Child pages
  • Mann Final Project

Cassie Mann

UEP 232

May 6, 2014

Final Paper


Affordability vs. Accessibility: Mapping the Walksheds of Affordable Housing in Portland, Maine


Project Description

My project evaluated the current inventory of federally-funded affordable housing in Portland, Maine, to measure how accessible this housing is to different community resources, including schools, grocery stores, and public transportation. For this project I used data from the Department of Housing and Urban Development (HUD) showing three types of federally-funded affordable housing: public housing buildings, Low-Income Housing Tax Credit (LIHTC) properties, and multi-family assisted properties (those with a subsidy like project-based Section 8).

I measured ¼ and ½ mile service areas – “walksheds” – around each housing point, and then counted the number of each amenity that falls within that service area. I then created a table showing each affordable housing property and the amount of each resource that exists within its ¼ and ½ mile walksheds, allowing me to show which housing has high and low levels of accessibility. I also counted the number of units in each of these properties, in order to demonstrate how many units – not just properties – have certain levels of accessibility.

Ideally a project like this could help identify which areas of the city are most in need of more affordable housing, and measure how well the existing supply is serving the population. A project like this could also inform future efforts to develop affordable housing that is more accessible to valued community resources for low-income residents, particularly seniors and persons with disabilities who may be less likely to drive.


Spatial Questions


  1. How accessible is Portland’s existing affordable housing to community amenities, including public transportation, grocery stores, and schools?
  2. Which affordable housing has the greatest accessibility to these amenities?
  3. How many units exist in each affordable housing property
  4. How does the walkability compare between different types of affordable housing – does one type of housing have greater accessibility to amenities than the others?


Data Sources




Key Attributes

Low-Income Housing Tax Credit (LIHTC) Properties

This dataset contains points for affordable housing properties funded using the federal Low-Income Housing Tax Credit (LIHTC)

HUD (2011)


Address, Number of units

Multi-Family Assisted Housing

This dataset contains points for federally subsidized multi-family affordable housing (ie. project-based Section 8)

HUD (2012)

Address, Number of units

Public Housing Buildings

This dataset contains points for federally-funded public housing buildings

HUD (2012)

Address, Number of units

NextGen911 Roads

This line dataset contains public road centerlines and road names for the state of Maine

Maine Office of GIS


Street type (local, secondary, primary)

Metro Bus Routes

This line dataset contains bus routes for Portland’s Metro Bus system

Greater Portland Council of Governments (not available online) (2014)

Route number

Metro Bus Stops

This dataset contains longitude and latitude for locations of bus stops for Portland’s Metro Bus system.

Greater Portland Council of Governments (not available online) (2014)

Longitude and Latitude


Grocery Stores

Table of grocery stores (includes convenience stores) from Reference USA

Reference USA (2014)

Longitude and Latitude


Point dataset contains name and address info for schools in Portland

Maine Office of GIS (2012)




Data Preparation & Analysis Steps

Data Preparation


  1. Projected all data sources into NAD_1983_UTM_Zone_19N


  1. Clipped all data layers to the mainland outline of Portland (While the City of Portland includes several islands in Casco Bay, none of the islands had any federally-funded affordable housing on them, so I chose to omit them from the analysis, so I could show a more detailed view of the mainland city in my maps.)


  1. Geocoded bus stop data by adding data from a spreadsheet listing longitude and latitude data for each stop to my map, then displaying XY data on the map.


  1. Geocoded grocery store data by adding spreadsheet from Reference USA to map and then displaying XY data.


  1. Clipped schools data set to just schools within Portland, then used Select by Attribute to remove two schools that are located on the islands in Casco Bay (one of which was erroneously showing up on mainland Portland, and thus would have skewed my results).


  1. Used Select by Line tool to divide large public housing developments into smaller clusters of points in order to create more accurate walksheds. Because some of the developments were very spread out, using the mean point of the entire development might have resulted in walksheds that were artificially small and thus not very accurate. To account for this, I divided several of the larger public housing developments (Bayside East, Front Street, Riverton Park, and Sagamore Village) into smaller clusters of buildings. With each selection, I exported data as a new layer with the name of the cluster (e.g. Riverton South) then added to my map.


  1. Use Mean Center tool (in Spatial Statistics toolbox) to calculate the mean center point for clusters of public housing buildings. Because many of the public housing developments included a great number of buildings, I did not want to conduct the network analysis of each individual building, but rather I wanted to get a better sense of the accessibility of the development itself. Calculating the Mean Center allowed me to use this point as the starting point for the walksheds, achieving a middle ground between calculating walksheds for all individual buildings (which would have made for a very cluttered map) and calculating one walkshed for the entire development, which would have been far less accurate.


Figure 1 shows the Riverton Park development, which I divided into two clusters, North (blue squares) and South (green squares). I then found the Mean Center points (purple circles) and created the walksheds based on these points rather than the individual buildings.



  1. Used Select by Attribute to select only those roads with a Road Class of either Local or Secondary, and then exported these roads into a new shapefile which I used for my analysis. I did this in order to omit highways and larger roads from my analysis, as these are less walkable for pedestrians and thus less relevant for my project.


Network Analysis

  1. Used this new Local and Secondary Roads dataset to create a new Network Dataset in ArcCatalog for use in my Network Analysis.


  1. Loaded the locations for the housing datasets, starting with Multi-Family Assisted, and ensured that all were located properly.


  1. Set the “impedance” for my analysis at 400 meters (1/4 mile) and specified that the analysis should create overlapping polygons for each facility. The resulting analysis showed the ¼ mile walksheds as well as the roads within that ¼ mile service area.


  1. Exported the lines (roads) and polygons (walksheds) as shapefiles, and then added them to my map.


  1. Repeated steps 10-12 for a ½ mile service area.


  1. Repeated 10-13 for LIHTC properties and public housing, although with the public housing mean center points, I had to repeat all steps for each mean center, because they are individual data layers.


Figure 2 shows the ¼ mile walksheds for LIHTC properties, including the roads within that walkshed.

  1. Used the Spatial Join tool to join the data points for grocery stores to the walkshed polygons, so the walksheds would take on the data of any grocery store that fell within them. This resulted in a new shapefile with an attribute table that had a new column, “Count” showing the total number of grocery store points within each walkshed polygon.


  1. Used Spatial Join to join data points for schools to these newer polygons, so the resulting shapefile had an attribute table with a “Count” column for grocery stores and another column with a count of schools.


  1. Used Spatial Join to join bus stops to the polygons with grocery stores and schools, so my final polygons included a count for all three amenities.


  1. Added a “TotalPoint” column to the attribute tables for these polygons, and then used Field Calculator to make the TotalPoint column the sum of the counts for grocery stores, schools, and bus stops.


  1. Repeated these steps for each type of affordable housing (and for all the individual mean center points for public housing), and then created tables showing the individual counts of each amenity and total points within the ¼ and ½ mile walksheds.


  1. Used the “Total Points” columns to select the most and least accessible walksheds for each type of housing, and then created new layers from these walksheds, showing the walksheds with the highest total points in blue and those with the lowest total points in orange.


  1. Calculated the mean acreage and mean total points of the ¼ and ½ mile walksheds for each type of housing, which gave a general sense of how the types of housing compare in terms of accessibility to key amenities, as well as the overall size of their walksheds. 


Difficulties Encountered

The greatest challenge I had was of my own creation. Breaking the public housing developments up into clusters and then finding the individual mean centers for each meant that I ended up with fifteen different datasets (fourteen mean center points and one dataset with the six individual buildings), so I had to run each analysis fifteen times. Any change I made – even small things like changing the color of a polygon or size of a symbol, had to happen fifteen times. Looking back, I’m sure there was a faster or easier way to do this, but once I started down this road it would have been very difficult to change course.

I was also limited by time in terms of how ambitious my analysis could be. While I initially planned on including maps of relevant Census data for background, I ended up not having the capacity (or space on my poster) to include these. This was unfortunate because they could have added important context about the demographics of Portland residents and the spatial distribution of poverty in the city, but ultimately I needed to limit my scope to the major walkability analysis.

Another limitation I encountered was with the accuracy of my data, specifically a lack of connectivity of my roads data set. While overall the roads dataset I used had greater connectivity than the Census Roads dataset I compared it with in Assignment 4, it lacked connectivity at a crucial intersection between a main road and the driveway for one of the properties I was mapping. As shown in the image below, the driveway (running east-west in blue) fails to connect with the main road (north-south in white). For a walkability analysis, this is a significant problem, as it results in the walkshed being inaccurate and far smaller than it is in reality.

Figure 3 shows the lack of connectivity in roads dataset that led to artificially small service area.


Because this walkshed was not accurate I chose to omit it from my analysis and from the tables I included in my final poster. I also did not include it in the calculations for mean walkshed area and mean total points because it would have skewed those results. 

I also found that some data points were inaccurately located. One LIHTC property showed up in the wrong place due to its address being misspelled in the attribute table (the property was listed as 52 Frederick Road in the attribute table, which is a house in South Portland, but the correct address is 52 Frederic St in Portland). I tried to edit the data in ArcMap, but despite changing the address, town, zip code, and longitude and latitude, I could not get the data point to move. I ended up omitting it from my analysis so that it wouldn’t skew the other results.

Lastly, one significant limitation to this project is that it only includes federally funded affordable housing, so it ignores any non-federally funded affordable housing that may exist. Due to time constraints I was unable to gather data on state-funded properties, but that would be a worthwhile exercise for future analysis.


Concluding Thoughts

As mentioned above, I think the most significant drawbacks of my project have to do with the data used in the analysis – a few inaccurate data points and lack of connectivity of my roads meant that I was unable to accurately assess two different properties, which likely influenced my outcomes. Looking back, I think it was very valuable for me to have completed Assignment 4 (GIS Data Quality Assessment) for my study area, because I was able to know in advance that the roads dataset I was using had better connectivity overall than the Census TIGER dataset. Had I not known that, I might have chosen to switch to the Census TIGER data mid-project due to the lack of connectivity described above – but doing so might have jeopardized the accuracy of my process elsewhere. Examining and assessing my data in advance was very helpful to this process, and I only wish I had used more of my final project data in that assignment!

The Network Analysis tool was extremely useful for this type of analysis, and provided a great visual to show something that is otherwise often intuitive but not usually explicit –ie. we can tell when an area isn’t very walkable/accessible to things but we don’t usually have visual proof of that. Likewise, the Spatial Join tool was very helpful for this analysis and easy to use.

I think my approach was an interesting first step at evaluating walkability for affordable housing in Portland. As mentioned above, I think a future project could expand on this to explore how the affordable housing stock aligns with demographic data, or to measure the walkability to a broader range of amenities beyond the three I chose. A future project could also compare the walksheds of affordable housing with those of newer market-rate developments, to gauge whether there are significant differences (in many cities I suspect there are!). A future analysis could also look more closely at the roads included in each walkshed – perhaps combining the GIS analysis with some ground-truthing to see whether the roads are truly walkable.

On the whole, I think this kind of analysis could be very useful for advocates to call out problem areas and for policymakers and planners trying to determine where to site affordable housing or different amenities. As many cities, including Portland, work to limit reliance on cars, it is crucial that policymakers and planners work to make housing more walkable for people at all income levels.



1. Shen, G. (2005). Location of manufactured housing and its accessibility to community services: a GIS-assisted spatial analysis. Socio-Economic Planning Sciences 39 (2005) 25-41.

In this article, the author used GIS analysis to evaluate the accessibility of manufactured housing to “public and community facilities” (PCF), which include both positive and negative facilities. Positive facilities include offices, stores, shopping centers, public transit, banks, churches, schools, etc. Negative facilities include heavy industrial or manufacturing, highways, airports, landfills, and others. This article was useful in guiding my thinking about how to measure the accessibility of Portland’s affordable housing. The authors used the buffer tool to measure accessibility of manufactured housing and compare it to that of other types of housing, but I used the network analysis tool instead to measure walkability and not just proximity.


2. Dawkins, C. J. (2011). Exploring the Spatial Distribution of Low Income Housing Tax Credit Properties. U.S. Department of Housing and Urban Development (HUD). .

This report completed by HUD examines the spatial distribution of Low-Income Housing Tax Credit properties within ten large U.S. cities, to evaluate whether these properties are consistent with HUD’s aims of de-concentrating the supply of subsidized housing and distributing it more evenly within cities. While this study does not utilize GIS, it was nevertheless useful in providing background for my project. The study found that LIHTC properties tend to be clustered in areas with higher poverty rates and higher non-white populations. While I did not end up having time/capacity to include Census data in my project, it would be interesting to compare this article’s findings with the reality in Portland in a future project.


3. Pendall, R., Theodos, B., & Franks, K. (2012). Vulnerable People, Precarious Housing, and Regional Resilience: An Exploratory Analysis. Housing Policy Debate 22:2, 271-296.

The authors explore ways to measure the resilience of households and communities to “shocks or strains” of various types, including the foreclosure crisis, long-term poverty and economic insecurity, as well as catastrophic events such as terrorist attacks or natural disasters. The authors discuss several different criteria for vulnerability of households and communities that make them more likely to live in “precarious housing situations.” This article was useful in informing my thinking about which demographic variables I wanted to include in the contextual maps I had planned to produce to accompany my accessibility maps.


4. Nicholls, S. (2010). Measuring the accessibility and equity of public parks: a case study using GIS. Managing Leisure , 6:4, 201-219.

In this article the author examines several approaches to measuring the accessibility of public parks, including the “straight-line radius” (or buffer) tool and network analysis. She compared the outcomes of both of these analyses, suggesting that the network analysis tool may be more accurate, but that the straight-line tool is more often utilized by public agencies attempting to measure accessibility. She also discusses how these different tools yield different results in terms of equity – in her study, equity relating to the number and demographics of people who are served by each public park. This article was useful for me in that it helped justify my decision to use network ana lysis for my project, rather than a simple buffer.