Journal of Transport Geography 111 (2023) 103664 Contents lists available at ScienceDirect Journal of Transport Geography journal homepage: www.elsevier.com/locate/jtrangeo E-scooter and bike-share route choice and detours: Modelling the influence of built environment and sociodemographic factors Jerònia Cubells a, *, Carme Miralles-Guasch a, b, Oriol Marquet a, b a Grup d’Estudis en Mobilitat, Transport i Territori (GEMOTT), Departament de Geografia, Universitat Autònoma de Barcelona, Carrer de la Fortuna s/n, 08193 Bellaterra (Cerdanyola del Vallès), Barcelona, Spain b Institut de Ciència i Tecnologia Ambientals (ICTA), Universitat Autònoma de Barcelona – Edifici ICTA-ICP, Campus de Bellaterra, 08193 Cerdanyola del Vallès, Barcelona, Spain A R T I C L E I N F O A B S T R A C T Keywords: Micromobility is often presented as a sustainable, affordable, and active urban transport option, in comparison to Route choice motorised modes. Understanding users routing preferences could help policymakers adapt and design facilities Bicycle that attract a myriad of micromobility users. Whereas previous research largely focused solely on the built Electric scooter infrastructure, the ways in which sociodemographic factors affect micromobility route choice and infrastructure Travel behaviour GPS preferences are unclear. This study examines how elements of the built environment and sociodemographic Built environment attributes influence the route selection of 115 e-scooter and bike-share users in Barcelona, Spain. We also compare participants’ GPS-tracked trips to the shortest path that they could have followed and develop a multilevel model to estimate how urban and sociodemographic factors affect the decision to deviate from the shortest path. The findings show that micromobility users rarely choose the shortest path since urban elements related to safety, accessibility and aesthetics seem to shape their wayfinding decisions. Results help us comprehend cyclists’ and e-scooter riders’ distinct route preferences and further illustrate how the gender identity of micromobility users influences route choice and detour. The models indicate that, on average, women take shorter detours than men. We observe gender differences in the way cyclists and e-scooter riders favour certain elements in their trips, such as parked cars and cycling infrastructure. Our findings offer valuable insights into how sociodemographic factors interact with infrastructure and built environment conditions to influence micromobility users’ route choice and open up the potential to use these results to manage micromobility flows within cities. 1. Introduction distances to avoid traffic and its impacts, such as injuries, noise and pollution (Desjardins et al., 2021b; Scott et al., 2021). This excess travel Cities have witnessed unprecedented growth in micromobility levels requires more energy consumption per trip and thus can thwart the over the last decade, which draws environmental and social challenges, attractiveness of micromobility modes by decreasing their convenience but also opportunities to the traditional transport system (Felipe-Falgas and extending travel times. Additionally, numerous detours that et al., 2022; Pucher and Buehler, 2017). Electric scooters (e-scooters outstretch travel distances may hinder these modes contribution to hereafter) and bicycles attract new users due to their innovativeness, reducing transport externalities, which can be translated into health and sustainability, affordability and speed (Bhandal and Noonan, 2022; economic extra costs (Gössling et al., 2019). However, deviating from Bretones and Marquet, 2022). This wave of popularity has brought an the shortest path may just be a consequence of the increased adaptability increasing interest in understanding how the built environment favours of these modes to the user’s needs and preferences. Because of their micromobility use (Codina et al., 2022; Cole-Hunter et al., 2015), along novelty, not much is known yet on how elements of the built environ­ with how specific urban elements affect riding behaviour. Among the ment differently impact multiple social groups in their micromobility latter group of studies, the issue of route choice has drawn significant detour decisions. While research recognizes that route choice is ulti­ attention. It has been seen that cyclists and e-scooter riders travel longer mately influenced by personal preferences and fears (Hardinghaus and * Corresponding author. E-mail addresses: jeronia.cubells@uab.cat (J. Cubells), carme.miralles@uab.cat (C. Miralles-Guasch), oriol.marquet@uab.cat (O. Marquet). https://doi.org/10.1016/j.jtrangeo.2023.103664 Received 28 December 2022; Received in revised form 8 July 2023; Accepted 11 July 2023 Available online 14 July 2023 0966-6923/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by- nc-nd/4.0/). J. Cubells et al. Journal of Transport Geography 111 (2023) 103664 Weschke, 2022; Ravensbergen et al., 2020a), sociodemographic factors choice (Heim LaFrombois, 2019; Ravensbergen, 2022; Ravensbergen such as gender tend to be overlooked, with only a handful of exceptions et al., 2020a). Similarly, Aldred et al. (2017) suggested that risk toler­ (Bernardi et al., 2018; Fitch and Handy, 2020; Prato et al., 2018). ance to traffic might also be different across age groups, which translates Filling this gap is important both to better assess the impacts of into older people having stronger preferences for separated infrastruc­ micromobility and to improve the design of urban spaces aimed to invite ture. Women and older adults also seem to take longer detours than men a myriad of cyclists and e-scooter users. With that objective, this paper to ride through parks and quiet neighbourhoods (Hardinghaus and examines how route choice is influenced by elements of the built envi­ Weschke, 2022; Nawrath et al., 2019). Notwithstanding, these route ronment and sociodemographic attributes using the GPS-tracked trips of preferences seem to take effect only during daylight, as fear and pros­ 115 e-scooter and bike-share users in Barcelona, Spain. We further pects of harassment and assault appear to override these paths during compare those trips to the shortest paths that participants could have nigh-time (Pellicer-Chenoll et al., 2021; Sersli et al., 2022). followed and build a multilevel model to estimate how urban attributes Previous literature has found that parents cycling with children are and sociodemographic factors affect the decision to detour from the also willing to deviate to run into green areas (Hardinghaus and shortest path. Moreover, we contribute to the literature by investigating Weschke, 2022). Indeed, cycling with kids or through pregnancy is e-scooter and bike-sharing route preferences within the same spatial perceived to entail an increased sense of responsibility that translates context. into seeking routes with protected bicycle lanes while eluding parked cars, traffic and narrow streets (Desjardins et al., 2021a; Heim 2. Literature review LaFrombois, 2019; Russell et al., 2021; Sersli et al., 2020). These route preferences have relevant social implications within the context of In recent years, several attempts have been made to analyse bicycle highly gendered travel behaviours. Combining care work and cycling route choice, whereas research on e-scooter wayfinding is still limited. entails developing specific strategies, not only for parents but also for Most of these studies acknowledge that cyclists tend to prefer short and cyclists that incorporate caring tasks such as shopping into their rides. As simple routes. Li et al. (2017), for instance, found that cyclists’ route such, some have suggested combining frequent bike routes with access choice is greatly influenced by distance, while Lu et al. (2018) observed to everyday and grocery shopping as means of facilitating complex that cyclists tend to avoid intricate routes. Nonetheless, Zimmermann travel schedules (Janke and Handy, 2019; Russell et al., 2021). Never­ et al. (2017) demonstrated that the built environment largely affects theless, Ravensbergen et al. (2020b) investigated how accommodating cyclists’ route preferences and their likeliness to deviate from the grocery shopping on cyclists’ routes might reproduce, and even rein­ shortest path. Extensive research has shown cyclists’ willingness to force, a gendered division of household labour and care work. detour to look for dedicated cycling infrastructure (Ghanayim and In comparison to bicycle route choice, the study of e-scooter route Bekhor, 2018; Rupi et al., 2019; Wang et al., 2022). It has been noticed preferences is still incipient. It is noteworthy that differences in route that cyclists often favour well-paved and well-lit bicycle lanes, separated preference between these vehicles might be due to the distinct socio­ from traffic and parked cars (Chen et al., 2018; Desjardins et al., 2021b; demographic characteristics of their riders. The fleet of e-scooter users is Majumdar and Mitra, 2019). In the same vein, cyclists tend to skirt on average younger, less affluent, and includes a higher share of men urban elements that slow down their pace such as crosswalks, traffic riders than the fleet of cyclists (Bieliński and Ważna, 2020). This finding lights and stops (McArthur and Hong, 2019; Prato et al., 2018). has also been noted in Barcelona, where cyclists also report higher ed­ With that, some cyclists are willing to pedal further to integrate parks ucation levels (Roig-Costa et al., 2021). The scarce literature on e- and green areas into their trips (Bernardi et al., 2018; Hardinghaus and scooters route choice shows that, overall, e-scooter riders share cyclists’ Nieland, 2021; Lin and Fan, 2020; Marquart et al., 2020). These findings route preferences. Similarly to cyclists, e-scooter users appear to favour are important because while recreational cycling has been linked with well-lit cycling infrastructure, but also sidewalks and one-way roads parks and green streetscapes, utilitarian cyclists are often highly exposed (Caspi et al., 2020; Yang et al., 2022; Zhang et al., 2021; Zuniga-Garcia to traffic-related pollutants (Lee and Sener, 2019; Sun et al., 2017). In et al., 2021). Furthermore, e-scooter riders seem likely to detour to go terms of route choice, no consensus yet exists on whether cyclists inte­ through green as well as commercial areas (Bai and Jiao, 2020; Yang grate shopping into their routes (Orellana and Guerrero, 2019; Sarjala, et al., 2022). 2019) or cycle away from shops and restaurants (Park and Akar, 2019; To date, only a handful of studies have simultaneously compared e- Skov-Petersen et al., 2018). In the same line, historical points of interests scooter and bicycle rides in the same spatial context. McKenzie (2019) and, presumably, the presence of tourist seem to repel bicycle rides of acknowledged slight differences in their spatial occurrence: while bikes local cyclists (Desjardins et al., 2022). dominated in the downtown, e-scooters presented a broader adoption in In addition to environmental factors, sociodemographic factors have the periphery of Washington, D.C.. In contrast, shared bikes had a more also been found to influence cyclists’ route choice (Fitch and Handy, widespread distribution in comparison to shared e-scooters in Singapore 2020). Schneider et al. (2021) observed that women are less likely to (Zhu et al., 2020). Haworth et al. (2021) found major variance in cycling detour from the shortest path than men. When asked about this, women infrastructure use, due to differentiated regulations on where each alleged that, pressed for time, balancing productive and reproductive vehicle should be ridden. Yet, cycling facilities use is similar between work constrained the possibility of extending their bicycle trips (Rav­ both vehicles in cities in which e-scooters and bicycles share a regulatory ensbergen et al., 2020b; Sersli et al., 2020). Although women on average framework on where they can be used (Cubells et al., 2023). Regarding present shorter detours than men, they seem willing to pedal further to vehicle ownership, most of the research gathered spatial data offered by use protected bicycle lanes (Hardinghaus and Weschke, 2022). To un­ e-scooter-sharing companies (Bai and Jiao, 2020; Caspi et al., 2020; derstand why women might outstretch their trips to use cycling infra­ McKenzie, 2019; Yang et al., 2022; Zhang et al., 2021). Therefore, route structure in a context of limited time availability, it is worth noting that choices of private e-scooter owners have, to our knowledge, only been cycling, in comparison to walking, has often been reckoned to magnify investigated using observational data (Haworth et al., 2021; Tuncer the visibility of women and people of colour, drawing unsolicited et al., 2020). Indeed, this approach let the authors reflect on the easiness attention and eliciting harassment (Lubitow et al., 2019). Sersli et al. of riders to dismount their vehicles and swap from the roadway to the (2022) found that cycling through traffic requires women to negotiate sidewalk, which might facilitate shortcutting (Tuncer et al., 2020). their presence in the public space and be exposed to aggressions, which However, it is worth noting that there could be differences in decision- some might be unwilling to do. Accordingly, this would trigger the de­ making processes between shared and private e-scooters. In the latter cision to travel greater distances to avoid specific areas even in a context case, vehicle availability and economic cost, could be relevant for users of time constraints. Consequently, safety concerns related to sexism and to decide their route. racism in public spaces are found to shape and limit cyclists’ route Recent attempts to assess preference for route attributes have greatly 2 J. Cubells et al. Journal of Transport Geography 111 (2023) 103664 benefited from GPS data. GPS tracking has the strength to infer travel all e-scooters ridden in Barcelona are individually owned and their trips behaviour indicators, such as route choice, and frame them within the represent 26% of the overall micromobility share. surrounding built environment (Park and Akar, 2019). Most studies investigating route choice from GPS-tracking data, however, have failed 3.2. Data source to disaggregate data by gender (Cho and Shin, 2022; Li et al., 2017; Lin and Fan, 2020; Lu et al., 2018; Pritchard et al., 2019; Sadeghinasr et al., In this context, we recruited 902 micromobility users through an 2021; Wang et al., 2022; Zhang et al., 2021; Zuniga-Garcia et al., 2021) intercept survey (more details to be found in Roig-Costa et al. (2021)). or tracked only a small sample size of cyclists who did not identify as Between September 2020 and July 2021, we surveyed 326 e-scooter men (Ghanayim and Bekhor, 2018; Lee and Sener, 2019; McArthur and owners and 325 bike-sharing cyclists through the CAPI (Computer Hong, 2019; Orellana and Guerrero, 2019; Sun et al., 2017; Zimmer­ Assisted Personal Interviewing) methodology. From the initial sample, mann et al., 2017), with very few exceptions (Bernardi et al., 2018; Fitch 65 e-scooter riders and 70 bike-sharing cyclists dwelling in Barcelona and Handy, 2020; Prato et al., 2018). Other approaches to assess route were further asked to wear a GPS device (QStartz BT-Q1000XT; QStarz choice determinants have used stated preferences surveys and in­ International Co., Ltd., Taiwan, R.O.C.) at all times during a week. The terviews. This body of literature has been particularly successful in GPS logged the location of the recruited cyclists and riders at each 15 s reviewing how multiple identities -including gender- influence cyclists’ epoch. This study was approved by the Ethics Committee on Animal and route preferences (Desjardins et al., 2021a; Hardinghaus and Weschke, Human Experimentation at Universitat Autònoma de Barcelona 2022; Marquart et al., 2020; Nawrath et al., 2019). In contrast to GPS (CEEAH-3656). Apart from GPS data, a second data set from the data, surveys ability to measure the use of cycling infrastructure and municipal open data repository containing characteristics of the built other elements of the built environment on the way is limited (Fitch and environment and exposure to pollutants was exploited (Ajuntament Handy, 2020). Ajuntament de Barcelona, 2022a). Both NO2 and acoustic pollution can be understood as proxies of traffic and road hierarchy. Further infor­ 3. Methodology mation on air and acoustic pollution measurement, modelling and mapping can be found elsewhere (Ajuntament Ajuntament de Barce­ 3.1. Study area lona, 2022b). Barcelona, with a population of 1.6 million residents (2020), is 3.3. Data processing located on the northwest coast of the Mediterranean Sea. The city has a compact and mixed urban planning, a uniform distribution of services GPS data were processed through the Human Activity Behavior and an extensive cycling infrastructure (Fig. 1) (Lind et al., 2021; Mar­ Identification Tool and data Unification System (HABITUS) software. quet and Miralles-Guasch, 2018). Micromobility, which includes shared This software identifies trips within trajectories and predicts their bikes as well as private bikes and e-scooters, constitute up to 4% of transport mode by calculating the distance and speed between consec­ Barcelona’s modal share (IERMB, 2021). Part of these bicycle trips is utive GPS points. HABITUS flagged trips with a 90th percentile speed performed with shared bikes from the city’s public operator Bicing. This between 1 and 10 km/h as walking. Following, trips with a 90th service can be acquired through an annual subscription and their users percentile speed between 10 and 25 km/h were targeted as micro­ perform 2 out of 10 of the micromobility trips registered in the city. This mobility and trips with a 90th percentile speed above the former speed dock-based bike-sharing system covers all city districts, and its threshold were classified as in vehicle (Kang et al., 2018). This software >100,000 subscribers undertake about 50,000 trips daily (Soriguera and has been validated to correctly assign micromobility 73% of the time it Jiménez-Meroño, 2020). On the contrary, the city does not provide an e- was actually used (Carlson et al., 2015). Walking and in vehicle trips scooter-sharing platform as it does not allow private e-scooter sharing were filtered out from the analysis. companies to operate within the municipality’s boundaries. Therefore, The remaining GPS points were map-matched to the street network Fig. 1. Barcelona cycling infrastructure and trip count at each street segment. 3 J. Cubells et al. Journal of Transport Geography 111 (2023) 103664 to contextualise the routes that micromobility users followed. GPS The 95 percentile of the detour percentage was excluded (40 routes) points were snapped to the street map using a dynamic buffer with an as they mainly accounted for outliers (Klein et al., 2021). This process upper search tolerance of 150 m (Li et al., 2020). A total of 1520 routes resulted in a final dataset of 696 routes (of which 51.6% were performed were then created with the Network Analysis toolset from ArcGIS Pro, by participants who self-identified as women) undertaken on 90 based on Dijkstra’s algorithm (Lu et al., 2018; Scott et al., 2021). To different days by 63 unique bike-sharing cyclists using conventional avoid outliers, we excluded (i) circular trips (174 routes), (ii) trips that bicycles (of whom 39.7% self-reported as women) and 52 e-scooter lasted <2 min or > 2 h (151 routes), (iii) trips outside the city limits (193 riders (of whom 46.2% self-reported as women). Each user participated routes), (iv) trips that contained unmatched points to the street network for an average of 2.7 days (sd = 1.5) and 6.3 trips (sd = 5.7) over the (17 routes), and (v) trips with observations >60 km/h (249 routes) week they were given the GPS device. Negative detour percentage (Clarry et al., 2019; Flügel et al., 2019; Zhang et al., 2021). values were kept in the analysis as they represent shortcuts, such as To answer whether micromobility users take detours, we compared riding through streets that are not catalogued in the OSM cycling chosen routes to the shortest path cyclists and e-scooter riders could network, sidewalks and/or contraflow (Fig.2). have followed without committing a traffic infraction. Shortest paths The next step was to measure the difference between chosen and were calculated between each trip origin and destination using the R shortest path. For each pair of routes, we calculated the percentage of package ‘openrouteservice’ (Oleś, 2022). The algorithm inferred the ride circulating through cycling facilities, average pollution exposure shortest route through the OpenStreetMap (OSM) cycling network, since and density of urban elements in a 50-m radius (Sarjala, 2019). Cycling riding on the sidewalk is generally not allowed in the city (Desjardins facilities included cycling lanes (segregated infrastructure away from et al., 2022; McArthur and Hong, 2019). The output of this process was traffic), bicycle sharrows (demarcated streets where the driveway is the shortest route between the first and the last observation of each shared with cars) and pedestrianised streets (low-speed areas where the chosen route. Regardless, it is worth noting that some chosen routes pavement is shared with both cars and pedestrians). We also evaluated covered a shorter distance than ‘openrouteservice’ inferred routes due to the average exposure to NO2 (μg/m3), daytime noise (dB) and Normal­ participants riding through streets that are not catalogued in the OSM ized Difference Vegetation Index (NDVI) of each trip, as well as the cycling network. Also, micromobility vehicles have been reported to density of the following urban elements: crosswalks, traffic lights and momentarily invade the sidewalk or ride contraflow, which might stops, streetlights, car parking places, historical points of interest, res­ explain the presence of shortcuts in our database (Lind et al., 2021; taurants, intersections, upslopes, shops, and trees. We then calculated Tuncer et al., 2020). The detour percentage, or Route Deviation Index, the difference in cycling infrastructure use, exposure and built envi­ was then computed as the ratio of chosen route distance to its shortest ronment between each pair of chosen and shortest routes. Due to the route distance (Park and Akar, 2019; Ta et al., 2016): diverse nature and magnitude of the variables, they were normalized ( ) using z-scores prior to model building. Chosen route distance Detour percentage = − 1 x 100 Shortest route distance Fig. 2. Examples of chosen and shortest paths between three origin and destination pairs. 4 J. Cubells et al. Journal of Transport Geography 111 (2023) 103664 3.4. Statistical analysis Table 2 Characteristics of chosen and shortest routes. Descriptive statistics and bivariate analysis were used to characterise Chosen route Shortest Difference Paired t-test trip attributes and the urban environment in the route vicinity. route (p-value) Furthermore, a set of multilevel linear mixed-effects models was built, mean (sd) mean (sd) (%) modelling the detour percentage. All the models included user-specific Built environment random effects to control for unobserved heterogeneity (Kang and 29.56 Fricker, 2018). Since each participant performed multiple routes, users Bicycle sharrowa 0.36 (1.25) 98.78 <0.001 (29.66) represent the highest level of the analysis. In contrast, dependent vari­ 151.10 128.02 Crosswalksb 15.27 <0.001 ables related to the built environment, pollutant exposure, transport (36.25) (25.70) Historical points of 18.66 17.769 mode and sociodemographic characteristics constitute the lowest level. 4.77 0.151 interestb (23.96) (25.58) The first model tested detour percentage as a sum of urban elements and 493.39 516.61 Intersectionsb 4.60 environmental exposure, while the second model added sociodemo­ (258.84) (278.85) <0.001 graphic variables. The third model ran the same variables as the second Upslope 15%a 1.28 (5.69) 1.32 (5.28) 3.08 0.627 model although stratified by transport mode. The sociodemographic b 3618.74 3645.14 Trees − 0.73 0.267 (1139.60) (1247.61) variable with the highest estimate that tested for statistical significance 358.16 363.05 b in the analysis of variance (type II Wald chi-square tests) was gender. Restaurants − 1.37 0.137 (182.75) (194.65) Hence, the fourth (only bicycle trips) and fifth models (only e-scooter 889.47 909.60 Streetlightsb − 2.26 <0.001 trips) were additionally stratified by gender. To facilitate the interpre­ (275.81) (299.80) 1195.51 1230.31 tation of the results, marginal effects were calculated and plotted for Shopsb − 2.91 <0.001 (549.55) (587.79) those variables that tested for statistical significance in the model Stops and traffic 4213.60 4377.91 − 3.90 analysis of variance. lightsb (1567.95) (1719.91) <0.001 Regardless the survey handled out to participants offered multiple Car parking places b 1990.58 2087.96 − 4.89 0.004 options to reflect their demographic diversity, particularly in terms of (1513.74) (1653.84) 30.35 33.46 gender identity (woman, man, non-binary), all the participants self- Cycling lanea (35.98) (39.55) − 10.25 0.002 identified as either women or men, and therefore the results struc­ Pedestrianised 31.49 36.83 − 16.96 tured gender following the two reported categories. <0.001 streeta (38.16) (42.21) 4. Results Environmental exposure NO2 (μg/m3) 32.86 (4.57) 32.08 (4.68) 2.37 <0.001 4.1. Trip characteristics Noise (dB) 42.00 (4.97) 41.19 (5.08) 1.93 <0.001 NDVI 0.02 (0.05) 0.02 (0.05) 0,00 0.452 a b Bike-share cyclists and e-scooter users rode for an average of 2.32 km Percentage of route conducted within; Density expressed in units/km2. (sd = 2.15) (Table 1). That was 1.16 km (sd = 1.65) longer than the average shortest path. A comparison between transport modes illus­ 4.2. Built environment and exposure trates that bike-share cyclists travelled a slightly greater distance, 2.48 km (sd = 1.97), than e-scooter riders, with 2.19 km (sd = 2.31). These The results of the first model show how the built environment, bicycle trips surpassed the shortest route by 1.06 km (sd = 1.44), while pollutant exposure and transport mode were associated with the detour e-scooter rides deviated an average of 1.25 km (sd = 1.83). Most trips percentage. The significance and meaningfulness of each of these urban included some sort of deviation since 87.93% of them increased the elements to influence detour percentage are captured in the coefficients shortest trip distance by at least 5%. Nevertheless, the detour percentage of Model 1 (Table 3). According to the model, crosswalks are the urban varied significantly between modes: half of the cyclists detoured at least element that explains detour the most. Hence, micromobility users seem 45.69% from the shortest path whereas half of the e-scooter riders, at to perform greater deviations to run into areas with a high density of least 71.80%. On the contrary, 7.61% of the trips included a negative crosswalks. On the contrary, pedestrianised streets, bicycle sharrows, detour. In this sense, some riders were able to shortcut and almost halve stops and traffic lights have the opposite effect as participants tend to the distance of the inferred shortest route (Fig. 2). dodge them. In terms of exposure, chosen paths were significantly more The routes participants followed presented different urban elements exposed traffic-related (nitrogen dioxide) and acoustic pollution. The than the estimated shortest paths (Table 2). Micromobility users seem to statistical analysis further illustrated that detour percentage varied be consciously choosing to travel through streets with a significantly significantly by transport mode, with e-scooters being associated with higher density of crosswalks, intersections, and bicycle sharrows, while higher detour percentages. avoiding stops, traffic lights and parked cars. Incidentally, cyclists and riders are also dismissing pedestrianised areas and streets with lots of 4.3. Sociodemographic factors and micromobility mode shops and streetlights. In the same vein, chosen routes hold higher NO2 and noise exposure levels than the shortest path. In comparison with the first model, the second model included sociodemographic variables. This model improved the estimation of the first one (p-value = 0.05) and revealed that sociodemographic attributes Table 1 Route distances and detour percentage. Chosen route (km) Shortest route (km) Detour (km) Detour (%) Mean sd Median Mean sd Median Mean sd Median Mean sd Median Bicycle 2.48 1.97 2.01 1.42* 1.11 1.19 1.06 1.44 0.63 117.82* 167.24 45.69 E-scooter 2.19 2.31 1.45 0.94* 0.81 0.63 1.25 1.83 0.57 161.14* 196.43 71.80 Total 2.32 2.15 1.67 1.17 1 0.83 1.16 1.65 0.61 139.98 183.91 54.65 * Significant p-values (<0.05) in the ANOVA test between transport modes. 5 J. Cubells et al. Journal of Transport Geography 111 (2023) 103664 Table 3 Detour percentage according to the built environment, exposure, transport mode and sociodemographic variables. Model 1: Model 2: Model 3: Built environment and Built environment, environmental Built environment, environmental Interaction term: Transport environmental exposure exposure and sociodemographic exposure and sociodemographic mode variables variables Fixed effects Coeffa CI 95%a Coeff CI 95% Coeff CI 95% Coeff CI 95% Intercept 126.8 [106.1–147.5] 127.4 [96.8–157.9] 122.5 [85.6–159.4] – – Built environment Crosswalks 40.0** [25.1–55] 41.6** [26.6–56.5] 36.2** [15.8–56.7] 9.9 [− 20.9–40.7] Streetlights 2.1 [− 13.2–17.5] 2.0 [− 13.3–17.3] − 16.7 [− 37.6–4.1] 40.5* [9.1–71.9] Car parking places 9.24 [− 4.8–23.3] 8.2 [− 5.9–22.3] 27.9 [5.2–50.5] − 30.6* [− 59.6–1.5] Historical points of interest 4.24 [− 10.5–19] 0.9 [− 14–15.8] − 6.4 [− 26.6–13.8] 14.5 [− 15.9–44.9] Restaurants − 6.0 [− 21.6–9.7] − 7.4 [− 23.1–8.3] − 9.0 [− 31.2–13.3] 10.7 [− 21–42.4] Stops and traffic lights − 31.3** [− 47.6–15.1] − 30.5** [− 46.6–14.3] − 29.2** [− 53.3–5.2] − 5.7 [− 38.9–27.5] Shops 7.89 [− 7.9–23.6] 7.1 [− 8.6–22.9] 18.0 [− 5.3–41.2] − 17.1 [− 49.3–15.2] Trees − 24.6** [− 39–10.3] − 24.2** [− 38.7–9.9] − 13.8** [− 34.5–6.8] − 19.6 [− 48.6–9.3] Cycling lane 7.97 [− 6.4–22.3] 6.5 [− 7.8–20.9] − 4.6 [− 22.5–13.3] 50.3* [18.6–82] Pedestrianised street − 38.2** [− 53.7–22.7] − 37.4** [− 52.9–22] − 38.9** [− 60.3–17.5] − 3.0 [− 34.5–28.4] Bicycle sharrow − 32.3** [− 46.1–18.4] − 33.1** [− 46.8–19.4] − 36.2** [− 56.1–16.4] − 5.1 [− 23.5–33.8] Intersections 10.6 [− 4.3–25.6] 9.8 [3.5–62.7] 10.5 [− 12–32.9] − 1.2 [− 32–29.5] Upslope 15% 0.54 [− 14.1–15.2] 0.3 [− 5.1–24.7] − 1.3 [− 27.2–24.7] 5.3 [− 26.5–37] Environmental exposure NO2 29.68** [11.8–47.5] 30.8** [12.9–48.7] 29.9** [4.8–55] 3.2 [− 32.6–39] Noise 32.5** [13.9–51.1] 31.8** [13.1–50.4] 31.0** [5.3–56.7] − 0.9 [− 38.5–36.8] NDVI 5.7 [− 7.7–19.2] 7.3 [− 6–20.7] 6.9 [− 13.8–27.6] − 1.4 [− 29–26.1] Transport mode Transport mode Bicycle =ref =ref =ref =ref E-scooter 30.2* [− 0.4–60.8] 33.1* [− 14.4–14.9] 23.2* [− 35.4–81.8] – – Sociodemographic variables Gender Woman – – =ref =ref =ref Man – – 31.5* [2.3–60.7] 20.2* [− 22.5–63] 25.7 [− 36–87.3] Parent of a minor child No – – =ref =ref =ref Yes – – − 26.8 [− 59.8–6.2] − 23.9 [− 71.9–24] 18.5 [− 51.4–88.4] Age 16–24 – – − 17.3 [− 58.3–23.7] − 18.3 [− 73.8–37.2] 21.9 [− 63–106.8] 25–34 – – =ref =ref =ref 35–44 – – − 19.0 [− 57.9–19.9] 12.1 [− 41.3–65.5] − 50.4 [− 131.1–30.2] >45 – – − 12.9 [− 54.5–28.6] 4.2 [− 55.4–63.9] − 18.6 [− 105.3–68] Random effects User-specific variance (sd) 916 (30.27) 506.8 (22.5) 572.9 (23.9) Residual variance (sd) 24,325 (155.97) 24,436.9 (156.32) 23,628.2 (153.7) Model parameters AIC 7155.2 7154.3 7156.1 BIC 7241.4 7262.1 7354.4 p-value b =ref 0.05 0.007 a Abbreviations: coefficient and confidence interval. * Significant p-values (<0.05) in the analysis of variance (type II Wald chi-square tests). ** Significant p-values (<0.001) in the analysis of variance (type II Wald chi-square tests). also influenced the detour percentage, together with the previously accounted for more streetlights and fewer car parking places, while the modelled urban elements, environmental exposure and transport mode opposite was observed for bicycle rides. (Table 3). From the second model, it is apparent that men performed Models 1 to 3, although adjusted by transport mode, included both significantly longer detours than women. The other sociodemographic bicycle and e-scooter trips. In the next set of models, the two vehicles variables, which were age group and being a parent, did not influence were split into different data sets and analysed separately. The variables detour percentage significantly. of the following pair of models were additionally intersected with Since detour percentage varies significantly and greatly by transport gender. In the only-bicycles model (Model 4) it is apparent that cyclists mode (an average of 43,32% detour difference between bikes and e- detoured longer distances seeking to use paths with numerous cross­ scooters), the third model intersected the transport mode with all the walks (Table 4). For e-scooters (Model 5), greater deviations from the previously run variables. As result, Model 3 statistically outperformed shortest path were associated with selecting routes that run through the fitness of the first model (p-value = 0.007). This last model tested dedicated cycling lanes and with a higher density of crosswalks. Both whether the built environment, exposure and sociodemographic char­ cyclists and e-scooter riders seemed to avoid routes with abundancy of acteristics explained cyclists’ and e-scooter riders’ deviation from the stops, traffic lights, pedestrianised streets and bicycle sharrows, as shortest path differently. Indeed, the third model presents that, for e- illustrated in Fig.3. Results also suggest that both cyclists and e-scooter scooter trips, the chosen path had a greater share of cycling lanes and riders would have been less exposed to traffic-related pollutants if they 6 J. Cubells et al. Journal of Transport Geography 111 (2023) 103664 Table 4 Detour percentage according to the built environment, exposure, transport mode and sociodemographic variables, stratified by transport mode. Model 4: Bicycles Model 5: E-scooters Interaction term: Gender Interaction term: Gender Fixed effects Coeffa CI 95%a Coeff CI 95% Coeff CI 95% Coeff CI 95% Intercept 106.0 [70.3–141.7] – – 86.7 [22.8–150.6] – – Built environment Crosswalks 24.8** [− 0.1–49.7] 21.4 [− 15.8–58.6] 43.1** [6.6–79.5] − 1.0 [− 53–51.1] Streetlights − 10.5 [− 36.8–15.8] − 10.0 [− 48–27.9] − 4.2 [− 41.4–32.9] 36.9 [− 17.5–91.3] Car parking places 7.7 [− 22.9–38.4] 37.4 [− 2.6–77.4] 29.9 [1.5–58.3] − 51.0* [− 92.2–9.9] Historical points of interest − 6.6 [− 36–22.9] 4.0 [− 33.6–41.5] 32.3 [− 3.7–68.3] − 43.3 [− 94.2–7.5] Restaurants − 8.3 [− 42.7–26] − 5.0 [− 47.7–37.7] − 5.9 [− 37.5–25.8] 21.9 [− 32.4–76.1] Stops and traffic lights − 47.9* [− 80.8–15.1] 26.0 [− 17.6–69.5] − 37.1* [− 73.3–0.8] 11.4 [− 39.8–62.6] Shops 4.3 [− 29–37.7] 19.0 [− 23.3–61.4] − 20.1 [− 56.2–15.9] 25.2 [− 27.3–77.7] Trees − 20.8 [− 46.4–4.9] 4.9 [− 34.4–44.1] − 34.4** [− 68–0.7] 5.4 [− 41.7–52.6] Cycling lane − 17.4 [− 39.6–4.7] 34.0* [1.9–66.2] 56.1* [15.5–96.8] − 38.8 [− 97.3–19.7] Pedestrianised street − 31.4** [− 64.6–1.8] − 16.5 [− 57.7–24.6] − 70.9** [− 104.6–37.3] 77.0** [22.0–132.0] Bicycle sharrow − 24.9** [− 48.1–1.8] − 23.1 [− 58.9–12.8] − 28.9** [− 63–5.2] − 7.0 [− 53.7–39.8] Intersections 18.3 [− 10.7–47.3] − 8.5 [− 48.6–31.5] 4.9 [− 27.9–37.8] 9.9 [− 39–58.7] Upslope 15% 6.2 [− 23.2–35.6] − 14.3 [− 63.6–34.9] − 7.9 [− 74.3–58.6] 22.1 [− 48.3–92.6] Environmental exposure NO2 64.9* [36–93.9] − 76.4* [− 120.8–32.2] 11.0 [− 28.9–50.9] 41.1 [− 20.8–102.9] Noise 12.1* [− 19.2–43.5] 46.4* [0.9–92] 45.4* [0.7–90.3] − 6.4 [− 71.2–58.3] NDVI 5.7 [− 19.9–31.2] 16.3 [− 21.1–53.8] 21.4 [− 5.3–48.1] − 28.8 [− 73.5–15.9] Sociodemographic variables Gender Woman =ref =ref =ref =ref Man 59.4 [4.3–114.5] – – 156.3 [64.2–248.4] – – Parent of a minor child No =ref =ref =ref =ref Yes − 6.5 [− 62.3–49.3] 22.4 [− 66.2–111.1] − 51.3 [− 146.1–43.5] 28.0 [− 101.5–157.6] Age 16–24 26.9 [− 43.8–97.5] − 120.8 [− 221.7–19.9] 116.1 [17.5–214.6] − 232.1* [− 386–78.2] 25–34 =ref =ref =ref =ref 35–44 21.3 [− 39.3–82] − 46.4 [− 135.9–43.2] 84.5 [− 35.4–204.5] − 167.1* [− 318.7–15.4] >45 − 12.0 [− 99.3–75.3] 8.5 [− 100.7–117.6] 64.2 [− 47.3–175.7] − 149.4* [− 309.4–10.5] Random effects User-specific variance (sd) 96.6 (9.8) 1260 (35.5) Residual variance (sd) 17,179.9 (131.1) 27,213 (164.96) a Abbreviations: coefficient and confidence interval. * Significant p-values (<0.05) in the analysis of variance (type II Wald chi-square tests). ** Significant p-values (<0.001) in the analysis of variance (type II Wald chi-square tests). had followed OSM route recommendations. 5. Discussion Results indicate that gender greatly explains detour, as it accounted for the highest coefficient in Model 2 among sociodemographic vari­ 5.1. Detour ables (Table 3). In addition, Model 3 further revealed that women took shorter detours than men regardless of the micromobility vehicle used. Our results indicate that only a minority of micromobility trips In Models 4 and 5, the variable gender was additionally intersected with actually used the shortest path available. In fact, almost 9 of every 10 route attributes to evaluate whether men and women favoured them trips deviated from the shortest route to some extent, which is consistent differently (Table 4). This analysis highlighted that bike-share cyclists with previous research (Park and Akar, 2019). However, micromobility seem to equally value the elements of the surrounding built environ­ users in Barcelona seem to have picked circuitous routes that, on ment, except for cycling lanes (Fig. 4). Results show that men cyclists average, surpass the detour percentage of other study cases (Chen et al., appear to pedal further to incorporate cycling lanes into their routes. In 2018; Fitch and Handy, 2020; Ghanayim and Bekhor, 2018; Lu et al., contrast, there are gender differences in the way e-scooter riders inte­ 2018). These differences might be indicative of a more fragmented grate specific route attributes in their rides. Compared to women, men network of dedicated infrastructure. Nonetheless, our sample being riding e-scooters deviate further from the shortest route to avoid car composed of both e-scooter and bike-sharing trips might also be skewing parking places while seeking to go across pedestrianised streets. our results towards longer and more frequent detours, as e-scooter trips Furthermore, detour percentages were significantly associated with the have been estimated to account for longer deviations than shared bi­ age of the e-scooter riders. The cohort that encapsulated 25 to 34-year- cycle. The reasons explaining e-scooters’ greater detour are threefold. old participants was the age group with the greatest gender gap in terms First, riding an e-scooter is not as physically demanding as it is to ride a of detour percentage, in which men travelled further from the shortest bicycle (López-Dóriga et al., 2022). Therefore, outstretching an e- path than women (Table 4). scooter trip does not result in a physical payoff. Second, e-scooter use has often been regarded to support riding for leisure, tourism or for fun sake (McKenzie, 2019; Roig-Costa et al., 2021; Weschke et al., 2022), which would side with not selecting the most direct route. Thirdly, variations in vehicle ownership may contribute to the observed differ­ ences in detouring behaviour between e-scooters and shared bicycles. In 7 J. Cubells et al. Journal of Transport Geography 111 (2023) 103664 Fig. 3. Route attributes and detour trade-off calculated using Models 3–5 marginal effects. Fig. 4. Route attributes of e-scooter trips and detour trade-off by gender calculated using Model 5 marginal effects. Barcelona, e-scooters are exclusively privately owned, while shared bi­ 5.2. Built environment and environmental exposure cycles operate under a fee structure that imposes charges on users exceeding a 30-min ride duration (Bustamante et al., 2022). This pricing Our analysis confirms previous studies finding that built environ­ mechanism could potentially be incentivizing bike-share users to pri­ ment conditions effectively impact wayfinding and route choices (Zim­ oritize shorter, more direct routes, in contrast to e-scooters. In addition mermann et al., 2017). In this regard, previous work has noticed the to detour, our analysis also revealed the presence of shortcuts. Shortcuts importance of dedicated cycling infrastructure to explain detour (Lu can represent wrong-way riding but also participants using infrastruc­ et al., 2018; Wang et al., 2022). As shown in our results, e-scooter riders ture that is not catalogued as cyclable by OSM (Dhakal et al., 2018). are willing to deviate significantly from the shortest path to travel Both e-scooters and shared bikes are light vehicles that make it possible through bicycle lanes. This behaviour might be related to the feeling of to engage in a series of small infractions such as temporarily invading insecurity when sharing space with cars, or the feeling of being a the sidewalk or riding contraflow, which might explain the presence of nuisance for pedestrians when riding through the sidewalk (Gibson shortcuts in our database (Lind et al., 2021; Tuncer et al., 2020). et al., 2022). Our findings corroborate those of Zhang et al. (2021) and Zuniga-Garcia et al. (2021) on e-scooter riders’ preference for cycling 8 J. Cubells et al. Journal of Transport Geography 111 (2023) 103664 lanes. However, neither bike-share cyclists nor e-scooter riders seem to 5.3. Sociodemographic factors detour to favour routes with a high proportion of bicycle sharrows or pedestrianised streets. For the case of bicycle sharrows, our results Drawing on the results of state preference surveys, several authors contradict earlier research (Park and Akar, 2019; Rupi and Schweizer, have hypothesised how research using spatial data could largely benefit 2018; Yang et al., 2022) although this might be due to OSM route en­ from including sociodemographic variables known to influence route gines tending not to recommend routes with bike sharrows (Table 2). choice (Hardinghaus and Nieland, 2021; Lu et al., 2018; Park and Akar, Avoiding pedestrianised areas, on the other hand, is in accordance with 2019; Scott et al., 2021). Of all the sociodemographic factors included in prior findings (Lee and Sener, 2019; Skov-Petersen et al., 2018). These our study, neither the age of participants nor their parental status authors argue that micromobility users avoid sharing the space with seemed to significantly influence route choice. These findings build on pedestrians because it slows down their pace. Aldred et al. (2017) conclusions that route choice might not stem Similarly, cyclists and e-scooter riders tend to plan their routing to directly from participants’ age, but rather from other factors such as skirt stops and traffic lights, as evidenced in our results and the literature previous cycling experience or familiarity with the cycling infrastructure (Prato et al., 2018). One could expect this behaviour to be a consequence network. Additionally, although neither shared bikes nor private e- of trying to maximize travel speed and flow (Cubells et al., 2023). scooters support travelling with an extra passenger, we could not Notwithstanding, this narrative is countered by the fact that micro­ exclude the possibility that parents might have been riding with children mobility users are also found to be selecting routes that have a higher by their side. In any case, elements of the built environment did not seem density of crosswalks. Previous research has also documented micro­ to shape parents’ route choice when, most likely, riding solo. mobility users’ preference for spaces with clear formal traffic rules that In contrast, our results reveal the importance of acknowledging increase space readability, and lower riding uncertainty (Cho and Shin, participants’ gender to understand route choice. It is apparent that 2022; Li et al., 2017). Streets with delimited spaces for different uses and women perform shorter detours than men in Barcelona, particularly a high density of designated crosses are perceived as safer by most cyclists. This is in line with earlier research that observed that women, micromobility users. This might explain micromobility users in our on average, tend to outstretch their cycling trips to a lesser extent than sample selecting routes with numerous crosswalks, which appears to men (Schneider et al., 2021). Preferring the straightest route to chain represent a safer way of interacting with pedestrians (Skov-Petersen trips might also be in accordance with time constraints. Women cyclists et al., 2018). Indeed, it is often claimed that perception of risk along with have reported feeling pressed for time, juggling care and productive risk-mitigating attitudes might be as relevant as distance in micro­ work, and this came across to curtail opportunities for bicycling (Rav­ mobility route choice (Kang and Fricker, 2018; Majumdar and Mitra, ensbergen et al., 2020b; Sersli et al., 2020). Apart from the likeliness to 2019). This is in line with our results showing that e-scooter riders detour and its extension, the findings also reflect gender differences in favour well-lit routes and avoid streets with numerous car parking pla­ route preferences. Although stated preference surveys surfaced women’s ces (Yang et al., 2022). willingness to pedal further to incorporate bike lanes into their rides, our Regarding environmental exposure, our results seem to suggest that GPS data shows that women might not be extending their trips to favour participants are not detouring to integrate urban greenness into their dedicated cycling infrastructure (Hardinghaus and Weschke, 2022). rides. While there is extensive literature on cyclists’ and e-scooter riders’ Among e-scooter riders, men’s greater deviation comes with the op­ willingness to ride further to include greenery (Bernardi et al., 2018; Lin portunity to elude parked cars while favouring pedestrianised streets. and Fan, 2020; Marquart et al., 2020; Yang et al., 2022), our results offer This finding is contrary to previous studies examining bicycle route significant nuance on a compact-city spatial context with low levels of choice. In those, women pedalled further to run into calm streetscapes, green areas but a homogeneous high tree coverage (Baró et al., 2019). while our results point to the opposite direction among e-scooter riders Thus, our findings point out that users are not choosing to detour in (Hardinghaus and Weschke, 2022). To the best of our knowledge, this is order to encounter higher tree density. Yet, other life-forms of vegeta­ the first time that gender route-preference differences among e-scooter tion, such as shrubs and grasses, or even bare soil were not accounted for users are considered. in the tree database and were only captured when analysing land coverage through NDVI. 6. Conclusions In terms of land use, results are not conclusive on whether micro­ mobility users integrate or avoid shopping while riding (Desjardins This study set out to comprehend how elements of the built envi­ et al., 2022; Orellana and Guerrero, 2019; Schneider et al., 2021; Soltani ronment influenced the trip route choice and detours of 115 bike-share et al., 2022). Neither restaurants nor historical points of interests have a and e-scooter users while acknowledging sociodemographic factors. clear positive or negative effect on route choice, as noticed by Sarjala Micromobility riders in Barcelona did not always choose the shortest (2019). Remarkably, Desjardins et al. (2022) and Fitch and Handy path since the presence of cycling infrastructure and urban elements (2020) reported that the presence of restaurants and historical points of together with personal preferences shaped their route choice. Hence, interests repelled micromobility users. Overall, this constitutes one of participants seemed to choose to extend their trips in order to select the first attempts to understand how e-scooter riders integrate both routes that had a higher density of crosswalks and bicycle sharrows, shopping and green elements into their rides (Bai and Jiao, 2020; Yang while avoiding stops, traffic lights and parked cars. Incidentally, riders et al., 2022). also avoided pedestrianised areas and streets with lots of shops and Results also reveal that wayfinding strategies of micromobility users streetlights. The results also revealed that cyclists and e-scooter riders might inevitably expose them to higher levels of traffic-related (NO2) shared preferences for certain elements of the built environment (such and acoustic pollution than those of shortest routes. This clearly con­ as crosswalks and stops), whereas they placed different value on others trasts with findings from previous studies that have used stated prefer­ (such as cycling lanes, streetlights, and car parking places). In addition, ence surveys and have observed a self-reported disposition to select gender appeared to influence participants’ willingness to detour as well routes with less traffic and thus less pollution exposure (Desjardins et al., as route choice preferences. These gender differences were more salient 2021a; Gössling et al., 2019). Our objective GPS tracking data finds that among private e-scooter users than among bike-share members. micromobility users are still highly exposed to pollution and noise, as This research is not without limitations. The generalisability of the noticed elsewhere (Lee and Sener, 2019; Willberg et al., 2023; Wu et al., results is subject to the urban environment and mobility schemes of 2021). Barcelona. We note that further investigation is required to assess the applicability of the findings elsewhere and to include other transport modes, such as private bicycles and shared e-scooters, as well as to ac­ count for the economic cost and gendered spatial occurrence of trips. 9 J. Cubells et al. Journal of Transport Geography 111 (2023) 103664 Moreover, since participants were asked to wear the GPS device all day and Research Grants (AGAUR, FI_B 00063, 2022) and O. Marquet is long, the study is limited by the accuracy of trip detection and transport funded by a Ramón y Cajal fellowship (RYC-2020- 029441-I), awarded mode inference methods. Future research will greatly benefit from ef­ by the Spanish Ministry of Science and Innovation. forts put into improving the GPS processing of micromobility data (Berjisian and Bigazzi, 2022; Lißner and Huber, 2021). While the sample References size is not large, this work accounts for an even gender representation of men and women and hence offers valuable insights into gender differ­ Aldred, R., Elliott, B., Woodcock, J., Goodman, A., 2017. Cycling provision separated from motor traffic: a systematic review exploring whether stated preferences vary by ences in travel behaviour. Being limited to participants’ answers on gender and age. Transp. Rev. 37, 29–55. https://doi.org/10.1080/ gender identity, however, this study could not regard route choice of 01441647.2016.1200156. non-binary riders and thus, further efforts ought to be made at exploring Bai, S., Jiao, J., 2020. Dockless E-scooter usage patterns and urban built environments: a comparison study of Austin, TX, and Minneapolis. MN. Travel Behav. Soc. 20, the travel behaviour of gender non-conforming people, but also other 264–272. https://doi.org/10.1016/j.tbs.2020.04.005. social groups, such as people with racialised or LGBTQI+ experiences Baró, F., Calderón-Argelich, A., Langemeyer, J., Connolly, J.J.T., 2019. Under one (Lubitow et al., 2020, 2019; Nello-Deakin and Harms, 2019). canopy? Assessing the distributional environmental justice implications of street tree benefits in Barcelona. Environ. Sci. Pol. 102, 54–64. https://doi.org/10.1016/j. The present study is among the first to explore micromobility route envsci.2019.08.016. selection and detour while accounting for multiple micromobility modes Berjisian, E., Bigazzi, A., 2022. Evaluation of methods to distinguish trips from activities and understanding their distinct detour triggers. Consequently, we in walking and cycling GPS data. Transp. Res. Part C Emerg. Technol. 137, 103588 https://doi.org/10.1016/j.trc.2022.103588. contribute to closing the gap between cyclists’ and e-scooter riders’ Bernardi, S., Geurs, K., Puello, L.L.P., 2018. Modelling route choice of dutch cyclists different preferences for urban attributes within the same spatial context using smartphone data. J. Transp. Land Use 11, 883–900. https://doi.org/10.5198/ (McKenzie, 2019). This is an important step given the increasing need jtlu.2018.1143. for scientific evidence that can help reliably forecast micromobility Bhandal, J., Noonan, R.J., 2022. Motivations, perceptions and experiences of cycling for transport: a photovoice study. J. Transp. Health 25, 101341. https://doi.org/ flows and demand within cities. Similarly, our study is also among the 10.1016/j.jth.2022.101341. first ones to incorporate the sociodemographic characteristics of the Bieliński, T., Ważna, A., 2020. Electric scooter sharing and bike sharing user behaviour users as predictors of detour and route choice. Our results clearly and characteristics. Sustain. Switz. 12, 1–13. https://doi.org/10.3390/su12229640. Bretones, A., Marquet, O., 2022. Sociopsychological factors associated with the adoption demonstrate the gendered nature of such decisions and reiterate previ­ and usage of electric micromobility. A literature review. Transp. Policy 127, ous calls on the importance of continuing to take an intersectional 230–249. https://doi.org/10.1016/j.tranpol.2022.09.008. approach in our understanding of micromobility use. Bustamante, X., Federo, R., Fernández-i-Marin, X., 2022. Riding the wave: predicting the use of the bike-sharing system in Barcelona before and during COVID-19. Sustain. Cities Soc. 83, 103929. https://doi.org/10.1016/j.scs.2022.103929. CRediT authorship contribution statement Carlson, J.A., Jankowska, M.M., Meseck, K., Godbole, S., Natarajan, L., Raab, F., Demchak, B., Patrick, K., Kerr, J., 2015. Validity of PALMS GPS scoring of active and passive travel compared with SenseCam. Med. Sci. Sports Exerc. 47, 662–667. Jerònia Cubells: Conceptualization, Data curation, Formal analysis, https://doi.org/10.1249/MSS.0000000000000446. Methodology, Software, Validation, Visualization, Writing – original Caspi, O., Smart, M.J., Noland, R.B., 2020. Spatial associations of dockless shared e- draft. Carme Miralles-Guasch: Resources, Supervision, Project scooter usage. Transp. Res. Part Transp. Environ. 86, 102396 https://doi.org/ 10.1016/j.trd.2020.102396. administration. Oriol Marquet: Conceptualization, Methodology, Chen, P., Shen, Q., Childress, S., 2018. A GPS data-based analysis of built environment Validation, Resources, Writing – review & editing, Supervision, Project influences on bicyclist route preferences. Int. J. Sustain. Transp. 12, 218–231. administration. https://doi.org/10.1080/15568318.2017.1349222. Cho, S.-H., Shin, D., 2022. Estimation of route choice behaviors of bike-sharing users as first- and last-mile trips for introduction of mobility-as-a-service (MaaS). KSCE J. Declaration of Competing Interest Civ. Eng. 26, 3102–3113. https://doi.org/10.1007/s12205-022-0802-1. Clarry, A., Faghih Imani, A., Miller, E.J., 2019. Where we ride faster? Examining cycling None. speed using smartphone GPS data. Sustain. Cities Soc. 49, 101594 https://doi.org/ 10.1016/j.scs.2019.101594. Codina, O., Maciejewska, M., Nadal, J., Marquet, O., 2022. Built environment bikeability Data availability as a predictor of cycling frequency: lessons from Barcelona. Transp. Res. Interdiscip. Perspect. 16, 100725 https://doi.org/10.1016/j.trip.2022.100725. Cole-Hunter, T., Donaire-Gonzalez, D., Curto, A., Ambros, A., Valentin, A., Garcia- The authors do not have permission to share data. Aymerich, J., Martínez, D., Braun, L.M., Mendez, M., Jerrett, M., Rodriguez, D., de Nazelle, A., Nieuwenhuijsen, M., 2015. Objective correlates and determinants of Acknowledgements bicycle commuting propensity in an urban environment. Transp. Res. Part Transp. Environ. 40, 132–143. https://doi.org/10.1016/j.trd.2015.07.004. Cubells, J., Miralles-Guasch, C., Marquet, O., 2023. Gendered travel behaviour in The authors would like to thank the participants who shared their micromobility? Travel speed and route choice through the lens of intersecting time for this project. We wish to express our gratitude to Oriol Roig- identities. J. Transp. Geogr. 106, 103502 https://doi.org/10.1016/j. jtrangeo.2022.103502. Costa for conducting the data-gathering process. Our research col­ Ajuntament de Barcelona, 2022a. Open Data BCN [WWW Document]. Serv. Dades leagues Zeynep Akinci, Laia Mojica, and Guillem Vich have also shared Obertes. URL https://opendata-ajuntament.barcelona.cat/ (accessed 9.20.22). valuable comments during the design and development of this research. Ajuntament de Barcelona, 2022b. Environmental data maps [WWW Document]. Ecol. Urban Plan. Infrastruct. Mobil. URL https://ajuntament.barcelona.cat/ecologiaurba Furthermore, the authors are grateful to the anonymous reviewers and na/en/environmental-data-maps (accessed 12.2.22). the editor for providing insightful feedback that helped to improve the Desjardins, E., Apatu, E., Razavi, S.D., Higgins, C.D., Scott, D.M., Páez, A., 2021a. “Going manuscript. This work was supported by the Spanish Ministry of Science through a little bit of growing pains”: a qualitative study of the factors that influence and Innovation (ECOMOV project, TED2021-129280B-I00 and the route choice of regular bicyclists in a developing cycling city. Transp. Res. Part F Traffic Psychol. Behav. 81, 431–444. https://doi.org/10.1016/j.trf.2021.06.005. MICROMOV project, PID2019-104344RB-I00); by the Social Observa­ Desjardins, E., Higgins, C.D., Scott, D.M., Apatu, E., Páez, A., 2021b. Using tory of the ”la Caixa” Foundation as part of the project: STEPP SR22- environmental audits and photo-journeys to compare objective attributes and 00147. J. Cubells is supported by a PhD grant by the Agency for Man­ bicyclists’ perceptions of bicycle routes. J. Transp. Health 22, 101092. https://doi. org/10.1016/j.jth.2021.101092. agement of University and Research Grants (AGAUR, FI_B 00063, 2022) Desjardins, E., Higgins, C.D., Scott, D.M., Apatu, E., Páez, A., 2022. Correlates of and O. Marquet is funded by a Ramón y Cajal fellowship (RYC-2020- bicycling trip flows in Hamilton, Ontario: fastest, quietest, or balanced routes? 029441-I), awarded by the Spanish Ministry of Science and Innovation. Transportation 49, 867–895. https://doi.org/10.1007/s11116-021-10197-1. Dhakal, N., Cherry, C.R., Ling, Z., Azad, M., 2018. Using CyclePhilly data to assess This work was supported by the Spanish Ministry of Science and Inno­ wrong-way riding of cyclists in Philadelphia. J. Saf. Res. 67, 145–153. https://doi. vation (ECOMOV project, TED2021-129280B-I00 and MICROMOV org/10.1016/j.jsr.2018.10.004. project, PID2019-104344RB-I00); by the Social Observatory of the ”la Felipe-Falgas, P., Madrid-Lopez, C., Marquet, O., 2022. Assessing environmental performance of micromobility using LCA and self-reported modal change: the case of Caixa” Foundation as part of the project: STEPP SR22-00147. J. Cubells is supported by a PhD grant by the Agency for Management of University 10 J. Cubells et al. Journal of Transport Geography 111 (2023) 103664 shared E-bikes, E-scooters, and E-mopeds in Barcelona. Sustainability 14, 4139. Marquet, O., Miralles-Guasch, C., 2018. Resilient territories and mobility adaptation https://doi.org/10.3390/su14074139. strategies in times of economic recession: evidence from the metropolitan region of Fitch, D.T., Handy, S.L., 2020. Road environments and bicyclist route choice: the cases of Barcelona, Spain 2004–2012. Eur. Urban Reg. Stud. 25, 345–359. https://doi.org/ Davis and San Francisco, CA. J. Transp. Geogr. 85, 102705 https://doi.org/10.1016/ 10.1177/0969776417703158. j.jtrangeo.2020.102705. McArthur, D.P., Hong, J., 2019. Visualising where commuting cyclists travel using Flügel, S., Hulleberg, N., Fyhri, A., Weber, C., Ævarsson, G., 2019. Empirical speed crowdsourced data. J. Transp. Geogr. 74, 233–241. https://doi.org/10.1016/J. models for cycling in the Oslo road network. Transportation 46, 1395–1419. https:// JTRANGEO.2018.11.018. doi.org/10.1007/s11116-017-9841-8. McKenzie, G., 2019. Spatiotemporal comparative analysis of scooter-share and bike- Ghanayim, M., Bekhor, S., 2018. Modelling bicycle route choice using data from a GPS- share usage patterns in Washington, D.C. J. Transp. Geogr. 78, 19–28. https://doi. assisted household survey. Eur. J. Transp. Infrastruct. Res. 18 https://doi.org/ org/10.1016/j.jtrangeo.2019.05.007. 10.18757/ejtir.2018.18.2.3228. Nawrath, M., Kowarik, I., Fischer, L.K., 2019. The influence of green streets on cycling Gibson, H., Curl, A., Thompson, L., 2022. Blurred boundaries: E-scooter riders’ and behavior in European cities. Landsc. Urban Plan. 190, 103598 https://doi.org/ pedestrians’ experiences of sharing space. Mobilities 17, 69–84. https://doi.org/ 10.1016/j.landurbplan.2019.103598. 10.1080/17450101.2021.1967097. Nello-Deakin, S., Harms, L., 2019. Assessing the relationship between neighbourhood Gössling, S., Humpe, A., Litman, T., Metzler, D., 2019. Effects of perceived traffic risks, characteristics and cycling: findings from Amsterdam. Transp. Res. Proc. 41, 17–36. noise, and exhaust smells on bicyclist behaviour: an economic evaluation. https://doi.org/10.1016/j.trpro.2019.09.005. Sustainability 11, 408. https://doi.org/10.3390/su11020408. Oleś, A., 2022. openrouteservice: Openrouteservice API Client. Hardinghaus, M., Nieland, S., 2021. Assessing cyclists’ routing preferences by analyzing Orellana, D., Guerrero, M.L., 2019. Exploring the influence of road network structure on extensive user setting data from a bike-routing engine. Eur. Transp. Res. Rev. 13 the spatial behaviour of cyclists using crowdsourced data. Environ. Plan. B Urban https://doi.org/10.1186/s12544-021-00499-x. Anal. City Sci. 46, 1314–1330. https://doi.org/10.1177/2399808319863810. Hardinghaus, M., Weschke, J., 2022. Attractive infrastructure for everyone? Different Park, Y., Akar, G., 2019. Why do bicyclists take detours? A multilevel regression model preferences for route characteristics among cyclists. Transp. Res. Part Transp. using smartphone GPS data. J. Transp. Geogr. 74, 191–200. https://doi.org/ Environ. 111, 103465 https://doi.org/10.1016/j.trd.2022.103465. 10.1016/j.jtrangeo.2018.11.013. Haworth, N., Schramm, A., Twisk, D., 2021. Changes in shared and private e-scooter use Pellicer-Chenoll, M., Pans, M., Seifert, R., López-Cañada, E., García-Massó, X., Devís- in Brisbane, Australia and their safety implications. Accid. Anal. Prev. 163, 106451 Devís, J., González, L.-M., 2021. Gender differences in bicycle sharing system usage https://doi.org/10.1016/j.aap.2021.106451. in the city of Valencia. Sustain. Cities Soc. 65, 102556 https://doi.org/10.1016/j. Heim LaFrombois, M.E., 2019. (re)producing and challenging gender in and through scs.2020.102556. urban space: women bicyclists’ experiences in Chicago. Gend. Place Cult. 26, Prato, C.G., Halldórsdóttir, K., Nielsen, O.A., 2018. Evaluation of land-use and transport 659–679. https://doi.org/10.1080/0966369X.2018.1555142. network effects on cyclists’ route choices in the Copenhagen region in value-of- IERMB, 2021. Enquesta Mobilitat en dia Feiner (EMEF) [WWW Document]. Enquestes distance space. Int. J. Sustain. Transp. 12, 770–781. https://doi.org/10.1080/ Mobilitat. URL https://iermb.uab.cat/ca/enquestes/enquestes-de-mobilitat/#144 15568318.2018.1437236. 7843451840-2-0 (accessed 4.5.22). Pritchard, R., Bucher, D., Frøyen, Y., 2019. Does new bicycle infrastructure result in new Janke, J., Handy, S., 2019. How life course events trigger changes in bicycling attitudes or rerouted bicyclists? A longitudinal GPS study in Oslo. J. Transp. Geogr. 77, and behavior: insights into causality. Travel Behav. Soc. 16, 31–41. https://doi.org/ 113–125. https://doi.org/10.1016/j.jtrangeo.2019.05.005. 10.1016/j.tbs.2019.03.004. Pucher, J., Buehler, R., 2017. Cycling towards a more sustainable transport future. Kang, L., Fricker, J.D., 2018. Bicycle-route choice model incorporating distance and Transp. Rev. 37, 689–694. https://doi.org/10.1080/01441647.2017.1340234. perceived risk. J. Urban Plan. Dev. 144, 04018041. https://doi.org/10.1061/(asce) Ravensbergen, L., 2022. ‘I wouldn’t take the risk of the attention, you know? Just a lone up.1943-5444.0000485. girl biking’: examining the gendered and classed embodied experiences of cycling. Kang, M., Moudon, A.V., Hurvitz, P.M., Saelens, B.E., 2018. Capturing fine-scale travel Soc. Cult. Geogr. 23, 678–696. https://doi.org/10.1080/14649365.2020.1806344. behaviors: a comparative analysis between personal activity location measurement Ravensbergen, L., Buliung, R., Laliberté, N., 2020a. Fear of cycling: social, spatial, and system (PALMS) and travel diary. Int. J. Health Geogr. 17, 40. https://doi.org/ temporal dimensions. J. Transp. Geogr. 87, 102813 https://doi.org/10.1016/j. 10.1186/s12942-018-0161-9. jtrangeo.2020.102813. Klein, S., Brondeel, R., Chaix, B., Klein, O., Thierry, B., Kestens, Y., Gerber, P., Ravensbergen, L., Buliung, R., Sersli, S., 2020b. Vélomobilities of care in a low-cycling Perchoux, C., 2021. What triggers selective daily mobility among older adults? A city. Transp. Res. Part Policy Pract. 134, 336–347. https://doi.org/10.1016/j. study comparing trip and environmental characteristics between observed path and tra.2020.02.014. shortest path. Health Place 102730. https://doi.org/10.1016/j. Roig-Costa, O., Gómez-Varo, I., Cubells, J., Marquet, O., 2021. La movilidad post healthplace.2021.102730. pandemia: perfiles y usos de la micromovilidad en Barcelona. Rev. Transp. Territ. Lee, K., Sener, I., 2019. Understanding potential exposure of bicyclists on roadways to https://doi.org/10.34096/rtt.i25.10958. traffic-related air pollution: findings from El Paso, Texas, using Strava metro data. Rupi, F., Schweizer, J., 2018. Evaluating cyclist patterns using GPS data from Int. J. Environ. Res. Public Health 16, 371. https://doi.org/10.3390/ smartphones; Evaluating cyclist patterns using GPS data from smartphones. https:// ijerph16030371. doi.org/10.1049/iet-its.2017.0285. Li, S., Muresan, M., Fu, L., 2017. Cycling in Toronto, Ontario, Canada: route choice Rupi, F., Poliziani, C., Schweizer, J., 2019. Data-driven bicycle network analysis based on behavior and implications for infrastructure planning. Transp. Res. Rec. J. Transp. traditional counting methods and GPS traces from smartphone. ISPRS Int. J. Geo-Inf. Res. Board 2662, 41–49. https://doi.org/10.3141/2662-05. 8 https://doi.org/10.3390/ijgi8080322. Li, W., Wang, S., Zhang, X., Jia, Q., Tian, Y., 2020. Understanding intra-urban human Russell, M., Davies, C., Wild, K., Shaw, C., 2021. Pedalling towards equity: exploring mobility through an exploratory spatiotemporal analysis of bike-sharing trajectories. women’s cycling in a New Zealand city. J. Transp. Geogr. 91, 102987 https://doi. Int. J. Geogr. Inf. Sci. 34, 2451–2474. https://doi.org/10.1080/ org/10.1016/j.jtrangeo.2021.102987. 13658816.2020.1712401. Sadeghinasr, B., Akhavan, A., Furth, P.G., Gehrke, S.R., Wang, Q., Reardon, T.G., 2021. Lin, Z., Fan, W., 2020. Bicycle ridership using crowdsourced data: ordered Probit model Mining dockless bikeshare data for insights into cyclist behavior and preferences: approach. J. Transp. Eng. Part Syst. 146, 04020076. https://doi.org/10.1061/ evidence from the Boston region. Transp. Res. Part Transp. Environ. 100, 103044 JTEPBS.0000399. https://doi.org/10.1016/j.trd.2021.103044. Lind, A., Honey-Roses, J., Corbera, E., 2021. Rule compliance and desire lines in Sarjala, S., 2019. Built environment determinants of pedestrians’ and bicyclists’ route Barcelona’s cycling network. Transp. Lett.- Int. J. Transp. Res. 13, 728–737. https:// choices on commute trips: applying a new grid-based method for measuring the built doi.org/10.1080/19427867.2020.1803542. environment along the route. J. Transp. Geogr. 78, 56–69. https://doi.org/10.1016/ Lißner, S., Huber, S., 2021. Facing the needs for clean bicycle data – a bicycle-specific j.jtrangeo.2019.05.004. approach of GPS data processing. Eur. Transp. Res. Rev. 13 https://doi.org/ Schneider, F., Daamen, W., Hoogendoorn, S., 2021. Trip chaining of bicycle and car 10.1186/s12544-020-00462-2. commuters: an empirical analysis of detours to secondary activities. Transp. Transp. López-Dóriga, I., Vich, G., Koch, S., Khomenko, S., Marquet, O., Roig-Costa, O., Daher, C., Sci. 1–24 https://doi.org/10.1080/23249935.2021.1901793. Rasella, D., Nieuwenhuijsen, M., Mueller, N., 2022. Health impacts of electric Scott, D.M., Lu, W., Brown, M.J., 2021. Route choice of bike share users: leveraging GPS micromobility transitions in Barcelona: a scenario analysis. Environ. Impact Assess. data to derive choice sets. J. Transp. Geogr. 90 https://doi.org/10.1016/j. Rev. 96, 106836 https://doi.org/10.1016/j.eiar.2022.106836. jtrangeo.2020.102903. Lu, W., Scott, D.M., Dalumpines, R., 2018. Understanding bike share cyclist route choice Sersli, S., Gislason, M., Scott, N., Winters, M., 2020. Riding alone and together: is using GPS data: comparing dominant routes and shortest paths. J. Transp. Geogr. 71, mobility of care at odds with mothers’ bicycling? J. Transp. Geogr. 83, 102645 172–181. https://doi.org/10.1016/j.jtrangeo.2018.07.012. https://doi.org/10.1016/j.jtrangeo.2020.102645. Lubitow, A., Tompkins, K., Feldman, M., 2019. Sustainable cycling for all? Race and Sersli, S., Gislason, M., Scott, N., Winters, M., 2022. Easy as riding a bike? Bicycling gender–based bicycling inequalities in Portland, Oregon. City Community 18, competence as (re)learning to negotiate space. Qual. Res. Sport Exerc. Health 14, 1181–1202. https://doi.org/10.1111/cico.12470. 268–288. https://doi.org/10.1080/2159676X.2021.1888153. Lubitow, A., Abelson, M.J., Carpenter, E., 2020. Transforming mobility justice: gendered Skov-Petersen, H., Barkow, B., Lundhede, T., Jacobsen, J.B., 2018. How do cyclists make harassment and violence on transit. J. Transp. Geogr. 82, 102601 https://doi.org/ their way? - a GPS-based revealed preference study in Copenhagen. Int. J. Geogr. Inf. 10.1016/j.jtrangeo.2019.102601. Sci. 32, 1469–1484. https://doi.org/10.1080/13658816.2018.1436713. Majumdar, B.B., Mitra, S., 2019. A study on route choice preferences for commuter and Soltani, A., Allan, A., Javadpoor, M., Lella, J., 2022. Space syntax in Analysing bicycle non-commuter bicyclists: a case study of Kharagpur and Asansol, India. commuting routes in inner metropolitan Adelaide. Sustainability 14, 3485. https:// Transportation 46, 1839–1865. https://doi.org/10.1007/s11116-018-9898-z. doi.org/10.3390/su14063485. Marquart, H., Schlink, U., Ueberham, M., 2020. The planned and the perceived city: a Soriguera, F., Jiménez-Meroño, E., 2020. A continuous approximation model for the comparison of cyclists’ and decision-makers’ views on cycling quality. J. Transp. optimal design of public bike-sharing systems. Sustain. Cities Soc. 52, 101826 Geogr. 82, 102602 https://doi.org/10.1016/j.jtrangeo.2019.102602. https://doi.org/10.1016/j.scs.2019.101826. 11 J. Cubells et al. Journal of Transport Geography 111 (2023) 103664 Sun, Y., Moshfeghi, Y., Liu, Z., 2017. Exploiting crowdsourced geographic information Wu, T.-G., Chang, J.-C., Huang, S.-H., Lin, W.-Y., Chan, C.-C., Wu, C.-F., 2021. Exposures and GIS for assessment of air pollution exposure during active travel. J. Transp. and health impact for bicycle and electric scooter commuters in Taipei. Transp. Res. Health 6, 93–104. https://doi.org/10.1016/j.jth.2017.06.004. Part Transp. Environ. 91, 102696 https://doi.org/10.1016/j.trd.2021.102696. Ta, N., Zhao, Y., Chai, Y., 2016. Built environment, peak hours and route choice Yang, H., Bao, Y., Huo, J., Hu, S., Yang, L., Sun, L., 2022. Impact of road features on efficiency: an investigation of commuting efficiency using GPS data. J. Transp. shared e-scooter trip volume: a study based on multiple membership multilevel Geogr. 57, 161–170. https://doi.org/10.1016/j.jtrangeo.2016.10.005. model. Travel Behav. Soc. 28, 204–213. https://doi.org/10.1016/j.tbs.2022.04.005. Tuncer, S., Laurier, E., Brown, B., Licoppe, C., 2020. Notes on the practices and Zhang, W., Buehler, R., Broaddus, A., Sweeney, T., 2021. What type of infrastructures do appearances of e-scooter users in public space. J. Transp. Geogr. 85, 102702 https:// e-scooter riders prefer? A route choice model. Transp. Res. Part Transp. Environ. 94, doi.org/10.1016/j.jtrangeo.2020.102702. 102761 https://doi.org/10.1016/j.trd.2021.102761. Wang, H., Moylan, E., Levinson, D.M., 2022. Prediction of the deviation between Zhu, R., Zhang, X., Kondor, D., Santi, P., Ratti, C., 2020. Understanding spatio-temporal alternative routes and actual trajectories for bicyclists. Transp. Find. https://doi.org/ heterogeneity of bike-sharing and scooter-sharing mobility. Comput. Environ. 10.32866/001c.35701. Urban. Syst. 81, 101483 https://doi.org/10.1016/j.compenvurbsys.2020.101483. Weschke, J., Oostendorp, R., Hardinghaus, M., 2022. Mode shift, motivational reasons, Zimmermann, M., Mai, T., Frejinger, E., 2017. Bike route choice modeling using GPS and impact on emissions of shared e-scooter usage. Transp. Res. Part Transp. data without choice sets of paths. Transp. Res. Part C Emerg. Technol. 75, 183–196. Environ. 112, 103468 https://doi.org/10.1016/j.trd.2022.103468. https://doi.org/10.1016/j.trc.2016.12.009. Willberg, E., Poom, A., Helle, J., Toivonen, T., 2023. Cyclists’ exposure to air pollution, Zuniga-Garcia, N., Ruiz Juri, N., Perrine, K.A., Machemehl, R.B., 2021. E-scooters in noise, and greenery: a population-level spatial analysis approach. Int. J. Health urban infrastructure: understanding sidewalk, bike lane, and roadway usage from Geogr. 22, 5. https://doi.org/10.1186/s12942-023-00326-7. trajectory data. Case Stud. Transp. Policy 9, 983–994. https://doi.org/10.1016/j. cstp.2021.04.004. 12