.. .. ._
...
u»..:..

 

.v;. <2.

.

.
..........

 

 

 

..
. . . ..

.‘....,‘.l.c..u .1 .

.A:.u..x, 4:...

It, .13:,.x....7
. q I .7

   

. . “Ar i... ....;.v:.s.$.. flauntuwar' .r?

 
 

2‘91. 1
.. . . 3.3219,
15,. u...~.........c nu . ,

   

 

 

 

THERE}

“WWI WWW WI 1,

3 1293 00794 9781

This is to certify that the

dissertation entitled
A METHOD FOR ESTIMATING THE LOCAL AREA ECONOMIC
DAMAGES OF SUPERFUND WASTE SITES '

presented by
DAVID RAY WALKER

has been accepted towards fulfillment
of the requirements for

Ph.D. Agricultural Economics

 

 

degreein

 

 

Date 3/22/7 L

MS U is an Afl'mnative Aca’on/Equal Opportunity Institution 042771

 

H-‘T ——\
um"
Michigan Inc.
University

L A

‘w__'

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.

ll DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ll

MSU Is An Affirmative Action/Equal Opportunity Institution
chma-pd

A METHOD FOR ESTIMATING THE LOCAL AREA ECONOMIC
DAMAGES OF SUPERFUND WASTE SITES

By

DAVID RAY WALKER

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Agricultural Economics

1992

ABSTRACT

A METHOD FOR ESTIMATING THE LOCAL AREA ECONOMIC DAMAGES
OF SUPERFUND WASTE SITES

By

DAVID R. WALKER

National Priority List (NPL) sites, or more commonly called Superfund sites, are
hazardous waste sites (HWS) deemed by the Environmental Protection Agency (EPA) to
impose the greatest risks to human health or welfare or to the environment. HWS are
placed and ranked for cleanup on the NFL based on a score derived from the Hazard
Ranking System (HRS), which is a scientific assessment of the health and environmental
risks posed by HWS.

A concern of the HRS is that the rank of sites is not based on benefit-cost analysis.
Because of this concern, the main objective of this dissertation is to develop a method
for estimating the local area economic damages associated with Superfund waste sites.
Secondarily, the model is used to derive county-level damage estimates for use in ranking
the county level damages from Superfund sites.

The conceptual model used to describe the damages associated with Superfund sites
is a household-firm location decision model. In this model it is assumed that households
and firms make their location choice based on the local level of wages, rents and
amenities.

The model was empirically implemented using 1980 Census microdata on
households and workers in 253 counties across the United States. The household sample

includes data on the value and structural characteristics of homes. The worker sample

David R. Walker

includes the annual earnings of workers and a vector worker attributes. The microdata
was combined with county level amenity data, including the number of Superfund sites.

The hedonic pricing technique was used to estimate the effect of Superfund sites on
average annual wages per household and on monthly expenditures on housing. The two
equations were specified in log-linear form and estimated using the seemingly unrelated
regressions model.

The results show that Superfund sites impose statistically significant damages on
households. The annual county damages from Superfund sites for a sample of 151
counties was over 14 billion dollars. In addition, the ranking of counties using the

damage estimates is correlated with the rank of counties using the HRS.

I dedicate this dissertation to my wife Hirae whose love and
inspiration was valuable beyond measure.

iv

ACKNOWLEDGMENTS

I want to thank Dr. John Hoehn for his guidance over the last several years. I
benefitted greatly from his urging to improve my skills in not only in economics but also
in writing, publishing and presenting papers. He expected a lot at times, but it is
apparent now how, important it is for an advisor to help his students reach their full
potential.

I also want to thank Dr. Luanne Lohr for her many helpful suggestions in writing
this dissertation. Her comments made for a more thorough and carefully written
dissertation.

I owe my gratitude to Dr. Lynn Harvey. Lynn was very supportive of me in so
many ways throughout my stay at Michigan State. I can never thank him enough for the
help he provided to me. I will always be proud to call Lynn my friend and colleague.

Finally, and most importantly, I want to thank my wife Hirae. She raised three
little boys, Stephen, Philip and Aaron, and still found time to complete a Ph.D. She
sacrificed a lot of things including her career to help me complete this dissertation. For
that I can never thank her enough. I praise God every day for making Hirae part of my
life.

TABLE OF CONTENTS

Page
List of Tables ...................................................... viii
List of Figures ...................................................... ix
CHAPTER 1. AN INTRODUCTION TO SUPERFUND WASTE SITES ........... 1
1. Introduction ................................................. 1
2. Recent Legislation and Regulation of Superfund Sites ................. 2
3. Economic Issues Regarding Existing Hazards ....................... 8
4. Research Objectives: Developing a Method for Estimating Economic
Damages ................................................... 9
CHAPTER 2. LITERATURE REVIEW OF STUDIES ESTIMATING THE
ECONOMIC DAMAGES OF SUPERFUND SITES USING THE
HEDONIC PRICE METHOD ............................... 12
1. Introduction ................................................ 12
2. The Hedonic Price Method .................................... 13
3. A Review of Superfund Waste Studies ............................ l4
4. Implications of These Studies .................................. 42
CHAPTER 3. A CONCEPTUAL MODEL FOR ESTIMATING THE LOCAL AREA
ECONOMIC DAMAGES OF SUPERFUND WASTE SITES ........ 46
1. Introduction ................................................ 46
2. Interregional Wage-Rent Model ................................ 46
2.1TheHousehold .......................... 48
2.2 The Firm ............................................ 49
2.3 Equihbfium .......................................... 50
3. Incremental and Aggregate Damages of Superfund Sites .............. 54

CHAPTER 4. ANALYTICAL METHODS FOR ESTIMATING THE LOCAL
AREA ECONOMIC DAMAGES OF SUPERFUND WASTE SITES . . . 56

1. Introduction ................................................ 56
2. Wage and Rent Equations ..................................... 56
2.1 Econometric Specification of the Model .................... 56

2.1.1 Correction for Hetereoskedasticity ................. 57

3. Data to Implement the Empirical Model .......................... 61
3.1 Rent Data ........................................... 62

vi

 

3.2 Wage Data .......................................... 62

3.3 Amenity Data ........................................ 65
3.3.1 Climatic Data ................................. 65
3.3.2 Social Data ................................... 65
3.3.3 Environmental Data ............................ 65
3.3.4 Superfund Data ................................ 66
CHAPTER 5. RESULTS AND DISCUSSION OF THE LOCAL AREA
ECONOMIC DAMAGES OF SUPERFUND WASTE SITES ........ 67
1. Estimation of the Local Area Economic Damages of Superfund Sites . . . . 67
2. Incremental and Aggregate Damages of Superfund Waste Sites ........ 76
3. A Restricted Model for Estimating Damages ....................... 78
CHAPTER 6. RANKING SUPERFUND SITES USING THE LOCAL AREA
ECONOMIC DAMAGE ESTIMATES ......................... 86
1. Introduction ................................................ 86

2. Household and County Level Economic Damages of Superfund Sites . . . . 86

3. Comparing the Ranking of Counties Using Aggregate County Damages
and the Hazard Ranking System ................................ 93
3.1 Implementation and Results of the Rank Comparisons ......... 94

CHAPTER ‘7. CONCLUSIONS, IMPLICATIONS AND FUTURE RESEARCH . . . . 96

1. Introduction ................................................ 96
2. Implications ................................................ 96
3. Future Research ............................................ 98
ENDNOTES ....................................................... 101
BIBLIOGRAPHY ................................................... 103
APPENDIX A: DERIVATION OF WAGE AND RENT GRADIENTS ........... 107

APPENDIX B: ANNUAL INCREMENTAL COUNTY DAMAGES ............. 109

LIST OF TABLES

Page

Table 2.1: Hedonic Models used for Estimating the Damages of

Superfund Waste Sites ....................................... 15
Table 2.2: Characteristics of Toxic Waste Sites in Kohlhase (1991) .............. 36
Table 4.1: Description and Mean of Variables in Wage and Rent

Data Set .................................................. 63
Table 5.1: Parameter Estimates and Standard Errors for OLS and SUR ......... 69
Table 5.2: Wage Parameter Estimates and Standard Errors for

Unweighted SUR and Weighted OLS and SUR .................... 74
Table 5.3: Aggregate and Incremental Damage Estimates .................... A. 77
Table 5.4: Parameter Estimates and Standard Errors for Restricted SUR ......... 80
Table 6.1: Annual Economic Damages of Superfund Sites .................... 87
APPENDIX TABLE:
Table 8.1: Annual Economic Damages of the Incremental Site ................ 109

 

LIST OF FIGURES

Page
Figure 2.1: Wage and Rent Equilibrium .................................. 53
Figure 5.1: Incremental and Aggregate Damages of Superfund Sites ............. 85

 

CHAPTER ONE: AN INTRODUCTION TO SUPERFUND WASTE SITES

1. INTRODUCTION

Hazardous waste is generated by industry, municipalities, mining operations and
by hospitals and laboratories among other sources (Andelman and Underhill, 1987). The
quantity of hazardous waste material generated annually by these groups is substantial.
For example, in 1981, industry in the United States generated 264 million metric tons of
hazardous waste (71.3 billion gallons) (CEP, 1986). In the past, these wastes would have
been disposed of in open dumps Or underground containers (Grisham, 1986). Presently
over 99 percent is eventually placed in the ground in deep wells, surface impoundments,
and lined landfills (CEP, 1986). In 1988, 2.3 billion pounds of toxic chemicals from
major manufacturing facilities were transferred or released to air, water, or land (EPA,
1990). As of December 1990, 32,506 potentially hazardous waste sites were identified
across the United States (EPA, 1990).

Among hazardous waste sites, National Priority List sites (NPL) or more
commonly called Superfund sites, are considered by the United States Environmental
Protection Agency (EPA) to be the most hazardous sites. These sites are considered to
pose the most threat to human health, natural resources and the environment.

Superfund sites are eligible to receive Federally mandated monies for cleaning up the

2

site and repairing, restoring or acquiring equivalent natural resources. There are over
1,200 sites listed on the NPL as of January 1991.

The main objective of this study is to develop a method for estimating the local
area economic damages caused by Superfund waste sites. Damage estimates can be used
to prioritize the cleanup of Superfund sites and to provide measures of the interim
damages caused by Superfund sites.

The remainder of this chapter describes recent legislation and regulation of
Superfund waste sites, economic issues regarding Superfund waste sites, and finally the

research objectives of this study.

2. RECENT LEGISLATION AND REGULATION OF SUPERFUND SITES

Congress _ in 1980 passed the Comprehensive Environmental Response,
Compensation, and Liability Act (CERCLA) to deal, inter alia, with the hazardous waste
problem. CERCLA enabled the federal government to respond to actual or threatened
releases of dangerous substances at sites and facilities by undertaking cleanup actions, to
administratively or judicially abate releases posing an imminent or substantial danger to
public health or welfare or to the environment, and to recover damages for the
destruction of or damage to natural resources. CERCLA also allows Trustees to conduct
a cleanup or remedial action at a site and then recover the costs from responsible parties
or from the Hazardous Substances Superfund or more commonly called Superfund
(Wolf, 1988).

CERCLA was amended by the Superfund Amendments and Reauthorization Act of

1986 (SARA). SARA expanded and toughened the cleanup authority of the federal

3
government and provided an increase in funds for the Superfund (Wolf, 1988). SARA

set new standards for cleaning up contaminated and polluted sites and mandated the
federal government to begin work at 375 sites within five years (Wolf, 1988). In
addition, SARA stressed the use of permanent cleanup methods such as detoxifying
hazardous wastes whenever possible, rather than burying wastes in landfills, or
transferring them from one site to another (Wolf, 1988).

SARA increased the power of the federal government (EPA) in a number of
ways. For example, the President may order a polluter to remove or control any
hazardous substance endangering public health, welfare, or natural resources. SARA
also retained and strengthened the authority of the federal government and members of
the public to enforce the act’s provisions and compel responsrble parties to pay the costs
of response actions and to reimburse the Superfund for initially financing these response
actions (Wolf, 1988).

CERCLA requires the EPA to maintain a National Priority List (NPL) of hazardous
waste sites with known or threatened releases. The NPL identifies abandoned or
uncontrolled hazardous waste sites that warrant further investigation to determine if they
pose a threat to human health or the environment (EPA, 1989).

The criteria for placing sites on the NPL is based on the hazard ranking system
(HRS). This system is used by the EPA and others to evaluate the relative risk to
human health and the environment posed by a site. The factors used in the HRS for
ranking sites include: relative hazard to public health or the environment, taking into
account the population at risk; hazardous potential of the substances at the site;
potential for contamination of drinking water supplies; direct contact or destruction to

sensitive ecosystems; damage to natural resources which may affect the human food

4

chain; ambient air pollution; and a state’s preparedness to assume the costs and
responsibilities of cleanups (Hall et al. 1987).

A total score is derived for each site using the HRS. A higher score implies
greater risk to human health and natural resources. Until recent regulation was passed,
a minimum score of 28.5 was required to place a waste site on the NPL’.

The purpose of Superfund is to finance government and private cleanup actions of
Superfund sites and to pay claims for damages to natural resources (Hall et al. 1987).
The claims for natural resource damage can only be claimed or recovered by a Trustee.
The Trustee may be a federal, state or local official, or representative of an Indian tribe
acting on behalf of the general interest of the public. Claims for natural resource
damages cannot be recovered by an individual. Up to 85 percent of Superfund is to be
devoted to the costs of removal and remedial actions while the remaining 15 percent can
be directed to natural resource damages (Hall et al. 1987).

Trustees are entitled to recover up to $50 million above response costs for
natural resource damages for each incident involving releases of hazardous substances.
However, Congress can prohibit the use of Superfund monies for recovering natural
resource damages if the EPA determines that all the money in the fund is needed for
response actions (Hall et al. 1987).

In addition, Superfund monies can only be used to recover natural resources that
are publicly held or for natural resources that are privately held but where the public has
a substantial statutory, common law, or regulatory interest (Department of the Interior,
1991).

To assess natural resource damages, CERCLA requires Trustees to follow a four

phase assessment procedure. The phases include preassessment, an assessment plan

5

(which should be consistent with a reasonable cost criterion), damage assessment, and
post-assessment. The damage assessment also has three phases: injury determination,
quantification, and damage determination (Department of the Interior, 1991).

In the preassessment phase the Trustee is required to determine if an emergency
exists with regards to potential injury of the natural resource”. If an emergency exists,
and no liable party has responded to the emergency, then the Trustee is authorized to
take limited action to abate the emergency situation.

Once an emergency action is completed, the Trustee performs a preliminary
screen of available data to determine whether the damage caused by a release justifies
the completion of a damage assessment. The Trustee can make this decision based on
information from the site or from information that is readily available from standard
research sources. If a damage assessment meets certain criteria the Trustee must
completely document the decision to continue the assessment (Hall et al, 1987).

The purpose of the assessment plan is to ensure that the assessment is performed
in a planned and systematic manner and at a reasonable cost (Hall et al, 1987). If the
type B assessment is chosen then a damage assessment requires a detailed three-step
approach’.

The first step is the injury determination phase which is to establish that an
injury has occurred to a natural resource and to link the injury to the release from a
waste site. For purposes of damage assessment, natural resources are divided into five
categories, including, surface water, ground water, air, geologic, or biological resources.
Once an injury determination is complete, the Trustee must decide which economic

methods will be chosen to estimate damages to the natural resource.

6

The second step in the damage assessment phase is to quantify the effects. That
is, the Trustee converts the natural resource injury to a dollar amount by measuring the
changes in the services provided by an injured resource as a result of the release.

The third step in the damage assessment phase requires the Trustee to estimate
the amount of money to be sought as compensation for the natural resource injury. The
measure of damages is the estimated cost for restoration, rehabilitation, replacement,
and/ or acquisition of equivalent resources, plus compensable value of the services that
will be lost to the public through the period of recovery to the baseline conditions
existing before the discharge or release. Compensable value encompasses all of the
public economic values associated with an injured resource, including use values and
nonuse values such as option, existence, and bequest values (Department of Interior,
1991).

Required _in the third step is a Restoration and Compensation Determination
Plan which should describe the restoration alternatives considered, the loss of services
associated with each, and the estimated period of recovery associated with each
alternative. Cost and valuation methodologies should also be described (Department of
Interior, 1991).

With respect to valuation methodologies, all standard methods are admissible.
Use values may be estimated using revealed preference methods: market price, travel
cost, and hedonic price methods. According to the Department of the Interior, the
contingent valuation method is the only nonmarket valuation methodology available that
is capable of explicitly estimating non-use values (Department of Interior, 1991).

However, Smith (1985) argues that the hedonic method can also be used to calculate

7

nonuse values, specifically option price. In addition, there is a rebuttable presumption
conferred upon natural resource assessments (Department of Interior, 1991).

Finally the post-assessment phase requires the Trustee to prepare a Report of
Assessment which includes documentation and support for decisions made during the
assessment (Hall et al, 1987).

Superfund provides no compensation for health effects (personal injury or death)
caused by hazardous waste but it can be used to cover the costs of related health studies
(Wolf, 1988). However, CERCLA has a provision that allows individuals who are
suffering from latent illnesses caused by exposure to hazardous substances to sue liable
parties without a time limit, since the damage caused by the exposure often occurs long
before the symptoms become apparent (Wolf, 1988).

Recently, however, a number of States have enacted or attempted to enact
Amendments or Acts that would provide funding for not only environmental damage but
also for personal injury (NCPA, 1986). In 1985 Minnesota enacted an Amendment to
the State Superfund Act that allows compensation for all damages for death, personal
injury or disease including medical expenses, rehabilitation costs, burial expenses, loss of
earning capacity, loss of past or future income, and damages for pain and suffering
(NCPA, 1986).

On the other hand, the Michigan legislature passed in 1982 the Michigan
Environmental Response Act (Public Act 307) which does not fund health related
damages. However, the Act allows for recovery of response activities, fines, and
exemplary damages, plus up to 50 million dollars in damages for injury to, destruction of,

or loss of natural resources resulting from the release or threat of release, including the

 

8

reasonable costs of assessing the injury, destruction, or loss resulting from the release or

threat of release.

3. ECONOMIC ISSUES REGARDING EXISTING HAZARDS

Hazardous wastes impose economic damages on individuals and households
(Kohlhase, 1991). Damages arise as the environment 'is contaminated by hazardous
wastes. A degraded environment may require cleanup of polluted water and soils and
repair to the natural environment. In addition, human health may be adversely affected
by hazardous wastes consumed in contaminated water supplies or by ingestion of food
contaminated by hazardous wastes.

Reducing or mitigating damages caused by hazardous waste benefits those individuals
who were affected by the wastes. However, an important question is whether the costs
of cleaning up a hazardous waste site and repairing the natural environment are
outweighed by the benefits. This leads to three economic questions related to Superfund
site assessment procedures discussed in section two.

First, in the assessment damage phase, the Trustee is required to quantify the
interim damages from Superfund sites. This implies that readily available methods be
available to estimate the interim damages. Thus tools for estimating the economic
damages caused by Superfund sites need to be developed.

Second, as mentioned earlier, in the preassessment phase a Trustee must
determine whether the damage caused by a release justifies the completion of a damage
assessment. This implies there is a need for estimating if economic damages from

Superfund waste sites are statistically significant, significant in size and if the estimates

9

are reliable. Readily available economic damage estimates of Superfund waste sites can
reduce the time and costs for assessing whether it is beneficial to complete a damage
assessment.

Third, Superfund sites are given a HRS score. Priority for cleanup is based
mainly on the HRS score. However, there is controversy over the relationship of the
HRS and the economic damages caused by Superfund sites (Hird, 1990). It could be that
a higher ranked Superfund site has smaller economic impacts than a lower ranked site or
the cost of cleaning up a lower ranked site is much smaller, resulting in greater net
benefits if the lower ranked site is cleaned up first. Thus to increase the benefits
generated from the use of Superfund monies, economies should be considered in ranking
sites for cleanup as well as appropriating money for restoring natural resources that were

damaged.

4. RESEARCH OBJECTIVES: DEVELOPING A METHOD FOR ESTIMATING

ECONOMIC DAMAGES

The first objective of this study is to develop a method for estimating the local area
economic damages of Superfund sites. The economic damage estimates obtained from
the method are one set of damages caused by Superfund sites - the local area economic
damages. Nonuse values as well as use values of noncounty residents are not estimated
using the method developed in this study. Estimates of local area economic damages can
be one set of damages used in the preassessment phase to estimate the interim damages

caused by a release of chemicals from a Superfund site.

10

The conceptual model used in this study for deve10ping the method is a
residential location model. The residential location model uses the change in individual
wages and land rents across space to estimate the damages of Superfund sites.

It is expected people will prefer to locate in high rather than low amenity counties.
In high amenity counties land rents are bid up and wages are bid down as households
move to these counties. Adam Smith (1776) noted over 200 years ago that many factors
affect the level of wages and rents paid. Wages are affected by the agreeableness or
disagreeableness of the job, the easiness and cheapness, or the difficulty and expense of
learning the job, the constancy or inconstancy of employment in the particular kind of
job, the amount of responsibility required of the job, and the probability of success in the
job. Rent is affected by things such as the fertility of the land. According to Adam
Smith, amenities also affect the level of both wages and rents. For two counties that are
identical except for the presence of a Superfund hazardous waste site, it is expected that
lower land rents and/ or higher wages would be necessary to induce households to remain
in the area, once the waste is discovered. This implies that the benefits of protecting
human health and natural resources can be estimated from this model by the
compensating changes in wages and rents due to different numbers of Superfund sites
across counties.

To estimate the changes in wages and rents across counties in response to the
presence of Superfund sites, the hedonic pricing technique is applied. The data to
empirically implement the residential location model is the 1980 US. Census Public Use
Microdata A Sample combined with county level amenity data.

A second objective is to use the economic damage estimates to rank the priority

of cleanup of Superfund sites. The ranking of Superfund sites using the damage

11
estimates is compared to the EPA’s ranking of Superfund sites using the HRS.

Prioritization for cleanup, repair and compensation for injury to natural resources is

based on the economic damages caused by Superfund sites.

CHAPTER TWO: LITERATURE REVIEW OF STUDIES ESTIMATING THE
ECONOMIC DAMAGES OF SUPERFUND SITES USING THE
HEDONIC PRICE METHOD

1. INTRODUCTION

The objective of this chapter is to review the hedonic price method as a
technique for estimating the economic damages of Superfund waste sites. The review is
divided into two sections.

Section one describes the hedonic price method and the welfare significance of
price measures derived from the hedonic technique. The literature provides evidence
that valid price measures of amenities can be derived from the hedonic price method.

Section two reviews research that has used the hedonic price method to estimate
the economic damages of Superfund waste sites. The data, methods, and results of these
studies are detailed.

By reviewing these studies characteristics common among the studies can be
identified and if appropriate incorporated into an empirical model. In addition, the
economic damages from Superfund sites estimated in the reviewed papers are compared

to the economic damages estimated in the present study.

12

13
2. THE HEDONIC PRICE METHOD

There is a large literature on the use of the hedonic price method to estimate
the value of amenities as well as other nonmarket goods‘. The hedonic pricing method is
simply a way to decompose the price of a good into the prices of the good’s attributes.
In this section, the analysis focuses on the decomposition of wages and rents into their
component prices.

The output of the hedonic price method is a price function, or a rent or wage
gradient which relates value to the quantity and / or quality of the amenity or disamenity
(Freeman, 1979a). The following example demonstrates how a marginal implicit price
for an amenity is derived. 7

Assume for example that the price of a home, r, is a linear function of n

characteristics (2,, i= 1,2,...,n). Regressing r on the n characteristics results in equation 1:
(1) r = a0 + alzl + + 3.1. + u,

where al through all are the estimated coefficients of the n characteristics and u is an
error term. The estimated coefficients are the marginal implicit prices of the house
characteristics’. For example, the marginal implicit price of 2,, Br/azl, is al. The
marginal price paid is the actual payment for the increased quantity of the characteristic.

Three assumptions are required to accurately estimate the value of an amenity
using hedonic analysis (Freeman, 1979a, 1979b and Bartik and Smith, 1987). First, the
amenity must be exclusive to the market transactions under examination. For example,

the value of the amenity must be captured only by changes in rents if the hedonic

14

property price model is used. Other markets, such as the labor market must not capture
any amenity value with changes in the amenity across space, otherwise the hedonic
property price model will provide biased results. The same is true if the labor market is
being used to capture the value of an amenity. The main point is that the value of an
amenity cannot be captured by looking at a single market (housing or labor) if more
than one price (wages and rents) is involved in compensation.

Second, the impact of the amenity must be observable by some portion of the
households in a locality. If all households are unaware of the existence of the amenity in
their community then price differentials will not occur in the housing or labor market.
In addition, some portion of households should be aware of the amenity in other
communities they would consider relocating to for price differentials to occur in the
labor and housing markets.

Third, some portion of the households must have the opportunity to choose
among various quality or quantity levels of the amenity. If no household can choose
among the amenity levels then price differentials are not expected to occur in the

markets.

3. A REVIEW OF SUPERFUND WASTE STUDIES

A number of studies have estimated the economic effects that Superfund waste
sites impose upon households using the hedonic price method. This section reviews
these studies and compares data, methods, and results (Table 2.1).

Adler et al. (1982) used the hedonic property price method to estimate the

economic damages caused by hazardous waste sites in two different communities. These

15

Table 2.1 - Hedonic Models used for Estimating the Damages of Superfund Waste Sites

 

 

Annual Benefit
Estimate of
removing a site
Objective and per household
Study Data Variables (1990 dollars)
Adler et al., 1982 Determine Dependent variable: natural In Pleasant Plains,
whether if the log of the selling price of the $1,992 if 1.75
Examined one hazardous social costs home. miles away to
waste site in Pleasant imposed by Independent variables: $7,269 if less than
Plains, NJ. The site hazardous wastes -structural diaractefisitcs half a mile. The
contaminated the upper are reflected in -neighborhood characteristics value of a
portion of two aquifers. property values. -date of sale residence 1.5 to
Chemicals at the site Sales of 675 -distance variables were used 1.75 miles from a
include aromatic homes in to proxy the effect of the site sold on
hydrocarbons, benzene Pleasant Plains hazardous waste sites. average 6 percent
and toluene among from 1968 to Distance variables expected more than a
others. They also looked 1981 and Sales to capture both health and residence within
at a hazardous waste site of homes in environmental risks. one-half mile. The
in Andover,MN. Barrels Andover from -Zone of contamination damage estimates
of solvents, inks, paints, 1978 to 1981. variables were used separately for Andover were
glues and geese were from the distance variables to not statistically
leaking. At Andover capture both health and significant from
contamination onlyy environmental effects of the zero.
occurred in wells located sites in Pleasant Plains only.
on the property’where
the site is located.
Harrison and Stodr, Estimate the Dependent variable: natural The damage
1984. benefits of log of the selling price of the estimates were
removing the home. based on
The authors included in hazardous Independent variables: coefficient
their analysis eleven sites chemicals from a employment access estimates that
located in the Boston, site. Sales of -structural were not
MA area. Sites contained 2,182 individual -the inverse square of the statistically
toxic organic compounds homes in the distance of a home to each significantly
or their equivalents. The Boston Metro site and the inverse square of different from
size of the sites ranged area for years the distance of a home to zero. $15 to $70
from one acre up to 400 1977-1981. each site weighted by the depending on the
acres. area of the site proxies health site. For a
effects. $100,000
environmental effects were residence
proxied by the the number of willingness to pay
nonhazardous and hazardous for cleanup of a
sites at one-half mile site 1.5 miles
increments around a home. away is 31,61!)
-neighborhood versus $13,500 if
one-half mile

away.

16

Table 2.1: Hedonic Models used for Estimating the Damages of Superfund Waste Sites, Continued

 

 

Annual Benefit
Estrm' ate of
removing a site
Objective and per household
Study Data Variables (1990 dollars)
Mendelsohn, 1987. Estimate the Dependent variable: sale Not applicable
economic price of the home since author
Waste site is the Boston damages of PCB Independent variables: found no
harbor where PCB’s pollution in New -structural significantly
were found in the Bedford harbor, -neighborhood different effect on
sediment. Portions of the Massachusetts. -access housing price
harbor were closed to Sale of single ~waste pollution zones after the
certain activities. family homes expected to proxy both contamination as
from 1964 to environmental and health versus before the
1984 within 2 effects. contamination
miles of the incident using the
habor for three hedonic property
cities. price model.
McClelland et al. 1990. Estimate the Dependent variable: Sale Results were
impact of a price of the home statistically
Examined impact of a Superfund site Independent variables: significant from
waste site that covered using a health -neighborhood zero. $11,897
190 acres and contained risk variable. The -structural before closure
30 million cubic yards of sale of 178 -an aggregated neighborhood and $5,822 after
refuse. homes from health risk belief variable to closure of the site
‘ August 1983 to capture health effects for a $135,000
November 1985 dollar residence.
in the Los
Angeles area.
Michaels and Smith. Estimate the Dependent variable: natural $248 to $300 for
1990. benefit of log of the sale price of the the removal of a
cleaning up or home. site. Results
Used same eleven sites removing a site. Independent variables: based a on full
as found in Harrison and That is, the site -neighborhood market analysis
Stock. would no longer marital which were
exist. The annual -town effect dummies statistically _
sales of homes in -time dummies significant.
the Boston area -distance to nearest waste site Estimates using
for years 1977- proxies both health and the submarkets
81. environmental effects bracket the
-distance to nearest site estimate from the
interacted with time dummies sample. The
to capture announcement coefficient
effect. estimates are
generally not
significant for the

submarkets.

17

Table 2.1: Hedonic Models used for Estimating the Damages of Superfund Waste Sites, Continued

 

 

Annual Benefit
Estimate of
removing a site
Objective and per household
Study Data Variables (1990 dollars)
Kohlhase, 1991. Estimate the Dependent variable: natural $377 for the
impactoflisting logofthehouseprice. removalofasite.
Looked at seven sites a site on the Independent variables: Estimate was for
containing various types NPL. The sale of -structural 1985 time period
of chemical homes in the -time period dummies data. Estimate is
contaminants. Size of Houston, Texas -neigborhood statistically
sites ranged from 1 acre area foryears -distance to nearest toxic significant from
up to 56 acres. 1976, 1980, and waste site proxies both health zero. The
1985. and environmental effects. coefficient
estimates of the
distance variables
were not
significant in 1976
but were
significant but of
the"wrong' sign in
1980.
Hoehn et al., 1987 and Derive unbiased Dependent variables: $168 to remove a
Blomquist et al., 1988. method for average hourly earnings and site from the
estimating monthly housing county.
Looked at Superfund amenity values expenditures. Superfund site
sites. and estimate the Independent variables: variable was
quality of life in structural significant.
selected U.S. -neighborhood However, licensed
counties. Data -climatic waste sites were
on 34,414 -environmental not statistically
households and socioeconomic different from
46,004 workers -number of Superfund sites zero.
form 1980 proxies health and
Census data environmental effects.
Nieves et al., 1991. Estimate the Dependent variables: Damage estimates
impacts on social annual wages plus other were not
Included Superfund sites, welfare caused income and the value of the statistically
and two arrrently by noxious home. sigtificant from
operating commercial facilities. Data Independent variables: zero for the
facilities for disposal of on 60,404 -structural hmrdous waste
low-lwel radioactive households and -socio-economic site category.
waste. 25,279 workers -climatic
from 1980 disequilibrium
Census data. -instrumental
density of the number of
hazardous waste sites proxies
health and environmental
risks. Density is the number
of these sites per 100 square

miles.

 

 

18
sites were not Superfund sites since the NPL did not begin until 1982. However, this

study is included since it is often quoted in the literature and later studies incorporated
many of the same methods used by Adler et al.

After a thorough search for an appropriate location with which to estimate the
damages of hazardous waste sites, Adler et al. chose two cities with hazardous waste
sites; Andover, Minnesota, and Pleasant Plains, New Jersey. These two cities were
chosen for several reasons. One reason is that their populations are relatively
homogeneous with respect to income, race and education. The authors point out that in
deriving welfare loss estimates it is assumed that tastes and income are identical for all
households. A second reason is that both cities have large residential populations
located close to the sites. They expect this would produce sufficient turnover of
residential property so as to have useable data. Third, the cities have no other major
source of disamenities, thus it is easier to isolate the effects of the hazardous waste sites.
Finally, households have information or knowledge of the site.

Another reason why Pleasant Plains was chosen is that actual widespread
contamination of private wells was discovered and announced in 1974. That is,
contamination of goundwater in the upper portion of two aquifers occurred. The
contamination occurred when an illegal dumping operation took place on a former
chicken farm near Pleasant Plains during a 10-month period in 1971. The chemicals
dumped at the site included aromatic hydrocarbons, benzene, toluene, styrene, xylene,
ketones, alcohols and phenolic resins.

The authors chose the Andover site because it is an example of a site for which
there was more of a threat of further contamination than actual contamination at the

time of the study. The waste site, owned by an individual, had many barrels of waste

 

19
solvents, paints, inks, glues, and gease that had deteriorated and had begun to leak

before they were moved to another location.

Data from county assessment offices was obtained for Pleasant Plains for the
years 1968 to 1981. There were 675 observations on the sale price of homes. For
Andover, 250 observations were collected for the years 1978 to 1981.

Analysis of the two cities differed. For Pleasant Plains, cross-section regessions
were estimated for before the contamination incident and after the contamination
incident (1974). This was done to test the hypothesis that the hedonic rent gradient
before the contamination was different from the hedonic rent gadient after the
contamination incident. For Andover, a regession was estimated using data after the
site was discovered by local authorities.

The dependent variable in the regressions is the sale price of the home. The
independent variables include a vector of lot and housing characteristics, locational or
neighborhood characteristics, date of sale and variables for distance from the site. The
regessions were estimated in log-linear form.

Distance variables were used to proxy the effect of the hazardous waste sites.
The distance variables were specified as 11 1/4 mile dummy variables. Each dummy
represents the observations inside one of 11 concentric circles each 1/4 mile apart. The
distance extended out two and a half miles from the site.

For Pleasant Plains, an alternative model was also specified where, rather than
using just the above discrete distance measures, the waste variable was a designated
contamination zone. Two zones were identified. Zone 1 was the area where households
were asked to seal their wells after the contamination was detected. Zone 2 was the

area where households were ordered to dig deeper wells.

"a

20

Thus, only one hedonic regession was estimated for Andover, which has a
distance variable proxyirng the hazardous waste site after it was discovered by local
authorities. On the other hand, four regessions were estimated for Pleasant Plains, a pre-
and post-contamination event for the two specified models, the distance model and
contamination zone model.

The results for Andover show that the coefficients on the dummy distance
variables were not statistically significant at the 95% level and not of the expected sign.
That is, the negative sign on the dummy distance variables imply that distance from the
site is negatively correlated with property prices. Apparently the existence of
contamination had triggered no local differences in property values. According to the
authors this was probably due to the fact that Andover draws on a different aquifer than
the aquifer at risk and the contamination that did occur was not substantial and was
limited to wells located on the same property as the waste site. In addition, the
eau'stence of a municpal landfill one-half mile north of the site reduced the ability of the
model to capture the disamenity effect of the hazardous waste site. Other variables in
the regession were generally significant. The adjusted R2 was 0.656. As noted earlier,
Andover was included in the analysis to test if the ”threat" of contamination to water
supplies causes property price differentials to arise in the housing market.

The results for Pleasant Plains with respect to the zone of contamination
formulation was not significant with an adjusted R2 of .911. The authors speculate that
this may be due to the boundaries of zones overlapping areas that are more highly

valued. That is, the zones are capturing effects other than that caused by the waste site.

21
With respect to the distance formulation for Pleasant Plains, the pre-

contamination results on the distance variables were not significant as expected, while
the post-contamination distance variables were significant. The adjusted R2 for the pre-
and post-contamination regessions were .788 and .892 respectively. In fact, the authors
speculate that the post-contamination estimates might have been even larger if not for
three events. One event was that the local government responded quickly to cleaning up
the contamination soon after the contamination was discovered. Second, public water
supply hookup for private well owners occurred one month after the date of discovery.
Third, no demonstration of contamination had occurred since 1976. In addition, the
existence of a landfill nearby may be reducing the size and significance of the hazardous
waste variable.

Using the post-contamination distance model results for Pleasant Plains,
economic damages in 1990 dollars were estimated to range from $7,269 per household if
the household is located close to the site (less than half a mile), down to $1,992 per
household if the household is located 1.75 miles from the site. In addition, a residence

I

1.5 to 1.75 miles from the site sold on average for six percent more than a residence
within one-half mile of the site.

Harrison and Stock [HS] (1984) applied the hedonic property value model to
estimate the benefits of cleaning up hazardous waste sites in the Boston area. Their
model attempts to measure health and aesthetic damages associated with hazardous and
nonhazardous waste sites. Health effects, caused by hazardous waste sites, result from
drinking contaminated water, breathing contaminated air, coming in contact with
contaminated soil, or experiencing the results of an explosion or fire at the site.

Aesthetic effects from hazardous and nonhazardous sites include unsightly visual impacts,

 

22
noise, traffic or odor. The aesthetic efiects would not be eliminated in their model even

if the hazardous wastes at the site is removed because even though the hazardous wastes
are removed the nonhazardous characteristics of the site remains, such as noise and
visual effects of the site. The benefits of cleaning up hazardous sites in their model are
therefore characterized by the elimination of adverse health effects. That is, the
hazardous site becomes nonhazardous.

Eleven sites in the Boston area were identified as containing hazardous material.
Most of these sites were on-site lagoons used to store process wastes. One of the sites
was identified as hazardous because it contained a variety of halogenated and aromatic
organic compounds which are listed as toxic under the Resource Conservation and
Recovery Act. The authors chose the other ten additional sites because they were
judged to be equivalent in toxicity to the first site based on the rating scheme used for
the Superfund. The size of the sites in acres were 1, 4, 5, 10, 10, 25, 30, 180, 200, 300,
and 400.

HS empirical results were based on housing transactions for single family
detached residences in the Boston Metropolitan area. Data for 2,182 individual housing
tracts for November 1977 to March 1981 were collected. In their analysis the dependent
variable was the actual selling price of the home in 1980 dollars. OLS was applied to a
log-linear hedonic price equation with the natural logarithm of the selling price of the
home as the dependent variable.

The independent variables included four employment accessibility variables,
fourteen structural attribute variables such as square feet of living space and lot size,
three neighborhood variables such as the full property tax rate, tWo amenity variables, a

number of health and aesthetic risk variables and fixed effect and time variables.

' 'l.“s-l_‘l-'l.'l‘

23

Two variables were constructed to represent the risk of health effects from
hazardous waste sites. One variable was the inverse square of the distance of an
individual’s home to each site (RISKI). The second variable was the inverse square of
distance of the individual’s home to the site weighted by the area of the site (RISK2).
According to the authors these variables were intended to proxy the decrease in chemical
mass from the site to distances further from the site and are expected to capture the risk
of human contact with the toxic chemicals found at a hazardous waste site.

To control for the aesthetic disamenity effects of waste disposal sites (noise,
traffic, odors) the authors included as independent variables the number of sites (non
hazardous sites, industrial sites and landfills) at one-half mile increments from 0.0 to 2.5
miles around the house. The sites include 41 non-hazardous industrial sites that stored
wastes on-site and 49 commercial and municipal landfills in the Boston area. The
industrial sites are similar to the hazardous sites except in the composition of the wastes
and thus represent a good approximation to a hazardous site after cleanup. That is, the
industrial character of a site would remain but no hazardous material would be present.
Thus the authors expect the disamenity effects of hazardous sites can be captured by the
distance to nonhazardous sites. That is, authors assumed that aesthetic efiects from
waste sites occur whether they are hazardous or not. Therefore, what they estimated
were the aesthetic impacts of non-hazardous waste sites on households.

One problem the authors failed to recognize is that their health risk variables
may also capture disamenity affects of hazardous waste sites. That is, as distance from
the site increases, it is expected that noise, dust etc. would decrease. Thus, their health
variables may capture some of the environmental damages of hazardous waste sites.

This may affect the size and sign of the estimated health effects.

24
Due to a lack of data on neighborhood characteristic variables, the authors

included town dummies in the equation to account for omitted neighborhood
characteristics. In addition, to control for the fact that their observations occurred over
a five-year period during which interest rates and other common influences on housing
prices varied widely, they included dummies for the quarter in which the sale of the
home occurred. In other words, they controlled for time effects.

Finally, to account for the fact that the presence of a hazardous waste site nnight
interact nonlinearly with the house price itself, they included interaction terms in which
the two hazardous waste health variables, RISKl and RISK2, were multiplied by a
predicted price obtained from an initial regression.

Using OLS, the adjusted R-squared statistic for the estimated equation was 0.80.
The results for the estimated coefficients displayed the expected sign for structural
variables and neighborth variables. The estimates were statistically significant.
Coefficients for the accessibility variables had the expected signs but were not statistically
significant. The estimate of the coefficient for the town dummy variable was significant
whereas the estimate of the fixed efiect (time) dummy was not significant.

The estimates for aesthetic coefficients were positive rather than negative, except
for waste sites 2.5 to 3.0 miles from the place of residence, however, the estimated
coefficients were not statistically significant (t-statistics ranged from -0.74 to 1.72). HS
explanation for this result was that proximity to waste sites proxies local accessibility
advantages that are not accounted for by their area-wide accessibility measures. They
concluded that advantages of proximity to industrial centers outweigln the aesthetic

disadvantages.

 

25

The health damage coefiicients were positive and insignificant (t-statistics for
RISKI and RISK2 were 0.46 and 0.36, respectively). The interaction terms were
negative as expected since the interaction terms reflected an increasing marginal value of
waste cleanup as predicted house price rises. However, these interaction terms were
insignificant with t-statistics of 0.46 and 0.38.

HS estimated the total benefit of cleaning up a single hazardous waste site in the
Boston area ranged from $3.6 to $17.4 million dollars (1980 dollars), depending on which
site was cleaned up. In 1990 dollars the benefit ranges from $5.7 to $27.6 million dollars.
The benefit was $15 to $70 (1990 dollars) per year per household depending on which
site would be cleaned up. The authors noted that for a $160,000 house, the willingness
to pay for cleanup of a site 1.5 miles away in 1990 dollars is $2,540; if the site is only
one-half mile away the estimated willingness to pay increases to $21,415. Note that these
willingness to pay estimates were calculated from their basic equation even though the
coefiicients were not significant. This implies that their benefit estimates probably
should be ignored.

HS point out that their benefit estimates are irnprecisely measured because of the
2,182 observations in the data set only 515 of the observations have a hazardous waste
site within four miles of the house. In addition, the benefit estimates vary due to
differences in the size of the site and the location of the site. Some locations have a
relatively dense population with expensive homes while other locations are less densly
populated with lower housing prices.

HS developed an alternative econometric approach to the health risk variables
since the specification of these variables implies a specific functional form for the

distance from the site and the willingness to pay to remove its toxic materials. Instead,

26

similar to the aesthetic variables, they estimated a less restrictive series of equations
adding variables based on the number of hazardous waste sites falling in half-mile rings.
To obtain a nonparametric estimate of the effect of distance on the willingness to pay,
they varied the distances at which the half-mile rings began. That is, they did not
assume a restrictive set of distributions concerning the health data, instead they allowed
the distance of waste sites from a home to vary. Four regessions were run, with the
second ring respectively beginning at 0.125, 0.250, 0.375 and 0.5 miles. The coefficients
on these new variables provide a semiparametric estimate of the benefits of cleaning up
a site at a given distance. For example, the benefit of cleaning up a site 1.5 miles away
is given by the coefficient on the waste variable representing the ring from 1.25 to 1.75
miles.

The results of the semiparametric estimation procedure confirmed the
implication of HS basic equation that the value of cleaning up a hazardous waste site is
substantial for houses near the site (no statistical results were provided). However, this
value declines sharply with distance and becomes negative for distances geater than one
mile. The authors suggested that this negative value occurs because the variables are
picking up the effect of omitted beneficial aspects of proximity to the sites. Therefore,
the authors concluded that the benefits estimated using the semiparametric technique
may underestimate the true value that households place on removing toxic material. HS
did not provide regession estimates for the semiparametric specification thus it is
difficult to make judgements about their results. The basic equation is the preferred
estimating technique.

The objective of Mendelsohn (1988) was to estimate the economic damages of

PCB pollution in New Bedford harbor (Massachusetts) on nearby households. PCB’s

27
were discovered in the harbor in 1976, and in 1979 the harbor was closed to numerous

uses. The author assumes that the "pollution" event date was January 1, 1981, since most
residents would have krnowledge of the event by that date‘.

Mendelsohn used three different techniques for estimating the efiect of
hazardous waste on housing values. The approaches were the hedonic approach, the
repeat sale approach and the fixed efiect approach. The repeat sale approach uses
previous sales as a point of comparison. That is, the repeat sale approach uses pairs of
sales for the same house to control for house to house variation. The premise of repeat
sale is that housing characteristics which do not change will continue to have the same
price.

The fixed effect approach uses the mean of the full set of sales for each house to
control for unwanted variation. ‘ Both the fixed effect and repeat sale approach take
advantage of their panel data structure and control for differences by examining changes
in prices for the same house. All three approaches assume: 1) Difi'erences in sales
values between homes can be explained by the difference in the qualities of the home
and 2) homeowners will pay more for homes which are closer to a valued amenity and
this price differential reflects the marginal value of the amenity. However, only the
regession estimates for the hedonic results from Mendelsohn’s paper will be discussed.

The data used to implement the hedonic model are sales of single family homes
from 1964 to 1984 within 2 miles of the harbor for the towns of Fairhaven, Dartmoutln,
and New Bedford. Hedonic regessions were estimated for before and after the

pollution event (1981).

 

28

The dependent variable in the regessions is the sale price of the home. The
independent variables include structural variables, neighborhood variables, access
variables and the waste pollution variables which are zones of pollution.

Three zones of pollution were identified. PCBZONEI is the most polluted zone
and is the inner harbor where no swimming. fishing, or lobstering is permitted. The next
most polluted zone is the proximate outer harbor PCBZONE2, where there are some
restrictions on fishing and lobstering. PCBZONE3 represents the outermost zone which
is considered unpolluted. In every case, the author tied houses to the quality of water
nearest them. Thus, as the author points out, the pollution variable does not reflect
pollution on each property but rather the proximity of the house to water sediments
which may contain PCB’s.

Mendelsohn’s hedonic model was estimated using OLS for both the linear and
semilog functional form. In addition, hedonic regessions were estimated for before and
after the pollution event.

The R2 on the linear and semi-log pro-pollution event (before 1981) regessions
were .51 and .48 respectively. In addition, the coefficients on PCBZONEl and
PCBZONE2 were significant. The t-statistics were 1.93 and 5.45 for the linear form and
1.23 and 6.48 for the semi-log specification. These results imply that the loss for a
household (1987 dollars) located near PCBZONEI is $6,194. The loss for being near
PCBZONE2 is $9,882.

The R’ on the linear and semi-log post-pollution event were .53 and .62
respectively. The t-statistics on PCBZONEI and PCBZONE2 for the linear form were

0.09 and 1.10 respectively and for the semi-log specification were 0.59 and 5.87

 

29
respectively. The regession results imply that the value of a home located near

PCBZONEI is $600. The loss for being located near PCBZONE2 is $3,720.

These results imply that damages actually decreased after the post-pollution
event. The author qweulates that the pre—pollution event variables may have been
capturing the influence of omitted variables or possibly another disamenity. Another
potential reason for this result is his specification of the pollution date. It may be that
the majority of households were aware of the site before 1981 and over time may have
adjusted to the contamination incident.

McClelland et al. (1990) estimated the impact of a Superfund site on households
using the hedonic property price model. The uniqueness of his study is that a
neighborhood risk variable was used to capture the effect of the site. This is in contrast
to most studies which attempt to‘ use some type of distance variable as a proxy for the
site.

Data was obtained on the sale price of homes in the Los Angeles metropolitan
area located near the site. The sale price of 178 homes from August 1983 to November
1985 were collected. The landfill site was closed in late 1984. At the time of its closing,
it was proposed for inclusion on the NPL. The site covers 190 acres and contains
approximately 30 million cubic yards of refuse. Nearby residents felt problems
associated with the site included possible health problems associated with the site,
leachate disposition, migating gas, landfill use after closure, and property devaluation.
However, California department of health experts found no indication of serious health
efi'ects caused by the site. In addition, they do not expect major health problems in the

future.

 

30
In the hedonic regession, the dependent variable is the sale price of the home.

Independent variables included proximity to a major freeway, square footage of the
home, sale date of the home, if the home had a pool, and a health risk variable to.
capture the effects of a waste site on the sale price of the home.

The health risk variable is an aggegate estimate of the collective neighborhood
risk judgnent of the site. The health risk estimate was obtained from a survey of 768
households in which each household was asked to assess the risks of living near a waste
site. The risk was the number of deaths per nnillion individuals. A risk ladder was used
to help the individual identify comparable risks. A response of 500 deaths per million
individuals at risk from the site placed the respondent in the high risk goup.

Based on the survey, the area surrounding the site was divided into
neighborhoods. Approximately 10 to 15 respondents from the survey were included in
each neighborhood. For each neighborhood, the proportion of responses from the
survey that fell into the high risk goup were calculated. This calculated percentage for
each neighborhood was used as the proxy for estimating the damages of waste sites. The
mean neighborhood risk proportion in the high risk goup before the site closure was 47
percent and after closure was 23 percent.

Using OLS for the semi-log specification of the hedonic equation, the R2 was .81
and the t-statistic on the neighborhood risk estimate was -2.73. This implies that for
each increase of 10 percent in the proportion of neighborhood respondents in the high
risk goup, house prices in the neighborhood decreased on average by about $2,084
(1985 dollars). Closing the landfill increased the average house value ($135,000) by
approximately $5,001. Even after closing the site, house prices are approximately $4,793
lower than they would be if there were no health risk beliefs.

31
The authors speculate that the health risk beliefs on sale price may have been

even greater if not for measurement error and if buyers were more aware of the landfill
and its problems.

Michaels and Smith [MS] (1990) used the hedonic property value approach to
estimate the benefits of removing hazardous waste sites from the geater Boston area. A
key point in their paper is the assumption of separate housing submarkets in the Boston
area rather than the assumption of a single market as had been done in previous studies.
In addition, MS contended that distance between a home and a landfill with hazardous
waste can serve as a proxy for two effects - the disamenity associated with landfills in
general and the heightened perception of risk when hazardous wastes are present.

The authors used the same eleven hazardous waste sites as described in Harrison
and Stock (1984). As of 1984, four of the sites had been included on the NPL. In
addition, they used the same data on house sales as Harrison and Stock - sales prices for
2,182 single-family homes beMeen November 1977 and March 1981. The main focus of
their study is not to repeat the Harrison and Stock study but to attempt to show that
segnentation occurs in the housing market which affects valuation of hazardous waste
sites.

In their estimated equation, Michaels and Snnith included the natural logarithm
of the deflated sale price (1977 dollars) of the house as the dependent variable.
Independent variables included linearly into the equation were structural characteristics,
distance to landfill, and neighborhood characteristics. Three variables were used to
specify the effect of hazardous waste sites on households. One of these waste variables
was the distance to the nearest waste site (MINDHW). The other two waste variables

were interaction variables.

32
One interaction variable is the multiplication of MINDHW by TIMEl, a dummy

variable, where TIMEl attempts to capture the short-term response to announcements
of hazardous waste sites. This short-term response was specified as six months after the
discovery of the waste site. The second interaction variable is the multiplication of
MINDHW by TIMEZ, where TIME2 is a dummy variable for sales after the end of the
six month discovery period.

The authors estimated one full sample hedonic price function, which included all
2,182 observations. They also estimated four separate hedonic price functions for housing
submarkets identified by housing realtors in the Boston area. The submarkets were
classified as premier, above average, average or below average.

Using ordinary least squares, the results showed that the full sample hedonic
price function performed best overall. It had an adjusted R-square of 0.626. The signs
on the coefficients were as expected and were generally statistically significant. The t-
ratios for MINDHW, TIMEI‘MINDHW, and TIMEZ‘MINDHW were 1.288, 4.213, and
6.901 respectively. The hedonic price function for the premier, above average, average,
and below average submarkets had adjusted R-squared statistics of 0.72, 0.67, 0.56, and
0.56 respectively. However, estimated coefficients tended to be not significant and not
always of the expected sign in the submarket hedonic functions. T-ratios for the four
submarkets for the variables MINDHW, TIMEI’MINDHW and TIME2‘MINDHW
ranged from -3.348 to 0.369, -0.877 to 2.26, and -1.195 to 7.229, respectively.

However, a Brown-Durbin-Evans test (cusum of squares statistic), an
independent statistical test for estimating the stability of the hedonic function, implied

that a single hedonic price function was not adequate for describing the determinants of

 

33
the real sales prices in suburban Boston. That is, the Brown-Durbin-Evans test implied

that they misspecified their equation by assuming a single housing market.

In addition, a Tiao-Goldberger test is used to determine if there are
distinguishable differences in the hedonic price functions across markets. The results of
the test suggested that most of the independent variables have significantly different
efiects on real prices in the different submarkets. In addition, all of their distance
measures had significantly different efi'ects across markets.

Finally, Michaels and Smith estimated the marginal willingness to pay to remove
a hazardous waste site in 1977 dollars for the full sample and for three of the submarkets
with plausible estimates. The estimates ranged from $38 to $1799 dollars per year per
household depending on the housing submarket. In this case, the removal of a site is the
equivalent of an increase in the distance to the nearest source of hazardous waste
exposure. They_ pointed out that a simple average of the benefit estimates across
submarkets is $139 versus $115 for the full sample estimate in 1977 dollars ($300 versus
$248 in 1990 dollars).

Based on their submarket estimation, Michaels and Snnith concluded that any
distance/ timing measure is a poor proxy for a household’s perceptions of the disamenity
and risk associated with hazardous waste sites. The distance measures may be capturing
other amenities or disamenities present in the towns that are difficult to measure. In
addition, if households expect the town to respond quickly and effectively to
contamination events there is less likelihood that the market will exhibit premia for
homes with increased distances from landfills with hazardous wastes. Also they
contended that identification of distinct submarkets can help characterize the influence

that specific housing or site attributes have on equih'brium prices.

34
One problem specifically analyzed in the Michaels and Smith article is the

assumption that there is one hedonic price function for an urban area, a contention
made by Freeman (1979b). If an urban area actually consists of several different
submarkets then one should be estimating hedonic price functions for each submarket
rather than one hedonic price function for an entire market. However, according to
Freeman at least two conditions must be met for different hedonic price functions to
exist in an urban area. First, there must be some barrier to the mobility of buyers in
each submarket so that buyers in one submarket cannot participate in another
submarket.

Second, either the structure of demand, the structure of supply, or both should be
different across submarkets. Even if buyer immobility exists, if the demand and supply
structures are similar among submarkets then the hedonic price functions will be similar.

Michaels _ and Smith found difierent hedonic price functions among their
identified submarkets in the Boston area. However, other studies have not found
significant differences in the hedonic price function among submarkets (Nelson, 1978).
Linneman (1980) provided evidence of a national housing market hedonic price function.
For three geogaphic regions in the United States (represented by Chicago, Los Angeles,
and a national sample of the largest 34 cities) Linneman could not find a significant
difference between the subsectors in Chicago and Los Angeles and the national sample.
The hypothesis that the functional forms of the hedonic price functions for the regions
are the same as the nation was not rejected.

On the other hand, Michaels and Snnith may be picking up difi’erent functional

forms, not different equilibriums. That is, the submarkets they estimated may be located

35

on difierent sections of the same non-linear hedonic price function or they are
estimating the equation using the incorrect functional form.

Kohlhase (1991) used the hedonic property price model to estimate the impact of
EPA announcements and policy actions on housing markets. Data on individual housing
sales in the Houston, Texas area were collected for the years 1976 (n= 1,969), 1980
(n=1,083) and 1985 (n=1,881). Three time periods were chosen so as to examine the
stages of environmental awareness in Houston concerning local area toxic waste sites.

Ten toxic waste sites, which are on the NPL were identified in the Houston area
(Harris county). Most of the toxic sites were used as waste disposal dumps by
manufacturing plants located on the site. Three of the sites were solely operated as
waste disposal pits. All sites caused significant contamination of goundwater, surface
water, soil and in some cases air.‘ In addition, drinking water wells are within 2,500 feet
of each site.

The final data set was based on home sales within a 7 mile radius of seven of the
sites. A more detailed description of the seven sites is shown in Table 2.2 and was
adapted from Table 2 in the Kohlhase paper.

The year 1976 was chosen because at this time there was no NPL or Superfund
list. 1980 was the period concurrent with the creation of the NPL. Finally, by 1985, all
seven sites were announced to be on the NPL. The author then estimates, using OLS,
three regession equations pertaining to the three years of interest.

The dependent variable in the regessions is the natural log of the selling price of
the home. Independent variables include a vector of housing characteristics, a vector of

neighborhood and location characteristics, a vector of quarterly time period dummies

k

36

Table 2.2 - Characteristics of Toxic Waste Sites in Kohlhase (1991)

 

 

Site Name Date Characteristics of Waste Site‘
Announced on
NPL

Brio 10.84 56-acre site; pollutants include copper, vinyl
chloride, flourene, styrene, ethyl benzene;
water well 2500 ft.

Crystal 7-82 5-acre site; arsenic contamination; emergency
capping of site with clay late 1982; water well
300 ft.

Geneva 9-83 l3-acre site; pollutants include PCB, vinyl
chloride, asbestos insulation; emergency
capping of site with clay late 1982; water well
900 ft.

Harris-Farley 7-82 2-acre site; pollutants include styrene tars
and its degadation products; Dow Chemical
began clean-up in 1984; water well on site.

North Calvacade 10-84 23-acre site; main pollutant creosote; water
well 200 feet.

South Calvacade 10-84 46-acre site; pollutants include polynucleor
aromatic compounds associated with
creosote, benzopyrene, chrysene,
flouranthene, anthracene; water well 1500 ft.

Sol-Lynn 10-84 l-acre site; pollutants include

trichloroethylene (TCE) and polychlorinated
biphenyls (PCBs); water well on site.

' Water well is distance of the site to sources of public water supply.

37
and the distance in miles to the nearest toxic waste site (TOXIC). She also included the

square of TOXIC in the equation because the quadratic formulation allows a nonlinear
price-distance relation and the computation of a range for the perceived effect of
TOXIC on house values.

For the 1976 regession the R2 was .89. The coefficients on TOXIC and TOXIC
squared were statistically not significantly different from zero.

For 1980 regession the R2 was .88. The coefficients on TOXIC and TOXIC
squared were significant but of the “wrong" sign, negative rather than positive. She
attributes this to other unmeasured economic trends. For example, between 1976 and
1980 an employment subcenter grew in the areas of the toxic sites. Thus distance to the
site could be proxying for distance to local employment.

Finally, for the 1985 regession, the R’ was .83. The coefficients on TOXIC and
TOXIC squared _were significant (t-statistics are 2.1 and 3.4 respectively) and of the
hypothesized sign (positive).

Using the estimates from the 1985 equation, the marginal price of TOXIC
evaluated at the means is $2,364. She finds the marginal willingness to pay to be on
average 1.08 miles farther from a hazardous waste site is $377 (1990 dollars). This
represents the benefit to the average household for removing a site. This compares to
approximately $250 for the Michaels and Snnith study.

In addition, she finds for the 1985 sample that TOXIC is significant for all
distances up to 6.2 miles from the site. In 1985, she finds that the price of a home would
likely be higher if it were further from the site, by as much as $3,310 per mile evaluated

at the means.

 

38

She concludes that the announcement effect of the EPA, that is, the listing of the
site to the NPL, is the primary cause of the depression in housing values observed. This
is based on the fact that in 1976 six of the seven sites were still operating and that in
1980 the Superfund was created and five of the seven sites were still operating yet the
regession results were not significant or of the “wrong sign“. However, by 1985 all seven
sites were announced to be on the Superfund list and the results were statistically
significant and of the expected sign.

Based on her results, she suggests that households have the ability to determine
whether or not a site will continue to be toxic, but households seem to be unable to
accurately distinguish between degees of toxicity. That is, the ranking of the site on the
NPL doesn’t seem to affect the size of damages, rather the listing of the site on the NPL
affects household damage estimates.

Finally she shows points out that the one site that was cleaned up during 1984~
1986, no depressive effect on housing is observed in the 1985 sample. She claims that
this provides evidence that consumers act on the information that is available to them,
and that government and private efforts to clean-up toxic wastes can enhance housing
values.

It is interesting to note that the announcement effect had an effect in the
Kohlhase study but not the Mendelsohn study. This can be due to the fact that the
announcement effect in Kohlhase had to do with the listing of the site on the NPL.
Mendelsohn’s announcement effect was based on the potential that household’s had
heard about the pollution problem in the harbor. Listing of a site to the NPL may cause
more concern than just hearing about the pollution problem. In addition, the type of

risks involved differ. NPL sites in Kohlhase have the potential of contaminating drirnking

39
water supplies, while the harbor site in Mendelsohn is not used for drinking water
PW-

Hoehn et al. [I-IBB] (1987) and Blomquist et al. [BBH] (1988) papers focus on
providing empirical evidence that the value of amenities are captured simultaneously by
both the labor and housing markets. Both papers argued that hedonic studies that focus
on a single market such as labor or housing estimate amenity prices that are only partial
prices and thus are unreliable measures of amenity values in an interregional context.
To correct this problem, the price of amenities in an interregional context may be posed
as the sum of the partial implicit prices estimated from the labor market and the housing
market.

These studies estimated a set of amenities using data on housing prices for
34,414 households and wages for‘46,004 workers in 253 counties. The studies estimated
two equations, a wage equation and a rent equation. Both estimated equations included
the same amenity variables. Amenities included the number of Superfund sites per
county and the number of licensed waste sites iii the county among others.

The dependent variable in the wage equation was the hourly wage for the worker.
In addition, other independent variables included in the wage equation were a number of
worker characteristic variables and climatic, environmental and neighborhood amenity
variables. I

The dependent variable in the rent equation was the monthly housing
expenditures. Other variables included in the equation were a number of structural
characteristic variables and climatic, environmental and neighborhood amenity variables.

HBB and BBH performed a Box-Cox search over the functional forms of the

equations. Both equations were linear in the independent variables. However, the

40
dependent variable was transformed in both cases (See either paper for a discussion of

the Box-Cox transformation). The equations were estimated using standard OLS
teclnniques. The R2 for the wage and rent equations were .3138 and .6624 respectively in
both the HBB and BBH papers.

The Superfund site variable was significant in both equations (t-statistics were
19.37 and 6.29 respectively for the rent and wage equation). The full implicit price
on the number of Superfund sites per county was -$ 168.24 (1990 dollars). This implies
that a household needs compensation of 168 dollars annually for each additional
Sniperfund site. The t-statistic on the full implicit price was -2.43.

Particularly interesting is the finding that the number of licensed waste sites in
the county was not significant. This implies that if households have some degee of
confidence that a site is well managed they do not feel threatened by the site. Superfund
sites on the other hand are mainly nonlicensed sites and thus household concerns are
expected to be higher than for licensed sites.

Finally, Nieves et al. (1991), using an interregional wage-rent model similar to
Blomquist et al’s, estimated the economic damages associated with noxious facilities,
including Superfund sites. A wage equation and a rent equation were estimated for
owners and renters of housing.

Eight different types of noxious facilities were included in the analysis including:
nuclear-powered electric generating plants, coal-fired generating plants, gas- and oil-fired
generating plants, military chemical weapons storage sites slated for decommissioning,
hazardous waste sites, petrochemical refineries, radioactively contaminated sites managed ‘
by the US. Department of Energy under the Formerly Utilized Sites Remedial Action

Program, and liquefied natural gas storage facilities.

41
The hazardous waste category includes both chemical waste sites, all of which

existed in 1980 and are listed on the Environmental Protection Agency’s NPL. Also
included in the hazardous waste category are two currently (at the time of this study)
operating commercial facilities for disposal of low-level radioactive waste. These sites
are not listed on the NPL.

Data on these noxious facilities were combined with 1980 Public Use Microdata
Sample B from the United States Census of Bureau. This data set contains information
on workers and households. There were 25,279 observations for workers. They confined
the sample to workers who earned calculated wages of more than $2 per hour. Because
of truncation in the PUMS data set, the income category of ”$75,000“ and up was
onnitted from the data set.

There were 60,404 observations on housing included in the Nieves et al. data set.
Observations omitted from the data set were rent data in the category "$999 and up”, as
well as estimated market values in the category “$175,000 and up".

Other data combined with the noxious facility data included human capital and
industry control variables in the wage equation, structural variables in the rent equation,
and local price variables, disequilibrium control variables, and county and city level
control amenity variables in both equations. In addition, instrumental variables were
included in both equations.

Three different specifications of the noxious facilities were tested. This included
the density of all noxious facilities in the area, the density of each type of facility in the
area, and a dummy variable signifying if the facility is located in the area. Density is in

number of facilities per 100 square miles.

42

The dependent variable in the wage equation is annual wages plus other income.
The dependent variable in the rent equation is the owner’s estimate of the market value
of the residence. The rent and wage equation were estimated separately for owners and
renters since they estimated that there were significant differences between owner and
renter characteristics.

The equations were estimated using two-stage least squares and the functional
forms of the equations were double-log. Nieves et al. based the estimation procedure on
the basis of Henderson’s (1982) finding that the amenity can be fully captured in
estimates for just one of the markets if effects on the other market are simultaneously
controlled. Their results imply that hazardous waste sites are an amenity rather than a
disamenity, however, the estimated coefficients were not significant at the 0.01 level.
The authors assume the unexpected results for hazardous waste sites may reflect either
lack of public information about these sites in 1980 or that Superfund sites are

associated with productive activities.

4. IMPLICATIONS OF THESE STUDIES

These studies attempted to estimate the damages that Superfund waste sites
inflict on households, although for the HBB and BBH studies this was not their main
objective. However, the results are not satisfactory for a number of reasons.

First, distance measures appear to be a poor proxy for exposure to hazardous
waste sites. As pointed out by Harrison and Stock and Michaels and Smith, distance
proxies tend to capture other effects such as distance to the central city and other

unaccounted for amenities and disamenities.

43
The papers by HS, Mendelsohn, Adler et al., and Michaels and Smith (MS) did

not find hazardous waste sites as significantly afl'ecting housing prices. In the case of
HS, Mendelsohn, and MS this is very likely due to their distance variables capturing the
value of other variables not accounted for in the Boston area. There are a large number
of other amenities and disamenities located in the Boston area. In the case of Adler,
other landfills may have affected the result for Andover.

Kohlhase on the other hand, had somewhat more success than other researchers
with respect to using distance as a proxy for hazardous waste sites. One reason for this
may be the level and number of other externalities affecting the distance variable across
the Houston and Boston areas. In any case, her 1980 results highlights some of the
problems associated with distance proxies.

In addition to the distance and other externality problems, the site themselves
may not be significant enough to find significant effects. For example, McClelland et al.
found significant effects on households from a waste site. This site however was
extremely large and noticeable by the public. Many of the sites in the other studies were
relatively small, many 1-10 acres.

The HBB and BBH papers point out the importance of simultaneously including
the labor and housing market in amenity valuation. Their results imply that the poor
results found in many of the other papers may be related to not only the distance and
externality problem but also to not including both the labor and housing markets in
measuring the value of hazardous waste sites. In addition, HBB and BBH provides
evidence that amenity values are captured by differences in amenities within and across
cities. Thus, they showed that to obtain unbiased amenity prices, the price diflerentials

which arise in the housing and labor markets should be included in the analysis. This is

 

44

an improvement over the other studies. The other studies did not take the amenity
valuation across cities into account.

There is one point that needs to be addressed when estimating the value of
amenities. The question concerns the degree of aggregation in housing and wage
markets. The HBB, BBH and Nieves et al. studies need to be concerned about the size
of the sample to be included in the analysis. More specifically, the number of
households and workers in each county used can affect benefit estimates. That is, they
need to be confident that the sample size is large enough in each county so that it is
representative of all the households in the county.

Significantly, in the paper by Kohlhase (1991) where the distance proxies seemed
to perform well, she specified the distance proxy in quadratic form. The other papers
either assumed a linear distance or used concentric rings at various distances from a
home. The quadratic form takes into account the possibility of nonlinearities between
distance from a site and economic damages. The effects of nonlinearity was not taken
into account in the HBB or BBH papers.

A final point is the difference in the results for the valuation of Superfund sites
between the BBH and Nieves et al. studies. One possibility is that Nieves et al. included
two currently operating commercial facilities for disposal of low-level radioactive waste
along with Superfund sites. Households may view radioactive sites differently from
Superfund sites. That is, the radioactive sites are licensed while Superfund sites are not.
Household perceptions of risks may be greater for unlicensed sites versus licensed sites.
As shown in HBB licensed sites do not have significant effects. Another possibility is
that the sample size for Nieves et al. was much smaller than in BBH, 84 cities/counties

versus 253 counties. Ninety-six Superfund sites and two radioactive sites were distributed

 

45

across the 84 cities. There were 195 other type of sites distributed across the 84 cities.
It could be that Nieves et al. did not capture the variation in Superfund sites across areas

due to a small sample size.

CHAPTER THREE: A CONCEPTUAL MODEL FOR ESTIMATING THE
LOCAL AREA ECONOMIC DAMAGES OF SUPERFUND WASTE SITES

1. INTRODUCTION

This chapter constructs a conceptual model that can be used to estimate the local
area economic damages of Superfund sites. The conceptual framework derived in this
chapter is a household-firm location decision model. In this study, households and firms
must decide which county to reside in. Households and firms are expected to make their

location decisions with respect to local wages, rents, and amenities.

2. INTERREGIONAL WAGE-RENT MODEL

Household and firm location decisions are best viewed as the choice of composite
bundles of wages, rents and amenities (Rosen, 1979). Households maximize utility and
firms minimize cost by choosing the optimal bundle of wages, rents and amenities. '

Local climatic, environmental and social conditions impact the economic activity
within the local area. Past studies have suggested that characteristics of a local area
(negative or positive) affect local area activities such as migration rates, business
investment, new business formation, and recreation activity (Graves and Waldman, 1991,

Greenwood et al., 1991 and Nieves et al., 1991). Changes in these activities can lead to

46

47
changes in local property values, local wage rates and thus cause changes in local

economic development.

Henderson (1982) and Graves and Knapp (1985) showed conceptually that the
value of regional amenities are captured simultaneously in both labor and property
markets. Roback (1982), Hoehn et al. (1987), Blomquist et al. (1988), and Nieves et al.
(1991) provide empirical evidence that labor and property markets simultaneously
capture the value of amenities.

The conceptual model which is used to estimate the economic damages of
Superfund sites is a household-firm location decision model. Economic damages are
measured by the compensating changes in housing (land rents) and labor markets
(wages) for households and firms located in a county which has Superfund hazardous
waste site(s). The model developed in this section is based on Roback’s (1982) and
Blomquist et al’s_ (1988) models. Similar to the Blomquist et al. paper, the model
assumes a fixed number of urban areas in which households and firms may locate and
that before location decisions are made, households and firms are freely mobile. In
addition, an urban area is assumed to be composed of two counties. Each county is
assumed to have a fixed amount of land and a different package of amenities available to
households and firms. It is assumed there is no cross-county commuting and work hours
are exogenous.

In the Roback (1982) and Blomquist et al. (1988) articles, the firm is included in the
analysis. The same applies here even though this paper concentrates on the household.
The importance of including the firm is that the wages firms pay must match the wages
that the workers receive. In addition, firms compete with households for land, thus the
price of land depends on firm and household demand.

48
2.1 The Household

Households are assumed to be identical in tastes and skills‘. Households
maximize utility with respect to a budget constraint. For the case of households making
a location decision choice, households gain utility through the use of a traded composite

good x, local residential land I, and amenities a. The representative utility function is:

(1) u(x,l;a)

where u(') is homogeneous of degree zero in prices and income, strictly quasi-concave,
strictly increasing and positive by nonsatiation.

A budget. constraint requires that the cost of the composite good and land
consumption do not exceed wages w. There is assumed to be no nonlabor income in this

model. The budget constraint requires:
(2) w=x+ (r‘l)
where r is the local rental cost of land.
Maximizing equation 1 subject to equation 2 and substituting in the resulting
demand functions yields the indirect utility function:

(3) v=v(w,r;a)

where the unit price of the composite good is suppressed since it always equals one.

The indirect utility function for a household located in county k is:

49

(4) vk'vk(whrl;ak)

A worker residing in county k demands residential land

(i.e. v: is avk and v: is 91:). The amount of land in county k is

61' 8w

.F"
ll
ail .5...

fixed and equal to L. The number of households in county k is thus N,,=:-=I..,,/lk if all land
is used for residences. It is assumed there is one composite worker per household and

that demand for residential land is the same for all households.

22 The Firm

Firms combine local labor and capital to produce the traded composite good x. The
prices of x and capital are fixed by international markets (i.e. assume an open economy
where labor and capital shift internationally). The price of capital and wages are
normalized on the price of x, and the price of x is set equal to unity. In addition,
production technology is assumed to exhibit constant returns to scale in both labor and
capital. The constant returns to scale assumption allows the use of the unit production

cost function. The unit production costs for a firm located in county k are:

(5) animus.)

where q is the firm’s unit production cost function, and the firm uses the labor provided
by each household. All other variables are as previously defined. The price of capital is

left implicit since capital is perfectly mobile and is uninfiuenced by amenities, its rate of

 

50
return will be equal in all places. Hence, capital input can be assumed to be optimized

out of the problem (Roback, 1982). It is expected that unit costs increase with an
increase in w and r, but the change in costs with respect to a depends on whether a is an

amenity or disamenity in the production process.

2.3 Equilibrium

For a spatial equilibrium to occur, households and firms cannot improve their utility
and firms cannot reduce their unit costs by relocating. That is, wages and rents have
adjusted so that a move cannot improve one’s present situation. More specifically, an
intercounty equilibrium occurs when all firm’s production costs are equal to the unit
product price and households across all counties have a common level of utility. There
has been some controversy over the assumption of equilibrium (i.e. Evans, 1990).
However, Graves et al. (1991) and Greenwood et aL (1991) have provided empirical
evidence that the assumption of equilibrium has no significant effects on the quantitative
or qualitative amenity valuation estimates.

For a given county, the set of wages and land rents that maintains an intercounty

equilibrium satisfies the following set of equations:

(6) 1=¢x(err;at)

(7) vo=vk(wvrk;ak)

where v" represents constant utility for all households across urban areas (counties) and

the unit cost function for all firms equals the product price, which is assumed to be unity.

51
Marginal implicit amenity prices, p, are estimated by taking the total derivative
of equation 7 and rearranging to find p.-Va‘/vw,. Thus the price of the amenity a for a

household is

(3) P. = tx(dr./dat)-th/dax

where t, is the equilibrium household demand for land, drk/da|K is the equilibrium rent
differential and dwk/da|l is the equilibrium wage differential. Thus, the marginal implicit
price of the amenity is the sum of the land expenditure and the negative of the wage
differential.

Aggregate marginal benefits of removing, say, a waste site, would require
summing the marginal implicit prices across all households.

Equilibrium wage and rent gradients are obtained by taking the total differential
of equations 6 and 7 and solving for drk/dall and dwk/dak. Comparative static analysis
can then be used to solve for the anticipated sign of the equilibrium wage and rent

differentials given specific assumptions regarding a,.

(9) deda. =- 1/B{-Ilat0rt + Caert}

(10) drk/da|I = 1/B{Va,cw, - CaIVwk}

where B is (wimpyI - Okark) > 0. (See appendix A for derivation of the gradients). The

signs on the wage and rent gradients depend on 1) the efiect of amenities on the

worker’s utility, and 2) the impact of amenities on production costs. For example, if the

52

amenity increases household utility and is a production disamenity then the sign on the
rent gradient is ambiguous, since households are willing to pay a higher rent to take
advantage of the amenity while firms would -need compensation in the form of lower
rents to reside in the county with the production disamenity. On the other hand, the
sign on the wage gradient is unambiguously positive since households are willing to
accept a lower wage to take advantage of the local amenity while firms pay lower wages
to offset the cost of the disamenity.

'Another way to understand how wages and rents are determined by the
interaction of the equilibrium conditions of the housing and labor market is seen in
Figure 3.1. In Figure 3.1 assume again that a is an unproductive amenity for firms and is
a desirable amenity for households. In addition, assume that a, the quantity of the
amenity in county two, is greater than a” the quantity of the amenity in county one. The
downward-sloping iso-cost curves are combinations of wages and rents which equalize
production unit costs at a given level of a. With a being unproductive, factor prices must
be lower in county two to equalize costs in both counties.

The upward-sloping iso-utility curves represent combinations of wages and rents
such that utility is equalized at given levels of a. In county two, households must pay
higher rents at every wage to be indifferent between the two counties. The equilibrium
level of wages and rents is found at the intersection of the iso-cost and iso-utility curves.
In county one, the equilibrium wage and rent is w1 and r1 respectively. In county two, the
high amenity county, the equilibrium wage and rent is w2 and r, respectively.

As seen in Figure 3.1, in the more amenable county two, wages are lower while
rents are only slightly higher. This is because with an unproductive amenity, firms prefer

low amenity counties while households prefer high amenity counties. Thus,

l

V

53
Figure 3.1 - Wage and Rent Equilibrium

vm: 5)

VM: 5)

GM: I1)
00m: 0,)

 

 

in county two firms will pay lower wages while households will be willing to accept lower
wages to reside in county two. On the other hand, firms require lower rents to reside in
county two while households are willing to pay higher rents to take advantage of the
greater quantity of amenity a. The result is that wages in county one are higher than in
county two, while the difference in rents is less clear”.

The analysis in Figure 3.1 becomes less clear when aggregation economies are
included in the model as was done in the HBB and BBH articles. Aggregation
economies or effects is when the population of an area affects the production costs of
local firms. Assuming an amenity is valued by households, a change in amenities in

county j results in a change in county population that in turn affects the cost of firms

54
within county j. The shift in costs induces a change in wages and rents. This implies
that the signs on the wage and rent gradients can vary depending on the affect of city
size on firm production costs.
Following Blomquist et al (1988) and Roback (1982) the last step is to replace the
land rent, r, by the price of housing, g, since it is the price of housing that is normally
observed, not land rents. This is important since housing prices are a function of the

characteristics of the house and its environment. Thus equation 7 becomes:

(11) Vo=Vr(wv&;ak)

and the equilibrium conditions are found using equations 6 and 11. The price of an

amenity a then becomes:

(12) P. = hk(dgk/dak)'(dwk/dak)
where h,I is the quantity of housing purchased by a household in county k.
3. INCREMENTAL AND AGGREGATE DAMAGES OF SUPERFUND SITES

The price of an amenity as estimated in equation (12) can be used to estimate
the incremental economic damages per household and the aggregate economic damages
per household of Superfund sites. Incremental economic damages per household are the
damages imposed on households from additional Superfund sites in a county and

aggregate economic damages per household are the sum of the incremental economic

55
damages per household. For a Superfund site, equation (12) can be interpreted as the

marginal damage of a Superfund site located in the county. The incremental economic

damages per household for Superfund sites are:

(13) D(s,z) = blag/as. + cut/dz.) - (was. + elm/dz.)

where s is the number of Superfund sites, D(s,z) is the incremental damage of the s
Superfund site and z= s’. The first term in parentheses is the effect of Superfund sites
on housing rents and the second term in parentheses is the effect of Superfund sites on
wage rates. Aggregate economic damages per household from Superfund sites is the

sum of the incremental economic damages per household:

(14) 3": D(s,z) = h.(dg./ds. + da/dzo - «was. + Men.)

34

where n is the number of Superfund sites located in the county. Aggregate economic
damages per county is estimated by multiplying aggregate economic damages per

household, equation (14), by the number of households in the county.

CHAPTER FOUR: ANALYTICAL METHODS FOR ESTIMATING THE
LOCAL AREA ECONOMIC DAMAGES OF SUPERFUND WASTE SITES

1. INTRODUCTION

The purpose of this chapter is to develop methods to estimate the local area
economic damages of Superfund waste sites. The chapter proceeds in two sections. The
first section describes the wage and rent equations as well as the econometric procedures
used in the empirical model. Section two describes the data used to empirically
implement the model. ,
2. WAGE AND RENT EQUATIONS
2.1 Econometric Specification of the Model

Assume for an individual i in household h (h=1,...,H) located in county k

(k=1,...,K), the wage (wn) equation is:

(1) wm=ao+atzw+a2ck+a3gk+a4ek+yw,

and the rent (1") equation is‘:

(2) 'M = 50 '°' 515» + 52": + 5331: + H481: + ”we

56

 

57
In addition:
zml is a vector of the individual attributes for worker ihk.
c,l is a vector of climatic conditions in county k.
g, is a vector of social conditions in county k
e is a vector of environmental conditions in county k.
s, is a vector of structural characteristics of the house for hk.
v... and u.“ are random error terms. (vml 2,“, ct, gvek, s“) are i.i.d. with mean 0
and variance of, and (“I z“, c” g,” e, s...) are i.i.d. with mean 0 and variance
2

0“-

a0, 02,, a, 01,, (1,, 5.. 3,, 13,, B, and 5, are vectors of coefficients to be estimated.

2.1.1 Correction for Hetereoskedasticity

In previous work (HBB, BBH and Nieves et al.) the possibility of
contemporaneous correlation between the error terms in the rent and wage equations
was not taken into account. However, it is possible that the errors in the wage and rent
equations are correlated, that is, for example, some amenity may significantly influence
county level wages and rents but was not included in either equation. If this occurs then
the equations may be related through nonzero covariances associated with the error
terms across different equations. Thus, any connection between the equations lies solely
in the error terms (Judge et al. 1988). This implies that if ordinary least squares is
applied separately to the wage and rent equations, a loss of efficiency occurs because
OLS equation by equation does not take into account the nonzero covariances.
Estimated parameters however, remain unbiased and consistent. Thus the standard
errors on the estimated coefficients are larger and hypothesis testing is less powerful.

To test for the effect of cross-correlated wage and rent errors, equation (1) is

reformulated by aggregating within the household. That is, instead of using equation (1),

 

58
the wage equation based on the individual level, the average household wage and the
mean worker characteristics of the household are used in the wage equation’. Equation

(1) can then be rewritten as

(3) wM=ao+agfl+azck+aagk+agk+§w

where w... is the average wage in household hk and 2,.ll is a vector of average
characteristics of the workers in the household. (full 2“, ct, g” e, s‘) is distributed
' 2

independently (but not identically) with mean 0 and variance 5' and N“, is the
h]:

 

number of workers in household h in county k.

However, a problem which arises from equation (3) is hetereoskedasticity, as
seen in the conditional distribution of {a from the average wage equation (eq. (3)).
Hetereoskedasticity occurs by aggregating within the household. Since the number of
workers varies by household, averaging causes the variance of the error term to differ
across households. As a general result, parameter estimates under heteroskedasticity are
consistent and unbiased but inefi'icient and the covariance matrix estimates are
inconsistent and biased.

To correct for this problem, the wage equation is multiplied through by the
square root of N“. A

The wage equation then becomes:

«NTWM=aoJN—hk—.+alNhk zM+a2JNTck+GSJiv—M—gk

+a‘N,‘ 6,445”,

(4)

59

where 51:]: a: M“): and (rail 2,, c, g, e” s“) are i.i.d. with mean 0 and variance

2
O"-

(homoskedastic) across households (i.i.d). Equations (2) and (4) represent the rent and

This weighting procedure causes the error term to again become constant

wage equations to be empirically estimated.
Following the assumption made before (ml 2,“, c” g, e, s_) are i.i.d. with mean
0 and variance 0% and (All zmp c, g, e, s.) are i.i.d with mean 0 and variance 02.

u
Further assume that Cov(v,,.u,,| z“, c,, g, e, s.)=pa,a,. It can then be assumed that:

(5) GOV“ Mull», ck’ 8k, ek’ sue) = nova“

where p is the correlation coefficient between the disturbances. If there is no
contemporaneous correlation between the disturbances then p is equal to zero and OLS
equation by equation is the BLUE. A p not equal to zero implies that use of OLS
equation by equation leads to less efficient parameter estimates and inconsistent
covariance matrix estimates (Judge et al, 1988).

The implication for this model is that more efficient parameter estimates and
consistent covariance matrix estimates of the local area economic damages of Superfund
sites are obtained if inter-equation correlation is taken into account. If OLS is used to
derive the economic damages of Superfund sites, the efficiency of the model is lower,
thus standard errors of the estimates are higher, thereby affecting statistical testing’.
Variables which are significant at the margin can affect a decision regarding whether the
variable is important in influencing some other variable. Thus, providing more efficient
estimates can be important for empirical and policy analysis.

To take account of the possibility that there may be correlation between the wage

and rent equations, the seemingly unrelated regressions (SUR) model is applied. In a

60

SUR model, equations (2) and (4) are estimated jointly using feasible generalized least
squares (FGLS) (F omby et al, 1984). The estimated covariance matrix used in
estimating the parameters takes into account the correlation between the error terms of
the wage and rent equations. This would be seen in the off-diagonals of the covariance
matrix. However, to apply F GLS, stronger assumptions are needed than are required for
015' That is’ (uhklzmrcklgkreklshk) and (”mklzihkvckvgkvekrsak)
are conditional on "all" independent variables. For OLS, p... is not conditional on 2m and
v“ is not conditional on s“. For FGLS to be consistent, the error terms cannot be
related to any of the independent variables. The efficiency gains from using SUR is
greater as the correlation between the error terms increase or as the correlation between
the independent variables decrease across the wage and rent equations (Judge et al,
1988).

To test for the appropriateness of the SUR model, a langrangean multiplier test
can be performed to test for Ho: p=0 (the covariance between the error terms equals

zero)(Judge et al, 1988). The test statistic is given by:

<6) x = ms.)

where T is the number of observations and r"u is the squared correlation:

(7) 2 012

 

where 0:2 is the square of the estimated covariance between the wage and rent

equation and 011 and 022 are the estimated variances for the wage and rent equation

 

61
’ respectively. Obviously, r32 = p2 . Under the null hypotheses, A has an asymptotic X"
distribution with 2 degrees of freedom. The null hypothesis is rejected if h is greater
than the critical value, implying that the covariances are significantly different from zero
and that the SUR model leads to consistent and efficient estimates.

To summarize the econometric model, the averaged data in the wage equation is
first weighted to correct for aggregation bias. The SUR model is then applied to
estimate the weighted average wage equation and the rent equation jointly as a system to

assess potential efficiency gains and to acquire a consistent covariance matrix.
3. DATA TO IMPLEMENT THE EMPIRICAL MODEL.

The data used to implement the empirical model in section 2 was obtained from
an earlier study by Blomquist et al. (1988). The data from their study was micro data on
workers and household obtained from the 1980 Census 1 in 1000 A Public-Use Sample.
The authors merged the microdata with county noncensus amenity variables by county.
The merged aggregate data consist of observations on 34,414 households and 46,004
workers who reside in the households. 253 counties are represented in the sample where
each county has a population exceeding 100,000 individuals.

The 46,004 workers were then matched to the household data. After determining
which workers matched with which household, the average hourly earnings and average
characteristics of the workers within each household were calculated. The matching of
workers with households resulted in 23,937 observations. This is less than the number of

households because for a number of the households there were no matches with workers.

62

3.1 Rent data

The Blomquist et al. housing sample includes all housing units on ten or fewer
acres for which value of the unit or contract rent is reported. For renters, monthly
housing expenditure is defined as gross rent including utilities. For owners, reported
house value is converted to monthly imputed rent using a 7.85 percent discount rate.
Monthly expenditures for utilities plus monthly payments for real estate taxes and
insurance are added to obtain gross imputed rent for owners. Monthly housing
expenditure is the dependent variable in the rent equation.

The rent equation as described above also includes a vector of structural housing
conditions. The vector of housing conditions, their description and mean value in the

data set is reported in table 4.1.

3.2 Wage data

The wage sample in the Blomquist paper includes all individuals aged 16 and
over who reported their earnings, hours and weeks, had nonzero wage and salary
earnings, and had positive total earnings. This includes part-time workers. The
dependent variable in the wage equation for the present study is average hourly earnings
for a household. Average hourly earnings for a single worker are calculated by dividing
annual earnings by the product of average hours worked per week and number of weeks
worked per year. The average hourly earnings for a household are the sum of average
hourly earnings for the household divided by the number of workers in the household.

The wage equation also includes worker attributes. The vector of worker

attributes, their description and mean value are also reported in table 4.1.

 

63

T“¢l-WMM¢INVM&W*.¢MMS¢

 

 

vm W Meal
ona Aw hoary—uh. lorthe MM (1900 dollars) as
am Wanamamaomn) 492.97
srra NnberofNPleeshthecouy 354
srraso wmduhmmwmz’) 3124
HDD Nunberofhadhgdepeedays mun
con 1»:deth 1152.09
m AwA-IIW(W) 31.31
HUMID Awm 0327
wmn Awwhdqacd(nfles/how) m
SUN mumm 61.28
cc mmbemmaympwmuawmm 030
chy,00¢berwhe)
rumor WWI-“o 13.02
cam. mmmwrummm) 05630
coasr IfcoudytoucbesanOwaaorGredImhllftoucheseiher) 034
mas mamammmmmemm 1.50
‘ did-urn
mox rmwmummmm) 472.02
rsr _ Tammany-nude. 7351
ms mung) 1535
within Sqmnfiuofmwderbthecomty 19.72
mm umamaumammmapmmmm) 2.40
A68 Apotthehouehyec 2253
mums Numberofmhthchome 2.30
aoous Nunberofmollhthehoue 5.64
sans Hmamhmw 3.55
anus Nunberofbdhroonshtheho-e 155
CONDO mmmmmprum.) 0.03
AIRCON mmumwmmpwm) 033
sawaa mmummmpwpm) oss
mwam mmuwmmpmm) 0.94
YARD mmusummrmormm9mom) 0.07
am mmummprumm 033
aunrr mummmum 1.77
mos mmmmmaoa 351
amounts mmmmmmw 1.02
aaoous mmmmmaoous 155

64

rum-anmavmushwmmmmseamm

 

 

Variable W Mean
RBEDS mmmmms-s 1.06
mums rmmmmmmmrs 0.45
110011130 mum-mammcouno 0.01
MOON 1m tern ham mraa and 111110024 0.10
asawaa Wmmmmm 0.36
arm 1mm m between m and YARD 0.01
EXPER Yard-BMW 18.40
sacs mmumpoumrmae) 0.15
sex mmumpweame) 0.44
MARRIED Dummibleformrhphlifwfied) 0.58
SCHOOL variances... 1292
DISABLED mmumprum) 0.05
132111011. mmmmmmowmmmr) 0.12
'racu demhuqd 11166111116166) 034
nor mammalian-111mb“) 025
CRAFT WMMpMW) 0.12
om Ocup‘lonaldunynflflleb-l am) 0.17
UNION _ mamamwm 2392
mm mamammfi 54931
smart mmammsax 731
swam, IrleractioavuiableomeRZandSEx 233.86
sues. mmamcamsax 0.0a
summer) mmammsax 021

SKIDS mmumnmormmu,msax 1.07

65
3.3 Amenity Data
A number of county level amenity variables were included in the wage and rent
equation. These amenity variables can be classified into three groups; climatic, social

and environmental variables.

3.3.1 Climatic Data

A number of county level climatic variables were included in both the wage and
rent equation. Climatic variables include heating and cooling degree days, annual
precipitation, average humidity, average wind speed, and percentage of possible sunny
days.

3.3.2 Social Data
A number of county level social variables were included in both the wage and
rent equations. Social variables include central city status, the pupil-teacher ratio in the

county, and the violent crime rate.

3.33 Environmental Data

Several county level environmental variables were included in both the wage and
rent equations. Environmental variables include a dummy variable which signifies if the
county touches a Great Lake or ocean, a number of pollution variables including the
number of pollution dischargers in the county, the quantity of toxic waste in the county,
the total suspended particulates that occurs on average in the county and the visibility in

miles within the county. A final variable is the area of surface water in the county.

66

3.3.4 Superfund Data

The number of Superfund sites in a county is included in both the wage and rent
equations. The existence of a Superfund site in a county is expected to capture the risk
to the environment and natural resources from the site. As the number of Superfund
sites in a county increases, it is expected that the risks to the environment and natural
resources increase. As shown in Kohlhase (1991), households value the presence of a
Superfund site but do not distinguish between toxicity levels of Superfund sites. This
implies that households view each Superfund site as being similar in risk as all other
Superfund sites. Therefore it is expected that the number of Superfund sites per county
should capture the economic damages imposed on local area households.

The data on Superfund sites was obtained from the EPA and includes all sites on
the NPL as of December 1990. The Superfund data set was merged to the data on

wages and rents.

CHAPTER FIVE: RESULTS AND DISCUSSION OF THE LOCAL AREA
ECONOMIC DAMAGES OF SUPERFUND WASTE SITES

1. ESTIMATION OF THE LOCAL AREA ECONOMIC DAMAGES OF

SUPERFUND SITES

The wage and rent equations are both estimated in log-linear form. The
dependent variables in the wage and rent equations, average hourly earnings and
monthly expenditures on housing respectively, were transformed using the natural log,
while the independent variables in both equations were included linearly. As pointed out
in chapter two, the log-linear functional form is typically used to specify rent equations.
In addition, the log-linear functional form is typically used to specify wage equations
(Dickie and Gerking, 1987; Gyourko and Tracy, 1989 and Roback, 1988).

The number of Superfund sites per county was included in both the rent and
wage equations quadratically. It is expected that the economic damages from Superfund
sites increase at a decreasing rate. The quadratic form allows one to statistically test the
hypothesis that Superfund sites affect wages and rents nonlinearly. This would be of
importance to policymakers making decisions on the use of monies to clean up
Superfund sites. If SITE is included in the model linearly, this implies that Superfund
sites impose constant marginal economic damages on households. On the other hand,

nonlinear specification of SITE implies nonconstant marginal damages.

67

68

The quadratic form can be thought of as using the second order Taylor’s
expansion to approach the "true" function and it is generally an appropriate method for
including variables in a model to allow for a more flexible functional form (Driscoll and
Boisvert, 1991). The other independent variables were included in the equation linearly
as a first order approximation, since the study is not specifically interested in the other
variables. Following Blomquist et al. some of the worker attribute and structural
characteristic variables were included nonlinearly.

Table 5.1 reports parameter estimates for the model. Column one lists the
variable names of both the wage and rent equations. Column two lists the OLS
parameter estimates for the rent equation, and column three lists the WI.S (weighted
least squares) for the wage equation. Columns four and five list the SUR parameter
estimates for the rent and weighted average wage equations respectively. The standard
error of the estimates are in parentheses below the parameter estimates.

As seen in table 5.1, the adjusted R”s for the rent and wage equations using OLS
and weighted least squares are 0.673 and 0.407 respectively. There were 23,937
observations in each equation. The adjusted R’ for the SUR model is 0.566.

The estimated coefficients for the structural and worker attribute characteristics
were generally significant different from zero at a significance level of 0.01 using a two-
tailed test in both the independent and SUR models. In addition, the estimated
coefficients for the amenity variables were statistically significant at a significance level of
0.01 in most cases.

Looking specifically at Superfund sites with SUR estimation, the coefficient on
SITE in the rent equation is 0.0129 with a standard error of 0.00134. The SITE

coefficient in the wage equation using SUR is 0.0106 with a standard error of 0.00204.

69

Table 5.1 - Parameter Estimates and Standard Errors for OLS and SUR‘

 

 

SUR Model
Independent Model In(RENT) ln(HAGE)
ln(RENT) ln(UAGE) (weighted average
Variable (OLS) (ULS) data)
INTERCEPT 5.834 ... 0.180 .. 5.815 ... 0.198 ..
(0.08255) (0.02111) (0.08209) (0.02101)
SITE 0.1281110?“ 0.1021110; 0.1291110?“ 0.1061110?”
(0.00134) (0.00204) (0.00134) (0.00204)
811850 4.1271110“1 4.3811110: 4.1331110“1 4.3971110:
(0.00008) (0.00013) (0.00008) (0.00013)
1100 4.2081110; 0.370x10‘ 4.2051110; 0.3901110"
(0.000002) (0.000004) (0.000002) (0.000004)
coo 0131:1101 -0.164x10'1 4.1301110: 0170:1101
(0.000005) (0.000008) (0.000005) (0.000008)
PRECIP 02751110: -0.113x10: 02701110: -0.108x10:
(0.00038) (0.00051) (0.00038) (0.00051)
1101410 -0.612x10: 0.1961110: 4.5921110: 0.2151110:
(0.00059) (0.00080) (0.00058) (0.00080)
111110 0.1741110?“ 0.1051110?“ 0.1741110; 0.1111110”;
(0.00211) (0.00314) (0.00211) (0.00313)
5011 0.2431110: -0.606x10'° 0.2601110: 0388:1104
(0.00057) (0.00081) (0.00057) (0.00081)
cc 4.8871110; 4.8821110; 4.9071110; 4.7021110;
(0.00811) (0.00947) (0.00810) (0.00947)
PUPTEACH 4.8811110: 0.7521110: 4.8411110: 0.7481110:
(0.00119) (0.00180) (0.00119) (0.00180)
cm: 0.8951110; 0.8371110; 0.8971110“... 0.8221110;
(0.000007) (0.00001) (0.000007) (0.00001)
c0451 0.850x101. -0.7571110‘2 0.8411110?“ 08511110“
(0.00589) (0.00914) (0.00589) (0.00913)
NPDES -0.157x10; -0.1501110'2 4.1575110; -0.180x10""
(0.00110) (0.00174) (0.00110) (0.00174)
010x 0.2841110“; 0.1301110“... 0.284x10‘m 0.1311110;
(0.000002) (0.000003) (0.000002) (0.000003)
rsp 4.1051110: -0.284x10'° 01031110: -0.284x10“
(0.00014) (0.00020) (0.00014) (0.00020)
v1s 4.2391110: -0.367x10’° 02341110: 4.2851110“
(0.00028) (0.00038) (0.00028) (0.00038)
was 0.3071110: 0.6311110“ 0.3001110: 0.5911110‘
(0.00007) (0.00011) (0.00007) (0.00011)
um 0.4331110" - 0.3881110" -
(0.00417) (0.00414)
AGE 4.5071110: - 04981110: -
(0.00023) (0.00023)

70
Table 5.1 - Parameter Estimates and Standard Errors for OLS and SUR, Continued

 

 

SUR Model
Independent Model ln(RENT) ln(HA6£)
ln(RENT) ln(UA6E) (weighted average
Variable (OLS) (HLS) data)
51011185 0.4011110?“ - 0.3941110;
(0.00443) (0.00441)
1100115 0.7811110?” - 0.7721110;
(0.00210) (0.00208)
8805 0.1841110?“ - 0.1851110?“
(0.00348) (0.00348)
8111115 0.228 ... - 0.223 ..
(0.00504) (0.00501)
CONDO —0. 199 .. - -0.202 ...
(0.01930) (0.01919)
111118011 0.118 .. - 0.113 ...
(0.00889) (0.00885)
5811811 0.3831110?“ - 0.3521110?“
(0.00928) (0.00921)
110811111811 -0.2301110'2 - -0.238:110‘2
(0.01150) (0.01144)
111110 0.180 ... - 0.159 ...
(0.01132) (0.01128)
118111811 -0. 170 ... - -0. 170 ...
(0.03032) (0.03015)
1101111 4.7331110;2 - -0.685x10'2
(0.00429) (0.00427)
RAGE 0.2511110“ - 0.2841110"
(0.00035) (0.00035)
1151011185 4.2381110; - 4.2381110;
(0.00488) (0.00485)
111100115 -0.118x10:. — 4.1181110:
(0.00488) (0.00485)
118805 0.1181110" - 0.1231110".
(0.00748) (0.00741)
RBATHS 4.5521110; - 05421110;
(0.00953) (0.00947)
11801100 0.271 ... - 0.271 ...
(0.03053) (0.03035)
11111118011 0.9021110?“ - 0.8941110?“
(0.01105) (0.01099)
115811811 078411101 - 4.8091110;
(0.02070) (0.02058)
11171110 -0.217 ... - -0.217 ...
(0.02223) (0.02210)

71
Table 5.1 - Parameter Estimates and Standard Errors for OLS and SUR. Continued

 

 

51111 11811111
Independent Model ln(REMT) ln(HAGE)
ln(RENT) ln(VA6E) (weighted average
Variable (OLS) (ULS) data)
EXPER - 0.3701110; 0.3631110?“
(0.00149) (0.00148)
RACE - 4.8841110; 4.6671110;
(0.01878) (0.01889)
58x - 4.4301110: 4.4491110:
(0.02114) (0.02101)
MARRIED - 0.219 .. 0.215 ...
(0.01423) (0.01415)
5811001 - 0.5781110; 0.5451110;
(0.00171) (0.00170)
DISABLED - -0. 131 ... -0. 118 ..
(0.01937) (0.01928)
81111011 - 05841110; -0.588x10:.
(0.01383) (0.01355)
18811 - 0.202 .. 0.194 ..
(0.01410) (0.01402)
PROF - 0.349 .. 0.337 ...
(0.01813) (0.01804)
CRAFT - 0.220 ... 0.215 ...
(0.01798) (0.01788)
OPER - 0.121 ... 0.119 ..
(0.01839) (0.01830)
0111011 - 0.4821110: 0.5001110:
(0.00025) (0.00024)
EXPERZ - 4.5591110: 4.5491110:
(0.00003) (0.00003)
SEXPER - 4.1511110; -0.153x10:.
(0.00233) (0.00232)
SEXPERZ - 0.2711110: 0.2781110:
(0.00004) (0.00004)
SRACE - 0.6051110: 0.8131110:
(0.02728) (0.02710)
SMARRIED - -0.247 ... -0.244 ..
(0.02384) (0.02371)
51(105 - 4.3101110; 4.2741110;
(0.00474) (0.00471)
Adjusted 11’ 0.873 0.407 0.588b

 

' Standard errors are in parentheses. Hypothesis is that estimates are significantly differentgfrom
zero. Levels of significance are denoted by asterisks. m“ means the estimate is significant at the
a-0.01 level. '“ means the estimate is significant at the a'0.05 level. ‘lmeans the estimate is
significant at the a-0.10 level. Based on two-tailed test.

System R2 for the wage and rent SUR system.

72
With respect to SITESQ, using SUR, the coefficients in the rent and wage equations are

-0.000133 (s.e.=0.00008) and -0.000397 (se=0.00013) respectively.

The annualized implicit price of a Superfund site is found by multiplying the rent
parameter estimates by the number of months per year (12) and multiplying the
estimated wage coefficients by 2725, the product of the sample means of workers per
household (1.61), mean hours per week worked (38.53) and the mean weeks worked per
year (43.92). Based on equation 2.13 and using the estimates from table 5.1, the price of

a Superfund site is estimated as follows‘:

(1.9.x. +---12..x...1

P5118 3 [151-.5118 1’ 2‘5r,511850“5n'5)12*° l

(0046,1051. . . $71619]
I

(5.1)
' [(311,8116 ‘* 2‘TBu'SITEsq*SITE)2725‘9

where the B’s are the SUR parameter estimates from table 5.1, r and w represent the B
coefficients for the rent and wage equations respectively, m is the number of variables in
the rent equation, 11 is the number of variables in the wage equation, and e is the
exponential. Consistent with Hoehn’s et al. and Blomquist’s et al. results, the annualized
implicit price of the Superfund variable is 3-107 in 1980 dollars.

To test if the SUR model increased the efficiency of the parameter estimates the
langrangean multiplier test described in Chapter four was done. With a chi-square

distribution with two degrees of freedom it was found that the estimated langrange
statistic, A , was 279, which exceeds the critical value at a significance level of 0.01. This

implies that the covariances between the wage and rent equation are not equal to zero

73

and that the SUR model uses the information in the covariance matrix to generate more
efficient estimates than using the independent modeL

To compare the effects of taking into account aggregation by households, table
5.2 includes nonweighted wage coefficient estimates for workers in column three and
weighted wage coefficient estimates for workers in columns two and four. Columns two
and four are reproduced from table 5.1. Column two is included as a comparison of
weighted OLS. The wage coefficients were estimated using the SUR model in columns
three and four.

As reported in table 5.2, the system weighted R2 for the nonweighted and
weighted SUR estimates are 0.541 and 0.556 respectively. The standard errors of the
estimated coefficients are generally lower for weighted SUR than for unweighted SUR.
This is because the covariance matrix and thus the standard error of the unweighted
SUR coefficient estimates are biased by the hetereoskedasticity. Because the estimated
variances of the coefficient estimates are biased, calculated confidence intervals and test
of significance are invalid. On the other hand, the estimated coefficients for the
unweighted SUR are unbiased and consistent but not efficient. Thus weighting increases
the efficiency of the coefficient estimates (Kmenta, 1986). For example, the coefiicient
estimate for SITESQ in the weighted SUR is almost one-third larger than for SITESQ in
unweighted SUR.

With respect to the weighted OLS and weighted SUR, the standard errors are
generally similiar for the county level amenities, however, the standard errors on the
worker attributes are generally lower for SUR. In this case weighted OLS has a biased
covariance matrix and thus standard errors as well as inefficient but unbiased and

consistent coefficient estimates. The reason the covariance matrix for weighted OLS is

Table 5.2 - Parameter Estimates and Standard Errors for Unweighted SUR
and Heighted OLS and SUR‘

 

 

015
01011188) ln(UA6E) 111M188)
(weighted average (nonweighted (weighted
Variable data) average data) average data)
INTERCEPT 0.180 .. 0.522 .. 0.198 ...
(0.02111) (0.14228) (0.02101)
SITE 0.1021110; 0.9251110: 0.1061110?“
(0.00204) (0.00230) (0.00204)
511850 03811110: 02881110? 0.3971110:
(0.00013) (0.00014) (0.00013)
1100 0.3701110‘ 0.1381110" 0.390111045
(0.000004) (0.000004) (0.000004)
800 01841110: 01501110: 01701110:
(0.000008) (0.000009) (0.000008)
PRECIP 01131110: 01791110: -0.108x10:
(0.00051) (0.000818) (0.00051)
11111110 0.1961110: 08491110‘ 0.2151110:
(0.00080) (0.00101) (0.00080)
111110 0.1051110“:.. 0.1031110?“ 0.1111110?“
(0.00314) (0.00359) (0.00313)
sun -0.606x10'° 02211110: -0.386x10'°
(0.00081) (0.00098) (0.00081)
88 08821110; 07231110; 0.7021110;
(0.00947) (0.01049) (0.00947)
PUPTEACH 0.7521110: 0.5561110: 0.7481110:
(0.00180) (0 00208) (0.00180)
8111118 0.8371110; 0.8161110; 0.8221110;
(0.00001) (0.00001) (0.00001)
COAST 07571110" -0.263x10’2 08511110"
(0 00914) (0.01015) (0.00913)
MPDES 01501110’2 02921110‘ 01801110“2
(0 00174) (0.00192) (0.00174)
010x 0. 13011101“ 0. 14211101 0. 13111101.
(0.000003) (0.000003) (0.000003)
1511 -0.284x10'° 05321110: 0.2841110"
(0.00020) (0.00024) (0.00020)
v15 -0.367x10'° 08301110“ -0.285x10’3
(0.00038) (0.00045) (0.00038)
11111811 0.6311110" 0.7881110“ 0.5911110"
(0.00011) (0.00012) (0.00011)
EXPER 0.3701110?“ 0.3721110“... 0.3831110?“
(0.00149) (0.00144) (0.00148)

75

Table 5.2 - Parameter Estimates and Standard Errors for Unweighted $08
and Heighted OLS and SUR, Continued

 

 

015
111(11488) 1n(11488) ln(UAGE)
(weighted average (nonweighted (weighted
Variable data) average data) average data)

11488 08841110; 07081110; 0.8871110;
(0.01878) (0.01884) (0.01889)

5811 04301110: 08281110; 0.4491110:
(0.02114) (0.02187) (0.02101)

1141111180 0.219 .. 0.211 .. 0.215 ..
(0.01423) (0.01321) (0.01415)

SCHOOL 0.5781110?“ 0.5371110?“ 0.5451110?“
(0.00171) (0.00175) (0.00170)

01548180 ‘0-131... 0114 ... 0118.“
(0.01937) (0.01927) (0.01928)

81111011 05841110; 07871110; 0.5881110;
(0.01383) (0.01441) (0.01355)

18811 0.202 ... 0.203 ... 0.194 ...
(0.01410) (0.01451) (0.01402)

PROF 0.349 ... 0.381 ... 0.337 ...
(0.01813) (0.01838) (0.01804)

CRAFT 0.220 ... 0.21680.“ 0.218 ..
(0.01798) (0.01819) (0.01788)

OPER 0.121 ... 0.127 ... 0.119 ...
(0.01839) (0.01879) (0.01830)

u111011 0.4821110: 0.4971110: 0.5001110:
(0.00025) (0.00025) (0.00024)

EXPERZ 05591110: 05871110: 0.5491110:
(0.00003) (0.00003) (0.00003)

SEXPER 01511110; 01451119: 0.1531110;
(0.00233) (0.00224) (0.00232)

SEXPERZ 0.2711110: 0.2771110: 0.2781110:
(0.00004) (0.00004) (0.00004)

511488 0.8051110: 0.8101110?“ 0.6131110:
(0.02728) (0.02573) (0.02710)

51141111180 0247.“ 0281 .. '0-2“...
(0.02384) (0.02130) (0.02371)

5x105 03101110; 02901110; 0.2741110;
(0.00474) (0.00471) (0.00471)
Adjusted 112 0.407 0.541” 0.588”

 

' Standard errors are in parentheses. Hypothesis is that estimates are significantly

different frgm zero using a two-tailed test.

asterisks.

Levels of significance are denoted by
means the estimate is significant at the a-0.01 level.

means the

estimate is significant at the a=0.05 level. . means the estimate is significant

at the a-0.10 level.
b System R2 for the wage and rent SUR system.

76

still biased is that it doesn’t take into account the correlation between the wage and rent
error terms. Thus the weighted SUR provides more efficient parameter estimates than

weighted OLS or unweighted SUR as well as an unbiased covariance matrix.

2. INCREMENTAL AND AGGREGATE DAMAGES OF SUPERFUND WASTE

SITES

The parameter estimates from table 5.1 can be used to estimate the incremental
and aggregate economic damages of Superfund waste sites per household. Table 5.3
reports the estimated incremental and aggregate damages per household of Superfund
sites using the coefficient estimates from table 5.1. Incremental and aggregate damages
per household are estimated for up to 11 sites in a county. Damages are annual
damages for a household and are in 1980 dollars. The standard errors for the damage
estimates are in parentheses.

The incremental damage per household for an additional site in a county
decreases from approximately 107 dollars per household for the first site to
approximately 7 dollars for a ninth site in the county. For additional Superfund sites
beyond nine, incremental damages become negative. This occurs as households don’t
place additional damages on large numbers of Superfund sites in a county.

Using a Wald test (Kmenta, 1986) it is found that incremental damages are
significantly different from zero at a significance level of 0.05 only for the first seven
sites. This implies that aggregate damages become constant for seven or more sites,
which is intuitively more appealing than decreasing aggregate damages for 10 or more

sites in a county.

77
Table 5.3 - Aggregate and Incremental Damage Estimates‘

 

 

Incremental
Number of sites Aggregate damages" damages”
1 106.73 106.73
(30.1) (30.1)
2 201.03 94.30
(60.3) (26.6)
3 282.84 81.87
(85.0) (23.3)
4 352.33 69.44
(106.4) (20.4)
5 409.34 57.01
(124.6) (18.0)
6 453.92 44.58
(139.9) (16.3)
7 486.07 32.15
(152.6) (15.6)
8 505.79 19.72
(163.0) (16.0)
9 513.07 7.30
(171.8) (17.5)
10 507.93 -5.14
(179.6) (19.8)
11 490.36 -17.57
(187.1) (22.6)

 

' Standard errors are in parentheses.
" Annual damages per household in 1980 dollars.

 

78
3. A RESTRICTED MODEL FOR ETIMATING DAMAGES

To take account of the fact that aggregate household economic damages level off
for seven or more Superfund sites in a county, the empirical model with respect to the
Superfund variables is reformulated. This was accomplished by imposing a number of
restrictions on the model.

First, a dummy variable was generated where the dummy equals one if there are
more than seven Superfund sites in the county and equals zero if there are less than
eight Superfund sites in the county. The next step was to interact the dummy term with
both SITE and SITESQ in the wage and rent equations. That is, the rent and wage

equations are reformulated as:

(53-1) rent - a, + 11,8le + fizsfl'ESO + 3,08”? + npSflESO + B.DUMMY + ...+p,,;rm

(5.2.2) wagg . .0 4 “SITE 4 1.25/T580 + “WITH + aflSlTESO + «SDUMMY + ...+a,;r,, ,

where DSITES is the interaction variable between DUMMY and SITE and DSITESQ is
the interaction variable between DUMMY and SITESQ and m and n are the number of
parameters in the rent and wage equations respectively. The last step is to place the

following set of restrictions on the reformulated empirical model described by equations

5.2.1 and 5.2.2:

79

12485 ,4. ‘1'” "TM“M’ - 2725485 ,4. ‘11” "m =

(50301) (12.31. I.“ (0+, ”W .7 - 2725‘31'u‘9 “no . $71M) ‘'7)

4 (124132 ,4. W" "”8““ 472 - 27254132 .4. (5° ””3"“ 472),

[1248, ,4. ‘1’” "“M’M’ - 2725413, .4. ‘1’ "“9""? 1
(5.3.2)

(B,+

4 [124193 ,4. ""5"“ - 27254133 .4. ‘1" "”W’ 1 4 o,

(B

[211211132 ,4. "$me — 24272541512 |,4. ‘9” ""‘n’n 1

(5.3.3)

“3‘” "”6"” — 2927251413,, ,4. W "413,11, ] = O.

4 [241243, ,4.
The first restriction limits aggregate economic damages per household for a
county with eight -or more sites to equal the aggregate economic damages per household
of a county with seven sites. This was done since it was found that aggregate economic
damages per household level off for seven or more Superfund sites in a county. The last
two restrictions restrict incremental economic damages per household for eight or more
sites to equal zero. This was done since it was shown previously using the Wald test that
incremental economic damages per household were insignificant for eight or more
Superfund sites.
The SUR model was reestimated with the above restrictions and the estimates
are shown in table 5.4. The full implicit prices for SITE and DSITE offset each other
for more than seven sites in a county, and the full implicit prices for STTESQ and

DSI'TESQ offset each other for more than seven sites in a county. Thus, DUMMY

 

h
5..

80

Table 5.4 - Parameter Estimates and Standard Errors for Restricted SUR‘

 

 

SUR Model
ln(RENT) ln(WAGE)
(wt. average
Variable data)
INTERCEPT 6.038 0.200
(0.08557)” (0.02101)”
SITE 0.220x10'2 0.7351(10’2
(0.00383) (0.00555)
SITESQ 0.2101(10‘2 0.2701(10'3
(000060)” (0.00077)
DSITE ~0132 -0.464x10'1
(0.01080)” (0.00634)”
DSITESQ 0245x102 0109x102
(0.00068)” (0.00077)
DUMMY 0.938 0.312
(006848)” (002285)”
- HDD -0.260x104 0396x105
(0000003)” (0.000003)
CDD -0.149x10'3 -0.214x10‘
(0000005)” (0000007)”
PRECIP -0.257x10'2 -0833x10'3
(000036)” (0.00051)
HUMID -0862x10‘2 0.157x10'2
(000064)” (000060)”
WIND 0.1801(10’1 0115x10’l
(000212)” (000312)”
SUN 0.299x10‘z 0.7701(104
(0.00058)” (0.00061)
CC 40.980x10'1 40744x10’l
(000611)” (000943)”
PUPTEACH -0.972x10‘2 0716x10'2
(0.00119)” (000179)”
CRIME 0.920x10" 0.866x10“
(0000007)” (000001)”

 

81

Table 5.4 - Parameter Estimates and Standard Errors for Restricted SUR, Continued

 

 

(0.01140)

SUR Model
ln(RENT) ln(WAGE)
(wt. average
Variable data)
COAST 0.991x10" -0.917x10'3
(0.00611)” (0.00906)
NPDES -0.140x10'1 -0.120x10'2
(0.00112)” (0.00175)
QTOX 0.271x104 0.124x10"
(0.000002)” (0.000003)”
TSP -0.115x10'2 -0.226x10'3
(0.00014)” (0.00020)
VIS -0.311x10‘z -0.438x10'3
(0.00026)” (0.00038)
WATER 0.970x10“ -0.113x10"
(0.00007) (0.00011)
UNIT 0.4041(10’2 -
(0.00413)
AGE -0.481x10'2 -
(0.00023)”
STORIES 0.3981(10'l -
(0.00439)”
ROOMS 0.771x10'1 -
(0.00208)”
BEDS 0.1801(10'l -
(0.00345)”
BATHS 0.221 -
(0.00499)”
CONDO -0.207 -
(0.01912)”
AIRCON 0.117 -
(0.00663)”
SEWER 0.4231(10'l -
(0.00919)”
PUBWATER 0.3201(10'2 -

 

82

Table 5.4 - Parameter Estimates and Standard Errors for Restricted SUR, Continued

 

 

SUR Model
ln(RENT) ln(WAGE)
(wt. average
Variable data)
YARD 0.159 -
(0.01 122)”
RENTER -0.164 -
(0.03004)”
RUNIT -0.724x10‘2 -
(0.00425)'
RAGE 0.171x10'3 -
(0.00035)
RSTORIES -0.252x10‘1 -
(0.00463)”
RROOMS -0.119x10’l -
(0.00463)”
RBEDS 0.120x10'l -
(0.00739)
RBATHS -0.530x10” -
(0.00944)”
RCONDO 0.271 -
(0.03023)”
RAIRCON 0.863x10'l -
(0.01095)”
RSEWER -0.808x10'1 -
(0.02050)”
RYARD -0.221 -
(0.02201)”
EXPER - 0.364x10"
(0.00148)”
RACE - -0.667x10'l
(0.01668)”
SEX - -0.441x10fl
(0.02101)”
MARRIED - 0.215

(0.01415)“‘

83

Table 5.4 - Parameter Estimates and Standard Errors for Restricted SUR, Continued

 

 

SUR Model
ln(RENT) ln(WAGE)
(wt. average
Variable data)
SCHOOL - 0.546x10'1
(0.00170)”
DISABLED - -0.118
(0.01926)”
ENROLL - -0.558x10’1
(0.01355)°"
TECH - 0.194
(0.01401)”
PROF - 0.337
(0.01603)”
CRAFT - 0.216
(0.01787)”
OPER - 0.120
(0.01629)”
UNION - 0.499x10"
(0.00024)”
EXPERZ - -0.550x10'3
(0.00003)”
SEXPER - -0.154x10‘l
(0.00231)”
SEXPER2 - 0.279x10’3
(0.00004)”
SRACE - 0.608x10'l
(0.02710)"
SMARRIED - .0244
(0.02370)”
SKIDS - -0.271x10'l
(0.00471)”
Adjusted R2 0.568”

 

“Wmmumwhngyuummmmmam
ulsehdchifimmmwm nangtheenhIehWatthea-onl
kn]. “thwuma-mm mthWatm
c-OJOIIRI.

’mn’umwmmmm

 

84

eaptures the aggregate damages for more than seven sites in a county, which equals the
aggregate damages of seven sites in a county. An F-test was done to test if there is a
statistical difference between this model and the unrestricted model. One cannot reject
the null hypothesis at the .05 significance level that the models are the same. Thus, the
more appealing restricted model is used to estimate the incremental and aggregate
economic damages of Superfund sites.

As with the unrestricted model, the incremental and aggregate economic damages
of Superfund sites can be estimated using the restricted SUR estimates from table 5.4.
Figure 5.1 shows in 1980 dollars the incremental and aggregate damages of Superfund
sites per household using the restricted SUR model estimates. Note the leveling off of
aggregate damages at approximately 480 dollars per household for seven or more sites in

a county.

 

Damages (1980 dollars)

85

Figure 5.1 - Incremental and Aggregate Damages of Superfund Sites

(per household)

 

500

,+-+-+-+-+-+-+-+--+--+--+-+-+---+
Aggregate Damages

 

450

 

400

 

350

 

300

 

250

 

200

 

150
100“

 

 

50

 

 

 

 

8 10 12 14
Number of Superfund Sites

16 18 20

 

CHAPTER SIX: RANKING SUPERFUND SITES USING THE LOCAL AREA
ECONOMIC DAMAGE ESTIMATES

1. INTRODUCTION

The EPA’s HRS ranking scheme may be undesirable from an economic
standpoint (Hird, 1990). The EPA ranks hazardous waste sites based on the total score
derived from the hazard ranking system. The basis for ranking is the threat to human
health and welfare and to natural resources and the environment (Wolf, 1988). Thus,
the total score derived is not based on markets in the economy but rather on scientific
assessment of risks. This implies that the basis for cleaning up sites is not the actual
damages imposed on households but rather the assessment of risks obtained from the
hazard ranking system (HRS).

The objective of this chapter is to provide a method for using the local area
economic damage estimates to rank the cleanup of Superfund sites. This market based

ranking can then be compared to the ranking of sites based on the HRS.

2. HOUSEHOLD AND COUNTY LEVEL ECONOMIC DAMAGES OF

SUPERFUND SITES

Table 6.1 shows in 1980 dollars the household and county annual aggregate

economic damages from Superfund sites. It should be emphasized that the damage

86

87

Tablefll-AnnlEcooonchO-apofwsaa

 

 

(1900 dollars)
HOUSEHOLD (DUNTY
ANNUAL ANNUAL
AGGREGATE AGGREGATE
ECONOMIC DAMAGES
NUMBER OP DAMAGES STANDARD PROM
NUMBER HOUSING PROM ERROR OP SUPERFUND
OP NPL UNrrs 1N SUPERFUND HOUSEHOLD srres
RANK COUNTY STATE smas THOUSANDS Sl'lES‘ DAMAGES (THOUSANDS)
. Sample Total . 499 32,100 14,472,000
1 Los Angeles CA 12 2.004 004 104 2,293,000
2 11m '17: 0 904 004 104 791.000
3 Mal-loops A2 7 000 004 103 432,000
4 Sun can CA 22 474 004 104 301,000
5 10.; WA 0 525 725 154 301.000
0 Name NY 12 432 004 105 347,000
7 3mm PL 0 477 725 133 340,000
0 Susan: NY 0 400 004 105 320,000
9 M MN 7 379 004 100 300,000
10 Cook 11. 1 1,993 130 09 300,000
11 Bergen NJ 9 307 004 103 247,000
12 St. Louh Mo 5 390 034 134 227,000
13 mm - FL 7 201 004 104 210,000
14 Drug CA 2 720 209 110 200,000
13 SM CA 3 324 034 130 205,000
10 Ella NJ 5 317 034 130 201,000
17 W PA 2 000 209 109 190,000
10 00110.0 MI 4 372 530 124 197,000
19 San Bernadino CA 4 300 530 12.3 194,000
20 Modpnery PA 15 232 004 105 107,000
21 Alameda CA 3 444 410 117 103,000
22 Dual PL 7 227 004 100 102.000
23 Allegheny PA 2 371 209 109 105,000
24 Middle-ex NJ 11 203 004 102 104,000
25 MW w: 3 370 410 119 157.000
20 Sal Lake UT 0 214 m 154 153,000
27 Flu-o CA 7 192 004 102 134,000
20 Dump 11. s 235 034 130 149,000
29 Place WA 7 100 004 104 149,000
30 Momma NJ 0 101 004 102 140,000

88

Table‘J-AnnlflMMolWSlqmud

 

 

(1900 doll-n)
HOUSEHOLD COUNTY
ANNUAL ANNUAL
AGGREGATE AGGREGATE
ECONOMIC DAMAGES
NUMBER OP DAMAGES srANDARD PROM
NUMBER HOUSING PROM ERROR OP SUPERFUND
OP NPL UNrrs 1N SUPERFUND HOUSEHOLD srrEs
RANK COUNTY srATE SITES THOUSANDS SITES‘ DAMAGES (THOUSANDS)
31 Wayne MI 1 075 150 09 131,000
32 10.01 MI 10 103 004 104 131,000
33 Mllon IN 3 309 410 120 129,000
34 Noam OH 4 227 530 123 121,000
35 Bach PA 0 105 725 152 120,000
30 Shay TN 3 200 410 119 119,000
37 Erie NY 2 307 209 100 112.000
30 Spokane wA 9 137 004 102 111,000
39 Su Diego CA 1 710 150 09 100.000
40 L000 11. 0 140 725 155 100,000
41 Hnnilou OH 2 343 209 111 99,000
42 Mano-b MI 3 230 410 110 90,000
43 Tum-1 _ Tx 2 300 209 110 97,000
44 w N1 13 121 004 102 97,000
6 Dana CO 3 220 410 119 95,000
40 D011- Tx 1 03 150 90 94,000
47 Qatar PA 9 110 004 102 09,000
40 0.0. PA 0 120 725 151 07,000
49 R0.” MN 3 177 410 119 73,000
50 Conn Costa CA 2 252 209 111 73,000
51 Ca-deu NJ 3 174 410 115 72,000
52 Am NJ 0 00 004 103 70,000
53 Hue-o- NI 2 221 209 100 04,000
54 Nug- NY 0 05 73 154 01,000
55 50030100 RS 3 140 410 117 00,000
50 Win-000.0 1L 5 93 034 140 59,000
57 Datum PA 2 201 209 109 50,000
50 Gnu-v01: sc 4 100 530 121 57,000
59 Mable AL 3 131 410 114 55,000
00 Ow FL 2 103 209 109 53,000
01 Dane WI 3 120 410 119 52,000

89

Tabletl-ADBNMMOCWSMCONDM

 

 

(1900 dollars)
HOUSEHOLD COUNTY
ANNUAL ANNUAL
AGGREGATE AGGREGATE
ECONOMIC DAMAGES
NUMBER OP DAMAGES STANDARD PROM
NUMBER HOUSING PROM ERROR OP SUPERPUND
OP NPL UNTIS n9 SUPERFUND HOUSEHOLD srm
RANK COUNTY STATE srTEs THOUSANDS SITES‘ DAMAGES (THOUSANDS)
02 Kalamazoo MI 5 79 034 139 50,000
03 Watch WI 4 91 530 120 40,000
04 San Pia-choc CA 1 310 150 09 40,000
05 WW NY 1 310 150 09 47,000
00 Gauge MI 2 103 209 111 47,000
07 Panic NJ 2 150 209 100 40,000
00 wm 1L 3 109 410 119 45,000
09 Ken CA 2 154 209 110 45,000
70 Palm Beach FL 1 207 150 07 43,000
71 Bmoue NY 4 01 530 123 43,000
72 Way CA ~ 3 103 410 110 43,000
73 010-10100- CA 3 102 410 117 43,000
74 10.1.. - MI 3 99 410 121 41,000
75 Stat OH 2 143 209 112 41,000
70 Galveston Tx 4 77 530 129 41,000
77 1m KY 1 200 150 07 40,000
70 Sun Joaquin CA 2 130 209 111 39,000
79 Jena-o0 AL 1 259 150 09 39,000
00 W sc 3 92 410 120 30,000
01 St. Joseph IN 3 91 410 121 30,000
02 um PA 2 129 209 100 37,000
03 Adam CO 3 09 410 123 37,000
04 Ana Amdel MD 2 127 209 110 37,000
05 Lack-mm PA 3 00 410 114 37.000
00 Ma MN 4 07 530 120 3,000
07 wane Nc 2 113 209 100 33,000
00 A2 1 210 150 00 33,000
09 on WA 3 73 410 124 30,000
90 Ede PA 2 102 209 107 30,000
91 van CA 1 103 150 00 27,000
92 Udon NJ 1 103 150 00 27,000

9O

TableéJ-AnnuachonouichmofWSfleLCOuinued

 

 

 

(1900 000.10)
HOUSEHOLD COUNTY
ANNUAL ANNUAL
AGGREGATE AGGREGATE
ECONOMIC DAMAGU
NUMBER OP DAMAGES STANDARD PROM
NUMBER HOUSING PROM ERROR OP SUPERFUND
OP NPL UNITS IN SUPERPUND HOUSEHOLD SITES
RANK COUNTY STATE SITES THOUSANDS STIPS‘ DAMAGES (THOUSANDS)
93 Yun- WA 3 05 410 122 27,000
94 01000410 NY 2 94 209 107 27,000
95 Bedon WA 5 43 034 141 27,000
90 B000: OH 2 92 209 100 27.000
97 W NJ 4 47 530 129 25,000
90 W NC 2 01 209 111 23,000
99 Rodi-Dd NY 2 00 209 112 23,000
100 Calhom MI 3 54 410 111 22,000
101 Eunm IN 3 52 410 110 22.000
102 Rock WI 3 51 410 117 21,000
103 Lexington SC 3 51 410 117 21,000
104 10001000 CO 1 137 150 07 21,000
105 St. Clarice Mo 3 50 410 121 21,000
100 W MD 3 49 410 122 21,000
107 [mu PA 1 134 150 90 20,000
100 E-I Bum Rouge LA 1 I34 150 90 20,000
109 Seam WA 1 129 150 05 19,000
110 Polk FL 1 127 I50 07 19,000
111 Polk IA 1 122 150 90 10,000
112 Von-I. PL 1 m 150 91 10,000
113 Abs-y NY 1 115 150 07 17,000
114 Brand PL 1 113 150 00 17,000
115 Amp-hoe CO 1 113 150 00 17,000
110 Allen IN 1 111 150 90 17,000
117 Hum TN 1 110 150 91 17,000
110 Johuou KS 1 103 150 00 15,000
119 cum SC I 99 150 91 15,000
120 Weber UT 2 50 209 119 15,000
121 Lonh OH 1 90 150 94 14,000
122 Dauphin PA 1 95 150 04 14,000
123 W VA 2 49 209 102 14,000

 

91

T0ble6.1-AnnualEcouonichwofSwedundSies,Continued

 

 

(1900 dollars)
HOUSEHOLD COUNTY
ANNUAL ANNUAL
AGGREGATE AGGREGATE
ECONOMIC DAMAGES
NUMBER OP DAMAGES STANDARD PROM
NUMBER HOUSING PROM ERROR OP SUPERPUND

OP NPL UNITS IN SUPERPUND HOUSEHOLD SITB
RANK COUNTY STATE srTES THOUSANDS SITES‘ DAMAGES (THOUSANDS)
124 Richmond City VA 1 91 150 07 14,000
13 St. Lorrie MN 1 07 150 92 13,000
120 Ector Tx 2 43 209 117 12,000
127 Santa Cruz CA 1 00 150 00 12.000
120 Lam NB 1 70 150 92 11,000
129 Boulder CO 1 73 150 95 11,000
130 Warunglon MN 2 37 209 109 11,000
131 Llnn IA 1 05 150 93 10,000
132 Richmond GA 1 05 150 02 10,000
133 Slearnr MN 2 34 209 170 10,000
134 Brown WI 1 02 150 97 9.000
135 Racine WI 1 02 150 01 9,000
130 Garlon _ NC 1 59 150 04 9,000
137 Alaelmr PL 1 59 150 05 9,000
130 Berkshire MA 1 50 150 90 0,000
139 Blalr PA 1 52 150 90 0,000
140 Jelleuou Mo 1 50 150 00 0,000
141 Lyoomhlg PA 1 45 150 00 7,000
142 Monroe MI 1 45 150 09 7,000
143 Vigo IN 1 43 150 93 0,000
144 Mlnnenann SD 1 43 150 94 0,000
145 Yellowstone MT 1 43 150 94 0,000
140 Roan-kc City VA 1 43 150 94 0,000
147 Calhoun AL 1 42 150 94 0,000
140 New Hanover NC 1 41 150 97 0.000
149 Aiken SC 1 40 150 101 0,000
150 Wooanny IA 1 39 150 77 0,000
151 Kanlnlwe IL 1 37 150 01 0,000

 

'Wmmmmmmmmwmmmmmme”ImmequationSJ.

92

estimates do not include existence, option, and other values of noncounty residents which
may be important.

There are 151 counties included in the sample. Of the 253 counties in the
original data set, only 151 had one or more Superfund sites as of December 1990. In
addition, there are other counties with Superfund Sites not included in this sample.
Those counties were not one of the 253 counties in the original data set. Column one is
the rank number. Rank is by county annual aggregate economic damages from
Superfund Sites. Columns two and three report the county and state name respectively.
Column four is the number of Superfund Sites located in the county. The number of
Superfund sites located in a county ranged fi'om one site up to 22 Superfund Sites in
Santa Clara county in California. Column five is the number of yearly housing units in
the county in 1980‘. This is used as a proxy for the number of households residing in the
county. Column six is the household annual aggregate economic damages from
Superfund sites in 1980 dollars’. Column seven is the standard error for the household
damage estimate. Finally, column eight is the county annual aggregate economic
damages from Superfund sites in thousands of 1980 dollars. County annual aggregate
economic damages was estimated by multiplying column five by column Six.

As seen in table 6.1 Los Angeles county had the greatest county annual aggregate
economic damages, over two billion annually. Harris county in Texas was second with
791 million dollars of damages annually from Superfund sites. Kankakee county in
Illinois had the smallest county annual aggregate economic damages from Superfund
Sites, only six million annually.

There are two factors driving the county annual aggregate economic damage

estimates. One factor is the number of Superfund Sites located in the county. The

 

93
second factor is the population Within the county. The effect of population is significant.

For example, Montgomery county, Pennsylvania has 15 Superfund sites yet is ranked 20
on the list, while other counties, such as Cook county, Illinois, are ranked higher even
though they have less Superfund Sites. The reason for this result is that Cook county has
a significantly greater population which is affected by Superfund sites than Montgomery
county, 2 million housing units versus 232 thousand housing units. The county annual
aggregate economic damages for all 151 counties is approximately 14.5 billion dollars.
Another method for ranking counties is to estimate the household and county
annual incremental economic damage of the nth site. This is Shown in table 3.1 in
appendix B. For example, Los Angeles county would have a household annual
incremental economic damage of zero for the 12th site, While Cook county has a
household annual incremental economic damage of 150.3 dollars. The county annual
economic damage from the incremental site is estimated by multiplying the household
incremental economic damage by the number Of households in the county. This
procedure can be done for all counties and the rank reestimated. It can be seen in the
first row that the county annual aggregate economic damages for all 151 counties from

the incremental site is approximately 3.3 billion dollars.

3. COMPARING THE RANKING OF COUNTIES USING AGGREGATE

COUNTY DAMAGES AND THE HAZARD RANKING SYSTEM

In this section, the ranking of Sites for cleanup based on the hazard ranking
system (HRS) is compared to the ranking of sites based on county annual aggregate

economic damages. The 151 counties with Superfund sites are ranked using the scores

94
obtained from the HRS and from the aggregate county damage as seen in table 6.1. It is

assumed that geater damages implies greater benefits if all the sites are cleaned up in

the county.

3.1 Implementation and Results of the Rank Comparisons

The first step is to rank the 151 counties from the sum of the highest HRS scores
to the lowest HRS scores. In the case of ties between counties, the counties were given
the same ranking.

Second, the 151 counties were ranked from highest county annual aggregate
economic damages to lowest county annual aggregate economic damages. As for the
HRS ranking, ties were given the same ranking.

The next step was to match the HRS and county annual aggregate economic
damage rankingsfor each county and to run a Spearman correlation test of rankings.
The results from the test Show that the value of the Spearman correlation was 0.689,
implying that there is strong correlation between the HRS and the ranking system based
on the economic impacts of Superfund sites.

To test if population was the driving force in the HRS rank being similar to the
damage estimate rank, the HRS rank was reestimated with the population factors
removed from the HRS scores and the new HRS rank is compared to the damage
estimate rank.

As mentioned before, the HRS score is based on four pathways which chemicals
from a Superfund site can potentially aflect human health, welfare and the environment.
These pathways include a groundwater route, surface water route, air route, and a direct

COHtfiCt route.

 

95

Within each route are a number of factors including; toxicity rating of the site, quantity
of hazardous waste at the Site, distance of the Superfund site to the nearest public well
and population among others. Many of these factors are included in each pathway.
Most of the factors are expected to capture the risk to human health. One specific
factor that is expected to capture environmental effects is distance to a sensitive
environment. Distance to a sensitive environment is found in the air migration route,
direct contact route and surface water route. Population factors are found in all four
routes.

To do a comparison of the HRS scores without the population factor, a smaller
subset of 104 counties were used in the analysis due to missing data and other
incompatibilities. First, for comparison, the full HRS rank was compared to the damage
estimate rank. The Spearman correlation was 0.644, slightly smaller than for all 151
counties. The Spearman correlation for the population factors taken out of the HRS
score resulted in a Spearman correlation of 0.632. Finally, with the population factors
only included in the HRS resulted in a Spearman correlation of 0.640. These results
imply that the population factors have a Slight affect on the correlation between the HRS

rank and the damage estimate rank.

 

‘ t

 

CHAPTER SEVEN: CONCLUSIONS, IMPLICATIONS AND FUTURE RESEARCH

1. INTRODUCTION

The objective of this study was to develop analytical methods for estimating the
local area economic damages of Superfund waste Sites. The conceptual framework was a
residential location model, where damages of Superfund sites are estimated by the
adjustment of wages and rents to Superfund sites across counties in the sample. The
damage estimates were subsequently used to measure county level aggregate damages
caused by Superfund sites. Additionally, the county damage estimates were used to rank
the cleanup of Superfund Sites and this rank was compared to the EPA’S Hazard
Ranking System.

2. IMPLICATIONS

Superfund Sites are not prioritized on the basis of benefit-cost analysis. There is
concern, though, that Superfund sites impose significant damages on households. This
implies that methods are needed for estimating the economic damages associated with
Superfund sites.

Previous studies that attempted to estimate the damages of Superfund Sites met

With mixed results. In most cases the method used to estimate damages was a hedonic

96

 

.- I

 

97

rent model where distance from a home to the Superfund site was used to proxy the
damages associated With the site. However, it was found that in most cases distance
variables performed poorly. In addition, these studies did not take into account the
possibility that economic damages may be captured in other markets.

An improvement in the methods was explored by Blomquist et aL 1988. They
provided evidence that the economic damages associated with Superfund sites were
captured Simultaneously in both the housing and labor market. In addition, rents are
affected intraregionally as well as interregionally.

The method developed in this Study was based on the Blomguist et al, 1988
model. In contrast to their study, the number of Superfund sites in a county was allowed
to affect rents and wages nonlinearly. In addition, the average household wage and
characteristics were used in the analysis. Finally, seemingly unrelated regressions rather
than Simple OLS was used to estimate the wage and rent gradients.

Using the model developed in this study, it was estimated that Superfund sites
are associated with Statistically significant damages. The economic damage estimates
were used to rank the cleanup of Superfund sites. The present method used by the EPA
for ranking the cleanup of Superfund sites is the Hazard Ranking System (HRS). It was
hypothesized that ranking sites using the county damage estimates would differ from the
HRS ranking. However, the HRS rank was found to be similar to the rank using the
county damage estimates.

The implication of the method used in this study to estimate economic damages
is that it detects statistically significant and large damages from Superfund sites. The
method appears to be useful in estimating the interim damages caused by Superfund

sites on local residents.

98

An important result that can effect the rank of cleaning up sites was the finding
that damages imposed on county residents are significant and non-zero up to seven
Superfund Sites. The implication is that it may not be beneficial to cleanup the first few
Superfund Sites in a county which have a large number of Superfund sites. Rather, the
benefits of cleanup will tend to be higher in counties with very few Superfund Sites
and / or large populations. However, it is noted that factors other than economies may

come into play for ranking the cleanup of sites.
3. FUTURE RESEARCH

It was estimated that the number of Superfund sites in a county captures the
local area economic damages of Superfund sites. However, it may be useful to examine
whether there are other factors or variables that could be used to estimate the economic
damages of Superfund sites.

One alternative is to use the HRS score of a Superfund site as a proxy for the
local area health and natural resource damages caused by the site. It was Shown that
ranking sites using aggregate damages was similar to the ranking using the HRS. This
implies that damage estimates using the HRS may be comparable. In addition, one
could compare the economic damage associated with licensed waste Sites versus
nonlicensed waste sites (Superfund sites).

A second research question is to estimate if there exists a single national housing
hedonic gradient. Since property is immobile this would seem to be more of concern
than that of a Single wage hedonic. Trade-offs among housing characteristics that differ

substantially across locations may indicate the existence of separate regional submarkets

99

(Nieves et al., 1991). If submarkets exist, this may imply that amenity value estimates
from a single national housing hedonic are unreliable.

Nelson (1978), Butler (1980), and Linneman (1980) provided evidence that the
assumption of a national housing hedonic has only a slight effect on the explanatory
power of the hedonic and only a Slight effect on the accuracy of the coefficients.
However, more recently Michaels and Smith (1990) provided evidence that submarkets
exist in the housing market and amenity valuations derived from the different
submarkets are significantly different. These results imply that more work needs to be
done to determine if amenity valuation can be done assuming a single housing hedonic
gradient.

Another research question is to estimate if workers have different wage hedonic
gradients. That is, test if workers in different occupations view the risks from Superfund
sites differently. __ The assumption in this study is that one wage hedonic can be
estimated. That is, it is assumed that the labor market is sufliciently homogenous to
estimate one model for the nation. However, workers tradeoff wages with the level of
risk in their occupation. This implies that adverseness to risk may vary across
occupations. If so, then workers in one occupation may perceive the risks from
Superfund Sites to be smaller than workers in different occupations. This could result in
different housing hedonies as well. In additiOn, it may seem reasonable that there may
be barriers to arbitrage across age groups in both the housing and labor markets.

Finally, the importance of the present study is that an analytical model was
constructed which can be used to estimate the local area economic damages of
Superfund waste Sites. The ranking of sites based on the damage estimates are only

preliminary estimates. To make the results more up to data and applicable to

 

“.I

 

100
policymakers, the wage and rent data should be updated using either the 1990 Census

data or more recent wage-rent data. Damage estimates obtained from the analytical

model using 1990 Census data could then be used to rank the cleanup of sites.

 

a.“

 

ENDNOTES

 

, 1)

 

ENDNOTES

CHAPTER ONE

1. For a discussion of the hazard ranking system see Haness and Warwick
(1991).

2. An emergency is defined as any situation related to a discharge or release
requiring immediate action to avoid an irreversible loss of natural resources

or to prevent or reduce any continuing danger to natural resources, or a situation
in which there is a similar need for emergency action (Hall et al, 1987).

3. A type A regulation does not require the detailed three-Step approach.
CHAPTER TWO

1. Ridker and Henning (1967) were the first to apply the hedonic property value
approach to estimate the value of air pollution. Since that time the

hedonic price method has been used to estimate the value of public safety (Clark
and Cosgrove, 1990), cultural amenities (Clark and Kahn, 1988), nuclear

power plants (Gamble and Downing, 1982; Nelson, 1981), and public parks
(Schroeder, 1982) among other things.

2. It is not necessary to assume linearity in the variables to derive marginal
implicit prices, though the prices will be derived in a manner different from
above if nonlinearity is assumed.

3. The pollution event terminology used by Mendelsohn is more accurately an
information event.

CHAPTER THREE

1. Roback (1988) extended her 1982 model by extending the assumption of
identical workers into two types of workers with different preferences. Her
results point out that the wages of on type of worker prove to be dependent on
the preferences of the other type.

2. By assuming identical workers Roback (1982) shows that the estimated wage
difference will be an underestimate of the true equalizing wage difference for
those with strong tastes for amenities and an overestimate for those With weak
preferences. However, She points out that estimates assuming identical workers
are kind of an average of the true gradients for the various type workers. Thus
the expected afl’ect on wages and rents are more diluted if nonhomogeneOus
work force is built into the model. In addition, Hoehn et al. (1987) has Shown
that by including aggregation economies into the model further dilutes the
expected effect on wages and rents from amenities.

101

 

“*1

 

102
CHAPTER FOUR

1. Within a household there is no need to distinguish i.

2. Instead of the average worker in a household the primary worker could have
been chosen, however, there are problems associated with using the primary
worker. In many cases, the household location decision is decided by the sum of
a couple’s income. To a much lesser extent, identifying the primary worker can
be a problem.

3. Due to inconsistent covariance matrix estimates, if a cross-equation restriction
is tested, it may lead to false results.

CHAPTERFIVE

1. Since the rent and wage equations are log-linear, the equations are linearized

as follows: In w = c + Bx implies that w = elm”! thus 3" _ a [4:4ij 652:

CHAPTER SIX
1. Yearly housing units was obtained from the 1980 Census of Housing.

2. Aggregate damages of Superfund sites are the sum of the incremental
damages. Incremental damages are estimated as:

ID - [12 0492 .97 (film-0282,1511?) 1-[0.75*2725 (01,,+202',,SITE)]

Where 5,, and 3,, are the rent coefficients and 6,, and 5,, are the wage
coefficients for SITE and STTESQ respectively. To annualize the damage
estimates the rent coefficients are multiplied by 12 (months per year) and the
wage coefficients are multiplied by 2725, the product of the sample means of
workers per household (1.61), mean hours per week worked (38.53), and the
mean weeks worked per year (43.92). Since the rent and wage equations were
estimated in log-linear form, the rent equation is multiplied by the average

rent paid per household (492.97) and the wage equation is multiplied by the
average wage received per household (8.75).

 

BIBLIOGRAPHY

BIBLIOGRAPHY

Adler, K.J., R.C. Anderson, Z. Cook, R. C. Dower, A. R. Ferguson, and MJ. Vickers.
-- ~ W - D V -

 

Estimator. Technical Report. Washington D. C. Public Interest EcOnonIies Center
Research Report.

Andelman, Julian B. and Dwight W. Underhill. 1987. W
W. Lewis Publishers, Inc. Chelsea, MI.

Bartik, Timothy J. and V. Kerry Smith. 1987. ”Urban Amenities and Public Policy," in

and 0 'cs. Volume II, Edited by ES. Mills.
Elsevier Science Publishers.

Blomquist, Glenn C., Mark C. Berger and John P. Hoehn. 1988. ”New Estimates of
Quality of Life In Urban Areas." W. 78. 89-107.

Butler, R.V. 1980. ”Cross-Sectional Variation in the Hedonic Relationship for Urban
Housing Markets.” MW. 20(4):439-453.

CERCLA. 1980. Public Law No. 96-510, 94 Stat. 2767.

Clark, David E. and James C. Cosgrove. 1990. "Hedonic Prices, Identification, and the
Demand for Public Safety.” Journal of Regional S_ciegce. 30(1):105-121.

Clark, David E. and James R. Kahn. 1988. 'The Social Benefits of Urban Cultural
Amenities.” Journg gf Regional gignge. 28:363-377.

Council on Economic Priorities (CEP). 1986. MW. Island
Press, Washington DC.

Department of the Interior, Natural Resource Damage Assessments, 43 CF R Part 2, April
29, 1991.

Dickie, Mark and Shelby Gerking. 1987. 'Interregional Wage Differentials: An
Equilibrium Perspective." MW. 27(4):571-585.

103

"x

 

 

104

Driscoll, Paul J. and Richard Boisvert.1991."Dual Second- and Third-Order Translog

Models of Production" WW
73(4). 1146-1160.

Environmental Protection Agency. 1990. WWW. EPA’S

review of progress and new directions in environmental protection. USEPA.
Paper no. 21k-2001.

Environmental Protection Agency 1989 W
W. U. S. E. P.A., Office of Emergency and Remedial Response,

Hazardous Site Control Division. EPA/540/8-89/011.

Evans, Alan W. 1990. "The Assumption of Equilibrium 1n the Analysis of Migration and
Interregional Differences: A Review of Some Recent Research.” W

W. 30: 515-531. November.

Fomby, Thomas B., R. Carter Hill, and Stanley R. Johnson. 1984. W
M. Springer-Verlay New York Inc.

Freeman. A- Myrick, III. 1979a. ECWWLM
Wig. Baltimore: The J ohnsHopkins University Press.

Freeman, A. Myrick, III. 1979b. “Hedonic Prices, Property Values and Measuring
Environmental Benefits: A Survey of the Issues.” mm
m. 81:154-173.

Gamble, Hays B. and Roger H. Downing. 1982. "Effects of Nuclear Power Plants on
Residential Property Value.” W. 22(4):457-478.

Graves, Philip E. and TA. Knapp. 1985. "Hedonic Analysis in a Spatial Context:
Theoretical Problems In Valuing Location-Specific Amenities. MM.
December. pp. 737-743.

Graves, Philip E. and Donald M. Waldman. 1991. 'Multimarket Amenity Compensation
and the Behavior of the Elderly." MW. 81(5):1374-1381.

Greenwood, Michael J., Gary L. Hunt, Dan S. Rickman, and George I. Treyz. 1991.
”Migration, Regional Equilibrium, and the Estimation of Compensating
DifferentiaIS" WW 81(5) 13324390-

Grisham. Joe W. 1986. W Persamon
Press Inc. Elmsford, New York.

Gyourko, Joseph and Joseph Tracy. 1989. "The Importance of Local Fiscal Conditions In

Analyzing Local Labor Markets " WM 97(5): 1203-
1231.

 

 

105

Hall, Ridgeway M. Jr., Tom Watson, Jeffrey J. Davidson, David R. Case and Nancy S.

Bryson. 1987. W. 2nd Edition. Government Institutes, Inc.
Rockville, Maryland.

Haness, Steven J. and John J. Warwick. 1991. ”Evaluating the Hazard Ranking System."
lemma 32: 165-176

Harrison, David Jr. and James H. Stock. 1984. W
W- Paper N 0- E-34-09-
Energy and Environmental Policy Center, John F. Kennedy School of

Government, Harvard University.

Henderson, J. Vernon. 1982. ”Evaluating Consumer Amenities and Interregional Welfare
Differences.“ WM. 11:32-59.

Hird, John A. 1990. ”Superfund Expenditures and Cleanup Priorities. Distributive Politics

or the Public Interest? WW 9(4) 455-
483.

Hoehn, John R, Mark C. Berger and Glenn C. Blomquist. 1987." A Hedonic Model of

Interregional Wages, Rents, and Amenity Values. W99.
27: 605-620.

Judge, George G. R. C. Hill, W. E. Griffiths, H. Lutkepohl, and TC Lee. 1988.
W John Wiley and Sons

Kmenta, Jan. 1986. W. 2nd edition. MacMillian Publishing
Company. New York, New York.

Kohlhase, Janet E. 1991. ”The Impact of Toxic Waste Sites on Housing Values.“ 19119131
91mm. 30:126-

Linneman, Peter. 1980. “Some Empirical Results on the Nature of the Hedonic Price
Function for the Urban Housing Market.” WW- 8:47-68.

McClelland, Gary H., William D. Schulze, and Brian Hurd. 1990. 'The Effect of Risk
Beliefs on Property Values: A Case Study of a Hazardous Waste Site." Risk

mm. 10(4):435.497.

Mendelsohn, Robert. 1987. "A Comparative Analysis of Three Methodologies for
Measuring Hazardous Waste Damages.” Working Paper. Yale School of Forestry
and Environmental Studies. New Haven, Connecticut.

Michaels, R. Gregory and V. Kerry Smith. 1990. "Market Segmentation and Valuing
Amenities with Hedonic Models: The Case of Hazardous Waste Sites.” W
W. 28:223-242.

Michigan Environmental Response Act. Public Act 307 of 1982 as amended.

 

“a

 

106

National Center for Policy Alternatives (NCPA). 1986. ' ° 0
Edited by David Jones and Jeffrey Tryens. Washington DC.

Nelson, John. 1978. ”Residential Choice, Hedonic Prices, and the Demand for Urban Air
Quality." MW. 5:357-369.

Nelson, Jon P. 1981. "Three Mile Island and Residential Property Values: Empirical
Analysis and Policy Implications." W. 57(3):363-372.

Nieves, L.A., R.C. Hemphill, and BE. Clark. 1991. “The Economic Impacts of Noxious
Facilities on Wages and Property Values: An Exploratory Analysis.“ Argonne
National Laboratory, Economics and Law Section, Environmental Assessment
and Information Sciences Division.

Ridker, Ronald and John A. Henning. 1967. 'The Determinants of Residential Property
Values with Special Reference to Air Pollution. WM

Statistics.
Roback, Jennifer. 1982. "Wages, Rents and the Quality of Life." W

m. 90:1257-1278.

Roback, Jennifer. 1988. "Wages, Rents, and Amenities: Differences Among Workers and
Regions." W. 26:23-39.

Rosen, Sherwin. 1979. "Wage-Based Indexes of Urban Quality of Life," in P. Mieszkowski
and M Straszheim (edS) WW Baltimore: John
Hopkins Press, pp. 74-104.

SARA. 1986. Public Law No. 99-499, 100 Stat. 1613.

Schroeder, Timothy D. 1982. "The Relationship of Local Public Park and Recreation

Services to Residential Property Values.” WW. Third
Quarter. pp. 223-234.

Smith, Adam. 1937. The Wealth gt Natigns. Edited by Edwin Cannan. Random House
Inc.

Smith, V. Kerry. 1985. “Supply Uncertainty, Option Price, and Indirect Benefit
Estimation." mm. 61(3):303-307. August.

Wolf, Sidney M. 1988. W. Quorom Books. Westport, Connecticut.

 

 

APPENDIX A

DERIVAT'ION OF WAGE AND RENT GRADIENTS

APPENDIX A

This appendix includes the derivation of the wage and rent gradients, equations
(9) and (10) in chapter three. The k subscript is left out in the derivations.

The first step is to totally difierentiate equations (6) and (7) from chapter three
and include on the right hand side the change in utility or costs with respect to
amenities. This results in the following:

(1) dew/da + Vrdr/da = -V.
(2) QdW/da + C,dr/da = -C.

Equations 1 thru 2 are then put in matrix form below:

”PM
V11 VT d6 40
{0,0,} dr =[-c,]
TE.

 

 

Set the first matrix equal to A, the second matrix equal to d and the third matrix equal
to b, where A'd=b. To solve for the wage and rent gradients, Shown in matrix (I, the
inverse of A is multiplied by b. That is A"*b=d.

The first step is to find the cofactor matrix of matrix A seen below:

0,. -c,,
-V, II"

The adjoint of the cofactor matrix is:

107

 

100
c, -v,
-CH Vi

The determinant of the matrix A, Det. A is:

 

Det.A=CV-CV>0 ,

TH WT

. -—r-y (h

Multiplying the adjoint of the cofactor matrix by 1 / detA results in the inverse of A, A".
So now the wage and rent gradients can be estimated by multiplying A’1 by matrix d. .
These result in the wage and rent gradients derived in chapter three. 4.

 

 

APPENDIX B

ANNUAL INCREMENT AL COUNTY DAMAGES

Appendix B

Table 8.1 - Annual Economic Damages of the Incremental Site
(1980 dollars)

 

 

COUNTY
HOUSEHOLD ANNUAL
ANNUAL DAMAGES
NUMBER OF DAMAGES STANDARD FROM THE
NUMBER HOUSING FROM ERROR FOR INCREMENTAL
OF NPL UNITS IN INCREMENTAL HOUSEHOLD SITE
Rank County State SITES THOUSANDS SITE DAMAGES (THOUSANDS)
- Sample Total - 499 32.186 - - 3,325,000
1 Cook IL 1 1,993 150.32 88.88 300,000
2 Wayne MI 1 875 150.32 88.88 131,000
3 San Diego CA 1 718 150.32 88.88 108,000
4 Orange CA 2 720 138.52 63.86 100,000
5 Philadelphia PA 2 685 138.52 63.86 95,000
6 Dallas TX 1 625 150.32 88.88 94,000
7 Allegheny PA 2 571 138.52 63.86 79,000
8 Alameda CA 3 444 126.71 44.03 56.000
9 Erie NY 2 387 138.52 63.86 54,000
10 Milwaukee HI 3 378 126.71 44.03 48.000
11 San Francisco CA 1 316 150.32 88.88 48,000
12 Uestchester NY 1 316 150.32 88.88 47.000
13 Hamilton OH 2 343 138.52 63.86 48.000
14 King HA 6 525 91.30 74.90 48.000
15 Maricopa AZ 7 600 79.49 100.99 48,000
16 Tarrant TX 2 338 138.52 63.86 47.000
17 Broward FL 6 477 91.30 74.90 44.000
18 Palm Beach FL 1 287 150.32 88.88 43,000
19 Oakland MI 4 372 114.91 38.48 43,000
20 San Bernadino CA 4 366 114.91 38.48 42,000
21 Jefferson KY 1 266 150.32 88.88 40,000
22 Marion IN 3 309 126.71 44.03 39.000
23 Jefferson AL 1 259 150.32 88.88 39.000
24 St. Louis MO 5 358 103.10 52.01 37,000
25 Shelby TN 3 286 126.71 44.03 36,000
26 Contra Costa CA 2 252 138.52 63.86 35,000
27 Sacramento CA 5 324 103.10 52.01 33.000

109

110

Table 8.1 - Annual Economic Damages of the Incremental Site, Continued
(1980 dollars)

 

 

COUNTY

HOUSEHOLD ANNUAL

ANNUAL DAMAGES

NUMBER OF DAMAGES STANDARD FROM THE

NUMBER HOUSING FROM ERROR FOR INCREMENTAL
OF NPL UNITS IN INCREMENTAL HOUSEHOLD SITE
Rank County State SITES THOUSANDS SITE DAMAGES (THOUSANDS)

28 Pima ' AZ 1 216 150.32 88.88 33.000
29 Essex NJ 5 317 103.10 52.01 33,000
30 Hudson NJ 2 221 138.52 63.86 31,000
31 Macomb MI 3 236 126.71 44.03 30.000
32 Hennepin MN 7 379 79.49 100.99 30,000
33 Denver CO 3 228 126.71 44.03 29.000
34 Delaware PA 2 201 138.52 63.86 28.000
35 Union NJ 1 183 150.32 88.88 27.000
36 Ventura CA 1 183 150.32 88.88 27.000
37 Montgomery OH 4 227 114.91 38.48 26.000
38 Orange FL 2 183 138.52 63.86 25,000
39 DuPage IL 5 235 103.10 52.01 24.000
40 Genesee MI 2 163 138.52 63.86 23.000
41 Ramsey MN 3 177 120.71 44.03 22.000
42 Camden NJ 3 174 126.71 44.03 22.000
43 Passaic NJ 2 158 138.52 63.86 22.000
44 Kern CA 2 154 138.52 63.86 21.000
45 Jefferson CO 1 137 150.32 88.88 21,000
46 Hillsborough FL 7 261 79.49 100.99 21,000
47 Luzerne PA 1 134 150.32 88.88 20,000
48 East Baton Rouge LA 1 134 150.32 88.88 20.000
49 Stark OH 2 143 138.52 63.86 20.000
50 Salt Lake UT 6 214 91.30 74.90 20,000
51 Snohomish HA 1 129 150.32 88.88 19.000
52 Polk FL 1 127 150.32 88.88 19,000

53 San Joaquin CA 2 136 138.52 63.86 19.000,
54 Sedgwick KS 3 146 126.71 44.03 18.000
55 Polk IA 1 122 150.32 88.88 18,000
56 Volusia FL 1 122 150.32 88.88 18.000
57 Lancaster PA 2 129 138.52 63.86 18.000
58 Duval FL 7 227 79.49 100.99 18,000

Table 8.1 - Annual Economic Damages of the Incremental Site, Continued

111

(1980 dollars)

 

 

COUNTY
HOUSEHOLD ANNUAL
ANNUAL DAMAGES
NUMBER OF DAMAGES STANDARD FROM THE

NUMBER HOUSING FROM ERROR FOR INCREMENTAL

OF NPL UNITS 1N INCREMENTAL HOUSEHOLD SITE

Rank County State SITES THOUSANDS SITE DAMAGES (THOUSANDS)
59 Anne Arundel MD 2 127 138.52 63.86 18.000
60 Albany NY 1 115 150.32 88.88 17.000
61 Arapahoe CO 1 113 150.32 88.88 17,000
62 Brevard FL 1 113 150.32 88.88 17.000
63 Allen IN 1 111 150.32 88.88 17.000
64 Mobile AL 3 131 126.71 44.03 17,000
65 Hamilton TN 1 110 150.32 88.88 17,000
66 Dane HI 3 126 126.71 44.03 16,000
67 Hake NC 2 113 138.52 63.86 16,000
68 Johnson KS 1 103 150.32 88.88 15.000
69 Fresno CA 7 192 79.49 100.99 15.000
70 Charleston SC 1 99 150.32 88.88 15.000
71 Bucks PA 6 165 91.30 74.90 15,000
72 Pierce HA 7 105 79.49 100.99 15,000
73 Lorain OH 1 96 150.32 88.88 14.000
74 Dauphin PA 1 95 150.32 88.88 14.000
75 Erie PA 2 102 138.52 63.86 14,000
76 Hill IL 3 109 126.71 44.03 14.000
77 Richmond City VA 1 91 150.32 88.88 14.000
78 Lake IL 6 148 91.30 74.90 14.000
79 Monterey CA 3 103 126.71 44.03 13.000
80 St. Louis MN 1 87 150.32 88.88 13.000
81 Oneida NY 2 94 138.52 63.86 13.000
82 Stanislaus CA 3 102 126.71 44.03 13.000
83 Butler OH 2 92 138.52 63.86 13,000
84 Ingham MI 3 99 126.71 44.03 13,000
85 Greenville SC 4 108 114.91 38.48 12,000
86 Santa Cruz CA 1 80 150.32 88.88 12.000
87 Richland SC 3 92 126.71 44.03 12.000
88 St. Joseph IN 3 91 126.71 44.03 12.000
89 Lancaster NB 1 76 150.32 88.88 11,000

112

Table 8.1 - Annual Economic Damages of the Incremental Site. Continued
(1980 dollars)

 

 

COUNTY

HOUSEHOLD ANNUAL

ANNUAL DAMAGES

NUMBER OF DAMAGES STANDARD FROM THE

NUMBER HOUSING FROM ERROR FOR INCREMENTAL
OF NPL UNITS IN INCREMENTAL HOUSEHOLD SITE
Rank County State SITES THOUSANDS SITE DAMAGES (THOUSANDS)

90 Adams CD 3 09 120.71 44.03 11.000
91 Cumberland NC 2 01 130.52 03.00. 11.000
92 Lackawanna PA 3 88 126.71 44.03 11,000
93 Rockland NY 2 00 130.52 03.00 11.000
94 Boulder c0 1 73 150.32 00.00 11.000
95 Berks PA 0 120 91.30 74.90 11.000
90 vaukeeha HI 4 91 114.91 30.40 11.000
97 Richmond GA 1 05 150.32 00.00 10.000
90 Linn IA 1 05 150.32 00.00 10.000
99 Hinnebago IL 5 93 103.10 52 01 10,000
100 Brown vi 1 02 150.32 00.00 9,000
101 Broome NY 4 01 114.91 30.40 9,000
102 Racine HI 1 02 150.32 00.00 9.000
103 Clark NA 3 73 120.71 44.03 9.000
104 Gaston NC 1 59 150.32 88.88 9.000
105 Galveston TX 4 77 114.91 38.48 9.000
100 Alachua FL 1 59 150.32 00.00 9.000
107 Berkshire MA 1 50 150.32 00.00 0.000
100 Yakima NA 3 05 120.71 44.03 0.000
109 Kalamazoo MI 5 79 103.10 52.01 0.000
110 Blair PA 1 52 150.32 00.00 0.000
111 Niagra NY 0 05 91.30 74.90 0.000
112 Dakota MN 4 07 114.91 30.40 0.000
113 Jefferson MO 1 50 150.32 00.00 0.000
114 Haber UT 2 50 130.52 03.00 7.000
115 Calhoun MI 3 54 120.71 44.03 7.000
110 Lycoming PA 1 45 150.32 00.00 7,000
117 Monroe MI 1 45 150.32 00.00 7.000
110 Chesterfield VA 2 49 130.52 03.00 7,000
119 Elkhart IN 3 52 120.71 44.03 7.000
120 Rock v1 3 51 120.71 44.03 7.000

 

113

Table 8.1 - Annual Economic Damages of the Incremental Site. Continued
(1980 dollars)

 

 

COUNTY
HOUSEHOLD ANNUAL
ANNUAL DAMAGES
NUMBER OF DAMAGES STANDARD FROM THE
NUMBER HOUSING FROM ERROR FOR INCREMENTAL
OF NPL UNITS IN INCREMENTAL HOUSEHOLD SITE
Rank County State SITES THOUSANDS SITE DAMAGES (THOUSANDS)
121 Lexington SC 3 51 126.71 44.03 7.000
122 Vigo IN 1 43 150.32 88.88 6.000
123 Yellowstone MT 1 43 150.32 88.88 6.000
124 Roanake City VA 1 43 150.32 88.88 6.000
125 Minnehaha SD 1 43 150.32 88.88 6.000
126 Calhoun AL 1 42 150.32 88.88 6.000
127 St. Charles MO 3 50 126.71 44.03 6.000
128 Harford MD 3 49 126.71 44.03 6.000
129 New Hanover NC 1 41 150.32 88.88 6.000
130 Aiken SC 1 40 150.32 88.88 6.000
131 Ector TX 2 43 138.52 63.86 6.000
132 Hoodbury IA 1 39 150.32 88.88 6.000
133 Kankakee IL 1 37 150.32 88.88 6.000
134 Cumberland NJ 4 47 114.91 38.48 5.000
135 Hashington MN 2 37 138.52 63.86 5.000
136 Stearns MN 2 34 138.52 63.86 5.000
137 Benton NA 5 43 103.10 52.01 4.000
138 Los Angeles CA 12 2.854 0.00 0 0
139 Santa Clara CA 22 474 0.00 0 0
140 Kent MI 10 163 0.00 0 0
141 Spokane HA 9 137 0.00 0 0
142 Nassau NY 12 ' 432 0.00 0 0
143 Monmouth NJ 8 181 0.00 0 0
144 Suffolk NY 8 406 0.00 0 0
145 Montgomery PA 15 232 0.00 0 D
146 Chester PA 9 110 0.00 0 0
147 Bergen NJ 9 307 0.00 O 0
148 Atlantic NJ 8 88 0.00 0 O
149 Burlington NJ 13 121 0.00 0 0
150 Harris TX 8 984 0.00 O O
151 Middlesex NJ 11 203 0.00 O O