c
_w.a .

.
3
1.. "I

:2; .
3.; m:
It:
,3 A
mtg... V...»
. ca.) 2.3.
..l .
. A:

J:
‘

1X \:
.yQ‘»vv>‘\.u
5 v

w .I‘
1.“‘
9...:

v» t...
l)..—
.: ;
. 3? »
fan‘s

$Praﬁama1.
: . \

AV ‘1
ix...
V 1:! 3

I I

.¢

,er .
1'2» Q
‘ ..:..

«nanny.
#55:? ...r.

;eysw
xdwm .
, Ln:
.... .
.5,
. R? n .

 

awmcn
mark:
N3»

.n a
$34..

d&.l‘~:
:11: 1.. a
If! . r. .QI
:51} [thwar-
. an”?! .45.»! .
yin»,

. crannzv
Awmmw

1.»

THESIS

2 I
Michigan 5.-
1 University

 

 

 

This is to certify that the

dissertation entitled

THE IMPACT OF PUBLIC LANDS AND PUBLIC LAND TIMBER
HARVESTING 0N PRIVATE PROPERTY VALUES IN MICHIGAN

presented by

David Michael Jones

has been accepted towards fulﬁllment
of the requirements for

PhD Forestry

degree in

 

 

020wa»

a jOI’ professor I

Date February 23, 2001

 

MS U is an Afﬁrmative Action/Equal Opportunity Institution 0-12771

 

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

 

:l :" l ' Ll E

 

DEC-‘1 ZCCl

 

”9*
MAP} 0,300.14

 

0
«,9.

 

"ii? g :5 509g

 

 

 

 

 

 

 

 

 

6/01 c-JCIRC/DateDuest-pJS

 

 

THE IMPACT OF PUBLIC LANDS AND PUBLIC LAND TIMBER
HARVESTING 0N PRIVATE PROPERTY VALUES IN MICHIGAN
By

David Michael Jones

A DISSERTATION

Submitted to
Michigan State University
In partial fulﬁllment of the requirements

For the degree of

DOCTOR OF PHILOSOPHY

Department of Forestry

2001

ABSTRACT
THE IMPACT OF PUBLIC LANDS AND PUBLIC LAND TIMBER
HARVESTING ON PRIVATE PROPERTY VALUES IN MICHIGAN
By

David Michael Jones

The inﬂuence that publicly-owned land has on nearby private property values has not
been studied. Externalities of public land may have an effect on nearby private
property values. This study applies hedonic regression analysis to real property
transactions of property near public land in Michigan to measure the effect of public
land and public land timber harvesting on private property values. The results
demonstrate that the percent of public land surrounding private property positively
impacts developed property values. The impact of distance to federal public land on

' private property is found to positively affect vacant property values. The closer that
vacant properties are to federal public land, the more they are worth. Adjacent vacant
property values are negatively affected by the presence of federal public land relative
to state land. Public land timber harvesting is not found to have any statistical impact

on adjacent property values.

ACKNOWLEDGEMENTS

First and foremost I would like to thank Larry Icefers. Dr. Leefers has a rare and
intangible mix of intelligence, patience, and kindness. He is an excellent educator,
researcher and mentor, and he deserves the most credit for my completion of graduate
school. I would also like to thank my wife Susan. Susan always gives me
encouragement, patience, and support whether I fail or succeed. Without her I would
not have finished graduate school. I would also like to thank my parents for all their
support and kind ears. I would like to thank my guidance committee members for
their insights, support and guidance: Dr. Patricia E. Norris, Dr. Karen Potter-Witter,
Dr. Daniel J. Stynes, and Dr. J. Michael Vasievich. I would also like to thank the
Department of Forestry at Michigan State University for their support and guidance.
Dr. Stephen Koontz deserves thanks for all his econometric advice and guidance.

Lastly, I would like to thank the McIntire-Stennis program for financial support.

u.»

Table of Contents

 

 

 

 

 

 

 

LIST OF TABLES .......................................................................................................... vii
LIST OF FIGURES ......................................................................................................... xi
CHAPTER ONE: INTRODUCTION ..................................................................... 1
Introduction ....................................................................................................................... 1
Problem Statement ............................................................................................................ 7
Objectives ........................................................................................................................... 9
Significance and Assumptions ...................................... 10
Organization of Study ..................................................................................................... 12
CHAPTER 2: LITERATURE REVIEW .............................................................. 13
Non-Market Valuation - - -- ..... - - ..... - - -- - l3
Hedonic Model Theory - - -_ -- - - - - 16
Demand .......................................................................................................................... 17
Supply ............................................................................................................................ 22
Market Equilibrium ....................................................................................................... 26
Applied Hedonic Models - - ................ - 29
Amenity and Disamenity Valuation .............................................................................. 29
Growth Control and Land-Use Impact Valuation ......................................................... 31
Other Related Literature _ ..................... - -- ..... _ _ -.-.32
Summary ..... -- - .................. _ 34
CHAPTER 3: METHODS .................................................................................. 35
Sampling Frame - _ - - - _- 35
Determination of Public Lands ...................................................................................... 36
Determination of Sample Areas .................................................................................... 37
Control of Inﬂuential Effects ......................................................................................... 40
Selection of Study Areas ............................................................................................... 42
Data Collection--- - -- -- _ ---------- - - -- . -- _- - - 4-6
Data Analysis..- - -- - ----- - _ -- ..... -- ----------- 47

 

Description of Data ........................................................................................................ 48

 

 

 

 

 

 

 

 

 

General Form of Model ................................................................................................. 54
Functional Form Considerations ................................................................................... 58
CHAPTER 4: ANALYSIS .................................................................................. 63
Descriptive Statistics--- -- -- - . ....... . - -- - - -------- 63
Entire Data Set ............................................................................................................... 64
Full Model Statistics ...................................................................................................... 67
Developed F ulI Model Data ....................................................................................... 68
Vacant Full Model Data ............................................................................................. 71
Adjacent Model Statistics .............................................................................................. 74
Adjacent Developed Model Data ............................................................................... 74
Adjacent Vacant Model Data ..................................................................................... 77
Full Models-- - -- - - - - -8l
Developed Property Full Models ................................................................................... 82
Vacant Property Full Models ......................................................................................... 95
Adjacent Models --------- - - - - -- - 104
Developed Property Adjacent Models ......................................................................... 104
Vacant Property Adjacent Models ............................................................................... 111
Removal of Outliers ..................................................................................................... 120
CHAPTER 5: CONCLUSIONS ........................................................................ 123
Research Summary- - - - ------- -- - -- -- 124
Results - - . - - -- - 127
Policy Implications - - ----------------- .. ---------- - 130
Recommendations for Future Research ---------- - -. - - -- 132
REFERENCES ................................................................................................. 135
APPENDIX A ................................................................................................... 138
Full Developed Model---- - . ---------- - -- -- - - - - - 138
Full Vacant Model - - - -- - ------------- . - . - - 146
APPENDIX B ................................................................................................... 159
Adjacent Developed Model - -- - - . -------- 159

 

Adjacent Vacant Model ................................................................................................ 167

vi

LIST OF TABLES

TABLE 1. ACRES AND PERCENT DISTRIBUTION OF PUBLIC LANDS IN

MICHIGAN ......................................................................................................... 36
TABLE 2. RANDOMLY SELECTED LIST OF 25 TOWNSHIPS IN MICHIGAN

THAT CONTAIN PUBLIC LAND, BY COUNTY. .......................................... 43
TABLE 3. STUDY AREAS, BY COUNTY. ........................................................... 44

TABLE 4. OUTLINE OF STEPS USED TO ASSESS AND IDENTIFY THE
FUNCTIONAL FORM OF EACH MODEL. ..................................................... 58

TABLE 5. DESCRIPTIVE STATISTICS FOR HOUSING AND MARKET
VARIABLES FROM THE ENTIRE DATA SET (N=l339). ............................ 64

TABLE 6. DESCRIPTIVE STATISTICS FOR PUBLIC LAND AND TIMBER
HARVESTING VARIABLES FROM THE ENTIRE DATA SET (N=l339)... 66

TABLE 7. DESCRIPTIVE STATISTICS FOR HOUSING AND MARKET
VARIABLES FROM THE DEVELOPED FULL MODEL SUBSET (N =667).68

TABLE 8. DESCRIPTIVE STATISTICS FOR PUBLIC LAND AND TIMBER
HARVESTING VARIABLES FROM THE DEVELOPED FULL MODEL
SUBSET (N=667). .............................................................................................. 70

TABLE 9. DESCRIPTIVE STATISTICS FOR HOUSING AND MARKET
VARIABLES FROM THE VACANT FULL MODEL SUBSET (N=672). ...... 71

TABLE 10. DESCRIPTIVE STATISTICS FOR PUBLIC LAND AND TIMBER
HARVESTING VARIABLES FROM THE VACANT FULL MODEL SUBSET
(N=672). .............................................................................................................. 73

TABLE 11. DESCRIPTIVE STATISTICS FOR HOUSING AND MARKET
VARIABLES FROM THE DEVELOPED ADJACENT MODEL SUBSET
(N=l34). .............................................................................................................. 75

TABLE 12. DESCRIPTIVE STATISTICS FOR PUBLIC LAND AND TIMBER
HARVESTING VARIABLES FROM THE DEVELOPED ADJACENT MODEL
SUBSET (N=134). .............................................................................................. 76

TABLE 13. DESCRIPTIVE STATISTICS FOR HOUSING AND MARKET
VARIABLES FROM THE VACANT ADJACENT MODEL SUBSET (N=179).78

vii

TABLE 14. DESCRIPTIVE STATISTICS FOR PUBLIC LAND AND TIMBER
HARVESTING VARIABLES FROM THE VACANT ADJACENT MODEL .
SUBSET (N=179). .............................................................................................. 80

TABLE 15. LINEAR DEVELOPED FULL MODEL ESTIMATION RESULTS. 83

TABLE 16. FINAL DEVELOPED FULL MODEL ESTIMATION RESULTS. 90

TABLE 17. LINEAR VACANT FULL MODEL ESTIMATION RESULTS. ....... 95

TABLE 13. FINAL VACANT FULL MODEL ESTIMATION RESULTS. ........ 100

TABLE 19. LINEAR DEVELOPED ADJACENT MODEL ESTIMATION RESULTS.
........................................................................................................................... 105

TABLE 20. FINAL DEVELOPED ADJACENT MODEL ESTIMATION RESULTS.
........................................................................................................................... 108

TABLE 21. LINEAR VACANT ADJACENT MODEL ESTIMATION RESULTS] 12
TABLE 22. FINAL VACANT ADJACENT MODEL ESTIMATION RESULTS.l 16

TABLE A1. LOG-LINEAR DEVELOPED FULL MODEL ESTIMATION RESULTS.
........................................................................................................................... 138

TABLE A2. TESTS OF SQUARED AND CUBED COEFFICIENTS OF FULL
DEVELOPED MODEL. ................................................................................... 140

TABLE A3. “BEST” DEVELOPED FULL MODEL ESTIMATION RESULTS. 141

TABLE A4. DEVELOPED FULL MODEL ARTIFICIAL REGRESSION
ESTIMATION RESULTS. ............................................................................... 143

TABLE A5. DEVELOPED FULL MODEL ARTIFICIAL REGRESSION TESTS —
QUADRATIC TERMS. .................................................................................... 145

TABLE A6. DEVELOPED FULL MODEL ARTIFICIAL REGRESSION TESTS —
LOGARITHMIC TERMS ................................................................................. 145

TABLE A7. LOG-QUADRATIC VACANT FULL MODEL ESTIMATION
RESULTS. ......................................................................................................... 146

TABLE A8. LOG-QUADRATIC VACANT FULL MODEL ESTIMATION RESULTS
CORRECTING HETEROSCEDASTICITY. ................................................... 148

viii

TABLE A9. TESTS OF SQUARED AND CUBED TERMS OF FULL VACANT
LOG-QUADRATIC MODEL. .......................................................................... 150

TABLE A10. “BEST” VACANT FULL MODEL ESTIMATION RESULTS. 151

TABLE Al 1. VACANT FULL MODEL ARTIFICIAL REGRESSION ESTIMATION
RESULTS. ......................................................................................................... 152

TABLE A12. ARTIFICIAL REGRESSION TESTS — QUADRATIC TERMS. .. 154
TABLE A13. ARTIFICIAL REGRESSION TESTS — LOGARITHMIC TERMS.154

TABLE A14. FINAL DEVELOPED FULL MODEL ESTIMATION RESULTS
WITHOUT OUTLIERS. ................................................................................... 155

TABLE A15. FINAL VACANT FULL MODEL ESTIMATION RESULTS WITHOUT
OUTLIERS. ....................................................................................................... 157

TABLE Bl. LOG-QUADRATIC DEVELOPED ADJACENT MODEL ESTIMATION
RESULTS. ......................................................................................................... 159

TABLE B2. TESTS OF SQUARED AND CUBED TERMS OF ADJACENT
DEVELOPED LOG-QUADRATIC MODEL. ................................................. 161

TABLE B3. “BEST” DEVELOPED ADJACENT MODEL ESTIMATION RESULTS.
........................................................................................................................... 162

TABLE B4. DEVELOPED ADJACENT ARTIFICIAL REGRESSION MODEL
ESTIMATION RESULTS. ............................................................................... 164

TABLE BS. DEVELOPED ADJACENT MODEL ARTIFICIAL REGRESSION
TESTS — QUADRATIC TERMS. .................................................................... 166

TABLE B6. DEVELOPED ADJACENT MODEL ARTIFICIAL REGRESSION
TESTS — LOGARITHMIC TERMS. ................................................................ 166

TABLE B7. LOG-QUADRATIC VACANT ADJACENT MODEL ESTIMATION
RESULTS. ......................................................................................................... 167

_ TABLE BB. TESTS OF SQUARED AND CUBED COEFFICIENTS OF ADJACENT
VACANT LOG-QUADRATIC MODEL. ........................................................ 169

TABLE B9. “BEST” LOG-LINEAR VACANT ADJACENT MODEL ESTIMATION
RESULTS .......................................................................................................... 170

TABLE BIO. VACANT ADJACENT ARTIFICIAL REGRESSION MODEL
ESTIMATION RESULTS. ............................................................................... 171

TABLE B11. VACANT ADJACENT MODEL ARTIFICIAL REGRESSION TESTS —
QUADRATIC TERMS. .................................................................................... 173

TABLE BIZ. VACANT ADJACENT MODEL ARTIFICIAL REGRESSION TESTS —
LOGARITHMIC TERMS ................................................................................. 173

TABLE B13. FINAL DEVELOPED ADJACENT MODEL ESTIMATION RESULTS
WITHOUT OUTLIERS. ................................................................................... 174

TABLE 814. FINAL VACANT ADJACENT MODEL ESTIMATION RESULTS
WITHOUT OUTLIERS. ................................................................................... 176

LIST OF FIGURES

FIGURE 1. CONSUMER BID FUNCTIONS FOR THE Z. CHARACTERISTIC AT

DIFFERENT UTILITY LEVELS. ...................................................................... 19
FIGURE 2. CONSUMER BID FUNCTIONS AND THE IMPLICIT PRICE
FUNCTION FOR CHARACTERISTIC Zl. ....................................................... 21
FIGURE 3. PRODUCER OFFER CURVES FOR CHARACTERISTIC 21 AT
DIFFERENT PROFIT LEVELS. ........................................................................ 24
FIGURE 4. PRODUCER OFFER CURVES AND THE IMPLICIT PRICE FUNCTION
FOR CHARACTERISTIC Z1. ............................................................................ 26
FIGURE 5. CONSUMER-PRODUCER EQUILIBRIUM AND THE IMPLICIT PRICE
FUNCTION FOR CHARACTERISTIC Z]. ....................................................... 28
FIGURE 6. MAP OF STUDY AREAS, BY COUNTY. .......................................... 45

xi

CHAPTER ONE: INTRODUCTION

Introduction

Throughout the history of the United States, public land policy has caused controversy
and debate. Central to much of the controversy are issues involving property rights,
property ownership, and values, both market and non-market, associated with public
lands. The property rights and ownership Structure, as well as the values associated with

public land, have changed Signiﬁcantly over the past 150 years.

From a historical perspective, Dana and Fairfax (1980) suggested three periods into
which public land policy can be divided: disposition, reservation, and management. The
disposition period occurred approximately between 1775 and 1891 , during which
Congress disposed of the public domain. The disposition period transferred property, and
' the associated property rights, from the federal government to, primarily, individuals.
Initially the goal of disposition was to accumulate revenue for the ailing Treasury. To
this end, Congress attempted to sell the public domain to settlers. After several
legislative attempts and failures, Congress changed its policy from raising revenue to
settlement and development. The Homestead Act of 1862 was the ﬁrst “free” land policy
in the United States. The Timber Culture Act followed in 1873. Both these acts offered -
land to individuals for agricultural and forest development of the public domain lands.
Both policies were marked with abuses. as the “terms and the spirit” of the “Acts were

widely violated” (Dana and Fairfax 1980). The intended development did not occur in

many cases. Much of the land was stripped of its timber and/or acquired for cattle
ranges. 1n the beginning of the disposition period, natural resources were abundant, but

as the period drew to a close many natural resources were perceived as becoming scarce.

From the developing perception of scarcity of many natural resources, coupled with
growing anti-trust sentiment, the reservation period emerged. The reservation period
occurred between approximately 1891 and 1905. This period overlapped in large part
with the “Golden Age of American Conservation” (1898-1910), and produced land
conservation policy unparalleled in US. history. Leaders of the day, such as Theodore
Roosevelt and Gifford Pinchot, believed that many of the country’s natural resources,
such as timber and minerals, were becoming scarce. Not trustingAmerican capitalism of
the time, the “trust-busters” felt it was the duty of the government to provide a long-term
supply of scarce natural resources for future generations. The idea of permanent public
ownership of land followed. Between 1905 and 1909 President Roosevelt withdrew over
92 million acres from the public domain and established it as public land. By the end of
the Roosevelt presidency over 150 million acres of land had been withdrawn for national

forests.

Besides the establishment of public land, the reservation period is marked by a split in the
conservationist movement. On one Side of the split were the utilitarians. Many,
including Theodore Roosevelt and Gifford Pinchot, believed the federally owned land
should be managed for human use or more appropriately “wise-use”. The wise-use

concept is best represented by the popular Pinchot slogan “the greatest good for the

greatest number for the longest time.” The greatest goods under this philosophy were
commonly market goods such as timber, minerals, and grazing range. On the other Side
of the split were the preservationists, led by John Muir. The preservationists believed no
“use” should occur on the public lands. The basis of their belief was that much of the
public land contains spiritual values, not just the market values associated with human
use. The preservationists marked the ﬁrst large movement that emphasized the existence
of non-market goods associated with public lands. The conservation split gave rise to
many debates over the years surrounding how certain public lands should be managed
and what values they Should contain. Developing from the Split were two different and

incompatible management philosophies: multiple use and preservation.

The last, and debatably the most controversial, period of public land policy is the
management period. This period began in 1905 and continues to the present. The
management period has repeatedly dealt with issues that formed land policy in previous
periods. One of the issues revisited in the management period concerns the values
included in the differing management philosophies. Another issue, that of public

ownership, has provided almost continuous conﬂict.

The period began with the formal establishment of the National Forest system, followed
by the National Park system. These federal public lands were managed for multiple use
and single use, respectively. The split system, which now includes among others the
Bureau of Land Management and the U. S. Fish and Wildlife Service, has been the

govemment’s answer to the split in the conservation philosophies. The single-use land

management philosophy has evolved to provide many non-market values such as non-
game species of wildlife and aesthetic beauty. The multiple-use management philosophy
has continued to provide market goods as well as some non-market values. The
evolution of non-market values in the differing management philosophies demonstrates a

demand for such goods and intangibles.

Another land policy issue, public ownership, has been questioned several times in the
management period. Challenges to federal ownership of land have occurred in four
notable movements. Interestingly, three challenges deal with grazing rights on public
lands. The ﬁrst Challenge followed the establishment of the ﬁrst grazing fee in 1906 by
Gifford Pinchot, then the head of the new Forest Service. A charge was made for all
livestock grazed on the forest reserves (Dana and Fairfax 1980). A storm of protest
followed. From 1907-1915, six public land conferences were held in the western states
providing a forum for ranchers demanding a change in federal management policy. The
ranchers invoked claims of states’ rights. Unanimous support never appeared and the

ﬁrst organized attempt to limit or stop federal control of lands ended.

The 1940’s held the second challenge against federal land ownership. From 1940-1943,
Senator Patrick McCarren of Nevada held committee hearings due to angered stockmen.
Stockmen were angered by proposed reductions in grazing on federal lands, and they felt
that too much regard was given to wildlife and that it might threaten established cattle
and Sheep grazing rights. The ranchers wanted to buy the public land with offers ranging

from $0.09 to $2.80 per acre. In October, 1946, the Joint Livestock Committee on Public

Lands developed a legislative program. The prOposaIS called for legislation that would
allow Forest Service grazing perrnittees to purchase the property allotted to them under
the Taylor Grazing Act. Also. the operators wanted the Forest Service administered
grazing lands turned over to the Department of Interior for disposition under the same
scheme. “Within months the opposition to the idea was so great that it was withdrawn”

(Dana and Fairfax 1980).

In more recent years, animosity over public land rights and non-market values led to the
third movement challenging federal ownership of land. In 1979 the Sagebrush Rebellion
erupted. The rebellion, a conservative western movement, centered primarily on property
rights issues related to public lands. Short (1989) stated “most observers agree that the
conservation movement’s crowning success, the Federal Land Policy and Management
Act of 1976, represented the last straw for many western ranchers, miners, and loggers.”
“In July, 1979, the Nevada State Legislature passed a resolution demanding the transfer
of 49 million acres of federal land to state control” and “Senator Onin Hatch of Utah
introduced a national sagebrush rebellion bill into the US. Senate...” (Short 1989). The
transfer of the land was justiﬁed, by rebellion leaders, based on the perception that the
states will be better land managers. The rebellion leaders felt the economic growth in the
west was stiﬂed by restrictive federal land management and that the states had the right to
control the land in their borders. “Some rebels even maintained that the state ownership
was only temporary and that sales to private individuals would soon follow” (Short
1989). The Reagan administration developed policies that were sympathetic towards the

western sentiment. Notably, the United States Department of Agriculture’s (USDA)

Asset Management Program (AMP) demonstrated the emerging federal govemment’s
pro-development stance demanded by the rebellion. The objectives of the AMP, as noted
by Douglas MacCleery (Deputy Assistant Secretary of Agriculture for Natural Resources
and Environment) were threefold: (l) to sell excess federal property and some public
lands that would have a higher and better use in private ownership, (2) to improve the
efﬁciency of government by selling lands that are costly and inefﬁcient to manage and
are not necessary to serve public objectives, and (3) to utilize sale revenues to pay off a
portion of the national debt (MacCleery 1984). The Sagebrush Rebellion soon lost
constituency, which was small in numbers but strong in interest group support, and ended

without any transfer of federal lands.

The Sagebrush Rebellion was immediately followed by the fourth movement to challenge
federal ownership: the privatization movement. The privatization movement, led by
“economists” John Baden and Steven Hanke. emphasized that private markets, not public
' bureaucrats, are the most efﬁcient managers of land (Short 1989). Their conclusion,
therefore, was that the public lands should be sold to private parties. Baden and Hanke
held that an economically efﬁcient allocation of land resources was more likely to occur
under private ownership than under public ownership. Utilization of the market values
on public land played a heavy hand in the privatization argument. Though the
privatization movement emphasized a more “scientiﬁc” basis than the emotional
Sagebrush Rebellion, it too soon disappeared from the policy spotlight without transfer of

public land.

Both the recent movements brought the issues of values, public ownership, and private
rights associated with public land back into the policy arena. From a historical
perspective, public ownership, private rights on public land, and the values emphasized
on public lands have been consistently active land policy issues. History provides
evidence that discontent over public ownership of land, over values emphasized on public
land, and over associated private rights has existed and is likely to continue. The ebb and
ﬂow of the continuous change of political climate coupled with the previously mentioned

discontent is likely to provide a platform for the same issues to resurface in the future.

Problem Statement

Baden and Lueck (1984) provided arguments for privatization, in which several
economic aspects are addressed. They noted “economic theory states that when property
rights to resources are well deﬁned, enforced, transferable, and privately held, owners
will tend to allocate the resources efﬁciently. . .”. Among others, one fault of their
defense lies in their interpretation of the above statement. As Runge (1984) stated
“economists always stress that the technical argument assumes strict caveats -- perfect
information and foresight, the absence of monopoly power, and none of the
interdependencies or third party effects called extemalities.” The privatization argument.
as presented by Baden and Lueck, ignores extemalities as well as associated welfare
implications. Public lands may offer several extemalities such as congestion associated

with recreation and amenities (or disamenities) offered to local residents. Many

extemalities and associated welfare implications of public lands are not well understood.

and therefore are not included in the privatization efﬁciency defense.

As stated above, public land may have associated extemalities. These extemalities take
the form of amenities or disamenities. Amenities such as recreational opportunities,
wildlife, and forested or pasture land are provided by public land. The amenities can
have positive local effects such aesthetic beauty, increased wildlife viewing, and
increased recreational opportunities. The disamenities can have negative local effects
such as congestion, wildlife damage, and timber harvesting impacts. Because these
effects are local, these amenities (disamenities) may affect nearby property values. The
economically efﬁcient allocation of public land requires the internalization of such
extemalities. Intemalization would require some type of agreement between those

affected and those who emit the extemality.

Not only do extemalities have efﬁciency implications, they also carry welfare
implications. An extemality that is not internalized may have positive or negative effects
upon public land users and local residents. Using the above example of timber
harvesting, the nearby residents may receive a negative impact on their utility. This
impact could be reﬂected in decreased property values. The local resident may realize a
decreased property value, while a non-resident will not see a property value decrease.

I The possible impact on local property values may then have welfare implications for

local residents.

In order to understand the full implications of privatization of public land, or cessation of
federal land to the states, the associated extemalities must be examined. Extemalities
related to public land may have large welfare and efﬁciency implications on local

residents.
Objectives ‘

It would be a logical assumption to believe that people are willing to pay a premium to
live near public land. Real estate advertisements commonly describe the existence of
nearby or abutting public land as a positive attribute. This premium would be equal to
the monetary value of the excess of beneﬁts over costs (or amenities over disamenities)
from the nearby public land. It would also be logical to further assume that timber
harvesting on the nearby public land could have a temporal negative impact on property
values. If a person bought a home near public land, and paid a premium, the possible
negative impact of future timber harvesting may temporally negate this premium. It is
important to examine the possible premium as well as the possible negative impacts of

timber harvesting on varying public lands.

The objectives of this research are: (1) to estimate the impact of Michigan’s public lands
on sales prices of adjacent private lands, (2) to estimate the impact of proximity
(distance) to public lands on private land sales values, (3) to compare impacts of various
types of public land use (ownership) on nearby private land values, and (4) to estimate

the impact of public land timber harvesting on adjacent private property values.

Signiﬁcance and Assumptions

In order for an efﬁcient level of public land to exist, the extemalities associated with
these lands need to be examined. Extemalities in the form of amenities (or disamenities)
and timber harvesting may have large effects on local property values and associated
property taxes. The sale or accumulation of public land may not be efﬁcient without
consideration for all the extemalities involved. Taxation of nearby residential property

may not be efﬁcient, or equitable, without the consideration of all extemalities.

Furthermore, any change in public land ownership, whether sold or given to the states
may involve widespread welfare changes. The sale of public land may have large
impacts on local property values and exhibit large welfare effects. If state and federal
land have different levels of impacts on property values or different tax subsidy policies.
the simple transaction of turning federal land over to states may also have large welfare

effects.

This research uses hedonic price regression models, which will be discussed in depth in
Chapter 2. By using hedonic price models, a researcher must accept two assumptions.
First, it is assumed that all property buyers are homogenous. Second, it is assumed that

all sellers are homogenous.

Unrelated to the models, further assumptions are made. Property owners are assumed to
be aware of nearby public land. It is assumed that property buyers and sellers are aware
of the amenities and disamenities provided by nearby public land and timber harvesting
on nearby public land. It is further assumed that buyers and sellers are aware of which
governing agency owns the nearby public land. Lastly, the data that was collected from
various government ofﬁces is assumed to be accurate and all-inclusive. None of these

assumptions are tested.

Buyers and sellers perceptions concerning the existence of public land, the distance from
public land, the ownership of public land, and timber harvesting on public land are not
the emphasis of the hedonic models in this study. Rather, the actual behavior of the
buyers and sellers is the emphasis. The perceptions are assumed but not detrimental to
the conclusions of the models with regard to actual behavior. The statistical signiﬁcance
of the public land related variables might be reduced due to a lack of awareness of both

' buyer and seller when compared to buyers and sellers that are completely aware. The
purpose of this study is to model the actual behavior of all buyers and sellers near public

lands in Michigan, not Speciﬁc subsets of different perception levels.

The purpose of hedonic models is to measure the implicit price of an attribute on the sales
price of a group of speciﬁc properties. The purpose of constructing hedonic models is
not for prediction. They should not be used for predicting property prices. Creating
predictive parsimonious models is not a goal of hedonic modeling and is not pursued in

this study.

Organization of Study

The second chapter reviews non-market valuation techniques. In particular, the hedonic
price model is reviewed extensively as this is the non-market valuation technique most
appropriate for addressing the objectives of this study. The methods used in this study
are covered in Chapter three. Chapter four provides and interprets the hedonic model

results. Chapter ﬁve offers conclusions and policy implications of the model results.

12

CHAPTER 2: LITERATURE REVIEW.

To address the study objectives, a review of theoretical and empirical literature was
conducted. The objectives require the measurement of non-market goods. This literature
review therefore focuses on topics related to theoretical and empirical non-market

valuation.

Chapter 2 begins with a review of common techniques used to place a monetary value on
non-market goods. Through this review, hedonic regressiOn analysis is selected as the
preferred technique. The following section explains the theoretical framework of hedonic
regression theory. A review of applied hedonic models then follows. The ﬁnal section of

Chapter 2 reviews other literature related to this dissertation.

Non-Market Valuation '

Economists have developed several techniques for placing a monetary value on non-
market goods. Each technique measures consumer willingness to pay for a speciﬁc non-
market resource. Three techniques in particular have been employed in the majority of
environmental valuation Studies: contingent valuation, travel cost, and hedonic pricing

(Freeman 1986, and Anderson and Bishop 1986).

The ﬁrst technique - the contingent valuation method (CVM) - is often referred to as a

“stated-preference” method. CVM “employs personal interviews, telephone interviews,

13

or mail surveys to ask people about the values they would place on environmental
commodities if ideal markets did exist or other means of payment such as taxes were in
effect” (Anderson and Bishop 1986). Consumers are, therefore, asked about their
willingness to pay for a good contingent on the existence of a market or other means of
supply. The CVM approach offers a broad framework to measure various non-market
values. CVM is unique in that it is the only non-market valuation technique that can
estimate existence values (i.e., “values people place on resources quite apart from any
desire they personally have to consume them or enjoy them in any conventional use” )

(Anderson and Bishop 1986).

The remaining two approaches, the travel-cost method and the hedonic-price method, are
often termed “market-related approaches”, “transaction-evidence” methods, and
“revealed preference” methods. Both these approaches attempt to form a linkage

between market transactions of private goods and services and non-market goods and

services.

The travel cost (TC) method measures consumers’ willingness to pay for a recreational
service. This method is “based on the premise that travel expenditures and travel time
ration access to some environmental commodities such as recreation sites” (Anderson
and Bishop 1986). Consumers are surveyed on the costs and time associated with their
trip to the recreation site. From the costs, time, user income, and user proximity from the
site, consumers’ willingness to pay for the recreation services of the particular site can be

inferred. The TC method is restricted primarily to valuation of recreational services.

 

The TC method is based on users traveling to a speciﬁc recreational Site. Many
environmental goods, including recreation, can. in some cases, be accessed in close
proximity to one’s home. Local landowners therefore incur little or no travel expense to
beneﬁt from the amenities (disamenities) associated with nearby public goods. Though
local landowners do not pay explicit travel expenses for public beneﬁts, they may pay
related costs in the form of land premiums. “Households often pay substantial premiums

to acquire homes that afford access to these amenities” (Anderson and Bishop 1986).

The hedonic price method attempts to measure consumers’ willingness to pay for these
public goods by measuring the premium inherent in their property values. The term
hedonic refers to pleasure, in economic language this is utility. Hedonic price, therefore,
refers to the price one is willing to pay for a speciﬁc level of utility or pleasure. Hedonic
price models estimate the marginal price the average person is willing to pay for an
additional unit of utility. Hedonic price models dissect real market prices and statistically
measure the hedonic price associated with each contributing factor and amenity

(disamenity).

Based on the objectives presented in Chapter 1, the appropriate non-market valuation
technique to be utilized in this study is the hedonic price method. The travel cost method
can not measure the impact of a resource on private property values, and therefore can
not address any of the objectives presented in Chapter 1. The CVM and the hedonic

price method can both address each of the four objectives of this study. The primary

difference between the CVM and hedonic regression approaches is that hedonic

regression requires actual market transactions while CVM does not.

With the hedonic regression approach the transactions are real, and consumers have to
determine their willingness to pay for the resources related to the nearby public land.
Consumers then have to weigh their willingness to pay against a premium, if any, on
property values. If the premium is less than or equal to their willingness to pay, the
consumers then actually make the transaction. If the CVM technique were used, the
consumer would be put through a choice exercise and asked to determine what their
willingness to pay without having to pay any of their own money. The hedonic price
model is more than a choice exercise for the consumer, it is an actual choice that requires
a large amount of money. An actual choice that requires large sums of money is more
likely to reﬂect consumers’ willingness to pay than a choice exercise that does not. As a
result, hedonic regression analysis is the better technique for addressing the objectives in

. this study. The theoretical foundations of the hedonic price method are presented next.

Hedonic Model Theory

Hedonic price theory develops arguments for determining the prices of implicit
characteristics of a class of highly differentiated products under perfect competition. The
implicit characteristics do not, generally, have explicit markets. In general, markets exist

only for the bundle of implicit characteristics sold together as a single market good. The

hedonic price theory described here closely follows the work of Rosen (1974) and

Anderson and Bishop (1986).

In order to describe the existing hedonic price theory, some structure and notation must
ﬁrst be presented. Consider a class of goods or products that are completely deﬁned by
an n-dimensional vector of product characteristics. Z = (Z ., Zz ..... Zn). Each Z, measures
the amount of the ith characteristic contained in each product. It is assumed that the
characteristics are perfectly divisible, and that they are positive. Because the vector Z
perfectly describes the products characteristics, one can assume a price function, P(Z)
exists, where P(Z) = P(Zl, Z; ..... Z,,). P(Z) is the hedonic or implicit price function. One
goal of hedonic modeling is to generate the implicit price function. As mentioned earlier,
the class of goods, Z, are assumed traded in a perfectly competitive market. In order to
generate an implicit price function, it must be assumed that demand and supply meet for
each characteristic. The theoretical derivation of demand and supply for the implicit
characteristics is described below. Following the demand-supply derivation are the

market equilibrium conditions.

Demand
The derivation of consumer demand begins at the basic level of utility maximization.
The consumer maximizes utility, U(X, Z), where Z is the bundle of characteristics of the

good purchased from the class of differentiated goods and X is the consumption of all

other goods measured in dollars. The maximization of utility is constrained by the

l7

consumer income, Y. If one assumes, for simplicity, the price of the X goods equals
unity, and that the consumer buys one unit of Z, then the income constraint becomes: Y =
X +P(Z), where P(Z) = P(Z., Zz ..... Z,,). The problem, in turn. becomes:

Max U(X, Z)

S.t. Y = X+ P(Z)

where the ﬁrst order condition can be expressed as:

aUZ 67P(Z)

The ﬁrst differential in equation (1) is the marginal rate of substitution (MRS) of good Z,
for X. The MRS is the rate that a consumer is willing to trade, or substitute, one good for
the other. The MRS holds Zj constant, for all j at i. The ﬁrst order condition ﬁnds the
MRS equal to the partial differential of the implicit price function, P(Z) = P(Zl, 22,...,Z,.),
with respect to Z,. The partial differential of P(Z) with respect to Z, is equal to P,-(Z), by
construction. Therefore the MRS of Z,- for X is equal to the price of Z,, P,(Z). The ﬁrst
order conditions listed in equation (1) fulﬁll the conditions oﬁen referred to as

consumption efﬁciency.

Because Z is a differentiated product containing n implicit characteristics, none of which
may have an explicit market, consumer bid functions for the characteristics must be
formed. Assume a bid ﬁmction in the form

0 = 6(Z;u, Y)
The consumer bid function for the implicit characteristics is a function of the quantity of

implicit characteristics in the differentiated product, Z, as well as utility (u) and income

18

(Y). As Rosen states, 6( Z; u, Y) represents “the expenditure a consumer is willing to pay

for alternative values of (Z i ..... Zn) at a given utility index and income” where u is any

given utility level (Rosen 1974). The bid function, 6(Z; u, Y) , represents a family of
indifference type curves relating the level of each characteristic, 2,, to money. Figure 1
depicts two bid functions for one consumer. The bid functionQl (Z , , Z 2 Zn , ul )

describes consumer l’s willingness to pay for alternative values of Z. at the level u,. The

bid function 6l1(Z1 ,Z Z n .112) describes the same consumer’s willingness to pay,

2,...,

but at a lower utility level, u 2 . The behavior of the bid function is assumed increasing

and concave in the characteristics, and decreasing in the given utility level.

 

0'(Z,,Zz,...,Z,,,u,)

0'(Z,,Z,,...,Zn,u2)

 

 

 

Z,

FIGURE 1. CONSUMER BID FUNCTIONS FOR THE Z. CHARACTERISTIC AT
DIFFERENT UTILITY LEVELS.

Implicitly, 6 is deﬁned

U(Y—6’,Z)=u

where Y — 6 is income minus the expenditure for the differentiated good, Z. Therefore,

Y — 6 represents the remaining income available for purchase of other goods, X.

The consumer maximization problem can be resolved with the inclusion of the bid
function. The bid function is substituted in the budget constraint for P(Z). The ﬁrst order

conditions with respect to X and Z yield

at]
6 2’" * Y - 21' f 11' 2
Zi( 9“ 9 )"' WX Ora I, ()

where Z ' and 14‘ are optimum quantities. Condition (2) states that the marginal bid for
the Zith characteristic must equal the marginal rate of substitution. The combination of
equations (1) and (2) provides an interesting result. As stated in equation (1), the MRS is
equal to the implicit price at the optimum, equation (2) states that the MRS is equal to the

marginal bid at the optimum. Combining equations (1) and (2) yields
auz §P(Z) It: It:
(OT/x: ﬁi=Pi(Z)=6Zi(Z :11 ,Y) (3)

Therefore, as depicted in equation (3), the MRS is equal to the implicit price and the

marginal bid. After solving for all 9; then the results yield
6(Z';u',Y)=P(Z‘) ' (4)
Condition (4) states that the optimal bid, at optimal utility and Z, must equal the optimal

price vector.

20

A graphical depiction of the demand-side derivation of the implicit price function is

illustrated in Figure 2. Figure 2 displays two consumers’ bid curves for Z l holding utility

and Z * Z ; constant at the optimal levels. The bid curves 61 (Z 1 , Z *

2* * d
29-", 2,..., n,u )an

62(Z1,Z*

III *
2 ,...,Zn,u ), represent two consumers. the latter of which demands more Z I at

their respective optimum quantities of Z ' and u. . From the two bid functions, one can
imagine several consumers bid functions in the Z I -P space. The numerous bid functions

lay out or trace the optimal price path, or implicit price function, for Z I. The implicit

price function, P(Z1 , Z * Z ; ) , is tangent to the bid curves at the optimal levels of

2,...,

*

Z 2 ,..., Z ; and u. . This tangency is the result of the equation (3) described above.

 

P(z,,z,‘,...z;)

 
  
 

62(Z,,Z;,...,Z,:,u )

6'(Z,,Z;,...,z,‘,u‘)

 

 

 

21

FIGURE 2. CONSUMER BID FUNCTIONS AND THE IMPLICIT PRICE
FUNCTION FOR C HARACTERISTIC Zl.

A similar analysis of bid curves can be conducted for all 2,. From these analyses
demand-side implicit price functions for all Z,- can be formed. The implicit price function
formed from consumer bid functions complete the demand side of the market. In order
for the market to clear, producers must have implicit price functions for the same

characteristics in bundle Z.

Supply

The analysis of the supply side is symmetric in many ways to the demand-side analysis.
Let M denote the number of units of differentiated product, Z, produced by a ﬁrm. Firms

are assumed independent of each other. The total cost function, C( M, Z; ,6) , is derived

by minimizing factor costs subject to a joint production constraint relating M, Z, and

factors of production. 6 is a shift parameter reﬂecting factor prices and production
function parameters. Three cost function assumptions should be presented at this point:

(i) C is convex and C(0,Z)=0, (ii) C M >0, and (iii) C Z >0. Assumption (i) states that
i

no production indivisibility’s exist. Assumption (ii) states that marginal costs of
production are positive and increasing. Lastly, (iii) assumes that the marginal costs of

increasing each characteristic, 2,, are positive and non-decreasing.

The economic objective of the ﬁrm is to maximize proﬁt,

I: = M * P(Z) — C(M,Z Zn;,6) , where Mand Zare the variables chosen for

1,...,

22

optimality and unit revenue is given by the implicit price function for all characteristics,
P(Z). The optimal choice of Z and M for a ﬁrm require

PI.(Z) = C7 (M,Zl....,Zn)/M, for all i, (5)

l

P(Z):CM(M’ZI’""Zn) (6)

At the optimum, marginal revenue from additional characteristics equals their marginal
cost of production per unit sold, as depicted in equation (5). Equation (6) states that
quantities are produced up to the point where unit revenues equal marginal production

costs. Equations (5) and (6) fulﬁll the conditions of production efﬁciency for good 2.

For simplicity, assume that the production of all other goods, X, fulﬁll the efﬁciency
conditions that are described above for 2,. It has already been assumed that the price of
X was unity. Based on these assumptions the Rate of Product Transformation (RPT) of

good Z,- forX can be determined. RPT (Z. for X) is equal to the ratio of their marginal

costs (MC). The MC of 2,, determined by equation (5), is equal to Pi(Z) , while the MC
of X is Px, which is equal to one. Therefore the RPT is equal to Pi(Z) . This fulﬁlls the

conditions for production efficiency for all goods.

Similar to the demand bid curve, assume a supply offer function, ¢(Z1 ,..., Z "m, ,6) ,

’ exists. Let ¢(Z Z ":1, 6) be deﬁned as a function indicating the unit prices a ﬁrm is

l””’
willing to accept for production of various characteristic levels, 2,. The offer function

assumes a constant proﬁt and the quantities produced, M, are optimal. Figure 3 depicts

23

two offer curves for the production of the implicit Z 1 characteristic. The offer curves are

for the same ﬁrm at different proﬁt levels, where the proﬁt level It is greater than it

 

 

 

 

1 2'
p
W (Z. ~---~Z.;”I~ﬂ)
¢'(ZIa---.-Z,,17Tz.~,3)
FIGURE 3. PRODUCER OFFER CURVES FOR CHARACTERISTIC 21 AT
DIFFERENT PROFIT LEVELS.

The proﬁt maximization can be resolved after inclusion of the offer function. By

substituting ¢(Z Z ”Mr, 6) for P( Z) in the proﬁt maximization problem, the ﬁrst

1 ,...,
order conditions produce

¢i(z)=Cz.(M,Z

l

1"°"Zn)/M’ for all i, ' (7)

‘ ¢ = CM (M,Z1,...,Zn) ‘ (3)

It follows from equation (7) that the marginal offer from additional characteristics is
equal to the marginal cost of production per unit sold. An important conclusion can be
drawn for equation (7). As determined earlier, the RPT (Z, for X) is equal to the MC of
2,. Equation (7) states that the MC of Z,- is equal to the marginal bid. Therefore the RPT
(Z,- forX) is equal to the marginal bid. Equation (8) ﬁnds the offer equal to the marginal

production cost at the optimum.

24

The combination of equations (5) and (7) ﬁnd the marginal offer and marginal price of

additional characteristics are equal at the optimum, or

PI.(Z )=¢zi (Z1 "”an 9.3) (9)

In addition, equations (6) and (8) ﬁnd, at the optimum, the offer is equal to the price, or

P(Z”) = ¢(Zl*,...,Z;;n*,/3) (10)

Figure 4 depicts the implicit price function for characteristic Z. traced by two offer

functions. The offer function ¢1 (Z 1 ,Z*

* It
2...,an ) describes a ﬁrms’ production and

III III *
cost conditions for Z. while optimizing Za ..... 2,, while ¢2 (ZI’ZZ’W’Zn’ﬂ ) describes

b

another ﬁrms’ offer function. The production and cost conditions described by ¢1 make
that ﬁrm better suited to produce lower levels of Z. than the ﬁrm described by ¢2 . In a

competitive market many ﬁrms exist, and many ¢i describing the many ﬁrms will trace

out the producers’ implicit price function for each characteristic inherent in the good Z.

25

 

P(Z,,Z;,...Z;)

 

 

 

 

Z.

FIGURE 4. PRODUCER OFFER CURVES AND THE IMPLICIT PRICE
FUNCTION FOR CHARACTERISTIC 2..

Market Equilibrium

The theoretical basis for the demand and supply for implicit characteristics of a market
good has assumed the existence of a market equilibrium. Assuming a market

' equilibrium, the price vector P(Z) can be considered to be determined by the market
process rather than a consumer and producer set of parameters. The equilibrium implicit
price function is determined through the interaction of consumers and producers. Proﬁt-
maximizing ﬁrms seek to reach the highest offer function. Utility maximizing consumers
seek the lowest bid function. Consumers and producers must now be assumed to meet in

the implicit Z,- market.

Conditions for an efﬁcient market ﬁnd that the MRS is equal to the RPT. Through the

consumers’ maximization process, the MRS (Z, for X) was found to equal the marginal

bid for Z, as well as the price of Z,, R(Z). These conditions were depicted in equation
(3). Through the producers’ maximization process, the RPT (Z, for X) was found to equal
the marginal cost of Z, as well as the marginal offer for Z,- both of which are equal to the
price of 2,, 13(2). By equating the MRS (Z, for X) and RPT (Z,- for X) the marginal bid is
found equal to the marginal offer and both are equal to the price, as presented in equation
(11)

MRS=92 (Z;u.y)=PI.(Z)-—¢Z (Z;ﬁ,ﬂ)=RPT (11)
i i

If the conditions stated in equation (1 1) occur then the conditions for market efﬁciency
are fulﬁlled. Through the demand and supply analysis provided above, the conditions of

equation (11) have been met.

Equation (11) requires the marginal offer to equal the marginal bid. Therefore, the slopes

of the offer and bid curves have to be tangent or equal. The various points at which this

occurs establishes the implicit price vector for characteristic 2,, Pi (Z) . A graphical

example of market equilibrium for the Z I characteristic is presented in Figure 5. Two
points of tangency are displayed in Figure 5. Many offer and bid curves can be imagined

in the Z. -P space and therefore many points of tangency.

27

 

P ¢3(Z,',...,Z;;7r')

 
 
  

P(Z,‘,Z,'....Z,j)
¢'(Z, ,....Z -

92(2,‘,....Z,j,u )

6' (Z:,...,Z,:,u.)

 

 

 

Z.

FIGURE 5. CONSUMER-PRODUCER EQUILIBRIUM AND THE IMPLICIT
PRICE FUNCTION FOR CHARACTERISTIC Z..
A similar analysis of offer and bid curves can be constructed for all Z,- characteristics.
From the analysis of all characteristics the entire implicit price function, P(Z), can be

derived.

Through hedonic price theory, the implicit price function for differentiated goods can be
assumed to exist and therefore estimated. Many studies have utilized this theoretical
structure and estimated various implicit price functions. The next section describes a

portion of the hedonic literature that is relevant to this study.

28

Applied Hedonic Models

The existing applied literature consists of a wide range of hedonic price studies. Hedonic
studies have been applied to numerous topics including labor wages (Clark and Nieves
1994), automobile prices (Arguea and Hsiao 1993), and computer services (Chow 1967).
In forestry, hedonic price models have been used to estimate how timber sale attributes

affect sales prices (see for example Bare and Smith 1999).

The largest segment of hedonic literature concentrates on land and housing values. The
hedonic land value stUdies can be divided into two broad topics: amenity (disamenity)

valuation and growth control and land use impact valuation.
Amenity and Disamenity Valuation

One frequent use of the hedonic framework has been to measure the inﬂuence that
amenities and disamenities have on property values. Environmental amenities such as
open green space (Correl et al. 1978), mature trees (Dombrow et‘ al. 2000), and proximity
to Lake Michigan (Diamond 1980) and other urban water bodies (Brown and
Pollakowski 1977) have been found to positively affect property values through the use
of hedonic modeling. The hedonic price method has also measured the impacts of
numerous disamenities. Disamenities such as aircraft noise (Abelson 1979), crime rates
(Smith 1978 and Diamond 1980), noxious facilities (Clark and Nieves 1994), and

hazardous waste sites (Kiel and McClain 1995, and Kiel 1995) were all found to have

29

negative impacts on local property values. Some perceived disamenities such as
proximity to high voltage electric lines (Hamilton and Schwann 1995) and proximity to
housing for the severe mentally disabled (Galster and Williams 1994) were found to have

no impact on nearby property values.

The hedonic amenity and disamenity literature has found that various factors can
inﬂuence pr0perty values. It is important for any hedonic study to take
amenities/disamenities into consideration. Amenities found in Michigan such as the

Great Lakes and other water bodies must be controlled for in this study.

Of all the amenities and disamenities measured in the literature, none are fully
comparable with those that may occur due to the presence of public land. The objectives
of this study, in part, focus on possible local amenities and disamenities provided by
public land. Public land offers amenities such as esthetic beauty and wilderness to local
residents. Public forestland may also produce possible disamenities such as timber
harvesting or trespass by public land hunters and other users. The amenity and
disamenity literature does provide evidence that public lands, and the activities on public
land, may affect nearby property values. The literature does not provide any concrete
evidence that property values are, in fact, affected by the associated

amenities/disamenities.

Growth Control and Land-Use Impact Valuation

The effect that land-use or growth control regulations have on property values has been
extensively studied. The growth control Studies of land-use policies in Chesapeake Bay
(Beaton and Pollock 1992), San Francisco (Katz and Rosen 1987), and Portland (Knaap
1985) have utilized the hedonic framework. Land-use restrictions in California (Frech
and Lafferty 1984) have been found to increase residential home values. A study of New
Jersey Pinelands land use policies found that developed residential property values went
up while vacant land values went down in some cases (Beaton 1991). The Chesapeake
Bay land-use restrictions were found to increase residential property values and have no
effect on vacant land values (Beaton and Pollock 1992). Environmental protection
zoning on selected rivers in Michigan (Leefers and Jones 1996) exhibited, post-zoning,

value increases for property with housing and no effect on vacant land.

The literature covering land use policy has demonstrated that growth control and land use
regulations do in fact impact land values. Therefore, hedonic studies need to control for

land use and growth control regulations when modeling land values.

Past hedonic land value Studies provide much guidance with respect to relevant variables
and methods. The literature demonstrates the need to control for a variety of
amenities/disamenities and land-use regulations. It also provides a vast array of parcel
Speciﬁc factors that inﬂuence property values. These factors are some of the variables

that the hedonic regression model requires to estimate the implicit price function. The

31

variables are, in essence, required data for hedonic modeling. This study also requires
various public land data. The next chapter, Methods, describes the sampling frame used
to determine where data were collected as well as a description of the type of analyses

conducted.

The aforementioned studies use several multivariate regression techniques relating (dis)
amenities, growth control measures, and land-use restrictions to property values.
Generally, cross-sectional and pooled cross-sectional/temporal data are used. Most of the
studies estimated regressions with sample sizes ranging between 100 and 1500
observations. The developed property models’ adjusted RZ’S varied widely, generally
between 0.30 and 0.70. Vacant property models’ adjusted Rz’s are generally lower,
ranging from 0.20-0.50. Common variables in the studies include: land area, distance
from central business district, distance from interstate highways, age of house, existence
of garage, exterior house material, and date of sale. Other variables used are study
Speciﬁc such as distance from a hazardous waste site, distance from Chesapeake Bay, and

dummy variables identifying if the property is located within a zoned area.

Other Related Literature

Two Objectives of this research are to measure the impact public land has on private
property values, and to measure the impact of timber harvesting on public land on private
property values. Public lands may affect private property values if differences exist

between the immediate public land and the surrounding private land and if property

buyers recognize the differences, or at least perceive that they exist. One difference that

may exist is vegetative treatment, or timber harvesting.

Daniel et al. ( 1973) studied whether observers could reliably discriminate between
various vegetative treatments. Further, Daniel et a1. (1973) studied if observers have
different esthetic responses to the various vegetative treatments. Six different vegetative
treatments were tested in ponderosa pine forest areas. The treatments used were: (1.) a
uniform stripcut, (2.) an irregular stripcut, (3.) a clearcut, (4.) a heavy thin, (5.) a
conventional harvest, which was an area selectively logged 15 years previous, and (6.) a
relict area, which represented an uncut natural area. They found that all treatments were
perceived as treated, with the uniform stripcut the least detectable followed by the
conventional logging. The raw “Perceived Esthetic Value” of each treatment was
adjusted to be relative to that of the relict. The adjusted “Perceived Esthetic Values” of
the heavy thin, irregular stripcut, and clearcut were all lower than the relict, thus less
pleasing. The uniform stripcut and conventional harvest were surprisingly perceived as
more pleasing than the relict. We may assume therefore that property buyers will be able

to detect treated areas, and that the type of treatment will determine the esthetic impact.

How important an esthetic value is relative to the several factors that people consider
when buying property can not be determined from Daniel et al.’s work. The research in
this dissertation incorporates the type of timber harvesting on public lands as an

explanatory variable of the sales price for property adjacent to public land. The

33

coefﬁcient Sign and statistical signiﬁcance reveal whether different types of timber

harvesting affect property values.

Summary

From the theoretical literature, a foundation exists such that the market sales price of a
good can be dissected. Using multivariate regression, the sales price of a good, such as a
property sales price, can be broken into its’ various attributes. Estimation of the
regression provides implicit prices in the coefﬁcients for all attributes. This method

assumes that the housing market is competitive and that all markets clear.

The empirical literature provides an array of modeling techniques as well as attributes
that are used as variables in the hedonic regression. The modeling techniques and
variables from previous studies guide the data collection and methods used in this

research.

34

CHAPTER 3: METHODS

This chapter describes the methods used for this dissertation. The data were gathered
from randomly selected regions in the State of Michigan. Within each randomly selected
region, every arms-length property transaction was recorded. The methods consist of
three general parts. The ﬁrst part of methods required a process for randomly selecting
regions in the state. This is presented in the ﬁrst section of this chapter, Sampling Frame.
The second part involved determining required sales and property data and the collection
of that data. The second section of this chapter discusses this part. Analysis of data is the

ﬁnal part of methods and is presented in the third section of this chapter.

Sampling Frame

The framework that determines the selection process of areas to be studied is often
referred to as the sampling frame. The sampling frame for this dissertation was
determined through four steps. First, the public lands for the study were chosen. The
second Step involved the determination of the sample areas, or regions from which a
random selection is made. Next, methods for controlling effects that have been
determined to inﬂuence property values were devised. Together, the ﬁrst three steps
determine the sample areas. The last step involved the determination of the speciﬁc
study areas from the available sample areas. All property sales observations within each

study area were collected. The four steps noted here are described in detail below.

35

Determination of Public Lands

Possible public lands considered for the purposes of this study include: all National and
State forests, National and State wildlife refuges, National and State parks, and State
game areas in the State of Michigan. These lands represent the majority of publicly
owned land in the State excluding correctional facilities, military installations, and
township and county level properties. Table 1 presents the acres and distribution of the
public lands in Michigan. The largest portion of Michigan’s public land, 87.8 %, is

National and State forest.

 

 

 

 

 

 

 

 

 

 

 

TABLE 1. ACRES AND PERCENT DISTRIBUTION OF PUBLIC LANDS IN
MICHIGAN.
Percent of Total Public

Public Land Type Acres (000) Land in Michigan
National forests 2,797 36.9

National parks and lakeshores 225 3.0

National wildlife refuges 111 1.5

State forests 3,857 50.9

State game and wildlife areas 299 3.9

State parks and recreation areas 261 3.4

 

 

 

Source: Travel and tourism in Michigan: A statistical proﬁle. 1991. Travel, Tourism,
and Recreation Resource Center. Michigan State University.

36

Determination of Sample Areas

Multiple factors contributed to the determination of sample areas. Data availability, the
objectives of this study, and data collection efﬁciency are just a few of these factors. To

best summarize the inﬂuence of these factors, the data that is available is described ﬁrst.

Private property sales data in Michigan is generally available from two sources. At the
county level, the Equalization Department records all sales transactions within the county
for usually the past ten years. Within the Equalization Department rolls, various pieces
of information are available. The county rolls state whether the sale is considered “arms-
length”, being a fair market transaction. The county rolls give the date and amount of all
fair market transactions. They also contain information on whether the property is a
primary residence or a secondary home. The second source of information is at the
township level. The township Assessor’s ofﬁce records information on various property
speciﬁc attributes. The lot Size, house size, assessor’s ranking, exterior material, and the
existence of a garage are but a few pieces of information available at the Assessor’s
ofﬁce. To record all required data, visits to both the county Equalization Department and

the township Assessors ofﬁce are needed.

The objectives of this study require data on private properties that vary in distance from
public land. Therefore, the geographical size of the sample areas is an issue. The sample
areas should demonstrate a consistent size, which will help assure that no weight or bias

is introduced which may favor particular sample areas. Besides being consistent, the size

37

should be large enough to capture the effects that this study attempts to measure. Three
units of boundary distinction are used in Michigan, they are: the county, the township,
and the section. Counties are the largest, generally they measure 36 miles by 36 miles.
Townships are the second largest and are found within counties. Generally, there are 36
townships per county, measuring six miles by six miles. The last and smallest area is the
section. Usually thirty-six sections make up one township. Each section is one mile by

one mile in dimension.

 

The county, as the sample unit, provides a distance from public land from 0 to 36 miles. N
Thirty-six miles may be too large. The possible impact of public land on private property

values is more likely to occur at distances nearer public land rather than farther. The

county as a sample unit may also be too large for another reason. The size of the county

restricts data to be collected from one, possibly two counties. Each county requires up to

thirty-six visits at the township level for data collection. Discussed later in this chapter,

ten townships were visited for data collection reasons. Therefore, sampling at the county

level greatly restricts the data to come from one or two counties, and this in turn restricts

the geographic regions.

In general, townships in Michigan measure 36 square miles. Due to the general
symmetry of township sizes, the township unit provides a fairly consistent sample area.
Though not all townships in the State are 36 square miles, the majority of townships are
consistent in size. The possible bias introduced using the township as the sample area is

negligible.

38

The township unit is likely to be large enough to capture the possible impact of public
lands, and timber harvesting on public lands, on private property values. The size of the
township allows the models to incorporate a proximity measure (i.e., distance of private
property from public property) ranging from zero up to six miles. This range is likely to
be adequate to determine the impact of proximity of public lands on private property

values. The range in Size for a section is probably too small.

The township also provides a data collection efﬁciency over that of the section. With one
visit to an Assessor’s ofﬁce, sales from thirty-six sections were collected. If the section
were the sample area, then more visits to more Assessors ofﬁces would have been

required for a Similar sample size.

All of the Objectives for this study could have been addressed using either county,
township, or section as sample units. Each possible sample unit would likely provide
adjacent private property sales, required by all objectives. Each possible sample unit
would provide differing distances of private parcels from public lands, though the county
unit has the potential for the distance range to be too wide while the section unit has the
potential for the distance range to be too small. Each geographic region could provide
differing public land ownership, as required for objective 3. Each sample unit measure
could provide occurrences of timber harvesting near adjacent private properties, required

for objective 4.

39

Due to the data collection efﬁciencies and the ability to best capture the data required to
address the objectives the township was determined the best sample area from which the
Study areas were drawn. The sample areas for this study consists of nearly all townships
that abut or contain public land. Some townships have been omitted from the sample
areas due to certain attributes. Townships exhibiting attributes such as neamess to
Metropolitan Statistical Areas and the Great Lakes were excluded. The consideration and

treatment of such attributes, or inﬂuential effects, are described in more detail next.

Control of Inﬂuential Effects

Past studies have determined that proximity of private property from speciﬁc land
features has inﬂuenced property values. Features such as the Great Lakes, metropolitan
areas, inland lakes, rivers, and wetlands have demonstrated impacts on private property
values. Two of the ﬁve features noted above, namely the Great Lakes and metropolitan
areas have been controlled for in the sampling frame of this study. A reason for
controlling for such land features in the sampling frame, rather than in the hedonic model,

is simplicity. Fewer explanatory variables will need to be included in the models.

Diamond (1980) demonstrated that Lake Michigan could have a positive impact on
private property values at a distance of up to ﬁve miles. Frech and Lafferty (1984) found
that the California coast from a distance of 13 miles affected housing prices. One
important difference between these studies involves the topography of the respective

study areas. The topography typically found in Michigan is not nearly as mountainous as

40

that of California. The View of a waterbody in California can, generally, take place at a
much farther distance than in Michigan. Therefore, all sample areas in this study will lie
at least ﬁve miles inland from Lake Michigan and other Great Lakes. A ﬁve-mile

distance from the Great Lakes Should greatly reduce the probability of their inﬂuence on

property values.

Land studies have demonstrated the positive inﬂuences that proximity to metropolitan
areas has on property values (see for example Beaton and Pollock (1992)). Proximity to
metropolitan areas serves as a proxy for various attributes including consumption and
labor activities. To control for the inﬂuence of large metropolitan areas, all the sample
areas should lie some distance from such an area. Metropolitan Statistical Areas (MSA),
as deﬁned by the US. Census Bureau, are areas that have populations greater than
150,000. Sample areas within 35 miles from MSA’S were omitted. The 35-mile distance
is somewhat arbitrary. Evidence suggests 20 miles is the average commute distance in
Michigan. Thirty-ﬁve miles should dispose of a majority of any impact that MSA’S may

have on property values.

Past land-use literature has found that zoning may impact property values. Due to the
multitude of types of zoning in Michigan, it is not feasible to classify all types of zoning
and include all classiﬁcations in the hedonic models without further congesting the
models with variables. Incorporated villages and cities generally administer more

speciﬁc and more inclusive zoning regulations than those not incorporated. Therefore,

41

properties that lie within incorporated Villages and cities were excluded during data

collection.

As mentioned above waterbodies such as inland lakes and rivers have been determined to
affect private property values. For reasons related to the efﬁciency of data collection,
proximity to inland lakes and rivers was not controlled for in the sampling frame.
Though not controlled for in the sampling frame, the inﬂuence of inland lakes and rivers
is accounted for in the hedonic models. This issue is addressed further in the Data

Analysis section at the end of this chapter.

Selection of Study Areas

The total number of sample areas, excluding the omitted areas, exceeded 300. Due to
time and monetary considerations the list of areas to be used in this research was reduced.
The areas to be used in this research are referred to as study areas. Study areas were
randomly selected from the sample areas. Twenty-ﬁve potential study areas were
randomly selected from the list of over 300 sample areas. The sequential list of potential

Study areas is presented in Table 2.

All arms-length property sales within ten of the randomly chosen Study areas were
collected. Data collection began at the top of the list and proceeded downwards with no
a priori knowledge of the number of sales within individual townships. Data were

collected from August 1998 through May 1999. Some townships were excluded from the

42

 

potential study areas due to data recording methods at the county level. Some counties’
recording methods were inconsistent with the methods used by the rest of the state. The
primary inconsistency was that these counties did not record private property sales over
the entire study period. Townships located within these counties were therefore
excluded. The exclusion of townships from the list of 25 potential study areas is further
described in the Data Collection section. Due to time and monetary considerations, the
list was not fully exhausted. The study areas used for this dissertation are presented in
Table 3. These 10 townships located in 10 different counties provided a good geographic
dispersal and mirrored the primacy of state forestlands, followed by national forestlands.

A map of the counties in which the study areas lie is presented in Figure 6.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 2. RANDOMLY SELECTED LIST OF 25 TOWNSHIPS IN MICHIGAN
THAT CONTAIN PUBLIC LAND, BY COUNTY.
Random County Township National State Other
No. Forest Forest

1 IOSCO Alabaster X

2 CHEBOYGAN Benton X

3 GOGEBIC Bessemer X X
4 SCHOOLCRAFT Hiawatha X X

5 MENOMINEE Cedarville X

6 DICKINSON Sagola X

7 LAKE Cherry Valley X

8 NEWAYGO Barton X

9 PRESQUE ISLE Krakow X

10 WEXFORD Slagle X

1 l OSCEOLA Hersey X

12 ONTONAGON Rockland X X

13 CHIPPEWA Pickford X

14 CHARLEVOIX Chandler X

15 CHIPPEWA Rudyard X X

16 MISSAUKEE Pioneer X

17 GRAND TRAVERSE Union X

18 MONTCALM Evergreen X

19 PRESQUE ISLE Bismark X

 

43

 

 

Table 2. (continued).

 

 

 

 

 

 

 

20 GRAND TRAVERSE Whitewater
21 BENZIE Weldon

22 OTSEGO Otsego Lake
23 WEXFORD Selma

24 CHIPPEWA Raber

25 HOUGHTON Laird

 

 

 

ABLE 3. STUDY AREAS, BY COUNTY.

County

EBIC
SCHOOLCRAFT
ICKIN SON

LA
NTONAGON
OD(
A
AUKEE

ownship

Sl
erse
Rockland

R
Pioneer

 

44

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

l

 

 

 

 

 

 

/

FIGURE 6. MAP OF STUDY AREAS, BY COUNTY.

 

 

 

 

 

 

 

 

45

Data Collection

This research required data that was available from several sources. The ﬁrst source was
various County Equalization Departments throughout the State of Michigan. Each
County Equalization Department recorded arms-length property transactions within the
respective county for the past ten years. Each County Equalization Department also
recorded whether each parcel was a primary residence or a secondary home. Some
County Equalization Departments did not record sales for the ten previous years. These
counties only had the sales from less than ten previous years available. The townships
located within these counties were excluded from the study areas. Inclusion of these
townships would have introduced some problems related to the study’s time period.
These townships would be represented in the hedonic models by only the latest year’s
sales while all other townships would be represented by the past ten year’s sales. A
possible bias could have occurred in the early years of the study period because the
models would not accurately reﬂect all the study areas over the entire study period.
Rather, the models would have represented only the study areas with complete sales data
in the earlier time periods, and then accurately represent all the study areas in the latter
years of the study period. The reliability of the earlier years of the hedonic models would

have been in question and all of the hedonic model results would have been questionable.

The second source of data was the Township Assessors office. Each assessor’s office has

a parcel record for each property in the respective township. The parcel records contain

46

information such as the square footage of a home, the acreage of the property, existence

of a garage, the exterior material of the home, and many other parcel speciﬁc attributes.

The third source of data was the United States Forest Service (USFS) Field Ofﬁces and
the State of Michigan Department of Natural Resources (MDNR) Field Ofﬁces. These
ofﬁces provided the forest types and harvesting activities of all the public land areas
researched in this study. The forest types information included descriptions Of forest
species composition and density of all public forests located within the study areas. The
harvesting activities information included the type of harvest and the year the harvest was

completed for all public land located within all the study areas.

The last source of data was a variety of maps. Proximity of properties from public land,
highways, and urban centers were measured from township maps. Township maps also
provided the identiﬁcation and proximity of inland lakes and rivers. County maps
allowed for density of public land measures to be calculated. USPS and MDNR maps

were used to measure proximity of properties from timber harvesting on public land.

Data Analysis

The ﬁrst part of this section provides a complete description of the data collected. The

second part of this section provides a general form of the models that are estimated. The

last part describes this studies approach to the problems related to functional form. All

regression analysis is estimated with Ordinary Least Squares (OLS). OLS is the

47

estimator of choice because of the desired properties under the assumptions of the
classical linear regression model (see Kennedy 1996). Under these conditions, OLS
produces the smallest sum of squared residuals, unbiased coefﬁcients with the smallest

variance, and asymptotically efﬁcient coefﬁcients.

Description of Data

The literature review, dissertation guidance committee suggestions, and data availability
were the guiding factors for identifying data collected. The empirical literature
demonstrated that various parcel speciﬁc attributes, such as square footage, lot size,
existing zoning, and existence Of a garage contributed to property values. To the extent
possible, all available parcel speciﬁc data were collected. Other possible inﬂuences, such
as proximity to prisons, forest type on public lands, and the density of public land were
thought to possibly affect property values by committee members. Data for these
variables were also collected. These variables are divided into three categories: Property
Attributes, Public Land Attributes, and Forest and Timber Harvesting Attributes. The

variable name is followed by a brief description.

Property Attributes:
PRICE The nominal US. dollar sales price of the property.
HOUSE The square footage of the home at the time of the sale.
LOT The Size in acres of the property at the time of the sale.
DATE Tire year the property sold. All sales are between 1988 and 1997.

48

 

CITY

HWY

R_IN T

GSP

REGEUP

REGWUP

REGNLP

VACANT

GARAGE

PRISON

ZON ED

BR

Proximity in miles from nearest metropolitan area with a population of
10,000 or more. This was measured from plat maps with a scale of l
1A inch equals one-mile.

Proximity in miles from nearest State or Federal highway. This was
measured from plat maps with a scale of 1 1%: inch equals one-mile.
The 30-year interest rate at the time of the sale.

The Gross State Product at the time of the sale.

A dummy variable indicating the property is located in the Eastern
Upper Peninsula.

A dummy variable indicating the property is located in the Western
Upper Peninsula.

A dummy variable indicating the property is located in the Northern
Lower Peninsula.

A dummy variable indicating that the property is vacant. not
developed.

A dummy variable indicating that a garage exists on the property.

A dummy variable indicating that a prison is located within 30 miles.
A dummy variable indicating that the property is zoned at either the
township or county level.

A dummy variable indicating that the home has a brick exterior.

A dummy variable indicating that the home has received the highest

quality ranking by the township assessor.

49

 

D A dummy variable indicating that the home has received the lowest
quality ranking by the township assessor.

MH A dummy variable indicating that the home is a mobile home.

PAVE A dummy variable indicating that the road the property abuts is paved.

HOME A dummy variable indicating that the home is a primary residence.

PLAT A dummy variable indicating that the property is platted.

H20 A dummy variable indicating that the property abuts a river or lake.

UTIL A dummy variable indicating that the home has publicly provided
electricity.

Public Land Attributes:

FED A dummy variable indicating that the nearest public land is federally
owned.

STATE A dummy variable indicating that the nearest public land is State
owned.

COU A dummy variable indicating that the nearest public land is county
owned.

F_DIS Proximity, in feet, of the property boundary from the federal land

boundary if federal land is the nearest public land. This distance was
measured off plat maps with a scale of 1 M. inch equals one-mile. The
distance measures were made along a straight line from nearest

boundary point to nearest boundary point.

50

S_DIS

PUBDIS

ABUT

FEDA

STATEA

COUA

PERPUB

PERADJ

PERDIS

Proximity, in feet, of the property boundary from the state land
boundary if state land is the nearest public land. This distance was
measured from plat maps with a scale of l 1A inch equals one-nrile.
The distance measures were made along a Straight line from nearest
boundary point to nearest boundary point.

Proximity, in feet, of the property’s nearest boundary from the nearest
public land boundary.

A dummy variable indicating that the property is adjacent to public
property.

A dummy variable indicating that the property is adjacent to federally
owned public property.

A dummy variable indicating that the property is adjacent to State
owned public property.

A dummy variable indicating that the property is adjacent to county
owned public property.

The percent of area within the property’s township and the
surrounding eight townships that is publicly owned. This was
measured from plat maps with a scale of 1 Mi inch equals one-mile.
The percent of the property that abuts public property.

An interaction variable that is the product of PERPUB and the
proximity, in feet, of the property from public land. The greater the
percent of public land, the less that proximity may affect the sales

price of the property.

51

Forest and Timber Harvesting Attributes:

CONIFER

Dummy variable indicating that the nearest forest is predominantly

coniferous.

HARDWOODDummy variable indicating that the nearest forest is predominantly

HARVEST

FEDH

STATEH

COUH

CC

hardwoods.

A dummy variable indicating that timber harvesting has occurred on
adjacent public property within one-quarter mile and within the ten
years previous to the sale of the property.

A dummy variable indicating that timber harvesting has occurred on
adjacent federal public property within one-quarter mile and within the
ten years previous the sale of the property.

A dummy variable indicating that timber harvesting has occurred on
adjacent state public property within one-quarter mile and within the
ten years previous the sale of the property.

A dummy variable indicating that timber harvesting has occurred on
adjacent county public property within one-quarter mile and within the
ten years previous the sale of the property.

A dummy variable indicating that a clear-cut has taken place within
one-quarter mile from adjacent property and within ten years previous

to the sale of the property.

52

CC_P

SEED

SEED_P

REM

REM_P

SEL

SEL_P

An interaction variable that is the product of CC and PER_PUB. The
greater the percent of public land, the less the effect that a clear-cut
may have on the sales price of the property.

A dummy variable indicating that a seed tree out has taken place
within one-quarter mile from adjacent property and within ten years
previous to the sale of the property

An interaction variable that is the product of SEED and PER_PUB.
The greater the percent of public land, the less the effect that a seed
tree cut may have on the sales price of the property.

A dummy variable indicating that a removal cut has taken place within
one-quarter mile from adjacent property and within ten years previous
to the sale of the property

An interaction variable that is the product of REM and PERPUB. The
greater the percent of public land, the less the effect that a removal cut
may have on the sales price of the property.

A dummy variable indicating that a selective cut has taken place
within one-quarter mile from adjacent property and within ten years
previous to the sale of the property

An interaction variable that is the product of SEL and PERPUB. The
greater the percent of public land, the less the effect that a selective cut

may have on the sales price of the property.

53

THIN A dummy variable indicating that a thinning has taken place within
one-quarter mile from adjacent property and within ten years previous
to the sale of the property

THIN_P An interaction variable that is the product of THIN and PER_PUB.
The greater the percent of public land, the less the effect that a

thinning may have on the sales price of the property.

The harvesting variables were not recorded for all properties, only adjacent properties.
The recording of harvesting variables for non-adjacent properties would have to include a
proximity measure. The decision of how many and which nearby timber harvesting
activities to include as attributes for each non-adjacent property would increase the
number of variables in the already congested hedonic models. It is likely that the price of
adjacent properties would experience the brunt of timber harvesting activities relative to
non-adjacent properties based on proximity alone. A simpler dummy variable for
adjacent property should provide evidence as to whether the speciﬁc harvesting activity

has any impact. Statistical information of each variable is provided in the next chapter.
General Form of Model

Given the data described above, it is now possible to represent the hedonic models in a

general form. The general linear form of a hedonic model is:

= * -‘
PRICEI. ﬁ0+§ﬁj X]. 4 e1.

54

where PRICE. is the sales price of private property i, 60 is the intercept, 2 ﬂ]. * X}. is
j

the sum of all coefﬁcients multiplied by the respective property attributes for each

attribute j , and 8i is the error for each property sale 2'.

Four hedonic models were estimated. The ﬁrst two included the entire data set, and are
referred to hereafter as the full models. One estimated a hedonic model for developed
private property while the other estimated a hedonic model for vacant or undeveloped
private property. Three reasons exist for dividing vacant and developed property models.
First, past studies have demonstrated that the relationship between developed property
and vacant property prices and the property attributes behave differently (see for example
Beaton and Pollock 1992). Second, the dependent variable (PRICE) mean values and
variances differ greatly between vacant and developed samples. Finally, there is no
reason to believe that public land and timber harvesting on public land will have similar
impacts, if any, on vacant and developed properties. Therefore, models of developed and
vacant property were estimated independently. The last two models included only the
properties that are adjacent to public land, and are referred to hereafter as the adjacent
models. As with the full models, the adjacent models have two separate models, one for

developed and another for vacant property.
The full models are used primarily to address the ﬁrst, second, and third objectives of this

research. The adjacent models are used to address the fourth objective. Disc‘ussion of the

relationship between speciﬁc models and objectives are discussed next.

55

The ﬁrst objective of this research is to estimate the impact of Michigan’s public lands on
sales prices of adjacent private lands. This objective is addressed with two full models,
which contain all observations for either developed property or vacant property. The full
model coefﬁcients of the F_DIS and S_DIS variables determine if a Statistically
signiﬁcant impact exists at increasing distances from public land. The distance variables
F_DIS and S_DIS will be replaced with FEDA and STATEA providing a simpler model
but with less information. The coefﬁcients to FEDA and STATEA will determine if any
impact caused by public land exists due to adjacency. Additionally, another variable,
PER_PUB will explain the effects that the amount or density of public land has on
private property. The coefﬁcient of the PER_PUB term in both the full models and
adjacent models indicate if the density of public land has any impact on private property.
The coefﬁcient of the PER_DIS term in the full models indicates if distance from public
land interacts with the percent of public land in the surrounding area to inﬂuence property

values.

The second objective of this research is to estimate the impact of proximity (distance) to
public lands on private land sales values. Full models, with all observations, are used to
address this objective. The coefﬁcients of the variables F_DIS and S_DIS from the full
models determine if any statistical signiﬁcance exists between proximity of public land

and private property values.

The third objective of this research is to compare impacts of various types of public land

use (ownership) on nearby private land values. Full models are used to address this

56

objective. The full model coefﬁcients of the F_DIS and S_DIS variables determine if
any statistical signiﬁcance exists between proximity of federal and State public land and

private property values.

The last objective of this research is to estimate the impact of public land timber
harvesting on adjacent private property values. Two adjacent models, with only adjacent
property observations, for developed and vacant properties are estimated to address this
objective. The coefﬁcients of the adjacent model CC, CC_P, SEED, SEED_P, REM,
REM_P, SEL, SEL_P, THIN, and THIN_P variables will determine if the respective
forms of timber harvesting have any statistically signiﬁcant impact on private property
values. The same variables are included in the full model, but the coefﬁcients are of
lesser importance there. The variables are included to control for the possible impacts of
timber harvesting, adding additional information to the model. Since the non-adjacent
observations do not have the timber harvesting information the comparison of these
variables between adjacent and non-adj acent observations is not as robust as that in the
adjacent model. The coefﬁcients of these variables in the full models represent the above
comparison between the adjacent and non-adjacent observations. An additional variable,
PERADJ, is included to control for the possible impact that the amount of public land

frontage may have.

57

......

Functional Form Considerations

The hedonic theory presented in Chapter 2 provided the framework of multivariate

regression as the analysis tool for hedonic pricing. Keen observers will note that the
theory does not provide any insights into the functional form relationship between price
and the attributes of the differentiated product. The speciﬁed functional form may have
an impact on the coefﬁcients of the variables of interest, especially if multi-collinearity

exists. Therefore, it is important to Specify the models functional form as theoretically

and empirically sound as possible.

Several steps were undertaken in this research to provide an adequate assessment and

speciﬁcation of the ﬁrnctional form of the dependent and independent variables. The

steps are outlined in Table 4.

 

 

 

 

 

 

 

 

 

TABLE 4. OUTLINE OF STEPS USED TO ASSESS AND IDENTIFY THE
FUNCTIONAL FORM OF EACH MODEL.

Step A Estimate all models with linear price and cubic continuous
independent variables.

Step B Estimate all models with logarithmic price and cubic
continuous variables. Compare to model results from Step A.
Determine proper dependent variable form.

Step C Correct models for heteroscedasticity where needed.

Step D Test the quadratic forms of the independent variables.
Determine “best” cubic form models.

Step E Test the logarithmic forms against “best” cubic forms for
independent variables.

Step F Determine “best” functional forms.

 

 

In Step A, the model is estimated with a linear dependent variable, PRICE, and cubic
continuous independent variables, such as house size, lot size, and proximity distances.

The Step A models are presented in the respective sections for both the Full Developed

58

'H “'7“. n-jl-q

Model and the Full Vacant Model. Step B estimates nearly the same model as Step A
except the dependent variable is the logarithm of price. The results of the Ramsey
RESET comparisons determine the form of the dependent variable in Step A and Step B
The logarithmic price variable is used for all further steps in all models. The models are
then re-estimated in Step B using a logarithmic dependent variable. Several model test
statistics (i.e., Ramsey’s RESET, F, and LM heteroscedasticity) of the Step A models are
compared to the Step B models. The Ramsey RESET test estimates the model with the
ﬁtted dependent variable squared and the ﬁtted dependent variable cubed as additional
independent variables. The statistical signiﬁcance of the additional squared and cubed
ﬁtted dependent variable terms coefﬁcients reveal whether additional systematic
curvature exists and is not being accounted for. The F -test jointly tests the signiﬁcance
of all independent coefﬁcients equal to zero. The LM (Lagrange Multiplier)
heteroscedasticity test is computed by regressing the squared residuals on the squared
ﬁtted values of the regression. Then the slope of the log-likelihood function with respect
to the coefﬁcient vector is tested to be signiﬁcantly different from zero. Step C corrects
for any heteroscedasticity present in the Step B estimates. The models exhibiting

heteroscedasticity were corrected using White’s variance-covariance matrix.

Once the functional form of the dependent variable up through Step C was deterrninied to
be logarithmic, the functional form of the independent variables needed to be decided.
Step D tests the statistical signiﬁcance of the squared and cubed terms of the quadratic
form of the independent variables. This step determines the “best” functional form for

the case of the logarithmic-cubic functional form. Step E challenges the functional form

59

 

of the “best” logarithmic-cubic model from Step D. In Step E, an artiﬁcial regression is
estimated which contains all the variables from the “best” logarithmic-cubic model as
well as logarithmic forms of all the continuous independent variables. In essence, the
artiﬁcial regression is attempting to capture all curvature between the logarithmic price
and each independent variable as explained by both logarithmic and cubic forms. Step E
then tests the statistical Signiﬁcance of the logarithmic forms jointly and cubic forms
jointly. These statistical tests determine if one form, logarithmic or “best” cubic,
accounts for a majority or all of the curvature between the independent and dependent

variables. Step F then estimates the model determined in Step E to be the better ﬁt.

The method provided by Steps A-F has not been conducted in any of the empirical
literature. This method offers two advantages over popular methods. First, the method
here allows for a more ﬂexible functional form that is less arbitrary than the methods
used in the current literature. Second, the method results here have a wider and more

useful interpretation compared to the method results that exist in the literature.

The ﬁrst advantage of the methods presented in Steps A-F, compared to the empirical
literature, is the more ﬂexible and less arbitrary functional form. Kiel (1995) uses a
hedonic model with a logarithmic price dependent variable and an arbitrary mixture of
quadratic, linear, and logarithmic independent variables. Beaton and Pollock (1992)
estimate a log-log functional form with one linear independent variable. Galster and
Williams (1994) estimate a log-linear functional form. Milon, Gressel, and Mulkey

(1984) estimate several functional forms. They estimate a Box-Cox ﬂexible functional

60

 

form, which restricts all continuous independent variables to have the same functional
form with the dependent variable. They also estimate linear, logarithmic, semilog,
inverse semilog, quadratic, inverse exponential, and simple Box-Cox functional forms for

comparison to the Box-Cox using a likelihood ratio.

AS seen from the literature, several functional forms have been estimated. Most of these
forms are either linear, log-linear, log-log, or some mixture of the three. The decision to
use these speciﬁc functional forms in the existing literature is usually arbitrary. The
ﬂexible Box-Cox is less arbitrary, but offers problems of its own. The arbitrary nature of
selecting functional form applies a restriction to the relationship between independent
and dependent variables. This may rule out the form that best ﬁts the data. The method
used here allows for variations of all of the four forms mentioned above with the addition
of the cubic form. The choice of functional form used here is not arbitrary but relies on a
series of statistical tests. The preferred log-cubic model (which is actually a mixture of

' cubic and linear independent terms) is statistically determined. This model is then
statistically compared to a log-log model using an artiﬁcial regression. The result
provides a unique relationship between each independent variable and the dependent
Variable. The Box-Cox form is restricted here. It assumes that all independent variables
have the same functional form with the dependent variable. As mentioned above, the

Box-Cox has a further problem, which is discussed below.

The second advantage of the method provided in Steps A-F is that the results are more

interpretable than those from some of the other functional forms. Milon, Gressel, and

6|

 

Mulkey (1984) point out, when using the Box-Cox transformation, that the implicit price
of amenities is dependent upon the levels of all other characteristics. This causes a
difﬁculty of attempting to “untie” amenities and therefore interpreting coefﬁcients.
Hamilton and Schwann (1995) mistakenly dismiss both log-linear and linear functional
forms. They then resort to a Box-Cox translog functional form. They make conclusions
on the absolute average level of impact of several characteristics without informing the If
reader how they came up with these ﬁgures and what assumptions they made while they

“untied” amenities. Because each characteristic’s impact on price is dependent upon the

 

level of other characteristics, the correlation of independent variables is important. The
problems associated with collinearity of independent variables, omitting relevant
variables, and including of irrelevant variables may be compounded in the Box-Cox
form. Collinearity is clearly a problem in all other functional forms, but the other

problems may have much more of an impact in the Box-Cox.

CHAPTER 4: ANALYSIS

This chapter describes the analysis of the data. The ﬁrst section reviews the descriptive
statistics of the data. Descriptive statistics of the entire data set are ﬁrst reviewed. Next,
data used in the Full Models and the Adjacent Models are reviewed. The second section
covers the model estimation and parameter testing of Steps A-F, outlined in the previous
chapter, for the Full Models. The third section does the same for the Adjacent Models.

The last section covers the removal of outliers.

Descriptive Statistics

Descriptive statistics provide useful insights pertaining to the data used to estimate any
model. This section provides a review of the data in several fashions. First, the entire
data set is reviewed. Then a review of subsets of the data follows. The breakdown of the
data into subsets is relevant to further analysis conducted in this chapter. The subsets are
ﬁrst divided into two primary sets: Full Model Data and Adjacent Model Data. The Full
Model Data subset contains all observations. The Adjacent Model Data contains only
observations that are adjacent to public land. These subsets are then further divided into
Developed Model Data and Vacant Model Data which, respectively, contain only
developed properties or vacant properties. The four subsets are those used in the four
hedonic models. A review of the data for each of the four subsets is provided after the

review for the entire data set.

63

“E‘s—n “...—n -wq

Entire Data Set

The entire data set contains 1,339 property sales. with mean sales price (PRICE) of

$22,769, average lot size (LOT) of 15.63 acres, and the mean year of sale (DATE) of

1993 (Table 5). One half of the data consists of vacant property (VACANT), while the

other is developed. The average distance of the properties from a city with a population

of 10,000 or more (CITY) is 38 miles. The average distance to a highway (HWY) is 2.58

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

miles.
TABLE 5. DESCRIPTIVE STATISTICS FOR HOUSING AND MARKET
VARIABLES FROM THE ENTIRE DATA SET (N=1 339).
Variable (unit) Mean St. Dev. Min. Max.
PRICE (8) 22769 23227 625 215000
HOUSE (sq ft) 435 504 0 2930
LOT (acres) 15.63 32.40 0.06 640.00
DATE (year) 1 993 3 1988 1997
CITY (miles) 38.35 36.97 8.75 193.60
HWY (miles) 2.58 4.52 0 19.35
R_INT (%*100) 8.75 1.01 7.40 10.45
GSP (S) 223.61 32.36 175.70 272.61
VACANT (0,1) 0.50 0.50 O l
GARAGE (0,1) 0.22 0.41 0 1
PRISON (0,1) 0.29 0.45 0 1
ZONED (0,1) 0.53 0.50 0 1
BR (0,1) 0.01 0.11 0 l
B (0,1) 0.00 0.04 0 l
D (0,1) 0.22 0.41 0 1
MH (0,1) 0.07 0.26 0 1
PAVE (0,1) 0.29 0.45 0 1
HOME (0,1) 0.26 0.44 0 1
PLAT (0,1) 0.33 0.47 0 1
H20 (0,1) 0.13 0.34 0 l
UTIL (0,1) 0.36 0.48 O l
REGEUP (0,1) 0.29 0.45 0 1
REGWUP (0,1) 0.18 0.38 0 1
REGNLP (0,1) 0.53 0.50 0 l

 

 

 

 

 

 

64

 

Approximately 53% of all properties are under some form of housing or property zoning
(ZONED). A minority of the properties, 26%, are primary residences (HOME). This is
not a surprise considering the amount of vacant property in the data set and is also a
function of the sampling. Most of the properties, 53%, are located in the northern Lower
Peninsula (REGNLP), while 29% are located in the eastern half of the Upper Peninsula
(REGEUP) and the remaining are located in the western half of the Upper Peninsula
(REGWUP). This result is also a function of the sampling, Since most of the southern
Lower Peninsula was omitted from the sample due to the inﬂuence of large metropolitan

areas.

Table 6 presents descriptive statistics for public land and timber harvesting related
variables for the entire date set. Of these properties, 23% abut public land (ABUT).
Eight percent abut federal land (FEDA), while 16% abut state land (STATEA). Eight
percent of the properties abutted public land that experienced timber harvesting within a
quarter mile of the private property previous to the sale date, as seen in the HARVEST
variable statistics. This average, though, is slightly misleading. The HARVEST
variable, among others, was recorded only for abutting properties. For non-abutting
properties, the HARVEST variable is zero. Therefore, the statistical description of the
HARVEST variable for all data reﬂects a mean and standard deviation lower than those
found for abutting properties and positive for non-abutting properties. Other variables
that were only'recorded for abutting properties are: CONIFER, HARDWOOD, BRUSH

OPEN, CC, CC_P, SEED, SEED_P, THIN, THIN_P, SEL, SEL_P, REM, REM_P, and

65

PERADJ. The relevant statistical descriptions for these variables are found in the

Adjacent Model Statistics section.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 6. DESCRIPTIVE STATISTICS FOR PUBLIC LAND AND TIMBER
HARVESTING VARIABLES FROM THE ENTIRE DATA SEHN=1339).
Variable (unit) Mean St. Dev. Min. Max.
ABUT (0,1) 0.23 0.42 0 1
HARVEST (0,1) 0.08 0.28 0 l
FED (0,1) 0.36 0.48 0 l
FEDA (0,1) 0.08 0.26 0 l
FEDH (0,1) 0.02 0.15 0 1
STATE (0,1 ) 0.62 0.49 0 1
STATEA (0,1) 0.16 0.36 0 l
STATEH (Oﬂ 0.06 0.24 0 1
COU (0,1) 0.02 0.15 0 1
COUA (0,1) 0.00 0.05 0 l
COUH (0,1) 0.00 0.00 0 0
CONIFER (0,1) 0.09 0.28 0 l
HARDWOOD (0,1) 0.18 0.38 0 l
BRUSH (0,1) 0.02 0.12 0 1
OPEN (0,1) 0.02 0.13 0 1
CC (0,1) 0.04 0.20 0 l
CC_P (0,1) 0.02 0.10 0 0.76
SEED (0,1) 0.00 0.05 0 l
SEED_P (0,1) 0.00 0.03 0 0.47
THIN (0,1) 0.02 0.15 0 1
THIN_P (0,1) 0.01 0.07 0 0.76
SEL (0,1) 0.01 0.11 0 l
SEL_P (0,1) 0.00 0.05 0 0.76
REM (0,1) 0.0] 0.09 0 l
REM_P (0,1) 0.00 0.05 0 0.76
PUBDIST (feet) 5840 6968 0 35571
F_DIS (feet) 2116 4665 0 35571
S_DIS (feet) 3526 6454 0 35571
PERADJ (%) 0.07 0.17 0.00 1.00
PERPUB (%) 0.41 0.19 0.08 0.76

 

 

The average distance from the nearest public land is 5840 feet, as seen in the mean of

PUBDIST. The descriptive statistics of the distance from federal land, F_DIS, and from

66

state land, S_DIS, variables in Table 6 are misleading. The mean values contain values
for proximity to their respective public land, and zeros otherwise. For instance, F_DIS
has proximity values for only those properties that are closer to federal land than state
land. F_DIS has zeros for all other properties. Therefore, because of the additional
zeros, F_DIS and S_DIS are not directly comparable to PUBDIST. F_DIS, for only
those properties that are nearer federal land (477 properties) has a mean of 5941 feet and
standard deviation of 6198. Similarly, S_DIS, for only those properties (833 properties)
nearer state land has a mean of 5668 feet and a standard deviation of 7405. These
numbers are directly comparable to PUBDIST. An average of 41% of the surrounding

nine-township area (PERPUB) is public land.
AS reported above, 1339 total sales observations were collected. The sample size is

primarily a function of monetary resources. AS much data was collected as possible,

given monetary considerations. The sample size was not predetermined.
Full Model Statistics
The Full Models use all observations in the respective developed or vacant category. The

data used in the Full Developed Model is presented ﬁrst, followed by the data used in the

Full Vacant Model.

67

Developed Full Model Data

Table 7 provides descriptive statistics for the Developed Full Model subset of the data.
This subset consists of all developed property in the entire data set. The Developed Full
Model subset contains 667 property sales. The average selling price for this data is
$31,044 (PRICE), with a smaller standard deviation compared to the entire data set. The
homes vary greatly in size (HOUSE) from 208 to 2930 square feet, with an average of
875 square feet. The average lot size is 11 acres with a large standard deviation of 21.29.
The variable averages for DATE, CITY, and HWY are similar to the entire data set

means.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 7. DESCRIPTIVE STATISTICS FOR HOUSING AND MARKET
VARIABLES FROM THE DEVELOPED FULL MODEL SUBSET (N=667).
Variable (unit) Mean St. Dev. Min. Max.

PRICE (S) 3 1044 24920 1200 210000
HOUSE (sq ft) 875 354 208 2930
LOT (acres) 1 1.03 21.29 0.06 176.00
DATE (year) 1993 3 1988 1997
CITY (miles) 39.64 39.42 8.75 193.60
HWY (miles) 2.59 4.74 0 19.35
R_INT (%" 100) 8.75 0.99 7.40 10.45
GSP (S) 222.73 31.83 175.70 272.61
VACANT (0,1) 0.00 0.00 0 0
GARAGE (0,1) 0.42 0.49 0 1
PRISON (0,1) 0.26 0.44 0 l
ZONED (0,1) 0.54 0.50 0 1
BR (0,1) 0.03 0.16 0 1
B (0,1) 0.00 0.05 0 l
D (0,1) 0.44 0.50 0 1
MH (0,1) 0.15 0.35 0 1
PAVE (0,1) 0.36 0.48 0 1
HOME (0,1) 0.52 0.50 0 l
PLAT (0,1) 0.39 0.49 O 1
H20 (0,1) 0.15 0.35 0 l

 

68

 

 

 

 

 

 

 

 

Table 7 (continued).

UTIL (0,1) 0.67 0.47 0 1
REGEUP (0,1) 0.26 0.44 0 1
REGWUP (0,1) 0.22 0.42 0 1
REGNLP (0.1) 0.52 0.50 O 1

 

 

 

 

 

Approximately 42% of homes have a garage, as shown in the mean of GARAGE. Three
percent have brick exterior (BR). F orty-four percent are ranked in the lowest housing
quality category of D. Fifteen percent of the developed properties have mobile homes
(MH). A small majority of these properties are primary residences (HOME). The
geographical distribution of these properties throughout the State resembles that from the
entire data set but with a slightly smaller percent of properties located in the eastern

Upper Peninsula and northern Lower Peninsula.

The timber harvesting related variables descriptive statistics are provided in Table 8. The
most notable differences of this data subset from the entire data set are found in the
ABUT, HARVEST, PUBDIST, F_DIS, and S_DIS variables. Fewer developed
properties abut public land than vacant, as shown by the mean of ABUT. The mean
percent. of developed properties that abut public land is 20%, down three percentage
points from the entire data set. The percent of developed properties experiencing timber

harvesting (HARVEST) is 7%, one percentage point lower than the entire data set.

The distance variables differ from the entire set. The developed properties are farther
from public land than vacant properties. Developed properties are an average of 6,614

feet (PUBDIST) away compared to 5,846 for all properties. The related variables, F_DIS

69

and S_DIS Show higher mean values for developed property relative to the entire data set
means, but again, these values are misleading. F_DIS has an average of 7031 feet and
standard deviation of 6723 for only those properties (222 properties) nearer federal land
than other public land. S_DIS averages 6201 feet, with a standard deviation of 7685, for

those properties nearer state land (424 properties).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 8. DESCRIPTIVE STATISTICS FOR PUBLIC LAND AND TIMBER
HARVESTING VARIABLES FROM THE DEVELOPED FULL MODEL
SUBSET (N=667).

Variable (unit) Mean St. Dev. Min. Max.
ABUT (0,1) 0.20 0.40 0 1
HARVEST (0,3 0.07 0.26 0 l
FED (0,1) 0.33 0.47 0 1
FEDA (0,1) 0.07 0.25 0 l
FEDH (0,1) 0.02 0.15 0 1
STATE (g1) 0.64 0.48 0 1
STATEA (0,1) 0.13 0.34 0 1
STATEH (0,1) 0.05 0.21 0 1
COU (0,1) 0.03 0.17 0 1
COUA (0,1) 0.00 0.05 0 l
COUH (0,1) 0.00 0.00 0 0
CONIF ER (0,1) 0.08 0.27 0 1
HARDWOOD (0,1) 0.15 0.36 0 l
BRUSH (0,1) 0.01 0.09 0 1
OPEN (0,1) 0.02 0.13 0 1
CC (0,1) 0.03 0.17 0 1
CC_P (0,1) 0.01 0.07 0 0.76
SEED (0,1) 0.00 0.04 0 l
SEED_P (0,1) 0.00 0.02 0 0.47
THIN (0,1) 0.02 0.15 0 l
THIN_P (0,1) 0.01 0.06 0 0.47
SEL (0,1) 0.01 0.12 0 1
SEL_P (0,1) 0.01 0.05 0 0.76
REM (0,1) 0.01 0.08 0 l
REM_P (0,1) 0.00 0.03 0 0.41
PUBDIST (feet) 6614 7326 0 35571
F;DIS (feet) 2340 5098 0 35571
Q18 (feet) 3942 6814 0 35571
PERADJ (%) 0.06 0.15 0.00 1.00
PERPUB (%) 0.41 0.19 0.08 0.76

 

 

 

 

 

70

Vacant Full Model Data

This section summarizes the data used to estimate the Vacant Full Model. Table 9
provides the descriptive statistics for housing and market variables in this model. The
average vacant property sales price (PRICE) is $14,555. The standard deviation of the
sales price ($17,993) is larger than the mean price, demonstrating a large variation in
vacant land prices. The proportionately larger variation in price probably contributes to

lower R2 values for the vacant models compared to developed models.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 9. DESCRIPTIVE STATISTICS FOR HOUSING AND MARKET
VARIABLES FROM THE VACANT FULL MODEL SUBSET (N=672).
Variable (unit) Mean St. Dev. Min. Max.

PRICE (S) 14555 17993 625 215000
LOT (acres) 20.19 40.01 0.06 640.00
DATE (year) 1993 3 ' 1988 1997
CITY (miles) 37.08 34.34 8.75 192.00
HWY (miles) 2.56 4.29 0 18.74

R_INT (%*100) 8.75 1.02 7.40 10.45

GSP (S) 224.49 32.87 175.70 272.61
VACANT (0,1) 1.00 0.00 1 1
GARAGE (0,1) 0.02 0.14 0 1
PRISON (0,1) 0.32 0.47 0 l
ZONED (0,1) 0.52 0.50 0 1
PAVE (0,1) 0.22 0.41 0 1
HOME (0,1) 0.00 0.00 0 0

PLAT (0,1) 0.27 0.44 0 1

H20 (0,1) 0.11 0.32 0 l

UTIL (0,1) 0.05 0.22 0 l

REGEUP (0,1) 0.32 0.47 0 l
REGWUP(0,I) 0.14 0.35 0 1
REGNLP (0,1) 0.55 0.50 0 1

 

 

 

 

 

 

Vacant lot sizes (LOT) also demonstrated a large variation, ranging from 0.06 to 640

acres, with an average of approximately 20 acres. This proportionately larger variation

71

may help to explain the large variation in price. The variables DATE, CITY, and HWY
are similar to the entire data set and the developed model data. Approximately 2 % have
a structure, such as a barn, shed, or garage, as found in the GARAGE variable Statistics.
Slightly more vacant property is in the eastern Upper Peninsula (REGEUP) and northern
Lower Peninsula (REGNLP), while fewer in the western Upper Peninsula (REGWUP)

relative to the entire data set.

Table 10 provides descriptive statistics for vacant property timber harvesting related
variables. Twenty-seven percent of all vacant property abuts public land (ABUT), four
percentage points higher than the entire data set, seven percentage points higher than
developed model data. Vacant properties experienced a one perCentage point higher level
of harvesting activities (HARVEST) than did the data set as a whole. The mean distance
from federal land (F_DIS) is 1,894 feet and 3,114 feet from state land (S_DIS). F_DIS
for only those properties nearer federal land has a mean of 4,993 feet with 255
observations. Similarly, S_DIS has a mean of 5,116 with 409 observations. The mean
distance from public land, 5071 feet, is smaller than the entire data set mean, 5840 feet,
and much smaller than the full model data mean of 6614 feet. The relatively smaller
distance from public property that the vacant property exhibits compared to the
developed property isn’t surprising since the percentage of abutting property is much

higher.

72

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 10. DESCRIPTIVE STATISTICS FOR PUBLIC LAND AND TIMBER
HARVESTING VARIABLES FROM THE VACANT FULL MODEL
SUBSET (N=672).

Variable (unit) Mean St. Dev. Min. Max.
ABUT (0.1) 0.27 0.44 0 1
HARVEST (0.1) 0.09 0.29 0 l
FED (0,1) 0.38 0.49 0 l
FEDA (0.1) 0.08 0.28 0 l
FEDH (0,1) 0.02 0.15 0 1
STATE (0.1) 0.61 0.49 0 1
STATEA (0.1) 0.18 0.38 0 l
STATEH (0.1) 0.07 0.26 0 1
COU (0,1) 0.01 0.11 0 l
COUA (0,1) 0.00 0.04 0 1
COUH (0,1) 0.00 0.00 0 0
CONIFER (0.1) 0.10 0.30 0 l
HARDWOOD (0,9 0.20 0.40 0 1
BRUSH (0,1) 0.02 0.15 O 1
OPEN (0,1) 0.02 0.13 0 1
CC (0,1) 0.06 0.23 0 1
CC_P (0,1) 0.03 0.12 0 0.76
SEED (0,1) 0.00 0.07 0 1
SEED_P (0,1) 0.00 0.03 0 0.47
THIN (0,1) 0.02 0.15 0 1
THIN_P (0,1) 0.01 0.08 0 0.76
SEL (01,) 0.01 0.10 0 1
SEL_P (0,1) 0.00 0.04 0 0.76
REM (0,1) 0.01 0.10 0 1
REM_P (0,1) 0.01 0.06 O 0.76
PUBDIST (feet) 5071 6508 0 31124
F_DIS (feet) 1894 41 84 0 28901
SADIS (feet) ‘ 3114 6052 0 31124
PERADJ (%) 0.08 0.18 0.00 1.00
PERPUB (%) 0.42 0.19 0.08 0.76

 

 

 

 

 

Overall, not many differences exist between the developed full model data and the vacant
full model data. The largest differences are found in the variation around the sales price,
the distance from public land, and the percentage abutting public land. Vacant properties
exhibit a higher variation around the sales price, a Shorter distance from public land, and

a higher percentage abutting public land as compared to developed properties. Slight

73

regional differences also exist. A higher percent of vacant properties are located in the
eastern Upper Peninsula and northern Lower Peninsula as compared to developed

properties.
Adjacent Model Statistics

This section reviews the descriptive statistics for the data used to estimate both the
Adjacent Models. First, the data used to estimate the Adjacent Developed Model is

reviewed. The Adjacent Vacant Model data review then follows.
Adjacent Developed Model Data

Table 11 presents the descriptive statistics for Adjacent Developed Property data. The
mean sales price (PRICE), $28,363, and the mean house size (HOUSE), 777 square feet,
are both smaller than the same means, $31,044 and 875 square feet respectively, for the
Full Developed Model data. The standard deviations of the two variables are fairly
consistent across the two subsets. The average lot size (LOT), 21 acres, is almost twice
as large as the mean lot size of all developed property, 11 acres. The relatively larger lot
size is accompanied by a larger, but proportionately smaller, standard deviation. The
distance to a city (CITY) is Shorter while the distance to a highway (HWY) is further
relative to all developed property. Fewer adjacent developed properties have garages
(GARAGE), compared to all developed property. Adjacent developed properties

demonstrate an increase in the D housing rating and the number of mobile homes (MH).

74

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 11. DESCRIPTIVE STATISTICS FOR HOUSING AND MARKET
VARIABLES FROM THE DEVELOPED ADJACENT MODEL SUBSET
(N=l34).

Variable (unit) Mean St. Dev. Min. Max.
PRICE (S) 28363 24071 1400 130000
HOUSE (sq ft) 777 352 208 1750
LOT (acres) 21.29 27.05 0.16 160.00
DATE (year) 1993 3 1988 1997
CITY (miles) 30.51 26.17 14.25 192.40
HWY (miles) 5.27 6.61 0 19.35
R_INT (%* 100) 8.78 1.01 7.40 10.45
GSP (S) 219.01 32.02 175.70 272.61
GARAGE (0,1) 0.33 0.47 0 1
PRISON (0,1) 0.12 0.33 0 l
ZONED (0,1) 0.48 0.50 0 1
BR (0,1) 0.01 0.09 0 1
B (0,1) 0.00 0.00 0 0
D (0,1) 0.51 0.50 O 1
MH (0,1) 0.21 0.41 0 1
PAVE (0,1) 0.16 0.37 0 l
HOMwJ) 0.36 0.48 0 1
PLAT (0,1) 0.06 0.24 0 1
H20 (0,1) 0.12 0.33 0 l
UTIL (0,1) 0.42 0.50 O 1
REGEUP (0,1) 0.12 0.33 0 l
REGWUP (0,1) 0.09 0.29 O 1
REGNLP (0,1) 0.79 0.41 0 1

 

 

 

 

 

 

 

Compared to all developed property, adjacent developed properties have considerably
fewer paved roads (PAVE), and primary residences (HOME). The adjacent developed
properties geographic locations are heavily skewed relative to all developed property.
Seventy-nine percent of adjacent developed property is in the northern Lower Peninsula
(REGNLP), compared to 52% of all developed property. Twelve percent of the adjacent
developed property is located in the eastern Upper Peninsula (REGEUP) with 9% in the
western half (REGWUP), compared to mean values for all developed property of 26%

and 22%, respectively.

75

Table 12 presents the adjacent developed property public land and timber harvesting-
related variables’ statistical descriptions. The timber related variables in Table 12 are not
directly comparable to those in Table 6 (Entire data set) or Table 8 (Developed Full
Model data set). AS stated earlier, many of these variables were recorded only for

adjacent properties.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 12. DESCRIPTIVE STATISTICS FOR PUBLIC LAND AND TIMBER
HARVESTING VARIABLES FROM THE DEVELOPED ADJACENT
MODEL SUBSET (N=134).

Variable (unit) Mean St. Dev. Min. Max.
HARVEST (0,1) 0.36 0.48 0 l
FED (0,1) 0.33 0.47 0 l
FEDA (0,1) 0.33 0.47 0 1
FEDH (0,1) 0.12 0.33 0 1
STATE (0,1) 0.66 0.48 '0 l
STATEA (0,1) 0.66 0.48 0 1
STATEH (0,1) 0.24 0.43 0 1
COU (0,1) 0.01 0.12 0 1
COUA (0,1) 0.01 0.12 0 1
COUH (0,1) 0.00 0.00 0 0
CONIFER (0,1) 0.38 0.49 0 1
HARDWOOD (0,1 0.75 0.43 0 1
BRUSH (0,1) 0.04 0.21 0 1
OPEN (0,1) 0.09 0.29 0 1
CC (0,1) 0.14 0.35 0 1
CC_P (0,1) 0.06 0.16 0 0.76
SEED (0,1) 0.01 0.09 0 1
SEED_P (0,1) 0.00 0.04 0 0.47
THIN (0,1) 0.11 0.32 0 1
THIN_P (0,1) 0.05 0.14 0 0.47
SEL (0,1) 0.07 0.25 0 l
SEL_P (0,1) 0.03 0.11 0 0.76
REM (0,1) 0.03 0.17 0 1
REM_P (0,1) 0.01 0.06 0 0.41
PERADJ (%) 0.27 0.22 0.00 1.00
PERPUB (%) 0.44 0.10 0.08 0.76

 

 

 

 

 

76

l’ 4'

AS seen in Table 12, thirty-six percent of all adjacent property experienced timber
harvesting (HARVEST) over the study period. Most occurred on state land (STATEH).
Hardwood forests dominate the public land (HARDWOOD), seventy-ﬁve percent of all
adjacent developed property has a public hardwood stand within a quarter mile.

Clearcutting (CC) and thinning (THIN) are the primary harvesting activities.

Various large differences exist between the Developed Full Model data and the
Developed Adjacent Model data. Large differences in sales price, house size, lot size.
distance to city, distance to highway, and geographic locations exist. The vast
differences between the subsets implies that the behavior of the Sales price may be very

different between the data sets.
Adjacent Vacant Model Data

The descriptive statistics for adjacent vacant property are provided in Table 13. Just as in
the case of the two developed property subsets, the vacant property subsets demonstrate
vast differences. Both the mean selling price (PRICE) of adjacent vacant land and the
average lot size (LOT) are higher relative to the respective averages over all vacant land.
The mean adjacent vacant property selling price, $16,218, is $1,663 higher than the mean
selling price of all vacant land. The standard deviation of adjacent property price is
actually smaller than that of all vacant property. The average lot size, 28.46 acres, is 8.37

acres larger than the average of all vacant property. The larger average lot Size is

77

accompanied by a much larger standard deviation. Adjacent vacant properties are farther,
3.99 miles, from highways than are all vacant properties, 2.56 miles. Adjacent vacant
properties demonstrate noticeably smaller percentages in other variables compared to all
vacant property. Neamess to prisons, local zoning, paved roads, platted properties, and
publicly provided water and utilities are less likely to occur for adjacent vacant properties
- relative to all vacant properties. A smaller percent of adjacent vacant properties are
located in the Upper Peninsula (REGEUP and REGWUP) than all vacant properties.

This is also reﬂected in the higher percentage of Lower Peninsula adjacent properties

(REGNLP) relative to all vacant properties.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 13. DESCRIPTIVE STATISTICS FOR HOUSING AND MARKET
VARIABLES FROM THE VACANT ADJACENT MODEL SUBSET
(N=179).

Variable (unit) Mean St. Dev. Min. Max.
PRICE (S) 16218 17133 1000 130000
LOT (acres) 28.46 61.75 0.14 640.00
DATE (year) 1993 3 1988 1997
CITY (miles) 36.68 29.47 15.25 183.00
HWY (miles) 3.99 5.13 O 18.74
R_INT (%"‘ 100) 8.83 1.06 7.40 10.45
GSP (S) 221.12 33.15 175.70 272.61
GARAGE (0,1) 0.04 0.19 0 1
PRISON (0,1) 0.25 0.43 0 1
ZONED (0,1) 0.41 0.49 0 1
PAVE (0,1) 0.11 0.31 0 1
HOME (0,1) 0.00 0.00 0 0
PLAT (0,1) 0.14 0.35 0 1
H20 (0,1) 0.08 0.27 0 1
UTIL (0,1) 0.01 0.07 0 1
REGEUP (0,1) 0.25 0.43 0 1
REGWUP (0,1) 0.06 0.23 0 1
REGNLP (0,1) 0.70 0.46 0 l

 

78

 

Table 14 presents the statistics for the public land and timber harvesting related variables
of the adjacent vacant properties. The percent of adjacent vacant property experienced
slightly lower timber harvesting activities (HARVEST) than did adjacent developed
properties as seen in Table 12. Thirty-ﬁve percent of adjacent vacant properties incurred
timber harvesting within one-quarter mile on public land, compared to 36% of developed
properties. The percent of adjacent vacant property abutting federal (F EDA), 32%, is
Similarly lower than the percent of adjacent developed property, 33%. Related, the
percent of vacant property abutting state (STATEA) public land is 68%, which is two

percentage points higher than developed property.

Less timber harvesting took place near vacant property on federal land than did near
developed property. Nine percent of vacant property experienced timber harvesting on
federal land (FEDH) compared to 12% of developed property. Conversely, more timber
harvesting took place on state land near vacant property than did near developed
property. Twenty-six percent of vacant property experienced nearby timber harvesting
on state land (STATEH) compared to 24% of developed property. The percent of
coniferous and hardwood forests did not vary much between adjacent vacant land and
adjacent developed land. The percent of brush on public land (BRUSH) is twice as high
for vacant properties than for developed properties, 8% compared to 4%. Clearcutting
occurred much more frequently near adjacent vacant property compared to adjacent
developed property. Twenty-one percent of vacant properties (CC) experienced

clearcutting compared to 14% of developed properties. Seed tree cuttings (SEED),

79

thinnings (THIN), selective cuts (SEL), and removal cuts (REM) all differed slightly

between vacant and developed properties.

The percentage of adjacent properties land that abuts public land (PERADJ) is slightly
higher for vacant properties, 29%, than for developed properties, 27%. Similarly, the
percent of public land that makes up the surrounding area (PERPUB) is higher for vacant

properties, 48%, than for developed properties, 44%.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 14. DESCRIPTIVE STATISTICS FOR PUBLIC LAND AND TIMBER
HARVESTING VARIABLES FROM THE VACANT ADJACENT MODEL
SUBSET (N=179).

Variable (unit) Mean St. Dev. Min. Max.
HARVEST (0,1) 0.35 0.48 0 l
FED (0,1) 0.32 0.47 0 1
FEDA (0,1) 0.32 0.47 O l
FEDH (0,1) 0.09 0.29 0 1
STATE (0,1) 0.68 0.47 0 1
STATEA (0,1) 0.68 0.47 0 1
STATEH (0,1) 0.26 0.44 0 1
COU (0,1) 0.01 0.07 0 1
COUA (0,1) 0.01 0.07 0 1
COUH (0,1) 0.00 0.00 0 0
CONIF ER (0,1) 0.37 0.49 0 1
HARDWOOD (0,1 0.75 0.43 0 1
BRUSH (0,1) 0.08 0.28 0 1
OPEN (0,1) 0.06 0.24 0 1
CC (0,1) 0.21 0.41 0 l
CC_P (0,1) 0.10 0.21 0 0.76
SEED (0,1) 0.02 0.13 0 1
SEED_P (0,1) 0.01 0.06 0 0.47
THIN (0,1) 0.09 0.29 0 1
THIN_P (0,1) 0.04 0.15 0 0.76
SEL (0,1) 0.04 0.19 0 l
SEL_P (0,1) 0.02 0.08 0 0.76
REM (0,1) 0.04 0.19 O l
REM_P (0,1) 0.02 0.11 0 0.76
PERADJ (%) 0.29 0.24 0.00 1.00
PERPUB (%) 0.48 0.13 0.25 0.76

 

 

 

 

 

80

AS is the case with the developed property data, the adjacent vacant property differed
from all vacant prOperties in numerous ways. The adjacent vacant properties demonstrate
a higher mean price, lot Size, and distance from highway when compared to all vacant
properties. Fewer adjacent vacant properties were in the vicinity of prisons, zoned, and
platted as compared to all vacant properties. Geographic location differences are also
found between the two vacant subsets. The large differences between adjacent and
developed data imply that the implicit prices, as a function of sales price, will behave

differently between adjacent and full models.

Full Models

This section describes the Full Developed and Vacant Models. As described previously,

the following steps are used to assess and identify the functional form of each model:

Step A Estimate all models with linear price and cubic continuous
independent variables.

Step B Estimate all models with logarithmic price and cubic
continuous variables. Compare to model results from Step A.
Determine proper dependent variable form.

Step C Correct models for heteroscedasticity where needed.

Step D Test the cubic forms of the independent variables. Determine

“best” cubic form models.

81

Step E Test the logarithmic forms against “best” cubic forms for
independent variables.

Step F Determine “best” functional forms.

All regressions presented in this dissertation are estimated using the Ordinary Least
Squares procedure. The models and tests used in Steps B-E are not presented in this
chapter, rather they are discussed in this chapter and the actual model estimates are
provided in Appendix A. The same information for Adjacent Models is presented in

Appendix B. Models for Step A and Step F are presented in this chapter.

Developed Property Full Models

The Step A, or linear-cubic, estimation results for the Developed Full Model are
presented in Table 15. The Developed Full Model contains 667 observations of all
developed property. The F-statistic, testing all coefﬁcients jointly equal to zero, is 12.61 ,
ﬁnding that not all coefﬁcients jointly are zero. The R-squared and Adjusted R-squared
terms are 0.53 and 0.49 respectively. The Ramsey RESET test statistic is 50.33,
indicating, within the 95% conﬁdence interval, that additional curvature exists than is
accounted for in this particular model. The Lagrange-Multiplier test for
heteroscedasticity is 80.27, indicating the presence of heteroscedasticity within the 95%

conﬁdence interval.

82

 

 

 

TABLE 15. LINEAR DEVELOPED FULL MODEL ESTIMATION RESULTS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Variable: Price

F -statistic: 12.6068 D-W: 1.9239

R-squared: .5316 RESET: 50.3314

Adj. R-Squared: .4894 LM Het. Test: 80.2653

N: 667

Variable (153332215; St;:f::d t-statistic P-value
C 156832.3000 13673790000 0.1147 [.909]
HOUSE -27.5295 15.6342 -1.7608 [079]”
HOUSE2 0.0411 0.0132 3.1126 [.002]*
HOUSE3 0.0000 0.0000 -3.4474 [.00ﬂ“
LOT 324.0123 175.8247 1.8428 [.066 **
LOT2 1.4706 3.5772 0.411 1 [.68u
LOT3 -0.0131 0.0166 -0.7866 [.432]
CITY -172.2364 713.9981 -0.2412 [.809]
CITY2 5.0140 7.2614 0.6905 [.490]
CITY3 -0.0195 0.0233 -0.8355 [.404]
HWY 658.4475 1415.8510 0.4651 [.642]
HWY2 65.6394 227.7525 0.2882 [.773]
HWY3 -2.9792 9.341 1 -0.3189 [.750]
GARAGE 6148.5790 1666.8590 3.6887 [.OOOt
UTIL 10643 .6700 2564.0620 4.151 1 [.000]*
PRISON -21236.2000 10863.7300 -1.9548 051]"
ZONED -2653.7280 4976.2680 -0.5333 [.594]
BR 85 79.03 10 4596.4240 1.8665 [062]* *
B 26410.7000 133 1 3 .7600 1.9837 [.048]*
D -14672.l800 1760.2210 -8.3354 [.000]*
MH -18326.2600 2472.3250 -7.4126 [.000]*
PAVE -292.0996 1893 .9290 -0.1542 [.877]
HOME 3304.8500 1622.2440 2.0372 [.0421‘
PLAT 2322.1690 2107.5450 1.1018 [.271]
H20 5344.0290 2590.8590 2.0626 [040]"
R_INT -288754.1000 341739.9000 -0.8450 [.398]
R_IN T2 31554.4600 38649.1000 0.8164 [.415]
R_INT3 -1 131.9350 1446.5540 -0.7825 [.434]
GSP 9889.6000 8604.0840 1.1494 4.251]
GSP2 -43.1703 37.2766 -1.1581 [.247]
GSP3 0.0640 0.0535 1.1970 [.23 2]
REGNLP -16I9.1720 5537.9530 -0.2924 [.770]

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

83

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 15 (continued).
Variable Estimated Standard t-statistic P-value
Coefﬁcient Error

PERPUB -511518.7000 426094.3000 -1.2005 [.230]
PERPUBZ 14933420000 12725810000 1.1735 [.241]
PERPUB3 -1 121776.0000 10291560000 -1.0900 [.276]
CC 5866.5340 40558.1700 0.1446 4.885]
CC_P -11501.2700 945412400 -0. 1217 [.903]
SEED -22490.2500 18630.5100 -1.2072 [.228]
THIN 100147.0000 45148.2600 2.2182 [.027]*
THIN_P -234923.5000 105082.4000 -2.2356 [.026K
SEL 15564.1300 35978.3600 0.4326 [.665]
SEL_P -2422.0040 93100.1900 -0.0260 [.979]
REM 17786.9700 101493.6000 0.1753 [.861]
REM_P —35845.4000 270232.2000 -0.1326 [.895]
F_DIS -0.0350 1.2901 -0.0272 [.978]
F_DIS2 0.0000 0.0001 -0.3 178 [.751]
F_DIS3 0.0000 0.0000 0.3372 [.736]
S_DIS -0.6980 1.1095 -0.6291 [.530]
S DIS2 0.0001 0.0001 1.2752 [.203]
S_DIS3 0.0000 0.0000 -1.5433 [.123]
PERDIS 0.4292 0.9999 0.4292 [.668]
PERADJ -8817.2340 32689.1000 -0.2697 [.787]
PERADJZ 63715.4100 100804.1000 0.6321 [.528]
PERADJ 3 -56436.7700 79066.9400 -0.7138 [.476]
CONIFER -854.6598 3625.8780 -0.2357 [.814]
HARDWOOD -1315.8210 3188.6870 -0.4127 [.680]

 

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

Due to near perfect collinearity with the GSP variable, the time trend variable, DATE,

was dropped from all models. The collinearity was severe enough to cause a Singular

X’X matrix and prevent estimation. The GSP variable has a strong time trend

component, as seen with the collinearity, and also has the economic trends from year to

year, which the DATE variable does not have. Therefore, the GSP variable is more likely

to account for more variation than the DATE variable and is left in the models.

84

Along with the highly signiﬁcant RESET and heteroscedasticity test statistics, many
coefﬁcient estimates are problematic. The HOUSE coefﬁcients, which are statistically
signiﬁcant, have a negative relationship with PRICE for values of 881 square feet and
below. The same coefﬁcients have a positive relationship for all values greater than 881.
The average house size for this data is 875 square feet. Therefore this model has a
negative relationship between price and house size for the average house. The ZONED
coefﬁcient is also negative, though not statistically signiﬁcant, implying that developed
zoned properties are worth less on average than non-zoned properties. This ﬁnding is
counter to most existing literature on zoning’s effect on property values. The REGNLP
is negative, implying that developed Lower Peninsula properties are worth less on
average than Upper Peninsula properties. These problems present a suspect model that

could be improved.

Though several problems exist with the linear model, a few variable coefﬁcients
demonstrate expected signs and statistical signiﬁcance. LOT, GARAGE, UTIL, BR, B,
D, MH, HOME, and H20 coefﬁcients have a priori expected signs. These variable
coefﬁcients all have positive Signs indicating a positive relationship with price. The three
variables are also statistically signiﬁcant. The PRISON coefﬁcient is negative and
statistically signiﬁcant. No a priori expectation was held for the Sign or statistical
signiﬁcance of this variable. Prisons are generally viewed as necessary but “not in my
backyard” institutions. The “not in my backyar ” attitude would lend credence to a
negative relationship with price. On the other hand, certain communities would probably

not exist without the local prison. This would have a positive impact with price. Which

85

 

impact is greater in absolute value has not been studied and is not known. The overall

relationship would likely change over time and geographical location.
Other variables have expected signs, but are not Statistically signiﬁcant. The coefﬁcients
for HWY, PLAT, GSP, and PERDIS all have expected positive signs while CITY,

F_DIS, S_DIS, and R_INT have expected negative signs. f

Still other coefﬁcients have unexpected signs and are not statistically signiﬁcant. The

 

PAVE, REGNLP, PERPUB, and PERADJ coefﬁcients are negative and insigniﬁcant. b

The coefﬁcient estimates for the public land and timber related variables provide little
evidence as to any impact on developed property values in this model. Because the
timber related data were only recorded for abutting properties, conclusions should not be
drawn from the Developed Full Models. These variables were included in the Full

Models to control for the possible effects they may have on adjacent property.

The Developed Full Models contain 667 observations, 134 of which are abutting
properties. Vegetation type (CONIFER, HARDWOOD), percent adjacent (PERADJ),
and harvesting variables (CC, SEED, THIN. SEL, REM) were included as additional
explanatory variables to control for the possible effects of these variables on abutting
properties. One harvesting term, SEED_P, was not included in any models. Due to the

extremely small variance in the variable. the coefﬁcient continuously demonstrated near

86

perfect collinearity with the intercept term. This caused singularity and prohibited

estimation. The variable was therefore drOpped from all models.

With respect to the variables of concern, no conclusions can be inferred at this point. The
model is plagued with heteroscedasticity and poor functional form. Until remedies to
these problems are attempted, no conclusions Should be drawn. The proximity (F_DIS,
S_DIS), percent public land (PERPUB), and interaction of proximity and percent public
land (PERDIS) variables are the only public land and timber related variables that
conclusions should be formed with this model. Other important variables related to
timber harvesting can not be correctly interpreted in this model because the variables do
not offer a complete comparison. The timber harvesting variables were only recorded for
adjacent properties. Their inclusion here serves to control for the possible effects on

adjacent properties.

The impact of the amount of surrounding public land, PERPUB, is negative and
insigniﬁcant in this model. The coefﬁcients of the proximity variables, F_DIS and
S_DIS, are both found to be negative and insigniﬁcant. The coefﬁcient of the interaction
between percent public land and proximity to public land (PERDIS) is positive and

statistically insigniﬁcant.
The Step B Developed Full Model, (Table A1 in Appendix A), are similar to the model

presented in Table 15, except the dependent variable is transformed to the logarithm of

PRICE. The logarithmic transformation brought the Ramsey RESET test statistic and the

87

n a .. 1 ii in‘i incl—“q

Lagrange-Multiplier heteroscedasticity test statistic down drastically. The RESET test
statistic dropped from 50.33 to 3.71 with the logarithmic transformation, with a P-value
of 0.06. The Lagrange-Multiplier heteroscedasticity test statistic fell from 80.27 to 0.89,
with a P-value of 0.346. The logarithmic transformation of the dependent variable
removed heteroscedasticity and reduced some missing curvature of the linear model from
Step A. Since the transformation greatly improved the model, the dependent variable for

all succeeding steps was the logarithm of price.

Since heteroscedasticity is no longer present, Step C was not needed in the Developed
Full Model process. Step D tested the Signiﬁcance of the squared and cubed terms of all
continuous independent variables (Table A2 in Appendix A). Two tests were conducted
in Step D. First, for each continuous independent variable, the squared and cubed terms
were tested together for signiﬁcance (independent tests in Table A2). Second, all squared
and cubed terms for all independent variables were tested jointly and are referred to as
joint tests. All independent tests can be rejected at the 95% conﬁdence level, meaning
that the coefﬁcients of all squared and cubed independent variables are not signiﬁcantly
different from zero independently. The only independent test that can not be rejected
within the 90% conﬁdence interval was the test of the squared and cubed terms of percent
of public land of the surrounding area (PERPUB). Both joint tests for all squared and
cubic terms together can be rejected within the 95% conﬁdence interval, as reﬂected by

the Chi-squared and F -test statistics.

88

The model with all squared and cubed terms removed, except for those of PERPUB, was
then estimated (Appendix A, Table A3). The removal of the insigniﬁcant quadratic and
cubic terms improved the RESET test statistic, Langrange-Multiplier heteroscedasticity
test statistic, and F -statistic, without affecting the adjusted R-squared term. Therefore,
removal of the quadratic and cubic terms further improved the model. This model is now

referred to as the “best” cubic model.

Step E began by estimation of an artiﬁcial regression. The artiﬁcial regression model
estimated the “best” cubic model from Step D (Table A3) with the addition of
logarithmic forms of all continuous independent variables (Table A4 in Appendix A).
Speciﬁc features of the artiﬁcial regression are not of interest here, so the discussion will
be limited to the tests related to functional form. The tests performed on the artiﬁcial
regression were similar to those in Step D. First, the coefﬁcients of the “best” cubic
model continuous independent variables were tested independently and then jointly to see
if they are equal to zero. Second, the coefﬁcients of the logarithmic form of the

independent variables are tested independently and jointly to see if they are equal to zero.

The independent tests for the LOT and HWY coefﬁcients are signiﬁcant within the 95%
conﬁdence interval, while none of the other independent tests are (Table A5). The joint
test of all the continuous variables from “best” cubic model must be rejected. The joint
test Chi-squared and F statistics carry P-values of 0.006. We can therefore conclude that
the Step D model explains the curvature between the independent variables and the

log(PRICE) signiﬁcantly better than a log-log model would.

89

Only the independent test of the log(LOT) is statistically signiﬁcant within the 95%
conﬁdence interval (Table 6 in Appendix A). The joint test that all coefﬁcients of
logarithmic form independent variables are equal zero can not be rejected. The P-values
of the Chi-squared and F statistics of the joint test are both 0.37. Therefore, the
logarithmic forms of the independent variables do not explain the functional form any
better than the Step D model. The functional form of the “best” cubic model is
considered the preferred and ﬁnal model for the Full Developed Data. The results of the

ﬁnal model are replicated in Table 16.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 16. FINAL DEVELOPED FULL MODEL ESTIMATION RESULTS.
Dependent Variable: Log(Price)

F -statistic: 17.9966 D-W: 2.1 181

R-squared: 0.5142 RESET: 2.6659

Adj. R-squared: 0.4857 LM Het. Test: 0.7864

N: 667

Variable 53:33:; SEE-1:? t-statistic P-value
C 8.7221 0.7374 1 1.8284 [.000]*
HOUSE 0.0003 0.0001 3.6168 [.000]*
LOT 0.0112 0.0014 8.2618 [.000]*
CITY 0.0012 0.0015 0.7876 [.431]
HWY 0.0353 0.0076 4.6319 [.000]*
GARAGE 0.3166 0.0555 5.7073 [.000]*
UTIL 0.3788 0.0839 4.5172 1.000?
PRISON -0.6406 0.2874 -2.2289 [026]”
ZONED 0.1589 0.1158 1.3722 [.170]
BR 0.1009 0.1533 0.6580 [.511]
B 0.3407 0.4371 0.7795 [.436]
D -0.5375 0.0581 -9.2552 [.000]*
MH -0.7670 0.0816 -9.4054 [.000]*
PAVE 0.0053 0.0612 0.0859 [.932]
HOME 0.1428 0.0532 2.6827 i.007]"‘

 

 

 

 

 

* signiﬁcant at 95% level, ** Signiﬁcant at 90% level

90

 

 

Table 16 (continued).

 

Estimated

Standard

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable Coe fﬁcien t Error t-statistic P-value
PLAT -0.0288 0.0618 -0.4666 [.641]
H20 0.2735 0.0852 3.2079 [.001]*
R_INT 0.0034 0.0366 0.0923 [926]
GSP 0.0069 0.0012 5.9469 [000K
REGNLP 0.0157 0.1383 0.1136 [.910]
PERPUB -12.2285 5.5863 -2.1890 [.029]*
PERPUBZ 32.0225 16.1273 1.9856 [048]“
PERPUB3 -21.4172 12.4046 -1.7266 [.085]M
CC 0.0004 1.3381 0.0003 [1.00]

' CC_P 0.3983 3.1202 0.1277 [.898]
SEED -0.2012 0.6207 -0.3240 [.746]
THIN 2.5997 1.4680 1.7709 [.077]* *
THIN_P -5 .7565 3.4126 -1.6868 [.092]* *
SEL 0.6090 1.1643 0.5231 [.601]
SEL_P -0.3006 3.0078 -0.0999 [.920]
REM -1.2100 3.3711 -0.3589 [.720]
REM_P 3.9361 8.9691 0.4388 [.661]
F_DIS -0.0000 0.0000 -1.2209 [.223]
S_DIS 0.0000 0.0000 ' 0.0151 [.988]
PERDIS 0.0000 0.0000 1.2134 [.225]
PERADJ 0.3510 0.2463 1.4247 [.1 55L
CONIF ER -0.1523 0.1173 -1 .2990 [.194]
HARDWOOD -0.0823 0.0964 -0.8544 [.393]

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

The Final Full Developed model has an F-statistic of 18.00, and an adjusted R-squared of
0.49. The Lagrange-Multiplier heteroscedasticity test statistic, 0.7864 (P-value=0.376),
ﬁnds no heteroscedasticity. The RESET test statistic, 2.6659(P-va1ue=0.103), ﬁnds that

within the 95% conﬁdence interval, all systematic curvature between the dependent and

independent variables is explained. These summary statistics demonstrate a much

improved model over the linear model in Table 15. The F, RESET, and LM

heteroscedasticity statistics have all greatly improved. The R2 values from the linear

9|

 

model and the ﬁnal model are not comparable since the dependent variable is not the

same form.

Many coefﬁcients exhibit the expected sign and are statistically signiﬁcant. The
coefﬁcients for HOUSE, LOT, HWY, GARAGE, UTIL, HOME, H20, GSP, and
PERPUB all have the expected positive signs and are statistically signiﬁcant. The
coefﬁcients for D and MH both have the expected negative signs and are statistically
signiﬁcant. The uncertain a priori PRISON coefﬁcient is negative and statistically

signiﬁcant.

Some statistically insigniﬁcant coefﬁcients had expected signs. The variables ZONED,
BR, B, PAVE, REGNLP, F_DIS, PERDIS, and PERADJ all have expected signs but are

insigniﬁcant. ZONED, BR, B, PAVE, REGNLP, PERDIS, and PERADJ coefﬁcients all

have expected positive signs, while F_DIS has the expected negative sign.

Other statistically insigniﬁcant coefﬁcients had unexpected signs. The coefﬁcients for
CITY, R_INT, and S_DIS are all positive while the expectation was negative. PLAT has

an unexpected negative sign.

The variables of concern in this model are PERPUB, F_DIS, S_DIS, PERDIS, and
PERADJ. Of these variables, PERPUB, S_DIS, F_DIS, and PERDIS provide mixed
results on developed private property values. The coefficients of the quadratic and cubic

terms of PERPUB are all signiﬁcant within the 90% conﬁdence interval, the linear and

92

squared coefﬁcients are signiﬁcant within the 95% conﬁdence interval. To measure the
impact of PERPUB on PRICE, take the partial derivative of the log(PRICE) function

6L0g(PRICE) .
, IS equal to:
6PERPUB

 

with respect to price. The partial derivative,

(-l2.2285 + 64.0405 * PERPUB - 64.2516 * PERPUBZ)

The average value for PERPUB is 0.4061 and for PERPUB2 is 0.1994. This equation has
a positive relationship with log(PRICE) at the average values. Using the mean values for
PERPUB, the impact, on average, on log(PRICE) is 0.9653. This represents the change
in log(PRICE) given a one unit change in PERPUB holding all other variables constant.
The coefﬁcients of PERPUB are decimal, representing a percentage. A one unit change
in PERPUB would be 1.00, or equivalent to 100%. Therefore, we must divide the
derivative, 0.9653, by 100 to represent the change in log(PRICE) for a one percent
change in PERPUB. The derivative divided by 100 is 0.009653. Kennedy (1981) states _
that the average percentage impact of a continuous independent variable on the dependent
variable in a semi-log model is the partial derivative. Therefore, the average percentage
increase in PRICE from a one-percent increase in PERPUB is 0.9653% or approximately
one percent. One percent of the average sales price of developed property, $31,044, is
$310. The mean value for PERPUB for the Full Developed Model data is 41%, and the

values ranged from 8% to 76%.

The coefﬁcients of the S_DIS, F_DIS, and PERDIS variables are not signiﬁcant within
the 90% conﬁdence interval. Therefore, proximity to either State or Federal public land
has no effect on developed property values, though increasing the percentage of public

land in the surrounding area has a small positive impact. The distance proximity

93

variables can be transformed in a number of ways, to offer conclusive evidence. The
aggregation of the separate S_DIS and F_DIS distance measures into one distance
measure, PUBDIS, creates a statistically insigniﬁcant and negative coefﬁcient.
Therefore, aggregation of the owner speciﬁc distance variables into one public land
distance variable does not change the result that distance has no impact. If the PERPUB
related variables are dropped from the model, the coefﬁcient for F_DIS becomes
statistically signiﬁcant within the 95% conﬁdence interval. Under these circumstances,
moving 1000 feet farther from federal public land will decrease a property value by
0.13%. Under these circumstances, however, the statistical signiﬁcance occurs at the cost
of a decreased amount of information in the model, and is therefore suspect. Another
possible transformation of the proximity variables would be into a dummy variable
indicating if a property is adjacent, ABUT. The ABUT coefﬁcient is not statistically
signiﬁcant (P value = 0.398) and has a negative coefﬁcient. Further, breaking the
PERPUB variable in to two separate variables, one for percent of surrounding federal
public land and one for percent of surrounding state public land, produces two positive

but statistically insigniﬁcant coefﬁcients.

Because of the incomplete information of the remaining timber harvesting variables, no
conclusions should be drawn from the Developed Models. Aggregation of all harvesting
related variables into one, HARVEST, does not negligibly affect the variables of concern.

Discussion of these variables will take place in the Adjacent Model sections.

94

Vacant Property Full Models

The Developed Vacant Full Model has 672 sales observations (Table 17). The model has
an F -statistic, testing all coefﬁcients jointly equal to zero, of 3.91. The R-squared and
adjusted R-squared terms are 0.22 and 0.17 respectively. The Ramsey RESET test
statistic is extremely high, at 93.67. The high RESET test statistic ﬁnds, within the 95%
conﬁdence interval, that unexplained systematic curvature exists in the model. The
Lagrange-Multiplier test for heteroscedasticity ﬁnds the existence of heteroscedasticity

within the 95% conﬁdence interval.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 17. LINEAR VACANT FULL MODEL ESTIMATION RESULTS.
Dependent Variable: Price
F-statistic: 3.9106 D-W: 1 .7098
R-squared: 0.2235 RESET: 93.6657
Adj. R-squared: 0.1663 LM Het. Test: 41.2833
N: 672

. Estimated Standard . .
Vanable Coefﬁcient Error t-statlsttc P-value
C -335076.9000 13208410000 -0.2537 [.800]
LOT 298.6792 57.0510 5.2353 [.000]*
LOT2 -0.4430 0.3675 -1.2053 [.229]
LOT3 0.0001 0.0005 0.2065 [.836]
CITY 399.4160 659.1241 0.6060 [.545]
CITY2 -0.4806 5.9501 -0.0808 [.936]
CITY3 -0.0047 0.0180 -0.2603 [.795]
HWY 1244.8560 1272.4690 0.9783 [.328]
HWY2 -246.3751 222.6943 -1 . 1063 [.269]
HWY3 9.8493 9.4909 1.0378 [.300]
UTIL 1148.7600 3475.1020 0.3306 [.741]
PRISON -1861 1.6700 10321.6800 -1.8032 [072]"
ZONED -73 .4926 4077.0020 -0.0180 [.986]
PAVE -4891 .0550 1836.5490 -2.6632 [.008]*
PLAT 1232.3470 2028.6630 0.6075 [.544]

 

"' signiﬁcant at 95% level, ** signiﬁcant at 90% level

95

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 17 (continued).
Variable 5:22.233; St;:f::d t-statistic P-value
H20 2125.8670 2278.5640 0.9330 [.351]
R_INT 154181.4000 3393 20.4000 0.4544 [.650]
R_IN T2 -18896.4500 383 74.8400 -0.4924 [.623]
R;INT3 756.8030 1436.2980 0.5269 [.598]
GSP -643.7474 7102.4800 -0.0906 [.928]
GSP2 2.4080 30.8015 0.0782 [.93 8]
GSP3 -0.0026 0.0442 -0.0588 [.953]
REGNLP 5589.3890 4582.8580 1.2196 [.223]
PERPUB -389399.9000 405928.6000 -0.9593 [.338]
‘ PERPUB2 10728900000 12149470000 0.8831 4.378]
PERPUB3 -766569.2000 981661.9000 -0.7809 [.435]
CC 12747.2800 1 1672.0800 1.0921 [.275]
CC_P -31520.0900 23086.5400 -1.3653 [.173]
SEED 1226.1800 10389.6500 0.1180 [.906]
THIN 28687.7500 17283 .6900 1.6598 [.097]
THIN_P -53526.4900 33082.1000 -1 .6180 [.106]
SEL 10529.9600 22090.2600 0.4767 [.634]
SEL_P 1064.7080 52126.1500 0.0204 [.984]
REM 3601.7230 19292.0600 0.1867 [.852]
REM_P -12402.8300 33831.8500 -0.3666 [.714]
F_DIS -1.5932 1.1768 -1.3538 [.176]
F_DISZ 0.0002 0.0001 1.2306 [.219]
F_DIS3 0.0000 0.0000 -1.2597 [.208]
S_DIS -0.3365 1.0998 -0.3060 [.760]
S_DISZ 0.0001 0.0001 0.8545 [.393]
S_DIS3 0.0000 0.0000 -1.2427 [.214]
PERDIS 1.7914 0.9114 1.9656 [.050]*
PERADJ -50206.6600 30341.7900 -1.6547 [.098 **
PERADJ 2 173260.2000 87323 .6500 1.9841 [.048]*
PERADJ 3 -133361.0000 66473 .2900 -2.0062 [.045]"‘
CONIFER -962.7979 2852.4290 -0.3375 [.736]
HARDWOOD 3738.0790 3034.2070 1.2320 [.218]

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

 

The Step A vacant model ﬁnds only six coefﬁcients statistically signiﬁcant. Of the six
signiﬁcant coefﬁcients, only LOT and PERDIS have the expected sign. The signiﬁcant

coefﬁcients for PAVE and PERADJ have unexpected negative impacts. PRISON, which

96

is signiﬁcant, has a negative impact on vacant property values in this model, while it had
a positive impact on developed property. The last signiﬁcant coefﬁcient, THIN, should

not be interpreted from this model. It has a positive sign.

Regarding the statistically insigniﬁcant coefﬁcients, ZONED, F_DIS, and S_DIS have
the expected negative signs. The coefﬁcients for LOT, HWY, UTIL, H20, PLAT, and

REGNLP carry the expected positive signs, though they too are insigniﬁcant.

Many variable coefﬁcients are statistically insigniﬁcant and carry the unexpected a priori
sign. Both CITY and R_INT coefﬁcients are positive in this model but were expected to
be negative. The coefﬁcients for GSP, PERPUB, and PERADJ are all negative while the
opposite was expected. This model, like the Step A Developed Property Full Model, has

many problems and can be improved.

The Vacant Full Model, Step B estimation results are presented in Table A7 in Appendix
A. Due to the transformation of the dependent variable to log(PRICE), the RESET test
statistic and the Lagrange-Multiplier test statistic dropped dramatically. The Lagrange-
Multiplier test statistic, 4.10, ﬁnds the existence of heteroscedasticity within the 95%
conﬁdence interval. The Ramsey RESET test ﬁnds that not all curvature is explained by
the model. Because of the large improvement in these test statistics, log(PRICE) is the

preferred dependent variable.

97

The Step C model, correcting for heteroscedasticity, is presented in Table A8. The White
variance-covariance consistent matrix is used to correct for heteroscedasticity (Gujarati

1995).

In Step D, the squared and cubic coefﬁcients of the continuous independent variables are
tested for statistical signiﬁcance. These tests are presented in Table A9. Independent
tests, testing whether particular independent variables’ squared and cubic coefﬁcients are
different from zero, are presented ﬁrst. Only the squared and cubic terms for LOT are
statistically different from zero within the 95% conﬁdence interval. The same
coefficients for the F_DIS are statistically signiﬁcant within the 90% conﬁdence interval.
All others are not statistically different from zero. The joint tests, testing if all squared
and cubic coefﬁcients together are different from zero ﬁnd that we must reject the
hypothesis that all squared and cubic terms are zero. A second joint test was then
conducted to see if all squared and cubic coefﬁcients, other than those for LOT and
F_DIS, are jointly different from zero. This hypothesis can not be rejected. The Chi-
squared test statistic for the test is 4.3211 with a P-value of 0.7421. The F-test statistic is
0.6173 with a P-value of 0.7438. Both tests are acceptable, and the coefﬁcients of the
squared and cubic terms for all continuous independent variables, other than LOT and

F_DIS, are all assumed equal to zero.

For the “best” cubic Full Vacant Model removal of the insigniﬁcant squared and cubed

terms increased the F-statistic and lowered the RESET test statistic (Table A10). The F-

statistic increased from 4.8575 to 6.7104 with the removal of the insigniﬁcant terms. The

98

RESET statistic dropped from 4.3054 (P-value = 0.038) to 2.5083 (P-value = 0.131) and
was no longer statistically signiﬁcant within the 95% conﬁdence interval. The Lagrange-
Multiplier heteroscedasticity test found the presence of heteroscedasticity. The model

presented in Table A10 was been corrected for heteroscedasticity.

The Step-E artiﬁcial regression model results are presented in Table A1 1. The tests of
the quadratic and cubed terms are presented in Table A12, while the tests of the
logarithmic terms can be found in Table A13. The tests of the quadratic and cubed terms
ﬁnd that LOT, HWY, and, PERPUB coefﬁcients are all independently signiﬁcant within
the 90% conﬁdenceinterval. The joint tests reject the hypothesis that all quadratic aqnd
cubed terms are jointly equal to zero. The tests of the logarithmic terms ﬁnd only
log(PERPUB) to be independently signiﬁcant within the 95% conﬁdence interval. The
joint hypothesis can not be rejected within the 90% conﬁdence interval, but can not be
accepted in the 95% conﬁdence interval. Because the joint test of the quadratic and
cubed terms can not be rejected within the 95% conﬁdence interval, and the same joint
test of the logarithmic terms can be over the same interval, the cubic model is considered

superior.

The Final Vacant Full Model results are presented in Table 18. This model is corrected
for heteroscedasticity using White’s variance-covariance matrix. The Final Model has an
F-statistic of 6.54 which is signiﬁcant within the 95% conﬁdence interval. The model

also has an adjusted R-squared of 0.21. The RESET test statistic, 2.28, is not signiﬁcant

99

within the 90% conﬁdence interval, therefore ﬁnding no unaccounted systematic

curvature.

 

TABLE 18. FINAL VACANT FULL MODEL ESTIMATION RESULTS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Variable: Log(Price)

F -statistic: 6.5357 D-W: 1.6848

R-scntared: 0.2466 RESET: 2.2818

Adj. R-squared: 0.2089 LM Het. Test: 7.7603

N: 672

Variable 32:33:; 5‘33"] t-statistic P-value
C 7.8812 0.8064 9.7728 [.000]*
LOT 0.0213 0.0026 8.1960 [.000] *
LOT2 -0.0001 0.0000 -4.493 1 [.000] *
LOT3 0.0000 0.0000 3.0792 [.002]*
CITY 0.0054 0.0016 3.3724 [.001]*
HWY -0.0268 0.0079 -3.3751 [.001]*
UTIL 0.1984 0.1789 1.1092 [.268]
PRISON 0.1073 0.1943 0.5523 [.581]
ZONED -0.1560 0.0980 -1.5920 [.112]
PAVE -0.1831 0.0905 -2.0233 [.043]*
PLAT 0.0063 0.1030 0.061 1 [.951]
H20 0.2589 0.0977 2.6494 [.008]*
R_INT -0.0240 0.0520 -0.4623 [.644]
GSP 0.0034 0.0016 2.0890 [.037]*
REGNLP 0.2633 0.1726 1.5256 [.128]
PERPUB 0.4345 0.4073 1.0670 [.286]
CC 0.6770 0.4690 1.4434 [.149]
CC_P -1.3838 0.9944 -1.3916 [.165]
SEED 0.3718 0.3449 1.0781 [.281]
THIN 1.4748 1.0355 1.4241 [.155]
THIN_P -2.6499 2.1445 -1.2357 [.217]
SEL 1.6468 0.6185 2.6623 [008]"
SEL_P -2.2343 1.1414 -1.9575 [.051]*
REM -0.0408 0.7089 -0.0576 [.954]
REM_P -0.2113 1.0811 -0.1954 [.845]

 

 

 

 

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

100

 

 

 

 

 

 

 

 

 

 

 

Table 18 (continued).

Variable 5:23:35 8:23:13.“ t-statistic P-value
F_DIS -0.0001 0.0001 2.5967 [.010]*
F_DISZ 0.0000 0.0000 1.9147 4056]“
F_DIS3 -0.0000 0.0000 -1.5300 [.127]
S_DIS 0.0000 0.0000 0.3920 [.695]
PERDIS 0.0000 0.0000 0.2772 4.782]
PERADJ 0.3872 0.2364 1.6375 [.102]
CONIF ER -0.1 729 0.1 196 -1.4460 [.149]
HARDWOOD -0.0232 0.0994 -0.2333 [.816]

 

 

 

 

 

 

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

The ﬁnal vacant full model has several statistically signiﬁcant coefﬁcients. Of these,
LOT, H20, and GSP have expected positive coefﬁcient signs. The signiﬁcant F_DIS
coefﬁcient has the expected negative sign. The signiﬁcant coefﬁcients for CITY, HWY,

and PAVE all have the opposite sign than expected.

Many coefﬁcients have the expected sign, though are not statistically signiﬁcant. The
coefﬁcients for UTIL, PLAT, REGNLP, PERPUB, PERDIS, and PERADJ all ﬁnd a
positive relationship with log(PRICE). The insigniﬁcant coefﬁcient of R_INT has the

expected negative sign.

The insigniﬁcant coefﬁcient for PRISON is positive in the vacant full model though it is
negative in the developed full model. The insigniﬁcant coefﬁcient for ZONED is
negative here, though it is positive in the Developed Full Model. Though having the
opposite sign compared to the developed model, the coefﬁcients for PRISON and
ZONED are plausible in the vacant model. Many areas are populated, such as Rudyard,

just south of Sault Ste. Marie, because of the presence of prisons. Development around

101

such areas may drive up vacant land values, while having the opposite effect on pre-
existing developed property. Vacant zoned property may incur a larger negative impact
from zoning regulations than developed property. This is because developed property
may be “grandfathered” and therefore receives, primarily, the beneﬁts associated with
zoning while not incurring the full costs. Vacant property, on the other hand, will not be
“grandfathered” and therefore incur both costs and beneﬁts. The costs of zoning, in this
case, are greater than the perceived beneﬁts, though not statistically. In either case,

neither PRISON or ZONED are statistically signiﬁcant in the Vacant Full Final Model.

Of the public land related variables of interest, the coefﬁcients for SEL, SEL_P, and
F_DIS are all statistically signiﬁcant within the 90% conﬁdence-interval. PERADJ is
signiﬁcant at a slightly lower interval of 89.8%. Because SEL, SEL_P, and PERADJ
contained observations for abutting properties only, the results should not be interpreted
from this model, rather the results of these variables should be interpreted from the

Adjacent Vacant Model.

The overall impact of F_DIS on log(PRICE) is negative. The partial derivative of the
log(PRICE) with respect to F_DIS is:

aging?) = —0.0001+(2.7540E-—08)* F_DIS—(1.0157E—12)* F_0182

 

At the mean values of F_DIS and F_DISZ, the derivative is equal to -0.0001. The
average percentage impact of moving 1 foot away from federal public land on vacant
property values is 000011, or -0.011%. This can be interpreted to mean that moving

1000 farther from to federal public land will decrease the average vacant property value

102

by 11%. The average vacant land price is $14,555 per parcel, and l 1% of this is $1,601.
The average vacant land price per acre is $720.90, and 11% of this is $79. The average
distance from federal public land for vacant property is 1,894 feet, with a range from 0 to
28,901 feet. The PERDIS interaction term is statistically insigniﬁcant. Dropping the
insigniﬁcant PERDIS from the model has no effect on the coefﬁcients of F_DIS, F_DISZ,

or F_DIS3.

To simplify the model, several aggregations and transformations of variables can be
completed. Several aggregations and transformations were examined in this model, after
the Final Vacant Full Model was estimated. One aggregation that may be of interest is
the combination of F_DIS and S_DIS, creating a variable, PUBDIS, that is not owner
speciﬁc. Another aggregation is the transformation of the distance proximity variables to
simple dummy variable, ABUT, indicating whether the property abuts public land.
Aggregation of all harvesting variables into one variable was also attempted. Lastly, the
harvesting related variables time frame was reduced. Initially, the models included
harvesting activities up to 10 years previous to sale. This reduction only models

harvesting activities up to 3 years previous to sale.

The aggregation and transformation of variables provided little differences from the ﬁnal
model. Transformation of the public land distance proximity variables did not change the
results signiﬁcantly. One transformation, dropping the F_DIS and S_DIS related
variables and adding the combined distance from public land variable PUBDIS, found a

positive and statistically insigniﬁcant coefficient. Another transformation of the distance

103

proximity variables to the ABUT dummy variable, found a negative statistically
insigniﬁcant coefﬁcient. Dividing the ABUT variable into two dummy variables, one
identifying properties that abut federal land and another that identiﬁes properties that abut
state land produced similar ﬁndings. The coefﬁcient for the federal abutting dummy
variable was positive but statistically insigniﬁcant. The coefﬁcient for the state abutting
dummy variable was negative and insigniﬁcant. Dropping the PERPUB related variables
from the model did not change the results. Breaking the PERPUB variable in to two
separate variables, one for percent of surrounding federal public land and one for percent
of surrounding state public land, produced two positive but statistically insigniﬁcant
coefﬁcients. Aggregation of all harvesting related variables into the HARVEST dummy

variable had no affect on the model results.

Adjacent Models

This section presents the results of the Adjacent Models. Similar to the Full Models
section, Steps A-E are reviewed for all Adjacent Models. .Only the Step A and Step F

models are presented in this chapter. The Step B through Step E models and statistical

tests are presented in Appendix B.
Developed Property Adjacent Models

The Step A linear-cubic model estimation results are presented Table 19. The dependent

variable is PRICE, and the model has 134 observations. The model F-statistic, 2.7985,

104

has a P-value of 0.00. Therefore, jointly, the coefﬁcients of all independent variables are
signiﬁcantly different from zero. The model’s R-squared and adjusted R-squared values
are 0.61 and 0.39. respectively. The Ramsey RESET test statistic, 27.3310, has a P-value
of 0.00, therefore ﬁnding curvature unaccounted for by the model within the 95%
conﬁdence interval. The Langrange-Multiplier test for heteroscedasticity ﬁnds the

presence of heteroscedasticity within the 95% conﬁdence interval.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 19. LINEAR DEVELOPED ADJACENT MODEL ESTIMATION

RESULTS.
Dependent Var.: Price
F -statistic: 2.7985 D-W: 1.9991
R-squared: 0.6125 RESET: 27.3310
Adj. R-squared: 0.3936 LM Het. Test: 32.6862
N: 134

. Estimated Standard . .
Vanable Coefﬁcient Error t-stattstlc P-value
C -389694.5000 33815840000 -0.1 152 [.909]
HOUSE 35.7070 57.9597 0.6161 [.539]
HOUSE2 -0.0446 0.0776 -0.5756 [.566]
HOUSE3 0.0000 0.0000 0.8242 [.412]
LOT -1 .6024 404.8094 -0.0040 [.997]
LOT2 10.5014 8.081 1 1.2995 L197]
LOT3 -0.0567 0.0376 -1.5072 [.135]
CITY -203.5976 2692.3540 -0.0756 [.940]
CITY2 9.3845 38.9383 0.2410 [.810]
C ITY3 -0.0394 0.1435 -0.2746 [784]
HWY 2504.0740 3692.2370 0.6782 [.499]
HWY2 -349.0210 527.2342 -0.6620 [.510]
HWY3 14.5612 20.2775 0.7181 [.475]
GARAGE 9889.2890 4778.6040 2.0695 [.042]*
UTIL 6420.1920 8395 .6920 0.7647 [.447]
PRISON -48698.2500 32178.4400 -1.5134 [.134]
ZONED -3212.3720 15982.3200 -0.2010 [.841]
BR -6516.5810 23814.2600 -0.2736 [.785]
DD -799l .9150 5115.4670 -1.5623 [.122]
MH -l2933.6300 6322.8130 -2.0456 [.044]*
PAVE 11844.7500 7011.9180 1.6892 [095]"

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

105

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 19 (continued).

Variable 5:211:32: St;:f::d t-statistic P-value
HOME 4136.9970 5676.6260 -0.7288 [.468]
PLAT 8318.9990 9818.4820 0.8473 [.399]
H20 7572.1890 7748.1370 0.9773 [.33 1]
R_INT 379745.8000 866303.6000 0.4384 [.662]
R_INT2 -44745.4300 97918.4100 -0.45 70 [.649]
R_INT3 1750.2340 3659.6550 0.4783 [.634]
GSP -10198.4200 23267.5400 -0.4383 [.662]
GSP2 53.3132 100.9942 0.5279 ].599]
GSP3 -0.0888 0.1453 -0.6108 [.543]
REGNLP -8572. l 350 15042.3800 -0.5699 [570]
PERPUB -947728.1000 10770770000 -0.8799 [.381]
PERPUBZ 28405230000 32039450000 0.8866 [.378]
PERPUB3 -2203608.0000 25779790000 -0.8548 [.395]
CC -59387.0000 576900500 -l.0294 [.306]
CC_P 134449.5000 132025.5000 1.0184 [.311]
SEED -46323.1500 26141.9500 -1 .7720 4080]"
THIN 99772.3 800 59198.9600 1.6854 [096]"
THIN_P -223020.7000 138005.3000 -1.6160 [.1 10]
SEL 43987.4900 46103.5600 0.9541 [.343]
SEL_P -7851 1.5000 1 19012.8000 -0.6597 [.51 1]
REM 33243.7300 1208670000 0.2750 [.784]
REM_P -62885 .2400 321361.6000 -0.1957 [.845]
FED 93.4613 1 2743.83 00 0.0073 [.994]
PERADJ 17445.2700 543 95.8500 0.3207 [.749]
PERADJ 2 -19063.3200 146215.1000 -0. 1304 [.897]
PERADJ 3 60.3124 107008.7000 0.0006 [1.00]
CONIF ER -624.3693 5225.9930 -0.1 195 [.905]
HARDWOOD 1023 .0600 5663.3200 0.1806 [.857]

 

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

Only three variable’s coefﬁcients were statistically signiﬁcant within the 90% conﬁdence
interval. GARAGE and PAVE both had positive signiﬁcant coefﬁcients, while MH had
a negative signiﬁcant coefﬁcient. None of the public land or timber harvesting related

variables were signiﬁcant within the 95% conﬁdence interval.

106

The Step B log-cubic model is estimated next. The estimation results are presented in
Table Bl. The transformation of the dependent variable from linear to logarithmic
greatly reduces the RESET and Lagrange-Multiplier test statistics. The RESET statistic,
0.0007, has a P-value of 0.98, therefore ﬁnding statistically no unaccounted curvature.
The Lagrange-Multiplier heteroscedasticity test statistic (P-value = 0.21) ﬁnds no
heteroscedasticity within the 90% conﬁdence interval. The F -statistic increased slightly
with the dependent variable transformation. Due to these improvements in the model, the

logarithmic form of dependent variable is hereafter used.

The Lagrange-Mulitplier test, discussed above, found no heteroscedasticity, therefore
Step C is not needed. The Step D tests of all continuous independent variables’ squared
and cubed coefﬁcients are presented in Table B2. The tests found no of the squared and
cubed coefﬁcients signiﬁcantly different from zero within the 90% conﬁdence interval.

Neither joint test can be rejected. All squared and cubed terms are dropped subsequently.

The “best” cubic model was then estimated with linear continuous independent variables.
The results are presented in Table B3. Dropping the insigniﬁcant quadratic and cubic
terms increased the F-statistic as well as the adjusted R-squared. The F -statistic increased
from 2.1136 (P-value = 0.010) to 3.1662 (P-value = 0.000). The adjusted R-squared
increased from 0.2867 to 0.3426. The RESET and Langrange-Mulitplier test statistics

found no unaccounted curvature or heteroscedasticity within the 90% conﬁdence interval.

107

The artiﬁcial regression, combining both logarithmic and “best” cubic independent
variables, is estimated next. The results are presented in Table B4. The “best” cubic
model continuous model independent variables are tested ﬁrst for independent and joint
signiﬁcance. The test results are presented in Table B5. The independent tests for LOT
and HWY are the only independent tests signiﬁcant within the 95% conﬁdence interval.
The joint tests rejects the hypothesis that all “best” independent variables are equal to

ZCl‘O.

The tests of the logarithmic terms in the artiﬁcial regression are presented in Table B6.
None of the independent tests are signiﬁcant within the 90% conﬁdence interval. Neither
joint test can be rejected. Considering the joint signiﬁcance of the “best” terms, and the
joint insigniﬁcance of logarithmic terms, the “best” cubic model is considered superior.

This is the Final Adjacent Model, and is presented in Table 20.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 20. FINAL DEVELOPED ADJACENT MODEL ESTIMATION
RESULTS.

Dependent Log(Price)

Variable:

F-statistic: 3.1662 D-W: 2.2039

R-squared: 0.5008 RESET: 0.1023

Adj. R-squared: 0.3426 LM Het. Test: 1.4453

N: 134

Variable (1333213223; 8:33;?“ t-statistic P-value

C 7.5194 1.5995 4.7010 [.000]*

HOUSE 0.0005 0.0003 2.0491 [.043]"‘

LOT 0.0120 0.0028 4.2860 [000]“

CITY 0.0002 0.0036 0.0552 [.956]

HWY 0.0214 0.0141 1.5202 [.132]

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

108

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 20 (continued).

Variable gig-12:; St;:f::d t-statistic P-value
GARAGE 0.3434 0.1613 2.1287 [.036]*
UTIL 0.0698 0.2586 0.2700 [.788]
PRISON -0.8160 0.5897 -1.3838 [.169]
ZONED 0.0271 0.3272 0.0828 [.934]
BR 0.6397 0.7856 0.8142 [.417]
D 02997 0.1756 -1.7060 [091]“
MH -0.45 84 0.2193 -2.0906 [.039]*
PAVE 0.4227 0.2356 1.7939 [.076]* *
HOME 0.0817 0.1675 0.4877 [.627]
PLAT 0.1109 0.3100 0.3578 [.721]
H20 0.3672 0.2568 1.4299 [.156]
R_INT -0.0033 0.1036 -0.0315 [.975]
GSP 0.0051 0.0037 1.3731 [.173]
REGNLP -0.3270 0.4301 -0.7603 [.449]
PERPUB ' 1.8242 1.2883 1.4160 [.160]
CC -1 .6914 1.9325 -0.8752 [.384]
CC_P 4.1620 4.4518 0.9349 [.352]
SEED -0.0927 0.7593 -0.1221 L903L
THIN 3 .0224 1.8653 1.6204 [.108]
THIN_P -6.3259 4.3757 -1.4457 H51]
SEL 1.8088 1.5672 1.154] L251]
SEL_P -3.2753 4.0656 -0.8056 4.422]
REM 0.6291 4.2261 0.1489 [.882]
REM_P -0.8067 1 1.2561 -0.0717 [.943]
FED 0.0658 0.3161 0.2082 [.836]
PERADJ 0.1890 0.3397 0.5564 [.579]
CONIFER -0. 1364 0.1747 . -0.7805 [.43 7]
HARDWOOD -0. 1275 0.1907 -‘0.6690 [505]

 

 

"‘ signiﬁcant at 95% level, ** signiﬁcant at 90% level

The ﬁnal developed adjacent model has a slightly higher F -statistic, 3.1662, compared to
the Step A model, 2.7985. The ﬁnal model also has improved Ramsey RESET and LM
heteroscedasticity test statistics compared to the Step A model. All the test statistics

provide evidence of an improved model relative to the initial Step A model.

109

Contributing to the improvements, several variable coefﬁcients are statistically
signiﬁcant and have the expected sign. The model ﬁnds the coefﬁcients of HOUSE,
LOT, GARAGE, and PAVE statistically signiﬁcant within the 90% conﬁdence interval.
These coefﬁcients also have the expected positive sign. The coefﬁcients of D and MH

are signiﬁcant and have the expected negative sign.

Many of the insigniﬁcant coefﬁcients have the expected sign. The insigniﬁcant
coefﬁcients for HWY, ZONED, BR, HOME, PLAT, H20, GSP and PERPUB are all
positive as expected. R_INT has an insigniﬁcant coefﬁcient with its expected negative

sign.

Three statistically insigniﬁcant coefﬁcients have signs opposite of those expected. The
coefﬁcient for CITY is positive, and the coefficients for PRISON and REGNLP are

negative.

The PERPUB variable is positive, but insigniﬁcant. (Note: the PERPUB variable is
positive and signiﬁcant in the Full Developed Model). The insigniﬁcant ﬁnding in the
Adjacent Developed Model seems plausible, since the all properties this models are
actually abutting public land while a large majority of those in the Full Developed Model

were not.

Only the timber harvesting variable coefﬁcient for THIN was near statistical signiﬁcance.

It is signiﬁcant within the 89.20% conﬁdence interval. The coefﬁcient is positive. If this

110

conﬁdence interval is considered acceptable, the impact a thinning has on an adjacent
developed property value appears to be positive. Aggregation of all timber harvesting
variables into one variable, HARVEST, produced no differences. The coefﬁcient for
HARVEST was positive and statistically insigniﬁcant within the 80% conﬁdence
interval. Reduction in the time since harvest from 0-10 years to 0-3 years had no impact
on the model. Models with time frames less than 3 years could not be estimated due to
the small occurrences of harvesting and therefore small variation in the harvesting

variables.

Vacant Property Adjacent Models

The Step A linear-cubic model estimation results of vacant adjacent properties are
presented in Table 21. The model has 179 observations. The F-statistic has a value of
2.9344 and a P-value of 0.00. The R-squared and adjusted R-squared are 0.4596 and
0.3030, respectively. The RESET test statistic, 63.7829 (P-value = 0.000), ﬁnds
unaccounted curvature within the 95% conﬁdence interval. The Langrange-Multiplier

test statistic, 40.7625 (P-value = 0.00) ﬁnds the existence of heteroscedasticity.

Ill

 

TABLE 21. LINEAR VACANT ADJACENT MODEL ESTIMATION

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

RESULTS.

Dependent Price

Variable:

F-statistic: 2.9344 D-W: 1 .7961

R-squared: 0.4596 RESET: 63.7829

Adj. R-squared: 0.3030 LM Het. Test: 40.7625
N: 179

Variable (13:33:33; St;:f::d t-statistic P-value

C 1 1234640000 23499640000 0.4781 [.633]
LOT 627.6558 105.2831 5.9616 [.000]*
LOT2 -2.6605 0.6535 -4.07 l 4 [.000]*
LOT3 0.0027 0.0008 3.4309 [.001]*
CITY ~990.1887 1360.5970 -0.7278 [.468]
CITY2 12.8426 16.7023 0.7689 [.443]
CITY3 -0.0436 0.0570 -0.7647 [.446]
HWY 3724.2030 2254.8630 1.6516 [.101]
HWY2 -588.3237 393.1330 -1.4965 [.137]
HWY3 22.9664 16.5041 1.3916 [.166]
UTIL -12390.0000 15839.7400 -0.7822 [.435]
PRISON 34273780 333 1 3 .3200 0.1029 [.918]
ZONED 20578.3700 8738.6800 2.3549 [.020]*
PAVE 9615.8910 4568.7000 2.1047 [.03 7]*
PLAT 4869.8170 5644.5920 0.8627 [.390]
H20 10586.5600 5195.2130 2.0378 [.043]*
R_INT -397610.0000 6205 66. 1000 -0.6407 [.523]
R_INT2 44648.9700 70504.9100 0.6333 [.528]
R_INT3 -1674.4410 265 0.5740 -0.6317 [.529]
GSP 328.5331 1 1865.8400 0.0277 [.978]
GSP2 -2.7662 51.4267 -0.0538 [.957]
GSP3 0.0061 0.0738 0.0824 [.934]
REGNLP 14524.9900 99820500 1.4551 [.148]
PERPUB 2973 68.3 000 309661 5.0000 0.0960 [.924]
PERPUB2 -597918.2000 651 18330000 -0.0918 [.927]
PERPUB3 43 1463 .2000 41997030000 0.1027 [.918]

 

signiﬁcant at 95% level, ** signiﬁcant at 90% level

112

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 21 (continued).

Variable giggle; Stg:f::d t-statistic P-value
CC -2288.9750 12684.9400 -0.1804 [.857]
CC_P -304.2508 24592.5800 -0.0124 [.990]
SEED -3289.5290 9584.3740 -0.3432 [.732]
THIN 916.1868 16905.2200. 0.0542 [.953
THIN_P 1296.7870 32759.5600 0.0396 [.968]
SEL -l 1815.4600 22148.8900 -0.5335 [.595]
SEL_P 50548.7100 52019.4700 0.9717 [.333]
REM ~16395.9900 18598.2000 -0.8816 [.380]
REM_P 24469.6300 32771.9200 0.7467 [.457]
FED -13623.5800 11771.5900 -l.1573 [.249]
PERADJ -10680.4300 34791.3600 -0.3070 [.759]
PERADJ 2 63 547. l 000 93 82 1 .9900 0.6773 [.499]
PERADJ 3 -58513.5600 68520.2000 -0.8540 Q95]
CONIFER -2664.6650 2985.7490 -0.8925 i374]
HARDWOOD 1 109.8590 3886.4980 0.2856 [.776]

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

The coefﬁcients for LOT, ZONED, PAVE, and H20 are the only coefﬁcients statistically

signiﬁcant within the 95% conﬁdence interval. Of these, LOT, PAVE, and H20 have the

expected positive impact on price. The ZONED coefﬁcient is also positive.

Many statistically insigniﬁcant variables exhibit expected signs. The insigniﬁcant

coefﬁcient for the R_IN T has the expected negative sign. The insigniﬁcant coefﬁcients

for PLAT, GSP, PERPUB, and PERADJ all have the expected positive signs.

While many statistically insigniﬁcant coefﬁcients have the expected signs, many did not.

The insigniﬁcant coefﬁcients for CITY and UTIL are negative, when a priori

expectations had them as positive. The insigniﬁcant positive coefﬁcients for HWY and

 

PRISON were expected to be negative. None of the public land variables of concern are

statistically signiﬁcant in this model.

The log-cubic Adjacent Vacant Model is estimated next, in Step B. The estimation
results are presented in Table B7. The logarithmic transformation increased the F-
statistic. The F -statistic in the Step A model is 2.9344, and is 4.2675 in the Step B
model. The RESET test statistic fell from 63.7829 (P-value = 0.00) to 1.3951 (P-value =
0.240) with the logarithmic transformation. The Lagrange-Multiplier test statistic fell
from 40.7625 (P-value = 0.00) to 0.5338 (P-value = 0.465). The logarithmic
transformation removed heteroscedasticity and accounted for all curvature, both within

the 95% conﬁdence interval. The logarithmic model is therefore used subsequently.

Since no heteroscedasticity is present, Step C is not needed. The tests of squared and
cubed coefﬁcients for all continuous independent variables are conducted next. The
tests’ results are presented in Table B8. The results of the independent tests ﬁnd only the
coefﬁcients for the squared and cubed LOT terms signiﬁcant within the 90% conﬁdence
interval. Both of the joint tests can be rejected. Another joint test is conducted to see if
all the squared and cubed coefﬁcients except for those related to the LOT term are jointly
equal to zero. The joint test results ﬁnd a chi-squared test statistic of 5.7950 (P-value =
0.4465) and an F-statistic of 0.9658 (P-value = 0.4470). This test clearly can not be
rejected. The “best” cubic model will retain the 1.0T2 and LOT3 terms, but will drop all

other squared and cubed terms.

114

Table B9 presents the estimation results of the "best” adjacent vacant cubic model. The
removal of the insigniﬁcant quadratic and cubic terms increased the model F —statistic
from 4.2675 to 5.7530. The adjusted R-squared increased by a very small amount, from
0.4234 to 0.4278. The RESET test statistic increase slightly but is still insigniﬁcant
within the 90% conﬁdence interval. The presence of heteroscedasticity is still rejected
within the 90% conﬁdence interval. Only the coefﬁcients for LOT, LOT2, LOT3, PAVE,

H20, and FED are statistically signiﬁcant within the 95% conﬁdence interval.

The artiﬁcial regression is estimated next. The results are presented in Table B10. The
results of the tests of the “best” model independent variables are presented in Table B1 1.
None of independent tests of the “best” model independent variables are rejected.
Neither joint test is rejected. The results of the logarithmic independent variables are
presented in Table B12. The independent test for the logarithm of LOT is not rejected,
while all others are. Neither joint test for all logarithmic independent variables are
rejected. The independent tests found that the log(LOT) is not statistically equal to zero,

and that the cubic form is.

None of joint tests are rejected, therefore neither the “best” model forms nor the
logarithmic forms are superior in explaining curvature. Therefore two ﬁnal model
options should be considered. First, the Final Adjacent Vacant Model contains the
log(LOT) variable and the other independent variables are those from the “best” model.
Second, the Final Adjacent Vacant Model contains all logs of continuous independent

variables.

115

The F -statistic‘s, adjusted R-square’s, RESET statistic’s, and Lagrange-Multiplier
heteroscedasticity test statistic’s between the two models are nearly identical. None of
the coefﬁcient signs are different between the models. All coefﬁcients that are
statistically signiﬁcant within the 90% conﬁdence interval in one model are statistically
signiﬁcant within the 90% conﬁdence interval in the other. Due to these similarities,
neither model provides evidence of superiority over the other. Therefore, for simplicity,
the ﬁrst model option will be considered the Final Adjacent Vacant Model. The results

of this model are presented in Table 22.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE 22. FINAL VACANT ADJACENT MODEL ESTIMATION

RESULTS.
Dependent Log(Price)
Variable:
F -statistic: 5.7726 D-W: 1.7379
R-squared: 0.4968 RESET: 1.9499
Adj. R-sglared: 0.4108 LM Het. Test: 0.2148
N: 179

. Estimated Standard . .
Vanable Coefﬁcient Error ' t-statlstlc P-value
C 6.6579 1.4669 4.5387 [.000]*
Log(LOT) 0.4002 0.0591 6.7669 [.000]*
CITY 0.0013 0.0034 0.3818 [.703]
HWY -0.0074 0.0131 -0.5628 [.5 74]
UTIL -0.7959 0.6817 -1.1675 [.245]
PRISON 0.2024 0.5528 0.3661 [.715]
ZONED 0.7057 0.2059 3.4266 [.001]*
PAVE 0.0799 0.1902 0.4202 [.675]
PLAT -0.0541 0.2471 -0.2191 [.827]
H20 0.3334 0.2074 1.6074 [.110]
R_INT -0.0809 0.0825 -0.9803 [.328]
GSP 0.0021 0.0026 0.7901 [.431]
REGNLP 0.4736 0.3817 1.2409 [.217]

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

116

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 22. (continued).
Variable Estimated Standard t-statistic P-value
Coefﬁcient Error
PERPUB 2.2767 1.5178 1.5000 [.136]
CC 0305] 0.5430 -0.5619 [.575]
CC_P 0.5571 1.0601 0.5255 [.600]
SEED 0.2708 0.4226 0.6408 [523]
THIN 0.0529 0.7242 0.0731 [.942]
THIN_P 0.1490 1.4107 0.1057 [.916]
SEL 0.1916 0.8832 0.2169 [.829]
SEL_P 0.4871 1.9881 0.2450 [.807]
REM -0.4673 0.811 1 -0.5761 [.565]
REMJP 0.5480 1.4230 0.3851 [.701]
FED -0.5592 0.1920 -2.9118 [.004]*
PERADJ 0.2912 0.2594 1.1227 [.263]
CONIFER -0.0445 0.1255 -0.3544 [.724]
HARDWOOD 0.2683 0.1489 1.8018 [071]"

 

 

* signiﬁcant at 95% level, ** signiﬁcant at 90% level

The model robustness statistics of the ﬁnal adjacent vacant model are all improved over
the Step A linear model. The ﬁnal model has an F-statistic of 5.77, higher than the linear
model’s 2.9344. The ﬁnal model has a RESET test statistic of 1.9499, which ﬁnds no
unaccounted systematic curvature while the linear exhibits unaccounted curvature. The

ﬁnal model also ﬁnds no heteroscedasticity that plagued the linear model.

Four coefﬁcients are statistically signiﬁcant within the 95% conﬁdence interval in the
ﬁnal model. The LOT, ZONED and HARDWOOD coefﬁcients are positive and
signiﬁcant. The coefﬁcient for FED is negative and signiﬁcant. The sign of the LOT
coefﬁcient is as expected. The positive impact of ZONED is not as expected but is also
not totally a surprise. Zoning has consistently demonstrated positive impacts on

developed property values while having inconsistent impacts on vacant property values.

117

Analogous to the Final Vacant Full Model, the Final Adjacent Vacant Model ﬁnds that
the only public land related variable coefﬁcient that is statistically signiﬁcant within the
95% conﬁdence interval is related to federal land. The Final Full Vacant Model found a
negative relationship between price and distance from federal public land (F_DIS). The
Final Vacant Adjacent Model ﬁnds a negative relationship between adjacent vacant
properties that abut federal land (FED), relative to state land, and the sales price. The
base case of the Final Adjacent Vacant Model is vacant property that abuts state public
land. Everything is in comparison to the base case. The negative F_DIS impact from the
Full Model seems at odds with the negative FED impact in the Adjacent Model. These
ﬁndings, though, may be stating the same result. The federal land abutting vacant
properties sold, on average, for less than similar vacant properties abutting state land.
The cheaper sales price may provide a large contrast between the sales prices of abutting
and non-abutting vacant properties near federal land. This contrast may be much larger

' for properties near federal land than for those near state land. If so, then the contrast
would likely have a negative impact on the distance from federal land, as found in F_DIS

in the Final Full Adjacent Model.

From the Full Vacant Model, the impact of state land on private property values is not
statistically signiﬁcant. Abutting state land is a special case of the Final Full Vacant
Model, where the distance from state land is equal to zero. We can then assume that

abutting state property has no statistical impact on vacant land prices. Assuming this, the

118

coefﬁcient of FED ﬁnds a negative relationship between vacant land prices of those

abutting federal land relative to those not abutting federal land.

To interpret the dummy variable FED coefﬁcient, as explained by Kennedy (1981), the
average percentage impact on the sales price is:

= Exp(c-.5*v(c))-1, where c is the coefﬁcient and v(c) is the variance of c

= (05576 — 1 = 0.4388
Therefore, the impact of vacant land abutting federal public land is a decrease in value of
44%. The average value of vacant adjacent land is $16,218, 44% of this is $7,136. The

average value per acre is $570, and 44% of this is $251.

Applying the same procedure to the HARDWOOD coefﬁcient, we ﬁnd that the average
percentage increase in adjacent vacant property values due to the presence of adjacent
hardwood forests is 29%. This is equivalent to $4,703 per vacant property and $165 per

acre.

Aggregation of the timber harvesting related variables into the dummy variable,
HARVEST, had an impact on the coefﬁcient of FED. The coefﬁcient is still negative
and signiﬁcant within the 95% conﬁdence interval. The aggregation slightly changed the
level of the coefﬁcient, and the standard deviation. After the aggregation, the impact of
federal land on adjacent properties is —62%. The impact prior to aggregation, -44%,

should be considered more accurate due to the model containing more information.

119

Reduction of the time since harvest from 0-10 years prior to sale to 0-3 years had no
effects on the model. Time frames of less than 3 years could not be estimated due to the

small occurrences of the harvesting variables and therefore small variances.

Removal of Outliers

Tests for outliers were conducted on all four ﬁnal models. The Studentized Residual
(SR) approach was used to identify outliers. The SR approach estimates a SR for each
observation by dividing each residual by its standard error. The ratio has a chi-squared
distribution. Rather than examining and removing extreme dependent variable

observations, the SR approach identiﬁes observations that do not ﬁt the ﬁnal model well.

Removal of outliers in the Full Models had very little impact on the results. The SR
approach identiﬁed 33 outliers in the Final Full Developed Model (Table A14). Of the
33 outliers, 24 had sales prices less than or equal to $8,000 and 2 were greater than
$100,000. Removal of the outliers in this model had little effect. The impact of the
PERPUB variable fell from 0.0097 to 0.0088. With rounding, the impact of increasing
the amount of surrounding public land by 1% was still a 1% increase in developed
property values. The impact of the F_DIS variable became statistically signiﬁcant within
the 90% conﬁdence interval with the removal of outliers, though not within the 95%

’ conﬁdence interval. The impact amounts to an approximate —1.6%, or $500, decrease in
developed sales price for every 1000 feet away from federal public land if the level of

signiﬁcance is deemed acceptable.

120

The SR approach identiﬁed 37 outliers in the Final Vacant Full Model (Table A15). Of
these, 15 were sales equal to or less than $3,500 and 2 were over $100,000. Removal of
the outliers had little effect. The only change was a decrease from -11% to -l 1.5% in the

sales price of vacant land for every 1000 feet farther from federal public land.

Removal of outliers in the Adjacent Models had very large impacts on the results. This is
due in large part to the extremely small sample size of these models to begin with. The
results of the Adjacent Models, relative to the full model, were questionable before any
outliers in the Adjacent Models were dropped. The SR approach found 8 outliers in the
Adjacent DeveIOped Model (Table B 1 3). Removal of outliers in this model required the
removal of the BR, UTIL, and SEED_T variables also. These variables had few
observations, removal of outliers developed a singular matrix with all the above
mentioned variables collinear with the constant term. The results of the model with the
outliers removed found a 700% increase in adjacent developed property sales prices if a

thinning occurs.

The SR approach found 12 outliers in the Adjacent Vacant Model (Table B14). Removal
of outliers also required the dropping of the UTIL variable in the Vacant Model due to
near perfect collinearity with the constant term. Model results with outliers removed
found a —3 7% impact on vacant properties abutting federal land relative to state land.
The results also found a 396% increase in vacant property sales prices for a 1% increase

in the percent of surrounding public land.

Removal of outliers in both of the Adjacent Models drastically changed the results of the
models. Further the results of the models before and especially after are not believable
and are inconsistent with the Full Models. The unrealistic behavior of the Adjacent
Models was probably due to the extremely small sample sizes. Reliable results from the

Adjacent Models would require much larger sample sizes.

122

CHAPTER 5: CONCLUSIONS

The history of United States public land policy, like most other policy, is marked by
controversy. Continuously, public land policy controversy has involved three issues: (1)
property rights, (2) public land ownership, and (3) the values associated with public land.
The property rights issue includes nearby private ownership. How our public lands are
managed may affect the property rights of nearby private property owners. The public
property ownership issue, revived by the sagebrush rebellion, has proponents that believe
that state or private ownership provides more net beneﬁts than federal ownership. The
values issue centers on what speciﬁc values are provided on public land. These values
may impact nearby residents. All three of the historic and controversial issues are tied to

private property near public land.

This research investigates some relationships between public ownership and nearby

' private property values. The speciﬁc relationships are related to the three common
issues. The speciﬁc relationships are contained in the objectives of this research. The
objectives are: (1) to estimate the impact of Michigan’s public lands on sales prices of
adjacent private lands, (2) to estimate the impact of proximity (distance) to public lands
on private land sales, (3) to compare impacts of various types of public land use
(ownership) on nearby private land values, and (4) to estimate the impact of public land

timber harvesting on adjacent private property.

123

This ﬁnal chapter concludes this research. Chapter ﬁve contains four sections. The ﬁrst
section sumarizes the steps taken in this research, from the literature review through the
methods. The second section reviews the results. The third section ties the results to
various policy implications. The ﬁnal section provides recommendations for future

research.
Research Summary

The ﬁrst step of this research involved a literature review. Primarily, the literature
provided the theoretical foundation for hedonic price (or regression) analysis.
Additionally, the empirical literature presented commonly used data, modeling reference

points, and solutions to existing problems for hedonic regression analysis.

The theory for hedonic price analysis provides the framework so that we can estimate the
implicit prices embedded in market goods. Theoretically, market sales prices can be
dissected into the various characteristics and attributes that make up the market good. In
the end, the contributing proportion of the characteristic or attribute of the market sales
price is estimated. Generally the characteristics and attributes do not have explicit
markets in which they are traded. The theory applies to highly differentiable goods that
are traded in a competitive market. Land property describes such a good. The supply
and demand of land is fairly competitive. The supply and demand for housing is also

fairly competitive. By assuming that indeed the markets for land and housing are

124

competitive then the theory provides a framework to break apart the sales price and

measure the implicit prices.

The empirical literature provided various examples of commonly used data. Data such as
acreage, square footage, distance from central business district, and distance from
interstate highway have been commonly used in past studies. The use of these variables,
in the existing literature, provided a list of suggested data. The use of these same
variables provided a reference point for their statistical signiﬁcance and directional
impact (i.e., the sign of the coefﬁcients). Further, the hedonic models in the empirical
literature provide reference points for the robustness (i.e., adjusted R2 and F-statistic) of
many property value models. The models presented in this study have similar levels of
robustness. Lastly, the empirical literature demonstrated that vacant and developed
property values behave differently and should be modeled separately. The vacant and
developed hedonic models in this study demonstrate different behavior. The models in

this study reinforce past practice of separating vacant models from developed models.

The last important provision of the empirical literature was found in the many
approaches to common estimation problems. Various approaches have been applied to
estimating the correct functional form. In many cases, the functional form problem was
ignored or assumed away. At the other extreme, the functional form problem is
’ addressed with purely data-ﬁtting techniques. Another problem associated with different
ﬁlnctional forms is the interpretation of the results. The empirical literature provided

interpretation of model coefﬁcients under different functional forms.

125

us-n. mane A...”

The second step of this research was to determine the sampling frame and gather the
required data needed to estimate hedonic models. The sampling frame was constructed
with a few factors in mind, namely: data sources and limitations, data collection
efﬁciency, size of sample areas, and controlling for inﬂuential factors. The sampling
frame contained over three hundred townships. Nine randomly sampled townships were
chosen and the data were collected from those townships. The data were collected from
two primary sources for each township: the County Equalization Department and the

township Assessors Ofﬁce.

The third step involved the estimation of the hedonic models. Four models were
required. Two models that contained all the adjacent and nonadjacent parcels were
estimated, one for developed property and the other for vacant property. These two
models are referred to as Full Models and addressed the ﬁrst three objectives.
Additionally, two models were estimated for adjacent parcels only, one for developed
property and the other for vacant property. These two models are referred to as Adjacent
Models and they were needed to address the third and fourth objectives. From the
empirical literature, a functional form determination and testing scheme (Steps A-F) was
developed and each of the four models was put through it. After each model was put
through the functional form scheme they became Final Models. Four Final Models were
therefore estimated: Final Developed Full Model, Final Vacant Full Model, Final

Developed Adjacent Model, and Final Vacant Adjacent Model.

126

Results

Four different models were required to address the objectives of this research. Two Full
Models estimated the impact of proximity from public land on private property values
and the impact of ownership of nearby public land on private property values. Two
Adjacent Models estimated the impacts of public land timber harvesting on adjacent
private property values and the ownership of adjacent public land on private property
values. The two Full Models contained large sample sizes while the two Adjacent
Models had much smaller sample sizes. The results of the Full Models were reasonable.
The results of the Adjacent Models were questionable and at times conﬂicted with the
Full Models. The reason for this is more than likely due to the small sample sizes of the
Adjacent Models. More reliable results, with respect to the impacts of public land timber

harvesting, would require larger sample sizes.

The two Full Models found, in general, that public land did impact both nearby
developed and vacant property values, but in different ways. Four conclusions can be
drawn from these models. First, the models found that Michigan’s public lands had no
impact on private adjacent developed property values but did affect vacant private land
values. Vacant property values decreased in value, on average, by 11% for every 1000
feet farther away from federal public land, holding everything constant. Therefore,
federal public land adjacent vacant property values are worth more than non-adjacent
property values. Eleven percent, on average, amounts to $79 per acre. Proximity and

adjacency from state public land had no impact on property values.

127

Second, proximity to public land had no impact on developed private property sales but
did affect vacant property values. As stated in the ﬁrst result conclusion, vacant property
values decreased in value the farther they were from federal public land. They decreased
in value by l 1% for every 1000 feet farther away. Proximity from state public land had

no impact on property values.

Third, the impact of federal public land on developed property values did not differ from
that of state ownership. The impact of federal ownership on vacant property values did
differ from the impact of state ownership. As stated in the ﬁrst and second result
conclusions, proximity from federal public land had an impact on vacant property values,

while proximity from state public land did not.

Fourth, the percentage of public land in the surrounding nine-township area did positively
' affect private developed property values. A one percent increase in the amount of
surrounding public land caused a 0.9653% increase in the average developed property
value. This is, on average, equivalent to $310. While this was true for develop property
values it was not for vacant property values. This last result was not public landowner

speciﬁc.

Due to the small sample sizes of the Adjacent Models, and the resulting conﬂicting

results, no conclusions can be drawn about the impact of public land timber harvesting on

private property values. Neither Adjacent Model found a statistically signiﬁcant

128

relationship between timber harvesting and the sales price of adjacent properties. The
Vacant Adjacent Model did ﬁnd a statistically signiﬁcant difference between the sales
prices of vacant parcels adjacent to federal and state land. The difference amounted to a
44%, on average, premium for state land adjacency. This was equivalent to $251 per
acre. This conclusion is in direct conﬂict with the Vacant Full Model. It is likely that

this result was due to small sample problems.

One potential ﬂaw in the models presented in this study is the lack of information
regarding nearby private forestland. Private forestland may have similar impacts on
nearby private property values as public land. Without this data, the models may have
behaved in ways that differ from how they may have behaved with the data. This could
be one reason why the adjacent models did not ﬁnd any impact from timber harvesting. ‘
Another reason could be related to the distance of the timber harvesting from adjacent
properties. This study recorded timber harvesting activities up to one-quarter mile away
from adjacent properties. This distance may have been too large, causing too large of a
variation in key variables to be found signiﬁcant. Another possible reason for not ﬁnding
any impact from timber harvesting could be the time that elapsed between harvest and
sale of the property. The data collection methods used in this study did not allow the
models with times between harvest and property sale of less than three years to be
estimated. The data set simply did not have enough observations, and therefore variation
for statistical estimation. If this study used different data collection methods targeting

small differences between harvest and property sale the results may have been different.

.129

In summary, the results in Michigan ﬁnd: the percentage of public property in the
surrounding nine townships positively impact developed property values and proximity to
federal public land positively impacts vacant property values. The results found no
statistically signiﬁcant impacts on private property values from public land timber

harvesting.

Policy Implications

Historically, federal ownership and management of land had been repeatedly challenged.
Interest groups have demanded the sale of federal public land and/or the disposition to
state ownership. There is no reason to believe that future challenges to federal ownership
of land will not be challenged. Further, there is no reason to believe that state ownership
will not challenged. The results from this study may have considerable implications if
the current structure of public land changes in Michigan. Changes in the public land
structure could change a nearby property owners wealth, primarily for large land holders.
Further, changes in the public land structure would change a regions property tax base.
Changes in the property tax base could therefore change property tax revenues and
property tax rates. The exact impact is speciﬁc to a geographic location, and is
dependent upon may factors including the amount of surrounding public land, the amount

of private vacant and developed property.

One policy-related implication to the results is the possible change in local property tax

bases as a result of changing the amount of public land or the ownership of public land.

130

The results ﬁnd that an increased percentage of public land increases the average
developed property value. This can also be interpreted to mean that a decrease in the
amount of public land would decrease nearby developed property values. A decrease in
the amount of public land would increase the amount of private property, and therefore
the amount of taxable land, and simultaneously lowering developed property values. If
the property tax rate did not change, this would also be accompanied by lower developed
property taxes (assuming updated tax records). The net effect here is indeterrninable.

The property tax base, property tax rate, amount of public land, and the number and value
of developed parcels in the nine surrounding townships would be needed to determine the
net effect. Either way, the increase or decrease in the amount of public land will affect

nearby developed property values.

The results demonstrate that vacant property values increase the nearer they are to federal
public lands. This would imply that increases in federal public land would cause
associated increases in nearby vacant land values. Increases in the amount of federal land
would decrease private land and simultaneously decrease the local property tax base.
Like the case above, the net effect can not be determined here,.but the change in the
amount of federal public land may have in impact on local vacant property owners.
Further, a decrease in the amount of federal public land would decrease vacant property

values.

A related result found that federal public land has a negative impact on adjacent property

relative to state public land. Here, replacing federal land with state land would increase

131

adjacent vacant property values. Federal public land has the opposite impact on non-
adjacent vacant land. Therefore, replacing federal land with state land would decrease
non-adjacent vacant property values. The net effect depends upon the amount of public
land and the amounts of adjacent and non-adjacent vacant properties. Similar to the
above conclusions, a change in the amount of federal land relative to state land will

impact adjacent property owners.

Either changing the amount of public land, or changing public ownership of land will
likely impact nearby local property values. The impact on property values is also likely
to affect property taxes. The impact on property values is an impact on private wealth,

especially to owners of large plots or more expensive property.

Recommendations for Future Research

One recommendation for future research would be to conduct a similar study that would
include information about surrounding private forested land. This study may have been
improved by controlling the possible effects that nearby private forested land may have
had on private property values. Generally, detailed inventory and timber harvesting
information for private forested land is difﬁcult to obtain. Inventory and timber
harvesting information for large areas of private forested land near both federal and state
public land would be ideal for a future study. With this information the effects of
surrounding private forested land could be incorporated and investigated in conjunction

with the effects of public land.

132

A second recommendation for future research would be to include information about the
existing trees and forests on the private property observations. Drombrow et al.
demonstrated that “mature” trees on private property increased property values on
average. It is possible that the trees and forests on private property could not only
increase the value but also interact with the impact of public land. Using information
about the existing trees and forests on the private parcels would allow the controlling of
the possible interactions. Though this information would be useful, consistent and

reliable data may be very difﬁcult to obtain.

A third recommendation for ﬁrture research would be to conduct a similar study at
intervals over time. Smith et al. ﬁnd that regression results applied to property values
provide reliable but individual coefﬁcients produced in the models may not produce
accurate measurements over time. Therefore, ﬁlture studies should be conducted to

observe changes in the coefﬁcients of concern.

Another recommendation for future research would be to determine property owner’s
perceptions surrounding nearby public lands. This study measured actual impacts on
property values assuming perceptions. Surveying land-owners to identify they’re
knowledge and perceptions about the public land and timber harvesting activities would
compliment this study. Modeling landowners segmented into different groups based on

the level of their perceptions could be a further extension.

The last recommendation for future research would be to investigate the impact of
different public land types, other than national and state forests, on private property

values. The results of this study have no implications for other public land types.

134

REFERENCES

Abelson. P. W. 1979. “Property prices and the value of amenities.” Journal of
Environmental Economics and Management 6:11-28. ‘

Anderson, G. D., and R. C. Bishop. 1986. The Valuation Problem. Chapter 3 in Natural
Resource Economics: Policy Problems and Contemporary Analysis. Edited by
Daniel W. Bromley. 234 p. Kluwer Nijhoff Publishing, Boston.

Arguea, N. M., and C. Hsiao 1993. “Econometric issues of estimating hedonic price
functions: with an application to the US. market for automobiles.” Journal of
Econometrics 56(1993): 243-267.

Baden, John, and Dean Lueck 1984. "A Property Rights Approach to Wilderness
Management." In Public Lands and the US. Economy: Balancing Conservation and
Development. G. M. Johnson and P. M. Emerson, eds. Boulder, CO: Westview.

Bare, B. B., and R. L. Smith. 1999. “Estimating stumpage values from transaction
evidence using multiple regression.” Journal of Forestry (July): 32-39.

Beaton, W. P. 1991 “The impact of regional land-use controls on property values: the
case of the New Jersey Pinelands.” Land Economics 67(May): 172-94.

Beaton, W. P., and M. Pollock. 1992. “Economic impact of grth management
policies surrounding the Chesapeake Bay.” Land Economics 68(Nov.): 434-53.

Brown, G. M., Jr., and H. O. Pollakowski. 1977. “Economic valuation of shoreline. ”
Review of Economics and Statistics 59(3):272-78.

Chow, G. C. 1967. “Technological change and the demand for computers. ” American
Economic Review 57(Dec.): 1117-1130.

Clark, D. E., and L. A. Nieves. 1994. “An interregional hedonic analysis of noxious
facility impacts on local wages and property values.” Journal of Environmental
Economics and Management 27: 23 5-253.

1 Correl, M. R., J. H. Lillydahl, and L. D. Singell. 1978. “The effects of greenbelts on
residential property values: some ﬁndings on the political economy of open space”.
Land Economics 54(May): 207-17.

Dana, ST. and S. K. Fairfax. 1980. Forest and Range Policy: Its Development in the
United States. McGraw-Hill. New York, NY.

135

Daniel, T. C., L. Wheeler, R. S. Boster, and P. R. Best Jr. 1973. “Quantitative evaluation
of landscapes: an application of signal detection analysis to forest management
alternatives”. Man-Environment systems 3(5): 330-44.

Diamond, D. B., Jr. 1980. “The relationship between amenities and urban land prices.”
Land Economics 56(Feb.): 21-32.

Dombrow, J ., M. Rodriquez, and C. F. Sinnans. 2000. “The market value of mature trees
in single-family housing markets.” The Appraisal Journal 68(1): 39-43.

Frech 111, H. E., and R. N. Lafferty. 1984. “The effect of the California Coastal
Commission on housing prices.” Journal of Urban Economics 16:105-123.

Freeman, 1986. Comment 1. Pages 139-149 in Natural Resource Economics: Policy
Problems and Contemporary Analysis. Edited by Daniel W. Bromley. 234 p.
Kluwer Nijhoff Publishing, Boston.

Galster, G, and. Y. Williams. 1994. “Dwellings for the severely mentally disabled and
neighborhood property values: the details matter.” Land Economics 70(Nov.): 466-
77.

Gujarati, D. 1995. Basic Econometrics. McGraw-Hill, Inc. New York.

Hamilton, S. W., and G. M. Schwann. 1995. “Do high voltage electric transmission line
affect property value?” Land Economics 70(Nov.): 426-44.

Johnston, G. M. and P. M. Emerson. 1984. Public Lands and the U. S. Economy:
Balancing Conservation and Development. Westview Press. Boulder, CO.

Leefers, LA. and D.M. Jones. 1996. Phase 11 ﬁnal report: assessing
changes in private property values along designated rivers in Michigan.
Report submitted to Forest Management Division, Michigan Department of
Natural Resources. Unnumbered mimeograph. 24 p.

Katz, L., and K. T. Rosen. 1987. “The interjurisdictional effects of growth controls on
housing prices.” Journal of Law and Economics 30(Apr.): 149-60.

Kennedy, Peter E. 1981. “Estimation with correctly interpreted dummy variables in
semilogarithmic equations.” Amer Econ Rev 71(Sept): 801-802.

Kennedy, P. 1996. A Guide to Econometrics. MIT Press, Cambridge, MA.

136

 

Kiel, K. 1995. “Measuring the impact of the discovery and cleaning of identiﬁed
hazardous waste sites on house values.” Land Economics 71(Nov.): 428-35.

Kiel, K. A., and K. T. McClain. 1995. “House prices during sitting decision stages: the
case of an incinerator from rumor through Operation.” Journal of Environmental
Economics and Management 28: 241-255.

Knapp, G. J. 1985. “The price effects of urban growth boundaries in metropolitan
Portland, Oregon.” Land Economics 61(Feb.): 26-35.

MacCleery, D. 1984. “The Public Interest in the Federal Lands and the Reagan
Administration’s Asset Management Program.” In Public Lands and the US.
Economy: Balancing Conservation and Development. G. M. Johnson and P. M.
Emerson, eds. Boulder, CO: Westview.

Milon, J. W., Gressel, J., and D. Mulkey. 1984. “Hedonic amenity valuation and
functional form speciﬁcation.” Land Economics 60(Nov.): 378-87.

Rosen, S. 1974. “Hedonic prices and implicit markets: product differentiation in pure
competition.” Journal of Political Economy 82(Jan.): 34-55.

Runge, C. F. 1984. “An Economists Critque of Privatization.” In Public Lands and the
US. Economy: Balancing Conservation and Development. G. M. Johnson and P. M.
Emerson, eds. Boulder, CO: Westview.

Shonkwiler, J. S., and J. E. Reynolds. 1986. “A note on the use of hedonic price models
in the analysis of land prices at the urban fringe.” Land Economics 62(Feb.): 58-63.

Short, C. B. 1989. Ronald Reagan and the Public Lands: America’s Conservation
Debate, 1979-1984. Texas A & M University Press. College Station, TX.

Smith, B. A. 1978. “Measuring the value of urban amenities.” Journal of Urban
Economics 5(July): 370-387.

Smith, C. A., M. E. Hanna, and S. C. Caples. 1999. “Residential appraising and the
stability of regression results over time.” The Appraisal Journal 67(4): 375-381.

137

APPENDIX A

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Full Developed Model
TABLE A1. LOG-LINEAR DEVELOPED FULL MODEL ESTIMATION
RESULTS.

Dependent Log(Price)
Variable:

' F-statistic: 12.4904 D-W: 2.1503
R-squared: 0.5293 RESET: 3.7064
Adj. R-squared: 0.4869 LM Het. Test: 0.8864
N: 667

. Estimated Standard . .
Varlable Coefﬁcient Error t-statlstlc P-value

C 10.5990 46.1274 0.2298 [.818]
HOUSE 0.0003 0.0005 0.5613 [.575]
HOUSE2 0.0000 0.0000 0.5088 [.61 1]
HOUSE3 0.0000 0.0000 —1.0202 [.308]
LOT 0.0122 0.0059 2.0496 [.041]
LOT2 0.0000 0.0001 0.3719 [.710]
LOT3 0.0000 0.0000 -0.6879 [.492]
CITY 0.0192 0.0241 0.7955 [.427]
CITY2 -0.0001 0.0002 -0.5545 [.579]
CITY3 0.0000 0.0000 0.4215 [.674]
HWY 0.0037 0.0478 0.0785 [.937]
HWY2 0.0055 0.0077 0.7153 [.475]
HWY3 -0.0002 0.0003 -0.6722 [.502]
GARAGE 0.3128 0.0562 5.5620 [.000]
UTIL 0.3713 0.0865 4.2929 [.000]
PRISON -0.8208 0.3665 -2.2397 [.025]
ZONED 0.0277 0.1679 0.1649 [.869]
BR 0.0612 0.1551 0.3950 [.693]
B 0.3774 0.4491 0.8402 [.401]
D 05163 0.0594 -8.6949 [.000]
MH -0.7608 0.0834 -9.1227 [.000]
PAVE -0.0022 0.0639 -0.0337 [.973]
HOME 0.1289 0.0547 2.3553 [.019]
PLAT 0.0086 0.0711 0.1215 [.903]

 

 

 

 

 

138

 

 

Table A1 (continued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable 553.1232: St;:g::d t-statistic P-value
H20 0.2813 0.0874 3.2190 [.001]
R_INT -10.0376 11.5283 -0.8707 [.384]
R_INT2 1.1203 1.3038 0.8593 [.391]
R_INT3 -0.041 1 0.0488 -0.8419 [.409
GSP 0.3781 0.2903 1.3025 [.193]
GSP2 -0.0016 0.0013 -1.2989 ]. 194]
GSP3 0.0000 0.0000 1.3217 [.187]
REGNLP 0.0927 0.1868 0.4961 [.620]
PERPUB -22.5354 14.3740 -1 .5678 [.117]
PERPUB2 63 .4994 42.9295 1.4792 [.140]
PERPUB3 -47.1712 34.7177 -1.3587 [.175]
CC -0.3765 1.3682 -0.2752 [.783]
CC_P 1 .4060 3.1893 0.4408 [.659]
SEED -0.1680 0.6285 -0.2672 [.789]
THIN 2.9350 1.5230 1.9270 [.054]
THIN_P -6.5963 3 .5449 -1.8608 [.063]
SEL 0.6545 1.2137 0.5393 [.590]
SEL_P -0.4702 3.1407 -0.1497 [.881]
REM -1.0484 3.4238 -0.3062 [.760]
REM_P 3.6441 9.1 161 0.3997 [.689]
F_DIS 0.0000 0.0000 -0. 1601 [.873]
F_DISZ 0.0000 0.0000 -0.0826 [.934]
F_DIS3 0.0000 0.0000 -0.0266 [.979]
S_DIS 0.0000 0.0000 -0.3617 [.718]
S_DISZ 0.0000 0.0000 0.8792 [.380]
S_DIS3 0.0000 0.0000 -1.1 146 [.265]
PERDIS 0.0000 0.0000 0.9715 [.332]
PERADJ 0.0870 1.1027 0.0789 [.937]
PERADJ2 1.4831 3 .4005 0.4362 [.663]
PERADJ3 -1.5502 2.6673 -0.5812 [.561]
CONIFER -0.1694 0.1223 -1 .3 846 [.167]
HARDWOOD -0.1000 0.1076 -0.9294 [.353]

 

139

 

 

TABLE A2. TESTS OF SQUARED AND CUBED COEFFICIENTS OF
FULL DEVELOPED MODEL.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Independent Tests

. Standard . .
Parameter Estimate t-statistic P-value

Error

HOUSE 0.0000 0.0000 0.5087 [.61 1]
LOT 0.0000 0.0001 0.3704 [.711]
CITY -0.0001 0.0002 -0.5549 [.579]
HWY 0.0053 0.0074 0.7164 [.474]
R_INT 1.0792 1.2550 0.8599 [.390]
GSP -0.0016 0.0013 -1.2989 [.194]
PERPUB 16.3281 8.4285 1.9373 [.053]
F_DIS 0.0000 0.0000 -0.0826 [.934]
S_DIS 0.0000 0.0000 0.8792 [.379]
PERADJ -0.0671 0.9808 -0.0684 [.945]
Joint Tests
Chi-squared(10): 8.8867 P-value: 0.5429
F(10, 1272): 0.8887 P-value: 0.5432

 

 

 

 

 

140

 

 

TABLE A3. “BEST” DEVELOPED FULL MODEL ESTIMATION
RESULTS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Log(Price)
Variable:
F-statistic: 17.9966 D-W: 2.1 181
R—sguared: 0.5142 RESET: 2.6659
Adj. R-sguared: 0.4857 LM Het. Test: 0.7864
N: 667

Variable 3:22:33; St£::::d t-statistic P-value
C 8.7221 0.7374 1 1.8284 [.000]
HOUSE 0.0003 0.0001 3.6168 [.000]
LOT 0.0112 0.0014 8.2618 [.000]
CITY 0.0012 0.0015 0.7876 [.431]
HWY 0.0353 0.0076 4.6319 [.000]
GARAGE ' 0.3166 0.0555 5.7073 [.000]
UTIL 0.3788 0.0839 4.5172 [.000]
PRISON -0.6406 0.2874 -2.2289 [.026]
ZONED 0.1589 0.1158 1.3722 [. 1 7g
BR 0.1009 0.1533 0.6580 [.511]
B 0.3407 0.4371 0.7795 [.436]
D 05375 0.0581 -9.2552 [.000]
MH -O.7670 0.0816 -9.4054 [.000]
PAVE 0.0053 0.0612 0.0859 [.932]
HOME 0.1428 0.0532 2.6827 [.007]
PLAT -0.0288 0.0618 -0.4666 [.641]
H20 0.2735 0.0852 3.2079 [.001]
R_INT 0.0034 0.0366 , 0.0923 [.926]
GSP 0.0069 0.0012 5.9469 [.000]
REGNLP 0.0157 0.1383 0.1136 [.910]
PERPUB -12.2285 5.5863 -2.1890 [.029]
PERPUB2 32.0225 16.1273 1.9856 [.048]
PERPUB3 -21.4172 12.4046 -1 .7266 [.085]

 

I41

 

 

Table A3 (continued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable 5:23:33; $2233” t-statistic P—value
CC 0.0004 1.3381 0.0003 [1.00]
CC_P 0.3983 3.1202 0.1277 [.898]
SEED -0.2012 0.6207 -0.3240 [.746]
THIN 2.5997 1.4680 1.7709 [.077]
THIN_P -5.7565 3.4126 -1.6868 Q92]
SEL 0.6090 1 . 1643 0.5231 [.601]
SEL_P -0.3006 3.0078 -0.0999 [.920]
REM -1.2100 3.3711 -0.3589 [.720]
REM_P 3.9361 8.9691 0.4388 [.661]
F_DIS 0.0000 0.0000 -1.2209 [.223]
S;DIS 0.0000 0.0000 0.0151 [.988]
PERDIS 0.0000 0.0000 1.2134 [.225]
PERADJ 0.3510 0.2463 1.4247 [.155]
CONIF ER -0.1523 0.1 173 -1.2990 H94]
HARDWOOD -0.0823 0.0964 -0.8544 [.393]

 

 

 

 

 

142

 

 

TABLE A4. DEVELOPED FULL MODEL ARTIFICIAL REGRESSION

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ESTIMATION RESULTS.
Dependent Log(Price)
Variable:
F-statistic: 12.4904 D-W: 2.1503
R-squared: 0.5293 RESET: 3.7064
Adj. R-squared: 0.4869 LM Het. Test: 0.8864
N: 667
. Estimated Standard . .
Variable Coefﬁcient Error t-statistic P-value
‘ C 17.6343 19.7937 0.8909 [.373]
HOUSE 0.0002 0.0002 0.9694 [.333]
LOG(HOUSE) 0.1321 0.1313 1.0066 [.315]
LOT 0.0080 0.0022 3.6842 [.000]
LOG(LOT] 0.0934 0.0470 1.9866 [.047L
CITY 0.0049 0.0039 1.2531 [.21 1]
LOG(CITY) 0.0981 0.3901 0.2515 [.801]
HWY 0.0450 0.0162 2.7767 [.006]
LOG(HWY) -0.0844 0.0960 -0.8790 [.380]
GARAGE 0.3154 0.0561 5.6264 [.000]
UTIL 0.3733 0.0849 4.3977 [.000]
PRISON -0.0904 0.5812 -0.1555 [.876]
ZONED 0.1946 0.1631 1.1930 [.233L
BR 0.1074 0.1538 0.6984 [.485]
B 0.3199 0.4469 0.7158 L474]
D -0.5120 0.0591 -8.6696 [.000]
MH -0.7523 0.0834 -9.0223 [.000]
PAVE 0.0083 0.0631 0.1308 [.896]
HOME 0.1 134 0.0547 2.0744 [.038L
PLAT 0.0657 0.0773 0.8507 [.395]
H20 0.2766 0.0864 3.2002 [.001]
R_INT 0.3276 0.5903 0.5551 [.579]
LOG(R_INT) -2.9445 5.0714 -0.5806 [.562]
GSP 0.0117 0.0208 0.5615 [.575]
LOG(GSP) -1.0725 4.7914 -0.2238 [.823]
REGNLP 0.2437 0.1891 1.2886 [.198]

 

I43

 

 

Table A4 (continued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable 53:32:; St;:f::d t-statistic P-value
PERPUB 2349.2590 1956.4290 1.2008 [.230]
PERPUBZ -942.0409 825.2016 -1 . 1416 [.254]
PERPUB3 262.0888 252.1448 1.0394 [.299]
LOG(PERPUB) -2404.8780 1985.6690 -l .21 1 1 [.226]
CC -0.0456 1.3658 -0.0334 [.973]
CC_P 0.5837 3.1807 0.1835 [.854]
SEED -0.1597 0.6258 -0.2551 [.799]
THIN 2.3040 1.5387 1.4973 [.135]
THIN_P -5.1462 3.5810 -1.4371 [.151]
SEL 0.2925 1.1847 0.2469 [.805]
SEL_P 0.3235 3.0626 0.1056 [.916]
REM -1.7962 3.3990 -0.5284 [.597]
REM_P 5.5495 9.0349 0.6142 [.539]
F_DIS 0.0000 0.0000 0.2152 [.830]
LOG(If_DIS) 0.0195 0.0233 0.8355 [.404]
S_DIS 0.0000 0.0000 0.2885 [.773]
LOG(S_DIS) 0.0309 0.0210 1.4703 [.142]
PERDIS 0.0000 0.0000 -0.0187 [.985]
PERADJ -1.8776 2.2033 -0.8522 [.394]
LOG(PERADJ) 3.0642 2.9072 1.0540 [.292]
CONIF ER -0.1345 0.1265 -1.0634 [.288]
HARDWOOD -0.0165 0.1325 -0. 1242 [.901]

 

 

I44

' ‘1’ p‘ﬁ’r-‘L' -.‘1

 

 

 

TABLE A5. DEVELOPED FULL MODEL ARTIFICIAL REGRESSION

TESTS — QUADRATIC TERMS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Independent Tests

. Standard . .
Parameter Estimate t-statlstic P-value

Error

HOUSE 0.0002 0.0002 0.9694 [.332]
LOT 0.0080 0.0022 3.6842 [.000]
CITY 0.0049 0.0039 1.2531 L210]
HWY 0.0450 0.0162 2.7767 [.005]
R-INT 0.3276 0.5903 0.5551 [.579]
GSP 0.0117 0.0208 0.5615 [.574]
PERPUB 1669.3070 1382.5890 1.2074 [.227]
F_DIS 0.0000 0.0000 0.2152 [.830]
S_DIS 0.0000 0.0000 0.2885 [.773]
PERADJ -1.8776 2.2033 -0.8522 [.394]
Joint Tests
Chi-squared(1 0): 24.5697 P-value: 0.0062
F(10, 1272): 2.4569 P-value: 0.0066

 

 

 

 

 

 

TABLE A6. DEVELOPED FULL MODEL ARTIFICIAL REGRESSION
TESTS - LOGARITHMIC TERMS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Independent Tests
. Standard . .
Parameter Estimate t-statlstlc P-value
Error
HOUSE 0.1321 0.1313 1.0066 [.314]
LOT 0.0934 0.0470 1.9866 [.047]
CITY 0.0981 0.3901 0.2515 [.801]
HWY -0.0844 0.0960 -0.8790 [.379]
R_INT -2.9445 5.0714 -0.5806 [.561]
GSP -1.0725 4.7914 -0.2238 [.823]
PERPUB -2404.8780 1985.6690 -1.21 1 1 [.226]
F_DIS 0.0195 0.0233 0.8355 [.403]
S_DIS 0.0309 0.0210 1.4703 [.141]
PERADJ 3 .0642 2.9072 1 .0540 [.292]
Joint Tests
Chi-squared( 1 0): 10.8802 P-value: 0.3669
F(10, 1280): 1.0880 P-value: 0.3679

 

 

 

 

 

I45

 

 

Full Vacant Model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE A7. LOG-QUADRATIC VACANT FULL MODEL ESTIMATION
RESULTS.
Dependent Log(Price)
Variable:
F-statistic: 4.85 75 D-W: 1.6749
R-squared: 0.2634 RESET: 4.3054
Adj. R-squared: 0.2091 LM Het. Test: 4.0981
N: 672
Variable 5:23;"; 5‘33?“ t-statistic P-value
C 28.2631 66.1024 0.4276 [.669]
LOT 0.0225 0.0029 7.8666 [.000]
LOT2 -0.0001 0.0000 -3.9568 [.000]
LOT3 0.0000 0.0000 2.7963 [.005]
CITY -0.0002 0.0330 -0.0064 [.995]
CITY2 0.0000 0.0003 0.1433 [.886]
CITY3 0.0000 0.0000 -0.1958 [.845]
HWY 0.0756 0.0637 1.1876 [.235]
HWY2 -0.0134 0.011 1 -1.2023 [.230]
HWY3 0.0005 0.0005 1.0244 [.306]
UTIL 0.1992 0.1739 1.1455 [.252]
PRISON -0.3376 0.5166 -0.6536 [.514]
ZONED -0.0025 0.2040 -0.0121 [.990]
PAVE -0.1994 0.0919 -2. 1697 [.030]
PLAT -0.0415 0.1015 I , -0.4083 [.683]
H20 0.2856 0.1140 2.5041 [.013]
R_INT -0.8245 16.9815 -0.0486 [.961]
R_INT2 0.0626 1.9205 0.0326 [.974]
R_INT3 -0.0014 0.0719 -0.0199 [.984]
GSP -0.2082 0.3554 -0.5857 [.558]
GSP2 0.0009 0.0015 0.5651 [.572]
GSP3 0.0000 0.0000 -0.5325 [.595]
REGNLP 0.0903 0.2294 0.3939 [.694]
PERPUB -0.4283 20.3150 -0.021 1 [.983]
PERPUBZ -2.3728 60.8029 -0.0390 [.969]
PERPUB3 5.6643 49.1280 0.1 153 [.908]

 

146

 

 

Table A7 (continued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable 53:22:: 82:19:13." t-statistic P-value
CC 0.7002 0.5841 1.1986 [.231]
CC_P -1.4000 1.1554 -1.2117 [.226]
SEED 0.4590 0.5200 0.8827 [.378]
THIN 1.4981 0.8650 1.7320 [.084L
THIN_P -2.7460 1.6556 -1.6586 [.098]
SEL 1.3040 1.1055 1.1796 [.239]
SEL_P -1 .3 837 2.6087 -0.5304 [.596]
REM -0.0289 0.9655 -0.0299 [.976]
REM_P -0.1849 1.6931 -0. 1092 [.913]
F_DIS -0.0001 0.0001 -2.3461 [.019]
F_DISZ 0.0000 0.0000 2.1325 [.033]
F_DIS3 0.0000 0.0000 -2.0126 [.045]
S_DIS 0.0000 0.0001 -0.6281 [.530]
S_DISZ 0.0000 0.0000 0.9422 [.346]
S_DIS3 0.0000 0.0000 -1.1782 [.239]
PERDIS 0.0000 0.0000 0.9879 [.324]
PERADJ -1.5513 1.5185 -1.0216 [.307]
PERADJZ 6.0106 4.3702 1.3754 [.170]
PERADJ 3 -4.6565 3.3267 -1.3997 [.162]
CONIFER -0.1441 0.1428 -1 .0094 [.313]
HARDWOOD 0.0741 0.1518 0.4878 [.626]

 

 

 

 

I47

 

 

TABLE A8. LOG-QUADRATIC VACANT FULL MODEL ESTIMATION
RESULTS CORRECTING HETEROSCEDASTICITY.

 

 

 

 

 

 

 

 

"‘1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Log(Price)
Variable:
F-statistic: 4.8575 D-W: 1.6749
R-squared: 0.2634 RESET: 4.3054
Adj. R-squared: 0.2091 LM Het. Test: 4.0981
N: 672
Variable 5:213:22: 8:32:33.“ t-statistic P-value
‘ C 28.2631 64.1 124 0.4408 [.659L
LOT 0.0225 0.0028 7.9954 [.000]
LOT2 -0.0001 0.0000 -4.4975 [.000]
LOT3 0.0000 0.0000 3.1952 [.00]]
CITY -0.0002 0.0341 -0.0062 [.995]
CITY2 0.0000 0.0003 0.1221 [.903]
CITY3 0.0000 0.0000 -0.1530 [.878]
HWY 0.0756 0.0595 1.2701 [.205]
HWY2 -0.0134 0.0112 -1 . 1971 [.232]
HWY3 0.0005 0.0005 0.9756 [.330]
UTIL 0.1992 0.1807 1.1027 [.271]
PRISON -0.3376 0.5457 -0.6187 [.536]
ZONED -0.0025 0.2127 -0.01 16 .991]
PAVE -0. 1994 0.0961 -2.0743 [.03 8]
PLAT -0.0415 0.1 103 -0.3 757 [.707]
H20 0.2856 0.1043 2.7377 [.006]
R_INT -0.8245 16.8428 -0.0490 [.961]
R_INT2 0.0626 1.9091 0.0328 [.974]
R_INT3 -0.0014 0.0716 -0.0200 [.984]
GSP -0.2082 0.3169 -0.6569 [.511]
GSP2 0.0009 0.0014 0.6330 [.527]
GSP3 0.0000 0.0000 -0.5958 [.552]
REGNLP 0.0903 0.2316 0.3 901 [.697]
PERPUB -0.4283 20.7392 -0.0207 [.984]
PERPUB2 -2.3728 62.3217 -0.0381 [.970]
PERPUB3 5.6643 50.5161 0.1121 [.911]

 

 

148

 

 

Table A8 (cgtinued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable gigging; 8:23:31 t-statistic P-value
CC 0.7002 0.4592 1.5246 [.128]
CC_P -1.4000 0.9762 -1.4340 [.152]
SEED 0.4590 0.2953 1.5543 [.121]
THIN 1.4981 1.0455 1.4329 [.152]
THIN_P -2.7460 2.1888 -1.2546 [210]
SEL 1.3040 0.7195 1.8124 [.070]
SEL_P -1.3837 1.5206 -0.9100 [.363]
REM -0.0289 0.6761 -0.0427 [.966]
REM_P -0.1849 1.0960 -0. 1687 [.866]
FJDIS -0.0001 0.0001 -2.0195 [.044]
F_DISZ 0.0000 0.0000 1 .6675 [.096]
F_DIS3 0.0000 0.0000 -1 .3 862 [.166]
S_DIS 0.0000 0.0001 -0.6038 [.546]
S_DIS2 0.0000 0.0000 1.0689 [.286]
S_DIS3 0.0000 0.0000 -1.4542 [.146]
PERDIS 0.0000 0.0000 0.9122 [.362]
PERADJ -1.5513 1.3830 -1.1217 [.262]
PERADJ 2 6.0106 4.0020 1.5019 [.134]
PERADJ3 -4.6565 2.9449 -1.5812 [.114]
CONIFER -0.1441 0.1 176 -1.2256 [.221]
HARDWOOD 0.0741 0.1288 0.5751 [.565]

 

 

149

 

 

TABLE A9. TESTS OF SQUARED AND CUBED TERMS OF FULL
VACANT LOG-QUADRATIC MODEL.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Independent Tests
. Standard . .
Parameter Estimate t-statistlc P-value
Error
LOT -0.0001 0.0000 -4.4990 [.000]
CITY 0.0000 0.0003 0.1220 [.903]
HWY -0.0129 0.0107 -1.2065 [.228]
R_INT 0.0612 1.8375 0.0333 [.973]
GSP 0.0009 0.0014 0.6330 [.527]
PERPUB 3.2915 12.0860 0.2723 [.785]
F_DIS 0.0000 0.0000 1.6675 [.095]
S_DIS 0.0000 0.0000 1.0689 [.285]
PERADJ 1.3541 1.2485 . 1.0846 [.278]
Joint Tests
Chi-squared(9): 25.7845 P-value: 0.0022
F(9, 1292): 2.8649 P-value: 0.0025

 

 

 

 

 

150

 

 

TABLE A10. “BEST” VACANT FULL MODEL ESTIMATION RESULTS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Log(Price)
Variable:
F-statistic: 6.5357 D-W: 1.6848
R-squared: 0.2466 RESET: 2.2818
AdLR-squared: 0.2089 LM Het. Test: 7.7603
N: 672

. Estimated Standard . .

Vanable Coefﬁcient Error t-statistlc P-value

C 7.8812 0.8064 9.7728 [.000]
LOT 0.0213 0.0026 8.1960 [.000]
LOT2 -0.0001 0.0000 -4.493 1 [.000]
LOT3 0.0000 0.0000 3.0792 [0%]
CITY 0.0054 0.0016 3.3724 [.001]
HWY -0.0268 0.0079 -3.3751 [.001]
UTIL 0.1984 0.1789 1.1092 [.268]
PRISON 0.1073 0.1943 0.5523 [Sim
ZONED -0.1560 0.0980 -1.5920 [.112]
PAVE -0.1831 0.0905 -2.0233 [.043]
PLAT 0.0063 0.1030 0.061 1 [.951]
H20 0.2589 0.0977 2.6494 [.008]
R_INT -0.0240 0.0520 -0.4623 [.644]
GSP 0.0034 0.0016 2.0890 [.037]
REGNLP 0.2633 0.1726 1.5256 [.128]
PERPUB 0.4345 0.4073 1.0670 [.286]
CC 0.6770 0.4690 1.4434 [.149]
CC_P -1.3838 0.9944 , -1.3916 [.165]
SEED 0.3718 0.3449 - 1.0781 [.281]
THIN 1.4748 1.0355 ' 1.4241 [.155]
THIN_P -2.6499 2.1445 -1.2357 [.2 ll]
SEL 1.6468 0.6185 2.6623 [.008]
SEL_P -2.2343 1.1414 -1.9575 [.051]
REM -0.0408 0.7089 -0.0576 [.954]
REM_P -0.21 13 1.081 1 -0.1954 [.845]
F_DIS -0.0001 0.0001 -2.5967 [.010]
F_DIS2 0.0000 0.0000 1.9147 [.056]
F_DIS3 0.0000 0.0000 -1.5300 [.127]
S_DIS 0.0000 0.0000 0.3920 [.695]
PERDIS 0.0000 0.0000 0.2772 [.782]
PERADJ 0.3 872 0.2364 1 .63 75 [.102]
CONIF ER -0. 1729 0.1 196 -1.4460 [.149]
HARDWOOD -0.0232 0.0994 -0.2333 [.816]

 

 

 

 

 

151

 

 

 

TABLE A1 1. VACANT FULL MODEL ARTIFICIAL REGRESSION

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ESTIMATION RESULTS.
Dependent Log(Price)
Variable:
F-statistic: 5.5089 D-W: 1.6953
R-squared: 0.2639 RESET: 5.8737
Adj. R-squared: 0.2160 LM Het. Test: 5.2077
N: 672

Variable 3:22:53: $2223" t-statistic P-value

C 35.9034 22.6439 1.5856 [.113]
LOT 0.0198 0.0050 3.9188 [.000]
LOT2 -0.0001 0.0000 -2.9614 [.003]
LOT3 0.0000 0.0000 2.2432 [.025]
LOG(LOT) 0.0377 0.0927 0.4061 [.685]
CITY 0.0024 0.0039 0.6211 [.535]
LOG(CITY) 0.0830 0.2336 0.3554 [.722]
HWY -0.0333 0.0198 -1.6800 [.093]
LOG(HWY) 0.1095 0.1173 0.9336 [.351]
UTIL 0.2162 0.1789 1.2080 [.227]
PRISON -0.3707 0.3249 -1.1411 [.254]
ZONED 0.0741 0.1228 0.6036 .546]
PAVE -0. l 821 0.0979 -1.8594 [.063]
PLAT -0.0352 0.1421 -0.2477 [.804]
H20 0.2612 0.1035 2.5247 [.012]
R_INT 0.5257 0.9277 0.5667 [.571]
LOG(R_INT) -4.8872 7.9425 -0.6153 [.539]
GSP 0.0262 0.0255 1.0301 [.303]
LOG(GSP) -5.0667 5.8998 -0.8588 [.391]
REGNLP 0.2166 0.2149 1.0075 [.314]
PERPUB 15.1798 6.8218 2.2252 [.026]
LOGiPERPUB) -19.2865 8.9459 -2. 1559 [.031]
CC 0.7126 0.4797 1.4856 [.138]
CC]P -1.3772 1.0275 -l.3403 [.181]
SEED 0.4072 0.3153 1.2917 [.197]
THIN 1.2803 1.0231 1.2514 [.211]
THIN_P -2.4069 2.1388 -1.1254 [.261]
SEL 1.6656 0.7717 2.1585 [.031]
SEL_P -2.4329 1.6463 -1.4778 [.140]
REM 0.0803 0.6096 0.1317 [.895]
REM_P -0.4906 0.9235 -0.5312 [.595]

 

152

 

 

Table All (continued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable 33:22:21 St;:?::d t-statistic P—value
F_DIS 0.0000 0.0001 0.0196 [.984]
F_DIS2 0.0000 0.0000 0.3041 [.761]
F_DIS3 0.0000 0.0000 -0.4442 [.657]
LOG(FADIS) -0.0326 0.0368 -0.8868 [.376]
S_DIS 0.0000 0.0000 -0.7084 [.479]
LOG(S_DIS) 0.0288 0.0256 1.1279 [.260]
PERDIS 0.0000 0.0000 0.3617 [.718]
PERADJ 0.6776 2.3636 0.2867 [.774]

' LOG(PERADJ) -0.4958 3.1663 -0.1566 [.876]
CONIF ER 01 143 0.1143 -0.9999 [.318]
HARDWOOD 0.1095 0.1438 0.7612 [.447]

 

153

 

 

TABLE A12. ARTIFICIAL REGRESSION TESTS — QUADRATIC

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TERMS.

Independent Tests
. Standard . .
Parameter Estimate t-stattstlc P-value
Error

LOT 0.0197 0.0050 3.9216 [.000]
CITY 0.0024 0.0039 0.6211 [.535]
HWY -0.0333 0.0198 -1.6800 [.093]
R_INT 0.5257 0.9277 0.5667 [.571]
GSP 0.0262 0.0255 1.0301 [303]
PERPUB 15.1798 6.8218 2.2252 [.026]
F_DIS 0.0000 0.0001 0.0196 [.984]
S_DIS 0.0000 0.0000 -0.7084 [.479]
PERADJ 0.6776 2.3636 0.2867 [.774]
Joint Tests
Chi-sguared(9): 29.4289 P-value: 0.0006
F(10, 1272): 3.2699 P-value: 0.0007

 

 

 

 

 

 

TABLE A13. ARTIFICIAL REGRESSION TESTS - LOGARITHMIC

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TERMS.

Independent Tests
. Standard . .
Parameter Estimate t-statlstic P-value
Error
LOT 0.0377 0.0927 0.4061 [.685]
CITY 0.0830 0.2336 0.3554 [.722]
HWY 0.1095 0.1173 0.9336 [.351]
R_INT -4.8872 7.9425 -0.6153 [.538]
GSP -5.0667 5.8998 -0.8588 [.390]
PERPUB -19.2865 8.9459 -2.1559 [.031]
F_DIS -0.0326 0.0368 -0.8868 [.375]
S_DIS 0.0288 0.0256 1.1279 [.259]
PERADJ -0.4958 3.1663 -0.1566 [.876]
Joint Tests

Chi-guaredﬂ 0): 14.7519 P-value: 0.0980
F( 10, 1280): 1.6391 P-value: 0.0986

 

 

 

 

 

154

 

 

 

TABLE A14. FINAL DEVELOPED FULL MODEL ESTIMATION
RESULTS WITHOUT OUTLIERS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Log(Price)
Variable:
F-statistic: 29.12 D-W: 1.9768
R-squared: 0.6443 RESET: 4.7596
Adj. R-sggred: 0.6221 LM Het. Test: I 1.4735
N: 667

Variable 53:22:; St;:f::d t-statistic P-value
C 9.75899 0.5923 16.48 .0001
HOUSE 0.0003 0.0001 5.08 .0001
LOT 0.0096 0.0010 9.18 .0001
CITY 0.0004 0.0012 0.30 .7642
HWY 0.0352 0.0060 5.83 .0001
GARAGE 0.3102 0.0447 6.94 .0001
UTIL 0.3027 0.0672 4.50 .0001
PRISON -0.6641 0.2278 -2.92 .0037
ZONED 0.1574 0.0934 1.69 .0925
BR 0.2080 0.1231 1.69 .0915
B 0.2559 0.3414 0.75 .4537
D -0.5 750 0.0466 -12.34 .0001
MH -0.7777 0.0655 -11.87 .0001
PAVE -0.0173 0.0492 -0.35 .7250
HOME 0.17023 0.0428 3.98 .0001
PLAT -0.0003 0.0003 -1 .00 .3 179
H20 0.1750 0.0669 2.62 .0091
R_IN T -0.03 60 0.0295 -1.22 .2230
GSP 0.0059 0.0009 6.31 .0001
REGNLP -0.0305 0.1099 -0.28 .7815
PERPUB -14.9154 4.4226 -3.37 .0008
PERPUB2 38.5905 12.7313 3.03 .0025
PERPUB3 -25.9767 9.7820 -2.66 .0081

 

 

 

 

 

155

 

 

Table A14 (continued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable 53:33:; 8:23:31 t-statistic P-value
CC 0.2603 1.0468 0.25 .8037
CC_P -0.2468 2.4398 -0.10 .9195
SEED -0.3777 0.4864 -0.78 .4377
THIN 2.9313 1.1516 2.55 .0112
THIN_P -6.8561 2.6814 -2.56 .0108
SEL 0.5295 0.9091 0.58 .5605
SEL_P -0.1796 2.3490 -0.08 .9391
REM 0.0209 2.6415 0.01 .9937
REM_P 0.5308 7.0278 0.08 .9398
F_DIS -0.0000 0.0001 -1.84 .0656
S_DIS -0.0000 0.0000 -0.37 .7098
PERDIS 0.0000 0.0000 1.37 .1697
PERADJ 0.2089 0.2000 1.04 .2967
CONIFER -0.0013 0.0962 -0.01 .9895
HARDWOOD -0.0060 0.0788 -0.08 .9397

 

156

 

 

 

 

TABLE A15. FINAL VACANT FULL MODEL ESTIMATION RESULTS

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

WITHOUT OUTLIERS.

Dependent Log(Price)

Variable:

F-statistic: 8.84 D-W: 1 .6488

R-squared: 0.3196 RESET: 6.9327

Adj. R—squared: 0.2834 LM Het. Test: 1.2878

N: 672

Variable 5:22:33; 8:13:13.“ t-statistic P-value

C 7.6614 0.7128 10.75 .0001
LOT 0.0220 0.0022 9.93 .0001
LOT2 -0.0001 0.0000 -4.94 .0001
LOT3 0.0000 0.0000 3.45 - .0006
CITY 0.0033 0.0013 2.50 .0125
HWY -0.0182 0.0079 -2.29 .0226
UTIL 0.3059 0.1480 2.07 .0391
PRISON 0.0215 0.1668 0.13 .8976
ZONED -0.2059 0.0839 -2.46 .0144
PAVE -0.0991 0.0767 -1.29 .1966
PLAT 0.0013 0.0007 1.93 .0540
H20 0.2603 0.0927 2.81 .0052
R_INT -0.01 18 0.0447 -0.26 .7920
GSP 0.0045 0.0014 3.24 .0013
REGNLP 0.1052 0.1503 0.70 .4840
PERPUB 0.4860 0.3482 1.40 .1633
CC 0.4923 0.4906 1.00 .3160
CC_P -1.1131 0.9781 -1.14 .2556
SEED 0.3606 0.4359 0.83 .4085
THIN 1.5521 0.7144 2.17 .0302
THIN_P -2.7451 1.3731 -2.00 .0460
SEL 1.8249 0.8447 2.16 .0311
SEL_P -2.4637 1.9564 -1.26 .2084
REM 0.0723 0.7959 0.09 .9276
REM_P -0.3863 1.3934 -0.28 .7817
F_DIS -0.0002 0.0000 -3.57 .0004
F_DIS2 0.0000 0.0000 2.96 .0032
F_DIS3 -0.0000 0.0000 -2.59 .0097
S_DIS 0.0000 0.0000 0.26 .7980
PERDIS -0.0000 0.0000 -0.08 .9346
PERADJ 0.4527 0.2537 1.78 .0749
CONIFER -0.1614 0.1171 -1.38 .1688

 

 

157

‘_.. 0:. I?“ '-

 

 

Table A15 (continued).

 

 

 

 

 

 

 

 

 

 

Variable 5:22.233; 8:21:33.“ t-statistic P-value
PERDIS -0.0000 0.0000 -0.08 .9346
PERADJ 0.4527 0.2537 1.78 .0749
CONIFER -0.1614 0.1171 -1.38 .1688
HARDWOOD -0.0646 0.1044 -0.62 .5366

 

158

 

 

APPENDIX B

Adjacent Developed Model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

TABLE Bl. LOG-QUADRATIC DEVELOPED ADJACENT MODEL
ESTIMATION RESULTS.
Dependent Log(Price)
Variable:
F-statistic: 2.1 136 D-W: 2.3004 [7
R-squared: 0.5441 RESET: 0.0007
Adj. R-squared: 0.2867 LM Het. 1.5183 1
Test:
N: 134 -
Estimated J
Variable 5:21:12." t-statistic P-value
Coefficient '
C -31.6059 129.6809 -0.2437 [.808]
HOUSE -0.0003 0.0022 -0.1471 [.883]
HOUSE2 0.0000 0.0000 0.3208 [.749]
HOUSE3 0.0000 0.0000 -0.2143 [.831]
LOT -0.0015 0.0155 -0.0998 [.921]
LOT2 0.0003 0.0003 1.0401 [.301]
LOT3 0.0000 0.0000 -1.1062 [.272]
CITY 0.0494 0.1032 0.4786 [.633]
CITY2 -0.0005 0.0015 -0.3313 [.741]
CITY3 0.0000 0.0000 0.2882 [.774]
HWY -0.1788 0.1416 -1 .2629 [.210]
HWY2 0.0234 0.0202 1.1569 [.251]
HWY3 -0.0007 0.0008 -0.8944 [.374]
GARAGE 0.4382 0.1833 2.3912 [.019]
UTIL -0.0572 0.3220 -0.1776 [.859]
PRISON -1.0289 1.2340 -0.8338 [.407]
ZONED 0.1908 0.6129 0.31 13 [.756]
BR 0.4766 0.9133 0.5218 [.603]
D -0.3011 0.1962 -1.5347 [.129]
MH -0.5226 0.2425 -2.1552 [.034]
PAVE 0.3832 0.2689 1.4251 [.158]
HOME -0.1 183 0.2177 -0.5436 L588]
PLAT 0.0758 0.3765 0.2012 [.841]

 

159

 

 

Table Bl (continued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Estimated Standard
Variable Error t-statistic P-value
Coefficient
H20 0.3504 0.2971 1.1792 [.242]
R_IN T 13.7464 33.2220 0.4138 [.680]
R_INT2 -1.5844 3.7551 -0.4219 [.674]
R_IN T3 0.0604 0.1403 0.4305 [.668]
GSP -0.0078 0.8923 -0.0088 [.993]
GSP2 0.0002 0.0039 0.0444 [.965]
GSP3 0.0000 0.0000 -0.0760 .940]
REGNLP 0.0852 0.5769 0.1477 [.883]
PERPUB -23.7614 41.3050 -0.5753 [.567]
PERPUB2 72.7663 122.8686 0.5922 [.555]
PERPUB3 -56.3423 98.8633 -0.5699 [.570]
CC -2.9969 2.2124 -1.3546 [.179L
CC_P 7.2462 5.0631 1.4312 [.156]
SEED -0.4224 1.0025 -0.4213 [.675]
THIN 3.9028 2.2702 1.7191 [.089]
THIN_P -8.3556 5.2924 -1.5788 [.118]
SEL 2.2045 1.7680 1.2469 L216]
SEL_P -4.5304 4.5640 -0.9926 [.324]
REM 0.3531 4.6351 0.0762 [.939]
REM_P 0.4048 12.3239 0.0328 [.974]
FED -0.1992 0.4887 -0.4077 [.685]
PERADJ 0.5229 2.0860 0.2507 [.803]
PERADJ2 0.071 1 5.6072 0.0127 [.990]
PERADJ 3 -0.6496 4.1037 -0.1583 [.875]
CONIFER -0.0863 0.2004 -0.4308 [.668]
HARDWOOD -0.0346 0.2172 -0.1591 [.874]

 

160

 

 

TABLE B2. TESTS OF SQUARED AND CUBED TERMS OF ADJACENT
DEVELOPED LOG-QUADRATIC MODEL.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Independent Tests
. Standard . .
Parameter Estimate E t-statistic P-value
rror
HOUSE 0.0000 0.0000 0.3208 L748]
LOT 0.0003 0.0003 1.0397 [298]
CITY -0.0005 0.0015 -0.3314 [.740]
HWY 0.0227 0.0195 1.1663 [.244]
R_INT -l .5240 3.6148 -0.4216 L673]
GSP 0.0002 0.0039 0.0443 .965]
PERPUB 16.4241 25.9339 0.6333 [.527]
PERADJ -0.5785 1.8396 -0.3145 [.753]
Joint Tests ,
Chi-squared—(S): 3.3069 P-value: 0.9137
F(8, 1269): 0.4134 P-value: 0.9134

 

 

 

 

 

I61

 

 

TABLE B3. “BEST” DEVELOPED ADJACENT MODEL ESTIMATION

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

RESULTS.
Dependent Log(Price)
Variable:
F-statistic: 3.1662 D-W: 2.2039
R-sqgared: .5008 RESET: 0.1023
Adj. R-squared: .3426 LM Het. 1.4453
Test: '
N: 134
Estimated
Variable 5:22;?“ t-statistic P-value

Coefficient
C 7.5194 1.5995 4.7010 [.000]
HOUSE 0.0005 0.0003 2.0491 [.043]
LOT 0.0120 0.0028 4.2860 [.000]
CITY 0.0002 0.0036 0.0552 [.956]
HWY 0.0214 0.0141 1.5202 [.132J
GARAGE 0.3434 0.1613 2.1287 [.036]
UTIL 0.0698 0.2586 0.2700 [.788]
PRISON -0.8160 0.5897 -1.3838 [.169]
ZONED 0.0271 0.3272 0.0828 [.934]
BR 0.6397 0.7856 0.8142 [.417]
D -0.2997 0.1756 -1.7060 [.091]
MH -0.45 84 0.2193 -2.0906 [.039]
PAVE 0.4227 0.2356 1.7939 [.07ﬂ
HOME 0.0817 0.1675 0.4877 [.627]
PLAT 0.1109 0.3100 0.3578 [.721]
H20 0.3672 0.2568 1.4299 [.156]
R_INT -0.0033 0.1036 -0.0315 [.975]
GSP 0.0051 0.0037 1.3731 [.173]
REGNLP -0.3270 0.4301 -0.7603 [.449]
PERPUB 1.8242 1.2883 1.4160 [.160]
CC -1.6914 1.9325 -0.8752 [.384]
CC_P 4.1620 4.4518 0.9349 [.352]
SEED -0.0927 0.7593 -0. 1221 [.903]
THIN 3.0224 1.8653 1.6204 [.108]
THIN_P -6.3259 4.3757 -1 .4457 [.151]

 

 

 

 

 

 

162

 

‘4“—

 

 

Table B3 (continued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Estimated Standard
Variable Error t-statistic P-value
Coefﬁcient

SEL 1.8088 1.5672 1.1541 [.251]
SEL_P -3.2753 4.0656 -0.8056 [.422]
REM 0.6291 4.2261 0.1489 [.882]
REM_P -0.8067 11.2561 -0.0717 [.943]
FED 0.0658 0.3161 0.2082 [.836L
PERADJ 0.1890 0.3397 0.5564 [.579]
CONIF ER -0.1364 0.1747 -0.7805 [.437]

‘ HARDWOOD -0.1275 0.1907 -0.6690 [.505L

 

163

 

 

 

 

TABLE B4. DEVELOPED ADJACENT ARTIFICIAL REGRESSION

MODEL ESTIMATION RESULTS.

 

 

 

 

 

 

 

 

 

 

 

“vs-57

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Log(Price)
Variable:
F-statistic: 2.6260 D-W: 2.2321
R-squared: 0.5304 RESET: 0.0860
Adj. R-squared: 0.3284 LM Het. 1.3434
Test:

N: 134

Variable giggle; St;:f::d t-statistic P-value
C -35.6355 62.4360 -0.5708 [.570]
HOUSE 0.0012 0.0006 2.1672 [.033]
LOG(HOUSE) -0.3860 0.3338 -1.1563 [.251]
LOT 0.0104 0.0048 2.1792 [.032]
LOG(LOT) 0.0394 0.1286 0.3065 [.760]
CITY 0.0060 0.0086 0.6977 [.487]
LOG(CITY) -0.1972 0.6405 -0.3078 [.759]
HWY 0.0836 0.0388 2.1558 [.034]
LOG(HWY) -0.4142 0.2727 -l.5190 [.132]
GARAGE 0.4059 0.1728 2.3489 [.021]
UTIL -0.0269 0.3005 -0.0894 [.929]
PRISON -0.5500 0.7824 -0.7029 [.484]
ZONED 0.2847 0.4659 0.6110 [.543]
BR 0.7712 0.8420 0.9159 [.362]
D -0.2611 0.1862 -1 .4025 [.164]
MH -0.4412 0.2286 -1.9304 [.057]
PAVE 0.3697 0.2529 1.4620 [.147]
HOME -0.0787 0.1919 -0.4102 [.683]
PLAT 0.1567 0.3566 0.4393 [.661]
H20 0.3351 0.2841 1.1795 [.241]
R_INT 0.7970 1.5688 0.5080 [.613]
LOG(R_INT) -6.4462 13.4948 -0.4777 [.634]
GSP -0.0470 0.0629 -0.7467 [.457]
LOG(GSP) 1 1.9344 14.5290 0.8214 [.414L
REGNLP -0. 1006 0.5078 -0.1981 [.843]
PERPUB 7.2700 18.1955 0.3996 [.690]
LOG(PERPUB) -7.8104 24.3361 -0.3209 [.749]

 

 

 

Table B4 (continued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable 53:23:; St;:f::d t-statistic P-value
CC -2.3825 2.0206 -1.1791 [.241L
CC_P 5.8771 4.6644 1.2600 [.21 1]
SEED -0.4627 0.7956 -0.5815 [.562]
THIN 3.7487 2.1220 . 1.7666 [.081]
THIN4_P -8.0556 4.9481 -1.6280 [.107]
SEL 2.1591 1.6296 1.3250 [.188]
SEL_P -4.1754 4.2258 -0.9881 [.326]
REM 0.2247 4.4004 0.051 1 [.959]
REM_P 0.6942 1 1.6745 0.0595 [.953]
FED -0.0870 0.4368 -0.1992 [.843]
PERADJ -1.9234 3.3028 -0.5824 L562]
LOG(PERADJ) 2.9221 4.5350 0.6443 [.521]
CONIFER -0.0999 0.1867 -0.5353 [.594]
HARDWOOD -0.1010 0.2047 -0.4937 [.623]

 

165

 

 

TABLE B5. DEVELOPED ADJACENT MODEL ARTIFICIAL
REGRESSION TESTS — QUADRATIC TERMS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Independent Tests
. Standard . .
Parameter Estimate t-statistic P-value
Error
HOUSE 0.0012 0.0006 2.1672 [.030]
LOT 0.0104 0.0048 2.1792 [.029]
CITY 0.0060 0.0086 0.6977 [.485]
HWY 0.0836 0.0388 2.1558 [.03 I]
R_INT 0.7970 1.5688 0.5080 [.611]
GSP -0.0470 0.0629 -0.7467 [.455]
PERPUB 7.2700 18.1955 0.3996 [.689]
PERADJ -1.9234 3 .3028 -0.5824 [.560]
Joint Tests
Chi-squared(8): . 16.5301 P-value: 0.0354
F(8, 1277):. 2.0663 P-value: 0.0352
TABLE B6. DEVELOPED ADJACENT MODEL ARTIFICIAL
REGRESSION TESTS - LOGARITHMIC TERMS.
Independent Tests
. Standard . .
Parameter Estimate t-statistic P-value
Error
HOUSE -0.3860 0.3338 -1.1563 [.248]
LOT 0.0394 0.1286 0.3065 [.759]
CITY -0.1972 0.6405 -0.3078 [.758]
HWY -0.4142 0.2727 -l.5190 [.129]
R_INT -6.4462 13.4948 -0.4777 [633]
GSP 11.9344 14.5290 0.8214 [.411]
PERPUB -7.8104 24.3361 -0.3209 [.748]
PERADJ 2.9221 4.5350 0.6443 [.519]
Joint Tests
Chi-squared(8): 5.8650 P-value: 0.6624
F(8, 1277): 0.7331 P-value: 0.6623

 

 

 

 

 

I66

 

 

 

Adjacent Vacant Model

 

TABLE B7. LOG-QUADRATIC VACANT ADJACENT MODEL

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ESTIMATION RESULTS.
Dependent Log(Price)
Variable:
F -statistic: 4.2675 D-W: 1.81 15
R-squared: 0.5530 RESET: 1.3951
Adj. R-squared: 0.4234 LM Het. 0.5338

Test:
N: 179
Estimated Standard
Variable . t-statistic P-value
Coefﬁcient
Error

C 1 19.2046 105.4498 1.1304 [.260]
LOT 0.0281 0.0047 5.9375 [.000]
LOT2 -0.0001 0.0000 -4.0673 [.000]
LOT3 0.0000 0.0000 3.4600 [.001]
CITY -0.0322 0.061 1 -0.5272 [.599[
CITY2 0.0008 0.0007 1.1032 [.272]
CITY3 0.0000 0.0000 -1.3200 [.189]
HWY 0.0700 0.1012 0.6921 [.490]
HWY2 -0.0219 0.0176 -1.2438 [.216]
HWY3 0.001 1 0.0007 1.4209 [.158[
UTIL -0.5375 0.7108 -0.7563 [.451[
PRISON l .0348 1 .4949 0.6923 [.490]
ZONED 0.7988 0.3921 2.0372 [.044]
PAVE 0.2220 0.2050 1.0829 [.281]
PLAT -0.1351 0.2533 -0.5335 [.595]
H20 0.5236 0.2331 2.2460 [.026]
R_INT -41.7587 27.8466 -1.4996 [.136]
R_INT2 4.8053 3.1638 1.5189 [.131]
R_INT3 -0.1833 0.1189 -1.5412 J.126]
GSP 0.0510 0.5325 0.0957 [.924]
GSP2 -0.0004 0.0023 -0.1640 [.870]
GSP3 0.0000 0.0000 0.2394 [.81 IL
REGNLP 0.9029 0.4479 2.0157 [.046
PERPUB 34.0567 138.9542 0.2451 [.807]
PERPUB2 -43.5348 292.2051 -0.1490 [.882
PERPUB3 1 1.9426 188.4530 0.0634 [.950]

 

167

 

 

Table B7 (continued).

 

Standard

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Variable Emma?“ t-statistic P-value
Coefﬁcient
Error

CC 0.1059 0.5692 0.1861 [.853]
CC_P -0.0994 1.1035 -0.0901 [.928]
SEED 0.4110 0.4301 0.9557 [.341]
THIN -0.1802 0.7586 -0.23 75 [.813]
THIN_P 0.8300 1.4700 0.5646 [.573]
SEL 0.0125 0.9939 0.0125 [.990]
. SEL_P 1.4314 2.3343 0.6132 [.541]
REM -0.7816 0.8346 -0.9365 [.351]
REM_P 0.9690 1.4706 0.6589 [.51 1]
FED -O.7523 0.5282 -1.4241 [.157]
PERADJ 0.7364 1.5612 0.4717 [.638]
PERADJ2 -1.0652 4.2101 0.2530 [.801]
PERADJ 3 0.5161 3 .0747 0.1679 [.867]
CONIFER -0.1247 0.1340 -0.9304 [.354]
HARDWOOD 0.2461 0.1744 1.4110 [.160]

 

168

 

 

 

 

 

 

TABLE B8. 'TESTS OF SQUARED AND CUBED COEFFICIENTS OF
ADJACENT VACANT LOG-QUADRATIC MODEL.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Independent Tests
. Standard . .
Parameter Estimate t-statistic P-value
Error
LOT -0.0001 0.0000 4.0680 [.000]
CITY 0.0008 0.0007 1.1024 [.270]
HWY -0.0209 0.0169 -1.2351 [.217]
R_INT 4.6220 3.0449 1.5180 [.129]
GSP —0.0004 0.0023 -0.1639 [.870]
PERPUB -31.5922 106.0576 -0.2979 [.766]
PERADJ -0.5491 1.3688 -0.401 1 .688]
Joint Tests
Chi-squared(7): 26.0561 P-value: 0.0005
F(7, 1294): 3.7223 P-value: 0.0006

 

 

 

 

 

169

 

 

TABLE B9. “BEST” LOG-LINEAR VACANT ADJACENT MODEL

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ESTIMATION RESULTS.
Dependent Log(Price)
Variable:
F-statistic: 5.7530 D-W: 1.7918
R-squared: 0.5178 RESET: 2.2086
Adj. R-squared: 0.4278 LM Het. 0.4532
Test:
N: 179
Variable 523333;: St;::::d t-statistic P-value
C 8.0181 1.4321 5.5989 [.000]
LOT 0.0291 0.0041 7.0804 .000]
LOT2 -0.0001 0.0000 -5.5176 [.000]
LOT3 0.0000 0.0000 4.7283 [.000]
CITY -0.0026 0.0035 -0.7456 [.457]
HWY -0.0043 0.0130 -0.3311 [.743
UTIL -0.7475 0.6726 -1 .1 1 14 [.268]
PRISON 0.1345 0.5493 0.2449 [.807]
ZONED 0.8706 0.21 19 4.1076 [.000]
PAVE 0.1 124 0.1873 0.6000 [.549]
PLAT -0.2441 0.2344 -1.0415 [.299]
H20 0.4121 0.2050 2.0101 [.046]
R_INT -0.1 198 0.0829 -1.4449 [.151]
GSP 0.0018 0.0026 0.6993 [.485]
REGNLP 0.3359 0.3780 0.8886 [.376]
PERPUB 2.1840 1.4845 1.4712 [.143]
CC -0.3600 0.5358 -0.6720 [.503]
CC_P 0.6625 1.0452 0.6339 [.527]
SEED 0.2538 0.4164 0.6095 [.543]
THIN -0.0342 0.7142 -0.0478 [.962]
THIN_P 0.4880 1.3918 0.3506 [.726]
SEL 0.1026 0.8777 0.1 169 [.907]
SEL_P 0.8400 1.9748 0.4254 [.671]
REM -0.8946 0.8045 -1.1 120 [.268]
REM_P 1.1395 1.4112 0.8075 [.421]
FED -0.8328 0.2058 -4.0459 [.000]
PERADJ 0.2608 0.2565 1.0167 [.311]
CONIFER -0.1280 0.1258 -1.0174 [.311]
HARDWOOD 0.1580 0.1538 1.0272 [.306]

 

170

 

 

TABLE B10. VACANT ADJACENT ARTIFICIAL REGRESSION MODEL
ESTIMATION RESULTS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Log(Price)
Variable:
F -statistic: 4.7501 D-W: 1.7237
R-squared: 0.5376 RESET: 0.5404
Adj. R-squared: 0.4244 LM Het. 0.1300
Test:
N: 179
. Standard
Variable Estimat ed Error t-statistic P-value
Coefficient
C 43.7754 45.4683 0.9628 [.337]
LOT 0.0140 0.0090 1.5482 [.124]
LOT2 -0.0001 0.0000 -2.0030 [.047]
LOT3 0.0000 0.0000 1 .9763 [.050]
LOG(LOT) 0.3050 0.1498 2.0360 [.044L
CITY -0.0040 0.0089 -0.4555 [.649]
LOG(CITY) -0.0821 0.7042 -0.1 166 [.907]
HWY 0.0054 0.0317 0.1713 [.864
LOG(HWY) -0.0517 0.2018 -0.2561 [.798]
UTIL -0.73 52 0.6967 -1.0553 [.293]
PRISON 0.1677 0.5907 0.2840 [.777]
ZONED 0.9378 0.3356 2.7943 [.006]
PAVE 0.0735 0.2002 0.3673 [.714]
PLAT -0.0395 0.2635 -0.1500 .881]
H20 0.3756 0.2106 1.7835 [.077]
R_INT -1.4363 1.3983' -1.0271 [.306]
LOG(RLINT) 1 1.0358 12.0029 0.9194 [.359]
GSP 0.0464 0.0464 1.0015 [.318]
LOG(GSP) -10.5568 10.7825 -0.9791 [.329]
REGNLP 0.3296 0.3826 0.8613 [.390]
PERPUB 17.8651 54.1568 0.3299 [.742]
LOG(PERPUB) -22.7048 80.9337 -0.2805 [.779]
CC -0.3488 0.5446 -0.6405 [.523]
CC_P 0.6668 1.0572 0.6307 [.529]
SEED 0.1 160 0.4266 0.2720 [.786]

 

 

 

 

 

171

 

 

Table BIO (continued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Estimated Standard .
Variable . Error t-statlstic P-value

Coefﬁcient
THIN -0.0304 0.7379 -0.0412 [.967]
THIN_P 0.3805 1.4242 0.2671 [.790]
SEL -0.0008 0.9268 -0.0008 [.999]
SEL_P 1.0671 2.1115 0.5054 [.614]
REM -0.6612 0.8243 -0.8021 [.424L
REM LP 0.9791 1.4457 0.6773 [.499]
FED -0.8651 0.2925 -2.95 80 [.004
PERADJ -1.1 143 2.5304 -0.4404 [.660]
LOG(PERADJ) 1.7937 3.3955 0.5283 [.598]
CONIF ER -0.0969 0.1299 -0.7459 [.457]
HARDWOOD 0.1323 0.1672 0.7912 [.430]

 

172

 

 

TABLE B11. VACANT ADJACENT MODEL ARTIFICIAL REGRESSION
TESTS — QUADRATIC TERMS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Independent Tests
. Standard . ,
Parameter Estimate t-statistlc P-value
Error
LOT 0.0139 0.0090 1.5458 [.122]
CITY -0.0040 0.0089 -0.4555 [.649]
HWY 0.0054 0.0317 0.1713 [.864]
R_INT -1.4363 1.3983 -1.0271 [.304]
GSP 0.0464 0.0464 1.0015 [.317]
PERPUB 17.8651 54.1568 0.3299 [.741]
‘ PERADJ -1.1143 2.5304 -0.4404 [.660
Joint Tests
Chi-squared(7): 3.8632 P-value: 0.7954
F(7, 1299): 0.5519 P-value: 0.7967

 

 

 

 

 

 

TABLE B12. VACANT ADJACENT MODEL ARTIFICIAL REGRESSION
TESTS — LOGARITHMIC TERMS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Independent Tests
Parameter Estimate Standard t-statistic P-value
Error
LOT 0.3050 0.1498 2.0360 L042]
CITY -0.0821 0.7042 -0.1 166 [.907]
HWY -0.0517 0.2018 -0.2561 [.798]
R_INT 1 1.0358 12.0029 0.9194 [.358]
GSP -10.5568 10.7825 -0.9791 [.328]
PERPUB -22.7048 80.9337 -0.2805 [.779]
PERADJ 1.7937 3 .3955 0.5283 [.597]
Joint Tests
Chi-squared(7): 6. 1 186 P-value: 0.5260
F (7, 1299): 0.8741 P-value: 0.5274

 

 

 

 

 

173

 

 

 

TABLE 813. FINAL DEVELOPED ADJACENT MODEL ESTIMATION
RESULTS WITHOUT OUTLIERS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Log(Price)
Variable:
F-statistic: 4.90 D-W: 2.1609
R-squared: 0.5969 RESET: 0.0038
Adj. R-squared: 0.4752 LM Het. 0.0826
Test:

N: 134

Variable (13:61:32: 5'33? t-statistic P-value
C 7.4079 1.5820 4.68 .0001
HOUSE 0.0006 0.0002 2.76 .0068
LOT 0.01 12 0.0023 4.89 .0001
CITY -0.0005 0.0029 -0.17 .8687
HWY 0.0277 0.0122 2.26 .0260
GARAGE 0.3930 0.1373 - 2.86 .0051
PRISON -0.6925 0.6354 -1.09 .2786
ZONED 0.0109 0.2548 0.04 .9660
D -0.3545 0.1501 -2.36 .0202
MH -0.4296 0.1816 -2.37 .0200
PAVE 0.3680 0.1817 2.03 .0456
HOME 0.0208 0.1424 0.15 .8843
PLAT -0.0026 0.0046 -0.56 .5739
H20 0.2408 0.2105 1.14 .2555
R_INT 0.0120 0.0848 0.14 .8873
GSP 0.0056 0.0031 1.82 .0714
REGNLP -0.4316 0.4088 -1.06 .2937
PERPUB 1.8569 1.7344 1.07 .2870
CC -1.3270 1.6702 -0.79 .4289
CC_P 3.4349 3.8346 0.90 .3726
THIN 3.6728 1.5573 2.36 .0204
THIN_P -8.0453 3.6573 -2.20 .0302

 

174

 

 

 

Table 813 (continued).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Estimated
Variable St;:f::d t-statistic P-value
Coefﬁcient
SEL 1.7800 1.3140 1.35 .1787
SEL_P -3.2191 3.4248 -0.94 .3496
REM 2.1433 3.5398 0.61 .5463
REM_P -4.9792 9.4368 -0.53 .5990
FED 0.1336 0.2696 0.50 .6213
PERADJ -0.0235 0.2897 -0.08 .9354
CONIFER 0.0103 0.1480 0.07 .9447
HARDWOOD -0. 1061 0.1547 -0.69 .4944

 

175

 

 

TABLE B14. FINAL VACANT ADJACENT MODEL ESTIMATION
RESULTS WITHOUT OUTLIERS.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Dependent Log(Price)
Variable:
F-statistic: 1 1.24 . 2.0520
R-squared: 0.6691 RESET: 0.0253
Adj. R-squared: 0.6096 LM Het. 0.0352
N: 179

Variable axial: 8:52:33.“ t-statistic P-value
C 4.0736 1.2336 3.30 .0012
Log(LOT) 0.4623 0.0430 10.76 .0001
CITY 0.0067 0.0035 1.95 .0535
HWY -0.0040 0.0105 -0.38 .7056
PRISON -0.2616 0.6427 -0.41 .6846
ZONED 0.4173 0.1997 2.09 .0385
PAVE 0.1 145 0.1641 0.70 .4866
PLAT 0.0035 0.0050 0.69 .4930
H20 0.2200 0.1851 1.19 .2369
R_INT 0.0223 0.0664 0.34 .7372
GSP 0.0053 0.0021 2.55 .0120
REGNLP 0.5958 0.4846 1.23 .2210
PERPUB 3.9618 1.4757 2.68 .0081
CC -0.6655 0.4477 -1.49 .1394
CC_P 1.3616 0.8862 1.54 .1267
SEED 0.0730 0.3278 0.22 .8241
THIN 0.7073 0.7789 0.91 .3654
THIN_P -0.9945 1.6255 -0.61 .5417
SEL 0.5503 0.7037 0.78 .4355
SEL_P 0.1200 1.5822 0.08 .9397
REM -0.6264 0.6765 -0.93 .3560
REM!P 1.4843 1.2734 1.17 .2458
FED -0.4400 0.2257 -1.95 .0533
PERADJ 0.4579 0.201 1 2.28 .0243
CONIF ER -0.0541 0.0985 -0.55 .5836
HARDWOOD 0.0955 0.1209 0.79 .4309

 

I76