What Drives or Hinders the Adoption of Sustainable Smart Logistics in Rural Areas?—A Mixed-Methods Analysis

Abstract

Under the impetus of digital rural development and the rapid advancement of smart logistics, intelligent terminal delivery technologies are gradually expanding into rural areas. This study employs a three-phase mixed research approach to systematically investigate the factors influencing and mechanisms underlying rural residents’ willingness to adopt smart logistics in Hebei Province. In the first phase, grounded theory is employed to identify seven key perceived factors: perceived usefulness, perceived ease of use, sensitivity to collective evaluation, cultural conservatism, infrastructure quality, facilitating conditions, and technological trust. In the second phase, integrating the TAM and the UTAUT, this study incorporates context-specific variables and conducts empirical analysis using SEM and the bootstrap method on 451 valid questionnaire responses. The results indicate that all factors except infrastructure quality significantly influence adoption willingness, with cultural conservatism exerting a negative effect. In the third phase, fsQCA is applied to identify eight configurations that lead to high adoption willingness, further supplementing and enriching the explanatory power of the SEM results. This research expands the theoretical understanding of smart logistics technology adoption mechanisms in rural areas and offers practical guidance for the promotion and application of related technologies.

1. Introduction

Against the backdrop of rapid advancements in automation technologies, continuous improvement of digital infrastructure, and growing emphasis on sustainable development, global logistics systems are undergoing profound transformations. These changes have not only enhanced the responsiveness of supply chains but also imposed higher demands on last-mile delivery, particularly driven by the rapid expansion of e-commerce [1]. Traditionally, last-mile logistics has focused on urban areas, relying on well-developed transportation networks and high population density. However, with the rise of rural e-commerce and the implementation of rural revitalization strategies, the importance of rural logistics is becoming increasingly prominent. Market research indicates that the volume of rural express deliveries in China reached 37 billion parcels in 2024, with the market size of agricultural products flowing into cities and industrial goods flowing into rural areas exceeding CNY 1.85 trillion [2]. Meanwhile, rural online retail sales have steadily increased, reaching CNY 2.56 trillion in 2024 [3]. This trend indicates that rural areas have become a key driver of national economic growth and consumption upgrading. Therefore, building an efficient, intelligent, and sustainable rural logistics system is of great significance for advancing agricultural modernization and promoting coordinated urban-rural development [4,5].

Rural end-point logistics in China faces significant challenges such as low population density and weak transportation and logistics infrastructure, which severely limit the accessibility and efficiency of traditional services [6,7]. To address these issues, the Chinese government has promoted the construction of a three-tier logistics network, aiming to integrate resources and enhance service capabilities [8]. However, problems such as insufficient transport capacity at logistics nodes, low levels of digitalization, and a shortage of skilled personnel remain prominent [9]. Moreover, the high energy consumption of current models conflicts with the national “dual carbon” strategy [10]. With the development of Industry 4.0, technologies such as smart parcel lockers, autonomous delivery vehicles, and drones have gradually been introduced in rural areas to alleviate infrastructure and labor shortages. For instance, JD.com began testing a drone-based logistics system in remote regions as early as 2016 [11]. Smart parcel lockers have significantly improved efficiency and user experience [12], while also promoting the development of green logistics [8,13]. However, intelligent logistics technologies in rural areas are still in the early stages and have yet to achieve large-scale deployment [14,15]. Their promotion continues to face multiple bottlenecks, including poor technological adaptability, immature business models, and insufficient policy incentives.

Existing research on smart delivery technologies has primarily focused on urban environments, with an emphasis on sustainability assessment [12], technological optimization [16], and comparative analysis of delivery models [17]. In contrast, systematic studies on user technology adoption behavior in rural contexts remain relatively scarce [11,18]. Rural areas possess unique sociocultural characteristics, and users exhibit significant differences and diversity in their willingness and cognition regarding the adoption of smart logistics technologies [19], highlighting the urgent need for in-depth exploration tailored to local realities. Moreover, mainstream studies often employ symmetric analysis methods such as SEM, which emphasize linear causal relationships and are limited in uncovering the complex interactive mechanisms and diverse configurational paths that influence user behavior. This restricts a deeper understanding of rural user adoption behavior [20]. To address these research gaps, this study adopts a multistage mixed-methods research design and clearly defines “users” as rural consumers in the context of smart delivery technology adoption. Given the ambiguity of the term “user”, this study specifies it as referring to rural consumers. In the first stage, key cognitive factors are identified through semi-structured interviews and a three-stage grounded theory coding process. In the second stage, an extended model is developed by integrating TAM, UTAUT, and rural context-specific variables. Based on 451 valid questionnaires collected from Hebei Province, SEM is used to test causal relationships among variables. In the third stage, the fsQCA method is employed to identify multiple configurational paths leading to behavioral intention, addressing the limitations of SEM in analyzing behavioral heterogeneity. This study focuses on the following research questions:

RQ1: What are the key perceived factors influencing rural users’ intention to adopt smart last-mile delivery technologies in China? How do these factors individually contribute to adoption intention through distinct pathways?

RQ2: Under conditions of complex interaction, which combinations of perceived factors effectively enhance rural users’ intention to adopt such technologies?

RQ3: How do the results of SEM and fsQCA compare in explaining the usage intentions of rural Chinese users?

This study makes three main contributions: it theoretically introduces new variables such as cultural conservatism and group evaluation sensitivity, extending the applicability of TAM and UTAUT in rural technology adoption and moving beyond the urban-centric perspective; methodologically, it integrates SEM and fsQCA to uncover configurational effects and path differences among multiple factors, addressing the limitations of traditional symmetric analysis; and practically, based on data from rural Hebei, China, it offers policy recommendations to promote the adoption of smart logistics technologies, supporting urban-rural logistics integration and the development of digital villages.

2. Literature Review

“Last-mile” delivery refers to the final leg of the logistics chain—from the distribution center to the end consumer—and is widely considered the most complex and cost-intensive segment of the supply chain [21,22]. This stage not only significantly influences customer satisfaction and loyalty but also bears critical implications for environmental sustainability, social equity, and economic efficiency [16,23]. With the rapid advancement of AI, automation technologies, and the Internet of Things, intelligent “last-mile” delivery systems are gradually taking shape. These systems encompass a range of advanced technologies, including autonomous delivery vehicles, drones, smart parcel lockers, and delivery robots. Such innovations are increasingly being deployed in urban logistics to enhance delivery efficiency, reduce carbon emissions, and minimize reliance on human labor. In urban environments, such technologies have demonstrated strong performance advantages, leveraging AI-driven route optimization and multimodal coordination mechanisms to significantly improve service quality and resource utilization efficiency [8]. Studies have shown that ground-based autonomous vehicles and drones are effective in alleviating traffic congestion, shortening delivery times, and reducing energy consumption in cities [21,24]. However, in rural areas where infrastructure is underdeveloped, network coverage is limited, and digital literacy varies widely, the implementation and adoption of intelligent delivery technologies face significantly greater obstacles [14,15]. Low order density, long delivery distances, and high per-unit costs—compounded by a growing digital divide—create significant disparities between urban and rural logistics systems, hindering the development of rural smart logistics [25,26].

While the application of smart delivery technologies in urban or well-connected areas has garnered extensive scholarly attention [12,16,17], systematic research focusing on their suitability and user acceptance in rural settings remains limited. Empirical studies specifically addressing smart last-mile delivery technologies and user intentions in rural areas are scarce and fragmented, providing only a partial understanding of the cognitive mechanisms underlying technology adoption. For example, ref. [27] found, in a study of rural South Africa, that smart lockers significantly reduce logistics costs and improve pickup convenience, and subsequently proposed the EUTAUT model to address theoretical gaps in UTAUT related to perceived value and privacy concerns. The authors of [28], integrating FAHP, ISM, and MICMAC methods, identified factors such as “return convenience”, “product integrity”, “scheduled pickup”, and “delivery cost” as key foundational elements influencing the sustainable development of rural e-commerce logistics in China, providing practical insights for service optimization and policymaking. Ref. [29] found that performance expectancy, facilitating conditions, and perceived price value positively affect rural residents’ intention to use smart lockers, while technology anxiety exerts a negative effect. Effort expectancy and social influence were found to be non-significant. Although these studies offer a preliminary theoretical foundation for understanding rural user adoption of smart delivery technologies, they generally lack a systematic and contextualized perspective on Chinese rural users’ acceptance mechanisms. On one hand, most existing studies rely on variable-centered symmetric analysis approaches, which are limited in revealing complex interactions and non-linear causal paths between variables. On the other hand, they fall short in capturing the nuanced perceptions and lived experiences of users within real usage contexts. Addressing these theoretical and methodological gaps is precisely what this study aims to achieve.

3. Extraction of Perceptual Factors Among Rural Residents

3.1. Data Collection

Grounded theory is particularly suitable for exploring under-researched phenomena and for inductively constructing new theoretical frameworks [30]. Its three-level coding strategy offers high flexibility and can be applied to various types of data sources [31]. To enhance the credibility and comprehensiveness of the research data, this study adopted a hybrid data collection approach that combined both online and offline interviews. The interviews focused on key topics, including respondents’ understanding of smart logistics systems in rural areas, their actual experiences, and willingness to adopt intelligent delivery technologies.

Given the wide regional user coverage of the Xiaohongshu platform [32], this study initially prioritized participant recruitment by posting online calls for participation on Xiaohongshu. Through this method, 10 rural users residing in areas with established smart logistics infrastructure were recruited and interviewed via phone. At the same time, recognizing the platform’s demographic skew toward younger users and aiming to mitigate potential sampling bias in terms of age structure [33], the research team conducted a supplementary round of fieldwork. Based on an analysis of the initial sample’s demographic characteristics, researchers visited Jiucheng Township in Xinji City, Hebei Province—an area where smart logistics technologies have been deployed—and conducted face-to-face interviews with 10 additional rural residents who had relevant experience with these technologies. In total, 20 valid interview records were collected (see

Table 1. Descriptive data of interviewers.

Section	Item	Frequency
Gender	Male	10
	Female	10
Age	18–30	6
	31–40	5
	41–50	5
	Above 50	4
Educational level	High school or below	10
	Vocational education	5
	Bachelor	4
	Master or above	1
Job	employee	7
	unemployed	3
	Student	5
Interview time	20–25 min	5
	26–30 min	8
	31–35 min	7

Source: authors.

for participant profiles). To ensure the integrity and authenticity of the data, all interviews were audio-recorded with prior informed consent. Following the interviews, the research team transcribed all recordings verbatim, generating approximately 90,000 words of raw textual material, which provided a robust empirical foundation for the subsequent three-level coding analysis.

3.2. Data Coding and Analysis

After completing data collection, the research team utilized two-thirds of the interview data for coding and analysis, while reserving the remaining one-third for theoretical saturation testing. The interview transcripts were organized and managed using NVivo version 11 software. Analytical procedures were conducted following the three-stage coding method proposed by ref. [34].

In the first stage, the researchers conducted a line-by-line examination of the interview texts, guided by the research themes and objectives. Initial concepts were extracted through careful interpretation and analysis. During this process, content with similar meanings was integrated, and concepts lacking substantive significance were excluded. This stage resulted in the identification of 39 initial concepts. In the second stage, the researchers inductively merged conceptually similar initial concepts into 17 representative subcategories. In the third stage, these 17 subcategories were further synthesized into 7 main categories, based on the characteristic features of smart logistics. Finally, the previously reserved one-third of the original interview data was employed for theoretical saturation testing. Using the established three-stage coding procedure, this subset of data was re-analyzed to extract and compare concepts and categories, enabling cross-validation. The results indicated that no new core concepts or categories emerged from the additional data, nor was there any evidence of content that would challenge the existing category structure. This confirms that the developed category system has a stable structure and satisfies the criteria for theoretical saturation. Detailed coding results are presented in Appendix A.

Based on the above procedures, this study preliminarily extracted several key factors influencing rural residents’ adoption of smart logistics services. These include perceived usefulness, perceived ease of use, sensitivity to collective evaluation, cultural conservatism, infrastructure quality, facilitating conditions, and technology trust. However, the underlying mechanisms and causal pathways through which these factors operate in real-world usage scenarios remain insufficiently understood and warrant further systematic investigation.

4. Empirical Analysis

4.1. Research Hypotheses and Models

The TAM proposed by ref. [35] has been widely applied to explain and predict individuals’ acceptance and usage behaviors toward information technologies. In this model, perceived usefulness refers to the degree to which users believe that using a particular technology enhances their work or life efficiency, while perceived ease of use denotes the extent to which users believe that the technology can be used with minimal effort [35]. Both constructs are established as key predictors and have been found to exert significant positive influences on users’ behavioral intentions. Notably, perceived usefulness demonstrates a stronger influence on behavioral intention than perceived ease of use, while perceived ease of use is treated as an antecedent of perceived usefulness rather than a direct determinant of behavioral intention. Perceived usefulness thereby mediates the relationship between perceived ease of use and behavioral intention [36]. In the context of logistics technology adoption, this mechanism has been empirically validated. Ref. [37] found that the ease of use of IoT systems significantly enhanced logistics practitioners’ perception of their utility, thereby increasing their intention to use the technology. Ref. [38] pointed out that AI-powered last-mile delivery solutions effectively reduce operational complexity and cognitive burden, making them particularly suitable for rural users with low digital literacy and more likely to be accepted. Ref. [39] further emphasized that perceived usefulness not only significantly affects user attitudes and behavioral intentions but also consolidates its central role in technology adoption decisions. Moreover, ref. [37] confirmed that both perceived usefulness and perceived ease of use significantly influence users’ intention to adopt IoT-based business intelligence technologies. Similarly, ref. [38] highlighted that AI-driven last-mile delivery systems optimize routing and improve service reliability, thereby increasing usage intentions. Ref. [39] also found that ease of use enhances users’ recognition of the system’s actual value and indirectly influences adoption intention through perceived usefulness. Based on the above discussion, the following hypotheses are proposed:

H1. 

PEU (Perceived ease of use) positively affects PU (perceived usefulness).

H2. 

PU positively affects UI (usage intention).

H3. 

PEU positively affects UI.

H4. 

PU mediates the relationship between PEU and UI.

Ref. [40] synthesized several classic theoretical models (e.g., TAM, TRA, and the motivation model) to develop the UTAUT framework. In this model, social influence refers to the degree to which an individual perceives that important others (e.g., colleagues, friends, or supervisors) believe they should use a particular technology, while facilitating conditions denote the extent to which an individual believes that sufficient organizational and technical resources are available to support the use of the technology. Both constructs are considered key direct determinants of behavioral intention [40]. In rural contexts where access to technologies is often limited and digital infrastructure is underdeveloped, individuals tend to rely on community-based experience sharing and peer demonstrations to assess the feasibility and credibility of new technologies [24]. These authors further noted that community demonstration projects led by local governments or agricultural cooperatives can effectively enhance users’ sense of trust and belonging, thereby encouraging users to adopt the technology. Numerous empirical studies have confirmed the importance of social influence in shaping user adoption intentions. For instance, ref. [41] demonstrated that social influence significantly affects user intention in mobile commerce contexts. Similarly, ref. [42] found that social influence was a key determinant of behavioral intention in the context of electric vehicle purchases. To more accurately capture rural sociocultural characteristics, this study modifies the UTAUT model by replacing the construct of “social influence” with “sensitivity to collective evaluation”. This construct is defined as the extent to which individuals are influenced by the approval and opinions of peers, neighbors, and authority figures during technology adoption. It more accurately captures rural residents’ sensitivity to group opinions, neighborhood judgments, and conformity-driven behaviors. This substitution provides greater cultural relevance and explanatory power than the original “social influence” construct. Moreover, facilitating conditions, another key variable in the UTAUT model, have also been widely validated in rural technology adoption studies. The authors of [29], based on an empirical investigation of rural users in China, found that facilitating conditions significantly influenced the intention to use smart lockers. Similarly, ref. [43] confirmed the positive effect of facilitating conditions on the adoption of online freight platforms in rural Thailand. Based on the above, the following hypotheses are proposed:

H5. 

SCE (sensitivity to collective evaluation) positively affects UI.

H6. 

FC (facilitating conditions) positively affect UI.

In supply–demand relationships, trust has consistently been considered a critical factor [44]. Rural users’ confidence in the reliability, safety, and performance of technologies such as drones and autonomous delivery vehicles greatly shapes their perception and willingness to adopt these technologies. Trust encompasses both the technical reliability of the system and the integrity of the service provider. When users believe the system can operate continuously, protect their data, and meet delivery demands without failure, they are more likely to adopt and use it [45]. Moreover, trust can reduce perceived risks and operational concerns, thereby fostering a positive attitude toward adoption [46]. Numerous empirical studies have validated the critical role of technology trust adoption models. The authors of [47] extended the meta-UTAUT model by investigating users in the Gulf countries and found that both trust and facilitating conditions significantly influenced usage intention, with trust having the stronger effect. The authors of [48], integrating UTAUT2, the diffusion of innovation theory (DOI), and online trust models, found that initial trust and facilitating conditions significantly predicted usage intention toward mobile health apps in Indonesia, with trust being the most influential factor. In [49], the UTAUT2 framework highlighted that facilitating conditions not only directly influence initial trust but also indirectly impact usage intention through trust. Similarly, the authors of [50], in the context of online flight ticket purchasing, found that facilitating conditions affected intention both directly and indirectly through trust. Based on these findings, the following hypotheses are proposed:

H7. 

FC positively influences TT (technology trust).

H8. 

TT positively affects UI.

H9. 

TT mediates the relationship between FC and UI.

Cultural factors have received increasing attention in technology adoption research, especially in rural societies with deeply rooted traditional value systems [51]. As an emergent construct derived from the grounded theory coding in this study, cultural conservatism is defined as villagers’ tendency to resist, doubt, or reject smart logistics technologies due to their alignment with social traditions, lifestyle norms, and interpersonal practices. From a technology acceptance perspective, cultural conservatism represents a deep-rooted sociopsychological barrier, akin to the “tradition barrier” described in the innovation resistance theory [52]. This resistance does not arise from technological flaws, per se, but, rather, from a clash between the modern characteristics of the technology and traditional life patterns [53]. In rural settings characterized by high-context interpersonal relationships and familiarity-based trust, this resistance becomes more pronounced. Empirical research supports the critical role of cultural factors in adoption behavior. For example, ref. [54] found that uncertainty avoidance in culture significantly reduced individuals’ intention to use wireless web communication technologies in the early stages. Similarly, ref. [55] reported that cultural biases had a more substantial effect than individual biases in diminishing users’ intention to adopt intelligent services, primarily due to identity and value conflicts. In the context of rural China, cultural conservatism not only manifests as value-oriented traditionalism but also involves vigilance and resistance toward the impersonal nature of smart logistics systems. This undermines villagers’ intention to adopt such technologies. Hence, the following hypothesis is proposed:

H10. 

CC (cultural conservatism) negatively affects UI.

In this study, infrastructure quality refers to factors such as road conditions, warehousing facilities, charging station distribution, and digital connectivity, all of which critically determine the operational efficiency, reliability, and scalability of smart delivery systems [56]. Existing studies indicate that limitations in infrastructure, such as safety risks, low loading/unloading capacity, and high construction costs, impede the effective functioning of autonomous delivery technologies [38]. These issues are particularly pronounced in rural areas due to inadequate logistics resources, thus weakening system resilience and constraining service coverage [24]. Although direct empirical evidence on the impact of infrastructure quality on smart delivery adoption is still limited, research in related domains offers valuable insights. For instance, ref. [57] found that system infrastructure quality significantly influences users’ intention to use open government data platforms. Ref. [58] also reported that the quality of technical infrastructure positively affects healthcare workers’ intention to use eHealth systems. Based on the above, the following hypothesis is proposed:

H11. 

IQ (infrastructure quality) positively affects UI.

Single-model frameworks often fall short of comprehensively explaining user adoption behavior. Integrating key constructs from multiple models enhances theoretical robustness and predictive power [59]. This study combines core variables from TAM and UTAUT, and incorporates innovative constructs derived from the qualitative phase, tailored to the rural Chinese context, to develop an integrated and contextually adaptive research model (Figure 1). A substantial body of research has demonstrated the applicability and explanatory power of TAM, UTAUT, and their extended models in rural contexts [27,29]. Moreover, prior empirical studies have provided both theoretical and practical support for the integration of TAM and UTAUT [60]. Specifically, TAM emphasizes the rational evaluation of functional value through perceived ease of use and usefulness [37], while UTAUT underscores the institutional and environmental enablers such as facilitating conditions [24]. Recognizing that the construct of “facilitating conditions” may insufficiently reflect the multidimensional support mechanisms in rural contexts, this study, following ref. [61] and ref. [62], decomposes it into two second-order dimensions: PS (policy support) and EI (enterprise incentives), to comprehensively capture the influence of both institutional safeguards and market-driven forces. Meanwhile, trust—frequently included in recent UTAUT extensions—has demonstrated substantial predictive power [48,49,50], supporting its role as a supplementary variable. Additionally, this study introduces three new context-specific constructs derived from grounded theory: sensitivity to collective evaluation, cultural conservatism, and infrastructure quality, to provide novel theoretical perspectives on the psychological mechanisms and social foundations shaping smart logistics adoption in rural China.

4.2. Research Design

4.2.1. Questionnaire Design

The measurement items were adapted from validated scales in the literature. To ensure the accuracy and reliability of the adapted items, two experts in the fields of technology acceptance theory and logistics were invited to review and revise the questionnaire. The final version of the scale includes 9 constructs and 35 items (Appendix A). The experts unanimously agreed that the revised scale demonstrated a high level of structural integrity and item quality.

4.2.2. Participants and Data Collection

The questionnaire employed in this study consists of three main sections: (1) an introductory section that briefly outlines the purpose of the research and assures respondents of the confidentiality of their data and the voluntary nature of their participation; (2) demographic questions, which collect basic information including gender, age, educational attainment, income level, and occupation; and (3) measurement items related to the constructs proposed in this study. All constructs were assessed using a five-point Likert scale, ranging from 1 (strongly disagree) to 5 (strongly agree).

The target participants of this study were rural residents aged 18 and above in Hebei Province, China. The choice of this sample was based on the following considerations: On one hand, existing studies on rural logistics often focus on macro-level trends [27,29,39], with limited investigation into the specific characteristics and current development of rural logistics in typical regions. On the other hand, under the ongoing Beijing–Tianjin–Hebei integration strategy, urban centers like Beijing and Tianjin have already developed mature logistics infrastructure and networks, whereas Hebei, as an inland province, faces challenges such as spatial fragmentation among prefecture-level cities and imbalanced supply–demand structures for agricultural products. These factors have contributed to the imbalanced progression of rural logistics across different areas of the region [63]. Therefore, focusing on rural residents in Hebei enables an in-depth understanding of the real challenges faced during the integration process and helps identify viable ways to improve rural logistics efficiency and promote the flow of goods and services between urban and rural areas.

Data collection for this study was conducted from 5 March to 30 March 2025, using a mixed-method approach that combined both online and offline surveys. The minimum required sample size was calculated using G*Power (version 3.1.9.7) statistical software following the guidelines proposed by ref. [64], with parameters set at f² = 0.15, α = 0.05, and power = 0.95. The calculation indicated that a minimum of 153 responses was required. Online recruitment was carried out by distributing survey invitations through WeChat groups and the Xiaohongshu platform (https://www.xiaohongshu.com, accessed on 5 March 2025). Offline data collection was conducted by the research team in two towns in Hebei Province where smart logistics systems have been implemented: Xigezhen in Tangshan City and Wolongzhen in Pingquan City. With the support of local village committees and volunteers, questionnaires were distributed to ensure broad participation, including residents with limited internet access. A total of 542 questionnaires were collected (282 online and 260 offline). After rigorous screening for completeness and response consistency, 451 valid responses were retained for analysis, meeting the minimum sample size recommended by ref. [64].

4.2.3. Data Processing and Analytical Methods

Data analysis was conducted using SPSS 27.0 and AMOS 23.0. SPSS 27.0 was used for descriptive statistics of demographic characteristics, reliability analysis (Cronbach’s α), assessment of common method bias (CMB), and mediation analysis (via PROCESS macro). AMOS 23.0 was employed to conduct confirmatory factor analysis (CFA) and to construct and estimate the SEM. This approach ensured construct validity and enabled robust testing of the hypothesized relationships. Moreover, prior studies have indicated that individual characteristics such as gender, age, and educational attainment may significantly influence technology adoption behavior [65]. Therefore, these variables were included as control variables in the model to improve the accuracy of estimations and to isolate the net effects of the main independent variables on the dependent variable.

4.3. Analysis Result

4.3.1. Demographic Data and CMB Test

The demographic data of the respondents are shown in

Table 2. Demographic data.

Section	Item	Frequency	%
Gender	Male	229	50.8
	Female	222	49.2
Age	18–30	81	18.0
	31–40	102	22.6
	41–50	98	21.7
	51–60	93	20.6
	Above 60	77	17.1
Educational level	Middle school or below	119	26.4
	High school	172	38.1
	Vocational education	93	20.6
	Bachelor or above	67	14.9
Average monthly income	2000 RMB or below	91	20.2
	2001–4000 RMB	133	29.5
	4001–6000 RMB	146	32.4
	Above 6000 RMB	81	18.0
Occupation	Farmer	113	25.1
	Individual business owners	103	22.8
	Government/enterprise staff	63	14.0
	Student	71	15.7
	Unemployment/retirement	88	19.5
	Other	13	2.9

Source: authors.

. In terms of gender distribution, male respondents had the highest number, totaling 229 people (50.8%). In terms of age, the majority of respondents were aged 31–40, with a total of 102 people (22.6%). In terms of education level, high school has the highest proportion, with a total of 172 people (36.8%). In terms of average monthly income, the highest number of respondents were between CNY 4001 and CNY 6000, with a total of 146 people (32.4%). In terms of occupation, the highest number was farmers, with a total of 113 people, accounting for 25.1%.

Given that the study relied entirely on self-reported data, there is a potential risk of common method bias (CMB) [66]. To assess this issue, Harman’s single-factor test was conducted [67]. The unrotated exploratory factor analysis extracted nine factors with eigenvalues greater than 1, accounting for a cumulative variance of 80.307%. The first factor explained only 23.668% of the variance, which is well below the critical threshold of 40%. According to the criterion proposed by ref. [68], this suggests that common method bias is not a serious concern in this study.

4.3.2. Reliability and Validity Analysis

To assess the internal consistency of the measurement structure, Cronbach’s alpha values were calculated for each construct. The results (

Table 3. Reliability and validity results.

Factor	Variable	Mean	Standard Deviation	Factor Loadings	α	AVE	CR
PEU	PEU1	3.59	1.046	0.917	0.913	0.731	0.915
	PEU2	3.54	1.039	0.935
	PEU3	3.59	1.063	0.747
	PEU4	3.62	1.023	0.806
PU	PU1	3.40	0.869	0.854	0.893	0.683	0.895
	PU2	3.39	0.818	0.844
	PU3	3.35	0.814	0.706
	PU4	3.40	0.825	0.890
SCE	SCE1	3.50	1.018	0.904	0.903	0.764	0.906
	SCE2	3.47	1.007	0.789
	SCE3	3.57	0.980	0.923
CC	CC1	3.26	1.082	0.901	0.875	0.708	0.878
	CC2	3.23	1.054	0.872
	CC3	3.24	1.064	0.743
IQ	IQ1	3.61	1.077	0.818	0.852	0.663	0.854
	IQ2	3.61	1.107	0.736
	IQ3	3.54	1.117	0.882
PS	PS1	3.35	0.978	0.806	0.905	0.768	0.908
	PS2	3.31	0.912	0.935
	PS3	3.30	0.913	0.883
EI	EI1	3.30	0.967	0.841	0.840	0.642	0.843
	EI2	3.39	1.008	0.735
	EI3	3.33	0.985	0.823
TT	TT1	3.21	0.825	0.886	0.870	0.696	0.873
	TT2	3.19	0.828	0.851
	TT3	3.23	0.815	0.761
UI	UI1	3.33	0.810	0.829	0.912	0.781	0.914
	UI2	3.31	0.797	0.905
	UI3	3.34	0.847	0.914

Source: authors.

) indicate excellent internal reliability, with all values significantly exceeding the commonly accepted threshold of 0.70 [69]. To evaluate the validity of the measurement model, CFA was conducted on all latent constructs. As shown in

Table 4. Discriminant validity results.

Section	UI	TT	EI	PS	IQ	CC	SCE	PU	PEU
UI	0.884
TT	0.518	0.834
EI	0.295	0.177	0.801
PS	0.285	0.123	0.469	0.876
IQ	0.122	0.054	−0.007	0.013	0.814
CC	−0.414	−0.285	−0.176	−0.087	−0.003	0.841
SCE	0.275	0.130	−0.021	0.040	0.045	−0.108	0.874
PU	0.428	0.278	0.123	0.132	0.212	−0.212	0.080	0.826
PEU	0.493	0.304	0.082	0.059	0.152	−0.185	0.178	0.290	0.855

Source: authors.

, all standardized factor loadings exceeded the recommended threshold of 0.70, indicating high indicator reliability. The composite reliability (CR) values ranged from 0.780 to 0.939, surpassing the minimum acceptable value of 0.70. The average variance extracted (AVE) values for all constructs were above the 0.50 threshold, demonstrating adequate convergent validity [70].

Discriminant validity was assessed using the ref. [70] criterion. As shown in Table 4, the square root of the AVE for each construct was greater than its correlations with other constructs, indicating sufficient discriminant validity among all latent variables.

According to the evaluation criteria adopted by ref. [71], the goodness of fit (

Table 5. Goodness of fit.

Goodness of Fit	Standard Value	CFA Model	SEM Model
χ2/DF	<3	1.082	1.843
RMSEA	0.05	0.014	0.043
CFI	>0.9	0.997	0.952
TLI	>0.9	0.996	0.948

Source: authors.

) further indicates that the measurement model has achieved a good fit with the data.

4.3.3. SEM and Mediation Analysis

The structural model was examined using the maximum likelihood estimation (MLE) method. The fit indices (Table 5) indicate a good model fit [71]. The factor loadings of the second-order latent variable are detailed in

Table 6. Factor loadings of the second-order latent variable.

Second-Order Latent	First-Order Latent	p	Factor Loading
FC (Favorable condition)	PS (policy support)→FC (policy support)	***	0.680
	EI (enterprise incentives)→FC (policy support)	***	0.690

Source: authors. ***, p < 0.001.

.

The significance test results of the SEM paths (

Table 7. Hypothesis test results.

Hypothesis	Path	B	b	C.R.	p	Result
H1	PEU→PU	0.261	0.289	5.699	***	Supported
H2	PU→BI	0.167	0.188	4.116	***	Supported
H3	PEU→BI	0.254	0.318	6.975	***	Supported
H5	SCE→BI	0.114	0.159	3.826	***	Supported
H6	FC→BI	0.330	0.281	4.255	***	Supported
H7	FC→TT	0.286	0.217	3.193	0.001	Supported
H8	TT→BI	0.262	0.294	6.082	***	Supported
H10	CC→BI	−0.201	−0.241	−5.382	***	Supported
H11	QI→BI	0.018	0.027	0.624	0.533	Rejected
H1	PEU→PU	0.261	0.289	5.699	***	Supported

Note. B: unstandardized coefficient, b: standardized coefficient. C.R.: critical ratio, *** p < 0.001.

and Figure 2) are as follows: perceived ease of use had a significant positive effect on perceived usefulness (β = 0.289, p < 0.001), supporting Hypothesis H1. Perceived usefulness had a significant positive effect on usage intention (β = 0.188, p < 0.001), supporting H2. Perceived ease of use also significantly positively influenced usage intention (β = 0.318, p < 0.001), supporting H3. Sensitivity to collective evaluation significantly positively affected usage intention (β = 0.159, p < 0.001), supporting H5. Facilitating conditions had a significant positive effect on BI (β = 0.281, p < 0.001), supporting H6, and also significantly affected technology trust (β = 0.217, p = 0.001), supporting H7. Technology trust had a significant positive impact on usage intention (β = 0.294, p < 0.001), supporting H8. Conversely, cultural conservatism had a significant negative effect on usage intention (β = −0.294, p < 0.001), supporting H10. The effect of infrastructure quality on usage intention was not significant (β = 0.027, p = 0.533); thus, Hypothesis H11 was not supported. In terms of standardized path coefficients, the strength of influence on usage intention among significant antecedents is ranked as follows: PEU (perceived ease of use) > TT (technology trust) > FC (facilitating conditions) > CC (cultural conservatism) > PU (perceived usefulness) > SCE (sensitivity to collective evaluation). Additionally, the results of control variable analysis indicated that age had a significant negative effect on usage intention (β = −0.137, p < 0.001), whereas other control variables did not show significant effects.

To test the mediation effects, this study employed the PROCESS macro in SPSS 27.0 and used the bootstrap method for estimation [72]. Specifically, a resampling procedure of 5000 iterations was conducted with a 95% confidence interval. A mediation effect is considered statistically significant if the confidence interval does not contain zero. As shown in

Table 8. Mediation effects analysis results.

Path	Effect Type	Effect	S.E.	95% CI		Type of Mediation
				LLCI	ULCI
H4: PEU→PU→UI	Total Effect	0.3659	0.058	0.264	0.489	Partial mediation
	Indirect Effect	0.0563	0.0147	0.0280	0.0855
	Direct Effect	0.3096	0.0548	0.2084	0.4215
H9: FC→TT→UI	Total Effect	0.3036	0.073	0.187	0.472	Partial mediation
	Indirect Effect	0.0692	0.0331	0.0067	0.1383
	Direct Effect	0.2344	0.0644	0.1098	0.3666

Note. LLCI: lower-limit confidence interval, ULCI: upper-limit confidence interval. S.E.: standard error.

, the bootstrap confidence intervals for the indirect effects of perceived ease of use and facilitating conditions on usage intention did not include zero, indicating significant mediation effects. Therefore, both hypotheses H4 and H9 are supported. Moreover, the results suggest that these mediation effects are partial.

5. fsQCA Analysis

Although SEM reveals unidirectional causal paths between variables, it primarily examines the net effect of individual variables on the outcome, making it difficult to uncover potential synergistic effects or complex causal structures among variables [73]. Therefore, this study further incorporates fsQCA to explore multiple causal configurations that drive rural residents’ behavioral intentions, thereby enriching the explanatory power and theoretical depth of the research [74].

5.1. Data Calibration

Data calibration is a fundamental and critical preprocessing step before implementing fsQCA [75]. In this study, perceived ease of use, perceived usefulness, sensitivity to collective evaluation, cultural conservatism, infrastructure quality, facilitating conditions, and technology trust were set as condition variables, with usage intention as the outcome variable for configuration analysis. First, the variable scores were calculated based on the mean values of the measurement items for each dimension [75]. Following ref. [76] recommendations, the 95th percentile, 50th percentile (crossover point), and 5th percentile of each dimension’s scores were set as thresholds for full membership, crossover, and full non-membership, respectively, serving as calibration anchors. Subsequently, direct calibration was conducted for each variable using the Calibrate function in fsQCA 3.0 software [76], which transformed the original variable scores into fuzzy-set membership scores ranging from 0 to 1. A membership score closer to 1 indicates a higher degree of set membership, whereas a score closer to 0 indicates a lower degree of membership. During the calibration process, it was observed that some variables exhibited a concentration of cases with a membership score of 0.05 near the full non-membership threshold, which may lead to potential misclassification of borderline cases. To address this issue and enhance the robustness and precision of the calibration, a minor adjustment of +0.001 was applied to all membership scores below 1, following the approach proposed by ref. [77]. To clearly distinguish between raw variables and calibrated fuzzy-set variables, the calibrated variables are labeled by appending the letter “F” to the original variable abbreviations (e.g., FPEU for the calibrated fuzzy set of perceived ease of use). Detailed calibration results are presented in

Table 9. Data calibration and descriptive statistics.

Before	Fuzzy-Set Calibration			After	Descriptive Statistics
	Full	Mid	Non		Mean	Standard Error
PEU	5.000	3.750	2.000	FPEU	0.486	0.308
PU	4.500	3.250	2.000	FPU	0.556	0.301
SCE	5.000	3.667	2.000	FSCE	0.474	0.303
CC	5.000	3.333	1.667	FCC	0.477	0.294
IQ	5.000	3.667	1.833	FIQ	0.522	0.297
FC	4.500	3.333	2.000	FFC	0.511	0.283
TT	4.333	3.000	2.000	FTT	0.566	0.291

.

5.2. Necessity Analysis

The purpose of necessity analysis is to determine whether any of the condition variables constitute necessary conditions for the presence of fuzzy usage intention. According to ref. [76] criteria, a condition is considered necessary for the outcome if its consistency score exceeds 0.90. As shown in

Table 10. Necessity analysis results.

Condition	Consistency	Coverage
FPEU	0.747	0.893
~FPEU	0.580	0.658
FPU	0.770	0.807
~FPU	0.530	0.695
FSCE	0.696	0.854
~FSCE	0.612	0.678
FCC	0.593	0.724
~FCC	0.741	0.825
FIQ	0.698	0.779
~FIQ	0.605	0.737
FFC	0.750	0.855
~FFC	0.610	0.726
FTT	0.830	0.854
~FTT	0.544	0.730

, none of the conditions meet the 0.90 consistency threshold. Therefore, it can be concluded that no single condition qualifies as a necessary prerequisite for fuzzy usage intention in this study.

5.3. Sufficiency Analysis

Prior to conducting sufficiency analysis, it is necessary to generate a truth table with 2^k rows using the fuzzy-set algorithm in fsQCA. The construction of the truth table requires the specification of two key thresholds: the frequency threshold and the raw consistency threshold [76]. Given that the sample size in this study exceeds 450 cases, the frequency threshold was set at 3 to ensure that each configuration is supported by a sufficient number of cases [78]. The raw consistency threshold was set at 0.95, which serves as the minimum acceptable level for a causal condition combination to be considered sufficient for the outcome [79]. In the sufficiency analysis, fsQCA produces three types of solutions: the complex solution, the intermediate solution, and the parsimonious solution. The complex solution, which retains all possible logical remainders, lacks substantive explanatory value. In contrast, the intermediate and parsimonious solutions enable differentiation between core and peripheral conditions [78]. Considering that the intermediate solution offers a better balance between theoretical relevance and empirical accuracy, this study adopts the intermediate solution as the basis for final interpretation [76].

Table 11. Configurations of conditions leading to usage intention.

Condition	Solution
	S1	S2	S3	S4	S5	S6	S7	S8
FPEU	●			●		★
FPU					●	●	●
FSCE					●	●		●
FCC		○	○	★
FQI		●		★		★	●	●
FFC			●				●	★
FTT	★	●	★		★		★	●
Consistency	0.952	0.950	0.956	0.956	0.946	0.972	0.970	0.972
Raw coverage	0.645	0.481	0.515	0.382	0.501	0.407	0.437	0.398
Unique coverage	0.049	0.021	0.019	0.015	0.016	0.014	0.008	0.002
Solution consistency	0.916
Solution coverage	0.846

Note. ● indicates the presence of a core condition, ○ represents the absence of a core condition, ★ signifies the presence of a peripheral condition, and a blank space suggests that the condition is optional or irrelevant.

presents eight causal paths leading to high usage intention. The overall solution consistency is 0.846, and the coverage is 0.916, indicating that these eight paths have strong explanatory power and high consistency in explaining the formation of high usage intention [80]. Among them, the raw coverage of path S1 is the highest (0.645), explaining the most cases, and its consistency reaches 0.952, making it the optimal explanatory path. This path highlights that high perceived ease of use and high technology trust are the key foundations for achieving high usage intention. Paths S2, S3, and S5 also cover nearly half of the cases (coverage rates of 0.481, 0.515, and 0.501, respectively), with high consistency (0.950, 0.956, and 0.946, respectively). In particular, S2 and S3 underscore the importance of cultural conservatism as a reverse condition, suggesting that even with low cultural conservatism, high infrastructure quality and technology trust, or high facilitating conditions and technology trust, it can still foster strong usage intention. Path S5 indicates that the combined effect of high perceived usefulness, high sensitivity to collective evaluation, and high technology trust is crucial for achieving high usage intention. Paths S6 to S8 demonstrate extremely high consistency (0.970–0.972), with coverage rates ranging from 0.398 to 0.437, showing the stability of the explanatory results. In these three paths, perceived usefulness, sensitivity to collective evaluation, infrastructure quality, facilitating conditions, and technology trust repeatedly emerge as core conditions, emphasizing their critical role in promoting the formation of high usage intention. Notably, while the coverage of path S4 is relatively low (0.382), its consistency remains high (0.956), suggesting that despite applying to fewer cases, this path remains reliable in explaining these specific situations.

6. Conclusions

6.1. Discussion

This study adopted a mixed-methods approach to explore the factors influencing rural residents’ adoption of smart “last-mile” delivery services. Grounded theory analysis identified seven key perceived factors: perceived usefulness, perceived ease of use, sensitivity to group evaluation, cultural conservatism, infrastructure quality, enabling conditions, and trust in technology. SEM and mediation effect analysis revealed that perceived ease of use significantly influences both perceived usefulness and usage intention, with perceived usefulness partially mediating this relationship, supporting findings from the literature [37,39]. Unlike the classical TAM model [35], ease of use has a stronger influence in the rural context, highlighting its critical role in bridging the digital divide. Enabling conditions not only directly influence usage intention but also operate through trust in technology. Policy support and commercial incentives further enhance both perceived usefulness and trust, contributing to the sustained adoption of smart logistics in rural areas. Moreover, the study identifies three rural-specific factors—sensitivity to group evaluation, cultural conservatism, and infrastructure quality—thereby extending the contextual scope of technology adoption research. SEM results show that sensitivity to group evaluation positively affects behavioral intention. Although this differs from traditional notions of social influence, it aligns with the sociopsychological mechanisms under collectivist cultures [41,42]. Cultural conservatism negatively influences intention, consistent with prior studies [54,55]. The effect of infrastructure quality was not significant, diverging from earlier findings [57,58], possibly because rural users prioritize personal relevance over general infrastructure quality. These results address RQ1.

RQ2 and 3 are also addressed. fsQCA identified eight sufficient configurations leading to high usage intention, validating the SEM results. Key variables—perceived ease of use, perceived usefulness, sensitivity to group evaluation, facilitating conditions, and trust in technology—recur across multiple configurations, indicating their importance. However, differences between methods exist: fsQCA reveals that in configurations S2 and S3, even with low cultural conservatism, strong infrastructure or facilitating conditions combined with trust in technology can still trigger usage intention. This contrasts with the linear conclusion from SEM that cultural conservatism negatively predicts intention. Moreover, although infrastructure quality is not significant in SEM, fsQCA identifies it as a key factor in five configurations, suggesting that its interactive role in complex causal pathways exceeds the analytical scope of SEM’s linear framework [73].

6.2. Theoretical Implications

The theoretical contributions of this study are as follows: First, by focusing on rural China (Hebei Province) [16,17], this research challenges the urban-centric bias prevalent in current studies. It reveals how inadequate infrastructure and cultural inertia act as barriers to technology adoption [11,18], identifying both psychological and structural constraints faced by rural users. This contributes to the development of integrated smart logistics theories that encompass both urban and rural contexts. Second, building on the TAM and UTAUT models, the study incorporates rural-specific variables such as sensitivity to collective evaluation, cultural conservatism, and infrastructure quality. It constructs a multidimensional model that includes cognitive, social, cultural, and institutional factors, thereby enhancing explanatory power and cross-context applicability. Finally, by employing the fsQCA method, this study uncovers the complexity, non-linearity, and multipath characteristics of rural logistics technology adoption behavior. Compared to traditional SEM approaches, fsQCA identifies multiple sufficient configurations leading to high adoption intentions [73], offering a methodological innovation for future research in logistics technology adoption.

6.3. Managerial Implications

Promoting rural consumption and establishing a three-tier logistics system are key measures for achieving rural revitalization [81]. This study uncovers the complex mechanisms behind rural residents’ adoption of smart delivery technologies, offering support for the optimization of initiatives such as “Express Delivery to Villages” [82,83]. It is recommended that local governments strengthen village-level logistics infrastructure, set up demonstration sites and provide training in high-sensitivity areas, and enhance publicity and guidance in culturally conservative communities. Logistics enterprises and e-commerce platforms should highlight the value of smart logistics, optimize user interfaces, and design incentive mechanisms that integrate economic and emotional factors to boost user willingness. At the national level, efforts should focus on setting standards, ensuring data security, and increasing transparency to build technological trust, while also fostering institutional and social trust in rural areas [81]. Given the heterogeneity of adoption paths, policies should be tailored to local conditions to avoid a “one-size-fits-all” approach, thereby enhancing adaptability and effectiveness.

6.4. Limitations and Future Study

This study has several limitations. First, it lacks behavioral data and longitudinal observations, relying solely on surveys and interviews to capture participants’ intentions, which makes it difficult to reflect the dynamic process of technology adoption. Future research could incorporate behavioral data or adopt a longitudinal design. Second, some variables in the model may have bidirectional causal relationships, leading to potential endogeneity issues. It is recommended that future studies use instrumental variables or other methods to improve the reliability of the estimates. Third, the sample is mainly drawn from Hebei Province, which limits the geographic scope and affects the generalizability of the findings. Expanding the geographic and demographic coverage would be valuable in future research. Finally, although participants were recruited through both online and offline channels, Xiaohongshu users tend to be younger and more digitally literate, which may underestimate the barriers faced by older or less tech-savvy individuals. Future studies should diversify sampling sources to enhance the representativeness and external validity of the findings.