Journal of Cycling and Micromobility Research 2 (2024) 100025 Contents lists available at ScienceDirect Journal of Cycling and Micromobility Research journal homepage: www.sciencedirect.com/journal/journal-of-cycling-and-micromobility-research Natural barriers facing female cyclists and how to overcome them: A cross national examination of bikesharing schemes Richard Bean , Dorina Pojani *, Jonathan Corcoran The University of Queensland, Australia A R T I C L E I N F O A B S T R A C T Keywords: Cycling Women Gender Bikesharing Intersectionality Transport disadvantage Worldwide, the gender gap in urban cycling is considerable, with most cyclists being young to middle-aged men. In the current study, we first capture the suite of cycling barriers facing women before empirically investigating whether and how much three natural barriers (inclement weather, hilliness, and darkness) impact female users of bikesharing systems. For the analysis, we spatially integrate gender for more than 200 million bikesharing trips with fine-grained weather, gradient, and sunset/sunrise data. Computing a suite of the generalized additive models for ten cities worldwide covering a period of 14 years, we find that wind and precipitation disincentivise cycling, and more so for women than for men. Similarly, steeper gradients are a significant barrier for female bikeshare users for many cities. In every city, women make fewer trips in the dark (i.e., before sunrise and after sunset) compared to men. In higher-cycling cities, regardless of natural barriers, cycling declines less with age for women compared to other cities. To overcome the barriers presented by inclement weather, hilliness, and darkness we recommend (a) partial electrification of bikesharing fleets, (b) reduced exposure along bicycle paths (through manufactured shelters or tree canopies), and (c) adequate nighttime lighting along cycling paths. In the spirit of open science, all data and code on which this paper is based have been provided on Mendeley: https://data.me ndeley.com/datasets/vmy42hywwx/1. Introduction The gender gap in urban cycling is considerable. While many women would like to embrace cycling due to environmental or health concerns (Ravensbergen et al. 2019; Prati et al. 2019), in most countries the majority of cyclists are young to middle-aged men. This skew in the cycling population is especially pronounced among commuter cyclists (Goel et al. 2022). In the Anglosphere (i.e., Australia, the United States, Canada, New Zealand, and the United Kingdom), women constitute just one-quarter of commuter cyclists and one-third of recreational cyclists (Garrard, 2021). Furthermore, in most European countries, there are more women than men who have never used a bicycle (see Prati, 2018). In high-cycling geographies (i.e., contexts where the cycling mode share is higher than 7 percent) women use bicycles as much as men (Goel et al. 2022). However, those settings are the exception rather than the rule and tend to be concentrated in north-western Europe (Germany, Netherlands, and Denmark) and north-eastern Asia (China and Japan). According to a study of 17 countries across 6 continents (Goel et al. 2022), women are equally represented across all age groups in cities with cycling shares greater than 13 %. Otherwise, only working-age women cycle while young girls and older women are excluded (Goel et al. 2022). Some cities in low-and-middle-income countries, such as New Delhi (India), Kisumu (Kenya), and Bogotá (Colombia), are more gender-imbalanced than could be expected from their overall cycling rates (Goel et al. 2022). Possibly, this owes to the fact that these countries are highly unequal and masculine in terms of culture (Hofstede Insights, 2022).1 Why should we care about the cycling gender gap? In fact, why should we care whether women cycle or not? While cycling is beneficial for all genders, women may stand to gain more from cycling because, according to the World Health Organisation, women tend to exercise less than men (WHO 2021). Yet, women may need more of the type of exercise like cycling that builds bone density, strengthens muscles, helps manage weight, and ameliorates mood, as women are at higher risk of osteoporosis, arthritis, anxiety, depression, and a variety of autoimmune diseases (NLM 2018). Women generally prefer physical activity that is * Corresponding author. E-mail address: d.pojani@uq.edu.au (D. Pojani). 1 Geert Hofstede defines masculine societies as having clearly distinct gender roles: “men are supposed to be assertive, tough, and focused on material success; women are supposed to be more modest, tender, and concerned with the quality of life.” https://doi.org/10.1016/j.jcmr.2024.100025 Received 30 January 2024; Received in revised form 14 April 2024; Accepted 15 April 2024 Available online 18 April 2024 2950-1059/© 2024 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 non-strenuous, incidental, and seamlessly incorporated into daily routines, and cycling meets that need (Garrard, 2021). A more implicit benefit of adequate cycling conditions is that children and older adults can travel independently by bicycle, saving mothers and other female caregivers much escorting time (Garrard, 2021). At the same time, where cycling conditions are adequate, many women may want to cycle alongside their children as a way to perform ‘good mothering’ and role-model a healthy behaviour (Ravensbergen et al. 2019). Finally, cycling may even help women to feel more empowered, free, and confident in their bodies and abilities. For these benefits to be accrued, we need to gain a better understanding of the relationship between gender and urban cycling. digital bikesharing systems rather than relying on surveys or observations of cyclists. Gathering longitudinal and representative survey data is excessively costly and time-consuming. Population censuses are only run once every five to ten years, and only report data on commute travel during a single workday of the year. Other useful information, such as on weekend and recreational travel, is not collected. Travel surveys run by transport agencies often rely on panels which may not be representative of the general population. Also, surveys tend to suffer from inaccurate completions. In contrast, some bikesharing schemes make data available on trip characteristics, users’ age, vehicle type, and, in a few cases, users’ gender.3 Given that most bikesharing systems no longer provide gender as part of their operational data, this study is particularly timely. Study objectives Background on gendered barriers to cycling In the current study we first draw on cycling scholarship to present the suite of barriers known to impact cycling among females. We group these barriers under six broad domains: Psychology & Identity; Culture & Society; Income & Poverty; Built Environment; Health & Physiology; and Policies & Institutions. Then we focus on the role of three natural barriers: inclement weather (precipitation, wind, and temperature), hilliness, and darkness. We analyse more than 200 million bikesharing trips in ten cities worldwide over the course of 14 years. These are matched with fine-grained weather, gradient, and sunset/sunrise data, which are available through various public agencies. Ideally, transport geographers of a feminist persuasion should consider both how gender shapes mobility and how mobility shapes gender (Ravensbergen et al. 2019; Hanson, 2010; Law, 1999). With this in mind, we have summarised the gender-specific barriers to cycling in Table 1. We expect most of these barriers to apply to bikesharing as well. The table has been developed based on foundational work by Loukaitou-Sideris (2020) on the transport-related barriers faced by women. To compile the table, we have relied on literature reviews (Hanson, 2010; Garrard et al. 2012; Ravensbergen et al. 2019; Garrard, 2021) or empirical studies that cover more than one country (Totaro Garcia et al. 2022; Le et al. 2019; Prati et al. 2019; Prati, 2018) or city (Aldred et al. 2016; Butterworth and Pojani, 2018), rather than attempting to account for every case study in this space. Note that sex and gender are different concepts. Sex refers to biological characteristics whereas gender is a socio-cultural construct which may or may not align with an individual’s biological sex. Table 1 combines ‘gender’ and ‘sex’ barriers. Some of the causes and effects noted in Table 1, especially in the Health & Physiology and Psychology & Identity domains, are speculative or based on anecdotal evidence and will need stronger empirical research evidence to substantiate. Several barriers in the Psychology & Identity domain may be due to how women are perceived and socialized. Also, some barriers are universal whereas others pertain to particular geographies, age brackets, ethnicities, and socio-economic groups (Ravensbergen et al. 2019). One of the strongest and ubiquitous barriers appears to be women’s heightened perception of risk while cycling (Prati et al. 2019; Halefom et al., 2022) – although the actual risks of cycling are generally higher for men (Prati et al. 2019). Many women will only cycle on fully segregated bicycle paths (Le et al. 2019). This means that low-cost solutions such as sharrows or bike lanes involving a simple line painted on the asphalt are ineffective in attracting women (Teschke et al. 2017, Garrard et al. 2008). Drawing on the academic literature, we hypothesise that certain natural barriers including inclement weather (Flynn et al., 2012), hilliness (Grudgings, et al., 2018), and darkness (Abasahl, et al., 2018) both individually and collectively are important and contribute towards the observed gender imbalance in cycling. First, we reason that women may potentially perceive precipitation, wind, and extreme temperature conditions as more daunting or uncomfortable. Second, hilly terrains require higher levels of physical exertion and fitness, potentially dissuading women who may feel less confident in their abilities compared to men. Finally, darkness not only raises safety concerns but also amplifies feelings of vulnerability, particularly for women cycling alone, impacting their willingness to ride at night. However, despite some evidence suggesting these barriers, a comprehensive empirical understanding is lacking. This deficit underpins the impetus of the current study. Identified research gaps Factors such as inclement weather, hilliness, and darkness fall within several of the six domains known to affect cyclists (see Bean et al. 2021). However, to date, few empirical studies have sought to examine whether and how these natural factors differentially act as barriers between men and women (see Le et al. 2019). With discussions advancing around the climate crisis and its impact on cities and sustainability, some natural barriers to cycling may increase or decrease among women (and other genders) depending on one’s geographic location alongside culture, health, income, and other contextual variables. Much contemporary cycling research employs data harvested from digital bikesharing systems to investigate bicycle travel patterns in cities. Research has already established the various ways in which weather impacts bikesharing usage dynamics (see, for example, Lepage and Morency, 2021; Bean et al. 2021; Tu et al. 2019; El-Assi et al., 2017; Heaney et al. 2019; An et al. 2019; Kim, 2018; Lu et al., 2017; de Chardon et al. 2017; Rudloff and Lackner, 2014; Gebhart and Noland, 2014; Corcoran et al. 2014; Wygonik et al. 2014) but most studies do not report findings by gender. The effect of hilliness on bikesharing rates is also well researched (Cervero et al. 2009; Iseki and Tingstrom, 2014; Parkin et al., 2016; Braun et al. 2016; Mateo-Babiano et al. 2017; de Chardon et al. 2017; Le et al. 2019; Eren and Uz, 2020; Scott et al. 2021; Tyndall, 2022; Cubells et al. 2023), the consensus being that steep gradients lead to less cycling.2 However, gender differences are not well known. The role of darkness in relation to bikesharing use remains nearly unexamined (see Noordzij, 1976; Cubells et al. 2023 for a few exceptions). Therefore, this study helps expand our understanding of bikesharing operations, in addition to exploring gendered travel patterns (Pojani et al., 2020). We chose to focus on bikesharing rather than cycling in general for two reasons. First, examining gender gaps in bikesharing is necessary because this activity is known to attract more men than women, but the reasons for this disparity are poorly understood (Pojani et al. 2020). Second, there are major advantages in employing data harvested from 2 Gentle slopes may be tolerated as they provide an opportunity for exercise or if they allow users to take a safer route (Halefom et al. 2022; Stinson and Bhat, 2003). 3 The obvious disadvantage is that bikeshare datasets do not capture the cycling trips of people who are not system members and users. 2 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 Table 1 Cycling barriers facing women. Mentions of natural barriers are in bold font. Barrier Causes Patterns and/or effects Psychology & Identity Fear of mugging and victimization Avoidance of bicycles / Cycling only during daytime / only in certain places / only accompanied by others Avoidance of bicycles, especially where secure bicycle parking is missing Children not allowed to travel independently by bicycle / Children (esp. girls) escorted everywhere mostly by mothers Avoidance of cycling in mixed traffic / cycling slowly and carefully (increasing trip length) / Avoidance of cycling along paths with potholes, gravel, debris Avoidance of fast-paced on-road cycling Culture & Society Income & Poverty Built Environment Health & Physiology Policies & Institutions Fear of bicycle theft Parental fear of stranger-danger Concerns over traffic safety / high sensitivity to risk (fatality/ injury/near misses) Women may not see themselves as ’sporty’ / have low confidence in their physical abilities Some cycling contexts are hypermasculine / cycling is used to perform masculinity Feminine body comportment is marked by restraint / hesitation Harassment/aggression from drivers and other male cyclists (e.g., for being too slow or clumsy) Women have higher wayfinding anxiety / prefer different wayfinding strategies Women receive less encouragement to cycle throughout their lives Some religious norms / practices around women’s presence in public space Fears that cycling causes virginity loss in girls Women are primary caregivers for children / parents / domestic chores Social ideals of appropriate female behaviour and appearance Dress codes in workplaces / women expected to wear skirts/ heels/makeup Lack of economic resources for bicycle ownership / modal affordability Lack of economic resources for expensive equipment / accessories / clothing / repairs Lack of economic resources for bikeshare scheme membership / cycling club membership Bicycle use as a symbol of poverty / low class Automobile-oriented urban form / sprawl / large distances between destinations Segregated land-uses / large distances between destinations Lack of adequate infrastructure for cycling Concentric-radial urban patterns with jobs concentrated in CBDs Preference for shorter cycling distances Cycling infrastructure often located in flood plains (along water bodies) Preference for moderate-intensity physical activity / less vigorous travel Less strength on average Pre-menstrual/menstrual symptoms / low energy and energy surges during those periods Female low-centre anatomy / different cycling style Women as primary caregivers for children & parents Women are more health-conscious / bear more responsibility for their family’s heath Pregnancy and longer life-spans for women Requirement to wear helmets while cycling interferes with hair coiffure Provision of cycling infrastructure with low separation from cars along major roads Women are excluded from cycling sports Slow response to increasing flooding events (due to climate change) Male-dominated cycling services (shops/repair points) Male-dominated planning/engineering institutions Women are uncomfortable cycling in public / girls fear being labelled a ’tom boy’ Women are reluctant to cycle Avoidance of cycling in mixed traffic Avoidance of cycling if infrastructure is confusing / lacks clear wayfinding signs Avoidance of cycling, especially in adolescence and old age Banning women from bicycle ridership Banning women from bicycle ridership Trip chaining / travelling with others / hypermobility / time poverty / carrying large items during travel Women avoid ’putting their body on show’ Avoidance of bicycle for commuting, especially in rainy / cold / humid weather or where showers are missing at work Lack of access to bicycles / shift to walking Avoidance of bicycles / shift to public transport Avoidance of bikeshare / cycling clubs, reduced opportunities for socialising Middle-class women avoid cycling Overreliance on faster and/or less strenuous modes such as private automobiles Overreliance on faster and/or less strenuous modes such as private automobiles Avoidance of bicycles Avoidance of long cycling commutes from middle/outer suburbs to the CBD Avoidance of long cycling commutes Avoidance of cycling in wet weather due to fear of floods Avoidance of bicycles in hilly topography / inclement weather Avoidance of bicycles if they need to be lifted to a storage space Avoidance of cycling during part of the month Avoidance of bicycles built for a male anatomy Avoidance of bicycles because carrying others (i.e., children) on a bicycle is more strenuous Avoidance of cycling in polluted air / along uncovered paths (sun exposure) Physical ability to cycle may vary substantially over the life course Avoidance of cycling to work Avoidance of cycling on main roads Perception that cycling is not for women Avoidance of cycling Avoidance of cycling services that do not understand women’s needs Gender issues in cycling given short shrift / Women uninvolved in planning processes Source: Table by authors based on Loukaitou-Sideris (2020); Hanson (2010); Garrard et al. (2012); Ravensbergen et al. (2019); Garrard (2021); Totaro Garcia et al. (2022); Le et al. (2019); Prati et al. (2019); Prati (2018); Aldred et al. (2016); Butterworth and Pojani (2018); Soltani et al. (2022). Methods 1. Mobility data from 229,413,658 individual bikesharing trips undertaken across ten cities worldwide, recorded between February 2010 and January 2023 inclusive (Table 2). Six of the cities are in the United States (New York City, Chicago, Columbus, San Francisco Bay Area, Boston, Minneapolis), with the remaining schemes located in Australia (Brisbane), Finland (Helsinki), and Mexico (Mexico City and Guadalajara). By far the largest two bikesharing systems are in New York City and Mexico City, representing 78 % of the total Datasets The data for the current study were drawn from three principal sources: 3 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 number of trips. Gender data is provided in all ten cities, but is possibly skewed by cultural factors.4 For this study, we consider only self-identified male and female users as there is insufficient information to distinguish between other possibilities (such as no answer, non-binary, intersex). Whether users had to report gender across cities is unknown. Across the data files, apart from male and female, gender can be blank, NULL, a number representing “Unknown”, “Other” (for Bay Area) or “Not available” (Brisbane short-term rentals). For Guadalajara (September 2016 to April 2019, and March 2021), Mexico City (February 2010 to March 2020), and for Brisbane long-term rentals, all trips are either given as male or female. 2. Sunrise and sunset data calculated using the suncalc package in the R programming environment (R Development Core Team, 2008). 3. Hilliness data captured from Japan Aerospace Exploration Agency’s Advanced Land Observing Satellite (ALOS), processed under the assumption that trips were “great circle” trips.5 on female riders, we matched hourly bikesharing trip data for each city individually with (a) hourly weather data, obtained through the European Centre for Medium-Range Weather Forecasts (ECMWF)7; (b) digital elevation data, obtained from Japan Aerospace Exploration Agency’s Advanced Land Observing Satellite (ALOS); and (c) sunset/ sunrise data, calculated using the suncalc package in the R programming environment (R Development Core Team, 2008). The Generalized Additive Model (GAM) (Hastie & Tibsharini, 1990) was used in the current study using the “mgcv” package (Wood, 2012) in the R programming environment. A GAM approach is preferable for examining the effect of multiple predictor variables on a response variable and has been successfully used in other weather-traffic studies (see for example, Becker, et al., 2022; Lepage & Morency, 2021; Chen & Shen, 2016). A GAM extends the concept of the Generalized Linear Model by modelling a response variable as a linear combination of “smooth” terms. A “smooth” term is a function with a continuous curve without sharp bends or changes in direction. A “link function” is used to transform the expected value of the response variable to linearize the relationship between the response variable and the predictors. In the case of the model here, the response variable is the hourly bicycle count data, which is modelled using a Poisson distribution and the link function chosen is the log, to ensure the linear relationship between the log of the expected count and the predictor variables. The predictor variables are weather and temporal variables. In this GAM, our response variables are measured hourly (male and female cyclists per hour) whereas the predictor variables are modelled non-parametrically (hour of day, day of year, temperature, precipitation, and wind) using splines or linear functions as necessary. The ability to model these “smooth” functions for predictor variables makes GAMs an excellent model choice for the current study. To explore how gender interacts with the predictor variables, self-reported gender is considered as a categorical variable. Where possible, we split the gender data by weekend and weekday classifications because weekday and weekend bikesharing use is notably different (Todd et al. 2021).8 For the analysis, we built GAMs using the “mgcv” package for hourly bicycle usage as follows:9 Usage=gender+s(JulianDate)+s(Hour)+te(Hour, JulianDate)+s (temp)+s(precipitation)+s(wind)+ε (1) Table 3 provides a summary of the years of complete data used in the study. The final column gives the percentage of trips (that is, rows in the data) with specified male or female gender. Some scholars (e.g., Totaro Garcia et al. 2022) have identified a significant link between the percentage of female cyclists and the modal share of cycling overall. Also, they have noted that bicycle use among women declines with age in low-cycling cities but not in high-cycling cities. For a more intersectional analysis, we plotted bikesharing users’ gender and age for each city (Fig. 1). As seen in the figure, adolescents of both genders are only minimally represented. Most bikesharing users tend to be younger adults, with use nearly halving after age 35. While women cyclists tend to outnumber men in the 18–35 age bracket, men dominate the older age brackets. Helsinki and Minneapolis, which are high-cycling cities, are notably different from the others. Here, older women (46−75) cycle more than men. Some cities provided information on the types of bicycles used in the system (classic vs electric, docked vs undocked).6 Presuming that some of the barriers discussed in the current study (such as steepness) may be removed or at least reduced through the use of an e-bike, we also examined trip data by bicycle type. Analysis 7 Since 1950, the Copernicus Climate Change Service (C3S) at the European Centre for Medium-ranger Weather Forecasts (ECMWF) has produced a reanalysis of the global climate, called ERA5. This service contains hourly estimates of climate variables useful for our purposes, such as two-metre temperature, total precipitation, and wind speed. 8 Everywhere outside of mainland China. 9 We used cyclic cubic splines for hour and day of year to avoid discontinuities between the beginning of end of days and years. This was based on advice from Simpson (2012, 2019) who builds a similar model with hour and day of year. See: To measure the effect of inclement weather, hilliness, and darkness 4 For example, the “Citi Bike Trip Histories” page describes the data format for the “Gender” field as “Zero=unknown, 1=male, 2=female”. A similar format is used across all the US cities studied; for example, the Boston data before 2015 used “Male”, “Female” and a blank field, changing to the numeric format from January 2015. In the Guadalajara data in 2015, there are 469,871 trips made by 5191 unique users, but only 236 trips where gender was not provided, which were made by two unique users. 5 The ALOS project has a Global Digital Surface Model (DSM) at 30 metre resolution with freely available data. For two cities, we have finer resolution data: New York City with 1 foot (30 cm) resolution and Brisbane with 1 m resolution. We obtained those data from the following sources: 1) https://stats.stackexchange.com/questions/308244/confidence-inter val-for-the-slope-of-a-gam; 2) https://stats.stackexchange.com/questions/32730/how-to-include-an-inte raction-term-in-gam 1) https://data.cityofnewyork.us/City-Government/1-foot-Digital-Elevati on-Model-DEM-/dpc8-z3jc 3) https://stats.stackexchange.com/questions/340387/gam-selection-whenboth-smooth-and-parametric-terms-are-present 2) https://qldspatial.information.qld.gov.au/catalogue/custom/viewMeta dataDetails.page?uuid=%7B5471036F-0ED8-41EE-BDBE-14EA846FC81E %7D 4) https://stats.stackexchange.com/questions/403772/different-ways-of -modelling-interactions-between-continuous-and-categori 6 Data on classic vs electric bicycles: New York City (April 2021 on), Chicago (July 2020 on), and Minneapolis (May 2020 on). Data on docked vs undocked bicycles: San Francisco Bay Area, New York City, Columbus, Chicago, and Minneapolis, from as early as April 2020. 5) cal-pred 6) https://stats.stackexchange.com/questions/413955/should-i-use-poissonor-gaussian-family-in-gam 4 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 Table 2 List of cities included in the current study, sorted by share of female users. City Start month End month Total male trips Total female trips Female trips % Link to data** Columbus San Francisco Bay Area* Chicago Boston Guadalajara New York City Mexico City Brisbane* Minneapolis* Helsinki* Total Mar 2018 Jun 2017 Jun 2013 Jul 2011 Dec 2014 Jun 2013 Feb 2010 Oct 2010 Apr 2011/18 May 2017 Mar 2020 Apr 2019 Dec 2019 Apr 2020 Jan 2023 Jan 2021 Jul 2022 Mar 2015 Nov 2012/19 Oct 2017 30,131 1902,810 12,235,830 6843,613 16,232,629 74,574,098 56,371,073 423,938 380,283 568,686 169,563,091 5524 637,069 4112,553 2334,078 5658,999 26,339,265 20,002,605 185,073 196,815 378,586 59,850,567 15.5 25.1 25.2 25.4 25.9 26.1 26.2 30.4 34.1 40.0 26.1 CoGo Bay Wheels Divvy Bluebikes Mibici Citi Bike Ecobici CityCycle Nice Ride City Bikes - * The Minneapolis system is open only April-November. The Helsinki system is only open May-October. For Brisbane, until 1 December 2013, bike hire was only available between the hours of 5 am and 10 pm. The San Francisco Bay Area includes San Francisco, the East Bay (Oakland and Berkeley), and San Jose. ** We only provide a web link where the data are publicly available. (inclement weather, hilliness, and darkness) were operationalised and present some descriptive statistics before proceeding to the model findings. Table 3 Bikeshare trip data used in the current study and percentage of complete gender data. City Complete years of data Total number of years of data % complete gender data New York City (NYC) Boston Chicago Minneapolis 2014–2020 7 90.0 % 2012–2019 2014–2019 2011–2012, 2018–2019 2015–2022 2011–2021 2011–2014 2019 2017 2018 8 6 4 82.3 % 77.8 % 66.1 % 8 11 4 1 1 3 99.9 % 99.5 % 87.1 % 34.7 % 63.0 % 92.6 % Guadalajara Mexico City Brisbane Columbus Helsinki San Francisco Bay Area Inclement weather We define ‘inclement weather’ as high rainfall, high wind speed, and high/low temperature. Table 4 shows the modelled decrease in male and female weekday ridership for a unit increase in observed rain (millimetres per hour) or wind (metres per second). This is assuming all other factors (e.g. hour of day, day of year) are equal. The calculations are time-weighted averages; that is, the modelling is performed annually and averaged for every year in the data. Hilliness Hilliness was operationalised as digital elevation. For each trip, we calculated the average gradient using the difference in elevation and the ‘great circle’ distance11 calculated using the “distGeo” function using the geosphere package in the R programming language. We were interested in finding out whether uphill gradients are a bigger barrier for female cyclists than male cyclists. For this purpose, we studied trips where the estimated ‘great circle’ average gradient exceeded 1 %. This cut-off value was chosen to demarcate “flat” from “non-flat” trips. A summary of national guidelines in several European countries indicates that a 1.75 % gradient is acceptable in Belgium and the Netherlands regardless of the difference in height between a given trip origin and destination (Buczyński, 2023). A descriptive analysis of our data in terms of hilliness by city and gender is shown in Table 5. In this equation, the ‘precipitation’ and ‘wind’ are linear terms, and every term also considers the interaction of ‘gender’ (noting that only trips with self-identified male and female riders are considered); ε represents an error term. As seen in Table 2, we have a total of 53 years of data. Given that weekdays and weekends needed to be considered separately, we built a total of 106 GAMs. We used the “gratia” package to plot the partial effects of each term. These are important to study the effects on the usage of holding one predictor variable constant. To test the difference in male and female cycling counts while simultaneously considering the effects of wind and precipitation, we also fitted a parsimonious model with linear terms for wind and precipitation as follows: Usage=gender+s(JulianDate)+s(Hour)+te(Hour, JulianDate)+s (temp)+precipitation+wind+ε (2) Modifying Equation (1), ‘precipitation’ and ‘wind’ are linear terms, and every term also considers the interaction of ‘gender’ while ε represents an error term. This parsimonious model (Equation 2) avoids possible overfitting of the precipitation and wind terms using smooth functions.10 Model 2 is fitted for whole calendar years of trips for each city so as to obtain parametric coefficients for the intercept, gender, precipitation, wind, precipitation interacting with gender (gender:precipitation) and wind interacting with gender (gender:wind). This enables us to estimate the change in usage based on changes in predictor variables (all else being equal). More importantly, it allows us to quantify and test for differences in usage by gender depending on precipitation and wind. Below we explain further how the predictor variables of interest Darkness Darkness was operationalised as sunrise and sunset. In each of our ten case study cities there was a strong pattern of trips taking place before sunrise and after sunset (Table 6). Results Modelling results are presented to draw attention to significant gender differences in the ridership counts for: (a) precipitation; (b) wind; (c) temperature; (d) gradient; and (e) the proportion of trips taken before sunrise or after sunset. 11 We were unable to estimate the actual trip lengths across all individual trips (n=229,413,658). Therefore, we chose to employ the ‘great circle’ distance, which is necessarily an underestimate of trip length. A lower value of a 1 % ‘great circle’ gradient was found to be appropriate across each of our case study cities, instead of attempting to choose a subset of the steepest trips in each city. We suggest that future studies split the ‘great circle’ gradients into more categories to examine gender differences. 10 Overfitting occurs where the model starts capturing the random error in each dataset as opposed to the relationships that exist between the variables. 5 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 Fig. 1. Gender versus age in nine cities. The male and female bikeshare cyclist numbers are standardised (the sum of each age is adjusted to be 1). Columbus is excluded due to low cyclist numbers in each bin. with associated confidence intervals. Partial plots permit visualisation of how a particular variable affects usage over the range of observed values. In the partial plots for New York (Fig. 2), we display the 95 % confidence intervals and where these are non-overlapping, this indicates a statistically significant difference for the relative interval. The “partial effect” values shown here refer to the log of the count data as we consider only hours with non-zero male and female usage. Further explanation of the meaning of the partial effects in Figs. 2 and 3 can be Inclement weather The effect of the weather factors – precipitation, wind, and temperature - are displayed through a visual examination of GAM models.12 We also checked the effect of temperature. We found that for the weekday model, all terms were significant with p-values less than 2e-16. Next, we plotted the partial effect of precipitation by gender for a range of values, 12 Some studies (e.g., Bean et al. 2021) have summarised weather variables into one variable, such as the UTCI (Universal Thermal Comfort Index). This effectively captures the effect of wind, humidity, solar insolation, and temperature into one variable. In this study, we consider wind separately. 6 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 approximately 6200 hours of data from weekdays in 2019, and of these only 14 hours have precipitation values greater than 5 mm. This makes it difficult to draw any clear inference for values of precipitation greater than 5 mm per hour. (In fact, adding in the term for precipitation without gender effects, the curve changes direction for values greater than 5 mm per hour, which suggests possible overfitting of the model.) However, the effect of precipitation is clearly different for male and female riders for values between 1 mm and 5 mm per hour. This accounts for 253 hours (or 4 % of the data coverage) of weekday riding in 2019. We apply the same technique for wind. We have 924 hours with wind greater than 5 m/s. Above this value, there is a significant difference between male and female ridership (see Fig. 2b). For temperature, the cross-over point is about 11◦ C, but the difference is not significant between men and women. Generally, below 11◦ C, the partial effect is greater for men; that is, women ride less below this temperature; however, the differences are very minor (Fig. 2c). The effect of hour-of-day (i.e., darkness) is clear (Fig. 3). In this plot, each line shows the general time of day effect added to the time of day by gender effect. The confidence intervals are too small to be visible, and the cross-over points are around 4:20 am and 6:30 pm. Before this and after this, women ride less. However, this plot does not capture the interaction term with day-of-year; therefore, it is simpler and more informative to examine the female and male riding percentage before sunrise and after sunset in each city. Instead of analysing the model for the effect of hour of day and day of year, which may depend on several causes, we simply split the data based on sunrise and sunset times. We now describe the results of fitting the parsimonious model with linear wind and precipitation terms. Generally, the precipitation and wind coefficients are negative.14 Note that these coefficients relate to female riders and can be adjusted by the “genderM:precipitation” and “genderM:wind” coefficients for male riders. Thus, it is possible, although unlikely, that a wind coefficient for a particular city / year / day-of-week model could be negative for female riders, but positive for male riders. If the “genderM:precipitation” coefficient is positive, then the slope of the precipitation line is higher for male cyclists, which indicates they ride more during precipitation. This coefficient is positive in 82 of 106 models and the results are statistically significant in 66 of those.15 Among the 24 models where the coefficient is negative, only 8 return statistically significant results.16 Similarly, if the “genderM:wind” coefficient is positive then the slope of the wind line is higher for male cyclists, which indicates they ride more in windier weather. This coefficient is positive for 72 of 106 models, of which 54 have statistically significant results.17 Of the 34 models where the “genderM:precipitation” coefficient is negative, the results are statistically significant in 13.18 To summarise, wind and precipitation are generally a greater barrier for female cyclists than for male cyclists. The exceptions to this rule are in Guadalajara and Mexico City during several years where the reverse was found. In New York City, each of the 14 models had a positive and statistically significant coefficient for wind and precipitation for male cyclists (p<0.05 for all). Similarly in Chicago each of 12 models had positive coefficients (p<0.05 for 23 of 24). Boston also has 16 models with positive coefficients (p<0.05 for 16 of 32 coefficients). The plots of Table 4 Inclement weather and bikeshare trips by gender. City Rain (male % drop) Rain (female % drop) Wind (male % drop) Wind (female % drop) New York City (NYC) Boston Chicago Minneapolis Guadalajara Mexico City Brisbane Columbus Helsinki San Francisco Bay Area (San Francisco) San Francisco Bay Area (East Bay) San Francisco Bay Area (San Jose) 35 37 25 20 11 21 32 7 42 51 39 41 29 21 13 18 28 6 45 56 5.4 4.0 3.6 3.9 -2.5 1.7 2.9 3.2 1.3 7.4 5.8 4.1 4.3 4.3 -2.4 2.1 2.1 1.4 1.7 6.9 55 57 10.0 8.9 25 21 5.0 4.0 Note: The table shows the percentage change for unit rain or wind change; weekday trips, time weighted average by year. Table 5 Hilliness and bikeshare trips by gender. City Male (%) Female (%) New York City (NYC) Boston Chicago Minneapolis Guadalajara Mexico City Brisbane Columbus Helsinki San Francisco Bay Area 24.5 11.4 1.9 11.1 13.4 6.6 6.7 8.5 13.6 20.7 23.8 9.6 1.5 10.4 13.2 6.7 5.7 11.9 11.5 19.2 Note: The table shows time-weighted average trips by month, with a gradient higher than 1 %. Table 6 Darkness and bikeshare trips by gender. City Male (%) Female (%) New York City (NYC) Boston Chicago Minneapolis Guadalajara Mexico City Brisbane Columbus Helsinki San Francisco Bay Area 26.6 24.8 21.8 18.5 22.1 22.1 22.2 25.5 16.4 21.8 23.2 22.0 19.8 14.9 20.6 20.8 20.2 22.4 13.5 18.4 Note: The table shows the percentage of trips at night, time-weighted average by month. In Minneapolis and Helsinki, the values are slightly lower than other cities as these bikesharing schemes are closed in winter. found in the Appendix and the reader may explore each model with data and code provided.13 The partial effect of precipitation for male versus female riders in New York is shown in Fig. 2a. Note that the model was built from 14 In 101 of 106 models and 82 of 106 models respectively. p < 2e-16 in 51 models, p < 0.01 in 5 models, p < 0.05 in 8 models, and p < 0.1 in 2 models. 16 p < 2e-16 in 3 models, p < 0.01 in 1 model, p < 0.05 in 3 models, and p < 0.1 in 1 model. 17 p < 2e-16 in 35 models, p < 0.01 in 4 models, p < 0.05 in 5 models, and p < 0.1 in 10 models. 18 p < 2e-16 in 3 models, p < 0.01 in 4 models, p < 0.05 in 5 models, and p < 0.1 in 1 model. 15 13 For example, the model contains three terms for precipitation: s(rain), s (rain):genderM and s(rain):gender(F). Evaluating these terms at, say, 2 mm of precipitation per hour results in estimates of −1.70, 1.07, and 0.926 respectively. Due to the log-link function used in the model, the multiplicative effect of this compared to 0 mm of precipitation is e−1.70+1.07=0.53 for males and e−1.70+0.926=0.46 for females. That is, all other factors being equal, this amount of precipitation reduces ridership by 47 % for males and 54 % for females. 7 R. Bean et al. Partial effect Journal of Cycling and Micromobility Research 2 (2024) 100025 2 gender F M 1 0 0 1 2 3 4 5 (a) Precipitation (mm) Partial effect 0.1 0.0 gender F M 0 3 6 9 12 (b) Wind speed (m/s) Partial effect 20 10 gender 0 F M 0 5 10 15 20 25 (c) Temperature (Celsius) Fig. 2. Partial effect of precipitation, wind, and temperature for male versus female riders, New York City, 2019, weekdays model. points).19 The general upward trend in cycling possibly reflects the expansion of the system to areas outside the original core. The annual seasonal effect is also clear and may be explained by several factors, including a different mix of recreational versus commuter trips, or the possible avoidance of hills in warmer months. In Boston, the percentage of uphill trips is somewhat lower, but the same annual pattern is observed despite the Digital Elevation Model resolution being much lower. In 63 of 64 months the percentage of uphill trips is lower for women than men.20 In other large cities in our sample, we see a more nuanced pattern. In Guadalajara, for example, in 91 months examined, the percentage of uphill trips for men exceeds of women’s in only 65 months.21 A seasonal effect, if present, is not readily apparent. Note, however, that Guadalajara’s elevation data has a the model outputs for the other cities look quite similar to New York’s and, therefore, are not shown here. Hilliness Gender difference in relation to hilliness can be most clearly seen by plotting the number of trips which exceed a 1 % gradient. Below we present the findings for five large cities. Due to space constraints, we only show the plot for New York City in the text. All the other plots are shown in Appendix 1. In New York City, the gender difference is very clear (Fig. 4). For the 92 months examined (June 2013 to January 2021), in 91 months the percentage of male trips with an average gradient greater than 1 % was greater than the percentage of corresponding female trips (although the difference is never more than a few percentage 19 Assuming a hypothesis of no difference for each month (that is, a probability of 0.5 for male/female ratio being higher), and independent observations, the estimated p-value for these observations is less than 2e-16 (two-sided binomial test). For the New York City figures, the differences range from −0.002–1.60 % with an average of 0.74 %. 20 p < 2e-16. 21 p = 5e-5. 8 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 1.0 Partial effect 0.5 gender 0.0 F M 0 5 10 15 20 Hour of day Fig. 3. Partial effect of hour-of-day for male versus female riders, New York City, 2014–2020. Fig. 4. New York City, June 2013 to January 2021: plot for male and female, percentage of trips with average gradient greater than 1 %. relatively coarse resolution.22 Mexico City presents a similar pattern to Guadalajara. In only 67 out of the 149 months examined did the share of uphill male trips surpass the share of uphill female trips. Also, the trips taken seemed noticeably flatter than for New York City and Guadalajara. The steep drop in the total number of trips in 2012 may be due to reductions in service or the 22 The resolution is 100 times lower than for New York City (30 ms vs 30 centimetres). 9 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 introduction of lighter bicycles. The analysis for Brisbane was based on much less trip data but the same fine-grained elevation model as in New York City. Of the 53 months studied, the percentage of uphill trips made by men exceeded percentage of uphill trips made by women in 43 months.23 The effect of fleet electrification could be investigated in some places as significant numbers of trips were taken with both classic and e-bikes. The San Francisco Bay Area and New York City schemes are illustrative (Fig. 5). In the San Francisco Bay Area, which is known for being hilly, ebikes were clearly used to overcome gradient issues. Overall, the percentage of steeper trips was 21 % for men, 19 % for women. However, classic push bicycles were used in only 19 % of the steeper trips compared to 23 % for e-bikes. In New York City, which is relatively flat, the opposite pattern was observed, with e-bikes used less than classic bicycles. However, men were still slightly more likely to cycle along steeper routes than women. Comparing the effect of hilliness in different cities is very difficult without knowing more about the bicycles used in each bikesharing system (i.e., their weights and gears). One study used standard deviation of station altitudes as a measure of scheme “hilliness” (de Chardon et al. 2017). As shown in Table 7, the “hilliness” of some schemes has changed considerably over time as each scheme expands beyond its original urban core. For example, New York City’s scheme has increased from a value of 19.6 m in 2013 to 38.1 m in 2022; in a sense, the scheme can now be considered twice as “hilly”. This helps to explain the upward trend over time seen in Fig. 4. - in the UK and Canada - have managed to increase cycling rates without any change in gender equity (Aldred et al. 2016; Winters and Zanotto, 2017). Conversely, in Australia, declines in cycling rates have been associated with reduced proportions of female cyclists (Munro, 2019). With decreasing levels of cycling, the representation of women deteriorates in every age group (Goel et al. 2022). Places such as Portland and Seville, which have increased cycling rates while also closing the gender gap (Garrard, 2021), are outliers. A set of recommendations to increase cycling participation among women has been provided by Garrard (2021). A summary follows: • Along main roads: create cycling infrastructure that provides maximum separation from motor vehicles. • Along neighbourhood streets: apply traffic calming devices such as speed limits, chicanes, and speed tables. • Improve off-road recreational trails and multi-purpose paths which appeal to women and families. • Raise awareness among all road users so that any interactions among drivers and cyclists are cooperative and respectful. • Modify road engineering and planning processes to prioritise bicycle travel over car flow. • Change traffic laws to presume driver liability in case of collisions with bicycles and generally protect cyclists. • Design and market sit-up bicycles with women’s comfort in mind. To this list, we add three recommendations designed to specifically help eliminate the three barriers investigated in this article (namely, inclement weather, hilliness, and darkness). We suggest that electric bikes offer much potential here. To this end, e-bikes require much less effort to ride than conventional bicycles thus enabling longer trips, more comfortable trips in hilly and/or windy places, and faster trips in the dark and/or rain. As seen in our analysis, e-bikes are already being used in the San Francisco Bay Area to overcome the gradient factor. However, e-bikes come with some risks and challenges. They are not lightweight due to their on-board batteries; this may not be an issue while riding but it does make it harder to lift the bike (for example, if needing to move upstairs to access a storage space). While costs are dropping, e-bikes remain relatively expensive, and users need to factor in the cost of recharging the batteries. Because starts and stops are more abrupt and dangerous than conventional bicycles, users may need some training to safely use an e-bike. Given these drawbacks, planners must carefully balance the costs and benefits of replacing all, or some of, a classic pushbike fleet with ebikes. For example, depending on the average gradients, prevailing wind directions, and precipitation patterns present in a city (de Chardon et al., 2017), a 50:50 mix of classic and electric bikes might provide the largest benefits by removing barriers for some segments of the population while keeping the costs low. Cities need to figure out what the correct ratio of e-bikes to classic bicycles is to help women overcome barriers. This type of calculation should factor in the safety advantage of e-bikes compared to e-scooters (Cox, 2016). Another desirable strategy is to minimise the exposure of bicycle paths, thus reducing the effect of heat, solar radiation, wind, and rain. Paths can be covered by manufactured shelters or tree canopies, and even artificially cooled by fans. Some good practice examples or ideas come from tropical or desert climes, such as Singapore and Dubai, rather than from temperate settings (see Lee and Pojani, 2019; Macmichael, 2023). To deal with (actual or perceived) safety and security issues in the dark, cities need to provide better lighting along cycling routes. While the cost of energy may be an issue, and the effect of streetlights on crime per se may be unknown, research has established that better lighting alleviates the fear of crime, especially among women (Boomsma and Steg, 2014). Women are in fact more likely to notice local improvements in street lighting compared to men (Painter, 1996). More broadly, a cyclable urban form should become the goal of Darkness Fewer women than men cycle before sunrise and after sunset (i.e., in the dark). To illustrate, we show the graph from New York City, the largest in the sample (Fig. 6); the plots for all ten cities are shown in Appendix 2. Our findings are in line with previously reported safety concerns (Cubells et al. 2023) but may also be due to the gendered patterns of most labour markets (for example, men may be more often engaged in shift-work, which requires late-night commuting). Future studies could further this analysis by extending the proposed GAM with a categorical variable for day and night, as with gender. This would allow researchers to examine whether and how nighttime riding has become “safer” over time in specific cities. Discussion In summary: we aimed to investigate whether and how much three natural factors (inclement weather, hilliness, and darkness) act as barriers to cycling for women. We started by integrating gender for more than 200 million bikesharing trips with fine-grained weather, gradient, and sunset/sunrise data. Then we built models for ten cities worldwide covering a period of 14 years. We found that, in most of our ten case study cities, wind and precipitation act as a disincentive to cycling, and more so for women than for men. Similarly, in many cities, steeper gradients are a significant barrier for female bikeshare users. In every city, women make fewer trips in the dark (i.e., before sunrise and after sunset) compared to men. Regardless of natural barriers, there is evidence that in higher-cycling cities (such as Helsinki and Minneapolis), cycling declines less with age for women compared to other cities. Given our findings, what ought to be done to overcome the gender gap in urban cycling? We argue that measures should be gender-specific rather than general (Lam, 2019) because increasing the overall share of bicycle trips does not automatically translate to improvements in female representation. While high-cycling countries have sustained a roughly equal representation of women over time (Garrard, 2021), several cities 23 p=6e-6. 10 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 Fig. 5. Steeper trips in the San Francisco Bay Area and New York City, electric vs classic bicycles. Conclusion planning. Beyond cycling-specific hardware, software, and orgware (Kong and Pojani, 2022), that involves compact development and mixed land-uses (Mateo-Babiano et al. 2017). In a sense, women may be considered as ‘indicator species’: cities are truly cycling-friendly when cyclists’ gender ratios are nearly equal (Baker, 2009). Our study adds to the growing literature on gender/sex specific cycling barriers. Due to data limitations, we could not explore intersectionality in detail. The combined effects of gender, age, race, income, and sexuality on cycling in general and bikesharing in particular deserve 11 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 Table 7 Hilliness measure for each scheme over time (standard deviation of station elevations in metres). Year/City New York City Boston Chicago Minneapolis Guadalajara Mexico City Brisbane Columbus Helsinki Bay Area (San Francisco) Bay Area (East Bay) Bay Area (San Jose) 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 15.1 12.8 13.2 12.8 10.9 6.1 10.9 6.5 28.1 11.3 16.7 14.3 19.8 12.8 30.3 11.0 16.7 14.5 20.0 12.8 33.8 12.2 18.3 15.8 20.9 12.8 30.9 12.3 17.5 16.2 21.0 12.8 38.1 12.9 7.2 21.9 11.9 11.1 14.7 15.6 12.8 7.2 37.7 12.9 7.2 19.7 11.5 6.0 15.2 10.1 12.8 7.2 34.7 16.1 19.6 11.8 6.0 15.4 21.0 12.8 21.6 12.8 21.6 12.8 21.6 6.8 6.9 22.2 24.8 12.3 28.0 30.6 9.2 27.2 24.1 8.5 26.1 27.4 7.7 26.8 26.0 7.3 27.0 27.4 7.4 8.7 21.4 22.8 9.9 2023 Note: Minneapolis 2012–2017 data from station lists. Fig. 6. New York City, trips by gender made before sunrise and after sunset. more research attention. Future research could also account for the weights of the shared bicycles alongside capturing the effects of headwinds, which have previously been shown to be an important natural barrier (Bean et al., 2021) with a gender dimension (Grudgings et al., 2018). To this end, an interesting explorative study could examine prevailing trip directions in weekday mornings with a compass rose versus the prevailing wind to visualise differences. In some cities, the bikeways with the highest quality infrastructure may be undermined by the direction of the prevailing winds during peak commute times. Examining the effect of tailwinds is probably less important. Dorina Pojani: Writing – review & editing, Writing – original draft, Conceptualization. Jonathan Corcoran: Writing – review & editing, Supervision, Software, Resources, Investigation, Conceptualization. CRediT authorship contribution statement The authors are grateful to Joseph Yang of OikoLab for the weather data services he provided and to Gavin Simpson for writing the mgcv advice. This study would not have been possible without them. Data availability All data and code have been provided on Mendeley: https://data. mendeley.com/datasets/vmy42hywwx/1 Acknowledgements Richard Bean: Visualization, Formal analysis, Data curation. 12 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 Appendices Appendix 1. Plot for male and female, percentage of trips with average gradient greater than 1 %. 13 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 Appendix 2. All cities, trips by gender made before sunrise and after sunset (i.e., in the dark). 14 R. Bean et al. Journal of Cycling and Micromobility Research 2 (2024) 100025 References Iseki, H., Tingstrom, M., 2014. A new approach for bikeshed analysis with consideration of topography, street connectivity, and energy consumption. Comput. Environ. Urban Syst. 48, 166–177. Kim, K., 2018. Investigation on the effects of weather and calendar events on bikesharing according to the trip patterns of bike rentals of stations. J. Transp. Geogr. 66, 309–320. Kong, W., Pojani, D. 2022. Low-Carbon Transport: Policies to Encourage Cycling in Sprawling Cities. In Brears, R. (ed.) The Palgrave Encyclopedia of Urban and Regional Futures, pp. 1-7. Cham, Switzerland: Palgrave Macmillan. Lam, T., 2019. How to get more women cycling in cities. Guardian March, 8. Law, R., 1999. Beyond ‘women and transport’: towards new geographies of gender and daily mobility. Prog. Hum. Geogr. 23 (4), 567–588. Le, Huyen T.K., Quinn, F., West, A., Hankey, S., 2019. Advancing cycling among women: an exploratory study of North American cyclists. J. Transp. Land Use 12 (1), 355–374. Lee, Q.Y., Pojani, D., 2019. Making cycling irresistible in tropical climates? Views from Singapore. Policy Des. Pract. 2 (4), 359–369. Lepage, S., Morency, C., 2021. Impact of weather, activities, and service disruptions on transportation demand. Transp. Res. Rec. 2675 (1), 294–304. Loukaitou-Sideris, A. 2020. A Gendered View of Mobility and Transport: next Steps and Future Directions. In I Sánchez de Madariaga, M Neuman (eds.), Engendering Cities: Designing Sustainable Urban Spaces for All, pp. 19-37. London: Routledge. Lu, C., Xu, F., Dong, S., Bie, J., 2017. Observations of public bikesharing: experiences from Ningbo, China. Transp. Res. Rec. 2662 (1), 93–101. Macmichael, S., 2023. 93km air conditioned covered cycleway proposed for Dubai. Road. cc February, 3. Mateo-Babiano, I., Bean, R., Corcoran, J., Pojani, D., 2017. How does our natural and built environment affect bicycle sharing? Transp. Res. Part A 94, 295–307. Munro, C. 2019. Australian Cycling Participation: Results of the 2019 National Cycling Participation Survey. Report, Austroads. Noordzij, P.C., 1976. Cycling in the dark: an analysis of fatal bicycle accidents in the Netherlands. J. Saf. Res. 8 (2), 73–76. Painter, K., 1996. The influence of street lighting improvements on crime, fear and pedestrian street use, after dark. Landsc. Urban Plan. 35 (2-3), 193–201. Parkin, J., Ryley, T., Jones, T., 2016. Barriers to cycling: an exploration of quantitative analyses. Cycling and Society. Routledge, London, pp. 83–89. Pojani, D., Chen, J., Mateo-Babiano, I., Bean, R., Corcoran, J. 2020. Docked and Dockless Public Bike-sharing Schemes: Research, Practice, and Discourse. In C. Curtis, Handbook of Sustainable Transport, pp. 129-137. London: Edward Elgar. Prati, G., 2018. Gender equality and women’s participation in transport cycling. J. Transp. Geogr. 66, 369–375. Prati, G., Fraboni, F., De Angelis, M., Pietrantoni, L., Johnson, D., Shires, J., 2019. Gender differences in cycling patterns and attitudes towards cycling in a sample of European regular cyclists. J. Transp. Geogr. 78, 1–7. R Development Core Team (2008) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna. ISBN 3-900051-07-0, URL http://www.R-project.org. Ravensbergen, L., Buliung, R., Laliberté, N., 2019. Toward feminist geographies of cycling. Geogr. Compass 13 e12461. Rudloff, C., Lackner, B., 2014. Modeling demand for bikesharing systems: neighboring stations as source for demand and reason for structural breaks. Transp. Res. Rec. 2430 (1), 1–11. Scott, D., Lu, W., Brown, M., 2021. Route choice of bike share users: leveraging GPS data to derive choice sets. J. Transp. Geogr. 90, 102903. Soltani, A., Allan, A., Pojani, D., Khalaj, F., Mehdizadeh, M., 2022. Users and non-users of bikesharing: how do they differ? Transp. Plan. Technol. 45 (1), 39–58. https://doi. org/10.1080/03081060.2021.2017215. Stinson, M.A., Bhat, C.R., 2003. Commuter bicyclist route choice: analysis using a stated preference survey. Transp. Res. Rec. 1828 (1), 107–115. Teschke, K., Chinn, A., Brauer, M., 2017. Proximity to four bikeway types and neighborhood-level cycling mode share of male and female commuters. J. Transp. Land Use 10 (1), 695–713. Todd, J., O’Brien, O., Cheshire, J., 2021. A global comparison of bicycle sharing systems. J. Transp. Geogr. 94, 103119. Totaro Garcia, M., Zapata-Diomedi, B., Herick de Sa, T., Tiwari, G., de Nazelle, A., Tainio, M., Buehler, R., Götschi, T., Woodcock, J., 2022. Cycling behaviour in 17 countries across 6 continents: levels of cycling, who cycles, for what purpose, and how far? Transp. Rev. 42 (1), 58–81. Tu, Y., Chen, P., Gao, X., Yang, J., Chen, X., 2019. How to make dockless bikeshare good for cities: curbing oversupplied bikes. Transp. Res. Rec. 2673 (6), 618–627. Tyndall, J., 2022. Cycling mode choice amongst US commuters: the role of climate and topography. Urban Stud. 59 (1), 97–119. Winters, M., Zanotto, M., 2017. 2032 - Gender trends in cycling over time: an observational study in Vancouver. Br. Columbia J. Transp. Health 5, S37–S38. Wood, S., 2012. mgcv: Mixed GAM Computation Vehicle with GCV/AIC/REML smoothness estimation. Wygonik, E., McCormack, E., Rowe, D. 2014. Bike-Share Planning in Cities with Varied Terrain. ITE Journal 84(7):31-35. Abasahl, F., Kelarestaghi, K.B., Ermagun, A., 2018. Gender gap generators for bicycle mode choice in Baltimore college campuses. Travel Behav. Soc. 11, 78–85. Aldred, R., Woodcock, J., Goodman, A. 2016. Does More Cycling Mean More Diversity in Cycling? Transport Reviews 36(1):28-44. An, R., Zahnow, R., Pojani, D., Corcoran, J., 2019. Weather and cycling in New York: the case of Citibike. J. Transp. Geogr. 77, 97–112. Baker, L. 2009. How to Get More Bicyclists on the Road: to Boost Urban Cycling, Figure Out what Women Want. Scientific American 1 October. Bean, R., Pojani, D., Corcoran, J., 2021. How does weather affect bikeshare use? A comparative analysis of forty cities across climate zones. J. Transp. Geogr. 95, 103155. Becker, N., Rust, H.W., Ulbrich, U., 2022. Modeling hourly weather-related road traffic variations for different vehicle types in Germany. Eur. Transp. Res. Rev. 14 (1), 16. Boomsma, C., Steg, L., 2014. Feeling safe in the dark: examining the effect of entrapment, lighting levels, and gender on feelings of safety and lighting policy acceptability. Environ. Behav. 46 (2), 193–212. Braun, L., Rodriguez, D., Cole-Hunter, T., Ambros, A., Donaire-Gonzalez, D., Jerrett, M., Mendez, M., Nieuwenhuijsen, M., de Nazelle, A., 2016. Short-term planning and policy interventions to promote cycling in urban centers: findings from a commute mode choice analysis in Barcelona, Spain. Transp. Res. Part A 89, 164–183. Buczyński, A. 2023. Quality Parameters for Cycle Infrastructure: Longitudinal Gradients. Available at: https://ecf.com/users/aleksander-buczynski/trusted-content/qualityparameters-cycle-infrastructure-longitudinal-gradients. Last accessed on 30 January 2024. Butterworth, E., Pojani, D. 2018. Why isn’t Australia a Cycling Mecca? European Transport 69(4):1-22. Cervero, R., Sarmiento, O.L., Jacoby, E., Gomez, L.F., Neiman, A., 2009. Influences of built environments on walking and cycling: lessons from Bogotá. Int. J. Sustain. Transp. 3 (4), 203–226. Chen, P., Shen, Q., 2016. Built environment effects on cyclist injury severity in automobile-involved bicycle crashes. Accid. Anal. Prev. 86, 239–246. Corcoran, J., Li, T., Rohde, D., Charles-Edwards, E., Mateo-Babiano, D., 2014. Spatiotemporal patterns of a public bicycle sharing program: the effect of weather and calendar events. J. Transp. Geogr. 41, 292–305. Cox, P., 2016. From e-bikes to e-scooters; a matter of safety. Pap. Presente Velo City Glob. Taipei, Taiwan, 27 February - 1 March. Cubells, J., Miralles-Guasch, C., Marquet, O., 2023. Gendered travel behaviour in micromobility? Travel speed and route choice through the lens of intersecting identities. J. Transp. Geogr. 106, 103502. de Chardon, C.M., Caruso, G., Thomas, I., 2017. Bicycle sharing system ‘success’ determinants. Transp. Res. Part A 100, 202–214. El-Assi, W., Mahmoud, M.S., Habib, K.N., 2017. Effects of built environment and weather on bike sharing demand: a station level analysis of commercial bike sharing in Toronto. Transportation 44 (3), 589–613. Eren, E., Uz, V.E., 2020. A review on bike-sharing: the factors affecting bike-sharing demand. Sustain. Cities Soc. 54, 101882. Flynn, B.S., Dana, G.S., Sears, J., Aultman-Hall, L., 2012. Weather factor impacts on commuting to work by bicycle. Prev. Med. 54 (2), 122–124. Garrard, J. 2021. Women and Cycling: Addressing the Gender Gap. In R. Buehler, J. Pucher (eds.), Cycling for Sustainable Cities, pp. 162-176. Boston: MIT Press. Garrard, J., Handy, S., Dill, J. 2012. Women and Cycling. In J. Pucher, R. Buehler (eds.), City Cycling, pp. 211-234. Boston: MIT Press. Garrard, J., Rose, G., Lo, S.K., 2008. Promoting transportation cycling for women: the role of bicycle infrastructure. Prev. Med. 46 (1), 55–59. Gebhart, K., Noland, R., 2014. The impact of weather conditions on bicycleshare trips in Washington, DC. Transportation 41 (6), 1205–1225. Goel, R., Oyebode, O., Foley, L., Tatah, L., Millett, C. and Woodcock, J., 2022. Gender Differences in Active Travel in Major Cities Across the World. Transportation, pp.117.Halefom, T., Pullar, D., Pojani, D., Frimpong, E. 2022. How much traffic stress can cyclists endure? Case Studies on Transport Policy 10(4):2251-2261. Grudgings, N., Hagen-Zanker, A., Hughes, S., Gatersleben, B., Woodall, M., Bryans, W., 2018. Why don’t more women cycle? An analysis of female and male commuter cycling mode-share in England and Wales. J. Transp. Health 10, 272–283. Halefom, T., Pullar, D., Pojani, D., Frimpong, E., 2022. How much traffic stress can cyclists endure? Case Stud. Transp. Policy 10 (4), 2251–2261. Hanson, S., 2010. Gender and mobility: new approaches for informing sustainability. Gender, Place &. Culture 17 (1), 67–80. Hastie, T., & Tibsharini, R. (1990). Generalized Additive Models. London: Chapman & Hall. Heaney, A.K., Carrion, D., Burkart, K., Lesk, C., Jack, D., 2019. Climate change and physical activity: estimated impacts of ambient temperatures on bikeshare usage in New York City. Environ. Health Perspect. 127 (3), 037002. Hofstede Insights. 2022. Country Comparison. Webpage, available at: https://www. hofstede-insights.com/country-comparison/colombia,india,kenya/. Last accessed on 16 March 2023. 15