Here at Conveyal, we do a lot of transport analysis using accessibility metrics, which are simply measures of how many destinations you can reach in a given amount of time (for instance, if you live downtown, you might be able to reach 500,000 jobs within an hour’s travel time on transit). This is a measure of your potential access, without having to model individual choices about what commute someone might actually choose. This makes the measure both simpler to calculate and more objective than the traditional metrics used in transport modeling, for instance number of people choosing to take transit to work.
This works very well for transit, and we’ve put a lot of effort into making sure we accurately represent the experience of taking transit. However, cycling has a somewhat unique problem in that you actually cannot disconnect behavior from potential access, due to different levels of tolerance for cycling in different environments. Most people who take transit are comfortable taking any transit vehicle, and most people who drive are comfortable driving on any road, so there is no need to incorporate a willingness to travel using a particular vehicle or road. However, the same is not true of cycling: there are streets where cyclists are legally permitted, but where many or most people would not ride. This has a very significant impact on accessibility: if you could bike somewhere legally, but it requires cycling on a very busy road with no bike lane, you don’t actually have access to that place (unless you’re one of the few cyclists willing to mix with traffic at high speeds).
Including willingness to bike in accessibility measurements allows you to measure the potential impact of bike project (for example, new lanes or paths) not only in terms of infrastructure, but also in terms of what it connects people to. For example, opening a bike lane on street that was previously very uncomfortable for cyclists actually opens up opportunities to reach places that previously weren’t reachable at all (since people would simply refuse to bike on that street). The World Bank’s Transport Group was interested in being able to accurately measure accessibility by bicycle and quantify the effects of bicycle projects, and partnered with us to investigate the problem.
We’re not the first people to observe this problem or to try to address it. There are many metrics of bikeability or bicycle level of service. Probably the best known is the Bicycle Level of Traffic Stress (LTS) methodology. This assigns each street an LTS score from 1 to 4, with 1 representing bike paths and low-volume residential streets where most anyone is comfortable riding, and 4 representing high volume, high-speed streets where only very few are comfortable riding. Level 2 represents a level of stress most people are comfortable with, including bike lanes on higher-volume streets.
This methodology is widely applied, and has data requirements that are available in most developed-world cities (e.g. lane widths, presence of a centerline, etc.). However, in the developing-world cities where the World Bank was interested in evaluating the effects of investment in bicycle infrastructure, these data are often not available. However, OpenStreetMap data is. While OSM doesn’t contain much of the information needed to construct the LTS metric, we suspected that we could infer most of the attributes relevant to LTS from other information in OSM, for example roadway classification. One concern with this inference is that you will mislabel some streets. If you completely prevent traversing those streets, you may inadvertently cause an enormous decrease in accessibility. We solve this problem by simply allowing people to walk their bikes through high-stress areas, which meants that the effect of accidentally mislabeling an important street is minimized, because if it is not possible to go around it, the router will just have users walk their bikes through it.
We developed a set of rules, which we call “Surrogate LTS,” to create similar classifications to the original LTS methodology based on OpenStreetMap data:
|Does not allow cars||LTS 1|
|Is a service road||Unknown LTS|
|Is residential or living street||LTS 1|
|Has 3 or fewer lanes and max speed 25 mph or less||LTS 2|
|Has 3 or fewer lanes and unknown max speed||LTS 2|
|Is tertiary or smaller road|
|Has unknown lanes and max speed 25 mph or less||LTS 2|
|Has bike lane||LTS 2|
|Is larger than tertiary road|
|Has bike lane||LTS 3|
|Unsignalized intersections receive the highest LTS of any cross street|
When we apply this methodology to Portland, we get the following result:
And we can also use it to make an isochrone, showing the area you can bike to within 20 minutes depending on your tolerance to traffic stress:
It’s extremely important that we accurately model willingness to cycle on particular streets when we build models, otherwise we will drastically overstate the accessibility afforded by cycling. By using this methodology, we can also quantify the impact of potential bike projects on the ability of citizens to access their city, which we can use to compare or justify investments in bicycle infrastructure. This project is still a work in progress, so we’re all ears—if you have feedback, please let us know! We’ll post updates here as we make progress.