🚲 Bike Share in the Time of COVID-19

The COVID-19 pandemic disrupted transportation patterns and routines, with many people shifting to low-emission transportation. This period provides a glimpse of a future shifting away from automobile dependence. In this project we compared bike share usage in San Francisco across three time periods - before and during the pandemic, and immediately following most re-opening.

To better understand our data, we performed exploratory analysis of the main Bay Wheels dataset.

We defined three discrete periods to compare in our analysis, based on San Francisco shelter-in-place and reopening mandates.

How did Bay Wheels behavior change from pre, peak, & “new normal” pandemic periods?

Rides became more dispersed, over both space and time. In the peak pandemic and new normal periods, trips more frequently began from bike stations away from the city center:

We also observe a shift in ride start times from regular “commuting hours” to the middle of the day:

Clustering of a network is a good indicator of how interconnected its nodes are. In this case, our nodes are our stations and connections are trips between station X and station Y. We see a decrease in clustering during the pandemic, followed by a subsequent increase during the “new normal” period. This supports the dispersion findings: trips were less clustered during the peak pandemic period, perhaps due to riders’ more irregular routines during this period.

Another common network metric is the average in-degree of nodes: in this example, the in-degree corresponds the number of unique trip origins for a given station. In-degree increases over the study period, though this is attributed to the growth in the number of stations over time.

As a last step, we wanted to look at bike share behavior by demographic segment. To do this, we conducted a Principal Components Analysis (PCA) on a station’s demographic variables to create clusters of stations located in areas with similar demographic characteristics. The top principal components distinguished between stations in largely white and high-income areas vs. stations in largely black and low-income areas.

Overall, the trips involving stations in the white/high-income cluster saw the biggest decrease in the number of rides during the pandemic period, and did not rebound to pre-pandemic trips compared to other clusters – perhaps a vestige of more knowledge workers working from home. Comparatively, there were much fewer rides taken from stations located in black/low-income areas, though those did appear to rebound to pre-pandemic numbers.

Datasets

• Bay Wheels
• ACS Census 5 year estimates
• SafeTrec Bike Crash Data
• SF Open Data

Methods

• NetworkX for network analysis
• Plotly, ArcGIS for mapping
• Average Shortest Path, PCA, K-Means Clustering
• Trip Distribution Modeling