Exploring the Role of Attitudinal Factors in Electric Vehicle Timeshare Rentals Adoption

Abstract

Electric vehicle timeshare rentals (EVTRs) have been recognized as promising solutions to growingly severe problems of traffic congestion, air pollution, and insufficient parking spaces. This study aims to explore the factors that affect the adoption of EVTRs. To achieve the research objective, the household survey is conducted to obtain the travelers’ attitudes towards their travel. Ten latent attitudinal factors are extracted based on the technology acceptance model (TAM) and the theory of planned behavior (TPB). The multi-index and multi-cause (MIMIC) method simultaneously estimates the correlations between the attitudinal factors. Two logit models with attitudinal factors or not are constructed to estimate the quantitative relationship between various factors and EVTR adoption. The results show that the accuracy of the mixed logit model with latent attitude variables is better than the binary logit model without attitude latent variables. This indicates that attitude latent factors could be well matched with the traveler’s travel behavior and could better reflect travelers’ travel demand. Perceived comfort, perceived efficient, subjective evaluation, use preference, and use willingness significantly impact EVTR use frequency. The inconvenience in travelers, rental stations, shared vehicles, and use modes have significant negative impacts. Finally, social pressure has no significant impact. Findings provide valuable insights regarding the efficient planning of the EVTR system and allow decision-makers to develop scientific and practical measures of EVTRs.

1. Introduction

Electric vehicle timeshare rentals (EVTRs) are emerging as a quasi-public transportation mode that is more flexible than transit and less expensive than traditional ownership [1,2,3]. EVTRs refer to the user self-service car rental service, which is the main application and development mode of carsharing. Users can flexibly book pickup stations, returning stations, and rental times according to their personal travel needs, and the charges are calculated based on the length of the rental time and the mileage [4,5,6]. In recent years, EVTR has been considered one of the solutions to urban transport issues such as air pollution [7], traffic congestion [8], and parking stress [9]. Therefore, a large number of EVTR projects are widely established around the world. However, the use intensity and scale of EVTRs have been relatively low. Autolib has less than 20% of the vehicle’s daily use time and was officially deactivated in Paris on 1 July 2018 [10,11]. EVTRs have been developed in China for more than 30 years, and there were more than 1000 in China in 2018, but the market penetration of EVTRs was still less than 1% in January 2019 [12]. Exploring the reasons that influence travelers to adopt EVTRs is the key to promoting the development of EVTRs [13].

Existing research mainly focuses on three aspects: characteristics of EVTR adopters, motives for EVTR, and factors influencing EVTR adoption. (1) For the characteristics of EVTR adopters, researchers focus on categorizing which socioeconomic attributes of travelers are more willing to use shared cars, for example, typically younger in age, having a higher social status, more financial lucidity, advanced education, and being more socially forward [14,15,16]. (2) For the motives for EVTR, the main motives travelers have to choose EVTRs are cost-saving and increased convenience, because they are time-saving and freedom-expanding compared with public transport [14,17,18]. With the challenge of sustainability, environmentalism has been one important reason to join EVTRs [19,20]. Additionally, choosing EVTRs to travel can be motivated by changes in personal circumstances, such as changing residence, work, etc. [21,22]. (3) For the factors influencing EVTR adoption, the sociodemographic attributes are the focus of early, extensive, and mature studies on this aspect [23], and most studies show these factors have a significant effect on travelers’ adoption of EVTR [24]. Further, a few studies analyze the influence of psychological factors on the adoption of EVTR [25]. Environmental concern and perceptions of financial saving possibilities positively influence EVTR adoption [13,26,27].

These studies on travelers’ characteristics, motives, and sociodemographic attributes of factors influencing EVTRs have been comprehensive. However, attitudinal determinants of EVTR adoption, use, and impacts, which are also recognized as subjective variables in the travel behavior literature, are not understood very well [26]. For example, existing studies focus on the effects of environmental sustainability, willingness, and social pressures [27,28,29] but ignore other basic psychological desires, e.g., comfortability, safety, sensitivity to time, flexibility, and so on. Therefore, it is necessary to analyze the effects of attitude and psychological factors on EVTR adoption from an overall perspective.

To the best of the authors’ knowledge, there is no research on the relationship between travelers’ attitudinal latent variables and EVTR choice behavior. Travelers’ attitudes are often unobservable and cannot be directly measured. Previously, the multi-index and multi-cause (MIMIC) method was commonly used to identify the latent attitudes towards travel behaviors from a series of attitudinal statements or questions [30,31,32,33,34]. For example, Outwater et al. [33] reported that the MIMIC method provided significantly better performance for identifying the travelers’ latent attitudes from the attitudinal statements compared to the logistic regression model with factor analysis. Shiftan et al. [34] used the MIMIC method to simultaneously identify the relationships among travelers’ attitudes, travel behaviors, and socioeconomic profiles. The latent attitudinal factors in the MIMIC were further used for the transit market segmentation. Thus, the MIMIC method was considered for the EVTR choice behavior analysis in the current study.

The current research aims to address this gap in the literature by empirically investigating the characteristics of users’ social economy, travel behavior, and use attitudes. We surveyed residents of Beijing to obtain a large urban sample. The emphasis on the urban population is warranted because most EVTR services target urban customers [35,36]. Ten latent attitudinal factors are extracted based on the technology acceptance model (TAM) and the theory of planned behavior (TPB). The multi-index and multi-cause (MIMIC) method simultaneously estimates the correlations between the attitudinal factors. Two logit models with attitudinal factors or not are constructed to estimate the quantitative relationship between various factors and EVTR adoption.

The rest of this study is organized as follows. Section 2 provides the methodological framework. The survey and some descriptive data are presented in Section 3. Section 4 reports the estimation results. The last section concludes the research and identifies opportunities for future research.

2. Methodology

2.1. Overall Framework

To explore the adoption of EVTRs, the latent factors are selected based on TAM and TPB. Then, the quantitative relationship between various factors and EVTR demand is explored. The overall framework of the approach is shown in Figure 1.

2.2. Factor System Construction

The technology acceptance model (TAM) and the theory of planned behavior (TPB) are both developed based on rational behavior theory. The homology of the two theories provides the possibility for integration. Chen and Chao [37] integrated the two theories to build models to study the public’s intention to change from motor vehicles to public transport. EVTR is a new way of travel, and EVTR adoption is also a purchase behavior, so it is suitable to use TAM and TPB to study travelers’ use attitudes and EVTR behavior.

• TAM: TAM is mainly used to analyze people’s acceptance of new technologies and products and their influencing factors. The theory holds that people’s use willingness determines adoption and use frequency, which is determined by people’s attitudes towards technology or system and perceived usefulness, while people’s attitudes are influenced by perceived usefulness and perceived ease of use;
• TPB: TPB is a theory explaining the relationship between attitude and behavior in social psychology. It holds that when an individual takes a positive attitude towards a particular behavior, thinks that it conforms to their subjective behavioral norms, and feels that they have mastered the skills and resources to adopt the behavior, they will have strong use willingness, and the probability of actual behavior will increase significantly.

Based on TAM and TPB, we selected ten latent factors (i.e., perceived usefulness (perceived comfort, perceived efficiency), subjective evaluation, use preference, perceived availability (station inconvenience, station inconvenience, use inconvenience, personal inconvenience), social pressure, and use willingness) to analyze travelers’ EVTR use attitude and behavior. The meaning of the attitudinal factors is shown in Table 1. Several hypotheses are constructed to estimate the relationship between attitudinal factors, as follows:

Hypothesis 1.

The higher the use preference is, the stronger the intention to use EVTRs;

Hypothesis 2.

The higher the perceived usefulness is, the stronger the intention to use EVTRs;

Hypothesis 3.

The higher the perceived availability is, the stronger the intention to use EVTRs;

Hypothesis 4.

The higher the social pressure is, the stronger the intention to use EVTRs;

Hypothesis 5.

The higher the subjective evaluation is, the stronger the intention to use EVTRs;

Hypothesis 6.

The perceived usefulness has a mediating effect between subjective evaluation and intentions;

Hypothesis 7.

The perceived availability has a mediating effect between subjective evaluation and intentions;

Hypothesis 8.

The social pressure has a mediating effect between subjective evaluation and intentions;

Hypothesis 9.

Use preference has a mediating effect between behavior attitude and intentions.

Table 1. Description of attitudinal variables.

Latent Variable	Index	Description
Perceived comfort	${\eta }_{AC}$	The respondents’ attitude towards the comfort of EVTRs
Perceived efficiency	${\eta }_{AQ}$	The respondents’ attitude toward the efficiency of EVTRs
Subjective evaluation	${\eta }_{AW}$	Investigators’ overall understanding and evaluation of EVTRs
Use preference	${\eta }_{AD}$	The direct preference of respondents to travel using EVTRs
Station inconvenience	${\eta }_{BS}$	Investigator perceived barriers to EVTR station
Station inconvenience	${\eta }_{BV}$	Investigator perceived barriers to EVTR vehicle
Use inconvenience	${\eta }_{BU}$	Investigator perceived barriers to the use of EVTRs
Personal inconvenience	${\eta }_{BI}$	Investigators’ personal barriers to using EVTRs
Social pressure	${\eta }_{SN}$	Social pressure sensed by investigators when using EVTRs
Use willingness	${\eta }_{CI}$	Investigators’ willingness and possibility to use EVTRs

Table 1. Description of attitudinal variables.

Latent Variable	Index	Description
Perceived comfort	${\eta }_{AC}$	The respondents’ attitude towards the comfort of EVTRs
Perceived efficiency	${\eta }_{AQ}$	The respondents’ attitude toward the efficiency of EVTRs
Subjective evaluation	${\eta }_{AW}$	Investigators’ overall understanding and evaluation of EVTRs
Use preference	${\eta }_{AD}$	The direct preference of respondents to travel using EVTRs
Station inconvenience	${\eta }_{BS}$	Investigator perceived barriers to EVTR station
Station inconvenience	${\eta }_{BV}$	Investigator perceived barriers to EVTR vehicle
Use inconvenience	${\eta }_{BU}$	Investigator perceived barriers to the use of EVTRs
Personal inconvenience	${\eta }_{BI}$	Investigators’ personal barriers to using EVTRs
Social pressure	${\eta }_{SN}$	Social pressure sensed by investigators when using EVTRs
Use willingness	${\eta }_{CI}$	Investigators’ willingness and possibility to use EVTRs

2.3. MIMIC Model

The MIMIC model aims to estimate the coefficients of latent attitudinal factors as functions of statement variables and identify the relationships between attitudinal factors. The MIMIC model used in this study includes statement variables and attitudinal factors. The attitudinal statement variables are directly measured from the survey data, inferring the latent attitudinal factors among the statement variable. The MIMIC model is used to calculate the scores of attitudinal factors for each individual, which are then used in the cluster analysis. The relationship between attitudinal factors in the MIMIC model estimates provides essential information in understanding the travelers’ considerations on the choice of EVTR for travel.

2.4. Mixed Logit Model

The utility of the first mode of travel for travelers $u_i$ is as follows:

$$u_i=a_{i}s+b_i{z}_i+c_i{\eta }_i+{\epsilon }_i$$

(1)

where $s$ is the vector of observable individual attributes of travelers; $z_i$ is the vector of attributes of travel modes, including travel time and expenses; $\eta$ is the vector of unobservable latent variables (i.e., perceived usefulness, perceived ease of use, attitudes, social pressures, perceived behavior control, and use willingness for EVTR); $a_i,b_i,c_i$ are unknown parameter vectors to be estimated; and ${\epsilon }_i$ is a random term.

$$d=\left\{\begin{array}{c}1, u=\max \left(u_i\right)\\ 0, other \end{array}\right.$$

(2)

The structural equation that reflects the relationship between latent variables is defined as:

$$\eta =\mathsf{\Gamma }x+\xi$$

(3)

where $\eta$ is the latent variable vector; $x$ is the exogenous variable vector that affects the latent variable $\eta$; $\mathsf{\Gamma }$ is the unknown parameter matrix to be estimated; and $\xi$ is the measurement error.

$$y=\mathsf{\Lambda }\eta +v$$

(4)

where y is the observable indicator vector of the latent variable $\eta$; Λ is the unknown parameter matrix to be estimated; and v is the error term.

Equations (1) and (2) constitute a discrete selection model, while Equations (3) and (4) constitute a MIMIC model. Therefore, Equation (3) can also be written as follows:

$${\eta }_i={\alpha }_{l1}{S}_{gender}+{\alpha }_{l2}{S}_{age}+{\alpha }_{l3}{S}_{education}+{\alpha }_{l4}{S}_{income}+{\alpha }_{l5}{S}_{car}+{\alpha }_{l6}{S}_{marriage}+\cdots$$

(5)

where $\eta$ is an individual’s latent attitudinal variables such as perceived usefulness, perceived availability, and subjective evaluations; and $\alpha$ is an estimated parameter. It also shows that the potential preference of travelers for car sharing is influenced by the social demographic attributes such as gender, age, and monthly income of travelers.

For the estimation of model parameters, the continuous two-stage estimation method can be used, which is mainly divided into two steps: Step 1, using the maximum likelihood method to estimate the latent variable model, that is, estimating the parameters of the MIMIC model; and Step 2, calculating the adaptation value of the latent variable according to the estimated value of the MIMIC model parameters, and substituting it as an exogenous variable into the discrete selection model for parameter calibration. This paper completes the above parameter estimation steps using AMOS v.24.0 and SPSS v.22.0 software. The conceptual diagram of the model is shown in Figure 2.

3. Data

The data resource was obtained through the household survey that was conducted in multiple traffic analysis zones in the metropolitan district, Beijing, 2021. Considering the complexity of the questionnaire, several confusing questions were redesigned based on the feedback from a pilot survey, including (1) the socioeconomic characteristics; (2) the daily trips information, i.e., trip length and mode choice; and (3) the attitudinal questions towards travels and EVTR. The attitude survey mainly refers to the mature research in TAM and TPB. A total of 868 questionnaires were distributed randomly, and 759 questionnaires were completed. A total of 619 valid questionnaires were obtained after excluding the questionnaires lacking key information. The Likert five-scale method is adopted: “1” means a strong disagreement, while “5” indicates a strong agreement. The summary statistics of latent variables and socioeconomic characteristics are shown in

Table 2. Statistics of latent variables of respondents.

Latent Variables	Observation Variables	Definition	Mean	Std.
${\eta }_{AC}$	AC1	The safety of EVTR is good.	3.04	1.19
	AC2	The comfort of EVTR is good.	2.97	1.56
	AC3	Traveling with a shared EVs is easy and convenient.	2.98	1.14
${\eta }_{AQ}$	AQ1	The accessibility of using EVTR to travel is higher.	3.07	1.63
	AQ2	The convenience of EVTR is good.	3.11	0.82
	AQ3	Using EVTR to travel is fast and saves time.	3.06	1.33
${\eta }_{AW}$	AW1	EVTR is environmentally friendly.	3.97	1.78
	AW2	The economic cost of using EVTR to travel is lower.	3.00	1.15
	AW3	EVTR is conducive to sustainable development.	3.93	1.41
${\eta }_{AD}$	AD1	EVTR can satisfy my travel needs.	3.04	1.64
	AD2	I am satisfied with the travel service of EVTR.	3.02	1.78
	AD3	EVTR is better than other traffic modes.	3.03	1.13
${\eta }_{BS}$	BS1	There are fewer EVTR stations around the origin.	3.91	1.68
	BS2	There are fewer EVTR stations around the destination.	3.91	0.88
	BS3	There are fewer shared EVs available at the station.	3.05	0.80
${\eta }_{BU}$	BU1	It is inconvenient to deal with EVTR traffic violations.	3.03	0.93
	BU2	EVTR is too expensive to use.	3.05	1.32
	BU3	It is inconvenient to rent or return shared vehicles.	3.13	1.29
${\eta }_{BV}$	BV1	The endurance mileage of EVs is too small to meet the travel needs.	3.04	0.85
	BV2	The type of vehicle available for EVTR is too single.	3.10	0.92
	BV3	The driving performance of shared EVs is poor.	3.01	1.04
${\eta }_{BI}$	BI1	I’m not good at driving because of my poor driving skills.	3.05	1.44
	BI2	My economic conditions cannot afford to use EVTR.	2.99	1.39
	BI3	My travel environment is not suitable for EVTR.	3.05	1.27
	BI4	My physical condition is not suitable for driving a car.	2.83	1.18
${\eta }_{CI}$	CI1	My friends think that I should use EVTR to travel.	3.02	1.48
	CI2	My friends often use EVTR to travel.	3.10	0.97
	CI3	I will use EVTR to travel because my friends use it.	3.03	1.03
${\eta }_{SN}$	SN1	Society advocates sharing travel and green travel.	2.99	1.23
	SN2	I’m willing to use EVTR to travel often.	2.96	1.18
	SN3	I will encourage friends to use EVTR.	3.00	1.10
	SN4	I am willing to transition from other traffic modes to EVTR.	3.81	1.40

and

Table 3. Statistics of socioeconomic characteristics of respondents.

Socioeconomic Characteristics	Valid Sample (N = 619)		EVS User (N = 188)		Non-EVS User (N = 431)
	Mean	Std.	Mean	Std.	Mean	Std.
Age (year)	3.15	0.29	3.14	0.28	3.15	0.09
Gender (female-1)	1.47	0.09	1.47	0.14	1.47	0.11
Education (college or higher = 1)	3.05	0.26	3.01	0.16	3.06	0.19
Individual income (1000 USD/year)	3.33	0.21	3.35	0.29	3.32	0.21
Marriage (yes = 1)	0.90	0.05	0.94	0.08	0.88	0.08
Child (yes = 1)	0.70	0.06	0.72	0.07	0.70	0.02
Automobile available (yes = 1)	0.32	0.02	0.39	0.03	0.29	0.02
EVTR Station density (low = 1)	2.33	0.18	3.80	0.25	1.69	0.12
EVTR fleet size (low = 1)	2.31	0.17	3.65	0.28	1.73	0.13
Familiarity with EVTR (low = 1)	2.72	0.24	4.02	0.31	2.16	0.15

, respectively.

To ensure the stability and reliability of the questionnaires and the collected data, Cronbach’s Alpha and Kaiser–Meyer–Olkin (KMO) sample measurements and Bartlett ball test were used to analyze the reliability and validity of the collected data before data analysis. The reliability of Cronbach’s Alpha in each dimension of the questionnaire is greater than 0.7, the factor load is greater than 0.8, the KMO is greater than 0.5, and the Bartlett is less than 0.05, which shows the reliability and validity of the questionnaire are ideal.

4. Results

4.1. Estimating Correlations between Variables Using MIMIC

4.1.1. The Correlation between Attitude Latent Variables

The EFA is first conducted to explore the proper groups of latent factors that explain all the original statement variables. The Cattell’s Scree Plot method is used to decide the number of factors. Based on the preliminary EFA, the CFA extracted 10 latent attitudinal factors from 32 statement variables using AMOS 24.0. The CFA goodness of fit index (GFI) is 0.953, indicating that the given model can reproduce 95.3% of the co-variation in the data. The CFA shows the ten-dimension latent factor structure is appropriate for the data of this study.

Based on prior information on the correlation structures from the CFA in this study and several studies [38,39,40], the model structure shown in Figure 3 is specified in the MIMIC model. The measurement model and structural model are simultaneously estimated. Figure 3 also shows the model estimates the correlations between the latent factors.

The goodness of fit index (GFI) of the MIMIC model is 0.953, the root mean square residual (RMR) is 0.028, and the root mean square error of approximation (RMSEA) is 0.012, indicating that these measures meet the acceptable criteria. The incremental fit index (IFI) = 0.925 and comparative fit index (CFI) = 0.917 exceed the ideal value of 0.9. The adjusted goodness of fit index (AGFI) = 0.944 is above the recommended value of 0.8. These indices indicate that the final MIMIC model is a good fit.

As shown in Figure 3, several attitudinal factors are significantly related to travelers’ use willingness towards EVTRs. Travel obstacles (including station inconvenience, station inconvenience, use inconvenience, and personal inconvenience) have a negative impact on the perception towards EVTRs. The other attitude latent variables have a positive impact on the perception towards EVTRs, as follows:

• The coefficient between use preference and use willingness in the MIMIC model is 0.745, indicating that use preference is positively related to use willingness, which proves that Hypothesis 1 is true;
• The coefficient between perceived usefulness and use willingness in the MIMIC model is positive (perceived comfort 0.512 and perceived efficiency 0.593), indicating that perceived usefulness is positively related to use willingness, which proves that Hypothesis 2 is true;
• The coefficient between travel obstacles and use willingness in the MIMIC model is negative (station inconvenience −0.393, station inconvenience −0.405, use inconvenience −0.501, and personal inconvenience −0.439), indicating that perceived availability is positively related to use willingness, which proves that Hypothesis 3 is true;
• The coefficient between subject norm and use willingness in the MIMIC model is 0.121, indicating that subject norm is positively related to use willingness, which proves that Hypothesis 4 is true;
• The coefficient between subjective evaluation and use willingness in the MIMIC model is 0.878, indicating that subjective evaluation is positively related to use willingness, which proves that Hypothesis 5 is true.

The travelers’ perception towards EVTRs has a significant positive impact on their willingness to use EVTRs. Other attitudinal factors are also related to this willingness. To further study the relationship between latent variables, we examined the mediating effect among potential variables in three steps through the mediating effect test method summarized by Lin W. [41,42], and the direct and indirect effects of attitudinal factors on the willingness to use EVTRs are summarized in

Table 4. Direct and indirect effects between latent attitudinal factors.

Dependent	Independent a	Direct Effect	Indirect Effect	Total Effect
${\eta }_{CI}$	${\eta }_{AW}$	0.878	/ b	0.878
${\eta }_{CI}$	${\eta }_{AD}$	0.745	0.078	0.823
${\eta }_{CI}$	${\eta }_{AQ}$	0.593	0.203	0.796
${\eta }_{CI}$	${\eta }_{AC}$	0.512	0.127	0.639
${\eta }_{CI}$	${\eta }_{BU}$	−0.501	/	−0.501
${\eta }_{CI}$	${\eta }_{BI}$	−0.439	/	−0.439
${\eta }_{CI}$	${\eta }_{BV}$	−0.405	/	−0.405
${\eta }_{CI}$	${\eta }_{BS}$	−0.393	/	−0.393
${\eta }_{CI}$	${\eta }_{SN}$	0.121	0.116	0.278

^a Descending order by total effect. ^b Insignificant at the 90% confidence level.

.

Figure 3. Path relationship between latent attitudinal variables in the MIMIC model.

Figure 3. Path relationship between latent attitudinal variables in the MIMIC model.

According to the system distribution and fitting results of the factors shown in Figure 3 and Table 4, it can be concluded that the latent variables have the following relationship:

• The coefficient of indirect effect between perceived comfort and use willingness is 0.078, indicating that perceived comfort has a positive impact on use willingness through subjective evaluation, which proves that Hypothesis 6 is true;
• The coefficient of indirect effect between perceived efficiency and use willingness is 0.203, indicating that perceived efficiency has a positive impact on use willingness through subjective evaluation, which proves that Hypothesis 7 is true;
• The coefficient of indirect effect between social pressure and use willingness is 0.127, indicating that social pressure positively impacts use willingness through subjective evaluation, which proves that Hypothesis 8 is true;
• The coefficient of indirect effect between use preference and use willingness is 0.116, indicating that use preference positively impacts use willingness through subjective evaluation, which proves that Hypothesis 8 is true.

4.1.2. The Correlation between Latent Attitude Variables and Exogenous Variables

In order to analyze the impact of latent attitudinal variables and exogenous variables, such as social demographic attributes, travel characteristics, and travel environment, on EVTR use demand in the next section, this section makes a detailed analysis of the quantitative relationship between exogenous variables and latent variables, and latent variables and their observed variables by constructing a MIMIC model. Since this section studies the direct relationship between latent variables and exogenous variables, and latent variables and their observed variables, the interaction between the latent variables of each attitude is not considered. After adjusting the structure between variables repeatedly, and screening and removing the correlations that are not significant, the correlation among variables is finally obtained, as shown in Figure 4.

The relationships between exogenous and latent variables were obtained from Figure 4, as shown in Table 5. All the coefficients in the MIMIC model are significant at a 95% confidence level. Gender (X_gen) and age (X_age) negatively impact personal obstacles, and education positively impacts personal obstacles, which indicates that all kinds of people have some inconvenience to the use of EVTRs. Income (X_inc) positively impacts travelers’ use preference, which might be related to the personalized travel requirements of high-income people. Travelers with children (X_chi) are more concerned with the comfort of travel. Travelers with private cars (X_car) tend to use private cars and are less interested in EVTRs. The station density (X_sta) and fleet size (X_veh) indicate the service intensity of EVTRs. The higher the service intensity is, the higher the service level perceived by the travelers. Therefore, expanding the infrastructure construction of EVTRs is fundamental to promote the development of EVTR use demand. The level of understanding of EVTRs (X_kno) has a negative impact on use obstacles and a positive effect on the social pressure, which indicates that improving the promotion might increase travelers’ acceptance of EVTRs. All car restriction measures (XIM1-XIM5) positively impact use willingness related to the competitive relationship between cars and EVTRs.

4.1.3. The Correlation between Latent Attitude Variables and Statement Variables

The relationships between latent variables and statement variables were obtained from Figure 3. Ten attitude factors with the factor loading, standard error, and t-statistics from the CFA are shown in Table 6. Each latent factor is a linear combination of several statement variables. For example, factor 1 includes three statement variables (AC1–AC3) related to the respondents’ perceived comfort. Thus, factor 1 reflects the need for comfort. Other attitudinal factors can be interpreted similarly: factor 2 reflects the respondents’ convenient attitude (AQ1–AQ3), factor 3 is related to subjective evaluation (AW1–AW3), factor 4 is identified as the use preference (AD1–AD3), factor 5 is the station inconvenience (BS1–BS3), factor 6 reflects the station inconvenience (BV1–BV3), factor 7 is related to the use inconvenience (BU1–BU3), factor 8 reflects the respondents’ personal inconvenience (BI1–BI4), factor 9 pertains to the respondents’ social pressure (SN1–SN4), and factor 8 is a measurement of the travelers’ use willingness (CI1–CI3).

Figure 4. The relationship between exogenous variables and latent variables.

Figure 4. The relationship between exogenous variables and latent variables.

Table 5. The relationship between exogenous variables and latent variables.

	${\mathit{\eta }}_{\mathit{A}\mathit{C}}$	${\mathit{\eta }}_{\mathit{A}\mathit{Q}}$	${\mathit{\eta }}_{\mathit{A}\mathit{W}}$	${\mathit{\eta }}_{\mathit{A}\mathit{D}}$	${\mathit{\eta }}_{\mathit{B}\mathit{S}}$	${\mathit{\eta }}_{\mathit{B}\mathit{V}}$	${\mathit{\eta }}_{\mathit{B}\mathit{U}}$	${\mathit{\eta }}_{\mathit{B}\mathit{I}}$	${\mathit{\eta }}_{\mathit{S}\mathit{N}}$	${\mathit{\eta }}_{\mathit{C}\mathit{I}}$
$X_{gen}$								−0.165 (2.122)
$X_{age}$								−0.127 (5.715)
$X_{inc}$				0.132 (3.072)
$X_{edu}$								0.145 (4.323)
$X_{chi}$	0.246 (2.721)
$X_{car}$				−0.313 (8.072)
$X_{sta}$		0.352 (4.602)	0.242 (6.045)	0.293 (2.883)	−0.176 (11.773)
$X_{veh}$		0.323 (6.101)	0.112 (2.262)	0.431 (12.292)		−0.145 (3.812)
$X_{kno}$							−0.115 (5.429)		0.213 (3.624)
$X_{IM1}$										0.244 (2.722)
$X_{IM2}$										0.123 (4.116)
$X_{IM3}$										0.122 (3.982)
$X_{IM4}$										0.112 (5.752)
$X_{IM5}$										0.157 (2.385)

Table 5. The relationship between exogenous variables and latent variables.

	${\mathit{\eta }}_{\mathit{A}\mathit{C}}$	${\mathit{\eta }}_{\mathit{A}\mathit{Q}}$	${\mathit{\eta }}_{\mathit{A}\mathit{W}}$	${\mathit{\eta }}_{\mathit{A}\mathit{D}}$	${\mathit{\eta }}_{\mathit{B}\mathit{S}}$	${\mathit{\eta }}_{\mathit{B}\mathit{V}}$	${\mathit{\eta }}_{\mathit{B}\mathit{U}}$	${\mathit{\eta }}_{\mathit{B}\mathit{I}}$	${\mathit{\eta }}_{\mathit{S}\mathit{N}}$	${\mathit{\eta }}_{\mathit{C}\mathit{I}}$
$X_{gen}$								−0.165 (2.122)
$X_{age}$								−0.127 (5.715)
$X_{inc}$				0.132 (3.072)
$X_{edu}$								0.145 (4.323)
$X_{chi}$	0.246 (2.721)
$X_{car}$				−0.313 (8.072)
$X_{sta}$		0.352 (4.602)	0.242 (6.045)	0.293 (2.883)	−0.176 (11.773)
$X_{veh}$		0.323 (6.101)	0.112 (2.262)	0.431 (12.292)		−0.145 (3.812)
$X_{kno}$							−0.115 (5.429)		0.213 (3.624)
$X_{IM1}$										0.244 (2.722)
$X_{IM2}$										0.123 (4.116)
$X_{IM3}$										0.122 (3.982)
$X_{IM4}$										0.112 (5.752)
$X_{IM5}$										0.157 (2.385)

Table 6. Latent attitudinal factors in the CFA.

Latent Factor	Statement Variable	Estimate	S.E.	C.R.	P
${\eta }_{AC}$	AC1	1
	AC2	1.017	0.140	9.597	0.009
	AC3	2.102	0.275	6.642	0.032
${\eta }_{AQ}$	AQ1	1
	AQ2	0.778	0.109	10.767	0.006
	AQ3	0.796	0.096	7.765	0.006
${\eta }_{AW}$	AW1	1
	AW2	0.246	0.029	5.978	0.021
	AW3	0.428	0.045	9.005	0.012
${\eta }_{AD}$	AD1	1
	AD2	0.658	0.084	5.610	0.009
	AD3	0.847	0.091	4.668	0.008
${\eta }_{BS}$	BS1	1
	BS2	0.340	0.041	10.433	0.031
	BS3	0.284	0.042	6.422	0.155
${\eta }_{BV}$	BV1	1
	BV2	0.677	0.079	2.946	0.033
	BV3	0.755	0.108	4.927	0.041
${\eta }_{BU}$	BU1	1
	BU2	4.373	0.468	5.888	0.038
	BU3	0.845	0.109	7.339	0.019
${\eta }_{BI}$	BI1	1
	BI2	0.722	0.086	8.173	0.030
	BI3	0.622	0.064	5.119	0.034
	BI4	0.595	0.068	7.068	0.039
${\eta }_{SN}$	SN1	1
	SN2	0.429	0.053	10.156	0.045
	SN3	0.351	0.046	3.151	0.032
	SN4	0.220	0.027	6.027	0.012
${\eta }_{CI}$	CI1	1
	CI2	1.566	0.230	8.830	0.005
	CI3	1.441	0.209	7.877	0.004

Table 6. Latent attitudinal factors in the CFA.

Latent Factor	Statement Variable	Estimate	S.E.	C.R.	P
${\eta }_{AC}$	AC1	1
	AC2	1.017	0.140	9.597	0.009
	AC3	2.102	0.275	6.642	0.032
${\eta }_{AQ}$	AQ1	1
	AQ2	0.778	0.109	10.767	0.006
	AQ3	0.796	0.096	7.765	0.006
${\eta }_{AW}$	AW1	1
	AW2	0.246	0.029	5.978	0.021
	AW3	0.428	0.045	9.005	0.012
${\eta }_{AD}$	AD1	1
	AD2	0.658	0.084	5.610	0.009
	AD3	0.847	0.091	4.668	0.008
${\eta }_{BS}$	BS1	1
	BS2	0.340	0.041	10.433	0.031
	BS3	0.284	0.042	6.422	0.155
${\eta }_{BV}$	BV1	1
	BV2	0.677	0.079	2.946	0.033
	BV3	0.755	0.108	4.927	0.041
${\eta }_{BU}$	BU1	1
	BU2	4.373	0.468	5.888	0.038
	BU3	0.845	0.109	7.339	0.019
${\eta }_{BI}$	BI1	1
	BI2	0.722	0.086	8.173	0.030
	BI3	0.622	0.064	5.119	0.034
	BI4	0.595	0.068	7.068	0.039
${\eta }_{SN}$	SN1	1
	SN2	0.429	0.053	10.156	0.045
	SN3	0.351	0.046	3.151	0.032
	SN4	0.220	0.027	6.027	0.012
${\eta }_{CI}$	CI1	1
	CI2	1.566	0.230	8.830	0.005
	CI3	1.441	0.209	7.877	0.004

Thus, according to Table 6, the adaptation value of each latent variable is obtained by the following Equations (6)–(15).

$${\eta }_{AC}=0.321\ast \mathrm{AC}1+0.326\ast \mathrm{AC}2+0.353\ast AC3$$

(6)

$${\eta }_{AQ}=0.388\ast \mathrm{AQ}1+0.303\ast \mathrm{AQ}2+0.309\ast AQ3$$

(7)

$${\eta }_{AW}=0.775\ast \mathrm{AW}1+0.032\ast \mathrm{AW}2+0.193\ast AW3$$

(8)

$${\eta }_{AD}=0.399\ast \mathrm{AD}1+0.263\ast \mathrm{AD}2+0.338\ast AD3$$

(9)

$${\eta }_{BS}=0.616\ast \mathrm{BS}1+0.209\ast \mathrm{BS}2+0.175\ast BS3$$

(10)

$${\eta }_{BV}=0.311\ast \mathrm{BV}1+0.426\ast \mathrm{BV}2+0.263\ast BV3$$

(11)

$${\eta }_{BU}=0.411\ast \mathrm{BU}1+0.278\ast \mathrm{BU}2+0.311\ast BU3$$

(12)

$${\eta }_{BI}=0.345\ast \mathrm{BI}1+0.249\ast \mathrm{BI}2+0.215\ast BI3+0.191\ast BI4$$

(13)

$${\eta }_{SN}=0.214\ast \mathrm{SN}1+0.500\ast \mathrm{SN}2+0.176\ast SN3+0.110\ast SN4$$

(14)

$${\eta }_{CI}=0.249\ast \mathrm{CI}1+0.391\ast \mathrm{CI}2+0.360\ast CI3$$

(15)

4.2. EVTR Travel Choice Prediction Using Two Mixed Logit Models

In order to explore the influencing factors of EVTR adoption and analyze the explanatory ability of exogenous variables and potential variables to shared car travel demand, the binary logit model without attitudinal variables and the mixed logit model with attitudinal variables are constructed, respectively.

4.2.1. EVTR Travel Selection Model without Latent Variables

The selection model without latent variables is constructed using the binary logit model, which takes the social population attributes, travel characteristics, travel environment, and car restriction measures as independent variables, and whether travelers choose EVTR travel as an independent variable. The SPSS is used to calibrate the model. After multiple modeling and screening to remove the independent variables that have no significant influence on the dependent variable, the parameter estimation and test results of the ordered logical model are finally established in Table 7. The number of samples was 619. The McFadden R2 = 0.273 > 0.2(41) shows that the precision of the ordered logic model is acceptable.

Gender significantly impacts the choice of shared electric vehicles (p = 0.038 < 0.05). The estimated value of the parameters is 0.052, and Or = exp(0.052) = 1.053 > 1, indicating that women are more willing to choose EVTRs to travel, which may be related to women’s higher personalization demand for EVTR travel. The age coefficient is −0.019, Or = exp(−0.019) = 0.981, indicating that the probability of choosing EVTRs to travel will decrease for each level of age increase in the decision-making process, which may be related to the fixed use habits of older passengers. Regarding education and income, the regression coefficients were 0.081 and 0.029, respectively. This shows that the higher the education level and income, the higher the probability of choosing EVTR to travel, which may be related to their higher acceptance of new things and higher demand for personalized services.

Whether the respondent owns a private car has the most significant impact on the use of EVTR. Its coefficient is −0.468, which indicates that respondents with private cars are more reluctant to use EVTRs. The number of vehicles available in the station and the transfer time between the origin/destination and rental station indicate the service strength of EVTRs. Increasing the number of vehicles available will help to improve travelers’ willingness to use EVTRs. Additionally, the longer the transfer time is, the weaker the willingness to use is. The level of understanding (EVTRs) is positively correlated with the demand for EVTRs, which indicates that travelers’ lack of understanding of EVTRs is an important factor limiting the development of EVTR demand. All car restriction measures are positively related to the use of EVTRs, which is related to the competitive relationship between EVTRs and cars (such as taxis, private cars, etc.).

Table 7. Parameter fitting results of binary logit model without attitudinal variables.

Variable	B	S.E.	Sig.	Exp (B)
Constant	−1.600	1.052	0.013	0.202
$X_{gen}$	0.052	0.180	0.038	1.053
$X_{age}$	−0.019	0.087	0.043	0.981
$X_{inc}$	0.029	0.100	0.027	1.029
$X_{edu}$	0.081	0.110	0.046	1.084
$X_{chi}$	0.154	0.199	0.064	1.166
$X_{car}$	−0.486	0.186	0.030	0.615
$X_{sta}$	−0.121	0.103	0.023	0.886
$X_{veh}$	0.012	0.081	0.028	1.012
$X_{kno}$	0.314	0.084	0.018	1.369
$X_{IM1}$	0.090	0.110	0.041	1.094
$X_{IM2}$	0.075	0.110	0.044	1.078
$X_{IM3}$	0.169	0.109	0.012	1.184
$X_{IM4}$	0.082	0.111	0.046	1.085
$X_{IM5}$	0.152	0.140	0.017	1.164
n	619
Log likelihood	742.995
McFadden’s R2	0.273

Table 7. Parameter fitting results of binary logit model without attitudinal variables.

Variable	B	S.E.	Sig.	Exp (B)
Constant	−1.600	1.052	0.013	0.202
$X_{gen}$	0.052	0.180	0.038	1.053
$X_{age}$	−0.019	0.087	0.043	0.981
$X_{inc}$	0.029	0.100	0.027	1.029
$X_{edu}$	0.081	0.110	0.046	1.084
$X_{chi}$	0.154	0.199	0.064	1.166
$X_{car}$	−0.486	0.186	0.030	0.615
$X_{sta}$	−0.121	0.103	0.023	0.886
$X_{veh}$	0.012	0.081	0.028	1.012
$X_{kno}$	0.314	0.084	0.018	1.369
$X_{IM1}$	0.090	0.110	0.041	1.094
$X_{IM2}$	0.075	0.110	0.044	1.078
$X_{IM3}$	0.169	0.109	0.012	1.184
$X_{IM4}$	0.082	0.111	0.046	1.085
$X_{IM5}$	0.152	0.140	0.017	1.164
n	619
Log likelihood	742.995
McFadden’s R2	0.273

4.2.2. EVTR Travel Selection Model with Latent Variables

The selection model with latent variables is constructed using the mixed logit model, which takes the social population attributes, travel characteristics, travel environment, and attitude latent variable as independent variables, and whether travelers choose EVTR travel as an independent variable. The SPSS is used to calibrate the model. After multiple modeling and screening to remove the independent variables that have no significant influence on the dependent variable, the parameter estimation and test results of the ordered logical model are finally established in

Table 8. Parameter fitting results of logit model without latent attitude variables.

Variable	B	S.E.	Sig.	Exp (B)
Constant	−2.700	1.611	0.014	0.067
$X_{gen}$	−0.023	0.184	0.090	0.977
$X_{age}$	−0.032	0.089	0.072	0.969
$X_{inc}$	0.018	0.102	0.186	1.018
$X_{edu}$	−0.076	0.112	0.060	0.927
$X_{chi}$	0.096	0.205	0.064	1.101
$X_{car}$	0.499	0.19	0.009	1.647
$X_{sta}$	−0.079	0.103	0.043	0.924
$X_{veh}$	−0.033	0.083	0.039	0.968
$X_{kno}$	−0.144	0.086	0.020	0.866
$X_{IM1}$	0.057	0.133	0.006	1.059
$X_{IM2}$	0.057	0.122	0.016	1.059
$X_{IM3}$	0.136	0.112	0.022	1.146
$X_{IM4}$	−0.097	0.114	0.040	0.908
$X_{IM5}$	0.117	0.114	0.030	1.124
${\eta }_{AC}$	0.387	0.181	0.032	1.473
${\eta }_{AQ}$	0.211	0.173	0.037	1.235
${\eta }_{AW}$	0.165	0.163	0.031	1.179
${\eta }_{AD}$	0.174	0.151	0.025	0.840
${\eta }_{BS}$	−0.005	0.137	0.027	0.995
${\eta }_{BV}$	−0.121	0.174	0.049	0.886
${\eta }_{BU}$	−0.277	0.187	0.014	0.758
${\eta }_{BI}$	−0.263	0.179	0.014	1.301
${\eta }_{SN}$	0.162	0.178	0.086	1.176
${\eta }_{CI}$	0.023	0.184	0.040	0.977
n	619
Log likelihood	727.120
McFadden’s R2	0.341

. The number of samples was 619. The McFadden R2 = 0.341 > 0.2 [43], which shows that the precision of the mixed logit model is acceptable.

For the observed variables, age, gender, education level, individual income, marriage, and child have no significant impact on EVTR adoption. Further, the automobile available, EVTR station density, fleet size, and familiarity significantly impact EVTR adoption. These findings suggest that travel characteristics influence travelers’ adoption of EVTRs more than other socioeconomic attributes. From the calibration results of model parameters containing latent variables, it can be found that not all latent variables have significant effects. Perceived comfort (0.387), perceived efficiency (0.211), subjective evaluation (0.165), use preference (0.174), and use willingness (0.023) have significant positive impacts on EVTR use demand. Station inconvenience (−0.005), station inconvenience (−0.121), use inconvenience (−0.277), and personal inconvenience (−0.263) have significant negative impacts. Social pressure (Sig. = 0.086 > 0.5) has no significant impact. Therefore, in addition to improving the coverage of rental stations and fleet size, increasing the publicity and promotion of EVTRs, enhancing the perception level of travelers on the perceived usefulness and availability of EVTRs, and reducing the perception level of travelers on travel inconvenience are essential contents to promote the development of shared car demand.

Lastly, we compare the results of the respondents’ choice obtained from the survey data and the logit model fitting and prediction. The goodness of fit of the binary logit model without latent attitude variables and mixed logit model with attitude latent variables is 0.237 and 0.341, respectively. In practice, it is considered that the accuracy of the model is higher when the value of R2 is above 0.2. Thus, this model could meet the reliability requirements of model fitting, and the analysis results of the above factors could be acceptable. Additionally, compared with the binary logit model, the goodness of fit (0.341 > 0.273) of the mixed logit model with attitude latent variables is better. This finding indicates that attitude latent factors could be well matched with the travelers’ travel behavior and could better reflect travelers’ travel demand.

5. Conclusions

This study explored the factors that affect travelers’ EVTR use demand. Individuals’ attitudes toward electric vehicle timeshare rental (EVTR) travel were extracted from the attitudinal questions in the household survey conducted in Beijing, China. A MIMIC model was used to identify the latent attitudinal factors and capture their interrelationships. Two logit models with attitudinal factors or not were constructed to estimate the quantitative relationship between various factors and EVTR adoption.

The results show that the accuracy of the mixed logit model with latent attitude variables is better than that of the binary logit model without attitude latent variables. This indicates that attitude latent factors could be well matched with the adoption behavior and could better reflect travelers’ travel demand. Perceived comfort, perceived efficiency, subjective evaluation, use preference, and use willingness significantly impact EVTR use demand. Station, vehicle, use, and personal inconvenience have significant negative impacts. Social pressure has no significant impact.

This study aimed to test whether attitudinal factors have a significant impact on EVTR adoption. The extracted attitudinal factors may not be complete. In the future, more attitude factors (e.g., environmental awareness) can be selected to explore their impact on the adoption and use frequency of EVTRs. Additionally, this study roughly divides travelers into EVTR users and non-users. However, public acceptance of an emerging concept requires a complex process that usually includes multiple cognitive stages, such as pre-awareness, awareness, preparation, action, and maintenance. Future research can explore the influence of attitudinal factors on different cognitive stages and stage transition behaviors of EVTR.

Due to the different cultures and environments in travel activities between China and other countries, travelers’ choice behavior regarding EVTR explored in this study may not be directly transferrable to other parts of the world. Other parts of the world should explore attitudes and other factors that affect EVTR adoption in combination with local culture, adaptability of electric vehicles, charging facilities, rental and sharing approaches, etc. Then, policies can be proposed according to the particular characteristics of travel environments and traffic facilities. Such studies can provide grounds for communities to strive for better carsharing facilities, amenities, and support.