GIS Analysis of Land Use on the

Rural-Urban Fringe: The Impact of Land

Use and Potential Local Disamenities on
Residential Property Values and on the
Location of Residential Development in
Berks County, Pennsylvania
by Richard Ready and Charles Abdalla
June, 2003

Rural Development Paper No. 18

©2003 The Northeast Regional Center for Rural Development

Located at: The Pennsylvania State University
7 Armsby Building, University Park, PA 16802-5602
Phone: 814-863-4656 FAX: 814-863-0586
E-mail: nercrd@[Link] URL: [Link]

“Contributing to the well-being of small towns and rural communities.”

 Staff Paper 364 June 2003

GIS Analysis of Land Use on the Rural-Urban Fringe: The Impact of Land Use and
Potential Local Disamenities on Residential Property Values and on the Location of
Residential Development in Berks County, Pennsylvania

Final Report

Richard Ready1 and Charles Abdalla2

1
Assistant Professor of Agricultural and Environmental Economics
2
Associate Professor of Agricultural and Environmental Economics
Department of Agricultural Economics and Rural Sociology
The Pennsylvania State University
University Park, PA 16802

Staff Papers are circulated without formal review by the Department of Agricultural Economics
and Rural Sociology. Content is the sole responsibility of the authors.

A shorter, less-technical version of this report, entitled “The Impact of Open Space and Potential
Local Disamenities on Residential Property Values in Berks County, Pennsylvania,” is available
at [Link]

This research was supported by the Northeast Regional Center for Rural Development
(NERCRD), under Internal Agreement No. 2093-TPSU-USDA-0496, with co-principal
investigators Drs. Ready, Timothy Kelsey, and Rick Day and consultant Dr. Nancy Bockstael.
Additional support that allowed collection of data on potential local disamenities came from
Citizens for Pennsylvania’s Future (PennFuture), under Grant in Aid No. 62114, with co-
principal investigators Drs. Ready and Abdalla.

The authors are grateful for technical and programming support from Stewart Bruce and Yanguo
Wang, for local knowledge of agricultural production provided by Clyde Myers, Donald Reinert,
Mena Hautau, and David Beyer, for invaluable help on assembling datasets from Martin Moyers
and Bill Kochan, and for the guidance and help provided by the Project Local Ad visory Group.
The authors are also grateful to four anonymous reviewers for their comments and suggestions.
The authors assume all responsibility for the statements made in this report.
 GIS Analysis of Land Use on the Rural-Urban Fringe: The Impact of Land Use and
Potential Local Disamenities on Residential Property Values and on the Location of
Residential Development in Berks County, Pennsylvania

Executive Summary

This research project examined two issues related to spatial interactions between residential
properties and nearby land use, using data from Berks County, Pennsylvania. First, the project
modeled the impact that surrounding land use and potential local disamenities have on residential
property values. If a land use is seen as an amenity by nearby residents, then the value of that
amenity will be capitalized into the market price of nearby homes. Similarly, a nearby land use
that is seen as a local disamenity will decrease house prices. A hedonic (or implicit) price
regression was used to estimate the marginal impact of surrounding land use on house prices and
measure the relative amenity values of different types of land use. Second, the project
investigated whether surrounding land use has on effect on the location of new residential
development. This was done by modeling the impact of surrounding land use on the probability
that a parcel that was undeveloped in 1996 would subdivide for residential development during
the period 1996-2002.

For the hedonic house price analysis, land use for each parcel in Berks County was categorized
as either open space, residential, commercial or industrial. For 8,090 single family houses sold
between 1998 and 2002, land use within 400 meters and within 1600 meters of the house was
measured, using a geographic information system (GIS). In addition, distances were measured
to potential local disamenities, including landfills, sewage treatment plants, the regional airport,
high-traffic roads, mushroom production facilities, and large-scale animal produc tion operations.
A regression analysis modeled variation in house prices, with explanatory variables including
structural characteristics of the house, measures of proximity to employment centers, and
measures of surrounding land use. An instrumental variables regression approach was used to
reduce possible bias due to endogeneity between land use and house prices.

The hedonic house price regression showed that, within 400 meters of the house, the land use
that has the most positive impact on house price was open space, followed by large- lot single
family residential land. Commercial, small- lot single family residential, and multi-unit
residential were less desirable. The least desirable land use within 400 meters of the house was
industrial. Of open space uses, land that is currently vacant but that has been zoned for
residential, industrial or commercial development was found to have a much lower amenity
benefit than other land in open space, possibly reflecting the negative impact of uncertainty over
the future use of that land and the potential for noise, dust and traffic during the building process.
Also, open space on parcels that are covered by conservation easements, including agricultural
conservation easements, has a less-positive amenity impact than open space not covered by such
easements. This does not necessarily mean that easements cause nearby property values to
decrease. It may be that farms with agricultural conservation easements tend to be managed
more intensively, which may be seen as less attractive by nearby homeowners.

Between 400 and 1600 meters away from the house, the land use with the most positive amenity
impact on house price was commercial, followed closely by large- lot single family residential.

i
Of open space uses, only land that is owned by Local, State or Federal Government and land that
is covered by conservation easements have a statistically significant positive amenity value. To
summarize, the ideally-situated house would be immediately surrounded by open space, with
commercial properties (stores and offices) located 400 to 1600 meters away.

Several potential local disamenities were found to have a negative impact on nearby house
prices. Of the potential local disamenities investigated, the impact of landfills on house price
was largest, and extended the farthest (up to 3200 meters). A landfill located 800 meters from a
house decreases that house’s sale price by an estimated 6.9%. The impact of a large-scale animal
production facility (over 200 animal equivalent units or aeu’s) on house price was about one half
to two thirds as large as that from a landfill (4.1% at 800 meters), and did not extend as far (up to
1600 meters). The impacts on house price from mushroom production and from the regional
airport were much less (0.4% and 0.2%, respectively, at 800 meters). The impact from high-
traffic roads was small, and extended only a short distance. No significant impact was found for
sewage treatment plants.

Additional analysis attempted to investigate whether different types of animal production

facilities had different impact on nearby house prices. Differences in the impact due to
differences in the size of the operation (number of aeu’s) were not statistically significant.
Further, medium-sized production facilities (200 to 300 aeu’s) were found to have a statistically
significant negative effect on house prices when considered apart from larger facilities.
Similarly, the impact did not vary significantly by species (poultry, swine, and beef/dairy). An
analysis of proximity of animal production facilities and residential properties showed that the
density of single family homes around animal production facilities was lower than the average
for rural parts of the county. An implication is that some potential for conflicts is avoided due to
the way in which these land uses are located on the land.

The total impact on surrounding house prices was calculated for a landfill, the regional airport,
and an animal production facility. The average impact on the value of 3342 houses located
within 3200 meters was $2442 (all values are in 2002 dollars). The total impact on all houses
was $8,162,000, which is 2.6% of the assessed value of the affected properties. The average
impact of the regional airport on 2256 houses located within 1600 meters of the airport runway
and its flight paths was $104, and the total impact on the value of these properties was $235,000,
or 0.1% of the assessed value of the affected properties. This calculation does not include 2391
properties located near the airport within the City of Reading. The average impact of a single
animal production facility on 119 single family residences located within 1600 meters of the
facility $1,803. The total impact on all 119 houses is $215,000, or 1.7% of the assessed value of
the affected houses. These figures are intended as illustrations, and should not be considered
averages for similar facilities. The impact from any given landfill, airport, or animal production
facility will depend on the number of houses located near the site, and on the market value of
those houses absent the facility.

An analysis of the location of new residential development was limited by the small number of
subdivisions that occurred during the study period. Still, the analysis showed that larger lots are
more likely to be developed, that lots located near existing residential areas are more likely to be
developed, and that proximity to government-owned open space has a small positive impact on

ii
probability of development. Potential house price, as predicted by the hedonic price function,
did not help explain the pattern of development. There was a tendency for new development to
locate near government-owned open space. However, for privately-owned open space, there was
not found a significant tendency toward leapfrogging, where new development locates in areas
with more open space.

The project demonstrated the utility and feasibility of using GIS and hedonic pricing analysis to
investigate spatial interactions between residential and other land uses. The study area chosen,
Berks County, was well suited to this type of analysis, in terms of data availability and the
diversity and dispersed spatial pattern of land uses and agricultural production. The research
method should be extended to more study areas, to see if differences in population density,
demographics, or type and amount of open space and agricultural production influence the
results. Until more research is conducted in more counties, care should be taken in extrapolating
the results from this research to other regions.

iii
 GIS Analysis of Land Use on the Rural-Urban Fringe: The Impact of Land Use and
Potential Local Disamenities on Residential Property Values and on the Location of
Residential Development in Berks County, Pennsylvania

I. Introduction and Project Objectives

Increased attention is being focused throughout the Northeast on how land use is changing over
time. Concerns over urban “sprawl,” with its patchy, diffuse pattern of development, include the
loads placed on the transportation infrastructure, the costs of delivering local services, the
impacts on natural systems, and the effects on the aesthetic and cultural value of the landscape.
Local authorities who manage and regulate growth and development need information on both
the relative desirability of alternative land use patterns, and on the forces generating those
patterns.

One means of assessing the relative desirability of alternative land use patterns in a region is to
measure the benefits and costs that the region’s citizens enjoy or suffer from those patterns.
Benefits and costs associated with land use changes can be measured using survey techniques
(for example Halstead 1984; Bergstrom, Dillman and Stoll 1985; Beasley, Workman and
Williams 1986; Ready, Berger and Blomquist 1997), hedonic pricing techniques that measure
differences in property values across different land use patterns (for example Garrod and Willis
1992a, 1992b; Ready, Berger and Blomquist 1997), or fiscal impact techniques (Burchell,
Listokin and Dolphin 1985). Survey and hedonic pricing studies have consistently found that
citizens place positive value on open space near their residence.

Because of data limitations, these studies have until recently tended to treat open space in a fairly
aggregated way, measured as percent of total area in the region, or total number of acres, in
forest or farmland uses. This simplification masks an important characteristic of the land use
pattern - its patchiness. The value to residents of open space in a region depends on how that
open space is distributed relative to residences and shopping and work centers. Fortunately, with
the development of detailed, more-or-less uniform, GIS- formatted land use data, it is now
possible to specify and estimate hedonic property price models that model spatial interactions
among land parcels in a much more detailed way.

Within the last few years, researchers have started using detailed GIS data to estimate hedonic
pricing models valuing open space and water quality (for example Bockstael 1996, Leggett and
Bockstael 2000). These studies represent an important advance in hedonic modeling because
they are based on much finer resolution data on land use, and therefore give a much more
detailed picture of the spatial interactions that occur among parcels. Using detail GIS-based
data, it is possible to differentiate, at a much finer level of detail, between open space located
close to a parcel and open space located farther away, and to estimate the marginal impact of
each on land prices.

Spatial interaction among parcels is important not only because it influences land values, but also
because it influences land use change. If residents value proximity to open space, then newly
constructed homes located farther away from high density development will have higher value
than new homes located in more-densely developed areas. This price differential may favor new

1
home construction in areas with more open space, leading to a tendency for patchy and/or
leapfrog development patterns (Irwin and Bockstael 2001a, Irwin 2000).

This project had three main objectives. The first objective was to develop a GIS database on
land use and residential property values for a county in Southeastern Pennsylvania, a region
characterized by highly productive agricultural land but also by continuing development pressure
and a rapid rate of farmland loss. This database included factors that would influence residential
property values, including both land use and location of several different types of potential local
disamenities.

The second objective was to estimate a hedonic pricing function that explains variation in single-
family residential properties. Explanatory variables include structural characteristics of the
properties (square feet of living space, lot size, etc.), factors that vary spatially related to local
government (school district quality, zoning, etc.), measures of surrounding land use, and
proximity to potential local disamenities.

The third objective was to characterize the pattern of recent residential development, to
determine whether spatial interactions among parcels were an important determinant of the
location of new home construction.

II. A Review of Theory and Previous Research

IIa. The Theory Behind Hedonic Pricing Analysis

The following discussion of the theoretical foundation for hedonic property price models follows
closely that of Palmquist (1991). Hedonic property price models stem from the work of Sherwin
Rosen (1974). Rosen’s model treats any market good (in our case, a single family home) as a
collection of attributes. In a competitive market, similar goods will exist with slightly different
levels of these attributes. The market price of a particular good will depend on its levels of those
attributes, which can be characterized as a vector, z. Goods with more attractive levels of the
attributes will sell at a higher price.

In a hedonic pricing analysis, the hedonic price function (also called the implicit price func tion),
P(z), is an empirical relationship that predicts the market price of a given good as a function of
the levels of its attributes. For single family homes, this function is the equilibrium set of house
prices that results given the population of hous e buyers and the available housing stock. House
buyers search the set of available houses, and choose the house that maximizes their indirect
utility function, given by V(W-P(z),z), where W is the wealth of the household. If houses with
sufficient variability in z are available in the market, then each household will choose a house
that maximizes their utility. For each single house attribute, zi, the first order condition for this
maximization is

∂P/∂zi = ∂V/∂zi .
∂V/∂W

2
The left hand side of this equality is called the marginal implicit price of attribute zi. The right
hand side is the household’s marginal rate of substitution between attribute zi and money. For
marginal changes in zi, then, the marginal implicit price of zi measures the household’s marginal
willingness to pay for additional zi.

For nonmarginal changes in zi, the implicit price function provides either an upper or a lower
bound on the willingness to pay of the household for an exogenous change in zi. If a
nonmarginal change from z0 to z1 is seen as an improvement, then ?P = P(z1 ) - P(z0 ) is an upper
bound on the willingness to pay of the household for that exogenous change. If the change from
z0 to z1 is seen as a worsening, then ?P = P(z0 ) - P(z1 ) is a lower bound on the amount the
household would need to be compensated to accept the exogenous change. However, as
Palmquist and others make clear, this is not the final benefit or cost to the household operating in
the housing market because households can relocate and market prices can adjust. If the change
affects only a small number of houses, then ?P will exactly equal the windfall gain or loss to the
house owner and no further welfare effects accrue as long as transactions costs of moving to new
equilibrium locations are minimal. For more widely dispersed effects the problem is much more
complex because the hedonic price function itself shifts as a result of the nonmarginal change.

Nonetheless, the marginal implicit price for an attribute is a reliable signal of the household’s
marginal willingness to pay for a change in that attribute, and provides information about the
benefits and costs from changes in those attributes. These benefits and costs can be of particular
policy relevance when the attributes are influenced either by government policy or by actions
taken by neighbors located near the parcel. In our context, z includes measures of land use near
the house, which is determined by the owners of nearby parcels who make decisions within the
context of local planning and zoning regulations. If surrounding land use is an attribute that
households care about, then a change in land use near a particular house will change the price of
that house, and will impose costs or benefits to the affected household that can be measured
using the estimated marginal implicit price for that attribute.

IIb. Previous Hedonic Pricing Studies of Land Use

Many papers have used hedonic pricing models to analyze the effects of open space on
residential property values (Cheshire and Sheppard 1995, Garrod and Willis 1992, Greghegan
etal 1997, Tyrvainen and Miettinen 2000, Irwin, 2002, etc.). However, results from these papers
are mixed due to different kinds of open space considered, specification of the open space
variables, and differences across study regions (Irwin 2002).

Garrod and Willis (1992) found that the effects of open space, specifically forests, on residential
property values depend on the species of forests. They conclude that deciduous trees within one-
km significantly increase house prices, but spruce conifers significantly decrease house price.
Geoghegan et al (1997) examine the effects of agricultural and forest land on the surrounding
residential housing values in a central Maryland region. However, their results vary with the size
of the neighborhood considered. Cheshire and Sheppard (1995) estimate the effects of open
space on residential property values in England. They found that the effects of open space
depend on the relative scarcity of open space surrounding the study area. Specifically, open
space has a positive and significant effect on residential property values only if the amount of

3
open space is sufficiently scarce. Tyrvainen and Miettinen (2000) found that the distance to the
nearest small area of forest has a negative effect on residential house price in Finland, while the
presence of a forest view from the housing unit has a positive influence. However, the relative
amount of forest land surrounding the housing unit and distance to the nearest large forest area
are found insignificant.

The study closest to this one in purpose and method is that done by Irwin (2002). In a hedonic
pricing analysis of residential properties in Maryland, Irwin measures the proportion of area
within 400 meters of each house in different land uses. Within the broad category of open space
uses, Irwin distinguishes among cropland, pasture, forest, permanently-conserved privately-
owned open space, and publicly-owned open space. Irwin found that compared to residential,
commercial or industrial uses, open space located within 400 meters of a residential property has
a positive impact on that property’s price. Further, pasture and cropland generate higher amenity
benefits to nearby residences than forested open space. Finally, both permanent conservation
through easements and public ownership increase the positive impact that open space has on
neighboring residential prices.

IIc. Endogeneity and Spatial Correlation in Land Use and Property Values

A unique issue arises when trying to estimate, with a hedonic price function, the impact of
surrounding land use on residential property values. That issue is the potential endogeneity of
open space in the hedonic price function. This can arise if there is a housing attribute that varies
spatially that is not observable to the researcher, but that influences house prices.

For example, consider two neighborhoods that are otherwise similar. The first suffers from a
localized odor problem from a facility that is unknown to the researcher. The second is desirable
because of its proximity to unique shopping or dining opportunities. Residential property values
will tend to be higher in the second neighborhood than in the first. Because of high resident ial
property values, any open space in the second neighborhood will be under development pressure,
and, over time, the amount of open space in the second neighborhood will decline. Conversely,
because residential property values are low, open space in the first neighborhood will be under
less development pressure, and will decline less over time. The amount of open space is said to
be endogenous to house price. That means that the amount of open space in the neighborhood
influences house price, but house prices in the neighborhood influence how much open space
remains.

The end result will be that the researcher will observe low property values and more open space
in the first neighborhood, and high property values and less open space in the second
neighborhood. If the researcher naively estimates a hedonic price function using ordinary least
squares (OLS) regression, he will conclude that open space is associated with lower house prices.
The danger is in concluding that open space is depressing house prices, when in fact it is the low
house prices that are allowing open space to survive.

One approach to estimating the structural relationship between open space and house prices in
the presence of this kind of endogeneity is to estimate a two-stage regression model using
instrumental variables (IV). In IV estimation, the right-hand-side variables in the hedonic price

4
function are divided into two groups, endogenous variables and exogenous variables. A third set
of variables, called instruments, is ident ified with the following characteristics: instrumental
variables are correlated with the endogenous right hand side variables, but are not correlated with
the error term in the hedonic price regression. In other words, the instrumental variables help
explain variability in the endogenous right- hand-side variables, but do not help explain
variability in house price.

IV estimation proceeds in two stages. In the first stage, the endogenous right-hand-side variables
are regressed, using OLS, on the instrume ntal variables and on some or all of the exogenous
right- hand-side variables. In the second stage, the hedonic price function is estimated using
OLS, but with the endogenous variables replaced by their predicted values from the first-stage
regressions. OLS standard errors from the second stage regression are biased, and must be
adjusted.

Irwin used IV estimation in her study on open space in Maryland (Irwin 2002, also Irwin and
Bockstael 2001b). She instrumented five endogenous variables: land in lo w density residential
use, land in medium and high density residential use, land in commercial and industrial use, land
in crops, and land in forested open space. Her instruments included physical features of the land,
such as slope, drainage potential of the soil, and soil quality for agricultural use. These
instruments are appropriate because they help explain variation in the endogenous variables, but
would not directly influence house prices. This project takes a similar approach, with a slightly
broader set of instrumental variables.

A second issue is that of spatial correlation among house prices. In addition to potential
endogeneity in land use, unobservable attributes that vary spatially and that influence house
prices can lead to spatial correlation in the house prices. Spatial correlation exists if the error
term in the hedonic price regression is positively correlated for observations that are located near
each other. If spatial correlation exists, OLS and IV regression will be inefficient, and estimated
standard errors on the parameters will be biased, affecting statistical inference in hypothesis
tests.

To date, no hedonic pricing study has estimated an implicit price function that accounts for both
endogeneity spatial correlation. Irwin (2002) uses an approach where observations are deleted
from the dataset until there are no two that are “close” to each other. However, she finds that
expanding the minimum separation among observations from 100 meters to 600 meters has little
effect on the IV regression results. This approach weakens statistical inference, however,
because information in the excluded observations is lost. In this project, all observations will be
retained. The model assumes, therefore, that spatial correlation in the errors does not exist. To
the extent that such correlation is present in the data, the estimated standard errors, and possibly
the parameters, may be biased.

IId. Previous Hedonic Studies of Local Disamenties

In this project, we investigate the impact on nearby residential property values of several
different types of potential undesirable facilities and land uses. These include sewage treatment
plants, landfills, high-traffic roads, airport runways, mushroom production facilities and large-

5
scale animal production facilities. Particular attention is paid to the potential disamenity
associated with animal production, as this is an activity that would be allowed on land that is
covered by an agricultural conservation easement.

Few hedonic studies have been conducted that specifically address the disamenity impact of
animal production on residential property values. Abeles-Allison and Connor (1990), in a study
of property values near large hog operations in Michigan, found that house values decreased by
$1.74 for each additional hog within a 2 kilometer radius of the house. They did not find
significant impacts outside of 2 kilometers. When comparing the impact per hog of large farms
versus small farms, no clear difference was found, though the dataset included very few
residential properties located near small farms, making statistical inference difficult. They did
not find any difference between residences located upwind of hog farms versus downwind. One
limitation of this study is that it only included eight hog operations that had received multiple
odor complaints. Property price impacts from these eight operations might well be greater than
those from other operations that did not receive complaints.

Palmquist et al (1997) measured the impact on residential property values of hog production in
the coastal plain of North Carolina, where some of the largest animal production facilities in the
nation are located. For each residential property, the total amount of hog manure produced
within ½ mile, within 1 mile, and within 2 miles was determined. Due to confidentiality
constraints, the authors were not able to determine how many individual operations combined to
give these totals, or where, exactly, those operations were located. Palmquist et al found that
house price was negatively affected by the concentration of hogs near the house, and that the
impact on house price from a single hog operation could be as large as 8.4%. Further, the impact
of an additional hog located within ½ mile of the house was found to be about 200 times greater
than that from an additional hog located more than ½ mile from the house, though the more-
distant hogs still did have a statistically significant negative impact on house price.

The Michigan study took as a maintained assumption that the negative impact from a livestock
operation on house prices increased linearly with the number of livestock increased. The North
Carolina study assumed that the impact from hog production was tied to the total tons of manure
generated within each ring around the house. One issue that is addressed in this project is the
relationship between the impact of animal production on house price and the scale of animal
production near the house. Second, both the Michigan and the North Carolina studies are
restricted to hog operations. This project includes poultry, swine, and beef and dairy operations.
Finally, the Michigan and North Carolina studies investigated the impact of animal production
on house price in isolation. This study estimates the impacts from several potential local
disamenities simultaneously, as well as from open space versus developed land use.

Several studies have investigated the impact of other types of local disamenities on residential
property values. Boyle and Kiel (2001) review several of these that show negative impacts on
property values from landfills, airports, and high-traffic roadways. Brisson and Pearce (1995)
review several studies that estimate the impact from hazardous and municipal waste facilities.
These studies show that municipal waste landfills, airports, and high-traffic roadways can have
localized negative effects on house prices.

6
IIe. Previous Studies of Location of Open Space Development

Many researchers have modeled the conversion of suburban and exurban land parcels from
agricultural or forest uses to residential use in different areas and different context. For the
purposes of this study, we are particularly interested in studies that have focused on the role
played by spatial interactions among parcels in the conversion process.

Irwin and Bockstael (2001a) model land development (conversion from open space uses to
residential uses) in Maryland, using parcel- level data. They track wha t happened to developable
parcels over a six year period in four Maryland counties, and model the probability of survival
(non-development) over the period, using a proportional hazards survival model. They find that
developable parcels located in areas with more surrounding open space are more likely to
develop. They conjecture that this is because of inter-parcel externalities. Parcels surrounded by
open space will be more highly valued in residential use, and will therefore experience higher
developme nt pressure, ceteris paribus, than parcels surrounded by developed uses. This
tendency to develop in areas with more open space could lead to classical “sprawl” features such
as patchy patterns of development and leapfrogging.

One key assumption underlying their explanation remains untested, however. Irwin and
Bockstael argue that parcels with higher potential residential value (after development) will tend
to be developed before parcels with lower potential residential value, holding constant
construc tion costs and opportunity cost of land. They then argue that inter-parcel externalities
generate higher potential residential values for parcels located in areas with more open space.
But do parcels with higher potential residential value develop first? In this project, we test this
assumption, by estimating a discrete choice model of development, but with potential residential
value as the key explanatory variable, rather than surrounding land use. In this way, we can test
whether spatial interactions influence development directly, or through their role in determining
potential residential value.

III. Selection of the Study Area

IIIa. Criteria for Selection of the Study Area

The focus of this project was on Southeastern Pennsylvania. Seven counties in Southeastern
Pennsyvlvania (Berks, Bucks, Chester, Delaware, Lancaster, Lebanon, and Montgomery)
account for 47% (by value) of the state’s crop production and 39% of the state’s livestock and
livestock products production (PASS 2003), even though they are located in close proximity to
the state’s largest city (Philadelphia). Several of these counties are experiencing rapid loss of
farmland at the same time as they are seeing increases in population (see Table 1).

7
 Table 1. Characteristics of selected Southeastern and Southcentral PA counties
% change in population % of area in farmland loss
population density farmland 1987-1997
2
County 1990-2000 (persons/mi ) (1997) (percent)
Bucks 10.4% 984.6 21.5% -1.9%
Montgomery 10.6% 1553.0 13.4% -26.8%
Chester 15.2% 573.4 36.2% -7.7%
Berks 11.0% 435.0 40.3% -8.9%
Lehigh 7.2% 899.4 41.2% -5.5%
Northampton 8.1% 714.1 32.7% -9.7%
Lancaster 11.3% 496.0 64.5% -3.0%
York 12.4% 422.3 45.1% -6.1%
Adams 16.6% 175.6 53.7% -4.4%
Lebanon 5.8% 332.4 47.7% -5.8%
Sources: U.S. Bureau of the Census and U.S. Census of Agriculture

In contrast to some other states (notably Maryland and Wyoming), where parcel- level GIS data
is available for the entire state in a consistent format, parcel- level GIS data is developed and
maintained at the county level in Pennsylvania, and is not even available for some counties.
Given the resources available for this project, it was feasible to develop a parcel- level database
for only one county.

Several criteria were identified for selecting a suitable county for study. These are

• Policy Relevance - The study area should be currently experiencing land use change,
but should have a fair amount of less-developed land remaining.

• Representativeness - This study is a demonstration project. In order for it to serve as a

useful model for application to other areas, the study area should not be unique in any
important way.

• Variability in land use patterns - The study area should have diversity not only in land
use, but in the spatial pattern of land use.

• Simplicity in growth patterns - It will be easier to discern the influence of land use
pattern on growth if there are fewer urban centers in or surrounding the study area.

• Data Availability - Geocoded data on land use should already be available, as well as
good data on property sales and characteristics.

• Willingness of local authorities to work with the project.

Berks County scored well in all criteria. Of particular importance were the following
considerations:

8
 • Berks County has been active in protecting farmland from development through
agricultural conservation easements and the use of zoning. Further, existing easements
are geographically dispersed, allowing statistical identification of their impact on
residential property values.

• Berks County has been experiencing a moderate level of land use change during the
past decade, but still has a fair amount of open space remaining.

• Parcel- level GIS maps of the county exist, with complete data on parcel characteristics
(from the county assessor’s office)

• Penn State’s Land Analysis Laboratory already had good working relationships with
several important county offices in Berks County.

IIIb. Characteristics of Berks County

IIIb1. Overview of Berks County Agriculture and Land Use

Berks County occupies an area (864 square miles) between Philadelphia and Harrisburg in
southeastern Pennsylvania. Between the Blue Mountain in the north and the South Mountain
Ridge, lies a valley with rich agricultural resources. In the early 18th century, the mild climate,
fertile soil and the Schuylkill River combined to attract immigrant groups, including the Amish
and Mennonites, to farming. The area became known as the “bread basket” for Philadelphia and
southeast Pennsylvania (Myers and Auchenbach 2002).

Today in Berks County, farming remains a very important sector amidst a suburbanizing
country-side and a diversifying economy. Currently, about 40% of the county’s land is devoted
to agriculture. As of 1997, Berks County had total farm sales of almost $248 million. It ranked
third in Pennsylvania in number of farms, cash receipts from agriculture products, layers, swine,
corn grain, soybeans, and apples. It ranked fourth statewide in dairy, broilers, cattle and calves,
peaches, nursery and greenhouse crops (includes mushrooms), and barley. Other selected
agricultural statistics for the county included: 221,511 acres in farmland, 187,645 acres in crop
production, and 1,586 farms. Animal agriculture is significant to Berks County’s agriculture.
52% of the market value of agricultural products sold is livestock. In addition 35% of the market
value is nursery and greenhouse, including mushrooms. Mushrooms are the largest market value
crop grown (US Census of Agriculture 1997).

Parallel to the agricultural growth in the 18th and 19th centuries was industrial development and
population increases in Berks County. The 1900’s found the county’s population doubling with
it concentrated in Reading and several larger towns. Recent growth patterns reflect suburban
sprawl outward from Reading as well as development in rural land beyond suburban areas.
Surbuban growth is related to economic activities in Philadelphia, Allentown and Lancaster
(Myers and Auchenbach 2002).

9
IIIb2. Berks Land Use and Farmland Protection Programs

Population shifts and suburbanization are affecting the viability of farming as well as the rural
landscape in Berks County. Recently, suburban sprawl and the changing agricultural industry
have been factors leading to increasing conflicts between rural residents and agriculture
production over issues such as odor, flies, chemical use, and farm traffic. To reduce rural- urban
conflict and increase the viability of the county’s agricultural industry, Berks County has
developed a suite of land use management tools to encourage landowners and municipalities to
use the laws available in Pennsylvania for protection of the farming and related industries (Myers
and Auchenbach 2002). These include:

1. Pennsylvania’s Right to Farm Law, which offers protection from nuisance action;
2. Clean and Green which assesses land according to its use—not according to developmental
value. (The county has 244,727 acres in this program.)
3. Agriculture Security Area (ASA) Program—a local municipality established area providing
benefits of protection from ordinances restricting normal farm structures and practices; land
condemnations; and the requirement of being located in ASA’s to be considered for the
purchase of conservation easements. (Thirty- four townships with 139,254 acres are enrolled
in this program.)
4. Purchase of Agr iculture Conservation Easements. Landowners are paid for their land’s
development rights permanently preserving it for agriculture production. In Berks County,
31,372 acres on 256 farms are covered by such easements
5. Effective Agriculture Zoning – A local land use planning tool that is authorized for
municipalities by state law. Local governments can provide for or allow land uses including
agriculture. Twenty townships in Berks County have Effective Agricultural Zoning,
covering 118,000 acres.

Though all of these acts play an important role for the preservation of farms and/or farmers and
the agriculture ind ustry, only the “Purchase of Agriculture Conservation Easements” and
development of “Effective Agriculture Preservation Zoning” are viewed as providing the
agriculture resource base needed for future production. In its 1991 county comprehensive plan,
the county set the goal of preserving 200,000 acres of farmland through these two programs.
Specifically, the county desired to preserve large contiguous areas (minimum of 500 acres) with
existing agricultural productivity. (Factors for inclusion in this area were prime/unique soils,
existing farms, land in programs administered by the Farm Service Agency, agricultural security
areas, agricultural eased parcels, effective agricultural zoning.) In addition, the Planning
Commission initiated an Agricultural Zoning Incentive Program in 1997 to encourage municipal
adoption of effective agricultural zoning by paying costs associated with amending or revising
the zoning ordinances (Myers and Auchenbach 2002).

IIIc. Project Local Advisory Group

The Project Local Advisory Group was formed to fulfill two functions. Early in the project, the
advisory panel provided input to the research team, including local knowledge of factors
influencing land use and values. The advisory panel also helped to identify the policy analysis
needs of the community. In addition to telephone and email exchanges between the researchers

10
and the advisory group, members of the research team met with the advisory group, both
individually and with the group as a whole.

Toward the end of the project, the research team presented preliminary research results to the
advisory panel. This meeting helped the research team refine its analysis to better meet local
policy analysis needs, and to refine the presentation of the research results to make them more
accessible and useful to local stakeholders.

After completion of the project, research team members will present final results to the advisory
group, and other interested parties in Berks County, both through presentations and through
extension materials developed from this final report.

The members of the Project Local Advisory Group are listed in Appendix 1 of this report.

IV. Data Sources and Manipulation

IVa. Data Based on Parcel Maps

The most important database for this analysis was the 2002 parcel map of Berks County. This
database, maintained by the Berks County Office of Assessment, includes a map showing the
boundaries of 152,809 parcels in the county, as well as of roads and streams. A second database,
contained in the Office Of Assessment’s CAMA file, includes information collected for each
parcel for assessment purposes, such as characteristics of built structures on each parcel, the
current use of the parcel, and the parcel’s owner.

IVa1. Single-Family Residential Properties

Special attention was paid to parcels that were identified as single family residences, as it is these
parcels for which the hedonic price function was estimated. For each of the 88,798 single- family
residential parcels, a parcel centroid was located. This point dataset was for calculating distances
to features in the County, and as centerpoints of circular buffers for measuring surrounding land
use.

From this set of parcels, a subset was chosen for use in the hedonic price model estimation.
First, because the analysis focused on property values on the urban/rural fringe, residences
located within the City of Reading were excluded. Second, in order that surrounding land use
could be completely characterized for each residence in the analysis, residences within one mile
of the county border were excluded. Third, because land use in New Morgan Borough is
somewhat unique (about ¾ of the Borough is zoned Industrial), properties located within one
mile of the Borough were excluded. Fourth, several restrictions were used to exclude “unusual”
properties that a hedonic price model would have difficulty valuing. To be included in the
hedonic price analysis, the property must meet the following criteria:

- lot size at least 0.1 acres - to avoid cases where the lawn surrounding a house is covered
by a separate deed

11
 - lot size no larger than 5 acres - to avoid cases where the land has multiple uses
- living area at least 600 square feet - to avoid structures misclassified as detached houses
- physical condition “poor” or better - to avoid condemned or damaged properties

Fifth, the analysis was limited to sales that occurred in 1998 or later. Finally, to assure that the
listed sale price included the house and not just the land, the analysis included only houses where
the assessed value was within 20% of the sale price. Also, only sales that were judged as “arms
length” by the assessor were included in the hedonic price analysis. This set of restrictions
resulted in 8,090 residential sales that were included in the hedonic price analysis.

For each of the 8,090 residential properties used in the hedonic price analysis, information was
extracted on the size of the house, the lot size, the number of bedrooms, the number of
bathrooms (including half baths), whether the house has a basement, whether some of the
finished area in the house is located in an attic, the exterior façade of the house, whether the
house has central air conditioning, the physical condition of the house (an index from 1 to 5, with
1 being the best, and 5 being the worst), the year of construction, the year sold, and whether the
house had public water and/or public sewer. The nominal sale price of the house was inflated to
2002 dollars.

IVa2. Land Use

Several features of parcel database were useful for our purposes. First, the assessor’s land use
codes served as a starting point for constructing a land use map of Berks County. For most
parcels, the land use code was sufficient for classifying each parcel into one of the following 12
land use categories:

Agricultural and other open space uses, including parks, golf courses and water bodies
Vacant land zoned for residential use
Vacant land zoned for commercial use
Vacant land zoned for industrial use
Single Family Residential with lot size < 0.2 acres
Single Family Residential with lot size > 0.2 and < 0.5 acres
Single Family Residential with lot size > 0.5 and < 1.5 acres
Single Family Residential with lot size > 1.5 acres
Other Residential (multi- family and various other types including mobile homes)
Commercial
Industrial
Transportation (Roads and Railroads)

Landfills were classified as an industrial use. Mushroom production facilities were classified as
an agricultural use. Parks, golf courses, and water bodies were classified as open space uses.
In some cases, the land use code was not specific enough for categorization. This occurred, for
example, for tax exempt properties, where the assessor did not need to keep track of land use.
For these parcels, other information was used to categorize the parcels use. For example, if the
parcel was owned by a school district, its use was assumed to be similar to a commercial use.
Likewise, parcels owned by religious organizations were categorized as commercial use. If the

12
parcel was owned by a private hunting and fishing club, it was assumed to be in an open space
use.
Where land use could not be classified based on the assessor’s land use code or other clues from
its ownership, its use was determined based on its land cover. The 1992 National Land Cover
Database categorizes land into 21 land use classes, of which 14 occur in Berks County, based on
satellite images collected during the early 1990’s. These 14 classes were divided into two
groups, developed-use land covers and undeveloped-use land covers (NLCD codes 21, 22, and
23). For each uncategorized parcel in the Berks County parcel database, the proportion of land
area in developed use was calculated. If more than 10% of the parcel was classified as
developed by the NLCD, then the parcel was considered to be a commercial parcel. It less than
10% of the parcel was classified as developed by the NLCD, the parcel was considered to be in
an open space use. This approach was applied to 1807 parcels (1.2% of the total), affecting
21,514 acres (3.9% of the total land in the parcel database). An additional 217 parcels totaling
39 acres could not be assigned using this approach. No land use code was assigned for these
parcels.

After each parcel was categorized into one of the 12 land use groups, transportation parcels were
dissolved into neighboring parcels. This approach was taken to better represent the amenity
impacts of different types of neighborhoods. A small- lot residential neighborhood may have as
much as 20-25% of its land in transportation uses. When considering amenity impacts, such a
neighborhood is not typically viewed as 75% residential and 25% transportation. A “residential
neighborhood” is viewed as a mix of houses, lawns, and streets. Similarly, from an amenity
viewpoint, a commercial area is viewed as a mix of shops, offices, parking lots, and streets. In
order to better match how these neighborhoods are perceived, land used in transportation was
assigned to the same category as the nearest non-transportation parcel. Thus, a street with
residential properties on both sides was categorized as residential land use. A street with
residential properties on one side and commercial properties on the other was split down the
middle.

The amount of land in Berks County categorized into each land use category using this method is
given in Table 2.
Table 2. Land Use in Berks County, 2002
Number of Percent
Land Use Category Parcels Acres Area
Ag and Open Space 13,032 383,159 69.1%
Vacant Commercial 531 3,156 0.6%
Vacant Industrial 108 1,444 0.3%
Vacant Residential 13,097 23,619 4.3%
Single Family Xlarge Lot 14,715 55,698 10.0%
Single Family Large Lot 15,747 16,745 3.0%
Single Family Medium Lot 28,999 11,394 2.1%
Single Family Small Lot 30,175 4,961 0.9%
Other Residential 25,278 9,170 1.7%
Commercial 8,733 24,362 4.4%
Industrial 2,177 20,933 3.8%
No Data 217 39 0.0%
Total 152,809 554,680 100.0%

13
Notice that open space uses (the first four categories) account for 74.3% of land in Berks County.
Of the “vacant” land, the majority is vacant residential parcels. These are typically parcels that
have been subdivided and zoned for residential use, but that have not yet had houses built on
them. The location of residential, commercial and industrial land is shown in Plate 1. Clear
areas in Plate 1 represent open space.

For each house in the hedonic price analysis, circular buffers were constructed at 400 meters and
1600 meters, and the total number of acres in each land use category was measured for each
buffer. These distances were chosen to represent two different conceptions of the
“neighborhood” the house is located in. A resident might easily travel 400 meters from his or
her house while taking a walk, while his or her child might walk up to 1600 meters to get to
school. Several previous studies used 400 meters as the definition of the neighborhood. We
investigate whether more distant land use also impacts house prices.

IVa3. Agricultural Security Areas and Eased Parcels

Two map databases were obtained from the Berks County Agricultural Land Preservation Board.
The first identifies all parcels in the County that were included in Agricultural Security Areas
(ASA). A total 131,647 acres in Berks County are in ASA’s. The second identifies all lands that
with Agricultural Conservation Easements (ACE). The ACE database included 29,543 acres of
land. All parcels with ACE’s must also be in an ASA. The ACE database was combined with
another map database that identified conservation easements he ld by other bodies, such as the
Berks County Conservancy (3,476 acres). The location of land in ASA’s and land with
conservation easements is shown in Plate 2.

For each residential property in the hedonic price analysis, the number of acres of eased open
space and the number of acres of open space contained within ASA’s within 400 meters and
between 400 meters and 1600 meters from the house were calculated.

IVa4. Government-Owned Open Space

It is important to identify open space land that is owned by the government because the use of
such land is less likely to change over time. Government-owned open space was identified using
both the Managed Lands Database developed in 1998 by the Penn State Environmental Resource
Research Institute (ERRI) and information contained in the County Assessement Office’s parcel
database. 1,623 parcels totaling 53,986 acres were identified as parcels that were owned by
Federal, State, or local government, and that were in open space use. The location of
government-owned open space is shown in Plate 3. Several large tracts accounted for the
majority of this land. An Army Corps of Engineers reservoir, Blue Marsh Lake, is located west
of Reading. A large water supply reservoir, Lake Ontelaunee, is located north of Reading.
French Creek State Park is located in the southeastern part of the county, and a large portion of
the ridge that defines the northwestern border of the county (Blue Mountain) is owned by the
Pennsylvania Game Commission.

14
For each residential property in the hedonic price analysis, the number of acres of government-
owned open space within 400 meters and between 400 meters and 1600 meters from the house
was calculated.

IVa5. Developable and Developed Parcels

The probability-of-development analysis to model residential land conversion started with the
County Assessment Office’s 1996 parcel map. A similar approach to that outlined above was
used to identify 7,399 parcels that existed in 1996 that could be developed. These were parcels
that were

- in open space land uses

- were not owned by a government body
- did not have a conservation easement on them
- were at least 5 acres in size
- were not located within 1 mile of the county boundary, or within one mile of New
Morgan Borough

To simplify the analysis, parcels that were eased after 1996 were also considered as “not
developable” in 1996. A more complete analysis that models the decision to develop versus the
decision to sell an easement would have to model competition for land between residential
developers and the Berks County Agricultural Land Preservation Board, a task that was beyond
the scope of this project.

A developable parcel was defined to have “developed into residential use” if its owner requested
subdivision, with at least three of the subdivided parcels rezoned for residential use. From a list
of maintained by the Planning Commission, 92 parcels were identified that met this criterion.
All of these parcels were considered to have developed, even though not all of them parcels had
completed the process by 2002. In this way, the analysis focuses on the factors that trigger the
subdivision request, rather than on the factors that result in successful subdivision.

The location of all developable parcels used in the analysis is shown in Plate 4. Parcels that
subdivided are shown in red.

IVb. Physical and Political Features

Supplementing the parcel-based databases were several GIS databases that included information
on physical features of the landscape and on boundaries of political jurisdictions.

The 1992 NLCD database was discussed briefly above. In addition to its use in categorizing
land use for parcels, this database provided information on type of open space. Land categorized
as open space was subdivided into three types based on the NLCD database, forests; grass,
pasture, and crops; and open water. For each residential property in the hedonic price analysis,
the number of acres of open space within 400 meters and between 400 meters and 1600 meters
that is in grass, pasture and crops was calculated.

15
Several databases were obtained from the Pennsylvania Geospatial Data Clearinghouse. These
included the USGS Digital Elevation Model for Berks County, from which were calculated
elevation and slope, a digital map of streams in Berks County developed by ERRI, and school
districts and municipal boundaries. Slope and elevation at the house site were included in the
hedonic price function, as was a measure of elevation relative to the surrounding terrain. This
measure was the difference between the elevation at the house site, and the average elevation
within 800 meters of the house site. Positive values of this difference indicate that the house sits
above the surrounding terrain.

A zoning map of the county was provided by staff from the Berks County Planning Commission.
Zoning regulations were grouped into six categories, Agricultural, Effective Agricultural (with
more stringent restrictions on the ability to subdivide parcels), Commercial and Industrial,
Multiuse/Village, Conservation (for erodible and other naturally sensitive areas), and
Residential.

Information on soils was obtained from a digitized soil map of Berks County, created by the
USDA Natural Resources Conservation Service. For each soil type, three relevant measures
were obtained, an index of the soil’s suitability for construction (ranging from 0 to 10, with 10
being soils least suitable for construction), an index of the soil’s suitability for septic systems
(ranging from 0 to 10, with 10 being soils least suitable for septic systems), and index of the
soil’s potential for agricultural production (ranging from 0 to 100, with 100 being soils best
suited for agricultural production). The septic suitability and construction suitability indices are
published with the county soil survey. The agricultural productivity index is developed by the
Pennsylvania State Office of the Natural Resources Conservation Service. For soil map
polygons with complexes of multiple soil types, area-weighted averages were used for the three
indices.

Three employment centers were identified based on conversations with the Local Project
Advisory Group. These are Reading, Allentown, and Philadelphia. For purposes of calculating
distance to downtown Reading, the city center was taken to be located at the corner of Penn
Street and Third Street. For Philadelphia, it was assumed that most commuters from the County
would travel through one of two points, where Route 422 crosses the County boundary heading
toward Philadelphia or where I-76 (The Pennsylvania Tollway) crosses the County boundary
heading toward Philadelphia. As these two points are about the same distance from Philadelphia,
for points within Berks County, commuting distance to Philadelphia was taken to be the linear
distance to the closer of these two points. A similar approach was taken for measuring the
distance to Allentown, with two waypoints, one located where Route 737 crosses I-78, the other
located where Route 222 crosses the northeastern border of the County.

IVc. Potential Local Disamenities

The project advisory group recommended that the analysis allow for measurement of the impact
of potential local disamenities, including landfills, mushroom production, and large-scale animal
production facilities. Additional financial support from Citizens for Pennsylvania’s Future
allowed collection of information on the location of several different types of potential local
disamenities.

16
IVc1. Landfills

Landfills can be a source of disamenities to nearby neighbors. These can include concerns over
groundwater contamination, noise from trucks and machinery, dust, blowing trash, birds, and
odor. There are four landfills located within Berks County (BFI Conestoga Landfill in New
Morgan Borough, Western Berks Refuse Authority Landfill in Cumru Township, Rolling Hills
Landfill in Earl Township, and Pioneer Crossing Landfill in Exeter Township). A fifth landfill
located just outside Berks County in Pottstown was not included in the database. The four
landfills were mapped from the parcel database. For each residential property in the hedonic
price analysis, linear distance was calculated to the boundary of the closest landfill (LFDIST).
The locations of the landfills are shown in Plate 5.

The amenity impact of landfills on residential properties is expected to be larger for properties
located nearer to a landfill. We assume that past some distance, K, the landfill has no amenity
impact on the residential property. An index of the landfill amenity impact (LFIND) was
constructed of the form

LFIND = 1/(LFDIST) - 1/K if LFDIST < K

=0 if LFDIST > K

This index decreases as distance to the landfill increases, and reaches value 0 at distance K from
the landfill.

To test the spatial limit of the amenity impact from landfills, the Landfill Index is further divided
into two pieces, one measuring impacts up to a distance K1 from the landfill, the second
measuring impacts between distance K1 from the landfill and distance K2 > K1 . These indices
take the form

LFIND1 = 1/(LFDIST) - 1/K 1 if LFDIST < K1

=0 if LFDIST > K1

LFIND2 = 1/K 1 if LFDIST < K1

= 1/(LFDIST) - 1/K 2 if K1 < LFDIST < K2

=0 if LFDIST > K2

LFIND1 is the same as LFIND, with K=K1 . LFIND2 is constructed so that

LFIND1+LFIND2=LFIND with K=K2 . In the hedonic price model, if the coefficient on
LFIND1 is negative and the coefficient on LFIND2 is zero, then the disamenity impact extends
only to distance K1 . If the coefficient on LFIND2 is equal to the coefficient on LFIND1, then the
amenity impact extends to distance K2 .

17
The choice of which values of K1 and K2 to investigate depends on a priori expectations about
the spatial extent of any amenity impacts. Nelson, Genereux and Genereux (1992) found
negative impacts on house prices from a landfill that extended to two miles from the landfill. We
use as starting points K1 =1600 meters and K2 =3200 meters, to test whether any impact of
landfills on house prices extends as far as two miles.

IVc2. Airports

Airports can be a source of noise problems to nearby neighbors. However, close proximity to the
airport may be an amenity for households who travel frequently. There is one commercial
airport in Berks County, located in Bern Township (see Plate 5). This airport is served by
propeller-driven commuter aircraft. The local disamenity impact from airport operation is best
measured by looking at the noise level at the residential property. Unfortunately, information
was not available on the distribution of aircraft noise surrounding the airport. Instead, the take-
off and landing flightpaths were approximated with a straight line extending two miles from
either end of the main runway. For each residential property in the hedonic price analysis, the
linear distance to this line was calculated. A set of airport indices, APIND, APINDK1, and
APINDK2 were constructed in the same way as for the landfill indices. As with landfills, we use
as starting points K1 =1600 meters and K2 =3200 meters.

IVc3. Mushroom Production

Mushroom production can also be associated with odor problems. From the County Assessment
Office’s parcel map, 74 parcels were identified where mushroom production occurred. The total
acreage of these parcels is shown, by township, in Plate 6. The mushroom production industry in
Berks County has been shrinking, and many of these parcels are no longer in active production.
Working with Mena Hautau, the county horticulture extension agent, and David Beyer,
Associate Professor of Plant Pathology at Penn State, an attempt was made to determine which
of the 74 operations were still active during the period of study, 1998-2002. For each residential
parcel in the hedonic price analysis, the distance to the closest parcel that was actively producing
mushrooms at the time of the residential property’s sale was calculated. A set of mushroom
production indices MRIND, MRINDK1, and MRINDK2 were constructed in the same way as
for the landfill indices. As a starting point for investigating the spatial extent of any potential
local disamenity, we use K1 =400 meters and K2 =1600 meters.

IVc4. Large-scale Animal Production

Large-scale animal production facilities can be a source of problems with odors, flies, and
concerns over ground- and surface-water contamination. Working with aerial photos, and
assisted by County Extension Agent Clyde Meyers and Donald Reinert, Nutrient Management
Specialist with the Berks County Conservation District, 71 large-scale animal production
operations (200 animal equivalent units or more) were identified. An operation was defined as a
cluster of buildings located within 400 meters of each other. A single operator could control
more than one building cluster. For each building cluster, the number of animal equivalent units
(aeu’s) was determined for three groups of livestock, poultry, swine, and cows (beef and dairy).

18
Animal equivalent units are as defined for each species by the Pennsylvania Nutrient
Management Act (Act 6 of 1993), and represent 1,000 pounds of live weight (Beegle, 1997).
Important changes in the scale of operation during the study period (1998 to 2002) were noted,
so that data on scale of operation (number of aeu’s) was recorded for each year. The number of
facilities located in each township is shown in Plate 7.

When modeling the amenity impact of animal production, it is not clear whether that impact
should be proportional to the scale of nearby operations. It is an empirical question whether a
building cluster with more aeu’s has a larger disamenity impact on nearby properties than a
building cluster with fewer aeu’s. Second, it is an empirical question whether having two
building clusters near the home is worse than having one. Finally, different species of animals
may have different disamenity impacts (e.g. air quality or odor emissions), as a result of animal
production methods or manure characteristics (e.g. wet versus dry). We construct several animal
production indices to investigate these issues. In the following definitions, ANDIST1 measures
the distance from the house to the nearest building cluster, ANDIST2 measures the distance to
the second nearest building cluster, etc.

IVc4i. Animal Indices Based on Building Clusters. The first animal production index is based
on the distance from the homes to each of the building clusters

ANBCIND = ? i 1/(ANDISTi) - 1/K

where the summation is over all building clusters located within K meters of the house. Again,
this index can be further divided by distance, into ANBCINDK1 and ANBCINDK2. As with
landfills, we use as starting points K1 =1600 meters and K2 =3200 meters. Under certain
conditions, odor problems from large-scale animal production can occur up to a mile from the
source (VanDevender 1998, Mikesell 2002), but it is not clear that property value impacts would
be detected at that distance.

To investigate whether the disamenity impact depends on only the closest building cluster, or
whether it depends on all building clusters within K meters from the house, the ANBCIND index
was divided into two pieces

ANBCINDC = 1/(ANDIST1) - 1/K if ANDIST1 < K

=0 if ANDIST1 > K

and

ANBCINDO = ANBCIND - ANBCINDC

where the last letter signifies C=closest building cluster, and O=all other building clusters within
K meters of the house. In the hedonic price estimation, if the coefficient on ABCINDO is zero,
then it is only the closest building cluster that generates disamenity impacts. If the coefficient on
ABCINDO is equal to the coefficient on ABCINDC, then all building clusters contribute equally
to the disamenity impact (given their distance from the house), and it is appropriate to use
ABCIND as the single index of that disamenity.

19
Finally, because animal production is of particular interest in this study, the functional form of
the index is investigated in more detail than for other potential local disamenities. A second
index is constructed that uses squared distances

ANBCIND2 = ? i 1/(ANDISTi)2 - 1/K 2

The two indices together provide a more flexible functional form for the relationship between the
disamenity impact and distance to the building clusters.

Additional analyses were performed using building clusters as the unit of observation. To
investigate size of operation, the ANBCIND index was further split into two size classes,
medium sized operations (> 200 aeu but < 300 aeu) and large operations (> 300 aeu). To
investigate managerial care related to conservation, ANBCIND was split into two groups based
on whether the farm had on file with the NRCS a detailed conservation plan. This is an
imperfect indicator of managerial care, but was the only objective measure available.

IVc4ii Animal Inidices Based on AEU’s. A second set of indices was developed based on the
number of aeu’s in each building cluster. The main aeu-based index was

ANAEUIND = ? i AEUi/(ANDISTi) - 1/K

where AEUi is the number of aeu’s in the ith closest building cluster in the year the house was
sold, and the summation is done over all building clusters located within K meters of the house.
This index retains the distance decay feature of the building cluster index, but assumes that the
disamenity impact is directly proportio nal to the number of aeu’s in each building cluster.

This index was further divided into three pieces by species of animal, POULTRYAEUIND,
SWINEAEUIND, and COWAEUIND, where each index includes only the number of aeu’s for
that species.

Finally, a hybrid index was constructed that used building clusters as the unit of interest, but that
weighted those building clusters by animal species. These took the form

POULTRYBCIND = ANBCIND * POULTRYAEUIND/ANAEUIND

SWINEBCIND = ANBCIND * SWINEAEUIND/ANAEUIND

COWBCIND = ANBCIND * COWAEUIND/ANAEUIND

These hybrid indices assume that the disamenity impact is proportional to the number of building
clusters, rather than to the number of aeu’s in those clusters, but allows consideration of whether
the impact is sensitive to the species housed in the clusters.

20
IVc5. Sewage Treatment Plants

Sewage treatment plants can also be associated with odor problems. From EPA and other
sources, 27 sewage treatment plants were identified in Berks County. These were located using
aerial photos, the parcel map, and information contained in the EPA NPDES database. For each
residential property in the hedonic price analysis, the distance to each sewage treatment plant
was calculated. A sewage treatment plant index, SPIND, was calculated using the same
approach as for ANBCIND. Only one distance limit, K=1600 meters, was investigated.

IVc6. High-Traffic Roads

High-traffic roads can be a source of problems with noise and air quality. At the same time,
access to main roads is an important amenity, because it reduces travel times to shopping,
recreation, and work sites. A database of Federal, State and County Roads was obtained from
the PennDOT. Roads in Berks County with more than 10,000 vehicles per day were identified,
and the distance to the nearest high traffic road was calculated for each residence in the hedonic
price analysis.

The potential noise and pollution disamenities from living in close proximity to a main road is
conjectured to extend only a short distance, likely less than 100 meters. Access and convenience
issues are likely to be relevant at all distances from main roads. To capture both effects, the
distance to the nearest high-traffic road (HTDIST) entered the hedonic price analysis in two
ways. First, the disamenity impact was modeled using a index similar to that used for the other
potential local disamenities in the study
HTIND = 1/(HTDIST) - 1/K if HTDIST < K

=0 if HTDIST > K

where K was set equal to 100 meters. Second, distance to the nearest main road was entered
linearly. If the coefficient on both terms is negative, then we confirm that proximity to main
roads is a disamenity for distances less than 100 meters, but an amenity for distances greater than
100 meters.

V. Hedonic Analysis of Residential Property Values

Va. Details of the IV estimation

The hedonic price function was estimated using both IV estimation and OLS regression. The
dependent variable in both cases was ln(real house price). In the IV estimation, the following
variables were treated as endogenous. Each entry in the list represents two variables, one for
land use within 400 meters of the house, the other for land use between 400 meters and 1600
meters from the house.

- acres of land in open space uses

- acres of land in residential uses

21
 - acres of land in commercial use
- the proportion of open space that is vacant land

The omitted land use category in all estimations is Industrial, so all coefficient values should be
viewed as the marginal difference between the land use in question and industrial use.

In addition to the variables listed above, SPIND is also treated as endogenous. This modeling
choice was motivated by the observation that in an OLS estimation of the hedonic price model,
the sewage treatment plant index, SPIND, entered the function with a significantly positive
coefficient, suggesting that sewage treatment plants are a positive amenity for home owners.
This even though service by public sewers is already accounted for in the model. This result
may be due to endogeneity in the decision of where to build sewage treatment plants. The choice
of where to build a new sewage treatment plant involves several considerations, one being
proximity to housing. If sewage treatment plants tend to be built in areas where demand for
housing is strong, then their location could be, at least in part, endogenous to house price.

Within the broad land use categories of Open Space and Residential Use, there are subcategories.
With the exception of Vacant Land, which is treated as endogenous, the following variables are
treated as exogenous

the proportion of open space that is government owned

the proportion of open space that is in an ASA
the proportion of open space that has a conservation easement
the proportion of open space that is in grass, pasture, or crops
the proportion of residential land in small- lot single family use (< 0.2 acres)
the proportion of residential land in medium- lot single family use
(> 0.2 acres and < 0.5 acres)
the proportion of residential land in large- lot single family use
(> 0.5 acres and < 1.5 acres)
the proportion of residential land in xlarge- lot single family use (> 1.5 acres)
the proportion of residential land in uses other than single family housing

Thus, we assume that how much land remains in open space use in a given location is
endogenous to house price in that area, but that the type of open space (forested vs grassy, eased
vs noneased, ASA vs nonASA) is exogenously determined. Similarly, the amount of land in
residential use is endogenous to house price, but the mix of types of residential use is taken as
exogenous. The exception is for vacant land. Most vacant open space land is zoned for
residential use. The speed with which that land is deve loped is likely determined in part by the
demand for houses (and the price of houses) in that area. It is reasonable to think, therefore, that
the proportion of open space in vacant use will be endogenous to house price.

Other exogenous variables included in the hedonic price regression are structural characteristics
of the house, distance of the house from downtown Reading, distance from commuting
waypoints to Philadelphia, and distance to commuting waypoints to Allentown, soil slope at the
house site, elevation, a measure of elevation relative to the surrounding terrain, zoning (with

22
dummies for zoning type, with Residential as the excluded type), and the disamenity indices
(other than SPIND, which as treated as endogenous).

Instrumental variables are exogenous variables that are not included as explanatory variables in
the hedonic price function. These are variables that help explain the endogenous variables, but
do not help directly explain house price. Each of the following instrumental variables was
calculated both for land within 400 meters of the house and for land between 400 meters and
1600 meters from the house,

- average slope
- average elevation
- average septic suitability index
- average building suitability index
- average agricultural productivity index
- proportion of open space surrounding the house that is in an ASA

Each of these instruments affect either the return from agricultural production (and thus the
opportunity cost of residential development) or the cost of residentia l construction. They should
therefore help explain variation in the amount of land in open space uses. In addition, distance to
the nearest stream was included as an instrumental variable, to help explain location of sewage
treatment plants.

A nonlinear IV approach was used (SAS PROC MODEL with N2SLS option). In this
procedure, the endogenous variables are first regressed, using OLS on all of the exogenous and
instrumental variables. The resulting OLS predicted values of the endogenous variables are then
used to calculate the acres of each different type of Open Space and Residential use in the
buffers surrounding each house. The second stage regression uses these calculated measures of
surrounding land use as explanatory variables.

While the estimation does account for endogeneity in land use, it does not model spatial error
correlation. The model results presented in this section assume that prediction errors are not
spatially correlated.

Vb. Regression Results - House Characteristics

Results of the IV and OLS regressions are presented in Tables 3a through 3c. The R-square for
the IV estimation was 0.8645. The R-square for the OLS regression was 0.8739.

Because it is theoretically more valid, we focus on the results of the IV estimation. All of the
coefficients for the structural characteristics of the house were significantly different from zero,
and of the expected sign. With the log-linear model used here, the coefficient values can be
interpreted as the percent increase or decrease in house price associated with a one-unit change
in the explanatory variable. So, for example, an increase in house size of 1 square foot increases
the value of the house by 0.023%. Likewise, one extra bathroom increases the value of the house
by 6.67%. The following characteristics are associated with higher house price: more
bedrooms, existence of a basement, a brick, stone or masonry exterior, central air conditioning,

23
better physical condition, newer construction, and location in a school district with higher 12th
grade PSSA average test scores. For a given size, a house is worth less if some of its finished
area is in an attic.

Houses on more sloped lots are worth less. Elevation in and of itself does not influence house
price, but elevation relative to the surrounding terrain does. Houses that sit above the
surrounding terrain are worth more than those that sit below the surrounding terrain. Public
water service increases house value, but public sewer service does not. There is high correlation
between the these two features, however, decreasing our ability to distinguish their individual
effects. House prices did not increase as fast as inflation during the study period (real prices
declined over time).

Shorter commuting distance to Allentown and Philadelphia are associated with higher house
prices, but lower distance to Reading was not seen as a positive amenity.

Of particular interest are the results related to lot size. To provide for a flexible relationship
between lot size and house price, lot size entered the hedonic price function both linearly and as
a natural log. The combined effect of these two terms is shown in Figure 1. In Figure 1, the
price of a house built on a 0.1 acre lot is normalized to equal 1.0. Figure 1 shows how the house
price will increase as the lot size increases. So, for example, a house built on a 1 acre lot will
cost 32% more than the same house built on a 0.1 acre lot. The marginal impact of additional lot
size decreases, however, so that a house built on a 5 acre lot is worth only a little bit more than a
house built on a 3 acre lot. This relationship can help inform developers and planners when
considering density of a new residential development.

1.6
(compared to same house on 0.1 acre lot)

1.5
Relative House Price

1.4

1.3

1.2

1.1

1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Lot Size (acres)

Figure 1. Relative house price as a function of lot size.

24
Vc. Regression Results - Surrounding Land Use

The coefficient estimates in Table 3b represent the marginal impact on house price of one more
acre of each type of land. The total impact of a land use that combines more than one feature
(for example eased open space) is given by the sum of the relevant coefficients. For each land
use type, these summed implicit prices are given in Table 4a and 4b, for both the IV and OLS
regressions. In each case, the marginal implicit price represents the percent impact on house
price from a one-acre change in land use, with industrial use as the baseline.

Within 400 meters of the house, the land use with the largest positive amenity impact is forested,
government-owned open space. However, forested, privately-owned open space has a similarly-
high amenity value, and the difference between the two is not statistically significant. Open
space in grass, pasture, and crops is less valued tha n forested open space, but again the difference
is not statistically significant. Eased open space is less valued than noneased open space, and
here the difference is statistically significant. Vacant open space is the least valued, and has in
fact a more negative impact on land values than does industrial land. A second regression
showed that whether surrounding open space was in an ASA had no impact on its amenity value.

Medium, Large, and Xlarge lot residential development has a positive impact on house price,
relative to industrial use. Differences among these three groups are not statistically significant.
Small- lot residential use and non-single family residential use have lower amenity value, and are
not statistically distinguishable from industrial use.

Commercial land use within 400 meters of the house has a more positive impact than small- lot
and non-single family residential, but the difference between commercial use and industrial use
is not statistically significant.

When comparing among these results, the marginal implicit price of privately-owned, forested
open space is significantly higher than for industrial use, commercial use, small and medium lot
single family residential use, and non-single family residential use. Conversion of privately-
owned forested open space located within 400 meters of a house to any of these uses would
reduce the price of a house. The marginal implicit price for privately-owned grass, pasture and
cropland is significantly larger than that for small lot single family residential, for non-single
family residential, for commercial, and for industrial. The marginal implicit price for eased,
privately-owned grass, pasture and cropland is significantly larger than that for small lot single
family residential, for non-single family residential, and for industrial.

To summarize, within 400 meters of a house, the surrounding land use that has the highest
amenity impact is open space. Whether that open space is forested or in grass, pasture or crops
does not matter much. Whether that open space is owned by the government does not matter
much. If the open space is eased, it has a smaller positive impact on house price. Among
developed land uses, the neighboring land use with the most positive amenity impact is medium
or larger lot single family residential. The land uses with the least positive impact on house price
are small lot residential, non-single family residential, commercial, industrial, and vacant land.

25
A comparison of the IV estimation results to the OLS regression results demonstrates the impact
of endogeneity on the estimated marginal implicit prices of open space uses. The differences
between the amenity value of open space and the amenity value of residential or commercial are
larger for the IV estimation than for the OLS regression. This difference is due to the
endogeneity of the open space.

One curious result is the difference in the value of vacant open space between the two
regressions. The IV estimation shows vacant open space within 400 meters to be a disamenty
relative to industrial use, a somewhat surprising finding. However, the OLS regression shows
the opposite result. An IV estimation that treats the proportion of open space that is vacant as an
exogenous variable also shows vacant open space to be an amenity relative to industrial use
(with the difference statistically significant at the 5% level). If the proportion of open space that
is vacant is endogenous to house price, this pattern suggests that higher house prices increase that
proportion. This could be the case if development is occurring at a higher rate in areas with
higher house prices, so that at any given time there is more land that has already been subdivided
but that has not yet been built upon. Why this land (which is almost all zoned residential) would
have lower amenity value than residential land that has already been built out is unclear. The
disamenity may stem from aversion to uncertainty over exactly how the neighboring residential
land will be developed, or aversion to the noise, dirt, and congestion associated with
construction.

Moving farther from the house (Table 4b) the picture changes somewhat. Marginal implicit
prices for land uses between 400 meters and 1600 meters from the house are generally an order
of magnitude smaller than those for land use within 400 meters. This makes sense not only
because the land use is located more distant from the house, and is therefore less noticeable to
the occupants, but also because one acre of land represents a smaller proportion of the total
located at that distance.

Still, land use between 400 meters and 1600 meters from the house does significantly impact
house price. At that distance, the land use with the most positive impact on house price is
commercial, closely followed by large and xlarge lot residential. Of open space uses, only eased
or government-owned open space has a significantly positive impact on price, relative to
industrial use. In fact, privately-owned, forested open space has a negative impact on house
price. Grass, pasture and crops have a significantly more positive impact than forested open
space, but the difference is small.

Comparing among land uses between 400 meters and 1600 meters from the house, even the
more-highly valued open space uses are significantly less attractive than large or xlarge lot single
family residential use or than commercial use. The only developed use significantly worse than
eased or government-owned open space is industrial.

Marginal implicit prices for land uses located farther than 400 meters from the house should be
interpreted with caution. Land use measures within 400 meters are highly correlated with land
use measures between 400 meters and 1600 meters from the house, making identification of the
marginal implicit prices difficult. One consequence of this collinearity is that the marginal

26
implicit prices for the outer ring are less robust to changes in modeling decisions than those for
measures of land use closer to the house.

To summarize, open space is an amenity to residential home owners, but the amenity value
generates only within 400 meters of the house. Outside of 400 meters, the preferred surrounding
land use is commercial. If open space does exist outside of 400 meters, then eased or
government-owned open space is preferred to privately-owned, uneased open space. The ideal
house, then, is located in the middle of forested open space, with commercial land located 400-
1600 meters away.

Vd. Regression Results - Potential Local Disamenities

Vd1. Roads, Sewage Treatment Plants, Mushrooms, Airports and Landfills

The coefficient on HTIND_100 is negative, and nearly significant, giving weak evidence that
proximity to high traffic roads is a local disamenity. At 20 meters from the road, this disamenity
impact decreases house price by 0.63%. The coefficient on HTDIST is negative and significant,
indicating that close access to major road arteries is an amenity. Past 100 meters from the road,
each additional km of distance from high traffic roads decreases house price by 0.47%.

Proximity to the airport is a local disamenity. When APINDK1 and APINDK2 are included in
the model, with K1=1600 and K2=3200, the former is negative and significant, while the latter is
negative but insignificant (t=-0.93). We conclude that the disamenity impact of the airport and
flight path extends to 1600 meters, but does not extend past 1600 meters.

When proximity to sewage treatment plants is modeled as an endogenous variable, SPIND is not
significantly related to house price. It remains to be determine why SPIND is positively related
to house price when it is treated as an exogenous variable.

Proximity to a landfill is a local disamenity. When LFINDK1 and LFINDK2 are included in the
model, with K1=1600 and K2=3200, both coefficients are significantly negative at the 5% level,
and they are not significantly different from each other. We conclude that the disamenity impact
from landfills extends to 3200 meters. Rerunning the analysis with K1=3200 and K2=4800, the
coefficient on LFINDK2 is no longer negative, and the two coefficients LFINDK1 and LFINDk2
are significantly different. We conclude that the disamenity impact from landfills does not
extend beyond 3200 meters.

Proximity to mushroom production is a local disamenity. It was not possible to conclusively

establish an outer boundary of the impact of mushroom production on house prices. Inclusions
of two mushroom indices gave unstable results. However, estimation using only one index gave
very similar results, regardless of the outer limit chosen. Estimations were run with MRIND
with an outer limit of 400 meters, 800 meters, 1600 meters, and 3200 meters. The coefficient on
MRIND is negative and significant for all outer limits, and ranges in size from -5.05 to -6.70. We
use an outer limit of 1600 meters, and conclude that mushroom production does represent a
disamenity to nearby homeowners, but we are not able to make any statements about how far the
disamenity impact extends.

27
Vd2. Large-Scale Animal Production

Three issues are relevant for accurately estimating the potential local disamenity from animal
production. First, is the disamenity impact, if it exists, proportional to the number of animals
near the house, or proportional to the number of building clusters near the house. Second, does
the disamenity impact depend only on the closest building cluster, or does it depend on farther
clusters as well? Finally, how far from a building cluster does the disamenity impact extend?

To address the first issue, the performance of the building cluster index, ANBCIND, was
compared to that of the aeu-based index, ANAEUIND. Because these two indices measure
similar things, they tend to be highly correlated, and it is difficult to distinguish which does a
better job explaining variation in house prices. However, in a model that includes both indices,
the estimated coefficient on ANBCIND is negative and nearly significant (t=-1.90), while the
coefficient on ANAEUIND is positive but insignificant. In a model that includes only
ANBCIND, the estimated coefficient on the index is negative and significant (t=-2.08). In a
model that includes only ANAEUIND, the estimated coefficient on the index is negative but not
significant (t=-1.42). Based on this set of results, we conclude that the building-cluster based
index does a better job explaining house price variation than the aeu-based index.

Next, we investigate whether only the closest building cluster generates a disamenity impact.
The AN BCIND is divided into two parts, one containing information only on the closest building
cluster, the other containing information on all other building clusters located within 1600 meters
of the house. In a regression that includes both of these indices, both estimated coefficients are
negative, and are not significantly different from each other. In fact, the estimated coefficient on
the closest building cluster is smaller in absolute size than the coefficient on the index for more-
distant building clusters. We conclude that all building clusters within 1600 meters can have an
impact on house price.

Next, we investigate the spatial extent of the disamenity impact from large-scale animal
production. A model was estimated that included two indices, ANBCINDK1 and ANBCINDK2,
with K1=1600 meters and K2=3200 meters. The estimated coefficient on ANBCINDK2 is
positive, indicating that the disamenity impact from animal production does not extend past 1600
meters. Similarly, Palmquist et al. (1997) found that hog operations located between ½ and 1
miles from the house had a statistically significant impact on house price (t=2.08), but that
operations between 1 and 2 miles from the house did not (t=0.26).

We conclude that the best index for measuring the disamenity impact of large-scale animal
operations is a building-cluster-based index and includes all building clusters within 1600 meters
of the house.

To investigate whether the disamenity impact varies due to the species housed in the building
clusters, a model was estimated with the three species-weighted building cluster indices,
POULTRYBC IND, SWINEBCIND, and COWBCIND. The coefficients on each of these three
indices are presented in Table 5. While the results suggest that poultry generate the largest
disamenity impact, and cows generate the least disamenity impact, the pairwise tests show that

28
estimated coefficients are not significantly different, though the cow vs. poultry comparison is
close to being statistically significant (t=1.84). These results are suggestive, but are not strong
enough to allow us to conclude that the disamenity impact varies by species.

Table 5. Disamenity impacts by species

Coefficient Standard
Species Index Estimate Error T-Stat
COWBCIND -8.071 16.442 -0.49
POULTRYBCIND -92.598 40.306 -2.30
SWINEBCIND -48.555 33.978 -1.43

To investigate whether managerial care affects the disamenity impact from animal production,
the ANBCIND index was split into two pieces, one including building clusters on farms that
have a detailed conservation plan on file with the conservation district, the other including
building clusters on farms without a conservation plan. Whether a farm has a conservation plan
is an admittedly imperfect indicator of the amount of care the operator takes in managing the
operation to minimize off- farm impacts. Fifty-three of the 71 building clusters are on farms with
conservation plans. The estimated coefficients on each sub-index are shown in Table 6. While it
appears that the disamenity impact from building clusters on farms that do not have a
conservation plan is larger than that from building clusters on farms that do have a conservation
plan, the difference between the estimated coefficients is not statistically significant (t=1.42).
Again, the results are suggestive, but not strong enough to warrant a definitive conclusion.

Table 6. Disamenity impact by conservation plan

Coefficient Standard
Facility Type Estimate Error T-Stat
No Cons Plan -67.198 30.999 -2.17
Has Cons Plan -18.283 14.446 -1.27

In order to model the disamenity impact from animal production more flexibly, a second index is
added, ANBCIND2. The combination of ANBCIND and ANBCIND2 allows the marginal
implicit price to vary with distance in a more flexible way, though it still is constrained to equal
0 for houses that have no building clusters within 1600 meters. This is the regression model
shown in Table 3c.

It is possible, using the coefficient estimates in Table 3c, to calculate the impact on a house’s
price from a single building cluster located at different distances from the house. This impact is
shown in Figure 2, along with 95% confidence intervals. A building cluster located 500 meters
from the house decreases its price by 6.4%. At 800 meters, the impact on house price is 4.1%.
At 1200 meters, the impact is 1.6%. An outer limit to the impact of 1600 meters is imposed, so
the impact is assumed to be zero past that point. Because very few houses are located within 500
meters of a building cluster, we have little confidence in using the model to predict impacts for
such distances. We would presume that the impact would be no less than 6.4%, but cannot,
based on our data, state how much greater it might be.

By way of comparison, Palmquist et al. (1997) found that the impact on house price of hog
operations located ½ mile to 1 mile from the house was less than 1% of the impact of operations

29
located within ½ mile of the house. While exact distances between the hog operations and the
houses is not known for their dataset, it may include more sales of houses located very close to
hog operations than we observed in Berks County.

0%
0 200 400 600 800 1000 1200 1400 1600 1800

-2%
% Impact on House Price

-4%

-6%

-8%

-10%

-12%
Distance From Building Cluster (meters)

Figure 2. Impact on house prices from a single building cluster

Finally, to investigate whether the scale of the operation at a building cluster influences its
disamenity impact, building clusters were divided into three groups: Medium (> 200 but <300
aeu’s), large (> 300 but less than 600 aeu’s) and very large (> 600 aeu’s). Of the 71 facilities
identified in Berks County, 32 fall into the medium size category, 30 fall into the large category,
and 9 fall into the very large category. No information is available on the location of small
operations (<200 aeu’s). The ANBCIND and ANBCIND2 indices were each divided into three
parts accordingly, and a hedonic price equation was estimated that included the three pairs of
sub- indices. From this estimation, the percent impact on house price from a facility of each size,
located at 800 meters from the house, was calculated, and shown in Table 7.

Table 7. Percent impact on house price from a facility located 800 meters from house
(with 95% confidence intervals)
% Decrease in
Facility Size House Price
-7.5%
Medium (-13.0, -1.9)
-0.9
Large (-4.5,2.7)
-15.0
Very Large (-31.0,0.9)

30
The estimated impact of a facility is largest for very large facilites, as might be expected.
However, this estimate is not significantly different from zero (t=1.84), probably due to the fact
that there are few very large production facilities in Berks County, limiting the statistical power
of the regression. There are more medium-sized facilities in Berks County, and the estimated
disamenity impact of these operations is negative and significantly different from zero.
Interestingly, the estimated disamenity impact from large facilities is less than that from medium
facilities. However, the difference between the two estimates (medium vs. large) is not
statistically significant (t=1.88). Based on pairwise comparisons among the three size classes,
we cannot conclude with statistical confidence that different-sized operations have different
impact on house prices. We can state with 95% confidence, however, that operations between
200 and 300 aeu’s do have a negative impact on house prices.

Because differences among facilities related to size, species, or presence of a conservation plan
were not statistically significant, we favor a model that does not distinguish among facilities.
The amenity impacts listed in Table 2 apply to all facilities larger than 200 aeu’s.

Ve. The Spatial Distribution of Housing and Potential Local Disamenities

The total impact that a potential local disamenity has on residential property values depends on
the location of the residences relative to the potential local disamenity, and on the value that the
residences would have absent the disamenity impact. In this section, we calculate the total
impact of a landfill, the regional airport, and an animal production facility on neighboring
residential property values. The numbers presented here are illustrations, and should not be
interpreted as averages. The informa tion on the location of mushroom production facilities is not
specific enough to allow a similar calculation for mushroom production.

Ve1. Housing and Landfills

Three of the four landfills located within Berks County are considered here. The Conestoga
landfill in New Morgan Borough is atypical, in that it is located in a borough that is almost
exclusively industrial. The other three landfills vary substantially in the number of single family
residences located within 3200 meters of the landfill. The landfill located in Earl Township is
located in a relatively less-populated area of the county. There are 341 residences located within
1600 meters of the landfill, and a total of 918 located within 3200 meters. In contrast, the
landfill located in Cumru Township is located in an industrial area, but near Reading.
Consequently, it has fewer residences located within 1600 meters (79) but more total located
within 3200 meters (4549). The landfill located in Exeter Township has 1561 residences located
within 1600 meters, and a total of 3342 located within 3200 meters.

The landfill in Exeter Township is chosen to serve as an illustrative example. For each house
located within 3200 meters of the landfill, the percent decrease in house price due to the
landfill’s presence was calculated. This was then multiplied by the total assessed value of the
house, to give the dollar impact on house price due to the landfill. This approach assumes that
the assessed value does not already include a discount due to the proximity of the landfill. If a
discount in the assessed values does exist, then the estimates presented here will underestimate
the true impact of the landfill on house prices. Because there were few house sales observations

31
included in the regression analysis where the house was located less than 500 meters from a
landfill, the predicted house price impacts are less reliable for such houses. To be conservative
for residences located very close to the landfill, the percent impact on house price is set equal to
the impact on a house located 500 meters from the landfill.

For the 1561 residences located within 1600 meters of the landfill, the average house price
impact from the landfill is $3,937 (all values are in 2002 dollars). For the 1781 houses located
between 1600 and 3200 meters away from the landfill, the average house price impact is $1,132.
The average impact on all 3342 houses is $2442, for a total impact on all houses within 3200
meters of $8,162,000, which represents 2.6% of the assessed value of those properties.

Ve2. Housing and the Airport

The disamenity impact from the regional airport was determined to extend 1600 meters from a
line extending two miles from either end of the main runway. This approach is a crude
approximation to the noise disturbance associated with take-offs and landings. There are 1246
single family houses located within 800 meters of the flightpath, and a total of 4647 single
family houses located within 1600 meters. Of these, however, 2391 are located with the City of
Reading. Because the hedonic price function was estimated only for residences outside the City
of Reading, it should not be used to calculate property value impacts within the City.

For each of the remaining 2256 single family residences located outside of the City of Reading
and within 1600 meters of the runway or flight path, the impact on property value was
calculated. For consistency with the analysis on landfills and animal production facilities,
houses located within 500 meters of the runway and flightpath are treated as if they were located
exactly 500 meters away. The average house price impact from the airport was $104, and the
total impact on the 2256 residences was $235,000, which is 0.1% of the assessed value of those
2256 houses. This estimate only counts the impacts on houses located outside of the City of
Reading, however.

Ve3. Housing and Animal Production

There are no setback requirements when constructing animal barns, though manure handling
facilities must be located at least 100-300 feet from property boundaries, depending on slope. In
this section, we measure how many houses are located close to animal facilities, and compare
this spatial distribution to what would be expected if animal facilities were randomly located.

The following analysis is done for the 60 animal building clusters that are located at least 1600
meters from the county border. For each building cluster, the number of single- family houses
located within 400 meters, within 800 meters and within 1600 meters was determined. The
average numbers of houses are given in Table 8.

32
 Table 8. Number of houses located near animal facilities
Animal Randomly-Chosen
Distance Facilities Points
400 meters 2.64 16.79
800 meters 16.70 60.32
1600 meters 105.86 238.24

For comparison purposes, 60 random points were selected in the county. These were the parcel
centroids of parcels that were in privately-owned open space use and that were at least 5 acres in
size. These are the types of parcels where animal operations are likely to located. Table 8 shows
that the actual animal facilities tend to be located in areas that have few houses. The number of
houses located within 400 meters and within 800 meters of actual animal facilities is about ¼ of
that which would be expected if these facilities were locating themselves randomly in the
landscape. The number of houses in the two columns is significantly different at all three
distances. We conclude that some process is working to minimize conflicts between residential
use and animal production by separating the uses spatially.

Part of the reason why there are fewer houses near animal facilities than would otherwise be
expected may be due to the effect of agricultural zoning and ASA’s. Fifty- nine of the 71 animal
production facilities (83.1%) are located in ASA’s. In contrast, of the privately-owned open
space parcels of at least 5 acres in size, only 37.6% of the land (by area) is located within ASA’s.
This difference is statistically significant, and we conclude that animal production facilities have
a tendency to be located in ASA’s. However, we do not know whether this is because animal
production facilities tend to locate in ASA’s, or if ASA’s tend to be created in areas that have
more animal production.

It is also interesting to look at the relationship between agricultural conservation easements and
location of animal production. Twenty-two of the 71 animal production facilities are located on
farms with ACE’s. In contrast, of the privately-owned open space parcels of 5 acres or more,
only 8.8% (by area) are under ACE’s. This difference is also significant, and we conclude that
animal production facilities have a tendency to locate on farms with ACE’s (or conversely, that
farms with ACE’s are more likely to have animal production facilities).

Finally, we calculate the total impact of one animal production facility on house prices. We
choose a production facility that is close to the average in terms of its location relative to houses,
with 119 houses located within 1600 meters and 17 houses located within 800 meters. For each
house, the percent decrease in price due to the animal production facility is calculated, and this is
multiplied by the house’s assessed value. No houses were located within 500 meters of this
animal production facility.

For this illustrative case, the average house price impact due to the animal production facility is a
decrease in value of $1,803. The total impact on all 119 houses is $215,000, which is 1.7% of
the total assessed value of the 119 houses. This total is intended as an illustration, and should not
be viewed as an average value for all animal facilities. The impact from any given facility will
depend on the number of houses near the facility, the location of those houses relative to the
facility, and the value of those houses.

33
It should be noted that this estimate of the impact on property va lues does not include amenity or
disamenity impacts that are not tied to house location. Price differentials for houses located near
a large-scale animal production facility would not capture benefits or costs experienced by
commuters or tourists who travel past such facilities, or any negative impact on water quality that
is experienced downstream from the facility.

Vf. Summary of Disamenity Impacts

Table 9 summarizes the disamenity impacts from landfills, the regional airport, mushroom
production and large-scale animal production facilities.

Table 9. House price impacts by distance from the house.

Distance from the House
500m 800m 1200m 2400m
Landfill -12.4% -6.9% -3.8% -0.8%
Airport Runway -0.3% -0.2% -0.1%
Mushroom Production -0.8% -0.4% -0.1%
Animal Production -6.4% -4.1% -1.6%

VI. Location of New Residential Development

If house price depends on location and on surrounding land use, and if residential development
tends to occur in places where houses are valued highest, then information on land use might
help explain the pattern of residential development. A probit analysis was used to model factors
that influence the probability that any given lot converts from open space to residential use.

The analysis treats as the unit of observation parcels of land that can be subdivided. The
population of such parcels was defined as every parcel that satisfied the following requirements

- was privately owned and in open space use in 1996

- is over 5 acres in size
- does not have an ACE at any time
- is not located within 1 mile of the county boundary, or within one mile of New Morgan
Borough

7,399 parcels were identified as potentially developable. From Planning Commission records,
92 of these parcels were identified where the owner sought to subdivide, with at least 3 of the
daughter parcels to be designated as residential. Our analysis then models factors that influence
the decision to subdivide for residential construction. It does not model commercial or industrial
development, and does not model home construction on parcels that are not subdivided.

A probit regression was used to model the probability of subdivision, conditional on measured
characteristics of the parcel. The measured characteristics are the same measures used in the

34
hedonic regression (with the exception of the house’s structural characteristics, which is
endogenously determined by the builder, and not a characteristic of the parcel prior to
development). So, for example, the land use observed in 2002 was used to measure prospective
land use surrounding the new residential parcel. An additional factor that might influence
development location is access to utilities. Unfortunately, digitized maps showing the location
of public utility service were not available. Based on the Assessment Office’s parcel database,
the distance from the developable parcel to a parcel that has pub lic water and/or public sewer
was calculated. A variable measuring the number of utilities located within 200 meters of the
developable parcel was used as an index of public utility availability.

Two probility-of-development models were estimated. The first model examined whether
prospective house price influenced the location of development. For each potentially-
developable parcel, a Prospective House Price Factor was calculated, equal to X’ß, where X is
the vector of location-specific attributes from the hedonic price analysis, and ß are the hedonic
price model coefficients for those attributes. X included all explanatory variables other than
structural characteristics of the house. While not a house price, X’ß will capture differences in
location that will generate differences in house prices.

Prospective house price is not the only factor influencing location of development. Cost of
development also matters. Here, the three soil indices (suitability for septic systems, suitability
for construction, and agricultural productivity) are used to reflect the costs of construction, and
the opportunity costs of taking the land out of agricultural production. Slope and elevation may
also influence construction costs, as will access to public utilities. Parcel size may matter if there
are economies of scale in subdivision. Finally, zoning should influence probability of residential
development, as should whether the parcel is in an ASA, if these policy tools are at all effective.
The individual effects of Effective Ag Zoning and ASA may not be additive, so an interaction
term between these two is included.

The results of the probit regression using these explanatory variables are shown in Table 9,
Model 1. The most striking result is that prospective house price has no impact on the
probability of subdivision. We have no evidence that location-specific differentials in house
prices are driving the pattern of development. The second most striking result is that large
parcels are much more likely to subdivide than small parcels. This makes sense, particularly if
small parcels tend to convert to residential use without first subdividing. Both Effective Ag
Zoning and ASA’s are associated with lower probability of subdivision, indicating that these
policy tools are effective in directing development to other areas. The only remaining
explanatory variable that had a significant effect on probability of development was elevation.
Here, higher-elevation parcels were less likely to subdivide. The coefficient on access to public
utilities was positive, consistent with expectations, but the significance level was marginal.

Model 1 does not show that lots with higher potential house price are more likely to be
developed. However, some caveats should accompany that result. First, Model 1 relies on a
two-stage modeling process, where the hedonic price regression is estimated in the first stage,
and the probability-of-development model estimated in the second stage, using the first stage
hedonic pricing model. It is not unreasonable to suppose that errors in the first stage model
would be spatially correlated with errors in the second stage model, but no attempt is made to

35
account for this possibility. Second, the hedonic price model is estimated for house prices,
whereas probability of development is more likely to depend on the value of a developable lot.

If we cannot show that location-specific house price differentials are driving development
patterns, can we show any spatial interactions in land use change? A second probit regression
model was estimated (Model 2) that included measures of surrounding land use out to 400
meters from the prospective house site. In contrast to Model 1, surrounding land use for Model 2
is measured prior to conversion, i.e. as it existed in 1996. Here, it was found that parcels that
have more residential land near them are more likely to subdivide. The estimated coefficie nt on
surrounding open space is positive, but is not significant (?2 = 2.30). However, parcels located
near government-owned open space are slightly but significantly more likely to subdivide.
These result are somewhat at odds with that found by Irwin and Bockstael (2001a) who found
that new residential development was more likely in areas with less development. We find that
new development tends to locate near existing residential land, with the exception that proximity
to government-owned open space tends to attract residential development.

We also examine what impact animal production has on the location of residential development.
We calculated our animal facility index, ANBCIND, for every potentially-developable parcel.
The estimated coefficient is negative, as would be expected if animal production is a local
disamenity, but is not significant (?2 = 1.99).

It should be remembered that while the regressions shown in Table 9 are based on several
thousand potentially-developable parcels, the observed number of subdivisions is small (only
92). Our ability to detect patterns having observed only 92 conversions is limited. Further, the
errors in the probit models may be spatially correlated, which would give biased parameter
estimates. However, Plate 4 shows no tendency toward clustering in the spatial pattern of
parcels that developed. If any spatial pattern exists, it is that parcels that develop may tend to be
isolated form one another. This is an issue that warrants further research.

To summarize, surrounding land use does appear to have some impact on the probability of
development, but the “leapfrog” model, where development occurs away from previously-
developed areas because new home buyers want to be located in areas with lots of open space,
was not supported by our results. In Berks County, it appears that new development is most
likely to occur near existing residential development, though proximity to government-owned
open space appears to have some effect.

VII. Conclusions and Future Research Directions

VIIa. Key Findings of This Research

We highlight several important conclusions from this research:

1) Spatial interactions among parcels do impact residential property values. In Berks County,
we found both nearby land uses and proximity to potential local disamenities impact the sale
prices of single family houses.

36
2) Endogeneity is important when modeling amenity impacts from surrounding land use. This is
demonstrated by the difference in parameter estimates between the hedonic price model
estimated using instrumental regression and the parameter estimates from ordinary least squares
regression.

3) Within 400 meters of a house, open space is the most desirable surrounding land use,
followed by large- lot residential use. Commercial and small and multifamily residential use are
less desirable. One implication is that conversion of open space to commercial, industrial, small-
lot residential, or multi- family residential will have a negative impact on house prices within 400
meters.

4) Within 400 meters, privately-owned open space with conservation easements have a less
positive impact on house price than privately-owned open space without easements. The act of
purchasing a conservation easement may not in itself drive neighboring house prices down.
Rather, it may be that conservation easements tend to be associated with a certain type of open
space (actively- farmed, productive farmland) that is less desirable as a near neighbor. Consistent
with this explanation is the finding that open space within 400 meters that is covered in grass,
pasture or crops has a lower amenity value than forested open space, though the difference in
estimated amenity values is not quite significant.

5) The impact of open space that is zoned for residential use, but that has not yet been built is
statistically indistinguishable from the impact of industrial use, and is significantly worse than
medium or large- lot residential use. This may be a short-term decrease in house price, reflecting
the uncertainty and disruption that accompany new residential construction.

6) Between 400 and 1600 meters from a house, commercial is the most attractive land use,
followed by large-lot residential, and then open space. Of open space uses, grass, crops and
pasture are preferred to forested open space and eased open space is preferred to uneased open
space, both results opposite to the results for open space within 400 meters of the house. Outside
400 meters, government-owned open space is preferred to privately-owned, uneased open space.
We can therefore characterize the ideal house as being immediately surrounded by forested open
space, but with commercial uses (offices and shopping) located within one mile of the house. At
all distances, small- lot and multifamily residential use is less attractive than large- lot residential
development.

7) The hedonic price regression was able to measure impacts on house prices from potential
local disamenities. Among the potential local disamenities investigated, landfills and large-scale
animal production facilities had the largest negative impact on house prices. Mushroom
production and the airport had smaller negative impacts. High-traffic roads had a small negative
effect that extended only a short distance from the road. No impact could be identified from
sewage treatment plants.

8) Specific to large-scale animal production facilities, we find a significant impact within 1600
meters from such facilities, but not farther than 1600 meters. We find that facilities with
between 200 and 300 aeu’s are large enough to have a negative impact on neighboring house

37
prices. However, we were not able to draw firm conclusions about whether the negative impact
varies by species of animal, size of operation, or whether the operator has developed a detailed
conservation plan.

9) Single family residences tend not to be located near large-scale animal production facilities.
It is not known whether this is the result of decisions made by animal producers to locate in areas
with fewer houses, by decisions made by developers not to build homes near animal facilities, or
whether each group is locating on land with different attributes, resulting in a natural separation.
Nor can we determine whether this separation is a result of policy measures such as Agricultural
Security Areas or Effective Agricultural Zoning. However it has occurred, this separation tends
to mitigate the impact that animal production facilities have on property values.

10) The total impacts of one landfill, the airport, and one animal production facility on nearby
house prices were calculated, as illustrations. The total impact of a landfill on the value of 3342
properties located within 3200 meters was calculated to be $8,161,771, or 2.6% of the assessed
value of the affected properties (in 2002 dollars). The impact of the regional airport on 2256
properties located within 1600 meters of the runway and flight path was calculated to be
$235,111, or 0.1% of the assessed value of the affected properties. The total impact of an animal
production facility on 119 properties located with 1600 meters was calculated to be $214,589, or
1.7% of the assessed value of those properties. These estimates capture only those impacts that
fall on residents who live near the facilities. They do not include costs of impacts that occur
farther from such facilities, such as impacts on downstream water quality, or positive or negative
amenity impacts on tourists or commuters who travel past such facilities. These calculations are
illustrative, but should not be viewed as averages for similar facilities. The total impact from a
given facility like these will depend on the number of houses located near the facility, the
distance between the facility and the houses, and the market value the houses would otherwise
have.

11) An analysis of the spatial pattern of development showed that larger lots are more likely to
be developed, that lots located near existing residential areas are more likely to be developed,
and that proximity to government-owned open space has a small positive impact on probability
of development. Of interest here is that potential house price, as predicted by the hedonic price
function, did not help explain the pattern of development. Also, other than for government-
owned open space, we did not find a significant tendency for new development to locate in areas
with more open space.

VIIb. Future Research Needs

This project has demonstrated the utility of using GIS analysis to investigate issues of spatial
interaction in land use, both interactions that affect house prices and interactions that affect the
pattern of development that occurs over time. The study area chosen, Berks County, was well
suited for this pilot study. First, Berks County has well-developed GIS data resources, and local
officials and their staff were very helpful to the project. Second, Berks County has a high
proportion of open space that is in ASA’s and a high proportion enrolled in ACE’s, and these
lands are spread broadly throughout the county. This is important because it allows us to
identify the impact of these lands on house prices independent of other factors that vary spatially.

38
Third, it was possible to map all animal production facilities in Berks County - a task that might
be somewhat more difficult in another county.

At the same time, performing the analysis in only one county has its limitations. Berks County is
still fairly well endowed with open space. It may be that the amenity value of open space near a
house will be larger in a county where open space is more scarce. Restricting the analysis to
only one county limited the number of animal operations included in the hedonic price analysis,
and limited the number of residential conversions in the probit probability-of-development
analysis. Extending this research approach to other counties will increase the amount of data,
allowing more-precise estimation of the hedonic price regression and land coversion probit
regression, and will allow us to determine to what extent the findings apply to other regions.

For these reasons, we recommend that the approach used in this study be expanded to a larger
region. Specifically, the hedonic price analysis should be broadened to include counties where
open space is more scarce, and where animal production is located closer to residential areas. It
is quite possible that the amenity value of open space will be higher in areas where open space is
more scarce, and that the marginal impact of eased open space will be positive, as it was found to
be in Maryland by Irwin (2002).

With more observations on house/animal interactions, we will be better able to distinguish the
relative impacts of different scales of animal operation, and different species. To the extent that
operation-specific information can be collected without violating the privacy rights of the
operators, that information could be used to help explain variation in the disamenity impact from
animal production. Similarly, more-detailed information on mushroom production facilities
would allow a more refined analysis of their impact on house prices. Finally, the land
conversion analysis requires observation of more conversions in order to be able to identify the
more subtle factors that influence the pattern of development.

A second extension to the research is to apply new estimation techniques that can simultaneously
account for endogeneity in land use and spatial correlation in house prices. This has not
previously been done for a hedonic pricing model.

39
References

Abeles-Allison, Mark, and Larry J. Connor. 1990. “An Analysis of Local Benefits and Costs of
Michigan Hog Operations Experiencing Environmental Conflicts.” Agricultural
Economics Report #536, Department of Agricultural Economics, Michigan State
University, East Lansing, Michigan.

Beasley, W., W.G. Workman, and N.A. Williams. 1986. Amenity values of urban fringe
farmland: A contingent valuation approach. Growth and Change 17(4):70-78.

Beegle, D.B. 1997. “Pennsylvania’s Nutrient Management Act: Who Will Be Affected?”
Pennsylvania State University Agronomy Fact Sheet #54, 4pp.

Bergstrom, J.B., B. Dillman, and J. Stoll 1985. Public environmental amenity benefits of private
land: The case of prime agricultural land. Southern Journal of Agricultural Economics
17:139-149.

Bockstael, N.E 1996. Modeling economics and ecology: The importance of spatial perspective.
American Journal of Agricultural Economics 78(5):1168-1180.

Boyle, Mellisa A., and Katherine A. Kiel. 2001. “A Survey of House Price Hedonic Studies on
the Impact of Environmental Externalities.” Journal of Real Estate Literature 9(2):117-
144.

Brisson, Inger, and David Pearce. 1995. “Benefits Transfer for Disamenity from Waste
Disposal.” Working Paper, CSERGE, University College-London.

Burchell, R., W.D. Likstokin, and W.R. Dolphin. 1985. The New Practitioner’s Guide to Fiscal
Impact Analysis. New Jersey: Center for Urban Policy Research.

Cheshire, P. and S. Sheppard 1995. On the price of land and the value of amenities. Economica.
62: 247-267.

Garrod, G.D., and K.G. Willis. 1992a. The amenity value of woodland in Great Britain: A
comparison of economic estimates. Environmental and Resource Economics 2(4):415-
434.

Garrod, G.D., and K.G. Willis. 1992b. Valuing goods’ characteristics: An application of the
hedonic pricing method to environmental attributes. Journal of Environmental
Management 34(Jan.):59-76.

Geoghegan, J, L. A. Wainger, and N. [Link]. 1997. Spatial landscape indices in a hedonic

framework: An ecological economics analysis using GIS. Ecological Economics. 23: 251-
264.

40
Halstead, J. M. 1984. Measuring the nonmarket value of Massachusetts agricultural land.
Journal of the Northeastern Agricultural Economics Council 13(Apr.):12-19.

Irwin, E.G. 2000. Economic theories of land use: How well do they explain fragmented patterns
of exurban development? Unpublished working paper, Department of Agricultural,
Environmental and Development Economics, Ohio State University.

Irwin, Elena G. 2002. “The Effects of Open Space on Residential Property Values.” Land
Economics 78(4):465-480.

Irwin, E.G. and N.E. Bockstael. 2001a. Interacting agents, spatial externalities, and the evolution
of residential land use patterns. Unpublished working paper, Departme nt of Agricultural,
Environmental and Development Economics, Ohio State University.

Irwin, E. and N. E. Bockstael. 2001b. The problem of identifying land use spillovers: Measuring
the effects of open space on residential property values. AJAE. 83(3), 698-704.

Leggett, C., and N.E. Bockstael. 2000. Evidence of the Effects of Water Quality on Residential
Land Prices. Journal of Environmental Economics and Management 39(2):121-144.

Mikesell, Robert E. 2002. Odor Remediation and Siting Considerations for Pennsylvania Swine
Farms. Unpublished Ph.D. Thesis. Department of Dairy and Animal Science, The
Pennsynvania State University.

Myers, C.A.B. and C. Auchenbach. "Effective Agriculture Preservation Zoning— One Tool to
Help Save an Industry." pp. 101-108 in Proceedings of a Conference on Protecting
Farmland at the Fringe: Do Regulations Work? Strengthening the Research Agenda, L.
W. Libby and C. Abdalla, eds. Columbus, OH: Swank Program in Rural-Urban Policy,
Ohio State University, 2002.

Nelson, A. C., J. Genereux and M. Genereux. 1992. “Price Effects of Landfills on House
Values” Land Economics 68(4), 359-65.

Palmquist, Raymond B. 1991. “Hedonic Methods,” in Measuring the Demand for

Environmental Quality, J.B. Braden and C.D. Kolstad, eds., Amsterdam: Elsevier Science
Publishing.

Palmquist, Raymond B., Fritz M. Roka, and Tomislav Vukina. 1997. “Hog Operations,
Environmental Effects,and Residential Property Values.” Land Economics 73(1):114-
124.

Pennsylvania Agricultural Statistics Service (PASS). 2003. Pennsylvania Agricultural Statistics

2001-2002. Pennsylvania Department of Agriculture, Harrisburg, PA.

41
Ready, R.C., M.C. Berger, and G.C. Blomquist. 1997. Measuring amenity benefits from
farmland: Hedonic pricing vs. contingent valuation. Growth and Change 28(Fall):438-
458.

Rosen, Sherwin. 1974. “Hedonic Prices and Implicit Markets: Product Differentiation in Pure
Competition,” Journal of Political Economy 82:34-55.

Tyrvainen, L. and A. Miettinen 2000. Property value and urban forest amenities. Journal of
Environmental Economics and Mangement. 39: 205-223.

US Department of Agriculture, 1997 Census of Agriculture, Pennsylvania State and County

Data, Vol. 1, Geog. Area Series, Part 38. Washington, DC : NASS. March 1999.

VanDevender, Karl. 1998. "Arkansas Swine Odor Survey." Agriculture and Natural Resources
Publication. Cooperative Extension Service, University of Arkansas, Little Rock, AK. 2
pp. [Link]

42
Appendix 1. Project Local Advisory Group

The research team is grateful to the Project Local Advisory Group for their contributions and for
the time and energy they donated to this effort. However, the research team takes sole
responsibility for the project’s results, for the contents of this report, and for any and all errors
contained within.

The members of the Project Local Advisory Group are:

Clyde Myers, Extension Agent, Berks County Cooperative Extension, Pennsylvania State
University

Judith Schwank, County Commissioner, Berks County

Glenn Knoblauch, Executive Director, Berks County Planning Commission

Duane Rashlich, Director or Real Estate, Berks County Assessment Office

Tami Hildebrand, Executive Director, Berks County Agricultural Land Preservation

Board

Kenneth Grimes, Chairman, Upper Tulpehocken Township Supervisors

Robert Behling, Cha irman, South Heidelberg Township Planning Commission

43
 Table 3a. Hedonic Price Regression Results - House Characteristics

IV Model OLS Model

Standard Standard
Variable Coefficient Error T stat Coefficient Error T stat

House and Location Characteristics

Intercept -1.7308 0.3237 -5.35 -1.08355 0.2526 -4.29
Living Area (sq feet) 0.00023 3.39 E-06 67.95 0.000227 3.11E-06 73.12
Lot Size (acres) -0.01799 0.00542 -3.32 -0.02087 0.00508 -4.11
ln(Lot Size) 0.1276 0.00579 22.03 0.140651 0.00517 27.22
# Bedrooms 0.03326 0.00314 10.58 0.035191 0.00298 11.79
# Baths 0.06666 0.00403 16.56 0.067237 0.00381 17.64
Basement (1=yes) 0.04959 0.00722 6.87 0.059085 0.00673 8.77
Finished Attic (1=yes) -0.01752 0.00669 -2.62 -0.02574 0.00621 -4.14
Brick Exterior (1=yes) 0.07353 0.00516 14.24 0.074897 0.00482 15.52
Stone Exterior (1=yes) 0.1132 0.0142 7.95 0.116432 0.0135 8.60
44

Masonry Exterior (1=yes) 0.04966 0.00571 8.69 0.050638 0.00534 9.48

Central AC (1=yes) 0.05050 0.00442 11.42 0.050064 0.00416 12.04
Phys. Cond. (1-5, 5 worst) -0.05849 0.00686 -8.53 -0.05844 0.00615 -9.50
Year Built 0.003422 0.000133 25.78 0.00311 0.000107 29.07
Year Sold -0.01216 0.00139 -8.75 -0.01368 0.00128 -10.71
Distance to Reading (km) 0.00104 0.000523 1.99 -0.00051 0.000394 -1.30
Distance to Allentown (km) -0.00161 0.000263 -6.14 -0.00158 0.000228 -6.94
Distance to Phil. (km) -0.00271 0.000396 -6.83 -0.00375 0.000305 -12.30
Slope (%) -0.00258 0.00103 -2.51 -0.00288 0.000799 -3.61
Elevation (m) -0.00002 0.000087 -0.20 0.000137 0.000066 2.07
Elevation Difference (m) 0.001152 0.000192 5.99 0.001203 0.000166 7.24
PSSA Score 0.000749 0.000101 7.41 0.000788 0.000088 8.94
Public Sewer (1=yes) 0.004988 0.00911 0.55 -0.00214 0.00777 -0.28
Public Water (1=yes) 0.02327 0.00867 2.68 0.02247 0.00765 2.94
 Table 3b. Hedonic Price Regression Results - Surrounding Land Use

IV Model OLS Model

Standard Standard
Variable Coefficient Error T stat Coefficient Error T stat

Surrounding Land Use

Open Space w/in 400m 0.00276 0.000431 6.41 0.002129 0.000278 7.65
Open Space 400-1600m -0.00008 0.000036 -2.12 1.53 E-06 0.00002 0.08
Gov Owned OS w/in 400m 0.00005 0.000374 0.13 0.000561 0.000222 2.52
Gov Owned OS 400-1600m 0.000199 0.000035 5.75 0.000064 0.000015 4.26
Eased OS w/in 400m -0.00075 0.000321 -2.34 -0.00045 0.000259 -1.73
Eased OS 400-1600m 0.000104 0.000023 4.49 0.00004 0.000018 2.25
Vacant Land w/in 400m -0.00367 0.000912 -4.02 -0.00066 0.000153 -4.33
Vacant Land 400-1600m 0.000058 0.00013 0.45 -0.00005 0.00003 -1.64
Pasture/Crops w/in 400m -0.00039 0.000222 -1.75 -0.00018 0.000144 -1.23
45

Pasture/Crops 400-1600m 0.000081 0.00002 4.00 0.000035 0.000013 2.63

Residential w/in 400m 0.002143 0.000351 6.11 0.00168 0.000291 5.77
Residential 400-1600m 0.000305 0.000053 5.75 0.000055 0.00003 1.84
Small-lot Res w/in 400m -0.00186 0.000306 -6.07 -0.0007 0.00021 -3.36
Small-lot Res 400-1600m -0.00022 0.000062 -3.50 0.00001 0.000035 0.29
Med-lot Res w/in 400m -0.00022 0.00022 -0.98 0.000193 0.000163 1.19
Med-lot Res 400-1600m -0.00027 0.000051 -5.31 -4.3 E-06 0.000027 -0.16
Large-lot Res w/in 400m 0.000262 0.000293 0.89 0.001104 0.000231 4.77
Large-lot Res 400-1600m -0.00001 0.000077 -0.16 0.000213 0.000058 3.67
Other Res w/in 400m -0.00176 0.000403 -4.36 -0.00208 0.00032 -6.48
Other Res w/in 400m -0.00017 0.000077 -2.27 0.000069 0.000057 1.20
Commercial w/in 400m 0.001089 0.00069 1.58 0.001909 0.000288 6.64
Commercial 400-1600m 0.000328 0.000056 5.90 0.000152 0.000023 6.65
 Table 3c. Hedonic Price Regression Results - Potential Local Disamenities and Zoning

IV Model OLS Model

Standard Standard
Variable Coefficient Error T stat Coefficient Error T stat

Potential Local Disamenities

HTIND -0.1575 0.0854 -1.84 -0.17366 0.0806 -2.15
HTDIST (km) -0.0047 0.00203 -2.32 -0.00687 0.001621 -4.24
MRIND_1600 -6.0360 1.8368 -3.29 -7.2706 1.6926 -4.30
LFIND_3200 -73.2097 19.328 -3.79 -99.8715 15.9685 -6.25
APIND_1600 -2.5287 1.2783 -1.98 -1.7303 1.2089 -1.43
ANBCIND_1600 -112.968 35.633 -3.17 -94.8982 33.9607 -2.79
ANBCIND2_1600 25391.44 9798.5 2.59 20280.35 9353.8 2.17
SPIND_1600 7.1859 12.029 0.60 18.3067 3.2129 5.70
46

Zoning
Zoning Agricultural -0.01116 0.013 -0.86 -0.01269 0.0124 -1.02
Zonging Effective Ag -0.01834 0.0132 -1.39 -0.00323 0.0114 -0.28
Zoning Comm/Indust -0.00793 0.00949 -0.84 -0.00821 0.00824 -1.00
Zoning Multiple 0.025325 0.00569 4.45 0.02686 0.00428 6.27
Zoning Conservation -0.01497 0.0118 -1.27 0.003911 0.0107 0.37
 Table 4a. Marginal Implicit Prices for Land Use Within 400m of House

IV Model OLS Model

Total Standard Total Standard
Land Use Type Impact Error T stat Impact Error T stat

Within 400m of House

Privately-owned Forested
Open Space 0.00276 0.000431 6.41 0.002129 0.000278 7.65
Govt -owned Forested
Open Space 0.00281 0.000433 6.48 0.00269 0.00032 8.41
Privately-owned Grass,
Pasture, and Crops 0.002373 0.000396 6.00 0.001953 0.000267 7.31
Eased, privately-owned
grass, pasture, crops 0.00162 0.000438 3.70 0.001504 0.000367 4.10
Vacant privately-owned
Open Space -0.00091 0.000789 -1.15 0.001465 0.000301 4.87
SF residential small lot 0.000284 0.000335 0.85 0.000975 0.000281 3.46
47

SF residential medium lot 0.001927 0.000306 6.29 0.001873 0.000258 7.26

SF residential large lot 0.002405 0.000378 6.37 0.002784 0.000302 9.21
SF residential xlarge lot 0.002143 0.000351 6.11 0.00168 0.000291 5.77
Other residential 0.000383 0.000477 0.80 -0.0004 0.000386 -1.02
Commercial 0.001089 0.00069 1.58 0.001909 0.000288 6.64
 Table 4b. Marginal Implicit Prices for Land Use Between 400m and 1600m from House

IV Model OLS Model

Total Standard Total Standard
Land Use Type Impact Error T stat Impact Error T stat

Between 400m and 1600m

Privately-owned Forested
Open Space -0.00008 0.000036 -2.12 1.53 E-06 0.00002 0.08
Govt -owned Forested
Open Space 0.000123 0.000029 4.28 0.000066 0.000021 3.08
Privately-owned Grass,
Pasture, and Crops 5.62E-06 0.000028 0.20 0.000037 0.000017 2.15
Eased, privately-owned
grass, pasture, crops 0.00011 0.000031 3.49 0.000076 0.000022 3.41
Vacant privately-owned
Open Space -0.00002 0.000121 -0.14 -0.00005 0.000034 -1.42
SF residential small lot 0.000087 0.000043 2.04 0.000065 0.000029 2.28
48

SF residential medium lot 0.000032 0.000033 0.96 0.000051 0.000023 2.20

SF residential large lot 0.000293 0.000062 4.70 0.000268 0.000046 5.79
SF residential xlarge lot 0.000305 0.000053 5.75 0.000055 0.00003 1.84
Other residential 0.00013 0.000075 1.72 0.000124 0.000055 2.23
Commercial 0.000328 0.000056 5.90 0.000152 0.000023 6.65
 Table 9. Probit Regressions of Probability of Subdivision
Model 1 Model 2
Standard Chi Signif. Standard Chi Signif.
Variable Coefficient Error Square Level Coefficient Error Square Level

Intercept -1.9967 0.9295 4.61 0.032 -4.7768 1.6112 8.03 0.003

Prospective House Price -0.0019 0.608 0 0.997
Parcel Size (acres) 0.0051 0.0008 42.78 0.001 0.006 0.0009 47.66 0.001
Utilities w/in 200m 0.0861 0.053 2.64 0.104 0.0836 0.0574 2.12 0.146
Elevation (meters) -0.0018 0.0009 3.95 0.047 -0.0027 0.001 7.22 0.007
Slope -0.0239 0.0175 1.86 0.173 -0.0163 0.0183 0.80 0.371
Ag Productivity Index 0.0016 0.0019 0.69 0.405 0.0026 0.002 1.60 0.206
Septic Suitability Index 0.0525 0.0423 1.54 0.214 0.0682 0.0443 2.37 0.124
Building Suitability Index -0.0057 0.0215 0.07 0.791 -0.008 0.0225 0.12 0.724
Dist to Philadelphia 0.0052 0.0047 1.20 0.273
Distance to Reading 0.0089 0.0077 1.32 0.251
Distance to Main Roads -0.0089 0.0273 0.11 0.746
49

Open Space w/in 400m 0.0197 0.0130 2.30 0.129

Commercial w/in 400m 0.0105 0.0156 0.45 0.503
Residential w/in 400m 0.0275 0.0131 4.39 0.036
Eased Open Space -0.0046 0.0066 0.48 0.491
Govt -Owned O.S. 0.0086 0.0038 4.98 0.026
ANBCIND -331.9 235.1 1.99 0.158

ASA -0.4419 0.1648 7.19 0.007 -0.4162 0.1696 6.02 0.014

Zoning Effective Ag -0.3232 0.1481 4.76 0.029 -0.2949 0.1582 3.47 0.062
Zoning Agricultural -0.2069 0.1673 1.53 0.216 -0.2061 0.1741 1.40 0.236
Zoning Comm/Indust -0.2047 0.1833 1.25 0.264 -0.0724 0.1975 0.13 0.714
Zoning Multiple -0.3619 0.3751 0.93 0.334 -0.3813 0.3960 0.93 0.336
Zoning Conservation 0.0204 0.1259 0.03 0.871 0.0298 0.1314 0.05 0.821
ASAxEff Ag Zoning 0.2726 0.265 1.06 0.304 0.3134 0.2698 1.35 0.245
Plate 1 - Developed land - Residential, commercial and Industrial

50
Plate 2. Open Space - Agricultural security areas and conservation easements.

51
Plate 3. Open space - Government-owned vs. privately-owned.

52
Plate 4 - Location of newly subdivided parcels.

53
Plate 5 - Location of landfills and regional airport.

54
Plate 6 - Location of mushroom production, by township.

55
Plate 7. Location of animal production facilities, by township.

56