ASSESSING GROWTH PREFERENCES ON THE RURAL-URBAN FRINGE

USING A DISCRETE CHOICE MODEL AND SPATIAL ANALYSIS METHODS

BY

LORRAINE R. STAMBERGER

THESIS

Submitted in partial fulfillment of the requirements

for the degree of Master of Science in Natural Resources and Environmental Sciences

in the Graduate College of the

University of Illinois at Urbana-Champaign, 2018

Urbana, Illinois

Master’s Committee:

Assistant Professor Carena J. van Riper

Professor William Stewart

Professor Amy Ando

ii

ABSTRACT

Landscapes on the rural-urban fringe are experiencing change and diversification, yet the

preferences of local residents for how these landscapes should develop is largely overlooked. My

thesis contributes to a growing body of research that explores residents’ preferences for

landscape change through two specific aims: (1) understand how preferences for growth are

influenced by multiple landscape-scale attributes, and (2) explore how these preferences are

distributed across spatial scales. My research takes place in two case study sites–Will County,

Illinois and Jasper County, Iowa–characterized as mixed-use landscapes with strong agrarian

roots, close proximity to large metropolitan centers, and conservation practices. I draw from

residential survey data to first determine how the impacts of land use and economic conditions

influence respondent choices for future growth scenarios using a discrete choice model that

includes six attributes: residential growth, protected grasslands, distance to recreation,

agriculture, bison presence, and unemployment. Next, using only the data collected from

respondents in Will County, I conducted posterior spatial tests to examine the spatial dependence

of preferences for these attributes at the individual household level. Global and local spatial

methods were applied together to understand overarching trends in spatial dependence and

regional clustering of preferences. My results showed that preferences varied within the sample

and across spatial scales. My research on the rural-urban fringe extends the geographic scope of

the choice modeling literature, and I look beyond preferences for a particular project or policy by

assessing landscape-scale attributes with implications for planning at a broader regional scale. At

the rural-urban fringe where change is inevitable, I help to pave the way for greater

understanding of stakeholder preferences and a more democratized planning process.

iii

ACKNOWLEDGEMENTS

This project would not be possible without funding provided by USDA National Institute

of Food and Agriculture (NIFA) and support from local partners involved in the project. A

special thanks goes out to the interviewees, focus group participants, and survey respondents in

both Will and Jasper Counties. Thank you for your participation and for sharing thoughtful

insights on the needs and visions for your local communities. I also acknowledge the many folks

that helped refine our survey instrument.

At this university, I have been fortunate to have worked with many wonderful people.

These include past and present graduate students – Eric, El, Kaitlyn, Katie, Vijay, Clint, Ben,

Nate, Dana, Elizabeth, Nikki, Doug, and John – and other faculty and staff – Max, Pushpendra,

and Sarah. Thank you for the laughs, mentoring, and encouragement along the way.

My committee members, Amy Ando, Bill Stewart, and Carena van Riper, have been

instrumental in arriving at this point. Amy has been a valuable sounding board for my analysis

ideas throughout the process. Bill’s boundless passion for the field has allowed many productive

and enjoyable conversations to ensue. Carena, my advisor, has provided me with many

wonderful opportunities – teaching several courses, winning a conference quiz bowl, and

spending a field season in Denali National Park just to name a few. But more importantly,

Carena has been a genuine and motivating mentor. I have learned so much from her, and am

grateful for her investment in my success throughout this program. I would also like to thank my

additional co-authors, Paul Gobster and Len Hunt, for their helpful reviews on previous

manuscript drafts.

Lastly, thank you to my other half, Rob, and our wild pup, Forrest, for your continued

moral support (Rob) and endless entertainment (Forrest).

iv

TABLE OF CONTENTS

CHAPTER 1: INTRODUCTION ................................................................................................... 1

1.1 Dynamics on the Rural-Urban Fringe ................................................................................... 1

1.2 Discrete Choice Experiments: Theory and design ................................................................ 3

1.3 Spatial Applications of Discrete Choice Experiments .......................................................... 5

1.4 Research Questions ............................................................................................................... 7

CHAPTER 2: ASSESSING PREFERENCES FOR GROWTH ON THE RURAL-URBAN

FRINGE USING A STATED CHOICE ANALYSIS .................................................................... 9

2.1 Abstract ................................................................................................................................. 9

2.2 Introduction ......................................................................................................................... 10

2.3 Background ......................................................................................................................... 12

2.4 Methods............................................................................................................................... 16

2.5 Results ................................................................................................................................. 25

2.6 Discussion ........................................................................................................................... 29

2.7 Conclusion .......................................................................................................................... 33

CHAPTER 3: ASSESSING SPATIAL PREFERENCE HETEROGENEITY IN A MIXED-USE

LANDSCAPE ............................................................................................................................... 35

3.1 Abstract ............................................................................................................................... 35

3.2 Introduction ......................................................................................................................... 36

3.3 Methods............................................................................................................................... 40

3.4 Results ................................................................................................................................. 48

3.5 Discussion ........................................................................................................................... 54

3.6 Conclusion .......................................................................................................................... 57

CHAPTER 4: CONCLUSION ..................................................................................................... 58

REFERENCES ............................................................................................................................. 61

APPENDIX A: WILL COUNTY SURVEY QUESTIONNAIRE ............................................... 77

1

CHAPTER 1: INTRODUCTION

1.1 Dynamics on the Rural-Urban Fringe

The rural-urban fringe, also referred to as ‘exurbia’ or the ‘urban hinterland,’ is the place

where rural and urban landscapes meet. An early definition of this region, provided by Pryor

(1968), asserted that “the rural-urban fringe is the zone of transition...lying between the

continuously built-up urban and suburban areas of the central city and the rural hinterland” (p.

206). Researchers have since attempted to delineate this region in a variety of ways using

location, population density, zoning regulations, socio-demographics, and other variables to set

thresholds for defining the region (Butler & Beale, 1990; Lichter & Brown, 2011; Tali &

Nusrath, 2014). However, this area is difficult to define because it is a dynamic and fluid

landscape that does not exist in isolation and is part of a larger network (Albrecht, 2007; Iaquinta

& Drescher, 2000; Tali & Nusrath, 2014). Research in the context of these fringe locations has

overwhelmingly centered around growth and its impacts on rural areas (Sonya Salamon, 2003;

Smith & Sharp, 2005; Soini, Vaarala, & Pouta, 2012; Taylor, 2011; von Wirth, Gret-Ragamey,

Moser, & Stauffacher, 2016), but less is known of how residents perceive this growth and prefer

these regions to change and develop in the future (Slemp et al., 2012). Because the rural-urban

fringe is continuously in a state of flux and adjustment (Masuda & Garvin, 2008), researchers

should prioritize studying the dynamics of people and environments in this setting.

Growth on the rural-urban fringe is occurring at a rapid pace, bringing with it many

changes to the landscape and human communities. This trend has been evidenced in both the

United States (Slemp et al., 2012; Walker & Ryan, 2008) and other countries (Haregeweyn,

Fikadu, Tsunekawa, Tsubo, & Meshesha, 2012; Valencia-Sandoval, Flanders, & Kozak, 2010),

and continues to occur as metropolitan centers push outward and residents seek desirable

2

locations to live. The movement of people is partly based on a locale’s natural, cultural, and

social amenities, known as “amenity migration” in the literature (Gosnell & Abrams, 2011). The

rural-urban fringe is sought after by prospective residents as a desirable location due to its

proximity to urban amenities and jobs while upholding a quasi-rural lifestyle (Smith & Krannich,

2000). As such, the fringes of many urban centers have witnessed unprecedented population

spikes. The American West, in particular, is one of the fastest growing regions in the U.S where

the region’s natural amenities, quaint downtowns, and proximity to larger metropolitan centers

have all been cited as reasons people migrate to these locations (Gosnell & Abrams, 2011;

Graber, 1974; Krannich, Petrzelka, & Brehm, 2006). While some evidence suggests a back-to-

the city movement in recent years (Hyra, 2015; Smith, 2005), growth on rural-urban fringe still

persists (Kasarda, Appold, Sweeney, & Sieff, 1997; Smith et al., 2017).

Multiple economic, social, and environmental changes stem from growth on the rural-

urban fringe. For example, these locations have witnessed economic restructuring in which new

industries (i.e., service and technology) have replaced old (i.e., farming and mining; Gosnell &

Abrams, 2011). The impacts of economic restructuring have been mixed as it can diversify and

strengthen the economic base (Strauser et al., 2018) but has potential to degrade local livelihoods

(Salamon, 2007). Gentrification processes have also occurred on the rural-urban fringe with land

prices increasing to the point of displacing long-term residents (Green, Marcouiller, Deller, &

Erkkila, 1996; Lichter & Brown, 2011). Concerning the social implications of growth on the

rural-urban fringe, it can mask unique, local identities but also has potential to bolster social

institutions, increase social learning through greater flow of ideas and cultures, and diversify the

demographic makeup of typically aging populations (Krannich et al., 2006; Salamon, 2007).

Lastly, growth can put pressure on important environmental resources (Haregeweyn et al., 2012).

3

For example, Strauser et al. (2018) found that focus group participants viewed growth as a threat

to existing open spaces and shared that grasslands, in particular, would be increasingly important

for providing ecosystem services such as storm water management in the future. Given the

implications of growth on the rural-urban fringe, there is a strong need for understanding how

residents respond and react to growth in these rapidly changing regions.

1.2 Discrete Choice Experiments: Theory and design

Discrete choice experiments have been used to analyze individual preferences in a

diverse range of contexts and fields of study. McFadden's (1974) early application of the method

examined people’s choices for different modes of transportation. Later Louviere & Woodworth,

(1983) began estimating choice models from hypothetical and intended (stated) instead of

reported (revealed) behaviors. Since then, researchers in other fields, such as recreation and

leisure (Reichhart & Arnberger, 2010; Van Riper, Manning, Monz, & Goonan, 2011),

environmental science (L. M. Hunt, 2005; Ivanova & Rolfe, 2011), and planning (Audirac, 1999;

Rambonilaza & Dachary-Bernard, 2007; Sayadi, Gonzalez-Rosa, & Calatrava-Requena, 2009),

have used discrete choice experiments for understanding preferences for specific policies,

conditions, and environments. For example, recent applications have assessed ‘willingness-to-

pay’ for grassland restoration (Dissanayake & Ando, 2014), understood residential location

choices (Liao, Farber, & Ewing, 2015), and analyzed recreational trail preferences (Reichhart &

Arnberger, 2010), thus demonstrating the wide-ranging utility of discrete choice experiments.

Further, this method has been useful in its ability to predict support for goods or services that do

not yet exist which has important implications for policy-makers and managers, such as

4

assessing the feasibility of future policies and projects (Audirac, 1999; Louviere, Hensher, &

Swait, 2000).

This well-established method captures preferences by instructing survey respondents to

choose between competing options or alternatives (Louviere et al., 2000). These alternatives are

defined by characteristics (or attributes) held at various levels, both specified by the researcher.

The choice behavior of respondents is understood through the theory of utility maximization

which posits that individuals make choices to maximize their utility (i.e., well-being; Ben-Akiva

& Lerman, 1985). Further, individuals are known to gain utility from the characteristics of the

alternatives rather than the alternatives themselves (Lancaster, 1966). These sets of alternatives

have been presented in survey instruments using a variety of formats. Researchers often use

tables that list each attribute level (Birol, Smale, & Gyovai, 2006; De Valck et al., 2014;

Espinosa-Goded, Barreiro-Hurlé, & Ruto, 2010), but have also represented alternatives using

visual images (Lizin, Brouwer, Liekens, & Broeckx, 2016) and written narratives (Hoehn, Lupi,

& Kaplowitz, 2010). Researchers have used a suite of logit models to analyze choice data,

operating under random utility theory which acknowledges researchers’ uncertainty in

understanding, measuring, and estimating utility for each individual (Manski, 1977; Thurstone,

1927). The ‘workhorse’ of choice models, the multinomial logit (MNL) model, has been

implemented widely by researchers (Louviere et al., 2000), but in recent decades, there has been

a shift to using more flexible models to analyze individual preferences (Hensher & Greene,

2003; Johnston et al., 2017; Train, 1998; Tu, Abildtrup, & Garcia, 2016).

The random parameter logit (RPL) model has been an important innovation for the

advancement of choice modeling research. The RPL model is superior to traditional choice

models in several ways. First, it allows for more complex error structures. Relaxing the

5

independent and identically distributed (IID) assumption, the error component of RPL models

can be correlated with alternatives and multiple iterations (Greiner, Bliemer, & Ballweg, 2014;

Hensher & Greene, 2003). Second, the RPL model accounts for heterogeneity in preferences,

contrasting previous models that assumed fixed utility of coefficients across a sample (Bliemer

& Rose, 2013; Cadavid & Ando, 2013; Greiner et al., 2014). Because individuals are known to

possess differences in their stated choices, the model parameters are specified as random rather

than fixed, thus, exhibiting a mean and standard deviation. Third, specifying the parameters as

random allows for estimates to be drawn from each individual in the sample (Hensher, Rose, &

Greene, 2005; Train, 2009). These ‘individual-specific parameters’ are conditional on

individuals’ known choices and can be used as variables in further analyses (Vollmer, Ryffel,

Djaja, & Gret-Ragamey, 2016). While the RPL model identifies heterogeneity in a sample, it

does not point to the sources of heterogeneity ( Hunt, 2005; Mieno, Shoji, Aikoh, & Arnberger,

2016). Thus, other techniques are needed in combination with basic RPL models to explore both

how and why preferences differ.

1.3 Spatial Applications of Discrete Choice Experiments

Understanding preferences as they differ across space is a prominent and growing area in

the choice literature. Bockstael's (1996) instrumental work spurred choice modelers to focus

greater attention on accounting for spatial heterogeneity in choice experiments as a way of

understanding how the local landscape influences people’s preferences. Subsequently,

researchers have incorporated space in choice experiments at a variety of geographic scales,

ranging from analyzing differences between large regions (Martin-Ortega, Brouwer, Ojea, &

Berbel, 2012) to assessing differences among neighboring households (Campbell, Scarpa, &

6

Hutchinson, 2008). To analyze preference heterogeneity at a broader geographic scale,

techniques such as segmentation by site (Liao et al., 2015; Martin-Ortega et al., 2012; Van Riper

et al., 2011) and interactions of site-specific variables have been used (Brouwer, Martin-Ortega,

& Berbel, 2010; Sayadi et al., 2009). In segmenting a sample, separate logit models are used to

better represent preferences of distinct regions. For example, Liao et al. (2015) used spatial

segmentation by county to analyze preferences for compact development on the Wasatch front in

Utah. Other studies have used interactions terms in one encompassing model to assess relative

preferences between sites. Brouwer et al. (2010) interacted specific regions of a river basin in

Spain with preference for water quality and found that local residents placed higher value on

water quality levels for their sub-basin than did non-residents.

Preferences have also been analyzed at the individual household level to better

understand localized patterns (Czajkowski, Budziński, Campbell, Giergiczny, & Hanley, 2017;

Johnston, Ramachandran, Schultz, Segerson, & Besedin, 2011; Meyerhoff, 2013; Yao et al.,

2014). To this end, researchers have connected individual-specific parameters to respondents’

address locations and applied this data using a variety of techniques (Johnston, Holland, & Yao,

2016). Vollmer et al. (2016), for example, analyzed how preferences for a river rehabilitation

project in Jakarta, Indonesia differed based on household location by plotting willingness-to-pay

(WTP) against a household’s distance from the river. Other researchers have used household-

level preferences as the dependent variable in regression models to understand how spatial

covariates can explain individual preferences (Abildtrup, Garcia, Boye Olsen, & Stenger, 2013;

Czajkowski et al., 2017; Yao et al., 2014). Another collection of studies implemented spatial

autocorrelation methods to examine local clustering of preferences within a study area (Johnston,

et al., 2011; Johnston, Jarvis, Wallmo, & Lew, 2015; Meyerhoff, 2013). Meyerhoff (2013), for

7

example, explored how preferences for wind power varied across the southeast region of

Germany and found that clusters of low WTP for wind alternatives existed within the city of

Leipzig, indicating that urban dwellers were less resistant to wind power than their rural

counterparts. These studies improve our understanding of spatial heterogeneity in preferences,

which is especially relevant for landscapes that vary across space. However, no known study has

applied these methods to the rural-urban fringe where diverse landscapes and preferences are

well-documented. The core of my thesis aims to address this intellectual and empirical gap.

1.4 Research Questions

Preferences for landscapes are diverse and warrant a research approach that allows for

preferences to differ across individuals. There is a particular need to better understand how

preferences vary based on the location where individuals reside. Using discrete choice modeling,

this gap can be analyzed at a broad geographic scale by using interaction terms to distinguish

study site differences, or at a finer scale, by examining preferences at the individual household

level. I analyze and compare growth preferences of residents in two distinct study sites (Chapter

2), spatially examine the patterns of these preferences (Chapter 3), and then draw conclusions

from the results to guide future research and policy (Chapter 4). My thesis addresses the

following two research questions and corresponding objectives:

1. How do changing landscape and economic conditions in Will and Jasper Counties

influence residents’ choices for future growth scenarios?

a. Determine the effects of the six attributes–residential growth, protected

grasslands, recreation, agriculture, bison presence, and unemployment rates–on

preferences for future growth.

8

b. Compare growth preferences between Jasper and Will County residents.

2. How do individual growth preferences vary across spatial scales in Will County?

a. Estimate resident’s preferences for the study attributes.

b. Assess the spatial dependence of individual preferences.

c. Analyze and map the local spatial patterns of preferences for the model attributes.

9

CHAPTER 2: ASSESSING PREFERENCES FOR GROWTH ON THE RURAL-URBAN

FRINGE USING A STATED CHOICE ANALYSIS1

2.1 Abstract

Increasing the capacity of communities on the rural-urban fringe to accommodate

sustainable growth is a key concern among resource management agencies. Decisions about the

future of these landscapes involve difficult tradeoffs that underscore the importance of

incorporating diverse stakeholder values and preferences into planning efforts. We assessed

residents’ preferences for exurban growth alternatives in two Midwestern counties–Jasper

County, IA and Will County, IL–that have strong agrarian roots and lie at the fringe of rapidly

expanding metropolitan areas. A random parameters logit model was employed to better

understand how residents responded to different growth scenarios. Specifically, we identified

how six landscape characteristics influenced respondents’ stated choices for growth scenarios.

Informed by previous research, focus groups, and pilot testing, our final model evaluated

preferences for residential growth, protected grasslands, recreation, agriculture, bison

reintroduction, and unemployment. Results from a county-wide survey mailed to 3,000 residents

indicated that five of the six landscape-scale attributes significantly influenced residents’

choices. Residential growth, increases in protected grasslands and agriculture, and greater access

to recreation positively predicted choices for hypothetical growth scenarios while residents

preferred future scenarios with low levels of unemployment. Further, the strength of preferences

for these land use and economic conditions differed between Jasper and Will County residents.

The study findings aid decision makers who face growth and urbanization pressures and provide

insight on how to integrate preferences of current residents into planning decisions at a regional

scale.

1 Formatted for potential publication in Landscape and Urban Planning.

10

Keywords: choice modeling; growth preferences; regional planning; stakeholder engagement;

exurbia, random parameters logit

2.2 Introduction

Growth in historically rural areas can take different forms but often happens in a rapid

and unplanned fashion. Areas along the fringes of large metropolitan centers, known as exurban

or peri-urban areas, are especially ripe for unplanned development due to their close proximity to

both urban amenities (i.e., city parks and public transit) and rural landscapes (Slemp et al., 2012).

While exurban growth can counter the decline of small towns that previously relied on farming

and other extractive industries (Krannich et al., 2006), it has potential to diminish natural

resource amenities (Albrecht, 2007), force long-term residents out through gentrification

(Gosnell & Abrams, 2011), and erase local symbols and identities (Tunnell, 2006). A variety of

public policy efforts at local, regional, and state scales have been implemented to manage

sprawling development patterns in the United States and other countries (Bengston, Fletcher, &

Nelson, 2004; Siedentop, Fina, & Krehl, 2016) and have been most successful when stakeholders

are involved in the planning process (Burby, 2003). Stakeholder participation for growth

management and open space protection increases the likelihood that policies will reflect local

values and conditions, facilitate a sense of ownership for community members, and minimize

social and land use conflicts (Burby, 2003). A stronger understanding of residential values and

preferences is especially critical for the rural-urban interface where landscape change is likely to

occur (Soini et al., 2012).

Previous research has underlined the importance of including stakeholders in the

decision-making process (Burby, 2003; Williams, Stewart, & Kruger, 2013). Stakeholders such

11

as residents and business owners have been engaged using a variety of techniques, including

participatory mapping (Brown & Raymond, 2007; van Riper, Kyle, Sutton, Barnes, & Sherrouse,

2012), semi-structured interviews (Slemp et al., 2012; Valencia-Sandoval et al., 2010), public

forums and workshops (Burby, 2003), and stakeholder surveys (Dillman, Smyth, & Christian,

2014). Within survey research traditions, stated choice modeling has been used to understand

individual preferences for specific choice alternatives (Hensher, Rose, & Greene, 2005; Johnston

et al., 2017; Louviere et al., 2000). In a stated choice experiment, an individual is asked to

choose an alternative from set of alternatives that are described by relevant characteristics or

‘attributes.’ While a large body of work is dedicated to using choice data in tandem with

planning efforts (Audirac, 1999; Cadavid & Ando, 2013; Sayadi et al., 2009), there is a strong

need to understand tradeoffs among stakeholder preferences for future growth in the context of

urbanization, particularly in changing landscapes on the rural-urban fringe. Choice modeling, a

rapidly advancing field, shows promise for quantifying growth preferences across a diverse

sample of stakeholders in rural contexts.

This study assessed residents’ preferences for future growth scenarios in two Midwestern

U.S. counties–Jasper County, IA and Will County, IL–both of which face urbanization pressures

from rapidly expanding adjacent metropolitan areas while working to preserve their strong

agrarian roots. As both counties are historically situated within a prairie ecosystem, conserving

and restoring native grasslands is crucial for their ecological vitality. Grasslands provide a

multitude of ecosystem services (Dissanayake & Ando, 2014) and counter the ecological impacts

of development (Slemp et al., 2012). Similar to previous work (Dissanayake & Ando, 2014;

Greiner, Bliemer, & Ballweg, 2014; Nassauer, Dowdell, & Wang, 2011; Rambonilaza &

Dachary-Bernard, 2007; van Riper, Manning, Monz, & Goonan, 2011), we gauged preferences

12

for conservation in future scenarios but distinguished between two relevant components of

Midwest prairie conservation–protected grasslands and bison reintroduction–as bison are integral

to the prairie ecosystem and are being reintroduced into the American Midwestern landscape.

Diverging from most planning-based stated choice experiments (see Arnberger & Eder, 2011 and

Bockstael, 1996 for exceptions), this study assessed landscape-scale preferences through the use

of attributes that reflected regional variation in land use and economic conditions. We also

transcended municipal boundaries to combat ‘leapfrog development’ and support landscape scale

decision-making (Bengston et al., 2004; Slemp et al., 2012). Our research on the desirability of

land uses and tradeoffs made by residents in two rural Midwestern counties contributes to an

increasingly important conversation on regional planning and growth in the face of change.

2.3 Background

Rural-Urban Landscape Trends

For the better part of the past century, many countries around the world have experienced

rapid and expansive urbanization in lands surrounding large metropolitan centers. In the U.S.,

most of the land-use conversion fueling this urbanizing trend is in the form of suburban

landscapes. Suburbs are markedly different than more established urban areas and their residents

are often characterized as car-dependent, affluent, and homogenous or ‘placeless’ (Jackson,

1985; Salamon, 2007). In these areas, residents benefit from natural amenities in the hinterland

while simultaneously maintaining employment in the city (Gosnell & Abrams, 2011). Given

improvements in transportation technologies and infrastructure, suburbs are able to expand at

rates that may be unsustainable for the region as a whole (Albrecht, 2007). Thus, as suburbs

13

continue to experience growth in population and capital, cities and rural communities alike suffer

from out-migration and loss of unique identities.

Growth in historically rural areas brings with it numerous social, economic, and

environmental implications. While growth in rural areas can enhance human capital, boost local

organizations, and increase household incomes (see Lichter & Brown, 2011), it can strain

existing institutions and transform held cultures and traditions (Krannich et al., 2006). Referred

to as the rural ‘growth machine,’ a new and growing amenity-based economy is viewed as

inherently good by local leaders (Green et al., 1996; Kunstler, 1994). However, economic

consequences of the rural growth machine model include diminishing agricultural livelihoods,

increasing vulnerability to national business cycles, increasing lower wage service-based jobs,

and displacing long-term residents through gentrification (Krannich et al., 2006; Lichter &

Brown, 2011). Rapid land conversion processes also have adverse effects on environmental

conditions such as increased storm-water runoff, habitat fragmentation, and air pollution

(Haregeweyn et al., 2012). Concerted planning at a regional scale has potential to address the

range of challenges along the rural-urban fringe (Davis, Nelson, & Dueker, 1994), particularly if

coupled with environmental social science research, grassroots community forums, and other

mechanisms for eliciting stakeholder input to make decisions about preferences for the future.

Choice Modeling

Individual preferences can be evaluated using a statistical technique referred to as “choice

modeling.” Choice models were first developed to address transportation-related problems

involving individual preferences for private and public modes of travel (McFadden, 1974).

Typically, a stated choice model presents pairs of hypothetical alternatives and asks respondents

14

to choose the most preferable alternative (Louviere et al., 2000). The alternatives alone do not

drive decisions, but rather the characteristics (or attributes) of the alternatives drive choices

(Lancaster, 1966). Attributes are often arranged in a series of levels that encompass a realistic

range of conditions found across a given study area. The researcher then assembles the attribute

levels into paired comparisons of alternatives (i.e., choice sets) using an experimental design.

These paired comparisons come in many different forms such as narratives describing the

alternatives (Hoehn et al., 2010), tables that list each attribute level (Cadavid & Ando, 2013), or

visual images illustrating different conditions (van Riper et al., 2011). To model choice data,

variations of the multinomial logit (MNL) regression model are most commonly used (Hensher

et al., 2005; McFadden, 1986). In particular, the random parameters logit (RPL) model has

gained traction in the stated choice literature due to its ability to account for heterogeneity in

preferences (Greiner et al., 2014; Hensher & Greene, 2003).

Choice Modeling and Planning

Stated preference models are commonly used in economics and marketing-based

applications, but they have also been incorporated into community planning research (Audirac,

1999; Dissanayake & Ando, 2014; Hunt & McMillan, 1994; Johnston, Swallow, & Bauer, 2002).

Researchers have investigated the types of future growth scenarios that residents desire. Audirac

(1999), for example, examined whether residents of Florida would be willing to trade off the

presence of a large yard for access to shared neighborhood amenities, while Bockstael (1996)

analyzed preferences in terms of access to city centers and desirable landscape features (i.e.,

waterfronts). Johnston et al. (2002) focused on specific types of land-uses by including protected

open spaces, residential development, and recreational facilities in an assessment of scenarios for

15

future growth in rural Rhode Island. Results from this study indicated that residents favored

larger areas of preserved open space and smaller areas of developed land with lower housing

densities. Recreational facilities were favored by some but were also seen to impact preserved

natural areas (Johnston et al., 2002). Other researchers have focused on evaluating preferences

for specific planning efforts, such as municipal storm water management (Cadavid & Ando,

2013), wetland valuation (Mahan, Polasky, & Adams, 2000), and prairie restoration

(Dissanayake & Ando, 2014). This body of past research has indicated stated choice experiments

carry relevance for landscape and urban planning and can be useful tools for informing decisions

about growth and development.

A stated choice model was developed for this study to evaluate residents’ preferences for

changing landscape and economic conditions of Midwestern U.S. areas on the rural-urban fringe.

Specifically, we employed a stated choice experiment in a county-wide survey sent to residents

of Jasper County, Iowa and Will County, Illinois in Spring 2018. We were guided by two

objectives: 1) determine the effects of the study attributes–residential growth, protected

grasslands, recreation, agriculture, bison reintroduction, and unemployment–on preferences for

future growth; and 2) compare growth preferences between Jasper and Will County residents.

This study provides insight on stakeholder preferences for planning at the regional level, which

is rare in the stated choice literature. Given that planning decisions are often dominated by

elected leaders and developers (Green et al., 1996), understanding the growth preferences of

diverse stakeholders, including long-term residents and people in minority groups, will represent

less powerful voices and democratize planning at the intersection of rural and urban life.

16

2.4 Methods

Study Context

Our two case study sites, Will County in Illinois and Jasper County in Iowa, are situated

near Midwestern U.S. metropolitan centers (see Figure 2.1). While both counties exhibit

similarities in terms of urbanization pressures, the two diverge in population dynamics and

economic conditions. Will County, located in the far southern part of the Chicago metropolitan

region, is the fourth most populous county in the state of Illinois (Data USA, 2018b). Its 700,000

residents are located primarily in the northern part of the county, which is characterized by

growing suburban and exurban landscapes. From 2000-2010, Will County experienced a 35%

population increase (U.S. Census Bureau, 2010). Several years ago, the county led the country in

the highest population shift, indicating the county experienced the biggest swing in the number

of people moving in and out of the area (Podmolik, 2015). Due to these episodes of rapid growth,

providing ample transportation infrastructure and amenities such as schools and emergency

services has been a challenge for Will County and local units of government. Economically, the

CenterPoint Intermodal Center has remained a regional hub, employing a segment of the county

in the transportation sector (4.3%) (Data USA, 2018b; “Will County Profile,” 2018).

Jasper County is located in central Iowa and home to 36,700 residents (Data USA,

2018a). While growth and landscape change in Jasper County has been less evident than Will

County, western portions of the county are experiencing urbanization pressures as the Des

Moines metropolitan area expands outward. In 2016, the Des Moines Metro was the fastest

growing area in the Midwest with a growth rate of 2% across a 12-month time period, outpacing

Fargo, ND (1.9%), Sioux Falls, SD (1.5%), and Madison, WI (1.3%) (Aschbrenner, 2017).

Recently, the economy in Jasper County has shifted as it recovers from a major industry,

17

Maytag, pulling out of the area (Margolis, 2017). At the time this research was conducted,

employment was primarily driven by the manufacturing sector but also supported by occupations

in farming, fishing, and forestry (Data USA, 2018a).

Figure 2.1. Map of Jasper County, IA and Will County, IL in the context of urban sprawl

A better understanding of regional preferences for future growth is strongly needed in

landscapes spanning rural and urban contexts where land use change is widespread. Particularly

in Will and Jasper Counties, there are uneven growth patterns outside of adjacent metropolitan

18

centers, and agricultural lands are rapidly being converted to new uses. Amidst land use change,

both counties have prioritized the protection of large tracts of land for conservation. Further,

federal properties within Jasper and Will Counties have initiated bison reintroduction in 1996

and 2015, respectively. Given changing socio-cultural, economic, and environmental conditions,

the future direction of Will and Jasper Counties could benefit from greater knowledge of their

residents’ preferences to promote growth that aligns with current interests. This context

motivated the present study to engage with county- and city-level planning to generate insights

on the preferences and tradeoffs residents were willing to make when considering their futures.

Survey and Choice Model Design

We developed an experimental design for a stated choice model. Mixed methods were

employed to engage stakeholders early on in the research process and build from qualitative and

quantitative data (Dissanayake & Ando, 2014; Greiner, Bliemer, & Ballweg, 2014; Johnston et

al., 2017; Ryan, Gerard, & Amaya-Amaya, 2008). The attributes of the choice model were

conceptualized through informal interviews (n = 20) and focus groups (two groups with eight

participants each) with community leaders, including planners, journalists, farmers,

conservationists, tourism professionals, and economic development representatives to identify

recent changes, key issues, and projected shifts in the region (Strauser et al., 2018). All

qualitative data were transcribed verbatim, thematically analyzed, and checked for inter-rater

reliability. This process maintained relevancy for local residents and ensured realistic attributes

were used in the experimental design (Johnston et al., 2002; Greiner et al., 2014). The focus

groups and past work (Bockstael, 1996; Johnston, Swallow, & Bauer, 2002; Nassauer, Dowdell,

& Wang, 2011) aided in formalizing six attributes–residential growth, protected grasslands,

19

bison presence, recreation, agriculture, and unemployment–that characterized how growth might

occur in Will and Jasper Counties (see Table 2.1). Each attribute was assigned between three and

five levels that encompassed a realistic range of conditions. Illustrative icons were then created

to reflect the attributes and levels (Dissanayake & Ando, 2014; Johnston & Ramachandran,

2014) using Adobe Illustrator CC 2017 software.

Table 2.1. Choice model attributes and levels for the survey instrument

Attribute Description Levels

1. Residential growth rate The annual population growth in the county 2% decrease No growth

2% increase 4% increase 6% increase

2. Amount of projected grasslands

The percent change of county land designated as protected grasslands

No change 5% increase 10% increase

3. Amount of bison The percent change in total number of bison in the county

No change 3% increase 5% increase

4. Distance to recreation area

The distance to the nearest recreation area from the resident’s home

20 miles 7 miles 1 mile

5. Amount of agriculture The percentage of land in the county used for agricultural production

30% land 50% land 70 % land

6. Unemployment rate The percentage of people unemployed in the county

2% unemployed 4% unemployed 8% unemployed

In the model, respondents’ choices were regressed on the six attribute variables. The

hypothesized relationships among these attributes were guided by evidence from previous

research with focus group dialogues with community stakeholders (see Table 2.2). For the

pooled data sample, we predicted that increases in unemployment would negatively influence

choice and that the coefficients of all other attributes would be positive. Separate hypotheses

were developed for each study site because the coefficient signs were thought to differ slightly

20

between Jasper and Will County respondents. Grasslands, in particular, were less salient in the

Will County focus groups. In the pooled sample, we predicted growth would positively influence

choices due to its potential to increase the county tax base. The second and third attributes,

grasslands and bison presence, were important for their ecological roles and attracting tourists,

especially for people outside of the two counties. Third, increased access to recreation was

positively regarded, because residents believed that these opportunities would make Jasper and

Will Counties more attractive to prospective residents. Fourth, stakeholders indicated the rural

qualities of each county, particularly related to a strong agricultural presence, were unique and

highly valued. Finally, we expected that low unemployment rates would be most desirable for

the future of these counties.

Table 2.2. Expected signs of variables in the model

Variable Pooled Sample Jasper County Will County

1. Growth + + + 2. Grasslands + + -

3. Bison + + + 4. Recreation + + + 5. Agriculture + + +

6. Unemployment - - -

The initial survey instrument and choice model were refined through two outlets prior to

data collection. First, the survey instrument was pre-tested with a convenience sample of

students, faculty, and staff at the host institution (n=8) following verbal protocol methods (Cahill

& Marion, 2007; Johnston et al., 2002). Next, the survey was pilot-tested at county fairs in Jasper

and Will Counties (n=120) using intercept sampling of adult fair-goers that were residents in the

two counties. Preliminary data provided insights on how best to revise the survey questionnaire

and were used to generate prior estimates necessary for producing an efficient design (Johnston

21

et al., 2017; Louviere, Hensher, & Swait, 2000; Rose & Bliemer, 2013). Obtaining priors from

previous knowledge (i.e., literature and pilot testing) enabled us to create an optimal design that

minimized error (Arlinghaus, Beardmore, Riepe, Meyerhoff, & Pagel, 2014; Rose, Bliemer, &

Hensher, 2008).

After refining our survey, the final experimental design consisted of 18 choice sets, and

each respondent was asked to evaluate nine paired comparisons that were organized into two

survey blocks. The ordering of paired comparisons was reversed for each of the survey blocks to

minimize information order effects (Johnston et al., 2017). Figure 2.2 illustrates an example

question from the survey instrument that asked respondents to choose between two hypothetical

scenarios (A, B) or to opt-out (C) if neither scenario was acceptable. In line with previous

research (Dissanayake & Ando, 2014; Greiner et al., 2014; Peterson, Taylor, & Baudouin, 2015),

an opt-out or “no preference” choice was included in the model. The inclusion of a “no

preference” choice did not pressure respondents into choosing either scenario (Johnston et al.,

2017), and in doing so, maximized the fit of the model (Haaijer, Kamakura, & Wedel, 2001).

NGene 1.1.2 software was used to generate the experimental design.

22

Figure 2.2. Sample choice question used in the survey instrument

Survey administration

Two county-wide, mail-back residential surveys were administered during spring 2018 to

a random sample of 1,500 residential addresses in each county. The survey was implemented

using an adaption of the ‘Tailored Design Method’ established by Dillman, Smyth, & Christian

(2014). There were five points of contact with residents over a three month period, including a 1)

hand-signed introductory letter endorsed by local partners, 2) questionnaire, 3) thank-you

reminder postcard, 4) second questionnaire to non-respondents, and 5) third questionnaire to the

remaining non-respondents. A cover letter and postage-paid return envelope were included in

each mailing. The survey process was administered by the Social and Economic Science

Research Center at Washington State University in cooperation with our institution. Monetary

23

incentives were included in the first survey wave with the inclusion of $2 to increase the

likelihood of response (Edwards et al., 2007). A total of 967 surveys were collected from

residents in Jasper and Will Counties, with a response rate of 37.3% in Jasper County and 30.6%

in Will County.

Analysis Approach

Discrete choice experiments are analyzed using a variety of logit models whereby

individuals are presented with several alternatives and assumed to choose the alternative that

provides the greatest utility (Hensher & Greene, 2003; McFadden, 1978). The utility of a given

alternative includes both deterministic (i.e., observed) and stochastic (i.e., unobserved) parts

(Aas, Haider, & Hunt, 2000; Dissanayake & Ando, 2014). The multinomial (conditional) logit

model (MNL), known as the “workhorse” of choice models, shows the relationship among the

observed attributes of the choice scenarios, unobserved variables, and observed choice outcomes

(Hensher et al., 2005; Louviere et al., 2000). Though widely used, the MNL model has been

heavily critiqued on the basis of its restrictive assumptions (Johnston et al., 2017; Ryan et al.,

2008). Specifically, the model assumes that the attribute effects are consistent across a sample

population and uncertainty is identically and independently distributed (Bliemer & Rose, 2013).

In recent decades, there has been a shift towards implementing more flexible models to

predict stated preferences (Bliemer & Rose, 2013; Boxall & Adamowicz, 2002; Hensher, Rose,

& Greene, 2005; Johnston et al., 2017; Train, 1998). A mixed logit model is a generalized form

of all possible choice models (McFadden & Train, 2000). Within this group, the random

parameters logit (RPL) model allows uncertainty to be accommodated in the estimation of each

parameter as a random variable (Greiner et al., 2014; Hensher et al., 2005; Hunt, 2005). The

24

distributions of these random parameters are commonly assumed to be normal but can account

for uniform, exponential, triangular, or other distributions as specified by the researcher (Bliemer

& Rose, 2013; Hensher et al., 2005). The RPL model offers significant advantages over a

traditional logit model, including the ability to account for (unobserved) preference heterogeneity

and more complex error structures (Greiner et al., 2014; Hensher et al., 2005). Though the RPL

model captures heterogeneity in preferences, it does not explain the variation among individuals

(Hensher et al., 2005; Hunt, 2005). Previous research has used unobserved segmentation (i.e.,

latent class analysis; Reichhart & Arnberger, 2010) and individual-specific variables (i.e.,

interaction of sociodemographic variables; Rambonilaza & Dachary-Bernard, 2007) to

understand potential sources of preference heterogeneity.

In the present study, the choice data were analyzed using a RPL model. Main effects and

main effects with interaction effects were calculated in two separate models using NLogit 6

statistical software. For the first model, all six parameters were specified as random with normal

distributions for the first two ‘unlabeled’ alternatives–Option A and Option B. A constant

represented the third no-preference alternative–Option C. Marginal willingness-to-accept higher

unemployment rates was also calculated to understand the tradeoffs residents were willing to

make between unemployment and land use goods and services. Though willingness-to-pay

(WTP) is a more common approach in choice experiments (see Sergio Colombo, Hanley, &

Louviere, 2009; Dissanayake & Ando, 2014; Train, 1998), the inclusion of county-wide

unemployment was more relevant than a price attribute tied to each future growth scenario. In

the second model, we incorporated a county-of-residence variable because we expected

differences in growth preferences between study sites. We specifically interacted a dummy-

25

coded variable (Jasper County = 0 and Will County = 1) with each of the six parameters to

understand preference heterogeneity on the basis of county-of-residence.

2.5 Results

Descriptive Results

Of the 967 respondents that returned the survey, 888 completed some or all of the choice

questions. A majority of respondents chose to mail back the completed surveys (n=761; 85.60%)

while 128 (14.40%) completed the online version of the survey. Table 2.3 describes the sample

population’s socio-demographic characteristics. A slight majority of respondents self-identified

as female (54.50%). The survey captured a wide age range (min=18; max=104) with the average

age being 57.70 years (SD=15.50; SE=0.53). In terms of education and income, just over half of

respondents earned at least a college degree (51.10%), and respondents primarily reported annual

household income to be within the middle-class income brackets. The majority of respondents

racially identified as being White (89.80%), followed by black or African American (2.80%) and

Asian (2.70%). Household size averaged about three people, and respondents reported living in

their current home 17.7 years (SD=14.70; SE=0.50).

26

Table 2.3. Respondent socio-demographic characteristics

Variable Mean (SD; SE) N (%)

Gender Female 467 (54.5) Male 388 (45.3) Age (years) 57.7 (15.5; 0.53) Education Some high school 26 (3.0) High school graduate 227 (26.0) Some college 174 (19.9) Two-year college degree 99 (11.3) Bachelor’s degree 164 (18.8) Some graduate school 52 (6.0) Graduate or professional degree 131 (15.0) Annual Household Income Less than $24,999 77 (9.3) $25,000 - $99,999 469 (56.8) $100,000 - $249,999 252 (30.6) $250,000 or more 27 (3.3) Race White 798 (89.8) Black or African American 25 (2.8) Asian 24 (2.7) Other 46 (5.2) Household Size Number of adults 1.9 (0.7; 0.02) Number of children 1.2 (1.4; 0.06) Years Lived In current home 17.7 (14.7; 0.50) In the county 29.2 (23.2; 0.80)

Choice Modeling Results

Responses to the multiple choice sets presented in the survey yielded 7,384 choice set

observations. Respondents who marked ‘Option C’ for all nine choice questions were identified

as ‘protest voters’ (n=50; 5.2%) and were removed from the analysis (Greiner et al., 2014;

Jürgen Meyerhoff, Bartczak, & Liebe, 2012). Two different models were estimated using the

RPL model to address our study objectives (see Table 2.4). In Model 1, choice (among Options

A, B, and C) was regressed on the six landscape-scale attributes. The impact of the attributes on

respondents’ choices is reflected in the coefficients, which showed the mean utility estimate

across the sample of respondents. All attribute coefficients, except for the bison attribute, were

27

significant predictors of choice at the 99% confidence level. The probability of choosing an

alternative increased with higher rates of growth (β = 0.024), more grasslands (β = 0.045), closer

recreation areas (β = -0.060), and more land in agriculture (β = 0.031). The probability of

choosing an alternative significantly decreased with higher unemployment rates (β = -0.394).

The standard deviation of the parameter distributions showed that preference for residential

growth, access to recreation, agriculture, and unemployment exhibited significant heterogeneity

(p < 0.01).

Tradeoffs between unemployment and the land use attributes were understood through

residents’ willingness-to-accept higher unemployment rates for increases in a particular good or

service. Marginal willingness-to-accept unemployment showed that residents were willing-to-

accept higher unemployment rates for increased residential growth rates (0.06%), more land in

grasslands (0.11%) and agriculture (0.08%), and closer recreation areas (0.15%). Across the

range of levels we measured, respondents were most willing to accept higher unemployment

rates for increases in agricultural land such that respondents would tradeoff higher

unemployment rates by 3.20% to have land in agricultural production increase from 30% to 70%.

Similarly, respondents would tradeoff increased unemployment rates by 2.85% to have the

closest recreation areas change from 20 miles to one mile away from their place of residence.

28

Table 2.4. Estimated random parameters logit (RPL) models

Variables MODEL 1: Attributes only MODEL 2: Including interactions

Coeff. (SE) SD (SE) Coeff. (SE) SD (SE)

Residential growth 0.024*** (0.009) 0.174*** (0.011) 0.167*** (0.027) 0.165*** (0.011) Protected Grasslands 0.045*** (0.010) 0.009 (0.026) 0.095*** (0.031) 0.023 (0.025) Distance to Recreation -0.060*** (0.003) 0.047*** (0.005) -0.024** (0.010) 0.049*** (0.005) Agriculture 0.031*** (0.002) 0.028*** (0.002) 0.061*** (0.005) 0.029*** (0.002) Bison 0.006 (0.005) 0.008 (0.013) -0.011 (0.015) 0.019 (0.012) Unemployment -0.394*** (0.016) 0.243*** (0.015) -0.467*** (0.043) 0.249*** (0.016) Constant -1.929*** (0.139) 2.858*** (0.136) 1.737*** (0.401) 2.453*** (0.117) Residential growth * Will Co.+ -0.096*** (0.017) N/A Protected Grasslands * Will Co. -0.032* (0.020) N/A Distance to Recreation * Will Co. -0.023*** (0.006) N/A Agriculture * Will Co. -0.020*** (0.003) N/A Bison * Will Co. 0.012 (0.009) N/A Unemployment * Will Co. 0.050* (0.027) N/A Constant * Will Co. 2.511*** (0.270) N/A

LL = -5,852; AIC = 11,732; N = 7,384; Pseudo R2 = 0.279

LL = -5,810; AIC = 11,663; N = 7,384; Pseudo R2 = 0.284

+ Dummy-coded site-specific variable where 0 = respondent from Jasper County and 1 = respondent from Will County Significance at 1% = ***, at 5% = **, and at 10% = * ++ LL=Log likelihood; AIC=Akaike information criterion

29

Table 2.5. Marginal willingness-to-accept higher unemployment rates

Variable Marginal Willingness-to-Accept Unemployment

Residential growth 0.0609 Protected grasslands 0.1140 Distance to recreation -0.1523 Agriculture 0.0787 Bison --

In alignment with our second objective, Model 2 illustrated the differences in preferences

between Jasper and Will County residents by interacting a site-specific variable with the six

attributes (see Table 2.4). When interacting the dummy-coded Will County variable with each

attribute, differences based on county-of-residence emerged. As illustrated by the interaction

effects in Model 2, choices made by Will County residents were less influenced by growth (β = -

0.096), grasslands (β = -0.032), agriculture (β = -0.020), and unemployment (β = 0.050).

Conversely, recreation (β = -0.023) was a strong driver of choice indicating Will County

residents preferred closer recreation areas more than Jasper County residents. Similar patterns

emerged between Models 1 and 2 when considering the main effects of the attributes, in that five

of the six attributes were significant predictors of choice with the expected signs. Moreover, the

heterogeneity displayed in the protected grasslands parameter distribution was no longer present.

That is, when an individual’s county-of-residence was accounted for, the SD for grasslands

became non-significant.

2.6 Discussion

This study advanced knowledge of how residents in two rural Midwestern counties

envisioned the future and made tradeoffs between competing landscape conditions. Results from

a stated choice experiment presented growth preferences in relation to different landscape and

economic conditions. Residents in Jasper County, IA and Will County, IL responded favorably

30

to hypothetical scenarios that included higher residential growth rates, more grassland areas

under protection, less distance between home and recreation areas, more land in agriculture, and

lower unemployment rates. Moreover, expanding existing bison herds did not significantly

influence respondent choices for future growth scenarios. Although the two study sites exhibited

similar characteristics (e.g., strong agricultural ties, emphasis on conservation, historic reliance

on industry), the attributes of our choice model were evaluated differently by residents in our two

case study sites. Generally, residents from Jasper County responded more positively to growth

than did residents in Will County.

Situating our results in the context of urbanization, this study offers insight on the

preferences reported by residents living in changing landscapes on the rural-urban fringe (Soini

et al., 2012). As both Jasper and Will Counties face pressure from adjacent, expanding

metropolitan centers, decision makers are increasingly challenged to respond to the needs of their

stakeholders. Our results showed that preferences for landscape change did not always align with

changes that often accompany urbanization. For example, urbanization is linked to less

dependence on agriculture (Krannich et al., 2006; Lichter & Brown, 2011), but respondents

preferred scenarios with more land in agriculture, regarding 70% of county land in agricultural

production as more preferable than 30% or 50%. Urbanization also has consequences for natural

environments such as grasslands, and similar to the work of Slemp et al. (2012), respondents

preferred more protection of these natural landscapes when envisioning the future of their

counties. In other words, respondents preferred both more agricultural land and natural

grasslands while at the same time preferred higher rates of residential growth.

Five out of six hypotheses for the pooled sample of respondents were supported. As

expected, respondents were more likely to choose scenarios with increased agriculture,

31

grasslands, and access to recreation (see Table 2.2). Previous studies have similarly

demonstrated the desire for open space protection in communities experiencing growth and

development (Lokocz et al., 2011; Slemp et al., 2012). The effect of residential growth, also a

positive predictor of preferred growth scenarios, was in accordance with qualitative findings in

which leaders indicated a need to attract prospective residents to their county (Strauser et al.,

2018). Additionally, greater unemployment rates were not preferred in future scenarios. The one

hypothesis not supported by our findings was that residents would have positive value for growth

of bison herds. However, the bison attribute in our model was not statistically significant. It

could be that the perceived benefits derived from the presence of bison may have been captured

in the protected grasslands attribute. Alternately, county residents may have been unaware that

bison existed in their county and, as a result, disregarded this attribute when making choices. An

implication of this finding is for planners and managers at the two case study sites to raise

visibility of existing bison herds given potentially limited public awareness.

Our comparison between study sites indicated that Jasper County residents had stronger

preferences for growth than those from Will County. Because Will County has experienced rapid

growth in recent years, residents might be more hesitant toward development and land-use

changes than those living in Jasper. However, Will County residents responded strongly toward

changes in access to recreation, preferring greater access to recreation areas in the future. They

were more driven by this attribute than were Jasper County respondents. The amenity migration

literature suggests that recreation and green space are natural amenities that attract people to

places on the fringe of urban centers (i.e., Will County) where residents can benefit from both

urban and rural amenities (Gosnell & Abrams, 2011; Tu et al., 2016). With many residents living

on the rural-urban fringe, Will County may have a greater demand for and capacity to support

32

recreation opportunities compared to its more rural counterpart, Jasper County. Finally, we

expected increases in protected grasslands to have a negative influence on Will County

respondents’ choices. While this effect was non-significant, grassland protection was

significantly less important for residents in Will County than in Jasper, suggesting that Will

County respondents were indifferent to grassland conservation when they envisioned their

futures.

The methodological approach we adopted to carry out the choice experiment produced

meaningful results. We inductively identified attributes for the experimental design (Dissanayake

& Ando, 2014; Johnston et al., 2017), followed by pilot testing with a representative sample to

strengthen the design (Louviere et al., 2000; Rose & Bliemer, 2013). Future studies

implementing choice experiments should strive to ground elements of the design (i.e., attributes

and levels) in site-specific contexts to maintain relevancy and credibility (Greiner et al., 2014).

In our quantitative assessment of the choice data, we expected differences in preferences across

our sample. Thus, a traditional model with fixed parameters estimates (i.e., multinomial logit)

was likely unsuitable. The random parameters logit (RPL) model accounted for respondent

heterogeneity and allowed parameter estimates to vary across individuals (Bliemer & Rose,

2013; Hunt, 2005). Though the RPL model was useful to assess preference heterogeneity, it did

not provide insights into the reasons why respondents’ preferences vary (Boxall & Adamowicz,

2002). We explained some preference heterogeneity using respondents’ county-of-residence;

however, other variables such as income (Rambonilaza & Dachary-Bernard, 2007), gender

(Hensher et al., 2005), age (Hunt, 2005), and distance to features (Dissanayake & Ando, 2014)

have been cited as sources of heterogeneity in landscape preferences.

33

Opportunities for Future Research

Our research was limited in several ways and thus created opportunities for future

research. First, including an opt-out option was consistent with previous literature (Greiner et al.,

2014; Peterson et al., 2015); however, this research approach did not allow us to understand why

respondents opted-out instead of choosing a growth scenario. Further, the verbiage used for the

opt-out option - No Preference - was ambiguous, making it difficult to interpret in the results. In

the future, more specific wording (see Dissanayake & Ando, 2014) or inclusion of reference

attributes and levels (see Lizin, Brouwer, Liekens, & Broeckx, 2016) is recommended. A second

limitation was related to potential biases in our sample of respondents. For example, individuals

that did not respond to the survey might have had systematic differences in preferences than did

the respondents. Future research should evaluate non-response bias and continue the quest for

maintaining high response rates. Finally, we did not consider attributes that may have been

ignored by respondents when choosing between growth scenarios. This ‘attribute non-

attendance’ should be empirically assessed in future work by asking respondents to state which

attributes they did not consider when making choices (Greiner et al., 2014).

2.7 Conclusion

Choice experiments are a useful tool for understanding stakeholder preferences and

democratizing the planning process. We engaged residents in two Midwestern counties

experiencing land-use changes by implementing a choice experiment that represented local

concern identified during an earlier, qualitative phase of this research. The study attributes

represent local priorities that warrant attention from resource planning and management

agencies. We not only advance the stated choice modeling literature but also provide practical

34

evidence for critically assessing the growth trajectory of rural communities in the coming years,

particularly around recreation, conservation, agriculture, and al opportunities alongside increases

in population. Unemployment rates are a particularly salient issue that should be carefully

considered in future communication about resource management. Our comparison between study

sites also illuminated preferences for growth in two different contexts, which broadened our

ability to generalize the findings of this research to other locales. As resources and lifestyles on

the rural-urban fringe continue to change, results from this research can be applied to enhance

regional scale plans for addressing growth challenges and inform strategies for stakeholder

involvement in decision-making.

35

CHAPTER 3: ASSESSING SPATIAL PREFERENCE HETEROGENEITY IN A MIXED-

USE LANDSCAPE2

3.1 Abstract

Discrete choice experiments are a well-known method for analyzing landscape

preferences. Although people’s preferences are known to vary across space, this body of work

imposes strong assumptions regarding the spatial distribution of preferences, often operating

under spatial homogeneity. Thus, localized approaches are needed to better understand spatial

heterogeneity in preferences. I analyzed landscape preferences in an American Midwestern

county–Will County, Illinois–using residential surveys. Drawing from the results of a discrete

choice model, I obtained and geo-located individual-specific parameter estimates for select land

use and economic attributes of Will County–residential growth, protected grasslands, recreation,

agriculture, bison reintroduction, and unemployment rates. Subsequently, I used both global and

local spatial autocorrelation patterns to analyze the spatial relationships of the landscape

preferences. Results showed that preferences for all model attributes were heterogeneous within

the sample. Local spatial autocorrelation revealed local clustering of high and low preferences,

especially apparent in the agriculture and residential growth attributes. This study gives insight

on how location of residence relates to stakeholder preferences for landscape attributes and

provides management implications for county leaders and resource managers tasked with

allocating resources for diverse and changing landscapes.

Keywords: discrete choice experiment, Moran’s I, growth preferences, hotspot analysis,

preference heterogeneity, spatial autocorrelation

2 Formatted in line with the requirements of the target journal, Land Use Policy.

36

3.2 Introduction

People’s preferences for landscapes are complex and vary based on an array of factors

that range from internal (i.e., biological needs) to external processes (i.e., previous experiences)

(Abildtrup et al., 2013; Arnberger & Eder, 2011; Boxall & Adamowicz, 2002). In particular,

spatial patterns in the landscape have been found to influence preferences for landscape

characteristics and benefits (Plieninger, Dijks, Oteros-Rozas, & Bieling, 2013; Schläpfer &

Hanley, 2003). In line with the idea of transactionalism (Zube, Sell, & Taylor, 1982), people

influence their environments, and in turn, environments influence people over time. Given that

individual preferences are influenced by the local landscape (Bockstael, 1996; Schläpfer &

Hanley, 2003), approaches that capture spatial preference heterogeneity are essential for an

assessment of landscape preferences. Yet, few studies address this gap (Bateman, Jones, Lovett,

Lake, & Day, 2002). Understanding landscape preferences and how they vary is important for

engaging policy makers and program administrators as well as advancing analytical techniques

used to capture public views on planning and management.

One tool that has been previously used to address the research gap noted above is discrete

choice modeling because it is a well-researched method for capturing individual preferences

(Louviere et al., 2000). Choice experiments were originally developed by economists to allow

researchers to understand people’s preferences for competing options (McFadden, 1986). Recent

advances in choice modeling have accommodated differences in individual preferences, known

as ‘preference heterogeneity,’ which identifies variation within a sample (Hensher & Greene,

2003; Sagebiel, Glenk, & Meyerhoff, 2017; Train, 1998). Although choice experiments have

effectively modeled heterogeneity in landscape preferences (Arnberger & Eder, 2011; Sayadi et

al., 2009), less attention has been dedicated to understanding spatial heterogeneity (Bateman et

37

al., 2002; Campbell et al., 2008). This is problematic because operating under preference

homogeneity overlooks how the local landscape influences people’s preferences (Bockstael,

1996; Schläpfer & Hanley, 2003). Discrete choice models that incorporate spatial relationships

can reveal localized patterns that are otherwise invisible (Johnston & Ramachandran, 2014;

Meyerhoff, 2013). These patterns have potential to uncover local clusters of high (low)

preferences within a spatial unit, such as a municipality or county, and to further a broader

understanding of how local landscapes influence preferences.

Spatial heterogeneity has been accounted for in discrete choice experiments through a

range of techniques, but most often, interaction terms have been applied to choice models to

reveal how spatial relationships influence decisions (Broch, Strange, Jacobsen, & Wilson, 2013;

Brouwer et al., 2010; Liao et al., 2015; Schaafsma, Brouwer, & Rose, 2012). For example, Broch

et al. (2013) studied how farmers’ preferences for improving ecosystem services on their land

through afforestation were influenced by spatial variables (e.g., forest cover). These authors

found that two spatial interaction effects were significant, in that population density negatively

influenced farmers’ willingness to provide recreation services while presence of hunting

increased the level of compensation farmers were willing to accept for agreeing to an

afforestation contract (Broch et al., 2013).

The integration of spatial variables in discrete choice models has also been applied to

understanding preferences as a function of distance to an asset such as recreation sites

(Schaafsma et al., 2012), restored grasslands (Dissanayake & Ando, 2014), and wetlands

(Bateman, Day, Georgiou, & Lake, 2006). Previous research has demonstrated the importance of

‘distance decay’ which describes how preferences decrease with increases in distance (Brouwer

et al., 2010; Dissanayake & Ando, 2014; J Meyerhoff, 2013). In other words, respondents tend to

38

place more value on conditions within close proximity to their place of residence. Often, distance

is interacted with respondents’ willingness-to-pay (WTP), which evaluates the amount that an

individual would pay for a public good or service (Hanemann, 1991). Bateman et al. (2006), for

example, conducted two case studies and found significant distance-decay in respondents’

willingness-to-pay for preserving wetlands and improving river conditions. While spatial

interaction terms in choice experiments can provide valuable information on how preferences

vary across space, generating these aggregate effects for the entire study area is not sufficient for

evaluating preferences that exhibit patchiness or clustering patterns (Campbell et al., 2008;

Schaafsma et al., 2012).

Knowledge of preferences at the individual-level, rather than in aggregate, has paved the

way to an expanding literature in spatial econometrics (Abildtrup et al., 2013; Campbell et al.,

2009, 2008; Czajkowski et al., 2017; Johnston et al., 2015; Johnston & Ramachandran, 2014;

Meyerhoff, 2013; Vollmer, Ryffel, Djaja, & Gret-Ragamey, 2016; Yao et al., 2014). Researchers

have utilized individual-specific outputs from logit models in various posterior analyses to

capture preference heterogeneity (Train, 1998). Specifically, individual-specific estimates have

been used in second-stage regression analyses to understand spatial factors contributing to

preferences (Abildtrup et al., 2013; Czajkowski et al., 2017; Yao et al., 2014) and in exploratory

spatial analyses testing the spatial dependence of preferences within a distinct study area

(Campbell et al., 2009, 2008; Czajkowski et al., 2017; Johnston et al., 2015; Johnston &

Ramachandran, 2014; Meyerhoff, 2013). The latter collection of studies has provided insight on

how individual preferences, based on location of residence, vary across space. Campbell et al.

(2008), in particular, assessed how preferences for landscape improvements were distributed in

Ireland. However, few discrete choice experiments have incorporated the spatial dynamics of

39

preferences and their influences despite the benefits that would emerge from recognizing the

inherent spatial arrangement of landscape patterns.

This study analyzed preferences for land use and economic conditions in an American

Midwestern county that has been historically dominated by agriculture but increasingly

accommodates other industries (e.g., manufacturing) and land uses. Given the county’s diverse

landscape, preferences were expected to vary within the county. I developed a discrete choice

experiment that allowed for preference heterogeneity using a random parameters logit model.

Similar to previous work (Campbell et al., 2009; Czajkowski et al., 2017), I explored spatial

heterogeneity in a posterior test that analyzed the spatial dependence of individuals’ preferences.

Diverging from most spatially-explicit choice experiments, I assessed preferences at a local

spatial scale (see Johnston, Jarvis, Wallmo, & Lew, 2015; Johnston & Ramachandran, 2014;

Meyerhoff, 2013 for exceptions). Therefore, I was able to identify spatial regional trends, local

outliers of preferences, and the locations where preferences were particularly high (or low). I

hypothesized that individuals living in close proximity would be more similar than those living

further away (Tobler’s first law of geography; Tobler, 1970), and that preferences would exhibit

local clustering.

Study Objectives

This study used discrete choice modeling and spatial analysis methods to examine

heterogeneity of landscape preferences in Will County, Illinois. Specifically, preferences were

analyzed using a random parameters logit model as well as both global and local spatial

autocorrelation tests. Three objectives directed my research: 1) estimate residents’ preferences

for county-wide landscape characteristics (referred to herein as “attributes”); 2) assess the spatial

40

dependence of individual preferences; and 3) analyze and map the local spatial patterns of

preferences for the model attributes. This is one of the few studies (e.g., Johnston, Jarvis,

Wallmo, & Lew, 2015; Johnston & Ramachandran, 2014; Meyerhoff, 2013) to test local spatial

autocorrelation using data from a discrete choice experiment. Moreover, my research advances

the discrete choice literature by illustrating heterogeneity in preferences for future landscape

scenarios and providing meaningful outcomes for decision-makers (i.e., targeting policy efforts)

working in mixed-use contexts.

3.3 Methods

Study Site

Will County is Illinois’s fourth most populous county and is situated within the Chicago

Metropolitan Statistical Area (see Fig. 3.1; “Will County, IL,” 2018). This site is a mixed-use

landscape; it is a productive agricultural region, transportation hub, economic engine, and place

for many valuable recreation and conservation activities (Strauser et al., 2018). A majority of the

county’s population and opportunities for employment are located in the northern region while

the south is largely characterized by agrarian land use practices (Chicoine, 1981). In 2012, farms

occupied approximately 43% of Will County and were mostly dedicated to growing field corn,

followed by soybeans and foraging crops such as hay (“2012 State and County Profiles,” 2018).

Due to its close proximity to Chicago and central location in the Midwest, Will County is an

important transportation center for the region. The county boasts multiple interstate systems,

well-developed rail lines, a national intermodal transportation facility, and an active river route

and continues to be at the forefront of the transportation industry, as evidenced by prolonged

plans for a new interstate system and major airport. Though these two projects could stimulate

41

employment opportunities, they have experienced opposition, primarily from rural residents who

have expressed skepticism of the benefits for local residents (Dolan, 2018; Steele, 2016).

Amidst the residential areas and industry presence, a patchwork of protected areas exist

in the county. The largest green space is Midewin National Tallgrass Prairie, which encompasses

over 72 square kilometers in the southwest region of the county. Managed by the US Forest

Service since 1996, the protected area embodies the idea of ‘multiple-use and sustained yield’

(Multiple Use Sustained-Yield Act of 1960) as it contains recreational trails, a restored tallgrass

prairie system, existing infrastructure from the land’s previous use as a federal arsenal, and a

bison herd on 1,200 acres of the prairie. Bison were reintroduced in 2015 and have increased

visitation threefold since their arrival (Lafferty, 2016). In addition to Midewin, the Will County

Forest Preserve District manages over 21,000 acres with one-third of the land in active

restoration (“Land Management,” 2017). Will County also has an extensive trail system that

includes the historic Illinois and Michigan (I&M) Canal Trail and the 23-mile long Wauponsee

Glacial Trail. These sites are increasingly important for providing recreational opportunities and

ensuring conservation of natural resources as the county continues to grow and develop.

42

Figure 3.1. Land cover in Will County, Illinois

43

Choice Modeling Experiment

The design of the choice experiment was guided by mixed methods through semi-

structured interviews and groups (see Strauser et al. (2018) for overview). After the scope of the

choice model was developed, pilot testing using a verbal protocol assessment (Cahill & Marion,

2007) and full administration of the survey to refine the survey questions and choice model

design (Campbell et al., 2009; Dissanayake & Ando, 2014; Garrod, Ruto, Willis, & Powe, 2012;

Greiner et al., 2014). The choice design was developed using NGene 1.1.2 software which used

priors from pilot testing to create an optimal design (Greiner et al., 2014; Wang & Swallow,

2016). My final choice design included 18 choice questions, blocked into two different survey

versions. Thus, each survey respondents evaluated nine choice questions. The six attributes were

Residential Growth, Protected Grasslands, Bison Presence, Distance to Recreation, Agriculture,

and Unemployment Rate, each of which had three or five levels that represented the range of

conditions that survey respondents may encounter in the study area (see Table 3.1). Responses

were collected through a mail-back survey that tasked respondents with choosing their preferred

scenario among two experimentally designed options and an ‘opt-out’ option (see Fig. 3.2).

Survey respondents were identified using a random address-based sampling approach and

surveyed using an adaption of the ‘Tailored Design Method’ (Dillman et al., 2014). Survey were

sent to 1,500 addresses in Will County. Individuals also had the option to respond to an online

version of the survey administered via Qualtrics.

44

Table 3.1. Choice model attributes and levels for the survey instrument

Attribute Description Levels

1. Residential growth rate The annual population growth in the county 2% decrease No growth

2% increase 4% increase 6% increase

2. Amount of projected grasslands

The percent change of county land designated as protected grasslands

No change 5% increase 10% increase

3. Amount of bison The percent change in total number of bison in the county

No change 3% increase 5% increase

4. Distance to recreation area

The distance to the nearest recreation area from the resident’s home

20 miles 7 miles 1 mile

5. Amount of agriculture The percentage of land in the county used for agricultural production

30% land 50% land 70 % land

6. Unemployment rate The percentage of people unemployed in the county

2% unemployed 4% unemployed 8% unemployed

45

Figure 3.2. Example choice question from the survey

Discrete choice experiments are guided by random utility theory which assumes that

individuals under identical conditions make different choices to maximize personal benefits

(Thurstone, 1927). Multinomial logit (MNL) models have been used most often to capture stated

choices (Brouwer et al., 2010; Louviere et al., 2000), though an increasing number of studies

have adopted more flexible models such as the random parameters logit (RPL) model (Brouwer

et al., 2010; Sergio Colombo et al., 2009; Hensher & Greene, 2003). The RPL model is preferred

to traditional approaches because it accounts for heterogeneity across individual preferences and

operates under less restrictive assumptions (Bliemer & Rose, 2013; Campbell et al., 2009;

Hensher & Greene, 2003; Train, 1998). Because individuals have different tastes and

preferences, not accounting for heterogeneity in a choice model can lead to biased results (Boxall

& Adamowicz, 2002). In allowing parameters to vary across individuals, each random parameter

46

has a distribution with mean and standard deviation values (Bliemer & Rose, 2013). A random

parameters logit model was employed in the present study to analyze the choice dataset using

NLogit 6 software. All study attributes were specified as random and followed normal

distributions (Johnston & Ramachandran, 2014; Sagebiel, Glenk, & Meyerhoff, 2017).

Another benefit of the random parameters logit model is its ability to estimate parameters

specific to each person in a sample. Individual-specific parameters are based upon the mean

parameter of a subgroup of individuals who chose the same option when faced with the same

choice set (Hensher et al., 2005; Train, 2009; Vollmer et al., 2016). These estimates are

conditional on individuals’ known choices and become increasingly accurate with a larger

number of choice questions (Johnston, Jarvis, Wallmo, & Lew, 2015). Empirically, individual-

specific parameters from RPL models are used to understand the shape of a parameter

distribution or to conduct posterior tests (Scarpa, Willis, & Acutt, 2005; Vollmer et al., 2016).

Using the latter approach, we derived individual-specific parameters for all six attributes to test

for spatial dependence of landscape preferences. Parameters were mapped in ArcMap 10.6

software based on geocoded respondent addresses (Johnston, Holland, & Yao, 2016; Meyerhoff,

2013). Respondents’ preferences were connected to their place of residence.

Global and Local Spatial Autocorrelation

Spatial autocorrelation methods were used to assess variation in landscape preferences

across Will County. This method measures the relationship between a variable and itself across

space. Similar to Campbell, Scarpa, & Hutchinson (2008), Johnston & Ramachandran (2014),

and others, I did not explain spatial variation in preferences but, rather, evaluated preferences in

a univariate exploratory analysis. In this study, I used both global and local methods to identify

47

spatial clustering and dispersion of individual-specific parameters. Global methods apply across

the study area while local methods depict trends around each observation in space (Fotheringham

& Brunsdon, 2010). I used univariate Moran’s I and Gi* analysis to analyze global and local

spatial autocorrelation, respectively. Global Moran’s I is the correlation of a value at location n

and its neighboring values. A Moran’s I value above zero indicates positive spatial

autocorrelation in which similar values are clustered together while a value below zero specifies

negative autocorrelation where nearby locations have dissimilar values (Campbell et al., 2008;

Haining, 2003).

In addition to analyzing global spatial autocorrelation of the individual-specific

parameters, local trends were analyzed. Gi*, a common indicator of local spatial autocorrelation,

was used to identify clusters of significantly low and high preference (Getis & Ord, 1992). In this

analysis, the average value for clusters of observations, that is, observation n and its neighbors,

was compared to the global average. Significant clusters existed if the local average was

significantly different from the overall sample (Johnston & Ramachandran, 2014). Both spatial

autocorrelation tests were completed using GeoDa spatial software (Anselin, Syabri, & Kho,

2006). Local ‘neighbors’ were defined using a spatial weights matrix, known as the ‘spatial lag.’

A queen contiguity first-order spatial weights matrix was used for the spatial analyses in this

study, and Monte Carlo methods were used for significance testing. The inference process

randomly re-assigned the values (i.e., individual-specific parameters) among the points (i.e.,

respondent households). This process was repeated 99,999 times to create a reference

distribution, and the actual Moran’s I and Gi* values were compared with the reference

distributions to understand how much different the spatial dependence of preferences was from

random. These values were generated for each of the six landscape attributes.

48

3.4 Results

Random address-based sampling yielded 440 survey responses (30.6% response rate)

from residents in Will County, Illinois. Of those surveys, 386 were used in our analysis after

removing incomplete data (n=37) and ‘protest votes’ (n=17; Greiner, Bliemer, & Ballweg,

2014). Over 80% of respondents returned a mail-back version of the survey compared to 19%

who chose the online option. Upon geocoding the respondents’ address locations, we found the

majority of respondents were from the northern part of the county. Also, respondent lived in

different environments. Over half of respondents resided in developed regions characterized by

low intensity development, followed by respondents living on medium-intensity developed land

(see Table 3.2).

Table 3.2. Respondent residential locations and land cover type (N=386) Land cover type1 N %

Developed, Open Space 13 3.4 Developed, Low Intensity 197 51.0 Developed, Medium Intensity 154 39.9 Developed, High Intensity 10 2.6 Deciduous Forest 1 0.3 Mixed Forest 1 0.3 Herbaceous 3 0.8 Cultivated Crops 4 1.0 Woody Wetlands 3 0.8 1National Land Cover Database (2011)

Background information was collected from Will County residents. The mean age of

respondents was 56.2 years (SD=14.7), ranging from 18 to 93 years old. The average household

size was two adults (SD=0.8) and one child (SD=1.3). A slight majority of respondents identified

as female (52.5%) and the majority identified as White (78%). Seventy-three percent of

respondents reported having completed at least some college, and the largest group of

respondents reported their yearly household income as $50,000-$99,999. On average,

49

respondents had lived in their current home for 16.7 years (SD=13.9) and in Will County for 19.3

years (SD=18.4). The majority of respondents stated they were currently employed (65.8%) with

education being the greatest employment sector for respondents (17%). In the first page of the

survey, respondents were asked about their knowledge on the attributes used in the choice model

with a five-point Likert Scale. Overall, knowledge of the attributes in Will County was low, but

respondents reported highest levels of knowledge on residential growth (M=2.3; SD=1.3),

followed by recreation and tourism (M=2.2; SD=1.2) and protected grasslands (M=2.0; SD=1.1).

Preferences for Landscape Scenarios

In line with Objective 1, I employed a random parameters model, revealing a range of

landscape preferences in Will County, IL (see Table 3.3). Results from the choice experiment

were generated from 3,421 choice set observations, and the McFadden’s pseudo R2 value of

0.306 indicated a well-fitting model (Hensher & Johnson, 1981). All six model attributes were

significantly different from zero (p<0.10) and had varying effects on the dependent variable of

“choice” (see Table 3.3). Respondent choices were negatively driven by higher Residential

Growth and Unemployment Rates while increases in Protected Grasslands, shorter Distances to

Recreational Areas, more land in Agriculture, and greater Bison Presence increased the

likelihood of a respondent choosing a given scenario. The model attributes also exhibited

significant standard deviations of the random parameter distributions, indicating that preference

heterogeneity existed among all parameters.

50

Table 3.3. The mean and spread of the six random parameters from the random parameters logit

model (N=386) Attributes Coefficients (st. err.) Std. Deviation (st. err.)

Residential Growth -0.031** (0.013) 0.181*** (0.017) Protected Grasslands 0.035** (0.015) 0.059** (0.027) Distance to Recreation -0.072*** (0.005) 0.060*** (0.007) Agriculture 0.023*** (0.003) 0.029*** (0.003) Bison Presence 0.013* (0.007) 0.028* (0.015) Unemployment Rate -0.374*** (0.023) 0.250*** (0.025) Constant -3.762*** (0.245) 2.883*** (0.200)

Log-likelihood = -2,609; Akaike information criterion (AIC) = 5,246; N = 3,421; Pseudo R2 = 0.306

*** = p<0.0001, ** = p<0.05, * = p<0.10 Spatial Dependence of Preferences

To understand the preference heterogeneity that existed among individuals in Will

County, I spatially located individual-specific parameter values based on respondents’ location

of residence, as indicated by Objective 2 of the study. A global spatial autocorrelation

assessment of the geocoded parameters revealed overall spatial dependence for each attribute

(see Table 3.4). Based on the Moran’s I statistics of individual preferences, there was a low

degrees of spatial dependence among the parameters. Further, patterns of spatial dependence

neither trended towards positive or negative spatial autocorrelation. The Distance to Recreation

attribute had a statistically significant trend (p<0.009), exhibiting negative spatial

autocorrelation. That is, preferences for Recreation were not similar among nearby households.

Individuals with high and low preferences for Recreation tended to be located near one another.

Conversely, the Agriculture attribute was positively correlated (p<0.068), indicating that nearby

households had similar preferences for Agriculture.

51

Table 3.4. Global Moran’s I statistics, z-values, and p-values for preferences of the six growth

attributes (N=386)

Attributes Moran’s I z-value p-value

Residential Growth 0.004 0.2164 0.408 Protected Grasslands -0.032 -0.9912 0.160 Distance to Recreation -0.072 -2.283 0.009 Agriculture 0.043 1.515 0.068 Bison Presence -0.004 -0.056 0.485 Unemployment Rate 0.009 0.386 0.342

Significance testing with 99,999 permutations; spatial weights defined using queens contiguity (first order)

Local Patterns of Preferences

Aligning with Objective 3, I analyzed landscape preferences using local spatial

autocorrelation tests to identify clusters of significantly high (or low) preferences. The number of

local cluster types are reported in Table 3.5 for each of the six landscape attributes. Significant

high-high clusters (or hotspots) and low-low clusters (or coldspots) existed for all of the

attributes. High-high clusters indicated an individual with high preferences surrounded by

neighboring high preferences, and low-low clusters were locations where an individual with low

preferences was surrounded by neighboring low preferences. Thus, individual preferences were

similar to neighboring individuals in specific areas of the county. Preference for Agriculture

exhibited the most significant local clusters (n=56), followed by Residential Growth clusters

(n=52).

Figure 3.3 illustrates the spatial clustering of landscape preferences across Will County.

The most prominent spatial clustering of landscape preferences existed in the Agriculture

attribute. Hotspots of preferences for Agriculture coincided with the southern region of the

county where the landscape was dominated by cropland (see Fig. 3.3d). Agricultural coldspots–

clusters of low preference–were found only in the northern half of the county. Spatial clustering

of the Residential Growth attribute also followed clear patterns within the county (see Fig. 3.3a).

52

Distinct bands of hotspots and coldspots emerged with a cluster of contiguous hotspots to the

northwest of Joliet and another hotspot band located in the center of the county. Coldspots for

Residential Growth were located primarily in a band extending northeast from Joliet and another

in the southeast region of the county. Spatial clustering of preferences for the other attributes

existed, but regional trends were less apparent. For example, preferences for Recreation were

mixed in the far northern part of the county, but a cluster of hotspots emerged in the center of

Will County (see Fig. 3.3c). Spatial patterns for the conservation-related attributes, Grasslands

(see Fig. 3.3b) and Bison (see Fig. 3.3e), were also mixed with less evident clustering of hotspots

and coldspots. Finally, clusters of preferences for lower Unemployment rates surrounded the

greater Joliet region while clusters of ‘willingness to accept’ higher Unemployment rates existed

in the central region of the county and in the northwest corner west of Bolingbrook (see Fig.

3.3f). These results suggest at least part of the preference heterogeneity present in respondent

preferences can be understood by local spatial differences.

Table 3.5. Local clusters1 of landscape preferences

Attributes High-High Clusters Low-Low Clusters Non-Significant

N (%) N (%) N (%)

Residential Growth 27 (7.0%) 25 (6.5%) 334 (86.5%) Grasslands 23 (6.0%) 21 (5.4%) 342 (88.6%) Recreation2 13 (3.4%) 14 (3.6%) 359 (93.0%) Agriculture 31 (8.0%) 25 (6.5%) 330 (85.5%) Bison 15 (3.9%) 16 (4.1%) 355 (92.0%) Unemployment 20 (5.2%) 19 (4.9%) 347 (89.9%) 1 Clusters are significant at p<0.05; spatial weights defined using queens contiguity (first order) 2 Recreation was reverse-coded so that high-high clusters represent preference for closer recreation areas

53

Figure 3.3. Gi* hotspot analysis for hypothetical growth preferences (99,999 permutations;

queens contiguity with first order effects); recreation was reverse-coded so that high-high

clusters represent preference for closer recreation areas

54

3.5 Discussion

The purpose of this study was to better understand the spatial patterns of landscape

preferences reported by residents in Will County, Illinois. Given the heterogeneous nature of the

county’s landscape, preferences were expected to vary across space. A discrete choice

experiment revealed preference heterogeneity for all model attributes–Residential Growth,

Protected Grasslands, Bison Presence, Recreation, Agriculture, and Unemployment–while

spatial autocorrelation methods illustrated spatial heterogeneity in individual preferences for

these attributes. Global spatial autocorrelation methods showed little evidence of overall spatial

dependence in individual preferences, but local methods illustrated significant patterns. The

Residential Growth and Agriculture attributes exhibited the most apparent spatial clustering

patterns with high and low preferences for these attributes clustered in distinct areas of the

county. Our results offer insight on preferences in a mixed-use landscape where land use types

and ecosystem services are diverse.

The random parameters logit model revealed that all model attributes were significant

predictors of choice, thus, validating the research approach in its ability to understand the drivers

of individual decisions. Will County residents were more likely to choose scenarios with more

land in Agriculture, more Grasslands, a greater Bison Presence, and closer Recreation

opportunities but less likely to choose scenarios with greater Residential Growth and higher

Unemployment Rates. Using this model, significant preference heterogeneity was captured for all

six attributes. Thus, the impact of these attributes on choice was not the same across individuals.

For instance, the mean coefficient of Residential Growth was positive, but preferences for

Residential Growth were significantly different, suggesting that a subset of the population is

negatively driven by population growth within Will County.

55

Global spatial autocorrelation indicated that overall spatial dependence of the six

attributes was weak. This finding extends other studies that have analyzed global spatial

autocorrelation of individual-specific parameters (Johnston et al., 2015; Johnston &

Ramachandran, 2014; Meyerhoff, 2013). Although the Moran’s I values for Distance to

Recreation and Agriculture preferences were significant at the 90% confidence level, both values

were within 0.1 from zero, indicating relatively weak autocorrelation as values above 0.3 or

below -0.3 are known to represent strong autocorrelation patterns (O’Sullivan & Unwin, 2014).

Weak global clustering might have been linked to the spatial resolution of individual households,

the units of observation. At this scale, global patterns did not demonstrate a strong relationship

between preferences of individuals and their neighbors’ preferences, as defined by the spatial

weights matrix. It could be that patterns existed at a broader spatial scale. Previous research has

indicated that aggregating preferences across spatial units (e.g., census blocks, electoral districts)

has shown stronger influence on global spatial dependence (Campbell et al., 2009; Czajkowski et

al., 2017). However, in line with Johnston et al. (2015), Johnston & Ramachandran (2014), and

Meyerhoff (2013), I chose individual households as the unit of analysis to not mask local

variability in preferences. Future research should carefully consider the unit of analysis for

understanding autocorrelation patterns of preferences.

My analysis approach allowed local, place-based patterns of preferences to surface which

has important implications for management and policy. The results of our study showed

clustering, regional trends, and outliers of preferences for specific landscape attributes. Such

information is helpful for targeting policy efforts and assessing the feasibility of policy proposals

or local projects (Campbell et al., 2008; Johnston et al., 2016; Sagebiel et al., 2017). Specifically,

the study’s results can assist in identifying where to focus opportunities for building future

56

recreational facilities, expanding existing open spaces and trail systems, and conserving natural

landscapes such as wetlands and prairies. Also, the outputs of the study have potential to

illuminate where resistance to these and other opportunities might be greatest. Spatial differences

in residents’ ‘willingness-to-accept’ greater Unemployment rates could illustrate locations that

are most vulnerable to economic downturns while patterns in Agricultural preferences identify

areas in which agrarian lifestyles are essential to local residents. In response to these results,

leaders in Will County and other counties in the Midwest, will be better able to apply their local

knowledge about phenomenon in the landscape that influence preferences in the locations

highlighted by my spatial analysis.

This study presents several opportunities for future research that applies spatial

autocorrelation methods. First, my assessment of both global and local spatial dependence of

preferences identified distributional patterns but did not explain why these patterns occurred.

Other methods such as spatial regression (Abildtrup et al., 2013; Czajkowski et al., 2017) and

latent class analysis (Scarpa et al., 2005) can be used in tandem with spatial autocorrelation for

future research to better understand the factors that influence the spatial distribution of

preferences. Second, the spatial weights matrix plays an important role in shaping the

interpretation of results and should be carefully considered in spatial autocorrelation tests. In line

with previous research (Johnston & Ramachandran, 2014; Raudsepp-Hearne, Peterson, &

Bennett, 2010), I defined ‘neighboring’ individuals using a queens contiguity spatial matrix.

Although this analysis approach yielded useful results, other spatial weights matrices are

available. Third, I analyzed uneven spatial units, which had implications for interpreting the map

outputs (see Fig. 3.3). Some spatial patterns were more easily observed in areas that were less

dense and had fewer observations. This challenge is particularly relevant for contexts such as

57

Will County that span a rural-urban gradient. Future research on the rural-urban fringe should

consider using a spatially stratified sampling approach to generate a more equal spatial

representation of preferences.

3.6 Conclusion

The use of spatial autocorrelation methods in a discrete choice experiment demonstrates

one approach for understanding spatial heterogeneity of preferences. I mapped preferences based

on place of residence within a mixed-use landscape in Illinois and found that individual

preferences varied across Will County. Moreover, these results illustrated clusters of high (and

low) preferences for six attributes measured at the landscape level. I therefore underline the

importance of allowing preferences to vary across space so that local variability in preferences

that are not easily explained by global statistics can be captured. Research in choice modeling

should continue to incorporate spatial methods for more representative and policy-oriented

efforts.

58

CHAPTER 4: CONCLUSION

The purpose of this thesis was to understand how preferences for growth on the rural-

urban fringe are influenced by landscape-scale attributes, and how these preferences vary across

space. Specifically, a random parameters logit model was implemented in the American Midwest

to analyze the effects of attributes on respondents’ choices for hypothetical growth scenarios. My

results showed that residents in Jasper County, IA and Will County, IL were more likely to

choose scenarios with higher residential growth rates, more grassland areas under protection, less

distance between home and recreation areas, more land in agriculture, and lower unemployment

rates. Though the two study sites had similar characteristics, the model revealed significant

differences between the two counties, and indicated that residents in Will County preferred lower

residential growth rates, less land in agriculture, and closer recreation areas when compared to

Jasper County residents. The second analytical component of my thesis explained how

preferences for growth varied across Will County’s diverse landscape. Tests of spatial

dependence showed little evidence of overarching, global patterns between individuals’

preferences and preferences of neighboring households but illustrated clear spatial clustering in

specific locations. For example, clusters of high agricultural preference were primarily located in

the southern region of the county while clusters of low preference for agriculture dominated the

northern half of the county. Thus, in specific locations of county, individual preferences were

strongly linked with the preferences of their neighbors. In summary, my study findings

illuminate that preferences are heterogeneous and that some of these differences can be

understood through spatial patterns.

This research provides opportunities for future research in the field of choice modeling.

First, my research found that implementing mixed-methods was a useful approach for informing

59

an experimental design. Future studies should continue using preliminary qualitative data and

pilot testing to develop relevant and meaningful discrete choice experiments (Cahill & Marion,

2007; Greiner et al., 2014; Ryan et al., 2008). Second, this research validated the use of the

random parameters logit model to account for individual heterogeneity. If differences in

preferences across a sample are anticipated, a RPL model is recommended because of its ability

to account for unobserved heterogeneity (Hensher & Greene, 2003; Ryan et al., 2008). Third,

few studies have analyzed the spatial dependence of individual preferences and have produced

mixed results. While several scholars found strong spatial dependence among individual-specific

parameters (Campbell et al., 2009; Czajkowski et al., 2017), my research aligns with others who

have discovered relatively weak spatial dependence among preferences across a given study area

(Johnston & Ramachandran, 2014; Meyerhoff, 2013). Thus, additional research on the spatial

dependence of preferences is needed to better understand how preferences within close proximity

are related. Finally, this study calls for similar research on the rural-urban fringe because these

areas are dynamic and, thus, preferences are likely to shift and change.

Extending beyond the pursuit of knowledge, this project has important implications for

management and policy. Broadly, my thesis prioritized stakeholder involvement at the regional

level (Nelson & Duncan, 1995). Studying preferences at this broader geographic scale is

especially important for planners and visionaries interested in combatting ‘leapfrog

development’ and supporting decision-making beyond the municipal-level (Bengston et al.,

2004). Further, managers and policy-makers can use these results to better understand the

differences that exist among stakeholders regarding their preferences for future growth and

development. Given that I showed stakeholder preferences were not homogenous, people may

respond differently to changes in policy or to proposed projects. My results can be used to

60

spatially target proposed policies based on spatial clusters of high and low preferences that were

identified. Specifically, the study findings strengthen understanding of where support for specific

projects might be greatest. Both researchers and managers can build upon this study to better

understand how preferences differ across dynamic landscapes such as the rural-urban fringe and

specifically how individuals prefer to grow and develop in the face of change.

61

REFERENCES

2012 State and County Profiles - Illinois. (2018).

Aas, Ø., Haider, W., & Hunt, L. (2000). Angler Responses to Potential Harvest Regulations in a

Norwegian Sport Fishery: A Conjoint-Based Choice Modeling Approach. North American

Journal of Fisheries Management, 20(4), 940–950. https://doi.org/10.1577/1548-

8675(2000)020<0940:ARTPHR>2.0.CO;2

Abildtrup, J., Garcia, S., Boye Olsen, S., & Stenger, A. (2013). Spatial preference heterogeneity

in forest recreation. Ecological Economics, 92, 67–77.

Albrecht, D. (2007). Small town in global society. Southern Rural Sociology, 22(1), 1–14.

Anselin, L., Syabri, I., & Kho, Y. (2006). GeoDa: An Introduction to Spatial Data Analysis.

Geographical Analysis, 38(1), 5–22. https://doi.org/10.1111/j.0016-7363.2005.00671.x

Arlinghaus, R., Beardmore, B., Riepe, C., Meyerhoff, J., & Pagel, T. (2014). Species‐specific

preferences of German recreational anglers for freshwater fishing experiences, with

emphasis on the intrinsic utilities of fish stocking and wild fishes. Journal of Fish Biology,

85(6), 1843–1867.

Arnberger, A., & Eder, R. (2011). Exploring the Heterogeneity of Rural Landscape Preferences:

An Image-Based Latent Class Approach. Landscape Research, 36(1), 19–40.

https://doi.org/10.1080/01426397.2010.536204

Aschbrenner, J. (2017, March 22). Des Moines is the fastest growing city in the Midwest, Census

says. Des Moines Register. Des Moines. Retrieved from

https://www.desmoinesregister.com/story/news/2017/03/23/des-moines-fastest-growing-

city-midwest-census-says/99502922/

Audirac, I. (1999). Stated preference for pedestrian proximity: an assessment of new urbanist

62

sense of community. Journal of Planning Education and Research, 19, 53–66.

Bateman, I. J., Day, B. H., Georgiou, S., & Lake, I. (2006). The aggregation of environmental

benefit values: Welfare measures, distance decay and total WTP. Ecological Economics,

60(2), 450–460.

Bateman, I. J., Jones, A. P., Lovett, A. A., Lake, I. R., & Day, B. . (2002). Applying

geographical information systems (GIS) to environmental and resource economics.

Environmental and Resource Economics, 22(1–2), 219–269.

Ben-Akiva, M., & Lerman, S. (1985). Discrete choice analysis: theory and application to travel

demand (Vol. 9). MIT press.

Bengston, D., Fletcher, J., & Nelson, K. (2004). Public policies for managing urban growth and

protecting open space: policy instruments and lessons learned in the United States.

Landscape and Urban Planning, 69, 271–286.

Birol, E., Smale, M., & Gyovai, A. (2006). Using a choice experiment to estimate farmers’

valuation of agrobiodiversity on Hungarian small farms. Environmental and Resource

Economics, 34(4), 439–469.

Bliemer, M. C. J., & Rose, J. M. (2013). Confidence intervals of willingness-to-pay for random

coefficient logit models. Transportation Research B, 58, 199–214.

Bockstael, N. E. (1996). Modeling economics and ecology: the importance of a spatial

perspective. Agricultural Economics, 78(5), 1168–1180.

Boxall, P., & Adamowicz, W. (2002). Understanding heterogeneous preferences in random

utility models: a latent class approach. Environmental and Resource Economics, 23(4),

421–446.

Broch, S., Strange, N., Jacobsen, J., & Wilson, K. (2013). Farmers’ willingness to provide

63

ecosystem services and effects of their spatial distribution. Ecological Economics, (92), 78–

86.

Brouwer, R., Martin-Ortega, J., & Berbel, J. (2010). Spatial preference heterogeneity: a choice

experiment. Land Economics, 86(3), 552–568.

Brown, G., & Raymond, C. (2007). The relationship between place attachment and landscape

values: Toward mapping place attachment. Applied Geography, 27, 89–111.

Burby, R. J. (2003). Making Plans that Matter: Citizen Involvement and Government Action.

Journal of the American Planning Association, 69(1), 33–49.

https://doi.org/10.1080/01944360308976292

Butler, M., & Beale, C. (1990). Rural-urban continuum codes for metro and nonmetro counties

(No. 9028). US Department of Agriculture, Economic Research Service, Agriculture and

Rural Economy Division.

Cadavid, C. L., & Ando, A. (2013). Valuing preferences over stormwater management outcomes

including improved hydrologic function. Water Resources Research, 49, 1–12.

Cahill, K., & Marion, S. (2007). Enhancing the interpretation of stated choice analysis through

the application of a verbal protocol assessment. Journal of Leisure, 39(2), 201–221.

Campbell, D., Hutchinson, W. G., & Scarpa, R. (2009). Using Choice Experiments to Explore

the Spatial Distribution of Willingness to Pay for Rural Landscape Improvements.

Environment and Planning A, 41(1), 97–111. https://doi.org/10.1068/a4038

Campbell, D., Scarpa, R., & Hutchinson, W. G. (2008). Assessing the spatial dependence of

welfare estimates obtained from discrete choice experiments. Letters in Spatial and

Resource Sciences, 1, 117–126.

Chicoine, D. (1981). Farmland values at the urban fringe: an analysis of sale prices. Land

64

Economics, 57(3), 353–362.

Colombo, S., Hanley, N., economics, J. L.-A., & 2009, undefined. (n.d.). Modeling preference

heterogeneity in stated choice data: an analysis for public goods generated by agriculture.

Wiley Online Library. Retrieved from

https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1574-0862.2009.00377.x

Colombo, S., Hanley, N., & Louviere, J. (2009). Modeling preference heterogeneity in stated

choice data: an analysis for public goods generated by agriculture. Agricultural Economics,

40(3), 307–322. https://doi.org/10.1111/j.1574-0862.2009.00377.x

Czajkowski, M., Budziński, W., Campbell, D., Giergiczny, M., & Hanley, N. (2017). Spatial

Heterogeneity of Willingness to Pay for Forest Management. Environmental and Resource

Economics, 68(3), 705–727. https://doi.org/10.1007/s10640-016-0044-0

Davis, J. S., Nelson, A. C., & Dueker, K. J. (1994). The New ’Burbs: The exurbs and their

implications for planning policy. Journal of American Planning Association, 60(1), 45–59.

De Valck, J., Vlaeminck, P., Broekx, S., Liekens, I., Aertsens, J., Chen, W., & Vranken, L.

(2014). Benefits of clearing forest plantations to restore nature? Evidence from a discrete

choice experiment in Flanders, Belgium. Landscape and Urban Planning, 125, 65–75.

Dillman, D., Smyth, J., & Christian, L. (2014). Internet, phone, mail, and mixed-mode surveys:

the tailored design method. John Wiley & Sons.

Dissanayake, S., & Ando, A. (2014). Valuing grassland restoration: proximity to substitutes and

trade-offs among conservation attributes. Land Economics, 90(2), 237–259.

Dolan, B. (2018, September 16). Lawmaker: Proposed new expressway key to relieving

gridlock, fatal accidents. The Times of Northwest Indiana. Retrieved from

https://www.nwitimes.com/news/local/lake/lake-newsletter/lake-news/lawmaker-proposed-

65

new-expressway-key-to-relieving-gridlock-fatal-accidents/article_5afac546-90db-53e4-

b0d1-3843a3cc4dca.html

Edwards, P. J., Roberts, I. G., Clarke, M. J., DiGuiseppi, C., Wentz, R., Kwan, I., … Pratap, S.

(2007). Methods to increase response rates to postal questionnaires. In P. J. Edwards (Ed.),

Cochrane Database of Systematic Reviews. Chichester, UK: John Wiley & Sons, Ltd.

https://doi.org/10.1002/14651858.MR000008.pub3

Espinosa-Goded, M., Barreiro-Hurlé, J., & Ruto, E. (2010). What Do Farmers Want From Agri-

Environmental Scheme Design? A Choice Experiment Approach. Journal of Agricultural

Economics, 61(2), 259–273. https://doi.org/10.1111/j.1477-9552.2010.00244.x

Fishwick, L., & Vining, J. (1992). Toward a phenomenology of recreation place. Journal of

Environmental Psychology, 12, 57–63.

Fotheringham, A. S., & Brunsdon, C. (2010). Local Forms of Spatial Analysis. Geographical

Analysis, 31(4), 340–358. https://doi.org/10.1111/j.1538-4632.1999.tb00989.x

Garrod, G., Ruto, E., Willis, K., & Powe, N. (2012). Heterogeneity of preferences for the

benefits of Environmental Stewardship: A latent-class approach. Ecological Economics, 76,

104–111.

Getis, A., & Ord, J. K. (1992). The Analysis of Spatial Association by Use of Distance Statistics.

Geographical Analysis, 24(3), 189–206. https://doi.org/10.1111/j.1538-

4632.1992.tb00261.x

Gosnell, H., & Abrams, J. (2011). Amenity migration: diverse conceptualizations of drivers,

socioeconomic dimensions, and emerging challenges. GeoJournal, 76, 303–322.

Graber, E. (1974). Newcomers and oldtimers: growth and change in a mountain town. Rural

Sociology, 39(4), 504–513.

66

Green, G., Marcouiller, D., Deller, S., & Erkkila, D. (1996). Local Dependency, Land Use

Attitudes, and Economic Development: Comparisons between Seasonal and Permanent

Residents1. Rural Sociology, 61(3), 427–445.

Greiner, R., Bliemer, M., & Ballweg, J. (2014). Design considerations of a choice experiment to

estimate likely participation by north Australian pastoralists in contractual biodiversity

conservation. Journal of Choice Modelling.

Haaijer, R., Kamakura, W., & Wedel, M. (2001). The “no-choice” alternative in conjoint choice

experiments. International Journal of Market Research, 43(1), 93–106.

Haining, R. (2003). Spatial data analysis: theory and practice. Cambridge University Press.

Hanemann, W. (1991). Willingness to pay and willingness to accept: how much can they differ?

The American Economic Review, 81(3), 635–647.

Haregeweyn, N., Fikadu, G., Tsunekawa, A., Tsubo, M., & Meshesha, D. T. (2012). The

dynamics of urban expansion and its impacts on land use/land cover change and small-scale

farmers living near the urban fringe: A case study of Bahir Dar, Ethiopia. Landscape and

Urban Planning, 106(2), 149–157.

Hensher, D., & Greene, W. (2003). The Mixed Logit Model: The State of Practice.

Transportation, 30(2), 133–176.

Hensher, D., & Johnson, L. (1981). Applied discrete-choice modelling. New York: Wiley.

Hensher, D., Rose, J., & Greene, W. (2005). Applied choice analysis: a primer. Cambridge

University Press.

Hoehn, J. P., Lupi, F., & Kaplowitz, M. D. (2010). Stated choice experiments with complex

ecosystem changes: the effect of information formats on estimated variances and choice

parameters. Journal of Agricultural and Resource Economics, 568–590.

67

Hoffman, S., & Duncan, G. (1988). Multinomial and conditional logit discrete-choice models in

demography. Demography, 25(3), 415–427.

Hunt, J., & McMillan, J. (1994). Stated preference investigation of influences on attractiveness

of residential locations. Transportation Research, 1466, 79–87.

Hunt, L. M. (2005). Recreational Fishing Site Choice Models: Insights and Future Opportunities.

Human Dimensions of Wildlife, 10(3), 153–172.

https://doi.org/10.1080/10871200591003409

Hyra, D. (2015). The back-to-the-city movement: Neighbourhood redevelopment and processes

of political and cultural displacement. Urban Studies, 52(10), 1753–1773.

https://doi.org/10.1177/0042098014539403

Iaquinta, D., & Drescher, A. (2000). Defining the peri-urban: rural-urban linkages and

institutional connections. Land Reform, 2, 8–27.

Ivanova, G., & Rolfe, J. (2011). Assessing development options in mining communities using

stated preference techniques. Resources Policy, 36, 255–264.

Jackson, K. (1985). Crabgrass frontier: The suburbanization of the United States. Oxford

University Press.

Jacobs, J. (1961). The Death and Life of Great American Cities. Random House.

Jasper County, IA. (2018). Retrieved March 7, 2018, from https://datausa.io/profile/geo/jasper-

county-ia/#intro

Johnston, R., Holland, B., & Yao, L. (2016). Individualized Geocoding in Stated Preference

Questionnaires: Implications for Survey Design and Welfare Estimation. Land Economics,

92(4), 737–759.

Johnston, R. J., Boyle, K. J., Adamowicz, W., Bennett, J., Brouwer, R., Cameron, T. A., …

68

Tourangeau, R. (2017). Contemporary guidance for stated preference studies. Journal of the

Association of Environmental and Resource Economists, 4(2), 319–405.

Johnston, R. J., Ramachandran, M., Schultz, E. T., Segerson, K., & Besedin, E. Y. (2011).

Characterizing Spatial Pattern in Ecosystem Service Values when distance decay doesn’t

apply: choice experiments and local indicators of spatial association. In Agricultural &

Applied Economics Association’s 2011 AAEA & NAREA Joint Annual Meeting, Pittsburgh,

Pennsylvania.

Johnston, R., Jarvis, D., Wallmo, K., & Lew, D. K. (2015). Multiscale spatial pattern in nonuse

willingness to pay: Applications to threatened and endangered marine species’. Land

Economics, 91(4), 739–761.

Johnston, R., & Ramachandran, M. (2014). Modeling Spatial Patchiness and Hot Spots in Stated

Preference Willingness to Pay. Environmental and Resource Economics, 59(3), 363–387.

https://doi.org/10.1007/s10640-013-9731-2

Johnston, R., Swallow, S., & Bauer, D. (2002). Spatial factors and stated preference values for

public goods: considerations for rural land use. Land Economics, 78(4), 481–500.

Kasarda, J. D., Appold, S. J., Sweeney, S. H., & Sieff, E. (1997). Central‐city and suburban

migration patterns: Is a turnaround on the horizon? Housing Policy Debate, 8(2), 307–358.

https://doi.org/10.1080/10511482.1997.9521257

Krannich, R., Petrzelka, P., & Brehm, J. (2006). Social change and well-being in western

amenity-growth communities. In W. A. Kandel & D. L. Brown (Eds.), Population change

and rural (pp. 311–331). Springer.

Kunstler, J. (1994). Geography Of Nowhere: The Rise And Decline of America’s Man-Made

Landscape. Simon and Schuster.

69

Lafferty, S. D. (2016). Bison are a big hit at Midewin. Daily Southtown. Retrieved from

http://www.chicagotribune.com/suburbs/daily-southtown/news/ct-sta-midewin-bison-

visitors-st-0120-20160120-story.html

Lancaster, K. J. (1966). A New Approach to Consumer Theory. Journal of Political Economy,

74(2), 132–157. https://doi.org/10.1086/259131

Land Management. (2017). Retrieved from

https://www.reconnectwithnature.org/conservation/land-management

Liao, F. H., Farber, S., & Ewing, R. (2015). Compact development and preference heterogeneity

in residential location choice behaviour: A latent class analysis. Urban Studies, 52(2), 314–

337. https://doi.org/10.1177/0042098014527138

Lichter, D. T., & Brown, D. L. (2011). Rural America in an Urban Society: Changing Spatial and

Social Boundaries. Annual Review of Sociology, 37(1), 565–592.

https://doi.org/10.1146/annurev-soc-081309-150208

Lizin, S., Brouwer, R., Liekens, I., & Broeckx, S. (2016). Accounting for substitution and spatial

heterogeneity in a labelled choice experiment. Journal of Environmental Management, 181,

289–297.

Lokocz, E., Ryan, R. L., & Sadler, A. J. (2011). Motivations for land protection and stewardship:

Exploring place attachment and rural landscape character in Massachusetts. Landscape and

Urban Planning, 99(2), 65–76.

Louviere, J., Hensher, D., & Swait, J. (2000). Stated choice methods: analysis and applications.

Cambridge University Press.

Louviere, J., & Woodworth, G. (1983). Design and analysis of simulated consumer choice or

allocation experiments: an approach based on aggregate data. Journal of Marketing

70

Research, 20(4), 350–367.

Mahan, B., Polasky, S., & Adams, R. (2000). Valuing urban wetlands: a property price approach.

Land Economics, 100–113.

Manski, C. (1977). The structure of random utility models. Theory and Decision, 8(3), 229–254.

Margolis, J. (2017, July 18). The most “broken” town in America is back on its feet. Public

Radio International. Retrieved from https://www.pri.org/stories/2017-07-18/most-broken-

town-america-back-its-feet

Martin-Ortega, J., Brouwer, R., Ojea, E., & Berbel, J. (2012). Benefit transfer and spatial

heterogeneity of preferences for water quality improvements. Journal of Environmental

Management, 106, 22–29.

Masuda, J., & Garvin, T. (2008). Whose Heartland?: The politics of place in a rural–urban

interface. Rural Studies, 24(1), 112–123.

McFadden, D. (1974). The measurement of urban travel demand. Journal of Public Economics,

3(4), 303–328.

McFadden, D. (1978). Modeling the choice of residential location. Transportation Research

Record.

McFadden, D. (1986). The choice theory approach to market research. Marketing Science, 5(4),

275–297.

Meyerhoff, J. (2013). Do turbines in the vicinity of respondents’ residences influence choices

among programmes for future wind power generation? Journal of Choice Modelling, 7, 58–

71.

Meyerhoff, J., Bartczak, A., & Liebe, U. (2012). Protester or non-protester: a binary state? On

the use (and non-use) of latent class models to analyse protesting in economic valuation*.

71

Australian Journal of Agricultural and Resource Economics, 56(3), 438–454.

https://doi.org/10.1111/j.1467-8489.2012.00582.x

Mieno, T., Shoji, Y., Aikoh, T., & Arnberger, A. (2016). Heterogeneous preferences for social

trail use in the urban forest: A latent class model. Urban Forestry & Urban, 19, 20–28.

Nassauer, J., Dowdell, J., & Wang, Z. (2011). Iowa farmers’ responses to transformative

scenarios for Corn Belt agriculture. Journal of Soil and Water Conservation, 66(1), 18–24.

Nelson, A., & Duncan, J. (1995). Growth management principles and practices. American

Planning Association.

O’Sullivan, D., & Unwin, D. (2014). Geographic information analysis. John Wiley & Sons.

Peterson, H. H., Taylor, M. R., & Baudouin, Q. (2015). Preferences of locavores favoring

community supported agriculture in the United States and France. Ecological Economics,

119, 64–73.

Plieninger, T., Dijks, S., Oteros-Rozas, E., & Bieling, C. (2013). Assessing, mapping, and

quantifying cultural ecosystem services at community level. Land Use Policy, 33, 118–129.

Podmolik, M. E. (2015). Will County leads the country in population shift. Chicago Tribune.

Pryor, R. (1968). Defining the rural-urban fringe. Social Forces, 47(2), 202–215. Retrieved from

https://academic.oup.com/sf/article-abstract/47/2/202/2228449

Rambonilaza, M., & Dachary-Bernard, J. (2007). Land-use planning and public preferences:

What can we learn from choice experiment method? Landscape and Urban Planning, 83,

318–326.

Raudsepp-Hearne, C., Peterson, G., & Bennett, E. (2010). Ecosystem service bundles for

analyzing tradeoffs in diverse landscapes. Proceedings of the National Academy of

Sciences, 107(11), 5242–5247.

72

Reichhart, T., & Arnberger, A. (2010). Exploring the influence of speed, social, managerial and

physical factors on shared trail preferences using a 3D computer animated choice

experiment. Landscape and Urban Planning, 96, 1–11.

Rose, J. M., Bliemer, M. C., Hensher, D. A., & Collins, A. T. (2008). Designing efficient stated

choice experiments in the presence of reference alternatives. Transportation Research Part

B: Methodological, 42(4), 395–406.

Rose, J. M., & Bliemer, M. C. J. (2013). Sample size requirements for stated choice experiments.

Transportation, 40(5), 1021–1041. https://doi.org/10.1007/s11116-013-9451-z

Ryan, M., Gerard, K., & Amaya-Amaya, M. (2008). Discrete Choice Experiments in a Nutshell

(pp. 13–46). https://doi.org/10.1007/978-1-4020-5753-3_1

Sagebiel, J., Glenk, K., & Meyerhoff, J. (2017). Spatially explicit demand for afforestation.

Forest Policy and Economics, 78, 190–199.

Salamon, S. (2003). From Hometown to Nontown: Rural Community Effects of

Suburbanization. Rural Sociology, 68(1), 1–24.

Salamon, S. (2007). Newcomers to old towns: Suburbanization of the heartland. University of

Chicago Press.

Sayadi, S., Gonzalez-Rosa, M., & Calatrava-Requena, J. (2009). Public preferences for

landscape features: The case of agricultural landscape in mountainous Mediterranean areas.

Land Use Policy, 26(2), 334–344.

Scarpa, R., Willis, K. G., & Acutt, M. (2005). Individual-specific welfare measures for public

goods: a latent class approach to residential customers of Yorkshire Water. In Econometrics

informing natural resource management (pp. 316–337).

Schaafsma, M., Brouwer, R., & Rose, J. (2012). Directional heterogeneity in WTP models for

73

environmental valuation. Ecological Economics, 79, 21–31.

Schläpfer, F., & Hanley, N. (2003). Do Local Landscape Patterns Affect the Demand for

Landscape Amenities Protection? Journal of Agricultural Economics, 54(1), 21–34.

https://doi.org/10.1111/j.1477-9552.2003.tb00046.x

Siedentop, S., Fina, S., & Krehl, A. (2016). Greenbelts in Germany’s regional plans—An

effective growth management policy? Landscape and Urban Planning, 145, 71–82.

Slemp, C., Davenport, M. A., Seekamp, E., Brehm, J. M., Schoonover, J. E., & Williard, K. W.

J. (2012). “Growing too fast:” Local stakeholders speak out about growth and its

consequences for community well-being in the urban-rural interface. Landscape and Urban

Planning, 106(2), 139–148. https://doi.org/10.1016/j.landurbplan.2012.02.017

Smith, J., Smart, L., Dorning, M., Dupéy, L. N., Méley, A., & Meentemeyer, R. K. (2017).

Bayesian methods to estimate urban growth potential. Landscape and Urban Planning, 163,

1–16.

Smith, M. B., & Sharp, J. S. (2005). Growth, Development, and Farming in an Ohio Exurban

Region. Environment and Behavior, 37(4), 565–579.

https://doi.org/10.1177/0013916505275312

Smith, M., & Krannich, R. (2000). “Culture Clash’’’Revisited: Newcomer and Longer‐Term

Residents’ Attitudes Toward Land Use, Development, and Environmental Issues in Rural

Communities in the.” Rural Sociology, 65(3), 396–421.

Smith, N. (2005). The new urban frontier: Gentrification and the revanchist city. Routledge.

Soini, K., Vaarala, H., & Pouta, E. (2012). Residents’ sense of place and landscape perceptions

at the rural–urban interface. Landscape and Urban Planning, 104, 124–134.

Steele, A. (2016, October 28). IDOT official says Peotone airport “still alive.” The Times of

74

Northwest Indiana. Retrieved from https://www.nwitimes.com/business/local/idot-official-

says-peotone-airport-still-alive/article_6c4f5c90-da0d-55c7-94e9-b19855f23f5c.html

Tali, J., & Nusrath, A. (2014). A literature survey on rural urban fringe. Journal of International

Academic Research for Multidisciplinary, 2(1), 504–517.

Taylor, L. (2011). No boundaries: Exurbia and the study of contemporary urban dispersion.

GeoJournal, 76(4), 323–339.

Thurstone, L. L. (1927). A law of comparative judgment. Psychological Review, 34(4), 273–186.

Tobler, W. R. (1970). A Computer Movie Simulating Urban Growth in the Detroit Region.

Economic Geography, 46, 234. https://doi.org/10.2307/143141

Train, K. (1998). Recreation demand models with taste differences over people. Land

Economics, 74(2), 230–239.

Train, K. (2009). Discrete Choice Methods with Simulation. Cambridge University Press.

Tu, G., Abildtrup, J., & Garcia, S. (2016). Preferences for urban green spaces and peri-urban

forests: An analysis of stated residential choices. Landscape and Urban Planning, 148,

120–131.

Tunnell, K. (2006). Socially disorganized rural communities. Crime, Media, Culture, 2(3), 332–

337.

Valencia-Sandoval, C., Flanders, D., & Kozak, R. (2010). Participatory landscape planning and

sustainable community development: Methodological observations from a case study in

rural Mexico. Landscape and Urban Planning, 94, 63–70.

van Riper, C. J., Kyle, G. T., Sutton, S. G., Barnes, M., & Sherrouse, B. C. (2012). Mapping

outdoor recreationists’ perceived social values for ecosystem services at Hinchinbrook

Island National Park, Australia. Applied Geography, 35(1), 164–173.

75

Van Riper, C. J., Manning, R. E., Monz, C. A., & Goonan, K. A. (2011). Tradeoffs among

resource, social, and managerial conditions on mountain summits of the Northern Forest.

Leisure Sciences, 33(3), 228–249.

Vollmer, D., Ryffel, A., Djaja, K., & Gret-Ragamey, A. (2016). Examining demand for urban

river rehabilitation in Indonesia: Insights from a spatially explicit discrete choice

experiment. Land Use Policy, 57, 514–525.

von Wirth, T., Gret-Ragamey, A., Moser, C., & Stauffacher, M. (2016). Exploring the influence

of perceived urban change on residents’ place attachment. Journal of Environmental

Psychology, 46, 67–82. https://doi.org/10.1016/j.jenvp.2016.03.001

Walker, A. J., & Ryan, R. L. (2008). Place attachment and landscape preservation in rural New

England: A Maine case study. Landscape and Urban Planning, 86(2), 141–152.

https://doi.org/10.1016/J.LANDURBPLAN.2008.02.001

Wang, H., & Swallow, B. (2016). Optimizing expenditures for agricultural land conservation:

Spatially-explicit estimation of benefits, budgets, costs and targets. Land Use Policy, 59,

272–283.

Will County, IL. (2018). Retrieved March 7, 2018, from https://datausa.io/profile/geo/will-

county-il/

Will County Profile. (2018). Retrieved from http://www.willcountyced.com/will-county-

profile.html

Williams, D., Stewart, W., & Kruger, L. (2013). The emergence of place-based conservation. In

Place-Based Conservation (pp. 1–17). Springer Netherlands.

Yao, R., Scarpa, R., Turner, J., Barnard, T., Rose, J., Palma, J., & Harrison, D. (2014). Valuing

biodiversity enhancement in New Zealand’s planted forests: Socioeconomic and spatial

76

determinants of willingness-to-pay. Ecological Economics, 98, 90–101.

Zube, E., Sell, J., & Taylor, J. (1982). Landscape perception: research, application and theory.

Landscape Planning, 9(1), 1–33.

77

APPENDIX A: WILL COUNTY SURVEY QUESTIONNAIRE

78

79

80

81

82

83

84

85

86

87