1. Introduction
How to achieve sustainable transportation for individuals is one of the key problems of future sustainability around the globe. Lately, the concept of sharing has become increasingly popular in the discussion regarding solutions to reduce private car use. This paper investigates the determinants of intention to use carsharing services by using an integrated model of psychological predictors of travel behavior.
Carsharing can be defined as a short-term car rental service that provides to its users access to vehicles for a short-term time interval [
1]. Carsharing services are different from car rentals, carpooling, and ride-hailing services, even though they may share some similarities. For car rentals, generally, the customer signs a designated contract with specifications of the service each time he/she rents a car for a certain number of days. In contrast, the members of carsharing services sign a service contract just once, and they may use one of the cars from the fleet whenever they want, given the availability of the fleet.
Carpooling can be defined as an arrangement in which two or more persons share a private vehicle for a trip or part of a journey, and the passengers contribute to the driver’s expenses [
2]. Some carsharing services allow carpooling among their members, but that is not the rule. The typical carsharing use is that one person uses the car for their trips, and they may be accompanied by a person they know.
Ride-hailing services became extremely popular with Uber, and they are categorized as a taxi-like service. In this case, the service user does not drive the car, and they need to order the service at the moment of the desired trip.
As a relatively new service on the market, different business models of carsharing have been used by different segments of the population, and these services vary in suitability for different trip purposes [
3]. The users and non-users of carsharing have been observed to vary substantially in terms of motivational aspects to use the services, as well as in their psychological attributes and patterns of behavior [
4]. Understanding these differences is a key aspect of fostering policies on carsharing and improving the services to benefit most users and those that want to start to use the service.
The most common classification of carsharing is dividing the services into (1) roundtrip (the vehicle is returned to the same station); (2) one-way, station-based (the vehicle is returned to a different designated station); (3) one-way, free-floating (the vehicle can be returned within a geographic area) [
5]; and (4) peer-to-peer (P2P) [
6]. P2P carsharing is a system that allows car owners to rent out their vehicle when they are not using it. Other drivers can rent the vehicle via an online platform that facilitates the exchange.
The evolving patterns of business models of carsharing also vary across countries, which is the case in Sweden and Italy. These two countries were selected for this investigation, because the users from both nationalities share some common motivational aspects of using carsharing, while the countries have some market particularities worth investigating. Additionally, funding-related considerations partially influenced the choice of countries for investigation.
First, both countries have an increasing market of different carsharing business models. Carsharing in Sweden is still a niche in big cities, with a dominant player owned by an incumbent car manufacturer. The free-floating model has not developed much, while station-based services are prevalent [
7]. In Italy, carsharing has consolidated as a recently increasing phenomenon, across 35 cities, with a big fleet and a big market share of free-floating business models [
8].
Second, there is some evidence that Swedish users of carsharing are more oriented toward practical and economical motives to use the carsharing services than to social and environmental aspects [
9]. Combined with an increased price for public transportation, that negatively affects the use of carsharing, since they are complementary to each other [
10]. Similarly, Italian users are also oriented toward convenience and cost savings, compared to the use of public transportation and owning a private car [
11].
Given the variations at individual and national levels previously described, the present study compares the determinants of intention to use carsharing among four groups: users and non-users of carsharing services in Italy and Sweden. To compare these groups, multigroup confirmatory factor analysis (MGCFA) in structural equation modeling (SEM) is performed (a detailed description of the analysis of invariance is described in the method section and the importance of assessing measurement invariance when interpreting differences between groups is discussed). The outcomes of this investigation increase the knowledge about the use of carsharing and help to identify the behavioral and psychological factors that mostly influence people’s intention to use it.
SEM is widely used in studies on attitudinal issues related to travel behavior and transportation; however, it has been mostly applied to single samples and more traditional modes of transportation, such as the private car, public transportation, and active modes of transport [
12,
13,
14,
15,
16]. This study adds to the literature an investigation of the psychological and behavioral aspects involved in the use of carsharing services across groups and nationalities.
Carsharing
Carsharing services became a disrupting innovation within the context of the sharing economy [
17] and a promise of a more environmentally friendly alternative for transportation. Without taking from its users the benefits typically associated with the instrumental values of cars, such as accessibility and convenience, the carsharing services represent an alternative to private vehicle ownership [
18]. Moreover, the carsharing services may have fleets of electric vehicles, promote the use of other modes of transportation such as cycling, walking, and public transport, and reduce the demand for parking areas in city centers [
5].
While research has shown evidence of carsharing services as an alternative driving personal travel choice toward a transition to sustainability, there is also evidence that sharing assets, such as cars, is reinforcing the social and economic unsustainable paradigm of overconsumption [
19]. The present study has the potential to give one more piece of information to explain this paradoxical puzzle by adding the psychological dimension to the investigation of sharing behavior.
There is strong evidence that psychology can make a difference in the research of climate change [
20,
21] and be a powerful tool to predict daily mobility behavior [
22]. For instance, psychological research has identified that carsharing’s benefits of lower prices and reduced environmental impact may be more attractive only for those who have lower psychological ownership profiles [
23]. There is also evidence that carsharing users do not necessarily own fewer vehicles than non-users or use the service primarily for environmental reasons [
24]. Moreover, it has been estimated that a mode shift to carsharing services may result in an unforeseen reduction of preferences for public transport and bicycle [
25].
Since carsharing is relatively new in the market and has not been fully investigated from a behavioral perspective, there is no strong support for one psychological theoretical framework instead of another to investigate the use of this service. Moreover, there is a need to further investigate the behavioral correlates of carsharing use [
26]. Therefore, the present research proposes a psychological framework, based on robust theoretical background and ease to communicate among non-academic professionals and the population in general. The main goals of this framework are to provide leads for policies on the topic of carsharing and its outcomes for mobility and to reduce the gap of knowledge of psychological correlates of carsharing use.
4. Results
4.1. Measurement Model
To investigate the validity of the measurement model, a six-factor model was tested by using CFA. Behavioral intention, habit, climate morality, subjective norms, trust, and control were modeled as interrelated latent variables.
The interrelatedness of the items within each latent variable was measured by Cronbach’s alpha [
60], and the lower bound to the reliability of the scales was reported by the λ
2 [
61]. They were assessed separately by each group.
Even though Cronbach’s alpha has been largely reported as a measure of internal consistency of a test [
62], other authors strongly advocate that this is a misunderstanding of the seminal work of Cronbach [
55] and that better measures of lower bound should be used instead [
63,
64,
65]. Given the tradition among psychologists to report alpha and the work proposed by [
62], this article reports both measures. Moreover, for the specific case of this study, there were no major differences between the Cronbach’s alpha and λ
2, and they were all under the acceptable range (see
Table 2). This is the case because all latent variables were unidimensional, and in these cases, alpha performs well [
63].
The missing data were treated by case-wise (or ‘full information’) maximum likelihood (FIML) estimation, and the standard errors were estimated by the robust method, as strongly advised by previous studies [
66,
67] (the percentage of missing values by each item of the scales are reported in the
Supplementary Material).
Four steps of analysis of invariance were performed across groups and nationalities (Italian users, Italian non-users, Swedish users, Swedish non-users). Configural invariance was tested in the first step to examine the goodness of fit of the factor structure across the groups. The factor loadings and item intercepts were unconstrained, and the model was fit onto the groups. The second step tested the metric invariance to examine if the factor loadings were equivalent across the groups. Therefore, each factor loading was constrained equally across groups. The purpose was to inspect if the construct of each latent variable that predicts behavioral intention had the same meaning across groups. Scalar invariance was assessed in the third and fourth steps to examine if the latent means can be compared across groups. The item intercepts of the factors habits and climate morality were constrained as equal across groups in the third and fourth steps, respectively [
68].
For each step of the confirmatory factor analysis (CFA), the chi-square statistic test was reported along with other fit indexes (CFI, RMSEA, SRMR) (see
Table 3). The ∆CFI values were all under 0.01, and therefore the most strict model (M4) was kept. The means and standard deviations for all items are reported in the
Supplementary Material.
Since the items of the factors subjective norms, trust, and control had equivalent intercepts between users and non-users, they were not tested for scalar invariance. The reasons are because these factors are strictly connected with the respondent’s previous experiences with carsharing and because the phrasing of the items was adapted for the reality of each group.
The chi-square (χ
2) tests the exact-fit hypothesis that there is no difference between the covariance in the tested model and the population covariance matrix. By rejecting this hypothesis, one may conclude that there is evidence against the given model explanation and that the discrepancies between the given model and the data should be further interpreted. The binary decision of rejecting or not the hypothesis does not imply that the model should be retained or rejected [
69]. This argument is based on the fact that the chi-squared test is sensitive to sample size. For instance, the sample size in this study is considered large for this kind of analysis, and small discrepancies between the model and the data may have led to this
p-value.
The four steps of the confirmatory factor analysis (CFA) indicated a good fit, although the chi-square test had a significant p-value on the four steps. The first step presented χ2 (964) = 5394.60, p < 0.001. The second step presented χ2 (979) = 5611.82, p < 0.001. The third step presented χ2 (1004) = 6520.62, p < 0.001. The fourth step presented χ2 (1012) = 6438.98, p < 0.001.
The values of the root mean square error of approximation (RMSEA) were acceptable, and the 90% confidence interval has a good narrow range, giving confidence in a good fit. This index is an indicator of a badness-of-fit statistic in which values closer to zero indicate better results. The fit of the analyzed path model (CFI) was about 94–95% better than that of the independence model (the null model). Moreover, the standardized square root of the average squared covariance residuals (SRMR) indicated a good fit [
69] (see
Table 3).
Given the results of the values of GIFs, it is possible to address the first research question (RQ1) and assume that the proposed model is a good model to predict carsharing use. Moreover, the results of the analysis of invariance allow group comparisons to answer the research questions RQ3 and RQ3a in the coming section, together with RQ2.
4.2. Structural Model
The results in
Table 4 and
Table 5 show the effect of habit, climate morality, subjective norms, control, and trust on the behavioral intention to use carsharing for users and non-users in Sweden and Italy. These results give support to answer RQ2, RQ3, and RQ3a.
Control was the main predictor of intention to use carsharing among users (βSweden = 0.77, p < 0.001; βItaly = 0.48, p < 0.001) and non-users (βSweden = 0.32, p < 0.001; βItaly = 0.43, p < 0.001). Driving habits negatively predicted the intention to use carsharing (βUsers Sweden = −0.21, p < 0.001; βNon-users Sweden = -0.07, p = 0.021; βUsers Italy = −0.79, p = 0.041), except for Italian non-users (β = −0.00, p = 0.743). Subjective norms positively predicted the intention to use carsharing among users (βSweden = 0.12, p = 0.014; βItaly = 0.37, p < 0.001) and non-users (βSweden = 0.15, p < 0.001; βItaly = 0.36, p < 0.001). Trust was a predictor of intention only for the Italian groups (βUsers Sweden = −0.00, p = 0.926; βNon-users Sweden = 0.05, p = 0.099; βUsers Italy = 0.25, p < 0.001; βNon-users Italy = 0.10, p = 0.002). Climate morality had a small negative effect on the Swedish groups only (βUsers Sweden = -0.09, p = 0.028; βNon-users Sweden = −0.14, p < 0.001; βUsers Italy = −0.01, p = 0.844; βNon-users Italy = 0.03, p = 0.342).
Five chi-squared tests were performed to test the null hypothesis that the regression coefficients were the same across groups for the five parameters. Each parameter was constrained at each time, and a χ2 difference test was performed to assess the difference between the model fits, one model unconstrained for the groups and one model constrained.
To compute an χ
2 difference test, the difference of the χ
2 values of the two models in question is taken as well as the difference of the degrees of freedom. If the χ
2 diff-value is significant, the model with more freely estimated parameters fits the data better than the model in which the parameters in question are fixed [
70].
The results indicate that there are differences between the unconstrained and constrained models for the parameters habit (χ2 diff (3) = 17.1, p < 0.001), climate morality (χ2 diff (3) = 9.2, p < 0.05), and control (χ2 diff (3) = 36.1, p < 0.001). No differences were identified for the parameters subjective norms (χ2 diff (3) = 3.3, p = 0.33) and trust (χ2 diff (3) = 7.7, p = 0.05). Therefore, comparisons among coefficients are only statistically significant at 95% confidence interval for the coefficients of the variables habit, climate morality, and control.
Table 6 summarizes the hypothesis testing. The results show that H1 was not supported. The effect of climate morality was statistically significant only between the Swedish groups, with a negative effect. H2 and H3 were supported among all groups. H4 was supported only between the Italian groups, and H5 was supported among all groups except for Italian non-users.
5. Discussion
The main purpose of this article was to propose and test an integrative structural model based on the frameworks of the TPB and TAM. This model specifies that intention to use carsharing depends on habits, climate morality (personal norms and environmental concern), subjective norms, control (perceived behavior control, ease of use, and perceived usefulness), and trust.
In this section, the research questions are firstly discussed, followed by a specific discussion of each latent variable modeled in this study and methodological aspects. The section is finalized with limitations and a proposal for future research.
The results of this study provide evidence to support both the TPB and the TAM as important frameworks to investigate and predict transport behavior, answering the RQ1 (to what extent does the proposed model, based on TPB and TAM, explain the intention to use carsharing in a near future?). They also show the importance of looking at behavioral aspects related to the use of new transport services, such as carsharing. Therefore, the results and discussion provided by this study have many implications for the transport research area and environmental psychology.
Control was the main positive factor to predict the intention to use carsharing. Among the factors with negative effects on behavioral intention, driving habits had similar negative effects between the Swedish groups, but did not affect Italian non-users. Climate morality had similar negative effects between the Swedish groups, but no effect between the Italian groups. These results successfully answer the RQ2 (What are the main incentives and barriers to the intention to use carsharing?) by indicating that perception of control was the main incentive to intention to use carsharing, while driving habits were the main barrier.
The estimates and CIs reported in
Table 4 and
Table 5 give support to answer the RQ3 (Do habits, climate morality, subjective norms, control, and trust differently affect users and non-users of carsharing regarding their intention to use this service?) and RQ3a (If there are substantial differences between users and non-users, to what extent do they differ depending on their cultural backgrounds?). Swedish users and non-users do not differ substantially in terms of psychological predictors of the intention to use carsharing. Between the Italian groups, driving habits did not affect the behavioral intention of non-users and had a small effect on users.
These results give evidence that users and non-users of carsharing in Sweden do not differ to great extent regarding the effects of driving habits, climate morality, and control on the intention to use carsharing. The greatest differences come when comparing Swedish and Italian groups. Climate morality was a negative predictor for Swedish groups, but not for Italians.
Driving habits had stronger negative effects for users of carsharing than for non-users. This result indicates that users with a high perception of a need for a car are less willing to use carsharing. Non-users may have a variety of preferences for traveling, and they do not necessarily have a strong link with a car to reach their destinations. Habitual behaviors are mental schemas linked to a goal to act [
71]. If there is no link between traveling to a certain destination and the vehicle “car”, it is unlikely that someone would choose a service based on cars to travel.
The concept of climate morality was hypothesized to have a positive relationship with the intention to use carsharing, given the support from the literature that carsharing may be a more sustainable way of traveling compared to private cars. However, climate morality had a small negative effect among the Swedish groups and no effect at all among the Italian groups.
This result could indicate a low perception of carsharing as a way of reducing environmental impact (since it is car-based). Despite the forecasted reduction in CO2 emissions and land use due to the use of carsharing instead of private cars [
72], the impact assessment of the carsharing is complex, and it depends on the time framings of methodological designs (if it is cross-sectional, ‘before–after’ or longitudinal designs) [
73]. Additionally, other factors can be more relevant to the decision to use carsharing, and concerns toward the environment are not always aligned with pro-environmental behavior performance [
74].
The comparison of the effect of climate morality between nationalities is challenging and interesting. Italy has been reported as a relatively environmentally friendly country in the European Union (EU) [
75], and the intention to reduce car use and travel in a more environmentally friendly way has been identified as quite low among Swedish residents [
76].
However, this small difference between countries should be carefully considered, both because it is not a large difference and because climate morality has been modeled as a direct predictor in this study and previous research has argued that variables measuring environmental concern perform better as an indirect predictor of a specific behavior [
77].
Subjective norms were a positive predictor across groups and nationalities. This result corroborates with previous studies that identified that the perception of approval of peers can be an influence for transport services, such as bike-sharing [
31] and public transport use [
78,
79].
The importance of investigating subjective norms is to highlight the cultural aspects of behaviors and develop strategies to promote the given behavior. Once the effect is estimated, one may advocate for more or less marketing strategies, such as a discount for social peers, club members, and sharing on social media.
The measure of control was assessed by a combination of perceived behavior control, perceived ease of use, and usefulness. This aspect is strictly connected to the services’ online platforms, vehicles, and parking schema. In this study, control was the main predictor of intention to use carsharing, corroborating previous studies. PBC has been shown to be among the main predictors of willingness to use shared parking systems [
80] and to re-use free-floating carsharing [
81].
The factor trust captured the predictability, trustworthiness, and quality of the carsharing service. This factor is strictly associated with people’s perception of the availability of vehicles and to what extent they can rely on the service to keep their regular travels without major incidents. Previous research has shown that this is a major component of decision-making and of the probability of enjoying a carsharing service [
82].
Interestingly, trust was only a positive predictor of behavior intention among the Italian groups. However, the χ
2 difference test did not indicate statistical differences across the groups. Previous research has identified that increased familiarity with the service platform has a positive impact on consumer’s trust in the service [
3]. However, the results from this study do not give support for that. Users and non-users of carsharing did not differ in levels of trust in the service.
It should be noted that in many previous studies with similar aims, the modeling of the variables has not always been under scrutiny of measures of invariance, which challenges the validity of the conclusions driven from the comparisons between groups. In this study, this kind of contribution can be done, and the effect of psychological variables in the intention to use carsharing can be compared between users and non-users of carsharing in Sweden and Italy. However, as with any cross-sectional study, the effects modeled are correlational and therefore should not be interpreted as causal.
In terms of limitations of this study, participants were recruited across European countries; however, it was only in Italy and Sweden that the data collection was successful, limiting the cross-cultural analysis to these two countries. It should be mentioned that there was no control for which kind of carsharing the participants have taken part in (e.g., roundtrip; one-way, station-based; one-way, free-floating; peer-to-peer). These differences in the service could be more or less attractive for different profiles of users and non-users and consequently influence their intention to use carsharing in the near future. Additionally, shifts in modal structure for less carbon-intensive vehicles are usually connected to transport infrastructure, urban planning, and population density [
83]. The present model did not control for these aspects; therefore, the conclusion drawn from this study to compare Sweden and Italy should be carefully considered.
6. Conclusions
Overall, the effects of the psychological predictor variables can be used as a source of information to understand the use of carsharing and provide insightful knowledge about the behavior of users of carsharing and the spread of the use of the service among those that have not had any experience with the service before. The results have shown that there are psychological common factors predicting intention to use carsharing between users and non-users, but also that there are some particularities for each group. Control was the main predictor of intention to use carsharing. Driving habits had stronger negative effects for users of carsharing than for non-users. Subjective norms positively predicted the intention to use carsharing among all groups. Trust was a predictor of intention only for the Italian groups. Climate morality had a small negative effect on the Swedish groups only.
Climate morality, that is, the personal norms to reduce the impact on the environment and the concern with it, is not affecting people’s intention to use carsharing to a substantial extent. This aspect of the model, the considerations taken based on environmental issues, should be further investigated in future studies. For instance, the connection between carsharing use and the use of public transportation could better explain to what extent the modal choice is oriented to convenience or environmental concerns. To what extent do people perceive carsharing services as an alternative to reduce their environmental impact due to their personal travel behaviors?
Future research could tackle this aspect by investigating the qualities of services that mostly satisfy different profiles of users and potential users and their demands. Moreover, in what circumstances, for instance, the travel purpose, are people more willing to take a carsharing? Previous research has shown that psychological predictors of travel mode choice differ depending on the travel purposes [
4,
84].
The results of this study indicated that driving habits are negative predictors, but with a relatively small effect. What other factors may be hindering the use of carsharing? Among those that have not had experience with this service, what aspects are impeding them to have the first trial? Therefore, other barriers to use carsharing services could be further investigated in connection with habits.