Measuring Access with Network Distances

Measuring Access with Network Distances


Now in its fourth year, Azavea’s Summer of Maps Program has become an important resource for non-profits and student GIS analysts alike.  Non-profits receive pro bono spatial analysis work that can enhance their organizational decision-making processes and programmatic activities, while students benefit from Azavea mentors’ expertise.   This year, three fellows worked on projects for six organizations that spanned a variety of topics and geographic regions.  This blog series documents some of their accomplishments and challenges during their fellowship.  Our 2015 sponsors Esri, Temple University Geography and Urban Professional Masters Studies Program and Betsy and Jesse Fink helped make this program possible.  For more information about the program, please visit the Summer of Maps website.

One of the projects I worked on this summer was an analysis of the needs of English for Speakers of Other Languages (ESOL) students at The Family Support Center in the Sunset Park neighborhood of Brooklyn. The Center is a part of the NYU Lutheran Family Health Centers system and provides non-medical services such as counseling, benefit enrollment assistance, and the ESOL program. For the past year, they have conducted an extensive needs assessment survey with each student who has enrolled in the ESOL program. One of the goals of this project was to compare the spatial distribution of people with particular needs to the locations of organizations with relevant services.

My initial idea was to make two density layers – one for need and one for services. I would then assign each a scale from 0 to 100 and subtract the resources from the need to find underserved areas. I started by calculating the density of students with various needs – below is the density of students planning to get a job (left) and the density of students planning to get their GED (right):


This is useful because it shows that the populations are spatially distinct – they have separate patterns and are concentrated in different areas. However, a new problem arose after adding GED classes and vocational training program locations onto those maps (Job seekers and resources on the left, GED seekers and resources on the right):


There are hardly any organizations in Sunset Park offering GED classes. This meant that making a density layer of organizations wouldn’t be helpful because it would be zero where most of the students live. On the other hand, there are many more vocational training programs than GED classes, so it seems like a density surface might be more meaningful in this case. But this raises some questions:

  • Is distance to a service representative of how accessible it is? For example, is someone who lives 1 block from a service really 4 times more served by it than someone who lives 4 blocks away? What is a meaningful radius of influence outside of which someone is not served at all?
  • Is someone more served by having three vocational programs near them than just one? Certainly they have more options, and in the likely case that there is limited capacity at each organization, having more nearby could be meaningful.
  • Living near a service doesn’t mean that’s the one someone will use. For example, most students with young children appeared to live near a childcare center, but who’s to say if that’s the one their child attends, or if it’s one near their work or somewhere else entirely?

In the process of thinking through all of these questions, I calculated one of these service density rasters and subtracted it from the student need raster to show underserved areas, and it looked like this:


In the map above we can see the areas that are under- and over-served, but it seems like it’s doing more to obscure trends than highlight them. For example, the area just to the west of the Family Support Center has a huge need for employment. However this isn’t clear when looking at the need minus services because those students are near two vocational programs as well as the Family Support Center, which also provides vocational services. In contrast, the need near the program just north of the Family Support Center looks vastly larger because there is a lower density of services in that area. It’s possible that the three programs to the south are meeting the area’s needs while needs are unmet in the north, but it seems hard to draw that conclusion based only on the information we have. Its also worth noting the relatively small distance between all four of these programs – they are all within about a half mile of each other, so even the programs further south are fairly accessible to people living near the more northern program.

Given all the complications above, my original analysis approach didn’t seem to be appropriate for the situation. While that partly has to do with the assumptions it would require, it largely had to do with their goals for the analysis. If the Family Support Center were looking for the best location to open a vocational training program where they could have the most impact on unmet need this would have been appropriate. Instead, they are trying to understand where raw need exists in order to begin to address it through their own programs. Creating rasters of need minus services was preventing them from seeing what they really wanted to see.

In the end we settled on showing the density of need with half-mile and one-mile street network walking distances to each program. This allowed for some quantification of accessibility while still allowing the viewer to easily see where populations are located. For example, 71% of students who want to get a job are within a half mile of a vocational training program while only 12% of students who want to get their GED are within that distance of a class! Using this approach, the Family Support Center will be able to begin to address gaps in service based on the geographic distribution of their ESOL students and the programs available to them in the neighborhood.