Spring 2019
Under Prof. Joe Ferreira Jr. and TA Rida Qadri
The experience of being a pedestrian can be frustrating and difficult when street systems are not designed with pedestrians in mind – for example, when crossing the road legally requires detouring some distance to get to the nearest crosswalk. Existing “walkability” metrics such as the Walk Score tend to measure “walkability” in terms of access to amenities such as restaurants and schools. In this project, I examine “walkability” at a more fundamental and granular level: For a given point, how much unnecessary distance is needed to walk to nearby points?
This project focuses on the development of an R script that uses the igraph library and OpenAddresses data to build a graph of addresses across the city with each address connected to its neighbors. The script queries the Google Maps Directions API for walking distance data, and then calculates a metric of walkability by comparing the actual walking distances to the straight-line distances for each address. The resulting map highlights regions of better or worse pedestrian convenience and may be useful as a tool for building a more complete understanding of walkability.
Earlier in the semester, I needed to visit an office building in Brighton (50 Soldiers Field Place) to buy a camera lens from a seller on Craigslist. It wasn’t too far from MIT, so I decided to ride my bike over. The bike route looked pretty straightforward on Google Maps:
However, at the very end of the route, Google Maps asks you to make a U-turn on Soldiers Field Road in order to get into Soldiers Field Place:
It looks innocuous enough, but this is what Soldiers Field Road looks like in reality – a four-lane road that is almost a highway. The blue arrow is what Google Maps is asking me to do.
I didn’t see very many good alternatives, so I reluctantly jaywalked across the four lanes. However, when it came time to head back home, I didn’t want to have to jaywalk again. Google Maps was not very helpful still, so I decided to walk to the nearest intersection on Leo Birmingham Parkway, thinking that I could probably cross the road there and be on my way.
However, as it turns out, at all the places on Leo Birmingham Parkway one might expect to find a crosswalk, there are no crosswalks:
(The red unlabeled marker is where I was standing.)
So how does one cross the street? This is how you get from 50 Soldiers Field Place across Leo Birmingham Parkway legally as a pedestrian:
What should really be a five-minute walk is actually a 31-minute, 1.5-mile detour if you are a pedestrian.
My first thought: “This is terrible!”
But after I got home and told this story to a few friends, I started to wonder: “Are there more places in the city like this, and could I somehow find them computationally?”
In this project, my overall methodology was as follows:
To build my set of points, I began with address data from the open-source and free dataset at https://openaddresses.io. First, the addresses are filtered so that only those that fall within the area of interest remain. The OpenAddresses data is unevenly distributed (there are more points where development is more dense); to create an evenly distributed set of points, I created a set of evenly-spaced points (using the st_make_grid function in the sf package for R), and then kept the OpenAddress point nearest to each grid point. I used a 100 meter spacing for the grid, and limited address matching to within 50 meters of each grid point.
Using the igraph package for R, I then built a non-directed network with vertices representing points and edges representing connections between neighboring points. The edges in this network were exported as a data frame in R, providing a list of all unique origin-destination pairs. The walking distance for each origin-destination pair was found using the Google Maps Directions API and then saved back into the igraph network.
To calculate the distance ratio (walking distances/straight-line distances) for a vertex, the distance ratio was first calculated for all incident edges of that vertex. These results were then averaged for each vertex.
The resulting point data, mapped. There are a little over 2000 points here, in and immediately bordering Cambridge:
Interpolated using inverse distance weighting (same map as at the top of this page):
The less walkable areas in Cambridge can be roughly classified into a few distinct types:
Campus, specifically MIT campus: Because Google Maps does not route through private buildings, and MIT’s main campus consists of several large and connected buildings, walking routes on main campus can be very long in order to detour around buildings. There also appear to be issues with how Google Maps handles walking directions along Memorial Drive.
Boundaries, most strikingly rail lines in Cambridge: Rail lines result in clear patterns of less walkable areas, most clearly in North Cambridge (the commuter rail line west of Porter Square) and along Vassar Street.
Long blocks or dendritic road designs: Typically residential, long blocks result in more walking needed to move from the middle of one block to the middle of the next block. This can be seen in many residential neighborhoods across Cambridge, such as immediately south of Inman Square.
This project was a valuable experience in learning to use R as a GIS tool. The resulting walkability metric differs from existing metrics in its focus on street networks (rather than amenities) and in its level of detail, and may be useful for identifying problem areas in pedestrian infrastructure in addition to other methods of evaluating pedestrian infrastructure quality. Future investigation could use this tool to map cities other than Cambridge, or compare this metric against other demographic or land use data.
Thanks to Prof. Joe Ferreira and TA Rida Qadri for all their help, and for a great semester!