A Methodological Review of the Center for Neighborhood Technology s Housing + Transportation Affordability Index

Transcription

1 A Methodological Review of the Center for Neighborhood Technology s Housing + Transportation Affordability Index Final December 8, 2010 Prepared for National Association of Home Builders Prepared by Abt Associates Inc Montgomery Ave Suite 800 North Bethesda, Maryland 20814

2 Executive Summary Research and common anecdote suggest that the cost associated with a household s transportation can be a significant share of household expenses and shape overall housing affordability in complex ways. Numerous organizations have studied and developed tools both to explore this connection and provide much-needed additional data for scholars, policymakers, and the general public. The Center for Neighborhood Technology s Housing + Transportation Affordability Index ( Index ) is one such tool. Beginning with the first application of its methodology in 2005, the Index has transformed into a robust, web-based, interactive map that provides estimates of transportation and housing costs for current residents down to the neighborhood level ( Partnerships with the Center for Transit Oriented Development, Center for Housing Policy, Urban Land Institute, Brookings Institution, and many local public and private entities have aided this development. The Index has increasingly been used in policymaking as its capacity has grown. Because of its interest and advocacy in housing affordability, the National Association of Home Builders is exploring the Index s methodological underpinnings, assumptions, and data sources as well as the reliability of its application as a policy tool or even a consumer information tool. The investigation described in this document reveals several key findings: A key concern about the Index is that it is not transparent. The Index contains several assumptions and logic calculations that are not disclosed. The Center for Neighborhood Technology (CNT) rightly notes that the Index is its intellectual property, and it is within its rights to obscure the inner workings of the Index. However, that complicates its use as both a scholarly work and a policy-making tool. The extent of peer review an important component of the process in scholarly research of the current version of the model is not clear. A previous version of the econometric models behind the Index s transportation cost estimates was published in a journal in 2008, a process that included review by anonymous peers. CNT reports that the current model s assumptions, calculations, and methods have been reviewed by researchers from several institutions; however, lack of publication in a peer-reviewed journal makes it difficult to assess the extent of peer review. Many of the data used for the Index s independent variables are out of date, or their respective sources are not regularly updated. This is particularly true for housing costs, which have undergone tremendous fluctuation in the past

3 decade. The Index provides an addendum for gasoline price sensitivity, but otherwise does not account for current data. The housing data used by the Index details the costs for current residents. Most other known affordability indices rely on home and/or rental market prices for potential residents. Effectively, this makes the Index applicable only to current residents and not prospective residents. The Index is a work in progress. Although the current release of the model is static, the development team plans for a new release in the spring of 2011 that will rely on almost entirely new data, include new features, and likely reflect changes to econometric specifications and variable definitions. Such change complicates the timing of policy being made that relies on the Index s outputs as justification. The Index does not account for dynamic interactions between transit development (and subsequent transit costs) and housing development (and its respective costs). To some extent, housing costs are high because of a neighborhood s proximity to public transit, and on the flip side, housing is less expensive further from public transit. The Index lacks a direct connection between housing costs and transit availability so cannot guide policy makers in determining whether housing that is currently affordable will remain affordable once new transit is built. Some of the regression models used for the transportation costs analysis may be unnecessarily complicated and contribute to the Index s lack of transparency. For example, CNT researchers estimate the number of vehicles per household with a proprietary estimation that is a function of household and location variables though data to represent that information is readily available from the 2000 US Census. In other cases, insufficiently detailed data are used. For example, auto use and ownership costs are estimated using weighted national fleet averages that may not be representative of actual neighborhood auto costs. The Index is somewhat limited as an index in both method and context. The mathematical and economic definition of an index is as a number (or price) that indicates the location of a unit (like a neighborhood) in a list, scale, or array of numbers or prices; indexes are also primarily employed because their simplicity allows for ease of comparisons. In its current map-based incarnation, the Index does provide visual comparisons within metropolitan regions of costs, but it does not do this between regions nor does it rank or apply a scale to neighborhoods within regions. Just as importantly, the Index s complexity limits its ease of use and therefore its usefulness to consumers as a decision-making tool. Perhaps of most interest are the limitations of the Index as a planning, underwriting, or policy tool. Virtually all of the findings listed above prompt a variety of questions regarding the Index s current and potential use in policymaking. In particular, the complexity and the proprietary nature of some of the methodological assumptions cast a shadow over all potential policy

4 applications. This obscurity prevents public transparency and even full involvement of local professionals in local planning regarding density changes, infrastructure investments, or land-use modeling. Because the Index produces general estimates for populations rather than modeling of financial capacity for individual households, its use in mortgage underwriting is also problematic. Additional analysis may suggest that the Index is an appropriate tool for certain policy processes (like overall neighborhood performance indicators) and not others (like the siting of housing or transportation developments). For example, it is not at all obvious that areas with high transportation costs are good candidates for public transit investments by definition, these are places where the average household is already well served by private transportation (owns two or more vehicles). They are also more likely to lack the density required to support public transit. Further, although it seems logical that affordable housing be located in areas with low transportation costs, this approach can work at odds with other goals, such as reducing concentrations of poverty and locating affordable housing near jobs, which are increasingly located in the suburbs. Ultimately, improving overall housing affordability for US households is a goal shared by both the Index s creators and the NAHB. While the Index is one tool for assessing the current state of affordability of the two largest household expenses, both its technical composition and its end purpose and uses are unclear. Given its current weaknesses, the Index s potential and feasible policy applications should be further explored before it is implemented for policy purposes.

5 Review of the H+T Index December 2010 Page 5 Table of Contents Executive Summary Introduction... 6 The Development of the Index... 7 Housing and Transportation Affordability in Policy Index Components Index Variables Cost Equations Strengths and Weaknesses Data Econometric equations Cost equations Use as an Index Use in Planning Practice Use in Research Implications for Policy Conclusions Citations... 33

6 Review of the H+T Index December 2010 Page 6 1. Introduction Several well-known indexes attempt to measure the affordability of housing around the country. These include: The National Association of Realtors Housing Affordability Index NAHB/Wells Fargo Housing Opportunity Index MIT s Housing Affordability Initiative (HAI) Index PMI s U.S. Market Risk Index In all of these measures, affordability is defined as the housing costs of a specific community (defined in turn as a neighborhood, city, state, or the nation) in relation to the average or median income of relevant households to the community. Relevance is defined differently for each measure, and can mean households within the same metropolitan area, region, or even the nation that are likely to move into houses in the community being considered, or indexed. Communities are assigned a score or value along an index based on that affordability to easily compare values across communities. While all of these measures have specific respective limitations, they are all most obviously limited by their focus solely on the costs or prices of housing as opposed to other household expenses or house characteristics. Other things that affect affordability more broadly include transportation costs, housing quality, local crime rates, costs of utilities, and school quality, among others. However, it has proven difficult to measure these aspects of housing for places across the nation, primarily because of a lack of consistent data. For example, MIT s HAI Index attempts to measure housing affordability more broadly across non-housing price indicators, but it is calculated only for the Boston metropolitan area. It is not surprising, then, that there is considerable interest in the newly developed Housing + Transportation Affordability Index (the Index ) which adds another category to the calculation of housing affordability based on the cost effects of a house s location: household transportation costs. The purpose of this report is to assess whether the Index introduces transportation costs into affordability measures successfully both in its methods for indexing household affordability and, also, as a tool for public policy. The first section of the report describes the development of the Index, the overall growth in public policy interest, and applications of the Index in assessing the impact of transportation costs on housing affordability. The second section documents the current Index itself including the econometric models used (to the extent that these were available), the variables, data, and cost equations used in the Index. The third section reviews the technical strengths

7 Review of the H+T Index December 2010 Page 7 and weaknesses of the Index and the final section lays out its strengths and weaknesses as a policy tool. A set of overall comments concludes the report. The Development of the Index The Index was developed by the Center for Neighborhood Technology (CNT) and the Center for Transit Oriented Development (CTOD) as a project of The Brookings Institution s Urban Markets Initiative in The tool now provides output for 337 metropolitan areas on the web ( and has become a centerpiece for policy discussions among advocates for transit-oriented development and smart growth. Numerous intellectual and advocacy projects preceded the Index s creation, however, and the Index itself has undergone many revisions since In 1994, one of the earliest studies linking location efficiency (or the minimization of household transportation demand) and transportation costs was published. 1 Published by the Natural Resources Defense Council, the study predicts annual automobile costs as a function of neighborhood density, transit access, and automobile ownership and use costs. The Index ultimately incorporates much of the conceptual approach outlined in this paper. In 2002, a paper by CNT s chief research scientist and others was published in Transportation Planning and Technology that develops a preliminary model for predicting auto ownership and vehicle miles traveled (VMT) per car and their associated costs. 2 The authors determined that these variables are primarily a function of the neighborhood s residential density, average per capita income, average family size, and the availability of public transit. 3 The 2005 pilot of the Index focused on the Minneapolis-St. Paul area, and was developed by CNT to estimate transportation costs by characteristics of neighborhoods and households. The model was documented in a 2005 working paper; a paper by CTOD with CNT that was published the next year by the Brookings Institution s Urban Markets Initiative expanded the Minneapolis-St. Paul metro area analysis and formally named the Housing + Transportation Affordability Index. 4 Numerous additional monographs have been released since then, often accompanying the application of the Index to a new metropolitan area. A more comprehensive, peer-reviewed paper was published in 2008 in the Transportation Research Record. 5 Along with its scholarly advance, the Index developed as an applied tool soon after its first application in In 2006, the Center for Housing Policy, CNT, and University of California Berkeley released a study that applied the Index s method to 28 metro areas. 6 By the 2008 publication, the Index had been expanded to cover 52 metropolitan areas whose output was available on the Index s first web-based release. The March 2010 release of the Index expanded coverage to 337 US metro areas and provided a more robust website that provided census block group-level views. This release was

8 Review of the H+T Index December 2010 Page 8 accompanied by the CNT s publication of its most recent study and other complementary documents. 7 Housing and Transportation Affordability in Policy The Index s development coincides with a growing recognition that factors other than house prices and rents, such as transportation, energy costs, and access to jobs affect housing affordability. Transportation costs, in particular, have come under scrutiny. At the same time, there is increasing awareness of the impact of expanded transportation demands on households (e.g., multi-hour daily commutes), on communities (most commonly epitomized by efforts against urban sprawl), and the local and global environmental impacts of automobile-based transportation (particularly increased greenhouse gas emissions). Combined with an emphasis during the Obama administration on sustainable and livable community development, this awareness has been translated into policy at the local, state, and federal level in several ways. Location-efficient mortgages introduced in the late 1990s were designed to be a financial vehicle for harnessing this awareness. Urban growth boundaries, inclusionary zoning, transit-oriented development requirements, and transit infrastructure investments are a few of the many of the national, regional, and local policy vehicles designed to make housing and transportation costs more affordable for a wider spectrum of city residents, as well. Parallel to the Index s technical development, then, are a variety of policy efforts that both informed the discourse over the role of transportation access and costs in housing development and set the context for it. With the Index s web launch in 2008, its inclusion in policy discussions and legislative actions have been noticeable. At the local, state, and federal levels of government, many initiatives are adopting the Index in different ways and for different purposes though often without a clear understanding of its data and methodologies. Locally, for example, the Bay Area Metropolitan Transportation Commission set a goal in 2009 to reduce housing and transportation costs for low- and moderate-income households by 10 percent in the region. While not using the Index as it currently exists, a customized Index created by CNT was used to develop baseline data for the goal. Another example of local policymakers use of the Index came in August of this year, when El Paso s City Council voted to use the Index when making affordable housing investment decisions to ensure that the proposed units will truly be affordable. The city will also use the Index in making lending decisions on low-interest loans to first-time home buyers and affordable housing developers in its ongoing affordable housing homeownership programs. 8 The processes for this use are not known at this time, and no evaluations of the legislation s effects on either housing and transportation affordability measures or overall housing and transportation access and quality have been implemented.

9 Review of the H+T Index December 2010 Page 9 The Index has received traction at the state level as well. In June 2010, the State of Illinois signed the Housing + Transportation Affordability Index Act into law. This act requires state agencies to use the Index, along with other measures of affordability, in screening and prioritizing public investments and in making state-funded site development decisions. Specifically, it adopts the Index as the state standard for measuring affordability. It also provides that the State Department of Commerce and Economic Opportunity, Department of Transportation, and Illinois Housing Development Authority use the Index as a development tool for consideration in funding allocation, distribution of incentives, and facility planning in Metropolitan Planning Organization areas. Last, it recommends that the State Housing Task Force consider the Index in its creation of a statewide definition for housing affordability, among other public matters. The federal government s interest in the Index has been equally strong. The US Department of Housing and Urban Development (HUD) and Department of Transportation (DOT) both issued notices of funding availability (NOFAs) in the summer of 2010 that require applications to provide information about the impact proposed projects would have on housing and transportation affordability, in some cases specifically suggesting that applicants use the Index as an indicator in proposals and providing guidance on how to do so. For example, the Sustainable Communities Regional Planning Grant program NOFA issued in June required applicants to report on the share of the population in neighborhoods where combined housing and transportation costs exceed 45 percent of AMI. Similarly, the DOT TIGER II Planning Grants and HUD Community Challenge Planning Grants NOFA included reducing the share of residents in a project area or development with a combined housing and transportation cost burden of more than 45 percent of income as an outcome for which grantees would be responsible. The NOFA referred applicants directly to the Index. Most recently, Representative Blumenauer (D-Oregon) introduced the Transportation and Housing Affordability Transparency Act on July 22, The Act, which has been referred to the House Committee on Financial Services, would require HUD to develop its own transportation affordability index (presumably using the Index as a point of departure) and to incorporate transportation costs into HUD s housing affordability measures and standards. In September 2010, HUD Secretary Donovan also stressed his interest in using the paradigm-shifting Index as a consumer and investment tool as well as one that we should get lenders to use in numerous press interviews, though he stressed that its use was not about regulation. 9 Whether the Index is an appropriate tool for any of these legislative, administrative, or market-based actions is an open question that merits further exploration of its inner workings.

10 Review of the H+T Index December 2010 Page Index Components On one level, the Index is quite simple. Average housing and transportation costs in a neighborhood are divided by a measure of area income to calculate costs as a percentage of income: Index = (Housing costs + transportation costs) / Income This calculation is very similar in concept to other housing affordability measures and measurement tools that exclude transportation costs and focus on housing prices rather than current costs. The Index also defines a new benchmark for affordability. Housing costs of 30 percent have long been considered affordable by housing advocates and policy makers, and this definition is also used in the Index. The Index goes further, though, and defines 45 percent of income as being an affordable combination of housing and transportation costs. 10 CNT notes that average transportation costs nationwide are about 19 percent of income, but that this varies from 15 percent to 25 percent. 11 Three income levels are used to determine where a household of a given income can afford to live: Regional 12 Typical Household: Regional median household income. Regional Moderate Household: 80% of regional median household income. National Typical Household: National median household income, or $41,994. Behind the simple equation describing the Index is a very complex set of econometric models and cost equations. The most critical distinction between the methods for calculating housing and transportation costs in the Index involve the use of censusreported current housing costs (rather than prices, like other affordability indices) for the former, and a heavy reliance on estimating for the latter. Both of these approaches have distinct strengths and weaknesses. Documentation for the current version of the index is available on CNT s website but lacks many key details, including: 13 Econometric specifications of models used to predict auto ownership, vehicle miles traveled (VMT), and transit use; Specific variables used in each of the econometric models; Coefficients of each of the independent variables and their standard errors; An indication of the goodness-of-fit of each of the econometric models, like R 2

11 Review of the H+T Index December 2010 Page 11 The authors spoke to CNT researchers on two occasions in September and October, and they provided additional information in both of these conversations about some aspects of the Index including the specific variables used in each of the econometric models and the data used. CNT refrained from elaborating on important details particularly related to the regression models used as well as decisions regarding use of certain data and data sources because of their concerns about revealing proprietary information. 14 The rest of this section provides available details about the Index assembled from our conversations with CNT researchers as well as documentation of the current version of the Index. The econometric models are described first, followed by the cost equations that use the predictions from the econometric models to calculate housing and transportation costs. Index Variables The Index incorporates the predictions from three econometric models used in predicting transportation costs. As shown in Exhibit 1, the auto ownership and transit use models incorporate the same nine variables: three household characteristics variables and six neighborhood characteristics variables. Auto ownership, auto use, and transit use are all predicted at the census block group level or smaller. Auto ownership and transit use regressions are both fit using data available from the 2000 decennial Census; the regression predicting VMT at the census block group is estimated using odometer readings in Massachusetts from the state s Department of Transportation. The source of data for each independent variable and the definition of each independent variable are shown in Exhibit 2. Most of the variables, such as average household income and average household size, are fairly straightforward and their data source and use are explored later in this study. The employment access index and transit connectivity index (TCI)/transit access index (TAI) warrant further explanation for informational purposes, though they do not necessarily represent a departure from accepted scholarship or practice. The employment access index is a gravity model designed to measure the likelihood of where a neighborhood resident will be employed. Employment proximity (E) for each census block group is calculated as: E = Σ p i /r i 2 It is equal to the ratio of the number of jobs in each census tract in a region (i=1-n) to the square of the distance from the center of that census block group to the center of the ith census tract. The equation essentially weights the quantity of jobs in an area by their distance from a particular neighborhood.

12 Review of the H+T Index December 2010 Page 12 The TCI/TAI are used to measure access to public transit in each census block group. The TCI measures the average number of public transit rides per week available within walking distance of a census block group. To obtain the TCI, the number of bus routes and train station within walking distance (.25 miles for bus routes,.5 miles for transit stations) are scaled by service frequency. Note that the location of bus routes is used rather than the location of bus stops, for which data is much less available. TCI is the preferable measure, and is used when sufficient data are available. When data on service frequency are not available, then the TAI is used, which is simply the number of bus routes and train stations within walking distance of a census block group. 15

13 Review of the H+T Index December 2010 Page 13 Exhibit 1. Dependent Variables Dependent Variable Independent Variables Source Geography Definition Comment Residential density Gross density Average block size Aggregate Number of Vehicles Available by Tenure defines total Auto ownership was estimated to determine Transit Connectivity number of vehicles; impact of urban form Auto ownership Index/ Transit Access Tenure defines universe rather than calculated (Vehicles per Index U.S. Census 2000, SF3, of occupied housing per block group from household) Job access (gravity tables H46 and H7 Block group units census data. measure) Both median and Average time to work average household (by auto or transit) incomes were used at Median household different times. Transit use (daily transit trips) income Average household size Average commuters per household U.S. Census 2000, SF3, table P30 Block group Means of Transportation to Work used to calculate a percent of commuters using public transit Formerly used FTA 2001 data for transit revenue data. Auto use (VMT per household) Vehicles per household Gross density Average block size Average household income Average household size Average workers per household Odometer readings for entire state from Mass. Department of Transportation for meter grid cell level Annual VMT per household Formerly used National Household Travel Survey (NHTS) to estimate total vehicle miles driven per household, but geocoded sample size was limited to 6,840 records.

14 Review of the H+T Index December 2010 Page 14 Exhibit 2. Independent Variables Independent Variable Source Geography Definition Income by tenure; both average and median have Census tract been used Average household income Average household size Residential density Gross density U.S. Census 2000, SF3, table HCT12 U.S. Census 2000, SF1, table P17 Total households: U.S. Census 2000, SF1, table P15; Residential block acres: U.S. Census TIGER/Line for blocks with at least one housing unit Block group Block group Total households: U.S. Census 2000, SF1, table P15 Block group Average block size (acres) U.S. Census TIGER/Line Block group Index measured for each block group; jobs are Employment access index U.S. Census Transportation per census (total jobs/mi2) Planning Package (CTPP) 2000 tract Transit Connectivity Index (TCI)/Transit Access Index (TAI) Vehicles per household Avg. journey to work time Avg. journey to work time: transit commuters Avg. journey to work time: non-transit commuters Average commuters per household Best available data from 400 different transit agencies U.S. Census 2000, SF3, tables H46 and H7 U.S. Census 2000, SF3, tables P33 and P30 Block group Block group Block group U.S. Census 2000, SF3, tables P30, H7, H15, P1 Block group Number of people living in household Households per residential acre Households per total acre = Total households divided by the total block group area in acres Total block group area divided by the number of census blocks within the block group A gravity model: the sum of the ratio of the size of the employment center (in # of jobs) to the square of the distance to that tract Service frequency is used to scale transit access (avg available rides within.25 miles per week) when available (TCI); otherwise use number of bus routes and train stations within walking distance (TAI) Number of cars per household from the decennial Census Calculated using Average Travel Time to Work (in minutes) by Travel Time to Work by Means of Transportation to Work (P33) to define universe of workers 16 and older who did not work at home; who use transit; and who commute by means other than transit. Total workers 16 years and over who do not work at home; calculated from Means of Transportation to Work (P30) and Tenure (H7) to define occupied housing units Because P30 includes workers living in group quarters, the ratio of Total Population in Occupied Housing Units (H15) to Total Population (P1) was used to scale the count of commuters to better represent those living in households

15 Review of the H+T Index December 2010 Page 15 Cost Equations Housing costs and three separate types of transportation costs (auto ownership costs, auto use costs, and transit costs) are calculated for use in the Index. These are summarized in Exhibit 3 and described below. Housing costs in the Index are calculated at the census block group level using a weighted average of costs between owners and renters from the 2000 decennial Census. Unlike many other affordability indices, the housing costs used in the Index do not represent the current value of housing in the area; instead they represent monthly rent or mortgage payments in addition to other expenses, as reported by owners and renters. These costs are likely to be fairly current for renters since rental rates change frequently with the exception of rent-controlled communities. Owner costs, however, may be quite different from the cost facing a prospective homebuyer the current market cost depending on recent property appreciation or depreciation rates in the neighborhood compared with mortgage amounts. Current market costs are used in almost all other housing affordability indexes. CNT researchers said they considered modeling housing costs, which would allow them to estimate market costs as well as account for other dynamic changes (such as the interaction of housing and public transit availability), but decided to use available data. The calculation of transit costs varies depending on the data available. Either the National Transit Database (NTD) or local transit agency data are used to calculate total transit farebox recovery 16 at the metropolitan level. Where neither is available, averages for all metro areas are assigned to the metropolitan area. This total spent on transit is distributed across census block groups using the number of transit trips taken modeled using the regression described above. 17 Auto use costs reflect expenses that vary based on the number of miles the vehicle is driven, including gas, maintenance, and repairs. Separate auto use costs for a variety of types of vehicles is calculated (for example, full-size van, minivan, full-size pick up, intermediate, and compact), and then weighted using the national fleet mix to obtain average auto use costs. Auto ownership costs are fixed costs that are generally unrelated to the level of use of the vehicle. These costs include depreciation, finance charges, insurance, and license, registration and taxes. Costs per type of vehicle are calculated, and a weighted average is calculated based on the national fleet mix.

16 Review of the H+T Index December 2010 Page 16 Exhibit 3. Cost Equations Cost Equation Source Definition Housing costs Geography: Census block group U.S. Census 2000, SF3, Selected Monthly Owner Costs for All Owner-Occupied Housing Units with a Mortgage and Gross Rent for Renters Paying Cash Selected Monthly Owner Costs include payments for debts on the property; real estate taxes; fire, hazard, and flood insurance on the property; utilities; fuels; any monthly condominium or mobile home costs. These data are categorical, so the average was obtained by aggregating SMOC value using the midpoint of each cost bin, and then dividing by owner-occupied housing units. Gross Rent is contract rent plus estimated average monthly cost of utilities and fuels if these are paid by the renter. Transit costs Geography: Urbanized area; where urbanized areas cross MSA/PMSA boundaries, total farebox allocated to MSA/PMSA level based on 2000 Census journey to work data. Auto use costs Geography: Gas prices are at the regional level; fleet mix is at national level; and VMT per household is at census block group level. Auto ownership costs Geography: National National Transit Database (NTD) 2000 and transit agency fare data Transportation Data Energy Book was used to obtain fleet mix using Vehicle Stock and New Sales in the U.S. Gas prices are for 2000 from the Energy Information Administration; maximum prices in 2008 are used in an alternative calculation. Average fuel efficiency assumed to be 20.3 miles per gallon. Costs per mile a multiplied by modeled results of VMT per household to get total auto use costs. FHWA 2001 Complete Car Cost Guide and Complete Small Truck Guide provide figures on cost of owning and operating vehicles; Transportation Data Energy Book was used to obtain fleet mix using Vehicle Stock and New Sales in the U.S. Average Selected Monthly Owner Costs and average Gross Rent (excluding subsidized housing and cleared mortgages) are weighted to calculate housing costs in each block group. Three calculations were used depending on available data: NTD 2000: Total Passenger Fares for Directly Operated Transit Service and Purchased Transportation Fare Revenues summed to identify total farebox. NTD 2000 and transit agency fare data: Total trips from NTD 2000 were multiplied by current transit agency average fare data to estimate total revenue where total revenue data were not available from NTD. Estimated revenue was aggregated to urbanized areas. Average transit revenue: Where NTD did not report total revenue or total trips, average transit revenue and fare data for all metropolitan areas with data available were assigned. Variable costs based primarily on the level of use of the automobile. Include gas, maintenance and repairs. Costs determined to be largely fixed and not determined by use. Includes depreciation; finance charges; insurance; and license, registration and taxes. Costs are estimated per mile and total ownership cost is calculated assuming 14,000 miles per year.

17 Review of the H+T Index December 2010 Page Strengths and Weaknesses A primary strength of the Index is its function in highlighting the importance of considering both transportation and housing costs in assessing the current costs associated with different locations and the costs potential implications for affordability. Although lenders and often landlords calculate whether housing costs represent a reasonable portion of a borrower s or renter s income, transportation costs associated with a housing unit are generally ignored. The Index has a number of weaknesses, however, related to the limited sources of data at the neighborhood level; the lack of documentation for the econometric equations; The use of housing costs for current residents instead of house prices; the use of national-level data to estimate important components of transportation costs; estimation of auto ownership rather than use of available data; and lack of dynamic interaction between housing and transportation costs. In addition, the Index is not truly an index, which makes comparisons between more two places at a time difficult. Its complexity and lack of full documentation also limits informed use of the Index for either planning purposes or research. Data Although a number of websites provide personal transportation cost calculators, 18 and a number of housing affordability indexes exist, we are aware of no other tool that attempts to combine housing and transportation costs. The Index s neighborhood level of geographic analysis is unique among housing affordability indexes due to the general lack of available data. This level of geographic detail is necessary for the Index to accurately assess available public transit options, which can vary dramatically within a metropolitan area or even a town. As shown in Exhibit 4, the smallest level of geography available in other indexes is at the town level; metro-area level geography is more common. The Index s small unit of geography as well as the combination of transportation and housing costs required an extraordinary effort to obtain and process detailed data from the 2000 decennial Census, local transit agencies, and the Massachusetts Department of transportation. Sources for data at the neighborhood level are quite limited, though, which as discussed in detail below, weakens the power of that neighborhood analysis in the Index.

18 Review of the H+T Index December 2010 Page 18 Exhibit 4. Level of geography used in housing affordability indexes Index Index (CNT) Housing Affordability Index (NAR) Housing Opportunity Index (NAHB/Wells Fargo) Amenity-based Housing Affordability Index (MIT) U.S. Market Risk Index (PMI) Level of geography Census block group (neighborhood) Region MSA Towns in Boston metropolitan area MSA Although the creators of the Index have made extensive efforts to collect a comprehensive set of data, there are inevitable limitations in the data that affect the use of the Index. These include the age of the data, incomplete transit service data, lack of data on parking cost and availability, and the use of data on auto use exclusively from Massachusetts. Age of data One weakness of the data is that much of it is from the 2000 decennial Census, which is now 10 years out of date. For some data, such as the number of vehicles per household, changes over the past decade have probably been minor in most areas. The impacts of old data are probably most important in the calculation of housing costs. Although housing prices have changed rapidly throughout the country over the last decade, the housing cost component of the Index would show no variation from one year to the next. The lack of frequent updates to census data is a key reason why all other housing affordability indexes rely on some other source of data for housing costs, such as sales transactions records from the Multiple Listing Service. These sources, however, do not allow housing affordability to be calculated at the neighborhood level. CNT told us that they plan to update the Index during the first half of The U.S. Census long-form data that were used for the 2010 release of the Index are no longer being collected, so the Index will be updated using the latest American Community Survey (ACS) data. These data are designed to replace census long-form data, and represent an improvement in that they are collected more frequently. ACS data at the block group level on a nationwide basis will be available for the first time in December 2010 based on data collected between 2005 and Given the volatility in house prices over the last several years, and because the data from the ACS are an estimate based on data collected over a three-year period, changes in housing costs will be reflected only gradually in ACS data. The housing costs used in the Index, therefore, will certainly be an improvement over the 2000 decennial Census data, but will continue to lag behind actual housing costs. The impact of this lag will

19 Review of the H+T Index December 2010 Page 19 depend largely on house price volatility. In periods of gradually changing housing costs, the impact may be small. When house prices rise or fall rapidly, as they have recently, the impact may be significant. Transit service data Several important aspects of transit service are lacking from some or all metro areas in the Index data. These include transit service frequency (available for some metro areas), data on bus stops (available for some metro areas; in others, routes are used), and public transit fares. The lack of data on transit service frequency in many metro areas of the country has important implications for the accuracy of predictions of auto ownership, auto use, and transit use for commuters. Several groups of researchers have published studies concluding that transit service quality has an important impact on transit use. For example, Woldeamanuel, et al. (2009), conclude that household satisfaction with transit availability is more important than actual proximity to transit in determining auto ownership. Similarly, Dargay and Hanly (2004) find that the frequency of bus service is more important than the distance to the nearest bus stop in determining public transit use. Schimek (1996) found that transit service was a significant contributor to the differences in transit use between Boston and Toronto. In addition, fares for public transit are excluded from the model predicting transit use, although research indicates that this is an important determinant of ridership. Taylor, et al. (2009) find that transit fares and service frequency together account for roughly a quarter of the observed variance in per capita transit ridership across urbanized areas. Winston and Maheshri (2007) also found transit fares to have a negative effect on demand. Last, data on bus routes are used instead of data on bus stops, because these data are far more widely available. However, as noted by Index creators, a bus line without a stop does not equate to transit accessibility. (Haas, et al., 2005) Data on auto use from Massachusetts While auto use generally plays a smaller role in the overall transportation costs than other variables (particularly auto ownership), the data used to generate that number may be problematic. Specifically, the data used to model auto use (vehicle miles traveled, VMT) are based on odometer readings from Massachusetts. Massachusetts drivers are not necessarily the typical of U.S. drivers overall (they drive about 15 percent fewer miles than U.S. households overall [CNT, 2010]), so a more representative sample of drivers may be an improvement in the Index. Drivers in Massachusetts also likely have better access to public transit than those in many other places, which could affect the predicted relationships between auto use and the independent variables used in the model.

20 Review of the H+T Index December 2010 Page 20 Data on parking The cost or availability of parking is not included as a predictor of travel behavior (for transit or auto use, as well as car ownership) in the Index, although research indicates that parking is an important determinant of travel behavior. For example, Cervero (1994) found that the strongest predictor of rail usage is whether residents living near stations have free parking at their place of work. Likewise, Hess (2000) found large impacts of a change in parking cost (from free to $6 per day) on commuters mode choice. An increase in parking costs is predicted to more than double transit use, and to reduce the share of commuters driving to work alone by 16 percentage points. A nationwide source of data on parking costs is probably not available, so it is not surprising that this factor is not included in the Index. Nevertheless, lack of data on parking costs may result in inaccurate estimates of auto use and transit use as well as auto ownership. Econometric equations Our evaluation of the strengths and weaknesses of the Index s econometric models is limited because very little is known about the econometric specification of the current release. Although econometric models used in previous versions of the Index are documented, 19 the current models are not the same as previous versions. 20 In interviews, CNT said that it does not have plans to publish the specifications of the current model, because they consider it to be CNT s intellectual property. CNT would share the details of the Index, however, if it were considered for use as a national requirement. 21 CNT did share some additional information about the econometric models used to create the Index that are not included in the publicly available documentation for the Index: The regression is non-linear. Each econometric model is estimated for all of the 160,000 block groups in the 337 metropolitan areas in one estimation procedure. That is, relationships between independent variables and the dependent variables, such as auto ownership, are not estimated separately for each of the 337 metro areas. The same relationships between the independent and dependent variables are assumed to hold across all metro areas. There is one important exception to the estimation approach: three different models are used depending on the metro-area transit information available. CNT noted that data are from 400 different transit agencies, which are not required to report data uniformly.

21 Review of the H+T Index December 2010 Page 21 Although this additional information is helpful, the lack of outsiders ability to evaluate the econometric equations remains, and is a serious weakness in the Index. Cost equations The Index s cost equations are better documented than its econometric equations. These cost equations calculate average housing and transportation costs for households in each census block group, and are sensitive to the assumptions used, as described below. Housing Costs The use of housing costs for current residents instead of house prices is an important weakness of the Index. As noted above, the housing costs used in the Index do not represent the current value of housing in the area; instead they represent monthly rent or mortgage payments in addition to other expenses, 22 as reported by owners and renters. This is a critical difference from other affordability indexes that are used as a tool for individuals considering a home purchase in a community, rather than a measure of housing costs for the residents who already reside there. As shown in Exhibit 5, other major affordability indexes use a measure of current home values rather than costs to existing occupants. Exhibit 5. Source of housing cost data used in housing affordability indexes Index Index (CNT) Housing Affordability Index (NAR) Housing Opportunity Index (NAHB/Wells Fargo) Amenity-based Housing Affordability Index (MIT) Housing Cost Data 2000 decennial Census, Selected Monthly Owner Costs and Gross Rent Sample of Multiple Listing Service (MLS) sales transactions for existing single-family homes Sales transaction records from CoreLogic, real estate taxes, and property insurance 2000 Census data house values updated using Case-Shiller-Weiss repeat sales indexes and data on price appreciation from Zillow.com Assuming the data are recent, housing costs for current residents are likely to be fairly up to date for renters. Rents typically change annually even for tenants who do not move and often reflect current market rents. Owner costs, however, may be quite different from the cost facing a prospective homebuyer depending on recent property appreciation or depreciation rates in the neighborhood, how recently the neighborhood has turned over, and changes in mortgage rates. As far as we know, no analysis of the magnitude of differences between costs for current owners and prospective buyers has

22 Review of the H+T Index December 2010 Page 22 been conducted, an important piece of further exploration necessary in order to fully understand the Index and whether it is appropriate in policy applications. Transportation Costs Transportation costs are a combination of auto ownership, auto use, and transit use costs. In most neighborhoods, transportation costs are overwhelmingly dominated by auto ownership as depicted by the Index. Two examples best demonstrate this. In a low-density neighborhood in Sherwood, Oregon, total annual transportation costs are estimated to be $12,905. In the neighborhood, households have an average of two cars and travel over 27,000 vehicle miles per year. There are no non-auto transit opportunities. Auto ownership costs make up $10,015 of this cost (78 percent), auto use accounts for $2,890, and transit use is, obviously, $0. On the other coast and the opposite end of the spectrum, in a high-density neighborhood in Cambridge, Massachusetts that is well served by public transit, there are an estimated 1.1 cars per household, and annual VMT per household is estimated to be about 10,670 miles per year. Here, total annual transportation costs are $7,130, of which $5,676 (80 percent) is auto ownership, $1,064 is auto use, and $390 is transit use. Although these neighborhoods are quite different, and have much different transportation costs, auto ownership accounts for roughly 80 percent of transportation costs in both places. Clearly, auto ownership is the most important component of the transportation costs equation, so accuracy of this component is critical. There are weaknesses to this component, however. Auto use costs are calculated using a national-level fleet data mix, which may not be representative of local drivers, and may under- or overestimate auto ownership costs and, in turn, overall transportation costs in a neighborhood. National-level data for a local calculation represent a limitation to the Index when applied to policy in specific locations with likely deviations from national averages. Auto ownership costs and other transportation costs are calculated as an average of the costs for households in each census block group. Although these averages suggest whether transportation costs in a neighborhood are high or low, there are some limitations to the usefulness of this information. Average auto ownership of 1.5 cars per household, for example, is not particularly meaningful information for a family making a location decision. Each individual household must make a judgment about whether they can meet their transportation needs with one car or two based on work locations, transit accessibility, access to local amenities, the number of people in the household and other factors. 23