cfik ..v.._._...,
«£3...
2 Km .

.

,fiwéam
3?. .:
munmxa, .f

3.
fl... 5
45......

t: s

 

 

 

 

 

 

‘0- It...‘

{flab};

.513! La.
.25..

 

 

«I I . V a . ‘ china. £5 ,
; . . . a ‘ . _ . V . .am . s.
V . . . Mum-wrfi. .r . . . . . . A gr. h. ~fllmfius‘0wn.’

,. . 1..anqu . V . . . . . . . . . . . a

35:22.!

 

 

’ m LIBRARY ‘“ ’l
. 62’ MIChlb . State
7.” 1 University

 

 

This is to certify that the
dissertation entitled

ESSAYS IN THE POLITICAL ECONOMY OF EMINENT DOMAIN
AND
EFFICIENT WATER RESOURCE MANAGEMENT

presented by

Dziwomu Kwami Adanu

has been accepted towards fulfillment
of the requirements for the

 

PHD degree in Agicultural Economics

 

W

/ Major Professor’s Signature

4Puc. Z56. 2.00?

Date

MSU is an Affirmative Action/Equal Opportunity Employer

-.-—o-o---.-.—---.-—--.-u-.--

o---—u----o--------o-.-—--— -

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5108 K:/Prolecc&Pres/CIRCIDateDue.indd

 

ESSAYS IN THE POLITICAL ECONOMY OF EMINENT DOMAIN AND
EFFICIENT WATER RESOURCE MANAGEMENT

By

Dziwomu Kwami Adanu

A DISSERTATION

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

DOCTOR OF PHILOSOPHY

Agricultural Economics

2009

 

 

Capyright by
DZIWORNU KWAMI ADANU
2009

ABSTRACT
ESSAYS IN THE POLITICAL ECONOMY OF EMINENT DOMAIN AND
EFFICIENT WATER RESOURCE MANAGEMENT
By
Dziwomu Kwami Adanu

The use of eminent domain power for economic development is an important
part of public policy in the US. Eminent domain is however a complicated policy with
divergent impacts on different segments of society. Two unresolved issues arising from
the use of eminent domain include first, how the perceived benefits and costs of
eminent domain affect people’s positions on the reform of eminent domain law. This is
addressed in the first essay of this dissertation by setting up and estimating a voting
model that explains voters’ decision on the reform of eminent domain and regulatory
compensation laws in the US. The second research issue involves the choice of owner
compensation levels that minimize the problem of holdouts and close the gap between
the theoretically proven effectiveness of eminent domain in resolving holdouts, and
observation of protracted eminent domain negotiations in practice. This is addressed in
the second essay using a two-period sequential game between property owners and local
governments. Finally, the third essay looks at the implications of functional form
choices for cost function estimations in the US water industry.

The first essay investigates voter responses to referenda in the 2006 midterm
elections on restricted use of eminent domain power, and regulatory takings
compensation. Results indicate that voters responded to these referenda on eminent

domain quite differently depending on whether the referenda included a requirement of

 

compensation for regulatory takings. A plurality of voters favored reforming eminent
domain law to limit its use for economic development purposes. Compensation for
regulatory takings was viewed less favorably. Combining these two issues on one ballot
therefore increases the proportion of voters rejecting the ballot on restricted use of
eminent domain. Further, county level socio-economic variables capturing the
perceived benefits and costs of eminent domain power were important for referendum
outcomes. Next, theoretical research findings by Miceli and Segerson (2007) indicate
that the threat of eminent domain resolves owner holdout problems in property takings.
Law and economics literature on eminent domain takings however abound of eminent
domain cases that end up in the courts because of disagreements between owners and
governments over compensation levels. The second essay reconciles the disparity
between theoretical predictions and actual observations about the effectiveness of
eminent domain in addressing owner holdouts. Using a two period sequential game
framework it is shown that the threat of eminent domain guarantees resolution of the
holdout problem only when owners have complete and perfect information about the
bargaining problem. These informational assumptions are later relaxed to model more
practical eminent domain bargaining problems. Finally, the third essay estimates total
variable cost functions for potable water facilities in the US. Cost functions are
parameterized using the Hyperbolically Adjusted Cobb-Douglas (HACD) and the
translog functional forms. The results show wide disparity in some of the estimated
efficiency parameters although the measure of fit is close for the two functional forms.
The results show the importance of using more than one functional form in cost

function estimations to allow for comparison and assessment of reliability of estimates.

 

 

To My Family
For Your Encouragement and Support

ACKNOWLEDGEMENTS

I would like to express my profound gratitude to my major Professor John P.
Hoehn for his support and guidance throughout my studies. I am thankful to Professor
Hoehn for his constructive criticisms and suggestions that have been immensely
valuable in improving upon my draft essays. I appreciate his patience and contributions
to my professional development especially in improving upon my professional writing
and thinking skills.

I would like to thank Professor Richard Horan for his contributions to this
dissertation. I am particularly grateful to him for suggesting the problem in essay 2 and
providing the necessary guidance to simplify, clarify, and to motivate the salient issues
discussed in this essay. I learned useful lessons about the importance of patience and
perseverance in developing good economic models while working with Professor Horan
on this second essay.

Professor Patricia Norris has been very helpful in providing me with guidance
on eminent domain and regulatory taking issues in the United States. She has helped
very much to improve and update my research literature by always remembering to
forward relevant literature and recent publications on the subject matter to me. Some of
these papers enabled me to reexamine my research results from other perspectives, and
to compare and contrast my findings with those from related studies.

I thank professor Emma Iglesias for helping with my econometric analysis. Her
suggestions on testing and model specifications issues helped address my model

specification problems. I particularly appreciate her prompt responses to all my

vi

“"7

 

 

requests. It is amazing how fast she responds to emails and reviews my research write-
ups.

I am grateful to Professor Runsheng Yin for agreeing to serve on the committee
even though my request was quite belated. I appreciate his contributions to my drafts.

I would like to express my appreciation to the faculty, staff, and students of the
Department of Agricultural Economics for their support throughout my time here. I
would especially like to thank Robert Myers, Eric Crawford, Scott Loveridge, and
Debbie Conway for various supports I received from them over the years. I thank Sarma
Aralas and Vandana Yadav for reading through drafts of my essays.

I would like to express my gratitude to my family for their unwavering love and
support. I am grateful to my parents for encouraging me every step of the way through
these very many years of education. I thank my siblings for their continued support and
encouragement.

Finally, I am very grateful for financial assistance provided by the Lincoln
Institute of Land Policy based in Cambridge, Massachusetts to support the completion

of the dissertation.

vii

TABLE OF CONTENTS

List of Tables ............................................................................................................ ix

List of Figures... .............................................................................................................. x

Essay 1: Voter Decisions on Eminent Domain and Regulatory Takings Referenda
1.1 Introduction. ................................................................................................. l
1.2 Conceptual Framework and Research Hypotheses ....................................... 5
1.3 Economic Model ...................................................................................... 18
1.4 Data .......................................................................................................... 29
1.5 Results .......................................................................................................... 32
1.6 Conclusions ............................................................................................... 41
References. . ...................................................................................................... 52

Essay 2: Information and Bargaining Breakdowns in Eminent Domain Takings

2.1 Introduction. . . ............................................................................................... 56
2.2 Bargaining under Complete and Perfect information .................................. 59
2.3 Bargaining Under Perfect but Incomplete Information ............................... 64
2.4 Summary and Conclusions ........................................................................... 74
References. . . ...................................................................................................... 8O

Essay 3: Cost Function Estimation in the Water Industry — Functional Forms and
Efficiency Measures

3.1 Introduction. . . ............................................................................................... 82
3.2 Theoretical Framework ................................................................................ 85
3.3 The Econometric Model ............................................................................... 89
3.4 Data.. ....................................................................................................... 98
3.5 Results .............................................................................. 101
3.6 Conclusmnle6
References. . . ..................................................................................................... l 13

viii

Table 1.1:
Table 1.2:

Table 1.3:

Table 1.4:
Table 1.5:
Table 1.6:
Table 1.7:

Table 3.1:
Table 3.2:
Table 3.3:
Table 3.4:

LIST OF TABLES

Summary of Results for all Eminent Domain Ballots in 2006 ............ 44
Estimated Logit Coefficients by Ballot Measure Type and Pooled Data
Sample .............................................................................. 45
Heckman Sample Selection Regression Results ............................... 46
Summary Statistic of Variables ................................................. 47
Estimated Logit, Odds, and Odds Elasticity Results ........................ 48
State-Level Odds of Yes Votes When Ballot Measure Type = 1 .......... 49
State Predicted Odds of Passing Eminent Domain Ballots by Ballot
Measure Type .............................................................. . 50 — 51
Summary Statistic of Main Variables ....................................... 109
Estimated TRANSLOG and HACD Model Results ....................... 110
Elasticity of Substitution and Input Elasticity Estimates ................. 1 ll
Economies of Scale Estimates ................................................ 112

ix

LIST OF FIGURES

Figure 2.1 Sequential Bargaining Under Complete and Perfect Information (When
Government Moves First) ...................................................... 78

Figure 2.2 Sequential Bargaining Under Complete and Perfect Information (When
Owner Moves First) .......................................................... 79

 

ESSAY 1
Voter Decisions on Eminent Domain and Regulatory Takings Referenda

1.1 Introduction

Eminent domain refers to the power of government to take private property for
public use without the owner's consent. Public use refers to purposes such as the
provision of public services like highways, public utilities, community centers, schools,
and other facilities that can be made available for use of the entire community (Merrill
1986). Court decisions have however gradually broadened the definition of public use
(Michigan, 1981, US. Supreme Court, 1954, US Supreme Court, 2005). By 1981, the
Michigan Supreme Court decided in favor of broadening public use to include takings
where public authorities condemn the properties of private owners and transferred
ownership to other private owners for purposes of economic development (Michigan
Supreme Court, 1981). Though the Michigan court later reversed itself on such indirect
public uses, other state courts made similar decisions to broaden the concept of public
use in their respective jurisdictions (Sandefirr 2006, Berliner 2003). Such decisions in
the Connecticut courts culminated in the US. Supreme Court’s June 23, 2005 decision
in the Kelo v. New London case (U .8 Supreme Court, 2005). In Kelo, the US. Supreme
Court endorsed the constitutionality of a broad concept of public use, ruling that, under
the US. Constitution, governments are permitted to use eminent domain to take
property and transfer its use to other private parties as long as there is a public benefit,
such as economic development (U .8 Supreme Court, 2005).

The Kelo case arose from the condemnation of 115 lots of private and

commercial properties by the City of New London in the Fort Trumbull neighborhood

 

area of New London. The owners of 15 of the 115 lots marked to be taken refused to
sell their properties citing violation of the fifth and fourteenth amendments of the US
constitution that govern the taking of private property for public use. In particular, the
plaintiffs in the case argued that taking private properties and transferring same to a
private developer to build new structures to increase the tax base of the city and
generate employment does not meet the “public use” requirement for the exercise of
eminent domain power. Led by the lead plaintiff in the case, Sussette Kelo, owners of
the 15 lots under contest argued and lost the case in the New London Superior Court,
and the Supreme Court of Connecticut before taking the case to the US Supreme court.
Closely related to eminent domain is regulatory taking. Regulatory taking refers
to the use of government police powers to limit land development by private owners
without depriving them of ownership rights over the property (Flick et. a1. 1995). For
instance to preserve open space or protect ecologically sensitive zones governments
may limit the percentage of a landowner’s property that can be developed. Although
eminent domain and regulatory taking are related in the sense that both institutional
mechanisms are used to provide public services they represent two different policy
tools. Eminent domain taking involves forceful transfer of property rights and requires
payment of compensation while property owners facing regulatory action retain
ownership rights over their properties and are entitled to no compensation (Flick et. a1.
1995, Goldstein and Watson 1997). Efforts to make compensation for regulatory
takings a legal requirement began in 1995 when the 104th US. Congress passed a
property rights bill calling for compensation to property owners whenever federal

agency regulatory actions decrease property values by more than 20%. The bill however

failed to pass the Senate (Goldstein and Watson 1997). This led to efforts at the state
level in November, 2006 to pass legislations that would require compensation for
regulatory takings.

Following the Kelo ruling several states passed referenda to ban the use of
eminent domain for economic development purposes or restrict the circumstances under
which such takings should be carried out (Orthner 2007, Sandefur 2006, Berliner 2003).
At the end of November 2006 the ten states included in this study (Arizona, Florida,
Georgia, Michigan, Oregon, South Carolina, Louisiana, New Hampshire, Idaho and
California) had presented special ballots on reforming eminent domain and regulatory
taking compensation laws to registered voters with all of them except Idaho and
California disapproving of unfettered use of eminent domain to take over private
property (see Table 1.1). In general, two main classes of ballot measure types are
identifiable from this data: eminent domain only ballots, and eminent domain and
regulatory taking compensation ballots. States with eminent domain only ballots
generally call for a ban or restricted use of eminent domain power for economic
development purposes while eminent domain and regulatory taking compensation
ballots combine restricted use of eminent domain power with compensation for
regulatory takings. The differences in type of ballot proposals also imply that data on
eminent domain and regulatory taking compensation election results cannot be pooled
across states for comprehensive empirical studies without appropriate adjustments to
account for differences in the type of proposition voters responded to in each state.

This paper analyzes the political response to the Kelo case by examining the

factors influencing the decisions of voters to support or reject initiatives on these

measures in Ten US. States. The paper focuses on the effect of ballot structure on vote
outcomes involving restricted use of eminent domain and regulatory taking
compensation. Voter preference over these two issues is explained using a rational voter
model (Deacon and Shapiro 1975, Downs 1957, Hess and Orphanides 1995). The
rational voter model explains how voter decisions at the polls depended on the
perceived net benefits expected from the vote choices. A cross-sectional limited
dependent variable model is estimated using a logistic regression functional form to
explain the vote outcomes.

The results indicate that the average voter supports imposing restrictions on use
of eminent domain power but opposes requiring compensation for regulatory takings.
Combining these two issues on one ballot increases the proportion of voters rejecting
the ballot relative to presenting a ballot on restricted use of eminent domain only. On
average, voters in economically weak counties are less supportive of restricted use of
eminent domain power and regulatory taking compensation. In particular, counties with
low income and/or high unemployment rates are less supportive of restricting the use of
eminent domain power and requiring regulatory taking compensation. Homeownership
rate fails to significantly explain the vote outcomes. This implies that renters reject
unrestricted use of eminent domain just as strongly as homeowners do. Finally,
education and income have a negative effect on increased property rights protection and
regulatory compensation. This finding indicates that when confronted with a choice
between more secure property rights and a healthier environment both educated and

high income voters lean towards protecting the environment.

The remainder of the paper is ordered as follows. The next section presents and
discusses the conceptual framework and research hypotheses of the paper. This is
followed by the economic model section which discusses the supporting theoretical and
econometric models of the paper. Discussion of the research data, results, and

conclusions then follow in that order.

1.2 Conceptual Framework and Research Hypotheses

Voting on referenda and ballot propositions can be considered as voter
preference revelation over the issues being voted upon. The analysis of vote outcomes
on eminent domain and regulatory taking compensation in this study is therefore treated
as one of revealing the demand for these two institutional mechanisms. This section of
the paper begins by outlining the conceptual framework of private demand for these two
institutions. The conceptual framework explains the relationship between the expected
benefits and costs fi'om voting (voter utility) and the ultimate voter decision made at the
polls. The conceptual framework is followed by the statement and description of the

research hypotheses to be tested.

ConceptualfiFramework

The rational voters model suggests that voters’ decisions on public good
provision can be treated as a derived demand of how much public good voters want to
consume at the optimum [Downs (1957), Deacon and Shapiro (1975), Matsusaka
(1993)]. This implies that voters make voting decisions on the provision of public goods
to maximize utility derived from the consumption of private and public goods subject to

an income constraint. This conceptual framework describes the preferences and

perceptions of benefits (direct benefits and ideological satisfaction), costs, and the
income constraint of voters facing propositions on eminent domain and regulatory
takings. The level of these benefits, costs, and constraints are then related to the model
variables to explain the motivation for including these variables in the model. The
underlying point of this analysis therefore is that the observable variables in the model
(ballot measure type, homeownership, income, education, unemployment, and
population density) affect the perceived benefits, costs, and income constraint of voters.
These variables can thus be used to develop testable hypotheses to explain the observed
vote outcomes.

The ideological positions of people on Kelo (property takings for development)
can be described as a continuum of views ranging fi'om outright rejection to wholesale
acceptance of government intervention in property markets to take properties for
economic development purposes. For instance, the November 2005 survey results by
the Saint Index polling organization [reproduced in Somin (2007)] indicate that the
position of respondents on Kelo range fi'om “agreement” to “strong disagreement”. This
background to ideological positions implies that there are voters on either side of the
property takings issue. For simplicity, the analysis here is restricted to two categories of
voters, voters supporting or opposing restricted use of eminent domain and regulatory
takings compensation. The proportional distribution of voters holding these two views
in a voting population would therefore be important in determining the likelihood of
passing propositions on these issues.

In addition to ideological satisfaction voters can expect direct benefits fiom their

vote choices (Sandefur 2006, Lazzarotti 1999). The direct benefits expected from voting

on restricted use of eminent domain power and regulatory taking compensation ballots
include the value at risk (e. g. home values) that voters seek to protect (Sandefur 2006,
Riddiough 1997) public goods (e.g. roads, and community centers) provided fiom
takings (Lazzarotti 1999, Munch 1976), direct transfers (e. g. regulatory taking
compensation) to landowners as a result of government regulatory action (Miceli and
Segerson 1994), and nonmarket values (e.g. open space) provided by regulatory actions
(Bengston et a1 2004).

On the other hand, there are costs attributable to vote decisions on these issues.
Such costs often take the form of higher tax obligations that can be expected to emanate
from some of these decisions (Deacon and Shapiro, 1975). For instance, in order to pay
the increased compensation for eminent domain and regulatory takings when the
average voter supports a ballot on unrestricted use of eminent domain, and a
requirement for regulatory taking compensation, voters may have to pay increased taxes
to raise the necessary revenue to provide compensation. The increase in tax obligation
reduces the disposable income of voters and changes the income constraint of their
utility maximization problem.

Explanatory variables included in the study control for differences in the ballot
measure types presented to voters and the probable incentives and disincentives
associated with vote decisions at the polls. For instance, the ballot type variable is
binary and is defined to equal 0 if the ballot question in a given state calls for restricted
use of eminent domain only and 1 if a requirement for regulatory taking compensation
is added to restricted use of eminent domain. This variable measures the effect of the

ballot question structure on voter choices and allows the model to capture the extent to

which the nature of the ballot question affects the chances of passing eminent domain
measures.

The next explanatory variable considered in the model is homeownership rate.
Homeowners can be expected to be more concerned about use of eminent domain
power and property regulatory actions than voters living in rented properties. This is
because homeowners have more value at risk than renters. The implication here is that
counties with high homeownership rates may be more supportive of the ballot measure
since their net benefits from voting yes to restricted use of eminent domain and
regulatory compensation exceed that for renters. Here, the difference in the expected
response of the two subgroups (homeowners and renters) of voters is influenced
substantially by the asymmetric expected effect of the ballot measure on these groups.

There are however equally relevant reasons to expect the average homeowner to
vote no as well. For instance, given that a common rationale for property takings for
economic development is to combat blight (Sandefur, 2006) the property price increases
that may be expected to come with neighborhood improvements associated with the use
of eminent domain to clean blight provides good reason for a class of property owners
to vote no to restricting use of eminent domain. The resultant effect of homeownership
rate may therefore be ambiguous.

The positive relationship between environmental quality and income has been
reported in several studies on vote behavior and environmental and resource
conservation [Deacon and Shapiro (1975), and Kotchen and Powers (2006), Khan and
Matsusaka (1997), Popp (2001)]. This implies that high income voters may vote in

support of regulatory compensation because of their relatively higher demand for

environmental quality and open space in urban and congested areas. If this finding holds
true in this study as well then it can be expected that high income voters would reject
regulatory taking compensation to promote the use of regulatory takings.

Past studies on factors affecting attitude towards the environment and natural
resource use consistently show that the level of education of voters positively affects
voters’ attitudes towards resource management [see, Deacon and Shapiro (1975), Khan
and Matsusaka (1997), Khan (2002), and Fischel (1979)]. This is because knowledge
about the value of environmental quality and open space, how these can be improved,
and exposure to research findings on the impact of environmental quality and open
space on property values and human health are important determinants of voters’
position on the environment. Education is therefore one factor that can affect the
ideological position and the subsequent choices of voters on natural resource-related
ballot measures. These findings can be extended to eminent domain and regulatory
compensation issues since eminent domain takings involve land resource use decisions
while government regulatory actions on land use often have implications for
environmental and ecological resource management.

Counties with high unemployment rates can be expected to be supportive of
eminent domain since use of eminent domain power for economic development
purposes can be valuable for economically depressed areas that are looking forward to
economic expansion and job creation [Clarke and Kornberg (1994), Bowler and
Donovan (1994), and Sandefur (2006)]. On the other hand, given that regulatory taking
does not involve any subsequent use of the property to provide jobs or any collective

good, unemployment rate may not have a significant effect on how voters react to

regulatory compensation ballots. This implies that high unemployment rate can be
expected to increase the proportion of no votes cast on restricted use of eminent domain
and regulatory taking compensation.

Population density is another variable that can be linked to the potential direct
benefits of eminent domain and regulatory takings. Limited land availability and high
land prices in high population density areas often imply that some public services may
only be provided by taking some existing properties and converting them to alternative
uses. For instance, single family homes at good locations may be taken and converted to
multi-story apartment complexes to serve more people and increase property tax
revenues. Lanza (2006) found that population density does not explain eminent domain
takings. However, Lanza’s study and this paper examine eminent domain from different
perspectives (actual eminent domain takings in Lanza (2006) as opposed to preference
for restricted use of eminent domain in this study). Further, the ballot question here does
not cover only eminent domain but regulatory takings as well; thus it is useful to still
consider the role of population density in explaining voter decisions here.

The next section presents and discusses the hypotheses to be tested. Because
each of these explanatory variables may affect the perceived benefits and costs of voters
in several complicated ways, building testable hypotheses based on these variables
requires explaining why some effects may be more influential than others. Ultimately
the data must be relied upon to verify these hypotheses and reveal the net effect of each

of these variables on vote choices.

10

Research Hypotheses

The hypotheses are founded on discussions in the conceptual framework and
results from the rational voter model. As previously discussed, support for eminent
domain and regulatory taking ballot measures varies across space, economic, and
demographic characteristics of voters. Statewide voting initiatives provide an avenue to
study how these characteristics affect support for resource-use ballot measures at the

state and sub- state levels. The research hypotheses follow.

Hypothesis 1: Support for the ballot measure declines as the ballot measure extends
fi'om restricted use of eminent domain to restricted use of eminent domain and
regulatory taking compensation

Summary results on eminent domain and regulatory taking ballots in the 2006
midterm election (see Table 1.1) suggest that voter support may be declining as the
ballot measure extends from restricted use of eminent domain to restricted use of
eminent domain and regulatory takings compensation. This is likely the case because
voters supporting restricted use of eminent domain power reject regulatory takings
compensation since regulatory takings do not really result in the loss of property rights
over the property in question.

Of course if this relationship turns out to be positive instead, then the assertion
that adding a requirement for regulatory takings compensation makes it less likely for a
restricted use of eminent domain ballot to pass is untrue. This result would be
suggestive of two things: that voters supporting restricted use of eminent domain power

also tend to support compensation for regulatory takings, and voters who are not

11

supportive of restricted use of eminent domain power tend to support compensation for
regulatory takings strongly enough to vote yes instead of no given that their decision on

these two issues conflict.

Hypothesis 2: Support for the ballot measure is increasing in homeownership rate

As previously discussed, property owners concerned about price declines that
may be associated with uncertainties introduced by property takings and the small
chance that their properties might be expropriated may be reluctant to support increased
property takings. This can be expected to result in a positive relationship between
homeownership rate and yes votes at the polls. On the other hand, if indeed eminent
domain takings for economic development purposes affect low-valued properties
disproportionately as a measure to deal with blight (Somin 2007, Sandefur 2006) then
property price increases that are expected to come with neighborhood improvements
provide good reason for a class of property owners to vote against restricted use of
eminent domain power. The observed sign on the coefficient for this variable would
therefore depend on the net effect of these two main influences. The resultant effect of
homeownership rate may therefore be ambiguous. Assuming that the incentive to
property owners to protect their property investments overrides any positive external
price effects obtainable from cleaning up blighted properties implies that
homeownership rate can be expected to have a positive net impact on the proportion of

yes VOICS.

12

Hypothesis 3: Support for the ballot measure is decreasing in level of knowledge/
education

Previous studies have consistently observed a strong correlation between
educational attainment and support for environment and resource management
measures [Press (2003), Salka (2001), Deacon and Shapiro (1975), and Kotchen and
Powers (2006), and Palfrey and Poole (1987)]. Education is therefore one factor that
can affect the ideological position and the subsequent decision of voters on natural
resource regulation and use. Looking at how these initiatives are written out on voting
ballots, it is clear that a fair level of education is necessary to understand the ballot
initiatives and be an informed voter. It can therefore be expected that the higher the
proportion of voters in a county with at least high school diploma the higher would be
the proportion of voters rejecting the ballot. Similarly, the higher the proportion of
voters in a county with at least a bachelor degree the higher would be the proportion of
voters rejecting the ballot.

If the results unexpectedly show that more educated voters are more inclined to
vote yes for more restricted use of eminent domain power and regulatory takings
compensation then that reveals an interesting and debatable result. It implies that more
educated voters tend to choose more secure property rights over possible environmental
quality gains from the use of eminent domain power and regulatory takings. Writing on
the social and ideological bases of support for environmental legislation Calvert (1979)
observed that “relatively high levels of support were found among college-educated,
white-collar professionals”. Thus what a counter finding under this hypothesis would be

indicative of is the relative importance of secure property rights and environmental

13

quality to educated voters. In particular, it would indicate that when confronted with a
choice between more secure property rights and a healthier environment educated

voters would lean towards securing property rights.

Hypothesis 4: Support for the ballot measure is decreasing in income

Given that the dependent variable in the model is made up of two main forms of
takings (regulatory takings and eminent domain), the decision of voters at the polls can
be expected to be motivated by two main factors: level of support for use of regulatory
action to preserve green space and protect ecologically sensitive zones, and level of
support for use of eminent domain for economic development and alternative public
uses. Several research findings on the environment have shown environmental and
resource conservation to be a normal good. The functional form of the relationship may
be specified in several ways. However, the most commonly studied form of this
income-environment relationship is that of the Environmental Kuznets Curve which
expresses an inverted-U relationship between income and environmental attributes.
[Dasgupta (2002), Grossman (1993), Harbaugh (2002), Kahn and Matsusaka (1997)]. If
high income is associated with support for the environment then higher income counties
can be expected to vote ‘no’ to requiring compensation for regulatory takings since this
limits the use of regulatory action. When it comes to use of eminent domain power, the
requirement to pay compensation for expropriated properties in itself draws local
government authorities to low-valued properties to reduce the outlay involved in paying
compensation. There is therefore good reason to expect support for a ballot measure

imposing additional restrictions on eminent domain takings to be declining in income.

14

In other words, higher income counties are again more likely to vote ‘no’ to limiting the
use of eminent domain power.

A counter finding of a positive relationship between income and yes votes
would not only invalidate the Kuznets curve relationship which suggests that high
income earners appreciate enviromnental resources better but also indicate that eminent
domain taking is really not a problem that is specific to low-valued property owners
only. A counter finding here would not be surprising given that the Kuznet relation is
still not a well established relationship. Several authors have found evidence to suggest
that this inverted U-shape relationship is highly unstable and fails to show up in several
studies that investigated this relationship. See Hettige et al. (1992), List and Gallet

(1999), Harbaugh, Levinson,and Wilson (2002), and Millimet, List, and Stengos (2003).

Hypothesis 5: Support for the ballot measure is decreasing in population density
Higher population density settlements generally have higher demand for urban
services like open and green spaces, housing, shopping centers, and car parking spaces.
Limited land availability and high land prices in high population density areas often
imply that some of these services can only be provided by taking some existing
properties and converting them to alternative uses. More densely populated counties are
thus expected to show more support for policy initiatives like eminent domain that
promises the provision of these much needed services. This effect is often reflected in a
strong positive relationship between urban communities and approval for resource-use
initiatives [Meddler and Mushkatel (1969)]. Lanza (2006) related population density

and eminent domain takings along the same line by noting that “to the extent eminent

15

domain helps solve the holdout problem, the incidence of taking should depend on
population density. As an area becomes more densely settled and ownership patterns
more fractured, bargaining is likely to grow more complex. If takings reduce transaction
costs, they ought to vary positively with population density”. On the other hand, voters
in high population density areas may react differently when it comes to requiring
compensation for regulatory takings. Since properties in urban areas tend to be much
more expensive than comparable properties in rural or low population density areas,
voters in high population density counties may be more inclined towards voting yes to
require compensation for regulatory takings. In summary, voters in counties with high
population density are likely to vote no on eminent domain but vote yes on regulatory
takings compensation. This implies that the impact of population density on the
dependent variable (logodds of yes votes) should depend on the ballot measure type
variable. This is captured by interacting ballot type and population density variables.

If population density turns out to vary positively with yes votes then this may be
evidence that voters place more weight on regulatory takings compensation than on
restricted use of eminent domain. If the reverse result is observed, then that may suggest

that voters place more weight on eminent domain than regulatory actions.

Hypothesis 6: Support for the ballot measure is decreasing in the level of
unemployment rate

Eminent domain would likely be a valuable tool for more economically
depressed areas that are looking forward to economic expansion and job creation than

otherwise. Some previous studies on the effect of economic conditions on vote

16

outcomes indicate that voter dissatisfaction with bad economic conditions tend to erode
support for ballot proposals because of low support for government [Clarke and
Kornberg (1994), Bowler and Donovan (1994)]. Since a common measure of economic
strength is the level of unemployment, it is expected that voters in high unemployment
regions would show more support for eminent domain than those in high grth areas.
Given that regulatory taking does not involve any subsequent use of the property to
provide jobs unemployment is not expected to have any significant effect on how voters
react to regulatory taking ballots.

If high unemployment rate induces yes votes instead of the expected no votes,
then the model may very well be picking up the possibility that voters are simply voting
their values of ensuring that appropriate compensation is paid to property owners for all
regulatory actions by government. An interaction term of unemployment and ballot

measure type should pick this effect up in the model.

Hypothesis 7: Support for the ballot measure is decreasing in voter turnout
Previous empirical studies indicate that low voter turnout correlates strongly
with approval of initiatives in referenda [Knox, Landry, and Payne (1984), Hadwiger
(1992), Stone (1965)]. As turnout rises the proportion of favorable votes decline. One
explanation offered for this result is that qualified voters who oppose ballot propositions
tend to express their protest by boycotting elections (Stone 1965). Hadwiger on the
other hand noted that this could be because of voting mistakes by voters that do not

realize that a ‘no’ vote in a referendum is a vote for reform and may be mistakenly

l7

voting no to show support for the status quo. As noted by Hadwiger, this is a result that

still requires further research to explain the rationale for the finding.

1.3 Economic Model

This section of the paper outlines the econometric model used to obtain the
estimated model parameters. The econometric model specification is prefaced by a brief
description of the voting behavior model upon which the econometric model is founded.
This voting behavior model is based largely on the individual voter preference
maximization model developed by Deacon and Shapiro (1975) to describe how self-
interest maximization may be integrated into voter decision-making to explain vote
outcomes.

The model begins by assuming a differentiable average voter utility function for

county i as,
U’=U’(X’,qk,h') (1)
wherex is a vector of private goods, q a vector of collective goods, and 11 represents

demographic and socio-economic variables (homeownership rate, education, and

population density) that characterize the voter. The set of policy alternatives available to
the voter at the polls is represented by k where k = [0,1] Here, k = 0 represents a no
vote and k = 1 represents a yes vote. The collective goods available to the voter
therefore depend on the choice made at the polls. Equation (1) thus indicates that a

voter’s utility is not only affected by the vector of private goods x and collective goods

q consumed but also by the listed set of demographic and socio-economic variables

y of the voter.

18

The tax liability faced by voters on the other hand is the expected tax funds

needed to compensate private property owners for takings and to invest in new public

. . i . .
infrastructure. After accounting for taxes, S k the disposable income of the consumer

is expended on a vector of private goods xl yielding the budget constraint,
1' i i
p k x = I k — S k (2)

. . . i . .
where p k rs a vector of private good prices and I k represents money income. Grven
that the budget constraint is satisfied, the indirect utility fimction for this problem is

written as,

maxUl(xl,qk,hl)=Vl(qk,Pk,[lk*Slkahl) (3)

x
Equation (3) gives the maximum utility obtainable by the voter for any given policy
choice made at the polls.

Thus the indirect utility outcomes under a yes (1 )/no (0) voting alternative are,
Vi(q0,p0,1i0 —Si0,hi)=VOi fornovotes (4)
Vi(ql,pl,[i1-Si1,hi)=Vjiforyesvotes (5)
Given that voters cast ballots to maximize their self-interest, the average voter compares
(4) and (5) and votes yes if Vli > Voi , ‘no’ if Voi > Vli and abstain if Vli = Voi.

Thus an average voter in county 1' votes yes if the indirect utility under this outcome is

perceptively greater than that of the alternative option. To simplify the arguments of

the model further, let 2 represent variables related to collective goods (q) , private good

19

prices (p) , and disposable income (I — S) . Then the vote decision result described in

equations (4) and (5) can be re-written in terms of differences in potential utility as,
Vli‘VOi =AVi(z’,h‘) (6)
The new decision rule thus becomes vote yes if AV1 > 0 , no if A V1 < 0 , and abstain

if A Vi = 0 . Thus far, both the functional form of the indirect utility and its arguments
vary across counties. However, a more realistic way to control for differences in
preferences across counties is to let such variations be explained by the arguments of
the utility function only. The difference in indirect utility function is therefore re-written

as,

AVi(zi,hi)=AV(zi,hi) (7)
It is assumed that AV is a random variable with a known distribution. Letting

E and h denote the mean of the vector of variables in Z and h respectively, the mean

and variance of AV may be represented by ,U(z, h) and 0'2 respectively. Given a

 

general distribution of AV for county i as, g <AV Z, h) , the proportion of yes votes

(Y) in a county can be written as,

P<Y|Ejli> = jg<AV

—CXD

 

2,2) d<AV> =A(flo +52 Nah) * (8)

where 5 refers to all real numbers, and A the distribution imposed on the variables in the

Z, h> . To restrict

 

model. The probability of voting no on an initiative is thus, 1 — P<Y

2O

the estimated proportion of yes votes to values that fall strictly between zero and one,
0 < A(S) < 1 the logistic function is used to describe the distribution of the vote data.

The observed vote data (proportion of yes votes per county, F ) to be modeled
represents an estimate of a true (population) vote outcome, 7! = /\(,60 + ,3: Z + ,th).
The model to be estimated can therefore be specified as,

F=A(,Bo+flzz+,6hh)+e=7r+e (9)

Where F = Number of Yes votes

 

, A is the logistic function, whilez and 11
number of valid votes

represent vectors of independent variables that explain the vote outcomes. Equation (9)
is however clearly nonlinear in parameters and requires nonlinear least squares
estimation to obtain the estimated parameters. The model can however be linearized by

taking the inverse of the logistic function to obtain,

A‘1(F)=po+pzz+phh+u, (10)

This transformation allows for the application of linear regression since equation (10) is

now linear in the parameters. Thus given thatF follows the logistic

exp (a) + a. 2+ an
1+6Xp(flo +flzz+flhhr

distribution, F —

 

Expressing the dependent variable in logodds form makes the right hand side of the

model linear in parameters and amenable to modeling by least squares estimation:

 

F
In[l Fj=fl0+flzz+flhh+u (11)

21

 

F
where (1_ F] represents the odds of voting yes. The dependent variable in equation
(11) is therefore the logodds of voting yes in the polls.

Imposing a logistic distribution implies the error term in equation (11) has zero

mean and variance given as,

1
nMFm—MF»

 

var(u) =

Obviously, this estimated variance depends on the observed proportions indicating that
the model is Heteroskedastic and requires the application of Weighted Least Squares
(WLS) estimation approach. The weight (w) applied is proportional to the inverse of the

estimated variance [thus each observation is weighted by [n/AHF )(1 — 0(F)]1/2 ].

Since the weights are functions of unknown parameters the estimation requires a two-
step procedure that uses simple least squares in the first stage to estimate the weights
that are applied to the WLS regression at the second stage. The application of WLS to
this model yields what is called the Minimum Chi-Square Estimator (MCSE) of ,3 (see
Greene 2003, Grizzle et al. 1969).

The variables in the model follow from the results in equation (6).
Homeownership rate, education, and population density represent demographic and
socio-economic variables (h) that measure variations in background and preference
patterns. These variables capture the change in perceived benefits from a given vote
choice across counties that is attributable to background characteristics of voters.

Unemployment rate captures the potential impact property taking is expected to have on

job creation and economic expansion across counties. The income variable corresponds

22

to the initial level of income at the time the votes were taken. The level of collective

goods (q) can be particularly difficult to measure empirically. Here, state dummy

variables are used to capture this effect. Following Deacon and Shapiro 2005, private

good price levels [9 are assumed to be uniform across counties and subsumed in the

constant term of the model.

Summary results on eminent domain and regulatory taking ballots in the 2006
midterm election (see Table 1.1) suggest that voter support may be declining as the
ballot measure extends from restricted use of eminent domain to restricted use of
eminent domain and requirements for regulatory takings compensation. This is likely
the case because voters supporting restricted use of eminent domain power reject
regulatory takings compensation since regulatory takings do not really result in physical
loss of properties. Besides, unlike eminent domain, determining the compensable value
for properties under regulatory takings can be difficult and imprecise. Paying more
taxes to make compensation for such regulatory actions possible may therefore be
unacceptable to some voters. The structure of the ballot question presented to voters can
therefore be the difference between passing and not passing eminent domain ballots.

Since the sample data being analyzed here may be classified by the ballottype
variable, two main options exist for estimating this cross-sectional regression model.
First, two separate regression models may be estimated per ballot measure type if the
model parameters are statistically different in the two models. The models for
subsamples one (covering eminent domain and regulatory takings compensation), and

two (covering eminent domain only) can be specified respectively as,

23

 

F
In[ [1:]=al+¢1home 1+nlhighschool |+Klbachelor 1+p1incomel+

to] unemp 1 + y/ 1 density 1+VI turnout 1+u I (12)

and

 

F

111(1 3:. J = a2 + (15 2 home2+ 772 highschool 2 + K 2 bachelor 2+ p 2 income 2
— 2

+ a) 2 unemp 2+ W2 density 2+ v2 turnout 2+ u2 (l 3)

where F represents the proportion of yes votes for county i, Home represents
homeownership rate which refers to the percentage of occupied housing units that is
owner-occupied. Highschool is percentage of population 25 and above that has at least
high school certificate, and bachelor is percentage of population 25 and above with at
least a bachelor degree. Income refers to median income, unemp to county
unemployment rate in 2006, while density is population density.

Alternatively, a pooled regression model that combines data across the two
ballot measures may be estimated. Because the higher sample size in the pooled
regression model enables more precise estimation of the model parameters, pooling the
data is certainly the better option if the model parameters are stable across the two
ballot types. The econometric model for the cross-sectional pooled regression is

specified as,

 

In(1 FF] 2 a + ,6 ballottype+ ¢ home+ 7] highschool+ K bachelor + p income

+a) unemp + t// density + v turnout + u (14)

The ballot measure type variable is a binary variable that equals 1 for state ballot

measures covering eminent domain and regulatory takings, and O for ballot measures

24

that cover eminent domain only. The choice between estimating separate regression
models for each data subsarnple as in equations (12) and (13), and estimating a pooled
regression model (equation 14) is made based on a test of parameter stability across the
two data subsamples using the Chow test (Chow 1960). In particular, the Chow test is
used to test the null hypothesis that parameter estimates in the regression model are
equal across the two data groups. The null hypothesis for the Chow test can therefore be

stated as,

a1=012a ¢1 =¢2,771=772aK1=K2aP1 =p2,601 =602M1=wzaand V1=V2
A rejection of this null hypothesis indicates that the estimated parameters vary
significantly across the two data groups. This calls for either estimating two separate
models for the two data subsamples or accounting for variation of the estimated

parameters across the two data groups using interaction terms if a pooled data is used.
The Chow test is based on the restricted and unrestricted sum of squared
residuals from separate regressions on the two data sub-samples, and the pooled data

respectively. The test statistic is written as,

_ [SSRp -(SSR1 +SSR2)] .[n—2(k+1)]
ss121+ss1t2 k+1

 

where n is the total number of observations in the pooled regression model, SSRP is the

sum of squared residual from regression on the pooled data, while SSR, and SSR2

refer to the sum of squared residuals from regressions on the two separate sub-samples.

The number of parameters (excluding the constant term) estimated in each regression is

25

represented by k. Table 1.2 below shows the estimated parameters for the three (two

separate sub samples and the pooled data sample) data samples.

Comparing the estimated parameters in Table 1.2 across the data samples by
inspection to determine if they are close enough to warrant the use a pooled data sample
for the analysis (without adjusting for differences across ballot measure type) would
yield unreliable results. Using the Chow test approach, the F-statistic for the test is

given as,

F _ [84.467 —(14.238 + 26553)] . [173]
14.238+26.553 8

=23.157

 

The returned F-statistic value of 23.157 can now be compared with a critical
value from the F -distribution table to make a decision on the hypotheses being tested.
Given the critical value at the 5% level of significance as 1.94, the null hypotheses that
the estimated parameters are equal across the two data sub-samples can be rejected.
This finding suggests that the modeling approach must account for this difference in the
two samples when the data is pooled for analysis. This is done by introducing ballot
measure type interaction terms that account for shifts in the impact of the explanatory
variables across the two sub-samples.

Next, extending the findings in this study to the US. population in general
requires verifying that the sample data being used is random. The statistical theory of
estimation and hypothesis testing indicates that random sampling is fundamental to
obtaining accurate estimates of population parameters (Wooldridge 2006). Violation of

the random sampling condition produces biased and inconsistent estimates that render

26

inferences drawn from this sample data valueless for purposes of explaining voter
decision choices outside the data sample.

States that have considered eminent domain and regulatory taking propositions
in the 2006 midterm elections may not represent a random sample of US. states. The
decision to reform eminent domain and regulatory takings law in a given state may be
influenced by several factors. For instance, the influence of special interest groups in a
given state’s legislative decisions may be important in explaining the decision to reform
eminent domain law. Also, whether a state already had some form of restriction on use
of eminent domain power prior to the 2006 elections may have affected the decision to
reexamine the law. It is therefore useful to verify if any set of variables systematically
affect the decision to reform eminent domain in a given state and county.

The typical approach to addressing this selection bias problem is to employ the
Heckman (1976) two-step procedure that involves estimation of a first stage equation
(selection equation) that explains the decision to hold a referendum. Results from the
first stage estimation are then incorporated into a second stage equation (outcome
equation) that explains the decision of the average voter to vote yes/no on the ballot
proposition at the polls.

The selection equation here is specified as,

W = 60 + 51 home + 5 2 highschool + 53 bachelor + (54 income +6 5 unemp
+ 56 density + 5 7ingroup+ 6 8takelaw + 5 9 incidence + r (15)

where W = 1 for states that placed eminent domain on the 2006 election ballots and 0
otherwise. For the Heckman procedure to be effective in addressing any non-

randomness in the sample data, the main variables in the outcome equation ,must also

27

appear in the selection equation. Here, the regulation type and voter turnout variables
are dropped from the selection equation because they are meaningless and counter-
intuitive as far as variables affecting the decision to reform eminent domain law are
concerned.

In addition to the main variables used in the outcome regression in equation
(16) two other variables are included in the selection equation to control for the effect of
special interest groups (ingroup) and existing restrictions on state eminent domain law
(takelaw). Both are dummy variables. Finally the incidence variable refers to per capita
property takings in a given state. Thus this value is the same for all counties within a
given state. It is expected that states with higher per capita takings would be more likely
to present voters with a proposition to restrict the use of eminent domain power and

require compensation for regulatory takings.

The Heckman procedure computes the so-called inverse Mills ratio (xi) which
is the ratio of the standard normal probability density function (pdf) to the standard

normal cumulative density function (cdf) evaluated at the estimated model parameters

and mean values of the explanatory variables. This new artificial variable (A) is

incorporated into the output equation as shown in equation (16) below,

In [Iii-5’) =fl0+fl1ballottype+fl2home +fl3home * ballotlype+ ,64 highschool+

,6 5 highschool“ ballottype+ ,66 bachelor+ ,67 bachelor“ balottype+ flgincome
+fl9 income“ ballottype+fl10 unemp+ ,611 unemfi‘ballottpe +,612 density+

,313 density" ballotorpe+fl14turnout+wzi +,6’15Ariz +fl16Calif +...,622 0re+§
(16)

28

The coefficient of the inverse Mills ratio can be tested using the regular t-ratio test. If

(U = 0 , then T and f are uncorrelated and the new variable A: should not appear in the

voter decision regression model. This means there is no evidence of selection bias in
estimation of the outcome equation. Results for the Heckman procedure are shown in
Table 1.3 below.

The standard error of 0.072 for the inverse Mills ratio parameterwhere
indicates that the Mills ratio variable is not statistically significant at the 10 percent
level. This implies that there is no need to control for sample selection bias in the

model. The model is thus estimated without the Mills ratio parameter.

1.4 Data

The paper uses cross-sectional county level data covering yes/no vote outcomes
on eminent domain and regulatory taking propositions in the 2006 mid-term elections.
The dependent variable in the model is the logodds of yes votes as defined in equation
(11) and the source of the vote data is the University of Michigan library government
documents centre (University of Michigan, 2006). The sample size of the data is 189.
Figure 1 below shows the contribution of the various states to the total sample size.
Florida has the highest county contribution of 18 percent followed by California with 17
percent. On the lower end New Hampshire and Idaho contribute 3 percent each to the

total sample size.

29

Figure 1 Composition of the data by contributing states

 

County Contribution of States to Total Data Sam ple

Mchigan, 25, 14% Georgia, 24, 13%
California, 31, 17% 1

, klaho, 6, 3%
S.Carolna, 20, 11%

  

Louisiana, 15, 8% *Ar‘uona. 10. 5%

N.Hanpshire, 5, 3%}

 

, l
0,690,142.,o Florida. 33, 18%

 

 

The variation in state contribution to total sample size is influenced by several
factors including the number of reporting counties in a state, and availability of data for
other variables in the model for the county concerned. Summary statistics for all
variables used in the model are shown in Table 1.4 below. The table provides
information on the sample size, unit of measure, mean, standard deviation, minimum,
and maximum values for each variable. Data summary statistic tables are useful in
identifying unusual observations and providing a quick impression about the
distribution of variables in the model.

The ballot measure type variable is a binary variable that equals 1 for state
ballot measures covering eminent domain and regulatory takings, and 0 for ballot
measures that cover eminent domain only.

The remaining independent variables included in the model are homeownership
rate, education, income, population density, unemployment rate, and turnout.
Homeownership rate is represented by the percent of occupied housing units that were

owner-occupied in 2006. This data is obtained from the 2006 US Census Bureau’s

30

American Community Survey. The data is limited to household population and excludes
population living in institutions, college dormitories, and other group quarters.

Level of education is represented by two variables: percent of people 25 years
and over who have completed high school education (includes equivalency) in 2006
(highschool), and percent of people 25 years and over who have completed bachelors
degree in 2006 (bachelors). The income variable is measured by the level of Median
Household Income (in 2006 Inflation-Adjusted Dollars). The source for both the
education and income variables is the 2006 American Community Survey data tables
(American Community Survey, 2006). Population density is measured by the number of
people living per square mile in each county and is computed using July 1 2006
population estimates and county area (land area in square miles) data. The source of this
data is the population division of US. Census Bureau (US. Census Bureau, 2006).

County level unemployment rate data is taken from the Bureau of Labor
Statistics’ local area unemployment statistics. The title of this data at source is “labor
force data by county, 2006 annual average (US department of labor, 2006)”. Election
turnout is computed as the ratio of votes cast to number of registered voters in a county.

To estimate the Heckman sample selection regression, the data sample size is
increased to 661 by including counties that did not consider eminent domain and
regulatory taking ballots. Additional explanatory variables included in the dataset to
allow for the implementation of Heckman sample selection regression are special
interest groups (ingroup), existing restrictions on state eminent domain law (takelaw).
and an incidence of takings variable that refers to per capita property takings in a given

state. Thus this value is the same for all counties within a given state. The interest group

31

variable is taken from Thomas and Hrebenar (2004). This variable is defined to equal 2
when the impact of interest groups in formulating laws is dominant, 1 when such groups
play just a complementary role in regulation formulation, and 0 when interest groups
are subordinate to other interests that influence law making in a given state. Takelaw is
a dummy variable that equals 1 when counties in a given state had some state
restrictions on the use of eminent domain power before the 2006 elections and equals 0
otherwise. The incidence of taking variable on the other hand is measured by per capita
property taking in a state. The source of the T akelaw and incidence data is the Castle

Coalition (2007).

1.5 Results

Results of the regression analyses are presented in Table 1.5 below. Three sets
of results are presented, the logit estimates, odds coefficient estimates, and the odds
elasticity values. The logit coefficients measure the change in logit or logodds of voting
yes for a one unit increment in the explanatory variables. The odds ratio estimates
indicate the value by which the odds of voting yes is multiplied for a unit increase in a
given independent variable. An odds ratio of 1.0 therefore corresponds to a zero
marginal effect of the logit coefficient suggesting that the independent variable
concerned has no effect on the logodds of voting yes. An odds ratio of 1.5 thus implies
the odds of voting yes is multiplied by 1.5 for a unit increase in the independent
variable concerned.

In general, odds ratios above 1.0 indicate an increase in the odds of voting yes as

the explanatory variable increases while odds ratios below 1.0 indicate that the odds of

32

voting yes decrease with an increase in the explanatory variable concerned. Finally, the
odds elasticity estimates offer a more direct interpretation of the estimation results. The
computed odds elasticityl values give the percentage change in the odds of a yes vote
for a one percent change in an explanatory variable. A detailed explanation of the effect
of right side variables on the dependent variable now follows.

The ballot measure type variable is statistically significant at the 5 percent level
in explaining the vote outcomes. The result from Table 1.5 indicates that if the impact
of ballot measure type on vote outcomes has nothing to do with the other independent
variables in the model (all interaction terms set to zero) then moving from a measure
that covers eminent domain only to one that covers eminent domain and regulatory
compensation decreases the logodds of voting yes by -2.888, and gives an odds ratio of
voting yes of 0.056. This implies that the odds of voting yes are multiplied by 0.056 as a
result of this change in ballot measure type. This finding is consistent with the
hypothesized relationship which suggests that voters appear to support imposing
restrictions on use of eminent domain power but oppose requiring compensation for
regulatory takings. Combining these two issues on one ballot increases the proportion of
voters rejecting the ballot relative to presenting a ballot on restricted use of eminent
domain only. This suggests that the strength of voter Opposition to regulatory taking

compensation exceeds support for restricted use of eminent domain.

 

i i . .
l The (direct) odds elasticity with respect to a given attribute X ,- (where [Z ,h j G X 1') rs given by,
6(a- ) / 6(I — Ft)
6X,-

    

(F‘)/(1 F ) =fliXi . See Fridstrom and Elvik (1997)
__r_-_-___:_i_

Xi

33

When interaction of ballot measure type and other independent variables are

accounted for, the unique effect of ballot measure type on vote outcome now becomes,

 

— 2.888+ 0.02 1* highschool— 0.022* income+ 0. 107* unemp+ 0.04 1* popdensty

Substituting the mean values of the interaction terms for highschool, income, unemp,
and popdensity gives the overall logit coefficient of ballot measure type as -2.548. The
corresponding odds value is 0.08. Overall, the impact of ballot measure type on the vote
outcome is negative as hypothesized. In summary the finding here indicates that
combining questions on restricted use of eminent domain power and regulatory
compensation on the same ballot reduces the odds of passing the ballot on eminent
domain.

Since the two ballot issues being analyzed here (eminent domain and regulatory
takings) affect property owners directly, homeownership can be expected to be
important in explaining the vote outcomes on these issues. Homeownership rate, which
refers to the percent of occupied housing units that was owner-occupied in 2006, is
however found to be statistically insignificant at the 5 percent level in explaining the
vote results. Variations in homeownership rate do not affect the chances that a voter
would vote yes at the polls. The non-significance of homeownership rate in explaining
the vote outcomes is inconsistent with the hypothesized positive impact. This finding is
however consistent with the November 2005 survey results by the Saint Index
organization (see Somin 2007) on the Kelo ruling which indicates that renters reject the
Kelo ruling almost as strongly as homeowners. In particular, 70 percent of renters
opposed the ruling while 83 percent of homeowners also opposed the ruling. This

implies that homeownership is simply not an important explanatory factor of voters‘

34

position on eminent domain reforms. The results from this regression analysis show that
this is true for regulatory taking reforms as well.

The capability of voters in understanding the costs and benefits associated with
eminent domain and regulatory takings may also be important in explaining the vote
choices. Two measures of education are included in the model (highschool and
bachelor) to control for knowledge of voters. The impact of highschool on the vote
outcome was found to be dependent on ballot measure type: highschool is statistically
significant at the 5 percent level in explaining vote outcomes on eminent domain and
regulatory takings initiatives but insignificant when the ballot measure involves eminent
domain only. In other words, when ballot measure type is set to 1, the odds elasticity
values indicate that the effect of a one percentage point increase in the percentage of
people 25 and over that hold at least a highschool certificate increases the odds of
voting yes by 1.8 percent. When ballot measure type is set to 0 highschool has zero
effect on vote outcomes. On the other hand, the impact of the higher measure of
education (bachelor) is independent of ballot measure type. A one percentage point
increase in the percentage of people 25 and over that hold at least a bachelor’s degree
decreases the odds of voting yes by 0.35 percent.

This finding is interesting: a more broad measure of high education (highschool)
correlates positively with yes votes while the finer and more rigorous measure
(bachelor) results in a decline in yes votes. This points to some threshold in the effect of
education on restricted use of eminent domain and regulatory compensation, and

possibly on the environment in general. It takes a level of education above highschool

35

for education to have a positive impact on environmental and resource conservation
measures.

In general, as the level of education of voters increases support for the ballot
measure declines. Thus counties with more highly educated voters appear to be less
supportive of more stringent property rights protection, and regulatory takings
compensation. High education affects actual takings as well. Lanza (2006) for instance
found that high education (per capita number of legal workers) positively affects the
number of takings at the state level in the US.

Strength of the economy is another category of variables that can be expected to
influence voter support for ballot propositions. Voter dissatisfaction with bad economic
conditions generally erodes support for ballot proposals because of low support for
government (Clarke and Kornberg 1994, Bowler and Donovan 1994). The regression
analysis here controls for two measures of economic strength, household income and
unemployment rate. Income is statistically significant at the 5 percent level in
explaining the vote outcomes irrespective of the ballot measure type under
consideration. Lower median household income counties are more likely to vote yes.

Since income is measured in thousands of dollars, a $ 1000 increase in median

household income increases the logit of voting yes by 0.016 -0.022 *W .
Setting ballot measure type to 1 gives a logit estimate of -0.006, and results in
multiplication of the odds of voting yes by 0.994. On the other hand, setting ballot
measure type to 0 gives a logit estimate of 0.016, and results in multiplication of the
odds of voting yes by 1.016. Using the computed odds elasticity values, a one

percentage point increase in income reduces the odds of voting yes by 0.28 percent

36

when ballot measure type is set to 1. On the other hand, when the impact of income is
independent of ballot measure type, a one percentage point increase in income increases
the odds of voting yes by 0.75 percent. This finding of negative relationship between
income and yes votes for restricted use of eminent domain and regulatory compensation
is consistent with the hypothesized relationship that benefits expected from property
takings are normal goods.

In summary, the results indicate that education and income have similar effects
on the dependent variable. Both have a negative effect on increased property rights
protection and regulatory compensation. This finding is consistent with the
hypothesized relationship which posits that both education and income would correlate
negatively with yes votes. It also indicates that when confronted with a choice between
more secure property rights and a healthier environment both educated and high income
voters lean towards protecting the environment. This implies that this class of voters
wony less about losing their property to government takings and probably confirms
previously discussed reports that property takings tend to affect poorer communities
disproportionately.

Unemployment rate, the second measure of economic strength is statistically
significant at the 5 percent level in explaining vote outcomes irrespective of ballot
measure type. Higher unemployment rates have a negative effect on the logodds of

voting yes at the polls. A one percent increase in unemployment rate decreases the logit

of yes votes by — 0.1 19 + 0.107 *ballottype . Setting ballot measure type to 1 gives a

logit estimate of -0.01, and results in multiplication of the odds of voting yes by 0.990.

Setting ballot measure type to 0 gives a logit estimate of -0.106, and results in

37

multiplication of the odds of voting yes by 0.888. The results therefore indicate that
unemployment has greater effect on eminent domain only ballots than it has on eminent
domain and regulatory takings combined.

Using the computed odds elasticity values, a one percent increase in
unemployment rate reduces the odds of voting yes by 0.06 percent when ballot measure
type is set to 1, and by 0.60 percent when ballot measure type is set to O. This implies
that voters in counties that are relatively weak economically tend to support relaxation
of restrictions on use of eminent domain power but are less supportive of regulatory
taking compensation. The effect of unemployment therefore re-enforces the finding for
income: High income and low unemployment rate both result in an increase in the odds
of voting yes for restricted use of eminent domain power and regulatory takings
compensation.

Population density is included in the model to capture rural-urban differentials
in the perceived benefits and costs of eminent domain power and regulatory takings.
Population density is statistically significant at the 5 percent level in explaining changes
in the vote outcome only when it is interacted with ballot measure type. When ballot
measure type is 0, population density has zero effect on the vote outcomes. This is
consistent with Lanza’s finding of no relationship between population density and
actual property takings (eminent domain takings).

When ballot measure type is set to 1, a 100 unit increase in population density is
associated with a 0.041 increase in yes votes and a multiplication of the odds ratio by
1.042. Using the elasticity estimates, a one percent increase in population density

increases the odds of voting yes by 0.1 percent when ballot measure type is set to l.

38

This means population density does not affect voters’ position on eminent domain as it
does on regulatory takings compensation.

Turnout in this model has the expected negative and statistically significant (at 5
percent level) effect on the vote outcome. Higher turnout at the polls is associated with
rejection of the ballot measure as hypothesized. In particular, a one percent increase in
turnout reduces the logits of yes votes by 0.01 and results in a multiplication of the odds
of a yes vote by 0.99. The computed elasticity value indicates that a one percent
increase in turnout results in 0.50 percent decline in yes votes. As noted by Hadwiger

this finding raises a question for future research.

Mean Respgnse

Evaluating the independent variables in the model at their mean values, the
model predicts that the logodds of voting yes for the average voter is 0.940. This value
is associated with an odds value of 2.56 and a probability2 of voting yes of 0.719. Thus
the model predicts that the average voter in this dataset is more likely to vote yes to call

for additional restrictions on use of eminent domain and regulatory takings than not.

Application of Results
The results from this research are insightful and useful in revealing how vote
outcomes at the county and state levels could change for a given change in each of the

statistically significant explanatory variables. First, it is evident from the results that if

 

. odds

2 Probabilities can be obtained from the computed odds using the formula, P = —
l + odds

39

the policy objective is to get voters to vote no to imposing additional restrictions on
eminent domain power, then adding a requirement for regulatory taking compensation
increases the chances of doing 50. Taking the results for Florida as an example, 69
percent of voters voted yes to pass the measure that calls for restricted use of eminent
domain power. This gives the odds of a yes vote of 2.226 as shown in Table 1.6 below.
The results from this analysis indicate that adding regulatory takings to eminent domain
on this ballot could have resulted in multiplication of this odds by 0.08 giving the new
odds value of 0.18. Note that this reverses the observed result from a “pass” to “fail”.
Similar analyses can be performed for the remaining explanatory variables at both the
state and county levels. Table 1.6 below presents new odds value for yes votes for the
states included in the data for this study when ballot measure type is set to 1. The
dramatic impact of combining these ballots is clear from the results in this table. The
percentage decline in the odds of passing the ballot is approximately 92 percent across
states.

The impact of ballot measure type on the odds of yes votes is however
significantly reduced when combined with the impact of other variables in the model.
Table 1.7 below gives predicted odds values from the estimated model for all 51 US.
states. Taking the Florida vote results as example again, the predicted odds of yes votes
when ballot measure type =1 is 0.499 suggesting that the ballot fails when restricted use
of eminent domain is combined with regulatory taking compensation. Consistent with
the actual vote results, the model predicts that the ballot in Florida passes when ballot

measure type = 0.

40

Results from Table 1.7 basically indicate that the ballot measure passes in all
states when presented as an eminent domain taking only ballot. Once restricted use of
eminent domain is combined with regulatory taking compensation the ballot fails in all
states except in DC. The computed predicted odds for DC are particularly high because
of the very high population density value for this state. The results in Table 1.7 still do
not tell the whole story about what affects the passage of restricted use of eminent
domain and regulatory compensation ballots since the ballot passed in Arizona (see
Table 1.1) even though these two issues were combined. The Arizona ballot result
shows that the negative impact of requiring regulatory taking compensation (although
generally high) may not be high enough to reverse the passage of restricted use of

eminent domain power measures in all cases.

1.6 Conclusions

The empirical analysis from this study has produced very interesting and
insightful results about voter decision-making on property takings in the US. It is clear
that voters do not look at eminent domain and regulatory taking compensation in the
same light. While the average voter leans more towards strengthening property rights
protection (restricting use of eminent domain) and providing at least market value
compensation for eminent .domain takings, adding compensation for property use
regulations reduces this support significantly. This implies that to obtain an accurate
estimate of demand for each of these two institutional measures it is important to

present them to voters in separate ballot questions.

41

Restoring landowners to their status quo welfare position however requires full
compensation for all takings (both eminent domain and regulatory takings). The results
here suggest that the average voter may be looking at taking compensation in a more
practical and cost effective way. For instance, while it is relatively easy to determine
compensation values for eminent domain takings, determining such value for regulatory
takings is a much more complicated and expensive task. Studies may have to be
initiated to determine property value losses imputable to the regulatory taking action.
Given that the cost of financing such studies may be substantial relative to the
compensable values, it makes sense that the average voter is side-stepping this
requirement.

Homeownership rate does not appear to significantly explain the vote outcomes
at all. This is an interesting finding since property right is central to the entire argument
on eminent domain and regulatory taking reforms. However, the November 2005
survey results by the Saint Index polling organization (Somin 2007) on the Kelo ruling
throws more light on the non-significance of homeownership rate in explaining the vote
outcomes. This survey result indicates that renters reject the Kelo ruling almost as
strongly as homeowners. This clearly weakens any distinguishing impact of
homeownership rate on vote choices.

Strength of the economy also matters in voter decision making on eminent
domain takings. Counties with higher average unemployment rates are more supportive
of relaxing requirements for eminent domain takings and regulatory compensation. This
is in line with the hypothesis that voters in weaker economies would tend to support

relaxation of the requirements for takings to promote economic activity and job

42

creation. The effect of unemployment also re-enforces the finding for income: high
income and low unemployment rate both result in an increase in the odds of voting yes
for restricted use of eminent domain power and regulatory takings compensation. This
finding is consistent with those from several other studies that indicate that voter
dissatisfaction with bad economic conditions erodes support for ballot proposals
because of low support for government (Clarke and Kornberg 1994, Bowler and
Donovan 1994).

The results indicate that education and income have a negative effect on
increased property rights protection and regulatory compensation. It indicates that when
confronted with a choice between more secure property rights and a healthier
environment both educated and high income voters lean towards protecting the
environment. This implies that this class of voters worry less about losing their property
to government takings and probably confirms previously discussed reports that property
takings tend to affect poorer communities disproportionately.

The results from this study as well as those from Lanza (2006) indicate that
population density is not an important variable in explaining both actual eminent
domain takings and voter decisions on eminent domain takings. The new finding here is
that population density does explain vote outcomes on regulatory takings compensation.
This is probably due to the relatively higher value of open space in highly populated
urban areas as compared to rural areas. Finally, consistent with the hypothesized
relationship in this study as well as previous findings on protest votes, election turnout
was found to negatively affect passage of restricted use of eminent domain and

regulatory takings compensation ballots.

43

TABLES

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 1.1
Summary of Results for all Eminent Domain Ballots in 2006
State Measure # Topic Area Pass/F ail
Arizona Prop. 207 Eminent domain & regulatory Pass (64.8%)
takings
California Prop. 90 Eminent domain & regulatory Fail (47.6%)
takings
Florida Amendment 8 Eminent domain Pass (69%)
Georgia Amendment 1 Eminent domain Pass (82.7%)
Idaho Prop. 2 Eminent domain & regulatory Fail (23.9%)
takings
Louisiana Amendment 5 Eminent domain Pass (55%)
Michigan Proposal 06-4 Eminent domain Pass (80.1%)
Nevada Question 2 Eminent domain Pass (63.1%)
New Question 1 Eminent domain Pass (85.7%)
Hampshire
North Measure 2 Eminent domain Pass (67.5%)
Dakota
Oregon Measure 39 Eminent domain Pass (67.1%)
South Amendment 5 Eminent domain Pass (86%)
Carolina
Washington Initiative 933 Regulatory takings Fail (41.2%)

 

 

 

 

 

Source: National Conference of State Legislatures: Property Rights Issues on the 2006 Ballot.

Washington State is not included in the data set considered for this study because the vote initiative

involves only regulatory taking. Nevada and North Carolina are dropped from the study due to missing

data problems for other variables in the model

44

 

Table 1.2

Estimated Logit Coefficients by Ballot Measure Type and Pooled Data Sample
(Dependent variable: logodds of yes votes)

 

 

 

Variable Estimated results Estimated results Estimated
when ballot when ballot results when
type=l type=0 data is pooled
Homeownership -0.005 0.010 0.03 8 *
(0.015) (0.007) (0.008)
High school 0.019 0.018 0031*
(0.018) (0.011) (0.013)
Bachelors 0.017 0.017 0.017
(0.019) (0.010) (0.012)
Income 0.017 0.027* -0.008
(0.013) (0.006) (0.007)
Unemployment 0.065 0.212“ 0138*
rate (0.051) (0.021) (0.034)
-0.052 0.006 0.045
Population density (0.062) (0.023) (0.027)
-0.008 -0.001 -0.019*
Turnout (0.017) (0.004) (0.027)
-1.335 -0.542 -4.175*
Constant (1.250) (0.717) (0.724)
R-square 0.162 0.432 0.308

 

 

 

 

 

45

Note: Values in parentheses are standard errors and * implies the estimate is statistically significant at 5%
level

 

Table 1.3

Heckman Sample Selection Regression Results
(Dependent variable: Selection Equation-Decision to hold referendum (W))
: Outcome Equation-logodds of yes votes

 

 

 

Variable Selection Equation Outcome Equation
Ballot type - 1.189
(1.549)
Homeownership -0.021 0.020*
(0.009) (0.007)
Homeownership*ballot type - -0.028
(0.019)
High school -0.014 -0.021
(0.009) (0.013)
High school*ballot type - 0.024
(0.010)
Bachelors 0.001 0.022*
(0.008) (0.010)
Bachelors*ballot type - -0.042
(0.023)
Income 0.008 -0.020*
(0.006) (0.006)
Income*Ballot type - -0.015
(0.016)
Unemployment rate 0.1 16 0.194*
(0.004) g (0.029)
Unemployment rate*ballot type - -0.018
(0.040)
Population density -0.029 0.011
(0.024) (0.024)
Population density*ballot type - -0.044
(0.040)
Turnout - -0.012*
(0.004)
Taking Law -0.708
(0.224)
Interest group 0.087
(0.094)
Taking Incidence 0.046
(0.011)
Mills ratio 0.543
(0.398)
Constant 1.1 19 -0.566
(0.978) (0.875)

 

 

 

Note: Values in parentheses are standard errors and * implies statistically significant at 5%

level

46

 

Table 1.4

Summary Statistic of Variables

 

 

 

Variable Unit Mean Standard Min Max
Deviation

Yes Votes (logodds) 0.931 0.806 -l.537 2.223
Yes votes Percent 69.600 16.300 17.700 90.200
Ballot type 0.249 0.433 0 1
Homeownership rate percent 69.740 8.247 45.200 87.800
High school percent 83.971 6.396 62.300 96.000
Bachelors percent 23.412 8.166 10.000 52.600
Income ‘000$ 46.576 9.551 23.119 81.761
Unemployment rate percent 5.048 2.014 2.400 15.300
Population density Pop/milesz (‘00) 2.523 1.988 0.063 9.462
Turnout percent 50.008 12.712 2.032 75.700

 

 

 

 

 

 

47

 

Table 1.53

Estimated Logit, Odds, and Odds Elasticity Results
(Dependent variable: logodds of yes votes)

 

 

 

 

 

 

Variable Logits Odds Odds elasticity

Ballot type 2888* 0.056 -
(0.976)

Homeownership 0.002 1 .002 0. 139
(0.004)

High school -0.006 0.994 -0.504
(0.009)

High school*ballot type 0021* 1.021 1.763

' (0.015)

Bachelors -0.015* 0.985 -0.351
(0.006)

Income 0016* 1.016 0.745
(0.004)

Income*Ballot type 0022* 0.978 -1 .025
(0.010)

Unemployment rate -0.1 1 9* 0.888 -0.601
(0.035)

Unemployment rate*ballot type 0107* 1.1 13 0.540
(0.040)

Population density -0.007 0.993 -1.766
(0.014)

Population density*ballot type 0041* 1.042 0.103
(0.023)

Turnout -0.010* 0.990 -0.500
(0.004)

Constant 1.514* 4.545 -
(0.658)

R-square 0.930

 

 

Note: Values in parentheses are standard errors and * implies the estimate is statistically significant at 5%
level

 

3 Results exclude nine State-dummy variables that were included in the regression

48

 

Table 1.6

State-Level Odds of Yes Votes When Ballot Measure Type = l

 

 

 

Variable Yes Odds for Actual Yes Odds for Actual Votes
Vote Results (When Ballot Type=l)
Florida 2.23 0.18
Georgia 4.78 0.38
Louisiana 1.22 0.10
Michigan 4.03 0.32
New Hampshire 5.99 0.48
Oregon 2.04 0.16
S. Carolina 6.14 0.49

 

 

 

49

 

Table 1.7
State Predicted Odds of Passing Eminent Domain Ballots by Ballot Measure [y e

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

No. States Predicted Odds when Predicted Odds when
ballot measure type=0 ballot measure type=1
1 Alabama 2.541 0.471
2 Alaska 2.177 0.433
3 Arizona 2.793 0.495
4 Arkansas 2.321 0.552
5 California 2.966 0.431
6 Colorado 2.360 0.421
7 Connecticut 3.774 0.529
8 Delaware 3.596 0.549
9 DC 94.550 18.580
10 Florida 2.871 0.499
1 1 Georgia 2.635 0.481
12 Hawaii 4.257 0.523
13 Idaho 2.756 0.526
14 Illinois 2.834 0.490
15 Indiana 2.804 0.581
16 Iowa 2.554 0.519
17 Kansas 2.391 0.498
18 Kentucky 2.122 0.492
19 Louisiana 2.874 0.542
20 Maine 2.164 0.488
21 Maryland 3.820 0.475
22 Massachusetts 3 .277 0.516
23 Michigan 1.736 0.447
24 Minnesota 2.363 0.413
25 Mississippi 2.750 0.754
26 Missouri 2.541 0.471
27 Montana 2.191 0.472
28 Nebraska 2.203 0.474
29 Nevada 2.725 0.504
30 New Hampshire 3.322 0.526
31 New Jersey 3.158 0.454
32 New Mexico 4.733 0.643
33 New York 2.352 0.470
34 North Carolina 3.088 0.531
35 North Dakota 2.744 0.553
36 Ohio 2.616 0.519
37 Oklahoma 2.363 0.538
38 Oregon 3.805 0.822
39 Pennsylvania 1.985 0.448
40 Rhode Island 4.219 0.849

 

50

 

Table 1.7
State Predicted Odds of Passing Eminent Domain Ballots by Ballot Measure Type

 

 

 

 

 

 

 

 

 

 

 

 

Continued
41 South Carolina 3.251 0.564
42 South Dakota 2.033 0.501
43 Tennessee 2.387 0.461
44 Texas 2.365 0.512
45 Utah 2.663 0.486
46 Vermont 3.352 0.553
47 Virginia 2.291 0.438
48 Washington 3.314 0.443
49 West Virginia 2.431 0.466
50 Wisconsin 2.622 0.612
51 Wyoming 2.367 0.477

 

 

 

 

51

 

References

Bengston D.N., Fletcher J .0, Nelson KC “Public policies for managing urban growth
and protecting open space: policy instruments and lessons” Landscape and
Urban Planning, 69, no 2-3(2004) 271-286

Bowler S. and Donovan T. “Economic Conditions and Voting on Ballot Propositions.”
American Politics Research, 22, no 27(1994): 27-39.

Calvert J .W. "The Social and Ideological Bases of Support for Environmental
Legislation: An Examination of Public Attitudes and Legislative Action", The
Western Political Quarterly 32, no 3, (1979): 327-337.

Castle Coalition, "50 State Report Card: Tracking Eminent Domain Reform Legislation
since Kelo"

Clarke H.D. and Kornberg A. "The Politics and Economics of Constitutional Choice:
Voting in Canada's 1992 National Referendum" The Journal of Politics, 56, No.

4 (Nov., 1994), pp. 940-962

Dasgupta, S., Laplante, B., Wang, H. and Wheeler, D. "Confronting the Environmental
Kuznets Curve." The Journal of Economic Perspectives 16, no. 1(2002): 147-
168.

Deacon, R., and P. Shapiro. "Private Preference for Collective Goods Revealed Through
Voting on Referenda." The American Economic Review 65, no. 5(1975): 943-
955.

Downs, A. An Economic Theory of Democracy. New York: Harper, 1957.

Fischel, W.A. "Determinants of Voting on Environmental Quality" A Study of a New
Hampshire Pulp Mill Referendum." Journal of Environmental Economics and

Management 6(1979): 108-1 18

Flick W.A., Barnes A., Tuft R. "A Public Purpose and Private Property: The Evolution
of Regulatory Taking" Journal of Forestry 93, No 6, (June 1995) , pp. 21-24(4)

F ridstrom and Elvik "The barely revealed preference behind road investment priorities"
Public Choice, 92: 145—168, 1997

Goldstein, J ., and W. Watson "Property rights, regulatory taking, and Compensation:
implications for environmental protection" Contemporary Economic Policy

15(1997):32-42.

Greene, W. Econometric Analysis. Delhi: Pearson Education, 2003.

52

Grizzle J .E., Starrner C.F., Koch G.C. "Analysis of Categorical Data by Linear Models."
Biometrics no. 25(1969): 489 - 505.

Grossman, G. M. and Krueger, A.B. Environmental Impacts of a North American Free
Trade Agreement. In "T he Mexico-US. fiee trade agreement", P. Garber, ed.
Cambridge, Mass: MIT Press, 1993.

Hadwiger, D. "Money, Turnout, and Ballot Measure Success in California Cities". The
Western Political Quarterly, 45(1992), No. 2, 539-547.

Harbaugh, B., Arik Levinson and Dave Wilson."Reexamining the Empirical Evidence
for an Environmental Kuznets Curve."Review of Economics and Statistics 84,
no.3 (2002).

Heckman J .J . (1976): The Common Structure of Statistical Models of Truncation,
Sample Selection and Limited Dependent Variables and a Simple Estimator for
such Models, Annals of Economics and Social Measurement 5, 475-492.

Hess GB. and Orphanides A., War Politics: An Economic, Rational-Voter Framework,
The American Economic Review 85, No. 4 (Sep., 1995): 828-846

Hettige, H., Lucas, R.E.B. and Wheeler, D. “The toxic intensity of industrial
production: global patterns, trends and trade policy”, American Economic
Review, Papers and Proceedings 82, 478-81

Kahn M.E. "Demographic Change and the Demand for Environmental Regulation. "
Journal of Policy Analysis and Management 21, no 1(2002):45-62

Kahn, M. E., and J. G. Matsusaka. "Demand for Environmental Goods: Evidence from
Voting Patterns on California Initiatives." Journal of Law and Economics 40,
no. 1(1997): 137-173.

Kotchen, M. J ., and S. M. Powers. "Explaining the appearance and success of voter
referenda for open-space conservation." Journal of Environmental
Economics and Management 52, no. 1(2006): 373-390.

Knox, J ., Landry, C. and Payne G. "Local Initiative. A Study of the Use of
Municipal Initiatives in the San Francisco Bay Area." San Francisco: Coro
Foundation, 1984

Lanza, S. "An Offer You Can’t Refuse: Why Do Connecticut and Other States Use
Eminent Domain?" The Connecticut Economy 9 (Winter 2006).

Lazzarotti J. "Public Use or Public Abuse." University of Missouri-Kansas City Law
Review 68 (1999), 49-53

53

List, J. A.,and Gallet, C. A. "The Environmental Kuznets Curve: Does one size fit
all?" Ecological Economics 31, no 3 (1999): 409-423.

Matsusaka J.G. "Election Closeness and Voter Turnout: Evidence from Califomia
Ballot Propositions." Public Choice 76 (1993): 313-34

Meddler, J., and A. Mushkatel. "Urban-Rural Class Conflict in Oregon land-use
planning." Western political quarterly 32(1969): 344-348.

Merrill T.W. "The Economics of Public Use." Cornell Law Review 61 (1986): 74-75
Millimet, D. L., List, J. A., & Stengos, T. The Environmental Kuznets Curve: Real
progress or misspecified models? Review of Economics and Statistics 85,
no 4 (2003): 1038-1047.
Michigan, S. C. Mich 616 (en banc) (1981).

Orthner D. "Toward a More “Just” Compensation in Eminent Domain" MC GEORGE
L. REV. 38(2007): 429- 455

Palfrey, T. R., and K. T. Poole. "The Relationship between Information, Ideology, and
Voting Behavior." American Journal of Political Science 31, no.3(1987):511-
530.

Popp D. " Altruism and the Demand for Environmental Quality" Land Economics 77 no
3(2001): 339 -349

Press, D. "Who Votes for Natural Resources in California?" Society and Natural
Resources 16(2003): 835-846.

Riddiough T.J. "The Economic Consequences of Regulatory Taking Risk on Land
Value and Development Activity" Journal of Urban Economics, 41 no.1 (1997):
56-77

Salka W.M. "Urban-Rural Conflict over Environmental Policy in the Western United
States." The American Review of Public Administration 31 no. 1 (2001): 3-48

Sandefur D. "The Backlash So Far: Will Americans Get Meaningful Eminent Domain
Reform?" Michigan State Law Review 709 (2006): 711-746

Somin, 1.. "Is Post-Kelo Eminent Domain Reform Bad for the Poor?". Northwestern
University Law Review, 101, No. 4(Fall 2007): 1931-1943

Stone C. N. "Local Referendums: An Alternative to the Alienated-Voter Model."
The Public Opinion Quarterly, 29, no. 2 (Summer, 1965), pp. 213-222

54

Thomas C. and Hrebenar R. "Interest Groups in the States" (2004). In, Politics in the
American States, Edited by Gray V. and Hanson R.

University of Michigan, "Election Results and Voting, Election 2006." link:
http://www.lib.umich.edu/govdocs/elec2006.html

US. Census Bureau, American Community Survey (2006). data link:
http://factfinder.census.gov/home/saff/main.html?_lang=en

US. Department of Labor, "Local Area Unemployment Statistics (LAUS) Program",
2006

US Supreme Court. Susette Kelo et al., Petitioners v. City of New London, Connecticut
et al. Vol. 545 U.S._2005, 2005.

US. Supreme, Court. "Berrnan v. Parker." 348 US. 26(1954)

Wooldridge J .M. "Introductory Econometrics, A Modern Approach" (Third Edition),
Mason, Ohio, 2006.

55

Essay 2

Information and Bargaining Breakdowns in Eminent Domain Takings

2.1 Introduction

The fifth amendment of the US. constitution is non-specific on the requirement
for just compensation in eminent domain takings (United States 1791). This non-
specificity about the definition of just compensation makes bargaining between owners
and local governments an integral part of eminent domain takings.

Eminent domain bargaining is however often plagued with owner holdouts.
Holdouts generally refer to the rejection of government compensation offers by private
owners in an attempt to extract higher compensation for loss of their properties (Fennel
2005, Parchomovsky and Siegelman 2004). Fennel (2005) and Parchomovsky and
Siegelman (2004) distinguish between two non-selling owners. They describe holdouts
as owners who refuse to sell in an attempt to extract a substantial share of government
surplus from property rights transfer. Holdins on the other hand refer to owners who
refuse to sell because they truly value their properties above the government
compensation offer. Holdouts as used in this paper cover both categories of non-selling
owners and the results derived here are relevant across these two classes of owners.

Holdouts introduce inefficiency into the economic system by imposing delay
costs on government and possibly preventing mutually beneficial trades from being
realized (Munch 1976, Bell and Parchomovsky 2005). The threat of use of eminent

domain power reduces delay costs resulting from such holdouts by reducing the

56

bargaining period through forceful taking over of properties involved in protracted
negotiation.

The holdout problem is further deepened by owner asymmetric information
problems about the true value govemment places on the property under threat of taking
(Menezes and Pitchford 2004, Fudenberg and Tirole 1983). This information, which is
private to the government, is important in determining the ceiling price offer (maximum
price government is willing to pay) of government. The main premise of this paper is
that extended holdouts and bargaining breakdowns result fi'om this asymmetric
information problem because owners attempt to bargain for prices that are beyond the
ceiling price of government.

The economic modeling of holdouts in eminent domain takings has received
little attention from economists. As noted by Miceli and Segerson (2007), although the
holdout problem has been discussed widely in the economics literature, few authors
have attempted to model the problem formally. Miceli and Segerson (2007) modeled
the problem of eminent domain holdouts by considering a two-period bargaining
problem between a developer and two owners. In this setting, they observed that
holding out is one of two possible sub-game perfect Nash equilibria, the other being
first-period sale of all parcels. When use of eminent domain power is introduced in
period two to take parcels that the deve10per was not able to acquire consensually in
period one and when compensation for takings is set at a property’s market value, they
found that the unique sub-game perfect Nash equilibrium is for all owners to sell in
period one. Thus, the mere threat of eminent domain is sufficient to overcome the

holdout problem.

57

In practice, the threat of eminent domain has not been sufficient in overcoming
the holdout problem. Eminent domain bargainings often end up in the courts due to
breakdowns in bargaining over acceptable compensation (Parchomovsky and Bell
2006). Clearly, the threat of use of eminent domain power appears to be less effective in
practice than theoretical predictions suggest.

This paper reconciles the theoretical and observed results by explaining
bargaining breakdowns. First, using a single owner and government bargaining model
the conditions under which the threat of eminent domain guarantees resolution of the
holdout problem are stated and explained. Next, the informational assumptions are
relaxed to explain how owner information affects the chances of bargaining breakdown
when the model accounts for uncertainty (incomplete information). Results indicate
that the less informed owners are about the payoff of government the greater is the gap
between the theoretical predictions of the effectiveness of eminent domain in resolving
holdouts and actual observations of bargaining breakdowns. This gap declines as the
owner’s subjective probability that government accepts a counter price offer increases
with improved owner information about the payoff of government.

In general, it is shown that eminent domain resolves holdouts completely when
negotiating owners engage in a complete and perfect information bargaining game.
Information is complete when the payoff of all parties in a bargaining game is common
knowledge (Gibbons 1992). This implies that both government and owner are privy to
each other’s exact payoff at every stage of the game. A party in the bargaining game has
perfect information when at every move of the game the history of the game up to that

point is known to this player (Gibbons 1992). This indicates that there are no

58

simultaneous moves at any stage of the game. Thus at the time each player makes a
move in the game, the player can review all previous moves in the game up to that
point.

The rest of the paper is ordered as follows. The next section presents and solves
a dynamic two period complete and perfect information game between an owner and a
government. This is followed by the individual owner bargaining problem under

incomplete information, discussion of results, and conclusions.

2.2 Bargaining Under Complete and Perfect Information

In this section of the paper, information about the payoffs of both agents in the
bargaining game as well as the history of the bargaining process is assumed to be
common knowledge to both agents. This information assumption guarantees early
resolution of the holdout issue by eliminating uncertainty and directing the equilibrium
price search rapidly towards a feasible price range.

To see this, consider a land market where a local government engages a property

owner in bargaining to acquire the owner’s property for a public project. Let the owner
and government valuation of the property be given as V L and VG respectively. Also let

the market value for properties identified for taking be represented bym. For
simplicity, it is assumed that court decisions on owner challenges to the right of
government to use eminent domain power always go in favor of government and courts
always award the market value of properties as the appropriate compensation. This

implies that owners can only look forward to market value compensation if they opt for

59

court settlement. The same is true when owners make overly high counter price offers
that compel government to seek court settlement.

Two types of costs are considered in the payoff functions of government and
owners. Government incurs delay costs of d by the time a court decision is made if the

owner holds out, while both government and owners incur court costs of 3 when court
settlement becomes necessary.

Bargaining between the owner and government occurs in two sequential rounds.

First, government offers the owner a price p0. The owner responds by choosing an
asking price p L . If p0 = p L , the owner accepts the govemment’s offer and the game
ends. If p L > [)0 , government enters the second round of negotiations and decides
whether p L is acceptable or not. Government accepts pL if p L S VG and proceeds to

evoke the use of eminent domain power if p L > V0. The bargaining game is

illustrated in the game tree in Figure 1.

The game tree in Figure 1 indicates that the bargaining game begins with the
government (G) making the offer [76 to the owner (L). If the owner accepts the offer
the owner gets the payoff pG - V L and the government gets VG - pG . If the owner
rejects this offer then the game enters the second period with the owner making a
counter offer denoted by p L- If the government accepts this counter offer then the

owner and the government get the payoffs p L — v L and VG - p L respectively. On

the other hand if the government rejects the counter offer the case is settled in court.

The payoff to government and the owner when there is a court settlement is given by

60

VG —m—d '4 and m — V L — 8 respectively. For simplicity, both government and

owner are assumed to be patient about receipt of the net benefits from the property
rights transfer and thus do not discount these values.

For a finite bargaining game of perfect information like the one described here a
pure strategy unique Nash equilibrium outcome can be derived using the backward
induction approach (Gibbons 1992, Mas-Collel A. et a1 1995). Beginning in the second
and final stage of the game, the government compares its payoff from rejecting the
owner’s counter-offer and settling the case in court to what is obtainable from accepting

the counter offer. For government not to go to court, the payoffs from accepting the

owner’s counter offer VG - p L must be equal or greater than that from a court
settlement, VG -m—d ~€ . Equating these two payoffs gives pL = m + d + f
implying that government goes to court only if p L > m +d + l . Here, assume that

VG > m+d +5 since government has no incentive to proceed with a taking

otherwise. The result shows that the maximum compensation that government is
willing to pay is equivalent to the sum of the market value of the property (m), delay
costs (61 ), and estimated legal costs from court litigation (£7 ). An owner counter offer
exceeding this value is rejected paving the way for a court settlement.

If government will pay p6 = m+d +5 in period 2 then it might as well offer

this amount as compensation in period 1 to end the bargain game. This is because there
is no benefit to government in allowing for protracted negotiations under this
circumstance. Introducing discounting into the problem makes it more evident that it

pays for government to end the bargaining early to maximize the discounted net

61

benefits from the project. The owner accepts this first offer in period 1 and the game
ends.
Turning to the owner’s moves in the game, it is obvious that the owner gets the

worst payoff (less than market value compensation) by going to court. Thus in principle
the owner would find an offer of p0 = m — g in period 1 acceptable if offered since it
is not worse than what is available to the owner after court settlement. However, the
owner would reject this offer (m - f ) and make a counter offer that is just less than or

equal to m+d +8 . This is because the owner is privy to the fact that any threat by

government to use eminent domain power is not credible until the owner’s counter offer

exceeds m+d +5. Like all dynamic games, credibility of the threat to use eminent
domain power is very important in arriving at the equilibrium outcome in this game. In
fact backward induction outcomes by definition must be devoid of all non-credible
threats.

In summary the backward induction equilibrium outcome of this bargaining

problem is for government to offer the owner a compensation of m + d + f in the first
period ( pG = m + d + E ). The owner accepts the offer and the game ends. The owner

thus gets well in excess of market value as compensation when the theoretical
information assumptions in this section are true. The finding that the owner receives
more than market value in the unique Nash equilibrium is quite consistent with what
may be expected in a free market exchange between government and the owner. Given
that the owner in question did not put the property on the market for sale at the going
market price before government expressed interest, it must be the case that the owner

values this property above market price.

62

This result is consistent with the findings by Miceli and Segerson (2007) in their
three-way bargaining problem that involves two owners and a developer. Miceli and

Segerson observed that when both owners bargain in the first period the optimal

Zr+6
3

 

compensation for each owner is given as, m+ where, Z' andd refer

respectively to the transaction/litigation costs, and delay costs incurred by the
developer. On the other hand, if only one of the two owners decides to bargain in the

first period (with the other holding out) then the optimal compensation for the

T
bargaining owner is, ”7+5 while the lone owner holding out receivesm. Given that

the payoff to each owner from bargaining in the first period exceeds that from holding
out it is evident that both owners would bargain in the first period.

The threat of eminent domain therefore clearly solves the holdout problem and
the optimal taking compensation exceeds market value compensation. However, it is
worth noting that this result is only obtainable under assumptions of complete and
perfect information where both government and owner are privy to information about
the property’s market value, delay costs, and legal costs. Without knowing these values
the theoretical finding about the guaranteed effectiveness of eminent domain in
addressing holdouts simply ceases to exist.

The game presented in Figure 1 does exhibit a last mover disadvantage. This
occurs because the penultimate mover in the game is able to choose a price offer that
makes the last mover indifferent between accepting the payoff from this offer and going
to court. Given that the payoff from court settlement is worst for both agents in the

game, being the last mover yields the lowest possible payoff to the last mover. To see

63

this, reconsider the compensation bargaining game described in Figure 1. However,

assume the sequence of moves is now as follows: owner makes offer p L which can be

accepted or rejected by government. Government makes the counter offer pG if p L is

found to be unacceptable. Finally, the owner chooses between the payoff

from [70 , pG - V L and opting for court settlement m - f - VL . Figure 2 illustrates

this version of the sequential bargaining game.

Since the owner can solve the govemment’s problem just as well as government

can solve its problem, the owner should offer p L = m - I in the first period to end the

game. Thus if the sequence of the game requires the owner to move last, then the
property is taken at less than market value instead of the original compensation that
exceeds market value.

The nature of the eminent domain taking problem however does not lend itself
very well to the latter representation of the interaction between government and the
owner. Since the owner and government are not equally interested in this trade, there is
no incentive for an owner to go to court if a compensation offer is unacceptable. This
calls for government to always make the last move of enforcing eminent domain law by

seeking a court action to take the property as shown in Figure 1.

2.3 Bargaining Under Perfect but Incomplete Information

Although the rather strong assumptions of perfect and complete information in
eminent domain bargaining substantially simplify the bargaining problem, it is not hard
to imagine instances where at least one of these assumptions is violated. Under this

section of the analysis, the complete information assumption is relaxed to allow for

64

uncertainty where both‘the government and owner have incomplete information about
the payoffs in the game.

Actual bargains between government and owners occur under significantly
limited information resources compared to the informational assumptions made in the
last section of this paper. First, both the government and owner are usually not privy to
information about the true value of the property to each other. Further, delay and court
costs are not common knowledge. These informational deficiencies affect the offers and
counter offers of the bargaining game and the occurrence of bargaining breakdowns that
result in court cases.

A starting point to modeling problems of this nature is finding a framework to
describe the problem and estimate unknown parameters. The most difficult issue to deal
with here is the ceiling price of government. The ceiling price of government is
assumed to be private information for the government. The owner’s reservation price on
the other hand is really irrelevant to reaching bargaining equilibrium. This is mainly
because of the overwhelming bargaining power of government in the bargaining game.

To analyze the eminent domain bargaining problem under uncertainty,
reconsider the land market problem described in the previous section where a
government wishes to acquire an owner’s property for a public project. Again, assume

bargaining between the government and owner occurs in two sequential rounds. First,

the government offers the owner a price, pG . The owner responds by choosing an
asking price p L . If pl, = [)0 , the owner accepts the govemment’s offer and the game

ends. If p L > [70 , the government enters the second round of negotiations and decides

65

if p L is acceptable or not. The government accepts p L if p L S pc , where pcis the

ceiling price of government and pc < V0. It is assumed here that pc is private

information to government. Again, let the expected legal expenses from court resolution

and costs incurred by government as a result of delays in acquiring an identified

property be represented by E and d respectively. Further, assume the market value of
properties m is common knowledge. This implies that the bounds of the equilibrium
price (market price and ceiling price of government) are known to government but only
the property’s market price is known to the owner. The main problem that needs to be

solved then is the optimal pricing rule for the owner when govemment’s price ceiling is

unknown.

Given that the price ceiling of government pc is unknown to the owner, an

owner needs to come up with an estimate of pc prior to choosing pL. Eckart (1985)

derived a similar but more general model for a land assembly problem where a
developer acquires contingent land parcels from n-owners for a development project.
Here, the developer acquires the land parcels if all owners involved in the bargaining
agree to sell and abandons the project otherwise. Eckart addressed the problem of
owners’ ignorance about the maximum price the developer can pay by assuming owners
know some ‘prohibitive’ price that exceeds the price ceiling of the developer with
certainty. This assumption works for the hypothetical case addressed in Eckart’s model
but may be difficult to adapt to practical problems. Strange (1995) on the other hand
assumed that owners have prior beliefs about the value of the land to the developer and

can update these using Bayes’s rule whenever this is possible. The Strange (1995)

66

approach is more practical in terms of how owners’ prior beliefs are updated over time

but uninformative about how the prior beliefs are formed.

Here, suppose the owner has a prior belief #(pc) about the true value of pc.

Ideally, this is based on some educated guess of pc from information available to the

owner. For instance, this prior belief could be constructed from some estimate of the

maximum compensation government is capable of paying under the complete and
perfect information scenario, m+d +3. Based on the prior belief #(pc ) about pc the

owner can obtain a probability that the counter price offer is accepted by government as

7f,- (pL). Once government reveals [70 the owner updates the prior belief

#(pclpg) about the location of pc. If the owner judges from the size of the initial

government offer that the government is willing and capable of paying a high
compensation then the price ceiling estimate is revised upwards accordingly. On the
other hand, the price ceiling estimated is revised down if the initial government offer is

low. Assume the owner can adjust the estimated price ceiling to all possible offers from

government. The updated probability that government accepts p L is now given

as ”(FLIPG).

The probability that government accepts a counter price offer from the

owner 721le p0) is assumed to be decreasing in the counter offer price p L- This is
because the higher the value of pL the more likely it is that the counter price offer

exceeds the true price ceiling pc.

67

The equilibrium of the taking game under uncertainty here emerges as follows.

In the last move of the game, government chooses between accepting an owner’s

counter price offer p L and opting for court settlement. Again, for government not to go
to court, the payoffs from accepting the owner’s counter offer VG - p L must be equal
or greater than that from a court settlement, VG - m — f - d . Equating these two payoffs
gives p L = m + I + d implying that government goes to court only if PL > m+ If + d .

Moving a step backwards in the game, the owner chooses a counter price offer
p L to the initial offer from government. There is however no way for an owner to

know when the condition pL > m+ 8 +61 is satisfied for government in the last period
because the owner does not know at least one component of government’s payoff. In
this case government’s delay cost d is unknown. Otherwise, the owner will

choose p L = m + f + d precisely.

Owners are assumed to be risk neutral and thus maximize the expected wealth
value from this transaction. The objective of the owner therefore is to choose an asking
price to maximize a linear combination of the owner’s payoff in the two states of

ownership, when owner retains property rights over the property,

[1 — 7! ( P L I pG )](m - g) and when property right is transferred to the local government

at the asking price, ”(lepG )pL.

The owner’s problem is thus represented as,

Max H(m,pL) = [l—7r(lepG)l(m -€)+7r(pL|po)pL (1)
PL

68

The solution concept in use here is the perfect Bayesian equilibrium for
incomplete information games (Gibbons 1992, Mas-Collel A. et a1 1995). The first

order condition for equation (1) is given as,

 

a H
—— = — 7r m — l — + 7r = 0
6p L p L ( p L ) (2)
Rearranging and simplifying equation (2) gives,
7!
pL*=(m-€)- (3)
7’ p L

where, 7! p L 9* 0. The optimal owner asking price pL *is given by two terms; a

constant term that represents the net market price after accounting for anticipated court

costs (m - f) and a second term that depends on the owner’s perceived probability of

7!
government accepting a counter price offer 7: . Clearly, the higher the perceived

PL

probability of government accepting a counter price offer 7! the higher is the owner’s
optimal counter asking price. A major driver of the owner’s counter offer here is the

sensitivity of the owner’s subjective probability to marginal increases in the asking

price 7! p L . The more insensitive the owner’s subjective probability that government

accepts a counter price offer p L , the greater is the optimal counter price offer. For ease

of interpretation, this optimal owner asking price offer is rewritten as a function of an

elasticity of the owner’s subjective probability of government accepting a counter

offern relative to p L . To do this, rewrite equation (2) as,

 

p1. (tn—g) pL “I (4)

69

PLflp

 

 

Letting 8 ”p L = 7r , equation (4) can be rewritten as,
e
pL*=(m-€) ”pL (5)
l + c 7: p L

where 871']; L is the elasticity of the owners subjective probability that government

accepts a bid relative to the owner’s stated acceptance price p L. The market value is

reduced by litigation costsf because of the owner’s uncertainty about government’s

walk-away price. Litigation costs I thus represent the cost of uncertainty. The owner
needs to make provision for this litigation cost when choosing the counter asking price
because the probability of overshooting the government ceiling price and litigating in
court is strictly positive.

Note that the elasticity of owner’s subjective probability that government
accepts a counter bid 8 7177 L relative to the owner’s stated acceptance price p L is
negative. This implies that the more elastic the subjective probability of government

acceptance of a counter price offer, It relative to the owner asking price p L the lower is

the optimal asking price. This latter result indicates that the owner’s perception of

government’s sensitivity to marginal increases in the owner asking price pL is an

important determinant of the size of the optimal owner asking price pL *. Here, a

relatively high elasticity of owner’s subjective probability of government’s acceptance

of marginal increases in p L emanates at least in part from the right that government

wields to use eminent domain power. Further, the result in equation (5) is meaningless

for elasticity values above negative one (871p L > "1 ). Therefore, for reasonable

70

interpretations of the owner pricing rule in equation (5) there is a more than
proportionate government response to a marginal increase in the owner asking price.

From equation (5) it is clear that the lowest owner’s optimal counter offer asking

price p L* = (m - E) is observed as the elasticity of owner’s subjective probability of
government’s acceptance of marginal increases in p L becomes highly elastic

(approaches — oo ). On the other hand, the highest owner payoff of p L * = m + l + d is

m+l+d

m is satisfied. This result is obtained by

observed when the condition 5741 =

equating the result in equation (5) to the owner’s optimal asking price under the

complete and perfect information assumptions. Thus for K < m , the owner’s subjective

probability of acceptance is bound from above by negative one, 871p L < —1 .

Once the owner reveals p L government decides on the optimal first period

compensation offer, [)0 . The objective of government is to pay the lowest possible
price for the property in order to maximize government surplus from the taking. Given

that government’s initial price offer pG affects the optimal counter price offer of the

owner, this linkage can be exploited to achieve government’s objective of maximizing
surplus from the property taking. This approach is consistent with that adopted by
Eckart (1985) in discussing the optimal strategy of a developer bargaining with
colluding owners for the acquisition of complementary land parcels.

Here, it is assumed that government can form a rational expectation about the
owner’s subjective probability of government accepting a counter price offertr. This

assumption suggests that government does not make systematic errors in predicting the

71

 

owner’s counter price offer given government’s initial offer. Any deviation from
government’s foresight of the owner’s choice of a counter price offer is purely random
with zero expected value. Although this is a rather strong assumption to make about
government’s knowledge of the owner’s subjective probability it makes two
contributions to the analysis. First, it simplifies the analysis substantially. Beyond that,
it provides an opportunity to design a mechanism that enables information exchange
between government and the owner. For instance, a neutral negotiator can stand
between government and the owner to collect information on government price ceiling
and owner’s subjective probability and then follow the results derived in this paper to
compute an equilibrium taking price that will be acceptable to both parties.

Given the optimal counter offer rule of the owner, the optimal strategy of

government in the first period of the game involves choosing pG to solve,

Min lpr<m.£.po)=(m—r)———”———l (6)
PG ”19L

Consistent with the owner’s prior beliefs it is expected that the owner’s counter offer

asking price Will be increasrng rn government 5 rnrtral price offer > 0 . Now,

ape
since higher initial government price offers bring the owner closer to the government

price ceiling, the rate of change in the owner’s counter price with government price

2
offer rs expected to decline With the rnrtral government offer, 2 < .

0p G

Differentiating equation (6) with respect to [90 gives the condition,

72

 

 

a it 7r 7r
PL=_ po+ PLPG >0

(7)
5190 ”p1. [”le2

where ”PLPG > 0 ’flPG > 0 . The assumption in equation (7) that ”PLPG > 0

implies that the rate of decline of the subjective probability of government acceptance

of a counter price offer with respect to the owner asking price p L is decreasing in the

initial government offer, [)0 . The two probabilities, p0 and p L , have opposing
effects on the subjective probability of government acceptance. Equation (7) indicates
that the higher the initial government offer, the higher is the counter price offer from the
owner.

The second order derivative of equation (6) is given as,

72' 7Z' +7! fl
321% = P1. PGDG PG PIPG ,flptpoflpo 2”lvzpo’mprpo <

1

606’ [4,1]2 [4,1]2 [4,113 >

 

 

where 71' p L < 0 , and 71' PG PG < 0. The assumption that 71' PG PG < 0 indicates
that the owner’s subjective probability of government accepting a counter price offer is

increasing in the initial government offer p0 and does so at an decreasing rate. The
result in equation (8) shows that the rate of increase in the owner’s counter offer may
increase or decrease with [)0 depending on the sign of the middle term in the
equation. In either case, the reaction of government is to choose the lowest possible
value of pa in order to minimize its outlay on the property and maximize the surplus

from taking. With the restriction that the government must pay at least the market value

73

 

of the property in question as compensation this result suggests that government will
begin the negotiation by offering the market value as the appropriate compensation
value.

Therefore in a perfect Bayesian equilibrium of this game government offers

[70* = min the first period and the owner responds by choosing the counter
offer p L * with a subjective probability that government accepts pL * of

71' ( P L *6 PG *). This result indicates that there is no guarantee of avoiding negotiation

conflict between government and owner in this case. Much depends on how close the
owner’s guess of government’s price ceiling is to the true price ceiling. Further, the
more sensitive owners are to the probability of government acceptance of owner’s

counter offer the lower is the optimal counter offer.

2.4 Summary and Conclusions

This paper investigates the problem of holdouts and compensation bargaining
breakdowns in eminent domain takings. The paper is in two main sections. The first
section demonstrates the value of information in making eminent domain power an
effective tool in resolving holdout problems. The results indicate that under the
assumptions of complete and perfect information the threat of eminent domain power
guarantees resolution of owner holdouts and prevent bargaining breakdowns that lead to
litigations. Owners also receive the maximum compensation government is willing to
pay in equilibrium. This finding clarifies recent findings on the effect of the threat of
eminent domain on protracted eminent domain negotiations in Miceli and Segerson

(2007). In particular, the findings indicate that the threat of eminent domain is effective

74

 

in preventing delays in eminent domain takeovers only under restricted information
requirements that are not explicitly specified in Miceli and Segerson (2007).

Next, relaxing the informational assumptions to allow for incomplete
information in the bargaining game, an optimal owner asking price rule is derived under
uncertainty. This pricing rule is shown to depend critically on the owner’s subjective
probability of overshooting the government price ceiling. It is evident from the analysis
that the plausible and straight forward way of closing the gap between observed and

theoretical effectiveness of the threat of eminent domain in resolving holdouts is to
require compensation levels to at least equal the market value of properties pG 2 m,

Taking compensation values exceeding market value weaken owner incentive to
holdout or litigate by reducing the perceived probability of government accepting a
counter price offer 7: .

Using property market value as the lower bound of taking compensation does
not only discourage excessive inefficient takings that have adverse effects on private
investments in properties, but also reduces the incentive to owners to pursue legal
actions to stop takings or extract higher compensation. In any case, given that private
owners often do not place their properties on the market for sale before government
initiates takings, it follows that owners value their properties to be at least equal to the
market value. Under the assumptions made in this paper about court-imposed taking
compensations, no property owner has the incentive to litigate if offered at least the
market value of the property in question as compensation. This is because owner
payoffs from court-imposed settlements are always worse-off than market value

compensations offered at the start of the bargaining process.

75

-! vr-n“—fi

 

Second, to reduce owner overshooting that sometimes lead to litigation it is
important to require detailed financial information disclosure on the part of government,
and provide essential professional help to owners to make good use of this information.
Information on projected net flows of funds from proposed projects as well as a
breakdown of these net flows across different subsections of the proposed project site
can allow for estimation of the value of a given piece of property to the proposed
project. At minimum a reasonable range over which government’s price ceiling is
located can be estimated from this information. Upon imposing an appropriate
distribution the probability of government accepting a counter offer from the owner as
well as the elasticity of this probability with respect to marginal changes in the owner
asking price can be computed.

There are several powers at play here. First, apart from the power to use eminent
domain power, government has substantial information power in the bargaining game
since the maximum price payable to the owner is known only to government. Second,
government suffers a last mover disadvantage in the bargaining game since the owner
can make a counter offer choice that makes government indifferent between the payoff
to government from court settlement and that from accepting the owner’s counter offer.
This to some extent offsets information rents from government’s information advantage
in the bargaining game. However, an owner can only make use of this structural
advantage in the bargaining game if there is reliable information about government’s
delay costs. Without knowing the delay costs the owner is unable to determine

government’s payoff from court settlement. Although some of this information may be

76

 

gleaned by the owner from government’s first offer this may be highly inadequate to
exploit for decision making in many cases.

The net effect of the interaction of these relative powers of government and
owner on the equilibrium taking price depends on the relative weight of each
informational advantage in affecting the equilibrium terms of exchange. Overall, it is
clear that the owner has very little to go on to improve the owner’s payoff. The only
effective action open to the owner under the circumstance is to threaten protracted
bargaining and litigation to compel government to cede more of the surplus fi'om the
taking. This explains a somewhat irrational decision making observed by some authors
about practical protracted bargaining problems. As noted by Ausubel et al (2002), the
central issue in protracted bargaining problems is to explain the decision by bargaining
agents to engage in lengthy bargains and legal battles even when it is evident that the
parties could settle at the same terms without the protracted dispute. The general
explanation in the economic literature for this behavior is that bargaining agents use
delays as a strategic signaling response to the presence of incomplete information
(Feinberg and Skrzypacz 2005, Bac 2000). The results from this study generally
indicate that delays in eminent domain taking bargainings are partly signals from
owners dissatisfaction with government’s compensation offer, and partly due to pure
mistakes made by owners in choosing counter price offers because of limited

information.

77

 

QR I U> 034 I U>
43 I N84 Q.» | UK

   

E3964 Eooo<

. ®
5» SEC emwoomomle on 6&0

wlhlfilbap
VISTE 4% Doomed

 

 

 

.EE ago—2 2585950 :2— :ouufiueufi Botan— 65 89. :50 .835 55a am :55: om A 95$...—

78

waflbm... t.d.r.vwfl__ L9: (IAV =.e-.)lu run-«allllllhll- l I

 

 

 

 

 

cab..— 852 .855 :2.

 

 

 

nets—Ecua— uootom one 80— :30 .325 3:? am :55: um .N charm

79

References
Ausubel L.M., Crarnton P., and Deneckere R.J., Handbook of Game Theory, Vol. 3,
Amsterdam: Elsevier Science B.V., chapter 50, 2002

Bac M. Signaling Bargaining Power: Strategic Delay Versus Restricted Offers, 2000, Vol.
16, issue 1, 227-237

Bell, A. and Parchomovsky, G., Bargaining for Takings Compensation 2005, U of Penn
Law

School, Public Law Working Paper No. 06-12, Available at SSRN:
http://ssm.com/abstract=806l 64

Eckart W., On the land assembly problem, Journal of Urban Economics, 1985, Vol.18,
Issue 3 (November), 364-378

 

Epstein R.A. Takings: Private Property and the Power of Eminent Domain, Cambridge I-
Mass. Harvard University Press, 1986, Pp xi + 362

Fennel L.A. Taking Eminent Domain Apart, 2004, Michigan State Law Review, 95 7

F einberg Y. And Skrzypacz, Uncertainty about Uncertainty and Delay In Bargaining,
Econometrica, Vol. 73, No. 1 (January, 2005), 69—91

F udenberg, D., Tirole, J .: Sequential bargaining with incomplete information. The Review
Of
Economic Studies 50(2), 221—247 (1983)

Gul F. and Sonnenschein H. On Delay in Bargaining with One-Sided Uncertainty
Econometrica, Vol. 56, No. 3 (May, 1988), pp. 601-611
Gibbons R., Game Theory for Applied Economists, Princeton University Press, 1992

Hoy M., Livernois J ., McKenna C., Rees R., and Stengos T., Mathematics for Economics,
Second Edition, MIT Press, Cambridge Massachusetts.

Mas-Collel A. Whinston M.D., and Green J.R. "Microeconomic Theory", Oxford
University
Press, USA (June 15, 1995)

Menezes F. and Pitchford R., A model of seller holdout, Economic Theory, Nolume 24,
Number 2 / August, 2004, 231-253

80

Miceli T.J. and Segerson K, A Bargaining Model of Holdouts and Takings, American Law
and Economics Review 2007 9(1): 1 60-1 74

Munch P. An Economic Analysis of Eminent Domain, The Journal of Political Economy,
Vol. 84, No. 3, (Jun, 1976), pp. 473-497

Parchomovsky G. and Bell A., Taking Compensation Private, 59 Stan. L. Review. 871,
2006

United States Constitution, Come] university law school, 1791,
http://www.law.comell.edu/constitution/constitution.overviewhtml.

 

Parchomovsky G. and Siegelman P. Selling Mayberry: Communities and Individuals in
Law

and Economics, 92 Cal. L. Rev. 75, 128-29 (2004)

Rubinstein A. Perfect Equilibrium in a Bargaining Model, Econometrica, Vol. 50, No. 1
(Jan, 1982), pp. 97-109

Strange W. Information, Holdouts, and Land Assembly, Journal of Urban Economics, 38,
1995, 317-332

81

 

'r-L

Essay 3

Cost Function Estimation in the Water Industry — Functional Forms and

Efficiency Measures

3.1 Introduction

Cost function estimation is an important component of efficiency analysis of
firms when multiple outputs are involved (Greene 1993). Managers often have to make
decisions on output expansion, input mix, and even location of plants based on the
interactive effects of output and input prices. Since there are usually many efficiency-
impacting factors at play in most production processes, relatively technical cost function
analyses are necessary to provide reliable information upon which managerial decisions
can be based. In the potable water provision industry for instance, while the per unit
cost of water extraction and treatment may increase with output as exploitation moves
to less accessible and lower quality water resources, the per unit cost of water
production may also decline with output expansion due to scale economies. The rising
cost of water extraction, treatment, and transmission may thus offset partially,
completely, or even more than offset cost-savings that may be derived from scale
economies. An estimated cost function for water provision therefore serves as an effect
aggregating tool that helps to extract the net effect of cost-impacting factors and
provides information to make decisions on output levels, efficiency-improving input
substitutions, and efficient system size.

Since the true production technology is unknown in most empirical estimation

problems and needs to be approximated, flexible functional forms play a valuable role

82

 

in cost function estimations (Tishler and Lpovetsky 2000, Salvanes and Tjotta, 1998).
A function is considered to be flexible when its shape is restricted only by theoretical
consistency (Sauer, et al., 2006). Some fiequently used flexible functional forms
include the Box-Cox, Box-Tidwell, Leontief, Minflex-Laurent, and the translog forms
(Shaffer, 1998). Among the class of flexible fmetional forms, the translog function
(Christensen et al. 1973) has emerged as one of the most popular flexible functional
forms used for efficiency analyses that involve cost function estimation (Salvanes and
Tjetta, 1998) 2001).

Recently however, Shaffer (1998) discussed a hitherto unknown weakness in the
translog’s ability to adequately model data that exhibit monotonically declining
average cost functions. Analytical and simulation results presented by Shaffer indicate
that the translog tends to produce spurious finite minimum efficient scale (MES) results
even when the true MES is infinite. This implies that application of translog fimctional
forms to data by researchers in empirical studies may be producing coefficient
estimates consistent with the imposition of U-shaped average cost structure on the data
when the true average cost represented by the data declines monotonically with output.
The biased estimates produced by such functional form misspecifications provide
misleading information upon which management decisions are based. Shaffer
introduced the Hyperbolically Adjusted Cobb-Douglas (HACD) as an alternative
functional form specification that is capable of differentiating between the regular U-
shaped average cost function and the monotonically declining average cost functions.
This suggests that the fit provided by the HACD can be expected to be at least as good

as that of the translog for data exhibiting a monotonically declining average cost. On

83

 

the other hand, when the data exhibits a U—shaped average cost these two functional
forms are expected to be competitive.

This paper estimates two multi-product total variable cost functions with two
inputs (capital and labor) using the translog and HACD functional forms. Estimates of
cost economies, input demand functions, and Allen-Uzawa partial elasticities of
substitution values are also computed for potable water provision across the US.

To assess the relative fit of the two functional forms to the data Vuong (1989)
and Mizon and Richard (1986) functional form tests are employed. Bontemps and
Mizon (2008) distinguished between these two competing functional form tests by
classifying the Vuong (1989) test as a model selection test and the Mizon and Richard
(1986) as a model comparison test. The model selection test procedure selects a winning
model to minimize or maximize a given criterion. This implies that the preferred or
winning model does not allow for the possibility that the alternative models considered
collectively contain information that could lead to the development of a better model.
On the other hand, the model comparison procedure uses an encompassing principle
that considers the effectiveness of each model in accounting for the explanatory power
of the other competing models. The preferred model in this case therefore incorporates
useful specific characteristics of the alternative models not selected.

The results generally indicate that the HACD provides a better fit to the data.
Results for the HACD indicate that a one percent increase in the price of capital and
labor results in 0.06 and 2.856 percent decline in the quantity of capital and labor
demanded respectively. Using the translog parameters, the same increase in price

results in a 0.05 and 0.09 percent decline in the quantity of capital and labor demanded

84

 

respectively. Finally, the HACD provides statistically significant cost economies
estimates that represent economies of scale to water provision. In particular, a one
percent increase in the quantity of water and population served increases costs by 0.48
and 0.43 percent respectively. The translog parameter estimates on the other hand point
to diseconomies of scale. A one percent increase in the quantity of water and population
served increases costs by 4.7 and 2.31 percent respectively.

The remainder of the paper is ordered as follows. The next section presents the
theoretical framework of the paper. This is followed by discussion of the data and

research hypothesis, results, and conclusions.

3.2 Theoretical Framework

A cost function represents the minimum cost of producing a given output with
given input prices (Mas-Colell, Whinston, Green 1995). Estimated cost functions
provide valuable information about the performance of firms. Useful performance
measures often extracted from estimated cost functions include pairwise input
elasticities of substitution, cost economies of scale values, and input demand functions.
These performance measures are common in cost function analyses partly due to the
difficulty of interpreting parameter estimates from flexible cost functions (Andrikopolos
and Loizides, 1998 Bhattacharyya, et al., 1995) as marginal effects.

The Pairwise input elasticity of substitution values shed light on how efficiently
the firm is using each input relative to the other. This represents a description of the
relationship between the various inputs employed in the production process and

indicates whether two inputs can be considered as substitutes or complements. The cost

85

 

economies values on the other hand measure the percentage change in total variable
cost as a result of a one percent change in outputs while the input demand function
indicates the percentage change in inputs used as result of a percentage change in the
input prices (Mas-Colell, Whinston, Green 1995).

The definition of water production outputs in cost ftmctions can significantly
affect the estimated cost economies values. Output in a network industry like water may
be defined in terms of volume of water produced, number of customers served, and
scope of services (Torres and Paul 2006). Thus in the water provision industry, output
increases may involve increases in the volume of water due to increased demand by
existing users, increase in number of water users, or increased scope of services. When
production increases due to higher demand of existing customers, then utilities may be
expected to enjoy some economies of scale. However this is not likely when the volume
increase is associated with an increase in the number of customers. This is because the
cost of extending services to additional customers may cause costs to increase more
proportionately than the cost economies attributable to output expansion.

Cost function analysis is usually based on the assumption that firms choose
inputs to minimize production cost. Determining whether empirical results conform to
cost minimization requires testing and verifying that the regularity conditions are

satisfied. These regularity conditions (Salvanes and Tjotta, 1998) are listed as follows,
1. Non-negativity of production costs, C( y, p) > O, V p > 0, y > 0. This
condition simply states that no positive output can be produced without
incurring some positive cost.

2. Monotonicity in prices, C(y, p') > 00’, P), for P'> P

86

 

3. Cost is homogeneous of degree one in prices,
c( y,tp) >tc( y, p), for 1‘ >0. This indicates that when all input prices

change by a given proportion total cost changes by the same proportion.

4. Cost is strictly increasing (monotonic) in output,
C(y', p) > C(y, p), for y'> y . In other words, marginal cost cannot be
negative.

5. Cost is concave, continuous and differentiable in prices (p)

F rom duality theory, functions satisfying 1-5 satisfy the requirements for a cost function
and for each of these cost functions there exist a production technology from which this
cost function can be derived (Hunt 1980).

Given that symmetry and linear homogeneity are imposed a priori on the cost
functional forms, the conditions left to be verified are non-negativity of costs,
monotonicity in input prices and output, and concavity of the cost function in input
prices. Since the dependent variable is the natural log of total cost, the non-negativity
condition on the cost function is automatically satisfied. Monotonicity is verified by
ensuring that all estimated marginal costs and cost elasticities are strictly positive.
Finally, to assess concavity of the cost function in input prices, the Hessian matrix must
be negative semi-definite.

The Hessian, H, which is a matrix of second order derivatives of the estimated

cost function with respect to the inputs is defined as,

87

 

lF,

 

 

 

 

 

 

( 32c 62c ‘
H: 6pK5PK 5PK5PL
620 620 (I)
\aPLaPK 5191,6101. 2

One common problem with cost function estimations in the water industry and
empirical studies in general is the violation of these regularity conditions (Diewert and
J ., 1991, Salvanes and Tjotta, 1998, F abbri and Fraquelli, 2000). In a study of the Italian
water industry, F abbri and Fraquelli (2000) observed that the technology underlying the
water industry is not characterized by the conditions of regularity in costs. In a
commonly cited study by Salvanes and Tjotta (1998), Salvanes and Tjetta reexamined
the US Bell cost function estimated in Evans and Heckman (1984) by calculating the
region where the cost function meets the regularity conditions. The study concluded that
the estimated function is not a valid cost function since it failed to meet the non-
negative marginal cost condition in most of the test region.

Failure to satisfy the concavity condition is particularly very common in
empirical cost fimction estimations (Christopoulos et al, 2001, Rao and Preston 1984,
Conrad and Jorgenson 1977). Violation of this condition generally implies that the data
being modeled does not exhibit the theoretical assumption of cost minimization. In
fact, in a survey of some recently published agriculture-related papers that made use of
the translog functional form Sauer et. al. (2006) found that the estimated translog
functional form in all seven publications failed to fulfill at least one local regularity
condition at the sample mean. Further, all the estimated functions fail to fulfill the

curvature requirement of quasi-concavity.

88

 

The econometric model presented in the next section is used to estimate the two
cost functions and to verify the theoretical consistency requirements described here in
the theoretical framework. The methods and steps taken to verify satisfaction of these

theoretical constraints are also described.

3.3 The Econometric Model

The econometric model considers water producing firms that use labor and
capital inputs to transform untreated water into outputs measured by volume of water
produced and number of customers served. Using the economic theory of duality
between production and cost functions allows for observable input prices and outputs to
be used in analyzing these production activities without knowing the underlying
technology of production (Mas-Colell, Whinston, Greene 1995). As previously noted,
the two functional forms used for the estimation are the Tanslog and the HACD.

Starting with the translog model, the model estimated is specified as,

InC= a+ 2 {3,112};- +%ZZ 5,-1- 1n);- 1an +Zwi1ng +%ZZw,-j1nP,-InPj +
1 l _] l l J

222 jiInPjInYi +5 (2)
j i
where C represents total variable cost of water production. The explanatory variables in
the model include data on a vector of input prices (P), and a vector of output definitions
(Y). The input price vector covers the costs of capital (k), and labor (I) while the vector

of outputs on the other hand comprises volume of water (q) and population served (5).

89

 

 

To impose continuity on the estimated translog cost function the following

symmetry conditions are imposed,
flij =fljiawij =(0ji, lij = ji forall ij.

Imposition of this symmetry condition is based on Young’s theorem (Jehle and Reny
2003) which indicates that these coefficients are the same (i.e. the order of the
interaction terms is irrelevant). In particular, Young’s theorem indicates that
differentiating this cost function with respect to labor and then with respect to capital
should give the same result as differentiating in the reverse order so long as both cross-
partial derivatives are continuous.

To satisfy theoretical assumptions of linear homogeneity of the cost function in

input prices the following additional restrictions are imposed,

Zn),- =1, Zwij =0 and 2’1]? =0

1' =1 1' =1 i=1
Using Shephard’s (1970) lemma, the derived input demand functions can be obtained

by differentiating equation (2) with respect to the input prices to obtain,

6111C " , m .
Mzzmzwi+2wljln Pj'i'Z/ijjlnYi (3)
I I

The vector of a given input used is therefore a function of the vector of other input
prices and the output vectors.

Two measures of input elasticity are employed, the Allen-Uzawa partial

elasticities of substitution between inputs 1' and j, Z I] , and the regular price elasticities

of input demand :1] The Allen-Uzawa elasticity of substitution measures the

90

 

1E:

percentage change in factor proportions due to a change in marginal rate of technical
substitution (input price ratios) while the price elasticity of input demand represents

elasticity of the ith input (X ij) with respect to the price of the jth input PJ-

alnX
[:6]: l aha]. As noted by Segerson and Ray (1989) these two elasticity measures

 

are not the same (except for the CBS and Cobb-Douglas). The elasticities are defined as 1"
follows,
2.. .._ (0)1'1'+Mi2 —Mi)/]Wi2’f0ri=j ziijwaor izj (4)

I — i' = . .

j (60,-!- +MiMj)/A/I,~Mj,f0ri¢j Zz'ij,f0"l¢J

where 51'; is own input price elasticity of demand while 6,-1- is the input cross partial

elasticity. The computed elasticity of substitution values may be positive (input
substitutability) or negative (input complementarity).

Finally, the estimate of economies of scale is computed as,

alnC n n
”y' =———=fl.-+Z flij 1“( Yj)+z ’4‘1'1'1“(PJ')(5)4

For proportional increases in volume of water produced and population served the

resultant economies of scale is given by the sum of cost economies associated with each

output type,

 

 

 

4 This can also be looked at in levels form as

U = U + U
y ‘1 5
Turning to the HACD model the estimated model is specified as,

_ l 1 1
MC: WZBilnl' +21% R + 2201] W +2vilnlg+E Z :2 :qunBInPj +
I 1 1¢jj¢i i i j

+8
InYi

 

2.2%
j l

The definitions of cost, output, and input vectors are exactly the same as defined for the
translog model. The symmetry conditions here are pij = ,0 ji, Uij = U jia #1)“ = ,Uji

for all ij where ii j and the linear homogeneity conditions

are,ZUi =1, 2% =0 and Zflji =0.
i=1 i=1 i=1

Comparing equation (2) and equation (6), it is clear that the difference between
the translog and HACD is in the representation of the nonlinear terms of output. While
the translog uses quadratic terms to accomplish this, the HACD makes use of inverse

output terms.

Using Shephard’s lemma (1970), the derived input demand fiinctions can again

be obtained by differentiating (6) with respect to the input prices to obtain,

aln C l

N-=——%u-+ u--InP-+ ~————-
’ aInP,» ’ 21:” J ZifllenYi (7)

The Allen-Uzawa partial elasticities of substitution between inputs 1' and j ay- , and the

corresponding price elasticities of input demands, 6 ,1 are computed as,

92

 

(an +N.-2 -M-)/Ni2,fori =1 ZizNiafori=j

(afiiviw/MAo-Jonm 1] 1’ J a»

where Eii is own input price elasciticity while 6 ij is the cross partial input price

elasticity.
Finally, the equivalent of the translog’s estimate of cost economies is derived

as,

 

 

BInC
gYi zaInY— — Qi-Zm— 2 2:ij ZZZ/11'1"]:— 2 (9 )

i (lnYi) i¢jj¢i (In Y')2 Ian j¢ii (In Y')

The estimation is done using the Seemingly Unrelated Regression Estimation
(SURE) method. This estimation approach is appropriate for analyzing a system of
equations with cross-equation parameter restrictions and correlated error terms as in this
paper. Although the equations here are not estimating the same dependent variable, they
share some independent variables, use the same data, and may have errors that are
correlated across the equations. Estimating the system of equations separately with OLS
(ignoring correlation of disturbances) yields inefficient but unbiased and consistent
estimates for each separate equation. SURE exploits the contemporaneous information
in correlated errors to achieve greater efficiency in the estimates. The cost functions are
estimated along with the capital share equation only to avoid the problem of singularity

of the variance-covariance matrix.

93

n. r.

 

r'l“o

Theoretical Consisteng

The first step in evaluating the estimated model results is to verify the
theoretical consistency requirements. Examining theoretical consistency of the
estimated model requires checking the regularity conditions. For both the translog and
HACD models, the regularity condition that costs be strictly positive is met through the

choice of functional form since exp(ln(C)) is strictly positive for all feasible(Y, P). The

estimated cost functions are also homogeneous of degree one in prices since this was
imposed apriori. The next set of conditions requires that marginal costs be nonnegative
and that the estimated cost function is non-decreasing in input prices.

The final regularity condition requires that the estimated function be concave in
input prices. The estimated cost function is concave in input prices if the Hessian matrix

VppC is negative semi-definite. This condition can be assessed using the computed

elasticity of substitution values. For the estimated cost fimction to be concave in prices,

the own partial elasticity of substitution values, (Z,,,a,,) should be negative

(Andrikopolos and Loizides, 1998). Alternatively, concavity may be assessed by
constructing the matrix of second order derivatives of cost with respect to input prices

(Chew et al. 2005),

2 .
a C 6x- 8x- P °x- C C

= l = I .1 pl 1 =6UMI.___ (10)
apiapj 5p} 5Pj xi C 17in 17in

 

 

Since equation (10) is a symmetric matrix the matrix that needs to be evaluated for

concavity can conveniently be presented in quadratic form as,

62C 1

1 , =—
” C(p)

H--= p
'1 C(p) apiapj

 

P'V2C(P)p=5szi (11)

94

 

The matrix in equation (11) is negative semi-definite when the HACD is
concave. A negative semi-definite matrix has non—positive diagonal elements and the

principal minors alternate in sign.

Imposing Quasi-Concavity

 

As previously noted, violation of the concavity condition is common in cost I.
function estimations. To obtain parameter estimates that are consistent with the
objective of cost minimization, concavity may be imposed locally on the cost function.
In this paper the results derived in Jorgenson and Fraumeni (1981) are followed. it

Imposing quasi-concavity here then requires replacing the elements in equation (1) by

the condition,

0,-1- = —-(DD'),-j + 01-6,]- + 1)in

where 5,1 = 1 if i = j and 0 otherwise and (DD'),j is the if —th element of the matrix

_(DD)_ (dkk 0],,[dkk dkl] =[’dkkdkk ‘dkldkk J (12)
dlk d1: 0 a’11 ‘dlkdkk ’dlkdkl -d11d11

Imposing this concavity locally requires choosing a point of approximation. The mean

is chosen as the point of approximation in this study. Substituting the results in equation

(12) into the HACD model specified in equation (6) gives the new model to be

estimated as,

95

_1_
Iné

+ UkInPk + 01 MP, -l—:(-dkkdkk+ Uk—Ukuk)lnPk2 +

1 +
Ian

 

InC=a +flq1an + ,BSInlg + éflqq + éflss

1
zfl 5‘1 1,111,1an

1 1
E(—dkldkl‘dlldll + ¢1-¢1¢1)lnP12 +§(-dk1dkk-¢k¢1)lnPkP1 +
xlqunPk Ian +illslnP11an + 8 (13)

The resultant model to be estimated (equation 13) is nonlinear in parameters. The model
is therefore estimated using the nonlinear estimation method in Stata. Satisfaction of all
the regularity conditions establishes a common ground for comparison of the relative

performance of the two functional forms to proceed.

Test of Functional Form Fit

The Vuong (1989) model selection test is employed to compare the performance
of the two functional forms. The Vuong test is a likelihood-ratio-based test that tests the
null hypothesis that the two estimated models are equivalent (i.e. are equally close to

the true model). The test statistic is given as,
In 6 114(1)
6 TRAN
W = ’ *
\/ n se { In |: g—i’iCDu—J }
g TRAN

where ’5 HACD represents the log-likelihood value from the HACD model, 5 TRAN , the

 

corresponding value from the translog model, andn is the sample size. Here, a positive
test statistic suggests that the HACD is closer to the true model than the translog. On the

other hand, a negative test statistic indicates that the translog is closer to the true model

96

 

than the HACD. Vuong (1989) has shown this test to be asymptotically distributed as a
standard normal under the null hypothesis. This implies that the null hypothesis can be
tested using critical values from the standard normal distribution.

Next, the Mizon and Richard (1986) [MR] test is used to test the fit of the two
functional forms. The MR test constructs a comprehensive model that contains one
model as a special case and then tests the restrictions that represent additional
parameters to the model being tested. For instance, testing whether the translog
functional form fits the data better than the HACD would require adding variables that
appear in the HACD model (but not in the translog model) to the translog model and
testing for the joint significance of the additional restrictions. Finding these new
variables jointly significant implies that the translog functional form is deficient in
adequately modeling the data.

Thus to evaluate the fit of the translog model the new model estimated is

specified as,

In C=a+ZBiInY 4226,71“;- InY +20),- InP +2220?- InPiJ-InP +

i j t J
1

)wInP-IY- -— —+ 14

€12. 1’ J "”12"” Ii. .1! ijInYIInY- JZZquInY- 8 ( )

This comprehensive model is made up of the translog model and four additional terms

from the HACD model. The Wald test is used to test the restriction,
77,- = ,0“- : .11}, = 0. This involves performing a joint test on the six additional

parameters introduced from the HACD fimction.

97

A well known characteristic of these non-nested functional form tests is that
rejection of one functional form does not necessarily mean the other functional form is
the correct model. In fact there may or may not emerge a winning functional form out
of the pair of functional forms being tested. Both functional forms may be rejected, one
rejected and the other accepted, or both may be accepted (Wooldridge 2006).

I The test is reversed with the HACD now being the base model augmented by
additional terms from the translog. The medel specification for this reversed test is

given as,

InC= a+§6i1nlf + g: mm); :20 ,1 _+ZU’1nYInY InP + $221,]. InPInP- +

#11:; i i j
”InP

j j 1'

Here, the HACD function is the nested functional form and the Wald test is used
to test the restriction, ,Bij = 417 = 0. Again, finding this joint restriction statistically

significant implies that the HACD functional form is deficient in adequately modeling

the data.

3.4 Data

Cross-sectional data covering 73 water and waste water utilities in the US is
employed for this analysis. The data is taken from the 2004 General Utility Information
and Basic Utility Operating database of the American Water Works Association
(AWWA, 2004). The survey data covers utilities that serve populations ranging from

1,200 to 9,000,000.

98

'.""';'"' "FL

 

 

Stratified random sampling is employed in the survey data collection. The data
sample includes states that voluntarily participated in the AWWA survey. The survey
list was later supplemented by wastewater utilities from the National Pollutant
Discharge Elimination System (NPDES) database. This list includes companies that
provide waste treatment services and may or may not be providing potable water
services. Extension of results from this study to the population of water and wastewater mm
services in the US. must therefore be cautiously done since it is not clear if any factor

systematically affected the decision of companies to respond to the survey.

 

Total variable cost is in 2002 US. dollars. Variables representing the input price 1.,
vector are capital and labor price vectors while variables in the output vector are volume v
of water, service population, and a dummy variable for scope of services (whether at
least one other service is provided along with potable water).
The input prices are computed in 2002 US. dollars. Price of labor is computed
as the ratio of total personnel expenses to the number of full time workers. Price of
capital on the other hand is defined as the weighted average of the cost of equity and
after-tax cost of debt (Modigliani and Miller 1958, Miller and Modigliani 1963, Miles
and Ezzell 1980). The weights applied are the respective ratios of equity and

accumulated debt to total capital. The cost of capital is therefore computed as,

price of capital = cos tof equity" (E / D+ E)) +After— tax cos tofdebt‘" (D/ D+ E))

where, E = equity of water utility (i.e. total assets less total liabilities)

D = debt of water utility (revenue bonds and financial notes)

99

 

Cost of equity refers to the opportunity cost of investment and is estimated using results

from the capital asset pricing model as,
cos t of equity 2 risk free rate + beta (risk premium) .

The average monthly discount window borrowing rate for 2002 is used to
represent the risk-free rate. This discount window borrowing rate is the rate at which the
Federal Reserve banks lend money to depository institutions like banks and US
agencies of foreign banks. This data was taken from the economagic database
(Economagic 2002). The estimated average value for 2002 used in this computation is
1.17.

Beta is a measure of the volatility of stocks in the water industry relative to the
rest of the stock market. The average beta for the water sector (0.73) estimated by
Damodaran is used (Damodaran 2007). It is estimated using the stock returns of 16 of
the largest investor-owned water companies in the US. A risk premium of 5.5% used
by Damodaran is retained. Cost of debt is computed in the data as total interest
payments / total debt, and debt is defined to include both short and long-term debt (but
not accounts payables). Since cost of debt expense is tax deductible, this adjustment is
made in the computation using the average effective tax rate (29.78%) computed by
Damodaran for the water sector.

The volume of water variable is represented by total gallons of water produced
(in millions of gallons) and/or purchased from other providers while population served
refers to total water consuming population served by the water company. For water

companies providing both retail and wholesale water services, total population

100

 

 

represents the sum of the population served by retail service and population in
communities purchasing bulk/wholesale water from them.

Other variables considered in the model are Water loss and Ownership. Water
loss refers to the percentage of treated water that is unaccounted for in the 2002
operating period. This is the difference between what is produced and what is used by
consumers. Ownership is a binary variable defined to equal 1 if the company in
question is government owned and 0 otherwise.

Table 3.1 below provides additional information on the variables in the model.
Summary statistics covering sample size, unit of measure, mean, standard deviation,

minimum, and maximum value of each variable are shown.

3.5 Results

Results of the model estimations appear in Table 3.2 below. Two sets of results
representing the translog and HACD parameter estimates are shown. The performance
parameters estimated from the regression results are presented and discussed. Further,
the models are evaluated for functional form fits, Heteroskedasticity, and theoretical
consistency requirements for cost fimctions.

Omitted from the final results are three dummy variables (ownership,
wastewater, and water loss). Ownership was initially included to measure the impact of
private water company ownership on cost levels relative to public ownership. The
variable ‘wastewater’ was included to measure the impact of joint service provision
(water and at least one other service) on costs. Most companies providing more than

one service provide water and waste treatment services. Water loss represents the

101

 

 

percentage of water lost in transit from the water company to the consumer. Such losses
are attributable mainly to pipe bursts. All three turned out to be statistically insignificant
and were dropped.

Allen-Uzawa and input demand elasticity values are computed from the estimated
results shown in Table 3.2 above. These values and their associated standard errors
appear in Table 3.3 below. The standard errors provide precision information about the
elasticity estimates and are used to evaluate the elasticity estimates for statistical
significance. These standard errors are computed following the derivations in (Toevs
1982).

The computed Allen-Uzawa partial elasticity of substitution values for the translog
model indicates that a one percent increase in the price of capital results in a 0.05
percent decline in the quantity of capital demanded. For labor, a one percent increase in
labor price results in a 2.099 percent decline in quantity of labor demanded. Thus
although both inputs face the conventional negatively sloped input demand function,
demand for labor is more elastic than that for capital. The positive estimated cross
partial elasticity of substitution (0.287) indicates that the two inputs are complements.
Here, a one percent increase in the price of capital results in 0.287 percent increase in
quantity of labor used. This is a reasonable finding since capital cannot be expected to
stand alone in water production. The estimates for own and cross input price elasticities

of demand, 5,7 and 6,-1- from the translog model confirm the earlier findings for the

Allen-Uzawa elasticity of substitution values. Although these values are smaller, they
also point to a negatively sloped demand for both inputs with the input demand for

labor being more elastic.

102

 

Turning to results from the HACD model, it is observed that a one percent
increase in the price of capital results in 0.066 percent decline in quantity of capital
demanded while a one percent increase in labor cost results in 31.822 percent decline in
quantity of labor demanded. Comparing these results to that of the translog, it is evident
that while the elasticity of substitution estimates for capital are quite identical for the
two models the corresponding estimates for labor are quite different. In particular, the
elasticity of substitution estimate for labor in the HACD model is much larger than the
corresponding estimate in the translog model. In summary, all the elasticity estimates
from the HACD are greater than the equivalent estimates from the translog model.
Given that the HACD provides a closer fit to the data than the translog the HACD
estimates can be considered to be relatively more reliable.

Estimated input share values for the two models indicate that the water and
waste industry may be very capital intensive. The estimated share of capital in total cost
is 0.956 in the translog and 0.913 in the HACD. The corresponding values for labor are
therefore 0.044 and 0.087 respectively.

Cost economies estimates computed from the estimated models are presented in
Table 3.4 below. The estimated parameters give conflicting results about economies of
scale to water provision. In particular, a one percent increase in the quantity of water
and population served increases costs by 0.48 and 0.43 percent respectively in the
HACD. These represent measures of economies of scale for increases in quantity of
water and population served. Considering the cumulative change in costs for changes in

both measures of outputs, a one percent simultaneous increase in quantity of water and

103

 

population served increases costs by 0.91 percent. This still constitutes cost economics
for size expansion.

The translog parameter estimates on the other hand point to diseconomies of
scale. A one percent increase in the quantity of water and population served increases
costs by 4.7 and 2.31 percent respectively. These two sets of results exemplify how the
choice of functional forms may drastically influence conclusions and policy
recommendations that are obtained from empirical cost function analyses.

To evaluate the models’ satisfaction of regularity conditions that are not

imposed apriori or met through the choice of fimctional form the first order partial
derivative of each estimated function with respect to lnYi and lnPi is computed for
each data point. The computed changes in cost with respect to the output measures and
input prices are strictly positive, satisfying the respective regularity conditions that the
estimated total variable cost function is increasing in prices and outputs. This satisfies
the requirement that the marginal cost is nonnegative and that the estimated cost
function is non-decreasing in input prices.

The final regularity condition requires that the estimated function be concave in

input prices. Looking at the partial elasticity of substitution values for the estimated
translog model below ( Z ij ), it is clear that the concavity condition is satisfied since the
principal diagonal values (as defined in equation 1) are negative. The equivalent

estimates for the HACD model (aii) indicated that the model failed to satisfy the

concavity requirement since the own partial elasticity value for capital turned out

positive. Concavity was imposed on the HACD function following the results in

104

 

equations (12) and (13). The new set of results is shown in Table 3.2 along side the

parameter estimates from the translog model.
Overall, the R 2 measures suggest that the HACD provides a better fit to the

data. For functional forms with the same dependent variable, the adjusted R 2 is an

appropriate basis for comparing the relative fit of non-nested functional forms

 

(Wooldridge 2006). Here, since the number of variables in the two models are equal, '2'
the R 2 provides an adequate basis for comparison of the two functional forms. Also,

the root mean square percentage error of 0.36 from the translog model as against 0.32

from the HACD shows that deviations of in-sample predictions from the HACD are i

smaller than those from the translog.

The Vuong (1989) test is employed to determine if the difference in fit of the two
models is statistically significant. The log-likelihood values obtained from the estimated
models are 116.861 and 109.75 for the HACD and translog models respectively. Given
the sample size and standard error of log-likelihood differences as 73 and 1.037
respectively the computed test statistic is 0.863. The positive test statistic here indicates
that the HACD marginally fits the data better. To determine if the difference between
the two models is statistically significant, the computed test statistic is compared to the
5 percent critical value from the standard normal distribution of 1.96. Since the
computed test statistic is less than the critical value it is concluded that the null
hypothesis that the two models are equivalent cannot be rejected.

Results from the Mizon and Richard (1986) test confirm the earlier findings
from comparing the R 2 from both models. The Mizon and Richard (1986) test

involves using the Wald test to test the restriction, 77,‘ = pij = ,uj,‘ = 0 in equation

105

(l 4) for the translog model, and flij = llvji = 0 in equation (15) for the HACD model.

Evaluating the fit of the translog involves performing a joint test on the six additional
parameters introduced from the HACD function. The computed chi-square statistic of
14.16 was obtained with an associated p-value of 0.028 indicating that the null
hypothesis that these additional restrictions jointly equal zero is rejected. This suggests
that the translog specification inadequately accounts for nonlinearities in output in the
cost function. The corresponding test for the HACD model gives the chi-square statistic
value of 10.97 and a p-value of 0.140. Here, the additional terms from the translog
model are jointly statistically insignificant. The Mizon and Richard (1986) test therefore
indicates that HACD provides a better fit to the data than the translog. Overall, it can be
concluded that the HACD marginally fits the data better.

The Breusch-Pagan Heteroskedasticity tests conducted on the residuals from
both models indicate that the null hypothesis of Homoskedasticity cannot be rejected at
the 10 percent level. The chi-square test statistic and associated p-value for the test on
residuals from the translog functional form are 0.039, and 0.844 respectively. The

corresponding test statistic and p-value for the HACD are 0.84 and 0.629 respectively.

3.6 Conclusions

This paper investigated the performance of two flexible functional forms
(translog and HACD) in multiple output cost function estimation for water and
wastewater facilities in the United States. Important performance measures like input
elasticity of substitution, economies of scale, and input demand functions are also

derived and compared.

106

 

Results for the HACD indicate that a one percent increase in the price of capital
and labor results in 0.06 and 2.856 percent decline in the quantity of capital and labor
demanded respectively. Using the translog parameters, the same increase in price
results in a 0.05 and 0.09 percent decline in the quantity of capital and labor demanded
respectively. Finally, the HACD provides statistically significant cost economies
estimates that represent economies of scale to water provision. In particular, a one
percent increase in the quantity of water and population served increases costs by 0.48
and 0.43 percent respectively. The translog parameter estimates on the other hand point
to diseconomies of scale. A one percent increase in the quantity of water and population
served increases costs by 4.7 and 2.31 percent respectively. Thus while the HACD
estimates suggest economies of scale to increases in quantity of water and population
served, the translog estimates suggest diseconomies to scale. Overall, the functional
form tests and analyses suggest that the HACD provides a better fit to the data. The
difference in fit of the two models is however quite small.

The contrasting result for the cost economies parameters is determined to be
attributable to the difference in structure of the two models. Once the derivative of the
estimated models is taken, the constant terms disappear and the computed cost
economies from the HACD becomes much smaller than those from the translog because
of the inverse output terms inserted to replace the quadratic terms in the translog.

In a purely economic sense, these cost economies estimates are decision-making
parameters that indicate whether expansion, contraction, or retention of current output

level should be pursued. Given that the choice of functional form may reverse such

107

 

decisions underlines the importance of using more than one functional form for studies

of this nature to allow for comparison and assessment of reliability of the estimates.

 

108

TABLES
Table 3.1

Summary Statistic of Main Variables

 

 

 

Variable Unit Mean Standard Min Max
Deviation
Total cost Million $ 38.50 54.8 2.70 327.00

Water production Million Gal. 22140.20 30566.92 1569.50 155125.00
Service population Thousands 385.75 522.34 29.99 2,390.00
Capital price percent 0.04 0.01 0.02 0.060

Labor price $/employee 67569.46 38516.26 8087.15 176,770.90

 

 

 

 

 

 

109

 

Table 3.2

Estimated TRANSLOG and HACD Model Results
Dependent Variable: Log Total Variable Cost

 

 

 

 

 

Translog Models Estimated Results HACD Estimated Results
Parameters Estimated coefficient Parameters Estimated coefficient
a (intercept) 13.220* 01 (intercept) -186.682*
(7.635) (80.053)
6,, -1.592 6,, 4589*
(2.334) (1.090)
,qu 0.643 6, 2.605
(0.508) (1.643)
6, 1.609 m, 1344.336“
(2.810) (349.379)
,8” 0.218 27, 528.033
(0.610) (600.201)
,qu -0.722 pqs -81 19.938*
(1.075) (2886.812)
(0k 1 .134'.‘ ”I: 1.006"‘
(0.1 19) (0.004)
(01d: -0.007* vkk 0024*
(0.004) (0.034)
(01 -0.l34 1)] -0.007
(0.119) (0.016)
a)” 0038* 011 -0. l 69*
(0.008) (0.009)
£01k -0.030* 01;, ~0.007
(0.007) (0.016)
Aqk -0.012 ,uqk -0.962
(0.018) (1.696)
liq, 0.103 ,qu -29.006*
(0.069) (1 1.663)
2.51 0.020 ,usk 2.027*
(0.019) (3.041)
is, -0. 1 10 ,usl 62.700
(0.070) (20.308)
R-square 0.879 0.904

 

 

 

 

- Values in parentheses are standard errors
* Values with asterisk indicate statistical significance at 10 percent level

110

 

Table 3.3

Elasticity of Substitution and Input Elasticity Estimates

 

 

 

 

 

 

 

Translog HACD
Elasticities Estimated Parameters Elasticities Estimated Parameters
2 Hr '0054* a“, -0.066*
(0.028) (0.038)
2” -2.099* a,, -31.822*
(0.910) (17.211)
Zlk 0287* a”, 0.912*
g (0.111) (0.511)
9"" -o.os1* 61* -0.061'*
6” (0.027) 6” (0.0352
-0.092* -2.856
in. (0.040) a, (1.497)
0.013 "‘ 0.079*
(0.005) (0.044)

 

* Values with asterisk indicate statistical significance at 10 percent level

111

 

 

Table 3.4

Economies of Scale Estimates

 

 

 

 

 

 

 

Translog HACD
Estimated Estimated value Estimated parameter Estimated
parameter value
5g 4700* ”q 0480*
(1.688) (0.264)
gs 2310* Us 0430*
(1.046) (0.210)

 

* Values with asterisk indicate statistical significance at 10 percent level

112

 

 

References

Andrikopolos, A., and J. Loizides. "Cost Structure and Productivity Growth in
European Railway Systems." Applied Economics 30(1998): 1625-1639.

AWWA "2004 Water and Wastewater Rate Survey." # 54001

Barnett, W. A., Y. W. Lee, and M. Wolfe. "The Global Properties of the Two Minflex
Laurent Flexible Functional Forrns,." Journal of Econometrics 36, no. 3(Nov.
1987): 281-98.

Bhattacharyya, A., et al. "Allocative efficiency of rural Nevada water systems: A
hedonic shadow cost function approach." Journal of Regional Science 35, no.
3(1995): 485-501.

Bontemps C. and Mizon G. "Encompassing: Concepts and Implementation." Oxford
Bulletin of Economics and Statistics 70, Supplement (2008): 721 - 750

Chew L.C., Hsein K., and Jongsay Y. "Airline Code-share Alliances and Costs:
Imposing Concavity on Translog Cost Function Estimation, Review of Industrial
Organization, 26 (2005): 461- 487

Christensen, L. R., J. D. W. and, and L. J. Lau. "Transcendental logarithmic production
frontiers." Review of Economics and Statistics 55(1973): 28-45.

Conrad K. and Jorgenson D.W. "Tests of a Model of Production for the Federal
Republic of Germany" European Economic Review 10 (1977):51-57.

Damodaran A. "Cost of Capital by Sector", Damodaran Online, intemet link:
http://pages. stern. nyu. edu/~adamodar/New_Home_Page/datafile/wacc. htm

Diewert, W. "An Application of the Shephard Duality Theorem: A Generalized
Leontief. Production Function." Journal of Political Economy 79, no. 3(1971):
481-507.

Diewert, W. E., and W. T. J. "Multiproduct Cost Function Estimation and Subadditivity
Tests: A Critique of the Evans and Heckman Research on the US. Bell System.
UBC." Department of Economics Discussion Paper 91—21(l 991): 1—19.

Economagic, "Discount Window Borrowing Rate (2002)" intemet link
http://www. economagic. com/em-cgi/data. exe/fedbog/a‘wb

EPA. "Community Water System Survey, 2000." (2002).

Evans, D. S., and J. J. Heckman. "A Test for Subadditivity of the Cost Function with an

113

 

 

Application to the Bell System." American Economic Review (Supplement)
74(1984): 615—623.

Fabbri P. and G. Fraquelli. "Costs and Structure of Technology in the Italian Water
Industry." Empirica 27, (2000): 65-82.

Feigenbaum, S., and R. Teeples. "Public Versus Private Water Delivery: A Hedonic
Cost Approach." The Review of Economics and Statistics 65, no. 4(1983): 672-
678.

Garcia, 8., and A. Thomas. "The structure of municipal water supply costs: Application
to a panel of French local communities." Journal of productivity analysis
16(2001): 5-29.

Greene W. H., "The Econometric Approach to Efficiency Analysis" in, "The
Measurement of Productive Efficiency: Techniques and Applications", by
Harold Fried, Knox Lovell, and Shelton Smith. Oxford University Press, 1993.

Hunt T.L. "The Structure and Changes of Technology in Prewar Japanese Agriculture:
Comment"American Journal of Agricultural Economics, 62, No. 4 (Nov., 1980):
826-827

Jehle, GA. and Reny P.J., "Advanced Microeconomic Theory" , Addison-Wesley , July
2000

J orgenson, Dale W. and Barbara M. Fraumeni (1981), “Relative Prices on Technical
Change,” in Modeling and Measuring Natural Resource Substitution, Ernst R.
Bemdt & Barry C. Field eds., MIT Press, Cambridge, MA, 17-47.

Mas-Collel A. Whinston M.D., and Green J .R. "Microeconomic Theory.", Oxford
University Press, USA (June 15, 1995)

McAllister, P. H., and D. McManus. "Resolving the Scale Efficiency Puzzle in
Banking." Journal of Banking and Finance 17(1993): 389-405.

Miles J. and Ezzell J. "The weighted average cost of capital, perfect capital markets and
project life: a clarification." Journal of Financial and Quantitative Analysis, 15

(1980), S. 719-730.

Modigliani F. and Miller M. "The Cost of Capital, Corporation Finance and the Theory
of Investment," American Economic Review, June 1958.

Modigliani F. and Miller M. "Corporate income taxes and the cost of capital: a
correction." American Economic Review, 53 (3) (1963), pp. 433-443.

114