Sharing Is Caring: An Economic Analysis of Consumer Engagement in an Electric Vehicle Sharing Service

Abstract

A growing population and increasing consumer demand have created unprecedented pressures on the environment and natural resources. The private vehicles market, one of the largest markets in the world, is associated with considerable environmental costs. Sharing electric vehicles, where consumers can enjoy the benefits of a greener vehicle without owning it, has emerged as an innovation that can reduce some of the environmental costs of ownership. However, uncertainty around the determinants of participation remain. This study employs an econometric model using survey and experimental data that were collected at the initial stages of the roll-out of an electric-vehicle-sharing service in Malta, in order to identify the psychological factors that determine the willingness to use and to pay for such a service, the propensity to walk to a car-sharing station, as well as the likelihood of scrapping a privately owned vehicle. The findings suggest that engagement in the car-sharing market is more likely to take place among those who have a lower psychological attachment to the private car, are already using multiple transport methods and are sharing a car with other household members. A large number of cars per household and a high use are negatively associated with uptake. The results also suggest that consumers who care about the environment are more likely to engage in car sharing.

1. Introduction

Growing populations and increasing consumer demand have created unprecedented pressures on the environment and natural resources [1]. The sharing economy offers a novel economic prospect whereby the benefits of goods and services can be enjoyed while reducing the environmental burdens that typically come with consumption [2,3]. Given that goods are typically used for just a fraction of the time they spend in the consumers’ possession, the sharing economy shifts the focus away from ownership to use [4,5]. This promises to use up excess capacity, to make more efficient use of resources and ostensibly to build social capital in the process [6]. Technological advancements have played their part in unlocking the potential of the sharing economy, as online platforms and smartphone applications create a market space for businesses and individuals [2].

In practice, the phenomenon has found considerable traction in recent years. In the accommodation sector, Airbnb enables people to offer their spare rooms for short let [2]. In the job market, TaskRabbit connects individuals with different skills who are willing to offer their services to people who need to get a job done [7]. In the transport sector, ride-sharing set-ups, such as Uber and Lyft, make it possible for individuals to transport others in their free time, while car-sharing services, such as Zipcar and Car2go, allow individuals to drive shared vehicles themselves, typically for short distances [8,9,10].

These developments have been rapid and not without controversy. There are considerable challenges in the legal and regulatory frameworks [11], mixed views on the benefits/costs to the environment [12], and cynicism in terms of the question of trust in the service providers. The disruptive element of the sharing economy also challenges the status quo for several competing businesses [2,6]. The scholarly literature is fast catching up with the developments in the field. This paper aims to contribute in this regard by assessing the determinants of engagement in a car-sharing scheme in a European Union country (Malta), at the initial stages of its roll-out; it focuses on the psychological determinants that are related to an attachment to the private vehicle. This has the potential to offer insights for other regions and countries experiencing similar conditions.

2. Literature Review

The private vehicles market is one of the largest markets in the world. In revenue terms, the leading ten automotive manufacturers generated over USD 1.6 trillion in 2017 alone [13], and sales of new passenger cars amounted to 79 million vehicles globally in 2018 [14]. Despite efforts to reduce the use of private cars and car emissions in the European Union (EU), [15] the registration of new vehicles has continued to increase, and consumers increasingly favor more polluting sport utility vehicles [16]. The economic problems related to the negative external costs on society are well theorized [17], and the real environmental costs of private car ownership and use are well documented [18]. In 2016, passenger cars accounted for 11% of the total carbon dioxide emissions in the EU and 41% of transport emissions [19]. Traffic congestion, defined as a state in which traffic flows at a very slow speed, presents another undesired cost in urban areas [20]. In Europe, road congestion is estimated to cost EU countries around EUR 100 billion [21]. In a recent study across 1360 cities, drivers were found to spend an average of 9% of their travel time at a stand-still [22]. Other environmental issues associated with car use include road damage, noise and accidents [20], and the associated social costs of pain and grief [23]. The manufacture and disposal of cars at their end-of-life poses additional costs [24].

While the literature is rich with proposals to address such externalities from private vehicles, it is the potential of the sharing economy that is of particular relevance to this study [25]. In the car-sharing market, licensed drivers typically pay a fee (per minute/hour) to unlock and use cars parked in various unattended reserved locations [26]. Such services are credited with having positive knock-on effects on the environment by reducing traffic, parking needs and private car ownership [10,27]. As car-sharing fleets are mostly electric vehicles, the service is often credited with having lower emissions [8,28]. The demand for car-sharing services may also lead users to sell or scrap their own private car, to actively consider doing so in the near future or, at least, to curb the need for additional household cars [29]. In Ulm, 29% of survey respondents stated that whey would strongly consider foregoing the purchase of a private car if car-sharing services were a worthy transportation option [8]. In Basel, 6% of members of a car-sharing service actually reduced their car-ownership [30].

Demand for car sharing is well documented to be influenced by price and convenience. For instance, demand for car sharing responded to price among university students in the Netherlands [31], while participants in Switzerland stated that their engagement in car sharing was based on considerations of practicality [32]. In Lisbon, individuals were more likely to engage in car sharing for certain tasks, such as shopping and errands [10]. Given that reserved parking is typically part of the package, car-sharing services thrive in areas where the availability of parking spaces is very limited [33]. In fact, the availability of a shared car space was found to be crucial factor in the decision to use car-sharing services in the Netherlands [34]. The distance to car-sharing stations was surprisingly not a significant explanatory variable of car-sharing membership in Switzerland [25], but the authors argue that this could be due to the dense network of car-sharing stations. In general, living in urban areas, especially in the city center, has been found to be positively correlated with the use of car sharing [35,36].

Psychological determinants also influence demand. In the domain of pro-environmental behavior, demand is often considered to be driven by those who care about the environment, sometimes theorized to be a warm glow that is derived from satisfying moral or normative preferences [37]. This appears to be the case in car sharing too. The feeling of forming part of a community was found to encourage engagement in car-sharing services among university students in Hamburg, Germany [9]. Similarly, the environmental benefits of car sharing were found to be an important determinant of car sharing among users in Amsterdam [2]. In line with the broader literature on pro-environmental behavior, females have been found to be more likely to cite environmental reasons for engaging in the sharing economy [2]. Trust in the service provider also emerged as a highly significant consideration in determining whether users are satisfied with the service among university students in Germany [9]. Users of shared cars were more likely to trust innovative mobility options, such as electric vehicles [38]. On the contrary, the anonymity of the service, short duration of usage and self-service access could foster norms of negative reciprocity and opportunism [4]. Finally, attachment to the private vehicle and the psychological effects of ownership are other factors that may inhibit people from giving up their private cars [31]. Individuals that incorporate car sharing as one of their transport options show other tendencies to manage their trip more actively, juggling between various modes of transport [10,25,30].

Finally, as with any service, population demographic characteristics can also forecast demand. The use of car-sharing services has been found to respond to higher educational attainment [25,34,35], while households with children are less likely to use car sharing as an alternative mode of transport [34]. Similarly, the possibility of sharing a private car among household members may reduce the need of a car-sharing service [25]. Indeed, it seems to be easier for single users to use car-sharing platforms than it is for families [35]. Males are more likely to use car-sharing services [35], much as they are also more likely to purchase a second private car [34]. Car sharing is more popular among younger cohorts [34,35], as is the sharing economy more generally [2]. Finally, while household income is often found to be positively correlated with car ownership [25,34], this is not the case with car sharing, where participation in the market is more equitable [34].

In summary, while the global increase in private car ownership has had considerable implications on resource use, waste, congestion and emissions, the electric car-sharing market has potential to reverse some of these effects through the more efficient use of resources, lower congestion and parking needs, and fewer emissions. It has already been determined that demand for a car-sharing service responds to price (relative to that of car ownership) and convenience (relative to alternative modes), that it is more popular among higher educated, male, single and younger segments of the population, but that is not necessarily responsive to income. Demand also seems to respond to more psychological preferences, such as environmental concern, trust in the service provider and attachment to the private vehicle; it is these preferences that this study sets out to examine in more detail.

3. Context

Europe dominates the car-sharing market, having an estimated 5.8 million users and 68,000 vehicles in 2016, and accounting for over half of the global market share [27]. The number of users is anticipated to grow to 15 million by 2020 [27]. This study is situated in a European country, Malta, where private passenger vehicles heavily dominate the transport mode. Here, the stock of vehicles stood at 385,326 by the end of 2018, of which 78% were passenger cars [39], a figure which grew by an average of 69 vehicles per day in the last quarter of that year [39]. For a resident population of less than half a million, this presented one of the highest passenger car:adult ratios in Europe, at 0.74 cars per every adult inhabitant [40]. Passenger cars in Malta also stay on the road longer than they do in other EU countries (14 years in comparison with 11) [40]. The cumulative impact of these vehicles includes not only increasing accidents, but also emissions [40]. The latter stood at 582 kilotons of CO₂ equivalent in 2016 and resulted in Malta having to purchase credits to offset the increase [41]. The impact of air pollution on health was linked to around 240 premature deaths in 2015 [42].

To address these problems, Malta has mainly relied on investing in a wider road network, including a EUR 700 million investment in major junctions, road widening and road alignments [43,44]. Despite evidence that this may induce further demand [45], there are approximately 762 km of roads per 100 km² in Malta, making it the densest road system in the EU [46]. With a land area of 316 km² [47], Malta is also the smallest nation within the European Union (EU), and its population density worsened considerably with tourist arrivals, amounting to 2.6 million in 2018 [48]. The resident population is also growing rapidly with a constant flow of foreign nationals seeking employment in Malta [49]. Consequently, the pressure on the transport infrastructures is immense [50]. A longer-term solution for the traffic situation in Malta envisages a shift towards alternative modes of transport. The public transport system also saw a considerable increase following an overhaul of the fleet and pricing incentives [51], but capacity constraints, infrastructure and reputational issues continued to hinder the uptake [52]. Several other proposals are in discussion, including a metro [53] and improved ferry services [54].

Car sharing is the latest venture courted by the Maltese Government as part of a plan to reach the electric vehicle and pollution targets set by the European Commission [55]. The sharing economy has become increasingly prevalent in Malta, with applications in the accommodation sector [56], rental goods sector [57], bicycle sector [58], car-riding sector [59] and car-pooling service at the University of Malta [60]. In November 2018, GoTo Malta, a subsidiary of the Israeli company car2go, started offering a car-sharing service, following a public call [61]. It closed its operations in Malta in 2022. With some 150 electric vehicles stationed around the island and 400 parking bays reserved specifically for the company’s cars [62], users could subscribe to their services by downloading their smartphone application, reserving a car nearby, unlocking it and driving it to their destination [63].

4. Method and Data Collection

Against the background provided by the literature review and in the context of Malta, we conceptualized the willingness to engage in the car-sharing market in four ways: i. willingness to use the vehicles, ii. Willingness to pay for the vehicles, iii. Willingness to walk to the vehicle stations, and iv. willingness to scrap/sell one’s vehicle and substitute it for car sharing. To generate data commensurate with the theoretical considerations outlined earlier, we designed a questionnaire (simplified survey flow in Figure 1).

To understand the extent of psychological attachment to their private vehicle, respondents were told to “Think of the car you normally use; i. Do others in your family drive it (No/Yes)? ii. For how many minutes is the car being driven per day on average (open)? iii. how much did your car cost when you bought it (open)?”. These questions generated the variables on the present extent of car sharing in the family (CARSHARE), the extent of car use (CARUSE) and the original cost of their car (CARCOST). In addition, respondents were asked the following: “In the last 3 months, did you use any of the following means of transport for a. work, b. leisure or c. errands?: i. car, ii. Bike, iii. Motorbike, iv. Public transport, v. walking, vi. Taxi”. Respondents could indicate a minimum of 0 and a maximum of 18 options, yielding a composite variable (MODALITY). Data on household size and the number of cars in the household were also used to generate the variable on cars per adult (CARSPA).

To assess the impact of trust in the service provider, the questionnaire contained an embedded experiment. Respondents were told the following: “Imagine car-sharing services were offered all over Malta this year. They would be parked in central locations with short walking distances and users pay per minute of use”. At this point, respondents were presented with one of four different treatments (randomly and equally assigned): T0: No further information; T1: The services would be operated by a private company; T2: The services would be operated by the Government; and T3: The services would be supported by environmental NGOs. These treatments were coded as T_PVT, T_GOV and T_NGO, respectively. As expected, each treatment absorbed around 25% of the sample. Following exposure to the treatment, respondents were asked four questions: “i. Would you be willing to use this service? (Yes, No, Maybe); ii. Would you actively consider selling or scraping one of your cars? (not to be replaced) (Yes, No, Maybe); iii. How much are you willing to pay for car sharing per minute of use? (in Euros); and iv. How many minutes would you be willing to walk to find your car-sharing destination?”. These generated the outcome variables Willing to Use (WTU), Willingness to Scrap (WTS), Willingness to Pay (WTP) and Willingness to Walk (WTW), respectively.

At the end of the interview, respondents were then asked about the extent to which they trusted Government, private companies and environmental Non-Government Organizations (NGOs), on a scale of 1 to 5, with 5 being the most and 1 being the lowest level of trust. This enabled us to construct three variables (TRUST_GOV, TRUST_PVT and TRUST_NGO). From this, we also created interaction variables between each trust variable and the respective treatment group for the purpose of testing.

The questionnaire also included questions on gender (male, female, other or prefer not to answer), age (open ended) and educational attainment, (the highest level of education successfully completed, whether up to primary level, up to secondary level, or beyond secondary level. Finally, an automatically generated dummy variable (DTIME) captured the time when the interview was undertaken—before or after the roll out of the service.

The questionnaire was distributed through social media platforms, six weeks ahead of the launch of the service (in November 2018) all the way through to three weeks after the launch of the service. The research followed the ethics procedure at the University of Malta and respondents were not identifiable. By collecting data both prior to the scheme’s roll out and at its initial stages of actual operation, we were also able to test the difference in responses as the market transitioned from hypothetical to real. Over this time period, we generated a convenience sample of 405 valid responses. To avoid multiple responses by the same person, a “ballot stuffing” control was adopted in data collection. Respondents were allowed to refuse to answer any questions, and, in such instances, the response was coded as missing. For most of the variables, the missing responses ranged from 30 to 50. The questions on the extent of car sharing, and modal-means (which were at the start of the questionnaire) had a higher response rate, while the open-ended variables on willingness to pay and willingness to walk, and the questions on trust, had a lower response rate. The majority of respondents refused to answer the question on income and this was dropped from the analysis. In regression analysis, the dataset becomes smaller the more variables are included, as the estimation is conducted only on those respondents with complete responses for all the variables.

The resultant dataset was broadly regionally representative [64], and the average household size was similar to the national average, at 3.1 persons per household compared to 2.7 for the population. As is typical in survey datasets [65], our sample was slightly female-skewed, with 58% of respondents being female (in comparison with 50% in the population). Respondents were also slightly younger than the population, averaging 35 years of age, compared to 40 years for the population. In contrast with the national population, our sample is, however, heavily skewed towards more highly educated people, with 90% stating that they have completed education beyond secondary level (in contrast with 20% of the Maltese population).

Descriptive data for the variables used in the analysis are presented in

Table 1. Descriptive Data.

Variable	Description	Min	Max	Mean	SD	N
WTP	Willingness to pay (euro per minute of car sharing)	0	1.5	0.233	0.277	326
WTW	Distance willing to walk to a car-sharing station (minutes)	0	20	7.191	3.929	329
WTU	Willingness to use car sharing (No = 0, Yes/Maybe = 1)	0	1	0.767		373
WTS	Willingness to sell car due to car sharing (No = 0, Yes/Maybe = 1)	0	1	0.300		370
GENDER	Respondent is female (Male/Other/No answer = 0, Female = 1)	0	1	0.577		357
EDUCATE	Highest level of education successfully completed (1 = Up to Primary; 2 = Secondary; 3 = Beyond Secondary)	2	3	2.902	0.297	348
AGE	Age of respondent (open ended)	17	74	34.717	12.38	357
CARSPA	Number of cars per adult (cars divided by adults per household)	0	4	0.919	0.441	374
CARCOST	Purchase price of private car (in euro)	0	100,000	10,256	10,871	403
CARUSE	Number of minutes car is driven per day	0	600	72.839	63.692	369
CARSHARE	Propensity to car share at home	1	5	2.851	1.435	389
TRUST_GOV	Trust in government (1 lowest, 5 highest)	1	5	2.801	1.152	352
TRUST_NGO	Trust in environmental NGOs (1 lowest, 5 highest)	1	5	3.256	1.126	352
TRUST_PVT	Trust in private companies (1 lowest, 5 highest)	1	5	3.279	0.989	351
T_GOV	Treatment Group Government	0	1	0.247		373
T_PVT	Treatment Group Private	0	1	0.249		373
T_NGO	Treatment Group NGO	0	1	0.244		373
DTIME	Data collected after car-sharing service started	0	1	0.202		405

. Examining the headline figures for the dependent variables reveals that all together, on average, 77% of respondents showed some interest in the hypothetical car-sharing service (WTU). Respondents were willing to pay an average of EUR 0.23 per minute of use. They were willing to walk an average of 7 min to the nearest car-sharing station. The potential willingness to sell or scrap a household car due to the introduction of the car-sharing service (WTS) stood at 30% of households. While these figures cannot be considered to be representative of the nation, they are well within the expected parameters given the existing prices of car-sharing services, distances to stations, and findings from the literature reviewed earlier. Our sample respondents have roughly one car for each adult household member. Their main household car cost just over EUR 10,000 and was driven for an average of 1 h and twelve minutes per day. The index for multimodality (MODALITY) sums the frequency of the use of all alternative modes of transport (namely bike, motorbike, public transport, walking or taxi for work, errands and leisure, thereby summing for each instance of use) as ranging from 1 to 15, while the variable CARSHARE reveals the propensity to share their household car with other household members. The TRUST variables indicate that households tend to trust the government somewhat less than they trust the private sector and environmental NGOs. The dummy variable DTIME indicates that a fifth of our respondents were interviewed after the introduction of the service.

With the data in hand, our intent was not only to predict car sharing uptake as a whole, but more importantly to examine the t-statistic of particular variables, namely those capturing psychological motives and barriers. To this end, we first estimated a basic model in which demand for car sharing (captured by (WTU, WTP, WTW and WTS) is a function of attachment to the private car (CARUSE, MODALITY, CARSHARE and CARSPA) and the control demographics (EDUCATE, GENDER, AGE), including income (proxied by CARCOST). In Model 1, i indexes respondents, Y_i represents the dependent variables and e represents the error term.

Model 1:

Y_i = b₀ + b₁CARUSE_i + b₂MODALITY_i + b₃CARSHARE_i + b₄CARSPA_i + b₅GENDER_i + b₆AGE + b₇CARCOST_i + b₈EDUCATE_i + e_i,

(1)

We then proceed to examine whether information about the service provider added further explanatory power to the estimation. Given the experimental nature of our data collection, we also employed a simple analysis of means to test the impact of different service providers. We then examine whether trust in the service provider (TRUST*) added further explanatory power to the willingness to engage, and, indeed, whether TRUST interacted with information on the service provider (TX*) to explain the stated outcomes. Finally, given that a portion of our responses were collected once the service started, we tested the extent to which this distinction helped to forecast the willingness to engage, noting that other studies document differences in behaviour between stated preferences in a hypothetical as opposed to an actual setting [66]. The methodological process is presented in Figure 2 below.

5. Analysis and Results

We first estimated a basic set of models (in line with Model 1) for willingness to use, to pay, to sell their car and to walk to a car-sharing station.

Table 2. Predicting Willingness to Engage in Car Sharing.

Variable	WT Pay (OLS)	WT Walk (OLS)	WT Use (LOGIT)	WT Sell (LOGIT)
CARUSE	−0.001 ***	−0.005	−0.000	0.001
	(0.000)	(0.003)	(0.000)	(0.000)
CARSPA	−0.093 ***	−0.508	−0.057	0.014
	(0.029)	(0.598)	(0.048)	(0.060)
CARCOST	0.004 **	−0.016	0.001	0.002
	(0.002)	(0.019)	(0.002)	(0.002)
MODALITY	0.002	0.204 *	0.028 **	0.023 **
	(0.009)	(0.114)	(0.011)	(0.011)
CARSHARE	0.009	0.332 *	0.063 ***	0.054 ***
	(0.012)	(0.177)	(0.016)	(0.018)
EDUCATE	−0.111	1.538 **	−0.130 **	0.040
	(0.090)	(0.714)	(0.059)	(0.087)
GENDER	0.065 **	−0.925 *	−0.036	−0.153 ***
	(0.029)	(0.474)	(0.046)	(0.054)
AGE	−0.001	0.018	−0.003	0.001
	(0.001)	(0.019)	(0.002)	(0.001)
Constant	0.369 **	4.514 ***
	(0.147)	(1.405)
Observations	283	286	316	316
R-squared	0.101	0.088
McFadden’s R-squared			0.109	0.065

Robust standard error in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

shows the coefficients obtained with heteroscedasticity-robust standard errors. In this table, willingness to pay and walk are estimated using Ordinary Least Squares, while the discrete dependent variables willingness to use and to sell/scrap are estimated using logit [67]. In OLS estimation, the coefficients measure the change in the mean of the independent variable for a one-unit change in the independent variable, ceteris paribus. In a logit model, however, the coefficients measure the probability that the dependent variable is equal to a particular category [67].

Our findings indicate that willingness to pay responds negatively to higher car attachment, captured by the minutes of car use and the number of cars per adult. Females forecast a higher willingness to pay, on average. Willingness to pay responds positively to the cost of the vehicle, which we consider to proxy income. We re-estimated the models, this time using willingness to use as the outcome variable. We found that this is predicted by higher multimodal preferences and the propensity to share a car between family members, while education level exerted a positive influence. In forecasting willingness to walk, we found again that multimodal preferences and a propensity to share the family car forecast a higher willingness to walk. There is also a gender and education effect: males are more willing to walk than females and more educated individuals are more willing to walk than others. When we ran the regression for willingness to sell/scrap a household car, we found, again, that males and respondents with a propensity to share, or multimodal preferences, are more likely to state an intent to scrap their vehicle.

The R-squared values obtained are rather low, suggesting that the X variables in Model 1 together only explain a small part of the dependent variables’ fluctuations. Indeed, the results in Table 2 do not control for the different information that respondents received as to who the service provider would be. A comparison of the means in each of the treatment groups suggested that there may be some diversity in outcome caused by information about the service provider, chiefly a higher willingness to pay (p = 0.054) if the service provider is in the private sector, (Mean = 0.285, S.D. = 0.314) relative to the control group (Mean = 0.203, S.D. = 0.215). We then proceed to test whether these results survive the regression analysis.

We noted that in re-estimating our model but controlling for the treatment effect, we obtained an improvement in the fit, as measured by the R-squared (

Table 3. Predicting Willingness to Engage in Car Sharing—Controlling for Service Provider.

Variable	WT Pay (OLS)	WT Walk (OLS)	WT Use (LOGIT)	WT Sell (LOGIT)
T_GOV	0.009	−0.347	0.058	−0.015
	(0.036)	(0.634)	(0.062)	(0.074)
T_NGO	0.065	−0.388	0.049	−0.084
	(0.046)	(0.598)	(0.063)	(0.070)
T_PVT	0.073 *	−0.249	0.068	−0.036
	(0.042)	(0.668)	(0.064)	(0.072)
Constant	0.330 **	4.667 ***
	(0.142)	(1.439)
Observations	283	286	316	316
R-squared	0.115	0.089
McFadden’s R-squared			0.114	0.069

Robust standard errors in parentheses. Regressions were estimated with the same variables as Table 2, that is, CARUSE, CARSPAC, CARCOST, MODALITY, CARSHARE, GENDER, EDUCATE, AGE. The table presents the abridged results; *** p < 0.01, ** p < 0.05, * p < 0.1.

). We also confirmed that respondents who were told the service would be provided by the private sector were willing to pay more than those who were not told who the service provider would be. This results in the respondent being willing to pay 7 cents more, a considerable change given that the average WTP obtained in the whole sample was 23 cents. We further noted a statistically significant and negative coefficient in the Willingness to Scrap model, among those who were told that the service would be provided by Environmental NGOs. The remaining explanatory variables remained stable following the introduction of the treatment variables.

We then proceeded to examine the effect of trust (

Table 4. Predicting Willingness to Engage in Car Sharing—Testing for Trust.

Variable	WT Pay (OLS)	WT Walk (OLS)	WT Use (LOGIT)	WT Sell (LOGIT)
CARUSE	−0.001 ***	−0.004	−0.000	0.001
	(0.000)	(0.003)	(0.000)	(0.000)
CARSPA	−0.087 ***	−0.443	−0.064	0.022
	(0.030)	(0.585)	(0.048)	(0.061)
CARCOST	0.003 **	−0.030	0.000	0.003
	(0.001)	(0.021)	(0.002)	(0.002)
MODALITY	0.004	0.194 *	0.026 **	0.019
	(0.009)	(0.113)	(0.012)	(0.012)
CARSHARE	0.007	0.390 **	0.052 ***	0.048 **
	(0.013)	(0.179)	(0.016)	(0.019)
GENDER	0.064 **	−0.967 **	−0.026	−0.157 ***
	(0.031)	(0.487)	(0.046)	(0.054)
EDUCATE	−0.112	1.453 **	−0.110 *	0.055
	(0.090)	(0.724)	(0.065)	(0.086)
AGE	−0.001	0.011	−0.002	0.0004
	(0.001)	(0.019)	(0.002)	(0.002)
TRUST_GOV	0.007	−0.276	0.013	0.009
	(0.019)	(0.224)	(0.024)	(0.028)
TRUST_NGO	0.008	0.259	0.045 *	0.052 *
	(0.012)	(0.256)	(0.023)	(0.027)
TRUST_PVT	−0.019	−0.265	−0.005	−0.057 *
	(0.020)	(0.295)	(0.029)	(0.034)
T_GOV	0.027	−0.207	0.0739	−0.064
	(0.101)	(1.401)	(0.135)	(0.180)
T_NGO	−0.024	0.213	0.184	−0.203
	(0.101)	(1.771)	(0.118)	(0.188)
T_PVT	−0.073	−0.431	−0.156	−0.143
	(0.143)	(1.888)	(0.202)	(0.192)
TXPG	−0.008	−0.094	−0.006	0.019
	(0.033)	(0.402)	(0.045)	(0.051)
TXPE	0.025	−0.219	−0.055	0.032
	(0.030)	(0.486)	(0.047)	(0.058)
TXPP	0.040	−0.011	0.056	0.032
	(0.042)	(0.498)	(0.050)	(0.055)
Constant	0.359 **	5.850 ***
	(0.169)	(1.928)
Observations	280	282	310	310
R-squared	0.119	0.125
McFadden’s R-squared			0.13	0.089

Robust standard error in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

). By including the original and interacted variables, we were able to explore the extent to which it matters that the car-sharing service is provided by government/the private sector/environmental NGOs (T_* coefficients); the extent to which trust in these broad entities predicts uptake (TRUST* coefficients), and the extent to which the interaction of these two phenomena predicts uptake (TX* coefficients). We note that while the other variables remain steady, neither the Treatment variables (T_*), nor the interaction variables (TX*) are significant. However, trust in environmental NGOs emerges as a positive and significant variable in explaining a willingness to sell and scrap a car. We consider it plausible that this variable may be acting as a proxy for pro-environmental preferences. In contrast, trust in the private sector is associated with a lower willingness to sell a vehicle.

Finally, we considered it useful to control for heterogeneity in responses collected prior and post roll-out. A cursory glance at the difference in the means between the 252 observations collected prior to the launch and the 74 observations collected after the launch (Figure 3) reveals a statistically significant difference (p = 0.000). We found that WTP was considerably higher prior to the launch (Mean = 0.259, S.D. = 0.306), than after (Mean = 0.146, S.D. = 0.107). We also noted that there was a statistically significant decline in the willingness to walk to the nearest station between the pre-implementation means (Mean = 7.48, S.D. = 4.02) and the post-implementation means (Mean = 6.17, S.D. = 3.41). No significant difference was found in the extent of willingness to use and willingness to sell/scrap between the before and after groups. One reason for this could be the fact that respondents were asked a simple binary question (Yes/No) for the latter, while they were asked to indicate a figure for the former, a process which is more likely to generate forecasting errors.

The inclusion of the dummy variable DTIME indeed improved the R-squared statistics in the regressions, as can be shown in abridged

Table 5. Predicting Willingness to Engage in Car Sharing—Pre and Post launch.

Variable	WT Pay (OLS)	WT Walk (OLS)	WT Use (LOGIT)	WT Sell (LOGIT)
DTIME	−0.068 ***	−1.703 ***	0.029	−0.036
	(0.024)	(0.531)	(0.058)	(0.063)
Constant	0.364 **	5.979 ***
	(0.168)	(1.869)
Observations	280	282	310	310
R-squared	0.129	0.155
McFadden’s R-squared			0.131	0.090

Robust standard errors in parentheses. Regressions were estimated with the same variables as Table 2, that is, CARUSE, CARSPAC, CARCOST, MODALITY, CARSHARE, GENDER, EDUCATE, AGE. The table presents the abridged results; *** p < 0.01, ** p < 0.05.

. The coefficient of the variable is negative, possibly indicating that once the real car-sharing service was experienced, this was associated with a decline in willingness to pay and willingness to walk. There were no differences in willingness to use or to scrap/sell the vehicle. While the time dummy should be interpreted with caution in view of the relatively small number of responses post-introduction of the scheme, the findings may suggest that willingness to pay was more sensitive to the characteristics of the actual market than willingness to use. The decline in the willingness to walk may well be attributed to the controversy and publicity around the guaranteed parking spaces for the service [68]. Another interpretation of the difference, however, relates to the possibility that simple binary questions (as were those on willingness to use and willingness to scrap) were easier for respondents to forecast with precision than open-ended ones. Indeed, willingness to pay questions have long been notorious for being difficult to answer [69].

6. Discussion

The analysis above rests on a number of assumptions and must be interpreted within the boundaries of its limitations. The dataset employed was based on self-reported intent in a hypothetical (and subsequently novel) market. Several respondents were unfamiliar with the concept of car sharing, a shortcoming that was mitigated (and exploited) by the fact that the survey ran before and after the launch, and by the fact that the good in question was explained as clearly and succinctly as possible. [70]. Nonetheless, a shortcoming of our engagement data is that they were based on intent. While this does predict behaviour, the relationship is not perfect [8]. Intent and behaviour may differ due to insincerity and poor forecasting, for instance.

Another limitation is the sample; the study used a convenience sample drawn from social media, which consisted exclusively of Maltese nationals whose level of interest in the car-sharing market may have been greater than the norm. Nonetheless, the data offered sufficient variation in order to gauge the direction of the different explanatory variables on the dependent variable using regression analysis, and allowed us to test how engagement responds to the varied characteristics of the participants under ceteris paribus conditions. Furthermore, the embedded experiment with randomly assigned treatments allowed us to infer the causal effect of such treatments on intent. A study with a representative sample would have allowed us to infer national averages. Moreover, given that tourists have the potential to be a major user group of this service, this would merit a separate study on the determinants.

A third limitation relates to the cross-sectional nature of the data. While it was not possible to undertake a longitudinal study in this context (given the interruptions caused by the COVID-19 outbreak and the closure of the service in Malta in 2022), doing so in other contexts could yield further useful insights. Furthermore, the R-squared levels in our regressions were quite low (albeit improving as the models became more comprehensive). A low R-squared could be the result of multicollinearity, where the linear dependence of one predictor variable with another results in inflated variances. A variance inflation factor (vif) test that was conducted on our OLS regressions (multicollinearity is a problem of the X-variables, not the link function) found none of the variables to have a factor higher than 1.7, except for the interaction variables (which is inevitable). Thus, while multicollinearity could theoretically have increased the SD and suppressed significance, this does not appear to be the case here. This, in turn, suggests that there are other potential explanatory variables that could help explain more of the variation in the dependent variables, which could be explored in future work.

Despite these limitations, our study provides insights that support those from the emergent literature in the field. We assessed the willingness to engage in the car-sharing market in terms of willingness to use and pay for the service, willingness to walk to shared-car stations or a willingness to sell/scrap a vehicle and use shared cars instead. We noted that willingness to use the service, to walk and to scrap a private vehicle were positively and significantly associated with the propensity to share the household car, and that pre-existing multimodal behaviour was positively associated with a propensity to walk and to use the service. These findings echo others in the literature, where multimodality was associated with car sharing [10,25,30,38]. They suggest higher engagement among those with a lower level of psychological ownership of the car [31]. Similarly, the higher the number of private cars in the household (per adult) and the longer the private car is used in a day, the less likely people were willing to pay for car sharing. Other works similarly found that the ownership of multiple vehicles negatively impacts uptake [25,34].

We also found that respondents who trust environmental NGOs are more likely to forecast scrapping their private vehicle, a finding that we interpreted as capturing the role of environmental preferences in stimulating demand. This, in turn, echoes Böcker and Meleen [2], who found environmental sentiment to drive demand in Amsterdam. Like Möhlmann [9] in the case of car sharing in Germany, we found some evidence to support the notion that trust in the service provider can matter. Finally, we noted that as we transitioned from the hypothetical to the real market, willingness to use and possibly scrap a vehicle remained stable, but both willingness to pay and willingness to walk declined considerably.

Turning to our control variables, in our sample, females tended to be less willing to walk, and less likely to sell their private vehicle (although they were willing to pay more on average), while those with a higher education tended to be more willing to walk but less willing to use the service. The fact that our sample was a highly educated sample can explain the difference between our results and those prevailing in the extant literature [25,34,35].

As in other countries, the introduction of the scheme and the loss of public parking bays caused controversy in Malta, particularly given the pressures on land [68]. In terms of the possible environmental impact of a car-sharing scheme, the percentage of respondents who considered selling/scrapping their car in this study was similar to that in other studies [8]; however, this figure is likely to be an overstatement of actual behaviour [30]. Again, it is worth noting that the respondents more likely to scrap a vehicle are those with stronger environmental preferences and pre-existing sharing behaviours.

7. Conclusions

It has been argued that the future of the sharing economy will likely depend on improvements in policy and strategic design [71]. The findings in our study offer some insights both for policy, as well as for the design of the service itself. High car ownership and use are negatively associated with car-sharing engagement, and this would suggest that efforts to reduce car dependence through other policy interventions (car-scrapping schemes, higher road licenses) would complement interventions in car sharing. Similarly, the links between car sharing and multimodality suggest that efforts to enhance the latter could support the former. A preference to share the household car is also linked with a propensity to engage in the car-sharing market, suggesting that initiatives to encourage car-pooling (such as incentives or preferential infrastructure) would help. Finally, the participants more likely to engage in the car-sharing market in Malta were more likely to be those who trust environmental NGOs. This positive relationship between environment preferences and the propensity to share an electric vehicle suggests that roll-out schemes could target this (growing) demographic for initial uptake. Sharing is about caring more for the environment, but less about one’s private vehicle.