E-bike to the future: Scalability, emission-saving, and eco-efficiency assessment of shared electric mobility hubs

In car-dominated urban areas, shared electric micro-mobility offers a sustainable alternative to decarbonise and reshape mobility paradigms. This paper presents a comprehensive framework for evaluating Dublin ’ s e-bike sharing system, comprising 12 stations (eHUBs). Using six months of real-world riding data, it employs data envelopment analysis to assess the eco-efficiency of each eHUB in utilising nearby infrastructure, population in the catchment area, and location to achieve desired economic, social, and environmental outputs. Results indicate an upward trajectory for the system ’ s eco-efficiency. The returns-to-scale analysis provides insights into the system ’ s scalability, suggesting that expanding the e-bike sharing system, along with infrastructural enhancements, would significantly increase ridership. Analysing decarbonisation, usage patterns, and spatial factors of eHUBs reveals the substantial potential of shared e-bikes if optimally used. The research highlights the need to reevaluate car-centric policies in favour of more inclusive and environmentally sustainable alternatives and proposes actionable policy recommendations to achieve this transformation.


E-bikes in the context of shared mobility
In the context of escalating climate change and in alignment with Sustainable Development Goals (SDGs) 10, 11 and 13, transforming the road transportation is crucial to render urban mobility more socially accessible, economically affordable, environmentally sustainable, and decoupled from greenhouse gas (GHG) emissions.Consequently, transportation planners and policymakers seek to pinpoint the most effective and sustainable strategies for investment, incorporate climate action into transport policies, and opt for mobility modes that are active, clean, and inclusive.Shared micro-mobility has emerged as one of the solutions to these challenges, with the capability to reduce private car dependency (Zhu et al., 2023), alleviate urban congestion (McKenzie, 2020), curtail air pollution (Domhnaill et al., 2023), and diminish GHG emissions (Martin and Shaheen, 2016).Central to this solution is the integration of shared electric vehicles into a mobility hub (eHUB).An eHUB is an on-street location that offers a single or a blend of electric mobility modes, frequently featuring electrically-assisted bikes (e-bikes), available for immediate hire on-demand within a conveniently accessible area.These hubs are well-suited for installing charging infrastructure, enabling stationary electric vehicles to recharge (Hosseini et al., 2023;Bösehans et al., 2023).When multiple eHUBs are interconnected and operate cohesively within a network, providing access to one or more mobility options, this integrated system is referred to as an e-bike sharing system or more broadly, an eHUB network.
Bicycles, e-bikes, and e-scooters, alongside e-cargo bikes, are commonly classified as micro-mobility vehicles.When comparing bicycles, e-bikes, and e-scooters from active and inclusiveness perspectives, traditional bicycles, while promoting physical activity, are less accessible to certain demographics due to the need for greater physical exertion, which also limits the distances bicycles can cover compared to e-bikes and e-scooters.E-scooters, however, cannot be categorised as an active transport mode, as they require minimal physical effort and, therefore, offer no health benefits.Additionally, the substantial incidence of accidents and injuries involving escooters (Félix et al., 2023) underscores the considerable safety concerns and perceived risks associated with their use in urban settings.E-bikes stand out as the only option that combines the benefits of active transport with inclusivity, as their electric-motor, pedalassist feature helps riders cover greater distances.The motor helps users overcome obstacles such as steep inclines and the need for high physical effort (Rérat, 2021;Schneider, 2023), making e-bikes suitable for a wide variety of users (Ji et al., 2014).This feature also allows e-bikes to traverse distances that would typically necessitate the use of private cars or multi-modal travel, such as both a bicycle and local public transport, establishing them as not only an active and inclusive mode but also a standalone travel option (Hosseini et al., 2023).
Understanding the trends and patterns in transportation choices is critical prior to unequivocally promoting the adoption of e-bikes (Sun et al., 2020) or any other mobility mode.While there is an abundance of qualitative and survey research on e-bikes both in the case of adoption (Jones et al., 2016;Cairns et al., 2017;Schneider, 2023;Andersson et al., 2021;Julio and Monzon, 2022;Bigazzi and Wong, 2020;Bruzzone et al., 2021) and decarbonising profile (McQueen et al., 2020;Hiselius and Svensson, 2017), quantitative studies that focus on the usage of e-bike sharing systems and Shared micro-mobility remain scarce (Krauss et al., 2024).This may stem from the fact that shared electric micro-mobility is a new phenomenon, and relevant data are either unavailable or confidential, often due to intense market competition.Although qualitative studies on e-bikes and shared e-bike systems provide valuable insights, they fall short of presenting a comprehensive view of the actual role and impact of e-bikes in urban mobility.Quantitative research utilising real datasets is a necessary component in filling this gap.Research using real-world data provides the benefit of measuring actual behaviour rather than stated preference surveys, and such studies for e-bikes are beginning to emerge.McQueen et al. (2020) investigated the role of e-bikes in reducing GHG emissions in Portland.They utilised a mode replacement model and local data using data from a survey of e-bike owners to calculate e-bike emission factors and estimate the impact on emissions.Their findings indicated that increasing e-bike usage could significantly lower CO 2 emissions, supporting e-bikes as an effective strategy for carbon reduction in urban transport.Few studies have examined the actual CO 2 emissions reductions from shared e-bikes.These savings are calculated using emission factors for both e-bikes and internal combustion engine vehicles (ICEVs), along with the distance travelled by shared e-bikes, which is tracked via GPS data (Hosseini and Caulfield, 2023;Liao and Correia, 2022;Li et al., 2023).Bigazzi and Wong (2020) carried out a meta-analysis incorporating observations of e-bike mode substitution documented in 24 studies.Their analysis synthesises those findings and reveals a median mode substitution rate of 24 % for private cars.Hosseini et al. (2023) developed a two-stage data envelopment analysis (DEA) framework to evaluate the performance of an e-bike sharing system in Inverness based on Global Positioning System (GPS) data.The objective of the study was to devise an evaluation framework for measuring the relative efficiency of these micro-mobility hubs and to discern the factors influencing their performance.The research determined that the population within the catchment area and weather temperature were pivotal factors affecting riding demand.Several works focus on estimating emissions savings resulting from the increased use of e-bikes as personal and shared vehicles.In their study, Philips et al. (2022) aimed to assess the maximum potential of e-bikes in reducing CO 2 emissions across all neighbourhoods in England using spatial microsimulation based on anonymous survey data.The research compared the capabilities of e-bikes with existing walking and cycling capacities.It was found that e-bikes possess the greatest CO 2 saving potential per individual and per small area, especially in rural regions and the rural-urban fringe.

Research objectives and contributions
The transport sector's contribution to emissions in Ireland has risen sharply from nearly 9 % of the total in 1990 to almost 20 % in recent times (EPA, 2022).Given this sector's growing carbon footprint, it is imperative to recalibrate mobility strategies to ameliorate the climate emergency, bolster urban resilience, and promote sustainable communal habitats.A 2022 report by the Organization for Economic Cooperation and Development (OECD) elucidates that, based on current trajectories, Ireland is poised to miss its stipulated carbon reduction benchmarks (OECD, 2022).The prevailing policies, which predominantly emphasise enhancements to private vehicular modalities, lead to increased forced car ownership (Caulfield et al., 2022;Curl et al., 2018) and have concurrently resulted in the underdevelopment of cycling infrastructure (Egan, 2022).These policies are not poised to effectuate transformative behavioural shifts, expedite substantive emission curtailments, or engender pronounced enhancements in societal well-being.Consequently, the salient challenge remains transitioning from an insular focus on private cars towards more avant-garde, low-carbon paradigms.This transition not only targets carbon mitigation but is also instrumental in fostering sustainable community vitality.This issue is not unique to Ireland; it mirrors a broader pattern observed at varying degrees across advanced and developing economies worldwide.Given the above picture, innovative mobility options such as shared e-bikes and shared electric micro-mobility hubs are increasingly being recognised as possible alternatives to private cars by policymakers and transportation stakeholders worldwide.
This study fills a critical gap in e-bike and shared micro-mobility research by employing real-world data to assess the expansion potential of an e-bike sharing system, to evaluate the relative eco-efficiency performance and location suitability of the stations, and to examine the decarbonising and environmental impact of e-bike sharing systems.The research concentrates on the six months of operation of the first e-bike sharing system in Dublin.Dublin, Ireland's capital, boasts a population of nearly 1,500,000 (Central Statistics Office, 2023).Situated along Dublin's M50, Ireland's most congested motorway, several pilot eHUBs have been positioned to create an e-bike sharing network aiming to serve local residents and commuters from surrounding towns.The primary objectives of the pilot project were to evaluate the performance of these eHUBs in incentivising motorists to switch from private cars to e-bikes for the remainder of their journey into the city centre and to quantify the prospective carbon offset resulting from such a modal shift.
It should be noted that implementing innovative mobility solutions faces challenges, particularly due to policy uncertainty (Cavallaro and Nocera, 2022;Lv et al., 2019), which can impact investment decisions, public acceptance, and the success of programs such as shared micro-mobility hubs.The examination of the e-bike sharing system presented in this study provides evidence that can assist in mitigating the current political contention surrounding cycling infrastructure, which oscillates between criticism for causing traffic congestion and not prioritising cycling sufficiently.Such insights might shift stakeholder perceptions, making what was once a divisive issue appear more equitable and forward-thinking, and highlighting the need to reassess traditional private car-focused policies (Hosseini and Stefaniec, 2023).The outcomes of this research could aid policymakers aiming to create supportive measures to encourage sustainable transport modes while reducing policy fluctuation risks.
This study develops a novel and comprehensive framework for the appraisal of e-bike sharing systems, utilising real-world riding data.The collected GPS data between March and August 2023 enabled a multifaceted analysis that includes data wrangling, inferential analysis of influencing external factors, assessment of emission savings, eco-efficiency evaluation of the network, returns-to-scale analysis, and spatial analysis of influencing factors.The analysis provided in this study is insightful, given Ireland's ranking as the second most car-dependent country in the European Union (Irish Independent, 2022).Its outcomes could enable policymakers to discern essential factors influencing the adoption of shared micro-mobility in heavily car-reliant metropolitan areas.Additionally, it enriches the body of research focusing on justice (Schwanen et al., 2015), aligned with SDGs 10 and 11, and decarbonisation, following SDG 13, regarding the impacts of shared mobility solutions, particularly shared e-bikes.

Data and method
Utilising DEA methodology, this study employed an eco-efficiency model to assess the suitability of e-HUB locations and to rank them according to their performance.Moreover, a returns-to-scale analysis was conducted to evaluate the potential for scaling the service in current locations.The analyses were supported by data derived from GPS tracking and emissions estimates, which were processed through data wrangling techniques and served as outputs to the model.The selection of inputs for the model was guided by an inferential analysis of spatial data.We also examined the characteristics of the areas surrounding the e-HUBs to understand the factors contributing to their varying levels of performance in the spatial analysis section by offering several geographical maps.That is to empower the service planners to discern where to expand or contract the system, and where best to allocate investments, thus optimising the entire system.
The subsequent sections introduce a multi-period model designed to evaluate the eco-efficiency of the e-bike sharing system in Dublin.By calculating the eco-efficiency scores of each station over six months, the model offers a dynamic assessment of the system's performance.Further, the returns-to-scale analysis provides decision-makers with insights into the scalability prospects of the system, as well as the profitability profiles of both the entire network and its individual stations.In this context, scalability is understood as the ability to efficiently and effectively manage increased demand, ensuring that the quality of service remains high even as the scale of operations expands.

Data wrangling
This study aims to quantitatively scrutinise the pilot eHUB network in Dublin, which exclusively offers e-bikes.The network was established by Electricity Supply Board (ESB) in September 2022 and is operated by the MOBY and BLEEPER companies.Of the 14 eHUBs in Dublin, 12 were functional during the study period, each supplying eight e-bikes and corresponding charging docks (Fig. 1).The dataset regarding the 12 eHUBs was collated from two operators over a six-month period from March to August 2023: MOBY, managing three stations, and BLEEPER, overseeing nine.The system was exclusively designed to support round trips.The dataset encompasses the start and finish dates and times for each trip, the trip's duration, and the eHUB from which the e-bike was rented.For journeys originating from MOBY's eHUBs, additional data captured included the distance of each trip and the users' names, whereas BLEEPER provided only a unique five-digit customer ID for each user, with no distance information recorded.Initially, trips shorter than five minutes were excluded, accounting for around 7 % of the total trips.Of the remaining trips, 57 % were from MOBY-operated eHUBs, while the rest were from BLEEPER-operated eHUBs.These e-bikes are equipped with speed controllers that discontinue power to the motor once a speed of 25 km/h is reached.The rental process for e-bikes involved using a mobile application offering two payment schemes.The 'pay as you go' option costs €0.15 per minute with a maximum charge of €10 and requires a €10 deposit.Alternatively, a monthly subscription was available for €15, with each ride billed at €3 and no deposit required.
In pursuit of an eco-efficiency assessment for Dublin's e-bike sharing network, the aforementioned raw data were transformed into structured and refined indicators.Utilising R software, these indicators include the total monthly duration of trips per station, the count of monthly trips per station lasting one hour or more, and the number of unique and frequent users.Frequent users were used as a variable in the study of Caulfield et al. (2017) for examining the usage patterns of a bike sharing system in Cork, Ireland.In the current study, frequent users are identified as individuals who undertake two or more trips per month.Additionally, employing a regression imputation method facilitated the generation of missing distance data, enabling the determination of monthly emission savings per station (Please see Section 3.3).These indicators were then applied as outputs within the proposed eco-efficiency model (Fig. 4).
Table 1 displays the proportion of trips ranging from five minutes to one hour, as well as those exceeding one hour, across all six examined months.The data reveals a predominance of longer-duration trips, whereas shorter journeys occur less frequently.Table 1 also outlines the proportion of e-bike usage by time throughout the week during the six-month study period of the eHUB network in Dublin.The bulk of trips occurred on weekdays from 7:00 to 18:00, hours often associated with nonrecreational or utilitarian travel.Fig. 2 charts the hourly distribution of e-bike rentals and returns at eHUBs in Dublin from March to August 2023, with departures showing a steady pattern during the morning (7:00-9:00) and evening (17:00-18:30) peak periods, maintaining a consistent level in the intervening hours.Although departures exhibit a steady pattern between 06:00 and 19:00 on weekdays, the average durations of trips that commenced between 06:00 and 09:00 were more than 7 h (Table 2).This duration was significantly longer than those starting between 09:00 and 19:00.Additionally, 81 % of trips that last more than seven hours begin between 06:00 and 09:00.Also, the post-work peak period (17-18:30) emerged as the most popular time for arrivals, followed by the later evening hours.This pattern signifies a propensity for users to retain e-bikes for extended durations as they meet their transportation requirements.The steady departure rate from 6:00 to 19:00 underscores the constant demand throughout the day, indicating that e-bikes are potentially fulfilling daily mobility needs for many users.Also, the prolonged retention of e-bikes for trips starting in the morning hours suggests a reliance on this travel mode for daily commutes, which could reduce private car usage and consequently decrease urban congestion.Concerning the gender distribution of e-bike riders throughout the study period, the data provided by MOBY's eHUBs indicate that the users consisted of approximately one-third females and two-thirds males.

Inferential analysis
In this study, the response variable was the number of trips, which is considered count data in modelling processes.Therefore, both Poisson and negative binomial distributions were considered for data modelling (Washington et al., 2020).The Poisson model posits that the data distribution's mean and variance are identical.Accordingly, the number of trip data in this study takes a form as indicated in Equations ( 1) and (2).
(1)  where, y j is the number of trips (non-negative integer), μ j denotes the Poisson parameter representing the expected value of the number of trips, X jk is the k th variable for station j, β is the coefficient, and K is the number of variables.The negative binomial distribution, however, was found to be a better fit based on both density and cumulative distribution function (CDF) analyses (Fig. 3).This choice is often considered when the variance of the count exceeds its mean (Washington et al., 2020).Here, the variance of the number of trips data (764.5)was significantly higher than its mean (25.14), indicating overdispersion.Hence, negative binomial regression is utilised to investigate relationships between confounding and predictor variables in count data, especially when overdispersion is present (Charly and Mathew, 2019;Truong and Currie, 2019;Washington et al., 2020).Additionally, since the data was repeatedly collected over six months for all stations, the unobserved heterogeneity in the number of trip data is accommodated by including random components, particularly addressing the temporal heterogeneity among the eHUBs.Accordingly, the random effect negative binomial model is formulated as Equation (3).
Here, η j is the randomly distributed term, e.g., a normally distributed term with mean zero and variance σ 2 and EXP ( ε j ) is a Gammadistributed error term with mean 1 and variance α.Equation (3) extends Equation ( 2) to consider the variance more than the mean, as presented in Equation ( 4). VAR To select pertinent input variables for the eco-efficiency model, negative binomial regression analysis was employed using IBM SPSS Statistics 28 software.This analysis investigated the association between the monthly frequency of trips from each station (dependent variable) and a range of potential predictors (independent variables) selected based on previous literature concerning shared mobility and mobility hubs.These determinants encompassed the cumulative length of bike lanes within a four km radius of the stations (Hong et al., 2020), the station's proximity to the city centre (Duran-Rodas et al., 2019;Nematchoua et al., 2023;Julio and  K. Hosseini et al. Monzon, 2022), the population within a 5-minute walkable catchment area of each station (Charly et al., 2023;Hosseini et al., 2023), and the number of public transport routes and public transport stops within a 400-meter radius (Oeschger et al., 2020).Data relating to the independent variables were estimated from open-source maps and geographical information systems, including Cyclosm (2023), Google Maps, Open Route Service (2023), and Transport for Ireland (2023).It should be noted that in the case of the length of bike lanes, all separate cycleways, paths designated for bicycles, and shared lanes with buses and taxis were considered.Table 3 summarises the descriptive statistics of predictor variables used in this study.
Table 4 displays two negative binomial models developed considering the three iterative processes of variable selection: the forward selection, the backward elimination, and the stepwise process (see Elliott and Woodward, 2007;Sil et al., 2020).The forward selection and the backward elimination processes provided the same model results, as mentioned in Model 1, while Model 2 showcased different results observed during the stepwise regression process.Both models show similar goodness-of-fit results (see Table 4).The pvalues of Model 1 indicate that distance to city centre and population were statistically significant at the 95 % confidence level.Additionally, Model 2 presents a significant effect of the length of bike lanes on the number of trips.Specifically, the odds ratio of Model 1 shows that an average increase of 100 people in the region increases e-bike usage by 8.1 %.The number of trips increased by 20.1 % with an average increase of 1 km distance from the city centre.Similar to Model 1, Model 2 presents that the number of trips was raised by 24.8 % for an average 1 km increase in distance to city centre.Additionally, the number of trips increased by 17.9 % for an average 10 km increase in the length of bike lanes.Overall, these models suggested an increment in the number of trips with the increase in the distance of stations from city centre, population, and length of bike lanes.With the limited availability of the number of trips data (only for six months of data), it is challenging to determine the best model among Models 1 and 2. Nevertheless, both of these models can be suitably used based on the data availability.Considering the odds ratio mentioned in Models 1 and 2, distance to city centre was more sensitive to the number of trips compared to length of bike lanes and population variables.
Both models did not show any significant association with public transport routes and public transport stops.The lack of a significant link between the number of nearby public transport routes and stops with e-bike usage, aligns with the findings of Hosseini et al. (2023), suggesting that e-bike stations can coexist with public transport options without negatively impacting their utilisation and the concern about service overlap.This observation indicates that shared e-bikes, particularly those from docking stations, may serve as an independent mode of transport rather than merely complementing public transport.
Further, predictor factors exhibiting a relationship with the dependent variable are selected as inputs for the proposed DEA model, which is developed to analyse the impact of external variables on the performance of e-bike sharing system (Fig. 4).The current research is unique in its incorporation of external factors for the evaluation of micro-mobility hubs, diverging from previous studies that focused on service provider investment and system capacity as inputs (Hosseini et al., 2023;Hong et al., 2020).The current study thereby uncovers different aspects of performance evaluation in micro-mobility hubs.

CO 2 emission-saving assessment of e-bike sharing system
To examine the impact of the e-bike sharing system on climate change within Dublin, the actual CO 2 savings monthly and over the entire period are quantified in this section.For a comprehensive view of the decarbonisation profile of eHUBs, three scenarios are considered (Table 5).In the first, it is assumed that all journeys undertaken by e-bikes would have otherwise been made using private cars.However, in practice, it is overly optimistic to assume that e-bike trips will exclusively replace journeys that are typically made by private cars.Thus, considering uncertainty in CO 2 quantification (Wei et al., 2021), we define the second and third scenarios based on Bigazzi and Wong's (2020) meta-analysis, which examined e-bike mode substitution instances from more than 20 studies and revealed a median substitution rate of 24 % for private cars.In the second scenario, it is posited that 24 % of journeys undertaken by e-bikes would substitute trips normally made by ICEVs.This is because in Dublin, less than one percent of private cars are electric vehicles (Department of Transport, 2022).In assessing the eco-efficiency of the e-bike sharing system, the emissions savings were determined using the second scenario as the basis.
It is noteworthy that 58 % of ICEVs in Dublin utilise diesel fuel, whilst the remainder rely on petrol (Department of Transport, 2022).The CO 2 emission factors for diesel and petrol cars in urban settings in 2020-2025 stand at 0.1501 and 0.1483 kg/km, respectively (Department of Transport, 2021).Therefore, the CO 2 emission factor of an ICEV (EF Private car ) in Dublin is computed to be 0.1493 kg/km, based on the weighted average calculation.Utilising this EF Private car allows us to quantify the CO 2 emissions produced by ICEVs.Furthermore, to ascertain the emission factor for e-bikes (EF e− bike ), one must multiply the emission factor for electricity Note: In the first and second scenarios, the fleet consists of 58 % diesel ICEVs and 42 % petrol ICEVs.In the third scenario, the fleet is composed of 41.13 % diesel ICEVs, 25.49 % petrol ICEVs, and 33.38 % EVs.In the first scenario, it is assumed that all trips undertaken by e-bikes would otherwise have been made using private cars.In the second and third scenarios, it is posited that 24 % of trips made by e-bikes would substitute journeys normally made by private cars.
K. Hosseini et al. generation (kg CO 2 /kWh) by the electricity consumption rate of e-bikes (kWh/km).Ireland's Provisional Greenhouse Gas Emissions report, published by the Environmental Protection Agency (EPA, 2023), highlighted that the emission factor for electricity generation is 0.2958 kg CO2/kWh in Ireland.The electricity consumption rate is set at 0.25 kWh/hr, with an assumed average speed of 20 km/hr for e-bikes within the eHUB network in Dublin.Based on these parameters, the EF e− bike for shared e-bikes in Dublin is calculated to be 0.0037 kg/km.This value aligns with the EF e− bike as mentioned by Kmet (2022).Note that the distance data is sourced exclusively from three MOBY-operated stations: Dun Laoghaire, Greenhills, and Sandyford.For the other nine stations operated by BLEEPER, the regression imputation method was employed to estimate the distance travelled (Table 5), using the duration of trips and distance data provided by MOBY as a basis.This missing data represents one of the limitations of our study, a common issue in research that utilises real-world data.Equation (5) calculates the CO 2 emissions saved by the shared ebikes in the first scenario, while Equation ( 6) computes the savings in the second scenario.
Here, CO2 e− bike represents the CO 2 produced using e-bikes with total distance travelled D at station j, taking the CO 2 emission factor as EF e− bike .Similarly, CO2 Private car represents the CO 2 produced using ICEVs with total distance travelled D at station j, taking the CO 2 emission factor as EF Private car , assuming that all the e-bike trips are replaced with ICEVs.Accordingly, Equation ( 5) calculates CO2 1st scenario Saved , representing the CO 2 saved, under the assumption that all journeys executed by e-bikes would alternatively be conducted using ICEVs in the absence of e-bikes.Equation ( 6) estimates CO2 2nd scenario Saved , based on the assumption that 24 % of the trips made by ebikes are substitutes from ICEVs.
It is noteworthy that the presence of electric vehicles (EVs) in the current fleet composition was minimal, thus considered negligible in the CO 2 offset estimates for the e-bike sharing system in the first and second scenarios.However, given the importance of assessing how CO 2 emission savings from eHUBs are sensitive to changes in vehicle composition, particularly with an increase in the proportion of EVs, a third scenario has been developed.This scenario aligns with the Government of Ireland's climate change mitigation strategy for significant EV uptake by 2030 (GoI, 2023).Projections anticipate a future fleet composition of 41.13 % diesel ICEVs, 25.49 % petrol ICEVs, and 33.38 % EVs in 2030 (Stefaniec et al., 2024).Assuming an 80 % share of renewables by 2030 (GoI, 2023), the projected emission factor for electricity generation in Ireland would be 0.1042 kg CO 2 /kWh (Stefaniec et al., 2024).With an average energy consumption for EVs set at 0.166 kWh/km (EEA, 2023), and presuming no significant technological advancements in EVs and e-bikes that reduce their energy consumption by 2030, the projected emission factors are 0.0173 kg/km for EVs and 0.0013 kg/km for e-bikes (EF e− bike− 2030 ).Hence, based on the abovementioned projected future fleet composition, the CO 2 emission factor for a private car in 2030 (EF Private car− 2030 ) is calculated to be 0.1053 kg/km, using a weighted average approach.Equation ( 7) is used to estimate the CO 2 emissions savings by the shared e-bikes in the third scenario.
In terms of sensitivity analysis, the reader should be aware that the estimates in the third scenario rely on the share of renewables in electricity generation and the penetration of EVs in the fleet, which are projections rather than actual figures.CO 2 emissions saved by e-bikes, when private cars consist solely of ICEVs (second scenario), amount to 394.01 kg (Table 5).With one-third replacement of EVs in fleet composition and an 80 % share of renewables by 2030 (third scenario), CO 2 emissions saved by e-bikes equal 281.43 kg, indicating that EVs save 112.58 kg of CO 2 .This implies that even if the government successfully meets its ambitious target of one-third EVs in the 2030 fleet, e-bikes will still contribute more to carbon savings compared to EVs.When the car fleet is cleaner, it will naturally reduce the carbon-saving impact of e-bikes.However, this is not yet evident in the third scenario (see the last row in Table 5), and there is even an observed increase in CO 2 reduction by e-bikes.Only if higher penetration of EVs or a cleaner grid is achieved (for instance, with 100 % EV replacement under an 80 % renewable grid) would this figure reach 92.48 %.This analysis shows that changes in travel behaviour and a modal shift towards e-bikes and other active travel modes are more effective than adopting EVs in reducing CO 2 emissions from urban transportation.Table 5 illustrates that, in the best-case scenario, 3,283.36kg of CO 2 per year could be conserved, given the current usage patterns and the eHUBs present in Dublin.Nonetheless, this is considerably distant from the system's optimal potential when eHUBs are employed to their fullest capacity.At present, each eHUB is equipped with eight e-bikes available for hire.Assuming each e-bike is leased for 16 h daily at a consistent speed of 20 km/h, this would culminate in an annual distance of 934,400 km covered by e-bikes.Such usage would lead to the conservation of 136,048.6 kg of CO 2 for a single eHUB.Considering that in 2022, road transport in Ireland was responsible for 11.0 Mt of CO 2 eq, with passenger cars contributing to 53 % of this total (EPA, 2023), counterbalancing 5 % of the passenger cars' CO 2 emissions would require a total of 2,143 eHUBs across Ireland.However, to optimise the utilisation patterns of this e-micro-mobility solution, a profound shift in travel behaviour must be orchestrated, incentivised, and bolstered at governmental, industrial, and individual tiers.The government's role is paramount in this transformation, given its capacity to implement public policies, thereby steering travel behaviour toward a more sustainable manner.

Eco-efficiency model
This section introduces an analytical framework designed to evaluate the eco-efficiency of e-bike sharing system in Dublin.The proposed multi-period framework is two-fold.Firstly, by measuring the eco-efficiency scores of each station over six months, the framework provides a dynamic appraisal of the system's performance.Secondly, the returns-to-scale analysis furnishes decisionmakers with insights into the decarbonising and profitability profile of both the entire chain and its individual stations.This empowers the service planners to discern where to expand or contract the system, and where best to allocate investments, thus optimising the entire system.
DEA is a technique anchored in data-driven evaluations assessing the relative efficiency of decision-making units (DMUs).This methodology is widely applied in diverse areas of the transportation sector for its robust efficiency evaluation capabilities (Stefaniec et al., 2021;Emrouznejad and Yang, 2018).DEA is adept at generating a singular performance metric for each DMU, considering multiple inputs used to yield multiple outputs.The foundational framework for DEA was established by Charnes et al. (1978), showcasing the application of linear programming techniques in assessing performance metrics.
To assess the e-bike sharing system using an empirical dataset, the objective function aims to optimise the proportional augmentation in outputs denoted by φ, incorporating slack variables s + r and s − i as delineated in Equation ( 8).These slacks capture the inefficiencies in the inputs and outputs.The constraints ensure that the weighted combination of inputs for the evaluated DMU does not exceed its actual input level, and that the weighted combination of outputs is at least the proportional increase times its actual output level.
This formulation allows for the relative eco-efficiency of DMUs to be evaluated based on multiple inputs and outputs.In this equation, φ is the efficiency score of DMU o , and the goal is to maximise the score.The j-th DMU has inputs x ij and outputs y rj , and the λ j are weights assigned to each DMU.The element ε > 0, characterised as a non-Archimedean component, is defined to be less than any positive real value.It ensures that the slacks have a minor but non-zero impact on the objective function, prompting the model to minimise them.The eco-efficiency of each DMU can be assessed using the envelopment form of the output-oriented Charnes-Cooper-Rhodes (CCR-DEA) model as outlined below (Charnes et al., 1978): (8) Consider a system spanning multiple periods, encompassing q distinct periods as illustrated in Fig. 5.The superscript p in X (p) ij and Y (p)  rj signifies the specific period in question.The aggregate quantities for the i-th input and the r-th output across all q periods for DMU j are represented as X ij = ∑ q p=1 X (p) ij and Y rj = ∑ q p=1 Y (p) rj , respectively.As illustrated in Fig. 5, the multi-period system spanning q time periods can be analogised to a parallel network system comprising q processes.This suggests that accounting for the operations of individual periods when assessing the cumulative efficiency across q Fig. 5. Configuration of the multi-period system.
K. Hosseini et al. periods can provide a more detailed understanding of the entire system.When viewing the multi-period system through the lens of a parallel network, the activities of each distinct period can be incorporated into the assessment of the system's overall efficiency.Park and Park (2009) introduced an aggregate efficiency model to gauge the performance of multi-period production entities, drawing upon the Bebreu-Farrell technical efficiency paradigm (Debreu, 1951;Farrell, 1957).Their output-oriented model, rooted in the Banker-Charnes-Cooper (BCC-DEA) approach (Banker et al., 1984), incorporated convexity constraints.Further, Kao and Liu (2014) adapted this model to an input orientation based on the CCR-DEA framework, and because the distance measure φ ECO connects all q periods, they named the result the connected network model.We modified this CCR-DEA model for the current output-oriented case study of an e-bike sharing system, using the linear program below: (9) In the above model, the efficiency score is bounded between zero and unity, reflecting relative performance among DMUs.A DMU is deemed fully efficient when its efficiency score is one, while scores between zero and one indicate varying levels of inefficiency, with scores closer to zero denoting greater inefficiency.The results of the eco-efficiency analysis will be presented in Section 4.1, detailing the performance metrics of the e-bike sharing system.

Returns-to-scale
Returns-to-scale, in economics, describe the effect of proportional increases in inputs on outputs.There are three recognised types: Increasing returns-to-scale (IRS) occur when a proportional increase in inputs results in more than a proportional increase in outputs; constant returns-to-scale (CRS) lead to exactly proportional increases and decreasing returns-to-scale (DRS) result in less than a proportional increase in output (Banker et al., 2004).The primary focus of discussions typically centres on qualitative characterizations, determining whether returns-to-scale are categorized as increasing, decreasing, or constant.These qualitative assessments help understand efficiency and cost dynamics in various economic contexts.
Returns-to-scale can explore the relationship between the scale of service provision (inputs) and its impact on efficiency and costs (outputs).In our case study, IRS suggests that as a shared mobility hub expands, it can offer transportation services more efficiently, potentially increasing riders and ridership and reducing per-user costs.Conversely, excessive expansion of a station characterised by DRS leads to declining efficiency and higher operational costs per rider.
In the case of e-bike sharing systems, IRS and CRS indicate that stations are located well, and possible increases of inputs, like an increase in bike lane infrastructure, can make individual stations more efficient, improving the system's cost-effectiveness.However, a station that experiences DRS is not in a suitable location, and carelessly increasing its scale could lead to diminished efficiency and higher operational costs.These results highlight the importance of careful expansion and resource allocation.
To determine returns-to-scale, Banker et al. (1984) proposed a model under variable returns to scale; this is known as the BCC-DEA approach.The output-oriented BCC-DEA model is as follows (Banker et al., 1984): To evaluate the returns-to-scale of focal DMU o based on the efficiency scores obtained from Equations ( 8) and ( 10), there are two cases (Seiford and Zhu, 1999).
In circumstances whereφ (Banker, 1984).The Most Productive Scale Size (MPSS) refers to the optimal operational scale for a DMU, where it maximizes efficiency by optimally utilising available inputs to produce outputs, considering factors such as costs, infrastructure, demographics, and market conditions.Operating at this scale ensures that resources are neither underused nor wasted (Assani et al., 2018).

Eco-efficiency and scalability analysis
The proposed eco-efficiency framework measures how well an e-bike sharing station takes advantage of available infrastructure, neighbouring demographics, and its location to generate desired economic, social, and environmental outputs.One of the contributions of this study is its integration of socio-economic and decarbonising indicators, providing a holistic evaluation of the transport sector's transition from private mobility to shared micro-mobility.
For the empirical assessment of Dublin's e-bike sharing network through the eco-efficiency model, three inputs and five outputs are employed.Table 6 presents the evaluation outcomes of 61 station-month units in Dublin, derived from the proposed multi-period ecoefficiency framework utilising R software.In the CCR model (Equation ( 8)), DMUs with a score of one are deemed efficient that is utilising available resources satisfactory.Greater efficiency scores indicate higher performance.Note that these scores are relative, reflecting each station's performance in comparison to others.Consequently, low scores do not signify poor system-wide performance but rather highlight specific stations that lag behind their peers within the network.Seven out of 61 units were relatively eco-efficient during the study period with φ CCR = 1.
During the assessment period, certain stations experienced vandalism, leading to them being non-operational or witnessing diminished riding demand.The Firhouse and Tallaght stations were affected over all six months, so they are excluded from our analysis.The Finglas station was affected for four months, the St. Lomans station for the last three months, and both the Park West and Santry stations for two months each.As a consequence, 11 station-months were removed from the analysis, resulting in a total of 61 station-months being considered for the study.This could be a significant factor contributing to the suboptimal performance of these stations.For instance, the Park West station, situated away from main roads and possessing low visibility, is more frequently subject to vandalism.The Swords eHUB, despite being operational throughout the study period, may have obtained a low performance score due to its remote location and distance from the city centre (Fig. 1).As an innovative, sustainable mobility initiative, these eHUBs tend to be focal points for antisocial activities, and a certain amount of time might be required before they seamlessly integrate into the standard street ecosystem.
From June onwards, the service provider introduced several virtual stations for the three eHUBs under the MOBY company to evaluate their impact on eHUB performance.Specifically, seven virtual stations were established around Sandyford, four around Dun Laoghaire, and two near Greenhills.Trips starting from each of these eHUBs could be concluded at one of these virtual stations, located no further than 2.6 km from the corresponding eHUB.This introduction of virtual stations is a potential reason for the high ecoefficiency observed at the Sandyford station from June onwards and also for Dun Laoghaire in June.
Returns-to-scale assesses whether the system is operating at an optimal size, providing crucial insights into the scalability and optimal sizing of the system (Equation ( 11)).It considers whether expanding, reducing, or maintaining the current scale of operations would be more efficient given the inputs and outputs.Nearly 60 % of the DMUs (station-month samples) exhibit IRS.Over 31 % display CRS or operate at their most productive scale size (MPSS), while fewer than 10 % (all at the Swords station) show DRS (Table 6).These findings suggest that a greater investment in expanding the e-bike system, coupled with infrastructural enhancements around the stations, could significantly boost the system's performance.In other words, for each unit of investment, whether from operators expanding the system or the government developing additional bike lanes, the return will exceed the initial input, benefiting the community economically, socially, and environmentally.
It is imperative to highlight that, of the 61 station-months analysed, MPSS was discerned in a mere seven instances.This underscores the fact that the system has yet to attain its optimal operational capacity.Augmenting both the financial investment and the infrastructural aspects can significantly enhance the system's eco-efficiency.Given that the majority of DMUs' returns-to-scale types are IRS, there is a compelling indication that proportionally larger benefits can be achieved for each unit of expanded investment.Intriguingly, MPSS predominantly occurred in the concluding months, notably in June and August, suggesting an overall upward trajectory of the system on a monthly basis.However, July exhibited a discernible decrement in performance for the majority of stations in comparison to June.This decline is potentially attributable to climatic adversities, with March and July 2023 being documented as Ireland's most precipitative March and July in recorded history by The Irish Meteorological Service (2023).Table 7 lists the aggregate eco-efficiency scores and rankings for the 12 considered eHUBs within the system, derived from the modified connected network model (Equation ( 9)).Higher eco-efficiency scores indicate a superior ranking.These scores range between zero and one.During the assessment period, the Sandyford station emerged as the most eco-efficient, followed by Howth and Dun Laoghaire.Conversely, Finglas, Swords, and Parkwest were identified as the least eco-efficient.The arithmetic mean of aggregate eco-efficiency scores was calculated to be 0.490.Leveraging this mean, stations were divided into two clusters: the five upper echelons with scores exceeding 0.490 and the seven others falling below this threshold, labelled as strongly eco-efficients and weakly ecoefficients, respectively.Further exploration of spatial patterns and their underlying reasons is provided in the spatial analysis section.

Spatial analysis
Interpreting why certain stations outperform their peers is a multifaceted and complex task.In this section, several geographical maps are presented, illustrating various characteristics of each eHUB's catchment area using QGIS version 3.28.3-Firenzesoftware (QGIS, 2023).These maps, combined with information about the locations of the two eHUB clusters, namely, strongly eco-efficients and weakly eco-efficients, offer illustrative insights into why some stations perform better than others.Consequently, this analysis can guide improvements for low eco-efficient stations and provide insights for service planners regarding potential locations for future eHUBs.Notably, stations located along the coastline demonstrated higher eco-efficiency, with all three coastal stations ranking in the strongly eco-efficient cluster.In contrast, the low eco-efficient stations are predominantly situated in the south-western or central northern regions of the city.
Fig. 6 displays the deprivation conditions of Small Areas Popluations (SAPs) overlapping with station catchment areas, using the Pobal HP Deprivation Index (Pobal, 2023) as a reference.This index assesses the overall affluence and deprivation of these SAPs, drawing from the 2022 census data.It incorporates three key dimensionsdemographic profile, social class composition, and labour market situationto offer a detailed perspective on the socioeconomic status of these areas (Pobal, 2023).Fig. 6 shows the stations in order of their aggregate eco-efficiency score from highest to lowest according to Table 7. Notably, stations with higher eco-efficiency scores are predominantly found in more affluent areas, in contrast to their lower eco-efficient counterparts, which are more frequently situated in disadvantaged regions.
Figs. 7 and 8 display the number of businesses and the number of persons at work in Dublin's SAPs, respectively.Some data on the number of businesses in SAPs were missing, as depicted in Fig. 7, which constitutes a limitation in this study.The Spearman's correlation coefficient calculated between the eco-efficiency scores and the available data on the number of businesses within the catchment areas of the stations was 0.508.It can be observed that strongly eco-efficient eHUBs are more likely to be located in areas with a higher concentration of businesses (Fig. 7).This is evident in locations such as Sandyford, Dun Laoghaire, Malahide, and Blanchardstown.This observation likely confirms that most trips taken by shared e-bikes were non-leisure journeys, probably commutes to workplaces.Similarly, though to a lesser extent, the catchment areas of high eco-efficient eHUBs are more likely to cover SAPs with a greater number of persons at work (Table 8).

Conclusions and policy implications
This research undertakes a quantitative evaluation of Dublin's pilot e-bike sharing system, employing six months of real-world GPS data.It pioneers a comprehensive framework encompassing data processing, inferential analysis of external factors, emission savings assessment, eco-efficiency evaluation of the network, and spatial analysis with geographical indicators.The proposed eco-efficiency framework evaluates the proficiency of an e-bike sharing station in leveraging its location and neighbouring infrastructure and adjacent population to yield the anticipated economic, social, and environmental outcomes.The studied system comprises 12 eHUBs placed around a congested motorway, aimed at encouraging both residents and commuters from satellite towns to switch from private vehicles to e-bikes for their city centre commutes.The analysis of the carbon-saving potential and usage patterns of these eHUBs provided insights into their effectiveness in promoting the shift from private vehicles to shared e-bikes.Importantly, the returns-toscale analysis provides insights into the system's scalability, determining the potential for the capacity growth while maintaining efficient operations and service quality, and the profitability prospects of both the entire network and individual stations.As a result, this study offers service providers essential information to make informed decisions regarding the improvement and expanding their services.
The study finds that approximately 90 % of the eHUB station-month units demonstrated increasing or constant returns when scaledup, denoted by IRS and CRS, respectively.Hence, the analysis of the e-bike sharing system highlights its high scalability and strongly suggests that expanding the system, alongside improvements in surrounding infrastructure, would significantly enhance overall performance.Essentially, any investment directed towards system expansion or infrastructural development, such as additional bike lanes, is likely to yield benefits surpassing the initial expenditure, contributing positively to the surrounding community.
The outcomes also reveal a link between the adoption of shared e-bikes and the presence of nearby bike lanes.To enhance e-bike usage, Dublin needs to focus on improving cycling infrastructure.The existing fragmented bike lanes and lack of dedicated paths and separate cycleways can deter potential users.Consequently, this evidence could contribute to resolving the ongoing political debate over expanding cycling infrastructure, particularly if the objective is to increase accessibility and social inclusion for disadvantaged groups in society (SDG 10) and to establish a safe, affordable, accessible, and sustainable transport system for all (SDG 11) and to integrate climate change measures into transportation policies and planning (SDG 13).Furthermore, to enhance road safety for vulnerable users like cyclists and pedestrians, Ireland could revise its legal liability system, possibly adopting 'strict liability'.This legal principle, widely implemented across continental Europe, presumes the more powerful vehicle (typically private cars) liable in traffic collisions, thus protecting vulnerable road users and encouraging more cautious driving.While it does not alter criminal responsibility, it establishes a civic duty for drivers to have insurance covering vulnerable victims, regardless of fault (Pooley et al., 2011).This measure could also address the current situation on Irish streets, where car drivers often behave as if they have more rights than other road users.Implementing such a system could incentivise motorists to drive more carefully, thereby protecting those most at risk on the roads.Combined with cycling infrastructure improvements and redistributive active travel measures (Egan and Caulfield, 2024), this approach could significantly boost active travel and e-bike usage.
During the study period, seven out of 61 station-month units were identified as relatively eco-efficient.Stations located along the coastline demonstrated notably higher performance, with all three coastal stations ranking in the strongly eco-efficient cluster.The results also indicate an overall upward trajectory for the system on a monthly basis.However, it is important to note that uncontrollable weather conditions, particularly wetter weather in July, negatively influenced this trajectory.Furthermore, The eHUBs, representing an innovative approach to sustainable mobility, often attract antisocial activities and may require a period of adjustment before they are fully assimilated into the typical urban street landscape.
The CO 2 savings assessment of shared e-bikes demonstrates significant potential for this mode of transport, aligning with findings from previous studies on e-bikes (Hiselius and Svensson, 2017;Philips et al., 2022;McQueen et al., 2020).For a comprehensive view of the decarbonisation profile of the e-bike sharing system, three scenarios were evaluated: full replacement of car journeys with e-bike trips, a more conservative estimate derived from the meta-analysis by Bigazzi and Wong (2020), and a sensitivity analysis scenario based on the projected higher adoption of EVs in the private car fleet and cleaner electricity generation.The outcomes of the third scenario indicate that even if the government successfully meets its ambitious target of one-third penetration of EVs in the 2030 fleet, ebikes will still contribute more to carbon savings than EVs.When combined with other benefits, such as reducing traffic congestion and enhancing well-being through increased physical activity, e-bikes should be given a more significant role in future transport policy formulations.Further analysis suggests that, to target a 5 % reduction in CO 2 emissions from passenger cars in road transport, approximately 2100 eHUBs would be needed across Ireland.However, this calls for a significant shift in travel behaviours, supported by efforts at various levelsgovernmental, industrial, and individual.The challenges to this transition are diverse, spanning from inadequate cycling infrastructure to deeply rooted social norms favouring private car usage.Governmental action plays a crucial role in facilitating this shift through the implementation of policies promoting and incentivising sustainable travel.
The spatial analysis reveals lower shared e-bike usage in less affluent areas despite lower car ownership rates in these regions.This could indicate that the current pricing model for shared e-bikes might not be affordable for more disadvantaged social groups.Additionally, for shared mobility solutions to gain traction among communities, concerted efforts from both private entities and government bodies are essential.Educational and awareness campaigns and both monetary and non-monetary incentives are crucial in promoting a shift towards shared mobility.Furthermore, the robust and positive correlation between the obtained eco-efficiency scores and both the number of businesses and the number of persons working in the catchment areas of e-bike sharing stations reinforces the conclusion that most trips taken by shared e-bikes were non-leisure, likely commutes to workplaces.
The analysis further shows that the majority of trips using the e-bike sharing system took place on weekdays, notably during and between the morning and evening peak hours, typically associated with non-leisure travel.This characteristic of the e-bike sharing system, along with its standalone nature as a transport mode (Hosseini et al., 2023), elucidates the potential to alleviate overcrowding on public transport, especially during peak times and along frequently used routes that struggle to accommodate all passengers.Moreover, during the study period, the user demographic comprised approximately one-third females and two-thirds males.To narrow this gender gap, it is crucial to challenge and alter the gendered perceptions and social norms surrounding cycling.Tackling the social stigma and negative connotations associated with women cycling, along with ensuring safer and more inclusive cycling infrastructure, could significantly contribute to bridging the gender gap in cycling (Egan and Hackett, 2022).
To conclude, the outcomes indicate that incorporating shared electric micro-mobility hubs in urban areas grappling with car dominance, such as Dublin, offers a sustainable alternative with the potential to reshape urban mobility paradigms.Essential steps for effective transport reform toward active and sustainable mobility include a combination of discouraging measures, such as introducing strict liability, and encouraging policies, like enhancing conditions for affordable, accessible, and safe cycling and e-cycling.The authors hope that this data-driven study lays the foundation for further research and practical endeavours, enabling a more thorough exploration of the efficiency and scalability of shared micro-mobility modes and the formulation of transformative policies for their successful implementation.Future research could explore user resistance to adopting sustainable mobility practices such as shared micro-mobility, a topic currently underrepresented in the literature.Additionally, similar to the approaches of Nocera et al. (2015) and Jiang et al. (2019), future research may consider estimating the economic impact of externalities mitigated by shared micro-mobility hubs, thereby supporting more informed decision-making for their scaling up.Also, investigating the detailed mapping of e-bike trip routes through GPS data could be a useful direction for future studies to enhance the understanding of their spatial usage patterns.Such research could provide practical insights into travel behaviours and preferences, aiding in the development of more targeted

Fig. 2 .
Fig. 2. Daily temporal distribution of e-bike trips by hour from the Dublin e-bike sharing system, March-August 2023.

Fig. 3 .
Fig. 3. Comparison between Poisson distribution and negative binomial distribution regarding the number of trips undertaken by shared e-bikes in Dublin in March-August 2023.

Fig. 4 .
Fig. 4. Illustration of the proposed DEA model for eco-efficiency assessment of e-bike sharing system.

Fig. 6 .
Fig. 6.Deprivation conditions of SAPs where 12 eHUBs catchment areas are situated (the first author obtained a Licence Agreement for the Pobal HP Deprivation Index for Small Areas from the developer (Dr Jonathan Pratschke) for the data used to create this map).

Table 1 E
-bike trip durations and weekly usage distribution from March to August 2023.
K.Hosseini et al.

Table 2
Average durations of e-bike trips across different daily time periods during March-August.Time periodAverage durations of e-bike trips (minutes)

Table 3
Descriptive statistics of predictor variables.

Table 4
Results from the random effect negative binomial model.Note: Sig.: level of significance, t: t values, AIC: Akaike Information Criteria, BIC: Bayesian Information Criteria, LL: Log-likelihood, DOF: Degrees of Freedom.

Table 6
Eco-efficiency scores and Returns-to-scale results for the eHUBs for the period March-August 2023.

Table 7
Aggregate eco-efficiency scores of stations based on the connected network model.

Table 8
Spearman's correlation coefficients between the obtained eco-efficiency scores and the number of businesses and persons at work in the eHUBs' catchment areas (March-August 2023).