/Bikeshare-Equity

Combining journey data with community ethnicity data to evaluate Capital Bikeshare's performance on inclusion goals in Washington D.C.

Primary LanguageJupyter Notebook

Captial Bikeshare - Ridership equity

Captial Bikeshare (CB) is a public-private company in, Washington DC, with a commitment to inclusivity as part of its mandate.

Project Goals

  • Assess performance on ridership diversity goals set by CB
  • Develop new KPIs and make predictions

Method

  • Use geographical demographic information to cross-reference with journey start/end points
  • Analyze trips based on residential demographics

About the data

Trip Data

Caiptal Bikeshare trip data for April 2023. The data includes:

 #   Column              Non-Null Count   Dtype  
---  ------              --------------   -----  
 0   ride_id             389243 non-null  object 
 1   rideable_type       389243 non-null  object 
 2   started_at          389243 non-null  object 
 3   ended_at            389243 non-null  object 
 4   start_station_name  360858 non-null  object 
 5   start_station_id    360858 non-null  float64
 6   end_station_name    358630 non-null  object 
 7   end_station_id      358630 non-null  float64
 8   start_lat           389243 non-null  float64
 9   start_lng           389243 non-null  float64
 10  end_lat             388555 non-null  float64
 11  end_lng             388555 non-null  float64
 12  member_casual       389243 non-null  object

Source: https://capitalbikeshare.com/system-data

Notes:

  • Excludes staff testing trips, and any trips lasting less than 60 seconds (potentially false starts or ends).
  • Dockless bikes and e-scooters (i.e., potentially without a named start or end station) are included after a certain point.

Cleaning:

  • Exclude trip data outliers (improbable or impossible real-world conditions such as very long or negative durations)

Demographic data

American Community Survey (ACS) 2022. Source

Processing steps:

  • Extract census tracts within DC, exclude census tracts with under 1200 population as non-residential and/or tourist areas
  • Find the proportion of white (‘Race: One Race: White’) residents per census tract

Combine data

  • Find the two bikeshare stations closest to the center of each tract
  • Compare trip data from stations serving areas with the highest and lowest proportion of white residents
  • Adjust data to allow comparison between the two groups studied (to capture 10,000 rides it took 5 stations in areas with the highest proportions of white residents, whereas it took 75 stations in areas with the lowest proportions)

Map: 75 stations primarily serving areas with the lowest % white residents: ~10000 journeysMap: 75 stations primarily serving areas with the lowest % white residents: ~10000 journeys

Findings

  • Fewer trips to and from stations not in predominantly white areas
  • Fewer stations
  • Shorter trip durations (inferred from net km)
  • Few changes in marketing/public information policy despite research into demographic reach

Recommendations

  • Extend use of regression model with feature importance investigation to predict where new stations will have most impact
  • A/B testing of marketing/public information strategies based on new combined dataset
  • Resume user monitoring