Objective This paper examines the relationship between the amount of bicycle use and the number of single-bicycle crashes (ie, only one cyclist involved) in Dutch municipalities. Previous research has focused on crashes between bicycles and motor vehicles; however, most cyclists admitted to hospital are victims of single-bicycle crashes.
Methods This correlational study used three data sets which included data relating to single-bicycle crashes and kilometres travelled by bicycle. Negative binomial regression was used to compare the amount of bicycling with the number of injuries incurred in single-bicycle crashes in Dutch municipalities.
Results The likelihood of single-bicycle crashes varied inversely with the level of bicycle use. The exponent for the change in the number of single-bicycle crashes in response to changes in bicycle volumes was <1 in all analyses (ie, the increase in the number of single-bicycle crashes in a given municipality is proportionally less than the increase in the number of bicycle kilometres travelled by its inhabitants). The value was reduced in analyses of single-bicycle crashes with more severe injuries.
Conclusions Cyclists are less likely to be involved in a severe single-bicycle crash in municipalities with a high amount of cycling. Given the large numbers of patients admitted to hospital as a result of single-bicycle crashes, it is important to include the risks of these in road safety and health effect evaluations, and to take into account the non-linearity of the relationship between single-bicycle crashes and bicycle use if road safety measures are to affect the level of bicycle use.
- road safety
- safety in numbers
- older people
Statistics from Altmetric.com
Several studies have shown that a motorist is less likely to collide with a cyclist (or pedestrian) where there is a higher incidence of cycling (or walking).1–3 These studies indicate that the average cyclist is safer in communities where there is more cycling because motorists adjust their behaviour in the expectation of encountering cyclists.1 To date, single-bicycle crashes (ie, only one cyclist and no other vehicle involved such as a fall or collision with an obstacle; see Nyberg and Schepers4 ,5 for a more extensive description of these crashes) have not been included in studies on the relationship between bicycle crashes and bicycle use in spite of the fact that, in countries with a high proportion of cyclists, most cyclists admitted to hospitals are single-bicycle crash victims—for example, three-quarters of all cyclist traffic incident victims and one-third of all traffic incident victims in the Netherlands.6 The lack of research on single-bicycle crashes may be due to heavy under-reporting in police statistics of this type of crash7 and the fact that most cyclists killed in traffic are victims of crashes between bicycles and motor vehicles (80% in the Netherlands).6 The most severe crashes tend to attract the greatest attention; however, large numbers of less severe crashes resulting in hospitalisation will generate a significant burden on health services.
To achieve a more complete understanding of the safety effects of increasing levels of bicycle use, this correlational study focuses on the relationship between the amount of cycling and the likelihood of single-bicycle crashes. Knowledge of the relationship between bicycle volumes and single-bicycle crashes is important for evaluating the effects of measures that intentionally or unintentionally affect the amount of cycling. Using multiple data sets, analyses will be presented at the level of both individuals and municipalities within the Netherlands.
There are at least two arguments to support the hypothesis that, at the level of municipalities, more cycling reduces the risk of single-bicycle crashes, meaning that the increase in the number of single-bicycle crashes in a given municipality is proportionally less than the increase in the number of bicycle kilometres travelled by its inhabitants. First, as the amount of cycling increases, authorities may improve infrastructure safety. According to Wegman et al,7 safe conditions for cyclists in countries with higher levels of cycling may be one of the explanations for their lower death rates. Such an explanation is possible for single-bicycle crashes as well. As the numbers of cyclists increase, politicians may be more likely to invest in the safety of bicycle facilities (eg, adequate bicycle path and lane width, even road surfaces). Providing attractive bicycle lanes and paths may even encourage more people to commute by bicycle.8
Second, more experienced cyclists are likely to have fewer single-bicycle crashes because, with time, they have better control of their bicycles, they know what to expect and they have greater physical fitness. Cycling requires the rider to stabilise the bicycle, which is particularly difficult at lower speeds. The skills of controlling and manoeuvring a bicycle must be acquired and automatised through extensive experience.9 Cyclists will have a better knowledge of where to expect hazards as a result of increased experience. For instance, it was found that collisions with bollards and road narrowing occurred more often to cyclists who were unfamiliar with the crash location.5 As cycling improves health10 (eg, by strengthening muscles and the skeletal system, reducing obesity and improving stamina), it may also improve the ability of cyclists to avoid single-bicycle crashes and reduce the severity of their injuries. For instance, avoiding an obstacle on the road at the last moment may be easier for cyclists with an athletic build than for obese riders.
The hypothesis will be tested by comparing bicycle use (based on a survey) and numbers of single-bicycle crash victims (deaths and inpatients recorded by the police and trivial injuries reported in a survey) between municipalities, referred to as ‘municipality level analyses’. Additionally, using a survey, it will be tested whether more experienced cyclists are less likely to be involved in single-bicycle crashes, referred to as ‘individual level analysis’. The argument that increased experience decreases the likelihood of single-bicycle crashes is tested by the individual level analysis. Age and population density are included as control variables. Age may affect skill and frailty, while population density may affect the design of bicycle facilities.
The relationship between the likelihood of single-bicycle crashes and the number of kilometres travelled by bicycle will be explored using negative binomial regression, which takes into account the problem of ‘excess’ zeros frequently observed in crash count data—that is, many individuals or municipalities without crashes. The basic form of accident prediction models is used11:(1)
The estimated number of crashes (S) is a function of traffic volume (V) and a set of factors (xi, where i=1, 2, 3, …, n)—that is, the control variables. Exponent β indicates the change in the number of single-bicycle crashes in response to changes in bicycle volumes. When β=1 the growth in crashes with increasing exposure would be linear, β<1 indicates that the growth in crashes would be less than linear and β=0 indicates that the number of crashes is not related to exposure. α is a scaling parameter. The effects of various factors that influence the probability of crashes is modelled as an exponential function (ie, ‘e’, the base of natural logarithms) raised to a sum of product of coefficients (yi) and values of the (control) variables (xi). Jacobsen1 applied the same model but without control variables.
Population density and age categories were used as control variables. Population density is important as it may influence both bicycle use8 and the likelihood of single-bicycle crashes—for example, where population density is low, more space may be available to design bicycle facilities and, where it is high, more money may be available to invest in facilities. Municipalities are classified into three population density classes: high (>692 inhabitants/km2), medium (263–692 inhabitants/km2) and low (<263 inhabitants/km2), each group representing 33% of all Dutch municipalities. Age influences both bicycle use12 and increased susceptibility to injury due to fragility.13 A low number of single-bicycle crashes in a college town may be the result of low cyclist age rather than high bicycle volumes. Respondents are classified into three age groups (<24, 25–64 and 65+ years). These categories differ substantially in terms of the number of single-bicycle crash victims per kilometre travelled by bicycle6 (cyclists in the 65+ category are more likely to be hospitalised) and are sufficiently large to achieve reliable results. Crashes and exposure per respondent are used for individual level analyses. For analyses at the municipality level, the numbers of crashes and kilometres travelled by bicycle per municipality were split among these age groups, resulting in one record per municipality per age group in the data file for analyses at the municipality level. Instances of missing values for one of the variables are excluded from the analyses.
Data collection methods
Three existing data sets commissioned by the Dutch Ministry of Infrastructure and the Environment were used in this study to gather data on crashes and exposure. The data sets are summarised in table 1. References to reports that describe the data collection methods and include the original questionnaire are included in table 1. Statistics Netherlands data on population density of Dutch municipalities were also used.14
Description of the data sets and data collection methods
Periodic Regional Road Safety Survey (PRRSS)
The Periodic Regional Road Safety Survey (PRRSS) is conducted every 2 years (up to 2005 by Traffic Test and since 2007 by TNS NIPO) for general monitoring of road safety and traffic behaviour. It includes questions on the number of bicycle crashes and bicycle kilometres travelled each year. The answer category ‘not applicable’ and unanswered questions were treated as missing values. Most single-bicycle crash victims sustained injuries that did not need to be treated at an Accident and Emergency department and were categorised as ‘trivial injuries’. Respondents are selected in two stages.
Using the 2005 survey as an example15:
Stage 1 (December 2005): a sample of addresses was drawn from across the Netherlands. Potential respondents aged 15 years or more were asked to return a reply card, indicating whether they were prepared to participate in the survey.
Stage 2 (February 2006): a sample of persons, stratified according to age and gender, was drawn from the Stage 1 respondents. They received the questionnaire.
Data obtained from the 2001, 2003, 2005, 2007 and 2009 surveys were combined. The average response rate was 61%.
National Travel Survey (NTS)
The National Travel Survey (NTS) describes the travel behaviour of the Dutch population. Every month a sample of households is drawn from the Borough Basic Administration (a government database of relevant personal information regarding residents of the Netherlands such as date of birth and address) to ensure all types of travellers and households and all days are proportionately represented. Each member of the household is requested to record all journeys made on a particular day. Respondents are telephoned if they have not responded or to clarify missing answers, otherwise the respondent is excluded from the final data set.16 Data between 2004 and 2009 were used to determine the numbers of kilometres travelled by bicycle in all 431 Dutch municipalities. A total of 140 852 households (317 258 persons) completed questionnaires, giving a response rate of 70.5%.
The study covered all police-recorded single-bicycle crash victims across all 431 Dutch municipalities over a 6-year period (2004–9), a total of 990 inpatients and 78 deaths. Besides details such as the crash location, the police records indicated whether the crash resulted in death or hospitalisation.
PRRSS and NTS were compared with Dutch population gender and age statistics.14 In the analyses, data from these surveys have been weighted to represent age and gender in the overall population. The weighting factor in the NTS data set enables the outputs to reflect the Dutch population more closely—that is, besides a ratio to correct for response biases (eg, the response rate of young adults is lower than that of older people), it includes a ratio of population size divided by sample size.
Use of the data sets for municipality level analyses
Single-bicycle crash victims in PRRSS (trivial injuries) and police-recorded victims (inpatients and deaths) were used for municipality level analyses. In PRRSS, respondents are asked to identify their municipality of residence. The sample of addresses for PRRSS is drawn so that citizens in all Dutch municipalities have about an equal chance of being invited to complete the questionnaire and report their involvement in single-bicycle crashes. The total number of victims from all 431 Dutch municipalities was subsequently distributed among the three age groups. The same process was carried out for crash victims whose details were recorded by the police.
The NTS was used to select exposure data for municipality level analyses for both sources of crash data because it is specifically designed to achieve reliable mobility statistics. As respondents taking part in the PRRSS reported their place of residence and not the location of the crash, the number of bicycle kilometres travelled by citizens of a given municipality were summed to serve as an exposure measure (for the analysis of trivial injuries in table 2). The study period of PRRSS (2001–9) differs from the study period of NTS (2004−9). It is assumed that the difference will not affect the outcomes as the average number of kilometres travelled between 2004 and 2009 was only about 2% higher than that between 2001 and 2009.14
The police-recorded crash data note crash locations. As an exposure measure for police-recorded crashes, the part of external trips (ie, leaving and arriving in different municipalities) that fell within the borders of a given municipality was added to the length of internal trips (ie, leaving and arriving within the same municipality—about 90% of all bicycle trips) for the analyses of inpatients and deaths in table 2. Some trips go through more than two municipalities. The part of those trips that fell between different municipalities could not be assigned to municipalities. The difference between the exposure measure for analysis of trivial injuries (bicycle kilometres of the citizens of a municipality) and the measure for police-recorded inpatients and deaths (bicycle kilometres covered within the borders of a municipality) is small as more than 90% of all bicycle trips start in a cyclist's municipality of residence.
Use of the data sets for individual level analyses
PRRSS contains the responses of individuals to questions on the number of single-bicycle crashes and the number of kilometres travelled by bicycle per year. These data were used for an analysis of single-bicycle crashes with trivial injuries at the level of individual cyclists (see table 3).
Tables 4 and 5 present descriptive statistics while tables 2 and 3 show the results of the regression analyses. Table 4 is based on crash and exposure data for municipality level analyses and table 5 is based on crash and exposure data for individual level analysis. Exposure data for municipality level analyses were based on NTS and exposure data for the individual level analysis were based on PRRSS. The number of kilometres travelled by bicycle in table 4 reflects how many kilometres are travelled by cities' inhabitants per year. Table 5 indicates how many kilometres are travelled per year by respondents who indicated that they use a bicycle.
The exponent for the growth in crashes in table 2 (ie, parameter β in equation 1) was <1 in all three analyses, indicating that the increase in the number of trivial injuries, inpatients and deaths caused by single-bicycle crashes in municipalities is proportionally less than the increase in the numbers of kilometres travelled by bicycle by the citizens of a municipality. This result is in accordance with the hypothesis stated in the introduction—that is, more cycling reduces the risk of single-bicycle crashes. The result of the analysis on inpatients (the analysis with the highest number of crashes) is shown graphically for the three age groups in figure 1. Regression lines are as described in table 2, with an exponent for the growth in crashes of 0.76 (leaving population aside) added to the scatter plots.
The CIs of the exponents for the growth in crashes in the analyses were compared. The coefficients were lower in analyses of more severe injuries—that is, lowest in the analyses of deaths and highest in the analysis of trivial injuries.
In addition to the municipality level analyses, one individual level analysis of trivial injuries was conducted (see table 3). Again, the exponents for the growth in crashes was significantly <1, indicating that more experienced cyclists are less likely to sustain injuries due to single-bicycle crashes. Improved safety of individual cyclists owing to increased experience is likely to be one of the explanations for the results found at the municipality level.
Of the control variables (ie, age and population density), only age was significantly correlated to the number of single-bicycle crashes. Older cyclists are more likely to sustain severe injuries in single-bicycle crashes. The RR of older cyclists tends to grow as the severity of such crashes increases—for example, the RR of older cyclists is highest in the analysis of deaths. The results of the municipality and individual level analyses of trivial injuries in tables 2 and 3 are less consistent. It can be concluded that older cyclists are more likely to sustain severe injuries if they are involved in single-bicycle crashes, but it cannot be concluded whether they are more likely to be involved in single-bicycle crashes.
While it is already known that crashes between bicycles and motor vehicles are less likely to occur where there is a greater incidence of cycling,1–3 this study shows that more cycling also reduces the risk of single-bicycle crashes. At the municipality level, the increase in the number of single-bicycle crashes in a given municipality is proportionally less than the increase in the number of bicycle kilometres travelled by its inhabitants. This result adds to the ongoing body of literature exploring the relationship between bicycle use and cyclist safety.
One explanation for this finding is that cyclists in municipalities with a higher amount of cycling are more experienced and have fewer single-bicycle crashes. This is due to the fact that, over time, they have better control, know what to expect and have greater physical fitness. This explanation is supported by the finding that the likelihood of sustaining injuries due to single-bicycle crashes was lower among more experienced cyclists. A second explanation is that authorities may improve infrastructure safety as the amount of cycling increases, as has been suggested by Wegman et al,7 and that attractive bicycle facilities may encourage more people to commute by bicycle.8 The link between the number of kilometres travelled by bicycle in municipalities and the quality of their bicycle facilities is outside the scope of this study and needs to be tested in future research.
Modelling the relationship between bicycle use and single-bicycle crashes as a power curve shows that, at the municipality level, the number of inpatients due to single-bicycle crashes will increase at roughly 0.75 power of the number of kilometres travelled by bicycle. Jacobsen found this parameter to be lower, roughly 0.4 for bicycle–motor vehicle crashes.1 This can be explained by factors that influence the likelihood of bicycle–motor vehicle crashes but not the likelihood of single-bicycle crashes. Jacobsen's main explanation is that the likelihood of bicycle–motor vehicle crashes decreases because motorists adjust their behaviour in the expectation of encountering cyclists.1 A second explanation is that there are fewer motorists to pose a risk to ‘vulnerable’ cyclists if there is a transfer of trips from motor vehicles to cycling (or walking), which is more likely if more people choose to travel from A to B by bicycle. According to Wegman and Aarts,17 for one car casualty there are 150 bicycle casualties in crashes between cars and bicycles. The risk to each cyclist decreases as the number of motor vehicles decreases.3
The results are generalisable to the Dutch population owing to the large sample of respondents represented by the underlying data sets, the application of weighting factors to correct for potential response biases and the selection of police-recorded crashes from all parts of the Netherlands. However, compared with other countries the Dutch are younger when they start cycling and they use bicycles more for utilitarian purposes such as commuting and shopping than for sport, so a comparison of these results with international findings is needed to further confirm their generalisability. The predominantly utilitarian use of bicycles in the Netherlands explains why the benefits of more experience and improved fitness are not undone by risk compensation such as riding faster. Cyclists do not want to be suffering the ill effects of overexertion when they arrive at their workplace or school. The conclusions found in this study apply when cycling is predominantly a means of transport from A to B. The beneficial effects of more cycling may be even greater when the share of utilitarian trips increases.
The results show that the exponent for the growth in crashes is lowest for crashes with severe injuries. This outcome may be due to different factors working at varying levels of crash severity. The advantages of being more experienced may be greater for more severe crashes as skills and fitness influence both the likelihood of being involved in a single-bicycle crash and the likelihood of sustaining injuries. However, more research is needed as some of these differences may result from the different research methods used to construct the underlying data sets.
Age was included as a control variable, but it is also an important factor for reliably estimating the results of increasing levels of cycling if the increase differs between age groups. Older cyclists sustain more severe injuries in single-bicycle crashes than younger cyclists. The results of the likelihood of being involved in a crash were less consistent. Susceptibility to injury due to fragility of older cyclists (in this study defined as >65 years of age) seems to be the most important explanation for the increased likelihood of sustaining severe injuries in single-bicycle crashes, a finding that is comparable to susceptibility to injury of older drivers.13
Recommendations for practitioners
Given the high numbers of single-bicycle crashes, it is important to include the risk of these crashes in road safety and health effect evaluations and to take into account the non-linearity of the relationship between single-bicycle crashes and bicycle use if road safety measures are to affect the level of bicycle use. Not doing so may result in biased conclusions. Single-bicycle crashes form part of the studies that should determine the health impact of changes within the whole transport system. A more complete estimation of road safety effects can be achieved using the models developed by Elvik,3 which can be applied to single-bicycle crashes, bicycle–motor vehicle crashes and others. The advantages of fewer motorists posing a risk to ‘vulnerable’ cyclists, where there is a transfer of trips from motor vehicles to cycling, are accommodated by including both bicycle and motor vehicle volumes in the models. A more accurate estimate of health effects can be achieved by taking into account health benefits from increased physical exercise, risks associated with higher exposure to air pollution, and decreased air pollution emissions when car trips are replaced by bicycle trips.10 As single-bicycle crashes often result in minor or even severe injuries but are only rarely fatal, the disability-adjusted life years and quality-adjusted life years measures (which include both mortality and morbidity) are more suitable for evaluating the disease burden from single-bicycle crashes than the life years gained measure which includes mortality only.18 ,19
What is already known on this subject
The likelihood of a cyclist colliding with a motorist is in inverse proportion to the incidence of cycling.
What this study adds
More cyclists are hospitalised as a result of single-bicycle crashes (eg, a fall) than collisions with motorists.
Cyclists are less likely to be involved in a severe single-bicycle crash in municipalities with a high amount of cycling.
I would like to thank Jean Smith for her excellent comments on this paper.
Funding This research was conducted by the Ministry of Infrastructure and the Environment and received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.