In recent years, the adoption of agricultural drone technology has emerged as a pivotal component in modernizing farming practices, particularly in enhancing precision agriculture, crop monitoring, and pest control. However, the diffusion of such innovations varies significantly across regions, with ethnic minority areas often lagging due to socioeconomic and cultural barriers. This study aims to investigate the factors influencing the adoption behavior of agricultural drone technology among farmers in the Tibetan-Qiang-Yi Corridor, a culturally diverse region in China, by integrating the Technology Acceptance Model (TAM) framework with a Probit-ISM analytical approach. The findings are expected to inform policy interventions that bridge the gap between technological advancements and grassroots implementation, thereby supporting rural revitalization and agricultural modernization.
The theoretical foundation of this research rests on the Technology Acceptance Model (TAM), originally developed by Davis to explain how perceived usefulness and perceived ease of use shape individuals’ intentions to adopt new technologies. In the context of agricultural drone technology, perceived usefulness refers to the extent to which farmers believe that using drones will improve their productivity and economic returns, while perceived ease of use relates to the convenience of accessing and operating these technologies. Previous studies have applied TAM in various domains, but its integration with econometric models like Probit and structural analysis tools like Interpretive Structural Modeling (ISM) remains underexplored, especially for minority regions. This gap motivates our study to develop a comprehensive framework that captures both the psychological and systemic factors affecting adoption decisions.
To contextualize our research, we review existing literature on agricultural technology adoption. Scholars have identified multiple determinants, including farmer characteristics, farm size, resource endowments, and government support. For instance, land scale has been shown to influence adoption behavior, but the direction of this effect is contested—some studies report positive correlations, while others indicate negative or nonlinear relationships. Moreover, most research focuses on plain areas in eastern and central China, with scant attention to ethnic corridors like the Tibetan-Qiang-Yi region. This oversight limits the generalizability of findings and underscores the need for localized studies. By employing a Probit-ISM model, we delve into the hierarchical structure of影响因素, distinguishing between surface-level factors (e.g., perceptions) and deep-rooted causes (e.g., household attributes), thus offering a nuanced understanding of adoption dynamics.
The core contribution of this paper lies in its methodological innovation. We combine a binary Probit model to quantify the impact of various factors on agricultural drone adoption with an ISM model to unravel the interrelationships and hierarchical ordering of these factors. This dual approach allows us to not only identify significant predictors but also map the causal pathways through which they operate. In the following sections, we detail our data sources, variable definitions, and analytical procedures, followed by a presentation of results, a discussion of implications, and policy recommendations. Throughout the article, we emphasize the term “agricultural drone” to highlight the specific technology under investigation, and we incorporate tables and formulas to summarize key insights.
Our data were collected through field surveys in three autonomous regions of the Tibetan-Qiang-Yi Corridor: Mabian Yi Autonomous County in Sichuan, Ganzi Tibetan Autonomous Prefecture, and Diqing Tibetan Autonomous Prefecture. These areas are characterized by unique geographical and cultural landscapes, with economies predominantly reliant on crop and fruit cultivation. We employed a stratified sampling technique complemented by simple random sampling to ensure representativeness. Specifically, we selected 2–4 administrative villages per region and conducted household-level interviews, distributing 230 questionnaires and obtaining 220 valid responses. The sample includes large-scale farmers, members of agricultural cooperatives, and family farm owners, who are key actors in the adoption of advanced agricultural technologies like agricultural drones.
The sample demographics reveal insights into the respondent profile. Approximately 61.37% of participants are over 36 years old, with a concentration in the 45–59 age group. In terms of education, 135 farmers have a junior high school education or below, while 85 possess a high school diploma or higher. Regarding farming experience, 30% have been engaged in cultivation for less than five years, indicating a trend of returning migrants who may be more receptive to new technologies. Economically, 81.82% report annual farming incomes below 100,000 CNY, suggesting limited financial capacity and risk tolerance. These characteristics are controlled for in our analysis to isolate the effects of perceptual and external factors.
To model the adoption behavior, we define the dependent variable as a binary indicator: 1 if the farmer has adopted agricultural drone technology, and 0 otherwise. The independent variables are categorized into four core dimensions based on the TAM framework: perceived usefulness, perceived ease of use, technology promotion services, and government support. Additionally, we include control variables for individual and household characteristics, as well as a mediating variable for scale of operation. The measurement of each variable is detailed in Table 1, which summarizes the definitions and descriptive statistics.
| Variable Category | Variable Name | Definition | Mean | Std. Dev. |
|---|---|---|---|---|
| Perceived Usefulness | Economic Benefit | Extent to which agricultural drone increases income (1–5 scale) | 3.45 | 1.12 |
| Support Level | Degree of support for agricultural drone technology (1–5 scale) | 3.78 | 0.98 | |
| Understanding Degree | Self-assessed knowledge about agricultural drone (1–5 scale) | 2.91 | 1.05 | |
| Promotion Willingness | Willingness to promote agricultural drone to others (1–5 scale) | 3.62 | 1.07 | |
| Perceived Ease of Use | Usage Effect | Perceived effectiveness of agricultural drone (1–5 scale) | 3.28 | 1.14 |
| Technical Cognition | Familiarity with agricultural drone operations (1–5 scale) | 3.15 | 1.08 | |
| Channel Accessibility | Ease of accessing information about agricultural drone (1–5 scale) | 3.51 | 1.02 | |
| Policy Attention | Awareness of government policies on agricultural drone (1–5 scale) | 3.22 | 1.11 | |
| Technology Promotion Services | Promotion Agents | Number of sources promoting agricultural drone (count) | 2.34 | 1.21 |
| Promotion Modes | Diversity of promotion methods for agricultural drone (1–5 scale) | 3.08 | 1.16 | |
| Government Support | Funding Support | Level of government subsidies for agricultural drone (1–5 scale) | 3.41 | 1.09 |
| Policy Publicity | Extent of policy宣传 for agricultural drone (1–5 scale) | 3.55 | 1.03 | |
| Control Variables | Age | Farmer’s age in years | 48.73 | 10.45 |
| Education | Education level (1=low to 5=high) | 2.67 | 1.22 | |
| Farming Experience | Years engaged in farming | 12.34 | 8.91 | |
| Farm Size | Total cultivated area in mu | 25.67 | 18.23 | |
| Household Income | Annual household income from farming (10,000 CNY) | 8.45 | 6.78 | |
| Number of Farmers | Number of family members engaged in farming | 3.12 | 1.45 | |
| Mediating Variable | Scale Degree | Extent of farm规模化 (1–5 scale) | 3.18 | 1.17 |
The empirical analysis proceeds in two stages. First, we estimate a binary Probit model to assess the marginal effects of each independent variable on agricultural drone adoption. The Probit model is specified as follows:
$$ P(Y_i = 1 | X_i) = \Phi(X_i \beta) $$
where \(Y_i\) is the binary adoption decision for farmer \(i\), \(X_i\) is a vector of explanatory variables, \(\beta\) is a vector of coefficients, and \(\Phi(\cdot)\) denotes the cumulative distribution function of the standard normal distribution. The marginal effect of variable \(x_j\) is computed as:
$$ \frac{\partial P(Y_i=1|X_i)}{\partial x_j} = \phi(X_i \beta) \beta_j $$
with \(\phi(\cdot)\) being the standard normal density function. We conduct diagnostic tests for multicollinearity using Variance Inflation Factors (VIF), with all values below 5, indicating no severe collinearity issues.
Second, we employ the Interpretive Structural Modeling (ISM) approach to dissect the hierarchical structure among the factors. ISM involves constructing a directed graph based on pairwise relationships, derived from expert judgment or empirical results. We define six要素: S1 (perceived usefulness), S2 (perceived ease of use), S3 (technology promotion services), S4 (government support), S5 (individual characteristics), and S6 (household characteristics), with S0 representing agricultural drone adoption behavior. An adjacency matrix \(A\) is created, where \(a_{ij}=1\) if factor \(i\) influences factor \(j\), and 0 otherwise. Through matrix multiplication, we derive the reachability matrix \(R\), which satisfies \(R = (A + I)^k = (A + I)^{k+1}\) for some integer \(k\), where \(I\) is the identity matrix. The levels are then extracted by partitioning the要素 based on their reachability and antecedent sets.
Before presenting the regression results, we assess the reliability and validity of our survey instrument. The overall Cronbach’s alpha coefficient is 0.757, and all variable-specific alpha values exceed 0.6, confirming acceptable internal consistency. The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy is 0.815, and Bartlett’s test of sphericity yields a p-value of 0.000, indicating that the data are suitable for factor analysis. To address potential common method bias, we apply Harman’s single-factor test, which extracts six factors with eigenvalues greater than 1, and the largest factor accounts for 26.07% of the variance (below the 40% threshold), suggesting that bias is not a serious concern.
The Probit regression results are summarized in Table 2, which reports coefficient estimates, p-values, and marginal effects. We estimate three nested models: Model 1 includes only control and mediating variables; Model 2 adds the four core TAM dimensions; and Model 3 incorporates all variables. The full model (Model 3) is discussed here, as it provides the most comprehensive insights.
| Variable | Coefficient | P-value | Marginal Effect |
|---|---|---|---|
| Perceived Usefulness | |||
| Economic Benefit | -0.059 | 0.729 | -0.012 |
| Support Level | 0.371 | 0.006*** | 0.073 |
| Understanding Degree | -0.372 | 0.030** | -0.073 |
| Promotion Willingness | 0.221 | 0.091* | 0.044 |
| Perceived Ease of Use | |||
| Usage Effect | 0.054 | 0.680 | 0.011 |
| Technical Cognition | 0.396 | 0.004*** | 0.078 |
| Channel Accessibility | 0.470 | 0.011** | 0.092 |
| Policy Attention | 0.094 | 0.500 | 0.019 |
| Technology Promotion Services | |||
| Promotion Agents | -0.073 | 0.520 | -0.014 |
| Promotion Modes | -0.059 | 0.730 | -0.012 |
| Government Support | |||
| Funding Support | 0.245 | 0.077* | 0.048 |
| Policy Publicity | 0.244 | 0.099* | 0.048 |
| Control Variables | |||
| Age | -0.052 | 0.775 | -0.010 |
| Education | -0.203 | 0.117 | -0.040 |
| Farming Experience | 0.012 | 0.919 | 0.002 |
| Farm Size | 0.347 | 0.048** | 0.068 |
| Household Income | 0.038 | 0.826 | 0.007 |
| Number of Farmers | -0.225 | 0.049** | -0.044 |
| Mediating Variable | |||
| Scale Degree | -0.051 | 0.709 | -0.010 |
| Notes: * p<0.1, ** p<0.05, *** p<0.01; N=220. | |||
The results indicate that within perceived usefulness, support level has a significant positive effect on agricultural drone adoption (p<0.01), while understanding degree has a negative effect (p<0.05). This suggests that farmers who endorse the technology are more likely to adopt it, but those with deeper knowledge may perceive complexities or costs that deter adoption. Promotion willingness is positively significant at the 10% level, highlighting the role of social advocacy. For perceived ease of use, technical cognition and channel accessibility are positively associated with adoption (p<0.01 and p<0.05, respectively), emphasizing the importance of familiarity and information access. Interestingly, usage effect and policy attention are not significant, possibly due to limited direct experience with agricultural drones among non-adopters.
Technology promotion services show no significant impact, reflecting the nascent stage of extension systems in the region. In contrast, government support, through funding and policy publicity, positively influences adoption at the 10% significance level, underscoring the enabling role of public interventions. Among control variables, farm size positively affects adoption, as larger operations may benefit more from agricultural drone efficiencies. Conversely, the number of farmers in the household negatively impacts adoption, likely due to labor abundance reducing the urgency for labor-saving technologies like agricultural drones. Other variables, such as age and education, are insignificant, aligning with some prior studies that found perceptual factors outweigh demographics in technology adoption decisions.
To delve deeper into the interrelationships, we apply the ISM methodology. Based on the regression insights and expert inputs, we construct the adjacency matrix A, where influences are coded as 1 or 0. For example, S5 (individual characteristics) influences S1 (perceived usefulness) because personal traits shape perceptions. The reachability matrix R is computed through iterative multiplication, and the level partitioning yields a four-tier hierarchy. The results are visualized in a directed graph, but since we cannot include images, we describe the structure: Level 1 consists of S0 (agricultural drone adoption behavior); Level 2 includes S1 and S2 (perceived usefulness and ease of use); Level 3 comprises S3 and S4 (technology promotion services and government support); and Level 4 contains S5 and S6 (individual and household characteristics). This hierarchy implies that surface-level perceptions directly drive adoption, while external support services act as intermediaries, and farmer attributes are foundational determinants.
The ISM analysis can be formalized mathematically. Let the adjacency matrix A be defined as:
$$ A = \begin{pmatrix}
0 & a_{01} & a_{02} & a_{03} & a_{04} & a_{05} & a_{06} \\
a_{10} & 0 & a_{12} & a_{13} & a_{14} & a_{15} & a_{16} \\
a_{20} & a_{21} & 0 & a_{23} & a_{24} & a_{25} & a_{26} \\
a_{30} & a_{31} & a_{32} & 0 & a_{34} & a_{35} & a_{36} \\
a_{40} & a_{41} & a_{42} & a_{43} & 0 & a_{45} & a_{46} \\
a_{50} & a_{51} & a_{52} & a_{53} & a_{54} & 0 & a_{56} \\
a_{60} & a_{61} & a_{62} & a_{63} & a_{64} & a_{65} & 0
\end{pmatrix} $$
where rows and columns correspond to S0 through S6. Based on our study, we assign values such that \(a_{01}=a_{02}=1\) (direct influences), \(a_{13}=a_{14}=1\), \(a_{23}=a_{24}=1\), \(a_{35}=a_{36}=1\), and \(a_{45}=a_{46}=1\), with others set to 0. The reachability matrix R is obtained by solving \(R = (A + I)^k\) until convergence. The level extraction process involves computing the reachability set \(R(s_i)\) and antecedent set \(A(s_i)\) for each要素, and then identifying those with \(R(s_i) \cap A(s_i) = R(s_i)\) for the top level. This yields the hierarchical model described above.
Our findings spark a discussion on the challenges and opportunities for promoting agricultural drone technology in ethnic corridors. The negative association between understanding degree and adoption suggests that mere knowledge dissemination may not suffice; instead, practical demonstrations and cost-benefit analyses are needed to align perceptions with realities. The positive role of government support underscores the necessity for tailored policies, such as subsidies for agricultural drone purchases or training programs. However, the insignificance of technology promotion services indicates a gap in extension mechanisms, urging a shift from top-down approaches to participatory models that involve farmers as co-creators.
To optimize adoption pathways, we propose a multi-pronged strategy. First, foster collective action through farmer organizations, such as cooperatives, to amplify learning and reduce risks. These groups can facilitate peer-to-peer knowledge sharing on agricultural drone applications, thereby enhancing perceived usefulness and ease of use. Second, adopt a farmer-centric approach by conducting needs assessments to tailor agricultural drone technologies to local contexts, such as developing drones suitable for mountainous terrains prevalent in the Tibetan-Qiang-Yi Corridor. Third, innovate socialized promotion systems by establishing digital platforms that connect government agencies, research institutions, and farmers, enabling real-time information exchange on agricultural drone maintenance, best practices, and policy updates.
Furthermore, we recommend integrating agricultural drone technology into broader rural development initiatives. For instance, linking drone adoption with precision agriculture projects can demonstrate tangible benefits like yield increase and input savings. Financial instruments, such as low-interest loans for agricultural drone acquisition, could alleviate cost barriers. Additionally, capacity-building workshops should target not only technical skills but also risk management, as farmers’ aversion to uncertainty often hinders adoption. By addressing these dimensions, stakeholders can create an enabling ecosystem where agricultural drones become a mainstream tool for sustainable farming.
In conclusion, this study elucidates the complex factors shaping agricultural drone adoption behavior in the Tibetan-Qiang-Yi Corridor. Using a Probit-ISM integrated model, we quantify the impacts of perceptual, promotional, and governmental variables, while revealing a hierarchical structure where farmer characteristics underlie all other influences. The adoption of agricultural drone technology is directly driven by perceived usefulness and ease of use, indirectly mediated by support services, and fundamentally rooted in household and individual attributes. These insights call for holistic interventions that combine psychological nudges, institutional reforms, and targeted subsidies. Future research could explore longitudinal dynamics or comparative studies across diverse ethnic regions to generalize findings. Ultimately, advancing agricultural drone diffusion in minority areas requires a synergy of technology, policy, and community engagement, paving the way for inclusive agricultural modernization.
To encapsulate the key relationships, we present a summary equation derived from our Probit estimates. The probability of adopting agricultural drone technology can be approximated as:
$$ P(\text{Adoption}) = \Phi(0.371 \cdot \text{Support} – 0.372 \cdot \text{Understanding} + 0.221 \cdot \text{PromotionWillingness} + 0.396 \cdot \text{TechnicalCognition} + 0.470 \cdot \text{ChannelAccessibility} + 0.245 \cdot \text{Funding} + 0.244 \cdot \text{Publicity} + 0.347 \cdot \text{FarmSize} – 0.225 \cdot \text{NumberFarmers}) $$
This formulation highlights the relative strengths of various factors, with perceptual and accessibility variables playing prominent roles. It serves as a tool for policymakers to simulate the effects of interventions, such as increasing funding support or improving technical training, on adoption rates.
In summary, the journey toward widespread agricultural drone adoption in minority regions is fraught with challenges but ripe with opportunities. By leveraging empirical evidence and systemic analysis, we can design pathways that not only boost technology uptake but also empower farmers as active agents in the agricultural innovation ecosystem. The agricultural drone, as a symbol of modern precision farming, holds the potential to transform livelihoods in the Tibetan-Qiang-Yi Corridor and beyond, contributing to food security and rural prosperity.