How to Use a Bike Share System?

An analysis of an Austin, Tx Bikeshare program from 2014–17

Introduction

Pre-Covid, bike share programs were on the rise in the U.S, with more than 60 cities adopting some form of bike-share system since 2007 according to the National Association of City Transportation Officials (NACTO). These programs have been a quiet success in many of the countries larger metropolitan cities with a record breaking 100,000 rides per day performed by New York cyclists in 2019.

But how exactly are consumers interacting with these systems? Were these 100,000 rides completed by avid cyclist biking to work everyday, or one time users looking for a new thrill? How do weather, and other environmental factors play a roll in this growing trend?

To analyze these questions, I used the Austin Bike Share Trips dataset, available on Google Public Data, in conjunction with weather data collected from the Austin KATT station.

The datasets contain information on over 649,000 trips taken between 2014 and 2017 along with corresponding climate data. The duration, start and end locations, subscriber type of each trip were captured by the study.

Below you can see the geographical layout of the 72 stations which were the focus of the study, overlaid on a map of downtown Austin, TX .

Figure 1: Austin TX, Bike Share Station Locations

Part I: How did weather conditions affect the frequency of trips?

First I analyzed the ways in which weather conditions affected the number of trips taken each day. As might be expected, the below heat map shows a high degree of correlation between related fields (ex. TempHigh, TempAvg, TempLow).

r

The two things of note:

1. Consumers were more likely to engage with the service on days that had higher temperatures and greater average visibility.
2. Less consumers utilized the service on days that had higher humidity, precipitation or wind gusts.

Breaking down the correlations with the number of trips taken per day by subscriber type, I found that the above trends were especially present in Local30 subscribers, consumers with monthly subscriptions.

Part II: How did different subscriber types utilize the service?

Here you can see a list of different subscriber type utilization rates in descending order. Here utilization is taken to be the total number of trips over the period.

Figure 2: Subscriber Type Utilization

Walk up users were by far the most common type of . We can also study the below box plot with outliers removed, to see that on average walk up users took longer trips than any other subscriber type.

Part III: What was the traffic distribution across all of the stations in the program?

Perhaps the question, that most drew my interest was how consumer traffic was distributed across the various stations. Which stations accumulated bikes? Were there any stations that would have run out of bikes, with no outside intervention?

Here we see a graph of station locations, where the size of each station is indicative of the in degree, bikes coming into the station, subtracted from the out degree, bikes leaving the station. Nodes, with a larger size indicate locations where bikes tend to pool.

In the above graph, we see that rather than be evenly distributed throughout the network, bikes tend to pool in a few key locations. Additionally, these key locations tend to be immediately surrounded by stations with a significantly lower degree.

Diving further into the analysis of the network of stations, we see that historically, the nature of most stations appears to be fairly constant over the three year period. Stations, that tend to have a net flow of bikes out, maintain that net negative flow, while the reverse is true for stations with a positive flow.

We see several blips of high activity in the 2D station activity map. One of these at station, 1007 with a net degree of 897, corresponds to a station near a voting station during the November of the 2016 elections. Although no direct causal relationship can be drawn from the data presented, a potential question for future analysis emerges.

Can station activity be used as a predictor of the timing and location of major events taking place in a city?

Conclusion

In this article we analyzed how users in Austin, TX interacted with their local bike share system.

1. We saw that adverse weather conditions such as humidity and precipitation had an adverse effect on the utilization of the bike share system, while warmer temperatures and higher visibility had positive effects. We also saw that this trend was more prevalent in monthly subscribers.
2. We looked at how different subscribers utilized the service. We saw that walk-up users, not only took the most trips, but also on average took longer trips,
3. Finally we looked at the traffic distribution across various stations. We saw that stations that tend to collect bikes tend to be surrounded by several stations that historically tend to lose bikes.

Our study also raises additional questions. Can we take the analysis a step further and begin to predict the location and timing of city events based on bike share traffic? How do the results seen in Austin, generalize to other cities?

To learn more about the analysis, see the link to my Github, here