Combining SEM, fsQCA and BNs to Explore E-Bike Riders’ Helmet Wearing Intentions under the Impact of Mandatory Policies: An Empirical Study in Zhenjiang

Abstract

China has become the largest country for e-bikes in the past decade, leading to a surge in road-related accidents. Chinese provinces and cities have successfully formulated local regulations, which set fines for not wearing helmets levying on e-bike riders. Without considering psychological resistance which is universal under the premise of compulsory legislation, the legislation and enforcement activities cannot continuously promote electric bikers’ helmet use. This study aims to investigate the predictors that influence e-bike riders’ intention to wear helmets by designing a research methodology that incorporates the theory of planned behavior (TPB), the protection motivation theory (PMT), and the psychological reactance theory (PRT). A multi-method analytical approach, including structural equation modeling (SEM), fuzzy-set qualitative comparative analysis (fsQCA), and a Bayesian Network (BN) with a sample dataset of 846 respondents. SEM and fsQCA explored the intentions to wear helmets from linear and nonlinear perspectives. BN verifies the degree of influence between different configurations resulting from fsQCA, to predict which mental configurations could get the most significant impact on helmet use. A total of 11 configurations lead to high helmet-wearing intention, and 4 lead to low helmet-wearing intention. Policymakers could design safety policies on our research, such as driving licenses and special lectures for e-bike riders.

1. Introduction

Electric bicycles (e-bikes), a popular travel mode in developing countries such as China, have advantages in mobility performance and speed compared with the pedal cycle [1] By the beginning of 2022, the annual sales of e-bikes in China exceeded 41 million [2], and the social possession was close to 350 million [3]. However, inadequate punishment and weak management raised many safety problems in China. In 2020, 8724 and 45,557 people were killed and injured in traffic accidents involving e-bikes, accounting for 13.51% and 17.43% of the total number of road traffic accidents in China [4]. A total of 81.5% of e-bike riders’ deaths are caused by head injuries [5], while wearing helmets could reduce the rate of head injuries by 70% and the fatality rate by 40% [6]. Although e-bike helmets could protect riders, the average proportion of e-bike helmet-wearing rates is less than 20% due to the lack of compulsory legislation and enforcement measures before 2020 in China [7].

The Chinese Traffic Administration Bureau implemented the “One Helmet, One Belt” Security operation to encourage e-bike riders to wear helmets. More than twenty provinces have formulated 50 local regulations requiring e-bike riders to wear helmets. By February 2021, the average proportion of e-bike helmet-wearing rates has risen to 54% in China [7]. Nearly half still resisted wearing helmets. The effect of compulsory legislation and enforcement measures might fade away and devaluate without individual motivation [8]. In other words, compulsory legislation and enforcement measures are insufficient to maintain long-term helmet-wearing behavior without other intervention in personal motivation. Behavioral intention is a vital construct reflecting personal motivation, which could be affected by individual psychology. Therefore, understanding the individual psychological factors affecting helmet-wearing behavior will help policy makers to formulate differentiated theory-based interventions and further promote the helmet-wearing behavior of e-bike riders in China.

Tang et al. used a layered regression model to construct the helmet-wearing intention model of e-bike riders based on the theory of planned behavior (TPB) and health belief model (HBM), generating fresh insight into helmet research after enacting Chinese helmet legislation [9]. However, e-bike riders could have psychological resistance with many forms of expression in the case of compulsory legislation. Psychological resistance can be divided into explicit or implicit, immediate or delayed, in word or deed. Resistance often hinders the implementation of policies or makes it difficult to carry them out thoroughly. Therefore, we adopted psychological reactance theory (PRT), intending to contribute to traffic safety researches by exploring e-bike riders’ psychological reactance.

Besides, the traditional linear approaches could not explain complicated and dynamic human behavior [10], such as e-bike helmet wearing. We introduced fuzzy-set qualitative comparative analysis (fsQCA), based on configuration analysis, to explore nonlinear relations and unique configurations between human behavior variables [11,12]. This study provided an opportunity to advance our perspective of helmet-related traffic safety issues under configuration analysis. In addition, the fundamental assumption of fsQCA is that each factor configuration is equal [13,14,15]. We could not distinguish the influence degree of each configuration. In practical implications, policymakers might wish to find the most appropriate configuration, which has the most significant impact on the dependent variable, to propose the most efficient policy. Bayesian network (BN) could demonstrate the influence of different configurations on the results by adjusting the values of each variable. Therefore, this research introduced and developed BN to fill a gap in the equal combination of fsQCA ’s research.

In summary, there are three gaps in current research.

(1)

The current study fails to analyze the impact of riders’ psychological resistance on their helmet wearing intention within the context of a mandatory policy.

(2)

Focusing on the linear symmetric relationship among the latent variables, the effect on helmet wearing intention of psychological factor configurations has been neglected.

(3)

In studies about psychological configurations, the various configurations are equal so that it is not difficult to identify the configuration which has the greatest effect on helmet wearing intention.

In order to refine the gaps, the objectives of this research are as follows:

(1)

Combined with TPB, PMT and PRT, this study constructed a framework of e-bike helmet wearing behavioral intentions and provided policy recommendations for the implementation of Chinese legislation.

(2)

This research effectively integrates SEM, fsQCA, and BN., exploring the relation-ships and combination of factors that affect e-bike riders’ helmets wearing intention. The combination extends the application scope of qualitative comparative analysis.

(3)

By further analyzing the results of SEM and fsQCA with BNs, this study found the optimal configuration for high helmet wearing intention.

The rest of this research is organized as follows. Section 2 shows an extensive literature review of the theoretical background and research hypotheses. Section 3 demonstrates the investigative procedure, research methodology, and significant results. Section 4 provides an overall discussion and implications. Finally, we draw the main findings, discuss the limitations of this research, and expound on potential future research in Section 5.

2. Literature Review and Research Hypotheses

In the past decade, several researchers have sought to determine helmet usage behavior of two-wheeler riders from a psychological perspective. To get a more comprehensive understanding of the research status, we reviewed the existing research.

Table 1. Helmet wearing intention in existing research.

Authors (Year)	Nation	Vehicles	Theoretical Model	Mathematical Model
Kang et al. (2021) [16]	USA	Bicycle	/	Maximum likelihood estimation, MLE
Tang et al. (2021) [9]	China	E-bike	HBM, TPB	Hierarchical regression analysis
Champahom et al. (2020) [17]	Thailand	Motorcycle	TPB, LOC	SEM
Fyhri et al. (2019) [18]	Norway	Bicycle	/	SEM
Kumphong et al. (2018) [19]	Vietnam	Motorcycle	T-LOC, HBM, TPB	SEM
Fallah et al. (2018) [20]	Iran	Bicycle	HBM	SEM
Ghasemzadeh et al. (2017) [4]	Iran	Motorcycle	TPB	SEM
Rosli et al. (2016) [21]	Malaysia	Motorcycle	/	SEM
Trinh and Le (2016) [22]	Vietnam	Motorcycle	TPB	CFA, SEM

summarizes previous studies on helmet usage behavior of two-wheeler riders.

In theoretical models, current research has explored riders’ helmet wearing intentions based on the theories of HBM, T-LOC, and TPB. However, at the critical period when China’s helmet policy is gradually rolling out, these theories fail to discuss psychological resistance in the mandatory policy context. In mathematical methods, SEM as the dominant mathematical model provides a wealthy of findings. However, helmet wearing behaviour as a complex decision-making process is necessary to provide new interpretations from an asymmetric perspective. Based on Table 1, we will conduct a review from two perspectives: mathematical model and theoretical framework.

2.1. Mathematical Model

Previous research on bicycle helmet wearing intention has focused on SEM to explore the relationship between variant variables, presented in a linear combination. SEM could quantitatively assess the combined effects of each influencing factor on target variables by measuring the correlations of independent variables [23,24]. SEM provides an effective tool to analyze the interaction of various psychological factors that may affect an individual ’s e-bike helmet-wearing intention after the “One Helmet, One Belt “Security operation. Nevertheless, SEM demonstrates the competitive relationship of independent variables rather than showing how independent variables combine to create dependent variables [15,25]. Traditional symmetrical approaches (e.g., SEM) usually focus on the distinctive contribution of a single variable while holding constant all other variables [26]. Traditional symmetrical approaches have difficulties dealing with variables as configurations and examining combinations of variables [25]. Configuration analysis and combinations of variables could help researchers recognize asymmetrical approaches to revisit previous conclusions established using symmetrical approaches [27,28,29]. Therefore, we introduce fsQCA to explore nonlinear relations and unique combinations between e-bike helmet-wearing behavior variables.

FsQCA is an asymmetric approach using Boolean algebra to implement comparison configuration based on set theory and fuzzy logic [26]. It has been widely applied to complex qualitative comparative problems, such as the subjective well-being of residents [30], greenhouse gas emissions of electric vehicles [31], customer-brand relationships [32], and male-typed leadership schemas [33]. FsQCA attains the inevitable outcome by combining case-oriented qualitative methods with variable-oriented quantitative methods for dealing with large cases, leading to more general analytical reasoning. Therefore, fsQCA is a “midway between exploratory and hypothesis approach [34]. This approach does not apply to hypothesis-testing but requires creating propositions that confirm the qualification in configurations [35].

FsQCA could only get multiple equal configurations to the same outcome. However, policymakers would prefer the most appropriate configuration, which has the most significant impact on the dependent variable, rather than a bunch of equal configurations in practical application. Hence, we introduced BN to explore the configuration most impacted the outcome. BN, a probabilistic directed acyclic graphical model, reveals the causal relationship between variables [36,37]. The directed acyclic graph of the BN consists of nodes and edges. Each node shows a variable of the network structure, following specific distribution, and each edge represents the causal relationship between two variables. BN compactly represents several variables‘ joint density by the conditional independence relation. Significantly, the BN could predict the variability of dependent variables in a unique configuration by adjusting the states of multiple conditions. This experimental work is the forerunner of the combination fsQCA with BN.

The three methods discussed above complement each other. Figure 1 shows the relationship between these methods. The SEM and fsQCA are mutually verified to reduce the error and prove the reliability of this research. The structural equations model provides a structural model for BN. In addition, BN demonstrates the degree of influence between different configurations resulting from fsQCA. Therefore, the three methods could be considered multi-method analytical approaches to study helmet-wearing intention in China.

2.2. Theoretical Framework

Table 1 summarizes studies exploring helmet-wearing behavior for two-wheelers by various theoretical models. Constructing a theoretical framework, an essential step to guarantee the measurement accuracy, could better reflect and explain the interaction between predictor variables in behavioral studies [38]. We propose an integrated theoretical model to investigate e-bike riders’ intention to wear helmets during legislation implementation.

Theory of planned behavior (TPB), derived from the theory of reasoned action (TRA), has been widely used in behaviors of various travel modes, such as advanced driver assistance systems [39], delivery riders’ red-light running behavior [40], and intention to take the new high-speed rail [41]. Besides, as shown in Figure 1, many studies have used TPB to research motorcycle or bicycle helmet issues, demonstrating the solid predictive ability of TPB. However, helmet wearing behaviour is also regulated by mandatory policies, which means that helmet wearing behaviour is closely related to whether or not the riders will be penalized. Interpreting helmet wearing behavioural intentions only in terms of Attitude, Subjective Norm, and Perceived Behavioural Control fails to respond to constraints on freedom and choices made for self-protective. Therefore, TPB might not be enough to explain the intention to wear a helmet during the policy implementation. Fear of head injury in e-bike accidents could be regarded as a health threat, prompting to establish a motivation to protect head safety by the helmet.

Protection motivation theory (PMT) includes a range of variables that initiate cognitive processes and argues that individuals are motivated to take protective measures through threat and response appraisal. Helmet wearing is just a protective action that allows e-bike riders to deal with the risk of head injuries in traffic accidents. PMT has been widely used to explore risky travel behaviors, such as adopting electric vehicles [42], protective behavior in ride-sharing [43], and travel fear after COVID-19 [44]. Therefore, we attempt to apply PMT to explore whether fear of head injuries in e-bike accidents and self-protective consciousness would prompt riders to wear helmets.

Current research has explored riders’ helmet wearing intentions based on the PMT and TPB. Within these two theoretical frameworks, riders are free to choose whether to wear a helmet or not, which ignores the influence of mandatory policies. However, during the Chinese helmet policy implementation, Traffic Administration Bureaus across China attempted to curtail no helmet behavior by launching enforcement campaigns [7]. The restriction on e-bike riding has induced people ’s psychological reactance and increased their desire to travel freely. The psychological resistance theory (PRT) explains why and how residents resist or combat powerful messages in persuasion. In PRT, freedom is considered a way of meeting the basic needs of human self-management, in which residents feel free if they could make their choices allodial [45]. When residents perceive that their freedom is violated by enforcement information, they experience psychological resistance, a negative motivation. The empirical evidence of e-bike riders’ psychological reactance is poorly understood. Understanding this psychological reactance could help transport agencies formulate more scientific control measures. This study, which adopted psychological reactance theory (PRT), aims to contribute to this growing area by exploring e-bike riders’ psychological reactance.

This research constructs a comprehensive theoretical model combined with TPB, PMT, and PRT to explore psychological variables of the e-bike riders’ helmet-wearing intention during helmet legislation implementation. Also, this study considers road users’ perceived legal consequences (e.g., punishment) for violating traffic rules. We extend perceived legal norms beyond the comprehensive theoretical model.

2.3. Research Model and Hypotheses

The theoretical framework we developed is shown in Figure 2.

Theory of planned behavior (TPB) is a generic model most widely used in predicting and describing social behavior [41,46,47]. According to the abovementioned theory, attitude is the degree to which a person has a positive or negative evaluation of the behavior in question. Attitude is an immediate determinant of behavioral intention, and human behavior is based upon behavioral intentions. Besides, a positive attitude will lead to a stronger behavioral intention. In the study, a positive attitude is a fundamental factor in an individual ’s choice to wear a helmet when riding e-bikes. Attitudes towards wearing helmets were positively correlated with intentions for helmet use when riding motorcycles or bicycles [48]. Therefore, we propose hypothesis 1:

H1: 

Attitude has a positive effect on Behavioral Intention.

Self-efficacy (SE) is also an essential variable in PMT referring to an individual’s perceived ability to perform specific conservation behaviors, which has a positive effect on attitudes [49]. By definition, self-efficacy (SE) in PMT has a similar meaning to perceived behavioral control (PBC) in TPB. In 1999, Compeau et al. first combined self-efficacy and perceived behavioral control into one variable [50]. In 2012, Ifinedo formally proposed that perceived behavioral control is a key variable in fusing TPB and PMT, and that self-efficacy can be equated with perceived behavioral control [51]. Zhang et al. considered both the positive effects of PBC on behavioral intentions in TPB and the positive effects of SE on attitudes in PMT when combining PBC and SE [52]. Considering the same meaning of self-efficacy (SE) and perceived behavioral control (PBC), we integrate perceived behavioral control into self-efficacy (SE).

Therefore, based on the above, the following hypotheses are proposed:

H2: 

Self-efficacy has a positive effect on Behavioral Intention.

H3: 

Self-efficacy positively affects attitude.

To provide more powerful reasoning to explain the intention to wear a helmet in riding e-bikes, we added more predictors from the Protect Motivation Theory (PMT). The PMT as a conceptual social cognition model is employed for understanding health behavior, and safety behavior. The theory has been the main subject in recent studies [53], it points out that passengers are stimulated by risk perception. They may protect themselves and others to avoid being injured. Risk perception and its predictors constitute the core part of the PMT, which is carried out based upon two parallel processes: “threat appraisal’ and “coping appraisal’.

In the threat appraisal, evaluate the extent of risk reception to which they: perceived susceptibility and perceived severity. Perceived susceptibility refers to individuals’ perceptions of risk or vulnerability to a health threat. Perceived severity refers to one ’s perception of the seriousness of the possible harms. In helmet usage, perceived susceptibility reflects the subjective perception of the likelihood of being involved in an accident without wearing helmets [54]. Meanwhile, perceived severity means the subjective perception of the seriousness and its potential consequences associated with crashes without wearing helmets [54]. In addition, the susceptibility and severity components, in combination, mobilized the individuals’ behavior. The research has found that perceived susceptibility correlated significantly with perceived severity in bicycle helmet usage [48]. Therefore, our study attempted to use the path for predicting intention to helmet usage in road safety when riding an e-bike.

H4: 

Perceived Susceptibility has a positive relationship with Perceived Severity.

Based on the above-discussed, if the e-bike user perceives that danger may occur around them, or perceived the anticipated seriousness about not using helmets, people may wear a helmet to avoid being injured. According to this subjective psychological mechanism, perceived severity may guide or transfer the individual ’s attitude toward wearing a helmet. As for perceived susceptibility, previous studies found it has significant effects on the intention to wear a helmet in motorcycle or bicycle riding behavior [55,56]. Moreover, the perceived susceptibility also directly influences the intention of e-bikes’ helmet usage.

The users will be emerging the behavioral intention of wearing a helmet for preventing the occurrence of injury, when the users perceive that a crash ahead. More risk perception has a relationship with more intention to wear helmets. Thus, the following hypotheses are proposed:

H5: 

Perceived severity has a positive effect on attitude.

H6: 

Perceived susceptibility has a positive effect on Behavioral intention.

Besides the above, in the coping appraisal, individuals assess the efficacy of a preventive response: “response efficacy’ and “self-efficacy”. Response efficacy refers to the perceived effectiveness of the recommended risk preventative behaviors, the likelihood that the action will actually reduce the risk of the health threat [53]. While self-efficacy is the degree of which is belief in one’s own capacity to take action in a certain situation [57]. In brief, response efficacy and self-efficacy evaluations are factors to increase the likelihood of making a response, in which the higher levels of these components produce a stronger probability of making an adaptive response. Thus, besides the variable of self-efficacy having a positive effect on the attitude, the response efficacy also has a significant relationship with the attitude. Both of these components may transfer the individuals’ attitude to perform the health-enhancing behavior. Our study attempted to manipulate these components to predict one ’s attitude toward helmet usage when riding e-bikes. For example, when the individual believes the coping response is effective (i.e., high response efficacy), the degree to which is that they believe wearing a helmet will decrease the risk of crash, he/she will transfer the attitude of opposition into the agreement, increase the intention to adopt the practice (wear a helmet). Thus, we proposed the following hypothesis:

H7: 

Response efficacy has a positive effect on attitude.

According to the PMT, when combing the processes between threat appraisal and coping appraisal, a certain interaction was emerging. There is an interaction may be between perceived susceptibility and self-efficacy. In PMT, it is assumed that if self-efficacy is high, the increase in susceptibility will positively affect behavioral intention. In contrast, if the proposed response was ineffective (i.e., low self-efficacy), the increase in susceptibility will reduce the intention to perform the behavior or have no effect on the intention [20,54]. It could be noted that a person with a strong sense of self-efficacy may easily overcome any barriers. In our study, this predicted interaction also is appropriated. When individuals perceive the recommended behavior of wearing a helmet as useful, the response is effective (i.e., higher self-efficacy), and the increase in perceived susceptibility will positively impact the intention to wear a helmet. Therefore, we proposed the following hypothesis:

H8: 

Self-efficacy has a positive effect on Perceived susceptibility.

“Why is it that a man sometimes does the opposite of what he is told?”, Miron and Brehm (2006) presented a theory of psychological reactance (PRT) to explain this phenomenon [58]. According to the theory, people are convinced that they have a great deal of freedom to take different attitudes and positions on any issue. “The freedom can pertain to what one does, how one does it, or when does it, “However, if individuals perceive that freedom is eliminated or threatened with elimination, the subjective response of psychological reactance will be aroused” [59]. Psychologic reactance being motivational arousal will be directed toward restoring one’s freedom, whose amount of reactance is based upon the threatened freedom [59]. In addition, PRT proposed that psychologic reactance also be aroused when people are pressured into being restricted by restrictive policies [60]. Proudfoot and Kay (2014) found that the public may react significantly negatively to policies that restrict their behavior [61]. In our study, considering the mandatory helmet usage law, people may generate the freedom threat by must be wearing helmets in riding e-bikes, and then arouse the reactance to wearing helmets. Thus, we propose hypothesis 9:

H9: 

Freedom Threat positively affects Psychological Reactance.

The message could be quite helpful in facilitating individual positive attitudes. As demonstrated by the found, motivational arousal such as reactance may hinder information acceptance or undermine the effectiveness of persuasive messages [62,63]. Psychological reactance may impede individuals from receiving or processing information when the threat to freedom emerges, resulting in negative attitudes, even indirectly determining behavior intention change into an opposite [64,65]. Beyond that, previous studies have shown that psychological reactance could be regarded as a composite index of self-report indices of anger and negative cognitions, such as counterarguments [62]. Working from the notion, the reactance may impact attitude through the above negative emotions. In addition, based upon PRT, an individual may experience reactance when he is pressured into making a choice in a two-sided behavior. Psychological reactance may cause the individual’s attitudes to change into antagonistic. In the study, due to the choice about whether to use helmets, the lack of messages by threats to freedom, and the generation of negative emotions, people may arouse the reactance to wearing helmets when riding e-bikes and change individual attitudes to without wearing helmets. As a result of that, we add this hypothesis:

H10: 

Psychological Reactance positively affects Attitude.

Perceived legal norm refers to one’s degree of perceived legal consequence (e.g., punishment) of law-violating behavior [9]. The norm can be understood as a person ’s attitude towards the required law, that is, people related to the understanding of legal norms on cognitive or affective [66]. The norm may still be applicable and binding in the behavior focused upon. Based on the definition of perceived legal norm, previous studies found that the norm had a short-term effect on people ’s behavior. For instance, Seddig (2016) revealed that legal norms were assumed to be a strong direct predictor related to school bonds over adolescence [67]. The legal norm also has been found to be related to the legal enforcement of youth ’s underage drinking [68]. In addition, Kim et al. (2021) examined that the legal norm does not statistically impact the behavior of texting while driving [69]. Besides the aforementioned studies, the perceived legal norm is also crucial to one ’s attitude toward wearing a helmet, after enacting the mandatory helmet legislation. This norm was found to be negatively correlated with traffic rule violating behavior [9]. In our study, mandatory helmet legislation may transfer some people into wearing behavior. For instance, when someone is punished for not wearing helmets while riding e-bikes, their perceived legal norms may increase, leading to a change in attitudes toward wearing helmets. Thus, we make the following hypothesis:

H11: 

Perceived legal norm positively affects Attitude.

3. Survey and Method

3.1. Data Collection and Sample

As mentioned in the previous section, the present study was conducted in the urban area of Zhenjiang, which is one of the central cities in the Yangtze River Delta region along the east coast of China, covering an area of 3843 km² and having an estimated population of 1.27 million. This study designed a questionnaire to investigate the impact of mandatory law on helmet usage. To measure latent variables, we developed questionnaire items and pre-tested on several small-scale random shopping malls with about 67 participants. The pre-test occurred on May 28th, was carried out between 8:00 and 17:00. This test determined the location and final structure of the questionnaire. After that, we further improved the interview procedure to minimize the participants’ impact and modified the questionnaire items. Subsequently, the data were collected in June 2021 and carried out in plazas, railway stations, and other crowded places. Before accessing the interview, participants were informed about the anonymous and confidential of their answers. Next, they were instructed to replay all of the items honestly, and the procedure took approximately 15 min. Dozens of well-trained observers were allotted at each location who cooperated to complete this survey. A total of 920 participants took part in the field survey. Of those, 74 samples contained missing values, which were discarded as invalid. The relative response rate was about 91.96%. More details on data analysis methods appear below.

The sociodemographic information is listed in

Table 2. Descriptive statistics of participant characteristics.

Demographic Variables		Frequency	Percentages
Gender	Male	418	49.41%
	Female	428	50.59%
Age	≤19	62	7.33%
	20–29	357	42.20%
	30–39	220	26.00%
	≥40	207	24.47%
Marriage	Yes	468	55.32%
	No	378	44.68%
Education	Primary school or below	23	2.72%
	Junior middle school	115	13.59%
	Senior high school	203	23.99%
	Associate college	172	20.33%
	Bachelor ’s degree	302	35.70%
	Master degree or above	31	3.66%
Income (RMB)	<2000 (below 317$)	161	19.03%
	2000-4000 (317–634$)	141	16.67%
	4001–6000 (634–950$)	279	32.98%
	6001–8000 (940–1268$)	155	18.32%
	>8000 (above 1268$)	110	13.00%
Job	Enterprise manager	117	13.83%
	Professionals	71	8.39%
	Service workers	251	29.67%
	Clerks	83	9.81%
	Students	158	18.68%
	Others	166	19.62%

. The sample consisted of 49.41% (n = 418) males and 50.59% (n = 428) females. The mean age was 32.39 (min = 13, max = 73, SD = 11.71) with 42.20% of the sample (n = 357) being between 20 and 29 years old. Around 18.68% of participants were students, and 44.68% (n = 378) of participants in the survey were single. 40.3% of respondents had not finished a college degree, while 35.7% had completed a bachelor’s degree.

3.2. Measurement Model

The research model consists of nine variables: self-efficacy, perceived susceptibility, perceived severity, attitude, response efficacy, psychological reactance, freedom threat, perceived legal norm, and behavioral intention. Reliability and validity indicate the preciseness of study processes and the trustworthiness of research findings (Appendix A).

The scale validity was decided by convergence validity and discriminant validity. Convergence validity is based on two conditions: item loading (λ) must exceed 0.7, and the average variance extracted (AVE) for each construct must exceed 0.5. In addition, the criterion of discriminant validity depends on the square root of AVE. As shown in

Table 3. Discrimination validity.

Construct	INT	SE	PSU	PSE	ATT	RE	PR	FT	PLN
INT	0.899
SE	0.250	0.797
PSU	0.516	0.344	0.884
PSE	0.373	0.217	0.630	0.858
ATT	0.437	0.266	0.547	0.578	0.867
RE	0.397	0.227	0.660	0.416	0.590	0.831
PR	−0.092	−0.037	−0.106	−0.067	−0.186	−0.124	0.820
FT	−0.144	−0.069	−0.201	−0.127	−0.205	−0.235	0.527	0.948
PLN	0.494	0.165	0.479	0.302	0.277	0.344	−0.160	−0.160	0.787

, the square root of the AVE for each construct must exceed its correlations with other constructs.

The scale reliability was decided by Cronbach ’s Alpha and composite reliability (CR). Cronbach ’s Alpha and CR’s threshold value was 0.7 and 0.6. As shown in

Table 4. Reliability and convergent validity test of the constructs.

Potential Variable	Item	Factor Loadings (λ)	Cronbach’s α	AVE	CR
Self-efficacy (SE)	SE1	0.615	0.827	0.635	0.836
	SE2	0.845
	SE3	0.902
Perceived susceptibility (PSU)	PSU1	0.854	0.914	0.781	0.914
	PSU2	0.914
	PSU3	0.882
Perceived susceptibility (PSE)	PSE1	0.871	0.915	0.737	0918
	PSE2	0.842
	PSE3	0.902
	PSE4	0.818
Attitude (ATT)	ATT1	0.801	0.922	0.752	0.924
	ATT2	0.895
	ATT3	0.917
	ATT4	0.852
Response efficacy (RE)	RE1	0.798	0.894	0.690	0.899
	RE2	0.768
	RE3	0.900
	RE4	0.851
Psychological reactance (PR)	PR1	0.721	0.889	0.672	0.891
	PR2	0.870
	PR3	0.852
	PR4	0.828
Freedom threat (FT)	FT1	0.927	0.963	0.898	0.964
	FT2	0.963
	FT3	0.953
Perceived legal norm (PLN)	PLN1	0.776	0.858	0.617	0.865
	PLN2	0.818
	PLN3	0.865
	PLN4	0.669
Intention (INT)	INT1	0.908	0.925	0.809	0.927
	INT2	0.946
	INT3	0.841

, Cronbach ’s Alpha was between 0.827 and 0.963, and CR was between 0.836 and 0.964. The current study’s measured Cronbach’ ’s Alpha and CR are satisfactory.

3.3. Structure Model and Hypothesis Tests

We constructed a structural equation modeling (SEM) to explore the relationships of variables of the bicycle riders’ helmet wearing intention during helmet legislation implementation. All of the hypotheses are supported at the significance levels of 0.05 and 0.001. We used a fitting index to test and modify the structural model and utilized maximum likelihood estimation to estimate the path coefficients. After repeated modifications, the model passed the fitting test. The structure of the model is shown in Figure 3.

The structural model results showed that the model provided a very good fit to the data. As seen in

Table 5. Results of the goodness of fit for the theoretical model.

Index	χ2/df	RMSEA	CFI	TLI	SRMR
Check critical value	<3	<0.08	>0.9	>0.9	<0.08
Model parameter value	2.171	0.037	0.960	0.955	0.047
Whether to accept	accept	accept	accept	accept	accept

, all fitness indexes of the measurement model seemed desirable (χ²/df = 2.171; RMSEA = 0.037; CFI = 0.960; TLI = 0.955; SRMR = 0.047). All factor-loading values of the items were acceptable.

3.4. Fuzzy-Set Qualitative Comparative Analysis

To further introduce the methodological characteristics of fsQCA, a configuration refers to a composition of positive, negative, or absent factors or conditions, while a condition is described as if an inevitable outcome cannot be attained without it [15,70,71]. The primary purpose of this approach is to identify which configurations are minimally sufficient or necessary to generate an inevitable outcome and to determine which groups of factors or conditions share a specific set of combinations [72,73,74]. FsQCA has the following three fundamental assumptions. First, fsQCA demonstrates positive and negative causality configurations, revealing an asymmetrical relationship [75,76]. Second, fsQCA identifies multiple equal configurations to the equifinality, demonstrating the same outcome [13,14,15]. Third, fsQCA determines that the effect of conditions on an inevitable outcome is determined by how conditions are coupled instead of the importance of individual conditions themselves, indicating complex causation [77,78].

Based on SEM results, attitude, perceived susceptibility, perceived severity, perceived legal norm, response efficiency, psychological reactance, freedom threat, and self- efficiency were considered independent variables (see Figure 4). High and low helmet-wearing intentions could be regarded as dependent variables in the fsQCA model. The implementation of fsQCA is divided into four phases: (1) calibrate and construct truth table; (2) necessary condition analysis; (3) truth table construction; (4) interpretation of the results.

Calibration of variables: Each variable needs to be converted into a set of 0-1 for calibration before fsQCA analysis. Both the independent and dependent variables in this paper are psychological variables. The original data of psychological variables tend to be distributed in specific areas rather than evenly across scales. Therefore, we use Fuzzy and Relative (TFR) to calculate the membership and correct the skewness of data distribution.

Necessary condition analysis: The analysis of necessary conditions determines whether causal conditions could be considered necessary for an outcome to occur [79] According to these recommendations, a condition is necessary when its consistency score is greater than the threshold of 0.9. Through the analysis of necessary conditions, we find no necessary conditions.

Table 6. Analysis of necessary conditions.

Outcome	INT Intention to Wear a Helmet		~INT Negated Intention to Wear a Helmet
Conditions Tested	Consistency	Coverage	Consistency	Coverage
SE ~SE	0.6971 0.4502	0.8020 0.6133	0.5292 0.7148	0.3674 0.5874
PSU ~PSU	0.8241 0.2907	0.8073 0.4988	0.5160 0.6741	0.3051 0.6981
PSE ~PSE	0.7486 0.3814	0.8036 0.5676	0.5185 0.6968	0.3359 0.6258
ATT ~ATT	0.7824 0.3592	0.8341 0.5398	0.4926 0.7420	0.3169 0.6730
RE ~RE	0.7773 0.3535	0.8155 0.5437	0.5083 0.7085	0.3218 0.6575
PR ~PR	0.4894 0.6530	0.6736 0.7447	0.6289 0.6071	0.5224 0.4178
FT ~FT	0.3934 0.7348	0.6505 0.7357	0.5626 0.6498	0.5614 0.3926
PLN ~PLN	0.7744 0.3767	0.8344 0.5578	0.5051 0.7452	0.3284 0.6659

presents the results of the analysis of necessary conditions considering both the presence and the absence (~) of the conditions for two outcome variables, intention to wear helmets (“INT”) and the negation of intention to wear helmets (“~INT”).

Truth table construction: A truth table was constructed based on the calibrated variables. There are k antecedent conditions and 2^k rows of data in the truth table. The value of k is 10 in this research. List all possible combinations of helmet wearing intentions via the truth table. Due to the randomness of a single sample, the case frequency threshold was increased to 4. This limitation implies that a combination with at least four samples could participate in the fsQCA analysis. Ultimately, 82% of the overall sample was retained. We screen out the sufficient antecedent condition combinations for helmet wearing intention.

Table 7. Results of truth table (partial).

SE	PSU	PSE	ATT	RE	PR	FT	PLN	Number	INT	Raw Consistency	PRI
1	1	0	1	1	0	0	1	10	1	0.954446	0.907515
1	1	1	1	1	1	0	1	24	1	0.952084	0.919038
1	1	1	1	0	0	0	1	9	1	0.951391	0.890251
0	1	1	1	1	0	1	1	4	1	0.949617	0.880861
1	0	0	1	1	0	0	1	4	1	0.948661	0.851596
1	1	0	1	0	0	0	1	5	1	0.948331	0.867685
1	0	0	1	0	1	1	1	4	0	0.905423	0.691095
1	0	1	0	0	1	1	1	5	0	0.904031	0.627985
0	1	0	0	0	0	1	0	4	0	0.846121	0.472036
1	0	0	0	0	1	0	1	7	0	0.842545	0.533918

Note: For each variable, the value of 0 indicates the low level, and 1 indicates the high level.

shows the partial rows of the truth table. The minimum threshold of Raw consistency and PRI for configuration were set to 0.90 and 0.75. Data analysis showed no contradictory combinations, allowing subsequent comparative analyses.

Interpretation of the results: 

Table 8. Configurations that produce high or low helmet wearing intention based on fsQCA.

Variable	High Helmet Intention Solutions											Low Helmet Intention Solutions
	S1		S2		S3			S4				S5		S6
	S1a	S1b	S2a	S2b	S3a	S3b	S3c	S4a	S4b	S4c	S4d	S5a	S5b	S6a	S6b
SE	●		●				●	●	●	●	●		⊗	⊗	⊗
PSU		●		●	●	●	●	●	●	●	●	⊗	⊗	⊗	⊗
PSE		●	⊗		●	●	●			●	⊗	⊗	⊗
ATT	●	●		●	●	●	●		●	●		⊗	⊗	⊗	⊗
RE			●	●	●	●	●		●		●	⊗		●	⊗
PR	⊗	⊗	⊗	⊗			⊗	⊗		●	●		●	⊗	●
FT	⊗	⊗	⊗	⊗	⊗			⊗			●		●	⊗	⊗
PLN	●	●	●	●		●		●	●	●	●	⊗	⊗	⊗	⊗
Consistency	0.93	0.94	0.93	0.94	0.92	0.93	0.93	0.93	0.93	0.94	0.93	0.89	0.92	0.90	0.92
Raw Coverage	0.36	0.38	0.36	0.39	0.49	0.51	0.35	0.37	0.44	0.26	0.11	0.45	0.37	0.36	0.44
Overall consistency			0.888026									0.859813
Overall coverage			0.708782									0.511593

Note: ● or ● indicates the presence of conditions, ⊗ or ⊗ indicates the absence of conditions, blank space indicates “do not care” conditions, ● or ⊗ means core conditions, ● or ⊗ means peripheral conditions.

offers the core and peripheral conditions to achieve high and low helmet-wearing intention. The overall consistency that leads to high helmet wearing intention is 0.888 (>0.8), and the all coverage is 0.709 (>0.5). Meanwhile, the overall consistency that leads to low helmet-wearing intention is 0.860 (>0.8), and all coverage is 0.512 (>0.5). This study’s condition consistency and coverage were in accordance with the standard requirements. The configurations show that the fuzzy set results in 4 first-order solutions (S1–S4) showing acceptable consistency (>0.80) to reach high helmet-wearing intention and two first-order solutions (S5–S6) to get low helmet wearing intention. Each first-order solution contains 2–4 s-order solutions with the same care conditions.

3.5. Bayesian Network

Bayesian network, a probabilistic directed acyclic graphical model [37], reveals the causal relationship between variables. A directed acyclic graph of a BN consists of nodes and edges. Each node shows a variable of the network structure, following specific distribution, and each edge represents the causal relationship between two variables. BN compactly represents the joint density of several variables by the conditional independence relation among them. The implementation of the BN is divided into four phases: (1) model construction; (2) BN test; (3) truth table construction; and (4) interpretation of the results.

Model construction: Structural learning and conditional probability estimation are the two indispensable steps of the BN [80]. In this study, the basic structure of the BN is the SEM hypothesis (Figure 5). Besides, we classified the number of states as low, medium, and high based on a seven-point Likert scale to determine the number of states [81]:

• Factor score is considered “low” between 1 to 2;
• Factor score is considered “medium” between 3 to 5;
• Factor score is considered “high” between 6 to 7.

Table 9. The prior probability of psychological latent variables.

Stata	Latent Variables
	ATT	PSU	SE	PSE	RE	PLN	PR	FT	INT
High	0.765	0.878	0.704	0.803	0.856	0.868	0.256	0.166	0.806
Medium	0.15	0.110	0.215	0.167	0.129	0.113	0.338	0.282	0.145
Low	0.085	0.012	0.081	0.030	0.015	0.019	0.406	0.551	0.049

shows each variable ’s prior probability distribution and state from the original data. For the “Intention (INT),” 11.1% of e-bike riders had “low” intention, 32.4% and 56.5% of them, respectively, had “medium” and “high” intention.

Estimating conditional probabilities depends on datasets and can be implemented by many different algorithms [80,82]. The expectation-maximization (EM) algorithm, a technique for learning latent variables, can automatically handle missing data and generate the conditional probability table (CPT) in the BN [83,84]. Therefore, we choose the expectation-maximization (EM) algorithm to estimate conditional probabilities and update the modeling. The updated BN model is shown in Figure 5.

BN test: BN models usually use error rates and confusion matrix to evaluate model performance [36]. In this study, the sample dataset of 846 cases was randomly divided into 80% data (n = 677) to train and 20% (n = 169) data to test switching behavior.

Table 10. Confusion matrix of the BN modeling.

Confusion Matrix				Error Rate	Total Error rate
Predicted			Actual
Low	Medium	High
4	1	0	Low	20%	12.15%
0	23	8	Medium	25.81%
0	13	132	High	8.97%

shows that the BN can predict 80%, 74.19%, and 91.03% accuracy of the cases with low, medium, and high switching behavior. In addition, BN models also use Spherical payoff (0–1), logarithmic loss (0–infinity), and quadratic loss (0–2) to test model performance [36,85]. In our study, the value of spherical payoff (close to one), logarithmic loss (close to zero), and quadratic loss (close to zero) are 0.877, 0.426, and 0.224, respectively. The above three indicators of this study follow the standard requirements. According to the above indicators, the high accuracy of this BN demonstrates this model can accurately predict e-bike riders to use helmets after the law is enacted.

BN Result different combinations: In the BN model, we could observe the change of the node of dependent variables by adjusting the node of independent variables. The changing size of the target node is the influence degree of different combinations. The results of fsQCA offer the 11 combinations to reach high helmet wearing intention and 4 combinations to reach high helmet wearing intention. When the condition is “presence”, we adjust the nodes in the BN to 100%. When the condition is “absence”, we adjust the nodes in the BN to 0%. Besides, when the condition is blank, we do not change the nodes of the BN.

Table 11. BN result different combinations.

Combinations	Int
	High	Medium	Low
S1a	90.1%	8.36%	1.54%
S1b	89.9%	8.34%	1.79%
S2a	68.3%	22.6%	9.04%
S2b	89.9%	8.31%	1.76%
S3a	90.0%	8.27%	1.73%
S3b	90.0%	8.26%	1.70%
S3c	89.9%	8.29%	1.80%
S4a	89.7%	8.24%	2.09%
S4b	92.2%	6.55%	1.21%
S4c	92.2%	6.55%	1.21%
S4d	75.3%	17.3%	7.37%

shows the influence degree of different combinations to helmet wearing intention. S4b and S4c have the greatest influence on helmet-wearing intention, with the highest probability (92.2%) of high intention.

Sensitivity analysis: To determine the most explanatory variable on helmet wearing switching behavior, the mutual information of sensitivity analysis could determine measures of the relationship between influencing and target nodes, where larger values stand for the more significant influence [86]. According to

Table 12. Sensitivity analysis of helmet wearing intention.

Variable	Mutual Information	Percent
INT	0.86668	100
ATT	0.07295	8.42
PSU	0.04486	5.18
SE	0.03903	4.50
PSE	0.02212	2.55
RE	0.00330	0.38
PLN	0.00166	0.19
PR	0.00028	0.03
FT	0.00005	0.01

, ATT, PSU, and PSE have a more substantial effect on switching behavior.

4. Discussion

This study presents a novel model based on self-protection and psychological resistance, incorporating TPB, PMT, and PRT. The integrated model proved the multi-analytical approach ’s feasibility which integrated SEM, fsQCA, and BN to better understand the mechanism of helmet wearing intention. The SEM result supported all of the hypotheses and proved the integrated model had a mighty explementary power. The FsQCA result gives eleven configurations leading to high helmet wearing intention and four configurations resulting in low helmet wearing intention. The BN result found that S4b and S4c had the most significant influence on helmet-wearing intention. These empirical findings provided a fresh perspective on helmet-wearing intentions’ psychological factors. These empirical findings have various implications for theory and practice, which are discussed in the following.

In Section 4.1, we briefly describe the theoretical implications in terms of both the theoretical framework and the mathematical model.

In Section 4.2, we explain how the specific psychology of riders in a mandatory policy affects their helmet wearing intentions by discussing the performance of two key variables (Psychological Reactance and Freedom Threat) of the PRT in this study.

In Section 4.3, we provide policy recommendations based on this study through a discussion of the relevant variables in PMT and legal norms.

We conclude our discussion with a brief analysis of the differences between Chinese and international helmet policies.

4.1. Theoretical Implications

The theoretical implications are mainly in two aspects. On the one hand, while China’s helmet wearing policy has gradually spread across the country, current research has paid little attention to the specific psychology of cyclists in the context of the mandatory policy. Combining the theoretical frameworks of TPB, PMT, and PRT provides the possibility to understand the helmet wearing intention of Chinese riders in the current stage. On the other hand, the mathematical model of SEM-fsQCA is widely used, however, the variable configurations obtained from current studies are equal. The integration of BNs found the configuration (S4b and S4c) with the highest probability of high helmet wearing intention.

These empirical findings of this research provide a series of theoretical implications to consider. First, this research integrates the theory of planned behavior (TPB), the protection motivation theory (PMT), and the psychological reactance theory (PRT) to build a theoretical framework for e-bike helmet-wearing behavior. The combination of TPB, PMT, and PRT expands the helmet wearing intent in mandatory policy, which is more relevant to the current situation in China. Also, this result demonstrates the conclusion of Veeramootoo et al. (2018) [87], who found that researchers could structure integrated models rather than relying on a single model to comprehend human behaviors better.

Second, we investigate significant predictive factors that influence the helmet wearing decision of e-bike riders in China. Influenced by Chinese helmet legislation enforcement, e-bike riders’ intentions for helmet wearing shows a notable change. Unlike Tang et al. (2021) study, this research demonstrates that psychological resistance and legal norms exist among some Chinese e-bike riders [9]. As far as we know, our research is one of the first explosions to excavate psychological resistance and legal norm among e-bike riders after enacting Chinese helmet legislation.

Finally, we adopted a multi-analytical approach, including SEM, fsQCA, and B.N., and demonstrated the applicability and correlation of these three methods. This research offers a new perspective and theoretical breakthroughs in enhanced analysis methods. SEM is an exquisite method to demonstrate the symmetric relationships among predictive factors. FsQCA provides us with the asymmetric relation between independent variables and dependent variables from a combinatorial perspective. BN offers which configuration could lead to the most significant influence on helmet-wearing intention. In addition, SEM and BN share the same structural model. BN is an essential supplement to the fsQCA results.

4.2. Psychological Resistance

‘Psychological reactance and freedom threat’ represents a motivational state that might arise when freedom is threatened and when individuals feel threats to their freedom. This research’s primary objective is to explore whether e-bike riders’ psychological reactance toward Chinese helmet legislation enforcement influences helmet wearing intention. Firstly, we construct an integrated model based on SEM. Psychological reactance negatively significantly influenced the attitude toward helmet wearing and indirectly affected the intention and is influenced by Freedom Threat in SEM. The SEM result shows that more excellent resistance does not generally cause people to wear helmets like ordinary e-bike riders. Do these findings illustrate that there is no psychological reactance in China? FsQCA result gives a more detailed conclusion.

Secondly, fsQCA results indicate that psychological reactance and freedom threat is the condition of S4d (SE*PSU*~PSE*RE*PR*FT*PLN). A total of 11.2% of the participants hold psychological resistance and freedom threat during the implementation of Chinese legislation. These e-bike riders might have psychological resistance to Chinese helmet legislation enforcement and show resistance to wearing helmets for freedom. In addition, we find these people did not exist PSE. These e-bike riders, who are at risk of the psychological resistance and freedom threat, might not be fully aware the consequences of the collision without helmet are very serious. Therefore, traffic safety education should focus on these people being aware of the protection of wearing helmets on their heads when there is an e-bike accident. What sociodemographic characteristics might this group have?

Thirdly, to determine which e-bike riders with sociodemographic characteristics also have psychological resistance and freedom threat, we conducted fsQCA on e-bike riders of different sociodemographic characteristics according to gender, age, education, and income. This research does not show detailed processes and results for each group since core conditions are more attractive. (More details are available from the authors). In Figure 6, each colored square indicates a variable regarded as a core condition in at least one configuration; otherwise, a blank square represents that the variable never appears as a core condition in any configuration. Regarding psychological resistance and freedom threat, there are many exciting results. Compared with college students and below, undergraduate students are more eager for freedom. Psychological resistance is not uniform across genders, whereas females consider they could not wear helmets properly. Younger (less than 30 years old) e-bike riders hold psychological resistance.

Highly educated, younger males are more likely to resist or desire freedom than the rest. This result provides practical implications for decision-makers. First, traffic police could pay more attention to these riders when they patrol the road. When intercepting highly educated, younger males, traffic police could mainly carry out legal education supplemented by safety education. Second, the government could popularize driving licenses for e-bike riders. The government would make it compulsory for residents to know about the current law on e-bikes during the riding license exam. Third, universities could organize special lectures on the rule of legislation for e-bikes since the younger well-educated group is the leading group of psychological reactance.

4.3. Protection Motivation and Legal Norm

4.3.1. Protection Motivation

Perceived susceptibility and perceived severity reflect the perceived risk of e-bike riding. Perceived susceptibility directly influences intention, and perceived severity indirectly affects the intention by the Attitude in SEM. Perceived susceptibility and perceived severity exist in nine and five configurations leading to high helmet wearing intention. However, two configurations contain no perceived severity. Thus, most people choose to wear a helmet when they understand the expected severity of the consequences of a crash involving an e-bike riding without a helmet. At the same time, some e-bike riders consider that the possibility of a head injury after riding an e-bike without a helmet is not high. Thus, accident videos of e-bikes can be used to demonstrate the seriousness of accidents without helmets in traffic safety education to motivate e-bike riders to wear helmets. Meanwhile, we could not exaggerate the possibility of an accident in traffic safety education.

Response efficiency shows personal confidence in the effectiveness of helmets. Response Efficiency indirectly affects the intention by the Attitude in SEM. Response efficiency exists in seven configurations, leading to high helmet wearing intention. E-bike riders still have doubts about the effectiveness of helmets. Therefore, governments could control helmet wearing rates in two ways. First, traffic safety education could focus on promoting the usefulness of helmets. Second, the quality of helmets on the market is uneven. Jiangsu market supervision and Administration Bureau found that the helmet failure rate was 38.5% in 2020 [88]. To keep e-bike riders safe, the government should strictly control the quality of helmets.

Self-efficiency refers to the subjective probability that one is capable of wearing a helmet. Self-efficiency indirectly affects the intention by the Attitude in SEM. Self-efficiency exists in seven configurations leading to high helmet wearing intention. Thus, traffic safety education and publicity could define wearing a helmet as common sense and provide a series of easy solutions to get a helmet. For example, one could deploy shared helmet access points within urban areas for people who do not wear helmets.

4.3.2. Legal Norm

Perceived legal norms significantly influenced the attitude toward helmet wearing and indirectly affected the intention by SEM. Meanwhile, perceived legal norm exists in 9 configurations leading to high-wearing intention and is the core condition of S4b and S4c, significantly influencing helmet-wearing intention. Thus, we believe that legal norm is an essential factor influencing helmet-wearing intention. This result is similar to the conclusion of García-Ferrer et al. (2006) [89], who considered that legislation relating to traffic accidents could deliberately make traffic participants avoid traffic violations. On the contrary, Tang et al. (2021) and Kim (2018) get disparate results [9,90].

Two main reasons might lead to differences in the Chinese unique e-bike environment. First, the penalty for not wearing helmets is 50 CNY (US$7) in China. A total of 50 RMB could be used to buy an e-bike helmet and is a total warning to e-bike riders who belong to economically disadvantaged subpopulations [9]. Second, at the beginning of Chinese helmet legislation, the local government increased the police force to maintain high helmet wearing rates. E-bike riders started wearing helmets due to traffic police on the street. Therefore, maintaining high helmet wearing rates needs enough police. However, China ranked 127th globally with 143 policies per 100,000 inhabitants, less than an average of 400 policies per 100,000 inhabitants In East Asia [91]. After the Chinese helmet legislation craze, China might not have enough police to maintain high helmet wearing rates. The government needs more cost-effective measures. For example, traffic police cannot check helmet wearing at a fixed place and time.

5. Conclusions

• Based on the current research, this study combines TPB, PMT, and PRT to construct a comprehensive theoretical framework considering riders’ resistance to mandatory policies. The results of SEM show that all hypotheses are valid.
• fsQCA results showed that a total of 11 configurations resulted in high helmet wearing intentions. To explore the differences between the 11 configurations, 2 optimal configurations (S4b and S4c) were identified within BNs.
• Through further analysis of socio-demographic characteristics, this study found that young men with higher education were more likely to resist or desire freedom than others.
• Limitation: The source of data is limited to one city, Zhenjiang, ignoring urban-rural differences and variations in different policy implementation areas. The raw data are cross-sectional data, but the helmet chapter wearing intention may change with the gradual implementation of the helmet policy.