Using mental mapping to unpack perceived cycling risk

Provided by the author(s) and NUI Galway in accordance with publisher policies. Please cite the published version when available. Title Using mental mapping to unpack perceived cycling risk Author(s) Manton, Richard; Rau, Henrike; Fahy, Frances; Sheahan, Jerome; Clifford, Eoghan Publication Date 2016-01-04 Publication Information Manton, Richard, Rau, Henrike, Fahy, Frances, Sheahan, Jerome, & Clifford, Eoghan. (2016). Using mental mapping to unpack perceived cycling risk. Accident Analysis & Prevention, 88, 138-149. doi: https://doi.org/10.1016/j.aap.2015.12.017 Publisher Elsevier Link to publisher's version https://doi.org/10.1016/j.aap.2015.12.017 Item record http://hdl.handle.net/10379/15058 DOI http://dx.doi.org/10.1016/j.aap.2015.12.017 Downloaded 2020-06-19T05:31:01Z Some rights reserved. For more information, please see the item record link above. 1 Abstract 2 3 Cycling is the most energy-efficient mode of transport and can bring extensive environmental, 4 social and economic benefits. Research has highlighted negative perceptions of safety as a 5 major barrier to the growth of cycling. Understanding these perceptions through the application 6 of novel place-sensitive methodological tools such as mental mapping could inform measures 7 to increase cyclist numbers and consequently improve cyclist safety. Key steps to achieving 8 this include a) the design of infrastructure to reduce actual risks and b) targeted work on 9 improving safety perceptions among current and future cyclists. 10 11 This study combines mental mapping, a stated-preference survey and a transport infrastructure 12 inventory to unpack perceptions of cycling risk and to reveal both overlaps and discrepancies 13 between perceived and actual characteristics of the physical environment. Participants translate 14 mentally mapped cycle routes onto hard-copy base-maps, colour-coding road sections 15 according to risk, while a transport infrastructure inventory captures the objective cycling 16 environment. These qualitative and quantitative data are matched using Geographic 17 Information Systems and exported to statistical analysis software to model the individual and 18 (infra)structural determinants of perceived cycling risk. 19 20 This method was applied to cycling conditions in Galway City (Ireland). Participants’ (n=104) 21 mental maps delivered data-rich perceived safety observations (n=484) and initial comparison 22 with locations of cycling collisions suggests some alignment between perception and reality, 23 particularly relating to danger at roundabouts. Attributing individual and (infra)structural 24 characteristics to each observation, a Generalized Linear Mixed Model statistical analysis 25 identified segregated infrastructure, road width, the number of vehicles as well as gender and 1 1 cycling experience as significant, and interactions were found between individual and 2 infrastructural variables. The paper concludes that mental mapping is a highly useful tool for 3 assessing perceptions of cycling risk with a strong visual aspect and significant potential for 4 public participation. This distinguishes it from more traditional cycling safety assessment tools 5 that focus solely on the technical assessment of cycling infrastructure. Further development of 6 online mapping tools is recommended as part of bicycle suitability measures to engage cyclists 7 and the general public and to inform ‘soft’ and ‘hard’ cycling policy responses. 8 9 Keywords: cycling; perceived risk; safety; mental mapping 10 11 1. Introduction 12 13 Cycling safety is receiving increased attention as researchers, transport planners and cycling 14 advocates seek to increase uptake of the mode. A Stop Killing Cyclists protest (or ‘die in’) by 15 more than 1,000 cyclists in London in November 2013 dramatically highlighted the continued 16 risk of fatalities (The Guardian, 2013), and called for more suitable roads for cycling. Cyclists 17 are classed as ‘vulnerable road users’; in 2010, 1994 cyclists were killed on the roads of 20 EU 18 countries. Although cyclist fatalities in Europe have declined over the last decade, cyclists 19 remain among the most vulnerable road users. Furthermore, the decline in cycling fatalities has 20 not been as steep as for other road users, and cyclists now account for a greater proportion of 21 overall road fatalities at 7% (ERSO, 2012). 22 23 Perceived cycling safety acts as a major barrier to increasing cycling (Pucher & Dijkstra, 2000). 24 According to Parkin et al. (2007a): “While actual, or objective risk, is relatively high for cycling 25 compared with other modes, the perceived risk, that is the risk that is assumed to exist by 2 1 existing and would-be mode users, is the important criterion in terms of behavioural response”. 2 This applies equally to people’s decision to cycle at all, their choice regarding particular routes 3 (e.g. avoiding roundabouts) as well as their actual behaviour (e.g. lane position). Consideration 4 of perceived safety is also central to successful cycling design (Parkin & Koorey, 2012), yet 5 there has been a lack of research into both the objective characteristics of cycling environments 6 as well as cyclists’ perceptions of these environments (Ma et al., 2014). 7 8 Mental mapping is a research method that offers ample potential for recording and analysing 9 safety perceptions but which has not yet been fully utilised. This paper uses mental mapping 10 as part of a mixed-method approach to capture perceptions of cycling safety and their 11 relationship to the physical environment. By matching qualitative data on the perceived quality 12 of the cycling environment to quantitative and qualitative data on the physical environment, 13 the paper ‘unpacks’ major determinants of perceived cycling risk. This is tested against a case 14 study carried out in Galway, a university city in the West of Ireland. The methodology and 15 results of this paper will be relevant to engineers, planners, policymakers and cycling advocates 16 as part of an interdisciplinary response to improving actual and perceived safety and increasing 17 sustainability in transport. 18 19 2. Literature Review 20 21 2.1 Environmental Perceptions and Travel Behaviour 22 23 The relationship between environmental perceptions and spatial behaviour has interested social 24 scientists for decades. In the field of transport studies, and traffic psychology, a body of worka 25 small, but emerging, body of literature contends that attitudes, perceptions, and preferences 3 1 strongly influence individual’s travel behaviour, including recent contributions from (Spears 2 et al., (2013) and; Gehlert et al., (2013). Indeed, recent several studies have indicated that 3 attitudes towards public transport as well as concerns about personal safety and traffic all play 4 a significant role in the decision to use public transport (Elias & Shiftan, 2012). 5 6 Within tTransport studies, researchers have applied attitude and behavioural theories from 7 environmental and cognitive psychology, such as Fishbein & Ajzen’s (1975) Theory of 8 Reasoned Action (TRA) and later Ajzen’s (1991) Theory of Planned Behavior (TPB), to 9 explore the psychological dimensions of travel behaviour and modal choice. The TRA and 10 related models from the field of cognitive psychology assume that individual variables such as 11 attitudes and perceptions are the dominant drivers of behaviour (this approach has been 12 advocated for promoting bicycle use by Bamberg (2012). A number of empirical studies 13 support this contention (e.g. Thogerson (2006)). 14 15 While often contested, the influence of perceptions cannot be ignored. Geographical and 16 sociological studies of crime in cities and perceptions of neighbourhood safety (Rengert & 17 Pelfrey, 1997; Austin et al., 2002) have shown that perception is often more important than 18 objective reality in shaping people’s use of the built environment, including transport 19 infrastructure and services. However, approaches derived from the TRA and similar theories 20 have increasingly been criticised for overstating the influence of perceptions and almost 21 completely neglecting of the role of structural and contextual factors in shaping individuals’ 22 behaviour (Nye & Hargreaves, 2009; Davies et al., 2014). As a result the past decade has seen 23 the growth in perception behaviour models which incorporate contextual and situational 24 factors. For example, the premise of Spears et al.’s (2013) Perception-Intention-Adaptation 25 (PIA) model is that both cognitive processes and the physical environment have a direct effect 4 1 on travel behaviour. Similarly, Kazig and Popp (2010) have argued for a practice-theoretical 2 approach to how people orient themselves in urban spaces which combines cognitive and 3 affective aspects as well as elements of the (infra)structural context. 4 5 2.2 Cycling Risk 6 7 2.2.1 Cycling Safety and Perceptions 8 Safety is the primary factor in choosing whether to commute by bicycle (Noland, 1995; 9 Whannell et al., 2012). The major cause of cycling collisions is interaction with motorised 10 vehicles: 82% of cyclist fatalities and 87% of cycling injuries occur in collisions with motorised 11 vehicles (ERSO, 2012). Junctions pose a particular danger to cyclists: 35% of cyclist fatalities 12 take place at junctions, compared to 20% for pedestrians and 17% for car users (ERSO, 2012). 13 The main injuries to cyclists are to the legs, head and arms and the most common types of 14 injury are fractures (34%), bruising (31%) and open wounds (13%). Injured cyclists spend, on 15 average, an extra day in hospital than those injured in car collisions (ERSO, 2012) and are 16 classed as ‘vulnerable road users’. An uptake in cycling is seen as particularly important from 17 a road safety perspective as the ‘Safety in Numbers’ theory holds that the likelihood of cycling 18 collisions is inversely related to levels of cycling (Jacobsen, 2003). 19 20 Perception of cycling safety may be more important than objective reality in determining 21 uptake of cycling. These perceptions are influenced by attitudes, social norms and habits 22 (Heinen et al., 2010; Ma et al., 2014). Drivers’ attitudes to cyclists, for example, present a 23 significant barrier to cycling (Lawson et al., 2013; Wooliscroft & Ganglmair-Wooliscroft, 24 2015). Cyclists themselves consider many more factors than users of other modes (Fernández- 25 Heredia et al., 2014). Horton’s (2007) ‘fear of cycling’ goes beyond that of collisions and traffic 5 1 to include the fear of being on show, of harassment or violence, and of seeming inept or unfit. 2 Many of these fears are culturally embedded and socialised, e.g. parents constrain the travel 3 behaviour of their children based on risk perceptions (Timperio et al., 2004; Carver et al, 2010). 4 Collective perceptions of risk also manifest in social pressure to wear disliked safety clothing, 5 such as high-visibility vests and helmets (Aldred & Woodcock, 2015; Deegan, 2015); however, 6 these do not improve perceptions of safety among cyclists (Lawson et al., 2013). 7 8 To date, few studies of perceived cycling risk have included the characteristics of the cyclist 9 (e.g. age, gender and cycling frequency) (Lawson et al., 2013; Black & Street, 2014; Bill et al., 10 2015), which is a gap that this paper seeks to address. The UK Department for Transport 11 considers the perception of cycling risk as a potential barrier to cycling and includes perceived 12 cycling safety in the British Social Attitudes survey (UK DfT, 2014). 61% of people in the UK 13 consider the roads to be too dangerous to cycle on and this varies with age (47% of 18-24 y/o, 14 76% of 65+ y/o), gender (69% of women, 53% of men) and cycling experience (48% of those 15 who cycled in the last year, 67% of those who did not) (UK DfT, 2014). Several studies 16 identified age and gender as factors which influence perceptions and which also shape 17 responses to segregated cycling infrastructure (Garard et al., 2008; Black & Street, 2014; Ma 18 et al., 2014; Dill et al., 2015). Cycling experience has also been shown to influence risk 19 perceptions and Frequent inexperienced cyclists are more likely to perceive road conditions as 20 hazardous (cyclists were found to have better perceptions of the cycling environment (Ma et 21 al., 2014; Bill et al., 2015). Sanders (2015) suggests that additional experience and skills gained 22 may make these cyclists more tolerant of risks, although even experienced cyclists are 23 concerned about a variety of possible causes of injury. 6 1 and are more likely to fear more commonly reported actual collisions, while infrequent cyclists 2 are more likely to be affected by near misses (which Sanders (2015) demonstrated to have a 3 stronger effect than actual collisions). 4 5 2.2.2 Infrastructure 6 Many authors, across various disciplines, have examined the connection between the built 7 environment and cycling behaviour. Key infrastructural and traffic factors that affect perceived 8 cycling risk include: motorised traffic volume and speed, presence of cycling facilities, driving 9 lane width, number of junctions and roundabouts, pavement surface, parked cars and traffic 10 mix (Lawson et al., 2013; Bill et al., 2015). Increased perception of cycling crash risk can be 11 found in areas of low density, non-mixed land uses as opposed to compact, mixed-use 12 neighbourhoods. This was even found when the latter areas experienced greater actual crash 13 risk (Cho et al., 2009). Bicycle-friendly neighbourhoods (connected streets, low-traffic etc.) 14 improve residents’ perceptions of the environment and these residents cycle more often due to 15 these positive perceptions (Ma et al., 2014). 16 17 Major streets with shared lanes are associated with greatest perceived risk while shared-use 18 paved paths are considered the safest form of infrastructure (Winters et al., 2012). Parkin et al. 19 (2007a) found that cycling facilities at roundabouts did not reduce the perceived hazard. 20 Cycling infrastructure on roads with heavy traffic marginally reduced perceived risk, while 21 completely off-road, traffic-free routes significantly reduced this perception. Cycle tracks are 22 perceived as the safest form of cycling infrastructure, preferred to raised cycle lanes, cycle 23 lanes, and on-road in traffic in Copenhagen (Jensen et al., 2007). Approximately 45% of 24 respondents felt ‘very safe’ cycling on cycle tracks, compared to 32% on cycle lanes and 11% 25 on road in traffic. These results confirm existing evidence of cyclists’ preferences for 7 1 segregated infrastructure, although there are limits to the additional travel time that cyclists are 2 willing to spend in order to use segregated infrastructure (Sener et al., 2009; Caulfield et al., 3 2012). 4 5 2.2.3 Existing measures of cycling risk perception 6 The landscape of existing measures of cycling risk perception shows clear tendencies towards 7 infrastructural and technical considerations for practical application in traffic engineering and 8 urban design, e.g. cycling level of service (LoS), facility suitability, friendliness and 9 compatibility. The empirical backgrounds of these measures typically model infrastructural 10 and traffic factors associated with perceived risk (e.g. road width, traffic volume). Such 11 measures are useful as road sections can be rated and mapped to assist cyclists in route choice 12 and identify route sections in need of improvement. To clarify inconsistent terminology and to 13 classify measures spatially, Lowry et al. (2012) proposes three definitions: 14  ‘bicycle suitability’ (perceived comfort and safety along a linear section of road) 15  ‘bikeability’ (comfort, coherence, and convenience of a bicycle network) 16  ‘bicycle friendliness’ (laws, policies, education, bikeability of a community) 17 18 Lowry et al. identified 13 measures of ‘bicycle suitability’ developed between 1987 and 2011 19 (e.g. Bicycle Compatibility Index (Harkey et al., 1998)), which vary according to 20 infrastructural characteristics considered, points system and weighting (see Parkin & Coward 21 (2009) for a review of cycle route assessments). Factors considered in these measures are: road 22 facility type; lane width, number and markings; cycle facility type and width; motorised traffic 23 volume and speed; cyclist traffic volume and speed; percentage of heavy vehicles; presence of 24 on-street car parking; number and type of junctions/driveways; pavement condition and 8 1 presence of a curb. The factors are weighted as adjustment factors and combined to yield a 2 score for bicycle suitability or perceived comfort or perceived safety. 3 4 The data collection methods for 13 perceived cycling safety studies have also been summarised 5 by Lawson et al. (2013) to include: video recordings, video simulations, completion of a test 6 course, interviews and questionnaires (see Doorley et al. (2015) for a novel application of heart 7 rate monitors in the assessment of perceived risk). However, only two of the studies reviewed 8 by Lawson et al. (2013) considered the characteristics of the cyclists: Møller & Hels (2008) 9 and Noland (1995). Møller & Hels investigated cyclists’ perception of risk at roundabouts, 10 finding that safety perceptions are determined by a combination of the characteristics of the 11 individual cyclist (age and gender), the design of infrastructure (e.g. cycle facility) and traffic 12 volume. 13 14 2.3 Mental Mapping: Visualising Cycling Risk Perceptions 15 16 To better understand road safety perceptions among cyclists requires a combination of methods 17 of data collection and analysis that can handle both quantity and quality. Importantly, the 18 successful application of videos, computer simulations, interactive maps and other visual aids 19 points towards the key role of visualisation in road safety research (cf. Prendergast & Rybaczuk 20 (2005) for a more general discussion of visualisation in spatial planning). Mental mapping, a 21 creative process that seeks to draw out and subsequently visualise people’s experiences of their 22 physical and social surroundings, deserves particular attention in this context. 23 24 Mental maps are defined as “an amalgam of information and interpretation reflecting not only 25 what a person knows about places but also how he or she feels about them” (Johnston et al., 9 1 1986). While all maps can serve as texts for exploring human perceptions of the landscape 2 (Soini, 2001), mental maps in particular have long been associated with cartography that 3 explores human perceptions of landscape. Lynch’s (1960) research on images in the city 4 represents an early landmark study in this field that reveals how different social groups view 5 and respond to the same environment in diverse ways. Mental maps have served to explore a 6 range of subjects including perceived desirability of neighbourhoods, orientation and way- 7 finding, perceptions of crime and migration propensities (Gould & White, 1993; Fahy & Ó 8 Cinnéide, 2009). 9 10 Growing interest across a range of disciplines in representations and the social construction of 11 places has coincided with an increased appreciation of mental mapping (Gregory, 2009). From 12 a land use planning perspective, approaches incorporating mental mapping offer significant 13 advantages over survey methods or other scale-based measures because of their place-specific 14 attributes (Brown and Raymond, 2007). Indeed, Brown and Raymond (2007: 108) argue that 15 “the mapping of landscape values and special places can provide an operational bridge between 16 place attachment and applied land use planning that seeks to minimize potential land use 17 conflict”. 18 19 Research into mental maps and travel behaviour is sparse and existing studies focus 20 predominantly on travel route choice. As noted by Mondshein et al. (2010:849): “the limited 21 research to date suggests that transport infrastructure and way-finding on overlapping, distinct 22 modal networks – sidewalks, bike lanes, transit routes, local streets and roads, and freeway 23 networks – affect the development of cognitive maps and, in turn, travel behaviour”. The 24 limited research on transport and mental mapping that exists suggests that mode of transport 25 influences level of detail and quality of maps, which has significant implications for transport 10 1 planning, accessibility, and wider public policy (Mondshein et al., 2010, 2013). For cyclists, 2 Snizek et al. (2013) used mental mapping to study route experience in a ‘high cycling’ 3 environment in Denmark, whereby an online questionnaire in Google Maps allowed 4 participants to award positive and negative experience points. Their approach points to a wider 5 field of online GIS-based platforms and sensors for crowd-sourcing perceptions of cycling 6 safety and identifying localised risks (cf. Loidl (2014), Nelson et al. (2015) and Zeile et al. 7 (2015)). However, Snizek et al. (2013) did not consider the individual characteristics of the 8 cyclists and the effect that these may have on route experience. The following section details 9 our own methodological approach which responds to both opportunities and gaps identified in 10 the literature review. 11 12 3. Methodology 13 14 This study combines mental mapping, a stated-preference survey and a transport infrastructure 15 inventory to unpack perceptions of cycling risk and to make visible both overlaps and 16 discrepancies between perceived and actual safety risks. The results of mental mapping and the 17 stated-preference survey captured perceptions of the cycling environment, while a transport 18 infrastructure inventory collected characteristics of the objective cycling environment. The 19 resulting qualitative and quantitative data were matched using Geographic Information 20 Systems and exported to statistical analysis software to construct a model of the individual and 21 structural determinants of perceived cycling risk. In this context this paper makes a significant 22 contribution to cycling safety research by exploring the perceptions of cycling risk through the 23 application of mental mapping as part of a larger mixed-method study. 24 25 3.1 Study Area 11 1 2 Ireland has established a national cycling target of 10% modal share by 2020, yet safety 3 concerns remain a major impediment to increasing cycling uptake (DTTAS, 2009a; 2009b). 4 Between 2013 and 2014, there was a 27% increase in vulnerable road user deaths; there were 5 12 cyclists killed in 2014, compared to 5 in 2013. Cyclists represent 6% of all road fatalities 6 despite accounting for only 2% of road users (RSA, 2014). Issues surrounding cycling safety 7 are gaining attention in the Irish media as shown by one recent current affairs programme 8 entitled ‘The growing war between cyclists and motorists, what’s happening on our streets?’ 9 (RTÉ, 2015). This discourse has centred on conflicts between the behaviour of cyclists 10 (breaking red lights, cycling on footpaths) and the behaviour of motorists (aggression, verbal 11 abuse, speeding, dangerous driving). Short & Caulfield (2014), for example, discuss the safety 12 challenge of increased cycling and the incorporation of safety in policy. 13 14 To achieve the national cycling target, small, compact urban areas with a young population are 15 deemed to harbour significant potential for modal shift away from the car and towards active 16 travel modes. The present study was conducted in Galway, a university city of 75,000 people 17 on the west coast of Ireland. The study area is affected by a number of issues that might impede 18 uptake of cycling and a recent qualitative study that investigated modal shift among the 19 workforce of a large employer found perceived safety risks in the city to be an important barrier 20 to walking and cycling (Heisserer, 2013). Galway experiences mean annual rainfall of 1193 21 mm and the mean annual temperature is 10°C (Met Éireann, 2015). The city has a cycling 22 modal share of 5%, while 57% residents travel to work by car, either as a driver or passenger 23 (CSO, 2012). Recent cycling-related developments include the installation of raised cycle 24 lanes, a series of greenways and a bike-share scheme. 25 12 1 3.2 Survey Sampling 2 3 In this study, people in Galway City who cycle to work, school or college make up the study 4 population. Convenience sampling was utilised by presenting the paper-based survey to 5 potential participants at large events in 2013; (random sampling techniques (e.g. simple 6 random, cluster or stratified sampling) could not be generated due to the lack of a sampling 7 frame; an intercept survey was also deemed unfeasible due to the time required to complete the 8 survey). The National University of Ireland, Galway campus was chosen for its central location 9 (1 km from Galway City centre) and relatively large cycling population (cycling modal share 10 12%, campus population 17,000 students and 2,000 staff (Manton and Clifford, 2012)). As the 11 sample was not randomly selected, it was not possible to make statistical inferences about all 12 cyclists or indeed the population of this study (Smith, 1983); however, the use of non-random 13 samples does not necessarily compromise the generality of the results, allowing for interesting 14 quantitative findings to be generated (cf. Chow, 2002). 15 16 3.3 Mental Mapping 17 18 While traditional mental mapping studies asked participants to draw a freehand sketch (Lynch, 19 1960), this study utilised a base-map of Galway City roads and streets as an assist. Participants 20 were provided with one map each (which included a brief written introduction, outlining the 21 task) and coloured pens. They were asked to draw their regularly used (at least weekly) cycling 22 routes and to colour each route section according to their perception of the safety of that section 23 of their route: Green for safe, Amber for unsafe, and Red for very dangerous. The use of this 24 traffic-light sequence allowed for easy expression of risk, compared to more complex rating 25 scales. Participants found their origin and destination on the base map and translated their 13 1 mental map into coloured ratings of risk along the route. The mapping task was undertaken 2 independent of any interaction with the researcher and there were no time restrictions placed 3 on any of the participants. Participating in this mental mapping exercise offered respondents a 4 chance to reflect on their everyday cycling practices and to offer some practical local 5 improvements. 6 7 3.4 Stated-Preference Survey 8 9 Following the mental mapping exercise, participants completed a stated-preference survey of 10 28 questions that reflected the findings of the reviewed literature. Questions on participants’ 11 general cycling experience and preferences (e.g. cycling frequency, trip purpose, self-ascribed 12 cycling skill, typical infrastructure used, preferred infrastructure) preceded questions on 13 cycling safety, including involvement in road collisions. The order of questions was designed 14 to invoke the memory of any previous cycling collision before the participant answered specific 15 questions on factors affecting cycling safety, including the volume of cars passing, volume of 16 trucks passing, roundabouts, adjacent car parking, speed limits, road lane width, cycle lane 17 width, and number of junctions. Due to the level of detail involved in these questions, 18 participants were challenged to carefully consider each factor before ranking them in order of 19 importance. Finally, participants were asked to provide demographic details including: age, 20 gender, years spent living in Galway, employment status, household composition, and car 21 availability. 22 23 3.5 Transport Infrastructure Inventory 24 14 1 Data on infrastructural and traffic-based factors affecting safety were collected using a 2 transport infrastructure inventory of Galway City. These included traffic volumes (cars and the 3 proportion of HGVs), on-street car parking, cycling facilities, road width, and junctions. The 4 roads in the study area were divided into sections of similar length(generally between junctions 5 and using named roads where possible) and data on each road section were collected through 6 desk studies and site visits. The volumes of light vehicles (predominantly cars), heavy vehicles 7 (predominantly trucks) were retrieved from Galway City Council (2013), based on annual 8 traffic counts conducted between 7am and 7pm on a standard day in November. The locations 9 of adjacent car parking were identified on site and by using Google Streetview. The speed limit 10 on all roads was 50 km/h, with the exception of the NUI Galway campus, which has a speed 11 limit of 20 km/h. The locations of segregated cycling infrastructure were identified from 12 Galway City Council (2012). The widths of road and cycle lanes were measured on site. The 13 number of junctions in each road section was counted from mapping. A shapefile of the road 14 network was imported to ArcGIS and the polylines were split according to road section and 15 inventory data were then added as attributes to each road section. Limitations to the assessment 16 of perceived safety include the under-reporting of cycling collisions, the avoidance of particular 17 routes and the variation in route types and location (Parkin et al., 2007a). 18 19 3.6 Data analysis 20 21 This final stage of the empirical part of the study constructed a model of perceived cycling risk 22 by matching the perceived environment (mental map) to characteristics of the physical 23 environment (inventory data). Mental maps were uploaded to ArcGIS by attributing the colour- 24 coded ratings of each participant (along with demographic information) to road sections (cf. 25 Boschmann & Cubben (2014) for sketch maps and qualitative GIS, and Snizek et al. (2013) for 15 1 map matching). This yielded a dataset in which each row represents one observation (the rating 2 given by one participant to one road section); this dataset was then imported into the statistical 3 software package SPSS (version 21) for analysis. The perceived risk rating is the response of 4 interest and is a qualitative variable with values Green, Amber, Red in order of increasing 5 perceived risk. Factors (qualitative/categorical input variables) and covariates (quantitative 6 input variables) include the physical characteristics of the road section and the demographics 7 of the individual participant. A statistical model was then developed to identify the significant 8 factors and covariates in perceived cycling risk. 9 10 A number of features associated with the study design posed challenges for the model. Firstly, 11 the response data are qualitative and ordinal. Secondly, as each participant rated several roads, 12 observations for any given participant may be correlated. Thirdly, interactions between several 13 of the variables can (as in any study) also arise. Of particular interest here are the interactions 14 between individual-level and infrastructural variables. The presence of a significant interaction 15 would imply that the effect of one independent variable (e.g. an infrastructural characteristic) 16 on perceived risk, which is a dependent variable, differs according to a second independent 17 variable (e.g. a characteristic of the cyclist). Also some variables can seriously mask the effect 18 of others (e.g. when present, multicollinearity may have such a masking or other adverse effect) 19 and it was considered appropriate to exclude certain variables (e.g. fitness) from the analysis. 20 Bearing in mind the design and goals of the study, it was decided to employ logistic regression 21 and to adjust the technique for the above mentioned possibility of correlations between 22 participants’ ratings and allow interactions between input variables. A Generalised Linear 23 Mixed model was applied to investigate multi-category responses that could accommodate the 24 within-subject correlation through random effects (McCullogh et al., 2008). Interaction terms 16 1 were introduced for all two-way interactions and then excluded on the basis of lack of 2 significance at the 5% level. 3 4 Red (dangerous) was chosen, arbitrarily, as the reference category for the response variable, 5 Rating. Following SPSS’s mixed model analysis for multinomial regression, the (multinomial) 6 logistic model employed models: 7 8 9 ln⁡( probability⁡that⁡a⁡random⁡person⁡will⁡respond⁡𝐺𝑟𝑒𝑒𝑛⁡or⁡𝐴𝑚𝑏𝑒𝑟 ) probability⁡that⁡the⁡person⁡will⁡respond⁡Red 10 as a linear function of variables representing the factors and of the covariates, along with a 11 random error term. The coefficient, βi, of a covariate, Xi, (such as age and road width) 12 represents the change in the above log-odds for a unit increase in that variable; while for a 13 binary input variable (such as gender or segregation) the coefficient of that variable represents 14 the expected change in the log-odds between the reference category of that variable to the other 15 category. For the only input variable which has three categories, cycling experience, there were 16 two parameters involved to represent changes from the reference to each of the other two 17 categories (i.e. from inexperienced to competent and highly skilled). 18 19 For most input variables, of interest is whether a change in levels of this variable increases the 20 log-odds (rather than changes the log-odds); that is, tests for which the alternative/research 21 hypothesis is one-sided, e.g. are women are more likely than men to perceive cycling risk (as 22 suggested by the literature) rather than simply whether there is any difference between men 23 and women in perceiving cycling risk. For other input variables (such as age), a two-sided 24 hypothesis test is applied (the p-value for a one-sided hypothesis test is half that of a two-sided 25 test). In practice, it may be easier for interpretation purposes to exponentiate the log-odds ratios, 17 1 so that then the linear function described above is replaced by an exponentiated version and 2 one can carefully interpret the corresponding coefficients as pertaining to changes in odds 3 rather than changes in log-odds. While the analyses illustrated in this study demonstrates the 4 potential major factors in determining perceived cycling safety, the fact that our data was not 5 strictly generated by a probabilistic sampling design method, and the fact that variations of 6 models that were fitted (e.g. different ways of modelling within-cyclist correlation) gave 7 slightly different results for the significance or non-significance of certain variables, it is 8 suggested that the hypothesis test results below may best be viewed as exploratory and as 9 suggestions of approaches to be pursued on new data by future researchers rather than as 10 ‘definitive’ statistical inferential conclusions. 11 12 4. Results and Discussion 13 14 4.1 Sample Characteristics 15 16 The number of survey participants was 104 and the total number of observations (i.e. perceived 17 risk ratings) was 484, an average of 4.65 observations per participant. The average distance 18 (subsequently included in the analysis) rated per participant was 1.95 km. Participants’ ages 19 ranged from 17 to 58 years (mean = 30.8 years; standard deviation= 10.7 years). The majority 20 of participants were male, 60.6%, and this reflects the national cycling gender gap – in Ireland 21 73% cyclists are male (CSO, 2012). The sample included 36% people at work, 36% 22 undergraduate students, 21% postgraduate students, and 6% other employment statuses. More 23 than half of the participants cycle everyday (51%), a further 29% cycle several times per week 24 and the remaining 20% cycle less often. 29% of cyclists in the study classified themselves as 25 highly skilled, 64% as competent and 7% as inexperienced. 14% of the sample classified 18 1 themselves as very fit, 51% as fit, 29% as of average fitness and 6% as unfit. The majority of 2 participants (61%) had not been involved in a collision as a cyclist. The most common 3 motivation for cycling purpose was commuting, followed by leisure, and health/fitness. 4 5 4.2 Perceived Environment 6 7 A total of 38 road sections in Galway City received a rating. Only road sections with a 8 minimum number of ten ratings were included (as road sections will be compared with respect 9 to a set of variables rather than compared to each other on the basis of rating, this sample size 10 was considered satisfactory), leaving 27 road sections in the final analysis. The average length 11 of these road sections was 419 metres and the total length of road network included in the 12 analysis was 11 km. The River Corrib divides Galway City approximately in half, east and 13 west. As the NUI Galway campus and the majority of residences are located west of the river, 14 road sections at that side of the city received the majority of ratings. The most frequently rated 15 roads were in the immediate vicinity of the university. Figure 1 shows a sample mental 16 mapping response across a route from Salthill, a seaside suburb, to the university at the banks 17 of the river. The start (residential roads) and end (canal towpath and university roads) are rated 18 as Green (safe), while one road section is coloured Amber and another Red. 19 19 1 2 Figure 1 – Sample mental mapping response (Male, 31 years old) 3 4 Of the 484 road section ratings, almost half (48.6%) were Green, 29% were Amber and 22% 5 were Red. This suggests that the majority of roads are perceived to be unsafe or very dangerous. 6 Furthermore, and route choice, whereby cyclists avoid dangerous roads, is likely tocould mask 7 the true extent of this perceived risk (Snizek et al., 2013). Of interest here is the relative 8 influence of individual and infrastructural factors in determining this ordinal rating. For 9 illustrative purposes in Figure 2, the three response colours have been weighted with values 1, 10 5 and 10 in order of increasing perceived risk. Averaging these values and forming three 11 equally-sized categories allows a rough comparison of perceived risk across the road network. 12 20 River Corrib 1 2 Figure 2 – Galway City road network, indicative perceived safety ratings and locations of 3 cycling collisions 4 5 Also shown in Figure 2 are the locations of the 32 reported collisions involving cyclists in 6 Galway City in 2005, 2006, 2007, 2008 and 2010 (RSA, 2014). There were no cyclist fatalities 7 in Galway in this period though it is believed that cycling collisions are subject to major under- 8 reporting (Short & Caulfield, 2014). In the absence of more reliable measures (e.g. collision 9 intensity), this source of cycling collisions was judged to be an acceptable but basic 10 representation of actual cycling risk. Of the 32 collisions, 23 occurred on road sections included 11 in this study. Four collisions align with the safe category, 15 with the unsafe category and four 12 with the very dangerous category (all at roundabouts). It is interesting that all of the collisions 13 on road sections perceived as very dangerous actually took place at roundabouts, though it 14 should be noted that the weighting system yielded just three very dangerous road sections other 21 1 than roundabouts. Roundabouts were rated as very dangerous by all participants and require 2 further research for cycling safety. Within the limitations of the arbitrary weighting of response 3 colours and the under-reporting of cycling collisions, this suggests that some perceptions of 4 risk align with location of actual collisions. This is envisaged as part of a complex connection 5 between perception and reality, whereby actual risks play some role in influencing cyclists’ 6 risk perceptions, although a linear relationship is not necessarily implied. 7 8 4.3 Physical Environment 9 10 The transport infrastructure inventory compiled the engineering and traffic characteristics of 11 the 27 road sections covered by mental mapping. Traffic volumes ranged between 0 (canal 12 towpath) and 14,791 vehicles per day, the proportion of HGVs between 0–4%, road lane width 13 between 2–4 m. There were two types of segregated cycling infrastructure: raised cycle lanes 14 and the canal towpath (Figure 3). On-street car parking is available in some areas and the 15 number of junctions ranged from two to nine. Images of typical types of road and cycling 16 infrastructure in Galway City are shown in Figure 3. 17 22 1 2 3 Figure 3 – Clockwise from top left: new raised cycle lane on main road, canal towpath, 4 typical roundabout, and a road without cycle facilities (Google, 2015) 5 6 4.4 Stated Preferences 7 8 Participants were asked to rank nine physical factors according to their impact on cycling 9 safety. Based on the number of 1st, 2nd and 3rd rankings, three of the major safety concerns were 10 found to be traffic-related: the number of trucks passing, speed of traffic and number of cars 11 passing. Infrastructure proved to be less of a concern than traffic; and cyclists consider the 12 presence of a roundabout, the width of the road lane and the presence of an adjacent car parking 13 lane to be the most concerning characteristics of infrastructure. Other factors expressed in 14 qualitative responses included road condition and driver behaviour. 23 1 2 Following the ranking of safety concerns, participants were then asked whether they felt two 3 types of traffic (trucks and cars) and two elements of infrastructure (roundabout and car 4 parking) compromised their safety while cycling, gauged on a 5-point Likert scale. 59.2% 5 agreed that the number of trucks passing compromised their cycling safety, while 54.5% agreed 6 that the number of cars passing was a major issue. 42.6% are deterred concerned by the 7 presence of a roundabout, but adjacent car parking, which can result in ‘dooring’, deterred 8 concerned just 14.9% of participants. The maximum speed limit of a road that most participants 9 (57%) would feel comfortable sharing with motorised traffic is less than 50 km/h, 26% said 10 50-60 km/h and 17% said 60-80 km/h. 11 12 Participants were asked to rank their frequency of use and preferred type of cycling 13 infrastructure or on-road cycling positions. Figure 4 shows the results of the participants’ actual 14 riding locations and shows that reasonable numbers always cycle on-road, mostly in the 15 secondary riding position (closer to the kerb, rather than ‘taking the lane’). Some participants 16 stated that they always cycle on the footpath, potentially indicating significant fear of 17 interaction with traffic. Figure 4 also shows the participants’ preferred cycling locations with 18 raised cycle lanes (footpath level), road-level cycle lanes and greenways receiving the highest 19 rankings. The disparity between this clear preference for segregated cycling infrastructure and 20 actual levels of on-road cycling suggests a deficit of dedicated cycling infrastructure, a finding 21 in line with Caulfield et al. (2012). 22 24 % Participants 50 45 40 35 30 25 20 15 10 5 0 47.5 Always use Prefer to use 23.2 23.0 5.7 21.3 15.4 16.4 14.0 9.2 1.6 4.9 On-road On-road Road-level Raised On the (secondary (primary cycle lane cycle lanes footpath pos.) pos.) 9.4 2.5 5.8 Shared Off-road bus-cycle greenway lane 1 2 Figure 4 – Actual and preferred cycling infrastructure 3 4 Finally, the impact of participants’ route choice must be considered. Cyclists may avoid roads 5 that they identify as dangerous, e.g. those with heavy traffic. This would lead to a disparity 6 between stated preference results and mental mapping results, as cyclists may not use the roads 7 they perceive to be most dangerous. However, this was not determined to be significant factor 8 in this survey as the mental mapping results show that the vast majority of participants chose 9 the most direct route between origin and destination, most likely due to the lack of route choice 10 in Galway City which does not have a grid pattern. Many cyclists will also temper safety 11 concerns with time and distance delays caused by alternative routing. 12 13 4.5 Modelling Perception of Cycling Risk 14 15 A Generalised Linear Mixed Model was built in SPSS, where the Subject was the participant 16 (using a unique participant number to identify repeated measurements) and the Target was the 17 perceived risk rating. The Measurements were the 484 observations, including associated 18 demographic and infrastructural data. The goal was to assess the extent to which the ordinal 25 1 variable Rating relates to nine main qualitative and qualitative effects (Table 1). The qualitative 2 variables are: gender, cycling experience [inexperienced/competent/highly skilled], 3 segregation [of cycling facility; yes/no], parking [adjacent car parking; yes/no]. The 4 quantitative variables are: age, LV [per 1000 light vehicles per day], %HV [percentage of heavy 5 goods vehicles], width[of road lane in metres], and number of junctions (Table 1). 6 Table 1 – Variable information 7 Variable Category Green Rating Amber Red Female Gender Male Highly Skilled Qualitative Cycling experience Competent Inexperienced Not Segregated Segregation Segregated No Parking Parking Parking Age (years) LV (1000 veh) Quantitative %HGV Width (m) Junctions (no.) n Percent Minimum Maximum 235 48.6 141 29.1 108 22.3 189 39.0 295 61.0 160 33.1 298 61.6 26 5.4 324 66.9 160 33.1 230 47.5 254 52.5 484 17 58 484 0 15 484 0 3.9 484 2 4 484 2 9 8 9 Figure 5a displays the percentage of participants for each category of gender. These results 10 suggest that female participants perceived more roads as very dangerous and fewer roads as 11 safe (of course, this is not a statistical inference and has not removed the effect of other 12 variables). Figure 5b illustrates the corresponding summary for segregation, which appears to 13 have a strong effect: dedicated cycling facilities received a larger proportion of safe ratings 14 than road sections that involve cycling in motorised traffic. Chi-squared tests showed that there 15 is a significant relationship between gender and rating (X2 = 6.632, p-value = 0.036) and 16 between segregation and rating (X2 = 48.033, p-value = 0.000) (of course, these tests have not 26 1 removed the effect of other variables). Both of these observations were also suggested by the 2 literature and the potential interaction of individual and infrastructural variables is also of 3 interest. For example, female participants rated a greater proportion of segregated infrastructure 4 than their male counterparts – potentially as they are more likely choose a route on segregated 5 infrastructure – as did older people and inexperienced cyclists. 6 7 8 Figure 5 – Rating plotted against Gender (left) and Segregation (right) 9 10 To account for interactions between pairs of variables, all two-way interaction terms were 11 initially included in the analysis and then systematically dropped according to their effect on 12 the significance of main effects. Some variables have the potential to mask the effect of others 13 and it was deemed necessary to exclude these. Fitness, for example, was dropped at an early 14 stage of the analysis as it was found to be highly correlated with, and masking the effect of, 15 Cycling Experience; this was also the case with Years Living in Galway and Age. Random 16 Effects were included to account for within-subject correlations. The fitted Generalized Linear 17 Mixed Model components are shown in Table 2. In this table, each coefficient, 𝛽̂ , estimates the 18 change in the log-odds of Green or Amber relative to Red for a unit increase in a quantitative 19 variable (units are denoted in parenthesis for quantitative variables) and as the change in the 27 1 log-odds between the reference and the other category (or other categories) for qualitative 2 variables. The exponentiated log-odds ratio, Exp(𝛽̂ ), then represents changes in odds; the 95% 3 confidence interval for the true underlying odds, Exp(𝛽̂ ), is also shown in Table 2. Significance 4 is implied by the magnitude of the p-value, displayed in Table 2 for two-sided hypothesis tests 5 and is halved for cases where the alternative hypothesis is one-sided. 6 7 Table 2 – Generalized Linear Mixed Model outputIndividual and infrastructural effects on 8 perceived cycling risk Ref=Red ̂ 𝜷 ̂) Exp(𝜷 Individual characteristics 0.022 1.024 9 Age (years) Gender [ref=Male] Female 1.526* 4.601 Cycling Experience [ref=Inexperienced] Highly Skilled -1.563* 0.210 Competent -1.694* 0.184 Infrastructural characteristics LV (1000 vehicles) 0.176** 1.192 HV (percent) 0.304 1.355 Width (m) -0.977* 0.377 Junctions (number) 0.006 1.006 Parking -0.521 0.594 Segregation -2.993** 0.050 Interactions Age*[Segregation] 0.070* 1.072 %HV*[Gender = Female] -0.500* 0.607 *Significant at the 5% level; **Significant at the 1% level 95% CI for Exp(𝛽) Lower Upper p-value 0.984 1.066 0.240 1.336 15.847 0.008 0.045 0.043 0.982 0.787 0.024 0.012 1.076 0.903 0.153 0.873 0.266 0.009 1.321 2.035 0.929 1.159 1.325 0.269 0.001 0.142 0.034 0.932 0.203 0.001 1.029 0.379 1.118 0.971 0.001 0.037 10 11 Individual characteristics 12 13 The coefficient for gender in the fitted model in Table 2 is 𝛽̂ = 1.526 and the corresponding 14 would increase by 1.526 for a female relative to a male (or equivalently, the estimated odds of 15 belonging to Red relative to the reference value Green or Amber is for a female 4.6 times larger exponentiated value is exp⁡(𝛽̂ ) = 4.6. This means that the estimated log odds of choosing Red 28 1 than its value for a male), when the other input variables are held constant. In other words, 2 female respondents are significantly more likely to rate a road section as dangerous than are 3 their male counterparts.1 Turning to cycling experience, being a highly skilled or competent 4 cyclist decreased the odds of perceiving risk by a factor of 0.18 (p-value = 0.024) and 0.21 (p- 5 value = 0.012), respectively, compared to inexperienced cyclists. Significant interactions were 6 found between age and segregation and between gender and %HV. These interactions confirm 7 the hypothesis that the effect of some infrastructural variables differs with individual 8 characteristics, but complicate the interpretation of the main effects. These results regarding 9 gender and cycling experience confirm the findings of several other studies (Lawson et al., 10 2013; Black & Street, 2014; Ma et al., 2014; Bill et al., 2015; Dill et al., 2015). Future transport 11 policymakers and planners should thus consider the roles of gender and the lack of cycling 12 experience in the promotion of cycling. 13 14 Infrastructural characteristics 15 Of the six infrastructural variables, the number of cars (LV), width of the road lane, and cycling 16 segregation were significant. The odds of rating a road section as dangerous decreased with 17 width by a factor of 0.38 (p-value=0.01) for each additional metre. The number of cars passing 18 increased the odds of perceptions risk by a factor of 1.2 (p-value <0.005) for each 1000 19 vehicles. Segregation had a particularly strong effect (Exp(𝛽̂ ) = 19.9, p-value <0.005): the 20 presence of a segregated cycling facility significantly increased perceptions of safety. These 21 findings confirm existing research on cyclists’ preferences for segregated infrastructure 22 (Caulfield et al., 2012; Lawson et al., 2013) as well as policy and advocacy for reduced 23 motorised traffic volumes and increased overtaking distancesroad space for cycling. However, When 𝛽 is the corresponding true log odds, consider testing the null hypothesis 𝐻0 : 𝛽 = 0 versus the (one-sided) alternative hypothesis⁡𝐻1 : 𝛽 > 0, or equivalently testing the alternatives 𝐻0 : exp(𝛽) = 1 versus 𝐻1 : exp⁡(𝛽) > 1, the p-value associated with the estimate is 0.008. 1 29 1 it is important to note that additional road lane width is unlikely to yield benefits for cycling 2 safety as motorists typically adapt their behaviour to these conditions by increasing speed (cf. 3 Lewis-Evans & Charlton (2006)). 4 5 Choice of model 6 The Generalized Linear Mixed Model (GLMM) correctly predicted 92% of Green (safe) 7 responses and the overall percentage correctly predicted was 67%. Two other models were 8 developed, namely multinomial logistic and ordinal logistic. Both of these models gave the 9 same results in terms of significance of the various factors and covariates but differed from the 10 mixed model multinomial logistic analysis in that segregation and the interaction between 11 %HV and gender each lost its significance. It is interesting to note that the mixed model 12 employed, a multinomial logistic, has allowed for possible correlation between observations 13 on the same person, whereas the (non-mixed) multinomial and ordinal logistic models assume 14 independence of all response observations. Future research could explore which model is more 15 appropriate for the analysis of data from this study design. 16 17 While the analyses illustrated in this study demonstrates the potential major factors in 18 determining perceived cycling safety, the fact that ourthe data wereas not strictly generated by 19 a probabilistic sampling design method, and the fact that variations of models that were fitted 20 (e.g. different ways of modelling within-cyclist correlation) gave slightly different results for 21 the significance or non-significance of certain variables, it is suggested that the hypothesis test 22 results belowabove may best be viewed as exploratory and as suggestions of approaches to be 23 pursued on new data by future researchers rather than as ‘definitive’ statistical inferential 24 conclusions. Overall, it is envisaged that the innovative methodology developed in this paper 25 has opened up a fruitful avenue for further mixed-method cycling safety research. 30 1 2 5. Conclusions 3 4 Perceived cycling risk has the potential to overshadow objective cycling risk as the major 5 barrier to increasing uptake of cycling. Perceptions of cycling have received substantial 6 academic attention over recent years; however, this work has focused on infrastructural 7 determinants of perceived risk and rarely considers the characteristics of the cyclist. This study 8 draws on attitude and behaviour theory to argue that cycling perceptions exist within a broader 9 model of attitudes, social norms and habits (Heinen et al., 2010) that need to be understood and 10 that new quantitative and qualitative methods are required to explore perceptions of risk. The 11 paper presents mental mapping, a stated-preference survey and a transport infrastructure 12 inventory to unpack perceptions of cycling risk and to make visible both overlaps and 13 discrepancies between perceived and actual characteristics of the physical environment. While 14 the more ‘traditional’ self-reported survey uncovered significant data related to perceptions of 15 cycling risk, we argue that the data derived from the mental mapping approach has the potential 16 to provide a more specific, placed-based assessment of these risks. 17 Upon critical reflection, the resulting maps display a snapshot of the geographical distribution 18 of selected elements but exclude cyclist’s in-depth cycling knowledge and experiences. Further 19 work is needed to include these qualitative aspects in analyses and debates regarding perceived 20 and actual cycling safety. 21 22 Participants’ mental maps (n=104) delivered rich perceived safety data (n=484) and initial 23 comparison with locations of cycling collisions showed alignment between perception and 24 actual conditions, particularly relating to danger at roundabouts. Attributing individual and 25 infrastructural characteristics to each observation, a Generalized Linear Mixed Model 31 1 subsequently identified segregated infrastructure, road width and traffic volume as well as 2 gender and cycling experience as significant. These results confirm previous research on 3 participants’ stated preferences and suggest interactions between the characteristics of the 4 cyclist and infrastructural conditions in the perception of cycling risk. Future data collection 5 could consider randomly-selected samples and more controlled physical environments to better 6 understand these interactions. 7 8 While the size and nature of the sample does not allow for inferences about the wider 9 population of cyclists, the findings nevertheless confirm observations made in cycling safety 10 documents and contributions to cycling policy by cycling campaigners and lobby groups in 11 low-cycling countries such as Ireland and the UK. Regarding cycling in traffic, these include 12 calls for reductions in traffic speeds and volumes, as well as for changes to legislation, such as 13 an increase in overtaking clearance distance to 1.5 m. This study also contributes to the 14 integration-segregation debate by demonstrating the importance of segregation for reduction 15 in perceived risk (cf. Parkin et al., 2007a; 2007b). Gaps between participants’ stated 16 preferences and actual cycling behaviour suggest a segregated cycling infrastructural deficit in 17 the city under study, whereby most would prefer to cycle in cycle lanes, yet in practice cycle 18 on road in traffic. Cyclists are a heterogeneous group however and characteristics such as 19 gender and cycling experience influence risk perceptions and infrastructure preferences. 20 Segregated infrastructure may well bring safety benefits for large sections of the population, 21 but space restrictions, indirect routes and junction requirements mean that sharing the road with 22 motorised traffic remains cyclists’ primary means of negotiating urban areas. A combination 23 of carefully-designed dedicated-space for cycling and making roads safer for cycling, for 24 example by reducing traffic speeds and volumes, is recommended for improving safety 25 perceptions among current and future cyclists. 32 1 2 Moving beyond a focus on infrastructural provision, the findings presented in this paper have 3 significant implications for future cycling policy. As previous research reveals, misconceptions 4 among different groups of road users continue to negatively affect the safety of vulnerable 5 groups and remain a source of tension. The Irish government's target for 10% cycling modal 6 share by 2020 requires a serious commitment to changing current attitudes and improving 7 interactions between motorised vehicles and cyclists. National policy initiatives could be 8 designed to both dispel prevailing perceptions of risks and raise awareness of the vulnerability 9 of non-motorised road users. Furthermore, interventions could be targeted at those user groups, 10 for example women, which are particularly sensitive to perceptions of cycling risk (cf. Garard 11 et al. (2012)) as part of broader policy of dismantling the ‘fear of cycling’. 12 13 The mixed method used in this study is a reflection of the interdisciplinary nature of the project 14 team, drawn from civil engineering, sociology, geography and statistics. There is clearly 15 potential to further develop the mapping and matching method as well as other mixed-method 16 approaches in transport studies in the future. Indeed, there is a dearth of research exploring how 17 transport brings individuals into cognitive and physical contact with their built environments 18 (Mondshein et al., 2013), and this study has shown that mental mapping has latent potential as 19 a research tool in this respect. Building on the success of this method, further research is 20 recommended on bicycle suitability measures and online mapping tools. Engaging cyclists and 21 the general public through GPS-based mobile applications and the crowd-sourcing of data, 22 including elements of mental mapping, can further unpack perceptions of cycling risk and feed 23 into ‘soft’ and ‘hard’ cycling policy responses. 24 25 Acknowledgements 33 1 This research was funded by NUI Galway through the College of Engineering & Informatics 2 Postgraduate Fellowship Scheme and by NUI Galway Students’ Union through the Explore 3 Innovation Initiative. 4 5 References 6 Ajzen, I. (1991).The Theory of Planned Behavior. Organizational Performance and Human 7 8 9 10 11 12 13 Decision Processes, 50, pp.79–211. Aldred, R. and Woodcock, J. (2015). Reframing safety: an analysis of perceptions of cycle safety clothing. Transport Policy, 42, pp.103-122. Austin, M.A., Furr, A.L. and Spine, M. (2002). The effects of neighbourhood conditions on perceptions of safety. Journal of Criminal Justice, 30, pp.417-427. Bamberg, S. (2012). Understanding and promoting bicycle use – insights from psychological research. Cycling and Sustainability, pp.219-246. 14 Bill, E., Rowe, D. and Ferguson, N. (2015). Does experience affect perceived risk of cycling 15 hazards? Scottish Transport Applications and Research (STAR) Conference. Glasgow, 16 UK. 20th May. 17 Black, P. and Street, E. (2014). The power of perceptions: exploring the role of urban design 18 in cycling behaviours and healthy ageing. Transportation Research Procedia, 4, pp.68- 19 79. 20 Boschmann, E.E. and Cubben, E. (2014). Sketch maps and qualitative GIS: using cartographies 21 of individual spatial narratives in geographic research. The Professional Geographer, 22 66(2), pp.236-248. 23 24 Brown, G. and Raymond, C. (2007). The relationship between place attachment and landscape values: towards mapping place attachment. Applied Geography, 27, pp. 89–111. 34 1 Carver, A., Timperio, A. and Hesketh, K. and Crawford, D. (2010).Are children and 2 adolescents less active if parents restrict their physical activity and active transport due to 3 perceived risk? Social Science & Medicine, 70, pp.1799-1805. 4 Cho, G., Rodriguez, D.A. and Khattak, A.J. (2009).The role of the built environment in 5 explaining relationships between perceived and actual pedestrian and bicyclist safety. 6 Accident Analysis and Prevention, 41, pp.692-702. 7 8 9 10 Chow, S.L. (2002). Issues in Statistic Inference. History and Philosophy of Psychology Bulletin, 14(1), pp.30-41. Central Statistics Office (CSO) (2012).Results of Census 2011 – Profile 10 Door to door – Commuting in Ireland. Cork, Ireland. 11 Davies, A, Fahy, F. and Rau, H. (2014).Challenging Consumption. In Davies, A.R., Fahy, 12 F. and Rau, H. (Eds.), Challenging Consumption: pathways to a more sustainable 13 future. London: Routledge, pp.3-19. 14 15 16 17 18 19 Deegan, B. (2015). Mapping everyday cycling in London. In: P. Cox (Ed.), Cycling Cultures, Chester: University of Chester, pp.106-129. Department of Transport, Tourism and Sport (DTTAS) (2009a). Smarter Travel. Dublin, Ireland. Department of Transport, Tourism and Sport (DTTAS) (2009b). National Cycle Policy Framework. Dublin, Ireland. 20 Dill, J., Goddard, T., Monsere, C.M. and McNeil, N. (2015). Can protected bike lanes help 21 close the gender gap in cycling? Lessons from five cities. Annual Conference of the 22 Transportation Research Board (TRB). Washington DC, USA. 11-15th January. 23 Doorley, R., Pakrashi, V., Byrne, E., Comerford, S., Ghosh, B. and Groeger, J.A. 24 (2015).Analysis of heart rate variability amongst cyclists under perceived variations of 25 risk exposure. Transport Research Part F, 28, pp.40-54. 35 1 2 3 4 5 6 7 8 9 10 11 Elias, W. and Shiftan, Y. (2012). The influence of individual’s risk perception and attitudes on travel behaviour. Transportation Research Part A, 46, pp.1241-1251. European Road Safety Observatory (ERSO) (2012). Traffic safety basic facts 2012 – cyclists. European Commission, Brussels, Belgium. Fahy, F. and Ó Cinnéide, M. (2009). Re-Mapping the urban landscape: community mapping – an attractive prospect for sustainability? Area, 41, pp.167-175. Fernández-Heredia, Á., Jara-Díaz, S. and Monzón, A. (2014). Modelling bicycle use intention: the role of perceptions. Transportation, published online. Fishbein, M. and Ajzen, I. (1975). Belief, Attitude, Intention, and Behavior: An Introduction to Theory and Research. Addison-Wesley, Reading, UK. Galway City Council (2013). Galway traffic counts 2012: Manual classified junction counts. 12 Conducted by Abacus Transportation Surveys for Galway City Council. 13 Galway City Council (2012). Walking and cycling strategy. Galway, Ireland. 14 Garrard, J., Rose, G. and Lo, S. K. (2008). Promoting transportation cycling for women: the 15 16 17 18 19 role of bicycle infrastructure. Preventive Medicine, 46(1), pp.55-59. Garrard, J., Handy, S, and Dill, J. (2012). Women and Cycling. In: J. Pucher and R. Buehler, (Eds.) City Cycling, Cambridge, MA: MIT Press, pp. 211-234 Gehlert, T., Dziekan, K., and Garling, T.(2013).Psychology of sustainable travel behaviour. Transportation Research Part A, 48, pp.19-24. 20 Google (2015). Google Maps. Available at: maps.google.com [Accessed 1st July 2015]. 21 Gould, P. and White, R. (1993). Mental maps (3rded.). Boston; London: Allen & Unwin. 22 Gregory, D. (2009). The Dictionary of Human Geography Wiley-Blackwell, Oxford (5th 23 24 25 Edition). Harkey, D.L., Reinfurt, D.W. and Knuiman, M. (1998). Development of the Bicycle Compatibility Index. Transportation Research Record, 1636, pp.13-20. 36 1 2 Heinen, E., Van Wee, B. and Maat, K. (2010). Commuting by bicycle: an overview of the literature. Transport Reviews, 30(1), pp.59-96. 3 Heisserer, B. (2013). Curbing the Consumption of Distance? A practice-theoretical 4 investigation of an employer-based mobility management initiative to promote more 5 sustainable commuting. NUI, Galway: Unpublished PhD thesis. 6 Horton, D. (2007). Fear of cycling. Cycling and Society, pp.133-152. 7 Hunt, J.D. and Abraham, J.E. (2007). Influences on Bicycle Use. Transportation, 34, pp.453- 8 9 10 11 12 13 14 15 470. Jacobsen, P.L. (2003). Safety in numbers: more walkers and bicyclists, safer walking and bicycling. Injury Prevention, 9(3), pp.205-209. Jensen, S.U., Rosenkilde, C. and Jensen, N. (2007). Road safety and perceived risk of cycle facilities in Copenhagen. Presentation to AGM of European Cyclists Federation. Johnston, R .J., Gregory, D., Pratt, G. and Watts, M. (eds.) (1986). The Dictionary of Human Geography. Oxford (Blackwell) 2nd edition. Kazig, R. and Popp, M. (2010). Unterwegs in fremden Umgebungen: Ein praxeologischer 16 Zugang zum „wayfinding“ von Fußgängern. Raumforschung und Raumordnung, 69(1), 17 3-15. 18 19 20 21 22 23 24 25 Lawson, A.R., Pakrashi, V., Ghosh, B. and Szeto, W.Y. (2013). Perception of safety of cyclists in Dublin City. Accident Analysis and Prevention, 50, pp.499-511. Lewis-Evans, B., & Charlton, S. G. (2006). Explicit and implicit processes in behavioural adaptation to road width. Accident Analysis & Prevention, 38(3), pp.610-617. Liang, K-Y.and Zeger, S.L. (1986). Longitudinal data analysis using generalized linear models. Biometrika, 73(1), pp.13-22. Loidl, M. (2014). How GIS can help to promote safe cycling. Proceedings of the 28thEnviroInfo Conference. Olderburg, Germany. 10-12th September. 37 1 Lowry, M., Callister, D., Gresham, M. and Moore, B. (2012). Assessment of Communitywide 2 Bikeability with Bicycle Level of Service. Transportation Research Record, 2314, pp.41- 3 48. 4 5 Lydon, M. (2003). Community mapping: the recovery (and discovery) of our common ground. Geomatica, 57, pp.131-44. 6 Lynch, K. (1960). The Image of the City. Cambridge MA: MIT Press. 7 Ma, L., Dill, J. and Mohr, C. (2014). The objective versus the perceived environment: what 8 matters for bicycling? Transportation, 41(6), pp.1135-1152. 9 Manton, R. and Clifford, E. (2012). A study of travel patterns to NUI Galway: lessons for 10 Smarter Travel in Universities. Irish Transport Research Network (ITRN) Conference. 11 University of Ulster, Jordanstown, UK. 29-30th August. 12 13 McCulloch, C. E.andNeuhaus, J. M. (2003). Generalized linear mixed models. John Wiley & Sons, Ltd. 14 Met Éireann (2015). Monthly data for Athenry weather station. Irish Meteorological Service 15 Online. Available at: http://www.met.ie/climate/monthly-data.asp?Num=1875 (accessed 16 2nd July 2015). 17 18 19 20 21 22 Møller, M. and Hels, T. (2008).Cyclists’ perception of risk in roundabouts. Accident Analysis and Prevention, 40, pp.1055–1062. Mondshein, A., Blumenburg, E. and Taylor, B. (2013).Going mental: everyday travel and the cognitive map. Access, 43, pp.2-7. Mondshein, A., Blumenburg, E. and Taylor, B. (2010).Accessibility and cognition: the effect of transport mode on spatial knowledge. Urban Studies, 47(4), pp.845-866. 23 Nelson, T.A., Denouden, T., Jestico, B., Laberee, K. and Winters, M. (2015). BikeMaps.org: a 24 global tool for collision and near miss mapping. Frontiers in Public Health, 3(53), pp.1- 25 8. 38 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Noland, R.B. (1995). Perceived risk and modal choice: risk compensation in transportation systems. Accident Analysis and Prevention, 27(4), pp.503-521. Nye, M. and Hargreaves, T. (2009).Exploring the social dynamics of proenvironmental behaviour change. Journal of Industrial Ecology, 14, pp.137-149. Parkin, J., Wardman, M. and Page, M. (2007a).Models of perceived cycling risk and route acceptability. Accident Analysis and Prevention, 39, pp.364-371. Parkin, J., Ryley, T. and Jones, T. (2007b). Barriers to cycling: an exploration of quantitative analyses. Cycling and Society, pp.67-82. Parkin, J. and Coward, A. (2009).Comparison of methods of assessing cycle routes. Proceedings of the ICE – Municipal Engineer, 162, pp.7-14. Prendergast, P. and Rybaczuk, K. (2005) Using visualisation techniques in planning to improve collaborative governance in Ireland. Paper presented at CORP 2005, Vienna. Pucher, J. and Dijkstra, L. (2000). Making walking and cycling safer: lessons from Europe. Transportation Quarterly, 54(3), pp.25-50. Raidió Teilifís Éireann (RTÉ) (2015). ‘The growing war between cyclists and motorists, what’s happening on our streets?’.Broadcast 26th May. 17 Rengert G.F. and Pelfrey, W.V. (1997). Cognitive Mapping of the City Center: Comparative 18 Perceptions of Dangerous Places, In: David Weisburd and Tom McEwen, eds. (1997) 19 Crime Mapping and Crime Prevention, pp.193-217, Willow Tree Press, New York. 20 Road Safety Authority (RSA) (2014). Provisional Review of Road Crashes 2014. Ballina, Co. 21 22 23 24 25 Mayo, Ireland. Sanders, R. (2015).Perceived traffic risk for cyclists: the impact of near miss and collision experiences. Accident Analysis & Prevention, 75, pp. 26-34. Sener, I. N., Eluru, N. and Bhat, C.R. (2009).An analysis of bicycle route choice preferences in Texas, US. Transportation, 36(5), pp.511-539. 39 1 2 3 4 5 6 7 8 Short, J. and Caulfield, B. (2014).The safety challenge of increased cycling. Transport Policy, 33, pp.154-165. Smith, T.M.F. (1983). ‘On the validity of inferences from non-random samples’. J. R. Statist. Soc. A, 146(4), pp.394-403. Snizek, B., Nielsen, T.A.S. and Skov-Petersen, H. (2013).Mapping bicyclists’ experiences in Copenhagen. Journal of Transport Geography, 30, pp.227-233. Soini, K. (2001). Exploring human dimensions of multifunctional landscapes through mapping and map making. Landscape and Urban Planning,57, pp.225-239. 9 Spears, S., Houston, D. and Boarnet, M.G. (2013).Illuminating the unseen in transit use: a 10 framework for examining the effect of attitudes and perceptions on travel behaviour. 11 Transportation Research Part A, 58, pp.40-53. 12 13 14 15 Sprinkle Consulting (2007). Bicycle level of service, applied model. Sprinkle Consulting Inc., Tampa, Florida, USA. Thogerson, J. (2006). Understanding repetitive travel mode choices in a stable context: a panel study approach. Transportation Research Part A, 40, pp.621-638. 16 Timperio, A, Crawford, D., Telford, A. and Salmon, J. (2004). Perceptions about the local 17 neighbourhood and walking and cycling among children. Preventive Medicine, 38, pp.39- 18 47. 19 The Guardian (2013).Over 1,000 cyclists stage die-in protest outside Transport for London 20 HQ. Available at: http://www.theguardian.com/environment/bike-blog/2013/dec/01/stop- 21 killing-cyclists-die-in-tfl-protest [Accessed 1st July 2015]. 22 23 UK Department for Transport (UK DfT) (2014).British Social Attitudes Survey 2013: Public attitudes towards transport. London, UK. 40 1 Whannell, P., Whannell, R. and White R. (2012). Tertiary student attitudes to bicycle 2 commuting in a regional Australia university. International Journal of Sustainability in 3 Higher Education,13, pp.34-45. 4 Winters, M., Babul, S., Becker, H.J.E.H., Brubacher, J.R., Chipman, M., Cripton, P.,Cusimano, 5 M.D., Friedman, S.M., Harris, M.A., Hunte, G., Monro, M., Reynolds, C.C.O., Shen, H. 6 and Teschke, K. (2012). Safe cycling: how do risk perceptions compare with observed 7 risk?. Can J Public Health, 103(9), S42-S47. 8 Wooliscroft, B. and Ganglmair-Wooliscroft, A. (2014). Improving conditions for potential 9 New Zealand cyclists: an application of conjoint analysis. Transportation Research Part 10 A: Policy and Practice, 69, pp.11-19. 11 Zeile, P., Resch, B., Dörrzapf, L., Exner, J. P., Sagl, G., Summa, A. and Sudmanns, M. 12 (2015).Urban Emotions –tools of integrating people’s perception into urban planning. 13 Proceedings REAL CORP 2015 Tagungsband. Ghent, Belgium. 5-7th May. 41