University of Southern California Dissertations and Theses

Three essays on housing demographics: depressed housing access amid crisis of housing shortage

(USC Thesis Other)

Three essays on housing demographics: depressed housing access amid crisis of housing shortage

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

THREE ESSAYS ON HOUSING DEMOGRAPHICS:
DEPRESSED HOUSING ACCESS AMID CRISIS OF HOUSING SHORTAGE

by

Jung Ho Park

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the Requirements for the Degree
DOCTOR OF PHILOSOPHY
(URBAN PLANNING AND DEVELOPMENT)

December 2019

Dissertation Committee Members:
Dowell Myers, Chair
USC, Sol Price School of Public Policy
Karen K. Kemp
USC, Dornsife Spatial Science Institute
T.J. McCarthy
USC, Sol Price School of Public Policy Copyright ® 2019 by Jung Ho Park
ii

Table of Contents Page

Abstract .......................................................................................................................... vi

Introduction ..................................................................................................................... ix
1. Outline of the Three Essays ................................................................................... ix
2. Housing Demographics Approach ...................................................................... xiii

Chapter I. Housing Shortage, Declining Household Formation, and Hidden
Dislodgements ................................................................................................. 1
1. Background ............................................................................................................. 1
2. Housing Shortage and Unmet Needs ....................................................................... 4
3. Data and Methods .................................................................................................. 11
4. Results ................................................................................................................... 14
5. Discussion ............................................................................................................. 38

Chapter II. Depressed Access to Affordable Housing Due to Higher-Income
Occupancy: Evidence from U.S. Metropolitan Areas, 2006 to 2016 .......... 41
1. Background ........................................................................................................... 41
2. Low-income Housing Access and the Great Recession ........................................ 46
3. Data and Methods .................................................................................................. 64
4. Results ................................................................................................................... 76
5. Discussion ............................................................................................................. 89

Chapter III. Geocoding Inaccuracies: A Case Study for Evaluation of the
Low-Income Housing Tax Credit Program Data ....................................... 91
1. Background ........................................................................................................... 91
2. Use of LIHTC Database and Geocoding Inaccuracy ............................................ 94
3. Data and Methods ................................................................................................ 102
4. Results ................................................................................................................. 118
5. Discussion ........................................................................................................... 127

Conclusion .................................................................................................................... 130

References ..................................................................................................................... 134

Appendix ....................................................................................................................... 155
iii

List of Tables Page

Table I–1. Actual and Expected Number of Households in 2017, United States,
by Tenure and Structure Type ................................................................. 16
Table I–2. Correlations between Dislodgements and Other Regional Factors ......... 36

Table II–1. Income Group and Corresponding Dollar Ranges of Income,
United States, 2016 ................................................................................. 47
Table II–2. Percent of Very Low-income (<50% of AMI) Renters that Occupy
Low-cost Affordable Housing, Ranked 1 to 15 (Least Available)
and 36 to 50 (Most Available) among the 50 Largest Metropolitan
Areas, 2006, 2011, and 2016 .................................................................. 60
Table II–3. Change in Percent of Very Low-income (<50% of AMI) Renters
that Occupy Affordable Housing, Ranked 1 to 15 (Biggest Increase)
and 36 to 50 (Biggest Decrease) among the 50 Largest Metropolitan
Areas, 2006 to 2011 and 2011 to 2016 ................................................... 63
Table II–4. Definition of Variables ........................................................................... 71
Table II–5. Descriptive Statistics for the 200 Largest Metropolitan Areas, 2016 ..... 77
Table II–6. Cross-sectional Regression Results, 200 Largest Metropolitan Areas,
2006, 2011, and 2016 ............................................................................. 81
Table II–7. Fixed-effects Panel Regression Results, 200 Largest Metropolitan
Areas, 2006 to 2016 ................................................................................ 87

Table III–1. Summary of LIHTC Studies Based on the HUD’s National Database,
By Areal Unit into Which Geocoded Points were Allocated ................ 99
Table III–2. List of the Refined LIHTC Projects, By Refinement Type,
Los Angeles County, 2016 .................................................................. 106
Table III–3. Location of Geocoded Address Point (Black Circle) of LIHTC
Project, Los Angeles County, 2016 ..................................................... 111
Table III–4. Three Cases of the Paired Geocoded Point and Land Parcel .............. 116
Table III–5. Point-in-Polygon Operation between Geocoded Address Points
and Land Parcel Polygons, Los Angeles County, 2016 ...................... 119
Table III–6. Geocoded Points/Parcel Centroids and Census Area Polygons,
Los Angeles County, 2016 .................................................................. 124

Appendix A. Data, Method, Study Area, and Study Period of Each Essay ............ 155

Appendix I–A. Summary of Housing Need Estimates, U.S. and California .......... 156
Appendix I–B. Housing Occupancy Rates by Age, United States, 2000, 2006,
2011, and 2017 ............................................................................... 157
Appendix I–C. Annual Estimates of Diversions and Dislodgements, by Tenure
and Structure Type, United States, 2006 to 2017 ........................... 158
iv

Appendix I–D. The 50 Largest Metropolitan Areas in the United States, 2017 ..... 159
Appendix I–E. Additional Tests: Actual and Expected Number of Households
in 2017, United States, by Tenure and Structure Type, Based on
Alternative 1980, 1990, and 2000 Base Years ............................... 160

Appendix II–A. Renter Households and Renter-occupied Housing Units, By
Income and Gross Rent, United States, 1980 to 2016 ................... 162
Appendix II–B. Percent of Low-income Renter Households Who Occupy
Low-cost Affordable Housing, Under Alternative Definitions of
Low-income, United States and 200 Largest Metropolitan Areas,
Ranked by 2016 Population, 2006, 2011, and 2016 ...................... 163
Appendix II–D. Influence of Metropolitan Observations on the Coefficient of
Government Subsidies on the Very Low-income Rental
Availability, Measured by Cook’s Distance, Ordered by
the Most Positive Influence at the Top, 200 Largest
Metropolitan Areas, 2016 .............................................................. 171
Appendix II–E. Influence of Metropolitan Observations on the Coefficient of
New Constructions on the Very Low-income Rental
Availability, Measured by Cook’s Distance, Ordered by
the Most Positive Influence at the Top, 200 Largest
Metropolitan Areas, 2016 .............................................................. 176

Appendix III–A. Variables and Descriptive Statistics of the HUD’s LIHTC
Database, LIHTC Projects Placed in Service Between 1987
and 2016 in the United States ...................................................... 181
Appendix III–B. Descriptive Statistics of LIHTC Projects, Los Angeles County,
2016 ............................................................................................. 183
Appendix III–D. Full List of the Refined LIHTC Projects, By Refinement Type,
Los Angeles County, 2016 .......................................................... 186

v

List of Figures Page

Figure 1. The Three Essays and Research Questions .................................................. x

Figure I–1. Annual Building Permits by Structure Type, United States,
1960 to 2018 ............................................................................................. 7
Figure I–2. Proportional Changes since 2000 in Housing Occupancy by Age,
United States, 2000, 2006, 2011, and 2017 .............................................. 9
Figure I–3. Conventional and Alternative Measures of Housing Needs ................... 12
Figure I–4. The Cascade of Diverted Owners and Dislodged Renters,
United States, 2000 to 2017 ................................................................... 20
Figure I–5. Annual Trajectories of Diverted Owners and Dislodged Renters,
United States, 2000 to 2017 ................................................................... 21
Figure I–6. Percent of All Householders, Renter Householders, and Dislodged
Renters, by Age and Race/Ethnicity, United States, 2017 ..................... 23
Figure I–7. Gains or Losses of Renters Relative to Expected, Within Ages and
Personal Income Quartiles, United States, 2006 to 2017 ....................... 26
Figure I–8. Living Arrangements of People Who Have Forgone Expected
Headship and Spouse Status, Excepting Institutionalized
Population, United States, 2000 to 2017 ................................................ 28
Figure I–9. Comparison between Actual Housing Growth and Expected Growth
of Households, 50 Largest Metropolitan Areas, 2000 to 2017 .............. 34

Figure II–1. Breakdown of Total Occupied Housing Stock, United States,
By Tenure and Income, 2016 ................................................................. 48
Figure II–2. Renter Households in 10%-interval Income-Rent-Group, U.S., 2016 ... 51
Figure II–3. Renter Households in 10%-interval Income-Rent-Group, By Level
of Rent Burden, United States, 2016 ...................................................... 53
Figure II–4. Rental Units and Renter Households in the United States,
Matched by Affordability and Income Categories, 2016 ....................... 55
Figure II–5. Disposition of the Rental Units that are Affordable to Very
Low-income (<50% of AMI) Renters in the United States,
By Income of Actual Tenant, 2016 ........................................................ 56
Figure II–6. Trend in Percent of Very Low-income (<50% of AMI) Renters
that Occupy Low-cost Affordable Housing, United States,
1980 to 2016 ........................................................................................... 58
Figure II–7. Metropolitan Areas with Least or Most Availability of Affordable
Rental Housing, 200 Largest Metropolitan Areas, 2016 ........................ 62

Figure III–1. General Workflow of Using LIHTC Database and Source of
Inaccuracies ......................................................................................... 97
vi

Figure III–2. Geocoded Address Points of LIHTC Project Placed in Service
between 1987 and 2016, Los Angeles County, 2016 ........................ 109
Figure III–3. Westlake Neighborhood in Downtown Los Aneles: Geocoded
Points of LIHTC Project (Circles Filled Black) and Paired
Land Parcels (Polygons Filled Orange) ............................................. 114
Figure III–4. Distribution of Distances between Geocoded Address Points and
Parcel Centroids of LIHTC Development, Los Angeles County,
2016 ................................................................................................... 122
Figure III–5. LIHTC Projects Within Certain Distances of Rail Stations,
Using Geocoded Points and Parcel Centroids ................................... 126

Appendix I–F. Additional Tests: Relationship between Young Headship Rate
of Householders Age 25 to 34 and Incidence of Rent-burden
among Young Adults (25 to 34), Largest 50 Metropolitan
Areas, 2017 ...................................................................................... 161

Appendix II–C. Metropolitan Areas with Least or Most Availability of
Affordable Rentals, Top 200 Metros, 2006, 2011, and 2016 ........ 169

Appendix III–C. Workflow Diagram of Refining HUD’s LIHTC Database and
Pairing Geocoded Address with Concordant Parcel ................... 185

vii

Abstract

My dissertation entitled “Three Essays on Housing Demographics: Depressed Housing
Access Amid Crisis of Housing Shortage” provides a comprehensive explanation about
how housing opportunities for the entire U.S. population, particularly lower-income
renters, have been disrupted during and after the Great Recession. In three independent
essays, I trace the pace of recovery after the Great Recession in terms of declined
household formation, occupancy losses by lower-income renters, and government-
subsidized affordable housing with lack of access to transit.
The first essay, “Housing shortages, declining household formation, and hidden
dislodgements,” examines how we know there is a shortage of housing and how much
housing is really needed for the most vulnerable population. Through a new
housing−demographic method for simulating otherwise expected housing occupancies in
recovery, what has disappeared from the housing stock is “made visible” through its
projected traces in population from metropolitan areas. Nationwide, 8 million renters
have been totally dislodged from running their own households in the housing market
between 2000 and 2017. Lower-income households and young adults, mainly
Millennials, have borne the brunt of housing shortage more acutely than others.
The second essay, “Depressed Access to Affordable Housing Due to Higher-
Income Occupancy: Evidence from U.S. Metropolitan Areas, 2006 to 2016,” investigates
how much of the lower-cost affordable housing that lower-income renters could afford
were “taken” by middle or higher-income households and what metropolitan factors have
limited or expanded this rental availability in some areas more than in others. A
viii

housing−household−classification method reveals that only 47 percent of very low-
income (earning half or less of their area’s median income level) renters successfully
occupied affordable housing in 2016, which is much lower than the rate of 60 percent in
the 1980s and 1990s. Despite the national average, a substantial variation in rental
availability exists across the largest metropolitan areas, with the least available areas
being concentrated in Southern California and Florida. Fixed-effects regression results
highlight the importance of the overall increases in housing supply, either government-
subsidized or market-rate, to ease rental competition in all rental brackets and open
availability of low-cost housing to the poorest renters.
The third essay, “Geocoding Inaccuracies: A Case Study for Evaluation of the
Low-Income Housing Tax Credit Program Data,” examines the geographic location of
government-subsidized affordable housing in Los Angeles County and its transit-
accessibility. A new parcel-level LIHTC database is developed to quantify the positional
accuracy of geocoded federal LIHTC database, which is essential for quality program
evaluation but has been often neglected in previous studies. Despite the generally high
accuracy of the geocoded data, I find that the level of accuracy is lower when examining
large-scale LIHTC projects at a finer geographic area than census tract. A demonstration
analysis on transit-accessibility shows that the inaccurate database overstates the transit-
accessibility of subsidized housing, particularly in the case of large-scale developments.
ix
Introduction

The introductory chapter contains the following sections: the overall outline of the three
independent essays that comprise this dissertation, the research questions for each essay,
and housing demographic approach. The overarching framework for the research is based
on housing demography, a specialized interdisciplinary field that links the dynamics of
population change and housing stocks, which is described in detail in the last part of this
chapter.

1. Outline of the Three Essays

A severe housing shortage is afflicting many metropolitan areas in the United States.
Intense competition for rental housing caused by the shortage crisis is driving up rents
and generating affordability problems, disrupting housing access of the entire population.
This limited access to housing has garnered interest especially from urban planners and
housing policymakers who became concerned that housing shortage would
disproportionately affect disadvantaged populations (HUD, 2017; JCHS, 2018).
Responding to this growing concern, this dissertation features three independent
essays related to the topic of depressed housing access in times of shortage crisis during
and after the Great Recession in the United States. The purpose of this dissertation is to
provide a comprehensive explanation and policy implications about how housing access
of the U.S. population, particularly lower-income renters, has been depressed during the
Great Recession and the stagnant recovery period afterward. Such research goal is
x
pursued from three fundamental dimensions of housing access: i) household formation,
ii) occupancy gains (or losses) by lower-income renters, and iii) housing subsidy
programs. These three dimensions are examined through the following research questions
as summarized in Figure 1, which are explained in further detail in each chapter.

Figure 1. The Three Essays and Research Questions

My dissertation is based on data from several sources, the most fundamental
being the U.S. Census Bureau’s 2000 Decennial Census and American Community
Survey Public Use Microdata Sample (PUMS) 2006-2017 from the Integrated Public Use
Microdata Series (IPUMS) provided by the Minnesota Population Center (Ruggles et al.,
2018). This Census/ACS−oriented data structure allows me to use unique methodologies
to trace annual trends in lower-income housing access in various geographic areas and
multiple time points before and after the Great Recession (see Appendix A for summary
of data, methods, study areas, and study period of the three essays).
xi
The first essay provides an explanation behind the shortage of housing and how
much housing is really needed in the largest metropolitan areas and the nation. A
housing-demographic method
1
is utilized for measuring the extent of shortage, which has
been significantly underestimated by conventional approaches that focus solely on
existing households that have survived the competition for an unduly restricted housing
supply. In contrast to the conventional approach, the housing-demographic method
focuses on the expected number of housing units to be occupied by the current
population, which when compared to the actual occupancies, provides a way of
estimating the number of dislodged households: households that fail to survive the
competition for a scarce supply and thus have been “made invisible.” This essay attempts
to identify the dislodged households as those with lowest income levels and those whose
household formation is the most flexible due to other accessible living arrangements,
such as staying with parents or doubling up with roommates.
The second essay investigates how lower-income housing opportunities have been
limited due to middle or higher-income occupancy of low-cost affordable housing during
and after the Great Recession and how this relates to new supply. A housing-household-
classification method
2
is used to identify whether a low-cost housing unit was occupied

1
The particular method for simulating otherwise expected housing occupancies was developed in Myers,
Painter, Lee, and Park (2016).
2
This method for categorizing households and units their income and rent respectively is commonly used
by scholars and practitioners who study rental affordability in a variety of geographies, including
neighborhood, municipality, county, metro area, state, and the nation. For the purpose of my dissertation, I
term it housing-household-classification method since there is not a specific name of the method.
xii
by a lower-income tenant or taken by a higher-income competitor. Aggregated across all
individual households and housing units in a metropolitan area, this measure of rental
availability provides a single-number average of all lower-income renters’ experience,
which appears to widely vary across places and to have sharply changed during and after
the Great Recession. Unlike previous studies that often focus on how much lower-income
households pay for rent (i.e., rent-burden), this essay emphasizes lower-income
occupancy itself, a more fundamental level of housing opportunity than rent-burden for
the poorest households, particularly in times of housing shortage. Regression models
have been used to estimate how much lower-income housing access can be expanded by
market-rate construction and how much will require housing assistance programs.
The third essay examines the geographic location of government-subsidized
affordable housing in Los Angeles County and its transit-accessibility. I attempt to
address two fundamental and important questions that have been often neglected in
existing literature: i) how accurate is geocoded
3
data in affordable housing research, and
ii) to what extent geocoding inaccuracy can affect program evaluation results? To
quantify geocoding accuracy in the Low-Income Housing Tax Credit (LIHTC) data, I
combine a common federal database with local county assessor’s land parcel map as an
ancillary data to measure accuracy.
4
With clearly quantified accuracy of geocoded
affordable housing, a demonstration analysis on transit-accessibility is conducted to show
the impact of geocoding inaccuracy on program evaluation results.

3
Geocoding is a common technique that associates an address in a tabulation with a point on the map.
4
I develop this method to quantify the geocoding accuracy of a federal database. My method can be easily
replicated to any municipality or county that has its own assessor’s parcel data.
xiii
2. Housing Demographics Approach

The overarching framework of the entire dissertation is deeply rooted in the field of
housing demography, a specialized interdisciplinary field that links demographic
structure and housing markets (Myers, 1990). The concept of housing demography is
broadly applicable, including not only the housing behavior of populations but also the
formation and composition of housing stocks. My dissertation follows three underlying
precepts that are common to research in the domain of housing demography.
First, housing-demographic behavior is conceptualized within a temporal
framework emphasizing longitudinal processes. The first essay of my dissertation
emphasizes the changes in housing occupancy behaviors over time, capturing differences
between a baseline year and a target year. The longitudinal perspective in this research
shows that household formation and tenure choice turn out very differently according to
demographic characteristics under different market conditions.
Second, the housing demography concept places emphasis on interconnections.
The housing stocks and resident populations are linked together at the individual
household and aggregate level. The interconnected frame adopted for the second essay of
my dissertation is vital to show how housing shortages lead to disruptions throughout the
entire population, particularly connecting rental hardships across all income groups.
Accordingly, the second essay provides an explanation behind specific interconnections
in changing housing stocks and populations in times of shortage rather than seeing these
as a set of isolated problems and separate interest groups.
xiv
A third concept, spatial patterns, is important for research of housing and
population behavior in small areas especially in regard to the immobility of housing units.
Houses are differentiated by type and fixed in space, and they attract particular types of
households matched to their attributes, which is especially true for income-restricted,
government-subsidized housing. By focusing the geographic scope to Los Angeles
County, the third essay of this dissertation visualizes how the housing stock, particularly
government-subsidized housing units, in Los Angeles accommodates the changing
population in times of shortage.

1 – Chapter I
Chapter I. Housing Shortage, Declining Household Formation, and Hidden
Dislodgements

1. Background

An acute housing shortage is afflicting many metropolitan areas, particularly those in
coastal regions and where the population is growing and supply is slow to expand
(Glaeser & Gyourko, 2018; JCHS, 2018). The large Millennial generation has arrived at
the normal age for household formation, typically seeking rental units (Myers, 2016). In
addition, the financial crisis leading into the Great Recession spurred a downturn in
homeownership rates that was accompanied by a massive shift of “diverted homeowners”
into rental housing (Gabriel & Painter, 2018; Myers et al., 2016). Thus, the demand for
rental housing is growing both for demographic reasons and because of the unexpected
number of former and would-be homeowners that have been thrown into rental
competition.
The magnitude of the current housing crisis is greatly underestimated because
conventional methods have proven inadequate for measuring housing needs in times of
shortage of housing. A fundamental limitation of conventional methods is that the
estimated needs cannot exceed the number of housing units available for occupancy. A
shortage of housing caps the number of households that can be surveyed and excludes
those that are dislodged by the shortage. The housing shortages also have caused a rising
cost of housing that then outruns households’ ability to pay and then prompt households
that are already stretched to give up their household headship (Brookings Institution,
2018; Gabriel & Rosenthal, 2015; Myers et al., 2016).
2 – Chapter I
The shortcoming of conventional methods for estimating housing needs is that
they all rely on the observed conditions of current households, which are located and
identified in occupied housing units (HUD, 2017; JCHS, 2018; NLIHC, 2019). However,
the conventional method only captures the survivors at the depressed, end result of the
housing shortage. Not accounted for is the gap between expected and actual numbers of
occupied housing units. Under conditions of shortage, there are many potential renters
whose needs cannot be observed. When I speak of the percentage of renters who suffer a
payment burden, not included are the potentially large number of households totally
dislodged from the housing market in the competition for a limited supply. In fact, those
dislodged households may be least able to pay and may have the most severe needs of all.
Thus, all the existing estimates of need are distorted to exclude dislodged households and
could be severe underestimates. This essay proposes a more inclusive method
5
and
demonstrates it for the 50 most populous metropolitan areas and the nation as a whole.
One particular finding suggests the importance of insights from this essay.
Analysis for the U.S. shows that between 2000 and 2017 the number of renter-occupied
housing units increased by 7.6 million. This falls behind what would have been expected
(7.8 million) by only 0.2 million, so it sounds like rentals are doing well. However, that
small change masks a very large transfer of previously expected homeowners diverted
into renting (7.8 million) and, because the total rental stock failed to increase by 15.6
million, a very large dislodgement must have occurred of expected renters out of
household status (8.0 million).

5
This method was developed in an earlier study for Research Institute for Housing America (RIHA) of the
Mortgage Bankers Association (MBA), together with Dowell Myers, Gary Painter, and Hyojung Lee.
3 – Chapter I
Even though the total number of renters has held fairly steady, growing slightly,
beneath the surface is a churn and intense competition caused by the rental shortage that
is driving up rents and forcing out others. Some quantity (8.0 million) of expected
households have been dislodged – either forced to double up, remain living with parents
or roommates, or otherwise dissolve. Thus, the true magnitude of the rental crisis has
been underestimated by conventional approaches because of the large scale of
dislodgements. Similar dislodgement patterns, in general, are found across the set of large
metropolitan areas, although, the magnitude of rental dislodgements was substantially
greater in some areas than others.
This essay has five objectives. First, I begin with a broad review of existing
estimations of housing needs. Second, I present a housing−demographic method which
was developed in Myers et al. (2016) for calculating how extensive the rental
dislodgement crisis is and the resulting housing needs, applying this to the U.S. as a
whole. Third, I assess the demographic profile and economic status of dislodged renters
to identify who were most vulnerable in times of shortage crisis. Fourth, I apply the
method to the 50 largest metropolitan areas to estimate the prevalence of housing
shortages in the metros by comparing actual growth in housing to my estimates of
expected housing needs. Lastly, I estimate the degree of rental dislodgement that has
occurred in each of the metros and examine how that correlates with other key
metropolitan factors.

4 – Chapter I
2. Housing Shortage and Unmet Needs

Conventional Estimates Based on Existing Households

In the United States, estimates of housing needs are generally prepared for geographic
areas administered by municipal or county governments, because it is local governments
that regulate land use and new construction (Baer, 1986; Myers et al., 2002). Although
housing construction occurs in localities, housing needs often have regional, state, or
national importance. State and regional agencies, or even nonprofit advocacy groups, also
depend on estimates of housing needs for particular subpopulation groups (e.g., lower-
income) as a guide for setting priorities for the use of limited public resources (NLIHC,
2019; SCAG, 2019).
The concept of housing needs, roughly characterized as the number and type of
housing units required to accommodate a population at a given standard of housing
occupancy, can be defined and estimated in many different ways (Myers et al., 2002;
Noll et al., 1997; SCAG, 2019). In the view of economists, there is no such thing as need,
only demand. Yet demand can be suppressed by inadequate supply or by insufficient
funds to express desired demand. “Need” often is normative: the amount and type of
housing that is expected based on established social standards (Struyk, 1987; Varady,
1996). But how do researchers define that in practical terms so that it can be measured?
Individual academic studies and professional reports on housing needs often
declare their own implicit definition and approach without recognition of alternatives.
They mostly focus on either of two (and sometimes both) essential dimensions of housing
needs. The one emphasizes the quality of the physical housing stock, as often indicated
5 – Chapter I
by such factors as age, plumbing conditions, and code violation. The other denotes the fit
between households and housing, most often indicated by the ratio of housing cost
payments to income (a measure of housing affordability) or the ratio of the number of
persons to the number of rooms in the unit (a measure of overcrowding) (Myers & Lee,
1996; Routhier, 2019). Measures of household fit, rather than physical quality, are much
more frequently emphasized in the United States. This is mainly because the large
improvement in housing quality since 1940 has reduced the severity of physical issues in
the housing stock (Clemmer & Simonson, 1983).
On the basis of varied definitions of housing needs, as reviewed above, I identify
a conventional approach widely used to measure unmet housing needs that are focused on
unaffordability (i.e., excessive rent burden) of existing households, rather than on
shortages that dislodge households completely. In this essay, the limitations of the
conventional approach are remedied by a new housing−demographic method.
Conventional estimates of housing needs have mainly focused on the current
deficit by which the existing housing conditions fall short of a normative standard of
housing affordability. The primary standard is the percent of renters paying more than 30
percent of gross household income for housing expenses. The most urgent cases of
housing need pertain to those with the lowest income, typically defined as falling below
30 or 50 percent of the median income in the surrounding area. Housing affordability
problems, however, extend to middle-income households as well.
National studies following the conventional approach largely find that some 6.4 to
8.2 million existing rentals fall short of a normative standard in the nation (see Appendix
6 – Chapter I
I–A for summary of existing housing needs estimates).
6
A study by the federal
government, HUD (2017), reports that 8.2 million very low-income renters (earning at or
below 50 percent of AMI) had “worst case” housing needs in 2015. Another national
study by National Low Income Housing Coalition (NLIHC, 2019) reports that the
shortfall between the demand for housing among very low–income households and the
available supply of market-rate units that these households could afford was 7.4 million
in 2017. An Urban Institute study (2017) also reports that the gap between the extremely
low–income renters (earning at or below 30 percent of AMI) and the number of adequate,
affordable, and available rental units was 6.4 million in 2014.
The serious omission of the conventional approach is that it is based on actual
households available to survey today. Those not present cannot be shown to have an
affordability problem. In fact, the presence of the extreme housing shortage of Los
Angeles and other large metropolitan areas implies that a large portion of housing needs
cannot even be recognized. This submerged need has been made invisible by the lack of
housing.

6
Two major enterprises sponsored by the United States government, Fannie Mae and Freddie Mac, also
report their own estimates of national housing needs. For example, Freddie Mac (2018) estimated that
about 2.5 million (based on baseline scenario) housing units are short of what is needed to match long-term
demand in the nation. Their estimates were not reviewed in this essay because it combines owner and rental
needs, which largely differs from other common estimates that mainly focus on rental housing needs.
7 – Chapter I
Great Recession Disruption and Unprecedentedly Low Construction

Since the beginning of the Great Recession, the United States has produced record low
amounts of new housing (see Figure I–1). The depth of the downturn in 2009 through
2011 was much lower than any years since 1960 or earlier, and the height of the upswing
in 2018 was still below the six-decade average (1.4 million) and slightly above the
troughs of previous construction cycles. Meanwhile, the giant Millennial generation
continues to push into a housing market offering scant opportunity for newcomers.

Figure I –1. Annual Building Permits by Structure Type, United States, 1960 to 2018

Sources: U.S. Census Bureau, 1960−2018, Annual Building Permits Survey.

8 – Chapter I
In this context, the shortage of housing has begun to dominate housing needs,
even as the shortage spurs rent increases that increase affordability problems. A serious
limitation of conventional methods is that the estimated housing needs cannot exceed the
number of housing units available for occupancy. A shortage of housing caps the number
of households that can be surveyed.

Depressed Household Formation and Diverted Housing Needs

The shortage effects can be disproportionately absorbed by certain population groups. In
practice, the greatest shortfalls are absorbed by the youngest age groups in the housing
market as plainly shown in Figure I–2. Previous studies also report that the recent
recession lowered the household formation of young adults, and in turn, depressed
housing demand (Dunne, 2012; Lee & Painter, 2012; Myers et al., 2016; Mykyta &
Macartney, 2011). This is logical because the youngest is the newest and the last in line
to find housing. The youngest are also most mobile, moving between units much more
commonly, while older households tend to hold tight to their previous homes.
7

7
Another possible explanation about the lowered household formation of young adults in the Great
Recession is changes in lifestyles that reduced independent living of young adults, particularly among
Millennials. It explains that Millennials delay marriage and parenthood, in turn independent living, because
they have different preferences about living arrangements than preceding generations. How much these
preferences have reduced household formation of young adults is unclear given the very poor job market
and the high cost of rental housing in the recent recession. However, findings of this essay need to be
carefully interpreted with regards to the possible explanation about preferences and changes in lifestyle
among young adults.
9 – Chapter I
Figure I –2. Proportional Changes since 2000 in Housing Occupancy by Age, United
States, 2000, 2006, 2011, and 2017

(a) Total Household Formation (HHs per capita, by Age)

(b) Formation of Renter Households (Renter HHs per capita, by Age)

(c) Formation of Owner Households (Owner HHs per capita, by Age)

Notes: The vertical axis is scaled as proportional to the base year rate (2000 = 1.0). See
Appendix I–B for the exact rates of household formation by age group.
Sources: Census 2000 5-percent IPUMS; 2006, 2011, and 2017 ACS 1-year IPUMS.
10 – Chapter I
Consider the contraction of housing occupancy revealed for the nation as a whole
in Figure I–2, comparing occupancy rates in 2006 as a proportion of those in the 2000
baseline, and further comparing those in 2011 and 2017.
8
The contraction of the youngest
is plainly evident across the formation of total households, owner households, and renter
households. When occupancy probabilities contract, those would-be households either
disappear into non–household status, or, if they are would-be homeowners, they most
likely shift into rental housing. It is this shift of expanded rental occupancy probabilities
of age 30 and over that is seen in the panel (b) of Figure I–2.
The lesson drawn from Figure I–2 is that formation of expected household
growth, in total, was concentrated among people under age 30. However, even when
household formation persisted at nearly the 2000 rate for people above age 30, there was
still a large shift of those ages toward a deficit of owner occupancies and a large
expansion of renter-occupied housing. It is a matter of reasonable expectation that those
expanding renter formations above age 30 were likely occupying space that otherwise
would have been taken by people in their 20s or by people of color who have lower
expenditures on average. The older households either held on to their previously acquired
units or outbid their younger and less-advantaged competitors for newly vacant units.
Any estimate of housing needs in the current period must be designed to take
account of the shortage effects. The conventional approach, however, omits broader
concerns of an adequate number of housing units available in the stock. The conventional
method, by focusing only on the affordability of existing households that have survived
the competition of shortages, underestimates housing need. Most likely, it is the

8
Appendix I–B of this essay reports the exact rates of household formation by age group.
11 – Chapter I
households with the greatest affordability problems that have been forced to dissolve.
Their lack of accounting in the conventional method should not imply that their needs do
not exist.

3. Data and Methods

My approach is based on the population residing in an area, not the existing occupied
housing units. It draws inspiration from methods for projecting future construction needs
to accommodate future population increases by applying headship rates. Instead, I use
this housing−demographic approach to infer present housing needs based on recent
population growth since 2000.
One way to illustrate the conventional underestimate of housing needs is to look
beyond the households observed in a survey or Census data, as diagrammed in Figure I–
3. An alternative to the conventional measure is based on the resident population, not just
the households, to estimate housing needs by applying baseline per capita housing
occupancy rates to the number of people in all the segmented groups in a population.

12 – Chapter I
Figure I –3. Conventional and Alternative Measures of Housing Needs

(a) Housing Needs Measured by Conventional Method

Met
per HH Housing Needs

(e.g., low–income HHer
living in an affordable
housing unit)

Unmet
per HH Housing Needs

(e.g., low–income HHer
living in an unaffordable
housing unit)
Data not Considered for
People Who Have Been
Dislodged from the
Housing Market

(b) Broader Housing Needs Measured by the Housing−Demographic Method

Met
per HH Housing Needs

Unmet
per HH Housing Needs

Unmet (Dislodged)
per capita Housing Needs
Based on Changes from
Baseline Expectations

(e.g., people who gave up
HHer status and are now
doubling up or homeless)
Notes: Not to scale. The shaded area was accounted for in a given method while the
unshaded area was not considered. POP = population. HH = household. HHer =
householder.

This housing−demographic approach follows methods for projecting future
construction needs but applies those to the current period. My focus is on current
shortages of housing, but I calculate these shortages over a time interval, as I would for a
projection, but my time interval is the period from 2000 to 2017. In brief, I project future
housing needs for 2017 based on the actual population surveyed in 2017 and a set of
expected housing occupancy rates that were observed at the baseline (2000). Thus, I
HH Data
POP Data
POP Data
HH Data
13 – Chapter I
estimate how many housing units would be normally expected under the baseline
standards (See Appendix I–E for results based on alternative base year). Given the
implicit assumption that the occupancy rates of 2000 were desirable to maintain, I
determine that the gap between the expected and actual housing occupancies of 2017
represents unmet housing needs.
The chosen baseline date of 2000 is commonly regarded as the last “normal” date
for the housing market.
9
It precedes the housing bubble that commenced in 2002,
following manipulations of the federal interest rate to abnormally low levels in 2001. The
full period of 2000 to 2017 includes years of housing boom, the Great Recession, slow
recovery years to 2011, and the revival of 2012 through 2017.
This method can estimate expected housing needs for a given year, 2017 in this
essay, by comparing actual housing occupied in 2017 to what was estimated by applying
the earlier, standard rates (my baseline from 2000) to the actual population by age and
racial/ethnic group in 2017. When the expected number of households is greater than the
actual number of households occupied by the same population, this shortfall is termed
dislodged housing needs. For this, I also embed the number of diverted homeowners into
the expected number of renters, so that I can estimate the number of dislodged renters.

9
The housing−demographic method can be especially sensitive to the choice of baseline year from which
changes are calculated. Sensitivity tests reported in Appendix E indicates very similar trends if 1990, a
preceding decennial census year, was selected for the baseline instead of 2000. Analysis based on 1980
housing occupancy rates, however, reports a much greater expected number of households in 2017 than
those based on 1990 and 2000, which is likely due to a greater occupancy in multifamily housing back in
1980. Based on the sensitivity test result, I select 2000 as my baseline.
14 – Chapter I
This essay calculates headship rates for each 10-year age group, including
separately for Latinos, and non–Hispanic Whites, Blacks, Asian or Pacific Islanders, and
other races. I then estimate how the expected households formed by each demographic
group will follow 2000 rates of occupancy to form owner-occupied and renter-occupied
households. Using the 2000 occupancy rates, I further subdivide each tenure into
structure type of residence, 1-unit structures, units in multifamily structures, or others.
My calculation yields household numbers that would have been expected based
on 2000 headship, ownership, and structure-type rates, which are a useful benchmark for
comparison with the actual rates from 2017. Given a set population, I thus can estimate
how many occupied housing units could be expected and compare each segment to its
actual number.

4. Results

As discussed above, a weakness of the nation’s housing accounting system is that I can
identify low-income households only if they occupy housing units. As the housing
shortage grows I may be accruing a growing undercount of low-income households
without housing. For lack of accounting, it is difficult or impossible to determine how
many low-income households, or how many payment burdened households, were
dislodged. However, it is possible to make some reasoned judgments.

15 – Chapter I
Expected Households and Unmet Housing Needs

I first show the expected number of households based on changing demographics in the
nation as a whole. I use per capita rates of householders who own and rent by age,
race/ethnicity, and structure type in the baseline year 2000 to estimate housing tenure
status of the population in 2017 and compare the estimates to the actual number of
households in 2017. The gap between actual and expected numbers represents unmet
housing needs.
The differences in the number of households (which the Census defines as
occupied housing units) in each category are displayed in Table I–1. The columns A and
B are the actual number of households by tenure and structure type, which only captures
the existing (or surviving) households. Column C displays the expected number of
households based on 2000 occupancy patterns and actual population of 2017, separately
by tenure and structure type. The last columns (D and E) are the differences between
actual and expected households, representing unmet housing needs. Note that the
absolute count of unmet housing needs is derived by the actual households less the
expected households (column D = column B – column C), collectively and separately by
tenure and structure type. Also reported is the unmet housing needs as a percentage of
actual housing needs in 2017 (column E = column D / column C × 100).

16 – Chapter I
Table I –1. Actual and Expected Number of Households in 2017, United States, by
Tenure and Structure Type (Unit: Thousands, %)

Actual Number of
Households
(C)
Expected
Number of
Households
in 2017
Difference between
Actual and Expected
(A) 2000 (B) 2017
(D)
in Absolute
Number of
Households
(E)
as a % of
Actual
Number
Total 105,480 120,063 128,040 -7,977 -6.6
Owner occupied 69,819 76,779 84,545 -7,766 -10.1
Single-family 60,073 67,895 72,565 -4,670 -6.9
Multifamily 3,804 4,102 4,966 -864 -21.1
Other 5,941 4,782 7,014 -2,232 -46.7
Renter occupied 35,662 43,284 43,495 -211 0.5
Single-family 10,612 14,924 12,508 2,416 16.2
Multifamily 23,479 26,433 29,200 -2,767 -10.5
Other 1,570 1,928 1,787 141 7.3

Notes: See Appendix I–E for results based on alternative base year. The expected number
of households in 2017 is calculated by multiplying the number of population in 2017 by
age and race/ethnicity by age−race/ethnicity−structure type-specific headship rates in
2000. ‘Other’ category includes mobile homes, boat, RV, etc.
Sources: Census 2000 5-percent IPUMS and 2017 ACS 1-year IPUMS.

The difference between the actual and expected number of households (column
D) is substantial, indicating numerous unmet housing needs in the nation. At 2000
headship rates, the country would have expected to have an additional 8.0 million
households in 2017 (column D). Unmet housing needs equal 6.6 percent of actual
households in 2017 (column E).
At 2000 (per capita) homeownership rates, the country would have been expected
to have an additional 7.8 million owners (column D). This consists of an additional 4.7
million owners in single-family structures and an additional 3.1 million owners in
multifamily and other structures in 2017.
17 – Chapter I
This essay finds 0.2 million fewer renters in 2017 than would have been expected
using 2000 occupancy rates (column D). As would be expected, the number of renters
increased during the recession. It is remarkable that there was such a large shift in the
composition of these renters. In 2017, there were 2.4 million more single-family renters
than would have been expected at 2000 rates (column D).
At the same time, there were 2.8 million fewer multifamily renters in 2017 than
would have been expected (column D). An increase of 5.7 million multifamily renters
had been expected (column C – column A), but only 3.0 million growth was realized
(column B – column A). This reflects the relative undersupply of new apartment
construction between 2000 and 2017, as shown previously in Figure I–1.
Nonetheless, even the above discussion underestimates the rental impacts,
because the 7.8 million would-be homeowners (column D) had to be accommodated in
the rental sector (unless dissolved). Only 2.4 million were accommodated by the new
increase in single-family renting (column D), thus leaving 5.4 million previously
unexpected households to rent multifamily housing or be dropped out of the housing
market. The multifamily rental sector did not expand fast enough to absorb all these
newly generated renters.

Cascade Effects

When households are expected in excess of the available occupied units, not all groups
have equal access, and the effects of the shortages become concentrated in less
advantaged sectors. Ultimately, dislodgements occur most likely at the bottom of the
housing market after households in the upper tiers have taken first choice. There are two
18 – Chapter I
different means by which the potential households are dislodged — first by diversion of
would-be homeowners into the rental market and then by dislodgement of renters entirely
out of the housing market. The greater the volume of diverted homeowners and the
greater the undersupply of rental housing, the greater the ultimate dislodgement of renters
and creation of unmet housing needs. On the surface, the changes among renters seem
much more modest than the compounding forces that become concentrated.
This essay describes this as a cascade of demand.
10
Although the rental supply
increased modestly, by 7.6 million occupied units, 0.2 million less than was
demographically expected, this proved inadequate. The downturn in the owner-occupied
sector diverted 7.8 million would-be homeowners into the rental market. This was added
on top of the 7.8 million growth in renters that was previously expected. The total growth
in renter households needed to fit into a rental supply that only grew by 7.6 million
occupied units, which included 4.3 million owner-occupied single-family units that had
shifted to renter occupancy. Clearly, the total rental occupied additions did not provide
room for all of the originally expected rental growth, plus the unexpected diverted
homeowners.

10
The key assumption in this estimation, namely that 100 percent of diverted owners enter the rental
market, may be subject to confirmation. There has not been a study that directly estimated the count of
diverted owners who end up being renters. Though a few studies reviewed the trend in living arrangements
of foreclosed owners, they have been limited to qualitative descriptions due to lack of national survey on
living arrangements of foreclosed owners (HUD (2012)’s Foreclosure Counseling Outcome Study; Urban
Institute (2009)’s The Impacts of Foreclosures on Families and Communities). Also noteworthy is that the
foreclosed owners account for only a portion of diverted owners that broadly include all types of potential
owners who would have been owners in 2017 based on baseline housing occupancy rates.
19 – Chapter I
At root, housing demand is growing from young adults. The coming of age of the
large Millennial generation is driving demand for more rental housing. Added to that is
the impact of the would-be homeowners (roughly 4.0 million victims of foreclosure and
also newly aspired homeowners) who have been diverted into rentals and held there for
lack of other opportunity. Meanwhile, existing homeowners and renters seek to hold on
to their current homes, so that more housing is required in total to accommodate the
growth. Without that growth, someone has to be dislodged.
The result I estimate is that 8.0 million renters have been dislodged from running
their own households in the housing market between 2000 and 2017. While these unmet
needs amount to 6.6 percent of total households existing in 2017, the unmet needs are a
far greater share of renters, 18.4 percent, due to the compounding effects of the cascade.
Using the national estimates specified in Table I–1, I can conceptualize the
cascade of changes that passed through the market sectors as diagramed in Figure I–4.
The diversion of 7.8 million would-be homeowners was split first with 2.4 million
landing in rented single-family units and the rest landing in other rentals (or dissolved).
Meanwhile, multifamily rentals already had been expected to grow by 5.7 million from
2000 to 2017, but only realized 3.0 million of that growth due to depressed housing
construction. The remaining diverted homeowners and the anticipated growth of
Millennial renters thus collided in a multifamily rental market that was severely
undersupplied. In the competition for scarce units I cannot be sure who occupied the
units, but those with the lowest incomes and highest payment burdens, and those who
were non-Hispanic White, as well as those who were youngest and newest in the market,
the Millennials, had distinctly lower chances (discussed in detail in the next section).
20 – Chapter I
Figure I –4. The Cascade of Diverted Owners and Dislodged Renters, United States,
2000 to 2017

Sources: Census 2000 5-percent IPUMS and 2017 ACS 1-year IPUMS.

A visual summary of annual trends in actual and diverted homeowners is
presented in Figure I–5 (left panel) as are the trajectories of actual and dislodged renters
(right panel).
11
It is noteworthy that the number of actual homeowners in the nation has
remained virtually constant since 2010 until the latest increase in 2017. Yet at the bottom
of the left panel, I see evidence of a 7.8 million loss of diverted homeowners (as
presented in column D in Table I–1). This stems from some decline in older age groups

11
Appendix I–C reports annual estimates of the diverted homeowners and dislodged renters. It shows that
the diversion of owners peaked in 2016 while total scale of dislodgements continues to grow since 2006.
Changes 2000 to 2017, Actual Population,
but Assuming 2000 Patterns of Housing Occupancy
Would-be homeowners
were diverted into rentals 7.8 million
Joining the expected growth of
renters from Millennials & others 7.8 million
Creating total POTENTIAL
growth in renters 15.6 million
Less the
ACTUAL increase in
renter-occupied units 7.6 million
LEAVES dislodged renters 8.0 million
Submerged demand equals 18.4% more than 2017 actual renters.
and equals 6.6% more than total 2017 households
21 – Chapter I
but more so from the failure of young adults in the Millennial and Gen X generations to
buy homes. In other words, most of the diversion after 2010 is not from foreclosures but
from foregone purchases based on what would have been expected based on the growing
size of the population and rates of housing behavior in 2000.

Figure I –5. Annual Trajectories of Diverted Owners and Dislodged Renters, United
States, 2000 to 2017

(a) Owners

(b) Renters

Notes: Diverted owners equal expected owners less actual owners. Dislodged renters
equal expected renters less actual renters plus diverted owners. See Appendix I–C for the
exact annual estimates of diverted homeowners and dislodged renters.
Sources: Census 2000 5-percent IPUMS; 2006 to 2017 ACS 1-year IPUMS.

The increase in actual renters in the right panel (line with filled circles) is not
sufficient to accommodate both the diverted homeowners (7.8 million in column D of
Table I–1) and the normally expected growth in renters (7.8 million as column B less
22 – Chapter I
column A of Table I–1) due to the rising number of young adults. Roughly 8.0 million
renters likely have been dislodged from household headship and are either doubled up or
forced completely out of the housing stock, all while still resident in the nation.

Demographic Profile and Economic Status of Dislodged Renters

Given the large scale of dislodgements (8.0 million), from this point forth I examine the
characteristics of dislodged renters on demographic and economic dimensions, firs
addressing age, then race and Hispanic origin. After that, I examine the income level of
dislodged renters by their age. Overall, those with the lowest incomes and presumably
highest payment burdens, and those who were non-Hispanic White, as well as those who
were youngest and newest in the market, the Millennials, were most likely to be
dislodged in the competition for scarce units.

Age Composition of Dislodged Renters

The age breakdown of the dislodged renters is displayed in Figure I–6 left panel, which
also shows separately the age profile of all householders and renter householders as
reference. Focusing on all householders, I find that young adults under age 35 made up
only 18.6 percent of households. Young adult share is greater among renter householders,
accounting somewhat more than one third (34.3 percent) of all renters.

23 – Chapter I
Figure I –6. Percent of All Householders, Renter Householders, and Dislodged Renters,
by Age and Race/Ethnicity, United States, 2017

(a) Age (b) Race/Ethnicity

Notes: Universe is all householders (120 million), renter householders (43 million), and
dislodged renters (8 million) in 2017 in the nation. HHers is householders. NH is non-
Hispanic and PI is Pacific Islanders.
Sources: Census 2000 5-percent IPUMS and 2017 ACS 1-year IPUMS.

When examining the dislodged renters, young adults are most likely to be
dislodged, comprising nearly a half (46.5 percent) of total dislodgements. As already
shown in Figure I–6, the youngest are the newest and the last in line to find housing. The
youngest are also most flexible with regards to pursuing independent living and moving
between housing units much more frequently. Older households, by contrast, tend to stay
24 – Chapter I
in their previous homes. Conversely, the early middle-age group (35 to 44) was least
likely to be dislodged, implying that they formed their own independent living
arrangements despite a shortage of housing and likely high rent burden.

Racial Makeup (Including Hispanic Origin) of Dislodged Renters

Among all householders (Figure I–6, right panel), non-Hispanic Whites accounted for
slightly more than two-third (67.5 percent) of households in 2017. The Hispanic and
Black shares followed accounting for 13.2 and 12.1 percent respectively. The non-
Hispanic White share of all renter householders was 52.5 percent while Hispanic and
Black shares were substantially greater (19.3 and 19.5 percent respectively), holding
Asian share similar.
I find that dislodged renters were most likely to be non-Hispanic White,
comprising 61.0 percent of 8.0 million dislodged renters. This non-Hispanic share is
between that of all renters (52.5 percent) and all householders (67.5 percent). Considering
that the dislodged renters were also likely to be the youngest, it can be inferred that a
large number of young non-Hispanic Whites gave up independent living and chose to live
at parents’ home or double (triple) up with roommates, which has been confirmed by
other studies (Pew Research Center, 2013; U.S. Census Bureau, 2017). Conversely,
Hispanic and Black residents comprised 15.3 and 17.5 percent of the dislodged renters,
which are smaller shares compared to all renters.

25 – Chapter I
Personal Income of Dislodged Renters by Age Group

I now turn to examine the economic status of dislodged renters by extending the basic
housing−demographic calculation. I segregate the entire population by personal income
quartile and add the income dimension into the basic housing−demographic calculation.
The extended analysis allows me to see that, over time, how each income quartile has
been growing. I particularly focus on changes between 2006 and 2017, the last decade.
Figure I–7 shows how each income quartile grew or shrunk since 2006, with age
on the x-axis and absolute count of renters that are greater (positive) or fewer (negative)
than expected on the y-axis. The top income quartile is highlighted dark green while the
bottom income quartile is highlighted dark red.
I find that a considerable loss of renters relative to the expected value was
concentrated among the two youngest age groups (15 to 24 and 25 to 34). In age 25 to 34,
two higher-income brackets (857 thousand) can be seen as diverted homeowners who
were thrown back into renting. For the regular renters who have below-median personal
incomes (highlighted orange and red), in contrast, 979 thousand (age 15 to 24) and 386
thousand (age 25 to 34) of people who should be normally expected to be renters gave up
renting and totally disappeared from the housing market. These dislodged people still
reside in the same area while living in their parents’ home or cohabiting to increase their
spending power. Obviously, it was the younger people with lower earnings who were
bumped out of the market in times of housing shortage.

26 – Chapter I
Figure I –7. Gains or Losses of Renters Relative to Expected, Within Ages and Personal
Income Quartiles, United States, 2006 to 2017

Sources: Census 2000 5-percent IPUMS; 2006 and 2017 ACS 1-year IPUMS.

Post-Dislodgement Living Arrangements of Dislodged Renters

In this section, I divide up the entire population by relationship to householder and add
the relationship dimension into the basic housing−demographic calculation. Thus, I
estimate how many non-head populations in different categories of relationship to
householder would be normally expected in 2017 under the baseline standards. This
extended housing−demographic method can estimate expected number of non-head
population in a relationship category for a given year, 2017 in this essay, by comparing
27 – Chapter I
actual non-head population surveyed in 2017 to what was estimated by applying the
earlier, standard rates (my baseline from 2000) to the actual population by age,
racial/ethnic group, and relationship to householder in 2017. When the expected number
of non-head population is smaller than the actual number of non-head population in the
same relationship category, this gap indicates how many dislodged renters chose the non-
head status instead of being household (either as owner or renter).
Figure I–8 layouts alternative living arrangements that the dislodged renters chose
instead of head status, with age on x-axis and percent of dislodged renters in each age
group who chose different non-head relationships on y-axis. Family relations, such as
Child (orange) and Other Relative (yellow), were located at the bottom and filled with a
warm color scheme while non-family relations were placed at the top and filled with cold
colors; Partner (sky blue), Roommate (blue), and Other Non-relative (dark blue). Note
that spousal status was not included as an adaptive strategy because being a spouse
cannot be considered a post-dislodgement living arrangement; instead, given that spouse
status is always subject to head by definition, I included declining spouse population
(similar scale as declining head population) into total count of dislodged household
population. Since I focus on the household population in the dwelling unit, the
institutionalized population were excluded in my analysis. Age 15-24 was the only age
group that had lower actual partner population than expected, appearing the only
downward bar in Figure I–8.

28 – Chapter I
Figure I –8. Living Arrangements of People Who Have Forgone Expected Headship and
Spouse Status, Excepting Institutionalized Population, United States, 2000 to 2017

Sources: 2000 Decennial Census IPUMS and 2017 ACS 1-year Estimates IPUMS files.

Overall, there is a clear pattern across age groups: the younger the dislodged
renter is, the more likely they end up living with their parents. Among dislodged renters,
74 percent of age 15-24 and 62 percent of age 25-34 went back to or continued living at
parents’ home; an even half (53 percent) of age 35-44 ended up living parents after being
dislodged. Living with parents is frequently reported as the most prevalent living
arrangement among young adults in the nation. However, none of the existing reports
explain how many or how much of young adults living with parents were dislodged.
Conversely, older dislodged renters tend to seek living with their siblings, uncles,
aunts, and other family members. The most prevalent strategy for dislodged renters aged
29 – Chapter I
55 or more was living with relatives. However, non-family people (colored cold color
scheme) appears to play a key role in accommodating dislodged older population,
particularly half of the dislodged elderly (65-74 years old) ended up living with non-
family head.
By comparing my 2017 estimate (absolute counts used for Figure I–8) to actual
2017 ACS data, I can separate dislodged sons and daughters from normally expected
children living with parents. 2017 ACS data shows that there were 44.3 million young
adults (age 25-34) in the nation and, among them, 8.8 million lived with parents. My
estimate reports 3.9 million young adults (age 25-34) were dislodged and ended up living
with parents, indicating 44 percent (3.9 million divided by 8.8 million) of young adults
living with parents in the U.S. are actually the dislodged renters. This means that nearly 9
percent (3.9 million divided by 44.3 million) of all age 25-34 population in the nation
was bumped out of the market and moved back to their parents’ home.

Unmet Housing Needs in Metropolitan Areas

An analysis parallel to the national estimation was conducted for the 50 largest
metropolitan areas. For simplicity of comparison, I focus solely on the scale of
dislodgements accumulated over the entire period of 2000 to 2017, the most recent year
for which data is available. Rental dislodgements are observed in every metropolitan
area, however, there were more dramatic shifts in some places than others.
The largest 50 metros provide a comprehensive geographic overview of housing
shortage problem and dislodged housing needs in terms of their share of population and
30 – Chapter I
occupied housing units in the nation. As of 2017, more than half (55.0 percent) of
population in the U.S. is concentrated in the largest 50 metros; similarly, more than half
(53.9 percent) of occupied housing units are concentrated in the same largest metros. I
select the most populous 50 metropolitan areas as delineated by the federal Office of
Management and Budget (OMB) in 2010. I rearranged data from the 2000 Census and
2017 ACS to conform to the 2010 OMB standard which is based on the 2010 Census.
A metropolitan area contains a core urban area of at least 50,000 population and
adjacent outlying areas that have a high degree of socioeconomic integration with the
urban area as measured through commuting. The largest is New York-Northern New
Jersey-Long Island metropolitan area with a population of 20.2 million in 2017. The
second-largest city, Los Angeles-Long Beach-Santa Ana Metro, has a population of 13.3
million including both Los Angeles and Orange Counties. While the Dallas-Fort Worth-
Arlington Metro includes both the City of Dallas and City of Fort Worth, San Francisco-
Oakland-Fremont Metro is separate from San Jose-Sunnyvale-Santa Clara Metro. The
full list of the 50 largest metropolitan areas is presented with 2017 population in
Appendix I–D.
I calculate the expected number of households as a reference point for judging
unmet housing needs in each of those 50 metropolitan areas. The actual population
recorded in 2017 ACS is used as a base for estimating the expected occupied housing,
with per capita occupancy rates supplied from the 2000 Census. My estimate of housing
needs is measured by the shortfall of actual occupied housing relative to what was
estimated for each metro.
31 – Chapter I
I can use the expected growth of households as a barometer for judging how
adequate was growth in the housing supply. In principle, one might assume that every
additional expected household might require one additional housing unit. However, given
the rate of obsolescence and demolition in the housing stock (on the order of 0.5 percent
per year over 17 years) and the presence of normal vacancies (roughly 4 percent) in any
stock of housing units, the total increase in housing units required over 17 years could be
12 percent greater than the actual increase in the size of the housing stock at the end of
the period. Thus, I can roughly use a 1.12 multiplier for the relation between construction
permits and actual increases in housing opportunity over 17 years (Nelson 2013). A
second, more direct measure would be a 1-to-1 increase between the actual increase in
occupied units and the expected growth in the number of households.
When I examine the actual increases in housing in a sample of the 50 largest
metropolitan areas, I expect that the rate of production falls well short of the idealized
ratios just expressed. The extent of that underproduction and the resulting housing
shortages is an empirical question. The housing−demographic method enables the
estimation of expected housing growth, or housing needs, which is independent of the
actual housing outcomes over the period from 2000 to 2017.
In fact, a comparison between the actual housing growth and expected growth in
households (equal to occupied units by definition) reveals a clear picture of housing
shortages prevalent in the large metros. A regression of 17 years of building permits on
expected household growth yields an estimate that 850 permits were issued for every
1,000 households that were expected to be added, well short (only 76 percent) of the
1,120 permits that might be expected after taking account of demolitions and vacancies.
32 – Chapter I
And when the regression switches to the actual growth in occupied units, only 677
households were added from 2000 to 2017 for every 1,000 expected households added.
Clearly, on either measure of production, the average relationship in the 50 metros
sample is for only 68 to 76 percent (two-thirds to three-quarters) of what would have
been expected for the period from 2000 to 2017.
Despite the prevalence of deficit, the magnitude of housing shortages was
markedly greater in some metros than in others. Figure I–9 shows scatterplots that place
actual housing growth on the vertical axis and the expected housing needs on the
horizontal axis. Housing growth on the y-axis is defined in two ways: summed number of
annual building permits from 1998 to 2016
12
in the left panel and changes in the number
of occupied housing units from 2000 to 2017 in the right panel. As presented above,
expected household growth from 2000 to 2017 was calculated by subtracting the actual
number of households in 2000 from the expected number of households in 2017. These
scatterplots have the advantage of showing which cities had less actual housing growth
than indicated by average relationship with the expected housing needs. Whereas these
trend lines hypothetically could have slopes of 1.12 (building permits) or 1.0 (actual
growth in occupied units), the trend lines fall well short of that. Thus, the lower a metro
falls below the trend line, the greater is the metro’s shortage relative to the average
expected shortage.
Housing growth fell short of the expected housing growth in more than half of
the largest metropolitan areas. In the left panel, 31 metropolitan areas fell below the trend
line indicating that those places authorized building permits at a much lower rate than

12
Building permits are lagged two nominal years to allow time for completion and occupancy.
33 – Chapter I
expected based on the collective experience of the 50 largest metros. The most serious
deficit of building permits was observed in the Los Angeles metro because it permitted
405 thousand housing units which amounts to only 67.9 percent of the expected permit
counts (597 thousand) based on the 50 largest metros’ average experience for the last 17
years. In contrast to Los Angeles, other sunbelt cities outperformed or at least closely
kept up to the expected housing growth in terms of permit authorization for the same
period, including Houston, Dallas, Atlanta, Phoenix, Miami, and Las Vegas.
A similar prevalence of housing shortages is confirmed in the right panel that
compares actual household growth (y-axis) to the expected household growth (x-axis).
Among 50 metros, 21 fell below the trend line, implying those cities added housing at a
lower rate than was expected based on the collective experience of the 50 largest metros.
Again, Los Angeles performed poorly in adding housing relative to other metropolitan
areas. It is noteworthy also how poorly New York added households (occupied housing),
even worse than on construction permit authorization. This possibly indicates a more
substantial volume of demolition of older units in New York.
In sum, the prevalence of housing shortages among large metropolitan areas was
estimated through comparisons between growth in housing and expected household
growth using two different measures of housing growth (construction permits and actual
increase in occupied housing units). Despite pervasive housing shortages in large metros,
this likely varies between renters and owners. In addition, some demographic groups may
bear the brunt of shortage more acutely than others.
34 – Chapter I
Figure I –9. Comparison between Actual Housing Growth and Expected Growth of Households, 50 Largest Metropolitan Areas, 2000
to 2017 (Unit: Thousands)

(a) Building Permits (Y-axis) versus
Expected Growth of Households (X-axis)

(b) Actual Growth in Occupied Units (Y-axis) versus
Expected Growth of Households (X-axis)

Notes: Dashed lines present an ideal housing market condition where actual housing growth, measured by either growth in occupied
units or summed residential building permits, exactly matches the expected housing growth. Metropolitan areas laid below the line
have shortage problems. Building Permits are lagged two nominal years to allow time for completion and occupancy.
Sources: U.S. Census Bureau, Annual Building Permits Survey; Census 2000 5-percent IPUMS and 2017 ACS 1-year IPUMS.
35 – Chapter I

Potential Factors Explaining Metropolitan Variation in Dislodgements

Several regional factors can help explain why the scale of dislodgements was greater in
some metropolitan areas than in others. Dislodgements could be related to a variety of
contextual factors, such as housing supply and demand, a sharp decline in
homeownership, rising rents and house prices, income changes, and rental affordability.
Relationships are investigated through correlation tests by which correlation
coefficients are computed between the dislodgement rate and the other regional factor.
Dislodgement rate is defined as actual renter households less expected renter households
as a percentage of “final” expected renter households in 2017. Dislodgement is a loss and
therefore negative correlations increase dislodgements. Comparison of such correlation
coefficients may lead to insights as to which factors correlate most strongly with
dislodgement among 50 largest metropolitan areas.
Table I–2 summarizes the correlations between dislodgements and other regional
factors by presenting correlation coefficients on the last two columns; relatively weak
coefficients are placed on the left of the two columns (highlighted grey) while stronger
coefficients are on the right.
The strongest factor is the sharp decline in homeownership rates. In the cascade
model, I showed that rental dislodgement was largely initiated by large spillovers of
would-be owners into the rental market. This essay found a very strong correlation (r =
−0.72) that indicates large cities with sharper decline in homeownership rates are likely to
have much greater dislodgements.

36 – Chapter I

Table I –2. Correlations between Dislodgements and Other Regional Factors

Regional Factors Correlated to the Scale of Dislodgements
(Dislodgement is a loss;
negative correlations increase dislodgement)
Correlation Coefficient
Weaker
(r < 0.30)
Stronger
(r >= 0.30)
Decline in
Homeownership
per capita Homeownership Decreases −0.72
per Household Homeownership Decreases −0.49
Supply
and Demand
More New Construction +0.46
Young Adults Growth −0.51
Employment Growth −0.50
Price
and rent
Median House Price Increases −0.37
FHFA’s Single-family Price Index
Increases
−0.28

Median Gross Rent Increases −0.25
Income
Household Income Increases +0.08
Owner Household Income Increases +0.13
Renter Household Income Increases −0.21
Rental
Affordability
Prevalence of 30%+ Cost-Burden in 2017 −0.45
Prevalence of 30%+ Cost-Burden in 2000 −0.30
Prevalence of 50%+ Cost-Burden in 2017 −0.19
Prevalence of 50%+ Cost-Burden in 2000 −0.12
Increases in 30%+ Cost-Burden (pp.) −0.25
Increases in 50%+ Cost-Burden (pp.) −0.18

Notes: See Appendix I–F for additional tests on the relationship between young headship
rate and incidence of rent-burden among young adults. Variables measure changes from
2000 to 2017 except for the measures of rental affordability level in 2000 and 2017
respectively.
Sources: Census 2000 5-percent IPUMS and 2017 ACS 1-year IPUMS; Census Bureau’s
Building Permits Survey; Bureau of Economic Analysis (BEA)’s Annual Employment
Data; Federal Housing Finance Agency (FHFA)’s Annual House Price Index (HPI).

This essay assumes metros with more new construction should have had fewer
dislodgements through expanded overall housing opportunities. On the contrary, demand
pressures derived from employment growth and a growing number of young adults who
normally seek entry-level rentals should have increased dislodgements. These supply and
demand factors have proven to have a strong relation with dislodgements as assumed.
37 – Chapter I

I hypothesize that metros with a faster increase in rents should have had a greater
dislodgement. Increase in prices is also expected to increase dislodgements by forcing
potential buyers to stay longer in the rental market and rising competition for limited
housing. Price (or rents) effects were largely confirmed as hypothesized.
Income changes are assumed to correlate with dislodgement in both ways. Metros
with greater increase in income may have households who have greater purchase power
and fewer dislodgements may occur. At the same time, income effect could have been
already offset by higher living costs. Income variables were found to have a weak and
mixed correlation with dislodgement indicating the need for multivariate statistical tests
to capture their true effects on dislodgement.
Lastly, I assume that higher rental cost-burden should have a positive association
with greater dislodgements. The rationale for the expectation is that a prevalent cost-
burden will prompt households that are already stretched to give up their householder
status and leave the housing market (dislodgement). Findings strongly confirm that
dislodgement is greater where the rental affordability problem is more serious. This
finding is constant for not only the prevalence level of rental cost-burden but its increase
over time. Even though housing shortages hide the dislodgements, the close relationship
based on the housing−demographic method suggests that more dislodgements occurred in
a metro with more serious affordability problems and housing shortages, which amplifies
the magnitude of the crisis in many metropolitan areas.
13

13
Additional finding is that household formation rates among young adults were lowest in metropolitan
areas that had serious affordability problems and also housing shortages. Appendix I–F presents a
scatterplot that places headship rate of householders (ages 25 to 34) on the vertical axis and the renters
38 – Chapter I

5. Discussion

This essay has estimated expected growth in occupied housing and made visible the
dislodged households due to housing shortages through the housing−demographic
method. Results also identified the dislodged households as those with lowest personal
incomes and adults under age 35 whose household formation is most flexible due to other
accessible living arrangements, such as remaining in parents’ homes or doubling up with
roommates. Prevalence of housing shortages was confirmed in the nation as a whole as
well as across the largest metropolitan areas by comparing actual housing growth to
expected household growth. Lastly, strong correlations between dislodgement and
regional factors of supply and demand have been also demonstrated.
Findings support several important implications for policies that try to address the
housing shortage and meet currently unmet housing needs. These include (1) the massive
diversion of would-be homeowners into rental market, (2) the shift in the composition of
renters amid a steady total number, (3) hidden housing needs neglected by conventional
measures, and (4) using estimates of dislodgement to augment traditional measures of
affordability that are based only on survived households. The combination of these
factors is clearly relevant for better describing the dynamics of the rental housing crisis.
Identifying where new rental demand comes from. One important finding of
this essay is that there was a huge (7.8 million) diversion of would-be homeowners into

(ages 25 to 34) with payment burden on the horizontal axis, both in 2017. It shows a strong negative
correlation (r = −0.60 for 30%+ rent-burden and r = −0.47 for 50%+ rent-burden) implying that household
formation rates are lower in metropolitan areas where rent burdens are higher.
39 – Chapter I

the rental market. The lower increase in renters (0.2 million) than expected shows that
rental supply was not sufficient to accommodate the unexpected massive demand
especially in times of shortage. This result has an important implication for current
housing policy direction that focuses on ad hoc treatment of observed symptoms (either
unaffordability or shortage) overlooking their root cause. Various efforts to boost housing
supply by federal, state, and local governments would help expand the overall capacity of
the rental sector. However, the rental shortage may last until the current massive spillover
of would-be homeowners is stopped or redirected to somewhere else.
Recognizing the importance of a massive shift in the composition of the
rental market. The second overall finding is that there are 2.4 million more single-
family renters than would have been expected while 2.6 million less multifamily (and
other structure type) renters than expected. Taken together, a small decrease (0.2 million)
in the number of renters compared to what would have been expected masks a massive
spillover (7.8 million) of would-be homeowners into renting and, hence, a very large
dislodgement (8.0 million) of expected renters out of market. Although the total number
of renters has held steady, beneath the calm surface is a violent churn and intense
competition where desperate home seekers scavenge and overpay, force out others,
double (or triple) up, or give up household status. HUD (2017) also emphasizes that the
most important cause of the recent (2013 to 2016) increase in worst case housing needs
was a “notable shift from ownership to renting.”
Accounting for hidden housing needs to estimate the full extent of the rental
crisis. The third finding is that 8.0 million renters likely have been dislodged from
household headship and are either forced to be doubled (tripled) up, remained living with
40 – Chapter I

parents or roommates, or otherwise completely disappeared out of the housing stock, all
while still resident in the population. It may be these dislodged and completely
disappeared households would be drawn from those with the worst case needs as termed
by HUD. As much as the dislodgements, the true magnitude of the shortage crisis should
be underestimated by any conventional methods.
Revealing more true variation in rental hardship across the large metros. The
final finding is that a strong correlation exists between dislodgement and affordability
measures, as well as with various regional factors, such as housing supply and demand.
In fact, I expect that this finding may complement conventional measures of rental
affordability that show little variation among places. Traditional methods likely
underestimate the rent burden because the housing shortage and rising rents progressively
eliminate the households with the greatest burden. With the most burdened households
removed, the prevalence of excessive burden among survived households may fail to rise
even as the crisis deepens in a metro. Incorporating the dislodgement rate that dampens
the true variation in rental hardship, augmenting traditional measures of affordability, can
thus provide a more complete indication of the extent of rental crisis across metros.
Further steps in the research require multivariate estimations that incorporate
measures of supply and demand, along with other regional factors. Explaining where
housing need is greatest and where the need was most underestimated by conventional
method remains to be tested.

41 – Chapter II

Chapter II. Depressed Access to Affordable Housing Due to Higher-Income
Occupancy: Evidence from U.S. Metropolitan Areas, 2006 to 2016

1. Background

A severe housing shortage has been afflicting metropolitan areas in the United States.
The shortage has resulted in intense competition for a limited supply of rental housing,
driving up rents, generating unaffordability, and disrupting access to housing by a large
share of the population (HUD, 2017; JCHS, 2018). This crisis has attracted considerable
interest from policymakers, who have become concerned about its disproportionate
impact on low-income households. In August 2017, the U.S. Department of Housing and
Urban Development (HUD) published a biannual report, Worst Case Housing Needs:
2017 Report to Congress, urging readers to “understand the evolving dimensions of a
persistently expanding shortage of decent and affordable rental housing for lower-income
households” (HUD, 2017). Underscoring the magnitude of the rental housing crisis, even
more than the HUD report, is excessive rent burden spurred by rental competition and
rapidly increasing rents (NYU Furman, 2017; Urban Institute, 2017). The excessive rent-
burden reached a peak in 2011, when 49.3 percent of renters paid more than 30 percent of
their incomes on housing, well above the 36.8 percent in 2000, prior to the housing
bubble (JCHS, 2017). Although rental affordability pressures started to ease steadily
since 2011, it remains widespread across incomes (JCHS, 2018).
The prevalence of cost burdens among low-income renters is due in part to the
fact that households with moderate or even higher incomes take the housing units that
42 – Chapter II

low-income renters could afford. In 2016, 44.2 percent of the affordable low-cost rentals
that Extremely Low-Income (ELI) households (earning less than 30 percent of area
median income (AMI), income definitions will be discussed in the following section)
could afford were occupied by higher-income households and consequently became
unavailable. Thus, the underlying challenge in rental affordability is not only the shortage
of affordable supply at the 30-percent-of-income standard but also the actual occupancy
by the poorest renters. Particularly, many large metro areas experienced a decline in their
stock of affordable low-cost rentals in both absolute count and relative shares during and
after the Great Recession (JCHS, 2018; Weicher et al., 2017). Given the declining supply
of low-cost rentals, examining rental availability is important for helping planners and
policymakers determine ways to expand housing opportunities for the poorest renters.
In an attempt to explain the depressed access to affordable low-cost housing by
low-income households, studies noted the gap between low-income demand and
affordable supply. Nationwide, for every 100 renter households earning 30 percent or less
of the metropolitan area median household income, roughly 82 affordable rental housing
units exist (NYU Furman, 2017; NLIHC, 2018). In addition, a number of metros have
experienced declines in their stocks of low-cost rentals, or referred to in this paper as
‘affordable rentals,’ in both absolute count and relative shares during and after the Great
Recession (HUD, 2016; Weicher et al., 2017). Worsening the gap between supply and
demand, much of the subsidized rental stock is at risk of loss, the result of either under-
maintenance or expiring affordability periods, with the number of lower-income renters
far outstripping the number of affordable units at the lowest end of the market (JCHS,
2018; Lens, 2017).
43 – Chapter II

Studies have also shown the efficacy of the supply gap approach, which directly
shows the shortage of affordable supply relative to demand, to describe declining access
to affordable low-cost housing for the low-income. However, as such an approach entails
the use of optimal sorting, a conceptual construct that assumes that the lowest-cost rental
units are filled with the lowest-income renters, it may fail to capture the nuances of the
actual housing options for lower-income households (Collinson, 2011; Joice, 2014;
NLIHC, 2018). Middle- or even higher-income households are more likely to be selected
over lower-income households when competing for the same affordable unit in times of
housing shortage; thus, affordable units occupied by better-off households may not be
truly available to lower-income households (JCHS, 2018; NLIHC, 2019).
The phenomenon termed ‘availability’ which can be defined as ‘occupied
14
by a
household at or below the lower-income threshold’ accounts for actual occupancy of
affordable housing by lower-income households (Collinson, 2011; Joice, 2014). While
higher-income households outbid middle-income households for the former middle-level
housing, the middle-income households must relocate to former lower-end housing,
which, in turn, leads to fewer opportunities for lower-class renters, who must then spend
more of their income on poorer housing. This long chain of events results in greater
problems for low-income renters.

14
Unoccupied housing units were not considered in this essay, which account for 12.4 percent of all
housing units (= 16,842,710 / 135,702,775 × 100) across the nation (2016 American Community Survey
B25002 summary table). ACS-based national vacancy rate was steady between 12.1 percent and 12.6
percent during this essay’s study period (2006 through 2016), save lower level of 11.6 percent in 2006
(peak of the housing market) and higher level of 13.1 percent in 2010 and 2011 (bottom of the market).
44 – Chapter II

Regarding the actual availability of affordable housing to low-income households,
most studies have verified the prevalence of rent burdens among lower-income renters
resulting, in part, from middle or even higher-income households occupying housing
units that lower-income renters can afford (Collinson, 2011; Joice, 2014). When the
dimension of availability is applied, only 43 affordable and available rental housing units
exist for every 100 renters making 30 percent or less of the AMI, suggesting that almost
half (47.5 percent) of affordable housing is occupied by middle or even higher-income
households. Thus, the source of the rental housing crisis lies not only in the shortage of
affordable rental housing at the 30-percent-of-income federal standard but also actual
occupancy by the poorest renters, particularly in metropolitan areas.
Existing studies on the availability of rental housing are limited with regard to the
number of observation years, and the use of general descriptive approaches and most
have focused on the nation as a whole, states, counties, or neighborhoods. For example,
although one study in Collinson (2011) examined housing availability in 2007 and 2009,
and the National Low Income Housing Coalition (NLIHC) report is updated annually,
their findings show trends only in specific years. In contrast, this study examines
metropolitan areas, which commonly represent the state of the housing market. I analyze
a sample of 200 metros, including the most populous areas, such as New York and Los
Angeles, and also mid- and small-sized areas like Buffalo and Greenville. In addition, the
data in this study cover 11 consecutive years from 2006 to 2016, including peak, bust,
and recovery of the housing market. Beyond general descriptive trends, I examine
contextual factors influencing rental unavailability in specific metros that remains
unexplored.
45 – Chapter II

This essay begins with a description of affordable low-income housing in the
context of total housing stock. As part of this background, I describe temporal trends and
geographic variation in low-income housing opportunities. Using American Community
Survey (ACS) microdata from 2006 to 2016, I measure the number of affordable rentals
occupied by households earning more than the low-income threshold and became
unavailable to the poorest renters in each metro. I estimate housing availability
specifically to the poorest renters because the bottom-income group is the least likely to
find housing and the primary target population for rental subsidy programs. I lastly run a
set of regression models to investigate associations between rental availability and
metropolitan contextual factors.
In sum, this essay examines how many affordable low-cost rentals that the poorest
renters were able to afford were occupied by middle or even higher-income households in
the study areas and what contextual factors have limited or expanded rental availability in
some areas more than in others. Although in the private market, middle and higher-
income households are free to occupy rentals that are affordable to the poorest
households, their doing so largely depends on the market-rate supply of housing, which
widely varies across metros. Subsidized rentals, however, are available only to the
poorest households through income restrictions. With a focus on market-rate
constructions and housing subsidies, this essay runs regressions to determine the factors
most strongly associated with rental unavailability. Determination of rental availability
problems should provide valuable information for policymakers faced with decisions
regarding the extent to which the market satisfies lower-income housing demand and
local, state, and federal governments need to fund housing subsidy programs.
46 – Chapter II

2. Low-income Housing Access and the Great Recession

In this section, I provide a background of low-income housing problems in the context of
total housing stock followed by an explanation of how access to affordable low-cost
housing by low-income households is limited by middle- and higher-income occupancy.
Then, I describe a national trend in access to low-cost affordable housing. Finally, I
examine the regional variation across metropolitan areas.

Low-income Housing in Context of Total Housing Stock

To describe low-income housing in context of all occupied housing units in the nation, I
first classify households into four exclusive income groups including high-income
earning 120%+ of median household income, middle-income earning 80-120% of
median, low-income earning 50-80% of median, and very low-income earning 0-50% of
median, all categorized on the basis of national median household income of $57,617 in
2016 (Table II-1). In order to reflect geographic variations in incomes and cost of living,
this essay adopts localized metro-specific median household income instead of the
national median income in the following sections on metropolitan analysis. Particular
interest of this essay is the very low-income households who earn $28,809 or less per
year and are in the greatest need to secure a place to live. As will be shown, this poorest
group is the last in line for housing.

47 – Chapter II

Table II –1. Income Group and Corresponding Dollar Ranges of Income, U.S., 2016

(a) Percent Range
Relative to the
National Median
Household Income
(b) Dollar Range
of Household
Income (2016$)
National Median HH Income = $57,617
High-income HH Above 120% of Median Above 69,140
Middle-income HH 80 to 120% of Median 46,094 to 69,140
Low-income HH 50 to 80% of Median 28,809 to 46,094
Very Low-income HH 0 to 50% of Median 0 to 28,809

Notes: Universe is occupied housing units in the United States. HH is a household.
Income groups are exclusive to each other, including high-income group earning 120%+
of national median HH income, middle-income earning 80-120% of median, low-income
earning 50-80% of median, and very low-income earning 0-50% of median. Thresholds
of income-to-AMI are based on HUD’s “at or below” definition.
Sources: 2016 ACS IPUMS Microdata files.

The low-income housing problem is not simply limited to the lower-end
submarket but is interconnected to middle- and even higher-end submarkets as well. The
left column of Figure II–1 below shows all households (118.9 million) in the United
States in 2016. These are distributed by tenure (owner versus renter) and income group.
At the bottom of the distribution in the left column of Figure II–1, 14.6 percent of total
households in the nation are found as renters with very low-income (earning at or below
50 percent of national median household income). They are the last in line for housing.
Stacked above them are other renters with higher incomes and who could pay more if
needed. At the top of the entire housing market competition are the homeowners.
48 – Chapter II

Figure II –1. Breakdown of Total Occupied Housing Stock, United States, By Tenure and
Income, 2016

Notes: National median household income is $57,617 according to 2016 ACS IPUMS
data. Thresholds of income-to-AMI are based on HUD’s “at or below” definition.
Estimates of government assistance include some occupants who are above the very low-
income levels, as discussed in text, implying that the market-rate share of very low-
income housing could be even greater.
Sources: HUD’s Picture of Subsidized Households Data; HUD’s National Low Income
Housing Tax Credit (LIHTC) Data; Public and Affordable Housing Research Corporation
(PAHRC) and NLIHC’s National Housing Preservation Database (NHPD); Housing
Assistance Council’s USDA Rural Housing Data; Schwartz’s Housing Policy in the
United States, 2015; ACS 1-year IPUMS, 2016.

49 – Chapter II

It is true that subsidized rental housing
15
is designed to be affordable and
restricted to low-income households. The public might assume that a majority of the low-
income households are housed by government-subsidized rentals. As shown in Figure II–
1, however, there are actually more low-income renters (9.2 million) in the market-rate
housing than in the government-subsidized housing (8.1 million).
The very low-income renters (17.3 million) in the nation are shown in the right
column of Figure II–1 by the source of their housing. The market-rate segment at the
bottom of the right column shows that more than half (54.2 percent) of housing occupied
by very low-income renter households in 2016 is actually market-rate, and slightly less
than half (43.9 percent) of that is subsidized by the Department of Housing and Urban
Development (HUD). The remaining (1.9 percent) is subsidized by the United States
Department of Agriculture (USDA), which is mostly located in rural areas and
encompasses a very small portion of very low-income housing in the nation.
Additionally, the estimate of government-subsidized housing occupied by the
very low-income group may be somewhat overstated because it includes the entire count
of Low-Income Housing Tax Credit (LIHTC) units and voucher assisted households
which are not exclusively restricted to renters in the very low-income bracket (<50% of
AMI).

15
Government rental housing programs include HUD-assisted Public Housing, Low-Income Housing Tax
Credit (LIHTC), Housing Choice Vouchers, Mod Rehab, Project Based Section 8, S236/BMIR, 202/PRAC,
and 811/PRAC. In addition, USDA-assisted rental housing program indicates section 515. Some units draw
upon multiple programs, so that the total of units supported in different programs may double count the
total scale of public assistance.
50 – Chapter II

In sum, the estimate of total 8.1 million assisted rental units in the nation as a
whole is likely overstated because a household (or a housing unit) may receive more than
one assistance, which then implies that I have understated the number of very-low
income households living in market-rate housing. Even with these margin of errors, low-
income renters housed in the market-rate housing (9.2 million) outweigh those in
subsidized units (8.1 million). Overall, it is clear that market-rate housing plays a much
larger role than government assistance in providing housing to the poorest Americans, as
observed also by Schwartz (2015) and Weicher et al. (2017).
16

Match Between Income Group and Rent Bracket

A common method to see how different income groups match different rental brackets is
to dissect all renter households (i.e., renter-occupied housing units) in terms of the
tenant’s income, rent payment, and rent-to-income ratio (or rent burden) (Collinson,
2011; Joice, 2014; JCHS, 2018; Lens, 2017; NLIHC, 2018).
To set the context, Figure II–2 presents the income and rent payment dimensions,
showing all renter households (44 million) in the nation in 2016. The rows show income

16
In the nation as a whole, Weicher et al. (2017, p.1) estimates at least 73.2% (= 13.4 million / 18.3 million
× 100) of very low-income renters (<50% of AMI) were housed in the private market in 2013. Similarly,
Schwartz (2015, p.9) estimates 8.3 million subsidized rental housing in the nation in 2012, which not only
very low-income renters but slightly better-off renters are eligible for. Given that there were 25.4 million
lower income renters (<80% of AMI) in 2012 (our ACS-based estimate), roughly 67.1% (= (25.4 million –
8.3 million) / 25.4 million × 100) of lower income renters (<80% of AMI) appears to be left in the market
without any subsidy.
51 – Chapter II

groups arranged in 10-percent-intervals of household income as a percent of national
median income ($57,617), while the columns show the 10-percent-interval gross rent
groups, arrayed on scale that shows rents equivalent to the corresponding income level.

Figure II –2. Renter Households in 10%-interval Income-Rent-Group, U.S., 2016

Notes: Universe is all occupied rental units (or renter households) in the nation in 2016.
Sum of all cell values matches the universe of 43,757,527 renter-occupied housing units.
Source: 2016 ACS 1-year IPUMS file.

The first row of Figure II–2 shows renters who earned less than $5,762 per year
(10 percent of national median household income of $57,617) while the first column
shows occupied rental units that cost $144 (= 30 percent of 10 percent of national median
household income divided by 12 months) which is less than 30-percent-standard of the
corresponding income group. The far upper left cell shows that there were 373,792 renter
households who earned the lowest income and occupied the lowest-rent-bracket unit.
This table matches incomes and the number of households paying corresponding
rents. Any renter with an income only 10 percent of the national median cannot afford to
pay rents equivalent to what is affordable to renters with income equal to 20 percent of
the national median. With this table structure, it is convenient to mark off the
combinations of cells that are affordable and those that are not.
0-10% 10-20% 20-30% 30-40% 40-50% 50-60% 60-70% 70-80% 80-90% 90-100% 100-110% 110-120% 120-130% 130-140% 140-150% >150%
0-10% 373,792 173,225 191,160 251,435 365,859 343,562 295,959 210,634 144,576 121,086 79,834 66,273 44,280 28,788 24,864 89,093
10-20% 302,909 708,609 383,619 431,592 516,493 408,592 321,654 204,519 121,291 106,460 65,901 42,634 30,908 26,818 19,457 42,656
20-30% 277,284 231,207 479,730 483,244 633,571 545,501 426,646 271,155 159,430 129,892 77,373 58,392 39,359 27,093 20,025 61,748
30-40% 217,228 91,597 219,822 445,999 629,473 619,593 473,482 331,098 200,187 157,191 98,231 70,155 52,226 35,415 22,576 69,752
40-50% 182,211 44,538 119,154 287,458 538,242 571,638 478,317 328,444 207,258 145,168 102,448 71,253 51,772 36,190 21,196 58,897
50-60% 156,015 33,022 84,592 229,830 460,761 538,244 476,065 346,072 210,929 170,164 109,825 77,226 54,462 36,056 27,380 63,459
60-70% 147,679 26,224 68,032 181,885 410,326 547,079 540,126 401,259 267,111 207,162 140,277 97,352 60,487 41,287 29,432 69,145
70-80% 114,248 15,590 38,882 112,897 260,312 358,487 393,036 317,935 201,938 168,085 121,106 80,709 54,217 34,307 30,830 64,836
80-90% 100,722 11,406 33,159 87,602 205,113 304,360 351,297 298,129 211,761 170,016 121,407 86,118 59,710 41,817 24,614 74,368
90-100% 77,893 8,863 25,484 72,794 164,241 253,352 304,071 265,448 191,768 167,823 115,036 80,038 50,576 35,124 22,054 74,647
100-110% 76,674 6,398 20,908 51,427 125,249 205,745 253,991 253,740 184,552 160,196 122,411 88,255 63,095 44,835 31,298 80,637
110-120% 60,236 5,689 14,295 40,055 90,500 157,245 201,461 188,489 153,298 129,301 103,806 83,073 54,825 35,349 24,037 68,498
120-130% 50,893 4,518 10,906 31,355 73,042 115,508 155,619 170,158 130,544 135,171 105,746 74,109 56,800 35,887 25,642 78,629
130-140% 50,089 3,834 9,975 28,927 64,264 105,018 160,487 161,152 142,922 140,266 114,143 90,963 67,241 47,707 37,378 98,244
140-150% 37,013 3,637 6,587 19,998 45,188 72,349 92,579 109,949 95,184 96,590 82,913 60,908 49,683 35,879 25,472 65,899
>150% 235,287 15,834 40,076 91,356 193,285 307,691 466,283 551,950 515,724 603,495 586,285 542,443 475,119 380,423 330,927 1,559,234
Renter HHs by Income
0-50% of AMI 50-80% 80-120% 120+% Total
Renter HHs
Rental Housing Units by Gross Rent
Affordable to 0-50% of AMI 50-80% AMI 80-120% AMI 120%+
52 – Chapter II

Figure II–3 shows all renter households in the nation by three levels of cost
burden. Panel (a) shows renters without rent burden by their income and rent payment.
Panels (b) and (c) show renters with moderate rent burden (30-50 percent of income paid
for rent) and severe burden (50 percent or more) respectively. I could identify 30 percent-
or-more burdened renters by summing up panels (b) and (c).
The upper left 5 × 5 cells in each panel of Figure II–3 reveal how many low-
income renters who successfully occupied low-cost rental units could be burdened. If I
sum up the 5 × 5 cells in panels (b) and (c), 5,191,518 low-income renters appear to be
cost-burdened (30%+) even if they occupied rental units that are considered to be
affordable on the basis of the national median household income. Those 5,191,518
renters account for slightly less than two-thirds (60.5 percent) of the 8,579,451 renters
who successfully occupied would-be affordable low-cost rentals.
Previous studies use this approach to report rental housing problems in different
ways. If I divide the sum of the first five columns of all three panels in Figure II–3
(13,214,518 rental units) by sum of the first five rows of all three panels (17,438,391
renters), it reports the estimate of affordable supply gap which will be discussed more
closely in the following section (Lens, 2017; NLIHC, 2018). Other researchers sum up
lower left 11 × 5 cells of panel (a) and report the number (4,635,067) of low-cost rental
units that were occupied by better-off renters (Collinson, 2011; Joice, 2014). It is clear
that all the higher-income households were not rent-burdened in terms of 30-percent-
standard by occupying lower-price units as marked off in the lower left 11 × 5 cells of
panel (a) while none of them appear in the same cells of panels (b) and (c).

53 – Chapter II

Figure II –3. Renter Households in 10%-interval Income-Rent-Group, By Level of Rent
Burden, United States, 2016

(a) No-Burden Renter Households

(b) Moderately Burdened (30-50%) Renter Households

(c) Severely Burdened (50%+) Renter Households

Notes: Universe is all occupied rental units (or renter households) in the nation in 2016.
Sum of all cell values matches the universe of 43,757,527 renter-occupied housing units.
Source: 2016 ACS 1-year IPUMS file.
0-10% 10-20% 20-30% 30-40% 40-50% 50-60% 60-70% 70-80% 80-90% 90-100% 100-110% 110-120% 120-130% 130-140% 140-150% >150%
0-10% 301,187 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
10-20% 302,909 423,395 0 0 0 0 0 0 0 0 0 0 0 0 0 0
20-30% 277,284 231,207 238,867 0 0 0 0 0 0 0 0 0 0 0 0 0
30-40% 217,228 91,597 219,822 204,904 0 0 0 0 0 0 0 0 0 0 0 0
40-50% 182,211 44,538 119,154 287,458 246,172 0 0 0 0 0 0 0 0 0 0 0
50-60% 156,015 33,022 84,592 229,830 460,761 243,054 0 0 0 0 0 0 0 0 0 0
60-70% 147,679 26,224 68,032 181,885 410,326 547,079 293,522 0 0 0 0 0 0 0 0 0
70-80% 114,248 15,590 38,882 112,897 260,312 358,487 393,036 182,480 0 0 0 0 0 0 0 0
80-90% 100,722 11,406 33,159 87,602 205,113 304,360 351,297 298,129 120,254 0 0 0 0 0 0 0
90-100% 77,893 8,863 25,484 72,794 164,241 253,352 304,071 265,448 191,768 86,752 0 0 0 0 0 0
100-110% 76,674 6,398 20,908 51,427 125,249 205,745 253,991 253,740 184,552 160,196 66,271 0 0 0 0 0
110-120% 60,236 5,689 14,295 40,055 90,500 157,245 201,461 188,489 153,298 129,301 103,806 42,174 0 0 0 0
120-130% 50,893 4,518 10,906 31,355 73,042 115,508 155,619 170,158 130,544 135,171 105,746 74,109 26,575 0 0 0
130-140% 50,089 3,834 9,975 28,927 64,264 105,018 160,487 161,152 142,922 140,266 114,143 90,963 67,241 24,844 0 0
140-150% 37,013 3,637 6,587 19,998 45,188 72,349 92,579 109,949 95,184 96,590 82,913 60,908 49,683 35,879 13,631 0
>150% 235,287 15,834 40,076 91,356 193,285 307,691 466,283 551,950 515,724 603,495 586,285 542,443 475,119 380,423 330,927 1,277,004
Rental Housing Units by Gross Rent
Affordable to 0-50% of AMI 50-80% AMI 80-120% AMI 120%+
Renter HHs by Income
0-50% of AMI 50-80% 80-120% 120+%
No Burden
(0-30%)
0-10% 10-20% 20-30% 30-40% 40-50% 50-60% 60-70% 70-80% 80-90% 90-100% 100-110% 110-120% 120-130% 130-140% 140-150% >150%
0-10% 14,550 18,084 0 0 0 0 0 0 0 0 0 0 0 0 0 0
10-20% 0 279,310 254,242 10,864 0 0 0 0 0 0 0 0 0 0 0 0
20-30% 0 0 240,863 408,115 175,513 0 0 0 0 0 0 0 0 0 0 0
30-40% 0 0 0 241,095 629,473 434,674 73,207 0 0 0 0 0 0 0 0 0
40-50% 0 0 0 0 292,070 571,638 471,108 164,602 6,025 0 0 0 0 0 0 0
50-60% 0 0 0 0 0 295,190 476,065 346,072 185,705 47,656 0 0 0 0 0 0
60-70% 0 0 0 0 0 0 246,604 401,259 267,111 207,162 96,188 20,873 0 0 0 0
70-80% 0 0 0 0 0 0 0 135,455 201,938 168,085 121,106 80,509 31,987 1,234 0 0
80-90% 0 0 0 0 0 0 0 0 91,507 170,016 121,407 86,118 59,710 39,453 9,183 0
90-100% 0 0 0 0 0 0 0 0 0 81,071 115,036 80,038 50,576 35,124 22,054 12,291
100-110% 0 0 0 0 0 0 0 0 0 0 56,140 88,255 63,095 44,835 31,298 38,508
110-120% 0 0 0 0 0 0 0 0 0 0 0 40,899 54,825 35,349 24,037 45,397
120-130% 0 0 0 0 0 0 0 0 0 0 0 0 30,225 35,887 25,642 60,427
130-140% 0 0 0 0 0 0 0 0 0 0 0 0 0 22,863 37,378 85,095
140-150% 0 0 0 0 0 0 0 0 0 0 0 0 0 0 11,841 61,040
>150% 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 278,770
Renter HHs by Income
0-50% of AMI 50-80% 80-120% 120+%
Moderate Rent
Burden (30-50%)
Rental Housing Units by Gross Rent
Affordable to 0-50% of AMI 50-80% AMI 80-120% AMI 120%+
0-10% 10-20% 20-30% 30-40% 40-50% 50-60% 60-70% 70-80% 80-90% 90-100% 100-110% 110-120% 120-130% 130-140% 140-150% >150%
0-10% 58,055 155,141 191,160 251,435 365,859 343,562 295,959 210,634 144,576 121,086 79,834 66,273 44,280 28,788 24,864 89,093
10-20% 0 5,904 129,377 420,728 516,493 408,592 321,654 204,519 121,291 106,460 65,901 42,634 30,908 26,818 19,457 42,656
20-30% 0 0 0 75,129 458,058 545,501 426,646 271,155 159,430 129,892 77,373 58,392 39,359 27,093 20,025 61,748
30-40% 0 0 0 0 0 184,919 400,275 331,098 200,187 157,191 98,231 70,155 52,226 35,415 22,576 69,752
40-50% 0 0 0 0 0 0 7,209 163,842 201,233 145,168 102,448 71,253 51,772 36,190 21,196 58,897
50-60% 0 0 0 0 0 0 0 0 25,224 122,508 109,825 77,226 54,462 36,056 27,380 63,459
60-70% 0 0 0 0 0 0 0 0 0 0 44,089 76,479 60,487 41,287 29,432 69,145
70-80% 0 0 0 0 0 0 0 0 0 0 0 200 22,230 33,073 30,830 64,836
80-90% 0 0 0 0 0 0 0 0 0 0 0 0 0 2,364 15,431 74,368
90-100% 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 62,356
100-110% 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 42,129
110-120% 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 23,101
120-130% 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 18,202
130-140% 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 13,149
140-150% 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4,859
>150% 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3,460
Rental Housing Units by Gross Rent
Affordable to 0-50% of AMI 50-80% AMI 80-120% AMI 120%+
Renter HHs by Income
0-50% of AMI 50-80% 80-120% 120+%
Severe Rent
Burden (50%+)
54 – Chapter II

A technical limitation is that a housing unit with rent that is affordable to one
household in the corresponding income range may not be affordable to a different
household in the same income range but the unit is still considered affordable for that
entire income range of households. Specifically, the sum of the upper left 5 × 5 cells of
panels (b) and (c) in Figure II–3 reports the number (5,191,518) of low-income renters
who successfully secured low-cost rentals but carried a rent burden of more than 30
percent of income due to the limitation. Consequently, a smaller number (3,387,933) of
final survivors are counted unburdened in the upper left 5 × 5 cells of panel (a).

Affordable Housing Occupied by Higher-income Households

Ideally, the lowest income renters match to the units in the lowest price bracket, while the
moderate-income renters match up to the next higher rent units. In fact, I find a serious
shortage of units available at the bottom. Based on national data in Figure II–4 below, for
every 100 very low-income renter households (dark red bar at the bottom of the right
column), roughly 74 affordable low-cost rental housing units (dark red bar at the bottom
of the left column) were supplied across the nation in 2016 (74 = 12.9 M / 17.3 M × 100).
However, not all of the units affordable to very low-income renters were actually
available to them. In fact, a sizable share of the 12.9 million most affordable units were
occupied by renters with higher income. The higher-income occupancy of low-cost
affordable housing worsens the issue of rental shortage for very low-income renters.

55 – Chapter II

Figure II –4. Rental Units and Renter Households in the United States, Matched by
Affordability and Income Categories, 2016

Notes: Universe is all occupied rental units (or renter households) in the nation in 2016.
HH is a household. National median household income is $57,617 according to 2016
ACS IPUMS data. Thresholds of income-to-AMI are based on HUD’s “at or below”
definition. This graph design is adapted from an annual report published by the National
Low Income Housing Coalition (NLIHC) and titled "The Gap: A Shortage of Affordable
Homes" (March 2018, Figure 1 on page 4). A rental unit is defined affordable to an
income group when it costs at or below 30 percent of top income threshold of the group.
Source: 2016 ACS 1-year IPUMS file.

The disposition of the units affordable to the very low-income group is portrayed
in Figure II–5 below. Only 65.0 percent (8,373,609 units) of rental units affordable to
very low-income renters of rental units affordable to very low-income renters were
occupied by very low-income renters while the remaining units were occupied by renters
of higher income who have scavenged downward into these rental units that are in a less-
expensive price range than what they are capable of affording. In other words, what was
ideally estimated as 74 affordable low-cost rental units per 100 very low-income renters
is reduced to only 48 that are both affordable and occupied by (i.e., available to) very
low-income renters (48 = 8.4M / 17.3M × 100).
56 – Chapter II

Figure II –5. Disposition of the Rental Units that are Affordable to Very Low-income
(<50% of AMI) Renters in the United States, By Income of Actual Tenant, 2016

Notes: Universe is 12.9 million low-cost rental units that were identified affordable to
very low-income renters in the nation in 2016. National median household income is
$57,617 in 2016. A rental unit is defined affordable to an income group when it costs at
or below 30 percent of top income threshold of the group ($720.21/month = $57,617 ×
0.5 × 0.3 / 12 months).
Source: 2016 ACS 1-year IPUMS file.

Renter households of middle income could find their way into low-cost rentals in
several ways. For example, relatively better-off renters might be saving their money to
accumulate a down payment for a future home purchase or another priority need. It’s also
possible they might have a friendly connection to the landlord and so are receiving a
private-market discount. Or the middle-income renters might have better access to units
restrained by state or local housing policies (e.g., rent-stabilized units) that are lower-cost
57 – Chapter II

because these are not income-restricted. For whatever reason, middle- or higher-income
households have a lot more resources and flexibility in the housing market than very low-
income households. When the better-off renters exercise those options, units left
available for low-income renters decreases and changes over time as will be shown in the
following section.

National Trends in Very Low-income Occupancy of Affordable Housing

Based on the findings above, I examine trends in the low-income occupancy rate for the
nation as a whole. Low-income occupancy rate equals the number of very low-income
(<50% of AMI) renters occupying affordable low-cost housing divided by the total
number of very low-income renters multiplied by 100.
Figure II–6 below shows a trend line (highlighted red) that represents the low-
income occupancy rate in each survey year during a 36-year period between 1980 and
2016.
17
Note that this estimate of low-income occupancy is simply based on occupancy
itself and therefore the tenant may or may become rent-burdened.
18

17
Appendix II–A shows exact estimates in absolute count between 1980 and 2016.
18
A technical detail is noteworthy here. Note that 12.9 million units are classified affordable low-cost
because they have gross rents lower than the upper limit of low-income threshold in the nation. Technically
speaking, not all of those 12.9 million units are affordable to actual tenants in terms of 30-percent-standard.
If I consider rent-to-income ratio (rent-burden dimension) of each of those 12.9 million units, the result
shows that the would-be affordable units are not always affordable. Therefore, I can consider the 12.9
million units as conservative estimate, understating the severity of rental housing affordability.
58 – Chapter II

Figure II –6. Trend in Percent of Very Low-income (<50% of AMI) Renters that Occupy
Low-cost Affordable Housing, United States, 1980 to 2016

Notes: See Appendix II–A for exact estimates in absolute count between 1980 and 2016.
Universe is very low-income (<50% of AMI) renter households in the nation in each
survey year. National median household income is used for each survey year according to
Census and ACS IPUMS data. Thresholds of income-to-AMI are based on HUD’s “at or
below” definition. A rental unit is defined affordable to very low-income renters when it
costs at or below 30 percent of top income threshold of the group.
Sources: 1980, 1990, and 2000 Decennial Census; 2006 to 2016 American Community
Survey (ACS) 1-year Estimate Integrated Public Use Microdata (IPUMS) Sample files.

During the 1980s and 1990s, a very steady trend in the low-income occupancy
rate is found, holding between 59.7 percent and 61.7 percent. In 2000, the low-income
occupancy rate started to sharply drop as the national economy expands. It declined by
59 – Chapter II

10.3 percentage points in only 6 years from 2000 to 2006. Despite a slight bounce in
2007 and 2008, the low-income occupancy plunged again until it hit the bottom at 43.8
percent in 2011. During the recovery period (2011 to 2016), progress in low-income
occupancy continues to rise but still much lower than the steady level back in the 1980s
and 1990s. This 36-year trend clearly shows that the very low-income households’ access
to low-cost affordable housing was sharply depressed during the economic boom (2000
to 2006) and bust (2006 to 2011) periods whilst its progress in recent recovery is
sluggish.

Metropolitan Areas with the Most and Least Availability of Affordable Housing

Table II–2 shows a closer analysis of trends for the 50 largest metropolitan areas on the
basis of metropolitan-specific area median household income (see Appendix II–B for the
full 200 largest metros).
19

19
The focus on only the largest 50 areas prevents potential noise from small- and mid-sized metros. The
least availability occurs primarily in metros with fewer households, creating instability in the
measurements, and serious unavailability of housing in smaller metros draw attention away from the largest
metros. For example, rental unavailability in the most populous New York or San Francisco would have
more consequences than the extreme unavailability registered in Wausau, Wisconsin or Decatur, Alabama.
Thus, simply comparing rental unavailability in metros of such widely varying sizes could be misleading.
60 – Chapter II

Table II –2. Percent of Very Low-income (<50% of AMI) Renters that Occupy Low-cost
Affordable Housing, Ranked 1 to 15 (Least Available) and 36 to 50 (Most Available)
among the 50 Largest Metropolitan Areas, 2006, 2011, and 2016
Rank
2006 2011 2016
Metropolitan Area % Metropolitan Area % Metropolitan Area %
15 Least Available Metropolitan Areas (Lowest Occupancy)
1 Orlando 27.8 Orlando 19.3 Orlando 19.6
2 Tampa 30.0 San Diego 24.6 Miami 25.2
3 Miami 31.3 Miami 24.7 San Diego 26.4
4 Los Angeles 32.1 Las Vegas 25.0 Las Vegas 27.8
5 San Diego 32.5 Los Angeles 25.2 Tampa 27.9
6 Las Vegas 33.9 Tampa 25.4 Los Angeles 28.0
7 Riverside-SB 36.7 Riverside-SB 27.8 Riverside-SB 31.4
8 Memphis 40.2 Atlanta 30.2 Memphis 33.6
9 Atlanta 41.5 New Orleans 32.0 New Orleans 34.0
10 Sacramento 41.5 Memphis 33.9 Virginia Beach 34.1
11 Austin 45.3 Jacksonville 34.2 Atlanta 36.2
12 New York 46.8 Sacramento 35.3 Sacramento 37.0
13 SF-Oakland 47.0 Portland 36.1 Jacksonville 38.0
14 Phoenix 47.1 Milwaukee 36.5 Phoenix 39.1
15 Dallas 49.3 Virginia Beach 36.9 Denver 40.0

United States 51.4 United States 43.8 United States 46.9

15 Most Available Metropolitan Areas (Highest Occupancy)
36 Seattle 55.1 Houston 47.9 Columbus 53.3
37 Charlotte 55.7 Providence 47.9 Oklahoma City 53.4
38 Baltimore 56.3 Nashville 48.3 Baltimore 53.4
39 Raleigh 56.8 Seattle 48.9 Boston 54.3
40 Nashville 57.1 Indianapolis 49.3 Louisville 54.7
41 D.C. 58.7 Louisville 49.4 Salt Lake City 55.0
42 Indianapolis 58.7 D.C. 50.2 Hartford 55.2
43 Louisville 59.4 Denver 50.4 Providence 55.8
44 St. Louis 59.8 Kansas City 50.9 Buffalo 55.9
45 Salt Lake City 60.4 Boston 51.0 St. Louis 57.0
46 Pittsburgh 61.2 Raleigh 53.1 Kansas City 57.5
47 Kansas City 62.6 Columbus 53.9 Raleigh 58.5
48 Hartford 63.6 Cincinnati 58.1 Minneapolis 59.8
49 Cincinnati 66.0 Pittsburgh 60.6 Pittsburgh 60.1
50 Minneapolis 66.6 Minneapolis 63.4 Cincinnati 65.0
Notes: Appendix II–B shows a full list of 200 largest metropolitan areas.
Sources: 2006, 2011, and 2016 American Community Survey (ACS) 1-year IPUMS data.
61 – Chapter II

I find that the differences among large metros with regard to rental housing
availability are substantial. Among the most populous 50 metros in 2016, only one-fifth
(19.6 percent) of very low-income renters in Orlando, FL succeeded in occupying low-
cost affordable housing while in Cincinnati, OH, nearly two-third (65.0 percent) of the
very low-income households secured an affordable place to live.
Acute rental availability problems were concentrated in California and Florida
metro areas, including seven least available metros in the nation: Orlando, Miami, San
Diego, Las Vegas, Tampa, Los Angeles, and Riverside-SB, showing an average
availability percent of 26.6. In contrast, many metros in the Midwest region were often
identified as the most available housing markets in the nation with an average availability
percent of 58.7, clearly surpassing the national average experience (46.9 percent).
At the peak and bottom of the market in 2006 and 2011 respectively, a similar list
of metros as in 2016 is shown as the least available metros. However, some available
metros turned into least or less available areas during the recovery period (2011 to 2016).
For example, Denver became one of the least available areas in 2016 after experiencing
housing market expansion during the recovery (2011 to 2016). Washington D.C. was also
available place for low-income households only until 2011. Another exception is New
Orleans whose housing market was seriously impacted by Hurricane Katrina and
subsequently became sharply unavailable for the poorest renters.
Figure II–7 visualizes the geographic concentration of the least available metro
areas across the nation in 2016 on a US national map. A Metro area is represented by a
polygon that shows the geographic boundary of the area where darker red color
representing metros with more severe availability problems. Figure II–7 clearly shows
62 – Chapter II

that the incidence of rental availability problems is not only prevalent across the country
but also concentrated in some parts of the country where we would expect them to be –
along the coasts and in areas of high income. A similar geographic distribution is found
in 2006 (peak of the market) and 2011 (bottom of the market), but a greater number of
largest metros appear to experience the lowest level of rental availability (at or less than
40 percent, highlighted the darkest red on the map) in 2016 compared to preceding years
(see Appendix II–C for 2006 and 2011 maps).

Figure II –7. Metropolitan Areas with Least or Most Availability of Affordable Rental
Housing, 200 Largest Metropolitan Areas, 2016

Notes: Appendix II–C for additional maps on rental availability. Universe is very low-
income (<50% of AMI) renter households in each of the largest 200 metropolitan areas in
2016. Metro-specific area median household income (AMI) is calculated for each
metropolitan area according to 2016 ACS IPUMS data. Thresholds of income-to-AMI are
based on HUD’s “at or below” definition. A rental unit in a metro is defined affordable to
very low-income renters when it costs at or below 30 percent of top income threshold of
the group in the same metro area.
Sources: U.S. Census Bureau, TIGER/LINE files; 2016 ACS 1-year IPUMS data.
63 – Chapter II

Table II –3. Change in Percent of Very Low-income (<50% of AMI) Renters that Occupy
Affordable Housing, Ranked 1 to 15 (Biggest Increase) and 36 to 50 (Biggest Decrease)
among the 50 Largest Metropolitan Areas, 2006 to 2011 and 2011 to 2016
Rank
2006 to 2011 2011 to 2016
Metropolitan Area % Metropolitan Area %
15 Biggest Increase (or Smallest Decrease) in Rental Availability
1 New Orleans -17.8 Denver -10.4
2 Jacksonville -17.3 Washington, D.C. -4.9
3 Salt Lake City -17.2 Houston -4.6
4 Hartford -15.8 Minneapolis -3.6
5 Portland -14.6 Virginia Beach -2.8
6 Charlotte -14.2 Dallas -2.7
7 St. Louis -13.4 San Antonio -1.6
8 Milwaukee -13.1 Richmond -1.2
9 Virginia Beach -12.6 Columbus -0.5
10 Baltimore -12.2 Pittsburgh -0.5
11 Kansas City, -11.6 Seattle -0.3
12 Atlanta -11.3 Memphis -0.3
13 San Jose -11.0 Indianapolis 0.0
14 Louisville -10.0 Orlando 0.3
15 Indianapolis -9.3 Philadelphia 0.4

United States -7.6 United States 3.2

15 Biggest Decrease (or Smallest Increase) in Rental Availability
36 Memphis -6.3 Atlanta 6.0
37 Sacramento -6.2 Kansas City 6.6
38 Seattle -6.2 Birmingham 6.8
39 Providence -5.6 Cincinnati 6.8
40 Buffalo -4.9 Oklahoma City 6.9
41 Tampa -4.6 Hartford 7.4
42 Oklahoma City -4.4 Providence 7.8
43 Houston -3.9 Buffalo 9.1
44 Raleigh -3.7 Cleveland 9.3
45 Denver -3.5 Baltimore 9.4
46 Minneapolis -3.1 Charlotte 9.9
47 Dallas -1.9 St. Louis 10.6
48 Pittsburgh -0.6 SF-Oakland 10.9
49 Boston -0.4 Salt Lake City 11.8
50 Columbus 0.7 Milwaukee 13.8
Notes: Appendix II–B shows full list of 200 largest metropolitan areas.
Sources: 2006, 2011, and 2016 American Community Survey (ACS) 1-year IPUMS data.
64 – Chapter II

Despite an overall consistency with regard to the rank of least available metros as
shown in Table II–2 above, rental housing availability changed in some places more
sharply than in others. As shown in Table II–3 below, rental availability declined across
all of the 50 largest metro areas in the bust period (2006 to 2011), except Columbus, OH
(gain of 0.7 percentage point).
Conversely, the availability of low-cost housing expanded across many largest
metros (fully 38 of 50 metros) in the recovery period (2011 to 2016). The reversing trend
of rental availability is also prevalent in the largest 200 metropolitan areas (see Appendix
II–B) most likely due to progress of recovery across the nation, which is controlled in the
empirical analysis of this essay.

3. Data and Methods

The national average of rental housing availability among the very low-income (<50% of
the national median household income) renters was 46.9 percent in 2016. However,
housing access by the very low-income household is substantially greater in some
metropolitan areas than in others. Widening differences across metropolitan areas are
caused by many regional specific factors such as job growth, changes in rent,
homeownership changes, new construction, and government assistance. These create a
volatile context for local chances of expanding low-income access to affordable housing.
The remainder of this essay focuses on how to explain reasons causing the metropolitan
differences in low-income housing access.

65 – Chapter II

Data

This essay collects data from several sources, most of which are derived from the U.S.
Census Bureau’s American Community Survey (ACS) 1 percent Public Use Microdata
Sample. The Integrated Public Use Microdata Series (IPUMS) data is provided by the
Minnesota Population Center (Ruggles et al., 2018). The ACS, an annual nationwide
survey of approximately 3.5 million households, provides timely data on the social,
economic, demographic, and housing characteristics of the entire U.S. population.
IPUMS contains individual ACS questionnaire records for a subsample of housing units
and households. Aggregating the detailed information of individual units and households,
this essay involves an aggregate-level analysis of metros and uses data from 2006 to 2016
because 2006 is the first year that the ACS gathered detailed data.
The primary analysis relies on the Microdata Samples due to certain advantages
of the Microdata Samples over summary tables that are tabulated and provided by the
U.S. Census Bureau through the American FactFinder website. First, the Microdata
Samples specify all the detailed demographic and housing data required for the study.
The aggregated format of summary tables from the American FacFinder is not sufficient
to identify rent, income, and rent burden
20
of individual households, which is crucial for

20
Small technical differences exist in how the 30-percent threshold is applied. Although the HUD
definition of excess cost burden is “greater than” 30 percent of income spent on housing, the published data
by the Census Bureau reports data that are “at or above” 30 percent. Analysis based on the “greater than”
standard yields a slightly lower incidence of rent burden than that derived from the published Census data.
A second difference concerns how to handle the category of renters for which complete data are not
66 – Chapter II

the entire analysis. Second, a metropolitan identifier within the Microdata Samples
enables utilization of the consistent and comparable geographic definitions in multiple
ACS years to measure low-income housing access. The geographic identifier will be
explained in the following section.
21

These ACS data are then supplemented by contextual data for each metropolitan
area. Government assistance data are taken from the Picture of Subsidized Households
database provided by HUD (HUD, 2018) to capture the prevalence of housing assistance

available. The most common approach ignores households with missing data and calculates the share of
renters paying excessive rent among cases with complete data only. By contrast, the Harvard JCHS
allocates the cases with missing data into two different rent burden groups. Units paying no cash rent
(roughly one-third of the “not computed” subgroup) are assigned to the no-burden group, whereas zero-or-
negative-income units (roughly two-thirds of the “not computed” subgroup) are assigned to the 50-percent+
burden group. This inclusive approach has the advantage of using available data to count all renters in the
nation. That has the effect of slightly raising the incidence of rent burden compared with when the not-
computed cases are excluded. Throughout this essay, I follow the Harvard JCHS’s approach to using the
“not computed” subgroup and also the “greater than” treatment of the 30-percent and 50-percent thresholds,
both of which differ from common analysis with the Census Bureau data.
21
A certain limitation to the use of the Microdata Samples should be noted. The smaller the geographic
area, the greater the deviation that may occur between the ACS summary table and the result of tabulations
from microdata samples. That is the result of the greater sampling error in the public use microdata than
that of the data files the Census Bureau uses to produce their summary tables. As an accuracy check on my
microdata-based household count, I compare microdata estimates of all renter-occupant households with
corresponding American FactFinder’s summary tables in each of multiyear ACSs. Trends in the two
indicators tracked very closely in both the nation as a whole and across the 200 largest metropolitan areas
in aggregated terms.
67 – Chapter II

programs in each metropolitan area. This is a database of reports from local housing
authorities to HUD on the number of households (or housing units) assisted by the federal
funds in their jurisdiction. This database is regularly updated and includes data across the
country since 2000, at multiple levels of geography as small as the census tract and as
large as the nation as a whole. Employment trends are taken from the Current
Employment Statistics (CES) Database provided by the U.S. Bureau of Labor Statistics
(BLS). Housing construction trends are taken from the U.S. Census Bureau’s Building
Permits Survey and C40 reports.

Sample

My sample is the most populous 200 metropolitan statistical areas in terms of 2016
population. This essay is an aggregate-level analysis of metropolitan areas, rather than an
individual-household-level analysis. A metropolitan area is a region consisting of a large
urban core together with surrounding communities that have a high degree of economic
and social integration with the urban core. The areas comprise one or more counties; New
England, however, consists of combinations of townships. In 2016, the sample of 200
metros contain 243 million people (75.2 percent of the U.S. population) and 88 million
households (74.2 percent of U.S. households).
Boundaries for the 200 metropolitan areas are specified in accordance with the
geographical definitions used in the 2010 census. The IPUMS database provides a
geographic variable, labeled MET2013, which identifies time-invariant and comparable
areas of residence using 2013 definitions for metropolitan statistical areas (MSAs) from
68 – Chapter II

the U.S. Office of Management and Budget (OMB). The 2013 MSAs are the first to be
based on 2010 standards and 2010 census data. Data from the 2000 census and the ACS
in 2006 and later were rearranged by the Minnesota Population Center to conform to
those 2013 definitions. The 2013 delineations used by MET2013 are entirely county-
based, even in New England. The MET2013 variable is available for 2000 and later
years. Use of MET2013 is essential to identify both gross rent of an individual rental unit
and income of its current occupant in each metropolitan area. Using the geographic
identifier, I conduct separate analyses of alternative low-income groups and occupancies
within the geographic sample.

Classification of Households and Housing Units

This essay measures rental availability in the 200 largest metropolitan areas through a
common approach that is used by scholars and practitioners who study a range of rental
housing problems in various geographies, including neighborhood, municipality, county,
metropolitan area, state, and the nation as a whole (Collison, 2016; Herbert et al., 2018;
JCHS, 2018; Joice, 2016; NLIHC, 2019; Urban Institute, 2017). This Census/ACS-based
method
22
classifies all households and housing units in an area by their income and rents
to examine matches between income groups and rental brackets.
ACS microdata provide information about individual rental housing units and
renter households, including the monthly gross rent and annual household income of

22
For the purpose of this essay, I term it housing-household-classification method since there is not a
specific name of the method.
69 – Chapter II

actual tenant. This full set of variables enables the measurement of the availability of
affordable low-cost housing for low-income households under two alternative income
definitions: a household income of 30 percent or less of the metro-specific area median
household income (AMI) and a household income of 50 percent or less of the AMI. This
essay focuses on the very low-income (<50% of AMI) group specifically because they
have long been the federal standard for identifying eligibility for various rental subsidy
programs (Schwartz, 2015; HUD, 2017).
For each housing market (metropolitan area), I calculate low-income access to
affordable housing. I begin by categorizing renter households in a metro area by their
incomes relative to their metro-specific area median household income (AMI) into
extremely low-income (ELI, 0-30 percent of the AMI), very low-income (VLI, 30-50
percent), low-income (LI, 50-80 percent), middle-income (MI, 80-120 percent), and high-
income (HI, above 120 percent), all based on the federal standard (HUD, 2017).
Separately from households, I categorize rental housing units that exist in each metro
according to the gross rents (i.e., contract rent plus utilities) that households in the same
metro need to pay without spending more than 30 percent of their income, a common
benchmark for measuring the housing cost-burden (HUD, 2017; JCHS, 2017). To
account for total rental stock in each metro, I account all renter-occupied housing units
either with or without a complete kitchen or plumbing facilities. Vacant housing units
thus were not accounted in my analysis. It is noteworthy that the categorization of
housing units takes place without regard to the incomes of actual tenant, indicating
separate categorization of housing units and households.
70 – Chapter II

Following the categorizations, I match housing units with households by counting
the extent to which households in each income category occupied housing units
categorized as affordable for that income level or other income levels. From the housing
unit perspective, for example, 0-30 percent of AMI households (equal to ELI
households), the process of matching results in the number of rental housing units
affordable to the 0-30 percent AMI renters and occupied by them or by other higher-
income groups. In other words, the estimate represents the absolute count of 0-30 percent
AMI households occupying rental housing units that are affordable to them or other
higher-income groups. Multiplying five tenant income categories by five affordable rent
categories, this essay derives 25 household-housing unit categories.
Dependent variable of this essay, low-income housing access, is derived from the
number of 0-50 percent AMI households who occupy affordable housing (numerator)
divided by the total number of 0-50 percent AMI households (denominator) multiplied by
100, and take the following form:

Availability of affordable housing to 0-50 percent AMI renters in metro i in year j =
number of 0-50 percent AMI renters who occupy affordable unit in metro i in year j /
total number of 0-50 percent AMI renters in metro i in year j × 100

The same calculation is replicated for the 0-30 percent AMI households as a
narrower definition of low-income households. These measures of rental availability
were regressed by new construction, government assistance, and other metropolitan
contextual factors. Definitions of all variables used to model low-income housing access
are given in Table II–4.
71 – Chapter II

Table II –4. Definition of Variables
Variable Description
Dependent variables

Very Low-income (VLI,
<50% of AMI) Rental
Availability
Number of 0-50 percent AMI households who occupy
affordable housing divided by the total number of 0-50
percent AMI households multiplied by 100 in each metro
Extremely Low-income
(ELI, <30% of AMI)
Rental Availability
Number of 0-30 percent AMI households who occupy
affordable housing divided by the total number of 0-30
percent AMI households multiplied by 100 in each metro
Independent variables

Federal Rental
Assistance per 1,000 HH
Absolute number of housing units subsidized by three
major federal programs (Public Housing, Voucher, and
LIHTC) in each metro area, per 1,000 households
Public Housing per HH Absolute number of public housing units in each metro
area, per household
Voucher per HH Absolute number of rental voucher households in each
metro area, per household
LIHTC per HH Absolute number of LIHTC rental units in each metro
area, per household

Construction Per HH Absolute number of total building permits (summed for
the last 1, 3, or 5 years) divided by the total number of
households in base year in each metro area, per HH

Percent NH-White Non-Hispanic White population divided by the total
population in each metro area
Percent BA+ Bachelor’s degree or above population divided by the
total population in each metro area
Ln(Employment) Natural logarithm of the total employment in each metro
Ln(Median HH Income) Natural logarithm of Median household income in each
metro area, inflation-adjusted to 2016$ (U.S. dollar)
Ln(Median Gross Rent) Natural logarithm of Median gross rent (including
contract rent and utilities) in each metro area, inflation-
adjusted to 2016$ (U.S. dollar)

Recovery 2012-2016 recovery period dummy (yes = 1, no = 0)
Notes: Census region dummy variables (Northeast as reference) were used only in the
cross-sectional models. Inflation is adjusted by BLS’s CPI-U all items for urban
consumers, AMI = Metro-specific area median household income, VLI = Very Low-
income (<50% of AMI), ELI = Extremely Low-income (<30% of AMI), HH =
Household, LIHTC = Low-Income Housing Tax Credit, NH-White = Non-Hispanic
White, BA+ = Bachelor’s degree or above, Gross rent is sum of contract rent and utilities.
72 – Chapter II

Model Specification

I estimate the level of housing availability for low-income renters in relation to the scale
of federal housing subsidies and of market-rate constructions, and characteristics of the
metropolitan areas in which they reside. The model to be described is associational and
cannot text causality. I build a set of cross-sectional models that identify the housing
market and the demographic characteristics of metros most strongly associated with the
scale of rental availability. The same models, tested in various market conditions,
including the peak of the housing market (2006), the bottom of the market (2011), and
the most recent year of the recovery period (2016), are ordinary least square models:

𝑅𝑒𝑛𝑡𝑎𝑙 𝐴𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦
i
𝑗 = 𝛽 0
+ 𝛽 1
𝑃𝑈𝐵𝐿𝐼𝐶
i
𝑗 + 𝛽 2
𝑉𝑂𝑈𝐶𝐻𝐸𝑅
i
𝑗 + 𝛽 3
𝐿𝐼𝐻𝑇𝐶
i
𝑗 +
𝛽 4
𝑀𝐴𝑅𝐾𝐸𝑇
i
𝑗 + 𝛽 5
𝐷𝐸𝑀𝑂𝐺
i
𝑗 + 𝛽 6
𝑅𝐸𝐺𝐼𝑂𝑁 𝑖 + 𝑒
𝑖 𝑗 (1)

where Rental Availability is the percent of 0-50 percent AMI renters who occupy low-
cost affordable housing in metro i in year j, as described above. PUBLIC, VOUCHER,
and LIHTC refer to public housing units, housing choice voucher households, and low-
income housing tax credit units, respectively, expressed as the proportion of all
households in each metro. These data come from the HUD’s Picture of Subsidized
Households, as described in the previous section.
I add a vector of housing market characteristics (MARKET) shown to affect
housing affordability and availability in a variety of ways, among them changes in
housing supply and demand. A proportion of housing demand relates to employment
73 – Chapter II

growth, data available from the U.S. Bureau of Labor Statistics. When new construction
lags behind employment growth, housing shortages elevate rents and create fierce
competition for higher-end rental brackets, thereby narrowing the availability of low-cost
affordable housing to the poorest renters. New construction data come from the U.S.
Census Bureau building permits survey. An additional element affecting metropolitan
rental demand is median household income and median gross rents.
Demographic characteristics (DEMOG) also play an important role in how renters
with different demographic profiles respond to market conditions. For example, the
percentage of non-Hispanic Whites or those with a bachelor’s degree capture racial and
educational differences. The natural logarithm of the total population controls the overall
effect of population size. To identify regional variation in rental unavailability, cross-
sectional models also include the region of the metro (REGION) as a control variable.
Cross-sectional models may not be capable of controlling for historical factors
that lead to both greater housing supply, either market-rate or government-subsidized,
and greater rental availability (Cameron and Trivedi, 2010; Wooldridge, 2010). For
example, old and low-cost housing is disproportionately located in older metros such as
New York and Chicago, which typically have a greater concentration of high-paying
jobs. Middle- and higher-income households are also concentrated in larger metros, often
accompanied by a greater number of subsidized housing that expands low-income
housing opportunities. Furthermore, the presence of subsidies in a metro could even
attract additional low-income households from neighboring jurisdictions.
In a determination of whether the observed associations in the cross-sectional
models are spurious, the above factors are important. To the extent that low-income
74 – Chapter II

households move to metros with greater government subsidy availability (or leave metros
with less subsidy availability), this would create a two-way relationship between
government assistance and rental availability. Thus, a cross-sectional model used in this
essay would more likely – all else being equal – find a positive relationship between
housing subsidy prevalence and low-income rental availability. In other words, I would
be more likely to conclude that government assistance makes housing more available to
low-income renters.
Annual ACS data can limit the extent to which the cross-sectional models are
problematic. Pooled annual ACS data across the same set of metropolitan areas allow to
fully control for the time-invariant regional differences as well as year-specific effects.
Thus, the fixed-effects panel model is adopted as the default specification as will be
described in the equation (2) below.
In addition, there is little reason to believe that low-income households could
anticipate annual changes in government subsidies and low-cost rental availability
sufficiently fast to make a move that would show up in the data. Low-income households,
however, may be more likely to respond to rental availability by overcrowding, doubling
up, or remaining in parents’ homes (Skobba and Goetz, 2015; Myers et al., 2016). To the
extent that they occur, such responses would underscore the important role that housing
subsidies play; that is, rental hardship leads to increased occupancy of low-cost housing
by higher-income households, which then reduces low-income occupancy. Furthermore,
despite localities’ limited resources for increasing housing subsidies in response to
housing market conditions on an annual basis, they could influence the federal
government (HUD) to provide additional funds or subsidies over a longer time horizon.
75 – Chapter II

Given these historical factors, which may lead to spurious correlations,
longitudinal data allow me to better isolate the effects of government assistance and
market-rate construction on rental availability. Accordingly, this essay constructs a panel
data of 11 consecutive years from 2006 to 2016 across the most populous 200
metropolitan areas. Controlling for both year-fixed and metro-fixed effects, the default
panel model fully absorbs time-invariant regional differences and metro-invariant year-
specific effects on rental housing availability. This model specification is based on the
assumption that a set of annual data is appropriate to track the volatile trend in rental
availability during and after the Great Recession (as shown in Figure II–6 above). I will
test the validity of this assumption in the result section that includes alternative tests, such
as RECOVERY dummy variable (if 2006 through 2011, RECOVERY = 0; if 2012 through
2016, RECOVERY = 1) instead of year-fixed effects and bi-annual data structures. Note
that the fixed-effects panel model omits the regional dummy variable (REGION) which is
still controlled and fully absorbed by the metro-fixed effects. Instead, it runs a set of
fixed-effects panel regression models that control for the time-invariant, either observed
or unobserved, characteristics of metros in addition to the above housing market and
demographic variables. Given the rental housing availability differs across metro areas
and varies over time as shown in the previous sections, the essay adopts the fixed-effects
estimator without testing a random-effects estimator (Cameron and Trivedi, 2010;
Wooldridge, 2010).

76 – Chapter II

The fixed-effects panel model to be estimated can be expressed as:

𝑅𝑒𝑛𝑡𝑎𝑙 𝐴𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦
i
𝑗 = 𝛽 0
+ 𝛽 1
𝑃𝑈𝐵𝐿𝐼𝐶
i
𝑗 + 𝛽 2
𝑉𝑂𝑈𝐶𝐻𝐸𝑅
i
𝑗 + 𝛽 3
𝐿𝐼𝐻𝑇𝐶
i
𝑗 +
𝛽 4
𝑀𝐴𝑅𝐾𝐸𝑇
i
𝑗 + 𝛽 5
𝐷𝐸𝑀𝑂𝐺
i
𝑗 + 𝑚
𝑖 + 𝑦
𝑗 + 𝑒
𝑖 𝑗 (2)

where 𝑚
𝑖 is metro-specific fixed effects and 𝑦
𝑗 is year-specific fixed effects. Fixed-
effects model based on the pooled annual ACS data would increase the confidence of this
study, that is, that observed effects are due to the impact of market-rate construction and
housing subsidies on the availability of low-cost affordable rentals, and not the other way
around. This model offers a more robust description of the relationship between
government assistance, new construction, and low-income rental availability.

4. Results

Descriptive Statistics of Variables

Table II–5 displays descriptive statistics for all variables for the 200 largest metropolitan
areas in 2016. On average, 43.9 percent of very low-income (<50% of AMI) renters
occupy affordable low-cost housing. In other words, more than half of the poorest renters
in the largest metros are occupying housing that they cannot afford. Given that the federal
government subsidizes roughly one fourth of all the 0-50 percent AMI renters in any type
of federal rental subsidy, I can infer that only one-fifth of all the 0-50 percent AMI
renters occupies affordable housing that are provided from the market, which is
confirmed by previous studies (JCHS, 2017; Weicher et al., 2017).
77 – Chapter II

Table II –5. Descriptive Statistics for the 200 Largest Metropolitan Areas, 2016

Variable N Mean S.D. Min Max
Dependent variables

VLI Rental Availability 200 43.86 11.13 10.81 76.39
ELI Rental Availability 200 25.86 8.97 3.22 52.23

Independent variables

Gov. Subsidy per 1,000 HH 200 826.54 2,108.86 35.03 26,392.24
Public Housing per HH 200 136.48 680.34 0.00 9,554.06
Voucher per HH 200 338.38 834.15 4.51 9,961.72
LIHTC per HH 200 351.68 705.67 3.40 6,876.46

Construction per HH 200 1.02 0.63 0.10 3.13

Percent NH-White 200 0.69 0.18 0.04 0.97
Percent BA+ 200 17.63 5.37 5.98 34.70
Ln(Employment) 200 12.47 1.13 10.51 16.07
Median HH Income 200 55,936 11,254 35,200 110,000
Median Gross Rent 200 922 227 510 2,050

Northeast 200 0.17 0.37 0.00 1.00
Midwest 200 0.20 0.40 0.00 1.00
South 200 0.38 0.49 0.00 1.00
West 200 0.26 0.44 0.00 1.00

Notes: Log form of median household income and median gross rent was used in the
following regression models. Inflation is adjusted by BLS’s CPI-U all items for urban
consumers, AMI = Metro-specific area median household income, VLI = Very Low-
income (<50% of AMI), ELI = Extremely Low-income (<30% of AMI), HH =
Household, LIHTC = Low-Income Housing Tax Credit, NH-White = Non-Hispanic
White, BA+ = Bachelor’s degree or above, Gross rent is sum of contract rent and utilities.

In the case of extremely low-income renters (<30% of AMI), much smaller share
(25.9 percent) of the lowest-income households are found to occupy affordable housing,
the vast majority of them should be subsidized by federal, state, and local governments.
The average metro contains 826 thousand housing units that are assisted by three major
federal programs – Public Housing, Voucher, and LIHTC.
78 – Chapter II

Regression Results

I report largely two sets of model estimation results with specific attention to the
associations between housing subsidy, new construction, and rental availability. First, I
present estimations of the cross-sectional models built for different market conditions,
including 2006 (peak of the market), 2011 (bottom of the market), and 2016 (most recent
year in recovery). Secondly, I present a panel regression result which differs from that of
cross-sectional models in meaningful ways. The results from either cross-sectional or
panel model are associational and cannot be treated as causal findings.

Cross-sectional Regression Results
Table II–6 below contains a set of cross-sectional regression results that explain the
substantial variation in rental availability across the largest 200 metros in 2006, 2011, and
2016 respectively. Two key factors that I examine to explain these variations of rental
availability are government subsidies and market-rate constructions.
The 2006 results show a positive association between government assistance,
particularly Public Housing, and rental availability for the very low-income renters (VLI
model 2) as well as the extremely low-income group (ELI model 3 and 4). In ELI model
3, the coefficient of 0.067 means that more government subsidies – 100 additional
government-subsidized housing units per 1,000 household – are positively associated
with an increase in the percent of extremely low-income households that occupy
affordable housing by 6.7 percentage points. Once divided into three major programs
such as Public Housing, Voucher, and LIHTC, however, the subsidy effect is only
significant in the case of Public Housing (VLI model 2 and ELI model 4).
79 – Chapter II

As for the relation between market-rate new construction and rental availability
for low-income households, I find a negative association between new construction and
low-income access to low-cost affordable housing in both income groups (VLI and ELI
model 1). This result suggests that, in the cross-sectional snapshot, low-income housing
access is likely to be lower in metropolitan areas with a greater new construction. Given
that new market-rate housing is generally not occupied by either very low-income or
extremely low-income renters, the negative association could be evidence suggesting
metro-specific historical factors that may associate with both greater new constructions
and high-paying jobs (or fewer new constructions and low-wage jobs). Therefore, the
negative association may not robust – it is observed only in 2006 and 2016 cross-
sectional models – and even reversed in the models using panel data (see the following
section for Panel Regression Results).
Other variables that appear to have a negative association with low-income rental
availability include log median gross rent, the percent of population with BA or higher
degree (implying a greater housing demand from highly-educated people), and living in
the West and Midwest regions. Variables that have a positive association with rental
availability are the homeownership rate (potentially reducing competition for rental
housing) and log metro-specific area median household income (implying an overall
increase in the capacity of housing consumption).
In 2011 and 2016, the observed relationships were consistent with those from the
2006 models. Several results, however, deviate from this pattern. The coefficient and
significance of the government subsidy effect is weaker in 2011 and much stronger in
2006 than 2016 for extremely low-income households, suggesting a much greater
80 – Chapter II

demand for subsidy than supply of government subsidies at the bottom of the housing
market. New construction effects appear to play a mixed role in expanding and reducing
the low-income rental availability. Least availability of affordable housing is found in the
West region in 2011 (ELI model 3), implying many metros in the coastal areas were
hardly hit by the Great Recession; however, this regional disparity is not observed in
2016.
In sum, a series of cross-sectional models confirm a positive relationship between
housing subsidies and low-income rental availability, but the effect is only significant for
the Public Housing program while insignificant for Voucher and LIHTC programs. New
housing construction, however, appears to have a mixed and inconsistent relationship
with rental availability, which is most likely due to potential biases uncontrolled in the
cross-sectional models.
81 – Chapter II

Table II –6. 2006 Cross-sectional Regression Results, 200 Largest Metropolitan Areas

Notes: Robust standard errors are used to account for heteroskedasticity. Standard errors are clustered by metropolitan area. + = p <
0.10, * = p < 0.05, ** = p < 0.01, *** = p < 0.001.

Coef. Sig. S.E. Coef. Sig. S.E. Coef. Sig. S.E. Coef. Sig. S.E.
Gov. Subsidies per 1,000 HH 0.027 (0.019) 0.067 * (0.032)
Public Housing per HH 2.111 * (0.907) 5.872 *** (1.337)
Voucher per HH -0.548 (0.569) -0.536 (0.822)
LIHTC per HH -0.323 (0.702) -0.379 (0.787)
Construction per HH -0.593 * (0.300) -0.566 + (0.326) -0.831 + (0.461) -0.647 (0.493)
% NH–White -6.281 * (3.025) -5.664 * (2.867) -3.013 (4.355) 0.273 (4.291)
% BA+ -0.059 (0.116) -0.049 (0.117) -0.345 + (0.185) -0.332 + (0.185)
per capita Homeownership 0.348 *** (0.103) 0.340 *** (0.099) 0.392 ** (0.144) 0.395 ** (0.141)
Median Gross Rent -66.560 *** (4.447) -65.100 *** (4.477) -5.312 (6.281) -1.879 (6.666)
Median HH Income 67.550 *** (5.308) 67.540 *** (5.503) 17.020 * (8.183) 17.840 * (8.456)
Ln(Employment) -0.311 (0.427) -0.288 (0.408) -0.496 (0.745) -0.427 (0.661)
Northeast (Ref.)
Midwest -4.009 *** (1.203) -3.102 * (1.329) -4.304 * (2.035) -1.854 (2.160)
South -1.762 (1.317) -1.086 (1.360) -1.234 (2.163) 0.908 (2.165)
West -0.660 (1.283) 0.965 (1.422) -5.485 * (2.222) -1.068 (2.346)
Constant -78.470 *** -11.03 -83.260 *** (12.17) -35.260 + (19.53) -55.790 ** (21.430)
Observations 200 200 200 200
Adj. R-Squared 0.761 0.766 0.196 0.269
Model (3) Model (4)
Very Low-income (<50% of AMI) Extremely Low-income (<30% of AMI)
Model (1) Model (2)
82 – Chapter II

(Continued) Table II –6. 2011 Cross-sectional Regression Results, 200 Largest Metropolitan Areas

Notes: Robust standard errors are used to account for heteroskedasticity. Standard errors are clustered by metropolitan area. + = p <
0.10, * = p < 0.05, ** = p < 0.01, *** = p < 0.001.

Coef. Sig. S.E. Coef. Sig. S.E. Coef. Sig. S.E. Coef. Sig. S.E.
Gov. Subsidies per 1,000 HH 0.024 (0.016) 0.045 * (0.021)
Public Housing per HH 3.044 *** (0.859) 3.271 ** (1.151)
Voucher per HH -0.492 (0.539) 1.022 (0.691)
LIHTC per HH -0.246 (0.487) 0.502 (0.610)
Construction per HH 1.526 (1.084) 1.826 + (1.006) -2.121 (1.479) -1.458 (1.506)
% NH–White -8.228 * (3.693) -6.452 + (3.531) -6.327 (4.739) -1.476 (4.634)
% BA+ -0.027 (0.115) -0.027 (0.112) -0.155 (0.157) -0.178 (0.147)
per capita Homeownership 0.270 ** (0.097) 0.287 ** (0.094) 0.377 ** (0.119) 0.460 *** (0.118)
Ln(Median Gross Rent) -58.570 *** (4.106) -55.590 *** (4.158) -5.463 (5.854) -1.139 (5.892)
Ln(Median HH Income) 64.870 *** (4.381) 64.380 *** (4.441) 18.810 ** (6.169) 17.200 ** (6.294)
Ln(Employment) -0.525 (0.477) -0.515 (0.421) -1.189 + (0.663) -1.024 + (0.592)
Northeast (Ref.)
Midwest -1.045 (1.358) 0.494 (1.446) -4.146 * (1.876) -2.560 (1.963)
South -2.991 * (1.352) -1.681 (1.441) -2.759 (2.055) -1.118 (1.984)
West -1.788 (1.470) 0.997 (1.691) -6.613 *** (1.819) -3.401 (2.161)
Constant -80.990 *** (12.920) -91.060 *** (12.79) -34.410 * (15.340) -55.870 *** (14.220)
Observations 200 200 200 200
Adj. R-Squared 0.724 0.740 0.255 0.310
Model (3) Model (4)
Very Low-income (<50% of AMI) Extremely Low-income (<30% of AMI)
Model (1) Model (2)
83 – Chapter II

(Continued) Table II –6. 2016 Cross-sectional Regression Results, 200 Largest Metropolitan Areas

Notes: Robust standard errors are used to account for heteroskedasticity. Standard errors are clustered by metropolitan area. + = p <
0.10, * = p < 0.05, ** = p < 0.01, *** = p < 0.001.
Coef. Sig. S.E. Coef. Sig. S.E. Coef. Sig. S.E. Coef. Sig. S.E.
Gov. Subsidies per 1,000 HH 0.021 (0.017) 0.069 ** (0.022)
Public Housing per HH 2.373 * (0.984) 5.201 *** (1.123)
Voucher per HH 0.441 (0.518) 0.277 (0.599)
LIHTC per HH 0.393 (0.623) 0.707 (0.543)
Construction per HH -1.739 * (0.698) -1.139 (0.750) -3.679 *** (0.826) -2.699 ** (0.894)
% NH–White -8.359 * (3.546) -5.932 + (3.130) -7.048 + (3.648) -2.659 (3.469)
% BA+ 0.097 (0.122) 0.089 (0.118) -0.277 + (0.151) -0.289 * (0.143)
per capita Homeownership 0.303 ** (0.100) 0.334 ** (0.102) 0.479 *** (0.107) 0.512 *** (0.106)
Ln(Median Gross Rent) -65.600 *** (4.492) -62.610 *** (4.435) -16.130 ** (5.516) -9.616 + (5.655)
Ln(Median HH Income) 70.820 *** (5.270) 69.700 *** (5.316) 34.330 *** (6.122) 31.790 *** (6.058)
Ln(Employment) 0.397 (0.518) 0.366 (0.456) -0.245 (0.607) -0.267 (0.526)
Northeast (Ref.)
Midwest -1.026 (1.550) 0.145 (1.707) -4.091 * (1.925) -1.833 (1.968)
South 0.061 (1.680) 0.919 (1.685) 0.687 (2.261) 2.237 (2.070)
West 1.156 (1.721) 2.714 (1.766) -3.004 (2.049) 0.086 (2.103)
Constant -102.400 *** (12.310) -113.000 *** (12.550) -85.970 *** (14.760) -104.300 *** (14.810)
Observations 200 200 200 200
Adj. R-Squared 0.721 0.732 0.371 0.443
Model (3) Model (4)
Very Low-income (<50% of AMI) Extremely Low-income (<30% of AMI)
Model (1) Model (2)
84 – Chapter II

Panel Regression Results
Panel regression model provides stronger controls than the cross-sectional regression
model for metro-specific characteristics such as historical factors that lead to greater
amount of old and low-cost housing as well as greater concentration of high-paying jobs.
These unique characteristics usually do not vary over time but maybe closely related to
low-income rental availability and therefore are ideal to be controlled.
Table II–7 below shows these panel regression results, which are different in
important ways from the previous cross-sectional results. New construction effect, which
was largely mixed and even insignificant in cross-sectional models, finally appears to
have a strongly positive association with rental availability for extremely low-income
households (ELI models 3 and 4). This positive effect of market-rate housing
construction supports my hypothesis that an overall increase in market-rate housing
supply could ease rental competition in the middle or higher-end brackets and expand the
availability of low-cost housing to the poorest renters. For the extremely low-income
group in model 4, the estimation of 0.726 implies that an additional building permit per
100 households in a metro area in the base year is positively associated with a 0.726
percentage point increase in rental availability for very low-income households, holding
remaining regressors constant. The positive construction effect was consistent and
somewhat smaller in the ELI model 3 but insignificant in VLI models 1 and 2.
Government subsidy effect appears to have a positive association with greater
occupancy of affordable rental housing for very low-income households, particularly for
LIHTC program (VLI models 1 and 2). It is noteworthy that the government subsidy
effect is significant even when market-rate construction is controlled. However, the
85 – Chapter II

positive subsidy effect is no longer significant and even reversed in the extremely low-
income models. Voucher program appears to reduce extremely low-income housing
occupancy (ELI model 4), which is counter-intuitive but has been found similarly in
some previous studies (Lens, 2017; Sinai & Waldfogel, 2005; Malpezzi & Vandell,
2002). The rationale behind the negative or insignificant effect of government subsidies
on low-income housing opportunities is that the subsidies could substitute or even crowd
out market-rate housing construction and consequently reduce overall housing access.
The negative association between vouchers and extremely low-income rental availability
(ELI model 4) suggests that vouchers may raise the overall level of rent for the
unsubsidized households and subsequently reduce access to affordable housing at the
metropolitan level as reported in previous literature (Eriksen and Ross, 2015; Susin,
2002).
The other variables largely confirm cross-sectional results. The per capita
homeownership rate is positively associated with rental availability in the very low-
income models, indicating that more potential homebuyers, presumably higher-income
renters, would move into the owner market leaving greater amount of available rental
units to low-income households. As in the cross-sectional models, the median rent is
negatively associated with rental availability because a higher level of rents may limit
low-income access to the rental market while potentially increasing the doubling-up and
non-household population. In the opposite direction, metro-specific area median
household income appears to be positively associated with rental availability, suggesting
an overall increase in housing consumption capacity among households. Evidence on
employment growth shows a positive association with rental availability in the extremely
86 – Chapter II

low-income models, which contrasts with cross-sectional findings on negative
associations. One possible interpretation is that employment changes in the largest
metropolitan areas have been mainly led by low-paying jobs between 2006 and 2016
(mostly decrease in the bust while rapid increase in the recovery), capturing increasing or
declining rental demand from low-income households.
Across the four sets of panel models, the most consistent association is found
between more market-rate new construction and greater rental availability (VLI and ELI
model 1) which contrasts to negative associations found in the previous cross-sectional
models. Metro-specific historical biases that may associate with both greater new
constructions and high-paying jobs (or less new constructions and low-wage jobs) were
controlled by metro-fixed effects in the panel regression approach. The year-fixed effects
in the panel model also addressed volatile market conditions across bust, turnaround, and
recovery periods.
Housing subsidy effect also appears to play an important role in aggregated
perspective (VLI model 1). LIHTC program is shown particularly effective to expand
very low-income housing opportunities (VLI model 2). When it comes to the extremely
low-income households, however, Voucher program has a negative association with
rental availability for extremely low-income group, which is a consistent finding on side-
effects on unsubsidized households and overall rent increases (Eriksen and Ross, 2015;
Lens, 2017; Malpezzi & Vandell, 2002; Sinai & Waldfogel, 2005; Susin, 2002).
87 – Chapter II

Table II –7. Fixed-effects Panel Regression Results

Notes: Robust standard errors are used to account for heteroskedasticity. Standard errors are clustered by metropolitan area. + = p <
0.10, * = p < 0.05, ** = p < 0.01, *** = p < 0.001.
Coef. Sig. S.E. Coef. Sig. S.E. Coef. Sig. S.E. Coef. Sig. S.E.
Gov. Subsidies per 1,000 HH 0.065 + (0.036) 0.029 (0.037)
Public Housing per HH -1.142 (1.038) -0.513 (1.549)
Voucher per HH -0.486 (0.419) -1.032 + (0.545)
LIHTC per HH 1.240 + (0.709) 1.119 (0.973)
Construction per HH 0.071 (0.296) 0.152 (0.301) 0.653 + (0.334) 0.726 * (0.324)
% NH–White -12.800 (14.780) -14.570 (14.690) -10.930 (19.000) -13.620 (18.930)
% BA+ 0.137 (0.161) 0.146 (0.160) -0.138 (0.189) -0.126 (0.188)
per capita Homeownership 0.149 + (0.077) 0.151 + (0.077) 0.129 (0.097) 0.130 (0.096)
Ln(Median Gross Rent) -56.370 *** (3.305) -56.560 *** (3.304) -20.430 *** (3.651) -20.670 *** (3.639)
Ln(Median HH Income) 48.450 *** (4.664) 48.080 *** (4.573) 15.810 ** (5.166) 15.490 ** (5.086)
Ln(Employment) 4.225 (4.466) 4.179 (4.491) 22.320 *** (5.590) 21.990 *** (5.736)
year = 2006 (Ref.)
year = 2007 -0.029 (0.476) -0.073 (0.480) 0.372 (0.586) 0.353 (0.594)
year = 2008 -0.174 (0.504) -0.283 (0.510) 0.083 (0.683) 0.033 (0.711)
year = 2009 -3.116 *** (0.624) -3.205 *** (0.648) 0.677 (0.799) 0.606 (0.819)
year = 2010 -3.482 *** (0.795) -3.645 *** (0.810) 0.094 (0.983) -0.022 (1.017)
year = 2011 -3.044 *** (0.824) -3.261 *** (0.832) -0.043 (0.930) -0.182 (0.980)
year = 2012 -3.207 *** (0.956) -3.503 *** (0.981) -0.021 (1.067) -0.230 (1.135)
year = 2013 -3.051 ** (0.992) -3.393 ** (1.027) 0.092 (1.215) -0.131 (1.305)
year = 2014 -3.520 *** (1.053) -3.944 *** (1.093) -0.887 (1.241) -1.179 (1.350)
year = 2015 -2.493 * (1.132) -2.977 * (1.171) -0.613 (1.536) -0.934 (1.628)
year = 2016 -1.585 + (0.809) -2.013 * (0.881) -0.909 (0.936) -1.084 (1.160)
Constant -70.490 (52.540) -67.240 (53.330) -265.200 *** (66.240) -257.300 *** (68.270)
Observations 2,200 2,200 2,200 2,200
Adj. R-Squared 0.452 0.454 0.106 0.109
Model (3) Model (4)
Very Low-income (<50% of AMI) Rental Availability Extremely Low-income (<30% of AMI) Rental Availability
Model (1) Model (2)
88 – Chapter II

Additional Test on Panel Models
As noted earlier, there might be some concerns about generalizing the results from the
200 largest metropolitan areas in the sample. For example, some may argue that the
results are mainly driven by the top metros like New York (population rank 1) and Los
Angeles (rank 2) or small metros like Hanford (rank 199) and Monroe (rank 200), given
their distinctive characteristics.
I measure Cook’s Distance for each of the 200 metropolitan areas to see how
much the estimated coefficient of government subsidies and new construction (two key
variables of interest in this essay) is sensitive to the presence of individual metro
observation. The Cook’s Distance was computed for very low-income renter (VLI) model
1 separately for each of 11 survey years (2006 through 2016). Appendix II–D lists metros
that have the most positive influence on the coefficient of government subsidies in 2016
Appendix II–E lists metros in terms of the influences on the coefficient of new
construction in the same year.
In general, the influence on the coefficient of government subsidies and new
construction appears similar across my sample metros ranging between 0.090 and -0.092
for government subsidies variable and between 0.279 and -0.193 for new construction
variable, except New York metro area with the strongest positive influence in the case of
government subsidies. Possible interpretation would be that the New York is one of the
oldest city that has many old government-subsidized housing units including Public
Housing, which in turn may have an upward influence on the coefficient of government
subsidies on very low-income rental availability.

89 – Chapter II

5. Discussion

This essay has noted, as have other studies, that the substantial hardships among very
low-income renters are due, in part, to middle or even higher-income households
occupying units that very low-income renters can afford. Nonetheless, the scholarship on
rental availability still has three major blind spots: i) most studies focus on only the top
50 largest metropolitan areas, neglecting small- and mid-sized areas, ii) temporal
trajectories of rental availability have not been annually traced during and after the Great
Recession, and iii) previous studies were focused on general descriptive approaches
leaving regional factors that expand or depress rental availability untested. In contrast to
previous work, this essay tracks rental availability in 200 largest metropolitan areas
during the recent housing bust and recovery periods and examines the contextual factors
that are associated with more limited or expanded rental availability in some metros than
in others.
Across the nation, I found a substantial variation in the degree of rental housing
availability. Acute rental availability problems are concentrated in the most populous
metropolitan areas in the West Coast, South Florida, and Northeast regions, where
income levels are high. Many small- and mid-sized areas with limited higher-end rentals
also experience worsening rental availability during the recovery from the Great
Recession.
The models in cross-sectional and panel analyses examined the relative
importance of federal housing subsidies and new market-rate construction in determining
rental availability across the metros. The finding of the positive effects of rental subsidies
90 – Chapter II

on low-income housing opportunities in this essay is consistent with those in the
literature. As a determinant of rental availability, the market-rate housing supply is
relevant to this essay, which has demonstrated that increases in building permits led to
the greater availability of low-cost rentals to the poorest households.
Of all the explanatory variables, housing subsidy rates are the most amenable to
policy influence, and so their significance to increasing rental availability to the low-
income deserves attention. The results of the analyses in this essay suggested that public
housing, vouchers, and LIHTCs may benefit the poorest renters by expanding their access
to available rentals, but they were not consistent across all models. Given the reality that
very few new market-rate housing is intended for the poorest households, the role of
subsidy programs that serve their interests was supported by the results of this essay.
In addition, the findings highlight the influence of new market-rate construction,
which eases rental competition among the higher-end brackets, thus creating more
availability of low-cost housing to low-income renters. Not only a majority of housing
policy and program efforts devoted to increasing new construction, wielded at the
municipal and county government levels, but also zoning, land use planning, growth
management, and other smart growth tools can all be used to increase the scale of
housing construction in the private market. A major obstacle to such efforts, however,
would be rising construction costs, land constraints, fewer construction workers, and
local political resistance. The scarcity of public sources for housing subsidies underscores
the need for new policy tools that stimulate new construction in the private market and
thus ease overall competition for rental housing.

91 – Chapter III

Chapter III. Geocoding Inaccuracies: A Case Study for Evaluation of the Low-
Income Housing Tax Credit Program Data

1. Background

The Low-Income Housing Tax Credit (LIHTC) is now the nation’s largest housing
assistance program designed to produce affordable rental housing for low-income
households (HUD, 2018; McClure, 2010). The nation’s stock of LIHTC units as of 2018
is about 3 million units (HUD, 2019; JCHS, 2018).
23
As the scale of LIHTC surpasses all
the other forms of federal housing assistance in 2017 first time since its onset in 1987,
data-driven quality evaluation of the program has been of particular interest among
housing researchers and policymakers (Eriksen & Lang, 2018; McClure 2019).
The LIHTC database, created and administered by the U.S. Department of
Housing and Urban Development (HUD), became available to the public since 1997. The
database contains information on all of the 46 thousand projects and 3 million housing
units placed in service since 1987. Although some data about the program have been
available through various sources, HUD’s database is the only complete national source
of information on the physical, financial, and geographic characteristics of individual
LIHTC projects (HUD, 2019). With the continued update of the national LIHTC database
by HUD, housing researchers commonly use the data in learning and evaluating the
program.

23
Since the commencement of the LIHTC program in 1987, over 46 thousand projects have been
completed until 2018. 3 million units represent those that still exist under affordability contracts.
92 – Chapter III

The most fundamental information in the database is the geographic location.
HUD emphasizes that the street-address records enable researchers to look at the
geographic distribution and neighborhood characteristics of individual LIHTC project.
HUD also explains that the address records may help show how incentives to locate
projects in low-income areas and other underserved markets are working. Accordingly,
virtually every LIHTC study uses the address information for at least some part of its
analysis. Yet as I show in the following section, the scholarship on LIHTC program still
has two major blind spots: i) we have yet to know how accurately the address records are
mapped on the map, and subsequently ii) we have an insufficient understanding of the
extent to which the positional accuracy affects program evaluation results.
This essay endeavors to fill those twofold gaps. To do so, I apply two empirical
approaches to Los Angeles County, California, the case study area of this essay. First, I
take advantage of geospatial analysis to quantify the accuracy of geocoded location of
LIHTC projects in the study area. Los Angeles County Assessor’s land parcel
24
dataset is
used as ancillary data for accurate measurement. I find that the positional accuracy of
geocoded LIHTC database is generally high, but the accuracy varies depending on the
project size and areal unit of analysis. This suggests that the existing results based on the
HUD’s database are mostly reliable except for a few analyses that focused on a large-
scale project at a fine geographic level.

24
A parcel is an identification for taxation purposes as well as a recognized subdivision of property with
a written legal description that addresses permissions or constraints upon its development. Local
government divides the land in its municipality into zones (e.g. residential, commercial) in which certain
land uses are permitted or prohibited (Los Aneles County Assessor’s Database).
93 – Chapter III

Second, I conduct a demonstration analysis on transit accessibility of LIHTC
projects to examine the extent to which geocoding inaccuracies affect program evaluation
results. I find that transit-accessibility of LIHTC projects are overstated when geocoding
is significantly inaccurate, which is likely to mislead housing researchers and
policymakers in judging the accessibility of LIHTC projects to transit stations.
This essay begins with a review of the LIHTC literature that relies on HUD’s
database. I start with a review of the ways that existing studies deal with geocoding
accuracies of the database. I also review the Geographic Information System (GIS)
literature to identify possible sources of geocoding inaccuracies with regards to the use of
HUD’s LIHTC database. Then, I turn to the case study area of Los Angeles County,
California, to quantify positional accuracy of HUD’s LIHTC database by using the Los
Angeles County Assessor’s land parcel map as an ancillary data for accurate
measurements. A demonstration analysis on transit accessibility follows to assess the
impact of geocoding inaccuracies on program evaluation results. I conclude by
recommending data collection and management efforts that could enhance the positional
accuracy of geocoded LIHTC database and discuss whether prioritizing this effort should
be a policy goal for the LIHTC program.

94 – Chapter III

2. Use of LIHTC Database and Geocoding Inaccuracy

LIHTC Program and Database

The federal government operates a variety of housing subsidy programs for low-income
households in the United States (Schwartz, 2015). Among many, the Low-Income
Housing Tax Credit (LIHTC hereafter) Program is the largest rental subsidy program that
serves nearly 2.5 million low-income households across the nation (HUD, 2018; JCHS,
2018). Created by the Tax Reform Act of 1986, the LIHTC program gives state and local
LIHTC-allocating agencies the equivalent of nearly 8 billion dollars in annual budget
authority to issue tax credits for the acquisition, rehabilitation, or new construction of
rental housing targeted to low-income households (HUD, 2018). Each year, real estate
developers in each state prepare proposals for development projects which compete to
receive tax credits awarded by each state housing finance agency. The state agencies rank
order the proposals according to criteria published in the agency’s Qualified Allocation
Plan (QAP) with the top-ranked proposals winning the tax credits. Through this process,
an average of 1,435 LIHTC projects and 108,810 low-income housing units were placed
in service each year since the program was created as a part of the Tax Reform Act of
1986.
The federal government administers a comprehensive national LIHTC database.
25

The LIHTC database, created by HUD and became available to the public since 1997,

25
The Department of Housing and Urban Development provides a national public data on LIHTC
developments via LIHTC Database Access portal (https://lihtc.huduser.gov/).
95 – Chapter III

contains information on 46,554 projects and 3.05 million housing units placed in service
between 1987 and 2016. Although some data about the program have been made
available by various sources, HUD's database is the only complete national source of
information on various characteristics and geographic location of individual projects.
Nearly all researchers rely entirely or at least partly on the HUD’s database in evaluating
the LIHTC program.
26

Variables of the HUD’s database include project street address, number of total
and low-income housing units, the year the credit was allocated, and many other
characteristics of individual LIHTC projects. In identifying the location of individual
LIHTC projects, researchers generally convert their street address records into points on
the map through the geocoding process which is essential for LIHTC analysis but often
involves inaccuracies.

Geocoding LIHTC Database and Source of Inaccuracies

Figure III–1 shows a general workflow by which researchers use HUD’s original data.
The first step is to access to HUD’s website and download the tabulated LIHTC database
in Excel or other statistical table formats. In this tabulated data, a row represents a
LIHTC project while columns provide different features of each sample project. Since

26
Despite prevalent use of HUD’s database in LIHTC research, it is noteworthy that some studies relied
solely on other local sources, including annual State Housing Finance Agencies (HFA) Factbook published
by National Council of State Housing Agencies (NCSHA) (Malpezzi & Vandell, 2002), state-level agency
data (Deng, 2005; Williamson et al., 2009), and syndicator’s data (Cummings & DiPasquale, 1999).
96 – Chapter III

HUD relies on survey responses from many local housing authorities across the nation,
the LIHTC database is subject to survey error. For example, some housing authorities
submit complete information on LIHTC projects in their area while others submit
incomplete result with missing values. Researchers generally address this survey
inaccuracy through the refinement process which involves comparison with other
supplemental documents and ground-truth field surveys.
A single LIHTC project usually consists of one or more building structures,
serving tens or even hundreds of low-income households. Housing researchers, however,
generally assigns a single point of street-address record to a project as if all the building
structures within the project were located at a single geographical point. The process of
associating a street-address record with a point on the map – geocoding – is a very
common technique and widely supported by most GIS software. This essay focuses on
how accurately those geocoded points are placed on the map.
Lastly, researchers aggregate the geocoded points and allocate them into spatial
units that are often a reflection of administrative jurisdiction boundaries (e.g., school
district, ZIP code area, and census tract) based on an implicit assumption that the points
were positioned accurately on the map. Following main analysis should also rely on the
accuracy of the allocation process. Very few studies, however, have questioned the
positional accuracy of geocoded address records and subsequent allocation accuracy,
which are fully examined in this essay.

97 – Chapter III

Figure III –1. General Workflow of Using LIHTC Database and Source of Inaccuracies

Notes: General workflow is drawn from a review of LIHTC literature that relies on
HUD’s LIHTC database. It does not describe an exhaustive workflow, but rather
summarizes fundamental steps that data users should follow to use address records.

Of particular interest among LIHTC researchers is the relationship between the
number of LIHTC projects and other socioeconomic attributes of the area. The areal unit
varies depending on research purpose, the convenience of enumeration, and
administrative jurisdiction boundary of interest. Table III–1 shows a list of selected
LIHTC studies that used different areal unit into which geocoded address points were
allocated. The most common area is census tract as a proxy of the neighborhood
(McClure 2006; Oakley 2008; Baum-Snow & Marion 2009; McClure 2010; Horn &
Geocoding Refined Data Geocoding Inaccuracy
Allocating Geocoded Data
into Census Area of Interest
Conducting Main Analysis
(Program Evaluation)
Data Users Inaccuracy Inaccuracy
Downloading Data
from HUD's Website
Survey Inaccuracy
Refining Data
General Workflow Recognized Neglected
of HUD's LIHTC and Resolved and Unresolved
Resolve
98 – Chapter III

O’Regan 2011; Ellen et al. 2016; Ellen et al. 2018). Particularly, researchers focused on
examining the question of whether LIHTC projects and units were geographically
concentrated in socioeconomically distressed areas that were designated by HUD as
Qualified Census Tracts (QCTs).
27

Geocoded LIHTC cases were also allocated into state (Malpezzi & Vandell 2002;
Adkins et al. 2017), metropolitan area (Deng 2005; Eriksen & Rosenthal 2010), county
(Eriksen & Rosenthal 2010; Williamson et al. 2009), local jurisdiction (Nedwick &
Burnett 2015), 10-mile-radius circle (Eriksen & Rosenthal 2010), and census block-group
(Baum-Snow & Marion 2009). To compare LIHTC projects in urbanized areas with those
in rural areas, the location of LIHTC projects was categorized into metropolitan,
suburban, and non-metropolitan areas (Cummings & DiPasquale 1999) or central city,
suburb, and rural area (O’Regan & Horn 2013).
Several web services even visualize the location of geocoded LIHTC addresses
project as a point feature on the map (see Table III–1). However, web users can easily
notice how low positional accuracy of the points is. While some points are located well
on the corresponding parcel, others are often found in the middle of a nearby street or
intersection.

27
A Qualified Census Tract is one in which 50% or more of households have incomes below 60% of the
area median income or the poverty rate is 25% or greater.
99 – Chapter III

Table III –1. Summary of LIHTC Studies Based on the HUD’s National Database, By
Areal Unit into Which Geocoded Points were Allocated
Author (Year) Study Area Areal Unit
Census Area Larger than Census Tract
Cummings & DiPasquale
(1999)
Nation Metro, suburb, and non-metro
Census Tract
Freedman & Ownes (2011) Nation Census tract
Oakley (2008) 5 Largest Cities Census tract
McClure (2006) Nation Census tract
McClure & Johnson (2015) Nation Census tract
Lang (2012) Nation Census tract
Census Area Smaller than Census Tract
Edkins Et al. (2017) 50 states Census block-group
Di & Murdoch (2013) Texas School district
Lang (2015) LA county Parcel; Zip code area; Census tract
Area within a Distance from a Point
Nedwick & Burnett (2015) Selected MSAs Within 1/2, 1/3, and 1/4 mile from
transit station
Baum-Snow & Marion
(2009)
Chicago Within 1 km from census block-
group centroid
Eriksen & Rosenthal (2010) Nation Within 10 miles from census tract
centroid; County; MSA
Williamson et al. (2009) Florida state Within 800 feet from geocoded
voucher holder address; County
Point Location
Freemark (2018) City of Chicago Point per building structure
HUD eGIS Nation Point per project
LowIncomeHousing.US Nation Point per project
Policy Map Nation Point per project
Affordable Housing Online Nation Point per project
Notes: A study that used two or more areal units was categorized in terms of the smallest
unit. This table includes LIHTC studies that explicitly mention the geocoding process as
a part of analyses.
100 – Chapter III

LIHTC research has been conducted in terms of various areal units as shown in
Table III–1 above. Nonetheless, very few studies raised a question about whether
geocoded LIHTC address records are located accurately. Instead, most LIHTC
researchers conducted geocoding functionality loaded in GIS software. Positional
accuracy might have been neglected mainly because researchers often aggregate
geocoded points into an area and did not have to pinpoint the exact location of LIHTC
projects. Particularly, a large areal unit of analysis such as state and county would not
require researchers to achieve a high level of positional accuracy. Yet the problem is that
an inaccurately geocoded point can be allocated into the wrong area and consequently
result in errors of program evaluation results. However, a more serious problem is that we
have not even known whether geocoded points are located accurately, and if not, how
inaccurate it is.
There are mainly three reasons why geocoded LIHTC addresses are located
inaccurately on the map. First, the LIHTC database itself has limitations that arise from
the survey process. Given that the database relies on the quality of survey responses from
multiple local housing authorities across the nation, a variety of issues can arise (McClure
2006; McClure and Johnson 2015; Oakley 2008).
28
For some developments, the address
was incorrect or incomplete and therefore could not be matched to the addresses stored in
GIS software. As a result, a significant percentage of the units could not be geocoded by

28
Climaco et al. (2006) indicate that the projects developed in the early years of the program were more
likely to be missing from the database, which had very high reporting accuracy in later years. It is possible
that developments awarded tax credits in the early years (pre-1995) were less likely to be in low-poverty
tracts because of the tendency to merge LIHTC subsidies with Section 8 Moderate Rehabilitation funds.
101 – Chapter III

data users. The successful coverage for the geocoding process was reported as roughly 90
percent in terms of LIHTC projects (McClure, 2006; Oakley, 2008) and 84 percent in
terms of housing units (McClure, 2006). This survey issue has been partly addressed by
individual researchers through a data refining process, which I conduct in the following
section.
The geocoding technique itself also has problems and subsequently affects the
positional accuracy of geocoded results. There are a number of potential problems with
geocoding technique and a number of different sources of error (Harries, 1990; Ratcliffe,
2001). Some of these include line simplification, geocoding imprecision, geocoding non-
address locations, out-of-date street directories, abbreviations or misspelling, local name
variations, address duplication, non-existent addresses, noise in the address file, and
ambiguous or vague addresses. It is noteworthy that the level of these errors varies
depending on which GIS software and geocoding technique a researcher adopts.
In addition, a spatial unit of analysis affects the positional accuracy of geocoded
addresses. A study of over 20,000 addresses in Sydney, Australia, using a TIGER-type
geocoding process suggests that 5.0 to 7.5 percent (depending on geocoding method) of
addresses may be misallocated to census tracts, implying a somewhat high level of
positional accuracy. However, more than half of the same addresses were given
coordinates within the land parcel of a different property. As will be discussed in the
following Results section, this essay finds a similar variation in the positional accuracy of
geocoded LIHTC addresses according to the choice of different census area.

102 – Chapter III

3. Data and Methods

This essay focuses on LIHTC projects that have been built and placed in service in Los
Angeles County between 1987, the beginning of the LIHTC program, and 2016, the latest
data year. Main sources of data include HUD’s LIHTC database and Los Angeles County
Assessor’s land parcel map, and U.S. Census Bureau’s TIGER/Line Shapefiles for
various census areas.
By combining those data sets, I quantify positional accuracy of geocoded LIHTC
address points in the Los Angeles County area in three alternative ways. First, I measure
whether the geocoded address point is positioned within the boundaries of concordant
parcel polygon, which is often called point-in-polygon operation. Second, I measure the
distance between geocoded point and centroid of concordant parcel either geocoded point
is identified located within the concordant parcel or not in the first measure. Lastly, I
measure whether a pair of geocoded point and concordant parcel centroid is allocated into
the same census area.
Based on the results of these accuracy tests, a transit-accessibility analysis is
conducted to demonstrate the impact of positional accuracy on program evaluation. Note
that this transit-accessibility analysis is intended for the purpose of simple demonstration
and is not intended to address issues of transit accessibility of LIHTC projects in the
study area, which would require much more rigorous analysis.

103 – Chapter III

Study Area

I focus this essay on developments within the county of Los Angeles, California, which
contains many LIHTC projects featuring a variety of built forms in a multitude of urban
settings, including downtown LA, coastal area, mountains, port, and suburban areas. Los
Angeles is famous for housing crisis created by supply shortages that reduce housing
opportunity and generate mounting affordability problems, particularly among low-
income households. Nearly 80 percent of extremely low-income households (earning less
than 30 percent of area median income) in Los Angeles are spending more than half of
their income on rent (JCHS, 2018; NLIHC, 2019). Federal, state, and local housing
programs, including LIHTC, help ease worsening rent-burdens of Angelenos, but the
county lost 64 percent of state and federal funding for low-income affordable housing
production and preservation from FY 2008-09 to FY 2018-19 (California Housing
Partnership Corporation, 2019). Together, the county encompasses conditions present in
many large American cities, from limited public funding, severe rent-burdens to housing
shortages, but it is important to note that the findings of this essay should only be
interpreted as directly reflective of the county’s circumstances.

Data

To explore conditions in Los Angeles, I take advantage of three main sources of data.
First, I use HUD’s LIHTC database, the only national data source of the program. The
federal database is regularly updated on the basis of survey responses by local housing
104 – Chapter III

authorities. The LIHTC data available in public through an interactive website
(https://lihtc.huduser.gov). A variety of variables are available for data users, which are
largely categorized into project identification, location, physical characteristics, LIHTC
program characteristics, and additional subsidies and other financial characteristics
(Appendix III–A shows the full list of variables).
I retrieve the HUD’s database which contains 954 individual LIHTC projects that
have been built and placed in service in Los Angeles County between 1987
(commencement year of LIHTC program) and 2016 (most recent data year), composed of
73,059 housing units (Appendix III–B shows descriptive statistics of the LIHTC projects
located in Los Angeles County). I refine the downloaded database and generate a final
dataset of 879 projects, which in turn is geocoded on the map. These data refinement and
geocoding process will be discussed in detail in the following sections.
Second, I utilize an enhanced Los Angeles parcel map which is created and
updated by the Southern California Association of Governments (SCAG), the largest
metropolitan planning organization (MPO) in the United States that governs six southern
California counties including Los Angeles. It contains 2,092,552 land parcel polygons
within the boundary of Los Angeles County in 2012. I use the map as an ancillary layer
to quantify geocoding accuracies. Local cadastral records are a catalog of interests in land
parcels and administered by local county assessor’s office.
29
Individual polygons in the
cadaster are identified by a unique local parcel number which is called Assessor’s Parcel
Number (APN) or Parcel Identification Number (PIN).

29
In general, these are retained as digital maps that contain descriptions of land parcels as well as unique
identifiers that can be used to identify who has ownership rights to the land as well as other legal interests.
105 – Chapter III

The third source of data is the polygon boundaries of the three census areas for
the study area: census tract, block group, and block. The boundary files were obtained
from the U.S. Census Bureau’s TIGER/Line system, a production of the U.S. Census
Bureau that includes census data along with boundary files in GIS file format. Census
block is the smallest geographical area defined by the U.S. Census Bureau, which
consists of one or more land parcels. A group of adjacent blocks consists of a block-
group and a group of block-groups makes up census tract which is often regarded as
neighborhood or community by housing and urban researchers. I use census tract as a
proxy for the neighborhood but also use block-group and block as an alternative areal
unit to test geocoding accuracies.

Refining HUD’s LIHTC Database

The HUD’s LIHTC database includes street address records of individual LIHTC
projects. The LIHTC database is critical because the street address records are used in the
geocoding process. Some LIHTC projects, however, cannot or should not be geocoded
because their address records in the HUD’s original database are incomplete, incorrect,
double- or triple-counted, or missing. These inherent data limitations occur mainly due to
a large variation in the quality of survey responses from local housing authorities. To
address the data issue, I perform data refinement process in three steps (see Table III–2
and Appendix III–D).

106 – Chapter III

Table III –2. List of the Refined LIHTC Projects, By Refinement Type, Los Angeles
County, 2016
Refinement
Type
Address Recorded
in HUD’s Data
Refined Address
Count of
Cases

Recovery of
Missing Address

Missing 1063 West 39th Place,
Los Angeles, CA 90037
7
Revision of
Incomplete or
Incorrect Address

658 676 South Ferris Avenue,
Los Angeles, CA 90022
658 South Ferris
Avenue, Los Angeles,
CA 90022
123
Drop of Double-
counted Cases
• Case 1: 535 East Carson
Street, Carson, CA 90745
• Case 2: 535 East Carson
Street, Carson, CA 90745

• Single Case: 535 East
Carson Street, Carson,
CA 90745
66
Drop of Triple-
counted Cases
• Case 1: 108 East 5th Street,
Los Angeles, CA 90013
• Case 2: 108 East 5th Street,
Los Angeles, CA 90013
• Case 3: 108 East 5th Street,
Los Angeles, CA 90013
• Single Case: 108 East
5th Street, Los Angeles,
CA 90013
6
Notes: Refined parts are highlighted grey. See Appendix III–D for the full list of refined
LIHTC projects. Universe is 954 LIHTC projects registered in the HUD’s original
database.
Sources: HUD’s LIHTC Database, 2018.

First, I identify and fill in missing address records by utilizing other variables
available in the HUD’s database, such as project name, property owner (real estate
development company) profile, and other physical characteristics of LIHTC building
structures. For example, I search a project name on the web search engines (e.g., Google,
Bing, and Yahoo) to find the address. Development companies often post a list of
completed projects and addresses on their website. Public administrative records
managed by the County of Los Angeles and other LA-based non-profit organizations are
also useful to find the missing addresses. None of these supplemental sources are
107 – Chapter III

complete by themselves, but in combination, they enable me to recover all of the missing
addresses. For verification of the recovered addresses, I search the addresses on the Los
Angeles County Assessor’s web service to find property information that are then
compared with variables available in the original HUD’s database, such as number of
housing units, year structure built, and structure type (single-family versus multifamily). I
find a perfect match between the County Assessor’s web service and HUD’s database
with regards to the recovered addresses.
Second, I identify and revise street address records that are incomplete or
incorrect for various reasons. For instance, address records may be incomplete or
incorrect when i) a typo is included, ii) two different street numbers are included in front
of street name, iii) spacing is missing between street number and name, and iv) street
name is abbreviated. Other minor revisions are also needed to prevent potential technical
errors in the geocoding process, such as changing “seventh avenue” to “7th

avenue.”
Lastly, I identify double or triple-counted cases and drop duplicated cases leaving
a single case per project. For example, 66 double-counted cases were reduced to 33 while
6 triple-counted cases were reduced to 2.
Data refinement results in 879 LIHTC projects with complete street address,
consisting of 66,573 housing units, which account 92.1 percent of 954 projects and 91.1
percent of 73,059 units in the original HUD’s database. One noteworthy point is that this
data refinement process improves geocoding results by increasing the number of LIHTC
projects geocoded on the map and preventing overlaps of double or triple-counted
projects.

108 – Chapter III

Geocoding Refined Data

I geocode the refined data by using Geocoding Toolbox available in ArcMap 10.1
program (see workflow diagram in Appendix III–C). Geocoding is a common GIS
technique that associates a street-address record with a point on the map. Point data for
urban locations are commonly created from geocoded address records. Current geocoding
tools in the United States are derived from US TIGER files, collections of street line
segments that hold street names, and the range of house numbers on each side of the
street as attribute data. The TIGER composition was originally developed to map US
census data and provides a US-wide address matching standard (Cooke, 1998).
Figure III–2 below displays 879 LIHTC projects within the boundary of Los
Angeles County. These projects were built and placed in service between 1987 and 2016.
One small grey circle represents a single LIHTC project. Many LIHTC projects appear to
concentrate in downtown Los Angeles where many multifamily housing structures were
built. LIHTC projects are also found along the coast, such as Santa Monica and Long
Beach, and in suburban areas, such as Lancaster and Palmdale.
Although Figure III–2 is useful to understand the overall distribution of LIHTC
projects across the county, it does not show exactly where the geocoded points are placed
on the map. Ideally, one might expect that all of the geocoded points are placed in
residential land areas on the map. However, some geocoded points were positioned in a
variety of non-residential areas.

109 – Chapter III

Figure III –2. Geocoded Address Points of LIHTC Project Placed in Service between
1987 and 2016, Los Angeles County, 2016

Sources: Esri’s World Topographic Map; HUD’s LIHTC Database, 2018; Los Angeles
County Assessor’s Land Parcel Map, 2016; U.S. Census Bureau’s TIGER/Line
Shapefiles, 2018.

Location of Figure III–3
Geocoded Point
of LIHTC Project
LA County Boundary
110 – Chapter III

Table III–3 below illustrates different kinds of location in which geocoded
LIHTC points were positioned as a result of the geocoding process in ArcMap. As will be
shown in the results section, most geocoded points were located within the boundary of
residential parcels (panel a in Table III–3). At this stage, I cannot tell whether the
geocoded points within residential areas are located within the concordant parcel or a
nearby irrelevant parcel. This distinction can be made after I pair each geocoded point
with the concordant parcel as discussed in the following section.
When a geocoded point is found in a non-residential parcel, there are many
different types of land. As shown in panels b through f in Table III–3, some geocoded
points are found on the road, which does not enable the researchers to determine the side
of the road that the LIHTC project actually locates (panel b). Other geocoded points are
found in green space where residential development is not allowed (panel c). Geocoded
points are also placed in vacant land parcel (panel d), public land parcel (panel e), or
commercial parcel (panel f). Table III–3 is not exhaustive but shows how inaccurate
geocoded locations could be.

111 – Chapter III

Table III –3. Location of Geocoded Address Point (Black Circle) of LIHTC Project, Los
Angeles County

(a) Residential Parcel

(b) On the Road

(c) Green Space

(d) Vacant Parcel

(e) Public Space

(f) Commercial Parcel

Sources: Esri’s World Topographic Map; HUD’s LIHTC Database, 2018; Los Angeles
County Assessor’s Land Parcel Map, 2016; U.S. Census Bureau’s TIGER/Line
Shapefiles, 2018.

112 – Chapter III

Matching Geocoded Point with the Concordant Parcel

As pointed previously, it is difficult to determine without further analysis if a geocoded
point found in a residential parcel is positioned in the concordant parcel. In this section, I
utilize both automated procedures and manual (or visual) methods to achieve complete
matching between geocoded points and concordant parcels (see workflow diagram in
Appendix III–C).
First, I perform a standard point-in-polygon operation to identify geocoded points
that are placed within residential parcel polygons. The LIHTC points identified within
residential parcels, however, still require verification. The verification is particularly
important for a large residential area where a number of residential land parcels are
concentrated and a geocoded LIHTC point was placed into a wrong nearby residential
parcel.
Next, I compare the information in the HUD’s database with the information in
the Los Angeles County Assessor’s database, which publishes information such as the
number of housing units in building structure, year structure built, structure type (single-
family versus multifamily), and government welfare exemption. The Los Angeles County
Assessor’s Office’s web service allows the public to find detailed property information
by point on a map or a search by street address. Through the comparison between the
assessor’s data and HUD’s database, I manually match geocoded points to their
concordant parcels to verify the results of the previous point-in-polygon operation.
Lastly, I use visual comparison method to match geocoded points that are
incorrectly placed outside of residential parcels to concordant parcels. I take advantage of
113 – Chapter III

Google Maps and Street View, which offer satellite and pedestrian-scale imagery,
respectively.
30
These tools allow me to visually identify appearances of all building
structures around the location of a geocoded point, such as structure type (single-family
versus multifamily), apartment complex name displayed at the front gate, and street
number displayed outside of a building. This approach is somewhat time-consuming but
very useful to check ground-truth. For example, I can search nearby residential buildings
for a geocoded point that is positioned on a road outside of a residential parcel. This
visual identification completes the entire process of matching geocoded points with
concordant resident parcel polygons.
Figure III–3 below shows the distribution of geocoded points (small black circles)
and their concordant parcels (orange polygons) in Westlake in downtown Los Angeles.
This neighborhood is a good example of geocoded points not placed within the
concordant parcel polygons and a similar pattern is found across Los Angeles County.

30
Satellite and street imageries in Google Maps and its Street View for Los Angeles County were taken
between 2014 and 2016. A similar approach has been used to accurately identify physical characteristics of
LIHTC building structures and their surrounding neighborhood environments (Freemark, 2018; Kelly,
Wilson, Baker, Miller, & Schootman, 2013; Rundle, Bader, Richards, Neckerman, & Teitler, 2011).
114 – Chapter III

Figure III –3. Westlake Neighborhood in Downtown Los Aneles: Geocoded Points of
LIHTC Project (Circles Filled Black) and Paired Land Parcels (Polygons Filled Orange)

Notes: Four different layers overlap in this map. The bottom layer is ArcMap’s World
Topographic Map which displays street and park name text, park and lake topography,
and building footprints. On top of the topographic map, LA county assessor’s parcel
boundary map is overlaid, displaying land parcel boundaries in dark grey. On top of those
two base maps, I create and place two layers, one for paired land parcels (polygons filled
orange) and the other for geocoded address records of LIHTC projects (small circles
filled black).
Sources: Esri’s World Topographic Map; HUD’s LIHTC Database, 2018; Los Angeles
County Assessor’s Land Parcel Map, 2016; U.S. Census Bureau’s TIGER/Line
Shapefiles, 2018.

Geocoded Point
of LIHTC Project
Concordant Land
Parcel
115 – Chapter III

Table III–4 below describes largely three possible cases of geocoded point
locations that I observed during the matching process: (a) point placed in concordant
residential parcel, (b) point placed in wrong residential parcel, and (c) point placed
outside a residential parcel. The second column shows the geocoded point location (black
circle). The third and fourth columns show land parcel layer (a set of void polygons
adjacent to each other) and concordant parcel location (polygon highlighted grey),
respectively.
The first and ideal case shows that a geocoded point of a LIHTC project (small
circle shaded black) that is located within its concordant parcel polygon (rectangle shaded
grey). It would be ideal for LIHTC researchers if all the pairs between geocoded LIHTC
points and land parcel fall into the case. However, as explained above, this is not always
the case. In the second case, a geocoded point is found within a wrong residential land
parcel across the road (case b of Table III–4). A geocoded point can also be located
outside a residential parcel. For example, it may fall on a road as shown in case c of
Table III–4, in a vacant land parcel, or on non-residential parcels (e.g., commercial,
industrial, etc.). Researchers will most likely be misled by these geocoded points not
placed in their concordant residential parcels due to identifying a nearby but incorrect
residential land parcel as a location of LIHTC project of their interest.

116 – Chapter III

Table III –4. Three Cases of Geocoded Point Location

Case
Geocoded Point
of LIHTC Project
Geocoded Point
Overlaid onto Local
Land Parcel Map
Geocoded Point and
Concordant Land
Parcel
(a) Point
Placed in
Concordant
Residential
Parcel

(b) Point
Placed in
Wrong
Residential
Parcel

(c) Point
Placed
Outside a
Residential
Parcel

Notes: Maps presented in this table are hypothetical.

Another noteworthy observation during the matching process is that there are 20
LIHTC projects that were built across more than one parcel.
31
For example, the Witmer

31
Among the 20 multi-parcel projects, 9 projects were built across 2 parcels, 4 projects across 3 parcels, 2
projects across 4 parcels and 5 projects across 5 parcels.
117 – Chapter III

Manor Preservation Project consists of 238 housing units that are located across 3
residential land parcels (APN 5153-015-001, 5153-015-002, and 5153-015-003). It means
that these multi-parcel projects have more than one concordant parcel. However, only 6
of the 20 multi-parcel projects were placed within at least one of the concordant parcels
during the geocoding process. In other words, 14 of the 20 multi-parcel projects were not
placed in any of the concordant parcels, suggesting a high geocoding inaccuracy at 70
percent (= 14 / 20 × 100). Geocoded points of these multi-parcel projects were matched
to the largest one of the concordant parcels. Including these multi-parcel projects, I
identified 147 (out of 879) geocoded points placed outside of their concordant residential
parcels and matched to their concordant residential parcels.

Allocating Geocoded Point and Concordant Parcel into Census Areas

Lastly, I allocate the geocoded points and concordant residential parcels into census
areas. As discussed above, many LIHTC researchers are most interested in the
relationship between the number of LIHTC projects in a given census area and other
socioeconomic attributes of the area. I use three different census areas: census tract, block
group, and block as a proxy for neighborhood (or community). This essay focuses on
whether a geocoded point and the concordant parcel is allocated in the same census area.
If all of the 879 LIHTC points are allocated into the same census area as their concordant
residential parcels, it means that the original HUD’s database is geocoded and positioned
on the map accurately enough to result in correct allocations into census areas. If not, it
means LIHTC researchers might need to consider a way to enhance geocoding accuracy,
as discussed in this essay.
118 – Chapter III

In sum, the final dataset of this essay consists of 879 LIHTC projects with four
new or revised features: i) complete street address through data refinement, ii) geographic
location through the geocoding process, iii) land parcel identification number through the
matching process, and iv) census tract, block group, and block code through allocation
process. The following section discusses the empirical findings on accuracy tests and
demonstration analysis.

4. Results

This section describes whether the geocoded LIHTC records in the study area are
accurately positioned on the map. Moreover, if they are found to be not accurate, I will
explore the degree of inaccuracy and how it affects program evaluation results. The
degree of positional accuracy varied depending on the measurement method and areal
unit of analysis. A demonstrative analysis shows that the positional inaccuracy overstates
transit-accessibility of LIHTC projects located in Los Angeles County, especially when
the transit-accessibility was measured in terms of the number of housing units close to
rail stations.

Accuracy of Point-in-Polygon Operation

For the first accuracy test, I performed a standard point-in-polygon operation to examine
the match between geocoded address points of HUD’s LIHTC data and their concordant
parcel polygons. Table III–5 below shows that 769 (87.5%) of 879 geocoded points in the
119 – Chapter III

final dataset were located within a residential land polygon from the parcel set. The
remaining 110 points (12.5%) were located outside of any residential parcels. Instead, the
110 points were geographically located on the road, walkway, and other vacant land.

Table III –5. Point-in-Polygon Operation between Geocoded Address Points and Land
Parcel Polygons, Los Angeles County, 2016

Count of Points Percentage (%)
Points with Complete Address 879 100.0

Points in Residential Parcel 769 87.5
Points in Correct Residential Parcel 732 83.3
Points in Wrong Residential Parcel 37 4.2

Points Not in Residential Parcel 110 12.5
Points on Road 88 10.0
Points on Vacant Parcel 6 0.7
Points on River 1 0.1
Points in Other Uses 15 1.7

Sources: Esri’s World Topographic Map; HUD’s LIHTC Database, 2018; Los Angeles
County Assessor’s Land Parcel Map, 2016; U.S. Census Bureau’s TIGER/Line
Shapefiles, 2018.

Among the 769 geocoded points within any parcel polygon, 732 points were
located correctly in the concordant polygon while 37 points were in the polygon for
another address. If we consider these correctly-located 732 points as a proportion of all
the 879 geocoded points, this equates to an accuracy level of 83.3% (= 732 / 879 × 100).
This accuracy rate would seem to raise a concern for LIHTC researchers and local
housing planners who are interested in micro-level mapping applications of address
records. In fact, the U.S. Census Bureau’s TIGER/Line system was not designed
originally along with a parcel basis but was created to provide a relatively accurate large-
scale mapping capability. It is therefore understandable that to some degree it does not
120 – Chapter III

hold up to a micro-level examination. However, the somewhat low percentage of records
that were located correctly in concordant parcel polygon (83.3%) suggests that further
work might be necessary to fully recognize the importance of the positional inaccuracy of
geocoded address points in LIHTC research. Visual inspection of several misplaced
points shows that a high number was near the correct polygon, often located in a
neighboring road; however, this was not always the case. The following analyses attempt
to get a more numeric measure of the inaccuracy level.

Distance Accuracy

For my second accuracy test, I measure distances from the geocoded address points to
concordant parcel centroids (“paired points”) to examine the distribution of distances.
The geocoded point for each address recorded in HUD’s LIHTC database, as extracted in
the preceding section, is used again while a new centroid is extracted for each concordant
parcel polygon. Then, I examine the distribution of the Euclidean distance, which is
computed as a distance between the paired points.
The computed centroid for each parcel polygon does not necessarily show the
location of the building structures that are within the LIHTC project parcel. The mean
area of the 879 parcels in the final dataset is 74,060 square feet. Examination of aerial
photography of the study area with parcel polygons used as an overlay indicated that the
majority of building structures dominate the whole land parcel with little room for a
pedestrian or, surprisingly for such an affluent area in the Los Angeles downtown, a
121 – Chapter III

parking space. Due to the dominance of the domestic building structures within the land
parcel, the centroid was located within the building perimeter the vast majority of times.
Figure III–4 below shows a histogram of the distances between geocoded points
and concordant parcel centroids. The x-axis represents distance in the 10-meter interval
while the y-axis represents the number of paired points that fall in each distance interval.
The mean distance is 55.6 meters with a standard deviation of 34.1 meters.
32
The
majority of distances is below the average, but many cases fall in distances that are
greater than the average. These cases, in which the distance is above the average, include
both geocoded points positioned outside of concordant residential parcels and many
points that were correctly positioned in larger or extended concordant parcels. This
finding implies that a careful test of the positional accuracy of geocoded address records
requires not only conducting simple point-in-polygon operation but measuring distances
between geocoded address points and parcel centroids.

32
The distances include a number of outliers with distances of over 200 meters. These outliers are
suspected errors in either the HUD’s LIHTC address record, the parcel data set, or the geocoding engine.
This was confirmed by plotting the points and comparing back to the original parcel files. The actual cause
of the error is not possible to determine without the resources to ground truth hundreds of addresses, and
when plotted on a map the erroneous addresses were scattered across the study area without any apparent
pattern. Therefore, as in other studies that have examined distances between estimated points and actual
locations, a ‘trimmed mean’ that does not include the smallest and largest 5% of values can be used as a
more accurate reflection of the data. The mean distance between the geocoded points and their
corresponding parcel centroid in the reduced set of 879 concordant points is 55.6m.
122 – Chapter III

Figure III –4. Distribution of Distances between Geocoded Address Points and Parcel
Centroids of LIHTC Development, Los Angeles County, 2016

Sources: Esri’s World Topographic Map; HUD’s LIHTC Database, 2018; Los Angeles
County Assessor’s Land Parcel Map, 2016; U.S. Census Bureau’s TIGER/Line
Shapefiles, 2018.

Allocation Accuracy

For my last accuracy test, I turn to the spatial relationship between the geocoded point,
paired parcel centroid, and concordant census area. This accuracy-test focuses on whether
a pair of geocoded point and parcel centroid is allocated into the same concordant census
area. Point-in-polygon operation was performed on both the geocoded points and the
parcel centroids for all 879 LIHTC projects in the study area.
I use three census areas, such as census tract, block group, and block. There are
many types of the areal unit used for socio-economic research, and the aggregation of
points within other types of polygons can be used as a cartographic technique to simplify
123 – Chapter III

the display of large numbers of points. Neither the research application nor the
cartographic requirement necessarily requires census tracts (or block-group or block),
though these have been used in this essay as they are a common choice of thematic
aggregator, giving a good indication of population density, and are a popular choice of
areal unit for LIHTC research. They are also readily available to researchers and the
public through American FactFinder administered by U.S. Census Bureau or Integrated
Public Use Microdata Series (IPUMS).
It is important to stress that the results of point-in-polygon searches depend on the
accuracy with which the boundaries of the polygons were digitized. In the case of the
census tract, the user’s trust is in the hands of the U.S. Census Bureau. Given the meter
resolution of geocoded points in this essay, it must be acknowledged in advance that
some degree of geographical uncertainty will exist about points that lie close to a census
tract boundary and that for compression purposes the general census area boundaries that
are available publicly have undergone a stage of line simplification. The following
section of the study examines this degree of uncertainty.
Table III–6 below, which summarizes the comparison of the unique identifiers for
each paired point, shows that 14 (1.6% of 879) geocoded points fell into different census
tracts from their concordant parcel centroids. Most users of geocoding tools, such as
LIHTC analysts or affordable housing developers who want to map existing LIHTC
projects, for example, will neither have time nor skills to question and analyze the
accurate distribution of hundreds of geocoded locations, and will proceed with their
analysis using the points automatically geocoded by their own GIS software.

124 – Chapter III

Table III –6. Geocoded Points/Parcel Centroids and Census Area Polygons

Count of Pairs Percentage (%)
Total Pairs of Geocoded Point and Parcel 879 100.0

Census Tract
Pairs in Different Census Tract 14 1.6
Pairs in the Same Census Tract 865 98.4

Census Block-group
Pairs in Different Census Block-group 23 2.6
Paris in the Same Census Block-group 856 97.4

Census Block
Pairs in Different Census Block 75 8.5
Pairs in the Same Census Block 804 91.5

Sources: Esri’s World Topographic Map; HUD’s LIHTC Database, 2018; Los Angeles
County Assessor’s Land Parcel Map, 2016; U.S. Census Bureau’s TIGER/Line
Shapefiles, 2018.

Though an inaccuracy of 1.6% at the census tract level would not seem high, this
figure is likely to be of most interest to LIHTC researchers concerned with address-level
data and their relationship to census variables. This essay shows that the census tract
analyses do not involve a serious level of inaccuracy in the geocoded addresses of LIHTC
project. At the block-group and block levels, however, the inaccuracy of geocoded
address rises to 2.6% and 8.5%, respectively. This higher inaccuracy at finer geographies
suggests cautions of LIHTC researchers when the results of a point-in-polygon process
indicate only slight statistical significance in micro-level analyses. The lack of high
variation in variables between adjacent block-groups and blocks goes some way to
alleviating the effects of this discrepancy for counts of LIHTC projects in polygons. The
effect could even increase at the housing unit level as will be shown in the following
demonstration section.
125 – Chapter III

In sum, the accuracy of the allocation process appears high particularly at the
census tract, which is the most common areal unit of LIHTC studies as reviewed in the
previous section. The high level of accuracy implies that LIHTC researchers may use the
HUD’s original database if the areal unit of analysis is at or larger than the census tract,
though the omission of data validation might reduce the quality of the analysis
undertaken.

Demonstration Analysis Result: Transit-Accessibility of LIHTC Project

As a way of demonstrating the impact of positional accuracy of the LIHTC database on
the program evaluation result, a relatively simple analysis is conducted in this section on
LIHTC projects located close to public transit stations in Los Angeles County. Data on
the locations of light rail transit stations of major lines in 2016 were acquired from the
LA Metro Database. In total, 96 stations were mapped and LIHTC projects (and housing
units with individual projects) were counted within a series of distances from the stations,
including 0.25, 0.5, 0.75, and 1 mile.
Figure III–5 below shows transit-accessibility, measured by how many LIHTC
projects are located within a range of distances from transit station, under two alternative
measures: one distance from the nearest station to geocoded point (highlighted red), and
the other distance from the nearest station to parcel centroid (highlighted black). Panel (a)
has distance from station on the x-axis and total number of LIHTC projects on the y-axis
while panel (b) has the number of housing units instead of the number of projects on the
y-axis.
126 – Chapter III

Figure III –5. LIHTC Projects Within Certain Distances of Rail Stations, Using
Geocoded Points and Parcel Centroids

(a) Number of LIHTC Projects

(b) Number of Housing Units within LIHTC Projects

Sources: Esri’s World Topographic Map; HUD’s LIHTC Database, 2018; Los Angeles
County Assessor’s Land Parcel Map, 2016; U.S. Census Bureau’s TIGER/Line
Shapefiles, 2018.
127 – Chapter III

Panel (a) shows that there is a very slight difference between transit-accessibility
results from the two alternative point identifications. This indicates that, in terms of
project level, LIHTC researchers may obtain nearly identical transit-accessibility results
regardless of which point identification method they use. However, panel (b) shows a
greater difference between results from two alternative point identifications. Particularly,
the difference widens when the buffer distance increases; 12.6% of the difference within
0.25 mile, 19.1% in 0.5 mile, and 17.6% in 1 mile.

5. Discussion

This essay offers new insight into the importance of data accuracy in affordable housing
research by clearly quantifying positional accuracy of geocoded address records and
demonstrating its impact on program evaluation. It finds that the geocoding accuracy of
the HUD’s LIHTC database is generally high, suggesting that the existing evidence on
the LIHTC program is not likely to be biased by data inaccuracies and therefore reliable.
However, I find that the level of geocoding accuracy varies depending on project scale
and areal unit of analysis, particularly when examining large-scale LIHTC projects at a
finer geographic area than census tract.
It is not in doubt that the point-in-polygon operation results in a correct matching
between geocoded point and concordant census area in the majority of LIHTC cases,
especially at the census tract level. Yet it is noteworthy that positional accuracy is lower
when an attempt is made to relate counts of cases within block-group or block to other
socioeconomic variables, or where areal comparisons are made between variables
128 – Chapter III

geocoded through two different processes. Given the spatial separation found in this
essay between geocoded point, parcel (or parcel centroid), and census areas, it would
seem prudent to verify the accuracy of any geocoded LIHTC data. Otherwise, researchers
might mistakenly validate any important findings that rely on geocoding process without
considerations on the positional accuracy of geocoded address points.
Accurate information on the location of the LIHTC project is important for
housing researchers and policymakers at the national, regional, and local levels. This
essay recommends that LIHTC researchers and local low-income housing planners to
attempt to improve the accuracy of LIHTC data in their locality by using local parcel
dataset as explored in this essay. Accurate positioning of LIHTC projects is especially
important in LIHTC research that utilizes location information at a finer aggregation,
such as census block-group and group, or focuses on the count of housing units instead of
projects. Accurate location of LIHTC records is also important when researchers examine
an urban area where LIHTC developments are densely concentrated.
This essay’s findings were made possible through a time-consuming refinement
of HUD’s database and matching it with a local parcel map. To date, there is not a
national parcel data that covers the entire U.S. territory. As such, it would be difficult to
replicate its approach at the national level with existing sources. The HUD’s database
does not include data at the parcel level; therefore, there is a lack of insight into the
positional accuracy of individual projects. The HUD could improve our understanding of
and scholarship on LIHTC through a revision of the database, such as requesting local
housing authorities to add local information on a land parcel.
129 – Chapter III

More research must evaluate several of the topics explored in this essay. One
avenue of potentially fruitful study would examine the location affordability of the
LIHTC program through developing a parcel-based LIHTC database as discussed in this
essay. Further examination is necessary to determine the geographic unit that is most
appropriate to identify the LIHTC project locations. Considering that a project often
consists of multiple building structures, depending on the granularity of the research,
some LIHTC researchers may want to investigate even further into how the location of
LIHTC projects could be identified at the building level. Additionally, to further extend
this essay’s analysis on large-scale LITHC projects that showed relatively low positional
accuracy, institutional research in support of local housing authorities is needed to test
whether and the extent to which an improved accuracy of LIHTC database helps evaluate
locations of existing large-scale LIHTC projects. A comparative approach that considers
conditions in various large counties could offer insight into how different housing
markets vary, and if so, which contextual factors are related to the variation.
130 – Conclusion

Conclusion

My dissertation featured three independent essays related to the topic of depressed
housing access in times of shortage crisis during and after the Great Recession in the
United States. This final chapter summarizes key findings of each essay and potential
contributions to policy debates.
Findings of the first essay (Chapter I) showed 8.0 million households (18.4
percent of all renter households) were dislodged and “made invisible” from the housing
market across the nation between 2000 and 2017. The prevalence of housing shortages is
estimated across the 100 largest metropolitan areas by comparing actual growth in
housing to my estimates of expected housing needs. Though dislodgements due to
shortage are extensive across metros, rental dislodgements appear substantially greater in
some places than in others. More dislodgements occurred in metros with more severe
affordability problems and housing shortages, which amplifies the true magnitude of the
housing crisis. Results also identified the dislodged households as those with lowest
personal incomes and adults under the age of 35 whose household formation is most
flexible due to other accessible living arrangements, such as staying with parents or
doubling up with roommates.
Results of the second essay (Chapter II) presented that only 46.9 percent of the
very low-income (earning half or less of their area’s median income level) renters in the
nation were successfully occupied in affordable housing, which is much lower compared
to the rate of 60 percent in the 1980s and 1990s. Despite the national average, I found a
substantial variation in metropolitan patterns when it comes to rental availability. Many
131 – Conclusion

metropolitan areas in Florida and Southern California turned out to have the least rental
availability for lower-income populations, while the Midwest region had the most
availability. Regression results highlight the imperative need for increasing overall
housing supply, either government-subsidized or market-rate, to ease rental competition
in all rental brackets and open availability of lower-cost housing to the poorest renters.
The most effective program to expand lower-income housing access was the Low-
Income Housing Tax Credit (LIHTC), the largest federal rental assistance program. In
addition, affordable rentals also were more available to lower-income households in
metropolitan areas with more market-rate constructions, specifically single-family
constructions that draw potential homebuyers into owner market and subsequently
expand availability of lower-cost housing to the poorest renters.
Evidences from the third essay (Chapter III) showed that the positional accuracy
of geocoded LIHTC database is generally high, suggesting that the existing evidences on
the program are not likely to be biased by geocoding inaccuracies and therefore reliable.
However, I also found that the level of geocoding accuracy varies depending on project
scale and areal unit of analysis, particularly when examining large-scale LIHTC projects
at a finer geographic area than census tract. This suggests that LIHTC researchers need to
carefully interpret the exact location of LIHTC projects when the project of interest is in
a land parcel that is very large or very lengthy along the road. A demonstration analysis
on transit-accessibility indicated that the inaccurate database overstated transit-
accessibility of subsidized housing, particularly in the case of large-scale developments.
My research findings will contribute to improving the public debate about the
housing shortage crisis and its meaningful consequences for everyone. Findings on the
132 – Conclusion

true magnitude of housing shortage and subsequently unmet housing needs (Chapter I)
would be particularly timely for public discussions about housing shortage problems
which have proven to be a major policy issue in gubernatorial elections across the nation.
Conventional approaches have underestimated housing needs because they only focused
on households that have survived the market and completely ignored displaced
households. My findings highlight the importance of using the entire local population and
a housing−demographic method to estimate the true magnitude of the housing shortage. It
is important for housing and related research to have more complete and thorough
estimates to inform policymakers of the “invisible” and actual reality of the shortage
crisis.
Results on rental availability (Chapter II) would contribute to facilitating the
policy debate about lower-income housing opportunities. My results suggest that not only
government subsidies but market-rate constructions are significant in expanding the
overall housing access by lower-income households in the largest metropolitan areas
across the nation. Regression results clearly show the positive association between
federal rental assistance and lower-income housing access. However, needs for rental
assistance outstrip the allocated resources which continue decreasing in today’s fiscal
climate. In the face of limited public resources, we should re-recognize the importance of
boosting market-rate construction as an indirect strategy that can help expand access to
lower-income housing. Under tight housing market condition, both lower and higher-
income households are forced to compete with one another for scarce homes. According
to my research findings, the more market-rate housing is constructed, the more it will be
accompanied by a consequent increase in lower-income occupancy of lower-cost
133 – Conclusion

affordable units. In other words, while it may not directly benefit lower-income
populations, new market-rate housing could ease the competition for housing.
Lastly, findings on geocoding inaccuracies (Chapter III) are anticipated to help
researchers to better understand the importance and need for data accuracy. I recommend
local affordable housing researchers to improve the accuracy of the federal LIHTC
database in their locality by combining local assessor’s parcel data with federal database
as explored in my dissertation. The enhanced parcel-level LIHTC database can benefit
researchers in largely three ways. First, it would increase the quality of program
evaluation especially at a finer geographic level, such as census block-group and group,
and when focusing on absolute count of LIHTC housing units instead of that of LIHTC
projects. Improved data accuracy can also benefit researchers when they examine central
cities of urbanized areas where LIHTC developments are densely concentrated, which
will require more accurate locational information of LIHTC projects. Furthermore,
improved parcel-level database may be used to allocate LIHTC projects in neighborhoods
with the most need. This is critical for the growing number of researchers who are
interested in examining the potential benefits of government-subsidized housing located
in disadvantaged neighborhoods, especially in how it will produce broader societal gains
for lower-income households.
134 – References

References

Introduction

Joint Center for Housing Studies of Harvard University. (2018). The State of the Nation’s
Housing. Cambridge, MA: Joint Center for Housing Studies of Harvard
University.
Myers, D. (1990). Housing Demography: Linking Demographic Structure and Housing
Markets. In Social Demography. Madison, WI: University of Wisconsin Press.
Myers, D., Painter, G., Lee, H., & J. Park. (2016). Diverted Homeowners, the Rental
Crisis and Foregone Household Formation. Washington, D.C.: Research Institute
for Housing America.
Ruggles, Steven, Sarah Flood, Ronald Goeken, Josiah Grover, Erin Meyer, Jose Pacas,
and Matthew Sobek. (2018). Integrated Public Use Microdata Series (IPUMS)
USA: Version 8.0 [dataset]. Minneapolis, MN: University of Minnesota.
https://doi.org/10.18128/D010.V8.0.
U.S. Department of Housing and Urban Development. (2017). Worst Case Housing
Needs: 2015 Report to Congress. Washington, D.C.: U.S. Department of Housing
and Urban Development.

135 – References

Chapter I. Housing Shortage, Declining Household Formation, and Hidden
Dislodgements

Annie E. Casey Foundation. (2000). Meeting the Housing Needs of Families. Baltimore,
MD: Annie E. Casey Foundation.
Baer, W. C. (1986). The Evolution of Local and Regional Housing Studies. Journal of
the American Planning Association, 52(2), 172–184.
Brookings Institution. (2018). Unpacking the “housing shortage” puzzle: How does
housing enter and exit supply? Retrieved from
https://www.brookings.edu/research/unpacking-the-housing-shortage-puzzle/
California Department of Housing and Community Development. (2018). California’s
Housing Future: Challenges and Opportunities - Final Statewide Housing
Assessment 2025. Sacramento, CA: California Department of Housing and
Community Development.
California Department of Housing and Community Development. (2019). Regional
Housing Needs Allocation and Housing Elements. Retrieved from
http://www.hcd.ca.gov/community-development/housing-element/index.shtml
California Housing Partnership Corporation. (2019). California’s Housing Emergency
Update. San Francisco, CA: California Housing Partnership Corporation.
California Legislative Analyst’s Office. (2015). California’s High Housing Costs:
Causes and Consequences. Sacramento, CA: California Legislative Analyst’s
Office.
136 – References

California Legislative Analyst’s Office. (2016). Perspectives on Helping Low-Income
Californians Afford Housing. Sacramento, CA: California Legislative Analyst’s
Office.
Dunne, T. (2012). Household Formation and the Great Recession. Cleveland, OH:
Federal Reserve Bank of Cleveland.
Ermisch, J. (1999). Prices, Parents, and Young People’s Household Formation. Journal
of Urban Economics, 45(1), 47–71.
Ermisch, J., & Di Salvo, P. (1997). The Economic Determinants of Young People’s
Household Formation. Economica, 64(256), 627–644.
Freddie Mac. (2018). The Major Challenge of Inadequate U.S. Housing Supply.
Washington, D.C.: Freddie Mac.
Gabriel, S., & Painter, G. (2018). Why affordability matters. Regional Science and Urban
Economics. Online first.
Gabriel, S. A., & Rosenthal, S. S. (2015). The Boom, the Bust and the Future of
Homeownership: The Boom, the Bust and the Future of Homeownership. Real
Estate Economics, 43(2), 334–374.
Glaeser, E., & Gyourko, J. (2018). The Economic Implications of Housing Supply.
Journal of Economic Perspectives, 32(1), 3–30.
Haurin, D. R., Hendershott, P. H., & Kim, D. (1993). The Impact of Real Rents and
Wages on Household Formation. The Review of Economics and Statistics, 75(2),
284–293.
Haurin, D. R., & Rosenthal, S. S. (2007). The influence of household formation on
homeownership rates across time and race. Real Estate Economics, 35(4), 411–450.
137 – References

Joint Center for Housing Studies. (2017). America's Rental Housing: Expanding Options
for Diverse and Growing Demand. Cambridge, MA: Joint Center for Housing
Studies of Harvard University.
Joint Center for Housing Studies. (2018). The State of the Nation’s Housing, Cambridge,
MA: Joint Center for Housing Studies of Harvard University.
Lee, K. O., & Painter, G. (2013). What happens to household formation in a recession?
Journal of Urban Economics, 76, 93–109.
McClure, K. (2010). Are low-income housing tax credit developments locating where
there is a shortage of affordable units? Housing Policy Debate, 20(2), 153–171.
McKinsey Global Institute. (2016). A Tool Kit to Close California's Housing Gap: 3.5
Million Homes by 2025. New York, NY: McKinsey & Company.
Miron, J. R. (1989). Household formation, affordability, and housing policy. Population
Research and Policy Review, 8(1), 55–77.
Mutchler, J. E., & Krivo, L. J. (1989). Availability and Affordability: Household
Adaptation to a Housing Squeeze. Social Forces, 68(1), 241–261.
Myers, D. (2016). “Peak Millennials: Three Reinforcing Cycles that Amplify the Rise
and Fall of Millennial Urban Concentration.” Housing Policy Debate 26 (6): 928-
947.
Myers, D., Baer, W. C., & Choi, S.Y. (1996). The Changing Problem of Overcrowded
Housing. Journal of the American Planning Association, 62(1), 66–84.
Myers, D., & Lee, S. W. (1996). Immigration Cohorts and Residential Overcrowding in
Southern California. Demography, 33(1), 51–65.
138 – References

Myers, D., & H. Lee. (2016). “Changing Demographics and Future Urban Development,”
chapter 2, pp. 11-58, in George McCarthy, Gregory Ingraham and Samuel Moody,
eds., Land and the City, Cambridge, MA: Lincoln Institute for Land Policy.
Myers, D., Painter, G., Lee, H., & J. Park. (2016). Diverted Homeowners, the Rental
Crisis and Foregone Household Formation. Washington, D.C.: Research Institute
for Housing America.
Myers, D., J. Pitkin, and J. Park (2002). “Estimation of Housing Needs Amidst
Population Growth and Change,” Housing Policy Debate 13 (3): 567-96.
Mykyta, L., & Macartney, S. (2011). The Effects of Recession on Household
Composition: “Doubling Up” and Economic Well-Being. SEHSD Working Paper
Number 2011-4.
Mykyta, L., & Pilkauskas, N. (2016). Household composition and family wellbeing:
Exploring the relationship between doubling up and hardship. SEHSD Working
Paper 2016-10.
National Low Income Housing Coalition. (2014). Aligning Federal Low Income Housing
Programs with Housing Need. Washington, D.C.: National Low Income Housing
Coalition.
National Low Income Housing Coalition. (2019). Housing Needs By State. Washington,
D.C.: National Low Income Housing Coalition. Retrieved from
https://nlihc.org/housing-needs-by-state
National Low Income Housing Coalition. (2019). The Gap: A Shortage of Affordable
Homes. Washington, D.C.: National Low Income Housing Coalition.
139 – References

Nelson, A. C. (2013) Reshaping Metropolitan America: Development Trends and
Opportunities to 2030, Washington, D.C.: Island Press.
NEXT 10. (2018). Current State of the California Housing Market: A Comparative
Analysis. San Francisco, CA: NEXT 10.
Noll, P. F., O’Dell, W., Smith, M. T., & Sullivan, J. (1997). Florida’s Affordable
Housing Needs Assessment Methodology. Journal of the American Planning
Association, 63(4), 495–508.
Pew Research Center. (2013). A Rising Share of Young Adults Live in Their Parents’
Home. Washington, D.C.: Pew Research Center.
Pierse, N., Carter, K., Bierre, S., Law, D., & Howden-Chapman, P. (2016). Examining
the role of tenure, household crowding and housing affordability on psychological
distress, using longitudinal data. Journal of Epidemiology and Community Health,
70(10), 961–966.
PolicyLink. (2005). Expanding Opportunity: New Resources to Meet California’s
Housing Needs. Oakland, CA: PolicyLink.
Routhier, G. (2019). Beyond Worst Case Needs: Measuring the Breadth and Severity of
Housing Insecurity Among Urban Renters. Housing Policy Debate, 29(2), 235–249.
Schwartz, A. F. (2015). Housing Policy in the United States (3rd ed.). New York, NY:
Routledge.
Sierra Club California. (2018). Sierra Club California Housing Policy: Meeting Our
Housing Needs and Protecting the Environment. Sacramento, CA: Sierra Club
California.
140 – References

Skobba, K., & Goetz, E. G. (2015). Doubling up and the erosion of social capital among
very low income households. International Journal of Housing Policy, 15(2), 127–
147.
Southern California Association of Governments. (2019). Regional Housing Needs
Assessment (RHNA). Retrieved from
http://www.scag.ca.gov/programs/Pages/Housing.aspx
Struyk, R. J. (1987). The Housing Needs Assessment Model. Journal of the American
Planning Association, 53(2), 227–234.
Susin, S. (2007). Duration of Rent Burden as a Measure of Need. Cityscape: A Journal of
Policy Development and Research, 9(1), 157–174.
UCLA Lewis Center for Regional Policy Studies. (2019). Not Nearly Enough: California
Lacks Capacity to Meet Lofty Housing Goals. Los Angeles, CA: UCLA Lewis
Center for Regional Policy Studies.
Urban Institute. (2009). The impacts of foreclosures on families and communities.
Washington, D.C.: Urban Institute.
Urban Institute. (2017). The Housing Affordability Gap for Extremely Low-Income
Renters in 2014. Washington, D.C.: Urban Institute.
U.S. Census Bureau. (2017). The Changing Economics and Demographics of Young
Adulthood: 1975–2016. Washington, D.C.: U.S. Census Bureau.
U.S. Department of Housing and Urban Development. (2012). Foreclosure Counseling
Outcome Study: Final Report Housing Counseling Outcome Evaluation.
Washington, D.C.: U.S. Department of Housing and Urban Development.
141 – References

U.S. Department of Housing and Urban Development. (2016). Rental Market Dynamics:
2011–2013. Washington, D.C.
U.S. Department of Housing and Urban Development. (2017). Worst Case Housing
Needs. Washington, D.C.
Varady, D. (1996). “Local Housing Plans: Learning from Great Britain.” Housing Policy
Debate, 7(2), 253–292.
Whittington, L. A., & Peters, H. E. (1996). Economic Incentives for Financial and
Residential Independence. Demography, 33(1), 82–97.
Wiemers, E. E. (2014). The Effect of Unemployment on Household Composition and
Doubling Up. Demography, 51, 2155–2178.
Wright, B. R. E., Caspi, A., Moffitt, T. E., & Silva, P. A. (1998). Factors Associated with
Doubled‐Up Housing—a Common Precursor to Homelessness. Social Service
Review, 72(1), 92–111.
Zhou, Y., & Myers, D. (2010). Misleading Comparisons of Homeownership Rates when
the Variable Effect of Household Formation Is Ignored: Explaining Rising
Homeownership and the Homeownership Gap between Blacks and Asians in the US.
Urban Studies, 47(12), 2615–2640.
Zuk, M., & K. Chapple. (2016). Housing Production, Filtering and Displacement:
Untangling the Relationships. Berkeley, CA: Institute of Governmental Studies at
the University of California, Berkeley.

142 – References

Chapter II. Depressed Access to Affordable Housing Due to Higher-Income
Occupancy: Evidence from U.S. Metropolitan Areas, 2006 to 2016

Bean, J. A. (2012). Renters more often burdened by housing costs after recession: Nearly
half of all renters spent over 30 percent of income on housing by 2010 (The
Carsey School of Public Policy at the Scholars’ Repository No. 49). Durham, NH:
Carsey Institute.
Cameron, A. C., & Trivedi, P. K. (2010). Microeconometrics Using Stata (Revised ed.).
College Station, TX: Stata Press.
Colburn, G., & Allen, R. (2018). Rent burden and the Great Recession in the USA.
Urban Studies, 55(1), 226–243.
Collinson, R. (2011). Rental Housing Affordability Dynamics, 1990-2009. Cityscape: A
Journal of Policy Development and Research, 13(2), 71–103.
DiPasquale, D. (2011). Rental Housing: Current Market Conditions and the Role of
Federal Policy. Cityscape: A Journal of Policy Development and Research, 13(2),
57–70.
Eriksen, M. D., & Rosenthal, S. S. (2010). Crowd out effects of place-based subsidized
rental housing: New evidence from the LIHTC program. Journal of Public
Economics, 94(11–12), 953–966.
Eriksen, M. D., & Ross, A. (2015). Housing Vouchers and the Price of Rental Housing.
American Economic Journal: Economic Policy, 7(3), 154–176.
Fannie Mae. (2018). Are Affordability Perceptions Reducing Household Mobility and
Exacerbating the Housing Shortage? National Housing Survey Topic Analysis.
Washington, D.C.: Fannie Mae.
143 – References

Feins, J. D., & Lane, T. S. (1981). How Much for Housing? New Perspectives on
Affordability and Risk. Cambridge, MA: Abt Books.
Green, R. K. (2011). Thoughts on Rental Housing and Rental Housing Assistance.
Cityscape: A Journal of Policy Development and Research, 13(2), 39–55.
Herbert, C., Hermann, A., & McCue, D. (2018). Measuring Housing Affordability:
Assessing the 30-Percent of Income Standard. Cambridge, MA: Joint Center for
Housing Studies of Harvard University.
Joice, P. (2014). Measuring housing affordability. Cityscape: A Journal of Policy
Development and Research, 16(1), 299–307.
Joint Center for Housing Studies of Harvard University. (2006). Middle Market Rentals
Hiding in Plain Sight. Cambridge, MA: Joint Center for Housing Studies of
Harvard University.
Joint Center for Housing Studies of Harvard University. (2017). America’s Rental
Housing. Cambridge, MA: Joint Center for Housing Studies of Harvard
University.
Joint Center for Housing Studies of Harvard University. (2018). The State of the Nation’s
Housing. Cambridge, MA: Joint Center for Housing Studies of Harvard
University.
Kroll, C. A. (2013). The Great Recession and Housing Affordability (Fisher Center
Working Papers) (p. 38). Berkeley, CA: UC Berkeley.
Kutty, N. K. (2005). A new measure of housing affordability: Estimates and analytical
results. Housing Policy Debate, 16(1), 113–142.
144 – References

Lee, K. O., & Painter, G. (2013). What happens to household formation in a recession?
Journal of Urban Economics, 76, 93–109.
Lens, M. C. (2018). Extremely low-income households, housing affordability and the
Great Recession. Urban Studies, 55(8), 1615–1635.
Myers, D., Painter, G., Lee, H., & Park, J. (2016). Diverted Homeowners, the Rental
Crisis and Foregone Household Formation. Washington, D.C.: Research Institute
for Housing America.
Myers, D., & Park, J. (2019). “A Constant Quartile Mismatch Indicator of Changing
Rental Affordability in U.S. Metropolitan Areas, 2000 to 2016,” Cityscape: A
Journal of Policy Development and Research, 21(1), 139–176.
National Low Income Housing Coalition. (2019). The Gap: A Shortage of Affordable
Homes. Washington, D.C.: National Low Income Housing Coalition.
National Low Income Housing Coalition. (2018). Out of Reach: The High Cost of
Housing. Washington, D.C.: National Low Income Housing Coalition.
NYU Furman Center. (2017). 2017 National Rental Housing Landscape: Renting in the
Nation’s Largest Metros. New York, NY: NYU Furman Center.
Quigley, J. M. (2011). Rental Housing Assistance. Cityscape: A Journal of Policy
Development and Research, 13(2), 147–158.
Quigley, J. M., & Raphael, S. (2004). Is housing unaffordable? Why isn’t it more
affordable? The Journal of Economic Perspectives, 18(1), 191–214.
145 – References

Ruggles, Steven, Sarah Flood, Ronald Goeken, Josiah Grover, Erin Meyer, Jose Pacas,
and Matthew Sobek. (2018). Integrated Public Use Microdata Series (IPUMS)
USA: Version 8.0 [dataset]. Minneapolis, MN: University of Minnesota.
https://doi.org/10.18128/D010.V8.0.
Schwartz, A. F. (2015). Housing Policy in the United States (3rd ed.). New York, NY:
Routledge.
Shlay, A. B. (2015). Life and Liberty in the Pursuit of Housing: Rethinking Renting and
Owning in Post-Crisis America. Housing Studies, 30(4), 560–579.
Sinai, T., & Waldfogel, J. (2005). Do low-income housing subsidies increase the
occupied housing stock? Journal of Public Economics, 89(11–12), 2137–2164.
Skobba, K., & Goetz, E. G. (2015). Doubling up and the erosion of social capital among
very low income households. International Journal of Housing Policy, 15(2),
127–147.
Stone, Michael E. 2012. “Shelter Poverty.” In The Encyclopedia of Housing, edited by
Andrew T. Carswell. Thousand Oaks, CA: SAGE Publications.
Susin, S. (2007). Duration of Rent Burden as a Measure of Need. Cityscape: A Journal of
Policy Development and Research, 9(1), 157–174.
Urban Institute. (2017). The Housing Affordability Gap for Extremely Low-Income
Renters in 2014. Washington, D.C.: Urban Institute.
U.S. Department of Housing and Urban Development. (2008). Trends in Housing Costs:
1985 - 2005 and the 30-Percent-of-Income Standard. Washington, D.C.: U.S.
Department of Housing and Urban Development.
146 – References

U.S. Department of Housing and Urban Development. (2016). Rental Market Dynamics:
2011–2013. Washington, D.C.: U.S. Department of Housing and Urban
Development.
U.S. Department of Housing and Urban Development. (2017). Worst Case Housing
Needs: 2015 Report to Congress. Washington, D.C.: U.S. Department of Housing
and Urban Development.
U.S. Department of Housing and Urban Development. (2018a). Comprehensive Housing
Affordability Strategy Data. Washington, D.C.: U.S. Department of Housing and
Urban Development.
U.S. Department of Housing and Urban Development. (2018b). Low-income Housing
Tax Credit Database. Washington, D.C.: U.S. Department of Housing and Urban
Development.
U.S. Department of Housing and Urban Development. (2018c). Picture of Subsidized
Households. Washington, D.C.: U.S. Department of Housing and Urban
Development.
———. 2018. Rental Affordability Index (RAI)—National Housing Market Summary and
Data. Washington, D.C.: U.S. Department of Housing and Urban Development.
———. 2014. “Rental Burdens: Rethinking Affordability Measures.” PD&R Edge.
Washington, D.C.: U.S. Department of Housing and Urban Development.
https://www.huduser.gov/portal/pdredge/pdr_edge_ featd_article_092214.html
———. 2015. “Learning about CHAS Data: An Interview with Paul Joice.” PD&R Edge.
Washington, D.C.: U.S. Department of Housing and Urban Development.
https://www.huduser.gov/portal/pdredge/pdr_edge_trending_042015.html
147 – References

———. 2016. “HUD’s New Rental Affordability Index.” PD&R Edge. Washington,
D.C.: U.S. Department of Housing and Urban Development.
https://www.huduser.gov/portal/pdredge/pdr-edgetrending-110716.html
———. 2018. “FY 2018 Income Limits Documentation System.” HUD User Portal.
Washington, D.C.: U.S. Department of Housing and Urban Development.
https://www.huduser.gov/portal/datasets/il/il2018/select_Geography.odn
———. 2017. “Defining Housing Affordability.” PD&R Edge. Washington, D.C.: U.S.
Department of Housing and Urban Development.
https://www.huduser.gov/portal/pdredge/pdr-edge-featd-article-081417.html
Weicher, J. C., Eggers, F. J., & Moumen, F. (2017). The Long-Term Dynamics of
Affordable Rental Housing. Washington, D.C.: Hudson Institute.
Williamson, A. R., Smith, M. T., & Strambi-Kramer, M. (2009). Housing Choice
Vouchers, the Low-Income Housing Tax Credit, and the Federal Poverty
Deconcentration Goal. Urban Affairs Review, 45(1), 119–132.
Wooldridge, J. M. (2010). Econometric Analysis of Cross Section and Panel Data (2nd
ed.). Cambridge, MA: MIT Press.

148 – References

Chapter III. Geocoding Inaccuracies: A Case Study for Evaluation of the Low-
Income Housing Tax Credit Program Data

Adkins, A., Sanderford, A., & Pivo, G. (2017). How Location Efficient Is LIHTC?
Measuring and Explaining State-Level Achievement. Housing Policy Debate,
27(3), 335–355.
Baum-Snow, N., & Marion, J. (2009). The effects of low income housing tax credit
developments on neighborhoods. Journal of Public Economics, 93(5–6), 654–
666.
Been, V., Ellen, I. G., & O’Regan, K. (2019). Supply Skepticism: Housing Supply and
Affordability. Housing Policy Debate, 29(1), 25–40.
Bonner, M. R., Han, D., Nie, J., Rogerson, P., Vena, J. E., & Freudenheim, J. L. (2003).
Positional Accuracy of Geocoded Addresses in Epidemiologic Research.
Epidemiology, 14(4), 408–412.
Cummings, J. L., & DiPasquale, D. (1999). The Low‐Income Housing Tax Credit: An
Analysis of the First Ten Years. Housing Policy Debate, 10(2), 251–307.
Dawkins, C. (2013). The Spatial Pattern of Low Income Housing Tax Credit Properties:
Implications for Fair Housing and Poverty Deconcentration Policies. Journal of
the American Planning Association, 79(3), 222–234.
Deng, L. (2005). The cost‐effectiveness of the low‐income housing tax credit relative to
vouchers: Evidence from six metropolitan areas. Housing Policy Debate, 16(3–4),
469–511.
Di, W., & Murdoch, J. C. (2013). The impact of the low income housing tax credit
program on local schools. Journal of Housing Economics, 22(4), 308–320.
149 – References

Dillman, K.-N., Horn, K. M., & Verrilli, A. (2017). The What, Where, and When of
Place-Based Housing Policy’s Neighborhood Effects. Housing Policy Debate,
27(2), 282–305.
Ellen, I. G. (2018). What do we know about housing choice vouchers? Regional Science
and Urban Economics.
Ellen, Ingrid G., Horn, K. M., & O’Regan, K. M. (2016). Poverty concentration and the
Low Income Housing Tax Credit: Effects of siting and tenant composition.
Journal of Housing Economics, 34, 49–59.
Ellen, Ingrid Gould, & Horn, K. M. (2018). Points for Place: Can State Governments
Shape Siting Patterns of Low-Income Housing Tax Credit Developments?
Housing Policy Debate, 1–19.
Ellen, I. G., & O’Flaherty, B. (2007). Social Programs and Household Size: Evidence
from New York City. Population Research and Policy Review, 26(4), 387–409.
Ellen, I. G., & O’Regan, K. M. (2011). How low income neighborhoods change: Entry,
exit, and enhancement. Regional Science and Urban Economics, 41(2), 89–97.
Ellen, Ingrid Gould, Horn, K. M., & Kuai, Y. (2018). Gateway to Opportunity?
Disparities in Neighborhood Conditions Among Low-Income Housing Tax Credit
Residents. Housing Policy Debate, 1–20.
Eriksen, M. D. (2009). The market price of Low-Income Housing Tax Credits. Journal of
Urban Economics, 66(2), 141–149.
Eriksen, M. D. (2017). Difficult Development Areas and the supply of subsidized
housing. Regional Science and Urban Economics, 64, 68–80.
150 – References

Eriksen, M. D., & Lang, B. J. (2018). Overview and proposed reforms of the low-income
housing tax credit program. Regional Science and Urban Economics.
Eriksen, M. D., & Rosenthal, S. S. (2010). Crowd out effects of place-based subsidized
rental housing: New evidence from the LIHTC program. Journal of Public
Economics, 94(11–12), 953–966.
Eriksen, M. D., & Ross, A. (2015). Housing Vouchers and the Price of Rental Housing.
American Economic Journal: Economic Policy, 7(3), 154–176.
Freedman, M., & Owens, E. G. (2011). Low-income housing development and crime.
Journal of Urban Economics, 70(2–3), 115–131.
Freeman, L. (2004). Siting Affordable Housing: Location and Neighborhood Trends of
Low Income Housing Tax Credit Developments in the 1990s. Washington, D.C.:
The Brookings Institution.
Freemark, Y. (2018). Challenges in the Creation of Mixed-Use Affordable Housing:
Measuring and Explaining Its Limited Prevalence. Housing Policy Debate, 28(6),
1004–1021.
Joint Center for Housing Studies of Harvard University. (2018). The State of the Nation’s
Housing. Cambridge, MA: Joint Center for Housing Studies of Harvard
University.
Horn, K. M., & O’Regan, K. M. (2011). The low income housing tax credit and racial
segregation. Housing Policy Debate, 21(3), 443–473.
Jones, R. R., DellaValle, C. T., Flory, A. R., Nordan, A., Hoppin, J. A., Hofmann, J. N.,
… Ward, M. H. (2014). Accuracy of Residential Geocoding in the Agricultural
Health Study. International Journal of Health Geographics, 13(1), 37.
151 – References

Khadduri, J. (2013). Creating Balance in the Locations of LIHTC Developments: The
Role of Qualified Allocation Plans. Washington, D.C.: Poverty and Race
Research Action Council.
Khan, S. (2018). Positional Accuracy of Geocoding from Residential Postal Codes
versus Full Street Addresses. Health Reports, 29(82), 8.
Lang, B. J. (2012). Location incentives in the low-income housing tax credit: Are
qualified census tracts necessary? Journal of Housing Economics, 21(2), 142–
150.
Lang, B. J. (2015). Input distortions in the Low-Income Housing Tax Credit: Evidence
from building size. Regional Science and Urban Economics, 52, 119–128.
Lens, M. C., & Reina, V. (2016). Preserving Neighborhood Opportunity: Where Federal
Housing Subsidies Expire. Housing Policy Debate, 26(4–5), 714–732.
Lens, M., McClure, K., & Mast, B. (2019). Does Jobs Proximity Matter in the Housing
Choice Voucher Program? Cityscape: A Journal of Policy Development and
Research, 21(1), 145–162.
Luque, J. (2018). Assessing the role of TIF and LIHTC in an equilibrium model of
affordable housing development. Regional Science and Urban Economics.
Malpezzi, S., & Vandell, K. (2002). Does the low-income housing tax credit increase the
supply of housing? Journal of Housing Economics, 11(4), 360–380.
McClure, K. (1990). Low and Moderate Income Housing Tax Credits Calculating Their
Value. Journal of the American Planning Association, 56(3), 363–369.
McClure, K. (2006). The low‐income housing tax credit program goes mainstream and
moves to the suburbs. Housing Policy Debate, 17(3), 419–446.
152 – References

McClure, K. (2010). Are low-income housing tax credit developments locating where
there is a shortage of affordable units? Housing Policy Debate, 20(2), 153–171.
McClure, K., & Johnson, B. (2015). Housing Programs Fail to Deliver on Neighborhood
Quality, Reexamined. Housing Policy Debate, 25(3), 463–496.
McClure, K. (2019). What Should Be the Future of the Low-Income Housing Tax Credit
Program? Housing Policy Debate, 29(1), 65–81.
McClure, K., Schwartz, A. F., & Taghavi, L. B. (2015). Housing Choice Voucher
Location Patterns a Decade Later. Housing Policy Debate, 25(2), 215–233.
National Housing Trust. (2015). Preservation and Opportunity Neighborhoods in the
Low Income Housing Tax Credit.
Nedwick, T., & Burnett, K. (2015). How Can the LIHTC Program Most Effectively Be
Used To Provide Affordable Rental Housing Near Transit? Cityscape, 17(2), 113–
137.
Oakley, D. (2008). Locational Patterns of Low-Income Housing Tax Credit
Developments: A Sociospatial Analysis of Four Metropolitan Areas. Urban
Affairs Review, 43(5), 599–628.
O’Regan, K. M., & Horn, K. M. (2013). What Can We Learn About the Low-Income
Housing Tax Credit Program by Looking at the Tenants? Housing Policy Debate,
23(3), 597–613.
Ratcliffe, J. H. (2001). On the accuracy of TIGER-type geocoded address data in relation
to cadastral and census areal units. International Journal of Geographical
Information Science, 15(5), 473–485.
153 – References

Ruggles, Steven, Sarah Flood, Ronald Goeken, Josiah Grover, Erin Meyer, Jose Pacas,
and Matthew Sobek. (2018). Integrated Public Use Microdata Series (IPUMS)
USA: Version 8.0 [dataset]. Minneapolis, MN: University of Minnesota.
https://doi.org/10.18128/D010.V8.0.
Schootman, M., Sterling, D. A., Struthers, J., Yan, Y., Laboube, T., Emo, B., & Higgs, G.
(2007). Positional Accuracy and Geographic Bias of Four Methods of Geocoding
in Epidemiologic Research. Annals of Epidemiology, 17(6), 464–470.
Schwartz, A. F. (2015). Housing Policy in the United States (3rd ed.). New York, NY:
Routledge.
Schwartz, A. F. (2016). The Low-Income Housing Tax Credit, Community Development,
and Fair Housing: A Response to Orfield et al. Housing Policy Debate, 26(2),
276–283.
Schwartz, A., McClure, K., & Taghavi, L. B. (2016). Vouchers and Neighborhood
Distress: The Unrealized Potential for Families With Housing Choice Vouchers
To Reside in Neighborhoods With Low Levels of Distress. Cityscape, 18(3), 207.
U.S. Department of Housing and Urban Development. (2013). The Feasibility of
Developing a National Parcel Database: County Data Records Project Final
Report. Washington, D.C.
U.S. Census Bureau. (2018). TIGER/Line Shapefiles and TIGER/Line Files. Washington,
D.C.: U.S. Census Bureau. Retrieved from https://www.census.gov/geo/maps-
data/data/tiger-line.html
154 – References

U.S. Department of Housing and Urban Development. (2006). Updating the Low-Income
Housing Tax Credit (LIHTC) Database: Projects Placed in Service through 2003.
Washington, D.C.: U.S. Department of Housing and Urban Development.
U.S. Department of Housing and Urban Development. (2019). Low-income Housing Tax
Credit Database. Washington, D.C.: U.S. Department of Housing and Urban
Development. Retrieved from https://lihtc.huduser.gov/
Ward, M. H., Nuckols, J. R., Giglierano, J., Bonner, M. R., Wolter, C., Airola, M., …
Hartge, P. (2005). Positional Accuracy of Two Methods of Geocoding.
Epidemiology, 16(4), 542–547.
Williamson, A. R., Smith, M. T., & Strambi-Kramer, M. (2009). Housing Choice
Vouchers, the Low-Income Housing Tax Credit, and the Federal Poverty
Deconcentration Goal. Urban Affairs Review, 45(1), 119–132.

155 – Appendix

Appendix

Introduction

Appendix A. Data, Method, Study Area, and Study Period of Each Essay
Data Study Area Study Period Method
Essay
#1
• Census and ACS
Public Use
Microdata
• Nation as a
Whole
• Top 50
Metropolitan
Statistical
Areas
• 2000, and
2006 through
2017
• Housing−Demo
graphic Method
• Correlation
Analysis
Essay
#2
• Census and ACS
Public Use
Microdata
• Subsidized
Housing Data
• Nation as a
Whole
• Top 200
Metropolitan
Statistical
Areas
• 2006 through
2016
• Housing−House
hold−Classificati
on Method
• Cross-sectional
OLS and Panel
Regressions
Essay
#3
• Census and ACS
Public Use
Microdata
• Subsidized
Housing Data
• Census Bureau’s
TIGER Shapefile
• LA County
Assessor’s Land
Parcel Data
• Los Angeles
County as a
Whole
• Neighborhoods
(Census Tracts,
Block-groups,
and Blocks) in
LA County
• 2016 • Collection of
Subsidized
Rental Housing
Data for LA
County
• Transit-
accessibility
Analysis
Note: In the second essay, explanatory variables for regression models are built from
multiple data sources in addition to main data as shown in this table. They include i) new
construction – Census Bureau’s Building Permits Survey, ii) employment growth –
Bureau of Economic Analysis (BEA)’s Annual Employment Data, and iii) housing
subsidies – U.S. Department of Housing and Urban Development (HUD)’s A Picture of
Subsidized Households.

156 – Appendix

Chapter I. Housing Shortage, Declining Household Formation, and Hidden
Dislodgements

Appendix I –A. Summary of Housing Need Estimates, United States and California
Author (Year) Current Needs
(Future Needs)
Current Year
(Future Year)
Features
(a) United States
U.S. Department of
Housing and Urban
Development (2017)
8.2 million 2015 Very low-income
(<50% of AMI)
National Low Income
Housing Coalition (2019)
7.4 million 2017 Very low-income
(<50% of AMI)
Urban Institute (2017) 6.4 million 2014 Extremely low-
income (<30%)
(b) California
California Department of
Housing and Community
Development (2018)
(1.8 million) (2025) Future needs only
California Legislative
Analyst’s Office (2016)
1.7 million 2014 Very low-income
(<50% of AMI)
California Housing
Partnership Corporation
(2019)
1.4 million 2017 Very low-income
(<50% of AMI)
UCLA Lewis Center for
Regional Policy Studies
(2019)
(0.7 million) (2025) Relative to
Governor
Newsom’s goal
McKinsey (2016) 2.0 million
(1.5 million)
2014
(2025)
New York standard
applied
NEXT 10 (2018) 2.3 million 2017 Based on
McKinsey’s
method
PolicyLink (2005) $29 billion
(plus $2.3 billion
per year ahead)
2004
(Future year not
specified)
Capital investment
estimated
Notes: This summary table is not an exhaustive literature review, but it includes most of
the recent estimates that are often referred to in public debates in California and the
nation as a whole.
157 – Appendix

Appendix I –B. Housing Occupancy Rates by Age, United States, 2000, 2006, 2011, and 2017

(a) Absolute Count (by age)

(b) Household Formation Rates (HHs per capita, by age)

Sources: Census 2000 5-percent IPUMS; 2006, 2011, and 2017 ACS 1-year IPUMS.
2000 2006 2011 2017 2000 2006 2011 2017 2000 2006 2011 2017 2000 2006 2011 2017
15-19 19,930,188 21,725,748 21,757,266 21,383,762 575,051 497,838 359,503 334,090 77,774 63,207 41,757 46,974 497,277 434,631 317,746 287,116
20-24 19,055,617 20,877,870 22,144,681 22,028,331 4,797,912 4,848,111 4,319,315 3,992,414 853,908 891,233 581,405 573,815 3,944,004 3,956,878 3,737,910 3,418,599
25-29 19,178,062 20,355,480 21,044,256 23,029,406 8,145,323 8,512,194 8,106,834 8,241,612 2,952,046 3,324,686 2,603,466 2,464,408 5,193,277 5,187,508 5,503,368 5,777,204
30-34 20,325,200 19,492,336 20,426,089 21,843,417 9,898,854 9,497,593 9,554,273 9,816,826 5,243,638 5,091,266 4,443,279 4,472,598 4,655,216 4,406,327 5,110,994 5,344,228
35-39 23,111,071 21,315,494 19,779,103 21,460,801 11,721,331 11,027,789 9,995,602 10,507,948 7,430,644 6,922,680 5,652,485 5,726,091 4,290,687 4,105,109 4,343,117 4,781,857
40-44 22,808,914 22,707,212 21,197,499 19,864,805 12,095,069 12,094,313 11,138,294 10,144,759 8,363,970 8,352,759 7,051,166 6,162,569 3,731,099 3,741,554 4,087,128 3,982,190
45-49 20,186,287 22,792,322 22,146,122 20,929,582 11,188,245 12,571,632 11,932,701 11,094,866 8,192,093 9,189,379 8,206,370 7,369,002 2,996,152 3,382,253 3,726,331 3,725,864
50-54 17,419,082 20,477,825 22,501,599 21,366,807 10,089,889 11,525,839 12,412,854 11,561,256 7,736,740 8,765,191 9,042,890 8,166,429 2,353,149 2,760,648 3,369,964 3,394,827
55-59 13,380,391 18,081,601 20,165,898 21,739,172 7,866,608 10,406,314 11,383,540 12,124,617 6,229,609 8,202,965 8,604,877 8,914,419 1,636,999 2,203,349 2,778,663 3,210,198
60-64 10,792,431 13,564,709 17,884,213 20,274,901 6,476,758 7,913,043 10,368,435 11,568,536 5,190,009 6,375,426 8,191,623 8,836,286 1,286,749 1,537,617 2,176,812 2,732,250
65-69 9,455,337 10,343,642 12,866,613 16,933,484 5,939,477 6,186,618 7,681,173 10,019,968 4,817,759 5,045,784 6,188,155 7,911,119 1,121,718 1,140,834 1,493,018 2,108,849
70-74 8,945,317 8,600,286 9,644,749 12,804,945 5,765,895 5,272,613 5,908,050 7,710,257 4,646,731 4,285,397 4,795,830 6,215,359 1,119,164 987,216 1,112,220 1,494,898
75-79 7,357,757 7,474,132 7,368,044 8,848,547 4,970,429 4,723,915 4,615,312 5,431,490 3,891,750 3,807,263 3,708,822 4,386,902 1,078,679 916,652 906,490 1,044,588
80-84 4,997,500 5,760,198 5,806,218 5,947,742 3,417,076 3,661,878 3,721,124 3,688,456 2,526,762 2,799,269 2,891,628 2,875,990 890,314 862,609 829,496 812,466
85+ 4,224,865 5,015,629 5,702,332 6,266,767 2,532,184 2,877,698 3,494,705 3,825,672 1,665,112 1,958,294 2,372,554 2,656,704 867,072 919,404 1,122,151 1,168,968
All 15+ 221,168,019 238,584,484 250,434,682 264,722,469 105,480,101 111,617,388 114,991,715 120,062,767 69,818,545 75,074,799 74,376,307 76,778,665 35,661,556 36,542,589 40,615,408 43,284,102
(a) Population (b) Households (c) Owner Households (d) Renter Households
Age
2000 2006 2011 2017 2000 2006 2011 2017 2000 2006 2011 2017
15-19 2.9 2.3 1.7 1.6 0.4 0.3 0.2 0.2 2.5 2.0 1.5 1.3
20-24 25.2 23.2 19.5 18.1 4.5 4.3 2.6 2.6 20.7 19.0 16.9 15.5
25-29 42.5 41.8 38.5 35.8 15.4 16.3 12.4 10.7 27.1 25.5 26.2 25.1
30-34 48.7 48.7 46.8 44.9 25.8 26.1 21.8 20.5 22.9 22.6 25.0 24.5
35-39 50.7 51.7 50.5 49.0 32.2 32.5 28.6 26.7 18.6 19.3 22.0 22.3
40-44 53.0 53.3 52.5 51.1 36.7 36.8 33.3 31.0 16.4 16.5 19.3 20.0
45-49 55.4 55.2 53.9 53.0 40.6 40.3 37.1 35.2 14.8 14.8 16.8 17.8
50-54 57.9 56.3 55.2 54.1 44.4 42.8 40.2 38.2 13.5 13.5 15.0 15.9
55-59 58.8 57.6 56.4 55.8 46.6 45.4 42.7 41.0 12.2 12.2 13.8 14.8
60-64 60.0 58.3 58.0 57.1 48.1 47.0 45.8 43.6 11.9 11.3 12.2 13.5
65-69 62.8 59.8 59.7 59.2 51.0 48.8 48.1 46.7 11.9 11.0 11.6 12.5
70-74 64.5 61.3 61.3 60.2 51.9 49.8 49.7 48.5 12.5 11.5 11.5 11.7
75-79 67.6 63.2 62.6 61.4 52.9 50.9 50.3 49.6 14.7 12.3 12.3 11.8
80-84 68.4 63.6 64.1 62.0 50.6 48.6 49.8 48.4 17.8 15.0 14.3 13.7
85+ 59.9 57.4 61.3 61.0 39.4 39.0 41.6 42.4 20.5 18.3 19.7 18.7
All 15+ 47.7 46.8 45.9 45.4 31.6 31.5 29.7 29.0 16.1 15.3 16.2 16.4
Age
(b / a x 100)
Total Household Formation Rates
(HHs per capita)
(c / a x 100)
Owner Household Formation Rates
(Owner HHs per capita)
(d / a x 100)
Renter Household Formation Rates
(Renter HHs per capita)
158 – Appendix

Appendix I –C. Annual Estimates of Diversions and Dislodgements, by Tenure and Structure Type, United States, 2006 to 2017

(a) Actual Number of Households

(b) Expected Number of Households

(c = a – b) Actual Less Expected Number of Households

Sources: Census 2000 5-percent IPUMS; 2006 to 2017 ACS 1-year IPUMS.
2000 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Total 105,480,101 111,617,388 112,377,963 113,101,361 113,616,192 114,567,449 114,991,715 115,969,580 116,290,974 117,259,422 118,208,212 118,860,053 120,062,767
Owner occupied 69,818,545 75,074,799 75,511,557 75,341,616 74,929,333 74,947,705 74,376,307 74,227,189 73,933,462 74,083,394 74,637,866 75,102,526 76,778,665
Single-family 60,072,901 65,460,815 66,019,909 65,978,708 65,663,599 65,863,753 65,430,322 65,385,655 65,164,415 65,307,680 65,832,282 66,378,521 67,894,737
Multi-family 3,804,353 4,204,162 4,258,483 4,215,779 4,154,211 3,978,524 3,918,141 3,900,147 3,882,237 3,976,280 4,010,432 3,963,701 4,102,085
Other 5,941,291 5,409,822 5,233,165 5,147,129 5,111,523 5,105,428 5,027,844 4,941,387 4,886,810 4,799,434 4,795,152 4,760,304 4,781,843
Renter occupied 35,661,556 36,542,589 36,866,406 37,759,745 38,686,859 39,619,744 40,615,408 41,742,391 42,357,512 43,176,028 43,570,346 43,757,527 43,284,102
Single-family 10,611,745 11,340,198 11,672,886 12,260,985 12,874,279 13,284,588 13,863,506 14,518,915 14,823,351 15,113,608 15,108,508 15,183,022 14,923,678
Multi-family 23,479,463 23,351,052 23,393,611 23,662,433 23,982,585 24,483,299 24,871,223 25,267,588 25,595,175 26,072,836 26,480,186 26,618,553 26,432,816
Other 1,570,348 1,851,339 1,799,909 1,836,327 1,829,995 1,851,857 1,880,679 1,955,888 1,938,986 1,989,584 1,981,652 1,955,952 1,927,608
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Total 113,650,566 114,819,231 116,060,642 117,161,032 118,596,503 119,848,365 121,165,147 122,403,711 123,937,412 125,436,048 126,560,742 128,039,896
Owner occupied 75,220,985 76,040,484 76,865,627 77,546,051 78,522,785 79,319,815 80,171,604 80,963,392 81,914,480 82,877,318 83,622,532 84,545,054
Single-family 64,740,587 65,442,044 66,137,442 66,705,913 67,539,679 68,206,606 68,913,216 69,569,723 70,365,572 71,174,336 71,793,774 72,564,515
Multi-family 4,184,296 4,242,586 4,306,229 4,360,069 4,443,045 4,509,432 4,584,444 4,654,751 4,737,555 4,818,248 4,884,412 4,966,483
Other 6,296,102 6,355,854 6,421,956 6,480,068 6,540,062 6,603,777 6,673,944 6,738,918 6,811,354 6,884,734 6,944,346 7,014,056
Renter occupied 38,429,581 38,778,747 39,195,015 39,614,981 40,073,718 40,528,551 40,993,543 41,440,319 42,022,932 42,558,730 42,938,211 43,494,842
Single-family 11,293,738 11,385,684 11,489,358 11,585,280 11,681,869 11,790,859 11,898,745 12,003,946 12,145,492 12,278,750 12,368,422 12,508,015
Multi-family 25,485,068 25,730,834 26,030,275 26,341,792 26,696,866 27,029,337 27,373,882 27,702,956 28,129,321 28,517,240 28,797,291 29,199,727
Other 1,650,776 1,662,228 1,675,382 1,687,909 1,694,983 1,708,355 1,720,916 1,733,417 1,748,119 1,762,740 1,772,498 1,787,099
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Total -2,033,178 -2,441,268 -2,959,281 -3,544,840 -4,029,054 -4,856,650 -5,195,567 -6,112,737 -6,677,990 -7,227,836 -7,700,689 -7,977,129
Owner occupied -146,186 -528,927 -1,524,011 -2,616,718 -3,575,080 -4,943,508 -5,944,415 -7,029,930 -7,831,086 -8,239,452 -8,520,006 -7,766,389
Single-family 720,228 577,865 -158,734 -1,042,314 -1,675,926 -2,776,284 -3,527,561 -4,405,308 -5,057,892 -5,342,054 -5,415,253 -4,669,778
Multi-family 19,866 15,897 -90,450 -205,858 -464,521 -591,291 -684,297 -772,514 -761,275 -807,816 -920,711 -864,398
Other -886,280 -1,122,689 -1,274,827 -1,368,545 -1,434,634 -1,575,933 -1,732,557 -1,852,108 -2,011,920 -2,089,582 -2,184,042 -2,232,213
Renter occupied -1,886,992 -1,912,341 -1,435,270 -928,122 -453,974 86,857 748,848 917,193 1,153,096 1,011,616 819,316 -210,740
Single-family 46,460 287,202 771,627 1,288,999 1,602,719 2,072,647 2,620,170 2,819,405 2,968,116 2,829,758 2,814,600 2,415,663
Multi-family -2,134,016 -2,337,223 -2,367,842 -2,359,207 -2,213,567 -2,158,114 -2,106,294 -2,107,781 -2,056,485 -2,037,054 -2,178,738 -2,766,911
Other 200,563 137,681 160,945 142,086 156,874 172,324 234,972 205,569 241,465 218,912 183,454 140,509
159 – Appendix

Appendix I –D. The 50 Most Populous Metropolitan Areas in the United States, 2017
Rank Metropolitan Area Full Name 2017 Pop
1 New York-Northern New Jersey-Long Island, NY-NJ-PA Metro 20,182,305
2 Los Angeles-Long Beach-Santa Ana, CA Metro Area 13,340,068
3 Chicago-Joliet-Naperville, IL-IN-WI Metro Area 9,550,108
4 Dallas-Fort Worth-Arlington, TX Metro Area 7,102,165
5 Houston-Sugar Land-Baytown, TX Metro Area 6,656,946
6 Philadelphia-Camden-Wilmington, PA-NJ-DE-MD Metro Area 6,069,875
7 Washington-Arlington-Alexandria, DC-VA-MD-WV Metro Area 6,098,283
8 Miami-Fort Lauderdale-Pompano Beach, FL Metro Area 6,012,331
9 Atlanta-Sandy Springs-Marietta, GA Metro Area 5,709,731
10 Boston-Cambridge-Quincy, MA-NH Metro Area 4,774,321
11 San Francisco-Oakland-Fremont, CA Metro Area 4,656,132
12 Detroit-Warren-Livonia, MI Metro Area 4,302,043
13 Riverside-San Bernardino-Ontario, CA Metro Area 4,489,159
14 Phoenix-Mesa-Glendale, AZ Metro Area 4,574,531
15 Seattle-Tacoma-Bellevue, WA Metro Area 3,733,580
16 Minneapolis-St. Paul-Bloomington, MN-WI Metro Area 3,524,583
17 San Diego-Carlsbad-San Marcos, CA Metro Area 3,299,521
18 St. Louis, MO-IL Metro Area 2,812,313
19 Tampa-St. Petersburg-Clearwater, FL Metro Area 2,975,225
20 Baltimore-Towson, MD Metro Area 2,797,407
21 Denver-Aurora-Broomfield, CO Metro Area 2,814,330
22 Pittsburgh, PA Metro Area 2,353,045
23 Portland-Vancouver-Hillsboro, OR-WA Metro Area 2,390,244
24 San Antonio-New Braunfels, TX Metro Area 2,381,828
25 Sacramento--Arden-Arcade--Roseville, CA Metro Area 2,274,194
26 Orlando-Kissimmee-Sanford, FL Metro Area 2,387,138
27 Cincinnati-Middletown, OH-KY-IN Metro Area 2,159,329
28 Cleveland-Elyria-Mentor, OH Metro Area 2,060,810
29 Kansas City, MO-KS Metro Area 2,088,269
30 Las Vegas-Paradise, NV Metro Area 2,114,801
31 San Jose-Sunnyvale-Santa Clara, CA Metro Area 1,976,836
32 Columbus, OH Metro Area 2,021,632
33 Charlotte-Gastonia-Rock Hill, NC-SC Metro Area 2,426,363
34 Indianapolis-Carmel, IN Metro Area 1,988,152
35 Austin-Round Rock-San Marcos, TX Metro Area 2,000,860
36 Virginia Beach-Norfolk-Newport News, VA-NC Metro Area 1,723,351
37 Providence-New Bedford-Fall River, RI-MA Metro Area 1,613,070
38 Nashville-Davidson--Murfreesboro--Franklin, TN Metro Area 1,830,298
39 Milwaukee-Waukesha-West Allis, WI Metro Area 1,575,747
40 Jacksonville, FL Metro Area 1,449,481
41 Memphis, TN-MS-AR Metro Area 1,343,572
42 Louisville/Jefferson County, KY-IN Metro Area 1,279,335
43 Richmond, VA Metro Area 1,271,142
44 Oklahoma City, OK Metro Area 1,358,452
45 Hartford-West Hartford-East Hartford, CT Metro Area 1,211,324
46 New Orleans-Metairie-Kenner, LA Metro Area 1,262,888
47 Raleigh-Cary, NC Metro Area 1,273,568
48 Buffalo-Niagara Falls, NY Metro Area 1,135,230
49 Salt Lake City, UT Metro Area 1,170,266
50 Birmingham-Hoover, AL Metro Area 1,145,647
Sources: U.S. Census Bureau, 2017 American Community Survey 1-year Estimate,
S0101 File.

160 – Appendix

Appendix I –E. Additional Tests: Actual and Expected Number of Households in 2017, United States, by Tenure and Structure Type,
Based on Alternative 1980, 1990, and 2000 Base Years (Unit: Thousands, %)

(a)
Actual Number of Households

(b)
Expected Number of Households
in 2017

(c = a – b)
Actual Less Expected Number
of Households
1980 1990 2000 2017
1980-
based
1990-
based
2000-
based

1980-
based
1990-
based
2000-
based
Total 80,467 91,746
105,48
0
120,06
3
130,397 128,066 128,040 -10,334 -8,003 -7,977
Owner
occupied
52,304 58,922 69,819 76,779 83,884 82,451 84,545 -7,106 -5,672 -7,766
Single-
family
45,053 50,452 60,073 67,895 71,859 70,693 72,565 -3,964 -2,798 -4,670
Multifamily 3,986 3,202 3,804 4,102 7,420 4,792 4,966 -3,318 -690 -864
Other 3,265 5,269 5,941 4,782 4,605 6,966 7,014 177 -2,184 -2,232
Renter
occupied
28,163 32,824 35,662 43,284 46,513 45,615 43,495 -3,229 -2,331 -211
Single-
family
8,601 9,845 10,612 14,924 13,850 13,175 12,508 1,073 1,749 2,416
Multifamily 18,798 21,356 23,479 26,433 31,634 30,336 29,200 -5,201 -3,903 -2,767
Other 764 1,624 1,570 1,928 1,029 2,104 1,787 899 -176 141
Notes: The expected number of households in 2017 is calculated by multiplying the number of population in 2017 by age and
race/ethnicity by age−race/ethnicity−structure type-specific headship rates in 1980, 1990, or 2000. ‘Other’ category includes mobile
homes, boat, RV, van, etc.
Sources: Census 1980, 1990, and 2000 5-percent IPUMS and 2017 ACS 1-year IPUMS.
161 – Appendix

Appendix I –F. Additional Tests: Relationship between Young Headship Rate of
Householders Age 25 to 34 and Incidence of Rent-burden among Young Adults (25 to
34), Largest 50 Metropolitan Areas, 2017

(a) 30%+ Rent-burden

(b) 50%+ Rent-burden

Sources: 2017 ACS 1-year IPUMS.

162 – Appendix

II. Depressed Access to Affordable Housing Due to Higher-Income Occupancy: Evidence from U.S. Metropolitan Areas, 2006 to 2016

Appendix II –A. Renter Households and Renter-occupied Housing Units, By Income and Gross Rent, United States, 1980 to 2016

Notes: HH is a household. HU is a housing unit. Universe is Very Low-income (<50% of AMI, VLI) or Extremely Low-income (<30% of AMI, ELI) renter households in the
nation in each survey year. National median household income each year is based on Census and ACS IPUMS data. Thresholds of income-to-AMI are based on HUD’s “at or
below” definition. A rental unit is defined affordable when it costs at or below 30% of the top income threshold of an income group.
Sources: 1980, 1990, and 2000 Decennial Census; 2006 to 2016 ACS 1-year Estimate Integrated Public Use Microdata (IPUMS) Sample files.
1980 1990 2000 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
All Renter HHs (or Rental HUs) 28,162,820 32,824,172 35,661,556 36,542,589 36,866,406 37,759,745 38,686,859 39,619,744 40,615,408 41,742,391 42,357,512 43,176,028 43,570,346 43,757,527
Renter HHs by Income
(a) 0-30% of AMI 6,245,700 7,625,075 8,408,642 9,353,186 9,373,847 9,696,292 9,752,499 10,059,564 10,510,969 10,602,593 10,600,283 10,652,106 10,485,778 10,576,181
(b) 30-50% of AMI 4,344,540 4,923,613 5,634,536 6,259,579 6,335,817 6,465,038 6,544,436 6,817,787 6,900,883 6,972,771 7,021,595 7,124,698 7,287,563 7,111,001
50%+ of AMI 17,572,580 20,275,484 21,618,378 20,929,824 21,156,742 21,598,415 22,389,924 22,742,393 23,203,556 24,167,027 24,735,634 25,399,224 25,797,005 26,070,345
Rental HUs by Gross Rent
Affordable to 0-30% of AMI 5,354,080 5,114,314 6,057,156 5,375,062 5,425,600 5,381,716 5,180,870 5,055,635 5,141,512 5,301,985 5,347,334 5,329,787 5,453,515 5,529,719
Affordable to 30-50% of AMI 6,872,900 7,362,931 9,069,653 7,167,588 7,461,707 7,626,652 6,559,328 6,167,279 6,197,245 6,501,634 6,582,046 6,581,793 6,879,325 7,177,311
Affordable to 50%+ of AMI 15,935,840 20,346,927 20,534,747 23,999,939 23,979,099 24,751,377 26,946,661 28,396,830 29,276,651 29,938,772 30,428,132 31,264,448 31,237,506 31,050,497
Matches between HHs and HUs
0-30% of AMI HHs
(c) HUs Affordable to 0-30% of AMI 2,228,380 2,814,451 3,029,344 2,975,021 3,011,760 3,028,068 2,883,533 2,773,865 2,863,956 2,957,813 2,985,049 2,966,308 3,026,252 3,087,602
(d) HUs Affordable to 30-50% of AMI 1,926,220 2,178,665 2,581,891 2,431,617 2,481,120 2,582,844 2,321,076 2,274,658 2,348,104 2,424,060 2,448,893 2,469,459 2,495,381 2,556,374
HUs Affordable to 50%+ of AMI 2,091,100 2,631,959 2,797,407 3,946,548 3,880,967 4,085,380 4,547,890 5,011,041 5,298,909 5,220,720 5,166,341 5,216,339 4,964,145 4,932,205
30-50% of AMI HHs
(e) HUs Affordable to 0-30% of AMI 918,980 808,089 1,028,256 843,012 860,182 851,503 812,181 830,091 822,834 824,791 838,694 819,707 872,466 859,371
(f) HUs Affordable to 30-50% of AMI 1,450,900 1,688,444 2,027,216 1,775,372 1,874,026 1,895,564 1,666,856 1,603,614 1,586,167 1,671,073 1,684,064 1,680,707 1,759,866 1,795,525
HUs Affordable to 50%+ of AMI 1,974,660 1,923,159 2,062,409 2,633,285 2,659,276 2,692,069 2,798,657 2,957,964 2,981,858 3,019,919 3,053,300 3,110,113 3,183,784 3,048,087
50%+ of AMI HHs
HUs Affordable to 0-30% of AMI 2,206,720 1,491,774 1,999,556 1,557,029 1,553,658 1,502,145 1,485,156 1,451,679 1,454,722 1,519,381 1,523,591 1,543,772 1,554,797 1,582,746
HUs Affordable to 30-50% of AMI 3,495,780 3,495,822 4,460,546 2,960,599 3,106,561 3,148,244 2,571,396 2,289,007 2,262,974 2,406,501 2,449,089 2,431,627 2,624,078 2,825,412
HUs Affordable to 50%+ of AMI 11,870,080 15,287,888 15,158,276 16,412,196 16,496,523 16,948,026 18,333,372 19,001,707 19,485,860 20,241,145 20,762,954 21,423,825 21,618,130 21,662,187
VLI Rental Availability (%) 61.6 59.7 61.7 51.4 52.4 51.7 47.1 44.3 43.8 44.8 45.2 44.6 45.9 46.9
(= (c + d + e + f) / (a + b) x 100)
ELI Rental Availability (%) 35.7 36.9 36.0 31.8 32.1 31.2 29.6 27.6 27.2 27.9 28.2 27.8 28.9 29.2
(= c / a x 100)
163 – Appendix

Appendix II –B. Percent of Low-income Renter Households Who Occupy Low-cost
Affordable Housing, Under Alternative Definitions of Low-income, United States and
200 Largest Metropolitan Areas, Ranked by 2016 Population, 2006, 2011, and 2016
2016
Pop
Rank
Metropolitan Area
2006 2011 2016
VLI % ELI % VLI % ELI % VLI % ELI %

United States 51.4 31.8 43.8 27.2 46.9 29.2

1
New York-Newark-Jersey City, NY-NJ-
PA
46.8 34.8 38.2 29.5 40.7 32.8
2 Los Angeles-Long Beach-Anaheim, CA 32.1 19.7 25.2 17.9 28.0 20.8
3 Chicago-Naperville-Elgin, IL-IN-WI 49.4 26.0 40.6 23.6 45.8 26.8
4 Dallas-Fort Worth-Arlington, TX 49.3 17.3 47.4 17.0 44.6 17.5
5
Houston-The Woodlands-Sugar Land,
TX
51.9 19.0 47.9 14.8 43.3 19.0
6
Philadelphia-Camden-Wilmington, PA-
NJ-D
52.1 29.9 44.6 26.8 45.0 26.6
7
Washington-Arlington-Alexandria, DC-
VA-
58.7 34.0 50.2 30.6 45.3 29.7
8
Miami-Fort Lauderdale-West Palm
Beach,
31.3 25.7 24.7 22.4 25.2 22.6
9 Atlanta-Sandy Springs-Roswell, GA 41.5 24.9 30.2 18.0 36.2 20.2
10 Boston-Cambridge-Newton, MA-NH 51.4 44.1 51.0 43.1 54.3 45.0
11 San Francisco-Oakland-Hayward, CA 47.0 31.4 40.1 29.2 51.0 35.8
12 Phoenix-Mesa-Scottsdale, AZ 47.1 24.2 38.0 13.5 39.1 17.8
13 Riverside-San Bernardino-Ontario, CA 36.7 19.1 27.8 17.6 31.4 19.4
14 Detroit-Warren-Dearborn, MI 50.4 28.8 41.1 21.8 47.1 27.4
15 Seattle-Tacoma-Bellevue, WA 55.1 32.1 48.9 25.8 48.6 28.2
16
Minneapolis-St. Paul-Bloomington,
MN-WI
66.6 31.1 63.4 29.9 59.8 34.3
17 San Diego-Carlsbad, CA 32.5 20.9 24.6 15.8 26.4 20.4
18 Tampa-St. Petersburg-Clearwater, FL 30.0 19.4 25.4 14.5 27.9 16.6
19 Denver-Aurora-Lakewood, CO 53.9 25.0 50.4 20.2 40.0 25.2
20 St. Louis, MO-IL 59.8 32.6 46.4 24.2 57.0 26.6
21 Baltimore-Columbia-Towson, MD 56.3 40.0 44.0 31.8 53.4 32.6
22 Charlotte-Concord-Gastonia, NC-SC 55.7 27.2 41.5 20.7 51.5 28.3
23 Portland-Vancouver-Hillsboro, OR-WA 50.7 26.6 36.1 20.2 42.0 25.0
24 Orlando-Kissimmee-Sanford, FL 27.8 17.0 19.3 11.4 19.6 12.1
25 San Antonio-New Braunfels, TX 52.5 32.3 45.9 28.8 44.3 30.6
26
Sacramento--Roseville--Arden-Arcade,
CA
41.5 20.7 35.3 19.6 37.0 19.4
27 Pittsburgh, PA 61.2 36.1 60.6 36.0 60.1 35.7
28 Kansas City, MO-KS 62.6 36.4 50.9 25.8 57.5 28.5
29 Las Vegas-Henderson-Paradise, NV 33.9 13.9 25.0 13.1 27.8 10.1
30 Cincinnati, OH-KY-IN 66.0 34.2 58.1 28.8 65.0 30.2
31 Austin-Round Rock, TX 45.3 14.3 37.4 18.8 41.7 22.4
164 – Appendix

2016
Pop
Rank
Metropolitan Area
2006 2011 2016
VLI % ELI % VLI % ELI % VLI % ELI %
32 Cleveland-Elyria, OH 51.7 29.0 43.5 26.8 52.8 29.0
33
Nashville-Davidson--Murfreesboro--
Frank
57.1 34.4 48.3 32.7 48.7 31.3
34 Indianapolis-Carmel-Anderson, IN 58.7 23.5 49.3 21.9 49.4 19.1
35 Columbus, OH 53.2 21.3 53.9 23.7 53.3 23.6
36 San Jose-Sunnyvale-Santa Clara, CA 53.7 25.7 42.7 28.0 46.2 31.1
37
Virginia Beach-Norfolk-Newport News,
VA
49.5 32.5 36.9 30.0 34.1 28.2
38 Providence-Warwick, RI-MA 53.5 44.6 47.9 39.5 55.8 44.3
39 Milwaukee-Waukesha-West Allis, WI 49.7 24.6 36.5 18.7 50.4 21.2
40 Oklahoma City, OK 50.9 24.2 46.5 18.4 53.4 30.5
41 Jacksonville, FL 51.5 33.2 34.2 25.5 38.0 22.3
42 Raleigh, NC 56.8 28.1 53.1 18.8 58.5 30.9
43 New Orleans-Metairie, LA 49.8 28.8 32.0 23.2 34.0 26.8
44 Richmond, VA 50.5 36.0 42.4 32.0 41.2 27.1
45 Louisville/Jefferson County, KY-IN 59.4 36.0 49.4 29.7 54.7 37.3
46 Salt Lake City, UT 60.4 29.8 43.2 21.7 55.0 27.8
47 Memphis, TN-MS-AR 40.2 29.4 33.9 18.4 33.6 20.0
48
Hartford-West Hartford-East Hartford,
C
63.6 40.8 47.8 30.7 55.2 34.9
49 Birmingham-Hoover, AL 54.2 31.0 45.3 29.1 52.1 38.6
50
Buffalo-Cheektowaga-Niagara Falls,
NY
51.7 27.2 46.8 24.7 55.9 30.6
51 Rochester, NY 48.3 24.5 44.0 20.8 48.5 24.5
52 Omaha-Council Bluffs, NE-IA 66.9 31.5 57.4 26.8 61.6 22.6
53 Tucson, AZ 47.5 15.6 38.5 15.0 40.4 18.8
54 Urban Honolulu, HI 48.8 35.8 41.8 33.7 40.5 36.8
55 Fresno, CA 39.8 22.0 30.1 20.4 32.4 18.7
56 Greenville-Anderson-Mauldin, SC 56.9 34.3 50.3 34.5 46.8 25.2
57 Knoxville, TN 63.8 45.5 53.2 38.3 56.8 44.1
58 Bridgeport-Stamford-Norwalk, CT 57.3 40.2 49.6 36.9 50.5 33.8
59 Worcester, MA-CT 60.8 42.3 55.2 38.6 59.1 39.3
60 Grand Rapids-Wyoming, MI 49.8 24.5 46.3 33.0 56.7 24.8
61 Albuquerque, NM 55.4 22.9 41.6 17.6 43.0 20.5
62 Bakersfield, CA 45.0 22.8 36.9 15.3 32.0 18.1
63 New Haven-Milford, CT 49.7 33.1 38.1 26.7 43.3 33.0
64 Albany-Schenectady-Troy, NY 58.7 25.8 47.5 23.4 51.0 33.0
65 McAllen-Edinburg-Mission, TX 46.1 30.4 40.3 34.3 41.4 26.3
66 Oxnard-Thousand Oaks-Ventura, CA 45.7 32.8 40.0 31.0 37.1 28.0
67 El Paso, TX 49.0 37.2 52.2 24.9 45.4 29.3
68 Allentown-Bethlehem-Easton, PA-NJ 58.6 33.6 42.3 26.5 46.7 34.0
69 Baton Rouge, LA 43.9 24.4 38.9 26.7 43.3 19.2
165 – Appendix

2016
Pop
Rank
Metropolitan Area
2006 2011 2016
VLI % ELI % VLI % ELI % VLI % ELI %
70 Greensboro-High Point, NC 50.3 33.0 43.3 32.3 39.1 18.6
71 Dayton, OH 52.4 29.4 41.2 26.6 51.6 28.3
72 North Port-Sarasota-Bradenton, FL 33.1 24.6 27.7 17.4 25.1 17.0
73 Charleston-North Charleston, SC 44.6 28.7 39.3 31.1 44.7 26.0
74 Stockton-Lodi, CA 40.2 20.1 34.3 18.8 37.6 23.3
75 Cape Coral-Fort Myers, FL 31.7 31.3 25.5 15.1 24.8 18.0
76 Boise City, ID 64.6 24.4 47.1 18.5 46.7 24.4
77 Colorado Springs, CO 62.8 21.5 51.7 19.1 45.8 14.6
78
Little Rock-North Little Rock-Conway,
A
51.5 20.4 37.4 16.5 47.9 23.7
79 Akron, OH 50.5 38.3 52.5 23.5 50.1 28.8
80 Syracuse, NY 53.8 28.5 49.9 28.6 49.4 26.4
81 Lakeland-Winter Haven, FL 38.4 28.4 29.8 21.7 28.8 14.9
82 Toledo, OH 58.3 30.5 35.8 17.9 51.9 27.5
83 Winston-Salem, NC 60.7 34.1 37.9 23.1 49.9 22.4
84
Deltona-Daytona Beach-Ormond
Beach, FL
38.5 37.7 26.7 24.0 22.7 17.8
85 Jackson, MS 36.6 26.6 42.9 28.8 51.9 36.6
86 Wichita, KS 61.9 13.8 58.6 20.7 51.4 19.9
87 Spokane-Spokane Valley, WA 55.5 25.8 52.7 21.1 53.5 26.3
88 Provo-Orem, UT 50.1 11.4 52.3 16.3 60.2 27.6
89 Ogden-Clearfield, UT 71.2 28.4 64.7 26.7 62.7 41.5
90 Springfield, MA 49.5 37.4 52.3 40.8 48.5 45.8
91 Palm Bay-Melbourne-Titusville, FL 29.0 15.8 31.4 28.9 31.2 27.7
92 Harrisburg-Carlisle, PA 63.5 33.8 54.8 29.9 57.0 35.2
93
Youngstown-Warren-Boardman, OH-
PA
65.6 32.3 52.2 32.9 51.5 24.7
94 Lafayette, LA 60.3 33.9 52.6 39.0 48.9 46.3
95 Augusta-Richmond County, GA-SC 54.8 39.4 43.5 34.2 37.5 24.0
96 Modesto, CA 38.9 18.9 20.5 11.2 28.4 26.8
97 Lancaster, PA 57.7 21.2 46.9 21.1 43.5 24.4
98 Chattanooga, TN-GA 66.5 40.0 40.4 23.0 42.5 36.0
99 Portland-South Portland, ME 50.4 39.7 51.1 27.7 60.5 52.2
100 Santa Rosa, CA 34.6 21.6 38.9 25.6 35.9 35.1
101 Salinas, CA 40.3 30.6 26.1 21.1 36.4 23.6
102 Scranton--Wilkes-Barre--Hazleton, PA 57.2 40.9 54.4 35.7 47.0 31.5
103 Corpus Christi, TX 41.4 28.6 48.7 33.2 49.4 35.3
104
Fayetteville-Springdale-Rogers, AR-
MO
46.6 22.5 53.7 13.6 47.3 25.2
105 Pensacola-Ferry Pass-Brent, FL 49.3 32.2 32.1 18.0 36.9 28.6
106 Lansing-East Lansing, MI 53.6 28.3 45.1 23.2 43.8 19.0
107 Port St. Lucie, FL 23.5 23.8 33.5 24.3 24.3 14.3
166 – Appendix

2016
Pop
Rank
Metropolitan Area
2006 2011 2016
VLI % ELI % VLI % ELI % VLI % ELI %
108 Visalia-Porterville, CA 53.6 22.0 33.8 17.6 35.9 30.8
109 Reno, NV 47.6 30.4 40.0 17.8 48.2 14.3
110 Santa Maria-Santa Barbara, CA 26.0 19.1 34.0 19.6 25.0 16.1
111 York-Hanover, PA 63.9 35.0 61.9 30.3 62.5 30.2
112 Vallejo-Fairfield, CA 36.0 29.2 30.2 23.2 35.1 24.0
113 Brownsville-Harlingen, TX 41.2 35.0 43.9 35.8 38.9 26.4
114 Huntsville, AL 73.6 28.9 57.0 32.7 63.9 33.1
115 Mobile, AL 47.7 31.3 33.2 28.6 42.4 27.3
116 Reading, PA 57.9 24.4 44.8 22.7 52.8 35.1
117 Shreveport-Bossier City, LA 45.5 31.4 43.0 28.9 35.8 27.0
118 Springfield, MO 51.8 21.5 45.9 14.0 46.4 18.3
119 Hickory-Lenoir-Morganton, NC 62.5 37.1 46.7 35.1 48.5 27.5
120 Canton-Massillon, OH 64.2 41.5 47.1 30.2 52.2 40.5
121 Salisbury, MD-DE 53.2 44.3 40.0 25.4 47.8 34.0
122 Beaumont-Port Arthur, TX 63.8 34.7 36.5 37.5 49.1 42.3
123 Kalamazoo-Portage, MI 51.9 24.9 42.0 30.3 58.1 34.5
124 Montgomery, AL 49.8 33.5 49.8 27.3 32.8 18.3
125 Trenton, NJ 55.9 39.7 53.1 34.5 51.4 44.9
126 Fort Wayne, IN 61.2 23.2 62.5 25.9 51.9 25.8
127 Eugene, OR 36.1 16.6 25.3 12.0 32.6 15.9
128 Naples-Immokalee-Marco Island, FL 32.1 27.9 35.1 20.7 25.1 20.9
129 Ann Arbor, MI 46.2 16.2 44.7 13.4 48.5 18.9
130 Ocala, FL 46.8 37.6 30.0 27.8 31.9 16.9
131 Manchester-Nashua, NH 55.3 33.6 54.3 31.1 55.2 27.7
132 Gulfport-Biloxi-Pascagoula, MS 53.8 42.0 26.5 27.4 34.3 19.2
133 Rockford, IL 49.0 27.9 41.4 15.4 40.7 18.5
134 Fort Collins, CO 52.7 17.8 31.5 18.6 35.7 14.7
135 Fayetteville, NC 45.7 22.8 33.8 22.9 29.8 20.4
136 Clarksville, TN-KY 56.4 36.9 38.6 18.1 52.3 30.7
137 Lincoln, NE 65.1 35.5 60.1 32.9 64.6 24.7
138 Lubbock, TX 36.4 19.9 33.4 10.9 31.0 19.9
139 Spartanburg, SC 55.7 45.3 45.0 28.9 40.8 30.4
140
San Luis Obispo-Paso Robles-Arroyo
Gran
32.0 16.8 32.8 24.0 34.8 18.4
141 Utica-Rome, NY 55.6 29.3 50.0 24.1 61.0 22.0
142 Erie, PA 47.6 21.0 49.5 22.9 55.6 21.8
143 Olympia-Tumwater, WA 50.3 18.5 39.3 23.6 36.9 15.7
144 Santa Cruz-Watsonville, CA 41.3 26.7 32.1 29.1 36.5 25.7
145 Laredo, TX 46.9 26.2 36.3 14.0 21.1 12.0
146 Norwich-New London, CT 61.9 37.3 50.3 43.1 53.5 36.1
167 – Appendix

2016
Pop
Rank
Metropolitan Area
2006 2011 2016
VLI % ELI % VLI % ELI % VLI % ELI %
147 Merced, CA 25.9 16.2 26.3 15.3 41.8 19.7
148 Bremerton-Silverdale, WA 50.6 36.8 47.3 26.3 40.8 31.8
149 Gainesville, FL 24.8 12.3 21.1 16.1 25.9 7.4
150 Lynchburg, VA 60.1 40.7 48.6 18.2 52.5 35.1
151 Atlantic City-Hammonton, NJ 41.6 37.6 36.8 32.8 49.6 33.6
152 Amarillo, TX 55.8 21.7 44.8 13.0 57.4 21.4
153 Yakima, WA 53.0 27.8 29.7 25.0 54.8 22.3
154 Waco, TX 38.3 24.8 33.8 21.9 34.5 18.9
155 Barnstable Town, MA 56.1 55.8 49.0 48.6 46.4 36.5
156 Houma-Thibodaux, LA 81.1 44.1 48.0 36.6 51.9 30.1
157 Binghamton, NY 60.9 32.5 48.5 18.4 53.3 18.1
158 Chico, CA 32.6 16.0 24.3 19.4 21.0 10.2
159 Tyler, TX 44.7 11.0 30.2 25.3 43.4 9.8
160 Prescott, AZ 40.5 21.4 34.2 16.4 26.3 22.3
161 College Station-Bryan, TX 21.5 10.0 11.1 6.6 10.8 8.6
162 Burlington-South Burlington, VT 53.0 23.3 44.8 29.8 53.6 30.8
163 Bellingham, WA 45.9 26.2 33.8 14.4 35.3 21.5
164 Medford, OR 43.9 42.0 23.3 23.9 33.6 24.0
165 Las Cruces, NM 46.1 30.0 33.3 18.1 30.0 9.8
166 Champaign-Urbana, IL 53.7 14.8 17.8 7.1 35.5 8.5
167 Yuma, AZ 60.4 28.6 41.7 21.0 45.5 31.4
168 Elkhart-Goshen, IN 63.4 40.0 43.1 17.2 53.6 40.0
169 Springfield, IL 59.3 38.3 59.3 29.0 42.7 21.1
170 Gainesville, GA 51.9 59.7 48.0 11.7 36.5 23.6
171 Racine, WI 58.2 27.1 48.5 32.8 31.3 18.9
172 Saginaw, MI 40.9 16.2 26.7 14.3 47.4 33.9
173
Blacksburg-Christiansburg-Radford,
VA
35.1 20.2 33.2 10.7 39.0 19.2
174 Bend-Redmond, OR 45.4 14.4 22.8 15.3 30.1 8.2
175 El Centro, CA 48.7 18.3 30.8 14.2 61.5 30.2
176 Redding, CA 26.8 14.3 31.8 26.7 27.3 18.6
177 Punta Gorda, FL 24.4 24.4 39.8 42.3 34.0 41.2
178 Joplin, MO 54.4 21.8 52.5 35.9 44.6 16.3
179 Greenville, NC 31.1 19.1 26.0 14.1 41.7 14.4
180 Columbia, MO 41.1 23.0 40.7 16.4 38.1 3.2
181 Dover, DE 42.2 24.4 39.1 23.3 38.0 21.4
182 Muskegon, MI 61.3 42.5 42.7 28.6 49.5 38.2
183 Bloomington, IL 57.0 26.4 67.9 37.9 58.5 18.2
184 Yuba City, CA 41.8 39.8 46.7 28.1 44.3 21.3
185 Midland, TX 72.2 19.5 28.7 11.2 43.2 22.7
168 – Appendix

2016
Pop
Rank
Metropolitan Area
2006 2011 2016
VLI % ELI % VLI % ELI % VLI % ELI %
186 Janesville-Beloit, WI 47.5 35.9 41.2 10.5 34.2 18.9
187 State College, PA 32.3 1.2 30.5 9.8 26.1 9.0
188 Burlington, NC 38.1 24.0 47.2 32.6 40.4 14.1
189 Jackson, MI 35.5 12.8 30.2 29.8 54.2 31.7
190 Auburn-Opelika, AL 47.8 17.5 41.6 14.5 46.4 40.4
191 Eau Claire, WI 68.9 44.1 56.9 37.4 53.2 21.5
192 Odessa, TX 61.4 23.1 59.1 35.5 34.0 20.8
193 Pueblo, CO 43.0 30.6 37.2 12.7 31.8 20.2
194 Monroe, LA 47.0 23.6 38.8 21.0 41.3 26.8
195 Madera, CA 43.0 37.2 48.4 27.8 28.4 19.3
196 Niles-Benton Harbor, MI 54.1 43.6 58.1 43.8 64.4 45.6
197 Decatur, AL 60.3 31.0 57.7 41.9 76.4 50.2
198 Bangor, ME 53.2 22.3 33.9 27.3 35.9 25.7
199 Hanford-Corcoran, CA 45.7 17.5 38.9 31.7 30.0 25.2
200 Monroe, MI 63.0 21.1 69.4 32.1 69.1 47.1
Notes: VLI is a Very Low-income group who earns 0-50% of area median income. ELI is
an Extremely Low-income group who earns 0-30% of area median income. Thresholds of
income-to-AMI are based on HUD’s “at or below” definition. Metropolitan areas are
based on 2013 definitions for metropolitan statistical areas (MSAs) from the U.S. Office
of Management and Budget (OMB), which is based on the 2010 Decennial Census.
Sources: 2006, 2011, and 2016 ACS IPUMS Microdata files.

169 – Appendix

Appendix II –C. Metropolitan Areas with Least or Most Availability of Affordable
Rentals, Top 200 Metros, 2006, 2011, and 2016

(a) 2006

(b) 2011

170 – Appendix

(c) Changes from 2006 to 2011

(d) Changes from 2011 to 2016

Notes: Universe is very low-income (<50% of area median income) renters in each area. A rental unit is
defined affordable when it costs at or below 30% of the top income threshold of the group in a metro area.
Sources: U.S. Census Bureau, TIGER/LINE files; 2006, 2011, and 2016 ACS 1-year IPUMS data.
171 – Appendix

Appendix II –D. Influence of Metropolitan Observations on the Coefficient of
Government Subsidies on the Very Low-income Rental Availability, Measured by
Cook’s Distance, Ordered by the Most Positive Influence at the Top, 200 Largest
Metropolitan Areas, 2016
Area
Code
Metropolitan Area Full Name
VLI Rental
Availability
(%)
ELI Rental
Availability
(%)
Cook's
Distance
Influence
on VLI
Rental
Availability
(dfbeta)
35620
New York-Newark-Jersey City, NY-NJ-
PA 40.721 32.797 0.353 1.710
33780 Monroe, MI 69.065 47.065 0.019 0.090
35660 Niles-Benton Harbor, MI 64.397 45.596 0.018 0.085
19460 Decatur, AL 76.394 50.199 0.017 0.081
20940 El Centro, CA 61.540 30.209 0.100 0.067
39460 Punta Gorda, FL 34.011 41.200 0.023 0.062
29820 Las Vegas-Henderson-Paradise, NV 27.838 10.104 0.005 0.058
36740 Orlando-Kissimmee-Sanford, FL 19.644 12.121 0.006 0.050
12060 Atlanta-Sandy Springs-Roswell, GA 36.195 20.246 0.007 0.046
12700 Barnstable Town, MA 46.433 36.486 0.013 0.037
32820 Memphis, TN-MS-AR 33.632 20.037 0.002 0.037
37980
Philadelphia-Camden-Wilmington, PA-
NJ-D 44.980 26.562 0.003 0.036
33700 Modesto, CA 28.375 26.831 0.011 0.035
47900
Washington-Arlington-Alexandria, DC-
VA- 45.257 29.742 0.015 0.031
17780 College Station-Bryan, TX 10.810 8.595 0.028 0.029
49740 Yuma, AZ 45.463 31.427 0.004 0.028
40060 Richmond, VA 41.189 27.088 0.002 0.028
12540 Bakersfield, CA 32.004 18.147 0.003 0.027
40900
Sacramento--Roseville--Arden-Arcade,
CA 37.043 19.449 0.001 0.025
27100 Jackson, MI 54.233 31.654 0.002 0.025
12220 Auburn-Opelika, AL 46.365 40.410 0.005 0.023
47260
Virginia Beach-Norfolk-Newport News,
VA 34.101 28.199 0.001 0.022
38860 Portland-South Portland, ME 60.501 52.233 0.005 0.022
34740 Muskegon, MI 49.536 38.237 0.001 0.019
45300 Tampa-St. Petersburg-Clearwater, FL 27.944 16.646 0.001 0.018
26620 Huntsville, AL 63.873 33.078 0.007 0.017
19820 Detroit-Warren-Dearborn, MI 47.114 27.361 0.001 0.017
49420 Yakima, WA 54.781 22.254 0.005 0.016
35380 New Orleans-Metairie, LA 34.036 26.759 0.001 0.016
26380 Houma-Thibodaux, LA 51.941 30.096 0.002 0.016
15540 Burlington-South Burlington, VT 53.564 30.789 0.002 0.015
12100 Atlantic City-Hammonton, NJ 49.558 33.585 0.008 0.015
49700 Yuba City, CA 44.293 21.308 0.016 0.014
172 – Appendix

20740 Eau Claire, WI 53.172 21.473 0.001 0.013
12940 Baton Rouge, LA 43.310 19.246 0.001 0.012
33340 Milwaukee-Waukesha-West Allis, WI 50.362 21.191 0.000 0.012
41740 San Diego-Carlsbad, CA 26.363 20.398 0.000 0.011
23420 Fresno, CA 32.396 18.688 0.000 0.011
38060 Phoenix-Mesa-Scottsdale, AZ 39.078 17.750 0.000 0.011
40380 Rochester, NY 48.546 24.548 0.001 0.011
28020 Kalamazoo-Portage, MI 58.136 34.519 0.001 0.010
19100 Dallas-Fort Worth-Arlington, TX 44.636 17.451 0.001 0.010
26420 Houston-The Woodlands-Sugar Land, TX 43.311 19.023 0.001 0.009
49620 York-Hanover, PA 62.514 30.153 0.001 0.009
24660 Greensboro-High Point, NC 39.133 18.595 0.000 0.009
19740 Denver-Aurora-Lakewood, CO 39.955 25.209 0.000 0.009
10580 Albany-Schenectady-Troy, NY 51.013 33.016 0.001 0.009
39340 Provo-Orem, UT 60.164 27.629 0.007 0.008
33260 Midland, TX 43.189 22.653 0.000 0.008
11100 Amarillo, TX 57.382 21.381 0.004 0.008
30700 Lincoln, NE 64.601 24.737 0.025 0.008
12420 Austin-Round Rock, TX 41.676 22.374 0.000 0.007
36100 Ocala, FL 31.946 16.949 0.011 0.007
29180 Lafayette, LA 48.850 46.253 0.004 0.007
45780 Toledo, OH 51.853 27.520 0.001 0.007
29460 Lakeland-Winter Haven, FL 28.773 14.921 0.002 0.007
14010 Bloomington, IL 58.504 18.160 0.000 0.007
46540 Utica-Rome, NY 60.952 21.975 0.001 0.007
13780 Binghamton, NY 53.330 18.140 0.001 0.006
45060 Syracuse, NY 49.432 26.390 0.001 0.006
10900 Allentown-Bethlehem-Easton, PA-NJ 46.675 34.046 0.000 0.006
35300 New Haven-Milford, CT 43.319 33.013 0.000 0.005
21500 Erie, PA 55.602 21.789 0.001 0.005
19660
Deltona-Daytona Beach-Ormond Beach,
FL 22.715 17.758 0.007 0.005
31340 Lynchburg, VA 52.550 35.126 0.000 0.005
29540 Lancaster, PA 43.464 24.388 0.001 0.005
42540 Scranton--Wilkes-Barre--Hazleton, PA 46.995 31.525 0.000 0.004
25060 Gulfport-Biloxi-Pascagoula, MS 34.292 19.231 0.000 0.004
32780 Medford, OR 33.627 23.999 0.005 0.004
12260 Augusta-Richmond County, GA-SC 37.542 23.987 0.001 0.004
15980 Cape Coral-Fort Myers, FL 24.766 18.001 0.004 0.004
27260 Jacksonville, FL 37.970 22.291 0.000 0.004
26900 Indianapolis-Carmel-Anderson, IN 49.374 19.111 0.000 0.004
25540 Hartford-West Hartford-East Hartford, C 55.151 34.867 0.000 0.003
17300 Clarksville, TN-KY 52.309 30.748 0.016 0.003
14860 Bridgeport-Stamford-Norwalk, CT 50.463 33.829 0.001 0.002
20100 Dover, DE 38.047 21.434 0.000 0.002
173 – Appendix

29620 Lansing-East Lansing, MI 43.832 18.951 0.000 0.002
15380 Buffalo-Cheektowaga-Niagara Falls, NY 55.890 30.643 0.000 0.002
32580 McAllen-Edinburg-Mission, TX 41.391 26.259 0.001 0.002
33860 Montgomery, AL 32.759 18.293 0.003 0.002
17020 Chico, CA 20.960 10.194 0.001 0.002
46060 Tucson, AZ 40.415 18.844 0.000 0.002
10740 Albuquerque, NM 42.975 20.458 0.009 0.001
48620 Wichita, KS 51.394 19.882 0.000 0.001
24340 Grand Rapids-Wyoming, MI 56.709 24.828 0.000 0.001
40420 Rockford, IL 40.654 18.521 0.005 0.001
39820 Redding, CA 27.340 18.572 0.000 0.001
42220 Santa Rosa, CA 35.920 35.055 0.000 0.001
16580 Champaign-Urbana, IL 35.467 8.494 0.000 0.001
45940 Trenton, NJ 51.356 44.885 0.001 0.001
41700 San Antonio-New Braunfels, TX 44.318 30.613 0.000 0.001
22220 Fayetteville-Springdale-Rogers, AR-MO 47.347 25.184 0.000 0.001
23540 Gainesville, FL 25.928 7.373 0.000 0.001
47380 Waco, TX 34.492 18.898 0.000 0.001
35840 North Port-Sarasota-Bradenton, FL 25.124 17.020 0.007 0.001
41180 St. Louis, MO-IL 56.981 26.640 0.000 0.001
42020
San Luis Obispo-Paso Robles-Arroyo
Gran 34.751 18.426 0.000 0.000
16860 Chattanooga, TN-GA 42.450 35.972 0.000 0.000
44180 Springfield, MO 46.352 18.336 0.000 0.000
35980 Norwich-New London, CT 53.511 36.109 0.000 0.000
25860 Hickory-Lenoir-Morganton, NC 48.470 27.457 0.000 0.000
49660 Youngstown-Warren-Boardman, OH-PA 51.481 24.716 0.000 0.000
27140 Jackson, MS 51.930 36.575 0.000 0.000
14260 Boise City, ID 46.723 24.437 0.000 0.000
37860 Pensacola-Ferry Pass-Brent, FL 36.868 28.649 0.000 0.000
31180 Lubbock, TX 31.001 19.913 0.000 0.000
29740 Las Cruces, NM 30.038 9.849 0.000 0.000
22660 Fort Collins, CO 35.748 14.687 0.000 0.000
33740 Monroe, LA 41.287 26.825 0.017 0.000
25260 Hanford-Corcoran, CA 29.981 25.232 0.061 0.000
49180 Winston-Salem, NC 49.923 22.383 0.000 -0.001
38940 Port St. Lucie, FL 24.294 14.292 0.000 -0.001
49340 Worcester, MA-CT 59.086 39.261 0.000 -0.001
19380 Dayton, OH 51.574 28.305 0.000 -0.001
43340 Shreveport-Bossier City, LA 35.800 26.988 0.000 -0.001
33660 Mobile, AL 42.352 27.263 0.000 -0.001
39740 Reading, PA 52.761 35.064 0.000 -0.001
17460 Cleveland-Elyria, OH 52.799 29.027 0.000 -0.001
17820 Colorado Springs, CO 45.762 14.607 0.000 -0.001
30780 Little Rock-North Little Rock-Conway, A 47.856 23.698 0.000 -0.002
174 – Appendix

44140 Springfield, MA 48.461 45.758 0.000 -0.002
36260 Ogden-Clearfield, UT 62.656 41.476 0.000 -0.002
27900 Joplin, MO 44.599 16.267 0.000 -0.002
22180 Fayetteville, NC 29.827 20.432 0.000 -0.003
13980 Blacksburg-Christiansburg-Radford, VA 38.997 19.211 0.001 -0.003
31700 Manchester-Nashua, NH 55.182 27.722 0.001 -0.004
11460 Ann Arbor, MI 48.549 18.891 0.004 -0.004
36420 Oklahoma City, OK 53.383 30.503 0.000 -0.004
33460
Minneapolis-St. Paul-Bloomington, MN-
WI 59.846 34.284 0.000 -0.004
41540 Salisbury, MD-DE 47.762 34.013 0.000 -0.004
17860 Columbia, MO 38.061 3.221 0.004 -0.004
15940 Canton-Massillon, OH 52.162 40.504 0.001 -0.004
13140 Beaumont-Port Arthur, TX 49.101 42.275 0.002 -0.004
16700 Charleston-North Charleston, SC 44.729 26.027 0.003 -0.004
46700 Vallejo-Fairfield, CA 35.088 23.963 0.001 -0.005
38900 Portland-Vancouver-Hillsboro, OR-WA 42.005 25.033 0.000 -0.005
13820 Birmingham-Hoover, AL 52.051 38.593 0.001 -0.005
37340 Palm Bay-Melbourne-Titusville, FL 31.208 27.701 0.005 -0.006
15500 Burlington, NC 40.434 14.108 0.000 -0.006
44700 Stockton-Lodi, CA 37.612 23.313 0.000 -0.007
13380 Bellingham, WA 35.340 21.480 0.002 -0.007
32900 Merced, CA 41.795 19.680 0.005 -0.007
23060 Fort Wayne, IN 51.877 25.771 0.002 -0.007
15180 Brownsville-Harlingen, TX 38.878 26.370 0.002 -0.008
16980 Chicago-Naperville-Elgin, IL-IN-WI 45.847 26.785 0.005 -0.008
10420 Akron, OH 50.096 28.796 0.001 -0.008
39580 Raleigh, NC 58.534 30.874 0.007 -0.008
47300 Visalia-Porterville, CA 35.907 30.839 0.002 -0.009
40980 Saginaw, MI 47.375 33.906 0.000 -0.009
18580 Corpus Christi, TX 49.421 35.314 0.006 -0.010
25420 Harrisburg-Carlisle, PA 56.974 35.227 0.001 -0.010
41500 Salinas, CA 36.448 23.594 0.001 -0.010
21340 El Paso, TX 45.390 29.333 0.009 -0.010
24860 Greenville-Anderson-Mauldin, SC 46.751 25.195 0.001 -0.011
37100 Oxnard-Thousand Oaks-Ventura, CA 37.053 27.959 0.002 -0.011
13460 Bend-Redmond, OR 30.081 8.249 0.002 -0.011
14740 Bremerton-Silverdale, WA 40.761 31.816 0.003 -0.012
46520 Urban Honolulu, HI 40.534 36.814 0.001 -0.012
38300 Pittsburgh, PA 60.132 35.732 0.001 -0.012
24780 Greenville, NC 41.701 14.382 0.004 -0.013
21140 Elkhart-Goshen, IN 53.612 39.992 0.004 -0.014
42100 Santa Cruz-Watsonville, CA 36.546 25.673 0.015 -0.015
31140 Louisville/Jefferson County, KY-IN 54.732 37.348 0.001 -0.015
14460 Boston-Cambridge-Newton, MA-NH 54.305 45.010 0.005 -0.015
175 – Appendix

44060 Spokane-Spokane Valley, WA 53.512 26.300 0.005 -0.015
39380 Pueblo, CO 31.781 20.184 0.002 -0.016
41620 Salt Lake City, UT 54.998 27.802 0.002 -0.016
18140 Columbus, OH 53.337 23.628 0.001 -0.017
29700 Laredo, TX 21.105 11.995 0.011 -0.017
39300 Providence-Warwick, RI-MA 55.755 44.322 0.001 -0.018
28140 Kansas City, MO-KS 57.544 28.533 0.001 -0.018
46340 Tyler, TX 43.364 9.777 0.002 -0.018
36540 Omaha-Council Bluffs, NE-IA 61.610 22.595 0.003 -0.018
23580 Gainesville, GA 36.504 23.582 0.006 -0.019
40140 Riverside-San Bernardino-Ontario, CA 31.384 19.415 0.001 -0.020
39900 Reno, NV 48.225 14.276 0.004 -0.020
12580 Baltimore-Columbia-Towson, MD 53.418 32.646 0.001 -0.020
34980 Nashville-Davidson--Murfreesboro--Frank 48.686 31.287 0.002 -0.020
21660 Eugene, OR 32.643 15.942 0.007 -0.021
34940 Naples-Immokalee-Marco Island, FL 25.076 20.949 0.010 -0.022
28940 Knoxville, TN 56.780 44.068 0.006 -0.022
43900 Spartanburg, SC 40.804 30.353 0.004 -0.024
17140 Cincinnati, OH-KY-IN 64.986 30.200 0.002 -0.025
42200 Santa Maria-Santa Barbara, CA 25.002 16.085 0.004 -0.026
31080 Los Angeles-Long Beach-Anaheim, CA 27.982 20.755 0.000 -0.028
41940 San Jose-Sunnyvale-Santa Clara, CA 46.204 31.130 0.006 -0.030
36500 Olympia-Tumwater, WA 36.908 15.680 0.007 -0.030
44300 State College, PA 26.082 8.996 0.024 -0.033
16740 Charlotte-Concord-Gastonia, NC-SC 51.454 28.349 0.004 -0.037
31460 Madera, CA 28.381 19.307 0.043 -0.038
42660 Seattle-Tacoma-Bellevue, WA 48.635 28.182 0.004 -0.041
33100 Miami-Fort Lauderdale-West Palm Beach, 25.167 22.579 0.013 -0.044
39140 Prescott, AZ 26.309 22.287 0.013 -0.062
39540 Racine, WI 31.254 18.910 0.046 -0.064
36220 Odessa, TX 33.979 20.781 0.033 -0.067
41860 San Francisco-Oakland-Hayward, CA 51.000 35.777 0.017 -0.067
27500 Janesville-Beloit, WI 34.184 18.860 0.022 -0.084
12620 Bangor, ME 35.903 25.708 0.017 -0.084
44100 Springfield, IL 42.653 21.090 0.028 -0.092

176 – Appendix

Appendix II –E. Influence of Metropolitan Observations on the Coefficient of New
Constructions on the Very Low-income Rental Availability, Measured by Cook’s
Distance, Ordered by the Most Positive Influence at the Top, 200 Largest Metropolitan
Areas, 2016
Area
Code
Metropolitan Area Full Name
VLI Rental
Availability
(%)
ELI Rental
Availability
(%)
Cook's
Distance
Influence
on VLI
Rental
Availability
(dfbeta)
30700 Lincoln, NE 64.601 24.737 0.025 0.279
39540 Racine, WI 31.254 18.910 0.046 0.225
39340 Provo-Orem, UT 60.164 27.629 0.007 0.205
39580 Raleigh, NC 58.534 30.874 0.007 0.166
27500 Janesville-Beloit, WI 34.184 18.860 0.022 0.147
16740 Charlotte-Concord-Gastonia, NC-SC 51.454 28.349 0.004 0.134
33700 Modesto, CA 28.375 26.831 0.011 0.130
40420 Rockford, IL 40.654 18.521 0.005 0.128
44100 Springfield, IL 42.653 21.090 0.028 0.127
16700 Charleston-North Charleston, SC 44.729 26.027 0.003 0.119
34980 Nashville-Davidson--Murfreesboro--Frank 48.686 31.287 0.002 0.116
44300 State College, PA 26.082 8.996 0.024 0.114
39900 Reno, NV 48.225 14.276 0.004 0.108
17300 Clarksville, TN-KY 52.309 30.748 0.016 0.102
10740 Albuquerque, NM 42.975 20.458 0.009 0.087
26620 Huntsville, AL 63.873 33.078 0.007 0.084
16980 Chicago-Naperville-Elgin, IL-IN-WI 45.847 26.785 0.005 0.082
31460 Madera, CA 28.381 19.307 0.043 0.073
37980
Philadelphia-Camden-Wilmington, PA-
NJ-D 44.980 26.562 0.003 0.068
41620 Salt Lake City, UT 54.998 27.802 0.002 0.066
42660 Seattle-Tacoma-Bellevue, WA 48.635 28.182 0.004 0.066
11460 Ann Arbor, MI 48.549 18.891 0.004 0.064
19660
Deltona-Daytona Beach-Ormond Beach,
FL 22.715 17.758 0.007 0.057
33860 Montgomery, AL 32.759 18.293 0.003 0.057
39380 Pueblo, CO 31.781 20.184 0.002 0.056
12220 Auburn-Opelika, AL 46.365 40.410 0.005 0.056
21340 El Paso, TX 45.390 29.333 0.009 0.055
37340 Palm Bay-Melbourne-Titusville, FL 31.208 27.701 0.005 0.052
47900
Washington-Arlington-Alexandria, DC-
VA- 45.257 29.742 0.015 0.049
24860 Greenville-Anderson-Mauldin, SC 46.751 25.195 0.001 0.047
44060 Spokane-Spokane Valley, WA 53.512 26.300 0.005 0.044
35380 New Orleans-Metairie, LA 34.036 26.759 0.001 0.042
35620
New York-Newark-Jersey City, NY-NJ-
PA 40.721 32.797 0.353 0.041
39460 Punta Gorda, FL 34.011 41.200 0.023 0.039
177 – Appendix

18580 Corpus Christi, TX 49.421 35.314 0.006 0.038
46340 Tyler, TX 43.364 9.777 0.002 0.038
19820 Detroit-Warren-Dearborn, MI 47.114 27.361 0.001 0.036
11100 Amarillo, TX 57.382 21.381 0.004 0.033
45060 Syracuse, NY 49.432 26.390 0.001 0.032
35300 New Haven-Milford, CT 43.319 33.013 0.000 0.032
20740 Eau Claire, WI 53.172 21.473 0.001 0.030
40380 Rochester, NY 48.546 24.548 0.001 0.030
40900
Sacramento--Roseville--Arden-Arcade,
CA 37.043 19.449 0.001 0.029
33260 Midland, TX 43.189 22.653 0.000 0.029
32820 Memphis, TN-MS-AR 33.632 20.037 0.002 0.028
29180 Lafayette, LA 48.850 46.253 0.004 0.027
45780 Toledo, OH 51.853 27.520 0.001 0.027
36540 Omaha-Council Bluffs, NE-IA 61.610 22.595 0.003 0.025
29540 Lancaster, PA 43.464 24.388 0.001 0.025
15940 Canton-Massillon, OH 52.162 40.504 0.001 0.025
29620 Lansing-East Lansing, MI 43.832 18.951 0.000 0.023
21140 Elkhart-Goshen, IN 53.612 39.992 0.004 0.023
24780 Greenville, NC 41.701 14.382 0.004 0.022
14860 Bridgeport-Stamford-Norwalk, CT 50.463 33.829 0.001 0.020
45940 Trenton, NJ 51.356 44.885 0.001 0.019
33340 Milwaukee-Waukesha-West Allis, WI 50.362 21.191 0.000 0.019
40060 Richmond, VA 41.189 27.088 0.002 0.019
10900 Allentown-Bethlehem-Easton, PA-NJ 46.675 34.046 0.000 0.019
24660 Greensboro-High Point, NC 39.133 18.595 0.000 0.018
32580 McAllen-Edinburg-Mission, TX 41.391 26.259 0.001 0.016
40980 Saginaw, MI 47.375 33.906 0.000 0.016
17020 Chico, CA 20.960 10.194 0.001 0.015
42540 Scranton--Wilkes-Barre--Hazleton, PA 46.995 31.525 0.000 0.014
33740 Monroe, LA 41.287 26.825 0.017 0.013
12540 Bakersfield, CA 32.004 18.147 0.003 0.013
31080 Los Angeles-Long Beach-Anaheim, CA 27.982 20.755 0.000 0.013
25540 Hartford-West Hartford-East Hartford, C 55.151 34.867 0.000 0.012
23420 Fresno, CA 32.396 18.688 0.000 0.012
15180 Brownsville-Harlingen, TX 38.878 26.370 0.002 0.011
41740 San Diego-Carlsbad, CA 26.363 20.398 0.000 0.010
42020
San Luis Obispo-Paso Robles-Arroyo
Gran 34.751 18.426 0.000 0.010
38940 Port St. Lucie, FL 24.294 14.292 0.000 0.009
20100 Dover, DE 38.047 21.434 0.000 0.009
13980 Blacksburg-Christiansburg-Radford, VA 38.997 19.211 0.001 0.009
39740 Reading, PA 52.761 35.064 0.000 0.008
15540 Burlington-South Burlington, VT 53.564 30.789 0.002 0.007
16580 Champaign-Urbana, IL 35.467 8.494 0.000 0.007
178 – Appendix

46060 Tucson, AZ 40.415 18.844 0.000 0.007
15380 Buffalo-Cheektowaga-Niagara Falls, NY 55.890 30.643 0.000 0.006
42220 Santa Rosa, CA 35.920 35.055 0.000 0.006
38900 Portland-Vancouver-Hillsboro, OR-WA 42.005 25.033 0.000 0.006
17820 Colorado Springs, CO 45.762 14.607 0.000 0.006
49740 Yuma, AZ 45.463 31.427 0.004 0.006
28140 Kansas City, MO-KS 57.544 28.533 0.001 0.004
18140 Columbus, OH 53.337 23.628 0.001 0.004
37860 Pensacola-Ferry Pass-Brent, FL 36.868 28.649 0.000 0.004
44700 Stockton-Lodi, CA 37.612 23.313 0.000 0.004
36420 Oklahoma City, OK 53.383 30.503 0.000 0.004
47260
Virginia Beach-Norfolk-Newport News,
VA 34.101 28.199 0.001 0.002
36500 Olympia-Tumwater, WA 36.908 15.680 0.007 0.002
31180 Lubbock, TX 31.001 19.913 0.000 0.001
41180 St. Louis, MO-IL 56.981 26.640 0.000 0.001
29740 Las Cruces, NM 30.038 9.849 0.000 0.001
31340 Lynchburg, VA 52.550 35.126 0.000 0.001
22180 Fayetteville, NC 29.827 20.432 0.000 0.000
23540 Gainesville, FL 25.928 7.373 0.000 0.000
41700 San Antonio-New Braunfels, TX 44.318 30.613 0.000 0.000
48620 Wichita, KS 51.394 19.882 0.000 0.000
35980 Norwich-New London, CT 53.511 36.109 0.000 0.000
16860 Chattanooga, TN-GA 42.450 35.972 0.000 0.000
30780 Little Rock-North Little Rock-Conway, A 47.856 23.698 0.000 0.000
26380 Houma-Thibodaux, LA 51.941 30.096 0.002 0.000
49660 Youngstown-Warren-Boardman, OH-PA 51.481 24.716 0.000 0.000
45300 Tampa-St. Petersburg-Clearwater, FL 27.944 16.646 0.001 0.000
44180 Springfield, MO 46.352 18.336 0.000 -0.001
27900 Joplin, MO 44.599 16.267 0.000 -0.001
25860 Hickory-Lenoir-Morganton, NC 48.470 27.457 0.000 -0.001
22660 Fort Collins, CO 35.748 14.687 0.000 -0.002
14260 Boise City, ID 46.723 24.437 0.000 -0.002
49180 Winston-Salem, NC 49.923 22.383 0.000 -0.002
26900 Indianapolis-Carmel-Anderson, IN 49.374 19.111 0.000 -0.002
19380 Dayton, OH 51.574 28.305 0.000 -0.002
33460
Minneapolis-St. Paul-Bloomington, MN-
WI 59.846 34.284 0.000 -0.003
17460 Cleveland-Elyria, OH 52.799 29.027 0.000 -0.003
24340 Grand Rapids-Wyoming, MI 56.709 24.828 0.000 -0.003
46520 Urban Honolulu, HI 40.534 36.814 0.001 -0.004
47300 Visalia-Porterville, CA 35.907 30.839 0.002 -0.004
43340 Shreveport-Bossier City, LA 35.800 26.988 0.000 -0.004
38060 Phoenix-Mesa-Scottsdale, AZ 39.078 17.750 0.000 -0.005
31140 Louisville/Jefferson County, KY-IN 54.732 37.348 0.001 -0.006
179 – Appendix

14740 Bremerton-Silverdale, WA 40.761 31.816 0.003 -0.008
14010 Bloomington, IL 58.504 18.160 0.000 -0.008
46700 Vallejo-Fairfield, CA 35.088 23.963 0.001 -0.008
19740 Denver-Aurora-Lakewood, CO 39.955 25.209 0.000 -0.010
27260 Jacksonville, FL 37.970 22.291 0.000 -0.010
31700 Manchester-Nashua, NH 55.182 27.722 0.001 -0.010
39820 Redding, CA 27.340 18.572 0.000 -0.010
44140 Springfield, MA 48.461 45.758 0.000 -0.010
36260 Ogden-Clearfield, UT 62.656 41.476 0.000 -0.011
49340 Worcester, MA-CT 59.086 39.261 0.000 -0.011
10580 Albany-Schenectady-Troy, NY 51.013 33.016 0.001 -0.012
38860 Portland-South Portland, ME 60.501 52.233 0.005 -0.013
13140 Beaumont-Port Arthur, TX 49.101 42.275 0.002 -0.014
27140 Jackson, MS 51.930 36.575 0.000 -0.014
23060 Fort Wayne, IN 51.877 25.771 0.002 -0.015
22220 Fayetteville-Springdale-Rogers, AR-MO 47.347 25.184 0.000 -0.016
49620 York-Hanover, PA 62.514 30.153 0.001 -0.017
13820 Birmingham-Hoover, AL 52.051 38.593 0.001 -0.017
12940 Baton Rouge, LA 43.310 19.246 0.001 -0.019
33660 Mobile, AL 42.352 27.263 0.000 -0.020
28940 Knoxville, TN 56.780 44.068 0.006 -0.020
15500 Burlington, NC 40.434 14.108 0.000 -0.022
29460 Lakeland-Winter Haven, FL 28.773 14.921 0.002 -0.022
38300 Pittsburgh, PA 60.132 35.732 0.001 -0.023
25060 Gulfport-Biloxi-Pascagoula, MS 34.292 19.231 0.000 -0.023
25420 Harrisburg-Carlisle, PA 56.974 35.227 0.001 -0.024
33780 Monroe, MI 69.065 47.065 0.019 -0.024
17140 Cincinnati, OH-KY-IN 64.986 30.200 0.002 -0.025
40140 Riverside-San Bernardino-Ontario, CA 31.384 19.415 0.001 -0.025
46540 Utica-Rome, NY 60.952 21.975 0.001 -0.026
12580 Baltimore-Columbia-Towson, MD 53.418 32.646 0.001 -0.026
41540 Salisbury, MD-DE 47.762 34.013 0.000 -0.030
41500 Salinas, CA 36.448 23.594 0.001 -0.033
32780 Medford, OR 33.627 23.999 0.005 -0.034
12060 Atlanta-Sandy Springs-Roswell, GA 36.195 20.246 0.007 -0.036
42200 Santa Maria-Santa Barbara, CA 25.002 16.085 0.004 -0.036
13780 Binghamton, NY 53.330 18.140 0.001 -0.036
12620 Bangor, ME 35.903 25.708 0.017 -0.038
47380 Waco, TX 34.492 18.898 0.000 -0.038
41940 San Jose-Sunnyvale-Santa Clara, CA 46.204 31.130 0.006 -0.039
21500 Erie, PA 55.602 21.789 0.001 -0.039
28020 Kalamazoo-Portage, MI 58.136 34.519 0.001 -0.039
37100 Oxnard-Thousand Oaks-Ventura, CA 37.053 27.959 0.002 -0.040
13380 Bellingham, WA 35.340 21.480 0.002 -0.041
29820 Las Vegas-Henderson-Paradise, NV 27.838 10.104 0.005 -0.042
180 – Appendix

34740 Muskegon, MI 49.536 38.237 0.001 -0.044
39300 Providence-Warwick, RI-MA 55.755 44.322 0.001 -0.045
32900 Merced, CA 41.795 19.680 0.005 -0.046
14460 Boston-Cambridge-Newton, MA-NH 54.305 45.010 0.005 -0.047
12260 Augusta-Richmond County, GA-SC 37.542 23.987 0.001 -0.049
49700 Yuba City, CA 44.293 21.308 0.016 -0.049
19100 Dallas-Fort Worth-Arlington, TX 44.636 17.451 0.001 -0.050
27100 Jackson, MI 54.233 31.654 0.002 -0.051
10420 Akron, OH 50.096 28.796 0.001 -0.052
43900 Spartanburg, SC 40.804 30.353 0.004 -0.052
12420 Austin-Round Rock, TX 41.676 22.374 0.000 -0.054
49420 Yakima, WA 54.781 22.254 0.005 -0.055
26420 Houston-The Woodlands-Sugar Land, TX 43.311 19.023 0.001 -0.056
19460 Decatur, AL 76.394 50.199 0.017 -0.059
36100 Ocala, FL 31.946 16.949 0.011 -0.061
17860 Columbia, MO 38.061 3.221 0.004 -0.063
33100 Miami-Fort Lauderdale-West Palm Beach, 25.167 22.579 0.013 -0.077
12100 Atlantic City-Hammonton, NJ 49.558 33.585 0.008 -0.077
21660 Eugene, OR 32.643 15.942 0.007 -0.086
36220 Odessa, TX 33.979 20.781 0.033 -0.086
12700 Barnstable Town, MA 46.433 36.486 0.013 -0.098
41860 San Francisco-Oakland-Hayward, CA 51.000 35.777 0.017 -0.099
29700 Laredo, TX 21.105 11.995 0.011 -0.107
13460 Bend-Redmond, OR 30.081 8.249 0.002 -0.109
35840 North Port-Sarasota-Bradenton, FL 25.124 17.020 0.007 -0.116
20940 El Centro, CA 61.540 30.209 0.100 -0.117
39140 Prescott, AZ 26.309 22.287 0.013 -0.120
23580 Gainesville, GA 36.504 23.582 0.006 -0.123
36740 Orlando-Kissimmee-Sanford, FL 19.644 12.121 0.006 -0.127
15980 Cape Coral-Fort Myers, FL 24.766 18.001 0.004 -0.130
42100 Santa Cruz-Watsonville, CA 36.546 25.673 0.015 -0.133
35660 Niles-Benton Harbor, MI 64.397 45.596 0.018 -0.161
17780 College Station-Bryan, TX 10.810 8.595 0.028 -0.187
34940 Naples-Immokalee-Marco Island, FL 25.076 20.949 0.010 -0.192
25260 Hanford-Corcoran, CA 29.981 25.232 0.061 -0.193

181 – Appendix

Chapter III. Geocoding Inaccuracies: A Case Study for Evaluation of the Low-
Income Housing Tax Credit Program Data

Appendix III –A. Variables and Descriptive Statistics of the HUD’s LIHTC Database,
LIHTC Projects Placed in Service Between 1987 and 2016 in the United States
Name Definition

Project Identification

HUD_ID Unique project identifier for the database
STATE_ID State-defined project ID
PROJECT Project name
CONTACT Owner or owner's contact
COMPANY Name of contact company

Project Location

PROJ_ADD Project street address
LATITUDE Latitude: degrees in 6 Decimal Places
LONGITUD Longitude: negative Degrees in 6 Decimal Places
DDA Is the project located in a Difficult Development Area (DDA)? (0=Not in
DDA, 1=In Metro DDA, 2=In Non-Metro DDA, 3=In Metro GO Zone DDA,
4=In Non-Metro GO Zone DDA)
QCT Is the project located in a Qualified Census Tract (QCT)? (1=In a qualified
tract, 2=Not in a qualified tract)

Project Physical Characteristics

N_UNITS Total number of units
LI_UNITS Total number of low-income units
N_0BR Number of 0-bedroom units
N_1BR Number of 1-bedroom units
N_2BR Number of 2-bedroom units
N_3BR Number of 3-bedroom units
N_4BR Number of 4-bedroom units
YR_ALLOC Allocation year
YR_PIS Year placed in service
TYPE Type of construction (1=New construction, 2=Acquisition and Rehab (A/R),
3=Both new construction and A/R, 4=Existing, 9=Not Reported)

LIHTC Program Characteristics

ALLOCAMT Annual dollar amount of tax credits allocated ($ per year)
BASIS Increase in eligible basis (1=Yes, 2=No, 3=Not Reported)
CREDIT Type of credit percentage (1=30% present value, 2=70% present value,
3=Both, 4=Tax Credit Exchange Program (TCEP) only, 9=Not Reported)
182 – Appendix

INC_CEIL Elected rent/income ceiling for low-income units (1=50% Area's Median
Gross Income (AMGI), 2=60% AMGI, 3=Not Reported)
LOW_CEIL Units set aside with rents lower than elected rent/income ceiling (1=Yes,
2=No, 3=Not Reported)
CEILUNIT Number of units set aside with rents lower than elected rent/income ceiling
TRGT_POP Targets a specific population with specialized services or facilities (1=Yes,
2=No)
TRGT_FAM Targets a specific population – families (1=Yes, 0 or blank=Not indicated)
TRGT_ELD Targets a specific population – elderly (1=Yes, 0 or blank=Not indicated)
TRGT_DIS Targets a specific population – disabled (1=Yes, 0 or blank=Not indicated)
TRGT_HML Targets a specific population – homeless (1=Yes, 0 or blank=Not indicated)
TRGT_OTH Targets a specific population – other (1=Yes, 0 or blank=Not indicated)
TRGT_SPC Targets a specific population – other as specified
NONPROG No longer monitored for LIHTC program due to expired use or other reason
(1=Yes)

Additional Subsidies and Other Financial Characteristics

NON_PROF Non-profit sponsor (1=Yes, 2=No, 3=Not Reported)
BOND Tax-exempt bond received (1=Yes, 2=No, 3=Not Reported)
MFF_RA HUD Multi-Family financing/rental assistance (1=Yes, 2=No, 3=Not
Reported)
FMHA_514 FmHA (RHS) Section 514 loan (1=Yes, 2=No, 3=Not Reported)
FMHA_515 FmHA (RHS) Section 515 loan (1=Yes, 2=No, 3=Not Reported)
FMHA_538 FmHA (RHS) Section 538 loan (1=Yes, 2=No, 3=Not Reported)
HOME HOME Investment Partnership Program funds (1=Yes, 2=No, 3=Not
Reported)
HOME_AMT Dollar amount of HOME funds ($)
TCAP Tax Credit Assistance Program (TCAP) funds (1=Yes, 2=No, 3=Not
Reported)
TCAP_AMT TCAP Amount ($)
CDBG Community Development Block Grant (CDBG) funds (1=Yes, 2=No, 3=Not
Reported)
CDBG_AMT Dollar amount of CDBG funds ($)
FHA FHA-insured loan (1=Yes, 2=No, 3=Not Reported)
HOPEVI Forms part of a HOPE VI development (1=Yes, 2=No, 3=Not Reported)
HPVI_AMT Dollar amount of HOPE VI funds for development or building costs ($)
TCEP Tax Credit Exchange Program (TCEP) Funds (1=Yes, 2=No, 3=Not
Reported)
TCEP_AMT Dollar amount of TCEP funds ($)
RENTASSIST Federal or state project-based rental assistance contract (1=Federal, 2=State,
3=Both Federal and State, 4=Neither, 5=Unknown whether Federal or State)
Notes: GO Zone is Gulf Opportunity Zone which is an area that is eligible for credits,
deductions, and incentives provided by the declaring of a disaster area in the locations
that were hit hardest by the catastrophic 2005 hurricane season.
Sources: HUD National Low-Income Housing Tax Credit (LIHTC) Database 1987-2016
Dictionary (May 2018) retrieved from https://lihtc.huduser.gov/
183 – Appendix

Appendix III –B. Descriptive Statistics of LIHTC Projects, Los Angeles County, 2016
Variable Count of
Non-missing
(N=954)
Minimum Maximum Mean Median

Project Identification

HUD_ID 954 - - - -
STATE_ID 954 - - - -
PROJECT 954 - - - -
CONTACT 954 - - - -
COMPANY 954 - - - -

Project Location

PROJ_ADD 948 - - - -
LATITUDE 936 - - - -
LONGITUD 936 - - - -
DDA 898 0 1 - -
QCT 896 1 2 - -

Project Physical Characteristics

N_UNITS 954 2 761 76.582 56
LI_UNITS 922 2 757 70.503 52
N_0BR 712 0 430 13.135 0
N_1BR 712 0 761 30.782 12
N_2BR 712 0 400 19.552 9
N_3BR 712 0 150 11.860 3
N_4BR 712 0 62 2.909 0
YR_ALLOC 954 1987 2015 - -
YR_PIS 954 1987 2016 - -
TYPE 910 1 3 - -

LIHTC Program Characteristics

ALLOCAMT 885 11,119 3,609,355 641,693 495,000
BASIS 910 1 2 - -
CREDIT 907 1 4 - -
INC_CEIL 904 1 2 - -
LOW_CEIL 341 1 2 - -
CEILUNIT 334 5 262 44.473 40
TRGT_POP 905 1 2 - -
TRGT_FAM 741 0 2 - -
TRGT_ELD 734 0 2 - -
TRGT_DIS 731 0 2 - -
TRGT_HML 731 0 2 - -
184 – Appendix

TRGT_OTH 718 0 2 - -
TRGT_SPC 135 - - - -
NONPROG 867 0 1 - -

Additional Subsidies and Other Financial Characteristics

NON_PROF 878 1 2 - -
BOND 907 1 2 - -
MFF_RA 243 1 2 - -
FMHA_514 347 1 2 - -
FMHA_515 908 1 3 - -
FMHA_538 346 1 2 - -
HOME 336 1 2 - -
HOME_AMT 334 0 11,400,000 1,004,664 0
TCAP 128 1 2 - -
TCAP_AMT 128 0 9,668,100 315,028 0
CDBG 335 1 2 - -
CDBG_AMT 335 0 4,500,000 83,720 0
FHA 287 1 2 - -
HOPEVI 335 2 2 - -
HPVI_AMT 335 0 0 - -
TCEP 207 1 2 - -
TCEP_AMT 207 0 20,000,000 527,763 0
RENTASSIST 320 1 5 - -
Notes: Categorical or non-numeric variables have “-” value. Universe is 954 placed in
service between 1987 and 2016 in Los Angeles County. All samples of LIHTC projects
registered in the HUD’s original database were included.
Sources: HUD’s LIHTC Database retrieved from https://lihtc.huduser.gov/

185 – Appendix

Appendix III –C. Workflow Diagram of Refining HUD’s LIHTC Database and Pairing Geocoded Address with Concordant Parcel

186 – Appendix

Appendix III –D. Full List of the Refined LIHTC Projects, By Refinement Type, Los
Angeles County, 2016
HUD LIHTC
Project ID
Address Recorded in HUD’s Data Refined Address

Recovery of Missing Address (N = 7)

CAA19980425 * - 6330 RUGBY AVE
CAA20130039 * - 6330 RUGBY AVE
CAA19870090 - 1063 W 39TH PL
CAA19880215 - 111 S AVENUE 63
CAA19950052 - 3730 W 27TH ST
CAA19980480 - 3460 HYDE PARK BLVD
CAA20050325 - 1122 W 37TH DR

Revision of Incomplete or Incorrect Address (N = 123)

CAA00000083 ** 10562 SANTA FE DR 3501 SANTA ANITA AVE
CAA20150083 ** 10562 Santa Fe Drive
3501 SANTA ANITA AVE
CAA20080790 ** 2001 RIVER AVE 2111 W WILLIAMS ST
CAA20000665 ** 2001 RIVER AVE
2111 W WILLIAMS ST
CAA20000190 ** 535 E CARSON ST 555 E CARSON ST
CAA20000625 ** 535 E CARSON ST 555 E CARSON ST
CAA19990150 ** 4352 CLARA ST 4343 ELIZABETH ST
CAA19990150 ** 4352 CLARA ST
4343 ELIZABETH ST
CAA00000116 ** 708 W CORREGIDOR ST 601 W CORREGIDOR ST
CAA00000116 ** 708 W CORREGIDOR ST
601 W CORREGIDOR ST
CAA00000244 ** 1929 E 122ND ST 1935 E 122ND ST
CAA20141065 ** 1929 E 122ND ST
1935 E 122ND ST
CAA00000051 ** 415 S BURLINGTON AVE 409 S BURLINGTON AVE
CAA20141011 ** 415 S. Burlington Avenue
409 S BURLINGTON AVE
CAA19960180 ** 1709 W 8TH ST 731 BEACON AVE
CAA19950044 ** 1709 W EIGHTH ST
731 BEACON AVE
CAA19980395 ** 2301 W AVENUE J8 2317 W AVE J8
CAA20150010 ** 2301 W AVENUE J8
2317 W AVE J8
CAA00000068 ** 43331 30TH ST W 43359 30TH ST W
CAA20141022 ** 43331 30th Street West
43359 30TH ST W
CAA20130062 ** 16111 PLUMMER ST 9700 WOODLEY AVE
CAA20130400 ** 16111 PLUMMER ST 9700 WOODLEY AVE
CAA00000192 ** 140 JESSIE ST 133 PARK AVE
CAA20132004 ** 140 JESSIE ST
133 PARK AVE
CAA19990080 ** 20727 VANOWEN ST 20717 VANOWEN ST
CAA20010095 ** 20727 VANOWEN ST 20717 VANOWEN ST
CAA00000229 ** 658 676 S FERRIS AVE 658 S FERRIS AVE
CAA19890230 *** 108 E 5TH ST 502 S MAIN ST
187 – Appendix

CAA19920155 *** 108 E 5TH ST 502 S MAIN ST
CAA20141055 *** 108 E 5th Street
502 S MAIN ST
CAA00000027 1651 E 117TH ST 1653 E 117TH ST
CAA00000040 1723 JAMES M WOOD BLVD 1723 W NINTH ST
CAA00000106 1001 PACIFIC COAST HWY 1011 BAYCREST LN
CAA19890105 800 E 6TH ST 802 E SIXTH ST
CAA19900090 430 S GRAND VIEW ST 428 S GRAND VIEW ST
CAA19900100 5169 HOLLYWOOD BLVD 5173 HOLLYWOOD BLVD
CAA19910124 916 N GARDNER ST 908 N GARDNER ST
CAA19910141 1010 E 7TH ST 1014 E SEVENTH ST
CAA19910144 110 S SAN PEDRO ST 112 JUDGE JOHN AISO ST
CAA19920185 627 E PALMER AVE 555 E PALMER AVE
CAA19920250 609 E 5TH ST 611 E FIFTH ST
CAA19920325 471 N LOS ROBLES AVE 473 N LOS ROBLES AVE
CAA19930035 372 LOMA DR 379 LOMA DR
CAA19930325 2423 VIRGINIA AVE 2425 VIRGINIA AVE
CAA19940090 1423 W 12TH ST 1412 W 12TH ST
CAA19940120 916 W FLORENCE AVE 910 W FLORENCE AVE
CAA19940275 10310 VALLEY BLVD 10130 VALLEY BLVD
CAA19940498 6224 LELAND WAY 6211 DE LONGPRE AVE
CAA19950072 2203 E 1ST ST 2201 E FIRST ST
CAA19950081 411 N VERMONT AVE 400 N VERMONT AVE
CAA19950118 266 E WASHINGTON BLVD 264 E WASHINGTON BLVD
CAA19960189 4494 186TH ST 4502 186TH ST
CAA19960192 4668 HUNTINGTON DR S 4648 HUNTINGTON DR S
CAA19970490 2816 ORCHARD AVE 1221 W 29TH ST
CAA19970540 6501 YUCCA ST 1805 WILCOX AVE
CAA19980320 9047 S VERMONT AVE 1011 W 91ST ST
CAA19980600 4080 S VERMONT AVE 4056 S VERMONT AVE
CAA19980615 511 S WESTLAKE AVE 509 S WESTLAKE AVE
CAA19990769 850 E GRAND AVE 858 E GRAND AVE
CAA19990781 12724 VAN NUYS BLVD 12700 VAN NUYS BLVD
CAA20000150 1355 W COURT ST 1355 W CT ST
CAA20000385 800 ORANGE GROVE AVE 626 ORANGE GROVE AVE
CAA20000495 1915 BATSON AVE 1945 BATSON AVE
CAA20000505 13700 SAN ANTONIO DR 13708 SAN ANTONIO DR
CAA20000640 11620 TOWNE AVE 11650 TOWNE AVE
CAA20010125 955 W ARROW HWY 965 W ARROW HWY
CAA20010165 5216 HOLLYWOOD BLVD 5226 HOLLYWOOD BLVD
CAA20010255 1001 S HOPE ST 1031 S HOPE ST
CAA20010280 14711 NELSON AVE 14730 PRICHARD ST
CAA20010285 12326 FOOTHILL BLVD 12500 FILMORE ST
CAA20010340 15126 MOORPARK ST 15100 MOORPARK ST
CAA20010375 12450 VANOWEN ST 6755 RHODES AVE
CAA20010495 11728 COHASSET ST 7507 SIMPSON AVE
188 – Appendix

CAA20010671 3431 S LA BREA AVE 5120 EXPOSITION BLVD
CAA20010684 350 W 3RD ST 360 W THIRD ST
CAA20020020 9229 SEPULVEDA BLVD 9229 SEPULV EWAY
CAA20020060 917 W CAMERON AVE 929 W CAMERON AVE
CAA20020090 1900 E CIENEGA AVE 1211 N LYMAN AVE
CAA20020165 522 W 127TH ST 523 1/2 W EL SEGUNDO BLVD
CAA20020175 2326 JAMES M WOOD BLVD 2300 JAMES M WOOD BLVD
CAA20030110 625 N HILL ST 601 N HILL ST
CAA20030135 151 E ORANGE GROVE BLVD 169 E ORANGE GROVE BLVD
CAA20030165 347 E 1ST ST 349 E FIRST ST
CAA20030230 1659 E IMPERIAL HWY 1651 E IMPERIAL HWY
CAA20030280 2375 CALIFORNIA AVE 2399 CALIFORNIA AVE
CAA20030295 10922 FULTON WELLS AVE 10902 FULTON WELLS AVE
CAA20040040 7800 S BROADWAY 254 W 78TH ST
CAA20040085 31990 CASTAIC RD 31978 CASTAIC RD
CAA20040470 710 S LOS ANGELES ST 700 S LOS ANGELES ST
CAA20040535 301 N ALVARADO ST 306 N ALVARADO ST
CAA20040680 1320 W 7TH ST 1304 W SEVENTH ST
CAA20040690 11900 CENTRALIA RD 11950 CENTRALIA RD
CAA20040705 231 WITMER ST 1501 MIRAMAR ST
CAA20050135 1301 W COURT STREET 1301 W CT ST
CAA20050340 2347 E EL SEGUNDO BLVD 2431 E EL SEGUNDO BLVD
CAA20050415 901 W 7TH ST 901 E SEVENTH ST
CAA20050511 772 N VAN NESS AVE 721 E EL SEGUNDO BLVD
CAA20050660 5451 N SUNSET BLVD 1516 N WESTERN AVE
CAA20060005 500 S SPRING ST 510 S SPRING ST
CAA20060130 635 S SAN PEDRO ST 643 S SAN PEDRO ST
CAA20060230 45024 10TH ST W 855 W JACKMAN ST
CAA20060635 105 N CHAPEL AVE 111 N CHAPEL AVE
CAA20060715 9158 TELFAIR AVE 11971 ALLEGHENY ST
CAA20060745 11919 S FIGUEROA ST 11917 S FIGUEROA ST
CAA20070005 2615 SANTA MONICA BLVD 1349 26TH ST
CAA20070110 1063 S EASTMAN AVE 1074 S ROWAN AVE
CAA20070680 240 S BROADWAY 242 S BROADWAY
CAA20080491 201 S LAKE ST 2220 W SECOND ST
CAA20080755 1509 S ST ANDREWS PL 1511 S ST ANDREWS PL
CAA20080865 1105 E AVE Q 4 1105 E AVE Q
CAA20090115 15711 S ATLANTIC AVE 1571 S ATLANTIC DR
CAA20090140 8553 COLUMBUS AVE 8613 COLUMBUS AVE
CAA20110308 14657 BLYTHE ST 14655 BLYTHE ST
CAA20110835 707 MILLING ST 725 W MILLING ST
CAA20120894 5525 KLUMP AVE 5539 KLUMP AVE
CAA20120971 1901 W 7TH ST 681 S BONNIE BRAE ST
CAA20130011 7238 CANBY AVE 7248 CANBY AVE
CAA20130019 8025 RESEDA BLVD 8039 RESEDA BLVD
189 – Appendix

CAA20130045 8101 SEPULVEDA BLVD 15301 LANARK ST
CAA20130096 1145 N LA BREA AVE 1151 N LA BREA AVE
CAA20130351 1229 N WESTMORELAND AVE 1229 S WESTMORELAND AVE
CAA20140439 716 S 5TH AVE 720 FIFTH AVE CT
CAA20080555 410 N HAWAIIAN AVE 450 KING AVE

Drop of Double-counted Cases (N = 66)

CAA19980425 * 6330 RUGBY AVE 6330 RUGBY AVE
CAA20130039 * 6330 RUGBY AVE
CAA20010095 ** 20727 VANOWEN ST
20717 VANOWEN ST
CAA19990080 ** 20727 VANOWEN ST
CAA00000192 ** 140 JESSIE ST
133 PARK AVE
CAA20132004 ** 140 JESSIE ST
CAA20130062 ** 16111 PLUMMER ST
9700 WOODLEY AVE
CAA20130400 ** 16111 PLUMMER ST
CAA00000068 ** 43331 30TH ST W
43359 30TH ST W
CAA20141022 ** 43331 30th Street West
CAA19980395 ** 2301 W AVENUE J8
2317 W AVE J8
CAA20150010 ** 2301 W AVENUE J8
CAA19960180 ** 1709 W 8TH ST
731 BEACON AVE
CAA19950044 ** 1709 W EIGHTH ST
CAA00000051 ** 415 S BURLINGTON AVE
409 S BURLINGTON AVE
CAA20141011 ** 415 S. Burlington Avenue
CAA00000244 ** 1929 E 122ND ST
1935 E 122ND ST
CAA20141065 ** 1929 E 122ND ST
CAA20141062 ** 708 W CORREGIDOR ST
601 W CORREGIDOR ST
CAA00000116 ** 708 W CORREGIDOR ST
CAA20150039 ** 4352 CLARA ST
4343 ELIZABETH ST
CAA19990150 ** 4352 CLARA ST
CAA20000625 ** 535 E CARSON ST
555 E CARSON ST
CAA20000190 ** 535 E CARSON ST
CAA20080790 ** 2001 RIVER AVE
2111 W WILLIAMS ST
CAA20000665 ** 2001 RIVER AVE
CAA00000083 ** 10562 SANTA FE DR
3501 SANTA ANITA AVE
CAA20150083 ** 10562 Santa Fe Drive
CAA00000229 ** 658 676 S FERRIS AVE 658 S FERRIS AVE
CAA20141093 658 S. Ferris Avenue
CAA20140406 7250 HAZELTINE AVE 7250 HAZELTINE AVE
CAA19980185 7250 HAZELTINE AVE
CAA19910132 11111 STRATHERN ST 11111 STRATHERN ST
CAA19910147 11111 STRATHERN ST
CAA00000089 45151 FERN AVE 45151 FERN AVE
CAA20141034 45151 Fern Avenue
CAA00000117 4827 S CENTRAL AVE 4827 S CENTRAL AVE
190 – Appendix

CAA20141043 4827 South Central Avenue
CAA19920205 1035 E 27TH ST 1035 E 27TH ST
CAA20150064 1035 East 27th Street
CAA00000059 811 S CARONDELET ST 811 S CARONDELET ST
CAA20150085 811 S. Carondelet Street
CAA00000199 2917 E FIRST ST 2917 E FIRST ST
CAA20141080 2917 E. 1st Street
CAA00000043 3553 BESWICK ST 3553 BESWICK ST
CAA20141008 3553 Beswick Street
CAA00000136 3414 MANITOU AVE 3414 MANITOU AVE
CAA20090260 3414 MANITOU AVE
CAA00000239 4125 WHITTIER BLVD 4125 WHITTIER BLVD
CAA20150100 4125 Whittier Blvd.
CAA00000197 3740 EVANS ST 3740 EVANS ST
CAA20141079 3740 Evans Street
CAA00000182 1515 N SAN FERNANDO RD 1515 N SAN FERNANDO RD
CAA20141069 1515 N. San Fernando Road
CAA00000224 16304 S VERMONT AVE 16304 S VERMONT AVE
CAA20150092 16304 S. Vermont Avenue
CAA00000037 13218 AVALON BLVD 13218 AVALON BLVD
CAA20141005 13218 Avalon Blvd.
CAA00000129 2114 LONG BEACH BLVD 2114 LONG BEACH BLVD
CAA20150044 2114 Long Beach Boulevard
CAA20080555 410 N HAWAIIAN AVE 410 N HAWAIIAN AVE
CAA20110110 410 HAWAIIAN AVE
CAA20150051 1500 PALOS VERDES DRIVE N 1500 PALOS VERDES DRIVE N
CAA00000148 1500 PALOS VERDES DRIVE N
CAA00000242 7025 FRIENDS AVE 7025 FRIENDS AVE
CAA20150101 7025 Friends Avenue

Drop of Triple-counted Cases (N = 6)

CAA19890230 *** 108 E 5TH ST
502 S MAIN ST
CAA19920155 *** 108 E 5TH ST
CAA20141055 *** 108 E 5th Street

CAA19970300 981 HARBOR VILLAGE DR 981 HARBOR VILLAGE DR
CAA20141039 981 Harbor Village Drive
CAA00000104 1068 BAYVIEW LN
Notes: Universe is 954 LIHTC projects registered in the HUD’s original database. *
Cases that were recovered but double-counted. ** Cases that were revised but double-
counted. *** Cases that were revised but triple-counted. The second group of cases under
the triple-counted is located at the intersection of Harbor Village Drive (west-to-east
road) and Hayview Lane (north-to-south road).
Sources: HUD’s LIHTC Database, 2018.

Abstract (if available)

Abstract My dissertation entitled “Three Essays on Housing Demographics: Depressed Housing Access Amid Crisis of Housing Shortage” provides a comprehensive explanation about how housing opportunities for the entire U.S. population, particularly lower-income renters, have been disrupted during and after the Great Recession. In three independent essays, I trace the pace of recovery after the Great Recession in terms of declined household formation, occupancy losses by lower-income renters, and government-subsidized affordable housing with lack of access to transit. ❧ The first essay, “Housing shortages, declining household formation, and hidden dislodgements,” examines how we know there is a shortage of housing and how much housing is really needed for the most vulnerable population. Through a new housing—demographic method for simulating otherwise expected housing occupancies in recovery, what has disappeared from the housing stock is “made visible” through its projected traces in population from metropolitan areas. Nationwide, 8 million renters have been totally dislodged from running their own households in the housing market between 2000 and 2017. Lower-income households and young adults, mainly Millennials, have borne the brunt of housing shortage more acutely than others. ❧ The second essay, “Depressed Access to Affordable Housing Due to Higher-Income Occupancy: Evidence from U.S. Metropolitan Areas, 2006 to 2016,” investigates how much of the lower-cost affordable housing that lower-income renters could afford were “taken” by middle or higher-income households and what metropolitan factors have limited or expanded this rental availability in some areas more than in others. A housing−household−classification method reveals that only 47 percent of very low-income (earning half or less of their area’s median income level) renters successfully occupied affordable housing in 2016, which is much lower than the rate of 60 percent in the 1980s and 1990s. Despite the national average, a substantial variation in rental availability exists across the largest metropolitan areas, with the least available areas being concentrated in Southern California and Florida. Fixed-effects regression results highlight the importance of the overall increases in housing supply, either government-subsidized or market-rate, to ease rental competition in all rental brackets and open availability of low-cost housing to the poorest renters. ❧ The third essay, “Geocoding Inaccuracies: A Case Study for Evaluation of the Low-Income Housing Tax Credit Program Data,” examines the geographic location of government-subsidized affordable housing in Los Angeles County and its transit-accessibility. A new parcel-level LIHTC database is developed to quantify the positional accuracy of geocoded federal LIHTC database, which is essential for quality program evaluation but has been often neglected in previous studies. Despite the generally high accuracy of the geocoded data, I find that the level of accuracy is lower when examining large-scale LIHTC projects at a finer geographic area than census tract. A demonstration analysis on transit-accessibility shows that the inaccurate database overstates the transit-accessibility of subsidized housing, particularly in the case of large-scale developments.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Insights into residential mobility and pricing of rental housing: the role of gentrification, home-ownership barriers, and market concentrations in low-income household welfare

PDF

Housing market crisis and underwater homeowners

PDF

A better method for measuring housing affordability and the role that affordability played in the mobility outcomes of Latino-immigrants following the Great Recession

PDF

Affordable south Los Angeles: survival, support, and different futures

PDF

Risks, returns, and regulations in real estate markets

PDF

Three essays on aging, wealth, and housing tenure transitions

Asset Metadata

Creator Park, Jung Ho (author)

Core Title Three essays on housing demographics: depressed housing access amid crisis of housing shortage

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Urban Planning and Development

Publication Date 12/06/2020

Defense Date 05/15/2019

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag geocoding,homeowner diversion,housing demography,housing needs,housing shortage,Low-Income Housing Tax Credit (LIHTC),OAI-PMH Harvest,rental affordability,rental availability,renter dislodgement

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Myers, Dowell (committee chair), Kemp, Karen K. (committee member), McCarthy, T.J. (committee member)

Creator Email jhpark.planner@gmail.com,junghopa@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-248054

Unique identifier UC11675087

Identifier etd-ParkJungHo-8025.pdf (filename),usctheses-c89-248054 (legacy record id)

Legacy Identifier etd-ParkJungHo-8025.pdf

Dmrecord 248054

Document Type Dissertation

Rights Park, Jung Ho

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA