Cycling provision separated from motor traffic: a systematic review exploring whether stated preferences vary by gender and age

ABSTRACT In this paper, we represent a systematic review of stated preference studies examining the extent to which cycle infrastructure preferences vary by gender and by age. A search of online, English-language academic and policy literature was followed by a three-stage screening process to identify relevant studies. We found 54 studies that investigated whether preferences for cycle infrastructure varied by gender and/or by age. Forty-four of these studies considered the extent of separation from motor traffic. The remainder of the studies covered diverse topics, including preferred winter maintenance methods and attitudes to cycle track lighting. We found that women reported stronger preferences than men for greater separation from motor traffic. There was weaker evidence of stronger preferences among older people. Differences in preferences were quantitative rather than qualitative; that is, preferences for separated infrastructure were stronger in some groups than in others, but no group preferred integration with motor traffic. Thus, in low-cycling countries seeking to increase cycling, this evidence suggests focusing on the stronger preferences of under-represented groups as a necessary element of universal design for cycling.


Introduction
Within countries with a low cycling mode share (approximately 5% mode share or less, herein referred to as low-cycling countries), cycling is demographically unequal, notably by gender and age (Pucher & Buehler, 2008). A policy concern to diversify cycling has been accompanied by a growth in academic literature on this issue. Aldred, woodcock and Goodman (2015) explored whether increasing cycle commuting (between 2001 and 2011) was associated with greater age and gender diversity in England and Wales.
The paper joins a growing number of publications in the transport field (e.g. Jothi Basu, Subramanian, & Cheikhrouhou, 2015;Vieira, Kliemann Neto, & Amaral, 2014;Wang & Notteboom, 2014) using a systematic review approach. Although stated preference studies have been common in transport research for some time (Hensher, 1994), they have rarely been synthesised using systematic reviews. Such synthesis is, however, increasingly common in other disciplines that make use of stated preference data, such as health economics (Whitty, Lancsar, Rixon, Golenko, & Ratcliffe, 2014).
Our choice of a systematic review approach means the paper benefits from the increasing robustness that comes with a more comprehensive search. However, the systematic, in-depth approach meant we had to choose a narrower question than narrative reviews can adopt. We would argue that this paper helps to demonstrate the value of systematically reviewing stated preference evidence in transport. We hope that it will be complemented by future systematic reviews of other topics and other types of evidence, including ecological studies and route choice studies.
In this review, an inclusive definition of "stated preference" is used. Traditionally in transport research, stated preference studies refer to techniques specifically used to estimate utility functions, used within choice modelling to predict change in use of transport infrastructure or services and/or to calculate cost-benefit ratios (Kroes & Sheldon, 1988). However, with the field becoming more interdisciplinary, health and social researchers (e.g. Winters & Teschke, 2010) are also conducting research asking about people's infrastructural preferences, although without the aim of creating utility models. Here, both types of study are included.

Methods for selection, appraisal and synthesis
Methods are outlined here: for more details on search terms, sources retrieved and screening procedures, please see Appendix. Two authors (RA and BE), the study appraisers, searched the academic databases (EBSCO, Web of Science, ProQuest, PubMed, TRID, ARRB) plus 11 websites (via Google) (end of March 2015), following a search protocol developed by the team with input from additional advisors at the Centre for Diet and Activity Research. We only included studies that covered preferences related to cycle routes and infrastructure; so not, for example, preferences for taking bicycles on trains. Studies were included that reported analysis of any similarities or differences by age and gender. BE screened abstracts and led initial study selection with RA checking wherever uncertainty was flagged. RA appraised the studies.
In the selected articles, separation from motor traffic was by far the most common infrastructural characteristic discussed (with a clear comparison made by age and/or gender in 44/54 studies). This mostly involved questions about the existence or not of some form of separate provision, but sometimes involved questions about motor traffic flows, where sharing takes place. Hence, in analysis, we focused on this issue. Other issues covered were diverse; for example, two studies covering preferences for winter maintenance of cycle infrastructure, and another covering preferences related to "quality of signage". It was not possible to synthesise similarities and differences related to these issues.
There are no established reporting guidelines for stated preference studies. We extracted data from each study on (i) issues that affect internal validity, (ii) issues that affect external validity or generalisability, and (iii) sample size. For internal validity, we focused on how preferences were elicited. Where little detail is given, participants may imagine quite different kinds of infrastructure when responding. Specifically, a "cycle lane" may be imagined as being effectively shared with motor traffic (an advisory painted lane), or separated by bollards, kerb or other barriers. More detail may allow more discrimination between different levels of separation from motor traffic.
(a) How situations were communicated to participants, for example, words only, images, video. (b) How specific the situations presented to participants were categorised as follows: . Low to very low specificity, for example, respondents choosing between "cycle lane present", and "no cycle lane". . Medium specificity, for example, respondents asked to choose between on-road segregated infrastructure, painted cycle lanes, and off-road tracks. . High specificity, for example, images of a range of different infrastructural types with differing degrees of separation from motorised traffic.
We considered external validity to refer to whether survey results represent broader population views about preferred cycling environments. We did not consider the wider issue of whether these stated preferences accurately predict subsequent behaviour change (Bradley, 1988) because we would argue that views about desired service provision are important in themselves.
Sampling methods were categorised as follows: . Convenience sample, for example, students, participants in cycle touring event.
. Purposeful convenience sampling, for example, potential cyclists, employees. . Representative survey, for example, randomly sampled national travel survey.
Studies with higher quality sampling methods of the general population are more likely to be representative of a potential cycling population. Inclusion of non-cyclists 1 was considered important as there are suggestions in the literature that cyclists' preferences, particularly in low-cycling contexts, may not represent the views of potential cyclists (Horton & Jones, 2015).
Three rounds of screening were carried out to filter the evidence, with data extracted into a bespoke table in Excel. Analysis in Excel and SPSS explored both headline findings (similarities and differences in preferences) and the extent to which these were associated with study design. We attempted to record information that could be used for meta-analysisquantitatively combining the results from multiple studiesbut in general, information such as sub-group means was not provided, meaning that meta-analysis was not possible.

Studies included and excluded
Our search strategy led to the identification of 54 separate studies, reported in 58 publications ( Figure 1). 2 About the studies Fifty studies examined stated preferences in relation to gender, with 33 covering age (adults) and only 2 discussing preferences related to child cycling (Table 1). A summary of study characteristics is presented in Table 3 and Figure 2.
It can be seen that this is a growing field, with 2009-2010 onwards providing a steady increase in the numbers of studies published. There is the potential to benefit from this growth by developing more consistent measures and/or sharing data for meta-analysis.
As noted above, the synthesis below includes 44 of 54 studies, with patterns similar to those for all 54 studies in terms of study composition and so on.

Country of origin
Over one-third of all studies were conducted in the U.S.A (19 studies), with eight from the UK, followed by Belgium and Canada (four each). Over two-thirds (39 of 56) were carried out only in high-income countries with low cycling rates. In terms of classifications, Australia, Brazil, Canada, Ireland, New Zealand, Spain, UK and U.S.A were judged to be low cycling. Other countries were judged to be medium or high cycling (Belgium, China, Denmark, India, the Netherlands and Sweden).

Study size and populations
Sample size varied considerably (35-3494, with one not stated). Most studies included more men than women. This was particularly true in studies set in low-cycling countries and drawing their sample from existing cyclists (in one study, "avid cyclists"). Only in two-fifths of the studies, at least 20% of the sample were non-or infrequent cyclists. A little over one-third of studies only sampled cyclists, while one only exclusively sampled non-cyclists. Overall, the proportion of regular cyclists included was far higher than for the general population, this being particularly true in studies in low cycling countries using convenience samples.

Reporting of results
Results were reported in diverse ways; for example, scores given out of five to different infrastructure types, or percentage of people agreeing that they would use a particular type of cycle route. Given the information available, a meta-analysis was not possible. For example, 13 of the 17 studies that reported no statistically significant gender differences in preferences for separation did not give subgroup means. 3 Sampling and elicitation methods Sampling methods varied widely from household surveys to convenience samples of cyclists attending specific rides. Nearly two-thirds used convenience sampling with around a quarter of studies using random sampling. Various study methods were used to elicit preferences (see Table 1). Almost half gave a text-based description of an infrastructure type (e.g. "painted lane"), conducted either using a paper questionnaire, on the phone, in person, or online. The participant would then be asked to rate the infrastructure type, although the type of rating would depend on the survey: including ranking preferences, assigning hypothetical monetary values, or asking people whether they would feel comfortable or safe.
The second most common type of elicitation method was to use images, either real or computer-generated. These were accompanied by questions about the desirability of the infrastructure type, as with studies using text-based elicitation. A less common method referenced existing infrastructure; for example, one study stopped cyclists in a series of sampled cycle lanes and asked them to rate the lane compared to other types of infrastructure. In other cases, researchers showed participants videos of infrastructure types, and then asked about preferences.
Finally, the situational specificity of the survey questions varied (see Table 1). Nearly half were very general (e.g. asking about "cycle lanes") with one in five very specific, for example, testing a range of infrastructure types with differing extents of segregation. The remainder were in between, for example, making trade-offs between different infrastructure situations and trip times; with situations including bus/cycle lanes, parks/quiet residential streets (combined option), on road cycle lane, and off-road track.

Infrastructural preferences
Findings: gender and preferences for greater segregation from motor vehicles Forty studies provided evidence as to whether preferences for separation from motor traffic differed by gender. Of these, 23 (57.5%) said women expressed stronger preferences for segregation from motor vehicles than did men (Table 2). Seventeen studies (42.5%) reported no statistically significant differences in gender preferences. No studies reported that men had stronger preferences than women for greater segregation from motor vehicles. Most studies that found no gender difference were small, and likely to have been insufficiently powered (see Figure 3) to detect a relevant difference. Among studies containing at least 200 participants, 20/29 (69%) reported stronger preferences in women than in men, whilst amongst studies containing fewer than 200 participants, only 3/10 (30%) did so. 4 Four-fifths of studies that found differences in preferences between men and women (19/24) also highlighted overall similarity in preferences across genders. For example, for both sexes, more people preferred complete separation from motor traffic compared with the presence of a painted lane but the gap in women was larger.  Females are more likely than males to self-identify as a "beginner cyclist who prefers to stick to the bike trails, paths and/or sidewalks" while substantially more males self-identified as an "advanced, confident cyclist who is comfortable riding in most traffic situations". .
Proportions of female students who chose "vehicular traffic" and "lack of bicycle lanes/paths/trails" as barriers to bicycling were significantly higher than that of the male students Study asked whether 10 aspects of the cycling environment would be suitable for riding on by (a) "most people" (b) on their own (c) carrying a child (d) with an 8-year-old (e) by a 12-year-old. Seven of these 50 differences were significant, with more support for segregation among females .
For 43 out of 50 situations, there were no significant differences. In particular, there were no significant gender differences for questions related to 8-and 12-year-olds Antonakos (1995) USA 552 cyclists at four recreational bike tours in Michigan .
For commuter cycling preferences, females rated bike lanes and bike paths higher on average, than males. On a scale from 1 (not at all preferred/not at all important) to 5 (very preferred/extremely important), females rated bike lanes and bike paths at 4.2 and 4.1, respectively, on average compared to 3.9 and 3.6 for males. Females and males both rated traffic safety as high importance, though females gave higher ratings (4.5 vs. 4.1) .
The hierarchy of "corridor type" preferences was similar with regards to commuting: males and females both rated bike lanes and wide curb lanes as high preferences Antonakos (1995) USA 552 cyclists at four recreational bike tours in Michigan .
For recreational cycling preferences, cycling experience and age were negatively associated with preference for bike paths, sidewalks, dirt roads, and trails for recreational cycling . For commuting cycling preferences, age and cycling experience were negatively associated with preference for bike paths, sidewalks, and dirt trails for commuting Females more often responded "much more likely" over "somewhat more likely" for every street characteristic by at least 9%. Safety characteristics of a bike route (e.g. markings, low traffic, traffic lights at arterials, etc.) were more important to female cyclists . Survey data did not indicate large gaps between the preferences of male and female cyclists when combining those who were "much more likely" and "somewhat more likely" to select a route based on a characteristic Bernhoft and Carstensen (2008) Denmark Pedestrians and cyclists aged 40-49 and 70+ in two provincial cities in Denmark .
A significantly higher proportion of females than males found cycle paths important for their comfort .
In the younger group, a higher proportion of females than males would choose a route with cycle path and signalised crossings .
Presence of cycle paths was the most important route attribute for both men and women Bernhoft and Carstensen (2008) Denmark Pedestrians and cyclists aged 40-49 and 70+ in 2 provincial cities in Denmark .
A higher proportion of older than younger respondents said that the presence of cycle paths was most important for their comfort . Females had a greater preference than males for "greenways" and "off road cycle lanes" .
Ordering of route types was the same for both gendersoff-road cycle lane, then Greenway, then on-road cycle lane, then shared bus lane, then no facilities Deenihan and Caulfield (2015) Ireland Tourists at two locations in Dublin .
Male tourists were more likely than females to choose a road without cycling facilities, with female tourists more likely to choose a road with cycle lanes or a segregated from traffic cycling facility .
Female tourists were very unlikely to select a road without any cycling facilities Deenihan and Caulfield (2015) Ireland   (2015) USA Cyclists and residents in five large US cities (Cyclists and non cyclists) .
On several measures of safety and comfort, female cyclists using protected bike lanes had more positive associations with the lanes than males. For example: -Females were more likely to agree that the new protected bikeways were safer than other city bikeways (93% vs. 87%) -Females were slightly more comfortable than males on paths or trails separated from the street and less comfortable on commercial streets without bike lanes . For potential female bicyclists (part of the residential group), protected lanes increased stated comfort levels significantly, though females still reported lower comfort levels than males . Among residents interested in bicycling more for transportation, 87% of females and 82% of males agreed that they would be more likely to ride if motor vehicles and bicycles were physically separated . Both males and female intercepted cyclists overwhelmingly felt that protected bike lanes increased their safety while riding in them . There were no gender differences found amongst intercepted cyclists in relation to some statements about protected bike lanes, for example: -The buffer section with parked cars makes me feel safe (asked on lanes with buffers) -The buffer effectively separates bikes from cars -The buffer does a good job of protecting bikes from cars -The lane design effectively separates bicyclists from pedestrians . Amongst intercepted cyclists, there were few gender differences with regards to facility design Emond, Tang, and Handy (2009) USA Random sample of residents in six small cities in the USA .
Males were more likely than females to report that they would ride with heavier traffic, despite similar levels of discomfort .
Males experienced approximately as much discomfort on average as females on facilities not separated from heavier traffic Gardner (1998) UK Leisure cyclists, non-cyclists and utility cyclists in different areas of England .
More males than females were willing to cycle alone, with some females indicating that they did not feel safe if cycle lanes were too isolated, or in parks or the countryside on their own Heesch, Sahlqvist, and Garrard (2012) Australia Adult cyclists in Queensland who were members of Bicycle Queensland (BQ) club .
For transport cycling, females were less likely than males to prefer cycling on the road .
For recreational cycling, females were less likely than males to prefer cycling on-road, but more likely to prefer cycling off-road .
Top constraints for at least half of males and females were perceived environmental factors, namely traffic and aggression from motorists, with females significantly more likely to report these .
For transport cycling, few males or females preferred to cycle on the road . On-road routes were even less preferred for transport cycling than recreational cycling by both males and females Hughes and Harkey (1997) USA Twenty-three casual and 12 experienced cyclists . Apparent reduction in sensitivity to risk on the part of the "younger cyclist" (under 20 years) Hunt and Abraham (2007) Canada Cyclists in Edmonton .
Indications that older people had less aversion to mixed traffic and that the very young had less aversion to paths were statistically weak Krizek, Johnson and Tilahun (2005) USA 292 current and potential cyclists in Minnesota . Females were willing to travel more additional minutes than men for a preferred facility. Assuming a 20 minute commute, males were willing to divert 5.43 fewer minutes than females .
Male and female cyclists were relatively similar in the proportion who valued specific facilities such as on-road lanes, separate paths, and a connected system of bicycle routes .
Both males and females were willing to travel longer for an off-road facility, followed by a bicycle lane with no onstreet parking Landis, Vattikuti, and Brannick (1997) USA 150 cyclists aged 13+ in Tampa  Compared with people aged 18-34, middle-aged (35-54) people are less likely to hold low perceptions in "high bikeable" neighbourhoods; while older people (55+) are nearly three times more likely to perceive "high bikeable" environments as low Majumda, Mitra, and Pareekh (2015) India Residents of two small Indian cities . Significant gender differences were found for the following statement for one of the samples: . I will not cycle because of safety hazard associated . No significant gender differences were found for the following statement for both samples: -I will not cycle because presence of other motorised vehicles makes it difficult for bicycle commuters especially in peak hours Majumda et al. (2015) India Residents of two small Indian cities . No significant age differences were found for the following factors/statements for both samples: -I will not cycle because presence of other motorised vehicles makes it difficult for bicycle commuters especially in peak hours -I will not cycle because of safety hazard associated Gender had an influence on rider type self-categorisation: -Male cyclists were more likely to categorise themselves as "strong and fearless" or "enthused and confident" than the female cyclists; -Females were more likely to classify themselves as "comfortable but cautious" than male riders . Riders across all cyclist types prefer dedicated cycling facilities and are opposed to high-speed traffic and highvolume traffic Misra et al. (2015) USA As above .
Age had an influence on rider type self-categorisation: -Older people were less likely to categorise themselves as "strong and fearless" -With increasing age people are more likely to group themselves into less confident groups .

Parkin, Wardman, and
Page (2007  Females gave "No traffic on the road", "Segregated cycle lane" and "Junction types" higher ranks than did males, while "Unsegregated cycle lane" was lower down the list compared to males . On average, females rated the absence of (clearly marked) cycle lanes as more unsafe than males. For example, the mean safety score given for "no cycle lanes" varied between 7.3 and 7.8 for different female age groups, compared to 6.5 to 7.1 for different male age groups (where 1 is "completely safe" and 10 "completely unsafe") .
Males and females both had stronger preferences for segregated cycle lanes (compared to no cycle lane or nonsegregated cycle lanes) and no traffic (compared to low volume of traffic) .
Men and women of all ages had similar infrastructural preferences, on average, when comparing kerbsegregated and off-road lanes (the two most preferred infrastructure types), clearly marked cycle lanes (less preferred) and no cycle infrastructure (least preferred) There was a significant difference in confidence between males and females saying that they felt confident to cycle on all roads (79% vs. 50%) .
Females were much more likely to prefer routes away from other traffic and difficult junctions. The average score for "Safety is the most important consideration when choosing a cycle route" for females was 0.89, compared to 0.53 for males (based on a five point scale from strongly agree (+2) to strongly disagree (-2)) .
In general, female respondents were slightly more likely to rate each junction as less safe than males and more prepared to detour .
Though male respondents agreed on average that they would avoid a route with difficult junctions, they were less certain they would avoid that particular route (0.66 compared to 1.04 for females) .
Female cyclists were slightly more willing to change their route in order to use a dedicated on-road cycle lane (56% vs. 48%) .
Generally, "route choice considerations" were in the same direction (where >0 indicates agreement and >0 disagreement) .
There was no variation between male and female cyclists in willingness to consider changing routes to use a cycle superhighway (Continued ) Older respondents associated a higher disutility for routes with car parking; however, as for roadway class, the impact of age was small .
Older people had a marginally higher preference for wide near-side lanes, and disliked major intersections more than younger people .
The differential preference for residential streets (compared to minor arterials) is small: even for a cyclist 100 years old, the magnitude on the minor arterial coefficient drops from -0.77 to -0.69 People aged 35+ (particularly over 50s) were more likely to report "more bicycle paths" would encourage them to cycle more Tiwari (2014) India Review including report of survey of current bicyclists and potential bicyclists in Pune, India .
Preferences of females and males for bicycle routes showed similar trends except a few variables; differences were slight Twaddle, Hall, and Bracic (2010) Canada Staff and students at University of Calgary, particularly potential or current cyclists .
Females were more concerned than males about safety issues. Significant differences were found for "I do not know a safe route" and "I feel unsafe riding on roads" .
Males were more likely than females to indicate a desire for wide curb lanes [shared with general traffic] .
There was no significant difference by gender in selection of any on-route improvements. Females were found to share similar preferences with males, high proportions of both wanting bicycle lanes, more pathways, and more direct routes.

Twaddle et al. (2010) Canada
Staff and students at University of Calgary, particularly potential or current cyclists .
Although all females were likely to indicate safety concerns prevent them from commuting by bicycle, the type of safety concern differs by age. Younger females are unsure about the route to take, with older females more concerned with feeling unsafe riding on the road For observed characteristics included in the "choice task": "evenness of the cycle path", "presence of new elements", "presence of historic elements" and "safety for crossing the street", associations with proportion of invitingness for transportation cycling were only found in females, with results not significant in males .
Moderating effects of gender on the association between the environmental characteristics and proportion of environmental invitingness for transportation cycling were absent in the final model for the choice task and the cognitive task Vliet ( Distance, pleasantness and traffic safety were the most important attributes for cyclists, but few differences were found in the importance of attribute weights between male and female subjects, or older and younger groups Winters and Teschke (2010) Canada 1402 adult current and potential cyclists, that is, the "near market" for cycling in Vancouver, Canada .
Females scored low preference routes even lower than males .
The two least preferred route types were major streets with no facilities, with or without parking (16% likely to choose). Only 79 respondents were "very likely" to choose to ride on major streets with parked cars. Virtually no differences in mean scores between males and females for the six most preferred route types (pavedoff street paths for bikes only, paved off-street multiuse paths, unpaved off-street multi-use paths, cycle path next to major street separated by barrier, residential streets marked as bike routes with traffic calming, residential streets marked as bike routes) (Continued) TRANSPORT REVIEWS Most results showed no significant differences by demographic group Differences were found by study type and composition. Smaller studies were less likely to report a gender difference, and some may have been underpowered to detect a meaningful difference. Of studies with larger sample sizes (> 100) and at least 20% non-or infrequent cyclists, 76.5% (17) found gender differences against 23.5% (4) who did not.
Of studies providing a high level of specificity, 78% (n =7/9) found a gender difference; this proportion was lower for studies with medium or low specificity (54% and 50%, respectively), but the difference was not statistically significant (p = .37 for trend).  Studies that contained at least 50% non-cyclists found gender differences in 58% (7/12) of cases, while studies that did not found gender differences in 62% (16/26) of cases (p = .85 for difference). 5 By contrast, study context made a difference to findings. Among studies conducted in low-cycling countries, 69% (n = 20/29) found gender differences in preferences for separated infrastructure, while only 27% (3/11) found differences in studies where some or all participants lived in medium-or high-cycling countries (chi-squared p = .02 for association).
Findings: age and greater segregation from motor vehicles Only 25 studies reported on age, with findings less consistent than for gender. While 9 studies (36% of those reporting on preferences for greater segregation and age) found that older people expressed stronger preferences for separation from motor vehicles, 13 (52%) found no differences, and 3 (12%) reported that older people had less strong preferences for separation from motor vehicles than younger people (Figure 4). Twentytwo out of 25 studies covering older people's preferences highlighted overall similarity in preferences across age groups.
The relationship between sample size and findings is less clear-cut than for gender, although smaller studies were more likely to find "no difference" (4 out of 5 of studies with a sample size of below 200, compared to 9 out of 20 for the larger studies).
It might be thought that (among participants who cycle) older cyclists' preferences do not stem from age per se, but from their having likely cycled longer than younger participants. We did not find support for this: among studies that mentioned controlling for cycling experience, an independent "age effect" remained in at least some of these, although reporting was sometimes unclear. However, some studies examining experience found an independent "experience effect" shaping perceptions of cycling infrastructure instead of, or as well as, age and gender effects (e.g. Ma & Dill, 2015;Ma, Dill, & Mohr, 2014).
Findings on child preferences or adult preferences for riding with or by children Because we only found two studies addressing preferences for infrastructure involving children (Aldred, 2015 andGhekiere et al., 2015), no attempt was made to quantitatively synthesise these. The former compared adults' preferences for infrastructure for themselves when riding alone, to their preferences for themselves when riding with children, or when deciding whether a child should cycle. The latter compared adults' preferences for child cycling with the children's own preferences. Both studies point to a stronger preference for separation from motor traffic where children are cycling. This goes beyond barrier separation and covers issues such as, in Aldred (2015), protection at crossings and reduction in rat-running (when drivers use residential streets as a short cut avoiding main roads), and in Ghekiere et al. (2015), the need for wide and even cycle paths.
The Table 3 summarises some key points from the literature on preferences for infrastructure separated from motor traffic, as synthesised in the tables above:

Other preferences
Studies highlighted some other similarities and differences by age and gender, but these proved too diverse to synthesise within the constraints of a systematic review. Research covered topics including preferences for cycling environments that minimise the impact of winter conditions (e.g. use of higher quality brine to maintain infrastructure), routes that are direct and ideally avoid hills, and routes that are well lit and overlooked.

Discussion
We have found good evidence that women express stronger preferences for greater segregation from motor vehicles than men. This is within a context of similar overall types of preference, that is, typically very similar hierarchies of preference across genders. As stated by Misra, Watkins, and Le Dantec (2015): "Riders across all cyclist types prefer dedicated cycling facilities and are opposed to high speed traffic and high volume traffic, with little variation based on the classification of the cyclist". In terms of age, again, there is an overall qualitative similarity between groups, but with some evidence suggesting that older people may have stronger preferences for separated infrastructure.
Gender differences were clearer among studies in low-cycling countries. In such settings, cycling is often perceived or experienced as risky, suitable only for the brave and confident (Horton & Jones, 2015). Men may be less concerned about risks than women, or more reticent about voicing their fears because these do not fit with dominant constructions of masculinity (Steinbach, Green, Datta, & Edwards, 2011). As such, the findings concerning gender differences in this review are arguably particularly relevant to places seeking to increase cycling from a low base.
Cycling speed may influence how views differ by age and gender. Cyclecraft, the UK's national guide to cycling (Franklin, 2009), recommends a speed of 20 mph (32 kph) in challenging traffic situations. This is far faster than the average cycling speed, the gap being even greater for women and older people. Our analysis of National Travel Survey data 6 indicates that in England, among those aged 18-29, the average speed was 11.3 mph men and 10.5 mph for women, while among those aged 60-69, it was 9.6 mph for men and 9 mph for women. Slower cyclists report more near misses per mile (Aldred and Crosweller, 2015). A stronger preference for separated infrastructure among older people could also stem from greater vulnerability to injury.
A few studies suggest that women may be more likely to be affected by barriers including the need to carry items, winter conditions, hills, and personal safety concerns (see also Damant-Sirois & El-Geneidy, 2015;Heinen, van Wee, & Maat, 2010). These issues merit further research, including how these factors might interact with infrastructural characteristics. Future stated preference work on gender could focus on the detail of infrastructure types (e.g. verge vs. kerb separated) and on how other factors, for example, cycling experience and cycling speed, affect preferences by gender. Another recommendation is for more research both on children's own views, and on adult views about infrastructure for child cycling. Understanding how infrastructural change might impact child cycling is crucial not just for children but also for carers, disproportionately affecting trips made by women .
Among studies covering age, definitions of older age varied considerably as did methods for evaluating its effects. While some studies used age in years as an independent variable within linear regression, or considered 3-6 categories, others used very different cut-offs for "older" cyclists. The use of harmonised age categories and/or treatment of age as a variable would have improved our ability to assess its impact. A recommendation that follows from this would be for stated preference studies to more routinely publish simple anonymised data sets (e.g. on the UK Data Archive or, given the non-sensitive nature of the data, on journal websites) suitable for individual-level meta-analysis. Comparability would also be enhanced by development of reporting guidelines.
The level of situational specificity varied substantially and this is worthy of further methodological investigation. Higher specificity potentially introduces more unobserved variation (e.g. related to path width or adjacent motor traffic), although this can be minimised or reduced (e.g. using manipulated photographs). However, higher specificity enables greater consistency in what people understand they are being asked to compare. Many less specific studies simply reference a "cycle lane", which could be assumed to be either a painted on-road lane or track with varying levels of segregation (Steer Davies Gleave, 2010a. More realistic representation of infrastructure allows greater discrimination between options and may help us estimate more realistically the type of infrastructure that may be required to substantially grow cycling levels. While there is not one right way to do things, future research should aim for comparability with published methods wherever possible. This is not to deny the need for innovation. Future research into infrastructure preferences may want to consider combining qualitative and quantitative approaches, and make greater use of video methods (see e.g. Ghekiere et al., 2014Ghekiere et al., , 2015. Finally, policy should focus on the infrastructural needs and preferences of under-represented groups, including older people, women, children and those cycling with children or making decisions about child cycling. Younger people, men, and those travelling without children also generally prefer separation from motor traffic, so building for under-represented groups should, if done well, suit others. Inclusive infrastructure is particularly important given evidence that some other barriers to cycling may be stronger for under-represented groups (van Bekkum, 2011;Bergström & Magnusson, 2003;Daley, Rissel, & Lloyd, 2007;Damant-Sirois & El-Geneidy, 2015;Finch et al., 1985;Steinbach et al., 2011). For example, women may have stronger concerns than men about safety from crime, while older people may struggle to cycle longer distances. Focusing on the needs and preferences of under-represented groups should be sensitive to these issues and, for example, take account of concerns about crime and route directness when planning the location of high-quality infrastructure. Notes 1. Definitions of "non-cyclists" varied between studies; here, we mean someone who did not cycle in the last week (or longer). It was not always possible to determine whether there were over 20% non-cyclists or not. 2. Note: in the case of the 17 publications where full text could not be found, both reviewers read the abstracts and considered almost all unlikely to be relevant: most were conference papers or policy reports, and often seemed relatively tangentially connected to the research question (we had erred on the side of inclusivity in Stage 3 where we could not initially find publications). 3. In two cases, graphs were given illustrating this, but not the precise figures. 4. One study did not report sample size. 5. Not all studies could be included due to difficulty in determining numbers of cyclists and noncyclists. 6. https://discover.ukdataservice.ac.uk/series/?sn=2000037.