Leveraging Fuzzy Set Qualitative Comparative Analysis to Explore Determinants of Intention to Use Self-Driving Vehicles in Ghana

Abstract

The transformative potential of self-driving vehicles (SDVs) in enhancing mobility and transportation safety is well documented, yet their adoption in developing countries remains understudied. While existing research has primarily focused on SDV adoption in developed nations using variance-based methods, limited attention has been given to understanding how multiple factors interact to influence adoption decisions in developing economies. This study addresses this gap by examining the determinants of SDV adoption intention in Ghana using fuzzy set qualitative comparative analysis (fsQCA). Drawing on the Technology Acceptance Model and incorporating additional constructs of perceived reliability, technological competence, and perceived risk, the study analyzed survey data from 1248 respondents across Ghana’s 16 regions. The findings reveal multiple pathways to high adoption intention, with the most effective combination being perceived reliability, perceived ease of use, and technological competence working together. For low adoption intention, two main configurations emerged, both highlighting how the combination of low technological competence and high perceived risk significantly hinders adoption. These findings provide valuable insights for policymakers and stakeholders in developing economies, emphasizing the need for targeted interventions that address both technological and socio-cultural factors influencing SDV adoption.

1. Introduction

The autonomous vehicles (AVs), which can sense their environment and operate with minimal or no human intervention [1], represent a potentially transformative shift in transportation systems. These vehicles, also known as driverless, self-driving, or robotic vehicles, offer numerous potential benefits, including enhanced mobility and accessibility [2,3], integration with shared mobility platforms, and the promotion of more sustainable transportation practices. By reducing human error, a primary cause of road accidents, AVs have the potential to improve traffic safety and efficiency significantly [4].

Recent years have witnessed remarkable advancements in autonomous driving technologies, with significant progress in sensing capabilities, decision-making algorithms, and control systems. The rapid development of deep learning approaches, particularly reinforcement learning frameworks, has enabled vehicles to navigate complex traffic scenarios with increasing reliability and safety. For instance, ref. [5] has demonstrated how integrated deep reinforcement learning frameworks can substantially improve high-speed cruising performance of autonomous vehicles, representing a critical advancement toward practical deployment in everyday transportation systems. Meanwhile, the global adoption of AVs, especially in developing countries, has progressed more slowly than expected [2].

A key challenge in this adoption is the significant infrastructural and regulatory disparities between developed and developing nations [6,7]. Developed countries typically have well-defined traffic laws and controlled road usage, whereas developing countries often face complex road environments characterized by mixed traffic patterns, informal economic activities, and limited infrastructure [8]. This divergence creates unique challenges for deploying AVs, necessitating careful consideration of context-specific factors that influence user acceptance and adoption.

Although the existing literature has primarily focused on AV adoption in developed contexts with robust infrastructure and regulation [9,10], the specific dynamics of adoption in developing economies remain largely unexplored. Furthermore, the dominant methodological approaches in this area have primarily relied on variance-based methods, such as regression analysis [11,12] and structural equation modeling (SEM) [13,14]. These approaches focus on the net effects of individual variables and often assume linear relationships, which may not adequately capture the complex interplay of factors that influence adoption decisions. Thus, less attention has been dedicated to investigating the pathways that influence high and low adoption of SDV. Again, while the Technology Acceptance Model (TAM) has been extensively validated in various technological contexts [15,16], its application to autonomous vehicle adoption in developing economies remains largely unexplored. Previous studies have primarily focused on TAM’s core constructs of perceived usefulness and perceived ease of use in developed nations [9,10], overlooking how these factors interact with context-specific challenges in emerging markets. Furthermore, the traditional application of TAM through variance-based methods may not fully capture the complex interplay of acceptance factors in environments with limited technological infrastructure and diverse socio-economic conditions [11,12]. This gap is particularly significant as developing nations present unique challenges for autonomous vehicle adoption, including mixed traffic patterns, informal economic activities, and limited infrastructure.

This research aims to address these critical gaps by examining the determinants of the intention to use AVs in Ghana, a rapidly urbanizing West African nation, using fuzzy set qualitative comparative analysis (fsQCA). While this study focuses specifically on Ghana, it offers valuable insights that may be representative of the broader sub-Saharan African context for several compelling reasons. Ghana exemplifies key characteristics shared across many developing economies in the region, including rapidly growing urbanization, mixed traffic patterns with formal and informal transportation systems, and varying levels of infrastructure development across urban and rural settings. As one of West Africa’s most stable democracies with a growing middle class and increasing technological adoption, Ghana represents an ideal testbed for understanding SDV acceptance in emerging markets that are experiencing similar socioeconomic transitions. Furthermore, Ghana’s diverse regional representation in this study (data from all 16 administrative regions) captures the heterogeneity typical across sub-Saharan Africa, from densely populated urban centers to remote rural communities.

fsQCA is a configurational research method particularly suited for exploring complex causal relationships and equifinality (i.e., multiple pathways leading to the same outcome) in emerging markets [17,18]. Unlike traditional variance-based approaches, fsQCA enables the identification of distinct configurations of conditions that lead to adoption, providing a more nuanced understanding of how various factors interact and combine to influence individual decisions. This approach is especially relevant for AV adoption in developing economies, where diverse socio-economic, experiential, and technological conditions may interact in complex, non-linear ways. The study precisely aims to address the following questions:

• What are the key conditions that are essential for determining high or low intention to use self-driving vehicles?
• What combinations of factors effectively explain high or not-high intentions to use self-driving vehicles?
• What is the most effective pathway to consider or avoid when explaining high and low intentions to use self-driving vehicles?

This study addresses a largely overlooked area by focusing on the adoption of autonomous vehicles (AVs) in Sub-Saharan Africa. The findings offer practical recommendations for governments, technology developers, and transportation stakeholders seeking to encourage the adoption of self-driving vehicles (SDVs) in similar emerging market contexts. Additionally, by using fuzzy set qualitative comparative analysis (fsQCA), this study goes beyond traditional quantitative methods to gain a deeper understanding of the interactions between various factors and their collective impact on technology adoption. This approach captures the complexity of real-world decision-making, especially in situations where multiple influences may affect user choices simultaneously.

We contribute to the existing literature in three ways: First, this study extends the Technology Acceptance Model (TAM) by integrating context-specific constructs: perceived reliability, technological competence, and perceived risk, offering a tailored framework for understanding self-driving vehicle (SDV) adoption in developing economies. Second, we employ fsQCA, a configurational methodology that reveals multiple, equally effective combinations of factors leading to both high and low intention to adopt SDVs, thereby moving beyond the linear assumptions of traditional variance-based models. Third, by focusing on Ghana, a representative sub-Saharan African context, and collecting data from all 16 administrative regions, this study captures regional heterogeneity and provides practical insights into the sociotechnical dynamics influencing SDV adoption in emerging markets. These contributions collectively advance theoretical development, methodological innovation, and policy relevance in the field of autonomous vehicle adoption.

2. Theoretical Background and Literature Review

The Technology Acceptance Model (TAM), developed by Davis in 1989, is one of the most influential frameworks for understanding and predicting user acceptance of technology. Its simplicity, adaptability, and focus on key constructs have established its importance in technology adoption research [15,19]. At its core, TAM posits that user acceptance is driven by two primary constructs: perceived usefulness (PU) and perceived ease of use (PEOU). PU refers to the extent to which a user believes that technology will enhance their performance, whereas PEOU reflects the degree to which the user believes that using the technology will require minimal effort. These constructs influence users’ attitudes toward technology, which, in turn, shape their behavioral intention (BI) to use the system and ultimately determine their actual usage.

The model’s parsimony and predictive capability have led to its validation and application across various fields, including education, healthcare, and financial systems [20,21]. A significant strength of TAM lies in its ability to provide actionable insights for both researchers and practitioners. The model clearly delineates the relationships between its constructs, allowing organizations to identify potential barriers to adoption, such as users’ perceptions of complexity or limited benefits. PU consistently emerges as a strong predictor of technology adoption, indicating that users are more inclined to adopt systems they perceive as beneficial [22,23]. For example, research in healthcare settings has shown that healthcare professionals are more likely to adopt electronic health records if they perceive these systems as enhancing patient care efficiency and effectiveness [24,25].

Similarly, PEOU is critical in shaping users’ intentions, as systems that are intuitive and user-friendly lead to higher acceptance rates. The relationship between PEOU and PU is particularly significant, as users often evaluate the usefulness of a technology based on how effortless it is to operate [26,27]. These insights highlight the importance of prioritizing usability and intuitive design during the development and implementation of new systems. TAM’s emphasis on behavioral intention (BI) further reinforces its utility as a predictive model. BI acts as a mediator between users’ attitudes and their actual system use, making it a crucial construct for understanding the transition from intention to behavior [28,29]. Empirical studies consistently validate BI as a strong predictor of actual usage across various domains, such as mobile payments, e-learning platforms, and healthcare applications [30]. Actual system use serves as the ultimate measure of TAM’s effectiveness. While PU, PEOU, and BI play essential roles in shaping intentions, other external factors such as organizational support, user training, and technological infrastructure also significantly influence actual usage. For instance, in the context of e-learning, it has been found that technical support and resource availability significantly impact students’ use of online platforms [31,32].

The Technology Acceptance Model is highly relevant for this study on the intention to use self-driving vehicles (SDVs). It provides a robust theoretical framework for understanding user acceptance of new technologies. TAM’s core constructs, perceived usefulness and perceived ease of use, align closely with the key factors influencing SDV adoption, such as users’ perceptions of the technology’s benefits (e.g., safety, efficiency) and usability. Additionally, TAM’s emphasis on behavioral intention as a precursor to actual usage offers valuable insights into the psychological and attitudinal factors driving SDV adoption. The model’s adaptability to diverse contexts and its demonstrated success in predicting technology acceptance across various domains, including transportation and advanced technology systems, make it particularly relevant for exploring SDV adoption in Ghana, where socio-demographic and infrastructural factors uniquely shape user perceptions. By leveraging TAM, this study can systematically assess the interplay of these factors and provide actionable insights for promoting public acceptance of SDVs.

2.1. Perceived Reliability

Perceived reliability represents the belief that an AV can consistently perform its intended functions without failure, instilling confidence in its safety features and reliability as a mobility service [33] and is a critical factor influencing technology adoption. This concept goes beyond mere functionality; it includes users’ perceptions of a system’s ability to deliver its intended outcomes consistently, without errors, failures, or unexpected disruptions. Within the framework of the Technology Acceptance Model [34], perceived reliability significantly shapes user confidence and trust in technology. This, in turn, affects perceived usefulness and perceived ease of use, ultimately influencing adoption intentions. A technology perceived as reliable instills a sense of security and predictability, thereby reducing user anxiety regarding possible malfunctions or inconsistencies. Consequently, this increases the likelihood that users will view the technology as useful for achieving their goals and easy to operate without frustrating technical issues.

Empirical research across various technological domains highlights the importance of perceived reliability. In the realm of financial technologies, ref. [35] found a positive relationship between perceived reliability and perceived usefulness in the adoption of Sharia fintech among micro-, small-, and medium-sized enterprises (MSMEs). This indicates that when users consider a fintech platform reliable, they are more likely to see it as effective in facilitating their business needs. Similarly, ref. [36] identified perceived reliability as a significant predictor of perceived usefulness concerning online savings behavior, suggesting that users tend to engage more with online financial services they regard as dependable. Ref. [37] also emphasized the importance of perceived reliability in e-banking adoption, showing how it influences perceived usefulness through the mediating roles of trust and security. Collectively, these findings suggest that perceived reliability is not just a technical characteristic of a system but also a vital psychological factor that influences user perceptions and behaviors.

Moreover, the impact of perceived reliability may vary across different contexts. Ref. [38] noted that cultural and social factors can moderate how perceived reliability affects technology acceptance. User demographics, prior experiences with technology, and prevailing societal norms can all shape how individuals perceive and value reliability. Therefore, a nuanced understanding of the specific context in which technology is deployed is essential for effectively addressing reliability concerns and promoting user acceptance. This is particularly evident in the context of self-driving vehicles, where safety and security are critical. Users must feel confident that SDVs can consistently and accurately perceive their surroundings, make appropriate driving decisions, and operate safely under varying conditions. Understanding perceived reliability as a multi-faceted construct influenced by both system characteristics and contextual factors provides a strong theoretical basis for exploring its role in SDV adoption.

2.2. Perceived Risk

Perceived risk, in the context of technology adoption, refers to an individual’s subjective evaluation of the potential negative outcomes associated with using a specific technology [39]. This concept is multifaceted, encompassing various dimensions of concern that can serve as significant barriers to adoption. These dimensions include, but are not limited to, financial risk (the potential for monetary loss), performance risk (concerns about whether the technology will function as intended), privacy risk (fears related to data breaches and misuse of personal information), social risk (worries about social disapproval or negative perceptions from others), and, particularly in the case of self-driving vehicles, safety risk (apprehensions about physical harm or accidents). High levels of perceived risk can foster uncertainty and distrust, ultimately hindering adoption across different technological domains. Empirical research across various sectors highlights the significant impact of perceived risk on technology acceptance. For example, in the realm of digital finance, concerns about financial security and the potential for data misuse have been identified as key obstacles to the adoption of digital assets and cryptocurrencies [40,41]. Similarly, in education, privacy risks associated with the use of AI technologies like ChatGPT have been shown to impede their acceptance [42]. In the healthcare sector, worries about data security and patient privacy greatly influence the adoption of Internet of Things (IoT) systems [43]. Moreover, research within the transportation domain has identified safety and reliability concerns as critical barriers to the adoption of electric motorcycles [44]. These findings consistently demonstrate that perceived risk is a strong predictor of technology adoption intentions across a wide range of applications. In the context of SDVs, perceived risk becomes particularly relevant due to the inherent concerns about safety and control. Potential users may feel anxious about a vehicle’s ability to navigate complex traffic situations, respond appropriately to unexpected events, and ensure passenger safety in the absence of a human driver. These safety concerns are often interconnected with perceptions of reliability, as users may equate a lack of reliability with an increased likelihood of accidents or malfunctions. Addressing these specific concerns is crucial for fostering public acceptance of SDVs.

2.3. Technological Competence

Technological competence (TC) refers to a user’s perception of technology’s sophistication, functionality, and reliability. It encompasses the belief that autonomous vehicle (AV) technology is not only advanced but also capable of performing its intended functions efficiently and dependably [45]. This concept goes beyond objective technical features to include the subjective confidence users have in the technology’s ability to operate without failure, aligning expectations with actual performance [46]. In the context of AV technology, TC embodies the perception that critical components such as sensors, machine learning algorithms, and safety mechanisms are designed with high precision and robustness. These elements work together to enhance reliability and foster trustworthiness [47]. Importantly, TC reflects the expectation that the system can adapt to diverse and dynamic road conditions, operate autonomously with minimal human intervention, and consistently deliver safety and efficiency in various scenarios [48]. Understanding TC is essential for grasping the adoption of AVs, as perceived technological competence directly influences user acceptance. Users are more likely to trust and use AVs when they believe the technology can effectively mitigate potential risks and perform safely under various conditions [45]. The relationship between perceived reliability, functionality, and trustworthiness is central to shaping users’ intentions to engage with autonomous systems, as highlighted by [45]. Therefore, demonstrating technological sophistication, seamless functionality, and a proven track record of reliability is crucial for building user confidence and accelerating the widespread adoption of AV technology [46,47].

3. Methodology

3.1. Sampling and Data Collection

This study focused on Ghana due to its emerging potential for the adoption of self-driving vehicles. This potential is supported by advancements in technology, government efforts to modernize infrastructure, and a growing interest in sustainable mobility solutions, such as electric buses and ride-hailing platforms like Uber and Bolt. Data were collected through face-to-face surveys between January 2023 and October 2023 [49], using a stratified random sampling method [50]. Surveys were conducted across all 16 administrative regions of Ghana, capturing a diverse population sample from urban, peri-urban, and rural areas, including cities such as Accra, Kumasi, Tamale, and Cape Coast. Predefined inclusion criteria ensured that respondents were at least 18 years old and capable of providing informed consent, with an intentional balance across age groups, educational backgrounds, income levels, and driving experience.

Ten trained researchers administered the surveys and were prepared to address issues such as participant disengagement and comprehension. Participants were briefed on level-5 SDV technology to provide a foundational understanding, ensuring informed responses. Participation was voluntary and uncompensated, with each survey averaging about 20 min in duration. Rigorous quality control measures were implemented, including real-time monitoring by researchers to detect disengagement, timing analysis based on recorded survey durations, and embedded attention-check questions to gauge participant focus. Only data from participants who demonstrated positive engagement and were observed to be attentive were included in the final analysis. Approximately 8.6% of initially approached participants (117 out of 1365) were excluded from the final analysis due to demonstrated confusion about the technology, disinterest during the interview process, or incomplete responses. This filtering process underscores our commitment to data quality and ensures the reliability of the findings presented.

3.2. Measurement and Survey Instrument

The constructs examined in this study were evaluated using a structured questionnaire based on established and validated scales from prior research. Each item was measured on a five-point Likert scale, ranging from 1 (strongly disagree) to 5 (strongly agree), to capture participants’ responses effectively. Perceived usefulness, perceived ease of use, and intention to use items were adapted from [34]. Perceived reliability was assessed using items adapted from [33], which focuses on users’ trust in the dependability and consistency of autonomous vehicle systems. Technological competence was measured using items adapted from [51], highlighting participants’ confidence in the capability and sophistication of AV technology. The perceived risk scale included items that addressed safety concerns and system reliability, drawing on the work of [52].

Figure 1 presents the QCA set-theoretic framework utilized in this study, offering a conceptual roadmap of how different conditions interact to influence the intention to adopt self-driving vehicles in Ghana. The framework is structured around six primary constructs: perceived usefulness (PU), perceived reliability (PRel), perceived ease of use (PEOU), perceived risk (PRisk), technological competence (TC), and intention to use (IU). Each of these constructs corresponds to a condition in the fsQCA (software Version 4.1), reflecting both theoretical relevance and empirical measurability. Figure 1 illustrates how these constructs are not treated in isolation but rather as interdependent factors that form unique configurations (or pathways) leading to either high or low intention to adopt SDVs. The directional arrows in the framework symbolize the configurational logic of fsQCA, highlighting that causality is conjunctural (conditions operate in combination), equifinal (multiple combinations can lead to the same outcome), and asymmetric (causal paths to the presence of the outcome differ from those leading to its absence). Importantly, the framework draws from the Technology Acceptance Model (TAM) and extends it by integrating perceived risk and technological competence, two critical dimensions particularly relevant in the socio-technological context of developing economies. As such, this framework guides the empirical analysis by identifying which combinations of conditions (presence or absence) are sufficient or necessary to explain the intention to adopt SDVs.

3.3. Asymmetric Data Analysis

Fuzzy set qualitative comparative analysis is a highly effective methodology for examining the factors influencing the adoption of self-driving vehicles in Ghana. Fuzzy set variables represent a sophisticated analytical approach that extends beyond traditional binary categorizations. Unlike conventional variables that assign cases to mutually exclusive categories (e.g., 0 or 1), fuzzy set variables capture degrees of membership in conceptual sets, ranging from 0 (full non-membership) to 1 (full membership), with gradations in between. The adoption of SDVs in developing countries like Ghana is a complex process shaped by interconnected socio-demographic, experiential, and technological factors; hence, fsQCA possesses the essential ability to capture configurational causality [18,53]. Rooted in complexity theory, fsQCA offers a detailed perspective on causal relationships, acknowledging the intricate, nonlinear interactions that characterize human behavior and decision-making in complex systems, particularly regarding the adoption of new technologies [53,54].

fsQCA is based on three main principles: conjunctural causality, equifinality, and asymmetric causality. These principles make it particularly suitable for understanding the factors that influence SDV adoption in Ghana. Conjunctural causality underscores that outcomes result from specific combinations of interacting conditions rather than from individual factors in isolation [53,54]. Equifinality recognizes that multiple pathways can lead to the same outcome, reflecting the diverse influences of socio-demographic, experiential, and technological factors on SDV adoption [55,56]. Asymmetric causality points out that causal relationships are not uniform or reciprocal [57,58]. By leveraging these principles, fsQCA provides a nuanced understanding of the complex, context-specific configurations that drive SDV adoption, offering valuable insights for policymakers and stakeholders as they navigate Ghana’s unique challenges [59].

This study employs fsQCA to analyze the complex configurations affecting SDV adoption intentions in Ghana. By examining combinations of socio-demographic characteristics, driving experience, and constructs from the Technology Acceptance Model and related theories, fsQCA delivers a deeper and more holistic understanding of causality [56,58,59]. This configurational approach is vital in the Ghanaian context, where factors such as infrastructure, driving conditions, and technological exposure may interact in complex ways. Unlike variable-oriented methods, fsQCA is specifically designed to handle causal asymmetry and nonlinear interactions [57,60].

Figure 2 illustrates the basic steps and workflow of fuzzy set qualitative comparative analysis (fsQCA), starting from sample data collection through contrarian case analysis, data calibration, truth table creation, and solution generation.

4. Results

4.1. Demographic Analysis

Table 1. Demographic profile.

	Profile	Number	Percentage (%)
Gender	Male	854	68.42
	Female	394	31.57
Age	18–25	181	14.5
	26–35	275	22.04
	36–45	454	36.38
	46–55	226	18.11
	56–65	112	8.97
Educational Qualification	Senior High School (SHS)	178	14.26
	Bachelor’s degree	628	50.32
	Master’s degree and above	320	25.64
	Other	122	9.77
Annual Income (Gh₵)	≤10,000	172	13.78
	10,001–20,000	317	25.4
	20,001–30,000	297	23.8
	30,001–50,000	386	30.93
	≥50,001	76	6.09
	Driving Experience
Car driving license holder	Yes	1009	80.84
	No	239	19.15
Years of driving experience	<1 year	52	4.17
	1–3 years	168	13.46
	4–6 years	287	23
	7–10 years	486	38.94
	>10 years	255	20.43
Electric Vehicles (EVs) Experience	Yes	333	26.68
	No	915	73.32
Cruise Control Experience	Yes	1002	80.29
	No	246	19.71

provides a detailed summary of the demographic profile of the study participants, including gender, age, educational qualifications, income levels, driving experience, and prior exposure to electric vehicles (EVs) and cruise control systems.

4.2. Measurement Model Assessment

To ensure the validity and reliability of the constructs, a confirmatory factor analysis (CFA) was conducted using Analysis of Moment Structures (AMOS) software version 29 before proceeding with the qualitative comparative analysis (QCA).

CFA is a statistical technique used to confirm that the measurement items accurately represent the underlying theoretical constructs in this case, TAM and its extensions. Establishing validity and reliability through CFA is especially important in studies using methods like fsQCA, which rely on calibrated scores derived from these constructs. A robust CFA ensures that the subsequent set-theoretic analysis is based on empirically sound and theoretically consistent measures [62]. This is particularly relevant here because the study critiques linear variance-based methods like traditional factor analysis, yet still leverages CFA as a foundational validation step, reinforcing the quality of the data inputs used for the more suitable fsQCA approach. The model fit indices demonstrated a strong alignment with the recommended thresholds, indicating an acceptable fit to the data. The chi-square value (χ² = 672.818, df = 216, p < 0.001) was significant; however, the χ²/df ratio of 3.115 falls below the recommended cutoff of 5.0, which supports model fit [63]. Additionally, the Root Mean Square Error of Approximation (RMSEA = 0.055) and the Standardized Root Mean Square Residual (SRMR = 0.048) meet the recommended thresholds of <0.08 and <0.50, respectively [64]. These fit indices represent established benchmarks in the structural equation modeling literature for assessing how well the proposed model represents the observed data. The comparative fit index (CFI = 0.957) and the Tucker–Lewis Index (TLI = 0.946) both exceed the recommended threshold of >0.90, further confirming the model’s goodness of fit [64]. The measurement model demonstrated strong factor loadings across all items, ranging from 0.682 to 0.981. Cronbach’s alpha values varied from 0.867 to 0.983, composite reliability (CR) values were between 0.921 and 0.977, and average variance extracted (AVE) values ranged from 0.792 to 0.961. These metrics meet the recommended thresholds, indicating excellent reliability, internal consistency, and convergent validity [65] (see

Table 2. Reliability and validity of the conditions.

Code	Measurement Items	Loadings	Cronbach Alpha	CR	AVE
Perceived Reliability
PRel	Autonomous vehicles are reliable.	0.868	0.917	0.941	0.883
PRel	I do not have suspicions about automated vehicles.	0.846
PRel	I would engage in other tasks while riding in an automated vehicle.	0.765
PRel	I feel hesitant about using an automated vehicle.	0.932
Technological Competence
TC	I believe autonomous vehicles are equipped with advanced technology that can handle various driving situations.	0.783	0.867	0.921	0.801
TC	I trust that the technology behind autonomous vehicles is reliable and efficient.	0.836
TC	I feel confident that autonomous vehicles can perform tasks without human intervention.	0.682
TC	I am convinced that autonomous vehicle technology is capable of adapting to different road conditions and environments.	0.941
Perceived Ease of Use
PEOU	Learning to operate an autonomous vehicle would be easy for me.	0.742	0.975	0.977	0.938
PEOU	I would find it easy to get an autonomous vehicle to do what I want.	0.956
PEOU	Interacting with an autonomous vehicle would not require much mental effort.	0.923
PEOU	I would find it easy to become skilled at using autonomous vehicles.	0.895
Perceived Risk
PRisk	Autonomous vehicles are more likely to lead me to a fatal accident.	0.732	0.929	0.928	0.824
PRisk	Autonomous vehicles might not perform well and could crash when faced with small problems.	0.922
PRisk	Using autonomous vehicles would be risky.	0.901
PRisk	I am concerned about equipment and system failures in autonomous vehicles.	0.881
Intention to Use
IU	I plan to use an autonomous vehicle in the future.	0.955	0.983	0.975	0.961
IU	I expect to use an autonomous vehicle in the future.	0.958
IU	I intend to use an autonomous vehicle in the future.	0.981
IU	I would recommend that my family members and friends ride in an autonomous vehicle.	0.978
Perceived Usefulness
PU	Using autonomous vehicles will boost my productivity.	0.682	0.884	0.926	0.792
PU	Automated vehicles will help alleviate traffic congestion.	0.837
PU	Automated vehicles will assist with parking.	0.853
PU	The use of AVs will lead to a decrease in accidents.	0.913

).

4.3. Discriminant Validity

The discriminant validity of the constructs was evaluated using both the Fornell–Larcker criterion and the heterotrait–monotrait (HTMT) ratio. The square roots of the average variance extracted (AVE) were greater than the inter-construct correlations, indicating sufficient discriminant validity [66]. Furthermore, all HTMT values were below the threshold of 0.85, which further supports the findings of discriminant validity [67] (see

Table 3. Discriminant validity.

Constructs	PU	PEOU	PRel	TC	PRisk	IU
PU	0.882	0.837	0.591	0.673	0.569	0.774
PEOU	0.628	0.953	0.537	0.692	0.516	0.793
PRel	0.736	0.812	0.830	0.707	0.537	0.763
TC	0.874	0.803	0.853	0.908	0.522	0.833
PRisk	−0.451	−0.371	−0.454	−0.547	0.883	0.587
IU	0.736	0.831	0.754	0.772	−0.491	0.983

Notes: Diagonal and bold are the square roots of the AVE. Below the diagonal elements are the correlations between the construct’s values. Above the diagonal elements are the heterotrait–monotrait ratio of correlation values.

).

4.4. Contrarian Case Analysis to Justify the Use of fsQCA

Contrarian case analysis investigates instances that traditional variance-based methods, such as correlation or regression, cannot explain [54,68]. While these traditional methods focus on average net effects, they often fail to account for complex interdependencies and outlier cases. This oversight is particularly significant in the social sciences, where context-specific factors heavily influence behavior. Understanding these is essential for identifying cases in which the relationships between variables deviate from expected main-effect predictions. Although contrarian case analysis aligns well with the configurational logic of fuzzy set qualitative comparative analysis, it is seldom utilized in fsQCA studies [62]. Traditional methods or initial fsQCA results may reveal dominant relationships; however, contradictory or inverse patterns often arise within specific subsets. This necessitates further examination to uncover hidden complexities [54]. For instance, two variables within the same dataset might show different relationships, such as positive, negative, or none, depending on particular conditions. These cases illustrate the limitations of relying solely on average effects and highlight the value of fsQCA’s configurational approach, which explicitly considers the interplay of multiple conditions and identifies distinct causal pathways. In this study, variables were divided into quintiles for a more detailed analysis, and cross-tabulations were conducted to reveal patterns that deviated from expected relationships.

Table 4. Contrarian case analysis.

Outcome: Intention to Use Self-Driving Vehicles
Perceived Usefulness							Perceived Reliability
	1	2	3	4	5			1	2	3	4	5
1	208	29	4	4	4		1	212	29	8	4	0
	16.7%	2.3%	0.3%	0.3%	0.3%			17.0%	2.3%	0.6%	0.3%	0.0%
2	37	106	66	28	3		2	34	115	72	30	8
	3.0%	8.5%	5.3%	2.2%	0.2%			2.7%	9.2%	5.8%	2.4%	0.6%
3	2	36	143	85	12		3	1	25	129	65	8
	0.2%	2.9%	11.5%	6.8%	1.0%			0.1%	2.0%	10.3%	5.2%	0.6%
4	0	8	37	151	39		4	0	11	39	167	33
	0.0%	0.6%	3.0%	12.1%	3.1%			0.0%	0.9%	3.1%	13.4%	2.6%
5	0	1	3	19	223		5	0	0	5	21	232
	0.0%	0.1%	0.2%	1.5%	17.9%			0.0%	0.0%	0.4%	1.7%	18.6%
Perceived Ease of Use							Perceived Risk
	1	2	3	4	5			1	2	3	4	5
1	216	27	6	4	0		1	181	28	6	4	0
	17.3%	2.2%	0.5%	0.3%	0.0%			14.5%	2.2%	0.5%	0.3%	0.0%
2	30	110	66	49	5		2	63	105	67	42	5
	2.4%	8.8%	5.3%	3.9%	0.4%			5.0%	8.4%	5.4%	3.4%	0.4%
3	1	29	140	56	7		3	2	35	148	62	13
	0.1%	2.3%	11.2%	4.5%	0.6%			0.2%	2.8%	11.9%	5.0%	1.0%
4	0	12	38	163	31		4	0	10	29	169	30
	0.0%	1.0%	3.0%	13.1%	2.5%			0.0%	0.8%	2.3%	13.5%	2.4%
5	0	2	3	15	238		5	1	2	3	10	233
	0.0%	0.2%	0.2%	1.2%	19.1%			0.1%	0.2%	0.2%	0.8%	18.7%
Technological Competence
	1	2	3	4	5
1	223	27	3	2	0		Cases in bold shapes represent contrarian cases. Cases in dotted shapes represent the main effects.
	17.9%	2.2%	0.2%	0.2%	0.0%
2	23	115	59	40	1
	1.8%	9.2%	4.7%	3.2%	0.1%
3	1	32	167	54	9		The sets of contrarian cases are counter to the main effect size.
	0.1%	2.6%	13.4%	4.3%	0.7%
4	0	4	22	171	35
	0.0%	0.3%	1.8%	13.7%	2.8%
5	0	2	2	20	236
	0.0%	0.2%	0.2%	1.6%	18.9%

presents these findings, demonstrating instances where observed relationships contradicted main effects. These deviations validate the use of fsQCA for capturing complex dynamics and provide empirical support for its application in understanding factors that influence the intention to use self-driving vehicles [61].

4.5. Calibration in Fuzzy Set Qualitative Comparative Analysis

Calibration is a critical step in fuzzy set qualitative comparative analysis, transforming multi-item survey variables, typically measured on Likert scales, into calibrated fuzzy set membership scores [53]. This process enables the analysis of how configurations of conditions influence the outcome of interest by translating nuanced ordinal data into the set-theoretic logic of fsQCA. In this study, the arithmetic mean of responses for each construct was calculated to simplify multi-item measures, following established practices in the fsQCA literature [56,61]. The aggregated scores were then calibrated using the direct calibration method, which defines three qualitative thresholds: full membership (0.95), the crossover point (0.5), and full non-membership (0.05). A small constant (0.001) was added to membership scores below 1.0 (full membership) to resolve ambiguities arising from calibrated scores of 0.5. These “borderline” cases, with a membership score of 0.5, are typically excluded from the truth table minimization process in fuzzy set qualitative comparative analysis because they represent maximum ambiguity (i.e., the case is neither clearly in nor clearly out of the set). Adding the small constant shifts, these scores are slightly above 0.5, allowing them to be included in the analysis as “mostly in” the set, thus avoiding their exclusion and potential loss of information [69]. These thresholds were chosen in alignment with best practices outlined by [61,70], ensuring methodological consistency and rigor. Direct calibration offers several advantages. It ensures transparency and replicability by explicitly defining and justifying the calibration criteria. The resulting membership scores, ranging from 0 to 1, provide a clear and interpretable measure of each case’s degree of membership in a fuzzy set, facilitating effective communication of findings.

4.6. Analysis of Necessary Conditions (Necessity in Kind)

The analysis of necessity identifies key conditions essential for having a high or low intention to use self-driving vehicles. According to [70], a condition is deemed necessary if its consistency score is >0.90. Consistency measures how consistently the outcome aligns with the presence of a condition, while coverage evaluates the condition’s empirical relevance.

For individuals with a high intention to use SDVs, three conditions exceeded the 0.90 consistency threshold: perceived reliability (0.908), perceived ease of use (0.901), and technological competence (0.879). Coverage scores for these conditions ranged from 0.91 to 0.94, indicating their strong empirical relevance in explaining high intention.

In contrast, for individuals with a not-high intention to use SDVs, no positive conditions met the 0.90 threshold. However, the analysis of negated conditions revealed that ~technological competence (~TC) had a high consistency score of 0.935, indicating that a lack of technological competence is a necessary condition for the absence of high intention to use SDVs (see

Table 5. Necessary condition analysis.

Outcome: High Intention to Use				Outcome: Not-High Intention to Use
Conditions	Consistency	Coverage		Conditions	Consistency	Coverage
PU	0.891553	0.912386		PU	0.456957	0.396507
PRel	0.908374	0.910356		PRel	0.463208	0.393611
PEOU	0.900674	0.913031		PEOU	0.460781	0.396057
PRisk	0.836975	0.933805		PRisk	0.421106	0.398363
TC	0.879382	0.940683		TC	0.437853	0.397135
~PU	0.410284	0.471193		~PU	0.899028	0.875453
~PRel	0.394929	0.464581		~PRel	0.894505	0.892214
~PEOU	0.404227	0.469249		~PEOU	0.898819	0.884697
~PRisk	0.460744	0.484181		~PRisk	0.830025	0.828681
~TC	0.436418	0.477969		~TC	0.934600	0.867897

Abbreviations: PU: perceived usefulness; PU: perceived usefulness; PEOU: perceived ease of use; PRel: perceived reliability; PRisk: perceived risk; TC: technological competence. Conditions in bold and italics indicate necessary conditions for the outcome.

).

4.7. Analysis of Sufficiency

The sufficiency analysis in fuzzy set qualitative comparative analysis identifies combinations of conditions sufficient to achieve the desired outcome, uncovering complex causal relationships beyond simple correlations [53]. This study followed established fsQCA protocols, beginning with the construction of truth tables (Table 6 and Table 7), which enumerate all possible combinations of conditions, their frequency within the dataset, and their consistency with the outcome.

To ensure robustness, configurations with fewer than three cases were excluded based on previous studies [69,71]. Additional criteria included a raw consistency (RI) threshold of >0.80 and a proportional reduction in inconsistency (PRI) score of >0.5, ensuring only stable and replicable configurations were analyzed [61,72]. Raw consistency measures the alignment between configurations and outcomes, while PRI assesses improvement over random chance. Using fsQCA software, logical minimization was applied to derive three solutions: complex, parsimonious, and intermediate. The complex solution retains all relevant conditions, offering detailed but intricate results. The parsimonious solution simplifies configurations by assuming “don’t care” conditions, which may result in the loss of important contextual information. The intermediate solution, which incorporates theoretical and empirical considerations, was prioritized for its methodological strength [53,69]. Conditions identified in both parsimonious and intermediate solutions are classified as core conditions, reflecting their critical role in achieving the outcome. Peripheral conditions, present only in the intermediate solution, support causal configurations but are not essential. This distinction helps policymakers and developers focus on the most important factors while addressing secondary elements when possible [61].

This study ensures theoretical rigor and empirical validity by integrating parsimonious and intermediate solutions. The findings provide a clear understanding of the relationships between necessary and sufficient conditions, offering practical guidance for promoting autonomous vehicle adoption in developing countries [61,62].

Truth Tables After Sorting

Table 6. High intention to use self-driving vehicles.

PU	PRel	PEOU	PRisk	TC	Number	IU	Raw Consist.	PRI Consist.	SYM Consist.
1	1	1	1	1	419	1	0.965434	0.944359	0.969274
1	1	1	0	1	21	1	0.962614	0.817092	0.81773
0	1	1	1	1	7	1	0.959368	0.742923	0.746762
1	1	0	0	1	3	1	0.95892	0.72032	0.720321
1	1	1	1	0	11	1	0.951248	0.715526	0.719293
0	1	1	0	1	7	1	0.955442	0.666658	0.666658
1	0	0	0	1	3	1	0.956265	0.654872	0.656809
0	1	0	1	1	3	1	0.956857	0.642587	0.642588
1	1	1	0	0	20	1	0.935009	0.640086	0.647627
0	0	1	1	1	4	1	0.950709	0.624993	0.624993
1	1	0	1	0	3	1	0.951694	0.610331	0.610332
1	1	0	0	0	3	1	0.931329	0.557212	0.557213
0	0	0	0	1	4	1	0.936451	0.536331	0.536332
1	0	0	0	0	9	0	0.911525	0.454325	0.454446
0	1	0	0	0	6	0	0.910579	0.446212	0.448355
0	0	0	1	0	3	0	0.924043	0.396408	0.397774
0	0	0	0	0	412	0	0.454247	0.069767	0.071651

Table 6. High intention to use self-driving vehicles.

PU	PRel	PEOU	PRisk	TC	Number	IU	Raw Consist.	PRI Consist.	SYM Consist.
1	1	1	1	1	419	1	0.965434	0.944359	0.969274
1	1	1	0	1	21	1	0.962614	0.817092	0.81773
0	1	1	1	1	7	1	0.959368	0.742923	0.746762
1	1	0	0	1	3	1	0.95892	0.72032	0.720321
1	1	1	1	0	11	1	0.951248	0.715526	0.719293
0	1	1	0	1	7	1	0.955442	0.666658	0.666658
1	0	0	0	1	3	1	0.956265	0.654872	0.656809
0	1	0	1	1	3	1	0.956857	0.642587	0.642588
1	1	1	0	0	20	1	0.935009	0.640086	0.647627
0	0	1	1	1	4	1	0.950709	0.624993	0.624993
1	1	0	1	0	3	1	0.951694	0.610331	0.610332
1	1	0	0	0	3	1	0.931329	0.557212	0.557213
0	0	0	0	1	4	1	0.936451	0.536331	0.536332
1	0	0	0	0	9	0	0.911525	0.454325	0.454446
0	1	0	0	0	6	0	0.910579	0.446212	0.448355
0	0	0	1	0	3	0	0.924043	0.396408	0.397774
0	0	0	0	0	412	0	0.454247	0.069767	0.071651

Table 7. Not-high intention to use self-driving vehicles.

PU	PRel	PEOU	PRisk	TC	Number	~IU	Raw Consist.	PRI Consist.	SYM Consist.
0	0	0	0	0	412	1	0.943641	0.903936	0.928349
0	0	0	1	0	3	1	0.949683	0.600157	0.602226
0	1	0	0	0	6	1	0.927178	0.549009	0.551645
1	0	0	0	0	9	1	0.926294	0.54541	0.545554
0	0	0	0	1	4	0	0.926492	0.463668	0.463668
1	1	0	0	0	3	0	0.913583	0.442787	0.442788
1	1	0	1	0	3	0	0.924339	0.389667	0.389668
0	0	1	1	1	4	0	0.917852	0.375007	0.375007
0	1	0	1	1	3	0	0.922433	0.357411	0.357412
1	1	1	0	0	20	0	0.882315	0.348269	0.352373
1	0	0	0	1	3	0	0.91664	0.34218	0.343191
0	1	1	0	1	7	0	0.910888	0.333343	0.333342
1	1	0	0	1	3	0	0.894199	0.279679	0.279679
1	1	1	1	0	11	0	0.87648	0.279238	0.280708
0	1	1	1	1	7	0	0.881767	0.251937	0.253238
1	1	1	0	1	21	0	0.832829	0.182127	0.18227
1	1	1	1	1	419	0	0.39736	0.029937	0.030727

Table 7. Not-high intention to use self-driving vehicles.

PU	PRel	PEOU	PRisk	TC	Number	~IU	Raw Consist.	PRI Consist.	SYM Consist.
0	0	0	0	0	412	1	0.943641	0.903936	0.928349
0	0	0	1	0	3	1	0.949683	0.600157	0.602226
0	1	0	0	0	6	1	0.927178	0.549009	0.551645
1	0	0	0	0	9	1	0.926294	0.54541	0.545554
0	0	0	0	1	4	0	0.926492	0.463668	0.463668
1	1	0	0	0	3	0	0.913583	0.442787	0.442788
1	1	0	1	0	3	0	0.924339	0.389667	0.389668
0	0	1	1	1	4	0	0.917852	0.375007	0.375007
0	1	0	1	1	3	0	0.922433	0.357411	0.357412
1	1	1	0	0	20	0	0.882315	0.348269	0.352373
1	0	0	0	1	3	0	0.91664	0.34218	0.343191
0	1	1	0	1	7	0	0.910888	0.333343	0.333342
1	1	0	0	1	3	0	0.894199	0.279679	0.279679
1	1	1	1	0	11	0	0.87648	0.279238	0.280708
0	1	1	1	1	7	0	0.881767	0.251937	0.253238
1	1	1	0	1	21	0	0.832829	0.182127	0.18227
1	1	1	1	1	419	0	0.39736	0.029937	0.030727

4.7.1. Configurations for High Intention to Use Self-Driving Vehicles

The configurations sufficient for high intention to use self-driving vehicles, expressed in fsQCA notation, are as follows:

PU * PRel * ~TC + PRel * PEOU * TC + ~PRel * ~PEOU * ~PRisk * TC + ~PU * PRel * PRisk * TC + ~PU * PEOU * PRisk * TC + PU * PRel * ~PRisk + PU * ~PEOU * ~PRisk * TC → IU

(1)

where “*” denotes logical AND, “+” denotes logical OR, and “~” denotes negation.

These configurations demonstrate that a high intention to use self-driving vehicles can emerge from multiple distinct pathways. For instance, the desired outcome can result from the combination of perceived usefulness and perceived reliability in the absence of technological competence or through perceived reliability, perceived ease of use, and technological competence. Other pathways involve combinations such as the absence of perceived risk with key factors like perceived usefulness and technological competence.

The consistency scores for individual configurations ranged from 0.924 to 0.957, indicating their reliability in producing the desired outcome. Raw coverage scores varied between 0.317 and 0.835, reflecting the proportion of cases explained by each pathway. The overall solution achieved a consistency score of 0.933, demonstrating the robustness of the identified pathways, and a solution coverage score of 0.891, indicating that 89.1% of cases with a high intention to adopt self-driving vehicles are accounted for by these configurations. These results highlight the effectiveness of the configurations in explaining the outcome (see

Table 8. Sufficiency analysis results of high intention to use self-driving vehicles.

Outcome: High Intention to Use Self-Driving Vehicles
Configuration/Solutions	1	2	3	4	5	6	7
Perceived Usefulness	⬤	◯	◯	⊗	⊗	⬤	⬤
Perceived Ease of Use	◯	●	⊗	◯	●	◯	⊗
Perceived Reliability	⬤	⬤	⊗	⬤	◯	⬤	◯
Perceived Risk	◯	◯	⊗	●	●	⊗	⊗
Technological Competence	⊗	⬤	⬤	⬤	⬤	◯	⬤
Consistency	0.924	0.956	0.935	0.956	0.951	0.929	0.955
Raw Coverage	0.373	0.834	0.319	0.316	0.321	0.388	0.322
Unique Coverage	0.004	0.437	0.008	0.0007	0.004	0.0023	0.00013
Overall Solution Consistency	0.933223
Overall Solution Coverage	0.891242

Notes: ⬤ indicates the presence of core condition; ● indicates the existence of peripheral conditions; ⊗ indicate the absence of core condition; ^⊗ indicate the absence of conditions; ◯ indicates the causal conditions may or may not be present or “don’t care”. Large circles = core conditions; Small circles = peripheral conditions.

).

4.7.2. Configurations for Not-High Intention to Use Self-Driving Vehicles

The fsQCA analysis identified two configurations sufficient to explain low intention to use self-driving vehicles. These configurations, expressed in fsQCA notation, are as follows:

~PRel * ~PEOU * ~TC * PRisk + ~PU * ~PEOU * ~TC * PRisk → ~IU

(2)

where “*” denotes logical AND, “+” denotes logical OR, and “~” denotes negation.

These configurations demonstrate that a not-high intention to use self-driving vehicles can result from specific combinations of absent and present conditions. For instance, the outcome can emerge from the absence of perceived reliability, perceived ease of use, and technological competence, combined with the presence of perceived risk. Similarly, it can also result from the absence of perceived usefulness, perceived ease of use, and technological competence, alongside the presence of perceived risk.

The consistency scores for both configurations were 0.934, indicating their reliability in producing the outcome. Raw coverage scores were 0.864 and 0.858, reflecting the significant proportion of cases explained by each configuration. The overall solution achieved a consistency score of 0.926 and a coverage score of 0.875, demonstrating the robustness and explanatory power of the identified pathways (see

Table 9. Sufficiency analysis results of not-high intention to use self-driving vehicles.

Outcome: Not-High Intention to Use Self-Driving Vehicles
Configuration/Solutions	1	2
Perceived Usefulness	◯	⊗
Perceived Ease of Use	⊗	⊗
Perceived Reliability	⊗	◯
Perceived Risk	●	●
Technological Competence	⊗	⊗
Consistency	0.933845	0.934296
Raw Coverage	0.863771	0.858445
Unique Coverage	0.0170264	0.0117004
Overall Solution Consistency	0.926261
Overall Solution Coverage	0.875471

Notes: ⬤ indicates the presence of core condition; ● indicates the existence of peripheral conditions; ⊗ indicate the absence of core condition; ^⊗ indicate the absence of conditions; ◯ indicates the causal conditions may or may not be present or “don’t care”. Large circles = core conditions; Small circles = peripheral conditions.

).

4.7.3. Predictive Validity (XY Plots)

Predictive validity, assessing a model’s ability to forecast the dependent variable in independent samples, is crucial for establishing robustness and generalizability [73]. To evaluate the predictive validity of the fsQCA model, the dataset was split into calibration (analysis) and holdout (test) subsamples, following [62]. This split mitigates overfitting and provides a more realistic estimate of predictive performance [73].

We calculated the sufficient configurations for the outcome(s) in the subsample (see

Table 10. Configurations indicating high intention to use self-driving vehicles for subsample (where “*” denotes logical AND).

Outcome: High IU
Models	Raw Coverage	Unique Coverage	Consistency
PU*PRel*PEOU	0.860004	0.0409976	0.943911
PRel*PEOU*TC	0.838461	0.0134037	0.959257
~PU*~PRel*~PEOU*~PRisk*TC	0.302896	0.0176091	0.943099
solution coverage	0.897068
solution consistency	0.932857
Outcome: Not-High IU
Models	Raw Coverage	Unique Coverage	Consistency
~PU*~PEOU*~PRisk*~TC	0.862213	0.862213	0.94008
solution coverage	0.862213
solution consistency	0.94008

), using established frequency and consistency thresholds to ensure robustness [62]. These configurations were then applied as fuzzy set variables to the holdout sample, using fuzzy set functions in the software. Consistency and coverage scores were calculated for these configurations in the holdout sample using the fuzzy XY plot method. As noted by [61], high consistency in one value implies a corresponding coverage score for the other. The fsQCA predictive validity test confirmed the model’s robust predictive capabilities, with consistency and coverage values in the holdout sample closely aligning with those from the subsample [62]. This alignment demonstrates the reliability and generalizability of the findings, indicating that the identified configurations represent genuine causal relationships applicable to new data. Figure 3 and Figure 4 show the XY plot. Figure 3 displays the fuzzy set XY plots demonstrating the predictive validity of the identified configurations from the subsample when applied to the holdout sample. The X-axis represents membership in the causal configurations (PU*PRel*PEOU, PRel*PEOU*TC, and ~PU*~PRel*~PEOU*~PRisk*TC), while the Y-axis represents membership in the outcome set (high intention to use self-driving vehicles). The distribution of cases above the diagonal line indicates consistent sufficiency relationships with high consistency scores (ranging from 0.92 to 0.96) and coverage values (0.83–0.94), validating that these configurations reliably predict high adoption intention across independent samples.

Figure 4 presents the fuzzy set XY plot for the configuration explaining the low intention to use self-driving vehicles. The X-axis shows membership in the causal configuration (~PU*~PEOU*~PRisk*~TC), while the Y-axis represents membership in the outcome set (not-high intention). The clustering of cases above the diagonal line with a high consistency score (0.93) confirms the robust predictive validity of this configuration across both samples, demonstrating that the absence of perceived usefulness, ease of use, and technological competence reliably predicts low adoption intention even when perceived risk is absent.

5. Discussion

5.1. Conditions Explaining High Intention to Use Self-Driving Vehicles in Ghana

The second configuration proved to be the optimal pathway in explaining a high intention to use self-driving vehicles and addresses the second research question raised. This configuration indicates that individuals are more likely to adopt SDVs when they perceive the technology as both reliable and easy to operate. Additionally, users’ confidence in the technology plays a crucial role in their willingness to embrace it. This finding aligns well with previous research on autonomous vehicles, which underscores the importance of perceived reliability and usability in establishing trust and fostering a readiness to adopt new technologies [13,45]. Furthermore, the emphasis on technological competence resonates with research conducted by [74], which elucidates the importance of users’ confidence in advanced systems. The high levels of raw coverage and consistency associated with this configuration illustrate that a combination of trust in the technology’s reliability, its intuitive design, and the user’s confidence in its capability provides a comprehensive explanation for a strong intention to use SDVs.

The first configuration highlights the influence of perceived usefulness and reliability in driving adoption, even when users may lack confidence in the technology. This finding supports the notion that users’ trust and the perceived benefits derived from the technology can effectively compensate for their insufficient confidence in the technology. Previous studies have demonstrated similar trends, indicating that perceived usefulness and trust often serve as stronger predictors of behavioral intention compared to confidence in the technology, especially in developing environments with limited technological exposure [13,75]. This configuration emphasizes the necessity of promoting the functional advantages and societal benefits of SDVs—such as enhanced safety, increased efficiency, and greater convenience—to mitigate challenges associated with confidence in the technology.

The sixth configuration underscores the pivotal role of perceived risk in shaping users’ intentions to adopt SDVs. When users consider these vehicles to be both useful and reliable while simultaneously having minimal concerns regarding their safety, their likelihood of adopting the technology significantly increases. This observation is consistent with existing research that identifies perceived risk as a formidable barrier to the adoption of SDVs [45,76]. For example, ref. [77] established that trust in the safety and reliability of autonomous vehicles is a major influencer of adoption intentions, emphasizing the need to address risk-related concerns in order to cultivate public confidence in the technology.

Configuration three provides an intriguing perspective by illustrating how technological competence can serve as a compensatory factor in the absence of perceived reliability and ease of use, provided that perceived risks are kept to a minimum. This pathway indicates that users with a higher level of technological confidence may be willing to tolerate lower standards of usability and reliability as long as they perceive the vehicles to be safe. This finding aligns with earlier research by [74], which posits that users with a greater perception of a technology’s competence are often more open to experimenting with it, especially when safety concerns are adequately addressed.

Similarly, the fourth configuration highlights the relationship between perceived reliability and technological competence in overcoming perceived risks. Even when individuals perceive low usefulness in self-driving vehicles, those who trust the reliability of these vehicles and the capabilities of the technology may still demonstrate a high intention to adopt them. This illustrates the complex dynamics of the factors influencing users’ adoption decisions and aligns with findings from [74], which indicate that users with greater confidence and perception of the capability of a technology are more likely to adopt SDVs despite perceived risks.

Additionally, the final configuration emphasizes the compensatory role of perceived usefulness, safety, and technological competence in addressing usability challenges. Users who view SDVs as beneficial, safe, and capable may choose to adopt them, even if they find the technology less intuitive or user-friendly. This idea is supported by conclusions drawn by [76], suggesting that users often overlook usability issues when they recognize clear benefits and feel confident in their ability to manage potential challenges related to the technology.

5.2. Conditions Explaining Not-High (Low) Intention to Use Self-Driving Vehicles in Ghana

Both configurations demonstrate very high consistency, indicating strong causal relationships. However, Configuration One has slightly higher raw and unique coverage, making it the more optimal choice for addressing the research question. This configuration explains a larger portion of the outcome, including some aspects that Configuration Two does not account for. The first configuration emphasizes that when potential users view SDVs as unreliable, find them difficult to operate, and doubt their capabilities, their likelihood of adopting such vehicles significantly decreases. The situation worsens when there are serious concerns related to safety and other risks. This trend aligns with previous research on the adoption of autonomous vehicles, which consistently indicates that a lack of trust in the reliability and safety of emerging technologies is a major barrier to user acceptance [45,74,75,78]. Additionally, perceived risk plays a crucial role in this configuration, highlighting how safety concerns can amplify the negative effects of missing enabling factors.

In the second configuration, a similar trend is observed. This configuration indicates that when users do not recognize any tangible benefits associated with using SDVs—meaning they view these vehicles as currently unhelpful—coupled with their lack of confidence in the technology and ongoing safety concerns, their intentions to embrace such innovations are significantly diminished. This finding aligns well with established research indicating that perceived usefulness is a fundamental element in Technology Acceptance Models [13,78], and its absence leads to a noticeable reduction in the intention to utilize new technologies. Furthermore, perceived risk has been consistently identified as a significant obstacle in various studies on the adoption of autonomous vehicles, mainly when safety concerns are inadequately addressed [76,79].

5.3. Theoretical Contributions

This study makes a significant theoretical contribution by utilizing the Technology Acceptance Model and incorporating fuzzy set qualitative comparative analysis to investigate the determinants of intention to use self-driving vehicles in Ghana. This research addresses a notable gap in the literature, as the adoption of autonomous vehicles in developing countries remains underexplored. By extending TAM with additional constructs such as perceived reliability, technological competence, and perceived risk, the study captures the intricate socio-technological and demographic dynamics unique to the Ghanaian context. These extensions enhance the broader technology adoption literature by illustrating how TAM-based constructs interact configurationally in developing markets, where infrastructural and technological diversity significantly shape behavioral intentions.

One of the primary contributions of this study lies in advancing the understanding of configurational causality through fsQCA. Unlike traditional regression-based approaches that isolate net effects, fsQCA identifies multiple causal pathways leading to high and low intentions to adopt SDVs. This approach captures equifinality, emphasizing that diverse combinations of conditions can lead to the same outcome [53,54]. For instance, while perceived usefulness and perceived reliability are key enablers of adoption, the study also demonstrates how technological competence and minimal perceived risk can compensate for other factors in fostering adoption. These findings underscore the configurational logic of TAM, offering deeper theoretical insights into the compensatory dynamics of factors driving SDV adoption in emerging markets.

Additionally, the research contributes to the literature on asymmetric causality, revealing that the determinants of high intention to adopt SDVs are not simply the inverse of those driving low intention. For instance, while perceived usefulness and perceived reliability strongly facilitate adoption, their absence does not directly lead to rejection. Instead, low technological competence and high perceived risk emerge as critical inhibitors of adoption. These findings highlight the asymmetric nature of causal relationships, illustrating the importance of examining both enablers and barriers to develop a comprehensive understanding of user behavior [53,57].

5.4. Practical Implications

This study offers several crucial practical implications for policymakers, technology developers, transportation planners, and businesses operating within the evolving landscape of autonomous vehicles, particularly in developing economies like Ghana. The configurational findings derived from the fsQCA analysis provide more granular and actionable insights than traditional variance-based approaches, allowing for targeted interventions tailored to specific user segments [53,61].

First, the study highlights the importance of fostering trust and confidence in SDV technology. The finding that configurations emphasizing perceived reliability and technological competence are strongly associated with higher adoption intentions suggests that efforts should be directed toward demonstrating the safety and dependability of these vehicles [45,74]. Public awareness campaigns, pilot programs in controlled environments, and transparent communication about the technology’s capabilities and limitations can be instrumental in building public trust. For instance, showcasing successful real-world applications of SDVs in similar contexts, addressing common safety concerns through rigorous testing and certification processes, and providing accessible educational resources on how the technology works can significantly enhance perceived reliability [76]. Furthermore, offering opportunities for potential users to experience SDV firsthand, perhaps through test drives or simulations, could bolster their technological competence and reduce anxieties surrounding their operation.

Second, the study underscores the need to emphasize the practical benefits and usefulness of SDVs. The configurations highlighting perceived usefulness as a key driver of adoption suggest that highlighting the tangible advantages of these vehicles is crucial [13,75]. This could involve focusing on potential benefits, such as reduced traffic congestion, improved accessibility for elderly or disabled individuals, increased productivity during commutes, and decreased transportation costs. In the Ghanaian context, where traffic congestion is a significant problem in urban areas, emphasizing the potential of SDVs to alleviate this issue could be particularly effective. Moreover, demonstrating how SDVs can improve access to essential services and opportunities for underserved communities could further enhance their perceived usefulness and promote wider acceptance.

Third, the study’s findings regarding perceived risk offer valuable guidance for mitigating public concerns. The configurations indicating that high perceived risk acts as a strong deterrent to adoption suggest that addressing safety and security concerns is paramount [76,79]. This could involve implementing robust cybersecurity measures to protect against hacking and data breaches, establishing clear regulatory frameworks for SDV operation and liability, and communicating transparently about safety protocols and accident prevention mechanisms. Public forums and engagement with community leaders can also help address specific concerns and build confidence in the safety of SDVs.

Fourth, the study’s focus on the interplay of multiple factors offers crucial insights for targeted interventions. The identification of distinct configurations leading to both high and low adoption intentions allows for tailored strategies aimed at specific user segments [53]. For example, individuals with a high perception of the technology’s competence might be more receptive to early adoption if the focus is on showcasing the advanced features and capabilities of SDVs, even if concerns about ease of use exist [74]. Conversely, for individuals with a lower perception of the technology’s competence, prioritizing ease of use and providing comprehensive training and support could be more effective. This segmentation approach allows for a more efficient allocation of resources and maximizes the impact of intervention strategies.

Finally, the study’s focus on a developing economy context offers important lessons for other similar markets. The unique infrastructural challenges, socio-cultural dynamics, and varying levels of technological exposure in Ghana highlight the need for context-specific approaches to SDV adoption. Policymakers and technology developers should avoid simply replicating strategies from developed countries and instead focus on understanding the specific needs and concerns of local populations. This could involve collaborating with local communities, conducting thorough assessments of existing infrastructure, and developing tailored solutions that address local challenges. By considering the unique context of developing economies, stakeholders can ensure that the introduction of SDVs is not only technologically feasible but also socially acceptable and beneficial.

5.5. Future Studies

While the study offers valuable insights into SDV adoption in Ghana, several limitations must be acknowledged. First, the reliance on self-reported survey data introduces the potential for social desirability bias, where respondents may overstate positive intentions. Future research should incorporate behavioral validation methods, such as field trials, virtual reality simulations, or app-based behavioral tracking, to complement self-reported intentions and enhance generalizability. While our rigorous survey design ensured high response quality, integrating behavioral measures would provide a more comprehensive understanding of actual adoption behavior, particularly in regions with limited exposure to advanced mobility systems. Second, although the use of fsQCA enables configurational insights, the cross-sectional nature of the data limits the ability to capture dynamic changes in perceptions and intentions over time. Longitudinal studies would provide a better understanding of how user attitudes evolve as awareness and exposure to SDV technology increase.

Additionally, the study’s focus on Ghana, while important, limits generalizability to other regions. Comparative studies across different developing countries, especially those with varying levels of infrastructure, digital literacy, and traffic regulation, could reveal broader or divergent adoption patterns. Finally, this study primarily examines demand-side factors (user perceptions and intentions). Future studies should also consider supply-side variables such as government policy, infrastructure readiness, industry engagement, and economic incentives, which are crucial in shaping the SDV adoption ecosystem in developing economies.

6. Conclusions

This study has employed fuzzy set qualitative comparative analysis to investigate the determinants of intention to use self-driving vehicles in Ghana, focusing on how different configurations of factors influence adoption decisions. The findings revealed multiple pathways to high adoption intention, with the most effective configuration combining perceived reliability, perceived ease of use, and technological competence. For low adoption intention, two primary configurations emerged, both emphasizing how the combination of low technological competence and high perceived risk significantly impedes adoption. These results highlight the complex, asymmetric nature of SDV adoption in developing economies and demonstrate that multiple-factor combinations can lead to similar outcomes.