The Influence of Digital Capabilities on Elderly Pedestrians’ Road-Sharing Acceptance with Autonomous Vehicles: A Case Study of Wuhan, China

Abstract

While autonomous vehicles (AVs) are increasingly integrated into urban mobility, little is known about how digital capability shapes elderly pedestrians’ willingness to share roads with these technologies. This is especially true in the absence of explicit vehicle–pedestrian communication mechanisms. To address this gap, we combine the Theory of Planned Behavior (TPB) with the Pedestrian Behavior Questionnaire (PBQ) and segment elderly pedestrians using Latent Class Analysis (LCA). A sample of 750 older adults in Wuhan, China, was divided into two latent groups: digitally disengaged (70.8%) and digitally engaged (29.2%). Classification was based on four indicators: smart device usage, online social interaction, online entertainment, and online economic behavior. We then applied ordered logit models to estimate group-specific determinants of AV road-sharing acceptance. Results reveal clear heterogeneity across digital capability levels. For digitally disengaged seniors, positive pedestrian behaviors significantly increased willingness (β = 0.316, p = 0.001). Prior accident experience reduced willingness (0 accident: β = 0.435, p = 0.021; 1–2 accidents: β = −0.518, p = 0.012). For digitally engaged seniors, perceived behavioral control showed a marginally positive effect (β = 0.353, p = 0.066). Errors had a significant positive effect (β = 0.540, p = 0.009). Positive behaviors had a significant negative effect (β = −0.414, p = 0.007). These patterns indicate that digital capability not only modulates the strength of TPB pathways but also reshapes behavior–intention linkages captured by PBQ dimensions. Methodologically, the study contributes an integrated TPB–PBQ–LCA–OLM framework. This framework identifies digital capability as a critical moderator of AV acceptance among elderly pedestrians. Practically, the findings suggest differentiated strategies. For digitally disengaged users, interventions should build digital literacy and reinforce safe walking norms. For digitally engaged users, strategies should prioritize transparent AV intent signaling and features that enhance perceived control.

1. Introduction

As autonomous vehicle (AV) technology advances, the global transportation system is experiencing profound changes. Wuhan has rapidly positioned itself as a leader in the autonomous vehicle industry, particularly in robotaxi deployment. As of 2023, the city operates nearly 500 autonomous vehicles across various districts, with some offering fully driverless services, including to Wuhan Tianhe International Airport. The city also boasts over 3300 km of open test roads, making it one of the largest autonomous driving operation zones globally. Wuhan has served over 900,000 passengers through autonomous travel services, and major companies like Baidu have expanded their robotaxi fleets from 5 to 300 vehicles in just one year, with plans for further expansion [1].

The adoption of AVs is expected to improve traffic efficiency, reduce accidents, and enhance overall travel experiences by minimizing human driving errors [2,3]. Nevertheless, AV deployment also raises new challenges, particularly in pedestrian–vehicle interactions, where issues of trust, behavioral prediction, and decision-making critically shape safety perceptions [4]. Whether pedestrians can accurately anticipate AV maneuvers, trust their safety, and ultimately accept road-sharing with AVs has become a central research question.

Recent scholarship emphasizes that transportation safety must be understood from an interdisciplinary perspective. For instance, Qiu et al. (2025) demonstrated how non-structural components significantly affect over-track building vibrations caused by train operations, underscoring the role of civil and structural engineering in assessing the broader impacts of transportation systems [5]. Similarly, Wu et al. (2025) showed that artificial intelligence and AIGC-driven data augmentation can enhance bolt object detection in automated construction inspection, highlighting how AI is reshaping infrastructure monitoring and safety management [6]. Together, these examples illustrate that transportation research extends beyond human–vehicle interactions to encompass structural resilience and AI-enabled safety innovations. Positioning our study within this broader interdisciplinary context, we focus on the human dimension of AV acceptance—specifically, how elderly pedestrians’ digital capabilities influence their willingness to share roads with AVs.

Elderly pedestrians represent a particularly vulnerable group in traffic environments. Compared to younger individuals, they may face limitations in vision, hearing, reaction speed, and technology acceptance, which compromise their ability to anticipate AV behavior and increase perceived risks [7,8,9,10]. In this study, the term “elderly population” refers to individuals aged 60 years and above, following the definition used by the National Bureau of Statistics of China and aligned with the World Health Organization’s classification in developing country contexts. This operational definition ensures clarity in interpreting the sample and consistency with prior research on aging and mobility.

In addition to age-related physiological and perceptual limitations, existing studies highlight several competing factors that influence AV acceptance across populations, such as perceived risk [11], trust in AV systems [12], social influence [13], and perceived usefulness or ease of use [14]. These factors are particularly relevant to elderly individuals, who may display heightened risk aversion, limited technological self-efficacy, or normative resistance to automation. Hence, a multidimensional approach is necessary to capture how these mechanisms interact with digital capabilities in shaping AV acceptance.

In this context, exploring the acceptance of AVs by elderly pedestrians is of great practical significance. In this study, we define digital capability as an individual’s ability to access, use, and integrate digital tools and services into daily life, which includes device ownership, digital skills, online interaction, and confidence in technology use [15,16]. As society shifts toward digitalization, disparities in digital capabilities among the elderly may shape their attitudes, subjective norms, and perceived behavioral control regarding AV technology [15]. However, existing studies on pedestrian acceptance of AVs have paid limited attention to the impact of digital capabilities differences within the elderly population, making this an important research gap.

This study examines how elderly pedestrians with varying levels of digital capabilities differ in their acceptance of AV road-sharing through Latent Class Analysis (LCA) and an ordered logit model. Four key variables are used to measure the digital capabilities of the elderly: ownership and use of smart devices (Dc1) [17], online social interaction (Dc2) [18], online entertainment activities (Dc3) [19], and online economic behavior (Dc4) [20,21]. Based on these variables, elderly pedestrians are categorized into two groups: digitally disengaged and digitally engaged. Utilizing the Theory of Planned Behavior (TPB) and the Pedestrian Behavior Questionnaire (PBQ), the study incorporates variables such as attitude, subjective norms, perceived behavioral control, behavioral intention, violations, errors, lapses, aggressive behavior, and positive behavior to analyze how digital capabilities influences elderly pedestrians’ road-sharing acceptance of AVs.

The contributions of this study are fourfold: First, it uniquely integrates the concept of digital capabilities with elderly pedestrians’ acceptance of autonomous vehicles (AVs), using Latent Class Analysis (LCA) to identify distinct subgroups based on digital capabilities, offering new insights into how digital capabilities influences technology adoption. Second, by incorporating variables from both the Theory of Planned Behavior (TPB) and the Pedestrian Behavior Questionnaire (PBQ), it provides a comprehensive model for understanding elderly pedestrians’ behavior, showing how digital capabilities shape their attitudes toward AVs. Third, the study reveals contrasting patterns of digital capabilities, with low-engagement groups driven by traditional factors like attitude and positive behaviors, while high-engagement groups are more influenced by perceived behavioral control and exhibit a more cautious approach to AVs. Finally, the findings offer practical insights for policymakers, suggesting that promoting digital capabilities and addressing the concerns of digitally savvy elderly individuals can enhance AV adoption, bridging the digital divide and ensuring equitable access to intelligent transportation technologies.

The remainder of this paper is organized as follows. Section 2 reviews the relevant literature on autonomous vehicles, pedestrian behavior, and digital capability, highlighting theoretical and empirical gaps. Section 3 introduces the theoretical framework and hypotheses, integrating the Theory of Planned Behavior (TPB) with the Pedestrian Behavior Questionnaire (PBQ). Section 4 presents the research design, including data collection and measurement strategy. Section 5 reports the results of the Latent Class Analysis (LCA) and ordered logit models. Section 6 discusses the key findings in relation to prior studies and emphasizes their theoretical and practical implications. Finally, Section 7 concludes the study by summarizing the contributions, acknowledging limitations, and outlining directions for future research.

2. Literature Review

2.1. Current Research on Autonomous Vehicles and Pedestrian Behavior

With the increasing deployment of autonomous vehicles (AVs) in mixed-traffic environments, understanding how pedestrians interpret AV behavior—especially in the absence of direct communication cues—has become a critical issue in transportation safety research. Unlike human-driven vehicles, AVs lack traditional social signals such as eye contact or hand gestures that pedestrians have historically relied on to assess vehicle intentions [22]. This absence introduces uncertainty in pedestrian decision-making, particularly at unsignalized crossings, leading to altered crossing behaviors and hesitation [23].

Pedestrians develop trust and make crossing decisions based not only on physical movement cues (e.g., speed, deceleration patterns) but also on their cognitive models of how AVs operate. Studies have shown that pedestrians adjust their behavior according to whether they believe a vehicle is manually or autonomously controlled [24], suggesting that the perceived predictability and reliability of AVs are crucial for establishing safe interactions. However, such predictability is not universally perceived—while some pedestrians assume AVs are programmed to yield, others express uncertainty due to the lack of interactive feedback [25].

These uncertainties are particularly problematic for elderly pedestrians. Age-related declines in sensory processing and decision-making speed [26] can impair the ability to accurately interpret AV kinematics and intentions. Furthermore, older adults tend to be more risk-averse and report lower confidence in navigating unfamiliar traffic technologies [27]. As a result, they may experience heightened stress or disengagement in AV-dominated environments, even in the absence of overt danger. Their behavioral adaptations—such as waiting longer to cross or avoiding AV lanes—reflect a coping strategy to mitigate perceived risk [28].

Despite the growing attention to AV–pedestrian dynamics, most existing studies focus on controlled or highly communicative settings. However, many real-world deployments of AVs still operate in environments where communication between vehicles and pedestrians relies entirely on kinematic and contextual cues. Therefore, it is essential to investigate pedestrian behavior in such non-communicative scenarios, where pedestrians must infer vehicle intent solely from motion and environmental context. This line of inquiry is especially relevant for older adults, who may struggle to adapt their behavior when facing vehicles that offer no social or interactive feedback.

2.2. Research on Digital Capability, Technology Acceptance, and the Elderly Population

Digital capability plays a pivotal role in shaping how elderly individuals understand, evaluate, and accept autonomous vehicle (AV) technologies. As the aging population becomes increasingly exposed to digital systems and smart mobility solutions, the disparity in their ability to engage with these technologies—commonly referred to as the “digital divide”—emerges as a significant barrier to AV acceptance.

Digital capability is broadly defined as the proficiency in accessing, understanding, and using digital technologies such as smartphones, online services, and interactive interfaces [29]. However, age-related declines in cognitive function, perceptual acuity, and adaptability often limit older adults’ ability to acquire such competencies [30]. As a result, elderly users may perceive AVs as too complex or intimidating, thus lowering their willingness to accept and adopt such technologies.

The Technology Acceptance Model (TAM) provides a useful framework for interpreting this phenomenon [31]. TAM posits that perceived usefulness (PU) and perceived ease of use (PEU) are key determinants of an individual’s intention to adopt new technologies. In the case of AVs, studies have shown that elderly individuals with higher digital capability report greater PU and PEU, which in turn enhances their intention to accept AVs [32]. Conversely, low digital capabilities may exacerbate feelings of anxiety or loss of control, undermining behavioral intentions.

Beyond individual cognition, social influences also play an essential role. Older adults’ decisions are frequently shaped by the behaviors and endorsements of peers, caregivers, or family members [33]. Observational learning—such as seeing peers successfully using digital devices—can significantly improve confidence and perceived self-efficacy, thereby promoting AV acceptance.

Furthermore, research indicates that digital capabilities is positively associated with technology trust, mobility satisfaction, and perceived security in technology-mediated transport systems [34]. Such factors are particularly critical in the context of AVs, which demand a high degree of trust in machine decision-making and algorithmic control.

To foster greater inclusivity in future mobility systems, public policy and AV interface design must account for the diversity in digital capabilities among older adults. Educational interventions, simplified interfaces, and age-sensitive training programs can help bridge this divide. Increasing digital capabilities in this demographic will not only improve AV acceptance but also promote greater equity in access to smart transportation services.

In summary, digital capability is a key enabler of AV technology acceptance among older adults. Higher proficiency in using digital tools enhances PU, PEU, and overall trust, while limited skills contribute to resistance and anxiety. Understanding the interplay between digital capabilities and psychological readiness is essential for accelerating AV adoption in aging societies.

2.3. Research and Applications of the Theory of Planned Behavior (TPB) and the Pedestrian Behavior Questionnaire (PBQ)

The Theory of Planned Behavior (TPB), introduced by Ajzen (1991), remains a foundational model for explaining intention-based human behaviors through three constructs: attitude, subjective norm (SN), and perceived behavioral control (PBC) [35]. In pedestrian research, TPB has been validated for predicting crossing behaviors, risk-taking decisions, and responses to traffic stimuli in emerging mobility contexts [36,37]. In the field of autonomous vehicle (AV) interaction, TPB has been applied to understand pedestrians’ willingness to share roads, with studies showing that attitudes toward AV safety and confidence in predicting AV motion (PBC) significantly shape intention [38,39].

Older pedestrians represent a vulnerable subgroup whose AV-related behaviors are influenced by both TPB constructs and age-related cognitive limitations such as slower risk assessment and reduced reaction time [40,41]. However, TPB alone cannot fully capture the behavioral tendencies that emerge in real-world traffic settings.

The Pedestrian Behavior Questionnaire (PBQ) complements TPB by classifying behaviors into violations, errors, lapses, aggression, and positive acts, thereby identifying risk profiles that may diverge from self-reported intentions [42,43,44]. This is particularly relevant for elderly pedestrians, whose lapses or errors may undermine perceived control and exacerbate risk perception.

Integrating TPB and PBQ is therefore not a simple combination but a necessary framework: TPB provides the cognitive-intentional basis of acceptance, while PBQ reveals behavioral patterns that reflect actual road-use practices. Their joint use enables the detection of intention–behavior gaps, segmentation of elderly groups by risk profiles, and the design of targeted interventions.

In summary, this integration is theoretically justified and practically essential for understanding elderly pedestrians’ AV acceptance, as it bridges the gap between psychological readiness and observable behavior.

3. Theoretical Framework and Hypotheses

3.1. Latent Class Analysis (LCA) and Digital Capability Grouping

Latent Class Analysis (LCA) is a statistical method that uncovers unobserved heterogeneity within a population by classifying individuals into mutually exclusive subgroups based on observed indicators. In behavioral and transportation research, LCA is particularly effective for revealing latent behavioral patterns that may be masked by aggregate analyses, especially among heterogeneous populations such as the elderly [45].

Digital capability—defined as the ability to effectively use digital technologies for communication, information access, and online engagement—has become a critical factor influencing technology acceptance and behavioral adaptation, particularly in aging populations [46]. Prior studies indicate that older adults with higher digital proficiency tend to report greater openness toward autonomous systems and demonstrate higher levels of trust and perceived control when interacting with emerging mobility technologies [41].

In this study, LCA is employed to classify elderly individuals into distinct digital capability groups using four behavioral indicators: (Dc1) Smart Device Usage, (Dc2) Online Social Interaction, (Dc3) Online Entertainment, and (Dc4) Online Economic Behavior. These dimensions reflect both operational fluency and depth of digital capabilities. For example, frequent use of smartphones or tablets (Dc1) and participation in social networking or video calls (Dc2) are associated with greater comfort in technology-mediated environments (Harada et al., 2013; Chen & Schulz, 2016) [17,18]. Similarly, engaging in digital content consumption (Dc3) and online financial activities (Dc4) indicate higher competence in navigating digital infrastructures [47,48].

By segmenting the elderly population according to these indicators, the LCA facilitates a more nuanced understanding of how digital capabilities moderates attitudes and intentions toward road-sharing with autonomous vehicles (AVs). This classification forms the empirical basis for subsequent multi-group comparisons within the theoretical framework.

3.2. Theory of Planned Behavior (TPB)

The Theory of Planned Behavior (TPB), originally proposed by Ajzen (1991) [35], posits that an individual’s behavioral intention is determined by three key psychological antecedents: attitude toward the behavior, subjective norms (SN), and perceived behavioral control (PBC). TPB has been widely applied in mobility and technology acceptance research and provides a theoretically grounded lens to examine elderly pedestrians’ willingness to engage in shared road environments with autonomous vehicles (AVs) [49,50].

However, increasing evidence suggests that the strength and direction of TPB pathways are moderated by individual-level characteristics—most notably, digital capability. High levels of digital capabilities are associated with greater confidence in technological environments, which can enhance both perceived control and the influence of attitude on behavioral intention [41,51]. Conversely, individuals with lower digital capabilities may rely more heavily on social influence cues and exhibit greater behavioral uncertainty [52]. In particular, older adults with limited exposure to digital interfaces may place disproportionate weight on subjective norms—mirroring heuristics found in low-familiarity technology contexts [53].

For conceptual clarity, in this study the terms “digitally engaged” and “digitally disengaged” are preliminarily defined according to four observed digital behavior indicators: (Dc1) Smart Device Usage, (Dc2) Online Social Interaction, (Dc3) Online Entertainment, and (Dc4) Online Economic Behavior. Elderly pedestrians with consistently high participation across these indicators are considered digitally engaged, while those with minimal participation are considered digitally disengaged. This conceptual classification is later validated through Latent Class Analysis (LCA).

To capture these differential cognitive patterns, this study employs a stratified hypothesis design based on latent digital capabilities classes (high vs. low), as identified via Latent Class Analysis (LCA). This approach allows for an empirically driven comparison of TPB pathways across groups, enhancing both the theoretical precision and practical relevance of AV acceptance research among aging populations.

(1)

Digitally Engaged Group

H1a. 

A positive attitude toward AV road-sharing will significantly and positively influence behavioral intention.

H2a. 

Subjective norms will positively predict behavioral intention, albeit with weaker effect strength than attitude.

H3a. 

Higher perceived behavioral control will be positively associated with behavioral intention.

H4a. 

Behavioral intention will strongly predict actual willingness to share roads with AVs.

(2)

Digitally Disengaged Group

H1b. 

The effect of attitude on behavioral intention will be weaker but still statistically significant, reflecting reduced internal confidence.

H2b. 

Subjective norms will be the dominant predictor of behavioral intention, due to heightened reliance on external guidance.

H3b. 

Perceived behavioral control will exert a limited but positive effect, moderated by digital anxiety.

H4b. 

Behavioral intention will remain a predictor of willingness to share roads, though its explanatory power may be attenuated.

In summary, the latent class analysis (LCA) conducted in Section 3.1 provides the structural basis for the stratification of TPB pathways described in this section. By dividing elderly pedestrians into digitally engaged and digitally disengaged groups, LCA enables us to empirically test whether the cognitive mechanisms proposed by TPB—attitude, subjective norm, and perceived behavioral control—operate differently across heterogeneous digital capability levels. Thus, Section 3.1 does not stand alone as a methodological step but rather directly informs the theoretical hypotheses in Section 3.2, ensuring a cohesive and logically integrated framework for subsequent empirical analysis.

3.3. Pedestrian Behavior Questionnaire (PBQ) and Hypotheses by Digital Capabilities Level

The Pedestrian Behavior Questionnaire (PBQ) is a validated tool that assesses five foundational dimensions of pedestrian behavior—violations, errors, lapses/distractions, aggression, and positive behaviors—which are closely linked to safety outcomes [54]. These dimensions are particularly germane when evaluating elderly pedestrians’ interactions with autonomous vehicles (AVs), as each behavior uniquely influences crossing decisions and risk exposure [55].

Emerging research indicates that digital capability moderates the influence of PBQ dimensions on AV acceptance. Specifically, older adults with low digital proficiency tend to exhibit more frequent lapses and errors, attributed to heightened cognitive load and reduced situational awareness in technologically complex environments. In contrast, digitally adept seniors are more likely to adopt safety-enhancing behaviors, driven by their comfort with and understanding of digital-automated systems [56].

Based on these insights, we propose the following stratified hypotheses:

Digitally Engaged Group

H5a. 

Higher violation rates will significantly reduce AV road-sharing willingness.

H6a. 

More errors will decrease willingness.

H7a. 

Increased distractions negatively impact road-sharing intention.

H8a. 

Aggressive behaviors will be negatively associated with AV acceptance.

H9a. 

Positive pedestrian behaviors will increase both willingness and perceived safety.

Digitally Disengaged Group

H5b. 

Violations will strongly reduce willingness, reflecting low rule awareness.

H6b. 

Errors will have an amplified negative effect due to limited error management capacity.

H7b. 

Distractions will significantly impair AV engagement.

H8b. 

Aggressive behavior will contribute to heightened mistrust of AVs.

H9b. 

Positive behaviors will modestly improve willingness, but the effect will be attenuated by lower digital confidence.

4. Methodology

4.1. Research Design

This study adopts a quantitative research design to examine the impact of digital capability on elderly pedestrians’ willingness to share roads with autonomous vehicles (AVs). The study employs Latent Class Analysis (LCA) to categorize elderly pedestrians into distinct groups based on their digital capability and utilizes an ordered logit model to explore how each group differs in its road-sharing acceptance of AVs. The Theory of Planned Behavior (TPB) and the Pedestrian Behavior Questionnaire (PBQ) provide the theoretical framework, which incorporates key variables such as attitude, subjective norms, perceived behavioral control, behavioral intention, and behavioral dimensions such as violations, errors, lapses, aggressive behavior, and positive behavior. Given that the dependent variable—elderly pedestrians’ willingness to share roads with AVs—was measured on a five-point Likert scale (1 = strongly disagree to 5 = strongly agree), this outcome is ordinal in nature. Therefore, the Ordered Logit Model (OLM) was selected as the most appropriate analytical technique. OLM accounts for the ordinal structure of the data without assuming equal distances between categories, unlike linear regression, and is widely applied in behavioral and transportation research where Likert-type scales are employed. This methodological choice ensures both statistical rigor and interpretability in analyzing heterogeneous behavioral responses.

4.2. Data Collection and Measurement

The data for this study were collected through a structured questionnaire consisting of three major sections:

(1)

Section 1: Socioeconomic Attributes

This section collects basic demographic and socioeconomic information, such as gender, age, education level, household income, employment status, car ownership, and past experience with pedestrian traffic accidents. These variables serve as control variables to account for potential demographic influences on road-sharing acceptance.

(2)

Section 2: Latent Variables

The second section measures latent variables derived from the TPB and PBQ models. Each variable, including attitude (ATT), subjective norms (SN), perceived behavioral control (PBC), behavioral intention (BIU), violations (VIO), errors (ERR), lapses (LAP), aggressive behavior (AGG), and positive behavior (PB), was measured using a 5-point Likert scale, with specific measurement items detailed in

Table 1. Latent Variables and Corresponding Measurement Items.

Latent Variables	Code	Variable Measurement Content
Violations	VIO1	I cross the street against a red light.
	VIO2	I cross diagonally at intersections to save time.
	VIO3	Even if a crosswalk is less than 50 m away, I cross outside the crosswalk.
	VIO4	I walk in prohibited areas to save time.
Errors	ERR1	I cross between parked cars during traffic jams.
	ERR2	I cross even when cars are coming because I think they will stop for me.
	ERR3	I walk on bike paths instead of sidewalks.
	ERR4	I cross the street without looking to save time.
Lapses	LAP1	I crossed several streets and intersections without observing traffic.
	LAP2	I crossed the street without observing traffic because I was distracted.
	LAP3	I crossed the street while talking to someone and didn’t observe traffic.
	LAP4	I crossed the street without observing traffic because I wanted to meet someone on the other side.
Aggressive Behaviors	AGG1	I yelled at another road user (pedestrian, driver, cyclist, etc.) out of anger.
	AGG2	I crossed the street very slowly to annoy drivers.
	AGG3	I made gestures at another road user in anger.
	AGG4	I hit a car in anger.
Positive Behaviors	PB1	I thank drivers who stop to let me cross the street.
	PB2	When walking with others, I walk in a single file to avoid disturbing other pedestrians.
	PB3	I walk on the right side of the sidewalk to avoid disturbing others.
	PB4	Even if I have the right of way, I let a car pass if there are no other vehicles behind me.
Attitude	ATT1	For me, walking on roads with autonomous vehicles is enjoyable.
	ATT2	For me, walking on roads with autonomous vehicles is positive.
	ATT3	For me, walking on roads with autonomous vehicles is desirable.
Social Norms (SN)	SN1	People important to me want me to walk on roads with autonomous vehicles in the future.
	SN2	People important to me (e.g., family and friends) support me walking on roads with AVs.
	SN3	If people around me walk on roads with autonomous vehicles, I would walk on such roads too.
Perceived Behavioral Control (PBC)	PBC1	Walking on roads with autonomous vehicles depends entirely on me.
	PBC2	For me, walking on roads with autonomous vehicles is very easy.
	PBC3	I have enough opportunities to walk on roads with autonomous vehicles.
Behavioral Intention	BIU1	I intend to walk on roads with autonomous vehicles.
	BIU2	I will walk on roads with autonomous vehicles.
	BIU3	I plan to recommend my family and friends to walk on roads with autonomous vehicles.

below. The questions were carefully designed to capture the degree of agreement or disagreement with each statement.

(3)

Section 3: Willingness to Share Roads

The final section assesses participants’ willingness to share the road with AVs using a 5-point Likert scale ranging from “strongly disagree” to “strongly agree.”

A two-step analytical approach was implemented to analyze the data:

(1) Latent Class Analysis (LCA):

LCA was employed to classify elderly pedestrians into latent groups based on the four digital capability variables: Dc1 (Smart Device Usage), Dc2 (Online Social Interaction), Dc3 (Online Entertainment), and Dc4 (Online Economic Behavior).

(2) Ordered Logit Model:

The ordered logit model was used to investigate how different groups (based on digital capability) vary in their willingness to share the road with AVs. The model incorporated socio-economic variables, latent variables from the TPB (ATT, SN, PBC, BIU), and pedestrian behaviors from the PBQ (VIO, ERR, LAP, AGG, PB).

4.3. Sampling Strategy and Data Collection

The sampling strategy for this study follows the guidelines for Latent Class Analysis (LCA) and Structural Equation Modeling (SEM). According to the recommendations of Hair et al. (2010), the sample size for SEM should be at least 5 to 10 times the number of estimated parameters [57]. Based on this criterion, the estimated minimum sample size ranges between 160 and 320 participants.

However, due to the complexity of the model and the inclusion of multiple latent variables, a more conservative approach was taken, aiming for a sample size of at least 500 participants. This allows for more robust parameter estimates and model fit.

The data were collected between 30 June 2024, and 30 July 2024, in Wuhan, Hubei Province, China. Wuhan was chosen due to its active engagement with autonomous vehicle services, with over 1.98 million individuals having experienced AVs either directly or through media exposure. Given the city’s advanced adoption of AV technology, it was an ideal location to study the perceptions of elderly pedestrians.

To ensure a diverse and representative sample, the study used both online (via Wenjuanxing) and offline (through in-person interviews) data collection methods. It should be noted that the choice of survey mode (online vs. offline) was employed solely to broaden participant accessibility and does not determine digital capability classification. The distinction between high and low digital capability groups was derived exclusively through Latent Class Analysis (LCA) of four observed behavioral indicators (Dc1–Dc4), independent of the data collection channel. The inclusion criteria required participants to be 60 years or older and to have at least heard of autonomous vehicles. The inclusion criteria required that participants be aged 60 years or older and possess at least basic awareness of autonomous vehicles (AVs). A total of 788 elderly individuals completed the survey. After applying data quality screening, 750 valid responses were retained for final analysis. Excluded responses (n = 38) were removed due to missing or incomplete data, repetitive or invariant answer patterns (e.g., selecting the same option throughout), and excessively short completion times (less than five minutes), which indicated low engagement or careless responding. These quality control procedures were implemented to ensure the integrity and reliability of the dataset. This study was approved by the ethics committee of Wuhan Polytechnic University (grant NO: BME-2024-1-28) and conducted in accordance with the principles of the Declaration of Helsinki. All participants provided informed consent before taking part in the study. Clinical trial number: not applicable.

4.4. Descriptive Statistics

The final sample consisted of 750 elderly individuals, with a nearly equal distribution of gender (48.8% male and 51.2% female). The age distribution was primarily concentrated between 60 and 75 years, with 41.33% aged 60–65, 28.93% aged 66–70, and 19.33% aged 71–75. The majority of respondents had low educational attainment (57.07% completed elementary school or less), and most were retired (83.73%). The respondents also represented a lower-income bracket, with 66.93% earning 3000 CNY or less per month. 72.27% of the sample owned a private car, and 66.93% had no traffic accident experience within the last two years. These statistics reflect a relatively low-income, retired, and educationally disadvantaged group of elderly individuals.

4.5. Reliability and Validity Analysis of Latent Variables

This study evaluates several latent variables related to pedestrian behavior, including Violations, Errors, Lapses, Aggressive Behaviors, Positive Behaviors, Attitude, Social Norms (SN), Perceived Behavioral Control (PBC), and Behavioral Intention. The reliability and validity of these constructs were assessed using Cronbach’s alpha (α), composite reliability (CR), and average variance extracted (AVE) [58]. The detailed reliability and validity results are summarized in

Table 2. Reliability and Validity Analysis of Latent Variables.

Latent Variables	Code	Means	SD	Standardized Factor Loading	Cronbach’s α	CR	AVE
Violations	VIO1	2.271	1.130	0.781	0.8509	0.837	0.562
	VIO2	2.212	1.116	0.747
	VIO3	2.363	1.104	0.763
	VIO4	2.296	1.124	0.707
Errors	ERR1	2.089	1.094	0.766	0.8615	0.848	0.583
	ERR2	2.067	1.067	0.762
	ERR3	2.165	1.092	0.758
	ERR4	2.241	1.156	0.767
Lapses	LAP1	2.045	1.044	0.766	0.8616	0.849	0.585
	LAP2	1.993	1.039	0.775
	LAP3	2.149	1.096	0.751
	LAP4	2.185	1.111	0.766
Aggressive Behaviors	AGG1	2.133	1.111	0.810	0.8604	0.848	0.583
	AGG2	2.032	1.085	0.776
	AGG3	2.128	1.087	0.702
	AGG4	2.248	1.146	0.763
Positive Behaviors	PB1	2.657	1.328	0.819	0.9068	0.898	0.688
	PB2	2.704	1.342	0.832
	PB3	2.767	1.323	0.830
	PB4	2.689	1.261	0.836
Attitude	ATT1	3.823	1.115	0.767	0.8308	0.802	0.574
	ATT2	3.827	1.139	0.781
	ATT3	3.728	1.127	0.724
Social Norms (SN)	SN1	3.767	1.112	0.763	0.8187	0.787	0.552
	SN2	3.804	1.105	0.730
	SN3	3.620	1.135	0.735
Perceived Behavioral Control (PBC)	PBC1	3.932	1.090	0.743	0.8146	0.782	0.544
	PBC2	3.893	1.076	0.731
	PBC3	3.745	1.117	0.739
Behavioral Intention	BIU1	3.644	1.169	0.783	0.8478	0.822	0.606
	BIU2	3.869	1.087	0.792
	BIU3	3.647	1.139	0.759

Notes: SD = Standard Deviation; CR = Composite Reliability; AVE = Average Variance Extracted; α = Cronbach’s Alpha.

.

4.5.1. Reliability Analysis

Cronbach’s alpha (α) was calculated to assess the internal consistency of the constructs. All latent variables exceeded the threshold of 0.7, demonstrating strong internal reliability [59]. Specifically, the α values for Violations (α = 0.8509), Errors (α = 0.8615), Lapses (α = 0.8616), Aggressive Behaviors (α = 0.8604), Positive Behaviors (α = 0.9068), Attitude (α = 0.8308), Social Norms (α = 0.8187), Perceived Behavioral Control (α = 0.8146), and Behavioral Intention (α = 0.8478) indicate satisfactory internal consistency across the items for each construct.

4.5.2. Validity Analysis

Composite reliability (CR) was calculated to further evaluate the reliability of the constructs, with all CR values exceeding the recommended threshold of 0.7 [58]. For example, the CR values for Positive Behaviors (CR = 0.898), Lapses (CR = 0.849), and Aggressive Behaviors (CR = 0.848) demonstrate robust construct reliability.

To assess convergent validity, the AVE for each construct was examined, with all AVE values surpassing the minimum threshold of 0.5, confirming that the majority of the variance is captured by the latent variables [60]. Constructs such as Positive Behaviors (AVE = 0.688), Lapses (AVE = 0.585), and Attitude (AVE = 0.574) demonstrate sufficient convergent validity.

4.5.3. Discriminant Validity

Discriminant validity was tested by comparing the square root of the AVE for each construct with its correlations with other constructs [61]. Discriminant validity is established if the square root of the AVE for a construct is greater than the inter-construct correlations.

Table 3. Results of Discriminant Validity Test.

Latent Variable	1	2	3	4	5	6	7	8	9
Violations	0.75
Errors	0.487	0.764
Lapses	0.453	0.425	0.765
Aggressive Behaviors	0.417	0.406	0.366	0.764
Positive Behaviors	0.056	0.114	−0.02	0.009	0.829
Attitude	−0.337	−0.374	−0.284	−0.279	−0.094	0.758
Social norms	−0.376	−0.337	−0.305	−0.263	−0.153	0.461	0.743
Perceived behavioral control	−0.356	−0.335	−0.246	−0.294	−0.195	0.445	0.395	0.738
Behavioral intention	−0.281	−0.28	−0.22	−0.254	−0.25	0.41	0.382	0.374	0.778

1—Violations; 2—Errors; 3—Lapses; 4—Aggressive Behaviors; 5—Positive Behaviors; 6—Attitude; 7—Social norms; 8—Perceived behavioral control; 9—Behavioral intention.

shows that the square root of the AVE for each construct (diagonal values) is greater than the correlations with other constructs, confirming good discriminant validity and indicating that the latent variables are distinct from one another.

The reliability and validity analyses confirmed that the latent variables were appropriate for further analysis, providing a solid foundation for investigating the relationships between digital capability, pedestrian behavior, and road-sharing willingness.

5. Results

5.1. Latent Class Analysis (LCA)

This study conducted a Latent Class Analysis (LCA) to identify subgroups within the elderly population based on their digital capabilities. Four key variables were used to capture digital behavior: Dc1 (Smart Device Usage), Dc2 (Online Social Interaction), Dc3 (Online Entertainment), and Dc4 (Online Economic Behavior). The objective was to classify the elderly into distinct latent classes that reflect variations in their digital capabilities patterns.

Several LCA models with varying numbers of classes were estimated to determine the best classification. To evaluate model fit, we used the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC), where lower values indicate better model fit.

Table 4. Model Comparison of Latent Classes.

Model	N	Log-Likelihood	df	AIC	BIC
One Class	750	−1876.362	4	3760.724	3779.204
Two Class	750	−1386.867	9	2791.733	2833.314
Three Class	750	−1373.855	14	2775.709	2840.390
Four Class	750	−1371.993	16	2775.987	2849.908
Five Class	750	−1371.993	18	2779.987	2863.148

presents the results of the model comparison.

As shown in Table 4, although the AIC value of the three-class model (2775.709) was slightly lower than that of the two-class model (2791.733), the Bayesian Information Criterion (BIC) of the two-class model (2833.314) was substantially lower than that of the three-class model (2840.390). Because BIC imposes a stronger penalty for model complexity and is particularly suitable for identifying the optimal number of classes in finite samples, it was prioritized over AIC in this study [62].

The LCA identified two distinct classes within the elderly population. Class 1 represents digitally disengaged, comprising 70.8% of the sample, while Class 2 represents digitally engaged, accounting for 29.2% of the sample. The estimated probabilities for each class are presented in

Table 5. Latent Class Marginal Probabilities.

Class	Probability	Std. Error	95% Confidence Interval
Class 1	0.708	0.021	[0.666, 0.747]
Class 2	0.292	0.021	[0.253, 0.334]

, along with their corresponding standard errors and 95% confidence intervals. For example, Class 1 (Digitally Disengaged) had an estimated probability of 0.708 (SE = 0.021, 95% CI = [0.666, 0.747]), while Class 2 (Digitally Engaged) showed a probability of 0.292 (SE = 0.021, 95% CI = [0.253, 0.334]). These intervals confirm that the subgroup classification is statistically robust.

Class 1 represents the majority of elderly individuals, with low levels of digital activity across all variables. Conversely, Class 2 represents a smaller but significant proportion of elderly individuals who are highly engaged in digital activities.

To further differentiate the two latent classes, the marginal means of digital behavior variables were estimated for each class.

Table 6. Latent Class Marginal Means for Digital Capabilities.

Variable	Class 1 (Digitally Disengaged)	Class 2 (Digitally Engaged)
Dc1	0.0741	0.8982
Dc2	0.0816	0.7795
Dc3	0.0944	0.8855
Dc4	0.1259	0.9096

summarizes the probabilities of engaging in various digital activities for both low and high engagement groups.

As seen in Table 6, individuals in Class 1 (Digitally Disengaged) show minimal engagement across all digital activities, while Class 2 (Digitally Engaged) demonstrates consistently high participation. Rather than focusing on percentages already provided in the table, the key implication is that Class 1’s limited digital engagement highlights barriers to AV readiness, whereas Class 2’s extensive digital activity suggests stronger integration into technology-mediated environments. This contrast underscores the digital divide among elderly pedestrians and its potential impact on equitable access to AV adoption.

The two-class model provides important insights into the digital behavior of elderly individuals:

Class 1 (Digitally disengaged): This group consists of elderly individuals with minimal engagement in digital activities. The low levels of participation in smartphone use, online interactions, and economic behaviors suggest that this group may face barriers such as limited digital capabilities, access to technology, or a lack of interest in digital tools.

Class 2 (Digitally engaged): In contrast, individuals in this group are highly engaged with digital activities. They use smartphones regularly and participate in online social interactions, entertainment, and economic activities. This group likely consists of elderly individuals who have a higher level of digital capabilities and are more integrated into the digital landscape.

These findings highlight the importance of addressing the digital divide among the elderly population to ensure equitable access to digital resources and opportunities.

5.2. Ordered Logit Model (OLM)

This study applies ordered logit regression to explore the differences in road-sharing acceptance between elderly pedestrians with digitally disengaged and digitally engaged. when interacting with autonomous vehicles (AVs). The analysis includes variables from the Theory of Planned Behavior (TPB), such as attitude (ATT), subjective norms (SN), perceived behavioral control (PBC), and behavioral intention (BIU), as well as pedestrian behavior variables from the Pedestrian Behavior Questionnaire (PBQ), including violations (VIO), errors (ERR), lapses (LAP), aggressive behaviors (AGG), and positive behaviors (POS). The results reveal distinct patterns of influence for elderly pedestrians with different levels of digital capabilities.

5.2.1. Digitally Disengaged Group (N = 532)

In the digitally disengaged group, several significant factors emerged as predictors of willingness to share roads with Avs, with Results shown in

Table 7. Ordered Logit Model Results for Digitally Disengaged Group.

Variable	Coefficient	Std. Error	z-Score	p-Value
ATT	0.266	0.150	1.77	0.076
SN	0.218	0.145	1.50	0.134
PBC	0.068	0.153	0.44	0.657
BIU	0.196	0.123	1.60	0.110
VIO	0.258	0.142	1.82	0.069
ERR	0.141	0.146	0.96	0.335
LAP	0.020	0.158	0.13	0.897
AGG	−0.029	0.125	−0.23	0.818
POS	0.316	0.092	3.44	0.001
Gender	0.116	0.088	1.32	0.186
60–65 years	−0.085	0.147	−0.58	0.563
66–70 years	−0.032	0.156	−0.21	0.835
71–75 years	−0.157	0.172	−0.91	0.361
Elementary school and below	−0.066	0.262	−0.25	0.802
Middle school	0.390	0.269	1.45	0.146
High school or technical school	−0.372	0.258	−1.44	0.150
≤3000 CNY	0.404	0.308	1.31	0.191
3001–5000 CNY	0.507	0.346	1.46	0.143
Full-time job	0.413	0.995	0.42	0.678
Part-time job	0.211	0.957	0.22	0.826
Retired	−0.079	0.454	−0.17	0.863
car	−0.106	0.109	−0.97	0.334
icard	−0.010	0.162	−0.06	0.953
0 accident	0.435	0.189	2.31	0.021
1–2 accidents	−0.518	0.206	−2.52	0.012
3–4 accidents	−0.088	0.268	−0.33	0.742
Threshold 1	−3.825	0.589
Threshold 2	−2.203	0.511
Threshold 3	−1.041	0.497
Threshold 4	0.361	0.495
Log-likelihood	−592.46
LR chi2	71.67
Prob > chi2	0.0000
Pseudo R2	0.077

Non-Significant Variables: subjective norms (SN), perceived behavioral control (PBC), behavioral intention (BIU), errors (ERR), lapses (LAP), aggression (AGG), demographic factors (gender, age), income, occupation, car ownership, and iCard possession did not significantly affect AV acceptance in this group.

:

Attitude (ATT): A marginally significant positive effect was found (β = 0.266, p = 0.076), suggesting that elderly individuals with a more favorable attitude toward AVs are more likely to accept road-sharing with them. Although not strongly significant, this indicates that attitudes play a moderate role in shaping acceptance in this group.

Positive Behaviors (POS): Positive pedestrian behaviors exhibited a robust and highly significant effect (β = 0.316, p = 0.001). This confirms that elderly pedestrians who comply with traffic rules are considerably more open to AV road-sharing, underlining the importance of safe pedestrian practices in acceptance.

Violations (VIO): Violations showed a marginally significant positive effect (β = 0.258, p = 0.069), implying that seniors who frequently engage in risky behaviors may paradoxically be more open to AVs, possibly because AVs are perceived as more predictable than human drivers.

Accident History: A different pattern emerged compared with earlier interpretation. Seniors with no accident history were significantly more willing to share roads with AVs (β = 0.435, p = 0.021). By contrast, those with 1–2 accidents were significantly less willing (β = −0.518, p = 0.012), suggesting that repeated accident experiences might heighten caution and risk aversion. The 3–4 accidents group did not show a statistically significant effect (β = −0.088, p = 0.742).

5.2.2. Digitally Engaged Group (N = 218)

In the digitally engaged group, different variables significantly influenced road-sharing acceptance, with Results shown in

Table 8. Ordered Logit Model Results for Digitally Engaged Group.

Variable	Coefficient	Std. Error	z-Score	p-Value
ATT	−0.036	0.201	−0.18	0.857
SN	−0.014	0.181	−0.08	0.937
PBC	0.353	0.192	1.84	0.066
BIU	0.224	0.177	1.27	0.205
VIO	−0.318	0.202	−1.57	0.116
ERR	0.540	0.207	2.60	0.009
LAP	−0.061	0.185	−0.33	0.740
AGG	−0.097	0.171	−0.57	0.569
POS	−0.414	0.154	−2.69	0.007
Gender	0.178	0.130	1.37	0.169
60–65 years	0.090	0.210	0.43	0.669
66–70 years	−0.018	0.221	−0.08	0.935
71–75 years	−0.474	0.263	−1.80	0.071
Elementary school and below	−0.860	0.496	−1.73	0.083
Middle school	−1.183	0.518	−2.29	0.022
High school or technical school	−0.548	0.503	−1.09	0.276
≤3000 CNY	0.247	0.434	0.57	0.570
3001–5000 CNY	0.822	0.451	1.82	0.068
Full-time job	−0.822	0.553	−1.49	0.137
Part-time job	0.080	0.629	0.13	0.899
Retired	0.108	0.314	0.34	0.731
car	0.307	0.163	1.88	0.060
icard	−0.085	0.186	−0.46	0.647
0 accident	0.356	0.221	1.61	0.107
1–2 accidents	−0.039	0.249	−0.16	0.875
3–4 accidents	−0.545	0.295	−1.85	0.065
Threshold 1	−3.022	0.459
Threshold 2	−1.507	0.417
Threshold 3	−0.064	0.411
Threshold 4	1.910	0.431
Log-likelihood	−305.20
LR chi2	43.24
Prob > chi2	0.0182
Pseudo R2	0.086

:

Perceived Behavioral Control (PBC): PBC showed a marginally significant positive effect (β = 0.353, p = 0.066). This indicates that elderly pedestrians with higher confidence in managing interactions with AVs are more likely to accept road-sharing, even though the effect is weaker compared to conventional thresholds.

Errors (ERR): Errors had a statistically significant positive impact (β = 0.540, p = 0.009). This suggests that digitally engaged seniors who are prone to making mistakes remain open to AV interactions, possibly reflecting their stronger trust in technological systems to compensate for human errors.

Positive Behaviors (POS): Unlike the digitally disengaged group, positive behaviors were significantly negatively associated with acceptance (β = −0.414, p = 0.007). This counterintuitive result may reflect heightened risk awareness among digitally literate seniors, who—despite safe walking practices—critically evaluate AV safety and reliability.

Education: Middle school education had a significant negative effect on acceptance (β = −1.183, p = 0.022). This indicates that moderately educated seniors with digital engagement may adopt a more skeptical stance toward AVs, potentially due to greater awareness of risks or information exposure. Elementary education also showed a marginally negative trend (β = −0.860, p = 0.083), though not fully significant.

Car Ownership: Car ownership was positively associated with willingness to share roads (β = 0.307, p = 0.060), albeit at a marginal significance level. This implies that those with driving experience and familiarity with road systems are more confident in coexisting with AVs.

Age and Accident History: Seniors aged 71–75 years exhibited a marginally significant negative effect (β = −0.474, p = 0.071), suggesting that the oldest segment of this group might be more cautious. Similarly, those with 3–4 accidents also displayed a marginally significant reduction in acceptance (β = −0.545, p = 0.065), reinforcing the role of accumulated risk experiences in shaping skepticism.

Non-Significant Variables: Attitude (ATT), subjective norms (SN), behavioral intention (BIU), violations (VIO), lapses (LAP), aggressive behaviors (AGG), gender, other age groups, income levels, employment status, and iCard ownership did not significantly predict AV road-sharing acceptance in this group.

To ensure model adequacy, we report key diagnostics including log-likelihood, pseudo R², and likelihood ratio (LR) tests. For the digitally disengaged group, the model yielded a log-likelihood of −592.46 with a pseudo R² of 0.077, and the LR chi² test was highly significant (p < 0.001), indicating satisfactory explanatory power. For the digitally engaged group, the log-likelihood was −305.20 with a pseudo R² of 0.086, and the LR chi² test was significant (p < 0.05), also suggesting that the model fit the data adequately.

5.2.3. Key Findings and Comparisons

Attitude and Perceived Behavioral Control (PBC): In the digitally disengaged group, attitude showed a marginally significant positive effect (β = 0.266, p = 0.076), suggesting that basic attitudinal support plays a moderate role in acceptance. By contrast, in the digitally engaged group, PBC was only marginally significant (β = 0.353, p = 0.066), implying that confidence in managing interactions with AVs matters more for digitally capable individuals, even though the effect did not reach full conventional significance.

Pedestrian Behaviors: Positive pedestrian behaviors exhibited opposite effects. For the digitally disengaged group, POS significantly increased acceptance (β = 0.316, p = 0.001), confirming that compliance with safe behaviors enhances willingness. In contrast, for the digitally engaged group, POS significantly decreased acceptance (β = −0.414, p = 0.007), suggesting that digitally literate seniors, despite safe behaviors, critically assess AV risks and reliability. This divergence is further illustrated in Figure 1: in the digitally disengaged group, the predicted probability of perceiving road-sharing as “very safe” increases steadily from 0.62 at POS = 1 to 0.85 at POS = 5, whereas in the digitally engaged group, the probability declines from 0.065 to 0.013 over the same range. Moreover, errors (ERR) were non-significant in the disengaged group but showed a significant positive effect in the engaged group (β = 0.540, p = 0.009), indicating that digitally confident seniors may rely on AVs to compensate for their own mistakes.

Education: Educational effects diverged across groups. In the disengaged group, middle school education was not significant (β = 0.390, p = 0.146), while in the engaged group, middle school education significantly decreased acceptance (β = −1.183, p = 0.022), and elementary education showed a marginally negative trend (β = −0.860, p = 0.083). This pattern suggests that higher education, when combined with digital engagement, fosters more critical or skeptical attitudes toward AVs.

Accident History and Car Ownership: For the disengaged group, having no accident history significantly increased acceptance (β = 0.435, p = 0.021), while 1–2 accidents significantly decreased acceptance (β = −0.518, p = 0.012), indicating that repeated accident experiences amplify caution. In the engaged group, car ownership had a marginally positive effect (β = 0.307, p = 0.060), showing that personal driving experience enhances confidence in coexisting with AVs. Additionally, seniors aged 71–75 years (β = −0.474, p = 0.071) and those with 3–4 accidents (β = −0.545, p = 0.065) were marginally less accepting, reflecting risk sensitivity among the oldest and most accident-prone individuals.

Note on Thresholds: In both groups, the ordered logit models estimated four threshold parameters (cut1–cut4). These thresholds represent the cut-off values on the unobserved latent scale of perceived safety, which divide the continuous distribution into the five ordered response categories (1 = very unsafe, …, 5 = very safe). Larger or smaller threshold values indicate shifts in how individuals transition between adjacent safety perception categories. This clarification ensures the correct interpretation of model outputs.

6. Discussion

This study aimed to investigate how elderly pedestrians’ digital capability influences their willingness to share roads with autonomous vehicles (AVs). Using Latent Class Analysis (LCA), respondents were segmented into two groups based on four dimensions of digital capability—smart device usage, online social interaction, online entertainment, and online economic behavior. An ordered logistic regression was then employed to evaluate how digital capabilities moderates key cognitive and behavioral constructs drawn from the Theory of Planned Behavior (TPB) and the Pedestrian Behavior Questionnaire (PBQ). The findings revealed distinct psychological mechanisms and behavioral tendencies across digital capabilities groups, offering both theoretical insight and practical implications for AV acceptance research.

6.1. Interpretation of Key Findings

Attitude vs. Perceived Behavioral Control (PBC): In the digitally disengaged group, attitude showed a marginally significant positive effect on AV acceptance (β = 0.266, p = 0.076), suggesting that basic attitudinal support plays a moderate role in influencing willingness to share roads. By contrast, in the digitally engaged group, PBC was marginally significant rather than fully dominant (β = 0.353, p = 0.066), implying that confidence in managing interactions with AVs matters more for digitally capable individuals, even though the effect did not reach full conventional significance. This aligns with TPB (Ajzen, 1991), which posits that PBC gains importance when individuals feel capable of executing the behavior [35]. High-digital-capability seniors likely possess stronger technological self-efficacy and familiarity, enhancing their perceived control over AV interaction, as corroborated by studies on digital capabilities and autonomy in older adults and the moderation of control in smart mobility adoption [63,64].

Pedestrian Behavior Variation (PBQ): Positive pedestrian behaviors significantly increased AV acceptance in the disengaged group (β = 0.316, p = 0.001), confirming that compliance with traffic safety norms enhances openness to AVs. Conversely, in the engaged group, positive behaviors had a significant negative effect (β = −0.414, p = 0.007). This seemingly paradoxical result can be theoretically explained by heightened risk awareness among digitally literate seniors, who—despite safe behaviors—may critically evaluate AV reliability and perceive automation as potentially fallible [65,66]. Moreover, prior research suggests that higher levels of digital competence may foster overconfidence in assessing technology risks, leading to more cautious or skeptical attitudes toward AVs [67].

Socio-Demographic Moderation (Education & Accident History): Education exerted opposite effects across groups. In the disengaged group, middle school education was not statistically significant (β = 0.390, p = 0.146), whereas in the engaged group, middle school education significantly decreased AV acceptance (β = −1.183, p = 0.022) and elementary education showed a marginally negative trend (β = −0.860, p = 0.083). This suggests that education, when combined with digital engagement, may foster more critical or skeptical attitudes toward AVs, aligning with risk-critical findings in higher-educated AV users [68,69].

Accident history also revealed distinct effects. Among disengaged seniors, having no accident history significantly increased AV acceptance (β = 0.435, p = 0.021), while 1–2 accidents significantly decreased acceptance (β = −0.518, p = 0.012). By contrast, the 3–4 accidents group did not show significance (β = −0.088, p = 0.742). In the engaged group, car ownership was a marginally significant positive predictor (β = 0.307, p = 0.060), suggesting that driving experience strengthens agency in AV interaction. Additionally, seniors aged 71–75 years (β = −0.474, p = 0.071) and those with 3–4 accidents (β = −0.545, p = 0.065) exhibited marginally lower acceptance, reflecting heightened risk sensitivity among older and accident-prone individuals.

6.2. Comparison with Previous Studies

The results of this study underscore the importance of digital capabilities as a boundary condition that shapes elderly individuals’ responses toward AV adoption—an insight supported by previous technology acceptance research.

Firstly, our finding that higher digital capabilities strengthens the relationship between perceived behavioral control (PBC) and AV acceptance aligns with the Unified Theory of Acceptance and Use of Technology (UTAUT2), where experience and prior digital exposure enhance self-efficacy and usage behaviors [51]. This study explicitly demonstrates this dynamic within elderly pedestrian populations—suggesting that digital proficiency can substitute for direct AV experience by strengthening users’ internal control beliefs—and Ahmad’s (2020) work, which indicates that digitally confident seniors maintain greater autonomy and receptivity toward emerging technologies [70].

Secondly, prior research typically establishes a straightforward positive link between safety-oriented behavior and technology adoption [71]. However, our high-digital group exhibited an unexpected negative relationship between positive pedestrian behavior and AV acceptance. This divergence emerges from theoretical constructs proposed in risk cognition literature, wherein individuals with higher familiarity with digital systems engage in heightened risk appraisal when assessing complex automation [72,73]. In effect, higher digital capability fosters critical awareness of AV limitations, which can override the typically positive influence of safety-oriented behaviors.

Thirdly, our observation that digitally savvy seniors tend to trust AVs more when they themselves make errors aligns with the psychological phenomenon of ‘delegation trust’—where individuals confident in themselves are more willing to delegate control to systems they perceive as reliable. Hegner et al. (2023) similarly highlight that perceived control mediates the acceptance of smart mobility solutions, indicating that digital familiarity predisposes users to trust AVs as compensatory mechanisms [74]. This phenomenon provides deeper theoretical grounding for why digital capability does not simply shift mean levels of acceptance, but transforms the cognitive processes linking behavior and intention.

6.3. Practical Implications

The findings of this study provide actionable insights for policymakers, urban planners, and autonomous vehicle (AV) developers. Importantly, the implications differ by digital capability level, highlighting the necessity of tailoring interventions to distinct subgroups of elderly pedestrians.

(1)

Enhancing Digital Literacy among Low-Digital-Capability Seniors: For elderly individuals with limited digital capabilities, interventions should prioritize the development of digital self-efficacy. Digital skills have been shown to significantly influence technology acceptance by enhancing confidence, perceived control, and willingness to engage with automated system [75,76]. Structured training programs (e.g., smartphone tutorials, internet workshops, or community-based digital inclusion initiatives) can improve older adults’ comfort with technology and, in turn, increase their readiness to coexist with AVs in road-sharing environments [77]. The digitally disengaged group (Class 1) represents the majority of the sample, indicating a strong need for targeted interventions to enhance digital inclusion, such as digital capabilities training and improved access to affordable technologies.

(2)

Designing User-Centric Interfaces for Digitally Literate Seniors: Among seniors with high digital proficiency, perceived behavioral control (PBC) was the strongest predictor of AV acceptance. This supports the Unified Theory of Acceptance and Use of Technology (UTAUT2), which posits that control and usability shape behavioral intention [51]. Accordingly, AV developers should prioritize intuitive, transparent, and adaptive interfaces that enhance users’ sense of control. Examples include real-time intent signaling, visual communication cues, and context-sensitive responsiveness. These features can enhance users’ sense of control, mitigate risk perception, and foster trust, especially among elderly users who are both tech-savvy and critically aware of automation limitations [78,79]. The digitally engaged group (Class 2), while smaller, still reflects a substantial portion of elderly individuals who are actively engaged in digital activities, suggesting that initiatives to support and expand their digital capabilities could be highly beneficial for sustained AV acceptance.

These differentiated strategies, grounded in empirical evidence and psychological theory, underscore the importance of integrating digital capability segmentation into AV policy and design to promote equitable and safe urban mobility.

7. Conclusions

This study investigated the impact of digital capabilities on elderly pedestrians’ willingness to share roads with autonomous vehicles (AVs) by using Latent Class Analysis (LCA) to categorize elderly individuals into digitally engaged and digitally disengaged groups. By applying the Theory of Planned Behavior (TPB) and the Pedestrian Behavior Questionnaire (PBQ) as theoretical frameworks, we utilized ordered logit regression to identify key factors influencing road-sharing acceptance across these groups. The findings provide significant insights into how elderly pedestrians’ digital capabilities influence their attitudes and behaviors toward AVs, contributing to the understanding of technology acceptance in an aging society.

For the digitally disengaged group, a positive attitude toward AVs and adherence to positive pedestrian behaviors were important drivers of road-sharing acceptance. Additionally, individuals with mid-level education and those with limited accident history showed higher acceptance of AVs. These findings highlight that elderly individuals with lower digital capabilities may be more reliant on their perceptions of safety and their past experiences when evaluating AV technology. Encouraging positive pedestrian behaviors and promoting education about AV benefits could be key strategies to enhance AV acceptance in this group.

In contrast, the digitally engaged group exhibited different behavioral patterns. Perceived behavioral control (PBC) emerged as the most significant predictor of road-sharing acceptance, suggesting that individuals with higher digital capabilities feel more confident in their ability to interact with AVs. Interestingly, positive pedestrian behaviors negatively influenced AV acceptance in this group, indicating a more cautious or critical stance toward AV technology among digitally engaged elderly individuals. This group also demonstrated that errors in pedestrian behavior did not diminish their willingness to share roads, possibly due to increased trust in the capabilities of AV technology.

The findings underscore the importance of tailoring AV-related policies and interventions based on the digital capabilities of elderly populations. For digitally disengaged individuals, enhancing digital capabilities, promoting positive pedestrian behavior, and providing accessible information about AV safety could boost acceptance. For digitally engaged individuals, efforts should focus on improving AV systems’ perceived control and addressing the specific concerns of digitally savvy users to foster greater confidence in AV technologies.

This study contributes to the literature by demonstrating that digital capabilities plays a critical role in shaping elderly pedestrians’ attitudes and behaviors toward AVs. As cities move toward integrating AVs into their transportation networks, understanding how digital capabilities affects road-sharing behavior among elderly populations is essential for designing inclusive and equitable mobility solutions. Further research could explore additional factors, such as cultural differences and the role of social influence, to develop a more comprehensive understanding of AV acceptance across diverse demographic groups.

Although the empirical data were collected in Wuhan, China—a city with advanced AV deployment and supportive policy frameworks—the theoretical mechanisms identified in this study, such as the moderating role of digital capability, are likely to be relevant in other contexts. However, cultural, infrastructural, and policy differences may influence the strength of these effects. Thus, the findings should be generalized with caution, and future cross-national comparative studies are encouraged to validate and extend these results.

Despite its contributions, this study is not without limitations. First, the data were collected in a single urban context (Wuhan), which may restrict the representativeness of the findings. Second, the reliance on self-reported survey data may introduce response bias. Third, although digital capability was the focal construct, other potentially relevant variables—such as health status, cognitive ability, or broader mobility habits—were not included. Future research should address these limitations by employing multi-site and cross-national samples, integrating objective behavioral indicators, and incorporating additional personal and contextual factors to provide a more comprehensive picture of elderly pedestrians’ AV acceptance.

This study opens several avenues for future research. First, the role of external human–machine interface (eHMI) systems could be explored to assess their potential in improving elderly pedestrians’ perceived safety and control when interacting with AVs. Second, future studies should investigate how varying levels of digital capabilities across different cultural contexts influence AV acceptance. Finally, longitudinal studies could provide deeper insights into how elderly pedestrians’ attitudes toward AVs evolve as they gain more exposure to these technologies in real-world environments.

By addressing these avenues, future research can provide a more nuanced understanding of how digital capabilities and pedestrian behavior interact in shaping elderly individuals’ road-sharing willingness in the era of autonomous transportation.