Cost Study Draft Research Design.pdf

About Global Documents

Global Documents provides you with documents from around the globe on a variety of topics for your enjoyment.
Global Documents utilizes edocr for all its document needs due to edocr's wonderful content features. Thousands of professionals and businesses around the globe publish marketing, sales, operations, customer service and financial documents making it easier for prospects and customers to find content.
COST house operate COST PUBLIC house Property Model GSD
About Global Documents

Global Documents provides you with documents from around the globe on a variety of topics for your enjoyment.
COST house operate COST PUBLIC house Property Model GSD
Prepared for the U.S. Department of Housing and Urban Development
PUBLIC HOUSING
OPERATING COST STUDY
HARVARD UNIVERSITY
GRADUATE SCHOOL OF DESIGN
Draft Research Design
July 9, 2001
Prepared For:
Office of Public and Indian Housing
U.S. Department of Housing and Urban Development
Cooperative Agreement K-PIH-99156
Public Housing Operating Cost Study
James G. Stockard, Principal Investigator
Gregory A. Byrne, Project Director
Gretchen A. Maneval, Senior Research Analyst
Lora A. Nielsen, Project Staff Assistant
Acknowledgements
The following individuals provided valuable review and comments:
Jill Khadduri, Ken Lam, and Rhiannon Patterson, Abt Associates Inc.
Charlie S. Wilkins and Lynda Ose, The Compass Group LLC
Glenn L. French, CPM, G. L. French & Associates
Leonard S. Jones, JMA Consultants Incorporated
Paul Lambert, Lambert Advisory
Jan Shapin, Jan Shapin and Associates
Judith E. Weber, J Weber Consulting
Cost Study Senior Advisors
Phillip L. Clay, Senior Protocol Advisor
Jeffrey K. Lines, Senior Public Housing Advisor
Georgia Murray, Senior Private Housing Advisor
Public Housing Operating Cost Study                                                                          Draft Research Design
Table of Contents
Table of Contents
Executive Summary................................................................................................................... i
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental
Housing .......................................................................................................... 1
1.1 Conceptual Framework................................................................................................. 1
A. Benchmarking the Operating Costs of Multifamily Housing Developments.............. 1
B. Costs of Assisted or Subsidized Rental Housing....................................................... 4
1.2 Development of Database ............................................................................................. 5
A. Operating Costs from FHA Annual Financial Statements ......................................... 6
B. Information on Property Characteristics from Other HUD Databases and the US
Census..................................................................................................................... 6
1.3 Approach to Modeling .................................................................................................. 7
A. Explanatory Variables............................................................................................ 10
B. Adjusting for Costs Associated with Particular Types of Housing .......................... 12
1.4 Testing the Accuracy of the Model.............................................................................. 14
A. Tests of How Well the Model Fits the Database on Which it is Based .................... 14
B. External Tests of the Model ................................................................................... 15
1.5 Applying the Model to Public Housing ....................................................................... 17
1.6 Distressed Properties................................................................................................... 19
Chapter Two – Case Studies ...................................................................................................20
2.1 Approach.................................................................................................................... 20
2.2 Number and Selection of Case Study Agencies ........................................................... 22
2.3 Supplemental Research............................................................................................... 23
Chapter Three – Debt Financing for Capital Improvements.................................................26
3.1 Research Objectives.................................................................................................... 26
3.2 Approach.................................................................................................................... 27
Chapter Four – Utilities ..........................................................................................................30
4.1 Activities Undertaken ................................................................................................. 30
4.2 Next Steps .................................................................................................................. 31
Chapter Five – Project Schedule.............................................................................................32
Appendix A – List of Regional Public Meetings .....................................................................34
Appendix B – List of Comments/Questions ............................................................................35
Appendix C – Case Study Pilot Sites: Research Team Instructions ......................................46
Appendix D – Comparison of Federal Rules and Regulations in Operating Public and
Assisted Housing.............................................................................................59
Public Housing Operating Cost Study                                                                          Draft Research Design
List of Exhibits
List of Exhibits
Exhibit E.1:  Cost Study Approach ............................................................................................. ii
Exhibit 1.1:  Variables that May Affect Operating Costs of Any Multifamily Rental Housing ....11
Exhibit 1.2:  Variables that May be Needed to Include and Exclude Particular Variations in
Operating Cost .....................................................................................................13
Exhibit 2.1:  Regulatory Differences Between Public and Assisted Housing...............................20
Exhibit 2.2:  Number of Proposed Case Studies, Based on Size of PHA .....................................23
Exhibit 2.3:  Comparison of Asset Management and Property Management Tasks .....................25
Exhibit 4.1:  Profile of Four PHAs Selected for Utility Cost Study.............................................31
Exhibit C.1:  Sample Statement of Operating Receipts and Expenditures ...................................57
Public Housing Operating Cost Study                                                                          Draft Research Design
Executive Summary
i
Executive Summary
Overview
Under a Cooperative Agreement with the United States Department of Housing and Urban
Development (HUD), Harvard University’s Graduate School of Design (GSD) is conducting a
study (Cost Study) to determine the cost to operate well-run public housing.  The results of this
Cost Study are intended to inform the development of a final rule implementing a new Operating
Fund formula for allocating public housing operating subsidies.
There are approximately 1.286 million units of federal public housing, administered by some
3,200 public housing authorities (PHAs).  Total operating expenditures for public housing are
approximately $5.4 billion annually, of which $2.2 billion is paid for with rents and other local
revenue, and $3.2 billion is funded with federal operating subsidies.
In April 2001, GSD published a document entitled Discussion of Research Issues and Initial
Recommendations for Review (April Report) that outlined GSD’s recommended research
approach to the Cost Study (this discussion document, as well as other related materials, may be
found at www.gsd.harvard.edu/phocs).  GSD then conducted a series of regional public meetings
soliciting comment (see Appendix A for the dates and locations of these Regional Public
Meetings).  Based on the input received on the April Report, inclusive of written comments,
public meetings, and other discussions with industry groups, GSD presents this Draft Research
Design.
Comments on the Draft Research Design may be directed to:
Lora Nielsen, Project Staff Assistant
Harvard Graduate School of Design, P.H.O.C.S.
48 Quincy Street
Cambridge, MA  02138
617-496-5065 (phone)
617-496-0052 (fax)
lnielsen@gsd.harvard.edu
Comments should be submitted by July 30, 2001, and can be received electronically, by
facsimile, or by mail.
Approach
In the April Report, GSD recommended that the non-utility cost to operate public housing be
benchmarked to the cost of operating federally-assisted multifamily rental housing with
mortgages insured by the Federal Housing Administration (FHA).1  FHA insures approximately
                                                  
1 As explained in the April Report, the choice of FHA assisted properties versus other assisted housing (such as
assisted housing financed by state housing finance agencies) was made primarily because of the FHA database’s
size, accessibility, number of property variables, etc.  The choice was not made based on any perceived differences
in characteristics between FHA assisted properties and assisted properties that are financed without FHA insurance.
Public Housing Operating Cost Study                                                                          Draft Research Design
Executive Summary
ii
1.0 million units of federally-assisted rental housing.2  This housing is part of a government
program in which owners of the housing are committed by contract to rent units to households
with incomes below a certain level and at rents specified by program rules.  The owners of FHA
properties are required to submit annual financial statements to HUD.  GSD recommended
developing a model that would estimate the cost to operate each public housing property as if it
were FHA assisted housing.  Using multiple regression techniques, this model would identify
determinants of costs according to key property characteristics, e.g., building type, location,
percentage of large-bedroom units, etc.  The results would then be adjusted to reflect reasonable
costs associated with public housing’s unique operating and regulatory environment, as
determined through case studies of selected PHAs.  Finally, GSD would add to this result the
costs of asset management and utilities, as determined through supplemental research.  Exhibit
E.1 illustrates this approach.
Exhibit E.1:  Cost Study Approach
Industry Comments
Appendix B includes a list of the major comments received on the April Report, along with
GSD’s response to those comments.  The comments can be grouped into seven main areas, as
follows:
• concern that the Cost Study is not doing enough to study “actual” PHA costs (as
opposed to studying FHA assisted housing costs);
• concern over the number of case studies to be conducted and the protocol for those
case studies;
• concern over the use of FHA-assisted housing as an appropriate benchmark;
• concern over the recommendations for a new financing program for public housing;
• concern over, and request for additional information regarding, the actual costing
model and how the FHA database, and other potential databases, will be used;
• concern that GSD may not have sufficiently captured the regulatory and other
differences in managing public housing; and
                                                  
2 FHA also insures 0.5 million units of unassisted housing.  As explained in the April Report, GSD plans to use the
data on these unassisted properties to understand the relationship between certain property characteristics and costs;
however, the "benchmark” cost produced by the model is one of operating FHA assisted housing.
FHA
Benchmark
Cost
Adjustment
for
Regulatory
and
Operating
Environment
Utilities
Cost to Operate
Well-Run Public
Housing
+
Asset
Management
Costs
+
+
=
Public Housing Operating Cost Study                                                                          Draft Research Design
Executive Summary
iii
• concerns of a miscellaneous nature, from the request for an appeal mechanism to
information on the costs of state public housing.
Discussion
GSD’s recommendation to benchmark the cost of operating public housing to the cost of
operating FHA assisted housing was based primarily on the issue of circularity (see April Report
for fuller discussion).  The Performance Funding System (PFS), now called the Operating Fund,
determines public housing funding levels and has been in place, essentially unchanged, for 25
years.3  PFS was based on the operating costs of a sample of 56 “well-run” PHAs in the early
1970s.4  As a result, public housing spending levels are dictated by a formula that has no real
relationship to performance or, necessarily, to operating need.  PHAs have accommodated their
operations to the spending levels allowed under this system.  To analyze PHA spending patterns
would only lead back to the formula, without explaining whether, perhaps, PHAs have been
systematically underfunded, overfunded or appropriately funded over the years.  Given this
situation, it seems critical to anchor operating costs to the verifiable operating costs of over a
million housing units serving a similar population and operating under similar regulations.   
GSD does not believe that there is any way around this issue of circularity.  Therefore, it
recommends moving forward with the proposed benchmarking approach and modifying it based
on input received, as explained below.  At the same time, and also based on industry feedback,
GSD has expanded the scope of the case studies as well as the number of agencies to be studied.
Although not mentioned in the April Report, GSD indicated at the regional public meetings that it
was contemplating conducting between 6 and 12 case studies.  GSD now expects to conduct
between 20 and 30 case studies.
A number of requests were made by industry groups to review the case study protocols before
they are finalized.  GSD agreed to such a review in its April Report.  To allow for even greater
participation by the industry groups in the development of the case study protocols, GSD has
decided to conduct a series of case study pilot sites.  Research team instructions for the case study
pilot sites are included as Appendix C.  GSD plans to meet with industry groups following these
pilot sites to share observations and accompanying recommendations for the final protocols.
With respect to whether FHA assisted housing is comparable to public housing in terms of
population served, location, building characteristics, or other features, GSD believes that FHA
assisted housing is sufficiently similar to public housing to build a reasonable costing model.
Moreover, although the statistical techniques that GSD plans to use do not require the costing
database to contain a distribution of property characteristics that exactly match the distribution of
characteristics in public housing, the properties in the FHA database do match up very well to the
public housing stock.  For example, the median household income in FHA assisted housing
averages 26.2 percent of the local median and in public housing it averages 24.2 percent.  Still, to
help evaluate the regression model’s predictive abilities, GSD will field test the model at a sample
of not less than 50 public housing properties.  At these sampled properties, GSD will employ
qualified property management professionals to prepare independent budgets for each of the
                                                  
3 Technically, the PFS was replaced by the Operating Fund as a result of the 1998 Quality Housing and Work
Responsibility Act.  However, the mechanics of the PFS remain essentially unchanged, pending the completion of
the Cost Study and subsequent rulemaking.
4 The History and Overview of the Performance Funding System: Evaluation of the Performance Funding System,
May 18, 1979, with clarification provided by the Office of Policy Development and Research, HUD, June 2001.
Public Housing Operating Cost Study                                                                          Draft Research Design
Executive Summary
iv
properties without knowledge of what the model-predicted costs might be and without reference
to what the current expenditures on these properties are (even if the PHA is able to provide
project-based cost accounting for these properties, the developed budgets will be prepared anew).
The developed budgets will then be compared against the model-predicted costs.  Among other
objectives, this field testing will indicate whether certain public housing property characteristics
have not been sufficiently captured in the model and, therefore, whether refinements to the model
may be necessary.
With respect to GSD’s discussion in the April Report of a new financing program for public
housing (Chapter Six, Building a Platform for Broader Change), GSD received strong support for
the ideas in this chapter from multifamily housing experts who work outside of public housing
but less support from within the public housing community.  A number of respondents questioned
why this discussion was included in the April Report and wondered whether these ideas reflected
a bias in the Cost Study.  There are a number of reasons why GSD included this chapter in the
April Report.  First, these ideas grew out of focus group discussions with both smaller PHAs and
with financial institutions.  Many of the smaller PHAs expressed the belief that they could operate
more effectively if they were treated more like other assisted housing, both in terms of regulations
and funding.  Second, one of the fundamental components to successful property management is
the ability to plan for the orderly replacement of building systems.  While the establishment of
replacement reserve accounts is now technically permitted in public housing, few PHAs can
afford to fund such activities because of the need to address capital backlog needs.  Implementing
a debt financing program for capital improvements would allow PHAs to address their backlog
needs and fund replacement reserves.  In the long run, such changes would almost inevitably lead
to a more effectively managed public housing program and a better preserved physical stock.
That being said, the model-building process that is the core task of the Cost Study is entirely
independent of the notions explored in Chapter 6 of the April Report.  As discussed in more detail
later in this document, GSD will examine the ideas for developing a new financing program more
fully by studying this issue at a select number of agencies.
With respect to the methodology for the costing model, and how the FHA and other databases
will be used, GSD has expanded the discussion in this document to explain the steps and
processes for developing the costing model in more detail.
Finally, with respect to whether GSD had adequately captured the regulatory similarities and
differences between public and assisted housing, GSD has prepared a new, more extensive chart
on the differences between these two programs (see Appendix D, Comparison of Federal Rules
and Regulations in Operating Public and Assisted Housing).  Based on industry feedback, this
chart contains items not previously included in the April Report, e.g., the requirement to allow for
pets in family public housing.  This chart is one of the primary tools that will be used during the
case studies; as with this entire document, GSD is seeking comment on this chart before the case
studies begin.  Several comments were received responding not to the differences in regulations
but the fact that public housing may be subject to differing standards of enforcement or that, as
public bodies, PHAs are often required to implement these requirements more extensively than
private operators.  One of the purposes of the case studies is to examine more fully the context of
local implementation.
Industry Review and Participation in Cost Study
As requested, GSD is examining ways to foster wider industry participation as the Cost Study
progresses.
Public Housing Operating Cost Study                                                                          Draft Research Design
Executive Summary
v
GSD shares progress reports at monthly public meetings in Washington, D.C.  These meetings
are typically attended by representatives from the major public housing trade/membership
associations, as well as other interested parties.  GSD is recommending, and soliciting comment
on, the establishment of a research working group that would include representation from
industry and related groups.  The purpose of the working group would be to provide more in-
depth feedback to GSD as the Cost Study progresses.  For example, GSD envisions several
meetings to review the initial regression model results or the observations from the case study
pilot sites.
GSD has also been asked to reconvene the Operating Fund Negotiated Rulemaking committee
that was originally organized to develop a new Operating Fund formula and that had
recommended that a cost study be conducted.5  Although GSD has no authority to reconvene or
reestablish the Operating Fund Negotiated Rulemaking committee – that authority rests solely
with HUD – GSD believes that the aforementioned research working group could serve much the
same purpose.  Further, GSD has prepared this draft research design so that the work produced
through the Cost Study will be useful to any future negotiated rulemaking efforts.
Organization of Report
The remainder of this report is organized into five chapters.
• Chapter One – Developing Benchmark Operating Costs for Assisted
Multifamily Rental Housing, discusses the recommended methodology for
developing the costing model using, primarily, the FHA database.
• Chapter Two – Case Studies, discusses how the GSD intends to conduct the case
studies of selected PHAs that will be used to adjust the model-predicted costs for the
different regulatory and operating environments between public and assisted housing.
• Chapter Three – Debt Financing for Capital Improvements, discusses the
additional research to be conducted to explore further the feasibility of converting to
a financing program for public housing that is more like other assisted housing.
• Chapter Four – Utilities, discusses GSD’s on-going feasibility study on
benchmarking utility costs to norms in private housing and the implications for
operating subsidies.
• Chapter Five – Project Schedule, includes a schedule for the completion of tasks
under the Cost Study.
                                                  
5 See March 16, 1999 Federal Register Notice establishing the Negotiated Rulemaking Committee on Operating Fund
Allocation.
CHAPTER ONE
Developing Benchmark Operating
Costs for Assisted Multifamily
Rental Housing
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
1
Chapter One
Developing Benchmark Operating Costs for
Assisted Multifamily Rental Housing
This chapter describes how GSD will proceed to develop a model to determine the benchmark for
the non-utility cost of operating assisted multifamily rental housing.  This benchmark cost will be
used, along with adjustments described in subsequent chapters, to determine the cost of operating
well-run public housing.  The treatment of utility costs is discussed in Chapter Four, Utilities.
1.1 Conceptual Framework
A.
Benchmarking the Operating Costs of Multifamily Housing Developments
The Cost Study will develop a model that uses the operating costs of FHA-insured multifamily
rental housing to produce benchmark operating costs for public housing.6  The premise behind
this approach is that there is an underlying core cost for operating any multifamily rental housing.
Most management and maintenance functions are common to multifamily rental housing,
regardless of ownership and tenancy.  However, operating costs, considered on a per unit basis,
will vary from property to property in some fairly well understood ways.
In the reconnaissance phase of the Cost Study (described in the April Report), GSD gathered
information on the sources of variation in operating costs in multifamily housing.  GSD tapped
many sources of information, as follows:
• GSD conducted a review of the literature on the determinants of operating costs in
multifamily housing, and specifically on the operating costs of public housing;
• GSD carried out field research in four cities that compared the costs of public
housing with the costs of multifamily assisted housing;
• GSD conducted focus groups with experts from the public housing and private
multifamily housing industries to discuss the factors that influence operating costs;
• Real Estate experts who are members of the Cost Study team provided advice based
on their experience as housing managers;
• GSD conducted a number of site visits to individual PHAs and examined various
PHA operating statements;
• at meetings in Washington, D.C., and around the country, public housing industry
groups and individual public housing administrators provided input based on their
experience as public housing managers; and
                                                  
6 The FHA database includes all properties with FHA-insured  (or formerly FHA-insured but now HUD-Held)
mortgage loans, as well as nonprofit housing for the elderly developed under Section 202.  Most of these properties
have five or more units at a single site, but there are some scattered site properties.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
2
• preliminary data analysis, based on cross-tabulations of several databases on
multifamily housing, suggested ways in which property characteristics (including
location) may affect costs.
The exploration of multifamily rental housing databases confirmed the basic premise that there
exists a core cost of operating multifamily rental housing.  While there is a wide range of per unit
operating costs for individual properties, large databases, populated by sets of properties that
overlap to a limited extent, demonstrate very similar average operating costs for properties in
similar geographic locations.
At the same time, all researched databases showed that assisted housing costs more to operate
than housing that is not part of a government program with restrictions on rent and income.
Therefore, while GSD will use FHA-insured housing that is not assisted as part of the database on
which GSD will build a cost model, the benchmark for public housing will be a cost benchmark
for operating assisted housing.
The reconnaissance phase of the Cost Study led GSD to the following hypotheses about sources
of operating cost variation.
• Costs will vary substantially from locality to locality.  For example, costs will vary
for different metropolitan areas or states, different types of locations (central cities,
suburbs, non-metropolitan areas), and different regions of the country.  These
variations in costs will occur because of differences such as wage rates, housing
market conditions (that could, for example, affect turnover rates), the difficulty of
doing business in different types of communities, the quality of local public services
and infrastructure and—perhaps—management functions that could differ in
different areas.
• Costs per unit will vary with the size of the multifamily property.  There are
expected economies of scale associated with the number of units in a property.
However, there may be diseconomies for certain types of properties beyond a certain
size.
• Costs per unit will vary with the mixture of number of bedrooms per unit, with
higher costs associated with more bedrooms per unit.  For example, units with
three or more bedrooms may house families with children who are old enough to
place demands on common space.
• Costs will vary with building type.  Each type of rental development (elevator high
rise, garden apartments, townhouses, etc.) has some unique management and
maintenance functions.  The costs of these different functions are likely to be
different.
• Costs will vary with the types of households occupying the property.  As already
suggested in connection with unit sizes, operating costs are expected to differ by the
age and number of residents in each unit and by the amount of time residents spend at
home, creating additional demands on management as well as more wear and tear on
units and common space.
• Costs may vary with the age of the property or with the date when the property
was most recently rehabilitated.  Some older systems may require additional
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
3
routine maintenance or may create major repair needs (e.g., leaking roofs or leaking
utility distribution systems).
• Costs may vary for properties operating in different types of neighborhoods.
There may be additional costs associated with the management of multifamily rental
housing in a “distressed” neighborhood—for example, because of additional needs to
secure exterior and common space from intruders or to repair damage that may be
caused by such intrusions.
The reconnaissance phase of the Cost Study also led GSD to understand that there will be cost
variations from property to property that are idiosyncratic.  These cost differences are the result of
choices made by owners and managers on such issues as how to carry out management functions,
the level of housing services to provide to residents, and how to treat non-recurring expenses for
accounting purposes.  It may not be possible to capture these cost differences in a model, but it
also may not be necessary or even appropriate to do so.  Instead of attempting to measure these
sources of cost variation, the model that creates benchmark cost estimates reflects the average
choices made by property owners and managers.
To further clarify, the choices made by owners and managers that create idiosyncratic cost
differences may include:
• differences in operating style, such as how many staff to hire based on the number
of units and whether to contract for management and maintenance functions or to
perform them with in-house personnel;
• differences in the level of housing services provided to residents, which may in
turn reflect the position of the property in the housing market or the motivations of
the owner; and
• differences in the investment and tax strategies of different owners, which means
that work that is on the borderline between routine maintenance and capital
improvements might be included in operating costs and “expensed” for income tax
purposes for some properties and not for others.
Finally, there are some sources of cost variation that should not be reflected at all, and not
included in the cost estimates that are used to benchmark the costs of operating public housing.
In developing the costing model, GSD will attempt to control for these sources of cost variation7
and to exclude the higher or lower costs associated with them from the benchmark costs.  Some
of these sources of inappropriate cost inflation/deflation are described below:
• Costs that have been systematically constrained by a funding system associated
with a government program.  If an entire group of properties has operating costs
that make it difficult to operate those properties effectively, the lower operating costs
of those properties should not be included in the benchmark.
                                                  
7 For example, by using “dummy variables” in the regression analysis that represent either the presence or the absence
of a particular property characteristic to measure its effect and exclude that effect on costs from the benchmark.
Alternatively, GSD may take the properties that exhibit the characteristic out of the database on which the model is
built.  See below for more on the planned use of dummy variables in the development of the model.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
4
• Costs that have been systematically inflated by the rules of a government
program that restrict the extent to which the property’s cash flow can be taken by the
owner as profit.  If an entire group of properties has higher operating costs than
otherwise similar properties because of an incentive created by “limited dividend”
rules to spend excess cash flow on the property, those additional costs should not be
included in the benchmark.
• Operating costs associated with “luxury” or very high rent housing that provide
additional services to residents (such as doormen or elaborate recreational amenities)
in return for the rent.
B.
Costs of Assisted or Subsidized Rental Housing
The reconnaissance phase of the Cost Study found that the operating costs of multifamily rental
housing are higher for housing that is part of a government program than for unassisted market
rental housing.  Given that public housing is similar to assisted housing (although not identical) in
its regulatory environment and rent and income limitations, the additional costs associated with
operating multifamily rental properties as assisted housing should be included in the benchmark
operating costs for public housing.
The higher costs of government assisted housing may be the result the following issues:
• Government regulation, which may impose costs by creating additional functions
for housing management.  An example is the requirement that managers collect
income information from residents every year.  Government regulation may also
impose limitations on business operations that could make them less efficient, require
additional record-keeping and reporting, or impose additional requirements for
services to residents.  There may also be additional costs incurred by the oversight
activities of public agencies, including extra inspections, substitutions of owner
judgements by oversight agencies, and various approvals.
• The mission of serving low income tenants, which may lead to management
choices that are different from the choices made for unassisted rental housing.  It is
possible that the operating costs of assisted housing differ from the operating costs of
unassisted housing in ways that are not explained either by the costs of regulation or
by easily defined and measurable household characteristics.  Characteristics such as
the greater number of children per unit in assisted housing and the greater likelihood
that non-elderly adults in assisted housing are not working and spend more time at
home can be measured and, therefore, can be used directly in cost benchmarks.
However, owners and managers may provide additional services to residents of
assisted housing that would not be needed by residents of market rate housing in
ways and for reasons that may not be entirely measurable through resident and
property characteristics.  The benchmark costs produced by the model will be based
on the assumption that the average levels of such services included as operating costs
for assisted housing are the appropriate level to be funded as part of the operating
costs of public housing.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
5
1.2 Development of Database
Given this conceptual framework, GSD sought to develop a database that:
• provides accurate information on the operating costs of multifamily rental properties;
• 
includes information on all of the characteristics of each property that are sources of
possible cost variation that should be taken into account when benchmarking the
operating costs of public housing; and
• consists of as large a number of properties as possible, so that (1) the model provides
accurate estimates of the effect of measurable property characteristics on costs and
(2) the sources of cost variation that cannot be specifically enumerated will balance
out into reasonable average costs.
An additional consideration was that property characteristics used to create a model for
benchmarking costs should be available for public housing developments, so that the benchmarks
could be applied to public housing.
As discussed in detail in the April Report, GSD concluded that a database that uses the Annual
Financial Statement data for multifamily properties insured by FHA or developed under the
Section 202 Housing for the Elderly program best meets these aforementioned criteria.  GSD
combined Annual Financial Statement data with data on property characteristics from other
sources to create an “FHA database” that will be used to develop a model of operating costs.
• The FHA database8 is large, with wide geographic coverage.  The 1999 Annual
Financial Statements data set includes data on more than 16,000 properties and more
than 1.5 million units.  All regions of the country and all types of locations—central
city, balance of metropolitan area (suburbs), and non-metropolitan—are well
represented.
• The properties in the FHA database have many characteristics that are also found in
the public housing stock and that are likely to influence operating costs.  In
particular, the database includes properties that have multiple bedrooms in each unit
and serve families with children, as well as properties with smaller units that serve a
predominately elderly population.  Residents of FHA properties include many
persons with disabilities and many single parent families.  The FHA database
includes all building types found in multifamily rental housing and properties with a
wide variety of sizes (total number of units in the property) and age (original
construction date and date of the current mortgage).  Many properties in the FHA
database are found in census tracts with high rates of poverty and other indicators of
social distress.
• The FHA database created by GSD includes data elements that can be used to
measure the effect of property characteristics on operating costs.  In addition to
property size, building type, unit mix, building age, and variables relating to the
                                                  
8 While Section 202 properties are not FHA insured (they were developed mostly with direct grants), the financial
statements on these properties are submitted annually to HUD.  For convenience, GSD is calling the database that it
has created the  “FHA database.”
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
6
property’s location, GSD has information on the demographic characteristics of
residents and indicators of the property’s financial soundness and physical condition.
• The FHA database has a high level of data integrity.  The operating costs of
multifamily properties based on FHA Annual Financial Statements are subjected to
independent audits and separate quality control checks.
A.
Operating Costs from FHA Annual Financial Statements
Annual Financial Statements (AFS) are full financial statements (including Statement of
Financial Position, Income Statement, Statement of Cash Flows, and various supplemental
schedules required by FHA) prepared under Generally Accepted Accounting Principles,
submitted by the property owner after the end of the property’s fiscal year.  Most owners are
required by FHA to submit AFS that have been audited by an Independent Public Accountant
before the data are submitted to HUD; the remaining owners are required to have a subsequent
audit and reflect any adjustments in the next year’s AFS.
Beginning with AFS for the fiscal year ending 12/31/98, owners have been required to submit
AFS electronically to HUD’s Real Estate Assessment Center (REAC).  Previously, AFS were
submitted in hard copy and entered into an electronic database by a HUD contractor.  AFS data
are subjected to extensive quality control checks by REAC, as they are used to determine whether
a property is financially troubled or at risk of becoming financially troubled.  All of the data from
fiscal years 1996 through 2000, whether submitted in hard copy or electronically, has been
captured in the database which GSD is using for modeling purposes.
The FHA-insured operating costs relevant for benchmarking the non-utility operating costs of
public housing properties is defined as the sum of the following items on form HUD-92410,
Statement of Profit and Loss (the items contained on form HUD-92410 have remained largely
intact over the years).
• Total administrative expenses (Line 6200/6300)
• Total operating and maintenance (Line 6500)
• Total taxes and insurance (Line 6700) minus real estate taxes (Line 6710)
Because more recent data are more complete and have been subjected to more rigorous quality
controls, GSD is likely to base the model for benchmarking public housing operating costs on
Annual Financial Statements for fiscal years1999 and 2000.
B.
Information on Property Characteristics from Other HUD Databases and the
US Census
The database containing FHA Annual Financial Statement data has been augmented to include
additional data elements from the following databases:
• HUD’s Office of Housing Real Estate Management System (REMS).  This
administrative database contains a wealth of information at the development level.
For example, it includes variables on the number of units in each property, the
distribution of units by bedroom size (i.e., one-bedroom, two-bedroom, etc.), building
type (high rise, garden, townhouse, etc.), mortgage sponsor type (for-profit, non-
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
7
profit, limited dividend), occupancy type (family, elderly/disabled), HUD program
(section of the act for FHA programs and what fraction of the units have Section 8
assistance), and the location of the property.
• HUD’s Office of Policy Development and Research (PD&R) A Picture of
Subsidized Households 1998 database.  From this database, GSD has added to the
FHA database variables describing the characteristics of the tenants occupying each
assisted property, including income, source of income, and size and structure of the
household.  Data on tenant characteristics are aggregated to the property level from
the Tenant Rental Assistance Characteristics (TRACS) system.  TRACS is a
household-level data system to which assisted housing property managers report each
month information from the certification of income and characteristics of household
members that is required at program admission and annually thereafter.
• 1990 Census of Population and Housing.  From the US Census, GSD has added
variables that can serve as proxies for the level of distress of the neighborhood
represented by the census tract in which the property is located.
• Real Estate Assessment Center (REAC) Physical Inspection Scores.  HUD’s Real
Estate Assessment Center has inspected each property in the FHA database at least
once.  GSD has obtained from REAC both the most recent overall physical
inspection score for each property, as well as a set of sub-scores associated with
capital needs.
• Office of Housing’s Field Office Multifamily National System (FOMNS).  This
database includes additional property characteristics, such as original age of
construction, square footage, heating and cooling system, building materials, and
management type.
• Office of Housing’s F-47 database.  This database contains the mortgage
endorsement date information, which is useful for determining the age of a property’s
mortgage.
1.3 Approach to Modeling
The following discussion is intended for non-technical readers who want to understand the details
of GSD’s approach to, and how to interpret the results of, this important part of the research.
However, in order to communicate GSD’s plans accurately and precisely, the discussion uses
(and defines) some technical statistical terms.
GSD will develop a cost model through the use of multiple regression analysis.  This is a
statistical technique that allows an outcome measure (called the dependent variable, in this case,
operating costs per unit month) to be expressed as the result of the combination of characteristics
that affect it (called the explanatory variables or independent variables, in this case, the factors
that drive operating costs) multiplied by their respective regression coefficients (called regression
weights).
The regression coefficients are the key product of the modeling exercise.  Each explanatory
variable has a regression coefficient that expresses in quantitative terms the extent to which the
explanatory variable is found by the model to determine the dependent variable.  Generally, an
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
8
explanatory variable that has a larger coefficient is more strongly correlated with the dependent
variable than an explanatory variable that has a smaller coefficient.9  If the sign of the coefficient
is negative, then the explanatory variable is negatively related to the outcome variable, meaning
that an increase in the value of the explanatory variable is associated with a decrease in costs; if
the sign of the coefficient is positive, then the explanatory variable is positively correlated with
the dependent variable.  The regression model holds other factors constant, while estimating each
coefficient.  Thus, the coefficient can be interpreted as the independent effect of each explanatory
variable on the outcome, holding constant the values of all other variables in the model.
There are three types of explanatory variables used in the model: dummy (or indicator) variables,
categorical variables, and continuous variables.
Dummy (or indicator) variables are “yes / no” variables: in the model, they take on a value of
zero or one.  An example is the variable “senior building”.  The coefficient for a dummy variable
is interpreted as the effect of having a “yes” value for the variable, relative to having a “no”
value.  Thus, if the model generates a negative coefficient for “senior building”, it would mean
that operating costs are lower in senior buildings than in non-senior buildings, all else held
constant.
Dummy variables are one type of categorical variable.  Categorical variables are variables that
refer to groups or categories of observations (in this case, properties).  In the case of dummy
variables, the category happens to have only two values: yes or no.
When GSD uses categorical variables, each explanatory variable includes a value that does not
receive a coefficient (called the omitted value, or reference category.)  The effect of the omitted
value is included in the statistical term called the intercept that becomes part of the estimate of
the value of the outcome measure or dependent variable.10  What the coefficients represent, then,
is the effect of being in one of the categories, relative to being in the reference category.  The
omitted value is often the value most typical of the population, since this makes it easier to
explain why the coefficients behave as they do.
For example, a variable on building type might have five values: high rise/elevator, garden/walk-
up, row/townhouse, detached or semi-detached, and other.  GSD might make garden/walk-up the
omitted value, since that is the most typical building type.  The coefficients then express the
extent to which the statistical analysis determines that each other building type has costs that
differ from the costs of garden/walk-up apartments.  Hypothetically, the model might show that
high rise/elevator buildings cost $10 more to operate per unit-month (after controlling for other
property characteristics), while detached and semi-detached units cost $2 less per unit-month,
compared in both cases with garden/walk-ups.11
                                                  
9 This statement assumes the two explanatory variables being compared have the same units of analysis.  For example,
both may be measured in dollars.  The statement does not hold true if the two explanatory variables have different
units of analysis.  Also, this statement assumes that both coefficients are statistically significant.  In addition to
coefficients, the model also generates a measure of the statistical significance of each coefficient, called the standard
error of the coefficient.  If a coefficient is not statistically significant, it means the relationship between the
explanatory variable and the outcome variable cannot be precisely estimated, and GSD cannot with confidence
interpret the coefficient as being different from zero, even though the coefficient might be very large.
10 For simplicity, the intercept can be thought of as the base value of the dependent variable, to which the quantities
indicated by the coefficients are added or subtracted.  But it is an arbitrary base value, since its size depends on the
choices made in structuring the model (such as which values of categorical variables to omit).
11 This statement assumes the dependent and independent variables are both entered linearly.  As noted below, GSD
may use a model in which the dependent variable is in log form, then GSD would be able to interpret the coefficients
as meaning that a high-rise/elevator costs five percent more to operate than a walk-up, while a detached building cost
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
9
Continuous variables are variables that take on a numerical value, rather than variables that refer
to categories of observations.  For example, property size (number of apartments) could be
entered as a continuous variable.  The coefficient for a continuous variable represents the effect
on the dependent variable resulting from a one-unit increase in the explanatory variable.  For
example, a coefficient on property size equal to - .02 means that for every one-unit increase in
property size, operating costs decrease by .02 dollars, or equivalently, for every 100 unit increase
in property size, operating costs decrease by 2 dollars per unit-month.12
GSD will likely use the most common form of regression, Ordinary Least Squares (OLS).  The
mathematical formula created by OLS predicts the dependent variable in a way that minimizes
the extent to which any actual value of the dependent variable (i.e., the operating cost of any
property in the database) differs from the predicted value.  More specifically, OLS is a rigorous
mathematical optimization algorithm that produces the regression coefficients for a given model
and a set of observations.  This estimation process is performed by high-speed computers and the
use of statistical software packages such as SAS and Stata.
GSD is considering the use of alternative regression estimators, including Least Average
Deviation (LAD)13 and iterative re-weighted least squares or robust regression.  In addition to
comparing results using alternative estimators, GSD will use influence statistics to identify
observations (properties) that have a very large effect on an OLS model and seek more
information about them.  In particular, GSD will look for factors external to the model or the
database that may be driving the costs of these properties.  It may be preferable to exclude certain
idiosyncratic properties from the model and to use OLS, rather than turning to a less commonly
used type of regression.
The value GSD is seeking to predict is operating cost per unit month.  Ultimately this must be
expressed as an absolute value (dollars).  However, in the development of the model, GSD is
strongly considering expressing this dependent variable as a logarithm.  One advantage of
transforming the dependent variable in this way is that it will enable GSD to estimate the effect of
the explanatory variables on the dependent variable as the percentage change created in the
dependent variable by a unit change in the explanatory variable.  The regression coefficients
produced by the model will be percentage changes.  To use the example of building type: it may
be more accurate to conclude that an elevator building costs 5 percent more to operate than to
conclude that it costs $10 more to operate per unit month regardless of the base cost to which that
$10 is added.
                                                                                                                                                      
one percent less to operate than a walk-up, all else equal.  Note that these examples are hypothetical and not to be
taken as actual results of modeling.
12 Once again, this statement assumes GSD is using a linear model.  In a log-linear model, a coefficient equal to A
means that a one-unit increase in the explanatory variable is associated with an increase of size A in the log of the
explanatory variable.  Fortunately, log-linear models are easy to interpret.  This is equivalent to saying that a one-
unit increase in the explanatory variable is associated with a percentage increase of size A in the outcome variable.
Please note that this is a hypothetical example.  GSD may decide to use size as a categorical variable.
13 The distinction between OLS and LAD is that OLS, by squaring and minimizing deviations from the predicted
value, accentuates the effect of extreme values or outliers.  This may make the model predict less well on average.
LAD, by not squaring deviations, may create a model that predicts more accurately for a larger number of
observations, but with greater inaccuracy at the extremes.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
10
A.
Explanatory Variables
The model building starts with the most simple models that include those property characteristics
most likely to affect costs and expressed in the most straightforward terms.  Additional variables
are then added or the form of existing variables is changed to determine whether those additions
or changes increase the explanatory power of the model.  In each case, either a new variable or a
new form of the variable must be based on a hypothesis about its relationship to operating costs.
The GSD project team must be satisfied that the data underlying the variable are reasonably
accurate and that the variable’s meaning is unambiguous.  Other desirable characteristics of any
additional variables are:
• 
the relationship between the variable and operating costs must be easy to explain; and
• 
the variable must already exist for public housing in one of the HUD databases, or
the cost of collecting it for both HUD and the PHAs must be modest.
The new or changed variable will be included in the model only if it is found to have a significant
statistical relationship to operating costs and only if it improves the overall ability of the model to
predict operating costs.
As an example of changing the form of a variable, take the age of the mortgage or building.  GSD
might start with age as a continuous variable.  Assuming that older properties cost more to
operate (which the model may or may not find to be the case), age as a continuous variable would
produce a coefficient showing how much each year of age adds to operating costs.  GSD might,
however, find the coefficient for this continuous variable does not meet a statistical test of
significance.  GSD might then try dividing age into categories (0-10 years, 10-15, 15-20, 20-25,
25-30, etc.), based on the premise that properties between 20 to 25 years old are all similar in the
effect that aging building systems have on operating costs.14
As GSD adds variables or changes their form (from continuous to categorical, shifting the
boundaries of the categories, etc.), GSD will watch for their effect on other variables.  It may be
that the new variable in effect measures the same thing as a variable already in the model, and
this causes the coefficients of the original variable to become statistically insignificant or difficult
to interpret in a reasonable way.  For example, a variable measuring the number of children per
unit in a property might measure the same thing as the average number of bedrooms per unit.
The addition of the “children” variable could make it appear that larger units are less expensive to
operate than smaller units (the confounding effect of variables that measure the same thing is
called multicollinearity.)  On the other hand, the number of bedrooms per unit and the number of
children per unit might well be found to have separate effects on operating costs, and in this case
GSD would want to include both in the model.
Exhibits 1.1 and 1.2 provide a snapshot of the variables being tested as GSD develops a model to
benchmark operating costs.  The tables indicate the source of data for each variable and the
theoretical assumption about how the variable affects operating costs.  The tables also include
comments explaining special issues or plans for the use of the variable.  Exhibit 1.1 includes
                                                  
14 GSD will also test whether any of the explanatory variables should be expressed in logarithmic form, which would
produce a change in the dependent variable associated with a percentage change in the explanatory variable.  It is
less likely that GSD will recommend such a model (a log-log model, rather than a log-linear model), because GSD
does not believe it will find much statistical advantage in expressing both dependent and independent variables in
percentage terms.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
11
Exhibit 1.1:  Variables that May Affect Operating Costs of Any Multifamily Rental Housing
Variable
Source
of Data
Expected Effect on Cost
Comment
Building type (high
rise, garden,
townhouse, semi-
detached, detached)
REMS
High rise/elevator more expensive
because of elevator maintenance and
costs associated with common hallways
and access.
Model may not work well for
scattered-site, single-family units.
Property size (number
of units)
REMS
In general, expect economies of scale,
so per unit cost should decrease with
property size.
Test for types of properties with
diseconomies:  e.g., interact with
unit mix or number of children.
Test if unassisted large and high
cost properties are luxury housing
(high rent).
MSA (for non-MSA
properties, state)
REMS
MSAs with high wage rates and other
local cost drivers (population density,
size of MSA, high rental turnover) will
have higher costs.
Test wage rates alone or
combined with other alternatives
to MSA variables.
Property is housing for
elderly
REMS
Housing for elderly is less expensive to
operate because elders create less
wear and tear on units and common
space; however, elders may place more
demands on management staff.
Test if unassisted high cost
elderly properties are luxury
housing (high rent).
Average bedrooms
per unit
REMS
More bedrooms in a unit are associated
with higher costs because of added size
of units, common areas, and building
systems; and because of use by larger
number of people.
Persons per bedroom
A Picture of
Subsidized
Households
Additional people associated with
higher costs.
Public and assisted housing may
have more intensive occupancy
because of rules matching
household types to unit sizes.
Percent units with 3+
bedrooms
REMS
Units with 3 or more bedrooms house
older children, creating wear and tear
on units and common areas.
Test alternatives:  children per
unit, people per unit.
Combined occupancy
by elderly and
persons with
disabilities
A Picture of
Subsidized
Households
Mixed population creates conflicts
managers must spend time resolving.
Test variable for assisted housing
not in the Section 202 program.
Mortgage age
F-47
Older mortgages associated with higher
operating costs as building systems
age.
Interact with assistance, because
unassisted properties may have
refinanced without rehabilitation.
Not available for public housing,
although could collect data from
PHAs on date and cost of
modernization by property.
Property age
FOMNS
Older properties are associated with
higher operating costs as building
systems age.
Date of substantial rehabilitation
(represented by mortgage age)
may be more relevant.
Capital backlog
REAC physical
inspection
results
Properties with backlog of capital needs
have higher operating costs.
Test variable based on inspection
results that suggest capital
backlog
Location type (central
city, suburb, non-
metro)
REMS
Central city properties cost more to
operate; non-metro cost less, both
because of wage rates and because
some management functions differ.
Test for separate effect of
location type, even when MSA
and state dummy variables are
used.
Turnover rate at
property (percent
move-ins in past year)
A Picture of
Subsidized
Households
Properties with high turnover rates have
higher operating costs because of unit
preparation and/or marketing.
Younger adults and
older children at home
A Picture of
Subsidized
Households,
Census
Properties with high percentages of
such households cost more because of
additional management attention, wear
and tear on units and common space.
Test several forms:  single
parents in property, single
parents in census tract, residents
on welfare, families in census
tract on welfare, etc.
Social distress of
neighborhood (poverty
rate, rate of single
parenthood in census
tract)
Census
Properties in distressed neighborhoods
cost more because of security needs,
vandalism.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
12
variables believed to be associated with variations in operating cost for any type of property,
whether it is privately or publicly owned, regardless of whether it has the restrictions applied by a
government housing assistance program, and regardless of whether it is luxury housing.
B.
Adjusting for Costs Associated with Particular Types of Housing
Exhibit 1.2 lists variables that may be used to make sure that the benchmark costs include all of
those cost variations appropriate to public housing and that they do not include inappropriate cost
variations.  There are several ways in which the model could be designed—or applied—to adjust
for costs associated with particular types of housing that have cost variations that should not be
included in the benchmark.
• GSD could limit the database on which the model is based to a subset of FHA
insured and Section 202 multifamily rental properties that do not have the particular
characteristic that affects costs in inappropriate ways.  GSD believes that it may be
preferable to include the additional properties that have the inappropriate
characteristics but remove the effect of the characteristic from the model.  Excluding
these properties altogether might unduly limit the number and types of properties on
which the model is based, losing valuable information about property characteristics
that the omitted subsets of the database have in common with public housing (e.g.,
location, property size, occupancy by families with children).  Instead, GSD will
define as narrowly as possible those properties that will be omitted from the database,
and will omit properties only in cases for which other options for adjusting for
inappropriate costs do not work well.
• GSD could use dummy or categorical variables to isolate the effect on operating costs
of a particular property characteristic.  GSD would then omit that cost effect from the
application to public housing.  For example, suppose GSD determines that unassisted
properties with very high rents have additional costs that should not be included in
the benchmark.  GSD could use a dummy variable that indicates whether a property
has this characteristic or not.  In applying the model to public housing, GSD will
assume public housing is not high rent/luxury housing and not apply the coefficient
associated with luxury housing.
• GSD could use combinations or interactions of variables in a way that isolates the
effect of a property characteristic that occurs only when that particular characteristic
is combined with another particular characteristic.  For example, GSD may find that
the model predicts costs more accurately if certain variables are associated with the
fact that a property is government assisted housing.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
13
Exhibit 1.2:  Variables that May be Needed to Include and Exclude Particular Variations in
Operating Cost
Variable
Source of
Data
Effect on Costs
Comment
Housing assistance
(whether the property is
assisted by a
government program or
unassisted)
REMS
Assisted housing costs more to
operate than unassisted housing
because of regulatory requirements
and constraints.
Public housing will be treated as
assisted housing in the application of
the model.  Test alternative definitions
of assistance (percent Section 8 only;
percent Section 8, Section 236, or BMIR
combined) for effect on operating costs
and for interaction with other variables.
Test both continuous and categorical
variables.
Mortgage sponsor type
(for-profit, non-profit,
limited dividend)
REMS
Some properties owned by non-
profits and limited dividend sponsors
may have inflated operating costs
because of inability to take out cash
flow as profit.
Examine cash flow and replacement
reserves.  Interact with older/newer
assisted, since this phenomenon may
affect only newer assisted properties
(which for many years received
automatic annual adjustments to rents).
Type of assistance
program (older or
newer assisted)
REMS
Older assisted properties may have
lower operating costs because
operating costs were constrained by
budget-based rents negotiated
between owners and HUD.
Examine cash flow and other indicators
of inadequate funding.  Test interaction
with ever troubled and REAC score
variables.
Ever troubled (property
is or was declared
troubled by HUD)
REMS
Troubled properties may have lower
operating costs because constrained
by rents; or owner decision to spend
little on property may have led to
troubled status.  Or they may have
higher costs because of deferred
maintenance or ineffective
management.
Test both effect on dependent variable
and other effects on model of omitting
troubled properties.
REAC physical
inspection scores
REAC
Low scoring properties may have
lower operating costs because
operating costs were constrained by
rents, or because owner decision to
spend little on property may have led
to physical distress.  Alternatively,
properties with low-scores may be in
poor condition and hence have
higher operating costs.
Test both effect on dependent variable
and other effects on model of omitting
low scoring properties.
Cash flow
AFS
Negative or low cash flow properties
may have lower operating costs
because constrained by rents.
Interact with housing assistance type,
because this phenomenon may affect
only older assisted properties, which
have had budget-based rents.  Test
both effect on dependent variable and
other effects on model of omitting
properties with low or negative cash
flow.
Properties with very
high rents
AFS,
REMS,
FMRs
Effect of property characteristics on
operating costs of luxury housing
may differ from effects of same
characteristic for modest, standard
housing.  Could confound effect of
property size, unit mix, etc.
Interact with housing assistance, since
this phenomenon only affects
unassisted housing.  Test effects on
model of omitting high rent properties.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
14
1.4 Testing the Accuracy of the Model
A.
Tests of How Well the Model Fits the Database on Which it is Based
GSD will test the accuracy of each version of the model under serious consideration, using
standard statistical measures of goodness of fit.  These include the following:
• R-squared.  This is a measure of the extent to which the model’s independent
variables and their coefficients explain all of the variation in the dependent variable
observed across the database—i.e., all the different values of operating costs per unit
month for all of the properties in the database.  The R-squared statistic can have
values between 0 (no variation is explained) and 1 (all of the variation is explained).
Following standard practice, GSD will adjust the R-squared for the number of
variables in the regression.  For reasons that can be demonstrated mathematically,
adding variables increases the R-squared statistic whether or not the additional
variables are significant.  The adjusted R-squared statistic corrects for this.
For reasons explained earlier—variations in operating costs that reflect management
style, inclusion of non-recurring costs in operating costs, and market positioning—
GSD does not expect the model to explain a very high percentage of the variation in
operating cost.  The objective of the modeling exercise is not to explain all the
variation in the operating cost of multifamily rental housing, but to produce
reasonable benchmarks for the cost of operating multifamily rental housing.
Nonetheless, a model with a higher R-squared should be selected over a model with a
lower R-squared, unless there are other considerations that make the model with the
lower R-squared superior.  For example, the R-squared statistic can shift substantially
with the exclusion or inclusion of a few influential observations (i.e., properties) in
the database (that is, observations with extreme values in the dependent or
independent variables).  Because of this “outlier effect,” R-squared statistics may not
capture the overall goodness of fit of a model.  For the same reason, as noted
elsewhere in this Draft Research Design, GSD will consider whether to eliminate
properties with extreme values from the database.
• Mean Absolute Percentage Error (MAPE).  This is a measure of how much the
individual values of the dependent variable (i.e., operating costs of individual
properties in the database) differ, on average, from the values predicted by the model.
For example, if MAPE is .12, the predicted operating cost for each property in the
database differs from its actual operating cost by 12 percent on average.  This statistic
may be more relevant than the R-squared statistic for the purpose of this modeling
exercise, which is to predict values for the dependent variable rather than to explain
variation.
Because there is a great deal of idiosyncratic variation in operating costs, GSD
expects that the MAPE for the model will be fairly large.
• Root Mean Squared Percentage Error (RMSPE).  The RMSPE measures the
deviation of predicted values from actual values in a way that accentuates the
importance of large individual errors more than MAPE.  This statistic may be a
useful diagnostic tool for identifying changes to the model needed to predict better
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
15
the operating costs of certain types of properties.  GSD will also look at individual
“worst case” errors to identify types of properties for which the model fits less well.
• Distribution of Percent Error (normalized residuals).  This statistic displays, for
all of the observations in the database, the difference between the predicted and
actual values of the dependent variable.  This is a very useful statistic for explaining
the model and for making decisions about how to apply it.  For example, it can be
summarized as:  “75 percent of the model’s predicted operating costs fall within –X
percent and +X percent of the actual value.”  The percentages of predicted values
falling above and below the actual value can be displayed separately.
• Percent Error for Subsets of Properties (prediction of cell means).  GSD will
create subsets of properties from the database that represent logical groupings of
properties—for example, high rise assisted elderly properties in suburbs by region of
the country or non-elderly assisted garden properties in central cities by region of the
country—and use statistical measures of goodness of fit to examine how well the
model predicts operating costs for that group of properties.  The analysis of how well
the model predicts operating costs for subsets of properties may be very useful for
making decisions on how to apply the model.
Such cell means could also be used directly for benchmarking costs; however, a
regression-based model is superior, because it enables the cost benchmarks to control
for a larger set of property characteristics and it permits the use of information from
observations that do not fit well into a predetermined cell.  Nonetheless, it will be
useful to test how accurately the model predicts costs for groups of properties within
the FHA database that are particularly similar to public housing in location, building
type, and so forth.
B.
External Tests of the Model
• Testing the model on a “hold-out” sample of FHA properties.  Because GSD has
a very large database of FHA insured and Section 202 multifamily rental properties
on which to base the model, GSD will be able to subject the model to a very powerful
test of accuracy.  GSD will select randomly 25 percent of the observations in the
database and hold that sample aside.  GSD will base the model on the remaining 75
percent of the observations and then test how well it predicts the reserved sample.
(GSD will select the sample randomly, but with some stratification to make sure that
certain property characteristics, such as age and size, are well represented in both
samples.)  GSD will then look at various statistical tests to compare how the model
performs for the hold out sample with the way in which it performs for the modeling
sample.
This is a particularly useful test for this modeling exercise because it mimics the
actual application of the model to a set of properties other than the set of properties
on which the model was based.
• Testing the model on other databases of multifamily rental properties.  GSD will
apply the model to other databases assembled as part of the reconnaissance phase of
the Cost Study: data provided by the National Apartment Association (NAA); a
database of properties owned by large private housing investors; and a database
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
16
provided by the United States Department of Agriculture’s Rural Housing Service on
multifamily rental properties assisted by the Section 515 program.  It may be
necessary to apply a reduced form of the model, since some of the variables used in
the model may not be available for those databases.  For example, none of the other
databases has information on the demographic characteristics of residents and none
has identifiers of property location as specific as the FHA database.  It will also be
necessary to adjust the model in other ways.  For example, the NAA database
contains very few assisted properties, so GSD will test the model against that
database using the coefficient for unassisted housing.
• Testing the model on individual private multifamily rental properties.  As part of
the model-building work, GSD will take advantage of the fact that the proprietary
databases from large housing investors include a number of properties that are also in
the FHA database and that GSD may be able to obtain additional information about
the “overlap” properties.  From among the properties that overlap, GSD will identify
two groups: those for which the model is predicting with a large error and those that
have a large effect on the model’s coefficients.  (GSD will use influence statistics to
make the latter determination.)  Members of the GSD team will then seek
information from the owners and managers of these “outlier” and “influential”
properties on the reasons for operating costs that (according to the model) are
extremely high or low.  This will help GSD decide on changes to the model—for
example, on changing the specification of variables, on interacting variables, and on
whether a subset of properties should be excluded from the model or have certain
characteristics controlled for by dummy variables.
• Testing the model on individual public housing properties.  While the model
development is proceeding, GSD will select not less than 50 public housing
properties, or 8 to 10 in each of 5 or 6 metropolitan areas (and nearby rural areas),
and will ask local property management experts who are experienced in managing
assisted housing in those markets to estimate the operating costs of those properties
as if they were privately owned assisted housing—that is, as if they had all the
regulatory requirements and constraints of project-based Section 8 properties with
budget-based rents.
GSD will select properties from a list of all public housing developments in each
metropolitan area (and nearby PHAs with rural service areas), with the following
selection rules:
− some properties, but not all properties, should be owned by the largest PHA in
the metropolitan area;
− 
the properties should represent a variety of building types and occupancy types
typical of both that metropolitan area and of public housing nationally;
− properties that have received HOPE VI grants, or for which the PHA intends to
apply for a HOPE VI grant, or that are subject to the viability analysis now
required for certain public housing developments should not be included;15 and
− properties determined (after consultation with the PHA) to have very atypical
operating costs for any other reason should not be included.
                                                  
15 Section 202 of the 1996 HUD Appropriations Act (subsequently amended by QHWRA) requires the viability
assessments and conversion to vouchers of large, substantially vacant public housing properties that have very high
rehabilitation costs.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
17
Utilizing a standardized protocol, these property-based budgets will be prepared
assuming that the properties are to be operated as well-run assisted housing (it will be
assumed that residents will be similar to the current residents or to those who would
be eligible to live and choose to live in a Section 8 project-based development in this
location and of this quality.  The real estate professional would be free to assume the
residents could be screened as permitted by program rules and evicted for cause as
permitted by program rules).  The model-predicted costs for these properties will not
be revealed to those preparing the budgets until after the field testing has been
completed.  These budgets will be supported with accompanying line-item narratives
and recommended staffing plans.  In the areas of security and resident programs, the
budgets will be prepared assuming what is normal and customary for well-run
assisted housing in those markets.  All of these operating budgets will be created with
the Chart of Accounts used for FHA that has been included in the GSD model of
FHA operating costs.  More specifically, these are the line items on form HUD-
92410:
− Total administrative expenses (Line 6200/6300)
− Total operating and maintenance (Line 6500)
− Total taxes and insurance (Line 6700) minus real estate taxes (Line 6710)
Additionally, because the need for deferred capital improvements may affect
operating costs, where warranted, these budgets will be developed under two
scenarios: (1) assuming current conditions and (2) assuming capital needs were to be
addressed.
In addition to testing the predictive abilities of the costing model, another purpose of
the field testing is to determine whether there are fundamental differences in the
physical design of public housing that affect operating costs, but are not captured in
the costing model.
When the model development process has produced a recommended model, GSD
will use it to estimate the benchmark costs for the same 50 properties.  GSD will then
convene a focus group including both the real estate professionals and the key staff
who worked on the model to discuss the benchmark estimates produced by the model
and the ranges estimated by the real estate professionals.  GSD will share these
results and observations with industry groups (either through the recommended
research working group or through the regular monthly public progress meetings).
This test of the model against specific public housing properties will produce
information for further refinement of the model and for decisions about its
application.  Any adjustments to be made to the model as a result of this field testing
process will be fully documented.
1.5 Applying the Model to Public Housing
When a recommended model (or some alternative models) has been selected, GSD will use the
model’s coefficients to estimate benchmark operating costs for individual public housing
developments (further adjustments will be made to those benchmarks outside the model, as
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
18
discussed below).  In order to apply the model to public housing at a development level, GSD has
obtained from HUD the database on public housing developments that is used to allocate capital
funds to PHAs.  With assistance from HUD’s Office of Policy Development and Research, GSD
has added to that database some additional information on the locations (street addresses and
geocodes) of public housing developments.  GSD will also match the database to information in A
Picture of Subsidized Households and the US Census.  If GSD finds that an indicator of physical
condition based on REAC inspection results enhances the predictive power of the model, GSD
will obtain parallel information from REAC for public housing developments and include these
data in the public housing development database.
For most of the variables in the model, GSD expects that the public housing development
database will have a precisely comparable variable—for example, property size, building type,
unit mix, MSA, and demographic and census variables—and GSD will apply the coefficients
from the model directly to those variables.  In the application of the model, it will be assumed that
public housing is assisted housing:  that is, the coefficients for assisted housing (and for any
variables that combine another variable with the fact that a property is assisted housing) will be
used.
In a few cases, there may be variables that GSD has determined should be in the model but that
are not available currently for public housing developments or for all public housing
developments.  The best example is the date of the most recent major rehabilitation of the
property, comparable to mortgage age for FHA properties.  If the use of this variable substantially
enhances the predictive power of the model, GSD will recommend that HUD collect data from
PHAs on the timing and cost of expenditures of capital or modernization funds at each property.
In the meantime, GSD will apply the model to public housing with that variable neutralized and
with information about the maximum percentage change in benchmarked per unit operating costs
that could occur once that variable is applied to public housing.
The following public housing developments will be excluded from the application of the model:
• developments that have received a HOPE VI grant; and
• developments that have failed the viability assessment required of certain public
housing developments or that are in the process of such an assessment.16
GSD may also exclude scattered site, single family developments, but that decision will not be
made until the end of the modeling exercise.
For each PHA, GSD will develop a composite operating cost benchmark based on the operating
cost (per unit month) of each development weighted by the number of units in each development.
This composite benchmark will be prepared both excluding the property types listed above and
including them.  The reason for including them is to compare better the model-predicted costs to
current Allowable Expense Levels (AELs) under the PFS/Operating Fund.
                                                  
16 Section 202 of the 1996 HUD Appropriations Act (subsequently amended by QHWRA) requires the viability
assessments and conversion to vouchers of large, substantially vacant public housing properties that have very high
rehabilitation costs.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter One – Developing Benchmark Operating Costs for Assisted Multifamily Rental Housing
19
1.6 Distressed Properties
The cost model will attempt to account for the effect of age and deferred capital needs on
operating costs.  It may also exclude the lower costs associated with operating distressed FHA
properties, if the modeling exercise determines that such a relationship exists.  However, public
housing may include some properties that are so severely distressed that the model cannot predict
their costs.17  Ultimately, severely distressed public housing properties should be targeted for
work-out action or should be demolished or removed from the inventory.  Once a property’s
distressed condition is treated, there is every expectation that the costing model should apply.
The issue, then, is the cost to operate a distressed property before it is treated, which should be a
temporary condition.
There are two basic approaches to determining the allowable operating costs for distressed
properties.  The first is to authorize current funding levels until the property’s condition is treated.
This presumes that the current funding to the property is higher than the model-predicted costs.
The second is to negotiate these budgets on a project-by-project basis based on informed
knowledge of property conditions.  In the event of the latter, and to assist in better understanding
what the operating costs might be for distressed properties, GSD plans to undertake an analysis of
the property-specific budgets for up to five distressed properties.  As indicated, these data will not
be incorporated into the costing model but will be used to inform policy makers of the options for
dealing with distressed properties.
While GSD acknowledges that distressed properties will need to be addressed separately from the
costing model, GSD has no ready way to identify which public housing properties are distressed.
From previous HUD research, the number of distressed units, variously defined, may be on the
order of 100,000.  To qualify as distressed, and therefore be subject to the aforementioned
negotiated budget process, HUD could either designate such properties through its local field
offices or it could leave such designation to the PHAs themselves.  In the case of the latter, and as
a condition of declaring a property distressed, and therefore being subject to the negotiated
budget process, a PHA might be required to commit to a set of remedial actions to correct the
underlying conditions.
                                                  
17 The FHA database does not include properties that are so severely distressed financially or physically that HUD has
assumed ownership of the property itself, not just the mortgages, i.e., HUD-owned properties.
CHAPTER TWO
Case Studies
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter Two – Case Studies
20
Chapter Two
Case Studies
The costing model that was described in Chapter One will estimate the cost to operate each public
housing property as if it were assisted housing, taking into consideration such property
characteristics as building design, location, number of three bedroom units, etc.  GSD will
conduct case studies of selected well-run PHAs to determine any adjustment in the model-
predicted cost as a result of differences in public housing’s regulatory or operating environment.
2.1 Approach
GSD’s earlier research suggests that the regulations governing public housing are similar (but not
identical) to those governing FHA assisted housing, and that the cost of those differences can be
reasonably estimated based on qualified professional judgement.18  Consequently, one of the
primary purposes of the case studies is to estimate these costs and determine whether they differ
for agencies of different sizes.  Exhibit 2.1 lists the major areas of differences in regulations
between public and assisted housing (a more detailed review of the comparative regulatory
environments is found in Appendix D).
Exhibit 2.1:  Regulatory Differences Between Public and Assisted Housing
Additional Regulatory Requirements
Imposed on Public Housing
Additional Regulatory Requirements
Imposed on Assisted Housing
• Annual Plan
• Community Service
• Cooperation with Welfare Agencies
• Deconcentration
• Designated Housing
• Grievances
• 
Income Disregards, Imputed Welfare
Income, and Flat Rents
• Pets in Family Housing
• Procurement
• PHAS
• Section 3
• 
Tenant Participation and Resident
Consultation
• Wage Rates
• Waiting Lists
• 
Financial Reporting
• Subsidy Transference
• Petition for Project Rent Increases
• Audit Requirements
The case studies will be conducted by teams of housing professionals that will include at least one
member who is an expert in operating assisted housing and one member who is an expert in
                                                  
18 The term “regulations” is used here to refer to the body of rules and regulations governing public and assisted
housing.  This would include not just items formally published in the Code of Federal Regulations but also HUD
handbooks, notices, and other instructions.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter Two – Case Studies
21
operating public housing.  Based on their observations, and in accordance with professional
judgement, the teams will attempt to answer the question, “What impact would these additional
regulatory requirements have on operating costs if imposed on an efficient operator of assisted
housing and assuming a competitive operating environment?”
As indicated, the point of reference is the cost to operate assisted housing.  The case study teams
will be attempting to determine the net burden of public housing’s unique regulatory
requirements, i.e., the added cost of regulations imposed or enforced only on PHAs less the cost
of regulations imposed or enforced only on operators of assisted housing.
Consideration will be given to both the direct and indirect costs of public housing’s unique
regulatory requirements.  For example, are there unique regulatory requirements that lead PHAs
to incur more extensive (and hence more costly) automation systems than typically found in
assisted housing?  Similarly, do these regulations give shape to organizational structures or
tendencies that are different than assisted housing and that have cost implications?
GSD is also interested in understanding the cost impact based on agency size.  For example,
federal procurement rules require formal competitive bidding for purchases of more than
$100,000.  A smaller agency would generally not have a single operating expenditure in any year
that exceeds that amount.  On the other hand, there may be certain regulations that require large
fixed costs and, therefore, are disproportionately burdensome on smaller agencies, who have less
ability to spread those fixed costs over a large number of units.
The other major area of research for the case studies is identifying differences in costs that are
driven not by federal regulations but by a PHA’s local operating environment.  GSD’s previous
research also suggests that these local factors may drive costs much more significantly than
differences in regulations.  Exclusive of differences in property characteristics or regulations, a
PHA may have costs that differ from assisted housing for reasons having to do with the
following:19
• Mission.  The cost to operate public housing could be higher/lower than the cost to
operate assisted housing depending on the mission of the PHA.  A PHA may decide
to provide a quality of housing that is substantially higher/lower than assisted
housing or it may decide to serve a tenant population that is substantially more
service-needy (for example, in the case of serving homeless families).  Additionally,
a housing authority may decide to provide substantially more property or resident
services to the residents of public housing than might be traditionally provided in
assisted housing.  Finally, certain PHAs may be expected to play roles in their
communities that extend beyond the actual operation of public housing and may
include housing planning or advocacy.
• Organization.  Public housing costs may differ from those of assisted housing
because of the way in which a PHA is organized.  For example, as GSD has
previously observed, PHAs tend to operate under more centralized arrangements than
private operators of assisted housing.  These organizational tendencies may have
fairly significant cost implications.
                                                  
19 A PHA may also incur costs for activities related to the development of new housing.  A discussion of these costs is
found under Asset Management, below.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter Two – Case Studies
22
• Local Law.  To the extent that local law (state or municipal) imposes requirements
on all operators of assisted housing, those costs should be reflected in the costing
model.  For the purposes of the case studies, the issue is whether local laws impose
unique constraints on the operation of public housing.  For example, local law may
require a PHA (but not a private operator of assisted housing) to provide a minimum
level of security coverage in senior buildings or to provide additional due process in
the case of evictions.
• Public Entity Costs.  Costs may be higher than assisted housing as result of a PHA’s
status as a public body.  This can include greater demands for public access and
participation in decision making.  As public bodies, PHAs may also have less flexible
work rules or pay higher wages and benefits, i.e., non-market-standard work rules not
otherwise required by local law.  Finally, there may be greater public expectations
and scrutiny of activities, leading PHAs to implement regulations more elaborately
than operators of assisted housing or to establish processes that place more emphasis
on equity and fairness over efficiency.
The above items raise a number of challenging policy issues.  Beyond the benchmark cost to
operate assisted housing, and beyond any differences in regulations, which of these items should
be funded with federal operating subsidies?  Should federal operating subsidies be used to pay for
locally-imposed mandates?  For supportive services and programs that exceed the level provided
in assisted housing?  For differences in organization (unless caused by regulation)?  Etc.
Before GSD completes the protocols for the case studies, it will conduct pilot case studies at not
less than six agencies.  These pilots will be conducted by two senior, experienced research teams,
each of which would visit at least one small, medium, and large agency.  The pilot case studies
will allow GSD to understand better the nature of the cost drivers, the information to collect, the
research questions to ask, etc.  Instructions to the research teams for these pilot sites are included
in Appendix C. These case study pilot sites will also help GSD to determine whether any such
adjustments should be expressed in absolute terms (dollars of additional funds) or in percentage
terms (X percent above the benchmark costs).
Importantly, following completion of these case study pilot sites, GSD will share its observations
with industry groups and accompanying recommendations for protocol development.
In addition to assisting in finalizing the case study protocols, the other advantage of conducting
these pilot case studies is that they will allow policy makers an opportunity to provide direction as
to which of the local cost drivers should be funded with federal operating subsidies and, hence,
should be included in the final case study analysis.  Following the case studies, GSD intends to
hold conversations with stakeholders (HUD, the Congress, industry groups) regarding which of
these functions should be funded with operating subsidies and the implications for the case study
process.
2.2 Number and Selection of Case Study Agencies
Including the pilot sites, GSD intends to conduct case studies of between 20-30 PHAs, as shown
in Exhibit 2.2.
Public Housing Operating Cost Study                                                                          Draft Research Design
Chapter Two – Case Studies
23
Exhibit 2.2:  Number of Proposed Case Studies, Based on Size of PHA
Small
PHAs
Medium
PHAs
Large
PHAs
Very Large
PHAs
Totals
Minimum
8
5
4
3
20
Maximum
10
8
7
520
30
The selection of case study agencies will be determined based on the following factors:
• Range of agency size.  Given that some of the potential cost drivers may have
differential impacts based on agency size, the case study agencies will represent a
range of agency size (small, medium, large, and very large).
• Geographic distribution.  Case study agencies should be geographically distributed,
representing different regions of the country but also representing city/suburban/rural
PHAs.
• Well-run status.  Agencies selected must score at least an 80 on the Public Housing
Assessment System (PHAS) but also be widely recognized within the industry as
well-run.  In the case of very large agencies, a lower threshold may be required in
that currently only five of 14 very large agencies have PHAS scores of more than 80.
• Willingness to share data and participate in the Cost Study.  GSD will not
undertake a case study at an agency that does not wish to participate.  Data collection
would be extremely difficult at such an agency.
• Sound and consistent financial management systems.  The presence of such
systems is essential to analyzing costs.
• Experience with assisted housing or private management or public housing.
Some, but not all, of the case study agencies will include agencies that have
experience in owning or managing assisted housing in addition to their public
housing.  The comparative operating experience of these agencies should prove
helpful in identifying and understanding differential cost impacts.  For example, it
will allow GSD to see how the different procurement rules play out in the two
programs.  For the same reason, GSD plans to include some agencies that have
experience with the private management of public housing.  This latter sub-set of
agencies may help GSD to distinguish what might be regulatory cost effects from
what might be organizational cost effects.
2.3 Supplemental Research
GSD intends to supplement the case studies with additional research on asset management,
mixed-finance public housing, and private management of public housing, as described below:
                                                  
20 In accordance with the Cooperative Agreement with HUD, at least one of the very large agencies to be included in
the case studies will be the Puerto Rico Housing Authority; however, their scope of work is likely to be expanded
beyond the areas studied in other case study sites.