2.1.1 Individual socio-economic attributes.

In terms of individual and household socio-economic attributes, gender, age, income, and education are all taken into consideration. Regarding gender, males exhibit a higher willingness to use non-motorized transportation and greater tolerance for its drawbacks [11]; however, females demonstrate stronger robustness to economic factors [13], and studies have shown that they are more accepting of transfer modes of transportation than males [8]. The main concerns for females are safety-related [47], and there is a lack of consideration for gender differences in various aspects of transportation, resulting in inherent gender barriers [10]. Young people have a more positive attitude toward bike-sharing [12], with affordability being an important factor [20]; however, they are also the group least willing to transfer multiple times [25]. Middle-aged individuals have more difficulty accepting bike-sharing as a new phenomenon [10], but there is still potential to be tapped into; they dislike long walks as a mode of transfer and are very sensitive to travel costs [8]. The biking rate among the elderly is very low, with the comfort and safety of biking facilities being their top concerns [11]. Income and education are negative factors for low-carbon travel modes [8], and some scholars believe that the positive correlation between income and education levels makes the preferences of these two groups tend to be similar [48], with electric bikes being more easily accepted by those with higher income and education levels [13].

2.1.2 Travel background.

Whether for leisure or commuting, people are willing to use bike-sharing or SE for their trips [19, 21]. However, the number of transfers has a significant negative impact on this mode of travel [26]. Non-motorized travel is generally affected by transfer distance, with a threshold of 3 km, showing a positive trend initially, followed by a negative one [17]. Some studies indicate that SE has a larger threshold than bike-sharing [13]. Travelers are more willing to walk for transfers within 200 meters, but this willingness drops to only 20% at 700 meters [17]. Travel cost is inversely related to the willingness to choose this mode of transportation [20].

2.1.3 Built environment.

The quality of riding infrastructure is a primary concern for riders [19]. The continuity and exclusivity of bike lanes, as well as the coherence of the riding experience, are crucial factors [49]. High-intensity land development, mixed-use areas, and the density of points of interest (POI) also serve as significant positive influences [23, 26]. Conversely, damaged bike lanes and the lack of safety measures like barriers greatly reduce the willingness to ride [26], especially at night [6]. Additionally, appealing environments such as parks, lakes, and greenery strongly attract users [32].

2.1.4 Attitudes and preferences.

Although users with a strong awareness of environmental protection and health tend to prioritize low-carbon transportation methods [34, 42], ongoing publicity can effectively enhance public awareness [17]. However, people are more sensitive to improvements in comfort and service levels [35, 38], and their need for safety outweighs the need for flexibility [37]. This is especially crucial for women [34].

2.1.5 Weather.

Weather is a significant factor affecting riding, with outdoor modes like SE being more impacted than enclosed travel methods [50]. Pleasant weather can boost users’ willingness to ride [14], while rain and snow have the most significant effects [51], causing up to 75% of rides to be canceled due to rain. Shared bike rides starting near metro stations are more susceptible to rain, leading to a significant decrease in riding frequency and duration [43]. This impact is even greater for inexperienced riders, as rain increases the risk by 2.5 times and snow by 4 times [49]. However, this influence mainly comes from immediate weather conditions, with subsequent weather changes being less important [44]. Temperature is another critical factor, with both summer and winter being less ideal for riding, while humidity amplifies the impact of high temperatures in summer [32].

The effects of wind vary; it can alleviate heat and increase the willingness to ride in summer [23], with riders often increasing their speed to create a cooling breeze. In contrast, frequent strong winds in autumn negatively impact riding, with each unit increase in wind speed decreasing the likelihood of riding by 0.8% [44, 45].Different groups have varying degrees of tolerance to weather, influenced by travel mode, purpose, environmental attitudes, and travel habits. This heterogeneity means that the impact of weather on travel behavior varies widely [46]. Understanding this variability is crucial for studying travel choices. However, as Nordbakke and Olsen point out, there is limited research on "weather tolerance" and its heterogeneous effects from the perspective of user diversity [9, 46]. Investigating this can highlight the sensitivity to weather impacts and underscore the travel behavior tendencies of different groups, which is of significant importance [24].

2.2.1 Behavioral theory.

User satisfaction and loyalty are overall evaluation metrics formed by users during purchase and use, primarily used to assess user experience and measure usage intentions. Satisfaction with the SE-metro transfer mode refers to SE users’ experiences compared to their expectations regarding the products, services, infrastructure, and environment involved in using this travel mode. It is an emotional response. Loyalty to the SE-metro transfer mode mainly reflects whether users continue to choose SE as a travel mode, which is a behavioral decision and an extension of satisfaction research.

Since travel mode choice is generally considered a rational behavior, current research in the travel field often uses the Theory of Planned Behavior (TPB) as the foundational theory for constructing user satisfaction models [52]. However, as an emerging travel mode, SE differs from more conventional travel modes like cars and buses and is still in the process of being gradually recognized and accepted by the public. Thus, TPB is not the optimal explanatory theory for SE. The Technology Acceptance Model (TAM), based on the Theory of Reasoned Action (TRA) and TPB, was first proposed by Davis in 1989. It is a theoretical model for explaining people’s adoption behavior, acceptance level, or usage intentions of new things [53]. TAM has the advantages of a simplified structure, robust model, and strong explanatory power for user behavior, making it more suitable for explaining travel choice behaviors related to SE.

As research has progressed, scholars have continually refined the Technology Acceptance Model (TAM) from three perspectives: the model framework, internal model factors, and the introduction of new theories into the model. Consequently, TAM evolved into TAM2 [54] and TAM3 [55]. Firstly, this evolution led to both a simplification and expansion of model variables. While removing attitude as a classic model variable and diminishing the influence of subjective norms, researchers also began to recognize the impact of user characteristics, environmental factors, and temporal factors on technology acceptance [55]. Secondly, the focus of research subjects has gradually shifted to specific groups. The scope of investigation has extended from internal model factors to external influence variables and from the information domain to multidisciplinary fields such as social, management, economic, and decision sciences [56]. These advancements have significantly broadened the applicability of the model.

2.2.2 Methodological approach.

The commonly used statistical analysis methods in existing travel mode choice research include the Discrete Choice Model (DCM), which primarily aims at prediction; the Structural Equation Model (SEM), which primarily aims at explanation; and the Hybrid Choice Model (HCM) [35, 43, 44]. Our focus is on the intention to choose non-motorized transportation, particularly in the context of its integration with public transit. Relevant literature is summarized in Table 2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Summary of literature.

https://doi.org/10.1371/journal.pone.0309953.t002

The Discrete Choice Model (DCM) is a non-aggregate model based on utility theory, which expresses travelers’ preferences for travel modes through utility value functions. It typically employs the principle of Random Utility Maximization (RUM), assuming complete rationality and commonly used methods include the Logit model and Probit model [60], with the Logit model being the most widely used. However, DCM generally assumes that the choice attitudes with individual heterogeneity are stable and consistent [38], and it explores less the impact of traveler heterogeneity [56], making it difficult to explain psychological factors such as individual choice preferences significantly related to behavioral intentions [61]. To address this issue, SEM is introduced. SEM typically incorporates unobservable psychological factors such as attitudes and perceptions into the structural model in the form of latent variables and quantifies them through measurement models in the form of questionnaire items. It is widely used in areas such as travel satisfaction and loyalty, increasing riding rates [36], and differences in policy influence [59], with overall robust research conclusions [62] but with fewer studies on attitudes towards transfer travel. The Hybrid Choice Model (HCM) is a combination of both models, but the increased complexity of the model limits its practicality.

4.3.1 Exploratory factor analysis.

The main purpose of EFA is to identify whether the latent variables align with the predetermined items in the conceptual model and to reduce the observed items while maintaining the model’s explanatory power. The study employed the Partial Least Squares method to test 32 items, of which COU5, TRA2, PEOU5, and LOF4 were excluded due to factor loadings less than 0.7. Consequently, 28 items were retained. The KMO value after excluding items was 0.877, indicating that the sampling was adequate. Bartlett’s test of sphericity was significant, and the rotated component matrix based on the maximum variance method showed that the absolute values of all elements exceeded 0.5. The questionnaire items could be divided into 8 different latent variables, namely: (1) Convenience of Use (COU). Includes items like "neat placement of vehicles" and "easy to scan and rent a vehicle”; (2) Perceived Safety (SAF). Includes items like "good traffic order" and "few interferences while riding”; (3) Comfort (COM). Includes items mainly related to the riding comfort of the e-scooter.; (4) Ease of Transfer (TRA). Defined by items such as "reasonable layout of metro entrances and exits," "more convenient than bus transfers," and "close proximity of pickup and drop-off points."; (5) Perceived Ease of Use (PEOU). Represents indicators related to perceived ease of use, such as "wide enough bike lanes" and "shaded paths."; (6) Weather Preference (WEA). Measured by different respondents’ tolerance to four weather conditions: "haze, light rain, heat, and cold."; (7) User Satisfaction (SAT). Composed of various evaluation factors such as "willingness to recommend to others" and "se has improved the quality of life."; (8) User Loyalty (LOF). Measured by users’ choice of se as a transfer mode in various hypothetical scenarios.

4.3.2 Simplification of questionnaire items.

We aim to use the PLS method to select the best-fitting model among M1~M3 of different questionnaire sizes based on the streamlining principle. R² and BIC values are both indexes for determining model fit, but R² tends to favor models with a larger number of variables, which is a drawback that can lead to unnecessarily complex models. At the same time, BIC is based on Bayesian Information Theory, which better represents the overall best goodness-of-fit. Distinguishing validity is a crucial criterion for determining whether a questionnaire is refined. HTMT (heterotrait-monotrait ratio) represents the ratio of between-trait correlation to within-trait correlation and is an indicator for assessing discriminant validity. An HTMT can be calculated between traits (i.e., latent variables) higher than 0.90, meaning poor discriminant validity [70]. In this study, HTMT was selected as the basis for simplifying questionnaire items, while BIC was used as the discriminant indicator for model comparison.

Deletions for streamlining are made in order of magnitude based on their model’s explanatory power. PU is jointly reflected by four second-order latent variables (COU, SAF, COM, and TRA), which are conceptually easily cross-repeated. In this paper, the simplification started from this point onwards, and we used the factor loading to measure their contribution’s magnitude. The questionnaire items screened by EFA were used as the Base group. Based on the principles of single variable control and not missing any possibility, we started from COU4 (0.739) and SAF4 (0.756), which have the smallest factor loading, respectively. The items were divided into two groups to list all the deletion schemes stepwise incrementally and then terminated when each latent variable’s observation variables were reduced to three. Finally, a total of 10 kinds of schemes were taken into account, as shown in Table 3. Except for schemes 8 and 9, which were eliminated due to not meeting the requirements of HTMT values, the values of the remaining schemes were relatively close, indicating that the model generally conforms to statistical criteria. In the most simplified schemes, 5 and 10, the former obtained a smaller BIC value, resulting in a final questionnaire with 23 items. This paper investigates how to improve user satisfaction and loyalty, combining BIC values of Base and 10 schemes with the principle of simplicity. It can be seen that M2 is the optimal model, as shown in Table 3. Therefore, this paper selects "M2+Scheme 5" as the optimal model for the following classification comparison and as the base model of M4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Comparison of model fitting and question item screening results.

https://doi.org/10.1371/journal.pone.0309953.t003

4.3.3 Applicability analysis of research methods.

For the TPB model, another alternative research method, when calculating the discriminant validity, it was found that the second-order factors PEOU and SAF, which constitute the ’subjective norms’ variable, do not meet the requirements. The correlation values of 0.822 and 0.873 in Table 4 far exceed the square root of the AVE value of this variable, which is 0.786. This indicates that the model structure of TPB cannot adapt to the psychological motives of users under the conditions of the SE-metro transfer mode. At the same time, the second-order latent variables COU, COM, and TRA, which constitute the ’perceived behavioral control,’ are all close to the critical value (0.665, 0.762, and 0.748), further proving that the TAM model is the appropriate research method for SE transfer travel.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Discriminant validity of the TPB comparative model.

https://doi.org/10.1371/journal.pone.0309953.t004

6.2.1 Weather tolerance.

The inclusion of weather tolerance helps enhance the model’s explanatory power. This latent variable can directly influence PEOU and LOF. Despite many related studies, few have considered subjective weather perceptions as influencing variables. It is also important to design questionnaire items more specifically to prevent overlooking survey information such as light rain.

Improving perceived ease of use can offset the impact of bad weather. Different regions should have different strategies based on factors such as precipitation, economic conditions, temperature, and air quality. For example, rubber seat cushions without sponges can be used for SE to improve drainage and anti-freezing performance. Planting large canopy trees as roadside trees can effectively alleviate the impact of haze, light rain, heat, and cold weather while ensuring continuous riding. Additionally, continuous rain shelters can be set up on roads with heavy non-motorized traffic, a measure already applied in cities like Singapore and Hong Kong, China. Maintaining the riding quality and safety of roads, ensuring continuous riding environments, and increasing riding speed can also effectively reduce the impact of weather.

The significant path coefficient from weather to loyalty implies that people are easily inclined to change their travel modes due to weather conditions, and they are only willing to use riding modes like SE in very comfortable situations. The impact of weather on loyalty can be categorized as follows: First, weather can impact personal image. Variable weather conditions may cause users to sweat or get wet in the rain, damaging their image at work and in daily life. Therefore, providing locker rooms and shower facilities in office buildings can facilitate users in cleaning their clothes, effectively enhancing their resistance to the effects of weather [75]. Second, weather can impact work or travel plans. Previous studies have suggested that facilities such as locker rooms can help avoid the constraints of riding in formal attire, providing conditions for changing from sportswear to work clothes [19]. Third, weather can hinder the choice of users with special needs. Equipping the vehicle with a mini disposable raincoat box and a foldable storage box can meet the needs of sudden rain and carrying large items.

In the model results, the impact of weather on loyalty (LOF) is greater than that on perceived ease of use (PEOU), indicating that weather’s influence on transfer choices is more emotional rather than a rational choice based on the real riding environment. It is worth noting that the time spent on transfer travel exposed to outdoor environments is short, and users’ inquiries focus on their sensitivity to minor weather changes. Overall, the fluctuation is small, so the numerical values of the weather-related paths are not significant.

6.2.2 Perceived Usefulness.

The significant path from Perceived Usefulness (PU) to Loyalty (LOF) implies that even under the assumption of suboptimal satisfaction, the construction department can still maintain users’ loyalty to SE by improving PU. The importance ranking of the four latent variables comprising PU, namely SAF, COM, TRA, and COU, is consistent with expectations, with all of them receiving high weights. This is in line with previous findings in shared bicycle research, where users are more concerned about the safety of SE. This is mainly due to the generally poor traffic safety conditions and road infrastructure, as well as the higher speed and greater weight of SE compared to bicycles.

We must also consider the influence of some underlying factors. Although all respondents have experience with the SE-metro transfer mode, the transfer rail function is only part of the purpose for choosing the SE. This may be an important reason for the relatively low TRA values.

6.2.3 Perceived Ease of Use.

Perceived Ease of Use (PEOU) not only directly affects satisfaction but also indirectly influences users’ loyalty to the SE-metro transfer mode. This indicates that the social reputation of SES is closely related to a good usage environment, including continuous and ample riding space, well-maintained road surfaces, shaded areas, and clear directional signs. Neglecting to optimize the riding environment may lead to the government making incorrect decisions regarding support for SES, consistent with a comparative study on the United States and Canada [76]. Most of the lower-rated items are related to government policies and infrastructure, indicating an urgent need for adjustment in Changsha’s government policies related to SES development. This includes infrastructure construction, including transfer functions, which should listen more to the opinions of riders.

Previously, the ban on SES by the Changsha government was based on complaints from citizens. However, these complaints, while reflecting some actual issues with SES, mostly highlight the deficiencies in the ease of use of riding space and facilities, such as mixed traffic with motor vehicles and illegal parking. In fact, these issues are mostly caused by private car owners. Still, based on the principle of fairness, low-income groups have the right to equal road rights as private car owners. At the same time, the mismatch between the number of parking lots for motor vehicles and non-motor vehicles has caused problems with nowhere to park SE, which also deserves attention.

6.2.4 User satisfaction.

Satisfaction is the most important psychological latent variable in the choice of SE, with few users giving extremely negative evaluations. They generally exhibit a more satisfied and concentrated attitude. Maintaining an affordable price is a decisive factor in satisfaction. Since the resumption of SE operations in Changsha, prices have doubled or even surpassed those of buses and subways. This indicates that there are many deficiencies in current operation management. Firstly, the policy of maintaining multiple operators and controlling the number of SEs put into operation has led to a lack of competition among operators. They are confident in maintaining high prices even when there is a high demand during peak hours to achieve higher profits. Secondly, the limited number of SE put into operation has led to high fixed costs for operators, which is also one of the reasons for the increase in prices.

Gradually increasing the number of operating vehicles and appropriately reducing the number of operators is key to future CSSE policies. The concept of low-carbon environmental protection is gradually becoming popular in China, which can promote the formation of satisfaction. However, as a developing country, Chinese people value the sense of honor represented by their means of transportation: using a car is considered more decent while using a non-motor vehicle is seen as a sign of a less successful career. This outdated notion is particularly evident among middle-aged and older people, which is also a major reason why SEs are difficult to promote. Therefore, it is important to reshape users’ social values, establish a good social image of riding, and actively promote environmental values. A good measure is to recommend the use of SEs to friends around them. The power of mutual recognition may be more effective and lasting.