Você está na página 1de 20

Journal of Land Use Science

ISSN: 1747-423X (Print) 1747-4248 (Online) Journal homepage: http://www.tandfonline.com/loi/tlus20

Parcelization in rural agricultural and forested


landscapes in Wisconsin, 19722007: evaluating
multiple dimensions of human decision-making
over time

Timothy T. Kennedy & Howard Veregin

To cite this article: Timothy T. Kennedy & Howard Veregin (2016): Parcelization in rural
agricultural and forested landscapes in Wisconsin, 19722007: evaluating multiple
dimensions of human decision-making over time, Journal of Land Use Science, DOI:
10.1080/1747423X.2016.1263686

To link to this article: http://dx.doi.org/10.1080/1747423X.2016.1263686

Published online: 05 Dec 2016.

Submit your article to this journal

Article views: 9

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=tlus20

Download by: [Fudan University] Date: 09 December 2016, At: 15:17


JOURNAL OF LAND USE SCIENCE, 2016
http://dx.doi.org/10.1080/1747423X.2016.1263686

Parcelization in rural agricultural and forested landscapes in


Wisconsin, 19722007: evaluating multiple dimensions of human
decision-making over time
a b
Timothy T. Kennedy and Howard Veregin
a
Geography and Geology Department, University of Wisconsin-Stevens Point, Stevens Point, WI, USA; bWisconsin
State Cartographer, University of Wisconsin-Madison, Madison, WI, USA

ABSTRACT ARTICLE HISTORY


Many land-use change studies have relied on geographical explanatory Received 7 April 2016
factors. Unfortunately, they are but a single perspective in the multi- Accepted 15 November
dimensional process of human decision-making. This project was 2016
designed to model the social, economic, geographic, and regulatory KEYWORDS
factors at the most appropriate unit of analysis, the landowner. By Humanenvironment
examining parcelization, a window is opened into the antecedent interactions; parcelization;
event of land-use change. A logistic regression model determined the human decision-making;
likelihood a tax parcel would split in three time periods between 1953 logistic regression; land-use
and 2007. Geographic variables showed expected relationships to a change; GIS
parcelization event, while economic and regulatory variables illustrated
some unexpected relationships. Social variables demonstrated scale
issues that challenged their ecacy. A temporal analysis showed that
historic parcelization was explained more robustly than more contem-
porary parcelization. The results could indicate that contemporary par-
celization is driven by new and more complex factors not yet
represented in similar models.

1. Introduction
In the United States, the landowner is the primary decision-maker regarding tax parcel subdivision
and land use. While cities, counties, and other agencies enact zoning and land-use plans to curtail
non-desired uses, the landowner makes the decision to maintain land in agricultural or forested
use, or to sell or subdivide. Most studies of land-use change focus on geographic or biophysical
factors, due in part to the ease of data capture through remote sensing. However, landowners are
aected by diverse factors from the social realm (e.g. population change, landowner age), the
economic realm (e.g. crop prices, commercial forest availability, fuel prices), the geographic/
biophysical realm (e.g. proximity to roads, water, urban services), and the regulatory realm (e.g.
zoning, subdivision restrictions, tax incentive programs). This study illustrates that factors from
these four dimensions of human decision-making can be incorporated into a single model to better
understand land-use change, although not without some challenges associated with data avail-
ability and scale.
Rural sprawl, also known as suburban or exurban sprawl, is a leading propagator of rural land-
use change and is characterized by the encroachment of low-density anthropogenic development
on rural or peri-urban areas (Brown, Johnson, Loveland, & Theobald, 2005; Radelo, Hammer, &

CONTACT Timothy T. Kennedy tkennedy@uwsp.edu Geography and Geology Department, University of Wisconsin-
Stevens Point, Stevens Point, WI, USA
2016 Informa UK Limited, trading as Taylor & Francis Group
2 T. T. KENNEDY AND H. VEREGIN

Stewart, 2005; Theobald, 2001). As parcel sizes decrease in sprawl-aected areas, strain between
rural working lands that contribute goods and services to society and rural lands that support less
tangible recreational or lifestyle options can occur, especially in amenity-rich rural areas (Gobster &
Rickenbach, 2004; Haines & McFarlane, 2012; McGranahan, 1999). Working lands provide economic
and social subsistence to farmers and forestland holders, which can be threatened by decreasing
parcel sizes and movement toward recreational and non-resource management of rural lands
(Gobster & Rickenbach, 2004; Haines & McFarlane, 2012).
Compatibility of land use does occur, but conict is more the norm. For example, rural
forested areas may support both a timber supply and the sporting traditions of hunters and
outdoor recreationalists (Gobster & Rickenbach, 2004). On the other hand, agricultural areas
with highly productive soils are prime targets for conversion to other land uses due to the
prevalence of at topography, absence of forests, and presence of well-drained soils capable of
supporting buildings and infrastructure (Daniels & Bowers, 1997; Haines & McFarlane, 2012;
Mariola, 2005). Beyond their initial construction usefulness, these traits do not carry long-term
meaning to a developer, but they can be critical to the economic success of farmers (Mariola,
2005). The long-term value of productive agricultural and forestry land does not always resonate
with sectors of society that view land only as a developable commodity. Unfortunately, once
land is lost to development it rarely, if ever, returns to its former state (Daniels & Bowers, 1997;
Mariola, 2005).
The rst step in the process of land-use change in rural working lands areas is often a tax parcel
split, also known as parcelization: the subdivision of larger landholdings into smaller landholdings
(Gobster & Rickenbach, 2004; Haines, Kennedy, & McFarlane, 2011). In rural amenity-rich areas
parcelization can stem from urban growth, second-home development for recreation or retirement,
or demand for residences in rural enclaves distant from urban centers (Gobster & Rickenbach,
2004). In any case the result is often a pattern of more development; development has been shown
to attract other development, increasing the likelihood of further land-use change (Haines et al.,
2011). Thus, parcelization is often a precursor to land-use change aecting rural working lands
including forest fragmentation and the loss of productive agricultural land (DeCoster, 1998;
Donnelly & Evans, 2008; Mehmood & Zhang, 2001).
There are several advantages to focusing on parcelization as an agent of land-use change. First,
in many U.S. states (including Wisconsin) tax parcel splits are recorded at the county level (LaPierre
& Germain, 2005). County tax records provide a precise, authoritative database of parcelization over
time. Second, the tax parcel is the most appropriate unit of analysis for microscale land-use change,
because decisions made by property owners occur at the parcel level, which directly aect the rural
landscape. While tax parcels are subdivided for multiple reasons, subsequent land sales to fund
high property tax or estate tax burdens can drive parcelization, introducing new and diverse
landowners to rural working lands areas (DeCoster, 1998; LaPierre & Germain, 2005; Stone &
Tyrrell, 2012). In the United States, one of the inherent rights owners receive with their property
is the right to alter their land use if subdivision regulations, zoning, and biophysical factors such as
soil quality will allow; strong private property rights are part of core American values (Jacobs &
Paulsen, 2009).
New owners bring dierent values and goals for rural lands. Some may adopt a hands-o
management style; some may actively manage their forests for harvesting purposes, while others
may manage their property for personal recreation and hunting. While the specics vary, the
ultimate result of parcelization is a net change in rural land-use change patterns as a reection of
the values and interests of the new owners. Moreover, the process of parcelization leading to land-
use change often begins imperceptibly. Parcel splits are eectively invisible to anyone not actually
monitoring them. This is especially true in rural areas with minimal planning and zoning regulation,
where a tax parcel split that falls outside of state minimum size requirements falls outside of any
planning or review processes (Ohm, 1999). The agents of land-use change can be in place long
before change is noticeable on the ground.
JOURNAL OF LAND USE SCIENCE 3

While parcelization and subsequent land-use change are linked to land conversion and the loss
of productive working lands in rural America, identifying the complex factors that impact the land
conversion process can be challenging (Haines & McFarlane, 2012; Lambin, Geist, & Rindfuss, 2006;
Turner II, Lambin, & Reenberg, 2007). Human decision-making is aected by a host of factors from
multiple and disparate dimensions of human life (Geist et al., 2006; Geoghegan et al., 2001; Irwin &
Bockstael, 2002). Not only are these factors often hard to dene and quantify but are also
intertwined in complex ways that do not lend themselves to easy explanation. Notwithstanding
other national and global inuences, the proximate and underlying factors of parcelization and
land-use change that represent the four critical dimensions of landowner decision-making include
the social, economic, geographical/biophysical, and regulatory realms (Agarwal, Green, Grove,
Evans, & Schweik, 2002; Bell & Irwin, 2002; Geist & Lambin, 2002).
Traditionally, models of land-use change often reect dierent disciplinary viewpoints: spatially
explicit, high-resolution models have been developed by geographers, and others with a biophy-
sical or remote sensing perspective, while models featuring socioeconomic factors to help under-
stand the underlying human decision-making process have been preferred by sociologists and
economists (Irwin & Geoghegan, 2001; Radelo, Hagen, Voss, Field, & Mladeno, 2000). Spatially
explicit models often make use of remotely sensed data, which is relatively easy to capture, archive,
and process. However, such data tend to favor biophysical and spatial attributes over socioeco-
nomic ones (Lillesand & Kiefer, 1994; Radelo et al., 2000). Unlike remotely sensed data, socio-
economic data are most often captured at the individual level and may be qualitative as well as
quantitative, increasing the challenge in evaluating the social and economic factors driving land-
use change (Irwin & Geoghegan, 2001; Verburg, Schot, Dijst, & Veldkamp, 2004). Combining spatial
and socioeconomic data can also be challenging given inherit disparities in resolution and
measurement units (Radelo et al., 2000; Veldkamp & Lambin, 2001; Verburg, Kok, Pontius, &
Veldkamp, 2006).
Despite these challenges, it is important to include socioeconomic factors in spatially explicit
land-use change models. The human dimensions of land-use change are inuenced by historic and
contemporary attitudes, which stem from social and economic inuences. These inuences are
some of the underlying root causes of land-use change, even though they may be more dicult to
capture and represent in an empirical model format (Lambin, Geist, & Lepers, 2003). While some
researchers choose to incorporate only easily obtainable data, others recognize the value of
including more challenging socioeconomic spatial data in land-use change models (Haines &
McFarlane, 2012). By failing to incorporate these additional dimensions into our models, we risk
attributing land-use change to a limited set of explanatory variables merely because they are easy
to capture. We also risk missing the insights and explanatory power added by the missing variables.
This study examines the social, economic, geographic/biophysical, and regulatory factors under-
lying the human decision-making process leading to parcelization. The specic research questions
examined are as follows:

(1) What are the signicant social, economic, geographic/biophysical, and regulatory drivers
and factors that lead to parcelization in rural amenity-rich areas in the United States?
(2) What is the likelihood of a parcelization event based on a parcels signicant social,
economic, geographic/biophysical, and regulatory attributes?

The study area encompasses six municipal towns within two counties in rural Wisconsin, represent-
ing a microcosm of rural, natural resource amenity-rich areas in the United States. The spatial unit
of analysis is the tax parcel, an appropriate level of analysis for human decision-making about
parcelization and land-use change. We adopt a spatial approach incorporating geographic expla-
natory variables and variations over space as well as a temporal approach to account for change
over a period of 35 years, starting in 1972 and continuing to recent time, across a mix of socio-
economic, environmental, and regulatory changes. The primary statistical method used in the
4 T. T. KENNEDY AND H. VEREGIN

study is logistic regression (LR) analysis, which allows for a probability-based assessment of
parcelization as a function of social, economic, geographic/biophysical, and regulatory factors
that aect landowner decision-making.

2. Methods
2.1. Study area
Two counties in rural Wisconsin contribute to the study area of this project (Figure 1). Bayeld
County, the most northern county in the state, represents a traditionally forested region with over
82% forest cover (Schmidt, 1997). With Lake Superior on two sides and the City of Bayeld as the
launching point to the Apostle Islands National Lakeshore, the county has also long served as a

Figure 1. Study areas in Wisconsin.


JOURNAL OF LAND USE SCIENCE 5

prominent tourist locale. The countys demographic history since 1900 diers from the state as a
whole. In the early twentieth century, following the northern forest cutover, the county experi-
enced a signicant drop in population, driven by the failure of mass agriculture from the countys
poor soils and northern climate (Gough, 1997). Since the 1970s, the county has experienced
population growth due to its attractiveness for recreation, tourism, and retirement. The nearest
area of large urban inuence is the Duluth/Superior area with a combined urban population of
over 131,000 (U.S. Census Bureau, 2010).
Columbia County, in the south-central portion of the state, represents a traditionally agricultural
region with over 87% of land in agricultural use or open space (Columbia County, 2013). The
countys demographic characteristics have generally mirrored statewide trends, with population
growing by an average of 6.1% per decade from 1930 to 2010. Between 1940 and 2010, housing
grew at a rate of 15.6% per decade. Geographically, Columbia County is adjacent to Dane County,
the second most populous county in the state and home of the state capitol in Madison.
Bayeld and Columbia counties represented a rural area under pressure from residential devel-
opment and a peri-urban area under pressure from urban growth (Haines & McFarlane, 2012).
Study selection criteria included active engagement in the comprehensive planning process,
interest from local ocials in maintaining favorable relationships with researchers, and the avail-
ability of geographic information system (GIS)-compatible digital parcel data (Haines & Olson,
2007). In each county, three towns were chosen for analysis: Bayeld, Barnes, and Delta in
Bayeld County and Lodi, West Point, and Springvale in Columbia County (Figure 1). In
Wisconsin, towns are legal administrative entities with some of the same roles and responsibilities
as incorporated cities and villages. Additionally, towns in each county were also selected for their
historic rates of parcelization with one each recently demonstrating high (Bayeld, Lodi), medium
(Barnes, West Point), and low (Delta, Springvale) rates of historic parcelization (Haines & Olson,
2007).

2.2. Parcelization
For each of these six towns, historic tax parcels were obtained from an earlier study that used
reverse parcelization methods (Haines et al., 2011). Reverse parcelization begins with a current tax
parcel layer in GIS format, provided by the county Land Information Ocer. For Bayeld County the
baseline year was 2007, while for Columbia County it was 2006. Parcels were then aggregated
going back in time based on the interpretation of tax assessment roll data. This allowed for the
construction of historic parcel layers for four snapshot years to represent three distinct time
periods. The initial time period 1953 (Columbia County)/1954 (Bayeld County)to 1972
represents parcelization histories prior to the enactment of shoreland zoning restrictions in
Wisconsin. The second time period 19721991 represents the immediate post-shoreland zoning
period when land subdivision was subjected to minimum size and water frontage requirements for
parcels within 1000 ft (304.8 m) of a lake or 300 ft (91.4 m) of a navigable stream (Ohm, 1999). The
nal time period 19912006 (Columbia County)/2007 (Bayeld County) represents signicant
residential growth during a prosperous economic period before the recession of 2008.

2.3. Explanatory variables


A literature search revealed signicant drivers of parcelization and land-use change from the
previously identied dimensions of human decision-making. Social factors include local and
regional demographics, education, lifestyle, landowner age, land tenure, deaths, and absentee/
resident ownership (Geoghegan et al., 2001; Gobster & Rickenbach, 2004; Mehmood & Zhang,
2001). Economic factors include taxes, crop and timber commodity prices, land value, o-farm
income, and availability of discretionary income (Geoghegan et al., 2001; Gobster & Rickenbach,
2004; Heimlich & Anderson, 2001). Regulatory or political factors include agricultural and forestry
6 T. T. KENNEDY AND H. VEREGIN

program incentives, environmental regulation uncertainty, and land subdivision regulations and
zoning (Daniels & Bowers, 1997). Geographical and biophysical factors include a parcels proximity
to desirable traits such as roads, water, urban services, agriculture, or public lands. They also
include the parcels slope, size, and neighborhood land uses (Gobster & Rickenbach, 2004; Haines
et al., 2011; McGranahan, 1999).
We chose to create explanatory variables from these four dimensions based on publicly avail-
able data, available project resources, and signicant variables revealed by our literature search
(Table 1). Digital tax parcels in GIS-compatible format were attributed with the applicable social,
economic, geographical/biophysical, and regulatory variables. We acquired social data from U.S.
Census records and encoded individual tax parcels based on municipal town values. We acquired
economic data from tax assessment roll data and encoded each tax parcel individually. Geographic
and biophysical data were created using various GIS tools, while regulatory data were acquired
from each countys Land Information Oce (for land-use zoning) or was created using GIS tools (for
shoreland zoning).

2.4. Multicollinearity
We examined all pairs of explanatory variables for multicollinearity that could aect LR model
tness. Non-orthogonal explanatory variables can cause uctuations in model parameters and
signicance, possibly resulting in erroneous interpretation (Craney & Surles, 2002). However, due
to the large sample size (all years N = 29,755) we found many correlation coecients were
statistically signicant, leading to the conclusion that little utility could be extracted from the
correlation analysis (Garcia, 2010).
Instead, we employed variance ination factor (VIF) analysis to measure the eects of multi-
collinearity on LR model tness (Craney & Surles, 2002; OBrien, 2007). A VIF analysis quanties how
much of the variability in an explanatory variable is explained by the other explanatory variables
(Craney & Surles, 2002). For each study year, we iteratively removed the explanatory variable with
the highest VIF value from the subsequent LR model. The nal LR model was selected when the VIF
values for all explanatory variables were below a value of 5.0 a commonly used heuristic
resulting in a set of explanatory variables with minimal multicollinearity impacts on model validity
(Craney & Surles, 2002).

2.5. Logistic regression analysis


Since a parcelization event occurs when a parcel splits, the event is a binary, or dichotomous,
response. LR models are widely used for such a dichotomous response event (Cho & Newman,
2005; Haines & McFarlane, 2012; Hosmer Jr & Lemeshow, 2004). We developed LR models for the
study area for 1972, 1991, and 2006/2007 using SPSS version 21. The dependent variable in this
model is a parcel split; a value of 1 indicates a parcel that split between the current and previous
study years, while a 0 indicates a parcel that did not split. To comply with model requirements,
certain explanatory variables developed, Managed Forest Law (MFL) program enrollment, and all
regulatory variables were classied as categorical variables, with data values of 1 or 0. All other
variables were entered in their raw data format.
The LR models in each study year provide a coecient for each explanatory variable, bi, which is
a measure of the contribution of that variable to a parcelization event, which is the dependent
variable. The standard LR model equation is as follows:
 
^
p
ln b0 b1 X1 b2 X2 . . .
1p ^

^ = the estimated probability of a parcelization event


where p
Table 1. Explanatory variables used in analysis.
Name Description Source* Dim.** Year(s)***
Parcel size Size of parcel (hectares) GIS G 1, 2, 3
Water frontage Quantity of parcel water frontage (meters) GIS G 1, 2, 3
Distance to water Distance from parcel to nearest open water (meters) GIS G 1, 2, 3
Percent forest cover Percent of parcel with forest cover GIS G 1, 2, 3
Percent agriculture cover Percent of parcel with agriculture cover GIS G 1, 2, 3
Percent unsuitable soils Percent of parcel with unsuitable soils GIS G 1, 2, 3
Developed Parcel developed status, yes = 1, no = 0 GIS G 1, 2, 3
Distance to nearest road Distance to nearest road (meters) GIS G 1, 2, 3
Nearest urban services Distance to nearest urban services (water/sewer) (meters) GIS G 1, 2, 3
Parcel average elevation Parcel average elevation (meters) GIS G 1, 2, 3
Road time to nearest major urban area Road travel time to nearest major urban area (minutes) GIS G 1, 2, 3
Managed Forest Law program enrollment Parcel enrollment in the Managed Forest Law program, yes = 1, no = 0 Tax E 1, 2, 3
Land value Land value ($) (not including improvements) as recorded from the tax assessment rolls Tax E 1, 2, 3
Town per capita value Total private land value per town divided by total population Tax/Census E 1, 2, 3
Town population Population of town taken from nearest census Census S 1, 2, 3
Town population increase Increase in town population from previous snapshot year Census S 1, 2, 3
Town population density Town population divided by town size (hectares) GIS/Census S 1, 2, 3
Town median age Median age of town residents Census S 1, 2, 3
Shoreland zoning Parcel location in a state regulated shoreland zone, yes = 1, no = 0 State R 1, 2, 3
Land-use zoning residential Parcel location in a residential zone, yes = 1, no = 0 County R 3
Land-use zoning agriculture Parcel location in an agriculture zone, yes = 1, no = 0 County R 3
Land-use zoning forest Parcel location in a forest zone, yes = 1, no = 0 County R 3
Land-use zoning commercial/industrial Parcel location in a commercial/industrial zone, yes = 1, no = 0 County R 3
Land-use zoning recreation/conservancy Parcel location in a recreational or conservancy zone, yes = 1, no = 0 County R 3
* Source (data source); GIS: explanatory variables created using GIS tools and techniques; Tax: variable data obtained from tax assessment rolls; Census: variable data obtained from U.S. Census;
County: variable data obtained from county government; State: variable data obtained from WI State Statutes.
** Dimension; G = geographical/biophysical, S = social, E = economic, R = regulatory.
*** Year(s) 1 = 1972, 2 = 1991, 3 = 2006/2007.
JOURNAL OF LAND USE SCIENCE
7
8 T. T. KENNEDY AND H. VEREGIN

bi = LR coecients (derived)
Xi = explanatory variables
Additionally, the LR model also calculates an odds ratio and the standard error. The odds ratio is
the increase in the odds of parcelization for every one-unit change in the explanatory variable,
holding all other variables constant. Following recommended practice, we report the LR model
coecients, standard error, statistical signicance, and odds ratio in the results section (Agresti &
Finlay, 2009; Hosmer Jr & Lemeshow, 2004; Peng, Lee, & Ingersoll, 2002).

2.6. Logistic regression model performance assessment


Assessment of model performance for LR can be complicated. One method to assess LR model
results is to validate the predicted probabilities in a classication table. A probability threshold or
cuto of 0.5 is used to classify the predicted probability as either a predicted parcelization event
(probability > 0.5) or a non-event (probability 0.5) (Hosmer Jr & Lemeshow, 2004, p. 156). The
overall accuracy is simply the number of correctly classied cases divided by the total number of
cases.
Finally, another method of model tness assessment involves the area under a receiver operat-
ing characteristic (ROC) curve. The ROC curve is useful because it summarizes the predictive power
of a model over all probability cuto values, not just 0.5. The ROC curve plots the trade-o between
Sensitivity and 1-Specicity for all possible cuto values. Each point on an ROC curve represents
the accuracy of a prediction result for a specic cuto value. A perfect prediction would occur at
the upper left corner of the graph. A random guess falls on the straight green line. An ROC curve
above the line is a better-than-random predictor; the larger the area under the curve, the higher is
the predictive power of the model (Fidell & Tabachnick, 2006, p. 461).

3. Results
3.1. Multicollinearity
Multicollinearity was a particular problem for the town-level variables, which showed signicant
correlations (p 0.01) among themselves and with other explanatory variables. We observed
strong positive correlations between town population and town population density (r = 0.967 in
1972, r = 0.988 in 1991, and r = 0.990 in 2006/2007), town population and town population
increase (r = 0.609 in 2006/2007), and to a lesser degree town population increase and town
population density (r = 0.532 in 2006/2007). We also observed strong positive correlations between
town per capita value and town median age (r = 0.904 in 1991, r = 0.920 in 2006/2007), although
not in 1972. The correlations for town population increase and town per capita value show a
change from positive to negative to positive over the three time periods of the study (r = 0.574 in
1972, r = 0.827 in 1991, and r = 0.320 in 2006/2007). Most other pairs of town variables show
strong negative correlations.
We also found the town variables to be strongly correlated with distance to nearest urban
services (r values range from 0.215 to 0.974), time to nearest urban area (r values range from
0.799 to 0.643), and mean elevation (r values range from 0.510 to 0.788). Town per capita value
and town median age generally showed positive correlations with these three variables (distance
to nearest urban services, time to nearest urban area, and mean elevation). Town population and
town population density showed negative correlations, and town population increase showed a
mixture of negative and positive correlations with these three explanatory variables. The inter-
correlations among these variables show the incompatibility of mixing town-level variables with
data collected at the parcel level.
The remaining economic, geographic/biophysical, and regulatory explanatory variables
show lower levels of intercorrelation, suggesting that they are measuring dierent dimensions
JOURNAL OF LAND USE SCIENCE 9

and phenomena at the parcel level. This includes MFL enrollment, land value, and all regula-
tory variables. Appendix 1 includes the descriptive statistics for all nonbinary explanatory
variables.

3.2. LR model analysis


The nal LR model coecients, standard error, statistical signicance, and odds ratio values for
1972, 1991, and 2006/2007 are summarized in Table 2. Most of the odds ratio results indicate a
slightly higher likelihood of a parcelization event (i.e. dependent variable) for every one-unit
increase in the individual explanatory variables or the presence of some characteristic of a binary
explanatory variable. For example, in the 1972 model the odds ratio for shoreland zoning is 1.15,
indicating that when shoreland zoning is present, the odds are 1.15 times greater for a parcel split
than a parcel non-split (i.e. the dependent variable) (Peng et al., 2002). However, there were some
notable exceptions. In 2006/2007, the odds ratio for all land-use zoning explanatory variables
(except for recreational land-use zoning) ranged between 0.53 and 0.64. These lower values are not
exhibited by any of the other explanatory variables in any study year. Alternatively, the odds ratio
for a parcels enrollment in the MFL program (1991) and the increase in town population (1991)
were very large at 11.5 and 183.9, respectively.
Signicant positive relationships (associated with an increase in the likelihood of parcelization)
occur for the following variables (Table 2): percent forest cover (1972 and 1991 only); percent
agriculture cover; developed (1991 only); mean elevation (1972 and 1991 only); MFL enrollment
(1972 and 1991 only); land value (1991 and 2006/2007 only); town per capita value (1991 only);
town population increase (1991 only); town population density (1991 only); and shoreland zoning
(1991 only). Signicant negative relationships (associated with a decrease in the likelihood of
parcelization) occur for the following variables: parcel size; distance to water (1972 and 2006/
2007 only); percent unsuitable soils (1972 and 2006/2007 only); distance to road (1972 and 1991
only); time to nearest urban area (1991 only); town median age (1972 only); and land-use zoning
for residential, agriculture, forest, and commercial/industrial (2006/2007 only).
In general, these results indicate that parcelization is more likely to occur in areas of forest or
agricultural land use, on land enrolled in the MFL program, in areas of higher elevation, and in
areas that are already developed and/or developing. Developed/developing areas experience
higher land values, higher incomes, higher population densities, and higher population growth
rates. In contrast, parcelization is less likely to occur when parcels are large, when they are far from
water, roads, urban areas, when unsuitable soils are present, and when the population has a high
median age. These results were mostly expected, as they reect the tendency of development to
attract new development, the desirability of land around lakes and within easy access of services
and amenities, and a preference for parcels containing agricultural or forested areas. With the
exception of shoreland zoning, the existence of zoning restrictions has a tendency to suppress
parcelization, even if the zoning is for residential uses. Paradoxically, shoreland zoning is associated
with higher probabilities of parcelization. These results are interpreted in more detail in the
discussion section.

3.3. LR model performance


Model tness assessment results demonstrated consistent trends over time. Classication table
measures decline from the past to the contemporary years as illustrated by the cross-tabular results
of the predicted probabilities of parcelization against actual (observed) parcelization events
(Table 3). Mirroring the general trend of the classication table assessment results, the ROC area
under the curve declined in 1991 with additional decline in 2006/2007 (Figure 2). The area under
each ROC curve is between 0.705 and 0.888 for all study years (Figure 2).
10

Table 2. LR model coecient (), standard error (S.E.), statistical signicance (Sig.), and odds ratio (Exp()).
1972 1991 2006/2007
Explanatory Variable S.E. Sig. Exp() S.E. Sig. Exp() S.E. Sig. Exp()
Parcel size (ha) 7.6 102 2.4 103 ** 9.2 101 5.9 102 2.5 103 ** 9.4 101 3.3 102 2.5 103 ** 9.6 101
Water frontage (100 m) 9.6 105 1.0 104 1.0 10+0 1.1 104 1.0 104 1.0 10+0 8.4 105 1.1 104 1.0 101
Distance to water (100 m) 7.7 105 6.0 106 ** 1.0 101 2.4 106 6.3 106 1.0 10+0 3.1 105 5.1 106 ** 1.0 101
Percent forest cover 5.7 103 1.2 103 ** 1.0 10+0 6.5 103 9.2 104 ** 1.0 10+0 5.2 104 8.9 104 1.0 10+0
Percent agriculture cover 9.6 103 1.3 103 ** 1.0 10+0 9.0 103 1.1 103 ** 1.0 10+0 7.3 103 1.3 103 ** 1.0 10+0
T. T. KENNEDY AND H. VEREGIN

Percent unsuitable soils 1.8 102 3.7 103 ** 9.8 101 2.1 103 1.5 103 1.0 10+0 6.0 103 1.4 103 ** 9.9 101
Developed 9.7 102 9.1 102 1.1 10+0 7.7 101 7.3 102 ** 2.1 10+0 9.7 102 6.1 102 9.0 101
Distance to road (100 m) 2.6 104 5.0 105 ** 1.0 101 2.5 104 4.6 105 ** 1.0 101 1.1 104 4.2 105 1.0 101
Nearest urban services (m)
Mean elevation (10 m) 3.1 103 8.4 104 ** 1.0 10+0 5.0 103 8.6 104 ** 1.0 10+0 9.7 104 5.6 104 9.9 101
Time to nearest urban area (min) 2.7 103 1.9 103 1.0 10+0 5.7 103 1.5 103 ** 9.9 101 6.8 104 1.8 103 1.0 10+0
MFL enrollment 7.1 101 3.1 101 * 2.0 10+0 2.4 10+0 3.9 101 ** 1.1 10+1 2.4 101 1.5 101 1.2 10+0
Land value ($1000) 1.1 105 1.6 105 1.0 10+0 1.2 105 2.5 106 ** 1.0 10+0 1.6 106 3.3 107 ** 1.0 10+0
Town per capita value 1.1 104 7.2 106 ** 1.0 10+0
Town population
Town population increase 5.2 10+0 2.9 101 ** 1.8 10+2 2.0 10+0 2.0 101 1.3 101
Town population density 2.6 103 2.8 103 9.9 101 2.8 102 2.7 103 ** 1.0 10+0
Town median age 7.1 102 6.2 103 ** 9.3 101
Shoreland zone 1.4 101 7.9 102 1.1 10+0 5.2 10+0 4.1 10+0 * 1.1 10+0 3.1 102 6.1 102 1.0 10+0
Residential LU zone 4.4 101 9.1 102 ** 6.4 101
Agriculture LU zone 4.7 101 1.1 101 ** 6.2 101
Forest LU zone 3.8 101 1.0 101 ** 6.8 101
Commercial/Industrial LU zone 6.3 101 1.7 101 ** 5.3 101
Recreational LU zone 8.6 102 1.0 101 1.0 10+0
Constant 6.1 101 4.7 101 1.8 10+0 9.7 10+0 5.5 101 ** 5.6 105 1.6 10+0 4.6 101 ** 5.3 10+0
Blank elds indicate explanatory variables not included in snapshot year LR model.
** Signicant variables at the p < 0.01 level
* Signicant variables at the p < 0.05 level
JOURNAL OF LAND USE SCIENCE 11

Table 3. LR model classication table results.


Predicted
Observed S NS Percent correct
1972 S 2308 629 78.6%
NS 963 4566 82.6%
1991 S 612 1469 29.4%
NS 471 7372 94.0%
2006/2007 S 21 2338 0.9%
NS 27 8979 99.7%
S: parcel split; NS: parcel non-split

Figure 2. ROC curves.

4. Discussion
4.1 Explanatory variable implications
The LR model results provide conrmation of expected relationships for some explanatory variables
and unexpected results for others. The signicance of individual variables diered over the three
time periods of the study, while the LR model goodness-of-t measures declined from historic to
contemporary time periods. Results also underscore the challenges associated with incorporating
social and economic variables at the parcel level.
In some respects, the results of this study align with ndings from other research. Distance to
water, distance to nearest road, and percent unsuitable soils show the expected negative relation-
ship to parcelization while percent forest cover and percent agriculture cover show the expected
positive relationship to parcelization (except in cases where these variables were not signicant).
This suggests that in certain years there may be a more limited supply of parcels available
whether parcels near water or with suitable soils in 1991 or parcels with forest cover in 2006/2007
due to local market conditions. In other cases, the results are more ambiguous. For example, parcel
12 T. T. KENNEDY AND H. VEREGIN

size was found to be negatively related to parcelization. This is similar to Ko and He (2011) but
contrary to Haines and McFarlane (2012).
Road travel time to the nearest metropolitan urban area is signicant in 1991 only, with a
negative relationship to parcelization. This suggests some ambiguity in the eect of commuting
times, with less of an impact felt in earlier (1972) and contemporary times (2006/2007), although
perhaps for dierent reasons. Mean elevation shows a positive relationship to parcelization, but not
in 2006/2007, again reecting an apparent change in the importance of this variable over time.
Likewise, the presence of existing development is positively linked to parcelization, but only in
1991.
Socioeconomic explanatory variables mostly gathered at the town level proved to be
complicated in terms of their eects on the nal model. Of the ve original town-level variables,
only town median age was signicant and remained in the nal model in 1972 (negative relation-
ship to parcelization). In 1991, town per capita value, town population increase, and town popula-
tion density all with a positive relationship to parcelization were signicant. In 2006/2007, none
of the ve variables was signicant. Despite the variation in these results, it is likely that demo-
graphic variables do have an eect on parcelization. Our results may be inuenced by the unit of
analysis, since they were derived at the town level, not the parcel level; thus, we cannot conclude
too much. However, at some level our results support other research that population change is
linked to land conversion (Gobster & Rickenbach, 2004; Heimlich & Anderson, 2001; Mehmood &
Zhang, 2001). Conversely, Carrion-Flores and Irwin (2004) and Irwin, Bell, and Geoghegan (2003)
found that total population was not signicantly related to the probability of parcel conversion.
The presence of urban services (e.g. sewer and water) despite being a spatial variable is
highly correlated with other variables in the database. This may indicate that the distance to
nearest urban services is linked to population density; however, town population and distance to
nearest urban services were both eliminated from the LR models by the VIF analysis, indicating
intercorrelation with other variables. Urban areas with sewer and water services typically maintain a
higher population density than rural areas (Heimlich & Anderson, 2001; Levy, 2009). Given the
challenges of collecting demographic parcel-level data, the spatial distance to nearest urban
services may provide a proxy population dimension for future studies.
The economic explanatory variables demonstrated some expected and unexpected results. Land
value showed an expected positive relationship to parcelization (except in 1972). This result is
supported in other studies showing that as land value increases, taxes also increase. Property taxes
are an important factor in an owners decision to sell and subdivide larger landholdings (DeCoster,
1998; Gobster & Rickenbach, 2004; Stone & Tyrrell, 2012). On the other hand, LR results show that
enrollment in a managed forestry incentive program (MFL enrollment explanatory variable) is
signicantly linked to a higher likelihood of a parcelization event for 1972 and 1991. The MFL
and Forest Crop Law (FCL) in Wisconsin are tax incentive programs that were designed primarily to
help landowners maintain sustainable commercial working forests. It should be noted that forest
landowners who enter tax programs like the MFL bring diverse backgrounds, values, and goals to
their reasons for enrollment; thus, there may be other explanations for the signicance of this
explanatory variable (Locke & Rissman, 2012).
Regulatory variables exhibited varied results. Shoreland zoning, designed to limit the amount of
development within a specied distance of water, was signicant only in 1991 and, surprisingly, is
linked to a higher likelihood of parcelization. This might indicate at that time a remaining inventory
of suciently sized parcels for further subdivision within regulatory requirements, despite their
location within a shoreland zone. This may also indicate the strong demand for parcels near water
(Haines & McFarlane, 2012; Lansford & Jones, 1995). In 1972 and 2006/2007, residence within a
shoreland zone was not a signicant factor in limiting parcelization.
Nonresidential land-use zoning (agriculture, forest, and commercial/industrial) is negatively
linked to parcelization in 2006/2007, the only year these data were available. This may indicate
that land-use zoning goals are being achieved by restricting compatible uses to planned areas.
JOURNAL OF LAND USE SCIENCE 13

However, results also show that residential zoning is negatively linked to parcelization, contrary to
zoning goals to restrict residential dwellings to residentially zoned areas. Caution is recommended
when interpreting the zoning results, however, as they represent only a single time period.

4.2 LR model implications


Examining the global LR model results illustrates the challenge of highlighting the individual drivers and
factors of parcelization with a limited set of explanatory variables to represent the multidimensionality of
human decision-making. In 1972, the multidimensional variables in the LR model correctly predicted a
parcel split 78.6% of the time, while correctly predicting a parcel non-split 82.6% of the time (Table 3). In
1991, the predictive capability of the explanatory variables substantially declined to correctly predict a
parcel split only 29.4% of the time; while increasing the correct parcel non-split prediction rate to 94.0%. In
2006/2007, the results indicate a continued and substantial decline in the correct prediction of parcel
splits, down to 0.9%; while the capability of the model to correctly predict parcel non-splits increased to
99.7%. In essence, the 2006/2007 LR model failed to predict parcel splits, but did predict parcel non-splits
more successfully. These results further support our argument that new explanatory variables are needed
to explain contemporary parcelization.
The nal LR model runs and accompanying VIF analysis support the utility of a consistent scale of
analysis among explanatory variables. Similar to other studies, spatial variables provide the primary
explanatory power for parcelization. The observed decline in model tness over time, however, suggests
that while spatial factors may have been the primary driving factors of parcelization historically in both
agricultural and forested landscapes in Wisconsin, new forces not yet captured in this study have more of
an eect driving contemporary parcelization. Factors driving parcelization are likely becoming more
complex over time, requiring the inclusion of new explanatory variables.
Caution is warranted when interpreting model tness using the ROC curve measures. Generally, ROC
area-under-curve values between 0.8 and 0.9 are considered to be excellent (Hosmer Jr & Lemeshow,
2004). Despite meeting the criteria for excellent based on Hosmer Jr and Lemeshow, the area under the
curve is dependent upon the threshold cuto value, which for this study is 0.5, and the interpretation of
the ROC results toward the purpose of the study. For comparison, Haines and McFarlane (2012) obtained
an ROC value of 0.771 using only spatial data to predict a parcel split, while Schneider and Gil Pontius
(2001) reported successful validation of LR models with ROC values between 0.65 and 0.70 to produce
deforestation suitability maps. The ROC area-under-curve results mirror the declining trend in model
tness from the earlier to most recent years in the present study.

4.3 New explanatory variables needed


What this research has merely hinted at is the essential need to incorporate personal decision-
making explanatory variables into our models. Stone and Tyrrell (2012) found from a survey of
landowners in the New York City watershed that diering personal circumstances, not diering
values or goals, distinguish landowners who decided to subdivide from those landowners who did
not subdivide. They found that landowners who subdivided were more likely to have less formal
education, more likely to be retired (and have higher health risk), and more likely to have lower
incomes. Historically, forest and farmland supported our economies; however, declining family
heritage or the lack of family cohesiveness may be a contributor to increasing parcelization. The
death of elderly working landowners combined with a lack of family interest in working the land is
also responsible for parcelization and change (DeCoster, 1998; Mehmood & Zhang, 2001). As
farming shifts to larger operations in the United States, small commercial farms are experiencing
decreases in size and quantity. Household factors such as o-farm income, farm income, farm
prot, and farm sales become important indicators of small farm success (Hoppe, MacDonald, &
Korb, 2010). The data required to represent these factors are located at the household or land-
owner level.
14 T. T. KENNEDY AND H. VEREGIN

Other studies have documented that the distance to urban areas plays a signicant role in parcelization
or land conversion (Cho & Newman, 2005; Gustafson, Hammer, Radelo, & Potts, 2005; Levia, 1998). The
changing nature of our global lifestyle along with infrastructure and technological advances may be
altering traditional perspectives of residential locations near urban centers and employers (McGranahan,
1999). Studies have noted that telecommuting workers often do so from greater distances than non-
telecommuting workers (Mokhtarian, Collantes, & Gertz, 2004; Zhu, 2013). Another study reported that
telecommuting can lead to residential locations farther from the workplace, which can allow more
exibility for residential location in suburban and rural areas (Lund & Mokhtarian, 1994). This may be
suggested in our study by the lack of statistical signicance of the commuting time to nearest metropo-
litan urban area in two of the three time periods.
We recommend that future research explore social and economic variables at the household or
landowner level to better explain parcelization. Geoghegan et al. (2001) utilized household surveys to
enhance a remote sensing analysis of tropic deforestation factors in Mexico. They reported increased
model explanatory value through the addition of variables such as household population, size of house-
hold land holdings, number of household individuals that are dependent upon agricultural output, total
value of livestock during the year, household education level, and amount of o-farm income generated.
Future parcelization studies in areas similar to this study may benet from including household-level
explanatory variables such as farm income, o-farm income, population, ages, education level, remote or
telecommuting employment, rural living values, and land stewardship values in explanatory models.
Household-level data like these can be collected through landowner surveys. This is supported by the
recent survey results of eld natural resource managers in the Lake States (MI, MN, WI), which found that
the respondents believe that parcelization drivers can be characterized as economic, familial, recreational,
and cultural (Kilgore & Snyder, 2016).

4.4 Parcelization continues


Parcelization is an ongoing process in both forested and agricultural areas of the state. During the
study period, the forested study area experienced a more than 50% decrease in private parcel sizes
(10.75.3 ha), while the agricultural study area experienced a more than 57% decrease in private
parcel sizes (12.15.2 ha) (Figure 3). These declines in private parcel sizes should concern planners

Figure 3. Private parcel sizes (hectares) over time in the study area.
JOURNAL OF LAND USE SCIENCE 15

and resource managers interested in larger parcel sizes to support forestry, agriculture, and other
natural resource activities.

5. Conclusion
This study illustrates the challenges of combining landowner parcel and aggregated data in a
meaningful way to produce statistical results that go beyond parcel-only analysis (Dungan et al.,
2002). The explanatory variable correlation analysis demonstrates the importance of the parcel as
the appropriate unit of analysis due to the highly intercorrelated nature of the social explanatory
variables represented at the town level. The primary challenge in this study is capturing explana-
tory variables at a consistent parcel scale along all dimensions of human decision-making.
Unfortunately, there is little social and economic data readily available at the parcel level for
many rural areas. Most publicly available population and economic data encompass a rural town
or census block, while regulatory data are often only available for administrative zones or munici-
palities. While some studies utilize individual surveys to obtain socioeconomic data at the parcel
unit, that was not feasible for a study with this large of a quantity of historic parcels. Thus, without
more resource-intensive surveys to collect data, the use of aggregated data as proxies for more
individualized socioeconomic data may be required if we want to incorporate multidimensional
data in singular models. Despite these challenges, we should strive to incorporate all dimensions of
the human decision-making process into land-use change models.
The temporal analysis paints a unique picture of the complex interplay of historic and con-
temporary parcelization drivers. Simple social, economic, and geographical/biophysical factors
robustly explain parcel splits and parcel non-splits in the earliest year examined (1972). Model
tness decrease in 1991 with an additional decrease and a failure to predict parcel splits in 2006/
2007, may suggest that the drivers of contemporary parcelization are more complex than in
previous decades. Adding to this complexity is the very low capability of the model to correctly
predict contemporary parcel splits, and in contrast, the capability of the model to robustly predict
contemporary parcel non-splits. We need to better represent landowner human decision-making
factors in our models, along with nding new methods of variable scale adjustment to incorporate
new and existing contemporary drivers of parcelization and land-use change in our models.
This study has demonstrated the impact of non-geographical/biophysical explanatory variables
in parcelization modeling. Despite the limitations of diering explanatory and dependent variable
scales, social and economic variables play an important role in explaining parcelization and should
not be overlooked. Additionally, this study has illustrated a way to evaluate the ecacy of
regulatory methods to control land use.
Planners and resource managers can draw some value from these results. First, the lack of
signicant explanation of the MFL program to maintain larger forested parcels may require additional
investigation. Locke and Rissman (2012), in a study of MFL/CFL enrollments in Wisconsin, found that
these types of programs enhance land conservation, especially near public lands, but they recom-
mend combining with other policy tools to prevent forest conversion to developed land uses. Thus,
additional policy or management measures may be needed to realize additional eects from con-
servation programs intended to provide property tax relief (LaPierre & Germain, 2005).
Additionally, the ever-changing impact of our connected and global lifestyle may be altering our
traditional consideration of urban inuences on rural and suburban areas, requiring a new per-
spective of parcelization in rural space. Increasing exibility in residential locations may contribute
to rural parcelization in areas not adjacent to urban centers. Whereas rural working lands were
traditionally used for agricultural or forested purposes, technology may allow workers to fulll their
rural or amenity-rich residential location dreams, while concurrently maintaining active and viable
employment outside of large urban areas (Caron, Germain, & Anderson, 2012; DeCoster, 1998). The
impact of these changes may not play out in the landscape via traditional patterns of urban sprawl,
but in more random and stochastic patterns.
16 T. T. KENNEDY AND H. VEREGIN

Generalizability is always a topic of concern for predictive modeling. Other researchers have tended
to use results from sample areas to generalize to larger regions (Gobster & Rickenbach, 2004; Haines &
McFarlane, 2012). In like fashion, the results from this study apply to some extent to rural forested or
agriculture working land areas in Wisconsin and the Upper Midwest. Generalizing beyond this regional
extent is not advisable due to the variability in the drivers of change and the resulting regional spatial
conguration from change (Sohl, Loveland, Sleeter, Sayler, & Barnes, 2010).

Acknowledgments
The authors would like to thank the College of Letters and Science, University of Wisconsin-Stevens Point for funding
support and the Center for Land Use Education at the University of Wisconsin-Stevens Point, Bayeld County,
Wisconsin, and Columbia County, Wisconsin for providing parcel and attribute data.

Disclosure statement
No potential conict of interest was reported by the authors.

Funding
The authors would like to thank the College of Letters and Science, University of Wisconsin-Stevens Point for funding
support.

ORCID
Timothy T. Kennedy http://orcid.org/0000-0003-3776-8606
Howard Veregin http://orcid.org/0000-0002-8988-9373

References
Agarwal, C., Green, G. M., Grove, J. M., Evans, T. P., & Schweik, C. M. (2002). A review and assessment of land-use change
models: Dynamics of space, time, and human choice. Newton Square, PA: USDA, Forest Service, Northeastern Research
Station.
Agresti, A., & Finlay, B. (2009). Statistical methods for the social sciences (4th ed.). Upper Saddle River, NJ: Pearson
Education International.
Bell, K. P., & Irwin, E. G. (2002). Spatially explicit micro-level modelling of land use change at the ruralurban interface.
Agricultural Economics, 27(3), 217232. doi:10.1016/S0169-5150(02)00079-8
Brown, D. G., Johnson, K. M., Loveland, T. R., & Theobald, D. M. (2005). Rural land use trends in the conterminous
United States, 1950-2000. Ecological Applications, 15(6), 18511863. doi:10.1890/03-5220
Caron, J. A., Germain, R. H., & Anderson, N. M. (2012). Parcelization and land use: A case study in the New York City
watershed. Northern Journal of Applied Forestry, 29(2), 7480. doi:10.5849/njaf.10-038
Carrion-Flores, C., & Irwin, E. G. (2004). Determinants of residential land-use conversion and sprawl at the rural-urban
fringe. American Journal of Agricultural Economics, 86(4), 889904. doi:10.1111/j.0002-9092.2004.00641.x
Cho, S.-H., & Newman, D. H. (2005). Spatial analysis of rural land development. Forest Policy and Economics, 7(5), 732
744. doi:10.1016/j.forpol.2005.03.008
Columbia County. (2013). Columbia County comprehensive plan 2030. Columbia County, WI: Author. Retrieved from
http://www.co.columbia.wi.us/columbiacounty/portals/3/County%20Plan/AmendedPlan/ccPlan.pdf
Craney, T. A., & Surles, J. G. (2002). Model-dependent variance ination factor cuto values. Quality Engineering, 14(3),
391403. doi:10.1081/QEN-120001878
Daniels, T., & Bowers, D. (1997). Holding our ground: Protecting Americas farms and farmland. Washington, DC: Island
Press.
DeCoster, L. A. (1998). The boom in forest ownersA bust for forestry? Journal of Forestry, 96(5), 2528.
Donnelly, S., & Evans, T. P. (2008). Characterizing spatial patterns of land ownership at the parcel level in South-Central
Indiana, 1928-1997. Landscape and Urban Planning, 84(34), 230240. doi:10.1016/j.landurbplan.2007.08.004
Dungan, J. L., Perry, J., Dale, M., Legendre, P., Citron-Pousty, S., Fortin, M.-J., . . . Rosenberg, M. S. (2002). A balanced
view of scale in spatial statistical analysis. Ecography, 25(5), 626640. doi:10.1034/j.1600-0587.2002.250510.x
Fidell, L. S., & Tabachnick, B. G. (2006). Using multivariate statistics. New York, NY: Harper and Row.
JOURNAL OF LAND USE SCIENCE 17

Garcia, E. (2010). On correlation coecients and cample cize. Retrieved from https://irthoughts.wordpress.com//?s=on
+correlation+coecients+and+sample+size&search=Go
Geist, H. J., & Lambin, E. F. (2002). Proximate causes and underlying driving forces of tropical deforestation. BioScience,
52(2), 143150. doi:10.1641/0006-3568(2002)052[0143:PCAUDF]2.0.CO;2
Geist, H. J., McConnell, W., Lambin, E. F., Moran, E., Alves, D., & Rudel, T. (2006). Causes and trajectories of land-use/
cover change. In Land-use and land-cover change (pp. 4170). Berlin: Springer-Verlag.
Geoghegan, J., Villar, S. C., Klepeis, P., Mendoza, P. M., Ogneva-Himmelberger, Y., Chowdhury, R. R., . . . Vance, C. (2001).
Modeling tropical deforestation in the southern Yucatn peninsular region: Comparing survey and satellite data.
Agriculture, Ecosystems & Environment, 85(13), 2546. doi:10.1016/S0167-8809(01)00201-8
Gobster, P. H., & Rickenbach, M. G. (2004). Private forestland parcelization and development in Wisconsins
Northwoods: Perceptions of resource-oriented stakeholders. Landscape and Urban Planning, 69(23), 165182.
doi:10.1016/j.landurbplan.2003.09.005
Gough, R. (1997). Farming the cutover: A social history of Northern Wisconsin, 1900-1940. Lawrence: University Press of
Kansas.
Gustafson, E. J., Hammer, R. B., Radelo, V. C., & Potts, R. S. (2005). The relationship between environmental amenities
and changing human settlement patterns between 1980 and 2000 in the Midwestern USA. Landscape Ecology, 20
(7), 773789. doi:10.1007/s10980-005-2149-7
Haines, A., Kennedy, T., & McFarlane, D. (2011). Parcelization: Forest change agent in northern Wisconsin. Journal of
Forestry, 109(2), 101108.
Haines, A., & McFarlane, D. (2012). Factors inuencing parcelization in amenity-rich rural areas. Journal of Planning
Education and Research, 32(1), 8190. doi:10.1177/0739456X11426781
Haines, A., & Olson, E. (2007). Factors inuencing land parcelization in amenity-rich rural areas and the potential
consequences of planning variables. Stevens Point, WI: National Research Initiative of the Cooperative State
Research, Education and Extension Services, USDA, Center for Land Use Education, University of Wisconsin-
Stevens Point.
Heimlich, R. E., & Anderson, W. D. (2001). Development at the urban fringe and beyond: Impacts on agriculture and rural
land (pp. 803). Washington, DC: U.S. Department of Agriculture.
Hoppe, R. A., MacDonald, J. M., & Korb, P. (2010). Small farms in the United States: Persistence under pressure.
Washington, DC: USDA.
Hosmer, D. W., Jr, & Lemeshow, S. (2004). Applied logistic regression. New York, NY: John Wiley & Sons.
Irwin, E. G., Bell, K. P., & Geoghegan, J. (2003). Modeling and managing urban growth at the rural-urban fringe: A
parcel-level model of residential land use change. Agricultural and Resource Economics Review, 32(1), 83102.
doi:10.1017/S1068280500002525
Irwin, E. G., & Bockstael, N. E. (2002). Interacting agents, spatial externalities and the evolution of residential land use
patterns. Journal of Economic Geography, 2(1), 3154. doi:10.1093/jeg/2.1.31
Irwin, E. G., & Geoghegan, J. (2001). Theory, data, methods: Developing spatially explicit economic models of land use
change. Agriculture, Ecosystems & Environment, 85(13), 724. doi:10.1016/S0167-8809(01)00200-6
Jacobs, H. M., & Paulsen, K. (2009). Property rights: The neglected theme of 20th-century American planning. Journal
of the American Planning Association, 75(2), 134143. doi:10.1080/01944360802619721
Kilgore, M. A., & Snyder, S. A. (2016). Lake States natural resource managers perspectives on forest land parcelization
and its implications for public land management. Land Use Policy, 59, 320328. doi:10.1016/j.
landusepol.2016.07.035
Ko, D. W., & He, H. S. (2011). Characterizing the historical process of private forestland ownership parcelization and
aggregation in the Missouri Ozarks, USA, from 1930 to 2000. Landscape and Urban Planning, 102(4), 262270.
doi:10.1016/j.landurbplan.2011.05.004
Lambin, E. F., Geist, H. J., & Lepers, E. (2003). Dynamics of land-use and land-cover change in tropical regions. Annual
Review of Environment and Resources, 28(1), 205241. doi:10.1146/annurev.energy.28.050302.105459
Lambin, E. F., Geist, H., & Rindfuss, R. R. (2006). Introduction: Local processes with global impacts. In Land-use and land-
cover change (pp. 18). Berlin: Springer-Verlag.
Lansford, N. H., & Jones, L. L. (1995). Marginal price of lake recreation and aesthetics: An hedonic approach. Journal of
Agricultural and Applied Economics, 27(1), 212223. doi:10.1017/S107407080001974X
LaPierre, S., & Germain, R. H. (2005). Forestland parcelization in the New York City watershed. Journal of Forestry, 103
(3), 139145.
Levia, D. (1998). Farmland conversion and residential development in North Central Massachusetts. Land Degradation
& Development, 9(2), 123130. doi:10.1002/(ISSN)1099-145X
Levy, J. M. (2009). Contemporary urban planning. Upper Saddle River, NJ: Pearson/Prentice Hall.
Lillesand, T. M., & Kiefer, R. W. (1994). Remote sensing and image interpretation. New York, NY: John Wiley & Sons.
Locke, C. M., & Rissman, A. R. (2012). Unexpected co-benets: Forest connectivity and property tax incentives.
Landscape and Urban Planning, 104(34), 418425. doi:10.1016/j.landurbplan.2011.11.022
Lund, J. R., & Mokhtarian, P. L. (1994). Telecommuting and residential location: Theory and implications for commute
travel in monocentric metropolis. Transportation Research Record, (1463), 10-14.
18 T. T. KENNEDY AND H. VEREGIN

Mariola, M. (2005). Losing ground: Farmland preservation, economic utilitarianism, and the erosion of the agrarian
ideal. Agriculture and Human Values, 22(2), 209223. doi:10.1007/s10460-004-8281-1
McGranahan, D. (1999). Natural amenities drive rural population change (Agricultural Economic Report No. 781). p. 32.
Washington, DC: U.S. Department of Agriculture, Economic Research Service.
Mehmood, S. R., & Zhang, D. (2001). Forest parcelization in the United States: A study of contributing factors. Journal
of Forestry, 99(4), 3034.
Mokhtarian, P. L., Collantes, G. O., & Gertz, C. (2004). Telecommuting, residential location, and commute-distance
traveled: Evidence from State of California employees. Environment and Planning A, 36(10), 18771897. doi:10.1068/
a36218
OBrien, R. M. (2007). A caution regarding rules of thumb for variance ination factors. Quality & Quantity, 41(5), 673
690. doi:10.1007/s11135-006-9018-6
Ohm, B. W. (1999). Guide to community planning in Wisconsin. Madison: University of Wisconsin-Extension.
Peng, C.-Y. J., Lee, K. L., & Ingersoll, G. M. (2002). An introduction to logistic regression analysis and reporting. The
Journal of Educational Research, 96(1), 314. doi:10.1080/00220670209598786
Radelo, V., Hammer, R. B., & Stewart, S. I. (2005). Rural and suburban sprawl in the U.S. Midwest from 1940 to 2000
and its relation to forest fragmentation. Conservation Biology, 19(3), 793805. doi:10.1111/cbi.2005.19.issue-3
Radelo, V. C., Hagen, A. E., Voss, P. R., Field, D. R., & Mladeno, D. J. (2000). Exploring the spatial relationship between
census and land-cover data. Society & Natural Resources, 13(6), 599609. doi:10.1080/08941920050114646
Schmidt, T. L. (1997). Wisconsin forest statistics 1996 (NC-183). St. Paul, MN: Forest ServiceU.S. Department of
Agriculture.
Schneider, L. C., & Gil Pontius, R. (2001). Modeling land-use change in the Ipswich watershed, Massachusetts, USA.
Agriculture, Ecosystems & Environment, 85(13), 8394. doi:10.1016/S0167-8809(01)00189-X
Sohl, T. L., Loveland, T. R., Sleeter, B. M., Sayler, K. L., & Barnes, C. A. (2010). Addressing foundational elements of
regional land-use change forecasting. Landscape Ecology, 25(2), 233247. doi:10.1007/s10980-009-9391-3
Stone, R. S., & Tyrrell, M. L. (2012). Motivations for family forestland parcelization in the Catskill/Delaware watersheds
of New York. Journal of Forestry, 110(5), 267274. doi:10.5849/jof.11-015
Theobald, D. M. (2001). Land-use dynamics beyond the American urban fringe. Geographical Review, 91(3), 544564.
doi:10.2307/3594740
Turner, B., II, Lambin, E. F., & Reenberg, A. (2007). The emergence of land change science for global environmental
change and sustainability. Proceedings of the National Academy of Sciences, 104(52), 2066620671. doi:10.1073/
pnas.0704119104
U.S. Census Bureau. (2010). Census of population and housing. Retrieved from http://www.census.gov/prod/www/
decennial.html
Veldkamp, A., & Lambin, E. F. (2001). Predicting land-use change. Agriculture, Ecosystems & Environment, 85(13), 16.
doi:10.1016/S0167-8809(01)00199-2
Verburg, P. H., Kok, K., Pontius, R. G., & Veldkamp, A. (2006). Modeling land-use and land-cover change. In Land-use
and land-cover change, 117135. Berlin: Springer-Verlag.
Verburg, P. H., Schot, P. P., Dijst, M. J., & Veldkamp, A. (2004). Land-use change modelling: Current practice and
research priorities. GeoJournal, 61(4), 309324. doi:10.1007/s10708-004-4946-y
Zhu, P. (2013). Telecommuting, household commute and location choice. Urban Studies, 50(12), 24412459.
doi:10.1177/0042098012474520
JOURNAL OF LAND USE SCIENCE 19

Appendix 1. Explanatory variable descriptive statistics

1972 (N = 8466) 1991 (N = 9924) 2006/2007 (N = 11365)


Min. Max. Mean Min. Max. Mean Min. Max. Mean
Parcel size (ha) 0.01 266.5 11.5 0.01 266.5 9.7 0.01 266.5 8.5
Water frontage (m) 0.0 3580 31.0 0.0 3580 26.6 0.0 3580 24.0
Distance to water (m) 0.0 15,351 1867 0.0 15,351 1744 0.0 15,351 1691
Percent forest cover 0.0 100.0 57.4 0.0 100.0 51.0 0.0 100.0 49.1
Percent agriculture cover 0.0 100.0 25.5 0.0 100.0 18.8 0.0 100.0 11.8
Percent unsuitable soils 0.0 100.0 2.2 0.0 100.0 5.3 0.0 100.0 7.9
Distance to road (m) 0.0 5790 163.0 0.0 5790 150.2 0.0 5153 135.6
Nearest urban services (m) 0.0 72,048 25,858 0.0 72,048 25,534 0.0 72,048 24,638
Mean elevation (m) 184 442.4 303.8 184.3 442.4 303.6 184.3 442.5 304.0
Time to nearest urban 20.4 313.5 60.8 29.9 313.5 68.8 29.9 313.5 68.1
area (min.)
Land value ($) 0.0 40,500 832.0 0.0 434,700 6961 0.0 1,215,600 36,143
Town per capita value ($) 401.0 5685 3036 980.4 45,764 18,608 5532 272,269 114,008
Town population 150.0 1357 629.8 215.0 1913 827.3 273.0 3273 1293
Town population increase 0.4 1.2 0.5 0.1 0.9 0.3 0.1 0.7 0.5
Town population density 2.1 47.0 15.0 3.0 66.3 21.3 3.8 113.5 33.6
Town median age 24.4 46.1 30.2 34.0 55.4 43.8 43.1 60.6 52.2

Appendix 2. Dependent variable descriptive statistics

1972 1991 2006/2007


Splits 2937 2081 2359
Non-splits 5529 7843 9006
N 8466 9924 11,365

Você também pode gostar