Articles



Can Cycling Apps be Used to Inform Smart Infrastructure Planning?

This article assesses whether crowdsourced data relating to cycling travel patterns can be used to inform investment decision-making on infrastructure funding to identify where it will have the most beneficial effects. Transport data is amongst the most popular data used by digital application developers, with many examples of innovative smartphone applications offering traffic information to avoid road congestion. The UK Department for Transport expects to see many customer-friendly services and applications being brought to market and emphasises the need to remove restrictions on commercial use of data, so that it can be more readily shared.1 In relation to cycling there has been an explosion of apps based on user-generated data, OpenStreetMap data, GPS, and open data designed to meet the needs of both more and less confident cyclists. Applications focus on different aspects to improve the cycling experience – journey planning (e.g., Cycle Streets, MapMyRide), personal performance (e.g., Endomondo, Strava Cycling), creating online communities (e.g., Social Cyclist) as well as reporting crime (e.g., Check That Bike!) or road hazards (e.g., Fill That Hole).

For this article, the smartphone app Strava is used to understand how cyclists move throughout the city, and make inferences about where infrastructure investment would be most beneficial. Strava user activity is aggregated into a single map interface which can be viewed online, highlighting the popular and less popular routes of cyclists. It does this through a feature called segments, which provides information on sections of a cyclist route, including the total number of users and journeys made, the demographics of the user, and average journey speeds, and which any user can access.

Strava provides readily available cycling data, but planners, developers, and campaign groups are uncertain whether it is representative of actual cycling patterns. If a correlation between Strava data and actual cycling levels were defined, the ability to use Strava segment data as an indication of actual levels of bicycle use on specific roads could be a very useful tool for these groups. Conducting cycling surveys can be prohibitively expensive, particularly if data is required over a large area or time period. The ability to use Strava to obtain instant estimates of cycle use on most roads in an area would allow stakeholders to make more informed decisions. The use of Strava segment data may be particularly useful when conducting city-wide studies such as developing a cycle network, or for a pre-feasibility study where there is a limited budget available for detailed investigation. This article uses the UK cities of Manchester and Bristol, and the US city of Portland, Oregon, to assess the extent to which Strava data provides a representative picture of cyclist behaviour and activity within the city.

Methodology

The study team performed two quantitative studies comparing bike counts and Strava segment data in Manchester and Bristol. Strava also provided data from Portland for a third comparison. Box 1 summarizes study locations and data collected. Two analyses were performed: a) a comparison between Strava and observational count data in Manchester (This comparison also included an analysis of additional data such as age, gender, speed, and journey purpose. Strava data includes this information, and the project team obtained on-site information from observations and questionnaires. Strava also provided the number of app users in Manchester between 2011 and 2014.) and b) a comparison between Strava and observational count data at Bristol and Portland. All bicycle count and Strava segment data was collected during widely accepted ‘neutral’ days (i.e., Tuesday to Thursday) and months to eliminate the effects of long weekends or other holidays on travel activity.

In Manchester, the study team collected and analyzed questionnaire and Strava segment data from 10 different segments around the Oxford Road Corridor (Box 2), which is known for its high level of cycling activity and its wide range in demographics that includes students and commuters. In Bristol, the study team performed a short quantitative study comparing bicycle trips recorded by Strava and by observation counts over a 12-hour period at 39 different sites (Box 3). These sites represent a mix of road types in the Bristol area. Finally, the short quantitative study of Portland data provided by Strava compared 24-hour cycle counts and Strava trips on two specific segments with relatively high bicycle volumes over a 12-month period.

Results from the Manchester Analysis

App Use

Strava’s userbase in Manchester has increased year-on-year. Box 4 shows the monthly average number of active users for the 10 monitored segments in Manchester between January 2011 and May 2014. Segment monitoring has shown an approximate 100 percent growth in the active Strava userbase in Manchester from February to May 2014.

Demographics

Questionnaire data indicated that women account for about one-third of cyclists in Manchester while they represent about a quarter of all cyclists on the 10 Manchester segments observed. However, the Strava data revealed that men account for more than 90 percent of observed Strava users (Box 5) on these same segments, significantly underrepresenting women. But this is changing. Women users increased by 3.6 percentage points when looking solely at new Strava users on the Manchester segments since data collection began in February 2014. When narrowing these segments to side roads only, the share of new users that are female is even higher at 19.5 percent. Strava users more closely match questionnaire results for those aged 25 to 44. For the segments analyzed, Strava users overrepresent cyclists under the age of 34 and underrepresent cyclists 35 and older.

Speed

Average speed for all Manchester segments for all time users was 24.3 km/h across all road types. Speeds were slightly faster on main roads (26.1 km/h) and slightly slower on side roads (21.8 km/h). For just new users between February and May 2014, the average speed for all segments was slightly slower at 23.2 km/h across all road types. Main road (24.5 km/h) and side road (20.9 km/h) speeds were also faster and slower than the average speed for new users, respectively, but overall slower for new users than all users. For reference, Copenhagen’s “green wave” synchronizes traffic signals for a 20 km/h speed.

Journey Purpose

Both the Manchester questionnaire and Strava data recorded journey types (commute or leisure) for two segments: a main road and a side road. The project team then compared both of these results to 2014 worldwide Strava data provided by Strava (Box 6). The results revealed that the majority of journeys made within the Manchester segments can be classed as a commute trip rather than a leisure ride. However, the worldwide Strava data, which includes both rural and urban areas, shows a smaller percentage of commute trips and leisure rides become more dominant. For city planners, who are primarily concerned with urban routes, the results suggest that Strava provides meaningful data about utility cycling patterns.

Results from All Analyses

Correlation of Strava and Count Data

Overall, Strava users in Manchester represented approximately 9 percent of cyclists that were counted in the field when averaged for all the segments counted. The results from the Bristol analysis show that there is a reasonable correlation between the actual numbers of cyclists compared to cyclists logging their ride with Strava (Box 7). In Bristol, Strava trips represented 3.9 percent of the actual 12-hour counts on average.

The Portland analysis found that the relationship between the cycle counts and Strava trips remained extremely constant over the 12-month period. At one of the two count sites in Portland, Strava trips consistently represented 1 percent of actual 24-hour trips; at the other count site Strava trips consistently represented 2 percent of the actual 24-hour trips. The Manchester samples indicated that the correlation between two data sets is likely to be stronger if both counts and Strava data are obtained on the same date.

Road Type Preference

Based on the results from Manchester and Bristol only, there does not appear to be any significant difference in the results due to road type within the city. This indicates that trips on both main roads and side roads are well represented in Strava.

Conclusions

Cycling journeys recorded in urbanised areas using Strava are predominantly for commuting and not leisure rides. From the Manchester, Bristol, and Portland data it is evident there are differences in Strava usage among the cycling population, both between and within cities. In addition, the demographics of Strava users continue to change over time as new users join the app. The Manchester analysis indicated that younger male cyclists are overrepresented in Strava datasets, but not to the extent commonly assumed. The active user trend shows an approximation toward cyclist demographics observed during on-site counts.

Overall, the results suggest that Strava correlates reasonably well with actual cycling activity, and could be used to indicate how cyclists are using a city’s infrastructure. For city planners who are primarily concerned with utility cycling within the urban environment, Strava data would provide a useful complementary tool for understanding cyclist behaviour. Short baseline studies could be undertaken to identify whether the 9 percent (the Oxford Road area of south Manchester) and the 4 percent rate (Bristol) is applicable to other UK cities.

In order to exploit this potential market opportunity, Strava have developed a product called Strava Metro, which can provide local authorities and other organisations with detailed aggregated data across large areas in a geographical information system (GIS) format. It is understood that Transport for London, Glasgow, and several authorities in the US and Australia are already using Strava Metro. The potential uses include analysis of minute-by-minute cycle flows, origin/destination data, speed and junction delays, and the impacts of infrastructure changes. As Strava Metro is based on the same logged rides as the data we have investigated in this article, it is reasonable to assume similar results.

 

Notes:

  1. Door to Door Action Plan: Progress Report. Department for Transport. July 2014.

Image Header Source: Tejvan Pettinger (Creative Commons)