The Societal Costs and Benefits of Commuter Bicycling: Simulating the Effects of Specific Policies Using System Dynamics Modeling

Background: Shifting to active modes of transport in the trip to work can achieve substantial co-benefits for health, social equity, and climate change mitigation. Previous integrated modeling of transport scenarios has assumed active transport mode share and has been unable to incorporate acknowledged system feedbacks. Objectives: We compared the effects of policies to increase bicycle commuting in a car-dominated city and explored the role of participatory modeling to support transport planning in the face of complexity. Methods: We used system dynamics modeling (SDM) to compare realistic policies, incorporating feedback effects, nonlinear relationships, and time delays between variables. We developed a system dynamics model of commuter bicycling through interviews and workshops with policy, community, and academic stakeholders. We incorporated best available evidence to simulate five policy scenarios over the next 40 years in Auckland, New Zealand. Injury, physical activity, fuel costs, air pollution, and carbon emissions outcomes were simulated. Results: Using the simulation model, we demonstrated the kinds of policies that would likely be needed to change a historical pattern of decline in cycling into a pattern of growth that would meet policy goals. Our model projections suggest that transforming urban roads over the next 40 years, using best practice physical separation on main roads and bicycle-friendly speed reduction on local streets, would yield benefits 10–25 times greater than costs. Conclusions: To our knowledge, this is the first integrated simulation model of future specific bicycling policies. Our projections provide practical evidence that may be used by health and transport policy makers to optimize the benefits of transport bicycling while minimizing negative consequences in a cost-effective manner. The modeling process enhanced understanding by a range of stakeholders of cycling as a complex system. Participatory SDM can be a helpful method for integrating health and environmental outcomes in transport and urban planning. Citation: Macmillan A, Connor J, Witten K, Kearns R, Rees D, Woodward A. 2014. The societal costs and benefits of commuter bicycling: simulating the effects of specific policies using system dynamics modeling. Environ Health Perspect 122:335–344; http://dx.doi.org/10.1289/ehp.1307250


Table of Contents Page
Summary of stakeholders 3 Table S1. Groups represented in the participatory system dynamics modelling process. 3 Sensitivity analysis 4 Table S2. Summary of formal validation procedures.
4 Table S3. Parameters tested and approach used in the sensitivity analysis of the simulation model. 5 Policy parameter sensitivity testing 9 Infrastructure costs 9 Table S4. Estimated costs per km of policy interventions. 9 Table S5. Sensitivity ranges for total and average annual costs of intervention policies (million NZ dollars). 9 Regional cycle network (RCN) 10 Figure S1. Upper and lower bounds for cycling injury outcomes under extremes for RCN component relative risk of cycle-vehicle collision.
11 Arterial segregated cycle lanes (ASBL) 12 Figure S2. Upper and lower bounds for cycling injury outcomes under extremes for ASBL relative risk of cycle-vehicle collision.
13 Self explaining local roads (SER) 14 Table S6. Sensitivity ranges for all mode shares under the self explaining roads policy.
14 Mixed universal policy (ASBL + SER) 15 Figure S3. Upper and lower bounds for cycling injury outcomes under extremes for ASBL + SER relative risks of cycle-vehicle collision and the proportion of cyclists travelling on arterial roads.
15 Figure S4. Range of injury outcomes for ASBL + SER policy seen under random simulation across normal distributions for the effect of components on collisions and the proportion of cyclists travelling on arterial roads. 16 Figure S5. Mode share outcomes for ASBL + SER policy simulated using random sampling from normal and uniform distributions of component criteria.
17 Table S7. Range of mode share and annual injury outcomes for all scenarios from the sensitivity analysis of policy assumptions.
18 Figure S6. Mode Table S1 summarises the stakeholders involved in the participatory system dynamics modelling, including the interviews and workshops to develop the qualitative model. In addition, a Māori steering group included 15 regional representatives. There was overlap in representation between the groups as some stakeholders represented more than one of the target groups. The groups represented were based on the requirements of the NZ Land Transport Management Act ([Anonymous] 2003). Table S1. Groups represented in the participatory system dynamics modelling process.

Groups represented Number of participants
People with disabilities 1 Māori communities 5 Pacific communities 3 Low income families 3 Young people 2 Regional transport policy makers 2 National transport agency 1 Public health 2 Local business association 1 Local tertiary institution 2 Local government 2 Regional government 2 Academics 3 People with disabilities self-identified as such and were represented by a member of the local council's disabilities steering group. Maori and Pacific representatives identified themselves as belonging to these ethnic groups and were drawn from a network of governmental and non-governmental organisations. Young people were defined as aged younger than 18. Public health representatives included professionals working at the Auckland Regional Public Health Service and public health academics. Adapted from Forrester and Senge (1980), Barlas (1996), Sterman (2000) and van den Belt (2004).

Infrastructure costs
The sensitivity of total and annual policy costs was tested to the range of infrastructure costs provided by Auckland transport (summarised in Table S4). The best estimate cost for the SER pilot study was used as an upper bound (range $100,000-300,000). A range of $100,000-300,000 was also used for the ASBL policy.
The ranges for total and annual average costs to 2051 under all policy scenarios are shown in Table S5. The model is order-of-magnitude sensitive to these assumptions for scenarios 2 and 5. numbers. In a worst case scenario, the same mode share is achieved, with a similar perception of safety, but 500 fatal and serious cyclist injuries and a large increase in the rate of serious injury to 10/1000 cyclists per year. The behavioural stability of injury outcomes under these upper and lower bounds is demonstrated in Figure S1.

Arterial segregated cycle lanes (ASBL)
Using upper and lower bounds from confidence intervals reported in the literature, a best case scenario for the combined effect of ASBL with best practice intersections on cyclist-vehicle collisions sees a cycling commute mode share of 18% by 2051 with 70% people considering cycling always/mostly safe. Five hundred fatal/serious injuries per year result by 2050 (similar to the worst case scenario for the RCN) but with a serious injury rate of 3.3/1000 cyclists per year. The worst case scenario has a similar effect on mode share and perception of safety but with 960 fatal/serious cyclist injuries per year and an injury rate of 6.4/1000 cyclists per year, slightly increasing over time. The behaviours of cycling injury outcomes over time for the best and worst cases are shown in Figure S2.
Analysing across 30 simulations using a normal distribution for these two relative risks again provides a narrower range of 233-563 serious and fatal injuries/year by 2051 and a serious injury rate of 3.57-5.75/1000 cyclists/year.
Random simulation across appropriate normal and uniform distributions for effects of the ASBL policy on cycling sense of safety and cycling good for work results in cycling mode shares between 10 and 24% by 2051. Figure S2. Upper and lower bounds for cycling injury outcomes under extremes for ASBL relative risk of cycle-vehicle collision.

Self explaining local roads (SER)
Worst and best case scenarios for the effect of SER on vehicle-cyclist collisions have similar effects on mode share. In a best case scenario this policy results in 158 serious/fatal cyclist injuries per year by 2051, and an injury rate of 2.9/1000 cyclists/year. At its worst the SER policy has similar effects, resulting in 182 serious/fatal injuries per year and an injury rate of 3.4/1000 cyclists per year by 2051.
Other effects of SER policy also have an impact on cycling injury outcomes. Best and worst case estimates for the effect of SER on the average speed of local roads result in a narrow range of annual serious/fatal cyclist injuries between 145-180 and a range of injury rates between 2.6 and 3.3/1000 cyclists per year. Testing the effect of SER on the proportion of peak time light vehicles on arterial and local roads using a range of reductions between 5 and 45% makes no difference to injury outcomes. Injury outcomes for this policy are most sensitive to assumptions about the effect of SER on the proportion of peak time cycling spent on arterial roads (baseline is 50%). A range of effects between a reduction to 45% and a reduction to 15% results in a range of serious/fatal injuries of 111-212/year and a range of injury rates between 1.8 and 4/1000 cyclists by 2051.
Random simulation across normal and uniform distributions for the effect of the SER policy on cycling perception of safety, cycling good for work and light vehicles hassle free results in the wide ranges for mode shares seen in Table S6.

Mixed universal policy (ASBL + SER)
Best and worst case scenarios were tested using the range of collision rates for the components of policies tested earlier, as well as the range for the SER proportion of cyclists on arterial roads. These two simulations result in a range of serious and fatal injuries between 223 and 1177/year by 2051 and injury rates between 0.7 and 3.6/1000 cyclists per year ( Figure S3). Simulating this policy over a range of effects using normal distributions for the relative risk estimates results in a narrower range of cyclist serious fatal injuries (201-755 per year by 2051) and injury rates (1.1 to 2.7/1000 cyclists per year by 2051), as shown in Figure S4.

Monte Carlo analysis
A Monte Carlo approach was used to randomly sample from distributions of variables determining mode share to provide a range for the mode share outcomes of each scenario.
The results of these analyses are summarised in Table S7. It can be seen that cycling mode share is order-of-magnitude sensitive to assumptions under scenarios 3 and 4. Some overlap between scenarios is also evident from the Monte Carlo analysis ( Figure ), but the model retains its ability to distinguish between scenarios 2, 3 and 5. Annual injury outcomes exhibit less overlap than the mode share outcomes overall, although a greater degree of overlap was again seen in the injury ranges for scenarios 3 and 5.
Assumptions about safety in numbers and variables influencing the effect of commuter cycling on all-cause mortality were tested separately. Changing the threshold for the safety in numbers effect to 5% cycling mode share did not alter the behaviour or order of magnitude of injury outcomes. However, simulating the power function from Jacobsen with no threshold changed the behaviour of injury outcomes. Simulating the range of relative risks of all-cause mortality for commuter cycling (using a Monte Carlo approach with the confidence intervals in the literature) altered the order of magnitude of all-cause mortality savings for all scenarios and disabled the ability of the model to distinguish between any of the active interventions.