fixed effects	Controlled	Controlled	Controlled
	R2	0.234	0.241	0.297
	Observations	504	504	504
Hypotheses 2	Technology demonstration intensity	0.199 *** (0.044)	1.328 *** (0.090)	0.212 ** (0.090)
	Technology cognition	0.621 *** (0.090)	1.109 *** (0.174)	1.218 *** (0.179)
	Mediation effect (α1×β2)	0.159 *** (0.037)	0.522 *** (0.125)	0.590 *** (0.135)
	Control variables	Controlled	Controlled	Controlled
	Fixed effects	Controlled	Controlled	Controlled
	R2	0.277	0.284	0.352
	Observations	504	504	504

“***” indicates significance at the 1% level, and “**” indicates significance at the 5% level.

reports the robustness of results. In all three specifications, the effect of demonstration intensity on CST adoption remains positive and significant, with coefficient magnitudes consistent with the baseline models. The mediating role of technology cognition also remains statistically significant. Notably, even after reclassifying the dependent variable, both the direct and indirect effects persist, indicating that the findings are not sensitive to the measurement scale. These results confirm the reliability of the main conclusion.

3.2. Effects and Mechanisms of Different Technology Demonstration Models

The estimation results for Hypothesis 3 are presented in

Table 10. Main Regression Results (Hypothesis 3).

Variables	CST Adoption Levels
	(1)	(2)
Government-led only	0.924 *** (0.155)	0.837 *** (0.150)
Enterprise-led only	−0.329 *** (0.114)	−0.255 ** (0.119)
Peer-driven only	0.258 * (0.151)	0.208 (0.157)
Gov × Ent	0.033 (0.054)	0.015 (0.055)
Gov × Peer	−0.520 *** (0.066)	−0.466 *** (0.066)
Ent × Peer	0.008 (0.074)	−0.006 (0.075)
All three types	0.094 *** (0.022)	0.010 *** (0.023)
Economic utility	0.083 (0.052)	0.086 (0.055)
Ecological utility	0.047 (0.046)	0.050 (0.049)
Operational utility	0.090 *** (0.031)	0.074 *** (0.029)
Control variables	Uncontrolled	Controlled
Fixed effects	Controlled	Controlled
R2	0.326	0.350
Observations	504	504

“***” indicates significance at the 1% level, “**” indicates significance at the 5% level, and “*” indicates significance at the 10% level.

. Among the three demonstration models, government-led and enterprise-led demonstrations significantly influence the adoption of CSTs, whereas the effect of peer-driven demonstrations is not statistically significant. Notably, the coefficient for enterprise-led demonstrations is negative, contradicting Hypothesis 3b. Field interview information helps explain this unexpected result. In the context of an increasingly competitive agricultural input market, many enterprises have failed to establish brand credibility in local communities. Consequently, cotton farmers exhibit a general lack of trust. Furthermore, enterprise-led demonstration activities are often sales-oriented rather than informative. In many cases, farmers are invited to attend demonstration sessions under misleading pretenses, only to realize the true objective is product promotion. This undermines farmers’ trust, fostering skepticism and resistance toward enterprise-led demonstrations. The combined effects of poor brand reputation and commercial motives lead to the observed negative association with CST adoption.

Moreover, the empirical analysis reveals notable interaction effects among demonstration models. Specifically, the interaction between government-led and enterprise-led demonstrations, as well as the three-way interaction among all models, shows significant effects. Of particular interest is the negative interaction between government-led and peer-driven demonstrations. A plausible explanation is that government-led demonstrations often prioritize ecological benefits, while farmers are primarily concerned with economic returns. In cases where both actors promote CSTs jointly, the selected technologies tend to emphasize environmental protection over profitability, weakening adoption incentives. Furthermore, among the three types of cognitive variables, operational utility emerges as the only significant mediator of adoption. Neither economic nor ecological utility shows significant effects. These findings imply that farmers prioritize whether a technology is easy to learn and apply, rather than its long-term profitability or environmental impact.

Overall, Hypothesis 3 is not supported by the empirical evidence. This result may be attributed to the multi-stage diffusion logic of CSTs, typically progressing through stages of R&D, commercialization, demonstration, cognition, and adoption. Different demonstration actors appear to operate independently, but a nested and hierarchical structure exists in reality. What appears to be peer-driven diffusion may, in fact, originate from government or enterprise-led promotion. This upstream-downstream relationship positions government-led demonstrations as initiators, with enterprise-led and peer-driven models following suit. As such, the three-way interaction demonstrates the strongest effect, while individual models appear less influential when considered in isolation.

Table 11. The Robustness of Results (Hypothesis 3). (1) represents the result of the ordered logit model was replaced with an ordered probit model; (2) represents the result of observations with extreme values in household income were removed to reduce potential outlier bias; (3) represents the result of the five-level adoption variable was re-classified into three categories.

Variables	CST Adoption Levels
	(1)	(2)	(3)
Government-led only	0.493 *** (0.081)	0.819 *** (0.151)	0.733 *** (0.181)
Enterprise-led only	−0.113 * (0.065)	−0.202 ** (0.119)	−0.292 ** (0.121)
Peer-driven only	0.167 ** (0.083)	0.206 (0.162)	0.096 (0.169)
Gov × Ent	−0.003 (0.031)	−0.011 (0.056)	0.062 (0.070)
Gov × Peer	−0.275 *** (0.036)	−0.457 *** (0.069)	−0.415 *** (0.077)
Ent × Peer	−0.026 (0.040)	0.002 (0.075)	−0.042 (0.087)
All three types	0.061 *** (0.013)	0.101 *** (0.023)	0.090 *** (0.028)
Economic utility	0.071 ** (0.030)	0.062 (0.057)	0.137 ** (0.066)
Ecological utility	0.008 (0.026)	0.064 (0.051)	0.053 (0.055)
Operational utility	0.047 *** (0.016)	0.065 ** (0.029)	0.067 ** (0.032)
Control variables	Controlled	Controlled	Controlled
Fixed effects	Controlled	Controlled	Controlled
R2	0.313	0.336	0.403
Observations	504	504	504

“***” indicates significance at the 1% level, “**” indicates significance at the 5% level, and “*” indicates significance at the 10% level.

reports the robustness checks. Across all specifications, the core results hold. Government-led demonstrations and their interaction with peer-driven demonstrations remain significant and directionally consistent. Notably, enterprise-led demonstrations continue to show a negative effect, and operational utility remains a significant mediator. Some variables show minor differences in statistical significance compared with the baseline regression, but the overall direction of the coefficients remains consistent. These differences may partly stem from changes in sample structure and variance distribution caused by excluding extreme observations and recategorizing the dependent variable. Compressing the five-level adoption variable into three levels merges some intermediate categories, which may weaken or amplify marginal effects across different groups. Another possible reason is that the influence of control variables and measurement errors may vary across sample structures, with significance levels more likely to weaken in smaller samples or when category boundaries are less distinct. In addition, some differences in significance may reflect genuine heterogeneity among farmers, as high-adoption and low-adoption groups may respond fundamentally differently to various demonstration modes. Therefore, despite fluctuations in significance levels under different robustness settings, the consistency of coefficient directions and the underlying causal logic further confirm the robustness of the core conclusions.

The results of the mechanism analysis are presented in

Table 12. Mediation Effects of Different Demonstration Models.

Models	Mediation Effect
	Economic Utility	Ecological Utility	Operational Utility
Government-led only	0.043 (0.028)	0.029 (0.029)	0.022 ** (0.11)
Enterprise-led only	0.040 (0.026)	0.023 (0.023)	0.048 ** (0.20)
Gov × Peer	0.015 (0.010)	0.008 (0.008)	0.013 ** (0.005)
All three types	0.003 (0.002)	0.002 (0.002)	0.003 ** (0.001)

“**” indicates significance at the 5% level.

. Among the three cognitive dimensions, operational utility consistently mediates the relationship between demonstration models and CST adoption. This effect is especially pronounced under enterprise-led demonstrations, which tend to promote technologies that are marketed as easy to operate and user-friendly, thereby lowering the psychological threshold for experimentation and increasing farmers’ willingness to adopt. These findings emphasize that while economic and ecological utility may be conceptually important, farmers’ adoption decisions are more directly shaped by perceptions of usability and control. Future extension programs should therefore prioritize hands-on experience, improve communication strategies, and foster interaction during demonstrations.

Despite the overall rejection of Hypothesis 3, heterogeneity analysis was conducted to explore whether farmers of different operational scales respond differently to combinations of demonstration models. Based on local agricultural practices in Xinjiang, farmers were divided into smallholders (<300 mu; mu is a traditional Chinese unit of area) and large-scale farmers (≥300 mu), considering differences in labor organization and managerial practices.

As shown in

Table 13. Heterogeneity Regression Results by Farm Size.

Models	Mediation Effect
	Smallholders	Large-Scale Farmers
Government-led only	0.976 *** (0.170)	0.853 ** (0.437)
Enterprise-led only	−0.128 (0.140)	−0.854 (0.747)
Peer-driven only	0.251 (0.175)	0.138 (0.529)
Gov × Ent	−0.050 (0.058)	0.103 (0.313)
Gov × Peer	−0.490 *** (0.081)	−0.558 ** (0.221)
Ent × Peer	0.027 (0.089)	−0.013 (0.362)
All three types	0.112 *** (0.027)	0.105 (0.110)
Control variables	Controlled	Controlled
Fixed effects	Controlled	Controlled
R2	0.373	0.389
Observations	414	90

“***” indicates significance at the 1% level, and “**” indicates significance at the 5% level.

, government-led demonstrations and three-way interactions significantly influenced smallholders, whereas the same interactions had no significant effect on large-scale farmers. This suggests that smallholders, with more limited access to information and weaker individual judgment, rely more on external signals from demonstrations. Conversely, large-scale farmers may make adoption decisions based on economic returns and institutional incentives. The negative interaction effect between government-led and peer-driven demonstrations is observed in both groups but is stronger for large-scale farmers. One explanation may be competition for policy resources. Early adopters, typically large-scale farmers, secure most of the subsidies or institutional rewards, reducing the marginal utility of adoption for others. Among smallholders, the joint demonstrations involving government-led and peer-driven models failed to deliver noticeable economic benefits. As a result, smallholders perceived the adoption of such CSTs as economically irrational, leading to a negative impact on their adoption decisions. This finding highlights the need for tailored extension strategies that take farm scale into account and optimize the allocation and transmission mechanisms of diverse demonstration resources.

Lastly,

Table 14. Heterogeneous Mediation Effects by Farm Size.

Models	Mediation Effect
	Smallholders			Large-Scale Farmers
	Economic	Ecological	Operational	Economic	Ecological	Operational
Government-led only	0.048 (0.033)	0.031 (0.035)	0.018 (0.12)	0.157 (0.153)	−0.015 (0.116)	0.026 (0.34)
Gov × Peer	0.016 (0.011)	0.009 (0.010)	0.010 * (0.006)	0.065 (0.064)	−0.005 (0.038)	0.036 (0.035)
All three types	0.004 (0.003)	0.002 (0.002)	0.002 * (0.001)	-	-	-

“*” indicates significance at the 10% level.

explores the heterogeneous mediation paths across farm sizes. Among smallholders, operational utility is the primary channel, with marginal statistical significance. Economic and ecological cognition are not significant. This suggests that smallholders’ adoption decisions are often passive responses to policy signals rather than deliberate assessments. In contrast, none of the three cognitive paths are significant for large-scale farmers, implying the presence of non-cognitive adoption logics such as policy expectations. Once government-supported demonstration projects are perceived as saturated, late adopters, especially large-scale ones, may opt out due to a perceived loss of potential rewards.

4. Discussion

4.1. Theoretical Contributions

Unlike existing studies that tend to focus on the adoption decision of a single technology or the effectiveness of a particular type of demonstration, this study adopts an integrated perspective on the application of CSTs. Five representative CSTs applicable to cotton production in Xinjiang are conceptualized as a unified technology bundle. By constructing a systematic analytical framework along the pathway of “demonstration–cognition–adoption”, the study offers a more comprehensive examination of farmers’ adoption behaviors. This approach addresses limitations in previous research that treated technology adoption in an overly fragmented or ambiguous manner.

Furthermore, this study situates technology adoption within a broader ecological and socio-economic context, thereby further advancing the theoretical understanding of agricultural green transformation. Recent global reviews of carbon footprint research have identified agricultural emission mitigation as an important emerging research frontier, indicating that the micro-level mechanisms driving farmers’ adoption behaviors are increasingly linked to macro-level climate governance and carbon neutrality objectives [55,56]. This trend suggests that agricultural technology adoption should not be regarded merely as a short-term economic choice driven by immediate returns, but rather as part of a long-term development agenda aimed at enhancing the resilience of agricultural systems and promoting the sustainable use of resources [57]. Against this backdrop, exploring how differentiated technology dissemination pathways can effectively align individual farmers’ adoption behaviors with regional and even global carbon mitigation strategies has become a critical direction for research on agricultural sustainability.

Against this backdrop, the present study challenges the dominant assumption that technology demonstration is a homogeneous and unidimensional intervention [34,36,37]. By categorizing demonstration types based on the behavioral logic of the actors involved, three distinct modes are identified and analyzed, and the differential mechanisms through which each mode influences adoption decisions are systematically investigated. This theoretical innovation not only revises the oversimplified treatment of demonstration activities in prior studies but also directly responds to recent calls for more nuanced understandings of how demonstration scenarios should account for farmer heterogeneity [33]. This perspective integrates ecological considerations, heterogeneous resource endowments, and differentiated behavioral motivations into a unified analytical framework, thereby enriching the theoretical understanding of how demonstrations shape technology cognition and adoption in agricultural contexts. It also provides a conceptual foundation for linking micro-level adoption behaviors with macro-level objectives of reducing agricultural carbon emissions and achieving sustainable rural transformation.

4.2. Practical Contributions

This study provides valuable empirical insights and actionable implications for the green transformation of Xinjiang’s cotton industry and the effective dissemination of CSTs. Against the backdrop of agricultural decarbonization becoming a national strategic priority, constructing an efficient technology dissemination system and enhancing farmers’ adoption levels have emerged as critical issues for achieving sustainable agricultural development. By quantifying the impact of CST demonstrations on adoption decisions among cotton farmers, and systematically identifying the influencing mechanisms of different technology demonstration modes, this study offers clear policy direction and practical guidance.

The findings suggest that technology demonstrations influence adoption behavior through the mediating mechanism of technology cognition. Among various demonstration types, government-led demonstrations play a central role in the dissemination system. While enterprise-led demonstrations also exert significant influence, their effectiveness largely depends on the credibility and reputation of the enterprises involved. Notably, farmers commonly regard economic utility as a major consideration in their adoption decisions, but the ease of operation proves to be a more decisive factor for the successful diffusion of technologies. Furthermore, the effects of demonstrations are significantly shaped by the characteristics of the target farmers, and their adoption decisions are conditioned by heterogeneous decision-making logics based on resource endowments.

Although the empirical results show that peer-driven demonstrations and the mediating effects of economic and ecological utility are not statistically significant, their practical value in CST dissemination should not be overlooked. In rural settings characterized by strong social ties, peer-driven demonstrations can enhance the credibility and acceptability of technologies through informal interpersonal communication. This form of demonstration is particularly persuasive in contexts with high uncertainty and information asymmetry, playing a vital role at the tail end of the dissemination chain. While economic utility cognition may not exhibit strong mediation effects statistically, it remains one of the most critical factors shaping farmers’ willingness to adopt and continue using CSTs. Strengthening farmers’ perception of expected cost–benefit outcomes and enhancing the visualization of economic benefits are essential strategies to increase adoption rates. Meanwhile, although ecological utility cognition exerts a weaker short-term influence on adoption, it holds long-term institutional value in improving agricultural resilience and resource sustainability, especially under the dual-carbon strategy. As such, these dimensions, despite their limited statistical significance, should be valued in practice and further explored in the design of dissemination strategies and incentive systems.

Based on these findings, the promotion of CSTs should emphasize multi-dimensional coordination between institutional design and implementation pathways. At the upstream stage of the dissemination chain, the central position of government-led demonstrations should be consolidated by strengthening the grassroots extension force, improving demonstration site management systems, and enhancing on-site technical services to ensure the continuity, professionalism, and broad coverage of demonstrations. During the dissemination process, the potential of enterprise-led demonstrations should be further leveraged by establishing a credit system for enterprise-based technology dissemination, clarifying technical standards and service protocols, and enhancing transparency to improve enterprises’ credibility among farmers. In parallel, the heterogeneity in farmer resource endowments should be acknowledged, and tailored demonstration strategies should be implemented to match specific farmer profiles. Emphasis should also be placed on demonstrating operational utility, especially through hands-on training, step-by-step video guides, and other approaches that reduce learning costs and boost adoption motivation. At the downstream stage, peer-driven demonstrations should be harnessed through mechanisms such as “model farmers + demonstration households”, enabling farmer-to-farmer knowledge exchange to complement formal dissemination in terms of coverage and psychological recognition. Moreover, efforts should be made to integrate CSTs with farmers’ income-enhancement goals by showcasing real-life input-output ratios and income gains from pilot adopters to strengthen their perception of economic returns. Lastly, although ecological utility may not directly drive adoption decisions, it is increasingly important in shaping green incentive systems. Therefore, it should be incorporated into long-term promotional messaging and linked with policies such as green subsidies and environmental performance evaluations, thereby achieving an organic integration of short-term incentives with long-term sustainability objectives.

4.3. Feasibility Analysis

Building on the technology extension strategies proposed earlier from the perspective of the technology promotion chain, the demonstration of CSTs must strike a balance between feasibility and sustainability under real-world constraints of limited budgets and restricted resources. Government-led demonstrations, as the upstream component of the promotion chain, play a critical role in shaping farmers’ technology cognition. However, due to their high organizational costs and fiscal dependence, large-scale or full-sector implementation is difficult. Under constraints of water, land, and fiscal resources, a strategy of concentrated pilot projects and targeted breakthroughs should be adopted. A set of CSTs with both economic and ecological benefits should be prioritized for government-led demonstrations in core areas with high adoption potential and strong spillover effects. A “small-scale, high-standard” pilot approach can then diffuse to surrounding regions, minimizing redundant investment and improving the marginal efficiency of promotion. Enterprise-led demonstrations, as the midstream component of the promotion chain, hold considerable potential. Compared with government-led demonstrations, enterprises possess stronger financial mobilization capacity and greater market sensitivity, with lower marginal fiscal costs of promotion. However, the demonstration effect of agricultural enterprises is currently limited by deficiencies in brand credibility and service standardization. To balance budget savings with promotion efficiency, a credit record system for enterprise technology promotion should be established. “Reputable enterprises” could be granted policy subsidies and demonstration incentives, while government–enterprise cooperation should define clear technical standards and information disclosure mechanisms to avoid excessive sales-driven bias that could undermine farmers’ trust. Such measures would reduce direct government spending on mid-term promotion while encouraging enterprises to develop a self-sustaining promotion mechanism through market-based incentives.

Peer-driven demonstrations, as the downstream component of the promotion chain, have the advantages of low cost and high psychological acceptance. They are particularly suitable as a supplementary and diffusion mechanism to government and enterprise demonstrations under resource-constrained conditions. However, their effectiveness depends on the quality of the demonstration source and peers’ demonstration capability. Thus, a “model farmers + demonstration households” cultivation mechanism should be established, providing small-scale subsidies and technical training to high-quality peer demonstrators within budgetary limits. This would foster sustained reputation effects within social networks, achieving large-scale diffusion with relatively low investment. Considering farmers’ heterogeneity in resource endowment and adoption willingness, targeted matching and tiered promotion strategies are recommended under budget constraints. For instance, limited fiscal and technical resources should be concentrated on large-scale farmers with high sensitivity to economic returns and strong adoption potential to maximize demonstration efficiency, whereas smallholders should primarily rely on peer demonstrations and simplified, easy-to-operate technologies to reduce promotion costs.

Overall, the effectiveness of demonstration models depends not merely on the intensity of investment in individual components but on the organic integration of government-led demonstrations for “efficient leadership,” enterprise-led demonstrations for “market-driven promotion,” and peer-driven demonstrations for “low-cost diffusion.” Such integration is essential to achieving sustainable and large-scale CSTs promotion under budget constraints.

4.4. Limitations and Future Directions

Despite the theoretical advancements and empirical contributions of this study to understanding the adoption mechanisms of CSTs, several limitations remain that warrant further investigation and refinement in future research.

First, the sample size and representativeness are limited. Due to constraints in research resources and sample accessibility, the data are mainly drawn from major cotton-producing areas in Xinjiang, with a relatively small sample size. In subgroup regressions and heterogeneity analyses, some subsamples are particularly small, potentially resulting in wider confidence intervals and limited statistical power. Moreover, as the sample is region-specific, regional particularities in agricultural policies, market conditions, and social network structures may limit the generalizability of the findings to other regions or crops. Future studies should conduct cross-regional and multi-wave tracking surveys to build panel or longitudinal datasets, which would not only enhance model robustness but also help uncover the dynamic evolution of technology adoption.

Second, causal identification strategies require further improvement. This study employs ordered logit and mediation effect models based on cross-sectional data to estimate the effects of demonstration and cognition on adoption behavior; however, the design cannot fully address potential endogeneity issues. Demonstration participation is unlikely to be completely random, and farmers’ willingness to adopt may, in turn, influence their probability of participating in demonstrations, introducing potential selection bias. Furthermore, unobserved individual traits such as risk preference and innovativeness may jointly affect technology cognition and adoption behavior, leading to omitted variable bias. Future research could adopt more rigorous causal identification strategies, such as instrumental variable approaches, propensity score matching, difference-in-differences, or quasi-natural experiments, to enhance the causal interpretability of the findings.

Third, the measurement precision of key variables can be improved. Technology cognition, as a core mediating variable, provides important insights into mechanisms but remains subject to non-negligible measurement errors due to its subjective nature. On one hand, farmers’ understanding of CSTs is highly heterogeneous, and their responses on cognition may be influenced by education level, information sources, and social desirability bias. On the other hand, the wording of cognition questions in surveys may also be affected by farmers’ language comprehension and social desirability, causing potential discrepancies between reported and actual cognition levels. Future research could adopt quasi-experimental methods to cross-validate the formation mechanisms of cognition from multiple dimensions, thereby strengthening both theoretical explanatory power and empirical validity.

Fourth, the measurement of demonstration modes remains insufficiently refined. Although this study categorizes demonstration modes mainly based on the attributes of demonstration actors, which is logically distinguishable, it does not fully capture the more critical procedural elements of “how demonstrations are implemented.” Factors such as demonstration frequency, interaction depth, and technology maturity can profoundly influence farmers’ willingness and behavioral decisions. Future research should develop a multi-dimensional index system for demonstration intensity to analyze the dynamic coupling among “demonstration content–cognition change–behavioral response,” providing an empirical basis for more targeted and differentiated promotion strategies.

Finally, external effects under resource constraints have not been fully explored. Although this study discusses feasible promotion strategies and resource allocation, it does not quantitatively analyze the optimal promotion pathways under different fiscal budgets and resource constraints, nor does it adequately consider potential externalities or diminishing marginal effects of large-scale promotion. Future research could integrate agricultural extension economics and policy simulation frameworks, combining cost–benefit analysis with scenario simulations to explore optimal combinations of promotion models under fiscal constraints and assess their long-term sustainability.

5. Conclusions

Against the backdrop of escalating global climate change and the deepening transition toward low-carbon agriculture, the effective diffusion and widespread adoption of CSTs has become a critical pathway toward achieving sustainable agricultural development. Focusing on this core issue, this study examines cotton farmers in Xinjiang and, based on 504 micro-level survey responses, constructs a theoretical framework linking technology demonstration, technology cognition, and technology adoption. It systematically investigates the influence mechanisms of different demonstration models on CST adoption behaviors, reveals the mediating role of cognition, and identifies the heterogeneity of demonstration effects and their underlying logic.

The findings show that technology demonstration indirectly influences CST adoption by shaping farmers’ cognition, with technology cognition serving as a key mediating mechanism. Among the various models, government-led demonstrations play a central role in the current extension system, while operational utility is identified as the most critical cognitive pathway in the adoption process. Although enterprise-led demonstrations also contribute to technology dissemination, their effectiveness largely depends on brand reputation and service quality. Farmers’ willingness to adopt is often determined by their level of trust in the enterprise brand.

In terms of technological characteristics, although economic utility is commonly viewed by farmers as an important criterion for adoption, empirical results indicate that perceived ease of use is the core factor throughout the diffusion chain. The simpler and more intuitive a technology is to understand and implement, the higher its likelihood of being accepted and replicated. Furthermore, the effectiveness of a demonstration is significantly influenced by the characteristics of the demonstration recipients. Farmers with different farm sizes, resource endowments, and management models respond differently to demonstration activities, and their underlying decision-making logic also diverges. Therefore, effective CST promotion strategies must account for farmer heterogeneity and be tailored to local conditions and farmer types, promoting a precise and tiered extension system to improve adoption levels and dissemination efficiency.

In summary, this study contributes theoretically by enriching the micro-level understanding of CST adoption mechanisms and expanding the knowledge of demonstration heterogeneity. Empirically, it provides insights for promoting CST diffusion and differentiated extension in Xinjiang and other regions. Furthermore, the promotion system for CSTs should be improved through multi-dimensional coordination of institutional design and practical pathways. Efforts should focus on reinforcing the linkage between demonstration mechanisms and farmers’ cognitive systems and encouraging multi-actor collaboration to implement targeted demonstration strategies. Future research should consider integrating larger samples, quasi-experimental designs, and finer-grained behavioral data to deepen the understanding and prediction of CST adoption dynamics, providing a more forward-looking and operational scientific basis for addressing climate change challenges in agriculture.