Analyzing Potential Areas For Divvy Expansion

Divvy is Chicago's bike share system which has been around since 2013. It is run by the Chicago Department of Transportation (CDOT) but was sold to Lyft in 2018. Today, it serves people through over 829 stations from all around the city.

• Classic Stations: Can support Classic Bikes, eBikes, and Scooters. The most common type of Station around the city.
• Lightweight Stations: Can only support eBikes and Scooters. Tend to be located on the fringes of the city.
• Divvy Public Racks: Certain public racks allow users to leave Dockless eBikes without having to pay fee for leaving a bike anywhere.

Many areas of the city still lack Divvy service and there are existing holes within the network. In order to be a truly equitable service, Divvy needs to serve everyone across the city.

The goal of this project is to identify different areas where Divvy could expand its service as well as identify different holes within the existing network of stations. GIS offers advanced tools that could be used to identify gaps within the Divvy network and determine good locations for potential expansion. This is a spatial problem and looking at it spatially helps reveal trends that might not be picked up.

They are in the process of trying to expand the network to the entire city and make sure that the service is available at an equitable basis.

This project details the development of GIS components in my undergraduate Senior Capstone project. I worked on a team for the project but what is included here was developed solely by me. We worked directly with the Chicago Department of Transportation (CDOT) who provided guidance on what they wanted us to work on.

The stated goal of the project was to "find spatial and socioeconomic barriers to accessing Divvy". We also developed a survey which was shared around, some of the results will be acknowledged in this project.

• Network Analysis
• Raster Calculator
• Hot Spot Analysis (Getis-Ord-Gi*)

The network analysis feature was used in ArcGIS based on the existing road network within the city of Chicago. The network analysis allows the software to calculate routes and create polygons based on a specified network. The polygon network feature is like a buffer, except it is based on the road network instead. Road data was downloaded from U.S. Government's data-sharing website, Data.gov. Therefore, the network analysis could be completed using the bike stations and the road network.

The network was calculated assuming a walking speed of three miles per hour, a moderate walking speed for an average individual (Schimpl et al., 2011). For the sake of this tool, it was necessary to include the villages of Norridge and Harwood Heights to capture the far northwest side of the city's road network. These suburbs are not part of the study area, but they are still built into the city's street grid.

The network does not include alleys, expressways, arterials, and ramps. This process involved deleting the alleys from the road datasets completely. Then, the expressways, arterials, and ramps were selected and exported as a new and separate data layer to keep them on the map for visual reference. Finally, they were deleted from the primary data layer, so all that remained were local roads and collectors. The population data came from Census Blocks highest level of detail available. The network was broken into tiers of five, ten, and fifteen minutes, using the dissolve method.

One of the report's main goals was identifying areas where new Divvy Stations could potentially go across the city. As a team, different factors were identified to measure areas where new stations could potentially go. We wanted to look through this in a variety of ways through different methods. One approach was to use the Raster Analysis and ultimately the raster calculator to add up layers.

What Was Measured:

• Distance From CTA Stations
• Distance From CTA & Pace Bus Routes
• Distance From Metra Stations
• Distance From Parks
• Distance From Existing Bike Routes
• Distance From Schools
• Business Density

In order to find infill areas, the "Distance Allocation" tool was used to measure how far the various attributes were from each other. The data was measured using the classify method in symbology using breaks of 750, 2,000, and 4,000 feet. The only exception to this was Business Density, which instead used the Kernel Density tool. For all rasters, Cell sizes were set at 30 by 30 feet. Raster data can only measure one data point at a time. Once all the distance layers were created, the Raster Calculator tool was used to sum up areas where all these features were shared.

Next, the newly summed-up layer was divided by the population data. This eliminated areas where nobody lives as well as normalize the factors. Finally, the areas beyond a 10-minute walk were clipped out of the map using the service layer created in the network analysis section. 10 minutes was chosen based on survey results in our capstone project. Respondents who lived more than 10 minutes were overwhelmingly less likely to use Divvy.

For hot spot analysis, I used all of the layers as listed above without dividing it by population. The Hot Spot Analysis tool only allows users to run it as a feature layer. I had to assign integers to the Raster using the "Int" tool which assigns a value to every cell, then use the "Raster To Polygon" tool to convert it to a feature class. The cells would remain 30x30, the same as the Raster Layer.

The polygon's table featured a column called "GridCode" which was a unique value that represented the value of the combined raster layer. Each cell has a unique value and would be used to calculate the quality of that area.

Next, this involved the use of Hot Spot Analysis (Getis-Or-Gi*). This was calculated using the "GridCode" field. I used Fixed Nearest Neighbor and Euclidean distance to measure this as well. I tried to incorporate the population using the Kernel Density tool, but the Hot Spot Analysis would not work with the population data incorporated into it. After running the tool with population, it created a checkered pattern that revealed nothing.

This section details the results that were found through the use of the Network Analysis tool.

This section is based off what was created using the Raster Calculator tool.

This was based off of :

• Distance From CTA "L" Stations
• Distance From Pace & CTA Bus Routes
• Distance From Metra Stations
• Distance From Parks
• Distance From Schools
• Distance From Existing Bike Routes
• Density of Businesses
• Population
• More Than 10 Minutes Away By Foot From Another Station

The areas in blue represent the best areas for Divvy usage based on trends as described above. The red areas are the worst areas for Divvy to go while the blue areas meet the factors most suitable for expansion. Much of the Southwest and Southeast sides are identified as areas where Divvy could work while this only confirms the potential that neighborhoods on the northwest side could offer.

The service areas only include roads and don't include bike paths or other natural walking paths people may take. Many roads don't have sidewalks or may not be safe for pedestrians. Many people walk at different speeds and cannot be assumed everyone walks at that 3mph pace.

This analysis doesn't reflect the nuances that reflect the on-the-ground situation. I did not go and visit the sites that were revealed in the final recommendations, and there may be more things going on to not make them good sites. There are more factors that could have been considered regarding new stations but whose data was not readily available. The density of restaurants, age, workplace population, locations of theaters, concert venues, and arenas could have been considered as factors that drive ridership as well. Also, using this calculated raster, it made the Loop appear like a bad location. It also showed up as a hot spot for where not to put Divvy in the Hot Spot Analysis. Due to time limitations, I was not able to identify what caused this.

When normalizing the data by population, it doesn't take into account industrial areas or business areas where no people live. This means that station recommendations do not take into account major trip generators where nobody happens to live. For example, the potential demand in the Loop is lower than it should be. This means a trip generator like a factory, park, or shopping center is not reflected very well. The new station recommendations should be seen as guidance for the general area they should go but, not a final recommendation. There are many different nuances of an area that should be taken into account.

These GIS tools helped identify different spatial trends and opportunities regarding holes and underserved areas within the Divvy network. These tools helped provide data to back up existing trends as well as identify some surprising gaps in service. The biggest takeaways of this analysis was that:

• More station density correlates with more ridership
• Much of the south side is undeserved when compared to the north side.
• It appears there are areas of the Northwest side that are primed for expansion

For our capstone project we made more formal recommendations about how to improve Divvy on a broader scale and wrote a formal research paper. Check them out below: