Staged Transition Process from Driving to Bicycling Based on the Effects of Latent Variables

Abstract

With the worsening of urban traffic congestion and environmental pollution, cities around the world have become conscious of the importance of the bicycle. The key to promoting the transition from driving to bicycling for short-distance travel is to understand the transition process and the effect of latent variables. This paper divides the transition from driving to bicycling into five stages according to the transtheoretical model of behavior change and analyzes the staged transition process in terms of influences from latent variables. First, based on survey data from ten communities in Beijing, China, a multiple-indicator and multiple-cause (MIMIC) model of transition intention was developed, and then the influential relationships were analyzed between latent variables such as riding attitude and the causal relationships between latent variables and exogenous variables such as infrastructure characteristics. Subsequently, four transition behavior hybrid-choice models for different stages were developed to describe the phased transition process. Finally, the key influencing factors in different stages were identified. The findings reveal the mechanism of the transition from driving to bicycling, which can support policy makers’ decisions at the level of bicycle travel promotion, guidance and facility planning.

1. Introduction

As a green transportation mode, bicycling plays an important role in easing traffic congestion, promoting energy savings and emission reduction, and so on [1]. Bicycling is also a lifestyle that enhances the life quality of residents. However, the number of cyclists was continually decreasing in most cities in China until the advent of bike sharing in 2019, which slowly brought the percentage back up [2,3,4]. Cherry et al. stated that e-bikes are efficiently replacing many urban car trips [3,4,5]. Yet since the right-of-way of e-bikes in Chinese cities is not yet clear (whether they are non-motorized or motorized), here we only discuss the role of bicycles as a substitute for cars. The reason for this decline lies in the transfer from bicycles to motor vehicles for short-distance travel that is suitable for bicycling. Therefore, the key to bicycling development is fostering the transition from driving to bicycling for short-distance travel.

The aim of transitioning motorists to bicycles has become a trend all over the world. Many cities in China have adopted several measures to improve bicycle infrastructure. However, these measures have not produced good results. Forsyth and Krizek [6] analyzed more than 300 empirical studies and found that community design, infrastructure, programs, pricing, and combined strategies all have an impact on the improvement of bicycle transportation. They also debunked common misunderstandings and reported several findings: (1) In the absence of information, people often overestimate the effects of proposed interventions. (2) Some promising interventions (such as preference, education, and marketing) that deserve attention have not been evaluated well enough to conclude that they really work. (3) Some interventions and rules of thumb (such as built bicycle roads) that may seem obvious are not backed by research evidence. Therefore, effective measures can only be developed if the transition process from driving to bicycling and the influence of latent variables are understood.

This paper is organized as follows. First, the theory of planned behavior (TPB) and the transtheoretical model (TTM) are introduced as conceptual models, and a literature review is presented. Second, a survey that was conducted in ten communities in Beijing and the resulting data are presented. Third, the methodological approach of the paper is described, including a multiple-indicators and multiple-causes (MIMIC) model and a hybrid-choice model (HCM). Finally, a transition intention MIMIC model is established to exhibit the causal relationships between exogenous variables and latent variables. A transition behavior hybrid-choice model is established to depict the staged transition process from driving to bicycling, and the key influence factors for different stages are reported.

2. Conceptual Model and Literature Review

Previous works in the literature have mostly studied bicycling from the perspective of travel mode choice. These studies have explored whether travelers choose bicycling or driving. Studies on travel mode choice have changed their methods from aggregate models to disaggregate models. Theories applied in these works include the random utility theory, the theory of planned behavior, and prospect theory.

Early studies on travel mode choice of bicycling used discrete-choice models to analyze the influences of personal characteristics, travel characteristics, and infrastructure characteristics. Scholars found that the decision to travel by bicycle was related to gender [7], age [8], income [9], vehicle ownership [10], and family composition [11]. However, the influence was different in different situations. A consensus has emerged that travel distance is one of the most important influence factors [12], but whether the influence of travel distance is weakened for short-distance travel is unclear. Previous studies considered the lack of a bicycle infrastructure to be the main barrier [13]. Different kinds of bicycle infrastructure have significantly different effects [14]. Hong et al. found that safe cycle paths could encourage people to cycle more [15]. Chapman and Larsson emphasized the importance of the quality of cycling infrastructure in winter to the attractiveness of cycling [16]. Saneinejad et al. modeled the impact of weather conditions on the use of active transport [17]. Early studies explored the influence of many exogenous variables. However, the latent psychological factor was regarded as a black box. These travel mode choice models cannot explain the relationships between attitude, preference, perception, and behavior.

With social economy development, the factors that affect travel mode choice have become more and more complicated. Thus, scholars have started to pay attention to latent factors. In recent years, TPB was widely used to study travel mode choice. From the perspective of psychology, Verplanken [18] and Xiong [19] analyzed the influences of attitude, emotion, norm, and preference. They found that latent factors affected behavior. McFadden [20] quantified latent factors and introduced latent factors into a discrete-choice model. Recker [21] found that the influence of attitude was more significant than travel time and cost. Paulssen [22] and Jing [23] found that a model that introduced latent factors fitted the data more precisely. Theodore [24] considered latent factors to have a more significant influence on bicycling than on motorized travel. Therefore, it is necessary to consider the effect of latent variables.

With the deterioration of urban traffic congestion, increasing numbers of scholars have begun to study the transition from motorized travel to non-motorized travel. Idris et al. [25] analyzed the possibility of transitioning from cars to buses for different levels of bus service. Yang and Qian [26] used evolutionary game theory to establish a transition model from car to bus. They found that it is necessary to reduce the opportunity benefits to travelers who do not intend to transfer.

Emerging from the field of psychology, Prochaska [27] put forward the transtheoretical model of behavior change to describe the process of individual behavior change. TTM has been widely applied in areas such as smoking cessation, alcohol consumption, dietary behavior, and exercise behavior [28]. This model is based on the premise that individuals move through a series of five stages when voluntarily changing; these stages are pre-contemplation (PC), contemplation (C), preparation (PA), action (A), and maintenance (M). Individuals in the pre-contemplation stage do not want to change their behavior in the future. Individuals in the contemplation stage are considering trying to overcome a behavior but have made no commitment to take action. Individuals in the preparation stage intend to take action in the future. Individuals in the action stage have recently initiated a modification of their behavior. Individuals in the maintenance stage have sustained an action.

Several studies have applied TTM to survey data and analyzed individuals’ readiness to use alternative transportation modes [29,30]. These studies suggest that attitudinal and geospatial factors are related to the stages of change. Bamberg established a theoretical model that combines elements of TTM, the norm-activation model [31], and TPB [32]. This model was applied to a car-use reduction intervention, and it described in detail the psychological factors contributing to stage progression. Several studies have applied TTM to analyze bicycling behavior [33,34]. These studies indicate that TTM is a useful model for designing and evaluating intervention strategies that aim to increase the use of bicycles. TTM has been applied for identifying barriers to and motivations for bicycling in Dar-es-salaam, Tanzania [35]. For each sequential stage pair, a series of binary logistic regression models was established to understand how differences in barriers and motivations among individuals explain their membership in different stages of change. These models can be used to examine the effect of barriers, motivators, and policy-related interventions. Thigpen et al. [36] used TTM to explore opportunities to increase bicycle commuting to the UC Davis campus in Davis, California. They constructed a Bayesian multilevel ordinal logistic regression model to understand the relationships between stage membership and sociodemographic characteristics, travel attributes, and travel attitudes.

On the basis of TPB, this paper proposes latent variables that affect bicycling. On the basis of TTM, this paper studies the staged transition process from driving to bicycling, and the effects of latent variables are explored.

3. Data Collection

3.1. Latent Variety

TPB explains the decision-making of individuals from the perspective of information processing. Attitude indicates individuals’ negative or positive assessments of some behavior; a subjective norm refers to the detailed feeling from other people on decision making; and perceived behavior control is similar to the concept of self-control over one’s own behavior. The theory indicates that behavioral intention is affected by attitude, subjective norm, and perceived behavior control. However, Ajzen [37] showed that behavioral intention was also affected by some other variables. Especially for bicycling, the influence of latent variables is larger, and the variety and structure of latent variables are more complex [24]. Frater et al. [38] used TPB and the prototype/willingness model to predict the intention to cycle to school. Their result showed that bicycling intention will improve through change at multiple levels targeting individuals, social environments, physical environments, and policies. Stark et al. [39] used structural equation modeling to analyze the effects of an intervention based on TPB. They found that the intervention was effective in changing attitudes, perceived behavioral control, and intentions to use non-motorized travel modes more and cars less.

On the basis of TPB, this paper expands the latent variables according to the characteristics of bicycling. Previous studies have shown that bicycling attitude differs from bicycling preference. Bicycling attitude refers to the safety, convenience, and comfort of bicycling as perceived by travelers [40,41]. Bicycling preference refers to a traveler’s like of bicycling [42,43]. Trapp [42] found that travelers who like bicycling have a stronger bicycling intention. Xing et al. [43] reported that people who have a positive attitude toward bicycling do not always like bicycling. Therefore, this study considered bicycling preference as a latent variable in addition to bicycling attitude. Moreover, Nkurunziza [35] found that the influence of perceived behavior control on bicycling was more complex. Perceived behavior control was composed of physical determinants and infrastructure barriers. Therefore, this study divided perceived behavior control into physical determinants and infrastructure barriers.

3.2. Variable Selection

Bicycling is affected by personal characteristics, travel characteristics, infrastructure characteristics, and psychological factors [42]. Therefore, this study obtained not only travelers’ personal characteristics, travel characteristics, and latent psychological factors but also infrastructure characteristics in communities.

Latent variables were determined on the basis of TPB and included infrastructure barriers, physical determinants, bicycling attitude, bicycling preference, and subjective norm.

For exogenous variables, this study considered personal characteristics, travel characteristics, infrastructure characteristics, and car-restrictive measures. Infrastructure characteristics included the bicycle level of service, the density of the bicycle road network, and the accessibility of amenities. The variables analyzed in this study are listed in

Table 1. Variables and Definitions.

Variable	Index		Abbreviations	Assignment
Personal characteristics	Gender		$X_{gen}$	0 = Man, 1 = Woman
	Age		$X_{age}$	(12~79)
	Income		$X_{inc}$	1 = below CNY 1500, 2 = CNY 1501~3000, 3 = CNY 3001~5000, 4 = CNY 5001~8000, 5 = CNY 8001~15,000, 6 = above CNY 15,000
	Educational background		$X_{edu}$	1 = Primary school, 2 = High school, 3 = University, 4 = Master’s or doctorate
	Number of children		$X_{chi}$	0 = 0, 1 = 1 or more
Travel characteristics	Travel distance		$X_{dis}$	1 = Below 1 km, 2 = 1~2 km, 3 = 2~3 km, 4 = 3~4 km, 5 = 4~5 km
	Usability of a car		$X_{car}$	0 = Not usable, 1 = Usable
	Usability of a bicycle		$X_{bic}$	0 = Not usable, 1 = Usable
Infrastructure characteristics	Bicycle level of service		$X_{los}$	BLOS within 1.5 km of home: 1 = Level 5, 2 = Level 4, 3 = Level 3, 4 = Level 2, 5 = Level 1
	Density of bicycle road network		$X_{den}$	Density of bicycle road network within 1.5 km of home
	Accessibility of amenity		$X_{acc}$	Accessibility of amenities within 1.5 km of home
Car restrictive measures	Car speed limit		$X_{IM1}$	Traveler’s transition intention after a car speed limit is imposed
	Increasing parking fees		$X_{IM2}$	Traveler’s transition intention after parking fees increase
	Reducing the number of parking bays		$X_{IM3}$	Traveler’s transition intention after the number of parking bays is reduced
	Levying congestion fees		$X_{IM4}$	Traveler’s transition intention after levying congestion fees
Latent psychological factors	Infrastructure barriers	Bicycle road barrier	${\eta }_{BL}$	Bicycle road barrier perceived by the traveler
		Bicycle parking barrier	${\eta }_{BP}$	Bicycle parking barrier perceived by the traveler
		Bicycle sharing barrier	${\eta }_{BS}$	Bicycle sharing barrier perceived by the traveler
	Physical determinants		${\eta }_{BI}$	Physical determinants of the traveler
	Bicycling attitude	Safety attitude	${\eta }_{AS}$	Safety attitude of the traveler
		Convenience attitude	${\eta }_{AQ}$	Convenience attitude of the traveler
		Comfort attitude	${\eta }_{AC}$	Comfort attitude of the traveler
		Awareness	${\eta }_{AW}$	Awareness of the traveler
	Bicycling preference		${\eta }_{AD}$	Bicycling preference of the traveler
	Subjective norm		${\eta }_{SN}$	Subjective norm of the traveler
	Transition intention		${\eta }_{CI}$	Transition intention of the traveler

.

3.3. Questionnaire Survey and Field Research

The questionnaire included four aspects: (1) personal characteristics, including gender, age, income, educational background, and the number of children; (2) travel characteristics, including the travel distance, the usability of a car, the usability of a bicycle, and the bicycling condition in the past, present, and future; (3) transition intention in the presence of car restrictive measures (imposing car speed limits, increasing parking fees, reducing the number of parking bays, and levying congestion fees); (4) latent variables, including infrastructure barriers, physical determinants, bicycling attitude, bicycling preference, and subjective norm. Latent factors were measured with 35 questions rated on 5-point scales.

A questionnaire survey was conducted from 12 to 18 April in 2017 in 10 communities. These communities, including Longyueyuan, Fangqunyuan, and Fangzhouyuan, are located in different geographical regions in Beijing and contain various infrastructure qualities. The locations where the questionnaire survey was deployed are shown in Figure 1. A total of 608 valid questionnaires were obtained using random sampling.

In order to ensure the quality of the data, the Kaiser–Meyer–Olkin (KMO) test, Bartlett’s test, and Cronbach’s reliability test were conducted. The KMO values for infrastructure barriers, physical determinants, bicycling attitudes, bicycling preferences, subjective norms, and transition intention were 0.814, 0.804, 0.836, 0.822, 0.818, and 0.789 respectively. The Cronbach’s alpha values for infrastructure barriers, physical determinants, bicycling attitudes, bicycling preferences, subjective norms, and transition intention were 0.814, 0.810, 0.843, 0.883, 0.876, and 0.861 respectively. The results show that the questionnaire design had high reliability and validity.

Field research was conducted from 24 April to 12 May in 2017 in 10 communities to determine the infrastructure characteristics. The bicycle level of service, the density of the bicycle road network, and the accessibility of amenities were calculated using data collected during the field research [44,45].

4. Methodology

The discrete-choice model is based on utility maximization, and it is widely used to study travel mode choice. The hybrid-choice model can solve the problem of the independence of irrelevant alternatives and acquire the alternatives’ utilities that are unobservable. It can effectively represent travelers’ psychological preferences and attitudes. Therefore, in this work, HCM was used to study the staged transition process from driving to bicycling and explore the effects of exogenous variables and latent variables on the mechanism of behavior.

A HCM is a combination of the MIMIC model and a discrete-choice model. The MIMIC model is used to analyze the impacts of exogenous variables on latent variables and the relationships between latent variables and measurement indexes. The MIMIC model can be divided into a structural equation and a measuring equation. The structural equation can be described as

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

(1)

where $\eta$ is a latent variable, $x$ is the vector of exogenous variable, $\mathsf{\Gamma }$ is the estimated parameter matrix, and $\zeta$ is the measuring error.

The measuring equation can be described as

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

(2)

where $y$ is the observable index vector of latent variables, $\mathsf{\Lambda }$ is the estimated parameter matrix, and $v$ is the measuring error.

In this study, the binary logistic regression model was chosen as the discrete-choice model; this model can be described as

$$P_{i+1}=1/[1+\exp (V_i-V_{i+1})]$$

(3)

$$P_i=1-P_{i+1}$$

(4)

where $P_i$ is the probability in stage i, $P_{i+1}$ is the probability transfer to stage j, $V_i$ is the utility in stage i, and $V_{i+1}$ is the utility transfer to stage j.

The transition from driving to bicycling was divided into five sequential stages in this study. Four transition stage models were established using HCM to describe the transition process. The conceptual model is shown in Figure 2.

The dependent variable is “whether a shift to the next stage occurs” and the independent variables are “personal characteristics, travel characteristics, infrastructure characteristics, vehicle restrictions, and potential factors”. If the utility of stage i is 0, then the utility of stage i + 1 can be described as

$$V_{i+1}={\displaystyle \sum _{j=1}^J\alpha X_j}+{\displaystyle \sum _{k=1}^K\beta X_k}+{\displaystyle \sum _{l=1}^L\chi X_l}+{\displaystyle \sum _{m=1}^M\delta X_m+}{\displaystyle \sum _{n=1}^N\epsilon {\eta }_n+}A$$

(5)

where $X_j$ is a personal characteristic, $X_k$ is a travel characteristic, $X_l$ is an attribute of infrastructure, $X_m$ is a car-restrictive measure, and ${\eta }_n$ is a latent variable. $\alpha$, $\beta$, $\chi$, $\delta$, $\epsilon$ are estimated parameters. $A$ is a constant term.

5. Results and Discussion

5.1. Stages of Change Classification and Characteristic Analysis

5.1.1. Stages of Change Classification

The transtheoretical model indicates that the change process of behavior should be divided into five stages: pre-contemplation, contemplation, preparation, action, and maintenance. Therefore, the transition process from driving to bicycling was divided into five stages in accordance with TTM. Using the stage classification method, which was proposed by Thigpen [36], 608 travelers’ behavior stages were determined according to four questions. The results (as shown in

Table 2. Survey questions for the stages of change classification.

Survey	Stages of Change
	Pre-Contemplation	Contemplation	Preparation	Action	Maintenance
Did you bike in the past week?	No	No	No	Yes	Yes
What mode of transportation do you usually use for short-distance travel?	Other	Other	Other	Other	Bicycle
Have you thought about bicycling for short-distance travel?	No	Yes	Yes	Not asked	Not asked
How likely are you to bicycle at least once in the next six months?	Not likely	Somewhat likely	Very likely	Not asked	Not asked
Percentage of stages/%	12.7	10.2	16.4	27.1	33.6

) show that 33.6% of travelers are in the maintenance stage and use bicycles as a common transportation mode; 27.1% of travelers are in the action stage and have started to ride a bicycle; 39.3% of travelers are in the pre-contemplation, contemplation, or preparation stage. These travelers do not ride a bicycle for short-distance travel. These figures indicate that it is necessary to promote bicycling.

5.1.2. Stages of Change Characteristic Analysis

According to the classification results from 608 travelers, the personal characteristics, travel characteristics, and latent variables were analyzed. The results are listed in

Table 3. Characteristics of individuals in different stages.

Factor	Pre-Contemplation	Contemplation	Preparation	Action	Maintenance	Average
Women/%	60	47	58	50	47	51
Age	40.87	33.63	35.92	34.00	33.32	34.92
Above university/%	51	65	59	69	69	63
Income/yuan	5000	4500	4400	3200	3600	4300
Families with children/%	56	60	55	53	50	44
Car usable/%	61	55	40	40	32	56
Bicycle usable/%	34	39	50	56	75	53
Travel distance/km	2.31	2.55	2.61	2.66	2.51	2.54
Bicycle road barrier	4.09	3.83	3.79	3.46	3.05	3.45
Bicycle parking barrier	3.59	3.82	3.48	3.23	3.10	3.37
Bicycle sharing barrier	3.78	3.72	3.71	3.66	3.67	3.69
Physical determinant	2.97	2.37	2.24	2.02	1.99	2.20
Bicycling preference	2.84	2.75	3.36	3.31	3.52	3.24
Safety attitude	2.19	2.13	2.51	2.63	2.66	2.50
Convenience attitude	2.67	2.69	3.13	3.57	3.46	3.29
Comfort attitude	3.04	3.12	3.32	3.56	3.61	3.55
Awareness	3.95	3.81	4.19	4.14	4.19	4.11
Subjective norm	3.11	3.27	3.67	3.57	3.59	3.50

.

The results show that personal characteristics, travel characteristics, and latent variables differ among the different stages, and the difference varies among the different stages. This means that the influence factor that promotes the transfer to the next stage is different for different stages. A similar result was found by Nkurunziza et al. [35]. For personal characteristics, there are higher proportions of women and seniors in the pre-contemplation stage. The income is higher for individuals in the pre-contemplation, contemplation, and preparation stages. For travel characteristics, the usability of a car is negatively correlated with the transition, and the usability of a bicycle is positively correlated with the transition, while the travel distance for short travel has no significant effect. For latent variables, bike lane barriers, parking barriers, and physical determinants present a descending trend in the five stages, while bicycling preference, safety attitude, convenience attitude, and comfort attitude present a rising trend in the five stages. These results indicate that the enhancement of perceived benefits and the reduction of perceived barriers are important for promoting the transition.

5.2. Transition Intention MIMIC Model

In this study, a transition intention MIMIC model was established. The model was calibrated using Amos. The model test results show that χ2/df = 1.16 < 2, GFI = 0.90 > 0.9, CFI = 0.97 > 0.9, AGFI = 0.91 > 0.9, and RMSEA = 0.02 < 0.08. These values mean that the model’s goodness of fit meets standard requirements. The calibration result of the model can be seen in Figure 3. The figure shows the influence relationships among the latent variables and the causal relationships between the exogenous variables and the latent variables.

The path coefficients for the impacts of exogenous variables on latent variables and T values (in brackets) are listed in

Table 4. Calibration results for the comprehensive structural model of the transition to bicycling.

Exogenous Variable		Bicycling Attitude	Bicycling Preference	Physical Determinant	Infrastructure Barrier	Subjective Norm
Personal characteristics	Gender	\	\	0.168 ** (4.051)	0.086 * (1.991)	\
	Age	\	\	0.214 ** (5.109)	\	\
	Income	\	−0.096 * (−1.982)	\	\	\
	Educational background	\	\	−0.104 * (−2.541)	\	0.078 * (1.960)
Travel characteristics	Usability of a car	\	−0.106 ** (−3.252)	\	\	\
	Usability of a bicycle	0.181 * (1.964)	\	−0.103 * (−2.514)	−0.130 ** (−3.000)	0.121 ** (2.834)
Infrastructure characteristics	Density of bicycle road network	0.139 ** (3.727)	\	\	−0.104 * (−1.961)	\
	Bicycle level of service	0.122 ** (3.281)	\	\	−0.201 ** (−4.865)	\
	Accessibility of amenities	0.074 * (1.963)	\	\	\	\
Car-restrictive measures	Car speed limit	0.098 * (2.301)	\	\	\	\
	Increasing parking fees	0.118 * (2.782)	\	\	\	\
	Reducing the number of parking bays	0.136 ** (3.048)	\	\	\	\
	Levying congestion fees	0.066 * (1.966)	\	\	\	\

Note: * means p < 0.05, ** means p < 0.01.

. Only the parameter estimates of variables that are significant at a 95% confidence level are listed.

From Figure 3 and Table 4, we observe the following:

(1)

The transition intention MIMIC model explains 73% of the variance in transition intention from driving to bicycling; transition intention can be effectively explained by infrastructure barriers, physical determinants, bicycling attitudes, bicycling preferences, and subjective norms. Transition intention is directly influenced by infrastructure barriers, physical determinants, bicycling preferences, and subjective norms, and it is indirectly influenced by infrastructure barriers and bicycling attitudes. Bicycling preference is the mediating variable between bicycling attitude and subjective norm.

(2)

Exogenous variables do not directly influence transition intention, but they indirectly influence transition intention through infrastructure barriers, physical determinants, bicycling attitude, bicycling preference, and subjective norm. The number of children and travel distance have no significant impact on latent variables.

(3)

Different exogenous variables of personal characteristics and travel characteristics affect different latent variables. The usability of a bicycle is positively correlated with bicycling attitude. This means that travelers whose bicycles are usable perceive bicycling as safer, more convenient, and comfortable compared with the perceptions of travelers whose bicycles are not usable. Income and the usability of a car are negatively correlated with bicycling preference [46]. This means that high-income groups do not like bicycling. They are used to driving, so they often have a bias against bicycling. Gender and age are positively correlated with physical determinants, while education background and the usability of a car are negatively correlated with physical determinants. This is because the bicycling skill and physical strength of women and senior men are relatively lower [47]. Gender is positively correlated with infrastructure barriers, while the usability of a bicycle is negatively correlated with infrastructure barriers. Education background and the usability of a bicycle are positively correlated with subjective norms.

(4)

The three infrastructure characteristics are all positively correlated with bicycling attitude. This means that improvements in the bicycle level of service, the density of bicycle road networks, and the accessibility of amenities help to enhance travelers’ perceptions of bicycles as agreeable. The bicycle level of service and density of the bicycle road network are negatively correlated with infrastructure barriers. This means that the perception of infrastructure barriers will decrease if the objective infrastructure is improved.

(5)

Car-restrictive measures are positively correlated with bicycling attitude. This indicates that the perception of bicycling not only depends on the service level of the bicycle but also depends on the service level of the car. Therefore, the implementation of car-restrictive measures can increase barriers to car travel. This is beneficial for improving bicycling attitudes.

5.3. Transition Behavior Hybrid Choice Model

In order to depict the staged transition process from driving to bicycling and explore the key influence factors in different stages, an HCM was used to establish four stage models, namely, the PC-C model, C-PA model, PA-A model, and A-M model. The models used transfer to the next stage or not as the dependent variable and used personal characteristics, travel characteristics, infrastructure characteristics, car-restrictive measures, and latent factors as independent variables.

The models were calibrated using SPSS. After eliminating non-significant factors, parameter estimates of the final models were determined, as shown in

Table 5. Parameters of the models.

Factor		PC-C Model	C-PA Model	PA-A Model	A-M Model
Variable	Index	Parameters (Sig.)	Parameters (Sig.)	Parameters (Sig.)	Parameters (Sig.)
Personal characteristics	Age	−0.037 (0.048)
	Income			−0.361 (0.001)
Travel characteristics	Usability of a car	−0.904 (0.037)			−0.917 (0.000)
	Usability of a bicycle				0.689 (0.004)
Infrastructure characteristics	Bicycle level of service			1.126 (0.001)
	Density of bicycle road network	0.475 (0.005)		0.605 (0.002)
	Accessibility of amenities	−0.011 (0.026)		0.021 (0.016)
Latent factors	Bicycle road barrier	−0.770 (0.025)	−0.410 (0.046)	−0.301 (0.025)	−0.306 (0.022)
	Bicycle parking barrier	1.078 (0.002)	−0.549 (0.022)	−0.627 (0.004)
	Physical determinant	−0.804 (0.004)		−0.423 (0.013)
	Safety attitude		0.729 (0.029)	0.556 (0.003)
	Convenience attitude		0.659 (0.045)
	Comfort attitude		0.715 (0.038)
	Bicycling preference		0.737 (0.027)
	Subjective norm		0.806 (0.049)
	Transition intention	0.684 (0.026)	1.875 (0.000)	1.103 (0.000)	1.118 (0.000)
Car restrictive measures	Car speed limit	0.690 (0.042)
	Reducing the number of parking bays		0.794 (0.047)
	Increasing parking fees		0.818 (0.037)
	Levying congestion fees	0.631 (0.049)	0.770 (0.041)
n		139	162	265	369
Log-likelihood		−69.648	−79.468	−114.773	−181.197
Pseudo R2		0.416	0.391	0.385	0.357

, which lists the parameters of the variables, odds ratios, and Sig. As shown in Table 5, the pseudo R² values of the models are between 0.357 and 0.416. It is generally acknowledged that a model has high precision if the pseudo R² is larger than 0.2, so these four stage models all have high precision [35].

The results of the PC-C model show significant negative correlations between age, the usability of a car, and the transition from pre-contemplation to contemplation. This means senior men and travelers whose cars are usable have difficulty transferring from pre-contemplation to contemplation. For infrastructure characteristics, the accessibility of amenities has a negative correlation with the transition. This is because travelers whose bicycling willingness is low prefer to walk to a destination in a community with high accessibility. The density of the bicycle road network has a positive correlation with the transition. This indicates that an increase in bicycle roads is beneficial for advancing the transition. For latent variables, there are significant negative correlations between bicycle road barriers, physical determinants, and the transition, while there is a positive correlation between bicycle parking barriers and the transition. The reason for this is that travelers who are in the stage of pre-contemplation do not pay attention to bicycle parking, so they perceive the parking barriers as low [48]. For car-restrictive measures, there are positive correlations between the car speed limit, levied congestion fees, and the transition. These measures can have good effects on the transition to bicycling.

The result of the C-PA model shows that only latent variables and car-restrictive measures affect the transition from contemplation to preparation. For latent variables, there are significant negative correlations between bicycle road barriers, bicycle parking barriers, and the transition, while there are positive correlations between bicycling attitude, bicycling preference, and the transition. This indicates that travelers who perceive more advantages of bicycling and fewer barriers to bicycling tend to transition [46]. The subjective norm has a positive correlation with the transition. This means that travelers who are in the stage of contemplation tend to be more influenced by social desirability and identity. For car-restrictive measures, increasing parking fees, reducing the number of parking bays, and levying congestion fees have positive correlations with the transition. This means that these measures effectively contribute to the transition from contemplation to preparation.

The results of the PA-A model show that income has a negative correlation with the transition from preparation to action such that high-income groups are less likely to transition. For infrastructure characteristics, the bicycle level of service and the density of the bicycle road network have especially large impacts on the transition. This indicates that the improvement of infrastructure is indispensable in prompting travelers to take action.

The results of the A-M model show that travelers whose cars are not usable or whose bicycles are usable prefer to maintain their bicycling behavior. For latent variables, there is a significant negative correlation between bicycle road barriers and the transition from action to maintenance. This indicates that decreasing bicycle road barriers is necessary for the maintenance of bicycling.

Summarizing the influence factors in the four stage models, we found that the influence factors in different stages were different. Thus, it is necessary to take different measures to promote the transitions between different stages. Specifically, (1) car-restrictive measures only act on the pre-contemplation and contemplation stages. These measures can only promote the production of bicycling intention. (2) For bicycle development, hard measures (community design and infrastructure improvement) and soft measures (programming and education) act on different stages, but both are indispensable. Subjective psychological factors are significantly correlated with the transition from contemplation to preparation. This indicates that the promotion of perceived benefits through soft measures is important for promoting the transition from contemplation to preparation. Objective environmental factors are significantly correlated with the transition from preparation to action. This indicates that improvements in the bicycle level of service, the density of the bicycle road network, and the accessibility of amenities are necessary for prompting travelers to take action. (3) Perceived environmental barriers act on every stage, but the objective environment only acts on the pre-contemplation and contemplation stages. Therefore, it is necessary to couple programming with education after improving the bicycling environment. Travelers are willing to transfer to a bicycling travel mode if they perceive an improved bicycling environment. Ma et al. [45] also found that the objective environment may only indirectly affect bicycling behavior by influencing perceptions.

6. Conclusions

Based on the benefits of bicycling for transportation, the environment, and public health, many cities around the world have set ambitious goals to increase the bicycle share of trips and reduce the car share of trips. Many cities have adopted measures to achieve this goal. However, these measures have not produced good results. Effective measures can only be developed if the transition process from driving to bicycling is understood. The transtheoretical model of behavior change was used to describe how positive and permanent change can be fostered in individuals, and it may shed light on how travelers stage their transfer from driving to bicycling. This paper divides the transition from driving to bicycling into five stages on the basis of TTM and studies the staged transition process based on the influences from latent variables. The results lay the theoretical foundations for the development of measures.

Bicycling is affected by personal characteristics, travel characteristics, infrastructure characteristics, and psychological factors. On the basis of TPB, this work expanded the latent variables according to the characteristics of bicycling. Bicycling attitude, bicycling preference, physical determinants, infrastructure barriers, subjective norm, and transition intention were used to represent bicycling psychology. Moreover, a survey was conducted to obtain data on personal characteristics, travel characteristics, psychological factors, and infrastructure characteristics in ten communities in Beijing, China.

In order to systematically research the impacts of latent factors on the transition from driving to bicycling, a transition intention MIMIC model was established to analyze the influence relationships among latent variables and the causal relationships between latent variables and exogenous variables. The results show that bicycling transition intention can be adequately explained by infrastructure barriers, physical determinants, bicycling preference, bicycling attitude, and subjective norm. Infrastructure barriers, physical determinants, bicycling preferences, and subjective norms directly affect transition intention, and infrastructure barriers, subjective norms, and bicycling attitudes indirectly affect it. Moreover, exogenous factors have no direct effect on transition intention, but they do have an indirect effect on transition intention through latent factors. Finally, infrastructure barriers and bicycling attitudes are complete intermediate variables between the objective environment and transition intention, and bicycling attitude and bicycling preference are complete intermediate variables between car restrictive measures and transition intention. These findings indicate that improving the perceived environment using intervening measures is very important. Meanwhile, improving bicycling attitude and preference by implementing car-restrictive measures is a crucial means of promoting the transition to bicycling.

On the basis of the transtheoretical model, travelers were divided into five stages of change: pre-contemplation, contemplation, preparation, action, and maintenance. Moreover, four staged transition models were established using the hybrid-choice model to examine the transitions between two adjacent stages. These models describe the staged change process and reveal the influence mechanisms of personal characteristics, travel characteristics, the objective environment, subjective psychology, and car-restrictive measure between different stages. The results show that car-restrictive measures can only promote behavioral intention. The stages influenced by subjective psychology and objective environment are different, and the stages influenced by soft policy and hard measures are different. The perceived environment affects every stage, but the objective environment only affects two stages. This means that making travelers perceive bicycling as agreeable by advocating and popularizing it may be the most important step. This study also had limitations. We only discuss the substitutability of bicycles for cars in this study, but in small and medium-sized cities, the substitution effect of electric bicycles for cars is actually irreplaceable. In addition, the sample of survey data used for the study is still preferred, and there is a need to conduct a wider range of questionnaires with GPS and other sensory devices in the future and to explore the influences of other potential variables such as habits and values on cycling behavior in order to improve the predictive accuracy of the psycho-cognitive and behavioral decision-making model.