VALUATION OF LAKE RESOURCES THROUGH
TIMOTHY DALE FEATHER
DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
Though this dissertation is my contribution to the field, there are many who have
served as important advisors and supporters during its development who I would like to
The efforts of my graduate committee chair, Edward Malecki,
, Cesar Caviedes, Timothy Fik and Warren Veissman are very much
Peter Waylen and James Heaney,
who were important advisors during the
initial conceptualization of this effort, are duly recognized.
note of Dr.
I would like to make special
Waylen's support during the period of my comprehensive examinations.
John Crissey of the Orange County Property
Appraiser's Office was a critical
supporter of this research and provided the necessary property value data for the analysis.
William Hansen and Michael Krouse of the U.S.
Army Corps of Engineers'
Water Resources are recognized for directing financial and technical backing for this
commitment to this endeavor it would have been very difficult to conclude.
would like to thank my family,
especially my wife and parents for their support, for
When I might be finished?
TABLE OF CONTENTS
Economic Analysis in Water Resource Management
Statement of the Research Problem .
Issues in Environmental Appraisal.. .
General Approach and Data Needs
. . 5
* S * . 5
* . S S S 5
Economic Techniques in Environmental Analysis
Impact of Environmental Features on Land Rent .
General Applications in Property Value Analysis .
Water Resource Applications in Land Value Analysis
Application to Flood Control .. . .
Application to Irrigation Projects ......
Application to Water Resources in General
. . 6
. S 9
S S S S a S S 51
S S S S S S S S S S S S S S S S S S S 5 26
S S S S S S S S S S S S S S S S S S S 5 5 28
S S S S S S S S S S S S S S S S S S S 28
Assumption of Like Willingness-to-Pay among Households
I I I I)I1X
LIS'I' OF FIGURES ................... ................... vi
LIST OF 1~ABLES ................... ................... viii
Lake Characteristics Data
Filtering the Data
. MODEL CALIBRATION AND RESULTS
Chapter Overview . . . . . . . 49
Lakefront-Nonlakefront . . . . . . . 50
Lake Characteristics Impact . .. .. . .. .. 52
Land Rent Gradient
DISCUSSION OF RESULTS ......
Hypotheses Results . . .
Beneficiaries of Lake Resources .
Performance of Trophic State Index
Hedonic Valuation as a Planning Tool
. * . * . 57
* . . *. a 88
. S S S 88
CONCLUSIONS AND RECOMMENDATIONS
. a * a . a S a a 96
Model Profiles of Past Work
Lake Quality Model Database
. . a A.. 117
Distance Program in BASIC . . . . 119
Land Rent Gradient Model Database . . . 123
BOX-COX Procedure Program in SAS ........ ... 128
BIOGRAPHICAL SKETCH . . ....
. . 132
LIST OF FIGURES
Techniques to Evaluate Environmental Change
General Principles of Environmental Economics.
. . .. 10
Theoretical Impact of Desirable Lake on Normal
Land Rent Gradient
* * 13
Marginal Price and Marginal Implicit Price Curves .
Intersection of Household Demand and Hedonic Pricing
Willingness to Pay for Change in Lake Characteristic .
Assumption of Aggregate Market Demand Curve
.* .. 30
.t f f S S 36
Sample Property Appraiser's Record
Variability of Temperature and Dissolved Oxygen in Lakes
Stages of Lake Eutrophication .....
The Florida Trophic State Index
Filtering Property Appraiser's Database
Models Evaluated in This Research Effo:
. . . . 43
.* S S . 47
rt . S. . 50
Sale Price versus Unit Price for Lake Quality Analysis
Sale Price versus Lot Size for Lake Quality Analysis
. .. 55
* S 5 54
Lake Regions in Study Area .
. .. . . 60
Residual versus Independent Variables
. . 74
Observed and Estimated Relationship between Property
Value and Distance
. S S * . . 0 78
Theoretical Land Rent Surface for Long Lake
Theoretical Isovalue Lines for Long Lake
. S 8
Observed and Estimated Relationship between Property
Value and TSI
TSI Stimulated Changes in Value for Long Lake . . 84
Theoretical Land Rent Surface for Four Lake Regions ........ 85
Theoretical Isovalue Lines for Four Lake Regions .......... 86
LIST OF TABLES
Selected Data from Property Appraiser'
Comparison of Mean Property Values of Lakefront
and Nonlakefront Parcels.
Descriptive Variables Used in Lake Characteristics Analysis
Pearson Correlation Coefficients and Significance Statistics for
Variables in Lake Characteristics Analysis
. * a a 56
Parameter Estimates for Lake Quality Model
Descriptive Statistics of Variables Used in Land Rent
Pearson Correlation Coefficients and Significance Statistics for
Variables in Land Rent Gradient Analysis
. a a a . S a 66
BOX-COX Procedure Summary for Land Rent Gradient Model
Parameter Estimates for Lake Quality Model
Collinearity Diagnostics from SAS ....
Summary of Hypothesis Results
. a 72
Sa a a . .. a a 76
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
VALUATION OF LAKE RESOURCES THROUGH
Timothy Dale Feather
many water resource
which are difficult or
impossible to associate with monetary worth.
With many water resource management
components of a project is desirable.
To partially accommodate this analytical need,
environmental goods is capitalized in proximate land values.
the value of the
environmental resource is implicitly assigned through analysis of land values.
The hedonic pricing technique is used to estimate the demand for various lake
The successes and failures of past empirical applications of hedonic valuation
collected for Orange County, Florida.
The Orange County Property Appraiser's database
was accessed to obtain property characteristic data and numerous sources were sought
to obtain physical lake characteristic data.
The locational attributes of the lakes and
parcels were placed into a geographic information system to elicit the impact of lake
amenities on theoretically normal land values.
Three separate but associated empirical analyses were conducted.
property values were compared to nonlakefront values and lakefront properties were
found to be 41
Next, lake quality and size were introduced to the
Trophic state index and lake size were shown to be negatively and positively related to
The last phase of the analysis shows parcel distance from
supports the general theory that lake impacts on property values diminish with distance
from the lakeshore.
The empirical hedonic relationships were shown through a series of graphics to
methodologies can be used to evaluate some of the benefits attributed to lake resources
that water resource planners should consider in management decisions.
On the one hand
, when used as an input to production,
water resource benefits
are derived as production cost savings.
For example, the benefits or savings can accrue
from irrigating instead of transporting water in some agricultural production settings.
valuation of flood control benefits from a water resource is also relatively
straightforward. A dollar value can be computed to represent foregone flood damage.
On the other hand, valuation of quantifications of certain benefits can elude traditional
The quandary occurs because society places worth on certain features of water
resources that have no direct monetary value, such as aesthetic value and some recreation
In other words
, these are perceived benefits that are not bought or sold in a
Therefore the challenge to water management decision makers is to
justifiably place value on unpriced goods.
What is the value of a water resource such as a lake?
be spent on cleaning up a waterway?
How much money should
These questions must be answered by those who
development and application of this technique are the topics of this research effort.
in Water Resource Management
Economic analysis in water resource management has evolved significantly since
Army Corps of Engineers was initially ordered
to keep account of project
benefits and costs as mandated by the Rivers and Harbors Act of 1902.
Control Act of
the Act of
of a project
contingent upon benefits outweighing costs.
decisions was generally accepted, making b
The idea of economically justifying policy
ienefit-cost analysis a popular and necessary
agent for allocation of government monies to water projects.
In the late
1940s, representatives from several federal water resource agencies
created a guide for planners and managers to benefit-cost analysis, referred to as the
"Green Book" (U.S.
in the late
Interagency Committee on Water Resources 1950).
1950s and early
1960s individual valuation procedures, as well as the total
process, underwent close examination and formalization by water resource economists,
and policymakers (Eckstein 1958; Krutilla and Eckstein 1958; McKean 1958;
Hirshliefer et al.
1961; Maass et al.
Techniques were redefined and became
rooted in widely acceptable economic theory.
The "Green Book" was revised in 1958,
and other similar methodological "guides" were written (Sewell et al. 1962; Howe 1971).
The 1960s and 1970s, often referred to as the "Environmental Era"
federal government mandated environmental impact statements
for all proposed
with the passage of the National Environmental Policy Act of 1969.
the intrinsic and extrinsic values of the environment were proclaimed by the federal
to set and
administer policy to maximize that value.
such as the Water Pollution Control Act of 19
(ratified in 1978 and 1981, respectively), en
Other important legislative advancements,
'72 and Executive Orders 10244 and 12291
iphasized water quality value and the need
for efficient management of water resources.
The federal government delegated greater water management responsibility to
state and local governments during the 1980s.
To support management decisions at all
Water Resources Council (1983) published the latest set of
"principles and guidelines"
, which continued the long line of guidelines that
and assumptions involved are, as they have been for decades,
under constant scrutiny and development.
Statement of the Research Problem
The purpose of this study is to identify the empirical relationship between lake
water resource attributes and the land values of surrounding residential properties.
the study will
benefit estimation of lake water
resources via hedonic pricing.
Specific statistical hypotheses are developed in Chapter
further work and application
The intent of this study is to show that hedonic pricing can successfully
be applied to lake resources, and that a strong relationship between the presence of water
resources and property values exists and should be considered in the planning setting.
This study, while sensitive to the successes and failures of past studies,
valid theory exists and that modeling applications are feasible, resulting in a versatile tool
for water resource managers.
Additionally, it provides a research background and defines the
The first section is a general discussion of the methodological proclivity
toward analyzing environmental change.
The second section details the role of land
provide critical appraisal
applications of land value analysis for evaluation of the general environment and water
Last, a brief statement of justification is made for this study
based on the stimuli of previous work.
Issues in Environmental Appraisal
General Approach and Data Needs
Management of environmental resources is a continuous and dynamic challenge.
Social and physical segments of the environment are affected and must be understood and
Thus the need for interdisciplinary research is quite apparent.
proffered a six-point framework for environmental analysis:
(1) estimation of available
(2) expansion of the
"range of choice," (3) assessment of technologies, (4)
The practical range of choice is set by the culture and institutions that
permit, prohibit, or discourage a given choice.
The actual selection within these limits
depends on the manager's perception of the elements of the decision.
Environmental managers find that an important limiting factor in developing this
data collection efforts should
similar plea has been
extended into the regional paradigm of geography (Johnston 1983).
collection efforts involve the synthesis of systematically collect
relational databases and development of expert systems. Westcoat (
role of description and data collection in water resource
contributed to each of the following steps of Westcoat'
xd data by means of
1984) emphasizes the
e planning. Geographers have
Survey of the perceived range of choice among different
social groups; integrated surveys of resource distributions, environmental
processes, and social phenomena affected by the problem
configurations in the landscape,
and their effects on the domain of choice
Expansion of the range of choice through identification of
unforeseen or misperceived linkages among resource users,
waste flows, and institutions
Economic Techniaues in Environmental Analysis
Management of environmental change falls within the realm of economic theory
and a series of evaluation techniques have been created.
This section summarizes some
P 4 >4>4>4>4
CbO >4 >4
'5 V LA LA
It I ~1 V
federal government defines seven types of water resource outputs (U.S.
flood control, navigation,
multiple purposes and therefore affect humans through a combination of the defined
Techniques used to measure the impacts of changes in the environment relating
to humans are broadly categorized as market value based,
value based (Bentkover et al.
1979; Hufschmidt et al.
1983) and are
determine value. Their use is dependent upon the existence of a market for the pertinent
environmental good. If prices determined by the market do not exist, survey-based
techniques can be used to create hypothetical markets.
Carefully worded questionnaires
are developed that ask respondents what they would be "willing-to-pay" for the set of
quantities of goods.
The last group of techniques-surrogate value based--use priced
complementary goods to determine the value of a good that is not defined in the market.
The selection of the appropriate technique is generally based upon project purpose,
shown in the third tier of Figure II-1.
environmental valuation and allocation analysis in the recent past.
above techniques, methodologies such as linear programming, option
In addition to the
uization models, and
trade-off analysis (Cohon 1978; Goodman 1984) are used to provide the best solution,
The gross economic benefit of a good is estimated as the area under the
demand curve and is designated by area (abc) in Figure II-2 (top).
to estimate demand
Assuming a price (or cost) P'
, the net benefit (or consumer surplus) of the good is
equivalent to the area (abc) under the demand curve less the area (decb) representing
If the quality of an environmental good is altered,
this will affect demand.
net benefit (or disbenefit) caused by a change in the environmental good is derived by
swimming is enhanced through cleaner lake water, the demand curve for swimming at
the lake will likely shift to the right.
shown in Figure II-2 (bottom), is equ
The economic gain from this increased demand,
livalent to the consumer surplus of the altered state
(abe) less the consumer surplus of the unaltered state (cde) represented by the shaded
Impact of Environmental Features on Land Rent
Following the suggestions for more data sources and development of economic
techniques for environmental analysis, research centered around the response of land
rents to the environment has taken place.
This section introduces the theory, and the
Natural and man-made features of the landscape affect household utility and are
considered during residential choice decisions.
If demand for the goods and services
Initial Consumer Surplus
of land values across the landscape is studied by many disciplines, including economics,
and regional science.
Most researchers cite Von Thunen's (1821) work on
agricultural land values as the seminal work in land rent theory. He postulates that land
rents decrease with distance from the central business district (CBD), ceteris paribus.
This gradient was the result of market compensation for increased transportation costs.
Early advances in urban land rent theory centered on a central business district were
as determinants of land rent.
Several formalized mathematical explanations
of urban land rent structure appeared in the 1960s.
or firm utility through a set of bid-rent functions.
* Alonso (1964) maximizes household
Muth (1969) discusses urban land rent
assumptions found in standard urban land rent models (e.g.
monocentric city and a
Empirical verification using the urban landscape of Chicago was
Richardson (1976) presents a comprehensive overview of the many extensions
to this formal mathematical approach to land value modeling.
Impacts of environmental features have been introduced into the development of
displaying uniform impact among all households is categorized as a
"public good" for
as a clean
has a spatially
nonuniform influence on households and is considered a special type of public good: an
an imaginary direct path from the central business district will inflate the land rents in
the immediate vicinity of the lake in a manner similar to what is shown in Figure II-3.
A related econometric technique, referred to as
external effects of the environment on land value to estimate value for the environmental
The value of the environmental good at hand (e.g., lake,
river, forest, air, sound)
differential due to the presence of the environmental good is assumed to be the surrogate
value of that environmental good.
Theoretical background and econometric explanation
are presented by Griliches (1971) and Rosen (1974).
Property value impacts and hedonic
valuation are the central methodological theories of the present research.
General Applications in Property Value Analysis
The value of many types of environmental goods has been measured through
The majority of studies have focused on
the value of air quality.
Ridker and Henning (1967) provided the first application of evaluating the land value-air
They reveal a positive relationship between the two:
as air quality
improves, so do property values.
Some researchers found this relationship to be just
marginally evident (Smith and Deyak 1975; Milliman and Sipe 1979).
A close look at
these studies reveals data definition problems or limited variability in air quality
are many other successful applications of the property value technique to air quality
many of which are summarized by Freeman (1979).
Harrison and Rubinfeld'
(1978a) analysis of air quality in Boston pays particular
Effect of Lake
Distance from Central Business District
Adapted from Thrall (1987)
Theoretical Impact of Desirable Lake on
Normal Land Rent Gradient
and socioeconomic class variables causes errors in the dependent
variable of between 20 and 30 percent.
benefits as compared with the aggregated "basic" equation.
Benefits from air aualitv improvements across income erouos are examined in
The variable indicated riverside tracts to be of significantly higher value than
those tracts not along the Charles,
thereby implying a positive demand for the amenities
of the river.
Havlicek et al. (1971) evaluate negative external effects on nearby land values of
waste disposal sites examined near sanitary landfills in Fort Wayne, Indiana.
considered are size of house and lot, number of bathrooms and bedrooms, age of house,
, year of sale, sale price.
Distance from the disposal site and
degrees (angle) from the prevailing wind describe the relationship between the parcel and
the disposal site.
Each degree from the direction of the prevailing wind is associated
with a $10.30 increase and a $0.61 increase in value is found for each foot away from
the disposal site.
Since the areal units are not reported,
it is unclear if these values are
in dollars per acre, dollars per front-foot, or some other unit.
Effects of hazardous waste dump sites have been the subject of many studies.
Payne et al. (1987) conduct an analysis of the property value response to proximity to
a radioactive waste site.
Awareness of the site was intensified by heightened publicity
in the region and resulted in a decrease in the value of older homes within a two-block
of the disposal
accidents are reviewed in Hageman (1981).
Using the Delphi technique, a panel of
experts reveal many cases where residents were compensated for decreased
as a result
documentation and evidence are convincing,
due to the nonempirical method of research
techniques as well.
Rubin and Yezer (1987),
who evaluated natural hazards in general,
report the land value response to the hazard to be significantly less in the case of an
expected disaster compared with an unexpected disaster.
Effects of flooding hazards are
Water Resource Applications in Land Value Analysis
Land value analysis as applied to water resource evaluation has been subject to
Some attention has been through simple description or recognition that
resource-impacted land value differentials do in fact exist.
Other applications have been
more aggressive empirically
This section summarizes these applications.
Because the empirical results of
past hedonic pricing studies are critical to the development of the models in the present
, model profiles were created that list the dependent and independent variables,
the functional form of the mathematical equation,
goodness of fit statistics.
the regression parameter estimates, and
These profiles are found in Appendix A and are cataloged
alphabetically by author.
Application to Flood Control
Flood control projects, in broad terms,
provide benefits related
land use intensification
, and location as outlined by the U.
Each of these categories of benefits can be measured through land
flood-prone land is an
obvious example of a land
differential caused by water resources.
Generally a flooding hazard is expected to be
found the rate of land value recovery to hinge upon the magnitude and frequency of
This confirms the findings of Rubin and Yezer's analysis of natural hazards
The federal government subsidizes residents of qualified floodplains through the
a price differential
between floodplain and nonfloodplain lands.
Thunberg and Shabman (1990) derived a
disruptions-these findings were developed while controlling for flood insurance impacts.
In evaluating a potential flood control project for the Passaic River in New Jersey, it was
discovered that nonfloodplain residential lands possess an average market value 30-40
percent higher than in the floodplain (U.S.
Army Corps of Engineers 1987).
These benefits are considered in the assessment of project-related
regional economic impacts (Apogee Research et al.
Antle (1977) presents a case study of the Chester Creek Basin in Pennsylvania.
The impact of flooding on average land value was estimated to be approximately $5,100
per floodplain parcel.
Another important result was the identification of other important
variables as determinants of land value.
Property size, township, number of floors, and
An apparent shortcoming of this study,
which is common to many
analyses of this sort, is data availability.
Various sources of property value data were
As part of a closer
look at the impact of flooding on
property net of flood
insurance, Donnelly (1989) finds an average floodplain parcel to be valued $6,000 less
than property outside the floodplain.
The statistical model appears to be very sound,
all parameter estimates are statistically significant and the r-square value is strong at
The author further explains how adjustments are made to some of the independent
variables to control for multicollinearity, though no empirical justification is made for
the linear functional form of the final model.
Annual flood insurance payments are
analyzed at a 10 percent interest rate for the average property-the resultant value being
The difference between the $6,000 and $3,500 is the residual
negative impact of floodplain property.
Donnelly refers to this as a "hassle premium,"
which echoes the findings of Thunberg and Shabman (1990) discussed above.
Application to Irrigation Projects
Milliman (1959) discusses the theoretical possibilities of measuring the primary
benefits of irrigation through increased agricultural land values.
require estimation of net returns from the crops being irrigated,
output factors, factor costs, and coefficients of production for future seasons.
assumptions as the "existing" methods, and that accuracy of the results may be adversely
influenced by data problems,
such as inaccurate land value and land
concludes that choice of the appropriate technique would have to be made on a case-by-
one for reservoir land and the other for nonreservoir land.
between the equations summed over all tracts of land near a reservoir are considered the
value enhancement attributed
to the reservoir.
reservoir presence enhances property values.
Generally the statistical results of the
model are encouraging.
structure of the Knetsch model
appears to have a few shortcomings.
the distance variable in the reservoir model
, each variable is stated as having a linear
influence on land values;
this does not allow for nonlinearities
, or "leveling off," of
influence on the dependent variable. Another problem lies in the use of two models to
estimate the influence of a reservoir on land values. A more desirable approach would
have been to use one function that allows the inclusion of both reservoir and nonreservoir
the reservoir model most likely possesses multicollinearity between
accuracy of the parameter estimates.
David (1968) expands on the work of Knetsch in a study of Wisconsin lakes.
Improvements in the independent variables employed include knowledge of water quality
quality parameters are based
Average lakefront slope is included as a measure of topography
and ease of access,
and the presence of swamp and other lakes are also
The "value-of-improvements" variable accounted for approximately
percent of the variance in
David's objective is to focus on the
"access to lake"
are found to be statistically significant.
David's study suffers from
unavailable, resulting in numerical aggregations and simplifying assumptions.
(1971) suggests important factors in lakeshore property appraisal are lake type, size,
nutrient content, depth,
The value of riparian rights might also
, as discussed by Holden (1973).
The physical characteristics of a lake probably have little affect on property
prices relative to other variables such as "value-of-improvements."
of lake characteristics
To identify the effect
, other variables that are capitalized into property values must be
identified and controlled for through inclusion in the model.
The omission of "value of
improvements" from the model appears to only lower the r-squared value (David 1968).
A formal econometric approach is used by Brown and Pollakowski (1977) in the
valuation of shoreline property.
They estimate implicit price functions via hedonic price
for waterfront-housing services.
Variables used in the model emphasize
housing structural characteristics of housing.
water is a distance to water variable (or setback).
The only parameter directly related to
By assuming identical utility functions,
the marginal implicit price function is used as a marginal willingness to pay curve.
household benefits with the general public's utility for open space.
Pollakowski model contained no water characteristic variables. Thou
of the study was valuation of open space (indicated by setback), w
The Brown and
igh the main intent
variables would have added considerable insight into the open space values.
Dornbush and Barranger (1973) perform a nationwide property value analysis and
find that abatement of pollution in all waters to a level "not inhibiting to desirable life
forms or practical users and which are aesthetically agreeable" would increase the capital
values of aggregate property value by approximately $1.3 billion.
They sampled twelve
improvement from 1960 to
A regression equation was developed in each area
using property value change as the dependent variable and independent variables of lot
influential to property values (e.g.,
distance to local park, distance to school,
to shopping center).
from a lack of
Variables, such as "distance to State and Commercial Street intersection," are
included that are unduly restrictive in geographic application.
Though recognition of a
transportation hub's influence on land values is desirable, the utilization of such variables
(in that form) severely limits the model's external validity.
A somewhat separate component of the same study is an examination of the public
perception of water quality.
Residential property owners were interviewed as to how
Epp and Al-Ani (1979) also evaluated the relationship between perceived and
technical water quality.
Perceived water quality was arranged through a survey asking
as to whether
recreational or aesthetic use of the water resource.
The parameter estimate indicated
negative perceptions of water quality were associated with lower property values.
biochemical oxygen demand,
nitrate, and phosphorous.
The only technical water quality
to be an
transformed into categories of 5.5 or lower and greater than 5.5.
The more acidic (5.5
lower property value.
perceived and technical water quality variables had a consistent impact on property value.
The value of urban water parks is measured by Darling (1973).
Two methods of
valuation are compared and contrasted: property value method and an interview method.
The study constitutes a respectable comparison of the two methods with actual empirical
, which is oftentimes absent from this type of analysis.
author provides valuable insights for further research.
The interview method
, which is often referred to as contingent valuation,
on survey data to develop a demand curve
for the water park.
The property value
method employed generally followed the approaches described above. Variables used
in the property value model are property value, improvements, size, crime, neighborhood
distance to water, and an inflation variable.
The general conclusion is that urban
However, results did not support this.
In two of the three areas analyzed, the property
value method produces a much higher value then the interview method.
In the third
area, the opposite is the case.
Questionnaire bias is a likely source of error.
questionnaire design is vital in estimating demand of a good.
Smith and Desvouges
(1986) provide a comprehensive discussion and application of interviewing techniques in
water resource valuation.
Data concerning property value transactions were difficult to obtain in Darling
(1973); thus a mix of assessed value and selling price was necessary to calibrate the
property value model.
This inconsistency would likely introduce additional errors in the
results and may have contributed to the misalignment of the demand estimates for the two
The area of greatest concern regarding Darling's study is variable selection (or
lack thereof). First, the study employs use of assessed value of property versus actual
market transactions. Actual sales price is certainly the metric of choice, as assessed
valuation techniques often inhibit inclusion of unique parcel characteristics and do not
inclusion of a technical measurement of water
quality would have been useful for water resource management application (which is
often based upon technical water quality goals).
Addressing the allocation of the Kissimmee River Basin in Florida among user
groups, Reynolds et al. (1973) measure the value of the river to proximate landowners.
Two analyses are conducted.
The first, in a similar manner to Darling (1973),
is a survey
the value of
When asked what they felt the value would be if the lake were drained,
price dropped 48 percent.
The authors attribute the apparent difference between the
results of the two analyses to the fact that the second analysis includes structures on the
property that "hide" the influence of water on the land value.
Another comparison of water resource valuation techniques is provided by d'Arge
technique with a contingent valuation approach.
They also interview realtors, in a third
tier of the analysis, to gather another perspective to the valuation question.
focus of this study was to evaluate the differences in demand around two glacial lakes
called East Okoboji and West Okoboji in Iowa.
The water quality in West Okoboji is
seemingly pure opportunity to compare the demand for higher quality lake attributes.
Estimates from the survey realtors attribute 23 percent of house value to water
while the hedonic price attributes 21
These two approaches were
expected to be close to one another, and the results support this hypothesis.
The authors also hypothesize that buyers are able to adjust the amount of water
quality they want as part of their bundle of goods by simply adjusting location,
causes the rent gradient for water quality to be concave downward.
, coupled with
the "thin" market in the Okoboji region,
causes the willingness-to-pay estimates for water
quality to be exceeded by the hedonic price estimates.
This hypothesis is supported, as
willingness-to-pay estimate that Darling (1973) found when comparing it with a hedonic
The purpose of a study by Rich and Moffitt (1982) was to determine a portion of
the regional benefits associated with a water pollution control program through hedonic
Regional benefits are calculated to be $600,000 for the 26.5 square miles that
were defined as the study area.
hedonic regression analysis, w
The total regional benefit is based on the results of the
which assigns $37 per acre for riparian land and $31
categorical variable that serves as the operational determinant of the $600,000 abatement
significant, since they have a limited number of observations (N
It may have
been worthwhile to replace the binary variable with a continuous distance to water body
, aligned with the discussion surrounding d'Arge and Shogren (1989),
technical measurement of water quality might have enhanced the engineering application
of the results.
Falcke (1982) closely follows the econometric theory and procedure of hedonic
pricing presented by Rosen (1974) to measure water resource benefits and also follows
the work of Dornbush and Barranger (1973) in derivation of a perceived water quality
Survey data show that laypersons and technical experts often have differing
conceptions of the conditions of a water body; that is,
in an extreme case, residents felt
equations are calibrated,
with the percent of property price change as the dependent
Each equation uses distance from the water body and perceived water quality
change as independent variables,
as well as a subset of the following variables:
, distance to shopping, location on busy street, location on corner lot, previous
property value, lot size, distance to new highway,
distance to nearest highway access,
distance to environmental nuisance, distance to other new facilities like a bridge, boat-
launching area, or country club.
The "distance-to-water-body" parameter estimate for
each site is regressed against perceived water quality change,
access, and region indicator.
water body type, public
This statistically meshed the site-specific equations into a
Some applications of hedonic valuation focus on the damaging impact of water
resources on property values.
The impact of flooding was discussed above.
Chetri and Hite (1990) examine the negative effects of reservoir regulation schedule on
the needs for hydropower
variability in reservoir stage,
which impacted the utility of waterfront property owners.
Each one foot decrease in stage from normal pool caused about $5,434 decreases in sales
price per acre.
Young and Teti
(1984) examined the impact of degraded
property values in the St. Albons Bay region of Vermont.
In comparison of two water
resource sites that provided a marked differential in
the lower water
quality caused approximately 20 percent lower property value.
Young and Teti did not
higher property value, Jack Faucett Associates (1991) related beach erosion to decreases
in property value.
This decrease in property value was felt not only by oceanfront
residents, but also residents throughout the community (but to a lesser extreme).
analysis was used to justify erosion control measures in oceanfront communities.
hedonic pricing using property values, the following issues are apparent:
* In nearly each study,
quality of data was a
* Applications of water resource valuation are
* Studies that did examine water resources
, for the most part, ignored hedonic
have not explicitly used water quality or characteristic data in the
have not attempted the development of a model that could be
applied in areas other then the study site.
rarely consider distance to water as a continuous variable.
The general exception to the above observations is the Falcke (1982) study.
study is designed to overcome the common data problems while working within the
examining cross-sectional data in an attempt to control for the shifting housing markets
and other exogenous forces that alter land values over time.
It also concentrates on a
single geographic area rather than sites throughout the
A considerable effort is made to compile the database for statistical analysis, both
in terms of the variables chosen and the data gathered to represent these variables.
, data, and results of this analysis are presented in the following chapters.
This chapter develops the formal model used in assigning hedonic value to water
as are the
assumptions required under the model.
theoretical principles supporting hedonic
The first subsection provides an overview of the
Pricing, based upon Freeman (1979) and Falcke
The second subsection discusses issues surrounding the calibration of the model.
estimate the price of environmental goods.
In particular, this study applies the technique
to the estimation of demand for various lake characteristics and permits positive and
negative benefits related to changes in these characteristics to be calculated.
The most empowering assumption of the hedonic technique is that the good being
is realized by the consumer and is part of
the bundle of
Two further assumptions about the housing market are made:
(1) a single
housing market dictates housing choice in the study area; and (2) the housing market is
in equilibrium (buyers and sellers are optimally satisfied with each transaction in which
characteristic at hand is derived.
equilibrium price of the good being measured.
The hedonic (implicit) price function can be stated mathematically as
land value at site i with lake
set of site characteristics at site i
with lake characteristics j
set of location characteristics
at site i with lake characteristics j
level of lake characteristics j
The form of this function varies but is generally multivariate.
Thrall (1988) provides an
appraisal of theoretical issues pertaining to land rent function development.
Box and Cox
(1964) make significant progress on proper functional form assignment, and Halvorsen
Value of a desirable water characteristic, such as water quality,
increases with the
level of quality up to a point where the benefits of increased water quality begin to "tail
off" as shown in Figure rI-1 (top).
Thus the hedonic price function (of a "desirable"
good) generally increases at a decreasing rate reflecting marginally diminishing utility.
Differentiating the calibrated hedonic price
with respect to
characteristics defines the marginal implicit price,
Figure l-1 shows the hedonic (imDlicit) price (too) and marginal implicit trice (bottom)
The second step is to derive willingness to pay curves, or inverse demand curves,
for lake characteristics.
Individual households or groups of households possess different
tastes and preferences for the good W.
The marginal implicit price denotes the aggregate
market value assigned to an additional unit of W; it does not directly account for
individual household demand for W. A sing
which, given its socioeconomic makeup, i:
estimate demand for that particular group.
;le observation for each household i is made,
s an insufficient number of observations to
Grouping of households by income class
(following Harrison and Rubinfeld 1978) provides an aggregated demand estimate by
group/individual household type.
A variety of possibilities exist as to the shape and empirical nature of willingness-
to-pay curves (Freeman 1974,
1979; Rosen 1974; Bartick 1988).
It is assumed here that
the lake characteristics are independent of a household's willingness to pay.
that lake characteristics are considered exogenous to their implicit price and can be
estimated without regard to a supply-side function (as assumed by Harrison and Rubinfeld
Thus the willingness-to-pay curve can be estimated by the function below.
= P(S.,Li,W, ,H)
willingness to pa:
P = willingness-to-pay function
Si = site characteristics at site i
L == location characteristics at site i
Wg = observed marginal implicit expenditure on lake characteristic
j by household/group i
Hi = set of household characteristics for household/group i
and the marginal implicit price function defines the equilibrium state for the household
in terms of lake characteristics (Figure III-2).
Households will buy quantities of W at
the aggregate marginal implicit price, moving along their willingness-to-pay curve to the
point where the two curves intersect.
This is the level of lake resource the household
will choose to obtain.
The benefits received by household/group i through a nonmarginal change in lake
characteristics from W1 to W2 are the integral of the willingness-to-pay function from W,
The aggregate benefits are the sum of this integral for each household.
economic benefit due
to the change in lake
characteristics for household/group i
initial lake characteristic level
lake characteristic level after change
willingness to pay for household/group i
number of households/groups in region
The economic benefit is depicted graphically as the area under the demand curve between
Assumption of Like Willingness-to-Pay among Households
Due to the complexity of revealing individual/group demand for various levels of
lake resources. an alternative annroach will be taken in this analvsis.
represents all individual households'
willingness to pay.
Falcke (1982) uses a form of
this assumption to assume that households at equal distances from the water resources
possess like willingness-to-pay functions.
Freeman (1979) recommends this method as
an approximation of benefits.
The assumption is graphically depicted in Figure III-4.
Assume that V(W), and
Pg represent the marginal implicit price and household willingness-to-pay functions for
state of water resource,
enhanced to W,, the actual household benefit would be represented by the area (abcd).
The assumption of this analysis,
given that P(W) will not be formally defined,
V(W), represents household willingness to pay.
Consequently the household benefit
resulting from a move from
potential errors may occur.
Wi to W, would be represented by the area abedd).
, if the demand is more responsive to price than the
Alternatively if the demand is less responsive to price following P,, the benefits will be
underestimates of benefits will cancel.
Data requirements for a hedonic valuation study are crucial to successful and
variable-some form of property value-can cause harmful bias in parameter estimates.
The vast requirements
for data oftentimes cause researchers
to shy away from
But in recent times
, development of geographic information systems (GIS) and
access to large databases, such as property assessors'
, allow integrated access to
a comprehensive vector of parcel level attributes.
Though the data used in the present research were not rooted in a GIS,
amount of effort went into gathering the best data available.
In fact, a GIS employing
of the same data used in
been developed by the agency
supplying the property value data.
This chapter describes the data used to test the model presented in the previous
The source, necessary formatting,
and filtering of the data are presented.
intent of this chapter is to clearly indicate the evolution of the data and to conclude with
the data set required to calibrate the models that are presented in the following chapter.
Chief Data Sources
Residential development has been quite significant, especially in the Orlando area, again
rendering a desirable sample of transacted residential land values.
Orange County officials are especially enthusiastic about this study and have
provided important insights and data.
Both the director of the Orange County Property
Appraiser's Office and the chairman of the Orange County Commissioner's Office were
contacted about data needs for this study and, subsequently,
collaborated in providing the
contribution to this study is certainly warranted,
the main point here is that identification
tremendous advantage to the research effort.
, because the hedonic approach is
considered an ex post justification for choice of study site.
Property Assessor's Database
The Orange County
Appraiser's database is stored on
Each of the 253.000 records in the database is
1,641 columns wide.
database-transactions are made and recorded daily.
shown in Figure IV-1
An example of a single record is
. These hard-copy records, kept on file at the County Appraiser's
Office, are made available to the general
This research effort requires data
describing the locational structures and economic features of property,
as noted in Figure
The data shown in Figure IV-1 are categorized in Table IV-1 according to these
Ir~~~n Irrn~ I
I,.; i i
Selected Data from Property Appraiser's Databases
Historical sale values
Historical sale dates
Assessed property value
Lake Characteristics Data
Associated with the property value data are characteristics of the environmental
resource at hand.
This study is aimed at describing the implicit value of lake resources;
therefore data describing lake resources are needed.
7.748 lakes in Florida
(Shafer et al.
In terms of data describing these lakes, many have only locational
(latitude and longitude) and surface area measurements.
In fact, only 3,261 of the lakes
Water quality data for 788 of Florida's lakes are compiled by Huber et al.
Another substantive source of water quality data is the annual water resource
assessment required of each state by the Federal Environmental Protection Agency (Hand
101 lakes in Orange County exist, ranging from simply a size
Heaney (1988) describes the difficulty in using single measures of water
quality to analyze water management effectiveness.
case in Florida,
Dierberg et al.
(1988) describes the
where lake management practices have had an impact on only
7 of 43
This small impact is attributed partially to ineffective lake management strategies,
but the main question raised concerns technical water quality measurement practices.
Looking at the seasonal variability raises concern as to the applicability of annual
averaging of water quality parameters.
Stratification and mixing cause seasonal trends
in temperature and dissolved oxygen (a common measure of water quality),
temporally but also by depth of sample,
as shown in Figure IV
Not only are there problems with the technical measurement of water quality
a layperson' s
perception of water quality adds another dimension of complexity (Falcke
1982; Dombush and Barranger
Water resource characteristics being purchased
as part of the "bundle of goods" constituting property value should be measured in terms
that laypersons can understand.
a change in concentration of dissolved
oxygen in a lake may not be recognized in terms of milligrams per liter (the scientific
unit measure of dissolved oxygen) by a layperson; but if it causes a change in the amount
of algae and weeds in the lake, this can easily be recognized.
suggest easily perceivable water quality characteristics as color, odor
Lant and Mullens (1991)
5 10 0 5 10 0 5 10
Dissolved Oxygen, g/m3
Tchobonoglous and Schroeder (1985)
Variability of Temperature and Dissolved Oxygen in Lakes
and the three main phases, or states, are shown in Figure IV-3.
oligotrophic state of eutrophication is the youngest.
It can be thought of as relatively
clean water, but so clean that it cannot support the threshold of food and nutrients to
sustain large populations of life forms.
natural growth and decay to take place that will cause food production to increase, and
the trophic state will move to mesotrophic.
A mesotrophic lake will support the largest
level of life in terms of oonulation and diversity: an oDtimal balance of nutrients and life
* balance between food
* high population
* low food/population
* nutrients dominate
Stages of Lake Eutrophication
causes dominance of algae and plant growth.
The highly variable oxygen availability
characterized by the eutrophic state can sometimes cause fish kills.
There are many facets of eutrophication that are easily perceived by laymen,
mainly because the eutrophic trend begins with a generally clean appearance and evolves
oligotrophic lake, and Lake Okeechobee,
to the common man.
a highly eutrophic lake, are certainly evident
Changes in the perceived water quality attributes suggested
above-color. aleae. odor. temnerature-can be detected by laymen and are indicators
Trophic state indices (TSI) are used to enumerate the level of eutrophication in
These indices typically range from 0 to 100, where 0 indicates very good water
100 very poor water quality.
TSIs are designed to reflect a doubling or
halving of algae biomass with each 10-unit change in index (Carlson 1977).
accepted as a representative indicator of lake trophic state.
Consequently water quality
management programs are often evaluated in terms of TSI (Dierberg et al. 1988).
Huber et al. (1982) provide an in-depth analysis of various constructs of TSI. A
combination of physical, chemical,
and biological parameters meshed through a statistical
weighting procedure laced with assumptions is the typical means of TSI development.
Huber et al.
evaluated the more popular indices,
paying particular attention to their
statistical validity in application to Florida lakes.
The recognized TSI configuration for
Florida lakes is provided in Figure IV-4.
justification are as follows:
first, lake eutrophication can be perceived by the general
combination of several technical water quality parameters that limit (but definitely does
not eliminate) the metric's volatility tied to sampling patterns; third, TSIs are used by the
scientific community in evaluating water quality management programs (e.g., Shannon
and Brezonik 1972).
Therefore the suggestion of Brezonik (1976) that
TSI is helpful in conveying lake quality information to the non- and
- LIMITED LAKES (TN/TP
= 1/3 [TSI (chl a) +
14.4 In chl a,
- 30.0 In SD,
= 23.6 In TP
-LIMITED LAKES (TN/TP
= 1/3 [TSI (chl a) +
= 59.6 +
21.5 In TN,
- BALANCED LAKES (10
1/3[TSI(chl a) +
+ 0.5(TSI(TP) +
= 56 +
19.8 In TN,
= 18.6 In TP
= 0.65TSI(Florida) +
Trophic State Index
Total Phosphorus (unfiltered)
Huber et al.
Filtering the Data
The property appraiser's database yields sales transactions and parcel description
associated lake characteristic data are not as comprehensive.
Therefore the sample
selected for analysis is controlled by availability of water quality data.
Three separate but tangential analyses are conducted in this research effort, each
using different data
follows in the next
Detailed documentation of these analyses is provided in the
Specific definition of the data set employed for these analyses also
chapter. Prior to the portioning design of the data set for the three
analyses, three levels of filtering of the property appraiser's database are conducted,
is diagramed in Figure IV-5. The first "global" filter isolates single-family residential
parcels from the population of parcels. This follows nearly all the studies directed at
hedonic valuation of water resources. Nonresidential users of land will in most cases
place a different value on land, depending on their business, industrial output, or other
, property tax schedules typically vary by land use type.
in land use purpose and resultant tax responsibility are capitalized into land value, and
this causes market value segmentation.
To control for this,
single family residential
properties are selected for the analysis.
Another reason for disaggregating parcels by land use type is that the exogenous
impact of water resources varies.
Residential occupants benefit from the recreation and
aesthetic opportunities that a water resource provides (assuming it is attractive and not
SOLD IN 1983
Filtering Property Appraiser's Database
Comparison of the value of a variety of water-related services would
be statistically unwieldy and is not the intent of this work; therefore only single-family
residential properties are analyzed.
Thus it should be noted that the application of the
results of this work should be applied to water resources surrounded by residential land
isolates properties sold in 1983.
, let us address the issue of selection of a single
Only vacant parcels are considered because the focus of the research is the
locational relationship with the lake resource.
Inclusion of developed parcels would
introduce variance in market price of property that is not needed.
Many other studies
include structural characteristics that typically are easily explained in terms of square
footage and age.
, inflates the explanatory power of the calibrated model.
By using only vacant parcels,
the present effort and resultant model will explain the
locational value of the lake resources.
A last note on the data is that only qualified market transactions are considered.
"qualified" sale is one that reflects on actual market transaction.
On the other hand,
a father may sell his daughter a parcel of land for $100.
is actually a gift, but the transaction is recorded in the property appraiser's database.
as an "unqualified" sale.
Actual sales price is used instead of assessed value.
(1976) for a discussion of these issues.
See Berry and Bednary
Though the goal of most assessment techniques
is to reflect market values
, they are sometimes biased to meet political goals.
uses actual qualified market sales to avoid this bias and get a true representation of what
the market bears for a particular parcel.
MODEL CALIBRATION AND RESULTS
The influence of lake resources is looked at incrementally in this chapter from a
very simple perspective to a multidimensional perspective,
as shown in Figure
Thus three hypotheses are tested with as many models:
Land value of lakefront property is greater than nonlakefront
The effect of lake characteristics
realized in land values.
Water resource related impact on land value will diminish with
distance from the water source.
First, the question of whether the present lake resources influence land value-yes
nonlakefront property values.
An affirmative answer to this question moves us to the
Is lake quality recognized in land values?
If so, lakes of varying
attributes are correlated to adjacent property values.
The final dimension builds upon the previous two while adding space or location.
influencing components of the urban landscape are considered (e.g.,
distance to central
The calibration of this third model constitutes the hedonic value
Non Inke front
Models Evaluated in This Research Effort
The three models
2. Lake characteristics
3. Lake influence land rent gradient
are presented in this chapter individually.
As mentioned in the previous chapter
model works from a separate data set.
as are the statistical arguments, and fi
The definition of the respective data set is given
final model results are presented for each.
the presence of lake resources exists at all must be settled.
It is hypothesized that a
relationship has been proven in the past (e.g.
1973) and is shown to hold true in this study.
means of the
lakefront parcels versus parcels not on lakefront. From the global filter presented in the
previous chapter, there are 3,241 single-family residential, vacant parcels sold in 1983.
A very convenient "special use" code in the property appraisers database allows specific
identification of lakefront property.
Of the 3,241 parcels, 174 are lakefront.
Analysis System's (SAS) PROC TEST procedure is used to test the means of the two
The results shown in Table V-l indicate a strong difference in property value
between the two groups.
The near lakefront parcel selling price is $26,085 compared
with $15,406 for nonlakefront property.
This difference, as shown by the t-statistic,
It should be noted that the t-statistic reported in Table V-l is an approximation
the case where the variances of the two groups are different.
Thus the t-statistic is approximated as:
- X / [(SF/n,
the samnle means of samnla
in thit per lalcefrnt and
In the case of (statistically) equal variances between the two groups, a pooled variance
term is used:
= pooled variance of the two groups
All other variables were defined above.
(1) yields a more conservative t-
statistic (more difficult to reject null hypotheses),
but strong significance is still shown.
Nonlakefront property is valued at about 59 percent of lakefront property.
(1964) found nonreservoir land to be 54 percent the value of reservoir land, a
Chetri and Hite (1990) indicated this value to be about 40 percent. The d
between lakefront and nonlakefront found in the present effort is probably conservative
because the presence of a lake influences more than lakefront parcels only.
theory describes (Thrall 1982),
As land rent
and is shown later in this report, a distance-decay effect
occurs over the landscape; so parcels not on the lakefront, but very near it, will receive
some inflationary influence by a lake's presence.
This distance-decay influence is not
realized in the statistics of Table V-1 because of the way the samples are defined-either
lakefront or not.
Therefore it is expected that the mean nonlakefront value, ($15,406)
harbors some of the distance decay impact that, in turn,
biases the nonlakefront mean
Lake Characteristics Imoact
Comparison of Mean Property Values of
Lakefront and Nonlakefront Parcels
Shogren 1989) illustrates a significant increase in demand for cleaner lake water quality.
This increased demand was found to be capitalized in land values.
For this portion of the analysis, the question of whether lake characteristics are
revealed in property values is examined. The lake characteristics examined are TSI and
lake size and are regressed on lakefront parcels only. Most of the 174 observations of
lakefront parcels (see Figure IV-5) were next to
lakes for which no
Therefore the original 91,314 single-family residential observations (shown
in Figure IV-5) were accessed to obtain observations for 1982 and 1984.
The final data
set used in this analysis contains 45 observations for lakefront parcels around 19 lakes
sold during the years 1982 through 1984.
This data set is provided in Appendix B.
plots of the raw
some obvious outliers.
Selling price versus selling price per square foot of lot and selling price versus square
plots occur because (1) there was an error made in recording the data in the property
appraiser's database; and/or (2) these observations represent property transactions that
In either case
, these data points are empirically separate from the remaining
points for reasons outside the realm of where this analysis is targeted.
are removed from the data set, causing the final database to contain 42 observations.
Variable name assignment and descriptive statistics of each are shown in Table
Simple correlation among all the variables are shown in Table V-3.
Each of the
, with the exception of the year-of-sale categorical variables, exhibits significant
correlation to ACTPR, at the 1 percent significance level.
The direction and magnitude
relationship that is verified here. The negative sign of the TSI variable indicates that TSI
is higher for lower-priced parcels. This is expected, as lake quality generally decreases
The SIZE correlation coefficients indicate selling price around
larger lakes is higher compared with selling price around smaller lakes.
, and Y84 simply reflect time-dependent inflation.
TSI is significantly correlated to parcel size.
There is no physical explanation for
There is also a somewhat significant relationship between lake size and TSI.
this is interesting,
there is no apparent reason why larger lakes would be more eutrophic
than smaller lakes.
8.1 Pdo (3) Fe Eqsr Pus
Sale Price Versus Unit Price for
Lake Characteristics Analysis
Descriptive Variables Used in Lake Characteristics Analysis
Description Variable Mean Std. Dev. Minimum Maximum
Parcel selling price ACTPR 66,445 51,107 5,000 175,000
Lake trophic state index TSI 49.5 10.9 33.0 75.0
Parcel footage FTSQ 24,070 18,792 2,436 88,305
Lake acreage SIZE 554 597 1 1,757
Pearson Correlation Coefficients and
Significance Statistics for Variables
in Lake Characteristics Analysis
ACTPR TSI FTSQ Y82 Y83 Y84 SIZE
ACTPR 1.0000 -0.5028 0.5456 -0.2648 -0.0187 0.2391 0.4215
0.0 0.0007 0.0002 0.0901 0.9060 0. 1271 0.0054
TSI -0.5028 1.0000 -0.6233 0.0822 -0.2155 0.1696 -0.2686
0.0007 0.0 0.0001 0.6045 0.1704 0.2829 0.0854
FTSQ 0.5456 -0.6233 1.0000 -0.3251 0.3067 -0.0697 0.4323
0.0002 0.0001 0.0 0.0356 0.0482 0.6609 0.0042
Y82 -0.2648 0.0822 -0.3251 1.0000 -0.4920 -0.2828 -0.2677
0.0901 0.6045 0.0356 0.0 0.0009 0.0695 0.0864
Y83 -0.0187 -0.2155 0.3067 -0.4920 1.0000 -0.6958 0.0837
0.9060 0.1704 0.0482 0.0009 0.0 0.0001 0.5980
VQlA n 3)IQI n 1^0^ _n nfhO7 _n 1V'Q _n flhl Q 4w n no
by the remaining variables.
sale variables are
, with 1983 being the base case and therefore part of the intercept.
The PROC REG procedure in SAS is used to produce the ordinary least-squares
parameter estimates to the multivariate model.
signs of each parameter estimate are as expected.
The results are shown in Table V-4.
TSI is the stronger of the two lake
In fact, the t-statistic for SIZE is only 1.2260, which makes it
Each unit of TSI increase causes the selling price to drop about
Unit change in FTSQ causes the selling price to increase $0.76.
for FTSQ reveals a moderately strong relationship at most.
It is significant at the 12
Y82 is insignificant,
is the strongest variable in the model,
picking up time-related variance in the selling price.
The functional form of the model shown in Table V-4 is linear.
semilog forms of the model were examined with marginal improvement in R-square (0.47
These configurations significantly reduce the strength of the t-
statistic for TSI.
Since the objective of this portion of the analysis is simply to prove the
existence of a relationship between lake characteristics and property value,
a few percent points of goodness-of-fit is considered warranted for stronger parameter
estimates for the variables of interest.
Land Rent Gradient
The final model evaluated incorporates locational variables in explaining property
Parameter Estimates for Lake Quality Model
Variable Estimate Error T-Statistic Prob > |I TI
INTERCEP 107,938.00 45,867.92 2.353 0.0242
TSI -1,549.22 749.07 -2.068 0.0459
FTSQ 0.76 0.48 1.602 0.1179
Y82 -2,697.02 18,705.98 -0.144 0.8862
Y84 32,102.00 14,694.66 2.185 0.0355
SIZE 14.51 11.84 1.226 0.2281
Dependent variable = selling price
N = 43
F-statistic = 6.2370
Prob > F = 0.0030
R-square = 0.4642
peak at the CBD (see the theoretical land rent gradient shown in Figure I1-3).
rent gradient reflecting reality has more peaks and valleys than the single CBD peak.
This portion of the analysis defines the peaks and valleys caused by water resources
while trying to control for other influences on property value.
In the previous two sections, the existence of a lake and its characteristics have
both been proven to impact land value.
The ability to indicate lakefront or nonlakefront
has been facilitated by a special code in the property appraiser's database.
The parcel number contains the township, range,
places the parcel in a square-mile area.
and section delineation, which
The expected influence of a lake may vary
locally, and therefore further locational definition is needed.
The parcel number also
contains subdivision, block, and lot specification that, unlike the township, range, and
section numbers, are not tied to a numerically consistent map location.
attributes serve as an index to the property appraiser's parcel maps that were supplied
by the Orange County Property Appraiser's Office on microfilm.
are each quarter-corners,
Maps on the microfilm
or one-half mile by one-half mile squares, a map scale that
allows identification of individual parcels.
defined in terms of X-Y
Thus each parcel of land considered was
Referring back to Figure IV-5,
it is indicated that 3,241 parcels are available,
which are located randomly throughout the study area. This defines the starting point
from which the final sample of observations is drawn. As expected, the lakes with TSI
data limit the number of applicable parcel observations. A region of at least one mile
were available was
specification followed the one-mile-square sections defined on the detailed county map.
The sections that bordered the lake of interest, plus one more section beyond the border
sector, make up the region associated with the particular lake of interest. The relatively
large band of surrounding land ensured consideration of all possible parcels that might
be impacted by the lake of interest.
regions of interest overlapped. The
Many of the regions blended together because their
e final lake regions considered for analysis are shown
Approximately 1,300 parcels were sought out in the parcel maps.
These were subsetted
further because (1) several parcels were located in an individual subdivision for which
a random number of parcels were selected; (2) the parcel was situated next to a lake for
TSI data were available;
(3) the parcel
was not found on the map.
information for approximately 570 parcels was taken from the parcel maps.
coordinates of the approximate center of each parcel were recorded.
The unit of size
is not consistent for all
parcels and in many cases cannot be converted to a consistent area metric.
many parcels are recorded as "one lot."
There is no way of knowing the size of this
Therefore while finding the X-Y location of parcels, lot areas were measured and
recorded in square feet.
All lakes within the region were considered because,
have TSI data recorded
considered in the statistical analysis.
though the lake may not
, it is a competing amenity within the region and needs to be
The result is a set of X-Y boundary coordinates for
the 96 lakes in the study area.
The intent of defining all objects of interest according to X-Y
coordinates is to
compile a simple geographic information system
This not only provides the
capability of determining relative distance among all the objects,
but graphic capabilities
Three other pieces of information are added to the GIS.
X-Y coordinates for
the major shopping malls are places on the database.
junctures are entered.
Similarly, major transportation
Major transportation and shopping hubs serve as proxies for CBD.
a, + ..
S e + -
0^ ^ "'
S9 a, U
3 A- + -4
I' + C-,
--- BBt+ +
A-~~ .~'6**IBB JH r
A computer program,
written in BASIC
between each parcel and its first,
was used to calculate the shortest distance
, and third closest lakes; nearest shopping center;
is found in
This set of distances for each parcel,
compiled with parcel size, sales
information, lake size, and TSI,
makes up the variables used in the analysis.
observations that were sold for $100 and one parcel for $114.
The next highest sales
price was $5,000,
and the prices increase continually from there.
One explanation for
this is that they may be unqualified sales,
such as sales to family members for a nominal
It should be noted, though,
that this was controlled for during the initial filtering
process (described in Chapter
IV) by including only
or qualified sales,
These may have been miscoded.
In any event,
these observations are
removed from the database.
Examination of selling price versus selling price per square foot, shown in Figure
V-6, reveals a couple of observations that are suspect.
These points are approximately
$18 per square foot, where the next highest values are about $10 per square foot and then
the values continually decrease.
As stated above, these points consist of effects outside
the intent of the model (most likely errors in coding) and are therefore deleted.
data set contains 153 points and is provided in Appendix D.
Variable name assignment and descriptive statistics are shown in
0 2 4
hk Pier NSm 01 N. F
Sales Price Versus Unit Price for
Land Rent Gradient Analysis
The distance to lake variables LAKE1D and LAKE2D both have strong
HUB and SHOP each indicate decreasing property value with increasing
The negative correlation coefficients of HUB and SHOP support this
and SHOP are all correlated with one another.
since SHOP and HUB possess the same tvoe of convenience benefit that
~~ 00 S\ C O0 O~C
en- en ott 00 00
'fl .-~~~~-0\Qft~ C'\8(M
In N~01 InOr tO000
oo-OVI-r--tnrI 0\ en-moo
OOt-4e 00%0 02 eno oC Onvro
N-'tmnennO\ 08 flO0 00 'n --O Nf
(10l (MOu cflO% 000'r
eq 'nO N\ t) CM CM 00'C
eq men Oe ne Me
\0 (1r ON en, 0O-40N
ICC O 1 I0\id
en mne en-' Cr rOO VOIn
0% Q5 0%0 N5N'OOn1 cit-'
tnI In CM (AS -\0c 0(1 Oen
In N eq In 0* NNO~na \
-o~d~6t6 66d66 6d
~~~~~ I;d;c idd 'dd9 d
LAKE1D and LAKE2D are correlated simply because, as distance to
distance to the second
nearest lake also increases.
correlations between TSI
, and FTSQ have no apparent physical justification.
statistically strong parameter estimates for the independent variables.
concern is directed at controlling for correct functional form,
The next level of
goodness of fit, and
multicollinearity and heteroskedasticity.
Inclusion of LAKE1D and LAKE2D
which were shown to be correlated
important variable, is dropped from tt
insignificant variables in the model.
SIZE1 and SHOP were found to be
The final model has ACTPR as the dependent
variable and FTSQ, TSI1
, and SHOP as independent variables.
Following the suggestions of Halvorsen and Pollakowski (1981) and Milon et al.
consideration of the Box and Cox (1964) procedures (referred to as BOX-COX)
for functional form selection is made.
Milon et al.
(1984) customize BOX-COX for
application to water resource amenities that is used in this research.
vector of nonwater-related attributes
vector of water-related attributes
number of independent variables
BOX-COX transfnnnormatinn fatortn
This is an iterative procedure that varies functional form by changing X and
-AK lno2 (X,0)
= log-likelihood statistic
K = number of observations
P = sale price
o2 = standard deviation
,\X = BOX-COX transformation factors
FTSQ and ACTPR are found to be linearly correlated and therefore share equal
transformations through the procedure.
for the dependent variable ACTPR and FTSQ.
The remaining independent variables are
transformed as follows:
4M 4 .a aa-
Box-Cox Procedure Summary for Land
Rent Gradient Model
0 -1 -1,399.0
0 0 -1,384.4
0 1 -1,356.5
0 2 -1,366.5
1 -1 -1,515.0
1 0 -1,508.9
1 1 -1,493.5
1 2 -1,496.9
2 -1 -1,703.2
2 0 -1,701.1
2 1 -1,694.3
2 2 -1,694.0
ordinary least-squares parameter estimates.
Programming in SAS is used to create the
BOX-COX mechanics and develop the log-likelihood statistic.
in Appendix E.
This SAS code is listed
The log-likelihood statistic for the iterations of 0 and X is provided in
The results indicate 0
as the best functional form.
Thus the land
rent gradient model takes the form:
Since increasing TSI is expected to cause ACTPR to decrease, the sign of #,
associated with FTSQ should be positive.
, and CBD are all expected
to cause decreases in ACTPR and
, therefore, f3,
and 0s should be negative.
The ordinary least square regression results for the final model,
shown in Table
, indicate all variables are strong indicators of
the t-tests indicate
The F-test and r-square values also indicate a strong model.
expected parameter estimate signs of each variable except CBD are revealed.
The sign of CBD indicates that increased distance from the city center causes
increased property value.
This is not a surprise because the correlation coefficient for
and ACTPR is positive.
measure of CBD
a convenience is
congestion has caused the traditional land rent gradient to be reversed in many cities.
This increasing land rent gradient with distance from the CBD is in the study area and
is statistically significant.
A couple of technical items regarding the final model should be noted.
was not considered as a functional form because the parameter estimates were
marginal, and the decision to sacrifice the "best" functional form for ease of application
and interpretation was considered justified.
Parameter Estimates for Lake Quality Model
T for H,:
= log of selling price
research, and its inclusion in the model is considered important for a prior reasons.
large sample size helps to minimize the collinearity impact on the parameter estimates.
The degree of collinearity gauged through derivation of a condition index,
Belsley et al.
(1980), indicates an acceptable level of collinearity.
shown in Table
The condition index
, is considerably less than the rule-of-thumb value of 30
recommended by Belslev et al.
The result is CRD is an imnnrtant cnntrihutnr in the final
Collinearity Diagnostics from SAS
Proportion of Variance
A necessary condition of a properly defined least squares regression model is that
Any pattern in
the error term indicates
causes specification errors in the parameter estimates.
Following Hannett and Murphy
(the model's prediction of selling price for each observation) is
plotted against the model's error term in Figure V-7.
very limited homoskedasticity.
No strong pattern exists, indicating
Plots of the error term versus the independent variables
are shown in Figure V-8, further indicating randomness in the error term.
A final consideration in evaluating the model's robustness involves examination
of autocorrelation based upon the location of each observation. This is termed spatial
0o o o o b
o 09 o0 0 oo
oo o yoo 0o
oo oo oo
o o oo
0 w0 000 00 0* o
U!dY Jo CO>tA pop jp ad
Ft)pd P ow'sQE~
randomness of the model error in terms of each observation's location.
If a pattern
exists, model specification problems may exist and appropriate statistical remedies are
The Moran coefficient (Moran 1948) is used to test for spatial autocorrelation.
Wg = inverse of the distance between points i and j
e1 = model error for observation i
ei = model error for observation j
The Moran coefficient can range between -1 and +1 where a value of 0 indicates a
purely random pattern. A z-statistic is used to determine if there is a non-zero indicator
of spatial autocorrelation. The Moran coefficient value for the present model is 0.05,
which is very close to zero, and the associated z-statistic indicates there is no significant
spatial pattern in the error term.
DISCUSSION OF RESULTS
The first two chapters provided a setting for the present effort by describing the
problem in environmental benefits estimation,
theory, and past empirical application of
Chapters II through V presented the data, methods,
and final models
as part of the present research.
results by examining
specific components of
the distance decay
These results are presented in terms of the theory
past work, and application
Three hypotheses were introduced at the beginning of Chapter V
The effects of lake characteristics (size and water quality)
are realized in land values.
Water resource-related impact on land value will diminish
with distance from the water source.
Each hypothesis was proven true through presentation of analysis and models in
Results relating to the hypotheses are summarized in Table VI-1.
lake characteristics are capitalized in proximate land values, and the magnitude of the
impact varies according to distance to lake-
and lake sirze
Summary of Hypothesis Results
Comparison of means
Nonlakefront property is
59 percent of lakefront property
TSI is negatively correlated
with property value; lake size
Land rent gradient
TSI and distance are negatively
correlated with property value
Beneficiaries of Lake Resources
or local taxing issues.
question is addressed.
The "who pays" part is often convoluted with political agendas
This issue is not addressed here, rather the "who benefits"
The types of benefits provided by lake resources are generally
recreation and aesthetics.
Water resources can also provide, for example,
but the focus here is on recreation and aesthetics.
It is important to emphasize here that the benefits calculated through hedonic
value represent only a part of the total benefits picture.
The magnitude of the benefit to
those in proximity, as measured through this hedonic valuation procedure, is valid and
a significant distance (approximately one mile or more) are not considered here.
any lake resource benefits derived from hedonic valuation are a partial estimate of the
total benefit value of the lake.
According to the parameter estimate on LAKE1D (distance in feet to the nearest
increases, the benefit received from the lake decreases. This is illustrated in the scatter
plot of observed land values to lake distance in Figure VI-1 (top). A line drawn through
this scatter plot would obviously have a negative nonlinear slope. Recognizing there are
other influences on
this line cannot be interpreted literally,
general trend exists.
Dombush and Barranger (1972) indicate the impact of lakes to be
negligible beyond 4,000 feet from the lake boundary.
trend of data shown in Figure VI-1.
calibrated land rent model, the lake
This appears to be the general
To look at the relationship more closely through the
: distance impact is shown in Figure VI-1 (bottom).
holding independent variables constant (at their respective
averages) while varying LAKE1D.
This plot shows a nonlinear-decreasing relationship.
The change in slope is less pronounced than expected.
At 4,000 feet on the x-axis, the
curve continues to decrease, where according to Dombush and Barranger (1972) the line
should be parallel with the x-axis.
These results indicate that the lake impact in Orange
Florida, goes beyond 4,000 feet.
This should be viewed cautiously,
Extrapolation beyond observed LAKE1D values should be done only for theoretical
2,000 4,000 6,000
Lake Distance, Feet
Lake Distance, Fe
The point along the curve in Figure
(bottom) at which the property value is 59
percent of the value of the lakefront property (lakefront property is where distance from
equals 0) is at approximately 4,400 feet.
This point on the curve compares
favorably with Dombush and Barranger (1972),
but the remaining portion of the curve beyond
4,400 feet indicates benefits greater than those reported by Dornbush and Barranger
be expanded three-
dimensionally to develop a surface, which is termed the land rent surface. This concept
is often referred to in theory but is rarely shown with actual empirical data. This is the
case because continuous data on distance are
typically not employed.
continuous data in the present effort,
empirical land rent surfaces can be explored.
land rent surface produced in the following figures are a means to present the concept
and results of the land rent gradient model.
The precise shape of the surfaces are a
function of the interpolation used in the mapping software.
Therefore any empirical
analysis should be based on the mathematical models versus data pulled from the land
rent surface maps.
The land rent surface for Long Lake is shown in Figure
The inset of
Orange County shows
the location of Long Lake.
Where limited precise empirical
information can be taken from the surface, the surface provided in Figure VI-2 is shown
two-dimensionally in Figure VI-3 through isovalue lines (where like land values are
connected through interpolation). The values associated with each line are in $1,000,
deflationary impact of distance is indicated by the lake appearing as a plateau at an
elevation higher than the remaining points.
Lake quality as measured through TSI has a negative relationship with property
The observed data for TSI and property values are shown in Figure VI-4 (top)
which indicates a decreasing trend.
Property value versus TSI is shown in Figure VI-4
(bottom) by holding all independent variables at their mean and varying TSI.
the inverse relationship according to the calibrated model.
The relationship is nonlinear
as was specified through the semilog functional form of the statistical model.
of 10 TSI results in about a 20 percent impact on price.
The change in land values associated with a changing TSI from 66 to the 54.5
(the sample average in this study) for the Long Lake region is shown in Figure VI-5.
This causes increases in property value throughout the Long Lake region that are more
pronounced near the lakefront.
The isovalue line
the lake indicates a $2,750
increase in property value associated with the enhanced water quality conditions.
isovalue lines decrease with distance from the lake
, which follows the trend of decreasing
lake impact with lake distance shown in the statistical model.
To further illustrate the impact of distance and TSI of lake resources on land
values, a four-lake region is shown in Figure VI-6 and Figure VI-7.
, Giles, and Arnold with a TSI range of 62 to
75 show the varying demand for
TSI for Lakes Underhill and Giles is 62 and is 65 and 75 for Lakes
Arnold and Como, respectively.
The higher plateaus are shown at the lesser eutrophic,
30 40 50 60 70
Trophic State Index
20 30 40 50 60
Trophic State Index
Performance of Troohic State Index
highly significant in
the lake quality
model and the
The difficulties in finding an appropriate water quality metric were
discussed in Chapter IV
and the decision to use TSI was made.
The results described
above indicate TSI is recognized in the residential property market and is therefore,
applicable in hedonic valuation framework.
As the engineering community becomes
more comfortable with TSI development, the exact form will likely change, which would
in turn affect the parameter estimates.
this study supports the use of
TSI as an indicator of water quality in the hedonic valuation framework.
Hedonic Valuation as a Planning Tool
Probably one of the more encouraging results to surface from this research is that
hedonic valuation can readily be used by water resource managers in benefit estimation.
This point has not received attention until now because the intricacies of the data and
statistical analysis have received a majority of the discussion. If the study area of
interest has an active GIS with parcel level property value information, the planner has
a tremendous advantage.
Provided below is a summary of the suggested procedure for
conducting hedonic valuation for lake resources.
Step 1: Determine the Purpose of Application
Generally, this method is used to estimate a portion of the benefits attributed to
lake resources as discussed earlier in this chapter.
If a particular application is intended,
this should be explored fully to determine exactly which benefits are being measured.
Determine Study Area
The study area is dependent on the purpose of the analysis (step 1).
is on a single site,
If the focus
then possibly a region of ten miles or so around the lake is required.
If a wider regional demand specification is derived, then a county or multiple county
study area is needed.
The most likely situation will be that the planner will be controlled
the amount of
level information are not
typically formed for more than one county.
Define Data Sources and Create Working Databases
, robust, and complete database is essential for application to hedonic
Property data should include sales price, size of parcel,
The relative distances between parcels and lakes are needed
type of sale, and time
I. It is also important
to include distance to other important factors that may influence property value (e.g.,
distance to shopping).
Technical lake characteristic data are required as part of the database.
TSI is a
recommended starting point based upon the successful application in the present effort.
Lake characteristics data are typically available through environmental or regulatory
agencies at the local,
or federal levels.
All avenues should be pursued,
characteristic data, especially describing water quality,
If a GIS is available and the property appraiser's information is part of it, a large
part of the data gathering is complete.
of nonmarket transactions. This resex
It is important to exclude the property value data
irch also shows that examination of vacant parcels
property value data.
If a relatively small region is being examined,
the property quantity
data needed may be quite manageable,
even if they are gathered from hard copy sources.
Given the database developed in step 3,
as the dependent variable.
run a statistical model using sale price
This may require careful statistical insight-the processes
presented in Chapter V (land rent gradient model) can be used as a guide.
estimates should be compared with past work.
check that the model
provides realistic results and recognize its limitations.
Examine Benefits of Alternative Projects
The effects of proposed engineering projects on the lake are then evaluated in
terms of the model.
If TSI is a parameter, the TSI values, before and after the project,
are plugged into the model and the difference is a measure of the project benefits.
assumptions and benefit calculations should be carefully documented.
This will allow
for reasonable application of the results to benefit-cost analysis and will also aid in future
application of the model.