People’s Intentions to Use Shared Autonomous Vehicles: An Extended Theory of Planned Behavior Model

Abstract

With the advancement of technology, sharing and autonomous driving will be the two major themes in the future transportation field, and SAVs (Shared autonomous vehicles) will combine the two things. When SAVs come to market, they will affect the transportation system, so the objective of this paper is to examine people’s intentions to use SAVs and clarify the factors affecting people’s intentions to use SAVs. Due to the application of the theory of planned behavior (TPB) in traffic travel research having important practical significance, this paper used an extended theory of planned behavior model to study people’s intentions to use SAVs. Some important findings are found that the intention to use SAVs is directly affected by attitude, subjective norm, perceived behavior control, barrier, and effects of a public health emergency, and indirectly affected by perceived risk, technical interest, government policy, and environmental awareness. Moreover, perceived behavior control has the mediating effect between government policy and intention to use SAVs, between technical interest and intention to use SAVs, and between subjective norm and intention to use SAVs. According to the influence degree of related influencing factors, the corresponding development recommendations on SAVs development are put forward. The research results of this paper contribute to the subsequent listing of SAVs, promote the further development of intelligent transportation, and provide the scientific basis for future travel policy formulation and traffic planning.

1. Introduction

Sharing travel and autonomous vehicles (AVs) are hot topics in the future urban transportation system [1,2]. Shared autonomous vehicle (SAV) is a new transportation mode derived from the combination of shared travel and autonomous vehicles. Namely, users submit their own location, destination, departure time, and other requests to the operator through mobile phone applications. Then, an SAV with the maximum utility is calculated to pick up the user at the departure point on time and take him/her to the destination [3].

The combination of AVs and sharing travel effectively saves time, improves operational efficiency, alleviates traffic congestion, further promotes green mobility, and also facilitates the landing of autonomous driving technology [4,5,6]. At the same time, SAVs combine the advantages of sharing travel and autonomous vehicles, which is efficient, convenient, safe, and comfortable and lower cost on-demand travel services [4,5,6,7]. The SAVs would effectively promote the construction of new traffic systems, avoid traffic accidents, and reduce the casualties and property losses caused by accidents [4,7].

However, reports of autonomous vehicle accidents and shared vehicle accidents frequently appear, even including deaths [8,9]. These to some extent affect people’s intentions to use SAVs. For example, scholars found that the number of serious accidents was highly significant to drivers of choice to SAVs [10].

In previous studies, AV acceptability has been widely studied [11,12]. Previous studies have shown that positive attitudes towards autonomous vehicles can be supported by giving people the possibility to try autonomous vehicles in a safe, real-life environment [11]. Furthermore, the willingness to pay for SAVs services was studied, which showed the main barrier preventing mass adoption of autonomous vehicles may also be psychological besides technological [13]. In addition, some studies have proved that positive use intention is beneficial to the promotion of SAVs and important in planning for a sustainable and intelligent transportation system [14,15,16]. Thus, to further guide the development of SAVs, it is very meaningful and necessary to study people’s intentions to use SAVs. The study of people’s intentions to use SAVs is conducive to the subsequent listing of SAVs, promotes the further development of intelligent transportation, and has important theoretical and practical significance.

To sum up, the objective of this paper is to examine people’s intentions to use SAVs, clarify the factors affecting people’s intentions to use SAVs, and put forward optimization strategies for the scientific development of SAVs. This paper is presented in five sections. The first section introduces the background. Section 2 introduces the theoretical framework and research hypotheses. Section 3 describes sample and data collection, survey instrument, and analytical method. Section 4 lists all the analytical results of the model analysis. Finally, Section 5 concludes this study and proposes future research recommendations.

2. Theoretical Framework and Research Hypotheses

2.1. TPB and Related Studies of Travel

The theory of planned behavior (TPB) was proposed by Ajzen as early as 1985 [17]. TPB is a psychological theory that connects beliefs with behaviors. According to the TPB, as shown in Figure 1, behavioral intention has the most direct influence on actual behavior, while attitude, subjective norm and perceived behavioral control have the influence on behavioral intention. Attitude refers to an individual’s positive or negative evaluation of the performance of a specific behavior. Subjective norm refers to an individual’s subjective perception of the social pressure that motivates them to adopt a specific behavior, mainly from others’ expectations towards the actor. Perceived behavior control refers to an individual’s perception of the ease or difficulty of executing a specific behavior, reflecting their perception of factors that promote or hinder the execution of the behavior. Accurate perceptual behavioral control reflects the state of actual control conditions, so it can serve as an alternative measurement indicator for actual control conditions to directly predict the likelihood of behavior occurrence (as shown by the dashed line in Figure 1). Moreover, Belief is the cognitive and emotional foundation of behavioral attitude, subjective norm, and perceived behavior control.

With the increasing popularity of traffic travel, the TPB has been applied to the study of traffic travel and has become one of the hot topics of researchers [18,19,20].

First, the TPB is used to study driving behavior. Researchers analyzed the impact of drivers’ attitudes and behavioral habits toward safe driving by investigating factors such as attitude, social norm, and perceptual control. For example, a questionnaire based on the TPB was designed to explore factors affecting fatigued driving behavior from the perspective of social psychology [21]. The key factors leading to the risky driving behavior of novice drivers were analyzed and quantified on the basis of the TPB and the protection motivation theory [22]. A TPB Questionnaire was conducted to explore the contributing factors of driving distraction and compare the contributing factors for three typical distracted driving behaviors: drinking water, answering a phone and using a mobile phone application (APP) while driving [23]. Based on the TPB, Dinh explored the factors influencing motorcyclists’ intention to drink and drive in Vietnam [24].

Secondly, the TPB is used to study public transport travel intention. By investigating travelers’ attitudes, social norms and other factors, researchers analyzed the impact of travelers‘ attitudes and travel habits on public transport travel. For example, a model of a comprehensive psychological process on an individual’s public transit use decision-making was presented, which is accomplished by the integration of the TPB and the customer satisfaction theory [25]. Through the TPB, an individual’s public transport travel behavior was studied, as impacted by travel and residential location experiences in their childhood [26]. If an extended version of the TPB is suited to predict subscription to a public transport ticket was investigated [27].

In addition, the TPB is also used to study pedestrian behavior. By investigating factors such as pedestrian attitudes, social norms and so on, researchers analyzed the influence of pedestrian attitudes and behavior habits on walking. For example, a TPB-based questionnaire was developed to measure road-crossing attitudes and potentially risky pedestrian behavior [28]. The TPB and the PWM (prototype willingness model) in pedestrian violations were compared by using structural equation modeling [29]. Child participants’ attitudes, beliefs, and perceptions were examined, which were psychological factors that influence their behavior when crossing roads [30].

With the popularity of shared travel and AVs, some scholars also use the theory of planned behavior to study travelers’ willingness to use AVs and SAVs. Based on an extended version of the TPB, the determinants that influence travelers’ behavioral intentions towards the use of AVs and SAVs were explored, which incorporates knowledge and perceived risk [31]. Scholars have found that the TPB constructs, namely attitude, subject norm, perceived behavioral control, along with its perceived facilitating conditions, are all effective predictors of intention to use SAVs [32]. User acceptance of fully automated vehicles in Iran was assessed by adopting and comparing three popular user acceptance models: the technology acceptance model (TAM), the theory of planned behavior (TPB) and the unified theory of acceptance and use of technology (UTAUT) [33].

In conclusion, the application of planned behavior theory in traffic travel research has important practical significance. Some researchers realized the importance of the intention to use SAVs and used TPB to study the intention to use AVs and SAVs. However, a comprehensive analysis of the use intention of SAVs has not been studied yet, so this paper focuses on people’s intentions to use shared autonomous vehicles by constructing an extended TPB model. Tring to provide the ide scientific basis for future SAVs travel policy formulation and traffic planning.

2.2. Research Model Based on TPB

According to the TPB, the intention is affected by three pre-variables, namely, individual behavior attitude, subjective norm and perceived behavioral control. So, these five latent variables are determined as the basic latent variable.

Hypothesis 1.

Attitude has a significant direct impact on the intention to use SAVs.

Hypothesis 2.

Subjective norm has a significant direct impact on the intention to use SAVs.

Hypothesis 3.

Perceived behavioral control has a significant direct impact on the intention to use SAVs.

Hypothesis 4.

Subjective norm has a significant direct impact on attitude.

Hypothesis 5.

Subjective norm has a significant direct impact on perceived behavioral control.

According to the development experience of carsharing, the government plays an important role in the promotion of carsharing [34]. For example, the Italian Ministry of the Environment has funded the carsharing initiative and formulated operational and technical standards to promote the dissemination of the Carsharing service. Furthermore, previous studies have shown that some government policies can affect people’s willingness to choose travel methods [35]. Thus, the following assumption is proposed:

Hypothesis 6.

Government policy has a significant direct impact on attitude.

Hypothesis 7.

Government policy has a significant direct impact on perceived behavioral control.

Shared transportation behavior is a green and environmentally friendly behavior that can reduce carbon emissions and help build an ecological civilization and environmentally friendly society [5]. SAV is also a way of shared transportation, and this paper speculates that environmental awareness also plays a key role in using SAVs. Thus, the following assumptions are proposed:

Hypothesis 8.

Environmental awareness has a significant direct impact on attitude.

Hypothesis 9.

Environmental awareness has a significant direct impact on perceived behavioral control.

Perceived risk is the prerequisite for understanding various objectively existing risks and identifying them. It will subjectively affect people’s attitudes toward things in the first place [30]. Perceived risk has a significant negative impact on attitudes toward transportation travel [30]. Thus, the following assumption is proposed:

Hypothesis 10.

Perceived risk has a significant direct impact on attitude.

When people have a strong interest in new technologies of new things, they will become very proactive and eager to try new things. Groups that are more interested in technology and support AVs are more likely to use SAVs [36]. Thus, the following assumptions are proposed:

Hypothesis 11.

Technical interest has a significant direct impact on attitude.

Hypothesis 12.

Technical interest has a significant direct impact on perceived behavioral control.

Hypothesis 13.

Technical interest has a significant direct impact on the intention to use SAVs.

Previous studies have shown that obstacles can affect people’s travel intentions, such as significant relationships between external obstacles, personal obstacles, and bicycle use [37]. Thus, the following assumptions are proposed:

Hypothesis 14.

Barrier has a significant direct impact on attitude.

Hypothesis 15.

Barrier has a significant direct impact on the intention to use SAVs.

During the outbreak of the COVID-19 epidemic, the traffic volume in Chinese cities declined significantly. Public health emergencies have had a profound impact on people’s travel [20]. After the epidemic, groups with private cars are more inclined to use private vehicles instead of public traffic for travel, while groups without private cars are more inclined to walk, bike, and share bicycles, which have less contact with others. The SAVs can meet the needs of private travel. Thus, the following assumption is proposed:

Hypothesis 16.

Effects of public health emergencies have a significant direct impact on the intention to use SAVs.

In conclusion, based on the TPB, combined with the existing research results, this paper extends the TPB by incorporating the government policy, environmental awareness, perceived risk, technical interest, barrier, use intention and effects of a public health emergency. In total, there are 10 latent variables (Figure 2).

3. Methodology

3.1. Sample and Data Collection

Two phases of this study were investigated: pre-investigation and formal investigation. Through reliability and validity tests of pre-investigation (150 questionnaires), questions were adjusted and deleted. Then the formal questionnaire was finally determined. Finally, in the formal questionnaire, there were 433 questionnaires distributed to respondents, and 364 valid questionnaires were collected, which met the minimum sample size requirement [38]. The questionnaire data is all from Beijing, based on the Internet.

3.2. Survey Instrument

The questionnaire includes four parts: travel characteristics, intention to use SAVs, potential variables and personal characteristics.

(1)

Travel characteristics, including the travel distance on workdays and weekends, respectively, travel frequency, travel mode selection, travel time period, parking fees, parking time and so on.

(2)

Intention to use SAVs, including familiarity with SAVs and possible travel purposes for using SAVs.

(3)

Potential variables include attitude, subjective norm, perceived behavioral control, environmental awareness, government policy and so on. The measurement of each variable is recorded using a five-point Likert rating scale, and the corresponding numbers are selected based on the strength of willingness. When filling out the questionnaire, choose from five options: “strongly disagree”, “disagree”, “neutral”, “agree” and “strongly agree”. The questions and abbreviations of potential variables and intention to use SAVs as shown in Appendix A.

(4)

Personal characteristics. This includes gender, age, highest education level, occupation, monthly income, annual family income, ownership of private cars, ownership of driver’s licenses, driving skills, marital status, number of people living together, whether there are children and the number of children. Refer to the Occupational Classification Code of the People’s Republic of China to classify occupation options. Classify age groups according to different educational levels and retirement time.

3.3. Analytical Method

The paper adopts a structural equation model (SEM) and uses IBM SPSS AMOS 26.0 software for data processing and analysis.

4. Results

4.1. Profile

The socio-demographic characteristics of respondents are shown in

Table 1. Socio-demographic characteristics of respondents.

Variable	Description	Frequency	Percentage (n = 364)
Sex	Male	196	53.85
	Female	168	46.15
Age	<18 years	8	2.20
	18–24 years	185	50.82
	25–35 years	85	23.35
	36–50 years	59	16.21
	51–60 years	23	6.32
	>60 years	4	1.10
Highest level of education	Senior high school and below it	20	5.49
	Junior college	56	15.38
	undergraduate	228	62.64
	Master’s degree or above it	60	16.48
Occupation	Professional and technical staff	63	17.31
	Business/Service occupations	8	2.20
	Employees of enterprises and public institutions/civil servants	54	14.84
	Military/police	2	0.55
	Production and transportation equipment operators	21	5.77
	student	189	51.92
	freelancer	6	1.65
	retirees	4	1.10
	others	17	4.67
Monthly income	<¥3000	175	48.08
	¥3001–¥5000	39	10.71
	¥5001–¥8000	74	20.33
	¥8001–¥12,000	54	14.84
	¥12,001–¥20,000	17	4.67
	>¥20,000	5	1.37
Annual household income	<¥0.1 million	108	29.67
	¥0.1–0.2 million	149	40.93
	¥0.2–0.3 million	46	12.64
	¥0.3–0.5 million	41	11.26
	¥0.5–0.7 million	10	2.75
	¥0.1–1 million	6	1.65
	>¥1 million	4	1.10
Car ownership	Yes	218	59.89
	No	146	40.11
Driving license	Yes	295	81.04
	No	69	18.96
Skillful driving	No Driving license	69	18.96
	Yes	205	56.32
	No	90	24.73
Marital status	married	137	37.64
	unmarried	227	62.36
The number of people living together	1 person	52	14.29
	2 persons	56	15.38
	3–4 persons	225	61.81
	>5 persons	31	8.52
Whether have any children	Yes	113	31.04
	No	251	68.96
Number of children	1	79	21.70
	2	26	7.14
	3	4	1.10
	≥4	4	1.10

. The male proportion of respondents is 53.85%, slightly higher than the female participation rate (46.15%). The age of the respondents is relatively young, with the majority being between 18 and 50 years old, accounting for 90.38%. More than half of the respondents have a bachelor’s degree (62.64%) or higher (16.48%). A total of 94.51% of the respondents have a college degree or above, which means the respondents have good reading ability and can understand the content of the questionnaire, which also ensures the reliability of the questionnaire.

Respondents are distributed across different occupations, including professional and technical staff, business/service occupations, employees of enterprises and public institutions/civil servants, military/police and so on.

Among them, students, professional and technical staff and employees of enterprises and public institutions/civil servants are the most, accounting for 51.92%, 17.31% and 14.84%, respectively.

Most respondents’ monthly income levels are below CNY 3000 (48.08%), which is related to a large number of student respondents (51.92%).

The majority of respondents’ households have an annual income of CNY 100,000 to 200,000 (40.93%). Among the respondents, those with private cars (59.89%) are higher than those without private cars (40.11%).

Most respondents have a driver’s license (81.04%), of which 56.32% are proficient in driving. It can be seen that whether there is a private car or not, there are still many people with a driver’s license, indicating that obtaining a driver’s license has become a basic phenomenon.

Among the respondents, more unmarried people (62.36%) than married people (37.64%). The vast majority of people live together with 3–4 people, accounting for 61.81%. The vast majority of people do not have children, accounting for 68.96%. Furthermore, most people with children have one child, accounting for 70.54%.

The above are the personal characteristics of respondents, which could indicate that these respondents may be representative of the Chinese people.

4.2. Reliability and Validity Test

(1)

Reliability test

Reliability is an indicator that tests the stability and consistency of a scale when measuring related concepts. Cronbach’s Alpha is currently the most commonly used reliability coefficient in reliability test [39]. Therefore, Cronbach’s Alpha is used for testing reliability in this paper. The formula of Cronbach’s Alpha is as follows.

$$\alpha =\frac{k}{k+1}\left(1-\frac{{\displaystyle {\sum }_{i=1}^k{\sigma }_i^2}}{{\sigma }_{total}^2}\right)$$

(1)

where $\alpha$ presents reliability coefficient; $k$ presents number of questions in measurement tools; ${\sigma }_i^2$ presents variance of the question $i$; ${\sigma }_{total}^2$ presents variance of overall score of measurement tools. The value range of $\alpha$ should be between 0 and 1, and the closer $\alpha$ is to 1, the higher the reliability.

As shown in

Table 2. Reliability tests.

Latent Variables	Abbreviation of Question	Correction Item-Total Correlation (CITC)	Cronbach’s Alpha If Item Deleted	Cronbach’s Alpha
Attitude	AT1	0.724	0.903	0.914
	AT2	0.65	0.907
	AT3	0.674	0.905
	AT4	0.641	0.907
	AT5	0.665	0.906
	AT6	0.574	0.911
	AT7	0.7	0.904
	AT8	0.702	0.904
	AT9	0.717	0.903
	AT10	0.515	0.913
	AT11	0.787	0.9
Subjective norm	SN1	0.749	0.897	0.898
	SN2	0.847	0.814
	SN3	0.805	0.851
Perceived behavioral control	PBC1	0.397	0.604	0.653
	PBC2	0.358	0.622
	PBC3	0.475	0.566
	PBC4	0.466	0.57
	PBC5	0.332	0.634
Government policy	GP1	0.766	0.851	0.887
	GP2	0.793	0.827
	GP3	0.779	0.84
Environmental awareness	EA1	0.63	0.764	0.81
	EA2	0.596	0.773
	EA3	0.676	0.748
	EA4	0.624	0.764
	EA5	0.467	0.812
Perceived risk	PR1	0.741	0.78	0.846
	PR2	0.676	0.808
	PR3	0.703	0.798
	PR4	0.618	0.833
Technological interest	TI1	0.764	0.79	0.863
	TI2	0.776	0.772
	TI3	0.688	0.861
Barrier	BA1	0.66	0.55	0.712
	BA2	0.429	0.706
	BA3	0.607	0.593
	BA4	0.349	0.733
Public health emergency	EE1	0.55	0.56	0.696
	EE2	0.51	0.606
	EE3	0.478	0.648
Intention to use SAVs	IU1	0.635	0.838	0.858
	IU2	0.664	0.831
	IU3	0.64	0.839
	IU4	0.707	0.819
	IU5	0.732	0.812

, the reliability test result shown that Cronbach’s α of the total scale is 0.846, and Cronbach’s α of every subscale range between 0.696 and 0.914. Furthermore, CITC values range between 0.332 and 0.847. Previous studies have shown that if $\alpha$ is higher than 0.8, it indicates high reliability; If $\alpha$ is between 0.7 and 0.8, it indicates good reliability; If $\alpha$ ranges from 0.6 to 0.7, it indicates that the reliability is acceptable. The correction item–total correlation (CITC) should be greater than 0.4, with a minimum acceptance value of 0.3. In this study, all Cronbach’s α greater than 0.6, and all CITC values are greater than 0.3. Therefore, the reliability of the data is high and can be used for further analysis.

(2)

Validity test

The validity test is to test its effectiveness. Validity is usually verified using KMO and Bartlett tests [40]. KMO test coefficient ranges from 0 to 1, and the closer it is to 1, the better the validity of the questionnaire. A range of 0.9 to 1 means perfectly suitable; A range of 0.8 to 0.9 means it is very suitable; Between 0.7 and 0.8 means suitable; Between 0.6 and 0.7 means basically suitable; Between 0.5 and 0.6 means barely suitable; Less than 0.5 means it is not suitable. The p-value corresponding to Bartlett’s spherical test should be less than or equal to 0.01. The formula of KMO is as follows.

$$KMO=\frac{{\displaystyle \sum {\displaystyle {\sum }_{i\ne j}{r}_{ij}^2}}}{{\displaystyle \sum {\displaystyle {\sum }_{i\ne j}{r}_{ij}^2}}+{\displaystyle \sum {\displaystyle {\sum }_{i\ne j}{\alpha }_{ij}^2}}}$$

(2)

where $r_{ij}$ presents simple correlation coefficient; ${\alpha }_{ij.1,2,3,\cdots k}^2$ presents partial correlation coefficient.

KMO and Bartlett tests are performed to verify the validity in this paper. It can be seen from

Table 3. KMO and Bartlett tests.

KMO Value		0.9233
Bartlett Test of Sphericity	Approx. Chi-Square	7778.157
	df	666
	Sig. (p)	0.000

that the KMO value is 0.933, which is greater than 0.9, indicating that the validity is very good. According to the significance of Bartlett test of sphericity, it can also be seen that the significance of this test is infinitely close to 0. The null hypothesis is rejected, so the questionnaire has very good validity.

In total, the results of the reliability and validity test shown that the questionnaire has satisfactory reliability and validity.

4.3. Tructural Model

(1)

Model fit

When building SEM, analyzing the model fit indices is a necessary step, which evaluates whether the fitting results of the model are reasonable. The higher the fitting degree of the model, the higher the degree of agreement between the theoretical model and actual data. If the model fitting is not ideal, the model needs to be corrected until meeting relevant standards. Measurement of model fit indices as shown in

Table 4. Measurement of model fit indices.

Fit Indices	Criteria [41,42]	Value (Initially Model/Before Model Correction)	Model Adaptation Judgment	Result (after Model Revision)	Model Adaptation Judgment
χ2	The smaller, the better	2966.346		752.519
χ2/DF	<3 Ideal, <5 Acceptable	3.106	Acceptable	1.271	Ideal
GFI	>0.8 Acceptable, >0.9 Ideal	0.701	No	0.903	Ideal
AGFI	>0.8 Acceptable, >0.9 Ideal	0.662	No	0.885	Acceptable
CFI	>0.8 Acceptable, >0.9 Ideal	0.793	No	0.978	Ideal
TLI	>0.8 Acceptable, >0.9 Ideal	0.775	No	0.976	Ideal
RMSEA	<0.08 Acceptable, <0.05 Ideal	0.076	Acceptable	0.027	Ideal
RMR	<0.08 Acceptable, <0.05 Ideal	0.06	Acceptable	0.04	Ideal

. The fitting results of the initially established model indicate that the model is unreasonable and needs to be adjusted and corrected. By using the MI (Model Fit Index) indicator adjustment method to modify the model, after model revision, χ² = 752.519; χ²/DF = 1.271; GFI = 0.903; AGFI = 0.885; CFI = 0.978; TLI = 0.976; RMSEA = 0.027; RMR = 0.04. The fitting values of all the model fit indices are within the recommended range (Acceptable and ideal), and the setting of the theoretical model is very ideal. The revised model and path coefficient are shown in Figure 3.

(2)

Path test

A path test is conducted on the structural model, and the results are shown in

Table 5. Path analysis.

Hypotheses	Relationship	Unstandardized Path Coefficient	Standardized Path Coefficient	t-Value	Result Value	Result
H1	Attitude → Intention to use SAVs	0.217	0.225	***	3.363	Supported
H2	Subjective norm → Intention to use SAVs	0.239	0.214	***	3.322	Supported
H3	Perceived behavioral control → Intention to use SAVs	0.232	0.216	0.004	2.894	Supported
H4	Subjective norm → Attitude	0.183	0.157	0.008	2.651	Supported
H5	Subjective norm → Perceived behavioral control	0.204	0.196	0.003	2.978	Supported
H6	Government policy → Attitude	0.223	0.170	0.004	2.891	Supported
H7	Government policy → Perceived behavioral control	0.473	0.403	***	5.880	Supported
H8	Environmental awareness → Attitude	0.174	0.182	***	3.358	Supported
H9	Environmental awareness → Perceived behavioral control	0.057	0.067	0.268	1.109	Not supported
H10	Perceived risk → Attitude	−0.158	−0.175	***	−3.336	Supported
H11	Technical interest → Attitude	0.200	0.165	0.003	2.924	Supported
H12	Technical interest → Perceived behavioral control	0.330	0.305	***	4.682	Supported
H13	Technical interest → Intention to use SAVs	0.077	0.066	0.336	0.962	Not supported
H14	Barrier → Attitude	−0.261	−0.218	***	−3.938	Supported
H15	Barrier → Intention to use SAVs	−0.153	−0.133	0.017	−2.383	Supported
H16	Effects of public health emergency → Intention to use SAVs	0.249	0.167	0.006	2.755	Supported

*** represents p < 0.001.

. The critical ratio (t-value) in Table 5 is usually used to test the significance of the structural path. When the absolute value of $t$ is greater than 1.96, $P$ is less than 0.05; When the absolute value of $t$ is greater than 2.58 and $P$ is less than 0.01; When the absolute value of $t$ is greater than 3.29 and $P$ is less than 0.001.

The results show that except for Hypotheses H9 and H13, all the other 14 Hypotheses are valid. The intention to use SAVs is directly influenced by attitude, subjective norm, perceived behavioral control, barrier and effects of a public health emergency, among which attitude, subjective norm and barriers are the most significant. The intention to use SAVs is indirectly influenced by perceived risk, technological interest, government policy and environmental awareness. Perceived risk and barriers have a great influence on attitudes. Government policy and technological interest have a great influence on perceived behavior control. The barrier has a powerful effect on attitudes. The unstandardized path coefficient between perceived risk and attitude, between barrier and attitude is −0.158 (p < 0.001) and −0.261 (p < 0.001), respectively, indicating that the higher the perceived risk or barrier, the lower the traveler’s attitude to the SAVs. The unstandardized path coefficient between barrier and intention to use SAVs is −0.153 (p = 0.017, p < 0.05), indicating that the higher the barrier, the lower the traveler’s intention to use SAVs.

(3)

Mediation effects

The Mediation effect refers to the explanatory or transitive effect that a variable (mediating variable) plays between the independent variable and the dependent variable [43]. There are three commonly used methods for testing mediation effects, which are Sobel’s method, the distribution of product method and the bootstrap method [43]. The Bootstrap method can be applied to medium to small samples and various mediation effect models. Thus, the Bootstrap method was used to test mediation effects in this paper. When testing the mediation effects, the Bootstrap method can be used to estimate the confidence interval and significance level of the mediation effect. If the confidence interval for the mediating effect does not include 0, then the mediating effect is significant. In addition, if the p-value is less than the significant level (such as 0.05), the same conclusion can be reached.

According to the results of the path test, most of the hypothesis tests are valid, which remains 14 significant paths (Figure 4). After removing the five paths with direct connections, the remaining nine paths with indirect connections were left. To explore whether there is a mediation effect in these nine significant paths, the Bootstrap method is run in AMOS 26.0 for verification. The test is repeated 5000 times, the confidence interval standard is 95% and the bias correction method is selected. In addition, this model is a multi-mediating model and individual analysis can only obtain the total mediating effect result, but not the specific mediating effect. Therefore, the grammar of AMOS 26.0 software is used to assign all the relevant paths, and the non-standardized and standardized specific mediating effects are calculated, respectively.

The results of mediating effects are shown in

Table 6. Mediation effects test.

Pathway Relationships	Indirect Effect Value	Lower	Upper	p	Conclusions
Government policy—Attitude—Intention to use SAVs	0.038	0.006	0.083	0.011	Supported
Environmental awareness—Attitude—Intention to use SAVs	0.041	0.009	0.084	0.004	Supported
Perceived risk—Attitude—Intention to use SAVs	−0.039	−0.079	−0.010	0.003	Supported
Technical interest—Attitude—Intention to use SAVs	0.037	0.007	0.080	0.007	Supported
Barrier—Attitude—Intention to use SAVs	−0.049	−0.095	−0.014	0.002	Supported
Subjective norm—Attitude—Intention to use SAVs	0.035	0.005	0.075	0.016	Supported
Technical interest—Perceived behavioral control—Intention to use SAVs	0.066	0.014	0.136	0.011	Supported
Government policy—Perceived behavioral control—Intention to use SAVs	0.087	0.019	0.161	0.011	Supported
Subjective norm—Perceived behavioral control—Intention to use SAVs	0.042	0.007	0.086	0.015	Supported

. The results show that the upper and lower intervals of the nine Mediation paths do not contain 0, and the p-values are less than the significant level of 0.05, indicating that the intermediary effects are established. That is, attitude is mediating between perceived risk and intention to use SAVs, government policy and intention to use SAVs, environmental awareness and intention to use SAVs, technical interest and intention to use SAVs, barriers and intention to use SAVs, subjective norms and intention to use SAVs. Wherein perceived risk and barrier have negative indirect effects on the intention to use SAVs through the mediating variables of attitude. Furthermore, perceived behavioral control has the mediating effect between government policy and intention to use SAVs, between technical interest and intention to use SAVs and between subjective norm and intention to use SAVs. Through the results of the path test and mediation effects test, the relationships between variables could be described in detail. It should be noted that identified relationships are associations and further longitudinal or experimental studies are needed to establish causal relationships.

5. Discussion and Recommendation

5.1. Discussion

Partial results are consistent with previous studies on SAVs’ use attention. In [15], the greatest receptiveness toward the introduction of SAV shuttles for public use is in part due to stronger perceptions that they will perform well and be easy to adopt. This is consistent with the findings in this paper: the higher the barrier, the lower the traveler’s intention to use SAVs. In [16], SAVs with dynamic ride sharing (DRS) usage intentions were significantly influenced by perceived usefulness, sharing attitude and perception of risk. This is consistent with the findings in this paper: The intention to use SAVs is directly influenced by attitude and perceived behavioral control, and indirectly influenced by perceived risk. In [32], attitude, subject norm and perceived behavioral control are all elective predictors of intention to use SAVs. This is consistent with the findings in this paper: The intention to use SAVs is directly influenced by attitude, subjective norm and perceived behavioral control. In [36], technical interests and attitudes have a significant impact on the timing of SAVs use. This is consistent with the findings in this paper: The intention to use SAVs is directly influenced by attitude, and indirectly influenced by perceived risk. In particular, the results of this study are more comprehensive than those of previous studies.

5.2. Recommendation

According to the analysis results and discussion, the following suggestions on the development of SAVs are put forward in this paper.

(1)

Improving people’s attitudes to SAVs

The government and other relevant departments should strengthen the publicity of SAVs, such as improving the impression of SAVs in terms of safety, environmental protection and social benefits, which could promote the intentions to use SAVs.

(2)

Improving subjective norms for SAVs

Efforts can be made in two aspects, one is to hold regular experience projects of SAVs to increase public participation. The second is to increase the number of SAVs so that more people can see the use of SAVs.

(3)

Increase the perceived usefulness of SAVs

Some measures can be taken to improve the availability of SAVs, such as setting up exclusive lanes for SAVs; SAVs can use bus lanes first during morning and evening rush hours. These measures can reduce the impact of SAVs on the outside world while increasing the public’s intentions to use SAVs. Furthermore, an integrated travel APP can be established for SAVs, carsharing, express cars, hitch cars, maps and so on, into one Application (APP) to reduce unnecessary operations.

(4)

Optimize in-vehicle facilities for SAVs

Due to the context of the COVID-19 pandemic, public health emergency is greatly affecting the intentions to use SAVs. In this regard, the SAVs can be disinfected at regular intervals, ventilated and sprayed with alcohol after each user uses it. At the same time, provide hand sanitizer, disinfectant tissue and disposable gloves in the SAVs to reduce the cross infection of the virus.

6. Conclusions

With the advancement of technology, sharing and autonomous driving will be the two major themes in the future transportation field, and SAVs will combine the two things. When SAVs come to market, they will affect the transportation system, so it is necessary to study people’s intention to use SAVs. Thus, this paper designs and carries out a questionnaire, and the validity and rationality of the questionnaire are tested. Then, the structural equation model constructed by SAVs using intention is fitted and modified, and the path relationship and hypothesis relationship are tested. Some important findings are found: (1) The intention to use SAVs is directly affected by attitude, subjective norm, perceived behavior control, barrier and effects of a public health emergency and indirectly affected by perceived risk, technical interest, government policy and environmental awareness. (2) Attitude has the mediating effect between perceived risk and intention to use SAVs, government policy and intention to use SAVs, environmental awareness and intention to use SAVs, technical interest and intention to use SAVs, barriers and intention to use SAVs and subjective norms and intention to use SAVs. (3) Perceived behavior control has the mediating effect between government policy and intention to use SAVs, between technical interest and intention to use SAVs and between subjective norm and intention to use SAVs. Lastly, according to the influence degree of related influencing factors, the corresponding development recommendations on SAVs development are put forward.

This research could help promote the development of SAVs and intelligent transportation. The results provide travelers’ understanding of SAVs, which is beneficial for improving the theory of travel behavior. Furthermore, SAVs operators can do a good job of corresponding positive publicity and guidance according to the research results in this paper, which could increase the acceptance of SAVs. However, there are some limitations and shortcomings. For example, this paper adopts the method of network survey, and the frequency of elderly people using the Internet is relatively low. The proportion of elderly people is still not satisfied. Furthermore, the proportion of students is too high. The unreasonable age ratio maybe has an impact on the research results. In the future, research on the willingness to use SAVs can be conducted at different age stages. In addition, every city or region is unique, so the geographical scope, cultural or regional variations in attitudes toward SAVs will be taken into account.