This page gives significant additional detail on previous research, the development of the Location Affordability Index (LAI), and important caveats for understanding and interpreting the data.
To access documentation about the LAI including research conducted as part of the development of the Index and modeling code for researchers looking to replicate the LAI or create modified versions using different data inputs, visit the Documentation page.
The Location Affordability Index (LAI) builds on an influential strain of research that has sought a better understanding of affordability by looking at transportation as well as housing costs.
In 2008, researchers from the Center for Neighborhood Technology (CNT) published a paper in the peer-reviewed Transportation Research Record describing a transportation cost model– the Housing and Transportation (H+T®) Affordability Index– which used variables representing neighborhood and household characteristics to estimate vehicle ownership, annual vehicle miles traveled, and transit use. This model built on CNT's earlier work with the Brookings Institution and Reconnecting America and featured estimates for housing and transportation costs in 52 regions; it was expanded to cover 337 metropolitan statistical areas in 2009. Results from the H+T® Index demonstrated that the location of a home can have a significant impact on transportation behavior and spending, serving to reframe the conversation about affordability in regions across the country. Learn more about CNT's H+T® Index in this guide: Penny Wise, Pound Fuelish: New Measures of Housing + Transportation Affordability.
The Location Affordability Index (LAI) builds on previous research while incorporating significant improvements based on new research, third-party peer reviews, and expert and stakeholder consultation. These efforts include:
The LAI was developed by HUD and the Department of Transportation (DOT), with support from Manhattan Strategy Group and Center for Neighborhood Technology (CNT). It draws on input from a variety of sources:
Version 2 of the Portal (launched in September 2014) implemented key improvements recommended by the Technical Review Panel and third-party research most significantly by moving to a Simultaneous Equation Modeling (SEM) approach to model housing costs, vehicle ownership, and transit usage for both owner and renter households. SEM better incorporates and accounts for interaction effects on the model's dependent variables than OLS regression, resulting in a model that has greater econometric validity.
Other improvements include:
These changes have resulted in a more sophisticated model with far more comprehensive geographical coverage. The inclusion of non-metropolitan areas allows the LAI to cover the entire populated area of the 50 states and the District of Columbia (contrasted to Version 1, which covered 94% of the U.S. population).
LAI Version 3 (published in March 2019) provides standardized household housing and transportation cost estimates at the Census tract level for the United States. Like in Version 2, auto ownership, housing costs, and transit usage for both homeowners and renters are modeled concurrently using simultaneous (or structural) equation modeling (SEM) to capture the interrelationship of these factors. The inputs to the SEM model include these six endogenous variables and 18 exogenous variables.
As with previous versions, the Version 3 model is used to estimate housing and transportation costs for eight different household profiles, to focus on the impact of the built environment on these costs by holding demographic characteristics constant. In addition to moving to modeling at the Tract level rather than block group, Version 3 incorporates a number of other updates, enhancements and tweaks:
|ACS data vintage||2016 5-Year ACS||2012 5-Year ACS|
|LEHD data vintage||
Note: LEHD data is unavailable for 2014 in Wyoming. So, LEHD data for 2013 is used instead.
|Level of geographical granularity||Tract||Block Group|
|Catchment area for Local Job Density and Local Retail Density (variables 6 and 7)||Used simply the number of workers in the tract and the land area||Used ½ mile buffer around centroid of Block Group and take the union with the BG and used that geography to get Employees and land area|
|National Transit Database (NTD) vintage||2014||2010|
|VMT data vintage||2013-2015||2008-2010|
|Allocation of Fare Box revenue from NTD||Used NTD Primary Urbanized Area||Used AllTransit™ stops and frequency to allocate revenue and trips|
|Region of Transit Service provider in NTD||Urbanized Area||Metro/Micro Area|
|When BG/Tract not in Transit service area how to estimate α and β||Used the values from the nearest urbanized area that had good data||Used the national average|
Choose the function that gave on average the best OLS fit for each of the six endogenous variables:
|Optimized each exogenous variable by finding which function made the variables distribution the most normal|
|Endogenous Variable Interactions||Included interactions if they improved goodness of fit (see Table 3)||Included interactions based primarily on significance|
|Top and Bottom Code housing costs||No bottom or top coding||Bottom coded to 10th percentile housing cost (owner or renter) within BG and top coded to 90th percentile.|
|Household Income as Percentile of Tract Income||Included for each household profile||Not included|
|VMT model||Used odometer readings averaged over Tract||Used odometer readings averaged over Block Group|
Version 3 was prepared by Avar Consulting, with assistance from the Center for Neighborhood Technology. For more information, please see the Version 3 Data and Methodology documentation.
A significant volume of research and analysis has gone into creating this site in order to try to capture all of the features of the built and demographic environment that contribute to transportation costs: population density, walkability, transit access, employment access, per capita income, family size, and number of commuters. This has engendered a great deal of complexity and nuance that should be taken into account when using the tools, particularly to make policy decisions. HUD and the Department of Transportation (DOT) have worked throughout this project to ensure that these details are captured in order to maximize transparency about the development and proper use of these tools. The following are important considerations for using and interpreting the information found on this site, many of which we are endeavoring to address in our ongoing work.
The Index is primarily intended for use by researchers, developers, planners, and policymakers to help enhance their understanding of combined housing and transportation cost burdens, analyze the relationships between the affordability landscape and other factors (e.g. transportation infrastructure, development pressure, etc.), and communicate this information to the public and stakeholders. One limitation of the Index is that the majority of the data on which it is based comes from the American Community Survey. While this is not an issue for transportation costs (which are heavily determined by features of land use and the built environment that change very slowly), it does mean that the housing cost estimates correspond to housing costs during that time period. Therefore, it is not appropriate to use the Index for applications that require real-time housing cost data.
To account for the fact that housing and transportation costs are jointly determined by locational and household demographic characteristics, the Location Affordability Index (LAI) provides estimates for eight different household profiles for any location. These household profiles allow the Index to hold household characteristics–i.e. income, number of family members, and number of commuters–constant across each metropolitan area, making the Index vastly more informative and easy to interpret. However, this does not imply that every block group contains families matching each household profile. Rather, housing cost estimates for any household type in a given location approximate what a particular household would be expected to pay for housing IF they had lived in that location between 2010 and 2014 (with some additional caveats, detailed in the following sections).
The LAI accounts for a number of qualitative factors that impact the cost of housing including neighborhood walkability, access to transit, and access to employment (which to some extent can also be considered a proxy for access to commercial amenities). However, there are some qualitative factors the Index does not take into account. These include the quality/condition of housing stock, school quality, public safety, and natural amenities (e.g. a view of or proximity to a beach, river, forested trails, etc.). They should be considered as part of the model's unexplained variation: the amount of variability in housing costs it does not explain. (See Sections 5.5 and 6.2 of the third-party review of the LAI's housing cost model for an expanded discussion of this and the following issue).
There are also quantitative factors not accounted for by the models used to generate the LAI. A major factor not covered by the model is housing subsidies, either through the Mortgage Interest Tax Deduction or any form of rental assistance. This is due to the fact that the housing costs reported in the American Community Survey–either Gross Rent or Selected Monthly Ownership Costs–do not include these subsidies.
The models for autos per household, annual vehicle miles traveled, and percent of commuters using transit also omit common subsidies for transit and parking costs. Teasing out parking costs is particularly tricky since they are so often subsidized by taxpayers or bundled with the cost of housing. An analysis of the feasibility of including parking costs in the Index found that it would require an inordinate amount of time and resources to explicitly incorporate the cost of parking since there is no existing source for this data. (Donald Shoup's The High Cost of Free Parking is the definitive source on this subject.)
The transportation cost model also does not account for the value of time spent travelling. This is difficult to capture for at least two reasons. First, people value their time differently. Second, different travel types (e.g. commuting versus driving to the beach) and modes (e.g. driving versus biking versus taking the bus or subway) produce qualitatively different experiences whose values would be impossible to generalize.
Finally, the transportation cost model does not take into account the rising prevalence of telecommuting.