Exploring Airport Access and Accessibility

Exploring Airport Access and Accessibility:

Regional and National Perspectives

A Dissertation Submitted to the Faculty of Drexel University by Fangwu Wei in partial fulfillment of the requirements for the degree of Doctor of Philosophy June 2016

Dedicated to my parents and my wife

Acknowledgments

There are a number of people who have helped and encouraged me a lot over the last five years. Dr. Tony H. Grubesic, my advisor and committee co-chair, is the first and foremost person I would like to thank for his patient guidance and consistent help to my

dissertation, other research work and career development. He supported my research ideas, gave invaluable advice, showed me how to perform solid research and taught me how to

write papers for publication. I am grateful to him for lifting me to a whole new level.

I would also like to thank my committee chair, Dr. Chaomei Chen, and my committee members, Dr. Ali Shokoufandeh, Dr. Daoqin Tong and Dr. Ran Wei, for their help to improve the quality of this dissertation and to offer professional advice for my career. I also want to express my gratitude to Marie Fazio for her consistent care and help to my dissertation.

To my parents and parents-in-law, thank you for encouraging and supporting me in my graduate studies. Finally, I want to thank my wife for accompanying and supporting me. That means a lot to me.

iii

Table of Contents 1. INTRODUCTION ...... 1 1.1 Background ...... 1 1.2 Objectives and Research Structure ...... 4 2. LITERATURE REVIEW ...... 8 2.1 Deregulation and Competition in the Airline Industry ...... 8 2.1.1 Deregulation ...... 8 2.1.2 Airport Competition ...... 10 2.1.3 Airport leakage and choice ...... 12 2.2 Airport Structure and Associated Network Evaluation ...... 16 2.2.1 Basic structures ...... 16 2.2.2 Advanced Network Evaluation ...... 19 2.3 Airport Performance and Efficiency ...... 23 2.3.1 Spatial Evaluation ...... 23 2.3.2 Operational Evaluation ...... 25 2.3.3 Interaction with local markets ...... 28 3. STUDY 1: EXPLORING DEHUBBING AT ONE AIRPORT ...... 32 3.1 Introduction ...... 32 3.2 Background ...... 35 3.2.1 Cincinnati/Northern Kentucky International Airport (CVG) ...... 35 3.2.2 CVG as A Hub ...... 37 3.3 Data and Methods ...... 40 3.3.1 Air Trafﬁc and Airport Data ...... 40 3.3.2 Street Network Data ...... 42 3.3.3 Business data and ZIP code areas ...... 42 3.3.4 Exploratory Spatial Data Analysis (ESDA) ...... 44 3.3.5 Least Absolute Shrinkage and Selection Operator (LASSO) ...... 45 3.4 Results ...... 48 3.4.1 Airport Market Structures ...... 48 3.4.2 The Dehubbing of CVG ...... 50 3.5 Discussion and Conclusion ...... 68 4. STUDY 2: EVALUATING GEOGRAPHIC MARKETS AT EAS AIRPORTS ...... 74 4.1 Introduction ...... 74

4.2 Background ...... 75 4.3 Data and Methods ...... 76 4.3.1 Business Data ...... 77 4.3.2 Demographic, Spatial, and Airport Data ...... 78 4.3.3 Catchment Area Generation ...... 79 4.3.4 Regression Models ...... 80 4.4 Results ...... 84 4.4.1 Regression ...... 86 4.5 Discussion and Conclusion ...... 90 5. STUDY 3: EVALUATING NETWORK ACCESSIBILITY AND CREATING A TYPOLOGY OF RURAL AIRPORTS ...... 96 5.1 Introduction ...... 96 5.2 Background ...... 99 5.2.1 Rural Airports ...... 99 5.3 Data and Methods ...... 103 5.3.1 Airline and Airport Data ...... 103 5.3.2 Network Analysis ...... 104 5.3.3 Cluster Analysis ...... 106 5.4 Results ...... 107 5.4.1 Rural Markets ...... 108 5.4.2 Rural Airport Peer Groups ...... 116 5.4.3 Transitory Peers ...... 122 5.5 Discussion and Conclusion ...... 124 6. STUDY 4: EXPLORING THE SPATIOTEMPORAL TRENDS IN AIR FARES AT U.S. AIRPORTS AND ASSOCIATED PAIRS ...... 130 6.1 Introduction ...... 130 6.2 Background ...... 132 6.2.1 Analytics to Capture Air Fare Trends ...... 132 6.3 Data and Methods ...... 133 6.3.1 Airline and Airport Data ...... 133 6.3.2 Index Development ...... 134 6.4 Results ...... 136 6.4.1 Average Airfares ...... 137

6.4.2 Variability of Airfares: A Pairwise Analysis ...... 143 6.4.3 Interaction of Average Fares and Fare Variability ...... 147 6.4.4 Putting it All Together ...... 153 6.5 Discussion and Conclusion ...... 158 7. CONCLUSION ...... 162 7.1 Summary ...... 162 7.2 Limitations ...... 166 7.2.1 Air traffic data ...... 166 7.2.2 Business data ...... 166 7.3 Future Research ...... 167 BIBLIOGRAPHY ...... 170 Appendix A Dependent and Independent Variables in LASSO Regression ...... 187 Appendix B Fuzzy Clustering Problem (FCP) ...... 190 Vita...... 192

List of Tables Table 3.1: Airline Origin and Destination Survey (DB1B) ...... 41 Table 3.2: Local market structure and catchment overlap ...... 50 Table 3.3: Top three carriers at CVG based on the number of passengers ...... 52 Table 3.4: Summary statistics of direct flight options from CVG ...... 61 Table 3.5: Shared CVG destinations reached by Southwest Airlines from competitor airports ...... 63 Table 3.6: Total number of passengers and associated market share of six airports (direct flights only) ...... 64 Table 3.7: Lasso regression model results ...... 67 Table 4.1: Dependent and Independent Variables ...... 82 Table 4.2 Descriptive Statistics ...... 83 Table 4.3: Regression model results ...... 89 Table 5.1: Top ten rural airports with passenger and frequency counts ...... 110 Table 5.2: Top ten rural airports with the shortest shimbel distance ...... 112 Table 5.3: Top ten rural airports and associated measures of centrality ...... 114 Table 5.4: Summary statistics of clusters for rural airports ...... 117 Table 5.5: List of transitory airports ...... 124 Table 6.1: Airports with high fare variability consistently from 2002-2013 ...... 145 Table 6.2: Airports in the Southeast with high itinerary yields by year ...... 150

vii

List of Figures Figure 1.1: Dissertation Structure ...... 5 Figure 3.1: Major airports of the tri-state area (Ohio, Indiana and Kentucky) ...... 33 Figure 3.2: CVG airport layout ...... 39 Figure 3.3: 90 minute catchment areas: IND, DAY, CMH, CVG, SDF, LEX ...... 49 Figure 3.4a: Average airport itinerary fares, CVG, CMH, DAY, IND, SDF and LEX...... 54 Figure 3.4b: Average itinerary fares to DFW...... 54 Figure 3.4c: Average itinerary fares to MSP...... 54 Figure 3.5: U.S. crude oil first purchase price, January 2000 – December 2013 ...... 58 Figure 3.6: Direct, nonstop flights from CVG, 2002 and 2013 ...... 62 Figure 4.1: 70 mile network catchment area ...... 80 Figure 4.2: Essential Air Service subsidies, May 2010 ...... 85 Figure 4.3: EAS airports and ﬂight totals for 2010 ...... 85 Figure 5.1: The spatial distribution of all rural airports with IATA codes in the United States, 2013 ...... 101 Figure 5.2: Rural airports in the United States with commercial traffic, 2013 ...... 109 Figure 5.3: Shimbel index of rural airports in the United States, 2013 ...... 112 Figure 5.4: Degree Measure of Rural Airports in the United States, 2013 ...... 115 Figure 5.5: Closeness measure of rural airports in the United States, 2013 ...... 115 Figure 5.6: Betweenness measure of rural airports in the United States, 2013...... 116 Figure 5.7: The spatial distribution of rural airport cluster groups in the United States, 2013 .... 119 Figure 6.1: Airports in the United States with commercial traffic, 2013 ...... 137 Figure 6.2: Inflation-adjusted average airfares by airport category, 2002-2013 ...... 138 Figure 6.3: Average one-way fares at airports, 2002-2013 ...... 141 Figure 6.4: Inflation-adjusted average one-way fares on flight pairs, 2002-2013 ...... 143 Figure 6.5: Variability of fares for all available flight pairs at origins, 2002-2013 ...... 147 Figure 6.6: Average fare prices vs. average fare variability for airports, 2013 ...... 154 Figure 6.7: Spatial patterns of airports associated with each pocket, 2013 ...... 155 Figure 6.8: Spatial distribution of fare asymmetries for the United States in 2002 and 2013 ..... 157

viii

Abstract Exploring Airport Access and Accessibility: Regional and National Perspectives Fangwu Wei

Airports, as ground facilities providing support for air traffic, strongly reflect the evolution of the airline industry. They play an important role in transporting people, freight and ideas. However, the role of an airport, both individually and as an element of the larger global network, can change over time. These changes are fueled by economic growth and recession, changing urban and demographic landscapes, variations in domestic and international transport policies, and the operational health of commercial carriers. Further, because most airports are inextricably tied to their regional markets, the benefits associated with locational access to an airport and its associated accessibility to the larger global air network can also vary substantially over time. More importantly, Deregulation in airline industry has significant impacts on changing structure of air transport network, functions and service levels at airports, interaction with local communities, airport competition and market leakage. All changes influence airport access and accessibility in the United State.

Although these changes have been widely recognized, several aspects that have impacted travelers, characteristics of airports and associated markets remain unexplored. The four papers included in this dissertation provide quantitative contributions to the literature on exploring and evaluating airport access and accessibility. The first paper explores the spatiotemporal dynamics of dehubbing process at Cincinnati/Northern Kentucky

International Airport (CVG). It highlights the operational, market and geographic factors that contributed to CVG's decline and use basic exploratory data analysis to provide perspective and deepen our understanding of this process. Results suggest that a

ix combination of commercial carrier strategies, operational efficiencies, hub structures, network topologies and regional competition contributed to the deterioration of CVG. The second paper examines the variations in local geographic markets and their impacts on service levels. Specifically, it uses exploratory and confirmatory statistical approaches combined with spatial analysis to examine the geographic, demographic, socio-economic and local business determinants that contribute to commercial flight activity from Essential

Air Service (EAS) airports. Results suggest that geographic proximity to larger hubs, subsidy levels and the local population of the EAS catchment area are critical factors in

ﬂight activity at EAS airports. The third paper explores the network of airports in the United

States, evaluating relative accessibility and using this information to develop a typology of rural airports for the U.S. Finally, the empirical work presented in this paper, particularly concerning peer groups, can serve as a powerful tool for enhancing planning and policy for rural air transport. The final paper explores the uneven spatiotemporal distribution of air fares, by airports and associated flight segments, examining both symmetries and asymmetries in fare patterns over time and across space. Results suggest that several significant pockets of pain still exist within the United States, and that asymmetries in air fares create a lopsided fare structure for many smaller markets, further aggravating the fare imbalances spawned by deregulation. In sum, the four studies comprising this research provide important information to understand and improve the airport access and accessibility, evaluate associated air transport polices and provide recommendations.

1. INTRODUCTION

1.1 Background

The first sustained and powered flight made by Orville Wright in 1903 made air

travel a reality (FAA, 2015a). Subsequently, World War I stimulated the growth and development of the commercial airline industry and during the 1930s, with four major airlines emerging to provide commercial air service in the United States: United, American,

Eastern, and Transcontinental and Western Air (FAA, 2010). Undoubtedly, the technical development of aviation has had a significant and positive impact on the growth of air transport since World War II (IATA, 2015). In fact, air transport has become one of the most important global transportation modes for long-haul travel.

Airports, as ground facilities providing support for air traffic, strongly reflect the

evolution of the airline industry. They play an important role in transporting people, freight

and ideas. However, the role of an airport, both individually and as an element of the larger

global network, can change over time. These changes are fueled by various internal and

external factors. More importantly, these changes have impacts on travelers, characteristics

of airports and associated local markets, and the larger air transport network.

As Bowen (2002) notes, globalization and liberalization of the airline industry have

enhanced the global air transportation system with fast expansion and dense networks, but

the system has been developed by focusing primary efforts on geographic centers with

large populations and good economies. Smaller peripheral communities are often ignored

during this process. For the United States, the Airline Deregulation Act of 1978 created a

mechanism for competition, but the liberalization of the European airline industry has been

2 a much longer process than the deregulation in the United States. It was a multi-phased, with the proposal of three aviation packages in 1987, 1990 and 1993 (Debbage, 1994).

Since deregulation provides more power and operational flexibility to commercial carriers, legacy airlines (e.g., Delta, American, etc.) have largely focused their service efforts on larger metropolitan areas through the formation of operational hubs. These hubs can benefit passengers, by providing them with more frequent flights to more non-stop destinations. Conversely, low-cost carriers (LCCs) approach both markets and operations somewhat differently by providing discounted air fares to a smaller subset of destinations.

In short, both types of carriers and associated operating strategies exert a significant influence on their local air transport markets, communities and the larger global airline network. Consider, for example, the impacts of dehubbing on a community. For example,

Memphis, TN (MEM) and Cincinnati, OH (CVG) have been dehubbed by Delta Airlines, drastically reducing local service levels. At one time, CVG was the second most important hub in the U.S. for Delta. It offered more than 600 flights per day to 148 unique destinations.

By 2013, this was reduced to 77 destinations with only a single, nonstop, international flight (Paris). Similarly, in Europe, the Clermont-Ferrand Auvergne airport was dehubbed by Air France, reducing seat capacity to the airport by 50% over a five year period

(Burghouwt, 2007).

In the meantime, LLCs have attempted to fill some of these service gaps left by legacy carriers. For example, in Cincinnati, Frontier Airlines has arrived and now offers non-stop service to Denver, Atlanta, Cancun and several other destinations in the U.S.

Although this does not offset the losses by Delta’s dehubbing, the emergence of LCCs at

CVG is promising.

The liberalization of the airline industry also leads to reduction and elimination of

air service in small communities (Flynn & Ratick, 1988). To better control and reduce the negative impact of deregulation on these locales, two federal programs have been developed to help communities maintain their air service in the United States. The first is a subsidized program known as Essential Air Service (EAS) (Grubesic et al., 2014). EAS was initially structured to help rural and remote airports connect to the national air transportation system (Grubesic & Matisziw, 2011). As of June 2014, there are 117 active

EAS communities, and in each location, several commercial flights are provided each day to nearby medium or large commercial hubs. The second program, the Small Community

Air Service Development Program (SCASDP) established by the Wendell H. Ford Aviation

Investment and Reform Act for the 21st Century (P.L. 106-181) and reauthorized by the

Vision 100-Century of Aviation Reauthorization Act (P.L. 108-176) (U.S. DOT, 2013a),

allows smaller airports to apply for federal monies that can be used to mitigate identified

service deficiencies (U.S. DOT, 2013b). During fiscal year 2013, SCASDP awarded 25

grants totaling $11.4 million. As detailed by Wittman (2014), the primary purpose of these

awards is to fund new service, although grants have been used to launch local and regional

marketing efforts, terminal improvements and parking expansion (U.S. DOT, 2013b).

Similar to the EAS program, the public service obligation (PSO) was developed in Europe

to subsidize transport routes and ensure that some level of service is maintained for city

pairs that would not be profitable for carriers in the free market (Williams & Pagliari, 2004).

Given the diversity of actors within the commercial air transport industry, from

airlines, to airports, to customers and their local markets, airport access and accessibility

can (and must) develop over time and across geography (Grubesic et al., 2009). Here,

airport access and accessibility are two different concepts. Geurs and Van Wee (2004, 28)

define accessibility as “the extent to which land use and transport systems enable (groups

of) individuals to reach activities or destinations by means of a (combination of) transport

mode(s)”, whereas access refers to the opportunity (and level of effort) associated with

entering a transportation system from individuals or areas demanding a service (Ellis et al.,

1974; Demetsky & Korf, 1980). That said, most airports are inextricably tied to their

regional markets. Further, it is also important to note that the benefits associated with locational access to an airport and its associated accessibility to the larger global air network can also vary substantially over time. Therefore, it is both important and necessary to examine the current state of airline networks, airport access and connectivity, and system accessibility for evaluating transport policies.

1.2 Objectives and Research Structure

Given the spatiotemporal dynamics of the airline industry, this research is to improve airport access and accessibility at national and regional spatial scales by evaluating current air transport policies and providing recommendations. Specifically, taking several important aspects (e.g. dehubbing, rural airports and air fares) into consideration, airport access and accessibility are examined, identified and explained from a smaller spatial scale (regional) to a larger spatial scale (national) (Figure 1.1). This research presents four studies, including exploring dehubbing process at a hub airport, evaluating interaction between local geographic markets and 106 EAS airports, evaluating network accessibility and creating a typology of 177 rural airports and exploring spatiotemporal trends in air fares at U.S. airports and associated pairs.

Figure 1.1: Dissertation Structure

The second chapter of this dissertation reviews current literature on airport structure

and topology, deregulation and competition in the airline industry, as well as airport

performance and efficiency, all of which are important measures/foci on airport access and

accessibility. The literature review consists of three sections. In section one, deregulation

and competition in the airline industry is examined. In particular, this section focuses on

how deregulation and its consequences influence airports and entire airline industry. In

section two, the current literature on airport structure and associated network evaluation is

reviewed. This includes an overview of network analysis methods that can be used to

evaluate airport accessibility. Finally, section three reviews the current literature on airport performance and efficiency. It examines how airport performance can be measured from three different angles; spatial, operational and socio-economic. It also reviews how airport access and accessibility can be included in airport performance evaluations.

The third chapter of this dissertation is a variation of a paper published in Journal of Transport Geography. Focusing on one single airport at a regional level, the purpose of this chapter is to explore the spatiotemporal dynamics of dehubbing process at

Cincinnati/Northern Kentucky International Airport (CVG). The service restructuring of a hub, specifically, partial or complete termination of service by the dominant carrier can be described as “dehubbing” (Bhadra, 2009; Redondi et al., 2012). Dehubbing is not fatal, but the path for attracting a good mix of alternative carriers and providing the market with appropriate levels of service after dehubbing is not easily traversed. It has been an important phenomenon than cannot be ignored in the U.S. airline industry. There are many mid-sized markets throughout the United States that are struggling to maintain commercial service. This chapter provides a detailed case study of the factors that contributed to CVG’s deterioration over the past decade, highlighting the operational, market and geographic factors that helped fuel its decline.

The fourth chapter of this dissertation is a variation of a paper published in Journal of Transport Geography. This chapter expands the study to 106 rural, isolated airports in

EAS program at a larger spatial scale (lower 48 states) and its purpose is to examine the

variations in local geographic markets and their impacts on service levels. Specifically, this

part of research uses exploratory and confirmatory statistical approaches combined with

spatial analysis to examine the demographic, socio-economic, geographic and local

business factors that contribute to commercial flight activity from EAS airports.

The fifth chapter of this dissertation is a variation of a paper published in The

Review of Regional Studies. The purpose of this chapter is to explore the network of 177

rural airports in the United States, evaluating relative accessibility and using this

information to develop a typology of rural airports for the United States. This is important because the airline industry is extremely dynamic and its operational topologies are in constant flux, changing through time and across geography (Grubesic et al., 2009).

Therefore, it is both important and necessary to occasionally re-examine the current state

of airline networks, airport connectivity and system accessibility for evaluating transport

policies.

The sixth and the final substantive chapter of this dissertation is a variation of a

paper submitted for publication consideration to Air Transport Management. The purpose

of this chapter is to examine the uneven spatiotemporal distribution of air fares, by airports including hubs and rural airports, and associated flight segments in the United States. As detailed previously, the existing transport literature captures variations in airfares over space and time using airport averages (Goetz & Sutton, 1997; Goetz, 2002; Goetz &

Vowles, 2009). However, significant variability of air fares exists amongst individual flight segments at any given airport. Thus, it is necessary to examine the distribution of air fares over unique flight segments to develop a deeper understanding of market trends and characteristics. Within this analysis, there are three distinct foci: 1) determining the variability of air fares among all available flight segments and how the variability interacts with the average air fare at a given airport; 2) determining if there are any symmetries (or asymmetries) in patterns of air fares over time and across space using data on arrival and departure flight segments for each O/D airport pair, 3) highlighting the implications of these fare structures for air transportation policy and planning.

2. LITERATURE REVIEW

2.1 Deregulation and Competition in the Airline Industry

2.1.1 Deregulation

There is no doubt that deregulation has significantly influenced the airline industry.

Goetz and Vowles (2009) note that 30 years after deregulation, the results for the U.S.

commercial air transport industry are somewhat mixed. Although deregulation prompted

higher levels of service frequency to the most popular destinations and lowered the

associated fares (on average), deregulation also increased financial instability within the

industry by reducing quality of service, increased passenger fees (e.g., baggage, seat), and

accelerated changes in network configurations via hubbing, dehubbing, mergers and

acquisitions. In sum, this process has “squeezed” service to/from smaller communities

through schedule reductions and higher fares, creating “pockets of pain” in the air transport

landscape (Goetz & Vowles, 2000). Specifically, through their empirical work, the authors

point out that notable pockets of pain existed in the southeastern United States, including

airports such as Columbia, SC (CAE), Memphis, TN (MEM), Montgomery, AL (MGM)

and Mobile, AL (MOB), extending northward to Cincinnati, OH (CVG) and select airports

in New England and the Upper Midwest. Bowen (2002) also states that the liberalization

in airline industry has resulted in negative impact (e.g. reducing or terminating the air

service) that is related to lack of accessibility on poor countries.

Subsidized programs have been established after the deregulation in order to ameliorate the negative impacts on rural and isolated communities. These programs have been studied to examine their impacts on airport access and accessibility, and associated local markets. Operationally, as outlined and detailed by the Office of Aviation Analysis

(U.S. OAA, 2009), community eligibility for the Essential Air Service (EAS) program is structured around three simple measures. First, communities must be 70 highway miles or more from the nearest large or medium hub airport (as defined by the FCC).1 Second, communities are not eligible if subsidies exceed $200 per passenger. Third, communities can receive a waiver on the subsidy threshold if they are more than 210 highway miles from a medium or large hub.2 As outlined previously, the program is structured to help facilitate access to/from rural and remote communities to the national air transport system.

Reynolds-Feighan (1995) examines the EAS program in the United States and the

European Union (EU) “Third Package” of liberalization, and compares the outcomes of airline liberalization in the United States and Europe. Taking effect in 1993, the third aviation package provides power to carriers and allows them to set itinerary fares and freely enter a new route within Europe. Results suggest that a subsidized program (“EU-wide

EAS program”) in Europe can help regional airports located in small and isolated communities connect to the national transportation network. Williams and Pagliari (2004) examine the use of the public service obligation (PSO), a subsidized program implemented in Europe, in different European countries and find that some countries such as France and

Norway have well accepted and implemented PSO mechanism, although they have unique challenges. Conversely, the mechanism has not been well used in UK.

Supportive and critical voices always exist simultaneously. Subsidized programs have been criticized because of associated high costs. Cunningham and Eckard (1987)

1 Large hubs board more than 1% or more of the total passengers per year in the United States, medium hubs board at least 0.25%, but not more than 1%. 2 For example, many of the EAS communities in Montana require more than $200 in subsidies, but are located over 210 miles from both Denver (DEN) and Salt Lake City (SLC) airports, their nearest medium and large hubs.

10 suggest that the EAS program has little effect on improving the quality of air service in small communities. Their analysis indicates that the subsidies have reduced itinerary fares but have no impact on other measures of airport quality other than reducing flight frequency. Within the context of EAS, it is important to note that the program and its participating communities are somewhat dynamic. Subsidy levels vary on an annual basis and a handful of airports come and go in the program. Regardless of this dynamism, many problems remain, including the overall efficiency of the EAS program (Grubesic et al.,

2014). Not only do EAS airports cannibalize their own markets (Grubesic & Matisziw,

2011), but they are also frequently located near relatively vibrant small hubs (as defined by the FCC), which offer a more robust range of air transport services, but are not considered in the EAS community eligibility criteria (Grubesic & Wei, 2013; Grubesic et al., 2012). In short, there is room for improvement in the EAS system and the rural air transport network, more generally.

2.1.2 Airport Competition

There is no doubt that deregulation stimulates airport competition that focuses efforts on attracting carriers and travelers, and it leads to different extents of catchment areas of airports associated with several parameters such as accessibility (Redondi et al.,

2011). Redondi et al. use 2008 global schedules data to investigate hub competition in a global air transport network. Their work highlights that hub competition exists globally and each of the major hubs has at least three competitors that cover more than 50% of origin-destination market provided by the major hub. The authors also find that the geographic location is an important factor for any of hubs and its competitors. For example,

European hubs have better locations to connect the global air transport network than North

American and Asian competitors.

Barrett (2000) presents 17 case studies of airport competition in the European

airline industry. The author points out that the busiest international route (Dublin-London)

in Europe has been highly influenced by airport competition caused by the European

liberalization. Passengers have benefited from this competition with lower itinerary fares

and more flight alternatives. In addition, results indicate that the collaboration of lesser-

used airports and low-cost carriers can significantly increase passenger volume.

One of the interesting geographic quirks of the commercial air industry, particularly

in the United States, is the presence of multiple airports in a metropolitan area (i.e., multi- airport regions). There are many regions in the U.S. that have more than one airport, including Los Angeles, San Francisco, Chicago, Houston, New York and Washington.

Derudder et al. (2010) point out that airports located within these mega-regions rarely engage in head-to-head competition for passengers. Instead, each airport and its associated carriers participate in the development of functional niches, where their geographical scales of operation (e.g. international, national and regional) and the speciﬁc role of each airport in the airline network (e.g. origin/ destination versus hub airports) dictate their place in the regional hierarchy of commercial air services. Consider, for example, of the six commercial airports located in Greater Los Angeles. Los Angeles International Airport (LAX) is the only airport that functions as a major international hub. Although both the John Wayne

Airport (SNA, in Orange County) and Ontario (ONT) do provide modest levels of international service, neither functions as a major hub airport. Instead, they serve as

important domestic gateways to Southern California. Burbank (BUR), Long Beach (LGB)

and Santa Barbara (SBA) are smaller regionally focused airports that process the bulk of

their trafﬁc as spokes to/from larger hubs.

However, airport competition is not always perfect. Goetz (2002) discusses the

relationship between “pockets of pain” and predatory behavior. The dominant carriers with

stable financial status exclude new entrants from the local market by sharply reducing

itinerary fares. Once the new entrants fail to operate their businesses in the local market,

the large carriers increase itinerary fares back to the original price. These “pockets of pain”

(Goetz 2002) manifested in many different ways, but included airports that were treated as

fortress hubs by carriers (Zhang 1996).3 Predatory pricing at these hubs was common, as was a general policy of not leasing gates to competitors, especially low cost carriers (Shaw and Ivy 1994; Goetz and Sutton 1997; Vowles 2000; Goetz 2002; Grubesic and Zook 2007;

Goetz and Vowles 2009). Brueckner and Spiller (1991) find that airport competition in a hub-and-spoke network may have negative impact on a market in terms of their results derived from a linear model. Starkie (2002) specifically discusses imperfect airport competition that can result in the emergence of dominant carriers and suggests that an ex- post regulation is necessary to be provided under normal commercial law in order to restrict the monopolistic phenomenon in airline industry.

2.1.3 Airport leakage and choice

An important consequence of competition between neighboring airports is market leakage. By deﬁnition, leakage refers to the process by which passengers do not select their

3 By deﬁnition, fortress hubs are airports where a single carrier has a local monopoly on spoke segments and/or airport gates (Zhang, 1996).

13 most proximal airport for commercial service. Instead, passengers travel to more distant airports to take advantage of lower fares, more convenient scheduling or service from a speciﬁc airline (Pels et al. 2000; Pels et al. 2001; Vowles 2001; Suzuki et al. 2004; Fuellhart

2007; Suzuki 2007; Tierney and Kuby 2008; Matisziw and Grubesic 2010; Grubesic and

Matisziw 2011; Marcucci and Gatta 2011; Lian and Rønnevik 2011). These types of

“trafﬁc shadows” (Taaffe 1956) are a common feature in large metropolitan regions with a dominant international hub. However, these shadows are often moderated by the functional niches within a region (Derudder et al. 2010), as well as the terrestrial travel costs (time, expense and aggravation) for making a journey from home to larger airports for commercial service. That said, the factors that influence a passenger’s choice are direct determinants of airport competition. Access time and cost for the passenger traveling to nearby airports, and level of service, including itinerary fare, frequency, and flying time

(direct or indirect flights), have been identified as significant determinants in the literature

(Harvey, 1987; Innes & Doucet, 1990; Phillips et al., 2005; Zhang & Xie, 2005). Here, flying time is measured as direct or indirect flights because the number of required connections to reach a destination is considered more important than distance (miles flown) from a passenger’s perspective (Grubesic & Zook, 2007).

For example, consider the San Francisco Bay area. Pels et al. (2001, 2003) developed nested multinomial logit models to examine the airport and airline choice for passengers departing from this region. Results identify access time and cost as significant factors that negatively influence a passenger’s choice. Specifically, business travelers have a higher tolerance for cost and are more sensitive to frequency of air service than leisure travelers. In part, this is because leisure travelers have more flexibility for trip planning

including departure and/or arrival time in order to find less expensive tickets. In addition,

the authors find that access time to departure airports is a more important determinant than

cost. In other work, Basar and Bhat (2004) construct a probabilistic choice set multinomial

logit (PCMNL) model to investigate the airport choice in the San Francisco Bay area. The

authors find that access time and flight frequency are two important factors that influence

airport choice. Further, demographic groups of passengers have different levels of

tolerance for the two determinants. Specifically, female and high-income passengers are not sensitive to flight frequency, instead caring more about access time. Hess and Polak

(2005, 2006) utilize logit models to examine airport choice in the San Francisco Bay area and the Greater London area. Similar to other studies analyzing airport choice, the authors also find that access time and cost, flight frequency and time are primary determinants for airport choice. Results also confirm the previous related work that leisure passengers are more sensitive to itinerary fares than business travelers. Ishii et al. (2009) identify that several factors have strong impact on airport choice including access time, service frequency, flight delays, and carrier mix.

Considering other study areas, Suzuki et al. (2003) examine airport choice in a single-airport region, Des Moines, Iowa, and find that leisure travelers are more likely to leak to larger metropolitan airports when compared to business travelers. Previous experiences (good/bad) also dictate airport choice for travelers in single airport regions.

Zhang and Xie (2005) study airport choice between a small community, the Golden

Triangle Regional Airport (GTR) in Mississippi, and distant major hubs. The authors use

2004 survey data, combined with a logistic regression model, to examine the choice behaviors of passengers. Several factors significantly influence passengers’ behaviors,

15 including itinerary fares, flight schedules and experience at the GTR airport. Dresner (2006) further studies the two types of passengers, business and leisure, and compares their characteristics. Results suggest that the proportion of leisure passengers will increase when low-cost carriers increase their market share because low-cost carriers provide more low- fare flights. However, the author identifies that characteristics of leisure and business passengers are similar when it comes to selected departure airports, parking requirements, and the number of checked bags. It implies that itinerary fare is a critical factor that affects airport competition and choice for some types of passengers. Fuellhart (2007) uses spatial econometric models that consider autocorrelation to investigate airport leakage in the Mid-

Atlantic region. The model includes three independent variables that are all related to population and proximity of a nearby airport. Results indicate that airport leakage possibly exists between Harrisburg International airport and Baltimore-Washington International airport.

Airport choice also receives attention outside of the United States. Innes and Doucet

(1990) examine a series of factors that are related to airport choice in New Brunswick,

Canada. Aircraft type is identified as the most important determinant on choosing an airport.

Specifically, travelers prefer to drive to a distant airport providing jet services. Additionally, flying time and non-stop flight are also criteria that affect passengers’ decisions. Lian and

Ronnevik (2011) use the data from 2003 and 2007 Norwegian Air Travel Surveys, combined with logistic regression analysis, to examine airport competition in Norway.

They find that airport leakage is highly related to a certain demographic group of passengers (leisure travelers) and proximity of a major hub.

2.2 Airport Structure and Associated Network Evaluation

2.2.1 Basic structures

Taaffe et al. (1963) thoroughly discuss the evolution of a transport network. Their discussion provides an essential understanding to the development of an air transport network. Currently, there are two types of basic structures utilized in the airline industry: hub-and-spoke and point-to-point. Since the 1978 deregulation of the airline industry in the

United States, hub-and-spoke structures have been adopted and implemented by carriers across the air transport landscape (Goetz, 2002). Major airports located in metropolitan areas become hubs and those in small cities and rural areas function as spokes (O’Kelly,

1998). The hubbing geography has helped carriers achieve economies of scope in the industry (Goetz, 2002; Grubesic & Zook, 2007). In hub-and-spoke systems, economies of scale develop when carriers to operate extremely large planes (e.g., Boeing 767s, 777s,

787s) to move passengers between hub cities. These high-volume passenger movements are more cost effective than running smaller planes (and requiring more ﬂights to move the same number of people) on inter-hub routes. Specifically, hub-and-spoke structures help minimize the cost of adding a spoke to the network and maximize the associated profit

(O’Kelly, 1998; Goetz, 2002; Grubesic & Zook, 2007; Grubesic & Matisziw, 2011). Thus, when airlines expand their network footprint by adding spoke routes, they beneﬁt from increased enplanements and revenues. The costs for spokes are usually nominal with respect to the additional revenues realized by a carrier. When combined within a hub-and- spoke system, economies of scale and scope form the foundation of an extremely efﬁcient and cost effective service topology. Higher levels of air service and more associated facilities have been provided in large cities. Conversely, services have been reduced, even

terminated, in small cities and rural areas. It is the natural selection of carriers who seek to

increase profits.

Hub-and-spoke systems, for better or worse, also exhibit spatiotemporal dynamism

within the commercial air transport industry (Grubesic et al. 2009). A variety of economic,

operational and regional market factors contribute to this dynamism. For example, the

emergence of low cost carriers (LCCs) in the late 1980s and early 1990s, including

Southwest Airlines, drastically changes the service landscape (Dresner et al. 1996; Vowles

2000; Goetz and Sutton 1997). Other factors such as terrorist attacks (Ito and Lee 2005),

disease (Bowen and Laroe 2006; Grubesic et al. 2009), increasing fuel costs and mergers

and acquisitions (Bhadra 2009) can also impact service. Part of this dynamism can also be

attributed to the dehubbing process. By deﬁnition, dehubbing refers to a dominant carrier

dismantling its structure of connections at an airport, or, in worst-case scenarios, the complete withdrawal of a dominant carrier from an airport (Bhadra, 2009). Bhadra (2009) argues that dehubbing is a strategy used by carriers to reduce the losses that multi-hub operations often accrue. In short, although a carrier may be operating more than one hub, rarely are all the hubs proﬁtable.

The implications of dehubbing for markets are varied. As detailed by Rodríguez-

Déniz et al. (2013), the dehubbing process implies a sudden change in connectivity for the dehubbed airport. Links that once existed between airport pairs are no longer available to passengers. This process is not uncommon. For example, Redondi et al. (2012) identify 37 instances of dehubbing in the international market between 1997 and 2009. This includes

11 cases in Europe, 11 in North America, 8 in Central-South America and 7 in the Asia-

Paciﬁc region. However, once the dehubbing process is complete, Bohl (2013) suggests

18 that “capacity dumping” can occur. In short, when routes are suddenly dropped by a dominant carrier due to dehubbing, potential future proﬁts along those routes increase for competitors because re-entry by the dominant carrier is unlikely. Thus, as consumers look for alternatives, the most attractive routes may entice new service by several airlines simultaneously, leading to market/route overcapacities. In turn, this can result in lower fares because of local competition and the glut of seats on certain routes after the dehubbing process is complete. It is important to note, however, that dehubbing is not always a death- blow to an airport, and “rehubbing” can occur. This is rare. Once large numbers of seats and/or ﬂights are lost, most airports never attain their peak levels again. This is especially true for airports where dominant carriers completely abandon their hubbing activity

(Redondi et al. 2012).

Alternatives to hub-and-spoke include point-to-point topologies, which are often implemented by low-cost carriers who provide the market with less expensive tickets than major carriers operating hub-and-spoke strategies (Doganis, 2001; Reynolds-Feighan,

2001; Williams, 2001; Burghouwt et al., 2003; Burghouwt & de Wit, 2005). This type of structure has been applied to the airline industries in the United States and Europe.

Specifically, Reynolds-Feighan (2001) states that Southwest Airlines is a typical U.S. carrier that utilizes point-to-point structure and benefits from this strategy. Using the Gini

Index to measure the air traffic distribution of carriers, carriers (e.g. Southwest Airlines) with lower Gini Index scores focus their efforts on short-haul markets by means of point- to-point strategy. Due to the spread of low-cost carriers and the development of efficient regional aircraft, the point-to-point structure is adapted and its strategy is applied to the indirect connections between cities in Europe (Burghouwt & de Wit, 2005). It is interesting

19 to note that the point-to-point structure is prioritized to maximize profit when the distance of a pair of cities is small (Lederer & Nambimadom, 1998; Grubesic & Zook, 2007). For example, non-stop connections operated by Southwest have an average distance less than

500 miles (Reynolds-Feighan, 2001).

Bania et al. (1998) examine and classify 13 major airlines according to four hypothetical route structures: mono-hub, dual-hub, diffuse system-1 and diffuse system-2.

The results highlight that most of major airlines adopted hub-and-spoke structures for air service, and in 1989, Alaska and Southwest Airlines implemented diffuse structures indicating point-to-point air service.

2.2.2 Advanced Network Evaluation

Given the basic structures of airline networks, connectivity and accessibility of airports have been further explored by utilizing different approaches and examining various factors. Chou (1993) uses spatial analyses of first- and second-order nodal accessibility and regression analyses to evaluate the impact of airline deregulation on nodal accessibility.

Bowen (2000) examines the temporal difference of hubs in Southeast Asia by evaluating

Shimbel indices of five airports, including Bangkok, Jakarta, Kuala Lumpur, Manila and

Singapore from intra- and inter-regional perspectives. Bowen finds that Kuala Lumpur and

Singapore are two airports with increasing Shimbel indices between 1979 and 1997, and the other three hubs have decreasing Shimbel indices, although connections at the three airports increased.

Fleming and Hayuth (1994) use centrality to identify spatial qualities of transportation hubs and delineate spatial distribution of airports that are strategically

important in a transportation system. Wang et al. (2011) utilize centrality measures,

including degree, closeness and betweenness to examine the air transport network of China

(ATNC) and measure the nodal accessibility of airports. Not surprisingly, Beijing,

Shanghai and Guangzhou are identified as three important airports in China in terms of degree and closeness measures. Airports with high betweenness are not constantly consistent with the first two centrality measures such as Urumqi in the northwest and

Kunming in the southwest with high betweenness and relatively low degree and closeness because of geographic and political reasons.

Zook and Brunn (2005) use travel time, cost and distance as measures to explore the connectivity of European cities in air transport networks and identify the hierarchies of

European airports. More importantly, the authors find the “shadow effects” that present that a large airport’s nearby competitors are usually relatively smaller. Zook and Brunn

(2006) also use these measures for global airports and explore airport accessibility at a comprehensive level. Further, the authors illustrate the structure of the global airline industry and delineate the variations among airports within each region. Air passenger flow is measured to identify and assess connectivity of U.S. cities in the world city network

(Derudder & Witlox, 2005; Derudder et al., 2007).

Efforts to disentangle network topology, connectivity and accessibility often result in the generation of airport typologies, which are useful for summarizing the wide variety of attributes found between airports. The Federal Aviation Administration (FAA) (2012) uses simple typologies to categorize airports by passenger enplanements. More importantly, airport typologies can also be identified using alternative analytical approaches such as the

Nystuen-Dacey (1961) method (Grubesic et al., 2008; Grubesic et al., 2009) and graph-

theoretic or statistical network-based approaches geared toward identifying hierarchies between airports and geographical regions (Reed, 1970; Wacht, 1974; Nader, 1981; Ivy,

1993; Shaw & Ivy, 1994).

Guimera et al. (2005) analyze the worldwide air transportation network and categorize airports with seven different roles: four roles for non-hub airports

(“ultraperipheral nodes”, “peripheral nodes”, “nonhub connector nodes” and “nonhub

kinless nodes”) and three roles for hubs (“provincial hubs”, “connector hubs”, “kinless

hubs”) in terms of values of the participation coefficient. Specifically, Guimera et al. (2005)

examine the worldwide airports by using betweenness centrality and find that some airports

with high betweenness are not highly connected such as Anchorage in Alaska and Port

Moresby in Papua New Guinea. Both airports provide main connections to the other cities

in their communities. Additional layers of context, such as geopolitical factors, including

distribution of locations of borders between states, can also be considered when evaluating

network community structure (Guimera et al., 2005).

Grubesic and Zook (2007) utilize a non-hierarchical clustering approach to classify airports and propose a new airport typology. Specifically, the new typology consists of two major categories (hubs and spokes) and each category has four specific hierarchies. The hub category includes “highly accessible national hub”, “moderately accessible national hub”, “highly accessible regional hub” and “moderately accessible regional hub”. The spoke category includes “highly accessible spokes”, “moderately accessible spokes”,

“geographically isolated spokes” and “geographically remote spokes/small markets”.

Derudder et al. (2007) examine U.S. cities in the world city network in terms of global origins and destinations of airline passengers, and observe a close relationship

between hierarchy and regionality. This work indicates that less large cities have fewer

connections with other cities and the connections are more regional. The authors highlight

that Miami is a special case, and it is different from other domestic hubs because it serves

as a gateway that connects Latin America. Only four of the top fifteen connections from

Miami are domestic, including New York, Atlanta, Los Angeles and Washington.

Neal (2010) employs a bivariate regression analysis in terms of degree centrality

with two types of weights, origin/destination centrality and hub/spoke centrality, and

highlights some interesting outliers. Specifically, Atlanta and Dallas are identified as hubs

for major airlines and New York and Los Angeles function as centers for resource exchange.

These airports can be further categorized into two types of large airports: international

gateways and domestic hubs. Domestic hubs that serve as focal points of hub-and-spoke

networks with high degrees of domestic connections are usually located in interior cities

(e.g. Atlanta) and international gateways serving as funnels for international connections

are usually located in large coastal cities (e.g. Los Angeles) (Goetz & Sutton, 1997).

Focusing on non-U.S. air transport networks, Burghouwt and Hakfoort (2001)

utilize cluster analysis with Ward’s method to classify airports in Europe. Specifically, they

examine schedules data, including average seat capacity, average number of destinations

and average number of intercontinental destinations during 1990-1998 for airport

classification and find that certain large hub airports only dominate the inter-continental

flights, not the air service within Europe. Lin (2012) notes that the hierarchical structure of

China’s air transportation system is consistent with the basic hub-and-spoke structure and further identifies the flight distance as a factor for classifying the China’s air transportation network into different hierarchies.

From a global perspective, Smith and Timberlake (2001) deploy centrality

hierarchies and clique analysis to examine the world cities and associated airports, and

identify their rankings in worldwide air transport network in terms of number of passengers.

2.3 Airport Performance and Efficiency

There is no doubt that access and accessibility are very important concepts for evaluating performance of an airport within an air transportation system. It has been studied by researchers from different perspectives and with various approaches (Chou, 1993;

Raynold-Feighan & McLay, 2006; Matisziw & Grubesic, 2010). Passenger volume and service frequency are two basic measures for evaluating airport accessibility and service quality (Goetz & Graham, 2004; Derudder et al., 2007).

2.3.1 Spatial Evaluation

As mentioned previously, access time and cost, flight frequency and number of direct flights are important factors that influence airport competition and choice. In fact, these identified factors delineate the characteristics of access to air transport and accessibility within an air transportation system. One outstanding research question in this domain is whether or not spatial efficiency plays a role in competition and choice.

Flynn and Ratick (1988) construct a multi-objective optimization framework to evaluate the Essential Air Service (EAS) program in North and South Dakota. The framework maximizes the total weighted population covered in the region and minimizes the total operational cost of the EAS program in the region. The proposed mathematical

24 programming model provides insights for decision makers to evaluate the federally subsidized program and make it more efficient.

Matisziw and Grubesic (2010) develop a new metric to examine access to air transport and accessibility within the air transport network together. The results present spatial heterogeneities in the 48 contiguous US states in terms of computed accessibility index of each census tract. Specifically, several regions are identified as high accessibility areas, including the Great Lakes region, Northeast Megalopolis, the Piedmont region and the Texas Triangle. Conversely, low accessibility areas include Nevada, southeastern

Oregon, northern Maine and the Great Plains states.

Grubesic and Matisziw (2011) also study the spatial distribution of EAS airports in the United States and identify coverage redundancy of air service with respect to overlapped market areas between EAS and non-EAS airports. Specifically, there is 35% overlap between EAS airports and large/medium hubs when driving distance from each airport is 70 miles, which is the distance criterion of the EAS program. When the driving distance between airports increases to 100-plus miles, the overlap between EAS and non-

EAS airports increases up to 56%. Including all non-EAS airports, the overlap rates are nearly 78% and 95% at the 70 and 100 miles driving distance, respectively. The authors suggest that the distance requirement for eligible rural airports could be changed to a larger distance threshold in order to improve spatial efficiency of the EAS program.

Matisziw et al. (2012) develop a multi-objective optimization model to investigate the performance of EAS airports from 2001 to 2006. Specifically, the model focuses on two objectives: minimizing the hub access cost and maximizing community accessibility.

The authors use a weighting method to provide a series of combinations of two objectives

with different weights and make a tradeoff. Eleven solutions are generated and classified

into three objective categories: cost minimizing, accessibility maximizing and intermediate.

Results show that only 5% of observed hubbing structures (EAS airports and their

connecting hubs) satisfy all three objective categories. This low level of overlap between

observed and modeled hubbing structure implies that there is potential to improve the

efficiency of the EAS structure.

Grubesic et al. (2012) develop a spatial optimization model to cover all census tracts served by EAS airports and minimize the total number of EAS airport in order to reduce the total costs. The authors utilize a sensitivity analysis to evaluate the program efficiency. Results suggest that the changes (e.g. increasing threshold of distance or including small hubs) would eliminate redundant EAS airports to save costs and improve efficiency. That said it would reduce the total amount of subsidy rates for all still existing

EAS airports. Grubesic et al. (2013) evaluate the spatial distribution of EAS airports by deploying an optimization model to maximize population coverage of census tracts served

by a certain number of EAS airports. Results show that the EAS program can reduce a

large amount of subsidies with limited impact on the provided service level. Specifically,

the authors emphasize that the program can reduce 20% of subsidies and eliminate 24 EAS

airports, but the population coverage is still greater than 98.5%.

2.3.2 Operational Evaluation

Operational evaluation focuses more effort on level of air service. Several studies

use data envelopment analysis (DEA) to evaluate airport performance. DEA is a linear

mathematical programming technique, and it determines efficiency of an entity that

26 contains inputs and outputs (Gillen & Lall, 1997; Grubesic & Wei, 2012). Output-oriented

DEA analyzes inputs to generate optimal outputs that can be used as threshold to evaluate efficiency (Grubesic & Wei, 2012).

Gillen and Lall (1997) examine 21 of the top 30 airports in the United States over a five-year period (1989-1993), combined with DEA and a Tobit regression model, to evaluate airport performance. First, the authors deploy DEA to obtain an overview of airport performance. Results show that Anchorage has a low level of performance and

Atlanta has a high level of performance, according to comparing terminal efficiency scores between 1989 and 1993. Gillen and Lall point out the development of long-haul aircraft and international gateways on the West Coast is the reason that Anchorage is less efficient in 1993 compared to 1989. Second, utilizing DEA output (efficiency index) as the dependent variable, they develop a Tobit regression model to further investigate the relationship between airport efficiency and airport characteristics (e.g. number of runways).

Their results identify two important variables that positively influence airport efficiency: greater proportion of international passengers and more gates. Adler and Berechman

(2001a) evaluate the efficiencies of airports that are mainly located in West Europe by using a combination of DEA and Principal Components Analysis (PCA) to examine subjective and objective airport data. Specifically, subjective data are collected from questionnaires indicating airlines' viewpoint to each airport, and objective data describe the characteristics of each airport, such as number of terminals, number of runways and distance to the nearest major city. Results show that Geneva, Milan and Munich have good performance with high efficiency scores, and Athens, Charles de Gaulle and Manchester have low scores. Based on a two-hub network, Adler and Berechman (2001b) use an

integer program to identify potential hub combinations in Western Europe in order to

minimize the total passenger distance traveled, and then an economic model with non-

linear program to maximize profits. Results suggest that the hub combination of London

Heathrow and Rome or Stockholm shows more potential to increase profits than other hub

combinations. Adler and Berechman further note that the large demand at Heathrow and

the geographic location of Rome or Stockholm explain the hub combination. Bazargan and

Vasigh (2003) evaluate the productivity of 45 U.S. commercial airports by using DEA.

Specifically, the authors examine financial and operational data to measure efficiencies of

15 large hubs, 15 medium hubs and 15 small hubs, and discover that small hubs have better

performance than large hubs in terms of efficiency scores. Grubesic and Wei (2012) use

DEA, combined with airport characteristics and local population, to evaluate the

performance of the EAS program. Several rural communities with low passenger volumes

and load factors are identified by DEA as efficient airports such as Glasgow, MT, Chadron,

NE, and Vernal, UT. As the authors explain, DEA presents a useful approach to evaluate

the performance of EAS program, but it should be used with other decision-making tools

to make a more informed recommendation.

Considering other methods, Vowles (1999) examines airport data indicating airport characteristics, combined with a logistic regression analysis, to measure the probabilities of U.S. communities that may lose air service. Rietveld and Brons (2001) develop a flight coordination coefficient to study four European airports by investigating the impact of frequency changes on waiting and rescheduling time. Heathrow and Charles de Gaulle have longer waiting times, and Frankfurt and Schiphol have shorter waiting times. One major reason is that the former two airports have low values of flight coordination coefficient. It

indicates that the two airports have less efficient flight coordination with carriers.

Reynolds-Feighan and McLay (2006) examine the accessibility of European airports by

considering the seat capacity and relative importance of destination for each pair of origin

and destination. Specifically, the authors use a weighted measurement to indicate the

importance of destinations in their new accessibility measure. The authors find that Dublin

and Manchester are two airports with relatively high accessibility scores.

2.3.3 Interaction with local markets

As detailed by Oum et al. (2008), an efficient airport can stimulate local and regional economies and then improve the quality of life in those areas. It indicates that the interaction between airports and local markets is important, and it further implies that the level of local business activities can be a measure for airport performance. Communities with easier access to the national transportation network would have greater potential for local economic growth and development (Keeble et al., 1982; Keane, 1984; Johansson,

1993).

Considering the city of Chicago as a single node, Taaffe (1959) finds that the tourism has a significant impact on increasing air passenger traffic by means of illustrating the difference of passenger volume, fares and revenues for connections between Chicago and other airports in the United States during the period of 1949-1955. Bowen (2000) argues that national governments use a variety of tools to shape the development of air transportation networks, such as airline industry liberalization and airport development.

Air transport services are related to manufacturing development, international businesses development, and tourism development.

Debbage (1999) identifies that there is a connection between “air passenger volume and air service connectivity” and “administrative and auxiliary employment” at airports in the Carolinas. In addition, there is a positive relationship between knowledge-based economic activities (information technology, biotechnology, electronics, and management services) and local air travel (Button & Taylor, 2000). In metropolitan areas, some business sectors can stimulate air transportation because of the importance of face-to-face communication (Debbage & Delk, 2001). Brueckner (2003) also examines the relationship between airline traffic and employment in metropolitan areas in the United States and notes that good air service can positively influence local economic growth. Results specifically present that a 10% increase in passenger volume in a metropolitan area can contribute about a 1% increase to employment in service-related industries. Here, the author emphasizes that the air service is only relative to local businesses in service-related industries, not all business sectors. Percoco (2010) studies the interaction between airport activities and local economic development in Italy in terms of the model developed by Brueckner (2003).

Percoco finds that local service-related businesses are more sensitive to air traffic than other types of business activities. Alkaabi and Debbage (2007) also focus their research on metropolitan areas. Based on their findings, there is a strong relationship between air transportation and economic development, especially in an area in which the major business is relative to professional, scientific and technical activities. Liu et al. (2006) also find similar results that confirm the relationship between airport activity and local businesses focusing on professional, scientific and technical activities.

Halpern and Brathen (2011) utilize a qualitative approach to study the impact of two airports in Norway in terms of regional accessibility and social development.

Specifically, the authors use a postal survey and sampling method to compare the two airports and find that the reasons for travel differ among the passengers at the two airports.

Results indicate that residents in the region that has an airport providing non-stop international service take more flights during holidays, and residents in the remote region with limited health service take more flights for health service.

Focusing on small communities and rural airports, Kanafani and Abbas (1987) analyze the correlation between local air service and economic impact of small airports.

Their research indicates that a short distance from a large hub airport can negatively influence local air service regardless of whether the local economy is growing. That said, a large hub airport is the first important factor to affect local air service, and a growing local economy is the second. Further, air travel provides a high-speed access to business people for movement and attracts businesses preferring face-to-face communication for small communities (Kanafani & Abbas, 1987). Kaemmerle (1991) builds an economic model to evaluate relationships between small community/air service variables and air service demand. Goff (2005) develops regression models to examine the determinants of air service to small cities with populations less than 500,000. The results present that market size measured by population and proximity of a resort have impact on air service in small markets. As detailed by Rasker et al. (2009), this is particularly true in the western portions of the United States, where enormous expanses of public lands are not available for development. This means that the distance between urban centers, towns and other communities are much larger, reinforcing the need for air transport to help get products to market and enable the movement of residents participating in knowledge-based professions

(Mathur & Stein, 2005).

Reynolds-Feighan (2000) examines the U.S. airport hierarchy and finds that small communities on the bottom of the hierarchy have variable relationships between air traffic and local population. As a focal point, a collection of small communities is further examined with local demographic and economic characteristics. The results suggest that small communities with EAS subsidies have a longer distance from the major hubs and lower economic growth than unsubsidized small communities. The author argues that the

EAS program solves the economic issues in rural and isolated communities. Özcan (2014) focuses on the EAS program and examines the relationship between the EAS and local economic growth by using a two-stage least squares estimation. Results indicate that a 1% increase in passenger volume in an EAS airport with a minimum of 1,000 annual passengers can contribute a 0.12% increase to local per capita income.

3. STUDY 1: EXPLORING DEHUBBING AT ONE AIRPORT

As detailed previously, several mid-sized airport hubs in the United States and

Europe have suffered dehubbing process in terms of various factors and more similar-sized

airports may experience the process in the future. At regional scale, this chapter focuses on

one single airport, Cincinnati/Norther Kentucky International Airport (CVG), to examine

the airport access and accessibility by exploring the dehubbing process that CVG has

experienced. It highlights the operational, market and geographic factors that contributed

to CVG’s decline and use basic exploratory data analysis and statistical analysis to provide

perspective and deepen our understanding of this process.

3.1 Introduction

The city of Cincinnati is located on the north bank of the Ohio River in

Southwestern Ohio. Although it remains the third largest city in the state, years of

population loss have reduced Cincinnati from approximately 500,000 residents between

1950 and 1960 to a city of 297,517 in 2013 (U.S. Census 2013). Its metropolitan area which

includes portions of Ohio, Kentucky and Indiana has a population of approximately 2.13

million, ranking it the 28th largest in the U.S., just ahead of Cleveland and behind

Sacramento.

The region is also home to Cincinnati/Northern Kentucky International Airport

(CVG), which is located in Hebron, Kentucky, approximately 15 min southwest of

downtown Cincinnati (Figure 3.1). At one time, CVG was the second largest hub in the

Delta Airlines network. For example, in 2005, it offered more than 600 ﬂights per day to

nearly 150 destinations in the United States and Europe and processed over 22.7 million

33 passengers. During 2013, CVG provided fewer than 180 daily departures to 77 destinations and processed fewer than 6 million passengers. The story of CVG's decline is a complex one, fueled by a combination of factors that include shifting commercial carrier strategies, challenges to operational efﬁciency, ineffective hub structures and network topologies, regional competition and a changing local business environment. The purpose of this chapter is to provide a detailed case study of the factors that contributed to CVG's deterioration over the past decade, highlighting the operational, market and geographic factors that helped fuel its decline.

Figure 3.1: Major airports of the tri-state area (Ohio, Indiana and Kentucky)

This type of case study is important for several reasons. First, Cincinnati is not

unique. There are many mid-sized markets throughout the United States that are struggling

to maintain commercial service or hub status. As a result, the factors that contributed to the

decline of CVG can serve as a cautionary tale for markets that have not yet suffered through

the dehubbing process, but may be vulnerable to this trend in the future. Further, CVG is

not removed from the global dynamics of the airline industry that are constantly facilitating

shifts in destinations served, seats available or the emergence/decline of gateway

operations. For example, O'Connor and Fuellhart (2015) suggest that improvements in

aircraft technology and rising demand for service to secondary and smaller airports serving

routes between Australia and Asia have changed the structure of gateways throughout

Australia. In particular, as the network moves away dominant trunk routes between major

gateways, secondary cities thrive. Although CVG did not beneﬁt from such shifts in the

United States, the factors which led to its decline remain important within the larger global

context. A second reason that a case study for Cincinnati is important is because geographic

market for CVG is an interesting one. It is sandwiched between ﬁve additional

small/medium-sized airports, all within a two-hour drive, that offer a similar palette of commercial air transport services. As a result, there are some instructive features related to market leakage and spatial competition for passengers between these markets that are worth detailing. Once again, the fortunes of secondary air transport gateways (O'Connor and Fuellhart 2015) remain salient, as do the market forces associated with deregulation both in the U.S. and abroad.

Lastly, the ﬁnal chapter for CVG is yet to be written. There are glimmers of hope for the airport, particularly with the emergence of several low cost carriers within the

35 market. In short, dehubbing is not fatal, but the path for attracting a good mix of alternative carriers and providing the market with appropriate levels of service after dehubbing is not easily traversed.

It is also important to note that this chapter departs, quite signiﬁcantly, from the standard academic fare in transport geography. Because it is structured as a case study focused on CVG and its competing regional airports, the small sample size (n = 6) precludes us from developing a detailed inferential framework to explain the dehubbing process. We view this as a strength of the chapter, rather than a limitation. This focus allows for the creation of a more comprehensive narrative that highlights the many political, operational and geographic subtleties that permeate CVG's situation, some of which are too nuanced to fairly capture with traditional conﬁrmatory approaches. This is not to suggest that this chapter is devoid of quantitative analysis, quite the opposite is true, but the analysis is decidedly exploratory in nature. Thus, the results of this chapter should be viewed as the beginning of a conversation, rather than the end of one.

3.2 Background

3.2.1 Cincinnati/Northern Kentucky International Airport (CVG)

Ironically, the story of CVG begins with Lunken Airport (LUK), which is located near the conﬂuence of the Little Miami River and the Ohio River in the city of Cincinnati

(elev. 147 m). Because of its location near the banks of the Ohio, LUK suffered from frequent ﬂooding and heavy fog, both of which signiﬁcantly impaired airport operations

(CVG 2013). Opportunistically, in 1942 the state of Kentucky lobbied Congress to establish an airﬁeld in Boone County, southwest of Cincinnati, on a plateau high above the

Ohio River (elev. 273 m), far less prone to fog and nearly impossible to ﬂood (CVG 2013).

By 1947 commercial carriers were using the airport and by 1960, the ﬁrst jet, a Delta ﬂight from Miami, Florida, touched down at CVG. By 1970, annual enplanements at CVG reached 2 million (CVG 2013).

Once the U.S. commercial air industry deregulated in 1978, CVG was poised for signiﬁcant growth. Delta Airlines made CVG its second largest hub. In turn, hundreds of domestic and international routes were added to the airport. By 2005, CVG was operating more than 100 gates, processing nearly 23 million passengers per year and offering 670 daily departures. Although this type of growth was not unique to Cincinnati,4 it certainly was not shared by the vast majority of commercial airports in the U.S.

It is also interesting to note that for Cincinnati, the traditional concept of a multi- airport region does not fit. Aside from Lunken Airport, which offers sporadic charter flights to Chicago, Charlotte and Morristown, NJ via Ultimate Air Shuttle, there are no traffic shadows, nor any intraregional competition between airports in the Cincinnati metropolitan area. Instead, Cincinnati suffers from fierce interregional competition, emanating from five small/medium hubs located in Indianapolis (IND), Dayton (DAY), Columbus, (CMH),

Lexington (LEX) and Louisville (SDF) (Figure. 1). Worth mentioning, but not included in subsequent analysis is Rickenbacker International (LCK), a non-hub located south of downtown Columbus that offers a limited set of commercial services.5

4 Similar growth trajectories occurred in places such as Pittsburgh and St. Louis, among others. 5 LCK is primarily a cargo airport, but it does offer limited commercial passenger service to Fort Lauderdale, Orlando/Sanford, Punta Gorda/Fort Myers, St. Petersburg/Clearwater, Myrtle Beach and New Orleans via Allegiant Air.

There are three reasons that interregional competition is so prevalent for CVG. The

ﬁrst centers on the zemblanity of its location. It is virtually equidistant to the medium-sized

hubs of IND (125 miles), CMH (127 miles) and SDF (104 miles), and the small hubs of

DAY (80 miles) and LEX (82 miles). This effectively puts ﬁve additional airport choices

within an easy drive of the Cincinnati metropolitan area. In fact, previous work suggests

that driving distances of 75–150 miles are reasonable standards for evaluating access to

airport alternatives (Lin 1977; Kanafani and Abbas 1987; Kaemmerle 1991; Matisziw and

Grubesic 2010; Grubesic and Matisziw 2011). Second, CVG has historically been the most

expensive airport in the country. In fact, it held this title for almost eight years straight in

the late 1990s and early 2000s (Pilcher, 2010). As will be discussed later, this was fueled

by the use of regional jets and an unsustainable regional hub structure operated by Delta

and its subsidiary, Comair. Third, where hubs are concerned, there is a long history of price

inﬂation at CVG because it was a fortress hub, with Delta controlling almost all of the gates

and its subsidiary (Comair) controlling virtually all of the spoke routes. Predatory pricing

and the lack of local carrier competition effectively provided Delta a monopoly on

operation at CVG. Even today, long after Delta's service reductions, CVG remains the most expensive airport in the country (McCartney, 2014).

3.2.2 CVG as A Hub

The emergence of CVG as a hub has as much to do with aircraft technology as it does with strategic planning. In 1977, a small regional carrier, Comair, opened for business at CVG, operating three turboprop airplanes. In 1986, Delta purchased 20% of Comair and began handling reservations and some of the marketing for the airline. By 2000, Comair

was processing over 5.3 million passengers (Bureau of Transportation Statistics [BTS],

2000) and by January 2001, it was acquired by Delta Airlines for $1.91 billion. This mega- acquisition effectively made CVG a hub for Delta's operations in the Midwest.6

The driving force behind the emergence of Comair as a major regional carrier was

the development of the Canadair Regional Jet. In the 1980s and early 1990s, regional

airlines, including Comair, relied on turboprop planes for servicing their short-haul route

networks. In addition to being noisy, the turboprop planes had low passenger capacity (n =

37), had a relatively short cruising range of 800 nautical miles and had a spotty safety

record (Simmons, 2000). Comair management understood the limitations of the turboprop

aircraft and believed that customers would be willing to pay a premium for ﬂying on a jet,

and that the higher price point for regional jet service would also make shorter routes more

proﬁtable (Pilcher, 2002).

In 1993, the Canadair Regional Jet 200 (CRJ-100/200) series aircraft were

introduced to the market by Comair. With a passenger capacity of 50 and a maximum range

of 1620 nautical miles for the ER model and 2003 nautical miles for the LR model, the

commercial market for regional jet travel in the U.S. was changed forever. Regional jet

travel on the new CRJ-100/200 aircraft proved to be so popular that during the late 1990s,

Comair was generating nearly $1 billion in revenue per year (Pilcher, 2002). Local

infrastructure investments also proved to be beneﬁcial to Comair and CVG. In 1994 Delta

spent $500 million to renovate Concourse A and construct Concourse B at CVG (Pilcher,

2010) (Figure 3.2). In addition, Delta built a new fueling facility and repair hangar for its

6 At the time, Delta's primary hub was ATL and its hub in the West was SLC.

ﬂeet (Pilcher, 2010). This also coincided with the construction of Concourse C ($50 million, funded by CVG), for providing gates to the Comair regional jets. Finally, CVG spent more than $350 million on two new runways, which was primarily ﬁnanced by federal grant monies, local tax bonds and ticket fees (Feiertag, 2002; Pilcher, 2010).

Figure 3.2: CVG airport layout

Unfortunately, the good (and proﬁtable) times for Delta and Comair at CVG did

not last for long. By October 2004, Delta was exploring options for restructuring under

Chapter 11 bankruptcy protection and by 2012, Comair was out of business and CVG was

undertaking a study to determine if demolishing Terminals 1 and 2 made ﬁnancial sense.5

What led to this? How could operations this innovative and proﬁtable decay so badly within

ten years? Moreover, how could Delta's hub at CVG simply wither away?

In the next section, this chapter explores the large constellation of factors that

contribute to the relative success or failure of an airport, all of which are dynamic in both

space and time. Although this is focused on CVG, the impacts of intra- and interregional

competition, market leakage and geographic context, to hub-and-spoke topologies, aircraft

ﬂeets, economies of scale and scope, market upheaval/disruption and dehubbing have the

potential to impact any airport. As mentioned earlier, the dehubbing of CVG may not prove

fatal, but the path to recovery is filled with challenges. Thus, CVG provides an important

template for deepening our understanding of what happens when markets are wounded by

factors within the dehubbing constellation, and the outlook for such airports to evolve

within a dynamic commercial travel environment.

3.3 Data and Methods

3.3.1 Air Trafﬁc and Airport Data

The Airline Origin and Destination Survey (DB1B) database is a 10% sample of airline tickets/itineraries from reporting carriers that are provided by the Bureau of

Transportation Statistics (Bureau of Transportation Statistics [BTS], 2013a) and is used

widely in the transport geography and economics literature (Goetz and Vowles, 2009;

Brueckner et al., 2013; Ryerson and Kim, 2013; Rodriguez-Deniz et al., 2013) (Table 3.1).

For the purposes of this paper, we limit our examination to the Market and Ticket databases.

The DB1BMarket dataset consists of records concerning air trafﬁc ﬂow attributes, origins,

destinations, number of passengers and a code that identiﬁes the number of steps required

to connect two airports (e.g., PHL:DEN:PDX). This feature allows analysts to track

origin/destination interactions across the system and provides interesting contextual data concerning passenger loads. For this analysis, only direct, nonstop ﬂights are aggregated and summarized because they are symbolic indicators of hub size and a relative measure of airport importance within the system (Grubesic and Zook, 2007; Grubesic and Matisziw,

2012). In conjunction with the Market dataset, we use the Ticket dataset to explore itinerary costs (based on ticket prices) for each route in the system. Data for 2002–2013 were obtained from the Market and Ticket datasets.7 In addition, commercial airport data were

obtained from the National Transportation Atlas (Bureau of Transportation Statistics [BTS],

2013b). This database includes the locations of all public airports in the United States.

Table 3.1: Airline Origin and Destination Survey (DB1B) Data Table Description Important Fields DB1BCoupon Providing coupon-specific information for Origin and destination airports, each domestic itinerary of the Origin and number of passengers, coupon type, Destination Survey fare class, and distance

DB1BMarket Containing directional market characteristics Origin and destination airports, of each domestic itinerary of the Origin and number of passengers, prorated Destination Survey market fare, and airport group DB1BTicket Containing summary characteristics of each Itinerary fare, number of passengers, domestic itinerary on the Origin and origin airports, and roundtrip Destination Survey indicator

7 DB1BMarket and DB1BTicket contained 248 million and 139 million rows during 2002–2013, respectively. Since the Market dataset contains more records concerning passenger volume and frequency, it is used for deﬁning route matrices.

3.3.2 Street Network Data

A national database of streets was obtained for analysis (ESRI, 2010a) and data for

Ohio, Indiana, Kentucky and Illinois were extracted. Speeds were assigned to each network segment using feature class codes. 8 For example, segments with a code of A10–A18

(primary roads with limited access or interstate highway) were assigned a speed of 65 miles per hour. In sum, 3,725,621 road segments were used for analysis and for generating airport catchment areas (to be detailed later).

3.3.3 Business data and ZIP code areas

Local and regional economic activity and the associated business environment is an important driver of air traffic volume. For example, some types of local economic activities and businesses, such as tourism, knowledge intensive industries and high- technology, rely on face-to-face communication, which in turn, stimulates air travel

(Debbage, 1999; Button and Taylor, 2000; Debbage and Delk, 2001; Storper and Venables,

2004; Agrawal et al., 2006; Alkaabi and Debbage, 2007; Dobruszkes et al., 2011). As detailed by Liu et al. (2006) and Grubesic and Wei (2013), there are four major industries that have a positive impact on business and commercial air travel. The four sectors are described by the Census Bureau (U.S. Census, 2007) as follows: 1) information technology

(IT), 2) finance, insurance and real estate (FIRE), 3) professional, scientific, technical services and management activities (PST) and 4) tourism (TOUR).

8 http://tinyurl.com/leblq6j

• IT – Represents North American Industrial Classification System (NAICS) sector

51 and includes publishing industries (e.g. software publishing and traditional

publishing), motion picture and sound recording industries, broadcasting industries,

telecommunications industries, Web search portals, data processing industries, and

information services industries.

• FIRE – Represents NAICS sectors 52 and 53 and includes establishments providing

financial transactions and/or facilitating financial transactions, real estate, rental

and leasing services.

• PST – Represents NAICS sectors 54 and 55 and includes establishments

performing professional, scientific, and technical activities, as well as

establishments administering, overseeing, and managing establishments of the

company/enterprise to achieve strategic and organizational goals.

• TOUR – Represents NAICS sectors 71 and 72 and includes establishments that

serve population groups depending on different cultural, entertainment, and

recreational needs. Establishments providing lodging and food services are also

included.

ZIP business pattern (ZBP) data were obtained for 2004-2012 (U.S. Census, 2012).9

These data contained counts of business establishments by ZIP code area, with each categorized by NAICS codes. Although these data do not provide an exact count of employees at the ZIP code level, the employment size class codes reflect a range of employees for each sector in each ZIP. Two measures were derived from the ZBP for

9 The ZBP and DB1B data are mismatched by two years (2002 and 2003), but there is sufficient coverage from the ZBP for identifying general spatiotemporal patterns for each catchment area.

44 exploring spatiotemporal trends in the local business environment for each airport catchment area:

• Employee Counts: Because employee counts are obfuscated within a range for each

establishment, the mean of each range is used to estimate employees for each sector.

• Establishment Counts: Because using aggregate establishment counts would make

small establishments equal in importance to large ones (with more employees),

establishments are classified in a three-tiered typology. Small establishments have

fewer than 100 employees, medium-sized establishments have between 101-499

employees and large establishments have 500 or more employees.

There are limitations associated with using ZIP code areas for this type of analysis

(Grubesic and Matisziw, 2006; Grubesic, 2008). Most notably, the use of ZIP code tabulation areas (ZCTAs) has proven to be especially problematic for spatial analysis.

Thus, this paper opts for the use of ZIP code boundaries derived from U.S. postal service routes (ESRI, 2014) rather than Census-based ZCTAs.

3.3.4 Exploratory Spatial Data Analysis (ESDA)

In an effort to deepen our understanding of the dehubbing process at CVG, several different exploratory data analysis (EDA) (Tukey, 1977) and exploratory spatial data analysis (ESDA) (Openshaw, 1991; Haining et al., 1998; Anselin, 1999) approaches are utilized. Some of the primary motivations for using EDA and ESDA include the identification of outliers, detecting patterns and trends, and refining hypotheses. Contrary to traditional hypothesis testing, which uses confirmatory statistical analysis such as regression, EDA and ESDA are typically done in a descriptive, rather than inferential

45 framework. Thus, much of the analysis in the next section includes basic tables and associated visualizations to detail the impacts of dehubbing to CVG.

One geocomputational approach used for analysis in this paper is the development of network-based, airport market catchment areas for all six focus airports (DAY, IND,

CVG, CMH, SDF and LEX). These catchment areas were based on the street network data detailed above. There is a long history and a variety of methods used for delineating airport geographic market areas. At the most basic level, previous work has used simple distance calculations (Lin, 1977; Kanafani and Abbas, 1987; Kaemmerle, 1991) to develop a measure of market and/or airport proximity. Other work has leveraged a combination of econometric methods (Cohas et al., 1995; Pels et al., 2001; Ishii et al., 2009) or consumer surveys (Fuellhart, 2007; Pantazis and Liefner, 2006). For the purposes of this paper, 90 minute drive time catchment areas were generated for each airport using the street database detailed previously. This represents a reasonable travel commitment for consumers seeking to access airport services (Kanafani and Abbas, 1987; Kaemmerle, 1991; Grubesic and Wei,

2013). Ultimately, the derived catchment areas were used to extract and summarize airport market data, including information pertaining to basic population patterns.

3.3.5 Least Absolute Shrinkage and Selection Operator (LASSO)

In order to better understand the relationship between the dehubbing process at

CVG, and air traffic attributes and local businesses at CVG during 2002-2013, regression analysis is utilized. The dependent variable for this analysis is a primary measure for dehubbing. Specifically, it is the yearly number of direct flight options between CVG and other domestic airports. It mainly reflects the airport activity. Since the business data were

collected during 2004-2012 with two types of estimations: employee counts and establishment counts, three regression models are developed in terms of different combinations of air traffic, airport and local business attributes: 1)the first model only includes air traffic and airport data covering entire period (2002-2013); 2)the second model examines air traffic and employee-based business attributes during 2004-2012; and 3)the third model measures air traffic and establishment-based business attributes during 2004-

2012. There are 12 observations in model 1 and 9 observations in models 2 and 3. The air traffic and business attributes are summarized as 22 independent variables in model 1, 26 independent variables in model 2 and 34 independent variables in model 3 (Appendix A10).

Since the number of variables is more than the number of observations, the ordinary

least squares (OLS) estimates are not appropriate considering prediction accuracy and

model interpretability (Tibshirani, 1996; James et al., 2013). Further, the unique least

squares coefficient estimate cannot be obtained because of infinite variance (James et al.,

2013). However, the variance can be reduced by sacrificing bias in order to improve

prediction accuracy. Some variables may be irrelevant to the research problem so including

these variables can result in unnecessary complexity and meaningless interpretation (James

et al., 2013). A solution is the coefficient estimates for some variables can be shrunk to low

values or set to 0 (ibid).

There are several techniques to improve OLS estimates including subset selection

and ridge regression (Tibshirani, 1996). However, both techniques have disadvantages.

Subset selection is a discrete process because it only focuses on a subset of variables each

10 Market shares of frequency counts are based on direct flights only.

iteration (James et al., 2013). It indicates that some relevant variables may be not included

so the model can be inaccurate. Ridge regression is a continuous process. It includes all

variables in a model and shrinks all coefficient estimates of variables towards 0 but not set

them to 0 so model interpretation can be a challenge according to a large number of

variables (James et al., 2013). Tibshirani (1996) proposes a new technique, least absolute

shrinkage and selection operator (lasso), that can shrink some coefficient estimates and set

others to 0 to keep advantages of both subset selection and ridge regression. Considering

the advantages of lasso, prediction accuracy and model interpretability, it is used to explore

the relationship between airport dehubbing and air traffic and other associated attributes.

In particular, lasso minimizes the value of

𝑛𝑛 ( 𝑝𝑝 ) + 𝑝𝑝 (3.1) 2 � 𝑦𝑦𝑖𝑖 − 𝛽𝛽0 − � 𝛽𝛽𝑗𝑗𝑥𝑥𝑖𝑖𝑖𝑖 𝜆𝜆 ��𝛽𝛽𝑗𝑗� 𝑖𝑖=1 𝑗𝑗=1 𝑗𝑗=1 The first component is the residual sum of squares (RSS) referring to the concept in a

multivariate OLS model. and are observed values of th dependent variable (Y) and

𝑖𝑖 𝑖𝑖𝑖𝑖 th independent variable (X)𝑦𝑦 for𝑥𝑥 the th Y, respectively. 𝑖𝑖 is the estimated value of the

0 𝑗𝑗constant term and is the least squares𝑖𝑖 coefficient estimates𝛽𝛽 in a standard multiple linear

𝑗𝑗 regression model. The𝛽𝛽 coefficient estimate for a certain independent variable indicates the

amount of change in the dependent variable for a change in that independent variable. A

positive coefficient means that the associated independent variable and dependent variable

have the same direction of amount change. If an independent variable x with a positive

coefficient increases, the amount of dependent variable y increases as well. The second

component is lasso penalty which prevents model overfitting and inaccurate prediction.

is a turning parameter to control the impacts of two components on the estimated values of𝜆𝜆

coefficients (James et al., 2013). When is 0, the penalty component does not have impact

on coefficient estimates and only RSS has𝜆𝜆 effect; when increases, the growing impact of

lasso penalty becomes significant so some lasso regression𝜆𝜆 coefficient estimates will shrink

and others will set to 0 in order to minimize formula (1) (ibid). The value of is determined

by selecting the smallest cross-validation error in order to extract the 𝜆𝜆best subset of variables with non-zero coefficient estimates. Cross-validation is an important method in statistical analysis to estimate the test error generated from another statistical method and further assess its performance (James et al., 2013). In this paper, lasso regression is implemented by glmnet algorithm (Friedman et al., 2010).

3.4 Results

3.4.1 Airport Market Structures

Before we delve into the dehubbing process at CVG, it is worth providing a quick overview of the local market context for CVG and its competing airports. Figure 3.3 displays the derived catchment areas and their associated overlap for all six airports and

Table 3.2 details the associated population distribution for these market areas in 2010.

There are a number of characteristics worth noting. First, where population is concerned, if CVG existed in isolation, it would uniquely serve over 3.75 million people in the Tri-

State area (Ohio, Kentucky and Indiana). However, as detailed in Figure 3.3, interregional

market overlap is signiﬁcant. When the catchment areas from nearby airports are

considered simultaneously, the unique population served by CVG is massively reduced, to

204,400 (− 94.56%). In other words, the 90 min catchment area for CVG shares 94.56%

of its population with the combined 90 minute catchment areas of DAY, CMH, IND, SDF

49 and LEX. The largest single overlap exists between CVG and DAY, which share 45.52% of the population in their respective catchment areas.

Figure 3.3: 90 minute catchment areas: IND, DAY, CMH, CVG, SDF, LEX

Table 3.2: Local market structure and catchment overlap Market(s) Total Population Total Shared Percent of Shared Population Population CVG 3,754,214 3,549,814 94.56% CVG and CMH 6,443,241 837,907 13.00% CVG and DAY 6,089,956 2,772,385 45.52% CVG and IND 6,583,518 56,092 0.85% CVG and SDF 5,339,027 644,528 12.07% CVG and LEX 5,161,038 945,709 18.32%

This is an important ﬁnding for several reasons. First, recall that regional airports

are heavily reliant on their local market for service demand. Unlike large hubs, such as

Chicago O'Hare (ORD) or LAX, which process huge numbers of passengers via connecting

flights, regional airport operations do not benefit from this connecting traffic or the

operational infrastructure needed to accommodate it. In short, catchment areas matter

because this is the geographic market from which small and medium-sized airports generate enplanements. Second, in regions where overlap is heavy and interregional competition is ﬁerce, market leakage can be problematic. Given the signiﬁcant overlap between the CVG catchment area and its neighboring airports, this is a huge concern.

3.4.2 The Dehubbing of CVG

There are a variety of factors that ultimately led to the dehubbing of CVG. In many instances, the individual factors within the constellation comingled, dramatically amplifying and/or accelerating the dehubbing process. Further, although some of these factors were unique to CVG, others were common to the entire airline industry. In this

subsection, each factor is detailed individually, but special attention is paid to instances

where factors such as pricing, competition, regional market structure, etc., are comingled.

Pricing

As detailed earlier in this paper, it is impossible to understand average itinerary

pricing at CVG without a close inspection of carrier market share over time. Table 3.3

highlights CVG's top three carriers, ranked by passenger enplanements, between 2002 and

2013. For clarity, Comair and Delta passengers are parsed individually. Collectively,

between 2002 and 2005, the peak of the Delta/Comair dominance at CVG, 85% or more

of CVG passengers were processed by these airlines. This is a textbook proﬁle of passenger enplanements at a fortress hub, and this monopolistic structure continued at CVG until

2012, the year that Comair went out of business (Table 3.3).

Table 3.3: Top three carriers at CVG based on the number of passengers Year #1 #2 #3 Carrier Number of Frequency Carrier Number of Frequency Carrier Number of Frequency Passengers Counts Passengers Counts Passengers Counts 2002 Delta 99,834 (63.30%) 56,403 Comair11 40,975 (25.98%) 29,533 ExpressJet 3,598 (2.28%) 1,887 (58.29%) (30.52%) (1.95%) 2003 Delta 89,621 (58.58%) 51,553 Comair 45,722 (29.89%) 32,557 ExpressJet 3,486 (2.28%) 1,830 (53.99%) (34.09%) (1.92%) 2004 Delta 100,643 52,579 Comair 50,687 (29.45%) 34,712 ExpressJet 4,262 (2.48%) 1,762 (58.48%) (52.90%) (34.92%) (1.77%) 2005 Delta 107,263 51,757 Comair 75,156 (35.06%) 47,847 Atlantic 9,663 (4.51%) 7,841 (50.04%) (43.61%) (40.32%) Southeast12 (6.61%) 2006 Delta 67,491 (42.04%) 31,394 Comair 66,055 (41.15%) 39,629 Envoy13 6,245 (3.89%) 2,674 (36.52%) (46.10%) (3.11%) 2007 Delta 54,133 (40.31%) 25,136 Comair 50,919 (37.91%) 31,313 Envoy 7,032 (5.24%) 3,206 (33.94%) (42.28%) (4.33%) 2008 Delta 45,145 (37.03%) 21,952 Comair 43,121 (35.37%) 27,539 Envoy 6,305 (5.17%) 3,295 (30.59%) (38.37%) (4.59%) 2009 Delta 51,861 (37.66%) 20,884 Comair 38,131 (27.69%) 20,778 Freedom 13,249 (9.62%) 8,774 (28.90%) (28.75%) (12.14%) 2010 Delta 57,264 (43.38%) 25,384 Comair 31,383 (23.77%) 22,169 Freedom 6,449 (4.89%) 4,986 (32.31%) (28.22%) (6.35%) 2011 Delta 56,433 (45.82%) 26,132 Comair 31,848 (25.86%) 23,125 Endeavor14 6,247 (5.07%) 5,337 (34.69%) (30.70%) (7.09%) 2012 Delta 47,209 (43.97%) 21,409 Comair 15,468 (14.41%) 10,816 ExpressJet 13,439 (12.52%) 9,418 (32.86%) (16.60%) (14.45%) 2013 Delta 42,579 (39.56%) 19,148 Endeavor 25,011 (23.24%) 16,530 ExpressJet 13,308 (12.37%) 8,610 (30.96%) (26.72%) (13.92%)

11 Ceased operation in 2012. 12 Merged with ExpressJet in 2011. 13 Formerly American Eagle Airlines. 14 Formerly Pinnacle Airlines.

Fortress hubs are price-problematic for several reasons. The ﬁrst is that dominant

carriers at fortress hubs often engage in predatory pricing. In 2005, low cost carriers had a

U.S. market share of 31.58% and by 2008, it reached 33.58% (Ben Abda, 2010). By 2008,

95 of the top 200 airports in the country had an aggregated LCC market share greater than

20% (Ben Abda, 2010). This was not the case at CVG, where the combined market share

for LCCs during 2008 was 2.91% and was much lower in previous years.15 The key factor

in LCCs failing to establish a lasting presence at CVG was that Delta forced the

competition out through predatory pricing. For example, Vanguard Airlines attempted to

enter the CVG market twice, once in 1996 and once in 2000. AirTran maintained a presence

at CVG from 1995 until 1998. In both cases, Delta aggressively responded to these LCC

entries by lowering fares and adding ﬂight frequency to the competing destinations

(Williams, 2013). Worse, as recently as 2013, when Frontier Airlines announced their entry

into the CVG market, Delta responded by cutting prices to all competing routes served by

Frontier, including CVG-SFO, CVG-LAS, and CVG-SEA (Williams, 2013). This lack of

sustained competition has, without question, spawned a long-term monopoly at CVG for

Delta. As a result, ticket prices at CVG have remained high for decades (Goetz, 2002;

Pilcher, 2003; Grubesic and Zook, 2007; McCartney, 2014).

15 Pilcher (2003) notes that ten LCCs attempted to establish a presence at CVG between 1993 and 2003, all of them failing. For a complete list of ICAO deﬁned LCCs, see http:// tinyurl.com/olc8seb.

Figure 3.4a: Average airport itinerary fares, CVG, CMH, DAY, IND, SDF and LEX. Figure 3.4b: Average itinerary fares to DFW. Figure 3.4c: Average itinerary fares to MSP.

Figure 3.4a highlights average itinerary pricing for CVG relative to its ﬁve proximal

competing airports (IND, DAY, CMH, SDF, LEX) from 2002 to 2013. During all but two

years, CVG was the most expensive airport in the region. Lexington was the most

expensive airport in the region during 2005 and 2009, but it is important to note that LEX

has scale, competition and pricing problems of its own (Hewlett, 2012). One of the more

intriguing features of Figure 3.4a is the average price differential for itineraries between

CVG and its neighboring airports. For example, CVG fares averaged $398 in 2002, while

the second most expensive airport, LEX, averaged $330, yielding a gap of $68. This gap

increased to $91 in 2003 and $100 in 2013. To be clear, these average fare differences vary

between routes (Figure 3.4b–3.4c), but in aggregate, CVG is frequently the most expensive

option in the region.16

The second reason that predatory pricing and fortress hubs are a problem is that

consumer response to high prices, especially at CVG, is market leakage to neighboring

airports. In a study commissioned by CVG in 2003, three interesting results emerged about

market leakage. First, only 72% of travelers in the Cincinnati market region used CVG

(Pilcher, 2003). At the time, Delta disputed this ﬁgure, arguing that the number was

potentially higher, and also suggested that some travelers from outside the region use CVG

because of its nonstop ﬂight options, offsetting a portion of the market leakage (Pilcher,

2003). Second, the report also estimated that Dayton (DAY) collected 12% of CVG's local

market. This situation was (and remains) particularly bad for CVG because the many of

16 Sensitivity analysis was conducted using a measure of fares/distance between all six airports. There is virtually no divergence between this derived metric and the average fares in absolute terms. CVG is often the most expensive, regardless of small differences in average stage length.

Cincinnati's northern suburbs are relatively afﬂuent (e.g. West Chester, Mason) and their location is almost equidistant to DAY and CVG. This means that many passengers with enough disposable income for leisure travel are likely drawn to DAY because of its lower fares. This is especially true when geographic market overlap (Table 3.2; Figure 3.3) and pricing are considered simultaneously. Third and ﬁnally, pricing asymmetries between

DAY and CVG were (and continue to be) pronounced. Grubesic and Zook (2007) noted the presence of multiple itineraries from CVG that were priced signiﬁcantly higher than

DAY. For example, August 2004, a roundtrip itinerary from DAY to Los Angeles on Delta airlines required a stop at CVG and was priced at $204. The same exact ﬂight (a continuation of the DAY journey), with a departure from CVG was priced at $323. This yielded a $119 price differential for the identical destination, via the same carrier. Is it any wonder that DAY continues to capture leakage from the CVG market? DAY is not alone, passengers from the CVG market routinely travel to nearby airports (e.g. SDF, CMH) to access lower fares (Williams, 2015).

Industry Upheaval

A second factor that contributed to the dehubbing of CVG relates to industry upheaval. This is a generic term used in this paper to refer to changing levels of travel demand, increasing fuel costs, systemic inefﬁciencies, the emergence of LCCs and many other minor factors. As noted earlier, some of these factors were common to the entire industry and all airports, while a handful impacted CVG only.

Consider, for example, the crippling Comair pilot strike in 2001 that lasted 89 days.

The strike grounded all of Comair's regional jets (~ 80) and effectively shut down a huge portion of the CVG market and portions of its feeder spokes (Leondardt, 2001). The strike

ended in June and Comair reinstated a limited level of service by July 2nd. Unfortunately,

three months later, the attacks of September 11th grounded all air travel in the United States

for several days, sending the industry (and consumers) in a tailspin that lasted for several

years. Ito and Lee (2005) note that September 11th resulted in a negative transitory shock

of over 30% and an ongoing demand shock of roughly 7.4%, that could not be explained

by economic or seasonal factors. This decline in demand lasted through most of 2002 and

2003.

During this time period, the regional jet market was undergoing several signiﬁcant

changes. In particular, the fuel inefﬁciency of CRJ and ERJ regional jets was becoming a

problem (Babikian et al., 2002). Specifically, regional jets are 10–60% less fuel efficient than turboprops and 40–60% less fuel efficient than larger, narrow- and wide-body

counterparts (e.g. Boeing 737s) (Babikian et al., 2002). During Comair's emergence as a

regional powerhouse (1993–1998), crude oil prices were below $20 a barrel. As detailed

by Figure 3.5, prices for crude began a relatively rapid ascent in 2001 and did not stop until

the crude market crashed in 2008.

This had signiﬁcant implications for CVG and Comair's ﬂeet of CRJs. Although

Bhadra (2003) notes that shorter distance air travel tends to be more fare-inelastic when

compared to longer trips, this was not the case at CVG. Travelers that were already fatigued

by CVG's premium prices were not willing to pay more as fuel costs increased (Pilcher,

2010). This put Comair in a dangerous position. They could not increase fares, but higher

fuel costs were eating up Comair's margins when serving routes with antiquated and fuel- inefﬁcient CRJs. When those planes/routes stopped generating revenue, there was no

money to purchase new equipment (Pilcher, 2010) and the Delta hub began to unravel at

CVG.

Figure 3.5: U.S. crude oil first purchase price, January 2000 – December 2013

Carrier Consolidation

In an effort to avoid bankruptcy, Delta began an aggressive restructuring of its

operations in 2004 and 2005. This included the August 2005 sale of Atlantic Southeast

Airlines (ASA) to SkyWest Airlines for $425 million (Delta Airlines, 2005a). One month

later in September 2005, Delta announced plans to “right-size” CVG as a hub within the

Delta system. This included a 26% reduction in ﬂights from CVG, effective December

2005 (Delta Airlines, 2005b). Operationally, this reduction meant that Delta redeployed many of the aircraft based at CVG to Atlanta and Salt Lake City. It also meant that nine of

59 the existing regional routes serviced by Comair (and ASA) were moved from CVG to

Atlanta (Delta Airlines, 2005b). One the heels of this reduction, Delta entered Chapter 11 bankruptcy and restructured. This included the layoff of several thousand employees (Delta

Airlines, 2005c), fending off a hostile takeover by U.S. Airways, and eventually surfacing from bankruptcy in 2007, only to merge with Northwest Airlines in 2008 (Associated Press

[AP], 2008). This merger had signiﬁcant implications for CVG. Northwest already had two large hubs in the Midwest (Detroit [DTW] and Minneapolis [MSP]). Northwest also had a smaller hub in Memphis (MEM).

Delta could not dehub MSP. In 1992, when Northwest Airlines was struggling, the

Metropolitan Airports Commission (MAC) reached a loan agreement with the airline.

MAC would provide Northwest with $245 million, but Northwest was required to keep a minimum of 10,000 employees in the State of Minnesota for 30 years (Lee, 2011). Because of this agreement, dehubbing MSP or moving related Northwest Airlines jobs to another

Delta hub would be expensive. Further, MSP was proﬁtable and there were no strong incentives to move the jobs immediately upon carrier consolidation (Lee, 2011). Similarly, the Northwest hub at DTW had just undergone a $3 billion renovation in 2001, including the opening of a fourth parallel runway and a $1.2 billion renovation of McNamara

Terminal with 97 gates (Schmeltzer, 2001). The North Terminal at DTW was also newly renovated and slated to open, adding 26 additional gates in late 2008 (Irwin, 2008). Again,

DTW was proﬁtable, it was growing, and there was no chance that Delta was going to close it.

This left CVG and MEM for Delta to deal with. In short, although it took some time, Delta completely dehubbed MEM after its merger with Northwest Airlines. At its

60 peak (circa 2009), MEM offered 240 flights per day, including an international connection to Amsterdam (Mutzabaugh, 2013). Northwest accounted for about 200 of these flights, and as recently as 2012, the merged Delta/Northwest provided 150 daily flights out of

Memphis (Ashby, 2013). The 2013 dehubbing reduced daily ﬂights from MEM, in aggregate, to approximately 60, with Delta providing about 50% of these. Delta also cut

230 customer service and cargo positions at the airport (Ashby, 2013).

Delta's dehubbing of CVG, although not official, left a similar footprint of reductions. At its peak in 2005, CVG offered more than 600 flights per day to 148 unique destinations via direct, nonstop flights. By 2013, this was reduced to 77 destinations with only a single, nonstop, international flight (Paris) (Table 3.4). Figure 3.6 highlights thins decline, geographically. Connections with dashed lines indicate that these itineraries were operated in 2002 but not in 2013. Perhaps most notable is the reduction of destinations in the Upper Midwest (e.g., Michigan, Wisconsin) and the Southeast (e.g., Alabama and the

Florida Panhandle). These destinations are now serviced through DTW and ATL, respectively. Table 3.4 also displays a dramatic increase in CVG's reliance on larger hubs and its abandonment of primary non-hubs, on a percentage basis. In 2002, CVG provided service to 30 large hubs, which accounted for 20.83% of its nonstop ﬂights. By 2013, CVG

flights to large hubs (n = 29) accounted for 37.66% of its nonstop routes. Further, CVG only provided service to 4 primary non-hubs in 2013, down 90% from its service footprint in 2005. Finally, the average miles flown from CVG has increased nearly 9% between 2002 and 2013 (Table 3.4). This is the classic profile of a dehubbed airport. The spokes are gone

(i.e. small- and primary non- hubs), drastically reducing the overall number of destinations.

Table 3.4: Summary statistics of direct flight options from CVG Year Number of Large Medium Small Primary Average Average Miles Number of Sharing Destinations Destinations Hubs Hubs Hubs Non-Hub Miles Flown Flown (48 states) CMH DAY IND SDF LEX 2002 144 30 32 49 33 685.62 682.75 85 58 83 66 29 (20.83%) (22.22%) (34.03%) (22.92%) (59%) (40%) (58%) (46%) (20%) 2003 140 32 29 43 35 674.1 667.96 84 50 77 65 21 (22.86%) (20.71%) (30.71%) (25.00%) (60%) (36%) (55%) (46%) (15%) 2004 144 29 30 48 36 710.23 684.78 86 60 84 68 29 (20.14%) (20.83%) (33.33%) (25.00%) (60%) (42%) (58%) (47%) (20%) 2005 148 29 30 47 41 723.38 702.33 83 54 84 67 34 (19.59%) (20.27%) (31.76%) (27.70%) (56%) (36%) (57%) (45%) (23%) 2006 140 29 30 47 33 761.73 730.83 88 62 81 70 34 (20.71%) (21.43%) (33.57%) (23.57%) (63%) (44%) (58%) (50%) (24%) 2007 130 28 28 42 29 763.89 743.85 74 48 78 70 32 (21.54%) (21.54%) (32.31%) (22.31%) (57%) (37%) (60%) (54%) (25%) 2008 127 27 25 42 32 732.26 728.14 73 52 79 67 32 (21.26%) (19.69%) (33.07%) (25.20%) (57%) (41%) (62%) (53%) (25%) 2009 115 27 26 40 22 737.85 737.85 71 61 72 69 33 (23.48%) (22.61%) (34.78%) (19.13%) (62%) (53%) (63%) (60%) (29%) 2010 106 28 25 35 18 736.02 736.02 71 47 76 71 33 (26.42%) (23.58%) (33.02%) (16.98%) (67%) (44%) (72%) (67%) (31%) 2011 86 27 21 29 9 746.33 746.25 52 36 60 54 17 (31.40%) (24.42%) (33.72%) (10.47%) (60%) (42%) (70%) (63%) (20%) 2012 80 27 16 27 10 727.24 727.24 52 40 59 53 15 (33.75%) (20.00%) (33.75%) (12.50%) (65%) (50%) (74%) (66%) (19%) 2013 77 29 17 27 4 (5.19%) 746.3 746.31 56 36 57 57 12 (37.66%) (22.08%) (35.06%) (73%) (47%) (74%) (74%) (16%)

Figure 3.6: Direct, nonstop flights from CVG, 2002 and 2013

There is an increased reliance on routes to/from large hubs and the average miles ﬂown

have increased. In effect, the transformation of CVG from a large hub, to a medium/small

airport is almost complete. One of the most important impacts of dehubbing process at

CVG on airport access and accessibility is significantly decreasing direct flight options to

small airports from CVG.

More damning for CVG was the slow erosion of unique routes served relative to

the competition between 2002 and 2013 (Table 3.4). For example, during 2002, CVG served 144 destinations with direct ﬂights. CMH identically covered 85 (59%) of those

ﬂights, IND 83 (58%), etc. By 2013, CMH increased its shared destinations/identical route

63 coverage to 73% and IND to 74%. Lexington is the only airport to see a decline in identical routes covered with CVG.

Is it any surprise that much of this erosion and associated route competition is due to the emergence of Southwest Airlines as a carrier in the region? Southwest entered the

Columbus, Indianapolis and Louisville markets in 2002. At this time, Southwest immediately began serving a handful (20–30%) of the destinations already shared by CMH,

IND and SDF with CVG (Table 3.5). By 2013, of the 56 routes identically shared by CVG and CMH, Southwest had a presence in 87.5% of them. In Indianapolis, it was 80.7% and

Louisville was 57.89%. In short, Southwest absolutely pummeled the CVG market via interregional competition.

Table 3.5: Shared CVG destinations reached by Southwest Airlines from competitor airports Year CMH DAY IND SDF 2002 25 (29.41%) ― 17 (20.48%) 20 (30.30%) 2003 24 (28.57%) ― 17 (22.08%) 19 (29.23%) 2004 26 (30.23%) ― 19 (22.62%) 17 (25.00%) 2005 23 (27.71%) ― 19 (22.62%) 16 (23.88%) 2006 24 (27.27%) ― 23 (28.40%) 19 (27.14%) 2007 31 (41.89%) ― 26 (33.33%) 24 (34.29%) 2008 40 (54.79%) ― 33 (41.77%) 27 (40.30%) 2009 42 (59.15%) ― 36 (50.00%) 33 (47.83%) 2010 44 (61.97%) ― 42 (55.26%) 33 (46.48%) 2011 47 (90.38%) ― 44 (73.33%) 33 (61.11%) 2012 50 (96.15%) 1 (2.50%) 46 (77.97%) 36 (67.92%) 2013 49 (87.50%) 5 (13.89%) 46 (80.70%) 33 (57.89%)

There was a similarly disheartening trend for CVG with respect to passenger counts within the region (Table 3.6). Between 2002 and 2013, every competing airport near

Cincinnati had witnessed growth in the number of passengers processed. LEX grew the

64 most (263.33%), but it remains a small airport and only processes a fraction of the passenger trafﬁc that airports such as CMH and SDF handle. That said, even relatively large airports such as IND displayed growth (+ 10.61%). Conversely, CVG declined (−

29.93%), a drastic reduction from 2002 and this decline would be even larger if the trend between 2005 and 2013 was calculated. More problematic is the reduction in regional market share for passengers processed at CVG. At its peak, CVG handled 32.13% of all passengers ﬂying to/from the region. As of 2013, it only processes 20.09% of regional passengers (Table 3.6). Again, it is the only airport in the region that lost market share over the study period.

Table 3.6: Total number of passengers and associated market share of six airports (direct flights only) Year Cincinnati Columbus Dayton Indianapolis Louisville Lexingto (CVG) (CMH) (DAY) (IND) (SDF) n (LEX) 2002 157,709 145,114 30,624 146,153 65514 4,926 (28.67%) (26.38%) (5.57%) (26.57%) (11.91%) (0.90%) 2003 152,989 120,067 34,327 161,558 60534 7,460 (28.49%) (22.36%) (6.39%) (30.09%) (11.27%) (1.39%) 2004 172,085 119,245 42,446 188,436 63753 8,771 (28.93%) (20.05%) (7.14%) (31.68%) (10.72%) (1.47%) 2005 214,339 134,389 40,958 194,136 72577 10,788 (32.13%) (20.14%) (6.14%) (29.10%) (10.88%) (1.62%) 2006 160,521 150,385 47,449 175,601 69433 9,844 (26.18%) (24.52%) (7.74%) (28.64%) (11.32%) (1.61%) 2007 134,305 178,452 48,249 174,291 67984 8,377 (21.96%) (29.18%) (7.89%) (28.49%) (11.11%) (1.37%) 2008 121,904 147,294 50,525 189,051 70188 8,851 (20.74%) (25.06%) (8.60%) (32.16%) (11.94%) (1.51%) 2009 137,725 142,000 42,127 174,378 58070 9,191 (24.44%) (25.20%) (7.48%) (30.95%) (10.31%) (1.63%) 2010 132,010 144,598 40,709 173,873 58387 13,322 (23.45%) (25.69%) (7.23%) (30.89%) (10.37%) (2.37%) 2011 123,168 146,968 40,873 165,923 56547 15,864 (22.42%) (26.75%) (7.44%) (30.20%) (10.29%) (2.89%) 2012 107,375 147,022 42,761 154,908 59508 15,387 (20.38%) (27.90%) (8.11%) (29.40%) (11.29%) (2.92%) 2013 107,620 146,294 40,425 157,474 66470 17,508 (20.09%) (27.30%) (7.54%) (29.39%) (12.41%) (3.27%) Trend -29.93% 3.49% 35.37% 10.61% 4.20% 263.33% (2000:2013)

Regression Models

Table 3.7 presents the variables with non-zero coefficient estimates and summary

statistics for three lasso regression models. Three models include four, six and six variables

with non-zero coefficient estimates, respectively. There are two things worth mentioning.

First, the four variables about direct flight options to/from CVG have a positive relationship

with the dehubbing process at CVG through all three models. As discussed earlier,

exploratory data analysis identifies that the decreasing number of direct flight options from

CVG primarily impacts small airports especially primary non-hubs, on a percentage basis.

In addition, the number of direct flight options connecting medium or small hubs also decreased significantly. These interesting findings are confirmed by LASSO regression analysis. That said, the number of direct flight options to non-large hubs can be considered as a parameter to indicate dehubbing process at a certain airport. Second, market share of frequency counts, market share of passengers and itinerary fares do not have a clear relationship with dehubbing process through all three models. It only shows that market share of passengers at LEX and itinerary fare at SDF have negative relationships with the dependent variable in Model 2. The three types of variables are complicated because they are not only affected by airport competition but entire air transport market and some external factors such as financial crisis and increasing fuel cost. That said, dehubbing can be reflected by changes of frequency counts, passenger counts and itinerary fares but it cannot be analytically measured by the three types of variables because they can be affected by other factors not dehubbing. In addition, only two business variables, number of estimated medium establishments in IT and large establishments in TOUR sector, are highlighted in model 3 and both negatively influence dependent variable. It is surprising

66 because there are less direct flight options when more medium establishment in IT and large establishments in TOUR sector exist in CVG market. There are two potential reasons.

Frist, the two variables had quite stable or opposite trend with the dehubbing at CVG.

Second, the impacts of business on airport activity may not be accurately identified as the numbers of and employees and establishments are estimated.

Table 3.7: Lasso regression model results Model 1 Model 2 (Employee-based) Model 3 (Establishment-based) Variables with non-zero Coefficient Variables with non-zero Coefficient Variables with non-zero Coefficient coefficient estimates coefficient estimates coefficient estimates INTERCEPT 14.794 INTERCEPT 16.405 INTERCEPT 33.595 LARGE_HUBS 0.586 LEX_PASSENGER -34.185 CVG_M_IT -0.077 MEDIUM_HUBS 1.084 SDF_FARE -3.438e-05 CVG_L_TOUR -0.046 SMALL_HUBS 0.837 LARGE_HUBS 0.523 LARGE_HUBS 0.143 PRIMARY_NON_HUBS 1.077 MEDIUM_HUBS 1.064 MEDIUM_HUBS 1.02 SMALL_HUBS 0.887 SMALL_HUBS 0.831 PRIMARY_NON_HUBS 1.046 PRIMARY_NON_HUBS 1.104 Lambda 0.664 Lambda 0.705 Lambda 0.705 Mean cross-validation error 4.056 Mean cross-validation error 3.62 Mean cross-validation error 3.771

3.5 Discussion and Conclusion

There are a number of points highlighted in this section that merit additional

discussion. First, as alluded to earlier, CVG's market region simply cannot support a large

hub. At best, Cincinnati is a middling market in a mid-sized metropolitan area that is not a

signiﬁcant tourist destination. Although the business community remains active, it too has

taken some damage from the reductions by Delta.17 During the late 1990s and early 2000s,

when Delta was aggressively protecting its market share via predatory pricing at CVG,

proximal airports such as IND, DAY, CMH and SDF were embracing LCCs. Again,

Southwest Airlines has a presence in all four of these cities. Further, even when CMH was

a mini-hub for America West Airlines, the airline never dominated CMH or its operational

proﬁle (American West Airlines [AWA], 2015). In fact, Southwest was active in the CMH

market when it was an America West hub. The resulting geographic ring of interregional

competition that surrounded CVG nurtured healthy regional markets, through competition.

Even in 1999, both the Louisville and Columbus airports were aggressively courting

passengers from the CVG market with “drive and ﬂy” campaigns (Dias, 1999) because of

their comparatively lower fares.18

Second, CVG continues to deal with the fortress hub hangover left by Delta. As

mentioned earlier, Delta continues to aggressively price routes served by LCCs in the

Cincinnati market. Delta also uses other means to keep LCCs away from CVG. WXIX

17 One notable example of the fallout associated with Delta dehubbing CVG is the loss of Chiquita Brands International, which moved to Charlotte, NC. Chiquita cited the relative health of the US Airways hub in Charlotte, versus the declining health of the Delta hub in Cincinnati as a major factor in its decision to relocate (Mecklenborg, 2011, Williams, 2014a). 18 Dias (1999) notes that business travelers who ﬂew during the week from CVG to Los Angeles were paying about $1900 in 1999. The same ﬂight from SDF was only $400.

(2013), a local television station in Cincinnati recently obtained the portions of the signed

lease agreement between Delta and CVG. In the lease document, Delta requires that only

gates B6 and B9 may be used for international ﬂights at CVG. Further, for airlines that use

gate B6, the planes must be comparable in size to a 767 (or less). Similarly, carriers that

use Gate B9 must not use aircraft larger than a 747. The lease agreement also stipulates that carriers using those gates must provide CVG with proof, in the form of a letter of credit or bond, that it has access to enough money for “three (3) months' worth of anticipated rates and charges” (WXIX, 2013). Needless to say, such cash reserves are difﬁcult to obtain for most LCCs. Although it is highly unlikely that LCCs would be seeking to serve international routes at CVG, it is the language used in the leasing agreement that is emblematic of Delta's tactics in the Cincinnati market. Delta has been, and continues to be, relentlessly aggressive in keeping competition away. This is bad for the health of any airport, including CVG.

Third, the worst (potentially) is yet to come. Some experts believe that CVG will eventually ﬁnd an operational equilibrium of 35 to 40 nonstop destinations (Wetterich,

2014). In part, this is because Delta has not formally dehubbed CVG. Although it is impossible to predict the actions of any commercial carrier, there is very little incentive for

Delta to remain. As detailed previously, Delta already operates two extremely large hubs in the Upper Midwest (MSP and DTW). DTW is only 229 miles from CVG and Delta's superhub at ATL is only 373 miles south. In short, any type of hub presence in CVG is geographically redundant for Delta. The international market for CVG is also soft.

Although the ﬂight from CVG to Paris runs near capacity during the peak summer travel season, it is only 70–75% full during the winter months (Williams, 2014b). One way to

70 offset the costs for carriers to operate additional daily, nonstop international ﬂights from

CVG is to use smaller planes (Wetterich, 2014). Unfortunately, the signed lease agreement between CVG and Delta prohibits this from happening (WXIX, 2013). It is also important to note that the daily nonstop from CVG to mainland Europe is something that much larger and healthier airports in the region do not have (e.g., Cleveland, Columbus, Indianapolis,

St. Louis, Nashville). This is a red ﬂag and it certainly suggests that CVG may not be able to support this level of service in the future. Lastly, all major domestic carriers maintain a presence at CVG, including United, US Airways/American and Delta. This is certainly a good sign. However, it is unlikely that other mid-sized carriers such as Alaska, JetBlue or

Southwest would consider CVG for service. JetBlue already has a presence in Detroit,

Cleveland (new), Pittsburgh, Chicago, Raleigh/Durham and Charlotte, but there may be an opportunity for it to provide ﬂights between CVG and BOS. As detailed previously,

Southwest already maintains service in Columbus, Dayton, Indianapolis and Louisville.

Alaska Airlines is more focused on cities in the West, but they too maintain service in

Detroit, Chicago, St. Louis and Atlanta.

Fourth, and ﬁnally, the comingling of these forces has clearly devastated CVG and its market. But CVG is not alone. As detailed previously, Memphis, Pittsburgh, St. Louis and many other cities have been victimized by dehubbing in the past and many more cities will suffer this fate in the future. Thus, where transport strategy is concerned, there is a path forward for CVG, but it is a challenging one. Attracting low cost carriers to the airport is critical. There is no federal legislation coming to save the day for CVG, such as the

Essential Air Service program, small community air service development grants, or the like

(Grubesic and Wei, 2012; Wittman, 2014). Consider the post-dehubbing success that

Pittsburgh International (PIT) had after U.S. Airways cut service in 2005–2006. Southwest and JetBlue immediately entered the market, along with several other carriers, including an independent AirTran Airways (now part of Southwest). Southwest may not be a good

ﬁt for CVG, but both Frontier and Allegiant have established a presence and continue to carve out market share. Regardless, more competition is needed.

Operationally, CVG also needs to be savvier with leasing agreements and long- term contracts. Rather than mortgaging its future with exclusive gate agreements with carriers like Delta, CVG should be making entry for LCCs easier, more proﬁtable and attractive. There is an opportunity for this when a portion of Delta's contract expires on

December 31, 2015 (Williams, 2014c). The current contract, which was signed in 1974, basically allows Delta full control of gate leases, airport improvements, all construction projects and incentive packages to attract competing carriers. Delta has veto power for everything (Williams, 2014c). The new contract, which is expected to be ﬁve years in length, should allow CVG to create more equity in their treatment of carriers. This includes a reconﬁguration of how airports assess landing fees, terminal rentals, concessions and parking revenues (Williams, 2014c). Further, Delta's stranglehold on Concourse B and the ticketing counter in the main terminal expires in 2020, which will provide an immediate opportunity to refresh and renew CVG into a competitive, multicarrier airport.

The dehubbing of an airport is a crushing blow to the host community and its surrounding region. Recent estimates for suggest that Delta's ﬂight reductions at CVG caused the loss of 33,000 jobs and nearly $1 billion in annual economic activity for the

Cincinnati region (Williams, 2014a). As detailed in this paper, Cincinnati is not alone, many communities suffered through dehubbing in the 2000s, both in the United States and

abroad, as terrorist attacks, industry upheaval, carrier consolidation, rising fuel prices,

maintenance costs and economic recession wracked the transport markets. Not surprisingly,

airline market resilience to these factors varies considerably. At the same time,

improvements in aircraft technology, the economic fortunes of regions, shifts in travel

demand and its associated destinations, the competitive dynamics between airports

(O'Connor, 2003), airline network strategies (Suau-Sanchez et al., 2015) and the lingering inﬂuence of deregulation largely dictate the operational outcomes for gateways. This is true for Cincinnati, the Tri-State region (Ohio, Indiana, Kentucky), the United States, Europe,

Asia and elsewhere. As detailed by Suau-Sanchez et al. (2015, 14), although advances in technology have improved the opportunities for long-haul service at secondary or tertiary gateways, the viability of these longer routes are still contingent on high levels of business demand and cargo transport. Further, the demand for such services depend on the economic potential of each region and are subject to local path dependencies. Cincinnati is struggling in both respects, as businesses, jobs and human resources shift away from the Ohio Valley toward growing metropolitan areas such as Raleigh, North Carolina, Austin, Texas, and

Portland, Oregon, as well as retrenching in the traditional strongholds of New York,

Washington and San Francisco.

That said, few, if any regions and/or airports share the geographic conundrum facing Cincinnati. It is within 90 min of four highly competitive airports (IND, DAY, CMH,

SDF) and a ﬁfth, smaller airport (LEX) that offers convenient, albeit somewhat expensive service to larger hubs. The presence of Southwest Airlines in IND, DAY, CMH and SDF is also a problem for the CVG market, softening it substantially, for both business and leisure travel. Interregional market leakage weakens operations. As detailed throughout

73 this paper, interregional competition has a distinctly different composition (and quality) than the intraregional competition found in large metropolitan areas with multiple airports

(e.g. Los Angeles, Chicago, etc.). Finally, Delta's historic stranglehold on CVG did the airport and the region no favors.

In sum, there is a silver lining for CVG. Portions of Delta's contract are expiring in

2015 and the remaining bits are set to expire in 2020. This provides CVG with an opportunity to rectify the awkward marriage between it and Delta. Also, a handful of LCCs have arrived and the community is embracing them. Clearly, there are many lessons to be learned from CVG. At the very least, the constellation of factors and associated process that contributed to the decline of CVG can serve as a case study for markets that have not yet suffered through the dehubbing process, but may be vulnerable to it in the future.

Dehubbing does not have to be fatal, but the path for attracting a good mix of alternative carriers and providing the market with appropriate levels of service after dehubbing is not easily traversed. CVG remains a work in progress.

This chapter suggests that a combination of commercial carrier strategies, operational efficiencies, hub structures, network topologies and regional competition contributed to deterioration of CVG. More importantly, it shows how the dehubbing process impacts airport access and accessibility. However, it only examines one airport and nearby airports (competitors) at a regional level. Next chapter further examines airport access and accessibility but at a national level by focusing on 106 rural, isolated airports in

EAS program.

4. STUDY 2: EVALUATING GEOGRAPHIC MARKETS AT EAS AIRPORTS

Questions of access and accessibility are clearly at the heart of the EAS program and have been a remarkably strong undercurrent in the air transport literature for some time

(Matisziw and Grubesic, 2011). Different from the previous one, this chapter extends study to a group of EAS airports that have similar characteristics at a larger spatial scale – national level, to explore the middle ground between structuring optimal EAS airport systems (Flynn and Ratick, 1988; Grubesic et al., 2012) and their routing structures

(Matisziw et al., 2012) and work relating to smaller communities, enplanements, and issues of accessibility (Cunningham and Eckard, 1987; Reynolds-Feighan, 1995; Kaemmerle,

1991). Speciﬁcally, this chapter seeks to identify the demographic, socio-economic, geographic, and local business factors that inﬂuence variations in service levels to EAS communities.

4.1 Introduction

Essential Air Service (EAS) is a federally subsidized program that connects small and/or isolated communities in the United States (U.S.) to the national transportation network (Grubesic and Matisziw, 2011). Many communities located in rural areas have beneﬁted from the program. For example, during ﬁscal year 2010, 107 community-based airports (in the lower 48 states) participated in the program and over $163 million were allocated to carriers to provide service to these EAS airports. West Yellowstone, Montana, received the smallest subsidy ($427,757) and Decatur, Illinois, received the largest subsidy

($3,082,403) (U.S. OAA, 2010). In 2010, 664,006 passengers and 103,291 commercial

ﬂight operations occurred in EAS communities (USBTS, 2010). In aggregate, this means

that $245.49 was spent by the federal government on each EAS passenger. Couched

somewhat differently, the federal government allocated approximately $1,578.16 to

commercial carriers for each EAS-related ﬂight. Perhaps most interesting is the fact that,

on average, EAS ﬂights carried six passengers per ﬂight in 2010. Not surprisingly, EAS

has been criticized for many years because of its high cost and low service utilization

(Gillies, 2004). However, rather than simply writing off the EAS program as wasteful

based on anecdotal observation, it is important to establish an objective evaluation process

and framework for determining which demand-side factors in a local EAS market inﬂuence service levels. In particular, this paper focuses on the number of ﬂights provided for each

EAS community. Generally speaking, these flights will occur as scheduled, regardless of passenger loads. Therefore, while questions of low service utilization are interesting, developing an understanding of service provision levels (i.e. flights) is a critical first step for unraveling the complexities of the Essential Air Service Program. This type of empirical evidence can be used to make more informed policy decisions as to which airports should be considered for subsidy elimination, or require additional subsidies to stimulate passenger use. This chapter utilizes basic spatial analysis techniques for modeling

EAS airport catchment areas and their associated demographic, socio-economic, geographic, and business compositions to identify the local characteristics of EAS markets that contribute to variations in service levels.

4.2 Background

Interestingly, prior to deregulation, there were fears of ‘‘cream skimming,’’ where private carriers would focus their infrastructure and service offerings in a select set of large

and proﬁtable market pairs (e.g. New York to Los Angeles) (Flynn and Ratick, 1988).

These concerns helped spawn the Essential Air Service Program. Speciﬁcally, rather than

see smaller and less proﬁtable markets lose their connections to the national air transport

system, the EAS program was created to provide federal subsidies for carriers to maintain

service to rural and remote communities (Reynolds-Feighan, 1995; Grubesic and Matisziw,

2011). Initially slated for 10 years worth of funding, post-deregulation, the EAS program

has been funded continuously for the past 34 years. While controversial, Grubesic and

Matisziw (2011) note that efforts to eliminate the EAS program have been rebuffed. There

are simply too many special interest groups (e.g. National Grange) and political leaders

that view these federal subsidies as critical to the economic vitality (and social fabric) of

their rural constituencies. These special interests are also echoed in much of the early

literature dealing with issues of accessibility and economic development in rural and

remote communities (Keeble et al., 1982; Keane, 1984; Johansson, 1993). Further,

although the allocation of subsidies totaling nearly $170 million to communities (FY 2011)

in the lower 48 states is substantial, when compared to the $6 billion in subsidies provided

by the federal government for ethanol in 2011 (Llanos, 2011), EAS monies represent a

relatively small component of the annual U.S. national budget.

4.3 Data and Methods

The empirical analysis for this chapter utilizes a combination of exploratory spatial

statistical approaches, proximity analysis, ordinary least squares, and spatial regression.

However, prior to introducing the modeling framework, details pertaining to the data used

in the analysis are outlined.

4.3.1 Business Data

As highlighted in Section 2, the local business environment can play a key role in

generating enplanements for commercial airports (Debbage, 1999; Button and Taylor,

2000; Debbage and Delk, 2001; Alkaabi and Debbage, 2007). For the purposes of this

chapter, four categories of businesses hypothesized to have a signiﬁcant and positive

inﬂuence on business travel are selected for inclusion: (1) ﬁnance, insurance, and real estate

(FIRE); (2) professional, scientiﬁc technical, and management services (PST); (3)

information technology (IT); and (4) tourism (TOUR). The ﬁrst three sectors, all of which

can be enhanced by face-to-face communication (Storper and Venables, 2004; Agrawal et

al., 2006; HBR, 2009), are known stimulants to business travel.19 The use of tourism-

related businesses and employment is necessary for evaluating Essential Air Service

because some (but not all) of the EAS communities are located near major tourist

destinations. For example, West Yellowstone, MT, serves as the western gateway to the

Yellowstone National Park and Crescent City, CA, serves as an access point to the

Redwood Coast of California. That said, although tourism can be a major determinant of

air passenger trafﬁc (Liu et al., 2006), many EAS communities are located in places where

tourism and its associated resources are sparse, at best (e.g. Kansas and Nebraska). In these

instances, resources might only amount to a few hotels along major interstate corridors.

Rather than using aggregate counts (e.g. by ZIP code or some other spatial unit),

business point data (ESRI, 2010b) were obtained for analysis. As mentioned in Chapter 3,

the speciﬁc breakdown of these business points, by North American Industry Classiﬁcation

19 Please note that the article from Harvard Business Review (HBR, 2009) was sponsored by British Airways.

System (NAICS) codes, is as follows: 1) information technology (IT), 2) finance, insurance

and real estate (FIRE), 3) professional, scientific, technical services and management

activities (PST) and 4) tourism (TOUR).

One last important point to make about the use of these data is that the statistical models constructed in this paper utilize employee counts in each sector (normalized by population), rather than firm counts associated with each community. Because most EAS communities are relatively rural and/or remote with smaller populations, they may be home only to one or two major employers. Thus, in an effort to avoid small number bias (i.e. firm count), the use of aggregate employee counts in each sector should ensure that the size of important local industries is accurately reflected for a community.20

4.3.2 Demographic, Spatial, and Airport Data

Demographic data for this study included total population for each EAS community

and average household median income for 2010 (ESRI, 2010c). The shortest network

distance between EAS communities and their nearest FAA-designated small, medium, or

large hub airport was also calculated (via TransCad) for use as an independent variable in

the models.

In sum, 106 of the 109 EAS airports were used for this analysis (U.S. OAA, 2010),

corresponding to all communities with at least one EAS subsidized ﬂight for 2010.21 Flight

20 Alternatives were also considered. For example, statistical results were also derived using location quotients for employee counts, by industry, in the regression models. There were no discernible differences in the modeled results between the LQs and the population normalized employment counts. 21 A subset of data were missing for Plattsburgh, NY, and we did not use the two airports located in Puerto Rico.

statistics (e.g. number of ﬂights and plane types) were gathered from the Research and

Innovative Technology Administration (RITA) (USBTS, 2010). Subsidy rates for each

carrier in each community as of May 1, 2010, were also obtained.22

4.3.3 Catchment Area Generation

A ﬁnal consideration for operationalizing these data in a regression-based

framework was the development of local EAS airport catchment areas. Although there are

many different approaches for delineating the geographic market areas of airports, ranging

from simple distance calculations (Lin, 1977; Kanafani and Abbas, 1987; Kaemmerle,

1991), to consumer surveys (Fuellhart, 2007; Pantazis and Liefner, 2006), and econometric

methods for evaluating consumer choice in multi-airport regions (Cohas et al., 1995; Pels et al., 2001; Ishii et al., 2009), this paper roughly follows the lead of Grubesic and Matisziw

(2011) and generates a 70-mile, network-based polygon around each EAS airport (Figure

4.1).

This catchment area represents an approximate drive-time of 1 h to each airport and also represents the programmatic distance threshold for EAS eligibility. In some cases,

EAS catchment areas overlapped. Therefore, to ensure statistical independence of the business data and employee counts in the regression models, businesses located within overlapping catchment areas were assigned to their nearest EAS community using a derived Voronoi tessellation (Grubesic and Zook, 2007).

22 Although the types of planes used for EAS service vary signiﬁcantly, all planes have at least nine, but no more than 50 seats.

Figure 4.1: 70 mile network catchment area

4.3.4 Regression Models

In an effort to deepen our understanding of the factors that contribute to EAS service levels, a series of ordinary least squares (OLSs) and spatial regression models for

EAS airport communities are developed. The dependent variable used for this analysis

reﬂects a combination of airport and demographic data. Speciﬁcally, a log-transformed

count of ﬂights per 1000 residents is used to model EAS viability. There is one caveat

worth noting regarding the use of this dependent variable. Because the EAS program

subsidizes carriers, not communities, the goal is to provide a minimum level of service to

each community, each week. Carriers must provide two daily round trips, 6 days a week,

with not more than one intermediate stop to the local hub airport (U.S. GAO, 2009).

Obviously, if demand levels exceed this threshold, carriers are free to increase the number

of ﬂights, although they will not be subsidized for them. However, if carriers fail to meet

the minimum threshold, government subsidies can be withheld (U.S. GAO, 2009).

Details on the variables used are presented in Table 4.1. The associated descriptive

statistics for each of the variables are found in Table 4.2. Finally, it is important to note

that the models presented in the next section were kept relatively simple, by design. In part,

because there is no internal competition within EAS communities for passengers (i.e. they

are served by a single airline), many of the usual supply-side considerations in modeling passenger enplanements at a typical commercial airport (e.g. fares) are not as relevant. This is not to say that competition does not exist and sensitivity to fares does not exist. As illustrated by Grubesic and Matisziw (2011), some EAS airports cannibalize each other, while others are impacted by their proximity to smaller hubs or nearby airports. However, because this paper considers EAS as a ‘‘closed system,’’ its goal is to isolate the effects of local market characteristics that contribute to EAS service levels, while controlling for the geographic effects potentially exerted by nearby airports.

Table 4.1: Dependent and Independent Variables Name Description LOG_FLIGHTS_POP Flights per 1,000 people in EAS catchment area (Dependent variable) SUBSIDY Federal subsidies, in dollars, for each carrier serving an EAS community POP Population within 70‐mile catchment for each EAS airport INCOME Average household median income within the 70‐mile catchment for each airport L_D Network distance from each EAS community to the nearest large hub airport M_D Network distance from each EAS community to the nearest medium hub airport S_D Network distance from each EAS community to the nearest small hub airport P_E Employed population in each catchment area P_IT Percent of employees in IT sector P_FIRE Percent of employees in FIRE sector P_PST Percent of employees in PST sector P_TOUR Percent of employees in TOUR sector WV Spatial regime variable for West Virginia KS Spatial regime variable for Kansas NM Spatial regime variable for New Mexico

Table 4.2 Descriptive Statistics Number Minimum Maximum Sum Mean Std. Skewness Std.

Deviation Error LOG_FLIGHTS_POP 106 -0.762 1.532071 42.694226 0.403 0.561 0.1 0.235 SUBSIDY 106 427757 3082403 1.61E+08 1507370 487199.1 0.536 0.234 POP 107 13104.6 10122629.66 89725542.78 838556.5 1367184 4.401 0.234 INCOME 106 24842.35 54504.25 3721380.93 34779.26 4944.946 1.13 0.234 L_D 106 71.1 771 31472.1 294.132 137.904 0.946 0.234 M_D 106 75.4 928 31568.9 295.036 204.474 1.602 0.234 S_D 106 32.2 541 19757.3 184.648 108.969 0.984 0.234 P_E 106 0.036 0.527 25.764 0.241 0.104 0.262 0.234 P_IT 106 0.003 0.050 2.086 0.019 0.008 1.166 0.234 P_FIRE 106 0.023 0.121 5.363 0.05 0.014 1.691 0.234 P_PST 106 0.008 0.095 3.773 0.035 0.016 1.409 0.234 P_TOUR 106 0.047 0.403 11.904 0.111 0.051 2.959 0.234 WV 106 0 1 5 0.05 0.212 4.357 0.234 KS 106 0 1 6 0.06 0.231 3.914 0.234 NM 106 0 1 4 0.04 0.191 4.947 0.234

4.4 Results

Figure 4.2 highlights the spatial distribution of EAS airports and their associated subsidy levels in the lower 48 states as of May, 2010. Of note is the complete absence of

EAS airports in many locations, including Florida, Indiana, Idaho, and the Carolinas. In other locations, such as Pennsylvania, Kansas, and Nebraska, there are multiple EAS airports in relatively close proximity. As detailed later, the spatial proximity of EAS airports in places like Kansas is an interesting feature of the system. Figure 4.3 provides some additional visual perspective on the locations of EAS airports and their nearest small, medium, or large hubs for the U.S. It also highlights the number of ﬂights associated with each EAS airport for 2010. Lebanon Regional Airport (LEB) had the most operations in

2010, with 2121, followed by Garden City, KS (1950), Marion, IL (1843), Quincy, IL

(1830), and Lancaster, PA (1753). The fewest operations were found in West Yellowstone,

MT (258), Ely, NV (291), Glendive, MT (498), Wolf Point, MT (502), and Havre, MT

(519).

Figure 4.2: Essential Air Service subsidies, May 2010

Figure 4.3: EAS airports and ﬂight totals for 2010

4.4.1 Regression

As noted in Sections 2 and 3, the demand-side determinants of EAS airport activity are multi-faceted. In a typical airport service region (i.e. catchment area or geographic

market), population and income are generally hypothesized to have a signiﬁcant and

positive impact on ﬂights. One would also hypothesize that a strong presence of employees

in the four economic sectors detailed previously (IT, FIRE, PST, and TOUR) will

positively inﬂuence airport activity. However, because EAS airports represent a special

case, where subsidies are required to keep commercial service active within each

community, the usual demand-side factors may be of reduced importance and at levels noticeably lower in EAS communities than their more metropolitan counterparts.

In an effort to reflect these special circumstances, we are also hypothesizing that as distance increases from EAS communities to their nearest small, medium, and large hubs, there will be a positive impact on the number of flights. Finally, we are also hypothesizing that there is a positive relationship between subsidy levels and flight activity for airports.

The results of the basic OLS regression analysis are interesting. Model 1 uses the log transformed count of ﬂights per 1000 residents in each EAS community as the dependent variable and all of the previously mentioned independent variables for analysis.

The model is significant (F = 39.30, p = 0.00) and explains approximately 76.4% of the variance (Table 4.3). However, only three of the independent variables are significant at the 95% level. Distance to medium hubs (M_D) and distance to small hubs (S_D) have a positive influence on EAS airport activity, while population (POP) has a negative influence.

Although none of the industrial composition/ employment variables are signiﬁcant, all but

FIRE retain a negative coefﬁcient value. The major problem with Model 1 is that the model

residuals display signiﬁcant and positive spatial autocorrelation (Table 4.3). 23 A

cartographic analysis of results derived from a test of local spatial association (not shown)

suggests that there are clusters of the airports where the model under-predicts ﬂights (e.g.

Kansas and Arizona) and clusters where the model over-predicts ﬂights (West Virginia).

In an attempt to control for the identified autocorrelation in the OLS model, a series of spatial regime variables (Anselin, 1992) are specified for airports in locations where the model displayed a systematic pattern of over- or under-prediction. In this instance, the definition for each spatial regime was relatively simple in that it corresponded to individual states. For example, all airports in the state of Kansas (KS) received a value of 1, while all other airports in the U.S. received a value of zero. A similar regime was also specified for the state of West Virginia (WV) and New Mexico (NM).24 When these variables were

added to the OLS regression model, several changes in model performance and viability

occurred. First, the standard error of the estimate decreased (Table 4.3) and the overall

explanatory power of the model increased nearly 5%. Distance from medium and small

airports remained signiﬁcant and positive, while population remained signiﬁcant and

negative. However, three additional variables also achieved signiﬁcance in Model 2,

including a positive coefﬁcient for subsidy levels (SUBSIDY) and mixed coefﬁcients for

the spatial regime variables. Speciﬁcally, while Kansas and New Mexico retained positive

coefﬁcients, West Virginia was negative. Perhaps most importantly, the level of

23 The weights matrix for the global and local tests for autocorrelation on the residuals is network distance between EAS airports.

24 Although not discussed or illustrated in this paper, a sensitivity analysis with spatial regime variables for nearly 25 states was conducted. The most relevant and interesting results are presented.

autocorrelation in the residuals decreased substantially, although the value for Moran’s I

remained signiﬁcant.

The last iteration of the conﬁrmatory tests was a spatially lagged regression model

(Anselin, 1988), Model 3. In addition to another increase in explanatory power, R2 = 0.84,

the standard error decreased, and spatial autocorrelation in the model is eliminated (Table

4.3).25 Of note is the relative importance of the spatial lag associated with EAS airports

and their ﬂights (Rho = .4891). Again, all of the industrial composition variables remain

insigniﬁcant and all retain negative coefﬁcient signs, while the distance variables and

subsidy levels remain positive and signiﬁcant.

25 The use of a spatial lag in Model 3 effectively eliminates the need for spatial regime variables. The lagged variable essentially picks up the source of spatial misspeciﬁcation from Model 1 and incorporates any additional autocorrelation not accounted for by the WV, KS and NM variables in Model 2.

Table 4.3: Regression model results Model 1 Model 2 Model 3 OLS Spatial Regimes Spatial Regression Coefficient t Value Significance VIF Coefficient t Value Significance VIF Coefficient t Value Significance VIF Constant -0.092 -0.278 0.163 0.51 0.1746 0.644 ― W_LOG_FLIGHT ― ― ― ― ― ― ― ― 0.4891 5.980 *** ― S_D 0.001 3.926 *** 1.530 0.001 4.955 *** 1.570 0.0001 3.993 *** ― M_D 0.001 6.835 *** 2.073 0.001 7.843 *** 2.107 0.0005 3.099 *** ― L_D 0 1.067 2.491 -2.22E-05 -0.079 2.683 0 1.293 ― POP -1.47E-07 -5.230 *** 2.111 -1.22E-07 -4.755 *** 2.191 ‐8.35E‐8 -3.344 ** ― INCOME -6.35E-08 -0.008 2.268 -9.38E-06 -1.121 3.054 ‐8.10E‐6 -1.265 ― SUBSIDY 1.07E-07 1.806 1.199 1.28E-07 2.367 * 1.246 1.44E‐7 2.905 ** ― TOUR -0.911 -1.502 1.368 -0.635 -1.142 1.433 ‐0.605 -1.203 ― IT -3.122 -0.820 1.299 -1.187 -0.341 1.354 ‐3.116 -0.995 ― FIRE 0.169 0.075 1.367 -0.974 -0.479 1.394 ‐1.498 -0.790 ― PST -2.261 -1.033 1.809 -0.967 -0.465 2.036 ‐2.299 -1.286 ― WV ― ― ― ― -0.36 -2.822 ** 1.303 ― ― ― KS ― ― ― ― 0.32 2.903 ** 1.161 ― ― ― NM ― ― ― ― 0.407 3.015 ** 1.182 ― ― ― - Adjusted R- 0.764 0.811 0.841 Squared Moran's I for 0.2291 **** 0.0698 * 0.0483 Residuals Rho ― ― 0.48912 Standard Error 0.27215 0.24373 0.22222 Breusch-Pagan ― ― 19.44 * *p<0.05. **p<0.01. ***p<0.001

4.5 Discussion and Conclusion

There are several interesting aspects of the regression models that merit discussion.

First, it is not surprising that as the distance between EAS airports and their nearest small

or medium hubs increases, so do the number of ﬂights per capita. This reﬂects a general

level of inaccessibility to commercial air transport options for many EAS communities

outside of their local subsidized service. Recall that many communities are hundreds of

miles from their nearest medium or large hub (Figure 4.3). This is especially true in places

like North Dakota, Minnesota, and Montana. For example, Dickinson, ND, is 523 miles

from its nearest large hub (MSP) and 683 miles from its nearest medium hub (OMA). Even

when small hubs are considered, vastly expanding the palette of air transport alternatives,

many EAS communities remain geographically isolated with respect to non-subsidized

commercial air service.

At the same time, many EAS airports are located in close proximity to each other.

In Pennsylvania, for example, the John Murtha Johnstown-Cambria Airport (JST) is

located only 36 miles from the Altoona–Bair County Airport (AOO). In Kansas, Garden

City (GCK) is only 45 miles from Dodge City (DDC). There are many other examples of

this throughout the United States. This is the primary reason that spatial autocorrelation

was detected in the regression models when evaluating EAS service levels. Although

Model 3 successfully accounted for this spatial dependence, it is important to acknowledge this geographic characteristic of the system. From an operational perspective, the clustering of EAS airports in some regions of the country works against the effectiveness of the program because carriers are actually cannibalizing market areas from each other (Grubesic and Matisziw, 2011). A more optimally structured network of EAS communities (Grubesic

et al., 2012) not only would save taxpayers money, but also would make the program more

efﬁcient without decreasing access and accessibility to rural constituencies.

A second point regarding the operational structure of the EAS system and its associated hubs is that all medium and large hubs are not created equally. For example, an

EAS hub connection to Denver (DEN) could be considered more valuable for a community than a connection to Albuquerque (ABQ). DEN one of the busiest airports in the world; it serves as a hub for United Airlines and the low-cost carrier Frontier Airlines. It is also a focus city for Southwest Airlines. Conversely, although ABQ is also considered a ‘‘large hub’’ by the FAA, it is a much smaller facility that is served by fewer carriers and serves a smaller array of destinations. Interestingly, several EAS communities have multiple hub connections. For example, in 2010, Boone County Regional Airport in Harrison, Arkansas, was provided service by SeaPort Airlines, with connections to both Memphis (MEM) and

Kansas City (MCI). These subtleties are important when evaluating the overall viability of the EAS program. For a more detailed discussion of hubs and route structures, see

Matisziw et al. (2012).

Third, returning to the discussion of more isolated communities, local population has a clear cut, statistical relationship with service levels. For example, of the 13 EAS communities located in North Dakota, Minnesota, and Montana, their average catchment area population is 78,051. When compared to the national average of 838,556 for EAS catchments, it is clear that geographic ‘‘remoteness’’ and local population are strongly linked in the reported models.26 Simply put, isolated communities with sparse population

26 Obviously, this relationship is not limited to Montana, North Dakota and Minnesota. There are many isolated and sparsely populated EAS communities throughout the Great Plains States and the Southwest.

92 are linked to higher, per-capita levels of EAS service. This is to be expected, given that the primary goal of the EAS program is to connect rural and remote communities to the national air transport system.

Another facet of the models worth noting is the lack of signiﬁcance associated with the variables for local economic composition. As highlighted in Sections 2 and 3, the local business environment, particularly industries associated with technology, banking, ﬁnance, insurance, and professional services, often helps generate air passengers for metropolitan areas and their airports – so why not for EAS communities? There are two potential reasons for this. As noted previously, many of the more rural and remote EAS communities simply do not have enough IT, FIRE, PST, or TOUR employees to generate a strong relationship between industrial composition and the need for commercial air service.

Second, the role of industrial composition and EAS viability is also impacted by communities that are located within the economic shadows of much larger metropolitan areas. The shadow effect is well documented in air transport (Taaffe, 1959), but it is also true that successful clusters of IT, FIRE, and PST industries can create agglomeration shadows that prevent the formation of another agglomeration (and possibly any related activity) close by (Krugman, 1994). When one considers many of the EAS communities in

Pennsylvania, Maryland, West Virginia, the Northeast, and portions of the Midwest, their proximity to major metropolitan areas (e.g. Pittsburgh, Philadelphia, Washington DC,

Boston, St. Louis, etc.) may suppress these industries from forming a signiﬁcant presence in EAS markets. In turn, this also means that air passenger trafﬁc remains suppressed, or at least unrelated to industries that typically demand increased face-to-face interaction.

For communities that are not within the shadow of a larger metropolitan area, such

as Grand Island, Nebraska, different factors are at play. Although Grand Island has a

relatively high level of EAS service (1192 ﬂights in 2010), it has been dubbed ‘‘economic

bizarre land’’ (Gandel, 2011) because of its rather non-traditional local economic structure.

For example, even during the recent U.S. economic recession, none of the local banks in

Grand Island failed, the largest local manufacturing plant (operated by Case IH to make combine harvesters) is at full capacity, and Global Industries, a company that makes grain storage bins and other building materials, has witnessed its sales increase nearly 130% since 2003 (Gandel, 2011). Clearly, the economic structure of the Grand Island market area, and many EAS communities for that matter, do not correspond to the more ‘‘traditional’’ economic profile that is associated with generating air travel. That said, perhaps alternative local economic profiles are now triggering demand for air travel in an increasingly globalized economy. For example, consider the presence and potential influence of a localization economy in smaller communities such as Grand Island. A cluster of firms in the same industry, such as commercial agriculture, benefit from labor pooling, increasing returns to scale for manufacturing inputs and enhanced communication (which can yield innovation, Malmberg and Maskell, 2002). Might these localization economies also be responsible for generating modest demand for air travel? It is difficult to say, given the structure of the statistical models presented in this paper, but it appears that such a relationship might exist. Grand Island obviously still requires a subsidy from the EAS

program to attract a carrier. However, if local economic growth continues apace, and

demand for air transport continues to increase, soon EAS subsidies may not be needed in

Grand Island. Further, it appears that more work is needed to disentangle the inﬂuence of

94 commercial agriculture, agricultural equipment manufacturing, and the exploding growth of extraction industries (e.g. oil and gas) on air transport demand in the United States.

A final aspect of the results worth noting is that subsidy levels are positively related to service levels in the EAS system. This is interesting for two reasons. First, this result should be interpreted with caution. As mentioned previously, subsidies are linked to carrier performance in the sense that a minimum service standard must be met. For carriers that provide all of the service agreed upon in their contracts, full payment is received – those that do not can have subsidies withheld. While there is no readily available information on subsidies withheld for 2010, the positive and significant relationship between subsidy and service levels holds. Second, it is important to remember that the expenses associated with providing service to EAS communities are quite variable. Although subsidies are structured to allow for a small cushion of profit for carriers (at least 5%) (U.S. GAO, 2009), flight length, associated fuel costs, and gate fees all impact the cost of doing business for carriers, and these expenses can be highly variable (Matisziw et al., 2012).

In conclusion, it is clear that the Essential Air Service Program is well intended, popular and will continue to link rural and remote communities to the national air transportation system into the foreseeable future. The results of this paper suggest that local service levels are primarily related to geographic isolation, local population, and subsidy levels. Again, this falls neatly in line with programmatic goals. Thus, these results are not particularly surprising, but the lack of any relationship between local economic composition and service levels is certainly notable.

One last point worth making, which is a major concern for transportation policy and planning, albeit beyond the scope of this paper, is determining and/or explaining why

service utilization is so low. As detailed in the introduction, the national average for

passengers on EAS ﬂights in 2010 was six. Considering that the planes used for this service

hold between 9 and 50 passengers, it is clear that the program is underutilized. While it

may seem as if there are relatively obvious answers to this (e.g. thin passenger markets,

self-cannibalization, leakage), regression models exploring the effects of local catchment

areas on service load factors conducted by the authors displayed extremely low statistical

signiﬁcance and explanatory power. As noted previously, EAS ﬂights occur as scheduled,

regardless of the number of passengers on board. From an operational perspective, this

effectively relegates passenger loads to an afterthought.

Finally, the politics of EAS should not be underestimated (Gillies, 2004; Grubesic

and Matisziw, 2011; Grubesic et al., 2012). There are many powerful lobbies and

politicians that favor EAS funding and believe it to be central to the economic and social

vitality of their constituencies. So, although there is signiﬁcant political interest in the EAS

program, the results of this paper clearly suggest that a mentality of ‘‘if you build it, they

will come’’ does not necessarily seem to apply for all EAS airports.

This chapter presents that highway distant to the nearest large hubs, subsidy levels

and the local population of the EAS catchment area are critical factors that determine EAS

service levels and further impact the levels of access and accessibility at EAS airports at a

national scale. Taking all rural airports into consideration including all EAS and non-

subsidized airports, next chapter directly measure the accessibility from a graph-theoretic perspective and develop a new typology of rural airports at a national scale.

5. STUDY 3: EVALUATING NETWORK ACCESSIBILITY AND CREATING A TYPOLOGY OF RURAL AIRPORTS

The last chapter identifies several critical determinants that can impact EAS service levels and airport access and accessibility. However, some rural airports operate commercial services without federal subsidies. It is important to consider all rural airports, including subsidized and non-subsidized, to evaluate accessibility. Creating a bigger picture, this chapter explores the network of airports at a national level, evaluating relative accessibility and using this information to develop a typology of 177 rural airports for the

United States.

5.1 Introduction

Since deregulation of the airline industry in 1978, commercial carriers have largely focused their service efforts on medium and large hub airports located in metropolitan areas with sufficient demand for air travel. These large, profitable locations often cross-subsidize service to smaller, feeder airports, which collect passengers for routing through larger hubs.

However, due to a variety of external and internal factors that affect the airline industry, including economic recessions, high fuel costs, and mergers and acquisitions between carriers, reductions and terminations of air service in smaller rural markets are common.

For example, between 2007 and 2012, domestic service at the largest 29 U.S. airports declined 8.2%. However, during the same time-frame, service at smaller U.S. airports declined by 21.7% (Wittman and Swelbar, 2013). To be clear, this is not a new development. Many rural and remote airports are all too familiar with service reductions.

Currently, there are two programs to help small communities maintain their air services in

the U.S. The first is a federally subsidized program known as Essential Air Service (EAS)

(Grubesic and Wei, 2013; Grubesic et al., 2014). EAS was initially structured to help rural

and remote airports offset the exodus of carriers to larger and more lucrative markets after

deregulation of the airline industry in the United States in 1978 (Grubesic and Matisziw,

2011). As of June 2014, there are 117 active EAS communities, and in each location, several commercial flights are provided each day to nearby medium or large commercial hubs. For example, Worland, WY (WRL)27 is served by Great Lakes Airlines with a 19 seat Beechcraft 1900 – providing connecting service to Denver, CO (DEN).28 The second program, the Small Community Air Service Development Program (SCASDP) established by the Wendell H. Ford Aviation Investment and Reform Act for the 21st Century (P.L. 106-

181) and reauthorized by the Vision 100-Century of Aviation Reauthorization Act (P.L.

108-176) (U.S. DOT, 2013a), allows smaller airports to apply for federal monies that can

be used to mitigate identified service deficiencies (U.S. DOT, 2013b). During fiscal year

2013, SCASDP awarded 25 grants totaling $11.4 million. As detailed by Wittman (2014),

the primary purpose of these awards is to fund new service, although grants have been used

to launch local and regional marketing efforts, terminal improvements, and parking

expansion (U.S. DOT, 2013b).

Although these types of marketing and infrastructure improvement efforts are

certainly important to the long-term vitality of small and rural air transport gateways, one

key challenge that nearly all of these airports face is a lack of accessibility. The reason that

27 The terminology used for referencing specific airports in this paper relies on a combination of city and state references, mixed with IATA codes when appropriate. IATA codes are an essential component of the travel industry. Not only do they help identify airports, hundreds of electronic applications are built upon the IATA codes and help make passenger and cargo operations more efficient (IATA, 2014). 28 Great Lakes Airlines is provided $1,987,148 in subsidies by the EAS program to serve this route.

the accessibility of rural airports is so important is twofold. First, there are noteworthy

tradeoffs for passengers when selecting rural airports as their gateway to the commercial

air transport network. When compared to most small, medium and large hubs, passengers

flying out of rural airports are often forced to accept an additional connection (or two)

when taking a trip. This means more time spent sitting at airports (waiting for connections

to arrive), a higher probability for delays (because of multiple connections), and more time

spent flying between intermediate locations before the final destination is reached. Second,

service frequency from rural airports is relatively low when compared to larger hubs. For

example, carriers participating in the EAS program are required to offer a minimum level

of service between rural communities and their assigned medium or large hub.29 The

Merced, CA (MCE) airport, for instance, offers 13 weekly roundtrip flights to Los Angeles,

CA (LAX). This is a severely reduced palette of destinations and associated flight

frequencies when compared to the neighboring Fresno, CA (FYI), which offers non-stop

service to multiple locales such as Honolulu, HI (HNL), Dallas/Fort Worth, TX (DFW),

Portland, OR (PDX), Las Vegas, NV (LAS), Seattle, WA (SEA), and many others. Clearly,

the accessibility of rural airports impacts their viability as travel options for many

passengers (Matisziw and Grubesic, 2010), but it also may have more severe economic

effects for regions (Allroggen and Malina, 2014; van den Heuvel et al., 2014).

Given the somewhat tenuous nature of commercial air transport in rural locations,

particularly with respect to accessibility, the purpose of this chapter is to explore the

network of airports in the United States, evaluating relative accessibility and using this

29 It is important to emphasize here that carriers never offer more service than the minimum requirement. There is no financial incentive to do so for EAS routes.

information to develop a typology of rural airports for the U.S. Further, this exploratory

analysis is structured to provide meaningful, empirical evidence to improve rural

transportation strategies, highlight the significant gaps in rural accessibility and illustrate

the implications of these gaps for rural communities and their airports.

5.2 Background

5.2.1 Rural Airports

Deepening our understanding of rural transportation challenges, particularly for

commercial air transport, requires a clear framework for measuring rurality and classifying

rural airports (Isserman, 2005). In the United States, rurality can be defined in several ways.

Consider the typology provided by the U.S. Census Bureau for urban and rural areas: 1)

Urbanized Areas (UAs) correspond to areas of 50,000 or more residents and 2) Urban

Clusters (UCs) represent regions with at least 2,500 residents, but less than 50,000. To

clarify, urbanized areas and urban clusters form the urban cores of metropolitan statistical

areas (MSA) and micropolitan statistical areas (μSA), respectively (U.S. Census, 2010).

Specifically, each MSA will contain a UA of 50,000 residents or more. Each μSA will contain at least one UC of at least 10,000 residents, but less than 50,000. Rural areas encompass all population and territory not included within an urbanized area or urban cluster (Federal Register, 2012).

To make matters more confusing, the Bureau of Transportation Statistics compiles an annual list of rural airports for the Internal Revenue Service (IRS). This list is purportedly used by commercial carriers for establishing air fares (U.S. Department of

Transportation, 2013c). Specifically, travel to and from airports defined as rural benefits

100

from reduced ad valorem ticket taxes30 and is exempt from segment fees (U.S. Department of Transportation, 2013c). These reductions in taxes and fees stem from the Taxpayer

Relief Act of 1997. The 2013 IRS definition for rural airports is any airport that has fewer than 100,000 commercial passengers departing from the airport by air during the second preceding calendar year31 and adheres to at least one of the following criteria: 1) The

airport is not located within 75 miles of another airport from which 100,000 or more

commercial passengers departed during the second preceding calendar year,32 2) the airport

was receiving EAS subsidies as of August 5, 1997, or 3) the airport is not connected by

paved roads to another airport. The most recent rural airport list provided by the U.S. DOT

to the IRS is published online (http://tinyurl.com/p6w667x) and contains 3,661 entries, the

bulk of which (80%) do not maintain an IATA identifier code. Figure 5.1 displays all 719

U.S. rural airports with IATA identifier codes. It is important to note that the majority of

these airports do not currently provide commercial service. Instead, most serve as public

gateways for private aircraft, agricultural operations (e.g., spraying crops), or small

freight/logistics hubs for a region.

30 The current ad valorem tax is 7.5% of the ticket price (FAA, 2013). 31 Second preceding calendar year is defined as two calendar years prior to the year in question. 32 Airport distances are calculated from a start and end latitude and longitude based on a constant compass course for a rhumb line between the points (loxodrome) using an Albers Equal Area Projection (IRS, 2013).

101

Figure 5.1: The spatial distribution of all rural airports with IATA codes in the United States, 2013

Historically, rural air transport options and associated levels of service have been a

function of local market size, the geographic proximity of larger hubs, and the proximity

of natural amenities (i.e., resort areas) or major military facilities (Goff, 2005). Simply put,

there is heterogeneity in commercial air service to rural locales. However, as noted

previously, deregulation prompted fears that massive service reductions would take place

in rural communities because commercial carriers would shift their focus to more profitable

markets and larger metropolitan areas (Flynn and Ratick, 1988; Matisziw and Grubesic,

2010). Again, this is the primary reason that the Essential Air Service program was established (Matisziw et al., 2012).

102

Within the context of EAS, it is important to note that the program and its

participating communities are somewhat dynamic. Subsidy levels vary on an annual basis

and a handful of airports come and go in the program. Regardless of this dynamism, many

problems remain, including the overall efficiency of the EAS program (Grubesic et al.,

2014). Not only do EAS airports cannibalize their own markets (Grubesic and Matisziw,

2011), they are frequently located near relatively vibrant small hubs (as defined by the

FCC), which offer a more robust range of air transport services, but are not considered in

the EAS community eligibility criteria (Grubesic and Wei, 2013; Grubesic et al., 2012).

In short, there is room for improvement in the EAS system and the rural air transport

network, more generally.

In sum, rural airports and their associated communities often face a wide array of

contextual challenges for obtaining adequate transportation services. These challenges

include the lack of spatial proximity to larger population centers, poor physical connections

to terrestrial transportation systems (e.g., road and rail), and weak demand-side markets for

air transport. 33 Although many rural airports survive on freight traffic and associated

demand from the agricultural sector, geographic isolation, poor physical connections to the

terrestrial network and the air transport system can impede regional economic development

for rural communities where small airports are located. This significantly weakens the

ability of such communities to attract and hold commercial air service. Matters are further

complicated by the operational challenges associated with providing service in rural and

33 Markets with low demand, especially rural ones, remain a challenge because enplanements suffer, even when the frequency of service remains relatively high.

103

remote locations, which include higher than average operating costs and diseconomies of scale.

In the next section, we outline the data and methods used for exploring the heterogeneities in the network of rural airports in the U.S. and evaluate their relative accessibility to the larger commercial system at the regional and national level. This is followed by the presentation of empirical results and a discussion regarding the implications of rurality for air transport planning and policy.

5.3 Data and Methods

5.3.1 Airline and Airport Data

The Airline Origin and Destination Survey (DB1B) database is a 10% sample of airline tickets from reporting carriers provided by the Bureau of Transportation Statistics

(BTS, 2013). Although these data are sampled, records are structured as individual flight segments. As a result, the information associated with each flight is highly accurate, as are the aggregate statistics generated from the DB1B database. The DB1B database includes three datasets: DB1BCoupon, DB1BMarket, and DB1BTicket (Table 3.1). Since the

DB1BMarket dataset is the primary source for records on air traffic flow attributes, including origins and destinations, number of passengers, a code that identifies number of steps connecting two airports (e.g. PHL:DEN:PDX), and market fare information that can be used to capture itinerary cost, it is selected as the primary data source for analysis. For the purposes of this paper, air traffic flow data for 2013 were obtained from DB1BMarket dataset and a total of 22,568,569 flights were retrieved.

104

Airport data were obtained from the National Transportation Atlas (BTS, 2011).

This dataset includes locations of all public airports in the U.S. Information for classifying

EAS airports in the contiguous U.S. (November 2013) was also collected from the U.S.

DOT (2013c), as were data for airports that received SCASDP funds between 2011 and

2013 (U.S. Department of Transportation, 2013b). Based on the rural airport classification

data obtained from the IRS (2013), the lists of EAS/SCASDP airports, and the typology

provided by the U.S. Census Bureau for urban and rural areas, a complete list of rural

airports (n = 177) with commercial air service was structured for 2013.

5.3.2 Network Analysis

A suite of graph theoretic, network, and statistical methods are used for exploring

the commercial air transport system for the United States in this paper. The graph-theoretic

methods include the use of a simple dispersion measure for the network, known as the

Shimbel Index ( ), which represents the sum of all shortest paths between each city pair,

34: 𝐷𝐷

𝑑𝑑𝑖𝑖𝑖𝑖 ( ) = (5.1) 𝑁𝑁 𝑁𝑁 𝐷𝐷 𝑉𝑉 ∑𝑖𝑖=1 ∑𝑗𝑗=1 𝑑𝑑𝑖𝑖𝑖𝑖 where the shortest path between node pairs is measured by the number of intervening arcs between an origin and destination pair. For this particular measure, smaller row sums

( = ) suggest that airports/communities are more accessible or have an efficient

𝑖𝑖 𝑗𝑗 𝑖𝑖𝑖𝑖 set𝐷𝐷 of connections∑ 𝑑𝑑 to the commercial air transport network in the U.S. Conversely, larger

34 If i=j, dij=0. If a pair of airports indicates a direct flight, dij will be 1 regardless of how close two airports are.

105

row sums suggest that more effort is needed to travel between cities and indicate a less

efficient set of connections between origin/destination (O/D) pairs.

Several measures of centrality (Freeman, 1979) are also used to capture the

characteristics of airports within the system. For instance, one popular and widely used

measure of centrality leveraged in this paper is the degree of node, . Given a matrix of destinations from a particular airport, , with elements , the degree𝛿𝛿 of node is obtained

𝑖𝑖𝑖𝑖 by summing the total number of destinations𝐴𝐴 served from�𝐴𝐴 node� :

𝑖𝑖 = (5.2) 𝑁𝑁 𝛿𝛿 ∑𝑗𝑗=1 𝐴𝐴𝑖𝑖𝑖𝑖 where higher degree nodes often exhibit more relevance to network operations. This often occurs when airports function as hubs, important gateways or as popular origin/destination locations (Fleming and Hayuth, 1994; Grubesic and Matisziw, 2012).

A second measure of centrality utilizes shortest path information to evaluate the probability that some type of interaction between an origin and destination will include a particular node (e.g. hub). For this particular measure, it is assumed that one of the shortest paths will be used in connecting the O/D pair. If there is more than one shortest path, it is assumed that there is an equal probability that each will be used. Therefore, if

𝑖𝑖𝑖𝑖𝑖𝑖 represents the number of shortest paths linking airports and involving airport𝑚𝑚 ,

captures the total number of shortest paths linking each𝑖𝑖 origin𝑗𝑗 and destination that𝑘𝑘

𝑘𝑘 𝑖𝑖𝑖𝑖𝑖𝑖 involves∑ 𝑚𝑚 airport . As detailed by Grubesic et al. (2008), the probability that one of these

paths is used can 𝑘𝑘be represented by the function, 1/ . Further, given this probability

𝑘𝑘 𝑖𝑖𝑖𝑖𝑖𝑖 function, the relative centrality of each node for ∑all𝑚𝑚 of the shortest paths can then be

measured as:

106

= . (5.3) 𝑚𝑚𝑖𝑖𝑖𝑖𝑖𝑖 𝐶𝐶𝐶𝐶𝑘𝑘 ∑𝑖𝑖 ∑𝑗𝑗 ∑𝑘𝑘 𝑚𝑚𝑖𝑖𝑖𝑖𝑖𝑖 𝑖𝑖 ≠ 𝑗𝑗 ≠ 𝑘𝑘 The minimum value for an airport is 0 and occurs if does not participate in any of the calculated shortest paths. It is also important to remember𝑘𝑘 that this measure of betweenness is sensitive to network size, which is why Freeman (1979) suggests the use of a standardized measure to mitigate bias.

Finally, closeness centrality (Freeman, 1979) is a relatively intuitive measure that also utilizes shortest path information to capture the relative network proximity of an airport to all other airports. Simply put, the more central an airport is to the network, the lower its total distance to all other nodes. Thus, closeness can be captured as follows:

= . (5.4) 1 𝜑𝜑 ∑𝑗𝑗 𝑑𝑑𝑖𝑖𝑖𝑖 Much like the betweenness centrality measure, the normalized closeness scores range

between 0 and 1. If an airport receives a score of 0, it is completely isolated from the system,

if an airport receives a score of 1, it is directly connected to all other airports in the system.

5.3.3 Cluster Analysis

A fuzzy clustering approach is used to synthesize collected air flow attributes from

the DB1B database and the network centrality measures specified previously to develop a typology of rural airports. The primary reason that a fuzzy-based approach makes sense for

this application is that the resulting clusters do not adhere to an all-or-nothing assignment structure for membership. Instead, fuzzy clusters allow observations to maintain a probabilistic membership profile (Kaufman and Rousseeuw, 1990). In this case, an airport may strongly identify with one particular cluster, but it also may have characteristics that

107

are similar to airports in a second (i.e., different) cluster. This allows for more flexibility

in the interpretation of generated typologies and remediates one of the most notable

weaknesses in non-hierarchical clustering, hard assignments (Grubesic, 2006). Although

space limitations prevent us from detailing the large literature on cluster analysis and its

applications, readers are referred to Kaufman and Rousseeuw (1990) for an excellent

review on statistical clustering, and Murray et al. (2014) for a review of spatial clustering.

For details on the fuzzy clustering method used in this paper, please see Appendix B.

5.4 Results

Based on the 2013 DB1B market data, 465 airports35 were initially used for analysis.

In total, there are 746,398 unique flight itineraries operating between the 465 airports.

Further, the entire air traffic network across the United States and its territories consists of

465 airports that provided commercial services in 2013 and includes all FAA-defined large,

medium, and small hubs, primary non-hubs, commercial service airports, general aviation

airports, EAS airports, and airports awarded SCADSP funding.36 Considering all possible

flight combinations, the resulting diameter of the DB1B network is 11 and involves a trip

from Marshall, AK (MLL) to Teller, AK (TLA). Without delving into the gory details of

the potential itinerary between these two locations, Anchorage (ANC), the largest airport

in Alaska, does not serve as a primary hub connector for this O/D pair. Instead, it serves as

a sub-connection, with a mandatory route via locations in the contiguous 48 states.37 When

35 Although over 3,300 airports are identified by the FAA’s National Plan of Integrated Airport System, 465 airports are identified as having commercial service in the DB1B database in 2013. 36 Passenger-based and subsidy/grant-based airport categories are not excluded.

108

one eliminates these types of unusual itineraries, a more realistic snapshot of network

diameter begins to emerge. For instance, when itineraries with 1 or fewer passengers are

eliminated, it is possible to book a trip from the Mid-America St. Louis Airport (BLV) to

Chefornak, AK (CYF), yielding a diameter of 6 for the network. Thus, although there is a

case to be made for removing outliers from the analysis, the sheer number of itineraries

processed in this analysis will effectively smooth the bias generated by the extreme

itineraries. Thus, this paper keeps the DB1B data intact for analysis.

5.4.1 Rural Markets

In an effort to deepen our understanding of these smaller airports, we leverage

information from the DB1B Market database using the final list of rural airports outlined

previously. Figure 5.2 displays the spatial distribution of 177 rural airports providing

commercial air service in the contiguous United States, its territories, and Alaska in 2013.38

It is important to emphasize that although this paper focuses on the subset of 177 rural

airports, all network measures are generated using the complete air transport network and

the 465 active commercial airports in the U.S.

Table 5.1 presents the top ten rural airports by passengers and flight frequency. Not

only does Kalispell, MT (FCA) serve the most passengers, it also provides the highest

frequency of service for any rural airport. To provide some perspective on these results, it

is important to note that FCA serves as the commercial air gateway to Glacier National

37 It is important to note for readers that the purchase of an 11 segment itinerary, while possible in the DB1B database, may be extraordinarily difficult for a typical passenger via standard booking engines. Realistically, this is a trip that could never be booked and service between this O/D pair would require a chartered flight. 38 Routes to Hawaii are included in the network analysis, but since Hawaii does not have any rural airports, it is not considered for the rural analysis, nor is it visualized in the associated figures.

109

Park, which had 2.19 million visitors in 2013 (NPS, 2013). Because of its geographically

remote location, driving to Glacier NP is not an option for many visitors. As a result,

flights to FCA are frequent during the summer season, with service provided between

Portland (PDX), Los Angeles (LAX), Oakland, CA (OAK), Atlanta, GA (ATL) and

Chicago, IL (ORD). When combined with year-round connections to Seattle (SEA), Las

Vegas (LAS), Minneapolis (MSP), Salt Lake City (SLC), and Denver (DEN), FCA is the busiest rural airport in the U.S, serving 198,91839 passengers on 3,225 flights for 2013.40

Figure 5.2: Rural airports in the United States with commercial traffic, 2013

39 191,918 is the actual number of passengers for all scheduled services in 2013 at FCA. Information was obtained from http://www.transtats.bts.gov/Data_Elements.aspx?Data=1. 40 These regular and semi-regular connections do not include infrequent charter flights which can appear as in the DB1B database for certain city pairs, especially in resort areas.

110

In addition to FCA, many of the top rural airports for passenger counts and flight

frequency are connected to tourism or fueled by their local economy. For example, Hilton

Head Island, SC (HHH) and Yuma, AZ (YUM) are prime outdoor recreation areas which

attract a large number of winter “snowbirds” seeking refuge from the harsh northern

climates. Williston, ND (ISN) is a central point in North Dakota’s Williston Basin and

associated Bakken formation, which is currently producing more oil than other site in the

United States (EIA, 2013) and generating significantly more demand for air travel to and

from the region.

Table 5.1: Top ten rural airports with passenger and frequency counts Ranking Total Number of Passengers Frequency Counts 1 Kalispell, MT (FCA) [37,084]41 Kalispell, MT (FCA) [28,475]42 2 Plattsburgh, NY (PBG) [25,658] Williston, ND (ISN) [16,333] 3 Williston, ND (ISN) [17,793] Tyler, TX (TYR) [15,481] 4 Casper, WY (CPR) [16,914] Abilene, TX (ABI) [14.032] 5 Tyler, TX (TYR) [16,410] Yuma, AZ (YUM) [13,733] 6 Abilene, TX (ABI) [15,634] Casper, WY (CPR) [13,125] 7 Yuma, AZ (YUM) [15,486] Salisbury, MD (SBY) [11,058] 8 Sitka, AK (SIT) [12,197] Hilton Head Island, SC (HHH) [10,851] 9 Hilton Head Island, SC (HHH) [11,867] San Angelo, TX (SJT) [10,752] 10 San Angelo, TX (SJT) [11,793] Manhattan, KS (MHK) [10,004]

When compared to raw passenger counts and flight frequency metrics, a network

analysis of the air transport system for the U.S. generates similar list of rural airports.

Figure 5.3 and Table 5.2 detail these results. FCA has the smallest Shimbel distance (1,095),

which suggests that there are 1,095 steps required to connect Kalispell to the other 464

41 This is the total count of passengers served from the DB1B sample (10%). 42 This is the total count/frequency of scheduled and non-scheduled flights (arrival and departure) from the DB1B sample (10%).

111

commercial airports in the U.S. This suggests that Kalispell is the most accessible rural

airport in the country. This is not a surprising result, given all the larger hubs that connect

to FCA (e.g. LAX, SEA, ORD, etc). Joplin, MO (JLN) has the second lowest D measure

(1,157) and Casper, WY (CPR) is a close third, requiring 13 additional steps. Of note is

that Joplin is an EAS community and served by American Eagle with a connection to DFW

during 2013.43 Again, this singular connection to DFW is the primary driver of a low D

value. Unlike JLN, CPR is able to maintain commercial air services without EAS subsidies, but it did receive a SCASDP grant in 2012. $100,000 in federal funds was provided to the

Casper airport, with an additional $70,000 of in-kind contributions from other local entities.

These monies were dedicated to marketing efforts to enhance service on the CPR-LAS

connection, support new service between CPR and SLC, and address high fare issues for

the airport (CNCIA, 2012).

43 This route is served by a 50 seat Embraer ERJ aircraft and has the lowest EAS subsidy rate in the system ($342,560) (U.S. DOT, 2013c).

112

Figure 5.3: Shimbel index of rural airports in the United States, 2013

Table 5.2: Top ten rural airports with the shortest shimbel distance Ranking Airport Shimbel Distance 1 Kalispell, MT (FCA) 1095 2 Joplin, MO (JLN) 1157 3 Casper, WY (CPR) 1170 4 Columbia, MO (COU) 1178 5 Williston, ND (ISN) 1198 6 Dickinson, ND (DIK) 1199 7 Manhattan, KS (MHK) 1206 8 Pueblo, CO (PUB) 1206 9 Grand Island, NE (GRI) 1207 10 Beaumont/Port Arthur, TX (BPT) 1208

113

At the opposite end of the spectrum are the most poorly connected airports, such as

Marshall, AK (MLL) (3,474 steps), Unalakleet, AK (UNK) (2,837 steps), Lewistown, MT

(LWT) (2,738 steps), and Ely, NV (ELY) (2,260 steps). Unalakleet is one of the most remote places in Alaska, located on the Norton Sound of the Bering Sea. UNK primarily serves as a cargo airport, although 13,430 passengers did use commercial air service during

2013.44 The issues associated with connectivity and EAS subsidies for Lewistown have

been documented thoroughly in the literature (Grubesic and Matisziw, 2011; Matisziw et

al., 2012; Grubesic et al., 2012), so its low ranking is somewhat expected. In fact, LWT is

no longer served by an EAS carrier because the per-passenger subsidies exceeded the

$1,000 limit. The same goes for Ely, NV, a former EAS community, which also lost commercial service because it exceeded the federally mandated per passenger subsidy maximum of $1,000 (Ely = $3,720).

Table 5.3 lists the rural airports with the highest centrality measures. Figures 5.4,

5.5 and 5.6 visualize the degree, closeness, and betweenness measures of rural airports in the United States, respectively. However, it is important to place these results in context

(see Table 5.3). Once again, consider FCA, which appears near the top of the list for degree and closeness, but it does not register highly for betweenness, suggesting that it is a true origin/destination (i.e. start point/endpoint) in the network without any hub functionality.

Even Plattsburgh, NY (PBG), which leads all rural airports for the betweenness measure, does not truly function as a hub in any way, but does maintain service to Fort Lauderdale

(FLL), Fort Myers (RSW), Seaford (SFB), Clearwater (PIE) (all in Florida), Boston, MA

44 UNK has a direct connection to both Anchorage and Nome.

114

(BOS) and seasonal flights to Myrtle Beach, SC (MYR). PGB is also close to the Canadian border and serves as a gateway to Canadian shoppers and as a connecting point for passengers traveling to Florida. In short, there is variety in the centrality measures for rural locations, all of which is fueled by operational quirks (e.g. EAS connections to hubs) or network geographic context (central locations in the U.S.).

Table 5.3: Top ten rural airports and associated measures of centrality Ranking Airport Degree Airport Closeness Airport Betweenness 1 Kalispell, MT 19 Kalispell, MT 0.428440 Plattsburgh, 0.000134 (FCA) (FCA) NY (PBG) 2 Yuma, AZ 17 Casper, WY 0.403829 Manhattan, KS 0.000085 (YUM) (CPR) (MHK) 3 Casper, WY 15 Joplin, MO 0.403829 Lewiston, ID 0.000066 (CPR) (JLN) (LWS) 4 Nantucket, MA 15 Williston, 0.402080 Rock Springs, 0.000049 (ACK) ND (ISN) WY (RKS) 5 Plattsburgh, 14 Columbia, 0.396581 Nantucket, MA 0.000047 NY (PBG) MO (COU) (ACK) 6 Vineyard 11 Cody, WY 0.389916 Yuma, AZ 0.000035 Haven, MA (COD) (YUM) (MVY) 7 Petersburg, AK 10 Dickinson, 0.389916 Casper, WY 0.000033 (PSG) ND (DIK) (CPR) 8 Wrangell, AK 10 Pierre, SD 0.389916 Columbus, GA 0.000027 (WRG) (PIR) (CSG) 9 Lewiston, ID 9 Pueblo, CO 0.387960 Salisbury, MD 0.000024 (LWS) (PUB) (SBY) 10 Sitka, AK 9 Manhattan, 0.387312 Clarksburg, 0.000022 (SIT) KS (MHK) WV (CKB)

115

Figure 5.4: Degree Measure of Rural Airports in the United States, 2013

Figure 5.5: Closeness measure of rural airports in the United States, 2013

116

Figure 5.6: Betweenness measure of rural airports in the United States, 2013

5.4.2 Rural Airport Peer Groups

In an effort to disentangle all of the derived network measures, data on passengers,

market fares, miles flown, and frequency of service for rural airports, cluster analysis is

used to develop an airport typology and to delineate rural airport peer groups. A total of

twelve variables were used for clustering. Table 5.4 presents the averages of the twelve

variables at the national level and three derived clusters. In addition, it is important to note

that the four key attributes associated with air traffic flow have two facets each: origin-side

and destination-side metrics. For example, consider the airport pair of FCA and SEA.

There is a market fare associated with each airport (both origin and destination) and these

fares are not necessarily identical. The fare from SEA to FCA may be more expensive than the fare from FCA to SEA. Thus, both sides of the market fare, for all interacting pairs,

117

are considered for each airport in the cluster analysis. In addition, where the process of

clustering is concerned, since fuzzy approaches are a generalization of partition-based

approaches, it is necessary to predefine the number of clusters (k) for analysis. However, rather than simply picking a static value for k, this paper uses a range of values (k = 3, 4, …,

15) for the purposes of sensitivity analysis in defining peer groups. A suite of diagnostic metrics were also used to make final classifications, including silhouette coefficients for evaluating the quality of the cluster results.

When the average silhouettes were evaluated for each k, the highest value was 0.493 for k = 3. As a result, three clusters were generated to define rural airport peer groups and each airport was assigned to the cluster corresponding with its highest probability score.

The following are the details of each of the three cluster groups and the associated rural airport typology:

Table 5.4: Summary statistics of clusters for rural airports Rural Cluster 1: Cluster 2: Cluster 3: National Rural Middling Rural Moribund Rural Average Overachievers Gateways Terminals Passengers (Departure) 1,638 6,379 1,990 261 Passengers (Arrival) 1,640 6,357 2,014 271 Market Fares45 (Departure) $378.33 $354.77 $385.31 $380.98 Market Fares (Arrival) $388.97 $358.81 $398.03 $392.27 Miles Flown (Departure) 1,732.81 1,756.22 1,933.56 1,630.41 Miles Flown (Arrival) 1,739.12 1,749.21 1,929.08 1,646.19 Frequency Counts (Departure) 1,292 4,907 1,571 236 Frequency Counts (Arrival) 1,298 4,926 1,582 247 Degree 3.7 7.730769 4.020408 2.231707 Closeness 0.352224 0.381053 0.365503 0.338629 Betweenness 3.32E-06 0.000018 1.64E-06 3.71E-07 Shimbel 1,364 1,235 1,289 1,432 Cluster Size 177 26 49 102 EAS Airports 106 2 24 80 EAS (AK) Airports 6 0 3 3 SCASDP Airports 12 5 4 3

45 Market fares are for one-way segments.

118

Cluster 1: Rural Overachievers

The Rural Overachievers group is the smallest of the three clusters, with 26 member

airports (Table 5.4; Figure 5.7). Representative members include FCA, CPR and PBG.

Perhaps the most notable characteristic of this group is the high volume of passenger

arrivals and departures, both of which average over 6,300 per year, much higher than the

national average for rural airports. Interestingly, the market fares associated with these

airports are the lowest, on average, between the three groups, even if the miles-flown are somewhat higher than Cluster 3. Frequency of service for the Rural Overachievers is also significantly higher than the other groups, as is the degree of node for these airports.

Perhaps the biggest surprise for this group is that two members are EAS subsidized – PBG and Grand Island, NE (GRI). The nuances of PBG were discussed earlier, but GRI is an interesting market. EAS subsidizes the GRI to DFW route via American Eagle for $2.2 million per year, however, Allegiant Air has also moved into the market, offering connections to Mesa, AZ (AZA) and LAS. In previous work, Grubesic and Wei (2013) note that GRI does not lie within the air traffic shadow of a larger metropolitan area (Taaffe,

1956), which helps GRI maintain a somewhat captive commercial passenger market. The economy is also vibrant in Grand Island, largely focused on agribusiness, which spurs additional demand for commercial service (Grubesic and Wei, 2013). 46 Finally, five of

the Rural Overachiever airports received SCASDP grants between 2011 and 2013,

46 During the recent U.S. recession (2007-2009), Grand Island’s economy was relatively resistant to the economic downturn. With an emphasis on agriculture and agricultural equipment production, unemployment rates remained low and factories were working at capacity (Gandel, 2011).

119

including Hailey, ID (SUN),47 San Angelo, TX (SJT), St. George, UT (SGU), Wenatchee,

WA (EAT), and CPR.

Figure 5.7: The spatial distribution of rural airport cluster groups in the United States, 2013

Cluster 2: Middling Rural Gateways

The Middling Rural Gateways group is an interesting mix of 49 airports from the

contiguous U.S. and Alaska, but it also includes Pago Pago (PPG), which is located on

American Samoa, an unincorporated U.S. territory in the South Pacific. Other

representative airports include Barrow, AK (BRW), Redding, CA (RDD) and Laramie,

47 SUN won a SCASDP award to explore the expansion of commercial air service from a large hub in the eastern U.S. (e.g., Cleveland, Columbus, Washington, New York City) to Hailey, ID. The plan is to attract more tourism dollars to the region during ski season (FMAA, 2013).

120

WY (LAR). Of note is that 24 non-Alaska EAS communities are included in this cluster.

This means that almost half of the airports in this group require subsidies to maintain

commercial service. This cluster has the highest market fare, on average (~$390), and trip

distances in this group are the longest of all three clusters, averaging over 1,900 miles. In

part, this can be attributed to the presence of PPG, where the only jet service available is

to HNL, 2,613 miles away. Where the centrality measures are concerned, the average

degree of node (4.02) is slightly higher than the national average (3.7), but many of these

airports (including most of the EAS communities) only maintain one commercial

connection. For example, the only commercial service available to and from RDD is to San

Francisco, CA (SFO) via United Express (SkyWest). RDD is not alone; many non-EAS airports are only able to support a single connection in this group, such as Laramie, WY

(LAR-DEN) and Twin Falls, ID (TWF-SLC). There are other airports in this group that support more connections, increasing the average degree of node measure for the cluster.

For example, Vineyard Haven, MA (MVY), maintains service to BOS, Nantucket, MA

(ACK) and New Bedford, MA (EWB) with seasonal service to New York, NY (JFK),

White Plains, NY (HPN), Teterboro, NJ (TEB) and Arlington, VA (DCA). Also, while many of the airports in Alaska included in this group (Petersburg [PSG], Wrangell [WRG],

Cordova [CDV], etc.) maintain multiple connections via Alaska Airlines, other airports in this cluster benefit from semi-frequent charter services when permanent commercial connections are unavailable.

Cluster 3: Moribund Rural Terminals

The final group of 102 airports is Moribund Rural Terminals, which includes 80

non-Alaska EAS communities. Representative locations include Muscle Shoals, AL

121

(MSL), Jonesboro, AR (JBR), Miles City, MT (MLS), and LWT. Commercial service in

these communities is sparse, with only several hundred passengers departing or arriving,

on average, during 2013. 48 Although the market fares correspond to the national

benchmarks for rural airports, it is important to remember that EAS subsidies reduce the

fare burden for passengers, keeping them artificially low and relatively close to the rural

averages. The true cost of operating these flights is very high and if portions of these

expenses were passed on to customers through the fares, the ticket prices would increase

dramatically. Further, it is likely that there would be no commercial service for the majority

of these cities without the EAS program (Grubesic et al., 2014). Perhaps the most revealing

statistic is the average D measure (1,432) for these locations. This suggests that, on average, the Moribund Rural Terminals require 143 more steps than Middling Rural Gateways and

197 more steps than Rural Overachievers to connect to all of the commercial airports in the U.S. This underscores exactly how remote and poorly served this cluster is, even when compared to their rural peers. For example, consider Topeka, KS (FOE), the state capital of Kansas and home to a metropolitan area population of over 234,000. In 2013, only a

handful of commercial flights operated to and from FOE according to the T-100 database.49

Specifically, there are twenty destinations connecting to FOE with commercial and non-

commercial scheduled service. For example, three charter flights between FOE and

48 Some rural airports may survive as centers for cargo transport or landing facilities but they do not provide any access to the national transportation network for passengers, a primary factor that keeps airports operating and maintains/develops local economic activities. Lack of passenger traffic is not only an important factor affecting rural airports, but also rural communities in general. 49 The T-100 database provides a 100% census of the domestic and international air traffic data. It provides monthly traffic and operational data for each carrier and for each city-pair market that the carrier operated (U.S. DOT, 2013d). However, the T-100 does not provide market fares or itinerary costs, limiting its overall appeal for the analysis in this paper.

122

Laughlin/Bullhead City, AZ (IFP) were documented, all of which were operated by Sun

Country Airlines and all passengers were destined for Don Laughlin’s Riverside Resort in

Laughlin, NV. Unlike GRI, detailed previously, FOE is located within the shadow of a much larger metropolitan area and its airport (77 miles from Kansas City, MO [MCI]).

This clearly presents challenges in attracting regular commercial service. However, this is exactly what happened in January 2014, when United Airlines began offering multiple daily roundtrips between FOE and ORD. Unfortunately, United cancelled this service only

6 months later because, on average, only 49% of the available seats were filled for this route, much lower than the average of 86% for United’s other Chicago routes and the third lowest for all flight segments in United’s network (Schleisman and Dulle, 2014). Today, once again, the only commercial service in/out of FOE is a charter flight to Wendover, UT

(ENV) or Laughlin-Bullhead (IFP), offered every 6 weeks from Allegiant Air. FOE is not unique; many of the Moribund Rural Terminals are served by infrequent charter flights and remain without regular commercial air transport service.

5.4.3 Transitory Peers

A final group worth mentioning includes the transitory airports which have a relatively equal probability of membership across multiple clusters. From a statistical perspective, although these airports are correctly assigned to their optimal peer group, there is statistical evidence that they also identify with airports in a different cluster. For example, consider Southwest Georgia Regional Airport (ABY), located in Albany, GA, which is a member of the Middling Rural Gateways cluster. ABY is located about 175 miles south of

Atlanta and 75 miles north of Tallahassee, FL. The probability of ABY’s membership to

123

Middling Rural Gateways is moderate, at 45.01% (Table 5.5). However, it also has a 37.11% probability of belonging to Rural Overachievers. If one digs a bit deeper into the characteristics associated with ABY, it becomes clear that average market fares (~$360) compare quite favorably to the national averages and Rural Overachievers, but the passenger enplanements (~2,990) are lower than the ~6,300 mark for Rural Overachievers.

In other instances, many of the airports that belong to Middling Rural Gateways are also closely related to Moribund Rural Terminals (Table 5.5). If nothing else, this highlights the razor-thin line between seemingly viable rural airports and those that are struggling. It also highlights potential pathways for middling airports to achieve more success. For example, can issues of low enplanements be addressed for airports like Albany, GA (ABY), helping them break into the Rural Overachievers group? The answer is a definitive yes, particularly if local managers are aspiring to expand airport operations, increase quality of service, and attract more passengers. In fact, this is exactly what SCASDP grants are for. In the next section, we detail how this type of empirical evidence can be used for airport planning and improving rural air transportation policy.

124

Table 5.5: List of transitory airports Airport City State Probabilities ID Cluster 1: Rural Cluster 2: Middling Cluster 3: Moribund Overachievers Rural Gateways Rural Terminals ABY Albany GA 37.11% 45.01% 17.88% ANI Aniak AK 15.85% 39.29% 44.87% BRW Barrow AK 35.38% 43.84% 20.78% CMX Hancock MI 38.48% 44.39% 17.13% CYS Cheyenne WY 8.47% 50.39% 41.14% DLG Dillingham AK 12.05% 46.06% 41.89% EAR Kearney NE 7.50% 45.20% 47.29% GAL Galena AK 17.54% 39.76% 42.71% GAM Gambell AK 18.05% 39.85% 42.10% HOM Homer AK 11.24% 44.13% 44.63% INL International MN 8.15% 49.33% 42.52% Falls MLL Marshall AK 20.29% 39.61% 40.10% PIR Pierre SD 7.39% 43.81% 48.79% PPG Pago Pago AS 30.34% 39.22% 30.44% ROW Roswell NM 41.01% 42.14% 16.85% UNK Unalakleet AK 16.63% 39.18% 44.20% YAK Yakutat AK 8.88% 44.14% 46.99%

5.5 Discussion and Conclusion

There are several facets of the results worth further discussion. First, many rural airports are struggling to attract passengers, provide competitive fares, and maintain a reasonable frequency of service. 58% (102 of 177) of rural airports in this analysis can be considered moribund – exhibiting limited frequency of service, few enplanements, low load factors,50 little network connectivity, and an over-reliance on infrequent charter flights.

In fact, the vast majority (81%) of these moribund airports are part of the Essential Air

Service program, relying on subsidies to help maintain their connection to the commercial air network.

50 Passenger load factors are a measure of how much passenger carrying capacity is used. Specifically, it tracks the number of passenger miles flown as a percentage of seat miles available.

125

As detailed throughout, the EAS program is an interesting one, but not without

controversy. As of November 2013, $219.8 million was allocated to carriers for providing

commercial air transport service to 106 communities in the lower 48 states. At the top end

of this subsidy spectrum, $3.8 million was provided to Presque Isle/Houlton, ME (PQI) for connecting service. As detailed by Grubesic et al. (2014), the hope is that these communities will eventually generate enough passenger demand and associated traffic to operate without subsidies, but this rarely occurs. Most communities remain subsidized because they are simply too small, too remote, or their market demand is too low to support regular commercial service.

What does this mean for the EAS program and rural communities dependent upon subsidies for commercial air transport? The future is not clear. EAS continues to be criticized for its high costs, low efficiency, and infrequent passenger use throughout the

U.S. (Grubesic et al., 2014), but it continues to survive as a program because of many vocal proponents in the U.S. Congress and Senate that represent rural constituencies (Taylor,

2012). In short, with a federal budget request that exceeds $3.9 trillion for 2015 in the U.S., cutting $220 million dollars from the EAS program barely registers as a concern. If anything, the potential fiscal savings achieved by cutting EAS would be offset by antagonizing members of Congress who are strong supporters of the program. As a result,

EAS perseveres.

On a much smaller scale, the SCASDP program awarded a paltry $11.4 million during 2013 and each recipient received less than $1 million. Unlike EAS, which provides subsidies directly to airlines, SCASDP is designed to encourage small communities to operate sustainably and to mitigate operational challenges through creative problem

126

solving and strategic decision-making. This may involve the aggressive pursuit of carriers

to provide service to new markets, or to improve advertising efforts within the region.

Regardless, it is this type of targeted funding that may help rural and remote communities

achieve sustainable commercial service in the long run, helping them avoid any and all

dependency associated with the pervasive EAS safety net that has existed since 1978.

In sum, while there are many rural airports that continue to struggle, all is not

gloomy. The Rural Overachievers group represents a healthy subset of rural markets that exhibit relatively low average fares when compared to their rural peers, a diverse set of

network connections and a relatively high number of annual enplanements when compared

to the rural national average. Although each of these markets is different, there are a few

commonalities amongst this group. Most of these airports serve relatively captive geographical markets, far removed from the spatial influence or “shadow” of larger hub airports. Again, Grand Island (GRI) is an excellent example of this – located 153 miles from the nearest major airport in Omaha (OMA). Many Rural Overachievers also have

vibrant local economies. In the case of Williston, ND (ISN), the oil industry is booming,

generating significant demand for business travel to and from the area. Finally, the best

rural markets have an excellent match between supply and demand in the structure of the

network serving a community. Although every rural airport authority would like to have

multiple connections, more connectivity does not always equate to better service. Too

many connections (or frequency of service) means that carriers are running planes with

fewer passengers and low load factors. Over time, this reduces profits and may negatively

impact carrier sentiment toward a market (e.g., United Airlines in Topeka). Conversely,

too few seats and overcrowded planes reduce profits and negatively impact passenger

127

quality of service. As detailed by Kanafani and Abbas (1987), an extensive

origin/destination survey is an absolute necessity for rural airports, especially for planning

purposes regarding network operations.

One option to break this cycle, especially for the many airports that are classified

as Middling Rural Gateways and Moribund Rural Terminals is the implementation of

irregular, “non-scheduled” service such as charter flights. For example, Page, AZ (PGA)

and Topeka, KS (FOE) served over 50% of their passengers from non-scheduled services

during 2013, including various charter services, air taxi, and air tours. Hattiesburg, MS

(PIB) and Pueblo, CO (PUB) also have large percent of passengers in non-scheduled

services.

In general, both EAS and SCASDP can be helpful to rural communities, but these

programs are really structured as short-term fixes, and should be treated as such. With that in mind, the typology of rural airports generated in this paper can serve as a foundation for beginning the process of strategically thinning EAS subsidies and better targeting SCASDP funds. For example, as detailed previously, airports in the Rural Overachievers (n = 26)

group generally have healthy rural markets and most are able to operate without subsidies

or grants. However, as of November 2013, Plattsburgh (PBG) and Grand Island (GRI), two

airports in this group, were still receiving $2,470,834 and $1,837,021 from the EAS

program, respectively. These airports may be good candidates for strategic reductions in

EAS monies given their local market structure.

Finally, the empirical work presented in this paper, particularly concerning peer groups, can serve as a powerful tool for enhancing planning and policy for rural air transport. Not only do peer groups allow managers, policy-makers, and airport authorities

128

to better understand how their airport statistics compare to similar markets, peer groups

can also provide intelligence as to what steps can be taken to improve local airport services.

Once again, airports like Plattsburgh (PBG) provide a good example. At one time, PBG

was served by a single EAS subsidized carrier. Today, both Allegiant Air and Spirit air provide seasonal roundtrip service to multiple destinations in Florida and South Carolina.

PBG is also offering free parking to passengers. Airports aspiring to break-out of the vicious cycle of local air service (Kanafani and Abbas, 1987), such as Topeka (FOE), may wish to leverage some of the strategies used by PBG to attract additional carriers and passengers. SCASDA grants are one option for helping this process along. 51 More

rudimentary strategies can also be ascertained from peer group information. For example,

a small investment in marketing efforts may be enough for some airports to attract more

passengers. However, without knowing the passenger counts and other relevant

performance data of peer airports, setting realistic goals and strategies based on peer

benchmarks may be difficult to develop.

In closing, rural air transportation in the United States continues to generate a

complex web of economic, operational, geographic and policy challenges. Rural

communities are diverse and so are their transportation needs. Clearly, a one-size-fits-all strategy for ameliorating the difficulties associated with rural air transport will not work for the United States, China, Brazil, or any other large and regionally diverse country.

51 Manhattan, KS (MHK) is another regional airport that was aided by the EAS program between 2003 and 2009. However, it was eventually eliminated from the EAS roster because it was able to attract a commercial carrier without subsidies in 2009. Further, MHK benefitted from a SCASDP grant in 2010 to maintain this commercial service as it struggled to become established. It also provides free parking and a rewards program to attract passengers. Currently, MHK operates scheduled commercial service to and from Chicago, IL (ORD) and Dallas/Fort Worth, TX (DFW).

129

However, the results of this paper provide a good first step in better understanding the

challenges at hand, while simultaneously providing planning officials and policy-makers an exploratory data analysis framework for improving decision making and helping craft higher-quality transport legislation. Although there is no easy way forward for many of the rural airports profiled in this paper, there is no reason such communities cannot be better prepared for lobbying for additional support or developing strategies for improving their

local transportation options.

This chapter provides a good understanding of network accessibility and develops

a typology of rural airports at a national level. More importantly, all rural airports including

EAS airports are classified, considering a set of network measures and air traffic attributes,

to capture the characteristics of different groups of airports. One interesting, important

finding is that peer groups provide better understanding how their airport statistics compare

to similar markets and intelligence as to what steps can be taken to improve local airport

services. The analysis can be used as a powerful tool for enhancing planning and policy for

rural air transportation. This chapter only focuses on rural airports so the key players

(large/medium airport hubs) in an air transportation system are not involved so next chapter

analyzes all airports in the United States including hubs and rural airports.

130

6. STUDY 4: EXPLORING THE SPATIOTEMPORAL TRENDS IN AIR FARES AT U.S. AIRPORTS AND ASSOCIATED PAIRS

As detailed in Chapter 2, airport competition and market leakage have significant

impact on air fares and itinerary pricing, and further airport access and accessibility.

Drawing a bigger picture than the last three chapters, this final substantive chapter

continues to evaluate airport access and accessibility by exploring the uneven

spatiotemporal distribution of air fares, by all airports and associated flight pairs, examining asymmetries in patterns of air fares over time and across space using air traffic

data at a national level.

6.1 Introduction

The deregulation and liberalization of the air transport industry, throughout the

United States, Europe and Asia, has dramatically enhanced portions of the global

transportation system through rapid expansion and densification of networks. However,

the bulk of this development has occurred in larger markets, with strong economies and a

mobile populace, often overlooking smaller, peripheral markets (Bowen, 2002). The

relative dynamism of the global air transport market is also important within this context.

As detailed by O’Connor and Fuellhart (2015), technological shifts, fluctuations in

consumer preferences and related market pressures (e.g. fuel prices, homeland security,

disease, etc.) are constantly spawning changes to destinations served, seats available,

itinerary pricing and the rise or decline in gateway operations. For example, the dehubbing

of airports (Redondi et al., 2012; Rodriguez-Deniz et al., 2013) can generate massive

changes in connectivity and pricing for a dehubbed gateway (Wei and Grubesic, 2015a).

131

Further, these changes are not limited to the dehubbed airport; they cascade through the

service network, impacting distant cities, operators and passengers – sometimes for the

better, sometimes not. In short, the role of an airport or a flight pair (i.e., origin/destination),

both individually and as an element of the larger global network, can change over time.

One way to capture all of these changes, either directly or indirectly, is air fare. From a

spatial perspective, we already know that the pricing of air fares is unevenly distributed

(Goetz and Sutton, 1997; Vowles, 2000; Fuellhart, 2003; Grubesic and Zook, 2007). It is

also well established that the uneven spatial distribution of air fares can impact passenger

choice for an origin airport and even a specific itinerary (Harvey, 1987; Innes and Doucet,

1990; Phillips et al., 2005; Zhang and Xie, 2005). From a temporal perspective, air fares

generally trend upward. For example, during 2007-2012, the average fares at the 29 largest

U.S. airports and 35 mid-sized airports increased by 8.7% and nearly 12% respectively

(Wittman and Swelbar, 2013).

The purpose of this chapter is to combine both the spatial and temporal elements of air fare analysis to explore the spatiotemporal distribution of fares in the United States by airports and associated flight pairs. This is a marked departure from much of the existing literature on airfares and airports, which tends to focus on average fares for ranking airport affordability and identifying discounted or overpriced gateways. In short, by using an average fare metric, the dynamism in fares for each airport is smoothed and/or masked.

For this paper, we drill deeper and use data for individual flight segments between origin/destination pairs to provide a more nuanced overview of market dynamics, particularly for smaller, more peripheral gateways. By examining the asymmetries in these patterns and detailing their underlying variability, we make some progress in achieving a

132

more holistic understanding of air transport accessibility, pricing and regional transport

forces in the United States.

6.2 Background

6.2.1 Analytics to Capture Air Fare Trends

There are many different analytical methods and underlying substantive contexts

that have been useful for capturing trends in air fares; from exploring the effects of code

sharing agreements (Oum et al., 1996) and horizontal mergers (Prager and Hannan, 1998),

to identifying the impacts of route competition (Brueckner et al., 2013), especially in

MARs (Cho et al., 2012). From a spatial and temporal perspective, much of the existing

work has focused on collecting snapshots of average itinerary pricing (in aggregate) for

individual airports (Goetz and Sutton, 1997; Oster and Strong, 2001; Goetz, 2002; Fuellhart,

2003; Tierney and Kuby, 2008; Goetz and Vowles, 2009), or the dispersion of fares for

particular origin/destination (O/D) pairs (Borenstein and Rose, 1991; Gerardi and Shapiro,

2009). However, the use of airport averages, or a strong focus on one or two routes for analysis obfuscates the dynamism in fares for each airport. In short, averages smooth variability. Moreover, focusing on a limited set of routes artificially reduces sample size and is too myopic to capture more general trends in the data. Worse, it can completely overlook additional, interesting routes that show variability. For this chapter, we attempt

to add clarity to the spatiotemporal dimensions of airfare and itinerary pricing for the

United States by exploring three distinct foci: 1) determining the variability of airfares

among all available flight pairs and exploring how this variability interacts with the average

airfares at a given airport; 2) identifying the existence of any symmetries, or asymmetries

133

in patterns of airfare over time and across space using data on arrivals and departures for

O/D pairs; 3) highlight the implications of these fare structures to develop a deeper

understanding of passenger accessibility, and regional transport forces in the United States.

6.3 Data and Methods

6.3.1 Airline and Airport Data

The Airline Origin and Destination Survey (DB1B) database is a 10% sample of

airline tickets from reporting carriers provided by the Bureau of Transportation Statistics

(BTS, 2013). The DB1B database includes three unique datasets: 1) DB1B Coupon, 2)

DB1BMarket, and, 3) DB1BTicket (Table 3.1). The DB1BMarket dataset contains records on air traffic flow attributes, including origins and destinations. The DB1BTicket and

Market datasets, include ~139 million itineraries over the span of 2002-2013. These are

the data we use for analysis in this chapter.

Airport data were collected from the Aviation Support Tables Master Coordinate

database (BTS, 2015). This table includes airport locations (latitude and longitude), as

well as historical information on each airport, including any changes to airport codes. The

data associated with the FAA airport typology were also collected (FAA, 2015b).

Specifically, six different categories are used for analysis, including large, medium and

small hubs, primary non-hub, non-primary commercial service and general aviation

airports.

134

6.3.2 Index Development

In an effort to disentangle average fares, fare variability and the pricing associated

with individual flight pairs, a suite of indices is constructed. Consider the following

notation:

index for origin airports (entire set denoted )

𝑖𝑖 index for destination airports (entire set denoted𝐼𝐼 ) 𝑗𝑗 index for itineraries on a specific flight pair (entire𝐽𝐽 set denoted ) 𝑚𝑚 average one-way fare from to 𝑀𝑀 𝑎𝑎𝑖𝑖𝑖𝑖 directional indicator for a flight𝑖𝑖 𝑗𝑗 pair (-1: unidirectional, only i to j available; 1: 𝐾𝐾 𝑖𝑖 − 𝑗𝑗 bi-directional, both to and to available) one-way fare at itinerary𝑖𝑖 𝑗𝑗 from𝑗𝑗 𝑖𝑖 to 𝑓𝑓𝑖𝑖𝑖𝑖𝑖𝑖 round-trip indicator from 𝑚𝑚 to at 𝑖𝑖itinerary𝑗𝑗 (0: No; 1: Yes) 𝑟𝑟𝑖𝑖𝑖𝑖𝑖𝑖 total number of flights departing𝑖𝑖 𝑗𝑗 from 𝑚𝑚 𝑁𝑁 𝑖𝑖 𝑖𝑖 More formally, for each of the binary elements in the indices:

(0,1) (6.1)

𝑖𝑖𝑖𝑖𝑖𝑖 𝑟𝑟 ∈ ( 1,1) (6.2)

𝑖𝑖−𝑗𝑗 𝐾𝐾 ∈ −

Average fares are calculated using a simple accounting procedure, where:

/( ) = (6.3) ∑𝑚𝑚∈𝑀𝑀 𝑓𝑓𝑖𝑖𝑖𝑖𝑖𝑖 1+𝑟𝑟𝑖𝑖𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖 𝑀𝑀 𝑎𝑎

The three indices are formally specified as:

/( ) = (6.4) 𝑓𝑓𝑖𝑖𝑖𝑖𝑖𝑖 1+𝑟𝑟𝑖𝑖𝑖𝑖𝑖𝑖 𝑎𝑎𝑎𝑎𝑎𝑎𝑖𝑖 ∑𝑗𝑗∈𝐽𝐽 ∑𝑚𝑚∈𝑀𝑀 𝑁𝑁𝑖𝑖

135

= (( ) ) / (6.5) 𝑓𝑓𝑖𝑖𝑖𝑖𝑖𝑖 2 𝑣𝑣𝑣𝑣𝑣𝑣𝑖𝑖 ∑𝑗𝑗∈𝐽𝐽 ∑𝑚𝑚∈𝑀𝑀 1+𝑟𝑟𝑖𝑖𝑖𝑖𝑖𝑖 − 𝑎𝑎𝑎𝑎𝑎𝑎𝑖𝑖 𝑁𝑁𝑖𝑖 � ( ) = × ( ) (6.6) 𝑎𝑎𝑖𝑖𝑖𝑖−𝑎𝑎𝑗𝑗𝑗𝑗 𝑣𝑣𝑣𝑣𝑣𝑣𝑖𝑖−𝑗𝑗 𝑎𝑎𝑖𝑖𝑖𝑖+𝑎𝑎𝑗𝑗𝑖𝑖 ⁄2 𝐾𝐾𝑖𝑖−𝑗𝑗

Subject to

( ) 0 ( ) (6.7) 𝑎𝑎𝑖𝑖𝑖𝑖−𝑎𝑎𝑗𝑗𝑗𝑗 𝑎𝑎𝑖𝑖𝑖𝑖+𝑎𝑎𝑗𝑗𝑗𝑗 ⁄2 ≥ 0 (6.8)

𝑖𝑖−𝑗𝑗 𝑣𝑣𝑣𝑣𝑣𝑣 ≥

The first index (6.4), or , captures the average airfare at each airport using all records

𝑖𝑖 in the DB1BTicket and 𝑎𝑎𝑎𝑎𝑎𝑎Market databases. The second index (6.5), or , captures the

𝑖𝑖 variability of airfares on all available flight pairs for a given origin. Finally,𝑣𝑣𝑣𝑣𝑣𝑣 index (6.6), or , tabulates the variation in airfares between departing and arriving flights for each

𝑖𝑖−𝑗𝑗 flight𝑣𝑣𝑣𝑣𝑣𝑣 pair. It is this metric that will be used to identify asymmetries in itinerary pricing between origins and destinations. One additional quirk worth mentioning with this index is found within the denominator. Because of the growing popularity of one-way fares in

the DB1BTicket database, we divide known round-trip itineraries for each origin by a value

of 2.52 This ensures that the underlying variability can be compared without bias for each

airport. Finally, there are two simple constraints for the index, namely, values for

𝑣𝑣𝑣𝑣𝑟𝑟𝑖𝑖−𝑗𝑗

52 The DB1B ticket database includes a roundtrip indicator to differentiate one-way and round-trip tickets. If it is a round-trip ticket, the difference between the two-direction segments is masked. Thus, for index (6), all flight pairs that have bidirectional flights ( and ) are extracted for identifying pricing asymmetries. Since each direction of these itineraries might have both one-way and round-trip fares associated with them, round-trip fares are divided 𝑖𝑖by− 2𝑗𝑗 to make𝑗𝑗 − 𝑖𝑖all the itineraries comparable.

136

the average fare variability calculations cannot be negative, nor can the resulting variability

index.

6.4 Results

Figure 6.1 highlights the locations of each airport (n=407) used for this analysis,

with symbology that reflects FAA airport designations (e.g., large hub, small hub, etc.).

As detailed previously, the DB1B Market and Ticket databases from 2002-2013 are used

to evaluate the spatiotemporal dynamics of airfares in the United States.53 It is important

to note that the measurements are based on the entire air transportation network in the U.S.,

including Alaska, Hawaii, and outlying territories. However, because of space constraints,

the primary focus of our analysis is the contiguous 48 states.

53 During pre-processing of the databases, outliers were removed, including itineraries that were $0, or greater than $5,000.

137

Figure 6.1: Airports in the United States with commercial traffic, 2013

6.4.1 Average Airfares

Figure 6.2 presents the index results for inflation adjusted, average one-way airfares between 2002 and 2013. In this case, lighter shades of blue indicate lower fares. Symbol size corresponds to flight activity (counts) within the FAA typology. The average fare for all airports (in aggregate) was cheapest in 2009 ($270.15), dropping approximately $23 from 2008 ($293.38). This was an interesting period for the commercial air transport industry, which was victimized by a huge increase crude oil prices between January 2007 and July 2008 (Grubesic and Wei, 2015a). Specifically, the U.S. crude oil first purchase price increased from $49.32 per barrel (January 2007) to $128.08 per barrel (July 2008).

Ultimately, oil prices dropped to $34.14 in February 2009. These savings in fuel costs

138

seem to have been passed on to passengers in 2009, but once again in 2010, as fuel prices

continued to increase, so too did airfares.

Figure 6.2: Inflation-adjusted average airfares by airport category, 2002-2013

One of the most interesting trends displayed in the data is the propensity for

medium hubs to exhibit significantly cheaper average fares than large hubs. For example,

even during 2009, where overall prices were the least expensive, the average price for a

one-way ticket at medium hubs was priced $31.37 less than large hubs ($216.78 versus

$248.15). Again, this relative discount at medium versus large hubs was consistent,

throughout all years of the study. As detailed by Lee and Luengo-Prado (2005), there are two general reasons for these differences. First, the average fares at large hubs tends to reflect differences in passenger mix (business and leisure travelers). With larger numbers of business travelers, who tend to place a high value on the time savings and convenience available from non-stop service to a wide variety of destinations, large hubs reflect the

139 premiums paid for these business itineraries. Lee and Luengo-Prado (2005) suggest that even when passenger mix is controlled for, large hubs are simply more expensive, especially for first-class, business-class and unrestricted coach tickets. Consider, for example, the most consistently expensive hubs in the U.S. from 2002-2013: Cincinnati

(CVG), Newark (EWR), New York (JFK), Dulles (IAD), San Francisco (SFO) and Los

Angeles (LAX). With the exception of CVG, which only maintained a handful of international routes (and was later dehubbed by Delta), the remaining cities functioned as large international gateways and offered many long-haul domestic itineraries. Ironically,

CVG was the most expensive of all. The itinerary yield54 at CVG was $0.408 in 2012, while the yields were less than $0.35 at EWR, JFK, IAD, SFO and LAX. Lee and Luongo-

Prado (2005) submit that large hubs in smaller cities, such as Cincinnati, routinely serve thinner routes using smaller and more costly aircraft. Thus, lacking the economies of scale present in larger cities, costs are inflated for hubs like CVG.

This is not to say, however, that all hubs are expensive. Many provide consistently low average fares, including Orlando (MCO), Tampa (TPA), Fort Lauderdale (FLL), Fort

Myers (RSW) and Chicago (MDW). Of course, all of these cities are served Southwest

Airlines, a low-cost leader in the United States for many years. The mere presence of

Southwest Airlines has been shown to both increase passenger traffic and lower overall fares for a region (Morrison, 2001; Vowles, 2001; Pitfield, 2008).

Finally, there are two categories with significantly higher average fares: non- primary and general aviation airports. Where the former is concerned, the Essential Air

54 Itinerary yield is the ratio between fare paid and itinerary miles flown.

140

Service program plays an important role. Specifically, 62 of the 70 communities in these

two categories have carriers subsidized by the EAS program. However, for many airports,

the costs associated with charter services are also important to consider. For example,

Topeka, Kansas (FOE) has struggled for many years to maintain consistent commercial

service (Wei and Grubesic, 2015b). Located less than 80 miles from Kansas City

International (MCI), FOE is squarely within the MCI traffic shadow (Taaffe, 1959; Goetz,

1992) of MCI. Although Topeka has, at various points in time, had commercial service to

Kansas City and Chicago (ORD), and has benefitted from occasional charter service to

casino resort communities in Nevada, commercial operations have been consistently

insoluble for the region.55 Page, Arizona (PGA) is another interesting case. PGA is an

EAS community, which had service to both Denver and Phoenix via Great Lakes Airlines

between 2002 and 2013. PGA is also a gateway to many of the national monuments and

parks in the Southwest. As a result, companies such as Westwind Air Service provide

charter flights and tours to the Grand Canyon, Lake Powell and Monument Valley. To put

the costs of these charters in perspective, a typical Grand Canyon flight tour, originating in

Page, AZ, is currently priced at approximately $400 per person (Westwind, 2015).

Figure 6.3 summarizes national trends in average itinerary pricing and its associated

spatial distribution. Specifically, Figure 6.3 highlights a cost surface for one-way fares for each year between 2002 and 2013. Each surface was interpolated using a geostatistical approach known as kriging (Oliver and Webster, 1990), similar to the method used by

55 Until 2003, Topeka was receiving EAS service. But, EAS eligibility was cancelled because subsidies over $200 per passenger are not allowed for communities within 210 miles of the nearest large or medium hub airport (Grubesic et al., 2012)

141

Fuellhart et al., (2016).56 There are no major surprises here. Significant portions of the

Mountain West (e.g., Colorado, Wyoming and Montana), the Upper Great Plains (e.g.,

North Dakota, South Dakota, and Nebraska) and several locales in the Southwest (e.g.,

Arizona and New Mexico) and Southeast (e.g., Georgia and North Carolina) had

consistently high average one-way fares for each year of the analysis. In particular, it is

the Southeast that corresponds to the original pockets of pain identified by Goetz (2002)

and the U.S. GAO (1996), and although the relative “pain” of expensive fares fluctuates

somewhat, it remains a relatively expensive region.

Figure 6.3: Average one-way fares at airports, 2002-2013

56 It is important for readers to notice that the legend breaks for each year are different. Thus, caution must be exercised when attempting to compare the spatial patterns of fares between years. Instead, it is more appropriate to compare spatial patterns and variations in pricing within each map and/or year.

142

At the same time, there are several relatively inexpensive regions, including the

Great Lakes Megalopolis (Doxiadis, 1968), Florida, the Pacific Northwest and

metropolitan Colorado (the I-25 Corridor). For example, Denver (DEN) and Fort Collins

(FNL) fares averaged $239.40 and $158.44, respectively, during 2013. However, their

smaller neighboring airports Yampa Valley (HDN), Eagle County (EGE) and Aspen-Pitkin

County (ASE), all of which are proximal to world-class skiing destinations, averaged over

$350.

Taking this analysis one step further, Figure 6.4 illustrates the inflation-adjusted

average fares on flight pairs (using the FAA typology) for all origins/destinations. Again,

2009 is an important touchstone, with the majority of pair categories exhibiting their lowest

itinerary prices for the study period. For example, the average itinerary price for all airport

pairs was $313.50. One of the least surprising results of the flight pair analysis is that

itineraries between any two large and/or medium hubs were consistently low. Again, this

is where the hub-and-spoke system really benefits passengers. The combination of high

demand, interhub linkages and large aircraft work together to manifest economies of scale

and lower average prices. Pairs that connect smaller airports or non-hubs, which are often focused on more rural or remote locations, are more expensive.

143

Figure 6.4: Inflation-adjusted average one-way fares on flight pairs, 2002-2013

6.4.2 Variability of Airfares: A Pairwise Analysis

Table 6.1 lists a subset of airports that were displayed consistently high levels of variability for their service pairs from 2002-2013. The airports can be divided into two basic groups: large coastal hubs and smaller airports located in popular tourist destinations.

Where the large hubs are concerned, high fare variability is generally reflective of the variation in service profiles for commercial carriers, which include both long-haul, interhub linkages and short-haul, regional spoke connections. The wide variety of service classes available at large hubs (e.g., first, business, coach, etc.) also contribute to variability in fares. For example, consider JFK airport, which maintained service to 341 destinations in the contiguous U.S., Alaska, Hawaii and outlying territories in 2013. When this variety of destinations is mixed between nearly a dozen domestic carriers, the resulting one-way fares

144

varied dramatically. This holds true for LAX, SFO, EWR and all of the other major coastal

gateways listed in Table 6.1.

The smaller, tourism-focused airports in Table 6.1, including Eagle, Colorado,

Telluride, CO, Aspen, CO, and Jackson Hole, WY are all major resort destinations, particularly during winter months when the alpine sports season is in full swing. The one exception to this in Table 6.1 is Monterey, CA, which offers a more temperate summer destination with a wide variety of seasonal activities along Monterey Bay. The underlying thread that connects all of these smaller airports to their associated fare variability are these seasonal trends. Fares tend to spike during peak tourism months and then deflate during the off-season.

145

Table 6.1: Airports with high fare variability consistently from 2002-2013 Origin 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Aspen, CO (ASE) 309.56 327.07 298.72 285.38 278.36 314.41 334.67 304.66 317.40 384.71 362.28 384.11 Arlington, VA (DCA) 242.47 219.67 205.99 183.97 205.77 215.07 235.78 219.97 220.10 231.42 235.59 231.65 Eagle, CO (EGE) 332.93 323.83 281.07 284.49 287.35 307.62 336.33 280.85 281.22 380.96 318.45 323.11 Newark, NJ (EWR) 281.73 270.71 246.14 210.26 223.48 238.86 253.00 233.61 263.71 290.41 289.62 301.11 Dulles, VA (IAD) 318.57 298.24 266.18 222.52 241.86 257.55 279.89 252.02 260.86 292.41 298.72 294.15 Jackson, WY (JAC) 267.18 248.34 274.54 228.10 260.29 282.66 286.53 246.79 242.98 271.32 280.30 289.62 New York, NY (JFK) 389.49 349.87 301.52 301.47 324.45 362.66 354.61 324.34 344.36 370.09 375.59 371.47 Los Angeles, CA 288.01 270.28 244.08 226.11 242.61 261.66 278.91 252.20 274.38 300.66 291.35 285.95 (LAX) New York, NY (LGA) 277.31 255.95 246.26 210.20 235.71 242.62 259.36 227.61 234.62 250.57 244.15 239.24 Miami, FL (MIA) 251.90 241.57 226.79 194.56 211.10 226.93 243.17 219.81 224.90 244.95 240.57 233.48 Monterey, CA (MRY) 242.68 219.94 213.16 206.59 219.01 221.55 244.03 216.77 228.01 273.30 244.33 230.28 San Francisco, CA 329.18 303.85 267.84 247.66 265.13 282.01 295.52 258.36 277.21 305.04 304.16 298.84 (SFO) Telluride, CO (TEX) 360.44 257.44 339.66 234.97 289.82 314.84 344.78 375.51 372.30 411.38 237.43 356.21

146

Figure 6.5 shows the spatial distribution of air fare variability for all available flight pairs and their origin airports between 2002 and 2013. Readers need to exercise caution when interpreting these results, because the variability range is different for each year.

That said, there are several trends worth noting. First, the Intermountain West displayed consistently high variability. Given the results detailed in Table 6.1, this is not a huge surprise. There are distinctive seasonal trends for travel to the smaller airports in this region that are driven by winter tourism and alpine sports. Southeastern portions of the U.S., extending northward through Appalachia also display high fare variability on flight pairs.

This closely corresponds to the original pockets of pain identified by Goetz (2002) and the

U.S. GAO (1996; 1999). Conversely, variability was low in portions of Nebraska, Kansas and Montana. Portions of West Texas and southeastern New Mexico also displayed low variability.

147

Figure 6.5: Variability of fares for all available flight pairs at origins, 2002-2013

6.4.3 Interaction of Average Fares and Fare Variability

Given the information provided in Figures 20 and 22, it is clear that there is a complex correspondence between air fare prices and their associated variability in both time and space for the United States. In an effort to disentangle these interactions, four basic types of “pockets” are identified from the geostatistical analysis.

Pockets of Chaos (High average fares, high variability)

A large, continuous swath of territory that includes portions of Montana, Wyoming,

Utah, Colorado, New Mexico and Arizona display high average fares and relatively high

variability. These pockets of chaos reflect both the seasonal nature of tourism to airports

148

such as Aspen (ASE), Telluride (TEX) and others, but also to warmer winter destinations

such as Phoenix (PHX). For example, the average fare for ASE during 2013 was $382.49,

but its variability was 314.41. Why is there such a huge variability in fares for ASE? More

than 65% of itineraries from were priced over $500 during 2013, with the most notable

price spikes during the winter quarter. As one might expect, travelers also pay a premium

for journeys to Phoenix during the winter months, but the prices for identical itineraries

drop dramatically during the summer when temperatures in Phoenix are routinely above

43 degrees Celsius (110 Fahrenheit).

The southeastern U.S. is identified as another pocket of chaos, including the states of Mississippi, Alabama, Georgia and Louisiana, as well as portions of Florida and West

Virginia. This, in some ways, contradicts the findings of the U.S. GAO (1996; 1999) and

Goetz (2002), which suggested that the region suffered from a persistently high average

fares when compared to the remainder of the United States. Here, although the results

certainly indicate that this region suffers from relatively high average fares, there is an

unexpected, underlying variability to their pricing. Again, there is likely a seasonal

component to this, with visitors seeking more temperate weather along the Gulf Coast

during the winter months. But, this variability also reflects the presence of many smaller,

more market-responsive airports scattered throughout the region, including the small hubs

of Jackson, MS (JAN), Baton Rouge, LA (BTR), Montgomery, AL (MGM), Birmingham,

AL (BHM), Pensacola, FL (PNS) and other primary non-hubs such as Albany, GA (ABY),

Columbus, GA (CSG), Dothan, Alabama (DHN). In every instance, the smaller hubs are

served by multiple carriers and function as spoke cities for larger hubs. This also means

that pricing at these spoke cities are subject to the typical revenue management strategies

149 employed by larger carriers (McGill and Van Ryzin, 1999). Interestingly, many of these airports have extremely high average itinerary yields (Table 6.2), suggesting that they are cash-cows for the legacy carriers, regardless of the underlying pricing variability. In short, they are very profitable routes.

150

Table 6.2: Airports in the Southeast with high itinerary yields by year Airport State 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 MCN GA 0.31446 0.34146 0.32953 0.34528 0.36598 0.36633 0.34743 0.61122 0.60190 0.43041 0.55528 0.55396 CSG GA 0.34406 0.39365 0.39457 0.44859 0.53842 0.52249 0.53594 0.54774 0.47131 0.45383 0.42675 0.48233 DHN AL 0.31139 0.32382 0.31580 0.32865 0.39485 0.44883 0.46437 0.46947 0.45003 0.42517 0.42782 0.41831 AHN GA 0.34990 0.38495 0.30252 0.32237 0.39420 0.40438 0.44350 0.44070 0.50512 0.51820 0.77180 NA GLH MS 0.27035 0.31525 0.30057 0.32756 0.37746 0.35225 0.45216 0.38054 0.34098 0.31341 0.34283 0.33232 MKL TN 0.31504 0.32782 0.33728 0.34832 0.32963 0.37453 NA 0.37550 0.40186 0.60836 NA NA MEI MS 0.30109 0.29711 0.29807 0.31434 0.34715 0.36135 0.33358 0.36685 0.38390 0.36763 0.38051 0.39274 HHH SC 0.30750 0.34946 0.30212 0.29890 0.39686 0.34225 0.34820 0.34930 0.34342 0.38755 0.39153 0.41031 TXK AR 0.28456 0.27745 0.29539 0.28449 0.30709 0.30219 0.36647 0.34556 0.37705 0.37234 0.37836 0.29533 SHV LA 0.23905 0.25011 0.26113 0.26727 0.29882 0.29568 0.35935 0.34434 0.36488 0.36656 0.35155 0.32517 CAE SC 0.32373 0.32675 0.29167 0.27063 0.33928 0.34311 0.39207 0.34179 0.35745 0.36574 0.36971 0.36081 ABY GA 0.34829 0.37062 0.36967 0.35411 0.38993 0.38501 0.35151 0.34139 0.34642 0.38389 0.37946 0.38147 TUP MS 0.29281 0.31924 0.33043 0.30841 0.34670 0.38488 0.42624 0.33993 0.33325 0.39592 0.41227 0.39523 MGM AL 0.24833 0.25672 0.26855 0.29270 0.31617 0.32620 0.35137 0.31818 0.29861 0.31320 0.32810 0.35016 MEM TN 0.29888 0.29876 0.29495 0.30137 0.32020 0.30494 0.34644 0.31750 0.33005 0.36053 0.36985 0.32517 GSP SC 0.31122 0.31646 0.29498 0.27158 0.33974 0.34400 0.38219 0.31735 0.33947 0.29770 0.30355 0.31440 CLT NC 0.38065 0.39747 0.35854 0.35380 0.35390 0.33644 0.37137 0.31652 0.34996 0.38876 0.38398 0.38471 LCH LA 0.27915 0.32818 0.33195 0.31941 0.32363 0.28099 0.37988 0.31469 0.32230 0.36056 0.39855 0.33793 AGS GA 0.32471 0.32222 0.31668 0.32739 0.39396 0.37526 0.35353 0.31248 0.29691 0.29468 0.30271 0.30861 National Average 0.25752 0.25815 0.24790 0.25356 0.27882 0.26941 0.28973 0.25091 0.26232 0.28058 0.28964 0.28978

151

Pockets of Pain (High average fares, low variability)

One of the more interesting results found when comparing fares and their variability

are the newly defined pockets of pain. These are some of the most remote and/or difficult

to reach geographic locations in the United States and include portions of the Upper Great

Plains (Kansas, Nebraska and eastern Montana), and parts of the Desert Southwest (New

Mexico). Here, service is both expensive and infrequent. For example, consider Miles

City, MT (MLS), Glendive, MT (GDV), Glasgow, MT (GGW), Kearney, Nebraska (EAR)

and Carlsbad, New Mexico (CNM). All of the carriers that serve these airports were

subsidized under the Essential Air Service program during the study period. For these airports, this means that carriers must provide at least two daily round-trip flights, six days

per week, without more than one intermediate stop on each flight to the connecting hub

airport. Unlike other small hubs or primary non-hubs in the U.S., airfares are not set using the typical, industry revenue management models. Rather, the USDOT provides a subsidy to cover the difference between a carrier’s projected costs of operations and its expected passenger revenues. Built into the subsidy is a profit margin equal to five percent of total operating expenses (Black, 2015). Given the relative geographic remoteness of these airports, the presence of the EAS program, and the inherent inflexibility in setting fares,

airports in these pockets of pain are always expensive, service is infrequent, and it is nearly

impossible to fly anywhere without at least one or two connections.

Pockets of Bliss (Low average fares, low variability)

The inverse to pockets of pain are pockets of bliss. These regions exhibit low

average fares and low fare variability. Between 2002 and 2013, the most notable pocket

of bliss was in/around the Chicago metropolitan area. Here, the presence of Chicago

152

O’Hare (ORD) and Chicago Midway (MDW), which is served by Southwest Airlines,

combine to make the region a consistent bargain for price-conscious travelers. Central

Florida, including Clearwater (PIE), Sanford (SFB), Tampa (TPA) and Orlando (MCO) is

also a major pocket of bliss for the United States. Tourism is extremely important to

regional economy in Florida, and places like Disney World (in the Orlando area) and the

beaches in the Tampa/Clearwater/St. Petersburg area are all-season destinations. As a result, prices are competitive and seats are generally available, regardless of the time of year. Interestingly, there are also “islands of bliss” in the Intermountain West. As detailed previously, much of this region suffers from high fares and high variability, but both

Denver (DEN) and Fort Collins (FNL) had low average fares and low fare variability. The

same can be said for many of the airports in the Pacific Northwest, including Seattle, (SEA)

and Portland (PDX).

Pockets of Diversity (Low average fares, high variability)

As one might expect, given the sheer scale and scope of commercial air transport

operations at the large coastal hubs in the United States, their relative assortment of carriers,

seating classes available and routes served spawn pockets of diversity, which are

exemplified by lower average fares, but high fare variability. Many of these pockets

correspond to multi-airport regions (MAR) (Fuellhart et al., 2013). For example, JFK,

EWR, LGA (all in New York Metro area) form a MAR, as do IAD, DCA and BWI (all in

D.C Metro area), SFO and OAK (in the San Francisco Bay Area), MIA, FLL, and PBI (S.

Florida) and LAX, SNA, ONT (in/around Los Angeles). As previously established in this

paper, we know that the larger hubs can be expensive, but they are still a bargain when

compared to the fares found in many of the smaller airports around the country (See Figure

153

6.2). We also know that there is a smoothing process that occurs when an airport serves

more than 100+ destinations in the U.S., as most of these do. In these cases, some fares

are expensive (e.g.., first class between JFK and LAX), but others are not (e.g. coach class

between BWI and CMH). All of these inputs are eventually captured by the average fare

metric, but their variations are reflected in the variability metric.

6.4.4 Putting it All Together

To provide some additional perspective, Figure 6.6 displays a scatterplot of average

fare prices and average fare variability for airports in 2013. However, it is important to

remember that this visualization is aspatial. In other words, airports located on the East

Coast can easily overlap with airports on the West Coast in the scatterplot. So, the results

are not directly comparable to the geostatistical analysis generated in Figures 20 and 22.

The takeaway from Figure 6.6 is that all four types of pockets are displayed here, with

some dimensional consistency. Specifically, blue points are representative of pockets of

chaos, red for pockets of pain, cyan for pockets of diversity and green for pockets of bliss.57

Figure 6.6 also displays a best-fit line (blue) and the gray area around the line

corresponds to a 95% confidence interval. The advantage for using the regression line here

is that it helps us to explore the functional form of the relationship between average fares

and fare variability. In this instance, the coefficient of determination (0.5239) suggests that

average fares are linearly related to fare variability. In other words, as average fares

increase, so too does variability. This should not be a surprise to readers, but the ability to

57 Again, these typologies were generated with geostatistical information from the kriging process, not from the scatterplot itself.

154 identify the airports where this dependence exists is helpful, even if the complex matrix of factors that lead to this dependence are not explored in this paper. We leave this for future work.

Figure 6.6: Average fare prices vs. average fare variability for airports, 2013

Figure 6.7 provides a complimentary visualization, displaying the spatial patterns associated with each pocket. As detailed previously, there are distinct regional trends, with large pockets of pain across the Great Plains states. Pockets of chaos can be found in the

Intermountain West, as well as the Southeastern U.S. Pockets of bliss and diversity are most closely aligned with multi-airport regions on both coasts.

155

Figure 6.7: Spatial patterns of airports associated with each pocket, 2013

Asymmetries in the Landscape

One of the more interesting findings in this analysis is that a large percentage of individual flight pairs in the United States have no significant variation between departing and arriving flights. These flight pairs are considered “fare symmetric”. For example,

13,076 out of 30,705 (42.59%) unique flight pairs58 for 2013 displayed a fare variation of less than 0.1 for departing/arriving flights. If the threshold is arbitrarily increased to 0.3, nearly 78% (28,817/30,708) of the flight pairs in the United States would be considered fare symmetric. This is remarkably high, given the wide variety of airports, carriers and origin/destinations throughout the country.

58 Each flight pair for symmetry/asymmetry analysis has both departure and arrival flights.

156

However, there are airports with fares that are asymmetric. Three criteria are used

to generate a subset of airports for the analysis of fare asymmetries. First, fare asymmetries

are detected by establishing cutoff points using a false non-discovery rate (FNDR)

approach (Genovesee and Wasserman, 2002; Strimmer, 2008). Specifically, fare

variations are measured and benchmarked against empirically derived (rather than arbitrary)

cutoff points. Those that fall below the benchmark are considered symmetric. Those above

the benchmark are considered asymmetric. For this analysis, the cutoff point for 2002 was

0.411 and for 2013 it was 0.304.59 Second, the airport must have a more than 50 passengers

for the year. This ensures that equipment movements (i.e., moving planes from one airport

to another) are not considered. Lastly, an airport must have more than 10 operations per

year. This effectively removes periodic and infrequent service from charter airlines.

The descriptive statistics for these results are not particularly compelling and due

to space constraints, they are not detailed. However, Figure 6.8 illustrates the spatial

distribution of fare asymmetries for the United States in both 2002 and 2013. Most airports

have at least a handful of asymmetries in both years, regardless of their size and the patterns

of asymmetric fares are relatively consistent across both time periods. Of note, are the

clusters of asymmetries in resort destinations, including the alpine resorts of Colorado,

coastal destinations along the Gulf of Mexico and the major business/tourism destinations

along both the Atlantic and Pacific coasts. Once again, an interesting example is Eagle,

Colorado (EGE) which provides access to both Vale and Beaver Creek ski resorts. EGE

59 This method assumes that a null model has truncated normal, truncated student t or a truncated correlation density (Strimmer, 2008). For the data used in this paper, the density of fare variation between departure and arrival flights for the same itinerary follows a truncated normal distribution.

157 displays a large number of asymmetric fare structures, given its relatively smaller size. In short, it is much cheaper to fly into EGE than it is to fly out. This is a general trend for most of the asymmetries, but as readers are certainly aware, all of this is contingent upon supply and demand, as well as the seasonality constraints associated with the O/D pairs in this analysis.

Figure 6.8: Spatial distribution of fare asymmetries for the United States in 2002 and 2013

158

6.5 Discussion and Conclusion

There are several facets of the results worth further discussion. First, this paper

highlights an important difference in the metrics used for exploring air fares and their

spatiotemporal distributions. Specifically, there is a difference between average fares and

fare variability for flight pairs, throughout the United States. There is no doubt that the use

of an average fare metric is useful for examining airport performance, but it fails to capture

the myriad of nuanced fare differences embedded within. For example, it is relatively easy

for two airports (A, B) to exhibit the same average fare. However, if Airport A has a range

of fares embedded within its flight pairs (e.g., some high, some low), while the fares for

Airport B are relatively homogeneous, the use of an average metric fails to capture these

differences – differences which are very important for understanding the underlying dynamics of air travel in the United States. Thus, when average fares for airport pairs and their variability are used simultaneously, a simple but informative typology of airports

(pockets of pain, chaos, diversity and bliss) are uncovered.

Second, the spatiotemporal patterns of air fares and fare variability in the United

States are relatively consistent between 2002 and 2013, even though the air transport

industry was undergoing massive changes, including a hangover from the attacks of 9/11,

carrier bankruptcies and consolidation, rising fuel prices and shifts in operational strategies.

To be clear, there is significant unevenness in the spatiotemporal distribution of air fares

throughout the U.S., but the landscape associated with these disparities is relatively

consistent and predictable. For example, the pockets of pain detailed in this paper have a

decidedly rural/remote geographic composition, and exhibit high average fares and low fare variability. Sadly, this is a direct result of deregulation in 1978 and has been a

159

consistent feature of the commercial air transport landscape for nearly 40 years. Rural

airports are more expensive, often requiring subsidies from the Essential Air Service

program to maintain operations, their carrier choice is limited and service frequency is low.

It is important to remember that the EAS program was never meant to exist in perpetuity.

It was designed as a temporary “bridge” so that rural areas could take some additional time

for establishing their local commercial markets (Wei and Grubesic, 2015b). Unfortunately,

this has not happened for many rural communities, and they are completely dependent upon

EAS for maintaining commercial service.

Third, the pockets of chaos detailed by this chapter share a similarly disconcerting

fate. Also primarily rural in geographic composition and located in the Intermountain West,

these airports exhibit higher fares and high fare variability. As a result, the permanent non- tourist residents of these communities, as well as the non-tourism oriented businesses, face a serious challenge when it comes to air travel. Moreover, because many of these communities are located in areas with especially mountainous topographies (e.g.,

Colorado), seeking alternative airports or travel choices is difficult, especially during the winter months when fares are highest and alternative transportation options (e.g., driving) can be most dangerous.

One of the more interesting results of this chapter is the identification of pockets of bliss. There are very few of these in the United States, but those that do exist have a unique combination of local factors that contribute to consistently low fares and low fare variability. Pockets of bliss are typically found in multi-airport regions, they are frequently located in popular tourist destinations and Southwest Airlines typically serves at least one of the airports in the region. Again, although the work presented in this paper is largely

160

exploratory, it is clear that these core local factors make for a very favorable (and

competitive) air transport market.

A final question worth considering is whether or not there are policy mechanisms that can be used to mitigate the identified pockets of pain or chaos for the United States.

Alas, there is no simple answer to this question. In many ways, air transport is a decidedly local problem. Local markets cannot be so thin that maintaining commercial operations for a carrier yields no return on investment, or worse, losses. In the United States, it is clear that some rural markets will never be soluble because there is simply not enough demand to attract commercial carriers. This, is a local problem. In many cases, the Essential Air

Service Program has proven to be a good mechanism for connecting thin markets to the air

transport network, but this is expensive, costing tax payers over $200 million per year.

Further, the benefits of these subsidies are not conferred equally to the public – only to

those traveling in/out of the EAS communities. This is not necessarily a huge problem, as

it certainly helps rural economies, but as detailed by Grubesic and Matisziw (2011),

Matisziw et al. (2012) and Grubesic et al. (2016), the EAS program needs a significant

overhaul for improving community eligibility guidelines, reducing market cannibalization

and reconfiguring EAS hubbing practices to make the program more efficient. Over time,

these changes might have a positive effect on the pockets of pain, especially if the program

is able to lower fares in these markets and stimulate some type of local competition.

However, this will not help the pockets of chaos. These are communities where EAS is

not needed, because the local markets are already served by commercial carriers, without

subsidies. Again, there is no easy solution to mitigating the pockets of chaos, air fare caps

are unrealistic, especially when there is a large, extremely affluent subsection of the market

161

that is willing to pay premium prices for travel to these destinations. Spurring competition

may be one way to help reduce fares in these locations. For example, the Small Community

Air Service Development Grant (SCASDG) program (Whitman, 2014) can provide

funding for these communities and the local airport authorities to expand carrier presence

through marketing and/or lobbying efforts.

The consequences of deregulation have been long-lasting. Many small airports in

rural and remote communities continue to exhibit high fares and low variability, or high

fares and high variability – both of which are relatively punishing outcomes for these

smaller locales. Beyond these rural problems, however, there are locations where air travel

is relatively inexpensive (i.e., pockets of bliss) or where the service offerings are highly

differentiated (i.e., pockets of diversity). These are relatively positive outcomes, also

fueled by deregulation.

Taking rural airports and airfares by itineraries into consideration, this chapter developed three types of indices to illustrate a more accurate and a more complete picture of spatial distribution of air fares in the United States. More importantly, identified airports, regions, four types of pockets and asymmetric patterns can provide insights to planners and policy makers on airport performance evaluation and development.

162

7. CONCLUSION

7.1 Summary

This research provided a quantitative contribution to the literature to explore and

improve airport access and accessibility. In general, this research reviews current literature

in airport access and accessibility, identifies related topics and issues and analyzes them

from a small spatial perspective, regional level with one single airport, to a large spatial

perspective, national level with all commercial airports in the United States including

large/medium/small hubs and rural airports. This research focused on major network pieces

individually and then evaluated the complete air transport network. Specifically, the first three studies explored and evaluate airport access and accessibility on individual parts of

U.S. air transport network. Major hubs and rural airports are two important parts of an air transport network. Study one explored dehubbing process at one hub airport and evaluated associated impacts on access and accessibility at this hub airport. Studies two and three evaluated airport access and network accessibility by focusing on rural, isolated airports.

Finally, the last study explored the spatiotemporal distribution of air fares on a complete network to achieve a more holistic understanding of air transport accessibility. Although previous studies have examined access and accessibility from different foci with a variety of analytical approaches and models, the air transport networks are dynamic on different spatial contexts. Specifically, the locational access to an airport and its associated accessibility to an air transport network can vary substantially over time. Also, some aspects of airport access and accessibility are still unexplored. Deregulation in airline

industry has significant impacts on changing structure of air transport network, changing

roles, functions and service levels at airports and interaction with local communities,

163

airport competition and market leakage, all of which influence airport access and

accessibility in the United States. That said it is both important and necessary to examine

the spatiotemporal dynamics of airport access and accessibility on different locational

contexts, and explore and reveal emerging patterns and phenomenon. As a whole, the four

substantive chapters comprising this research not only enhance the understanding of airport

access and accessibility, air transport network and associated various factors and impacts,

but provide meaningful implications for policy and transportation planning.

Chapter 3 highlighted the operational, market and geographic factors that

contributed to CVG’s decline and used basic exploratory data analysis and LASSO regression to provide perspective and deepen our understanding of this process. Results

suggested that a combination of commercial carrier strategies, operational efficiencies, hub

structures, network topologies and regional competition contributed to deterioration of

CVG. Furthermore, dehubbing process negatively impacts airport access and accessibility

especially connections to small airports.

Focusing on rural, isolated airports receiving federal subsidies, chapter 4 developed

three regression models to examine the geographic, demographic, socio-economic, and local business determinants that influence EAS service characteristics. Results indicated that distance to the nearest hubs, local population and subsidy levels are critical factors that impact EAS service levels. More importantly, these factors identified in this chapter have impacts on access and accessibility of EAS airports. All the findings discussed in this chapter can be used as a guide for transportation policy making, planning and recommendation.

164

Chapter 5 took a different tract to evaluate relative accessibility and use this information to develop a typology of rural airports, combined with graph-theoretic and statistical techniques. Results showed that the combination of high-resolution operations data from airports, network analysis approaches and statistical clustering techniques allows for an extremely detailed snapshot of airports and their associated context and also provided planning officials and policy-makers an exploratory data analysis framework and insights for improving decision making and helping craft higher-quality transport legislation.

The final substantive chapter developed three types of indices to explore the uneven spatiotemporal distribution of air fares, by airports and associated flight pairs. The results highlight a fascinating spatiotemporal landscape of air fares and fare variability for the

United States, through the use of three simple and easily reproducible indices, and identify regions with high/low levels of airport access and accessibility. The reported results are a good first step in deepening our understanding of fare variability and asymmetries for local markets.

In general, deregulation of airline industry and consequential airport competition have resulted in a dynamic air transportation system and associated issues that have been studied in this research. In light of the findings from the four studies, these issues that operational hubs and rural airports have faced cannot be simply solved by the mechanism created by deregulation because the nature of the mechanism is competition that created these issues and negatively impact them at some point. That is why it is necessary to establish post regulations to limit the power of carriers and restrict their monopolistic behaviors in a deregulated industry. More importantly, these regulations can ensure that

165

company strategies made by carriers not only focus on supply side but consider demand

side. As focused on in this research, post regulations also need to consider hubs, rural airports and a complete network, respectively.

As the consequence of deregulation and airport competition, dominant carriers can dehub their operational hubs only considering their needs and not taking a passenger’s perspective and a government’s perspective into consideration. Local residents may lose

the accessibility if the dehubbing issue is not handled very effectively. More holistically,

fewer major hubs would become more important to a national air transport network. It can

make entire network more vulnerable to unexpected events such as hurricane storms or

snow storms. A regulation can limit the exclusive gate agreement between airport and

carrier to allow other carriers (e.g. LLCs) to enter the market. Providing service by new

entrants can help a dehubbed airport still keep the level of accessibility with similar even

lower level of cost (ticket price). Also, the regulation can specify benefits to attract LLCs

to enter a dehubbed airport and associate market. It can further help the whole air transport network more balanced and less vulnerable than simply dehubbing hubs and excluding new entrants by dominant carriers.

As discussed in previous chapters, there are two federal programs to help small, rural airports in the United States. EAS program is facing a dilemma of maintaining activity of rural airports. It has been criticized for years due to its high cost. On one side, EAS program subsidizes eligible rural airports to keep active in the national transportation network. On the other side, EAS tries to reduce cost. Saving money and helping rural airports become two conflicting purposes of the EAS program. The dilemma and critical voice are questioning the program and its underlying characteristic. At some point,

166

although some airports really need subsidies to maintain connections, but the program

cannot help rural airports and associated local communities develop and improve

commercial air services. This program should focus more efforts on specific groups of

airports sharing similar characteristics, not generally all rural airports. Compared with frequently academic exposure of the EAS program, the SCASDP only receives little attention but its essential characteristic is very important to some rural airports that are seeking externally financial support to improve air services. The two government programs may be integrated to focus on different levels of rural airports and provide more effective help.

7.2 Limitations

7.2.1 Air traffic data

Itinerary fare is an important air traffic attribute that was used widely in this research. The DB1B ticket database includes a roundtrip indicator to differentiate one-way and round-trip tickets. However, the fare difference between the two-direction segments is masked if it is a round-trip ticket because the only data reported is the final booked, round- trip price. When one-way air fares are concerned and studied, the averaging fare for each round-trip flight cannot capture the fare variation between segments. The fare details would certainly provide more information about the fare structure in the United States.

7.2.2 Business data

ZIP business pattern (ZBP) data used in this research were only available from 2004 to 2012 so the data could not cover the whole temporal spectrum for analysis. In addition,

167

the ZBP data do not provide employee information such as actual number of employees

within each industry category so employee counts used in Chapter 3 were estimated in

terms of a parameter that indicates a range of employees for each sector in each ZIP. This

lack of data completeness and details prevents us from identifying a more thorough

interaction between airports and associated local markets.

7.3 Future Research

While the work presented in this research increases our knowledge and

understanding of the airport access and accessibility, including changing roles and

functions of individual airports and associated pairs, dynamic air transport network,

deregulation and competition, and airport performance, it shows opportunities for future work beyond the scope of this research. A potential extension to the analysis in Chapter 3 will focus on more airports that have suffered dehubbing process in the United States, such as Memphis, TN (MEM) for Delta, Cleveland, OH (CLE) for United and Pittsburgh, PA

(PIT) for US Airways (now merged with American). Based on more dehubbed or dehubbing airports, similarity or difference of factors and associated process that contributed service decline can be captured between these airports to help better understand the dehubbing process in the United State and changes of access and accessibility in these local markets. Although Chapter 4 identified several important factors determining service levels in EAS communities, statistical averages used for regression models in this chapter can often obscure important variations in data and this might be the case for passenger loads at certain EAS airports. As a result, the extension of the analysis will focus more efforts on detailed information from spatial perspective and temporal perspective. There

168

may be some EAS routes that are more easily and proﬁtably served than others in the

system. It will be necessary to explore these subtleties between EAS routes. Also, there

may be strong seasonal trends in passenger loads for some locations, which are effectively

lost by the use of the aggregate and/or averaged BTS data for the program. There seasonal

variations will be further examined by using actual load information for each flight. The

extension of the analysis in Chapter 5 will explore detailed characteristics of airports within

each cluster. Considering geographic location, demographic information and local business

pattern, each rural airport may have unique characteristics that cannot be identified by the

three general clusters. Case studies may provide more detailed information and a better

understanding of rural airports. In Chapter 6, the extension of the analysis will focus on

spatiotemporal distribution of air fares for each important carrier (e.g. United, Delta or

Southwest) to find similarity or difference and further evaluate their impacts on national

air fare distribution by capturing the variations and comparing them. Further, additional

confirmatory statistical analysis will be incorporated on the specific factors that may be

driving these differences.

Current research focuses efforts on studying airports and associated flight pairs in

the United States. Future work will also include analysis for airport access and accessibility

at international, intercontinental and even global levels. As reviewed in Chapter 2, airport connectivity and accessibility have been measured by using traditional graph-theoretic or statistical network-based approaches, as well as newly developed metrics that emphasize the importance of individual destination(s). However, most current studies limit their focus to individual airports within a certain region (e.g. Asia, Europe or the United States), and these analyses fail to include basic elements of airport interaction. In this context, it will

169 be important to explore connectivity and accessibility of airports and accessibility trends for airport pairs at a larger spatial scale, such as global level, from a spatiotemporal perspective.

170

BIBLIOGRAPHY

Adler, N., & Berechman, J. (2001a). Measuring airport quality from the airlines’ viewpoint: an application of data envelopment analysis. Transport Policy, 8(3), 171-181.

Adler, N., & Berechman, J. (2001b). Evaluating optimal multi-hub networks in a deregulated aviation market with an application to Western Europe. Transportation Research Part A: Policy and Practice, 35(5), 373-390.

Agrawal, A., Cockburn, I., & McHale, J. (2006). Gone but not forgotten: knowledge flows, labor mobility, and enduring social relationships. Journal of Economic Geography, 6(5), 571- 591.

Alkaabi, K. A., & Debbage, K. G. (2007). Air passenger demand and skilled labor markets by US metropolitan area. Journal of Air Transport Management, 13(3), 121-130.

Allroggen, F., & Malina, R. (2014). Do The Regional Growth Effects of Air Transport Differ Among Airports?, Journal of Air Transport Management, 37, 1–4.

American West Airlines [AWA]. (2015). America West in Columbus. Retrieved 02/25/15, from http://www.awa-history.org/?q=about_awa/columbus.

Anselin, L. (1988). Spatial Econometrics. Kluwer, London.

Anselin, L. (1992). Spatial data analysis with GIS: an introduction to application in the social sciences. National Center for Geographic Information and Analysis, Santa Barbara, CA.

Anselin, L. (1999). Interactive techniques and exploratory spatial data analysis. Geographical Information Systems: Principles, Techniques, Management and Applications, eds., P. Longley, M. Goodchild, D. Maguire, and D. Rhind. Cambridge: Geoinformation Int.

Ashby, A. (2013). Will Delta’s departure open window for lower fares? Memphis Business Journal. Retrieved 02/25/15, from http://tinyurl.com/mpjozdt.

Associated Press [AP]. (2008). Delta, Northwest deal gets DOJ approval. Retrieved 02/25/15, from http://tinyurl.com/m8eufbz.

Babikian, R., Lukachko, S.P., & Waitz, I.A. (2002). The historical fuel efficiency characteristics of regional aircraft from technological, operational, and cost perspectives. Journal of Air Transport Management, 8(6), 389–400.

Bania, N., Bauer, P. W., & Zlatoper, T. J. (1998). US air passenger service: A taxonomy of route networks, hub locations, and competition. Transportation Research Part E: Logistics and Transportation Review, 34(1), 53-74.

Barrett, S. D. (2000). Airport competition in the deregulated European aviation market. Journal of Air Transport Management, 6(1), 13-27.

171

Başar, G., & Bhat, C. (2004). A parameterized consideration set model for airport choice: an application to the San Francisco Bay area. Transportation Research Part B: Methodological, 38(10), 889-904.

Bazargan, M., & Vasigh, B. (2003). Size versus efficiency: a case study of US commercial airports. Journal of Air Transport Management, 9(3), 187-193.

Ben Abda, M. (2010). Evolution of domestic traffic and fares at the top 200 US airports between 1990 and 2008 (Doctoral dissertation, Massachusetts Institute of Technology).

Bhadra, D. (2003). Demand for air travel in the United States: bottom-up econometric estimation and implications for forecasts by origin and destination pairs. Journal of Air Transportation, 8(2), 19-56.

Bhadra, D. (2009). Race to the bottom or swimming upstream: performance analysis of US airlines. Journal of Air Transport Management, 15(5), 227-235.

Black, F. M. (2015). Essential Air Service. Regional Airline Association. Retrieved 02/16/16, from http://www.raa.org/?page=EAS.

Bohl, P. (2013). The consequences of de-hubbing for airports and tourism-a case study. European Journal of Business and Management, 5(25), 168-178.

Borenstein, S., & Rose, N. L. (1991). Competition and price dispersion in the US airline industry (No. w3785). National Bureau of Economic Research.

Bowen, J. (2000). Airline hubs in Southeast Asia: national economic development and nodal accessibility. Journal of Transport Geography, 8(1), 25-41.

Bowen, J. (2002). Network change, deregulation, and access in the global airline industry*. Economic Geography, 78(4), 425-439.

Bowen, J.T. (2010). The Economic Geography of Air Transportation: Space, Time, and the Freedom of the Sky. Routledge.

Bowen, J. T., & Laroe, C. (2006). Airline networks and the international diffusion of severe acute respiratory syndrome (SARS). The Geographical Journal, 172(2), 130-144.

Brueckner, J. K. (2003). Airline traffic and urban economic development. Urban Studies, 40(8), 1455-1469.

Brueckner, J. K., & Spiller, P. T. (1991). Competition and mergers in airline networks. International Journal of Industrial Organization, 9(3), 323-342.

Brueckner, J. K., Lee, D., & Singer, E. S. (2013). Airline competition and domestic US airfares: A comprehensive reappraisal. Economics of Transportation, 2(1), 1-17.

Bureau of Transportation Statistics [BTS]. (2000). Retrieved 02/24/15, from http://tinyurl.com/ob72o7uN.

172

Bureau of Transportation Statistics [BTS]. (2010). Research and Innovative Technology Administration. Retrieved 04/18/2012, from http://www.transtats.bts.gov/Data_Elements.aspx?Data=1.

Bureau of Transportation Statistics [BTS]. (2011). National Transportation Atlas Databases 2013. Retrieved 05/09/2015, from http://www.rita.dot.gov/bts/sites/rita.dot.gov.bts/files/publications/national_transportation _atlas_database/2013/points.html.

Bureau of Transportation Statistics [BTS]. (2013). Airline Origin and Destination Survey. Retrieved 05/08/2015, from http://www.transtats.bts.gov/DatabaseInfo.asp?DB_ID=125&DB_Name=Airline%20Ori gin%20and%20Destination%20Survey%20(DB1B).

Bureau of Transportation Statistics [BTS]. (2015). Aviation Support Tables: Master Coordinates. Retrieved 02/16/16, from http://www.transtats.bts.gov/Fields.asp?Table_ID=288.

Burghouwt, G. (2007). Airline network development in Europe and its implications for airport planning: Ashgate Publishing, Ltd.

Burghouwt, G., & de Wit, J. (2005). Temporal configurations of European airline networks. Journal of Air Transport Management, 11(3), 185-198.

Burghouwt, G., & Hakfoort, J. (2001). The evolution of the European aviation network, 1990– 1998. Journal of Air Transport Management, 7(5), 311-318.

Burghouwt, G., Hakfoort, J., & van Eck, J. R. (2003). The spatial configuration of airline networks in Europe. Journal of Air Transport Management, 9(5), 309-323.

Button, K., & Taylor, S. (2000). International air transportation and economic development. Journal of Air Transport Management, 6(4), 209-222.

Button, K. & Yuan, J. (2013). Airfreight Transport and Economic Development: An Examination of Causality, Urban Studies, 50, 329–340.

Casper-Natrona County International Airport [CNCIA]. (2012). Airport Receives Small Community Air Service Development (SCASDP) grant. Retrieved 05/15/25, from http://tinyurl.com/n3jam9k.

Cho, W., Windle, R. J., & Hofer, C. (2012). Route Competition in Multi-Airport Cities: An Analysis of US Air Fares. Transportation Journal, 51(3), 265-288.

Chou, Y.-H. (1993). Airline deregulation and nodal accessibility. Journal of Transport Geography, 1(1), 36-46.

Cohas, F.J., Belobaba, P.P., & Simpson, R.W. (1995). Competitive fare and frequency effects in airport market share modeling. Journal of Air Transport Management, 2, 33–45.

CVG. (2013). From humble beginnings…. to international hub. Retrieved 02/24/15, from

173

http://tinyurl.com/pganhedN.

Cunningham, L. F., & Eckard, E. W. (1987). US small community air service subsidies: essential or superfluous? Journal of Transport Economics and Policy, 255-277.

Debbage, K. G. (1994). The international airline industry: globalization, regulation and strategic alliances. Journal of Transport Geography, 2(3), 190-203.

Debbage, K. G. (1999). Air transportation and urban-economic restructuring: competitive advantage in the US Carolinas. Journal of Air Transport Management, 5(4), 211-221.

Debbage, K. G., & Delk, D. (2001). The geography of air passenger volume and local employment patterns by US metropolitan core area: 1973–1996. Journal of Air Transport Management, 7(3), 159-167.

Delta Airlines. (2005a). Delta Airlines Reaches Definitive Agreement to Sell Atlantic Southeast Airlines to SkyWest. Retrieved 02/24/15, from http://tinyurl.com/kxqhsrw.

Delta Airlines. (2005b). Delta Airlines Strengthens Domestic Hubs, Offers Customers More International Choices with Winter Schedule. Retrieved 02/24/15, from http://tinyurl.com/q7r7jww.

Delta Airlines. (2005c). Delta Airlines Steps Up Transformation Plan to Accelerate Path to Profitability. Retrieved 02/24/15, from http://tinyurl.com/qx23vss.

Demetsky, M., & Korf, J. (1980). Specification of rapid transit access mode requirements. Journal of Advanced Transportation, 14(3).

Derudder, B., & Witlox, F. (2005). An appraisal of the use of airline data in assessing the world city network: a research note on data. Urban Studies, 42(13), 2371-2388.

Derudder, B., Witlox, F., & Taylor, P. J. (2007). US cities in the world city network: Comparing their positions using global origins and destinations of airline passengers. Urban Geography, 28(1), 74-91.

Derudder, B., Taylor, P., Ni, P., De Vos, A., Hoyler, M., Hanssens, H., Bassens, D., Huang, J., Witlox, F., Shen, W., & Yang, X. (2010). Pathways of change: shifting connectivities in the world city network, 2000–08. Urban Studies, 47(9), 1861–1877.

Dias, M. (1999). Cincinnati/Louisville airport fare war heats up. Cincinnati Post. Retrieved 11/02/15, from http://tinyurl.com/ox278wx.

Dobruszkes, F., Lennert, M., & Van Hamme, G. (2011). An analysis of the determinants of air traffic volume for European metropolitan areas. Journal of Transport Geography, 19(4), 755-762.

Doganis, R. (2001). The airline industry in the 21st century. Londres, 1º Ed. Routlege.

Doxiadis, C. A. (1968). The emerging great lakes megalopolis. Proceedings of the IEEE, 56(4), 402-424.

174

Dresner, M., Lin, J. S. C., & Windle, R. (1996). The impact of low-cost carriers on airport and route competition. Journal of Transport Economics and Policy, 30, 309-328.

Dresner, M. (2006). Leisure versus business passengers: Similarities, differences, and implications. Journal of Air Transport Management, 12(1), 28-32.

Drucker, P. F. (1994). The Age of Social Transformation, The Atlantic Online. Retrieved 05/15/15, from http://www.theatlantic.com/politics/ecbig/soctrans.htm.

Efron, B., Hastie, T., Johnstone, I., & Tibshirani, R. (2004). Least angle regression. The Annals of statistics, 32(2), 407-499.

Ellis, R. H., Bennett, J. C., & Rassam, P. R. (1974). Approaches for improving airport access. Journal of Transportation Engineering, 100(Proc. Paper 10737).

Energy Information Administration [EIA]. (2013) North Dakota Oil Production Reaches New High In 2012, Transported by Trucks and Railroads. Retrieved 05/15/15, from http://www.eia.gov/todayinenergy/detail.cfm?id=10411.

ESRI. (1998). How to: CFCC code lookup table. Retrieved 05/11/2015, from http://support.esri.com/cn/knowledgebase/techarticles/detail/11966.

ESRI. (2010a). Data and Maps for ArcGIS. Retrieved 02/24/15, from http://tinyurl.com/bsbekgtN.

ESRI. (2010b). Business Locations and Business Summary. Retrieved 05/11/2015, from http://doc.arcgis.com/en/esri-demographics/data/business.htm.

ESRI. (2010c). Demographics. Retrieved 05/11/2015, from http://doc.arcgis.com/en/esri- demographics/.

ESRI. (2014). USA Zip Code Boundaries. Retrieved 05/11/2015, from http://www.arcgis.com/home/item.html?id=8d2012a2016e484dafaac0451f9aea24.

Federal Aviation Administration [FAA]. (2012). Airport Categories. Retrieved 5/7/2015, from http://www.faa.gov/airports/planning_capacity/passenger_allcargo_stats/categories/.

Federal Aviation Administration [FAA]. (2015a). A Brief History of the FAA. Retrieved 5/7/2015, from https://www.faa.gov/about/history/brief_history/.

Federal Aviation Administration [FAA]. (2015b). Passenger Boarding (Enplanement) and All- Cargo Data for U.S. Airports – Previous Year. Retrieved 02/16/16, from http://www.faa.gov/airports/planning_capacity/passenger_allcargo_stats/passenger/previous_ years/.

Federal Register. (2012). Qualifying Urban Areas for the 2010 Census. Retrieved 05/15/15, from https://www.federalregister.gov/articles/2012/03/27/2012-6903/qualifying-urban-areas- for-the-2010-census.

175

Feiertag, J. (2002). Cincinnati Business Courier. Retrieved 02/25/15, from http://tinyurl.com/pfyzjutN.

Fleming, D. K., & Hayuth, Y. (1994). Spatial characteristics of transportation hubs: centrality and intermediacy. Journal of Transport Geography, 2(1), 3-18.

Flynn, J., & Ratick, S. (1988). A multiobjective hierarchical covering model for the essential air services program. Transportation Science, 22(2), 139-147.

Freeman, L. C. (1979). Centrality in social networks conceptual clarification. Social networks, 1(3), 215-239.

Friedman, J., Hastie, T., & Tibshirani, R. (2010). Regularization paths for generalized linear models via coordinate descent. Journal of statistical software, 33(1), 1.

Friedman Municipal Airport Authority [FMAA]. (2013). Friedman Memorial Airport Hailey, Idaho. Retrieved 05/15/15, from http://www.iflysun.com/assets/pdfs/airservice- studies/SUN-SCASDP-Application-2013.pdf.

Freeman, L. C. (1979). Centrality in Social Networks Conceptual Clarification. Social networks, 1, 215–239.

Fuellhart, K. (2003). Inter-metropolitan airport substitution by consumers in an asymmetrical airfare environment: Harrisburg, Philadelphia and Baltimore. Journal of Transport Geography, 11(4), 285-296.

Fuellhart, K. (2007). Airport catchment and leakage in a multi-airport region: The case of Harrisburg International. Journal of Transport Geography, 15(4), 231-244.

Fuellhart, K., O’Connor, K., & Woltemade, C. (2013). Route-level passenger variation within three multi-airport regions in the USA. Journal of Transport Geography, 31, 171-180.

Fuellhart, K., Ooms, K., Derudder, B., & O'Connor, K. (2016). Patterns of US air transport across the economic unevenness of 2003–2013. Journal of Maps, 1-5.

Gandel, S. (2011). Want to Make More than a Banker? Become a Farmer! Time. Retrieved 04/18/2012, from http://www.time.com/time/magazine/article/0,9171,2080767,00.html.

Genovese, C., & Wasserman, L. (2002). Operating characteristics and extensions of the false discovery rate procedure. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 64(3), 499-517.

Gerardi, K. S., & Shapiro, A. H. (2009). Does competition reduce price dispersion? New evidence from the airline industry. Journal of Political Economy, 117(1), 1-37.

Geurs, K. T., & VAn Wee, B. (2004). Accessibility evaluation of land-use and transport strategies: review and research directions. Journal of Transport Geography, 12(2), 127- 140.

Gillen, D., & Lall, A. (1997). Developing measures of airport productivity and performance: an

176

application of data envelopment analysis. Transportation Research Part E: Logistics and Transportation Review, 33(4), 261-273.

Gillies, A. (2004). Flying Pork. Forbes. Retrieved 04/12/12, from http://www.forbes.com/2004/01/21/cz_ag_0121beltway.html.

Glacier Park International Airport. (2014) Major Cities Served. Retrieved 05/15/15, from http://www.iflyglacier.com/cities-served.html.

Goetz, A. R. (1992). Air passenger transportation and growth in the US urban system, 1950– 1987. Growth and Change, 23(2), 217-238.

Goetz, A. R. (2002). Deregulation, competition, and antitrust implications in the US airline industry. Journal of Transport Geography, 10(1), 1-19.

Goetz, A. R., & Graham, B. (2004). Air transport globalization, liberalization and sustainability: post-2001 policy dynamics in the United States and Europe. Journal of Transport Geography, 12(4), 265-276.

Goetz, A. R., & Sutton, C. J. (1997). The geography of deregulation in the US airline industry. Annals of the Association of American Geographers, 87(2), 238-263.

Goetz, A., & Vowles, T. (2000). Pockets of Pain’across the deregulated landscape: The geography of US airline fares and service in the 1990s. Paper presented at the annual meeting of the Association of American Geographers, Pittsburgh.

Goetz, A. R., & Vowles, T. M. (2009). The good, the bad, and the ugly: 30 years of US airline deregulation. Journal of Transport Geography, 17(4), 251-263.

Goff, B. L. (2005). Estimating determinants of passenger air service to small markets. Urban Studies, 42(3), 557-565.

Graham, A. (2008). Managing Airports: An International Perspective, 3rd ed. Taylor & Francis: United Kingdom.

Grubesic, T. H. (2006). On the application of fuzzy clustering for crime hot spot detection. Journal of Quantitative Criminology, 22(1), 77-105.

Grubesic, T. H., & Matisziw, T. C. (2011). A spatial analysis of air transport access and the essential air service program in the United States. Journal of Transport Geography, 19(1), 93-105.

Grubesic, T. H., & Matisziw, T. C. (2012). World cities and airline networks. The international handbook of globalization and world cities, 97-116.

Grubesic, T. H., & Wei, F. (2012). Evaluating the efficiency of the Essential Air Service program in the United States. Transportation Research Part A: Policy and Practice, 46(10), 1562- 1573.

Grubesic, T. H., & Wei, F. (2013). Essential air service: a local, geographic market perspective.

177

Journal of Transport Geography, 30, 17–25.

Grubesic, T., & Zook, M. (2007). A ticket to ride: Evolving landscapes of air travel accessibility in the United States. Journal of Transport Geography, 15(6), 417-430.

Grubesic, T. H., Matisziw, T. C., & Murray, A. T. (2012). Assessing geographic coverage of the essential air service program. Socio-Economic Planning Sciences, 46(2), 124-135.

Grubesic, T. H., Matisziw, T. C., & Zook, M. A. (2008). Global airline networks and nodal regions. GeoJournal, 71(1), 53-66.

Grubesic, T. H., Matisziw, T. C., & Zook, M. A. (2009). Spatio-temporal fluctuations in the global airport hierarchies. Journal of Transport Geography, 17(4), 264-275.

Grubesic, T. H., Murray, A. T., & Matisziw, T. C. (2013). A strategic approach for improving rural air transport in the United States. Transport Policy, 30, 117-124.

Grubesic, T., Wei, R., Murray, A., & Wei, F. (2014). Essential Air Service in the United States Exploring Strategies to Enhance Spatial and Operational Efficiencies. International Regional Science Review, 1012690214532653.

Grubesic, T. H. M., Timothy C. (2012). World cities and airline networks. The International Handbook of Globalization and World Cities, 97-116.

Guimera, R., Mossa, S., Turtschi, A., & Amaral, L. N. (2005). The worldwide air transportation network: Anomalous centrality, community structure, and cities' global roles. Proceedings of the National Academy of Sciences, 102(22), 7794-7799.

Haining, R., Wise, S., & Ma, J. (1998). Exploratory spatial data analysis. Journal of the Royal Statistical Society: Series D (The Statistician), 47(3), 457-469.

Halpern, N., & Bråthen, S. (2011). Impact of airports on regional accessibility and social development. Journal of Transport Geography, 19(6), 1145-1154.

Harvey, G. (1987). Airport choice in a multiple airport region. Transportation Research Part A: General, 21(6), 439-449.

HBR (Harvard Business Review), 2009. Managing Across Distance in Today’s Economic Climate: The Value of Face-to-Face Communication. Retrieved 04/12/12, from http://www.hbr.org/hbrmain/resources/pdfs/marketing/15426_HBRAS_BA_Report.pdf.

Hess, S., & Polak, J. W. (2005). Mixed logit modelling of airport choice in multi-airport regions. Journal of Air Transport Management, 11(2), 59-68.

Hess, S., & Polak, J. W. (2006). Exploring the potential for cross-nesting structures in airport- choice analysis: a case-study of the Greater London area. Transportation Research Part E: Logistics and Transportation Review, 42(2), 63-81.

Hewlett, J. (2012). Blue Grass Airport ‘more fortunate’ than others in a changing airline climate. Lexington Herald-Leader. Retrieved 02/25/15, from http://tinyurl.com/6sfsmb8N.

178

IATA. (2014). IATA Codes. Retrieved 05/15/15, from https://www.iata.org/services/pages/codes.aspx.

IATA. (2015). Growth and Development. Retrieved 05/07/2015, from http://www.iata.org/about/Pages/history_3.aspx.

Innes, J. D., & Doucet, D. H. (1990). Effects of access distance and level of service on airport choice. Journal of Transportation Engineering, 116(4), 507-516.

Internal Revenue Service [IRS]. (2013). Communications and Air Transportation Taxes. Retrieved 05/15/15, from http://www.irs.gov/publications/p510/ch04.html.

Irwin, J. (2008). New terminal taking shape at Detroit Metropolitan Airport. USA Today. Retrieved 02/25/15, from http://tinyurl.com/qcb4cvm.

Ishii, J., Jun, S., & Van Dender, K. (2009). Air travel choices in multi-airport markets. Journal of Urban Economics, 65(2), 216-227.

Isserman, A. M. (2005). In the national interest: Defining rural and urban correctly in research and public policy. International Regional Science Review, 28(4), 465-499.

Ito, H., & Lee, D. (2005). Assessing the impact of the September 11 terrorist attacks on US airline demand. Journal of Economics and Business, 57(1), 75-95.

Ivy, R. L. (1993). Variations in hub service in the US domestic air transportation network. Journal of Transport Geography, 1(4), 211-218.

James, G., Witten, D., Hastie, T., & Tibshirani, R. (2013). An introduction to statistical learning. New York: Springer.

Johansson, B. (1993). Infrastructure, accessibility and economic growth. International Journal of Transport Economics/Rivista internazionale di economia dei trasporti, 131-156.

Kaemmerle, K. C. (1991). Estimating the demand for small community air service. Transportation Research Part A: General, 25(2), 101-112.

Kanafani, A., & Abbas, M. S. (1987). Local air service and economic impact of small airports. Journal of Transportation Engineering, 113(1), 42-55.

Kaufman, L. R., & Rousseeuw, P. J. (1990). Finding Groups in Data: An Introduction to Cluster Analysis. New York: John Wiley.

Keane, M. J. (1984). Accessibility and urban-growth rates: Evidence for the Irish urban system. Economic and Social Review, 15, 125-139.

Keeble, D., Owens, P. L., & Thompson, C. (1982). Regional accessibility and economic potential in the European Community. Regional Studies, 16(6), 419-432.

Krugman, P. (1994). Complex landscapes in economic geography. The American Economic

179

Review, 84(2), 412-416.

Lederer, P. J., & Nambimadom, R. S. (1998). Airline network design. Operations Research, 46(6), 785-804.

Lee, W. (2011). Delta pays back $175 million loan to MAC, sparking job fears. Retrieved 02/25/15, from http://www.startribune.com/business/132537688.html.

Lee, D., & Luengo-Prado, M. J. (2005). The Impact of Passenger Mix on Reported" Hub Premiums" in the US Airline Industry. Southern Economic Journal, 372-394.

Leonhardt, D. (2001). Small Jets’ Big Stake in a Strike. New York Times. Retrieved 02/25/15, from http://tinyurl.com/manw85z.

Lian, J. I., & Rønnevik, J. (2011). Airport competition–Regional airports losing ground to main airports. Journal of Transport Geography, 19(1), 85-92.

Lin, F.B. (1977). Airport choice in low demand region. Transportation Engineering Journal, 103, 711–727.

Lin, J. (2012). Network analysis of China’s aviation system, statistical and spatial structure. Journal of Transport Geography, 22, 109-117.

Liu, Z.-J., Debbage, K., & Blackburn, B. (2006). Locational determinants of major US air passenger markets by metropolitan area. Journal of Air Transport Management, 12(6), 331-341.

Llanos, M. (2011). $6 billion-a-year ethanol subsidy dies – but wait there’s more. Retrieved 04/12/12, from http://www.tinyurl.com/dyxy5jh.

Lockhart, R., Taylor, J., Tibshirani, R. J., & Tibshirani, R. (2014). A significance test for the lasso. The Annals of Statistics, 42(2), 413-468.

Malmberg, A., & Maskell, P. (2002). The elusive concept of localization economies: towards a knowledge-based theory of spatial clustering. Environment and Planning A, 34 (3), 429– 449.

Marcucci, E., & Gatta, V. (2011). Regional airport choice: consumer behaviour and policy implications. Journal of Transport Geography, 19(1), 70–84.

Mathur, V. K., & Stein, S. H. (2005). Do amenities matter in attracting knowledge workers for regional economic development?*. Papers in Regional Science, 84(2), 251-269.

Matisziw, T. C., & Grubesic, T. H. (2010). Evaluating locational accessibility to the US air transportation system. Transportation Research Part A: Policy and Practice, 44(9), 710- 722.

Matisziw, T. C., Lee, C.-L., & Grubesic, T. H. (2012). An analysis of essential air service structure and performance. Journal of Air Transport Management, 18(1), 5-11.

180

McCartney, S. (2014). The most expensive airports to fly to. Wall Street Journal. Retrieved 02/25/15, from http://www.wsj.com/articles/SB10001424052702303980004579576012567760336.

McGill, J. I., & Van Ryzin, G. J. (1999). Revenue management: Research overview and prospects. Transportation science, 33(2), 233-256.

Mecklenborg, J. (2011). Lagging air service at CVG may mean more trouble than just Chiquita’s departure. UrbanCincy.com. Retrieved 02/25/15, from http://tinyurl.com/naf4krj.

Morrison, S. A. (2001). Actual, adjacent, and potential competition: Estimating the full effect of Southwest Airlines. Journal of Transport Economics and Policy, 239-256.

Murray, A. T., & Grubesic, T. H. (2002). Identifying Non-Hierarchical Spatial Clusters. International Journal of Industrial Engineering, 9, 86–95.

Murray, A. T., Grubesic, T. H., & Wei, R. (2014). Spatially significant cluster detection. Spatial Statistics, 10, 103-116.

Mutzabaugh, B. (2013). Delta to pull plug on Memphis hub after Labor Day. USA Today. Retrieved 02/25/15, from http://tinyurl.com/mwyvcl9.

Nader, G. A. (1981). The delineation of a hierarchy of nodal regions by means of higher-order factor analysis. Regional Studies, 15(6), 475-492.

National Park Service [NPS]. (2013). National Park Service Press Release. Retrieved 05/15/15, from http://www.nps.gov/news/release.htm?id=1568.

Neal, Z. (2010). Refining the air traffic approach to city networks. Urban Studies, 47(10), 2195- 2215.

Nystuen, J. D., & Dacey, M. F. (1961). A graph theory interpretation of nodal regions. Paper presented at the Papers of the Regional Science Association.

O'Connor, K. (2003). Global air travel: toward concentration or dispersal? Journal of Transport Geography, 11, 83–92.

O'Connor, K., & Fuellhart, K. (2015). The fortunes of air transport gateways. Journal of Transport Geography, 46, 164–172.

O'Kelly, M. E. (1998). A geographer's analysis of hub-and-spoke networks. Journal of Transport Geography, 6(3), 171-186.

Oliver, M. A., & Webster, R. (1990). Kriging: a method of interpolation for geographical information systems. International Journal of Geographical Information Systems, 4(3), 313- 332

Openshaw, S. (1991). Developing appropriate spatial analysis methods for GIS. Geographical Information Systems: principles and applications, 1, 389-402.

181

Oster, C., & Strong, J. (2001). Predatory practices in the US Airline Industry. US Department of Transportation, Office of the Assistant Secretary for Aviation and International Affairs. Available from

Oum, T. H., Yan, J., & Yu, C. (2008). Ownership forms matter for airport efficiency: A stochastic frontier investigation of worldwide airports. Journal of Urban Economics, 64(2), 422- 435.

Özcan, İ. Ç. (2014). Economic contribution of essential air service flights on small and remote communities. Journal of Air Transport Management, 34, 24-29.

Pantazis, N., & Liefner, I. (2006). The impact of low-cost carriers on catchment areas of established international airports: the case of Hanover Airport, Germany. Journal of Transport Geography, 14, 265–272.

Pels, E., Nijkamp, P., & Rietveld, P. (2000). Airport and airline competition for passengers departing from a large metropolitan area. Journal of Urban Economics, 48(1), 29-45.

Pels, E., Nijkamp, P., & Rietveld, P. (2001). Airport and airline choice in a multiple airport region: an empirical analysis for the San Francisco Bay Area. Regional Studies, 35(1), 1- 9.

Pels, E., Nijkamp, P., & Rietveld, P. (2003). Access to and competition between airports: a case study for the San Francisco Bay area. Transportation Research Part A: Policy and Practice, 37(1), 71-83.

Percoco, M. (2010). Airport activity and local development: evidence from Italy. Urban Studies.

Phillips, O. R., Weatherford, L. R., Mason, C. F., & Kunce, M. (2005). Passenger leaks and the fate of small community air service. Economic Inquiry, 43(4), 785-794.

Pilcher, J. (2002). Comair's History Sometimes Turbulent. Cincinnati Enquirer. Retrieved 02/25/15, from http://tinyurl.com/q7bxglo.

Pilcher, J. (2003). Curse of High Fares has Economic Upside. Cincinnati Enquirer. Retrieved 02/25/15, from http://tinyurl.com/pk7sstx.

Pilcher, J. (2010).Why CVG lost half of all flights. Cincinnati Enquirer. Retrieved 02/25/15, from http://tinyurl.com/lct6ltx.

Pitfield, D. E. (2008). The Southwest effect: A time-series analysis on passengers carried by selected routes and a market share comparison. Journal of Air Transport Management, 14(3), 113-122.

Prager, R. A., & Hannan, T. H. (1998). Do substantial horizontal mergers generate significant price effects? Evidence from the banking industry. Journal of Industrial Economics, 433-452.

Rasker, R., Gude, P. H., Gude, J. A., & Van den Noort, J. (2009). The economic importance of air travel in high-amenity rural areas. Journal of Rural Studies, 25(3), 343-353.

182

Redondi, R., Malighetti, P., & Paleari, S. (2011). Hub competition and travel times in the worldwide airport network. Journal of Transport Geography, 19(6), 1260-1271.

Redondi, R., Malighetti, P., & Paleari, S. (2012). De-hubbing of airports and their recovery patterns. Journal of Air Transport Management, 18(1), 1-4.

Redondi, R., Malighetti, P., & Paleari, S. (2013). European Connectivity: The Role Played by Small Airports. Journal of Transport Geography, 29, 86–94.

Reed, W. E. (1970). Indirect connectivity and hierarchies of urban dominance. Annals of the Association of American Geographers, 60(4), 770-785.

Reynolds-Feighan, A. J. (1995). European and American approaches to air transport liberalisation: some implications for small communities. Transportation Research Part A: Policy and Practice, 29(6), 467-483.

Reynolds-Feighan, A. (2000). The US airport hierarchy and implications for small communities. Urban Studies, 37(3), 557-577.

Reynolds-Feighan, A. (2001). Traffic distribution in low-cost and full-service carrier networks in the US air transportation market. Journal of Air Transport Management, 7(5), 265-275.

Reynolds-Feighan, A. J. (2007). Competing Networks, Spatial and Industrial Concentration in the US Airline Industry. Spatial and Economic Analysis, 2, 237–257.

Reynolds-Feighan, A., & McLay, P. (2006). Accessibility and attractiveness of European airports: A simple small community perspective. Journal of Air Transport Management, 12(6), 313-323.

Rietveld, P., & Brons, M. (2001). Quality of hub-and-spoke networks; the effects of timetable coordination on waiting time and rescheduling time. Journal of Air Transport Management, 7(4), 241-249.

Rodríguez-Déniz, H., Suau-Sanchez, P., & Voltes-Dorta, A. (2013). Classifying airports according to their hub dimensions: an application to the US domestic network. Journal of Transport Geography, 33, 188–195.

Ryerson, M.S., & Kim, H. (2013). Integrating airline operational practices into passenger airline hub definition. Journal of Transport Geography, 31, 84–93.

Schleisman, N., & Dulle, B. (2014). United Airlines to End Topeka Service. KSNT.com, Retrieved 05/15/15, from http://ksnt.com/2014/07/23/united-airlines-to-end-services-at-topeka- airport/.

Schmeltzer, J. (2001). O'Hare may feel pains of DetroitMetro's gains. Chicago Tribune. Retrieved 02/25/15, from http://tinyurl.com/pfjh4v7.

Shaw, S.-L., & Ivy, R. L. (1994). Airline mergers and their effect on network structure. Journal of Transport Geography, 2(4), 234-246.

183

Simmons, T. P. (2000). The regional jet, cancer or cure? A trend analysis detailing the effects of the regional jet on the quality of air service offered at small community airports. Naval Postgraduate School: Monterey, CA.

Smith, D. A., & Timberlake, M. F. (2001). World City Networks and Hierarchies, 1977-1997 An Empirical Analysis of Global Air Travel Links. American Behavioral Scientist, 44(10), 1656-1678.

Starkie, D. (2002). Airport regulation and competition. Journal of Air Transport Management, 8(1), 63-72.

Storper, M., & Venables, A. J. (2004). Buzz: face-to-face contact and the urban economy. Journal of Economic Geography, 4(4), 351-370.

Strimmer, K. (2008). fdrtool: a versatile R package for estimating local and tail area-based false discovery rates. Bioinformatics, 24(12), 1461-1462.

Suau-Sanchez, P., Burghouwt, G., & Fageda, X. (2015). Reinterpreting EU air transport deregulation: a disaggregated analysis of the spatial distribution of traffic in Europe, 1990–2009. Tijdschrift voor economische en sociale geografie. http://dx.doi.org/10.1111/tesg.12133.

Suzuki, Y. (2007). Modeling and testing the “two-step” decision process of travelers in airport and airline choices. Transportation Research Part E: Logistics and Transportation Review, 43(1), 1-20.

Suzuki, Y., Crum, M. R., & Audino, M. J. (2003). Airport choice, leakage, and experience in single-airport regions. Journal of Transportation Engineering, 129(2), 212-218.

Suzuki, Y., Crum, M. R., & Audino, M. J. (2004). Airport leakage and airline pricing strategy in single-airport regions. Transportation Research Part E: Logistics and Transportation Review, 40(1), 19-37.

Taaffe, E. J. (1956). Air transportation and United States urban distribution. Geographical Review, 219-238.

Taaffe, E. J. (1959). Trends in airline passenger traffic: A geographic case study. Annals of the Association of American Geographers, 49(4), 393-408.

Taaffe, E. J., Morrill, R. L., & Gould, P. R. (1963). Transport expansion in underdeveloped countries: a comparative analysis. Geographical Review, 503-529.

Taylor, A. (2012). House Rejects Bid to Slash Rural Airline Subsidies. cnsnews.com. Retrieved 05/15/15, from http://cnsnews.com/news/article/house-rejects-bid-slash-rural-airline- subsidies.

Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society. Series B (Methodological), 58, 267-288.

184

Tierney, S., & Kuby, M. (2008). Airline and airport choice by passengers in multi-airport regions: the effect of southwest airlines*. Professional Geographer, 60, 15–32.

Tukey, J.W. (1977). Exploratory Data Analysis. Ma, Reading, pp. 231–232.

United States Census Bureau [U.S. Census]. (2007). North American Industry Classification System. Retrieved 05/11/2015, from http://www.census.gov/eos/www/naics/2007NAICS/2007_Definition_File.pdf.

United States Census Bureau [U.S. Census]. (2010). 2010 Census Urban and Rural Classification and Urban Area Criteria. Retrieved 05/15/15, from http://www.census.gov/geo/reference/ua/urban-rural-2010.html.

United States Census Bureau [U.S. Census]. (2012). Zip Business Patterns – ZBP. Retrieved 05/11/2015, from http://factfinder.census.gov/faces/nav/jsf/pages/searchresults.xhtml?refresh=t.

United States Census Bureau [U.S. Census]. (2013). State & county quickfacts. Retrieved 02/25/15, from http://tinyurl.com/6p7rlyf.

United States Department of Transportation [U.S. DOT]. (2013a). Small Community Air Service Development Program. Retrieved 05/07/2015, from http://www.dot.gov/sites/dot.dev/files/docs/FY2013_SCASDP_RFP.pdf.

United States Department of Transportation [U.S. DOT]. (2013b). Small Community Air Service Development Program (SCASDP). Retrieved 05/07/2015, from http://www.dot.gov/policy/aviation-policy/small-community-rural-air-service/SCASDP.

United States Department of Transportation [U.S. DOT]. (2013c). Essential Air Service Reports. Retrieved 05/08/2015, from http://www.dot.gov/office-policy/aviation-policy/essential- air-service-reports.

United States Department of Transportation [U.S. DOT]. (2013d). Form 41, Schedules T100 and T100(f) Air Carrier Data. Retrieved 05/15/15, from http://2bts.rita.dot.gov/programs/statistical_policy_and_research/source_and_accuracy_c ompendium/form41_schedule.html.

United States Government Accountability Office [U.S. GAO]. 1996. Airline Deregulation: Changes in Airfares, Service, and Safety at Small, Medium-sized, and Large Communities. Report GAO/RCED-96-79, Washington, DC, April.

United States Government Accountability Office [U.S. GAO]. 1999. Airline Deregulation: Changes in Airfares, Service Quality, and Barriers to Entry. Report GAO/RCED-99-92, Washington, DC, March.

United States Government Accountability Office [U.S. GAO]. (2009). Options and analytical tools to strengthen DOT’s approach to supporting communities’ access to the system. Retrieved 04/18/2012, from http://www.gao.gov/new.items/s0753.pdf.

United States Office of Aviation Analysis [U.S. OAA]. (2009). What is Essential Air Service

185

(EAS)? Retrieved 05/07/2015, from http://www.dot.gov/sites/dot.gov/files/docs/easwhat.pdf.

United States Office of Aviation Analysis [U.S. OAA]. (2010). US Subsidized EAS Report. Retrieved 04/18/2012, from http://ostpxweb.dot.gov/aviation/x- 50%20role_files/essentialairservice.htm.

van den Heuvel, F. P., Rivera, L., van Donselaar, K. H., de Jong, A., Sheffi, Y., de Langen, P. W., & Fransoo, J. C. (2014). Relationship between freight accessibility and logistics employment in US counties. Transportation Research Part A: Policy and Practice, 59, 91-105.

Vowles, T. M. (1999). Predicting the loss of commuter air service in the United States. Journal of Air Transport Management, 5(1), 13-20.

Vowles, T. M. (2000). The effect of low fare air carriers on airfares in the US. Journal of Transport Geography, 8(2), 121-128.

Vowles, T. M. (2001). The “Southwest Effect” in multi-airport regions. Journal of Air Transport Management, 7(4), 251-258.

Wacht, W. F. (1974). The Domestic Air Transportation Network of the United States. University of Chicago, Department of Geography.

Wang, J., Mo, H., Wang, F., & Jin, F. (2011). Exploring the network structure and nodal centrality of China’s air transport network: A complex network approach. Journal of Transport Geography, 19(4), 712-721.

Wei, F., & Grubesic, T. H. (2015a). The dehubbing Cincinnati/Northern Kentucky International Airport (CVG): A spatiotemporal panorama. Journal of Transport Geography, 49, 85-98.

Wei, F., & Grubesic, T. (2015b). A Typology of Rural Airports in the United States: Evaluating Network Accessibility. The Review of Regional Studies, 45(1), 57-85.

Westwind. (2015). Grand Canyon Aerial Tour. Retrieved 02/12/16, from https://www.westwindairservice.com/tours-from-phoenix-scottsdale/grand-canyon-south-rim- air-tour/.

Wetterich, C. (2014). Is it high time for Cincinnati to end its CVG dreams? Cincinnati Business Courier. Retrieved 02/25/15, from http://tinyurl.com/pk5lume.

Williams, G. (2001). Will Europe's charter carriers be replaced by “no-frills” scheduled airlines? Journal of Air Transport Management, 7(5), 277-286.

Williams, J. (2013). Can Frontier survive where others dared fly? Cincinnati Enquirer. Retrieved 02/25/15, from http://tinyurl.com/mb28cla.

Williams, J. (2014a). CVG loses ground as peers gain. Cincinnati Enquirer. Retrieved 02/25/15, from http://tinyurl.com/qj7prew.

186

Williams, J. (2014b). CVG ‘can't give up’ on more international flights. Cincinnati Enquirer. Retrieved 02/25/15, from http://tinyurl.com/n92xj3u.

Williams, J. (2014c). 5 crucial transportation decisions coming this year. Cincinnati Enquirer. Retrieved 02/25/15, from http://tinyurl.com/p42m4ch.

Williams, J. (2015). Low-cost carriers help CVG end yearslong slide. Cincinnati Enquirer. Retrieved 02/25/15, from http://tinyurl.com/mkzyvm4.

Williams, G., & Pagliari, R. (2004). A comparative analysis of the application and use of public service obligations in air transport within the EU. Transport Policy, 11(1), 55-66.

Wittman, M. D. (2014). Public funding of airport incentives in the United States: The efficacy of the Small Community Air Service Development Grant program. Transport Policy, 35, 220-228.

Wittman, M. D., & Swelbar, W. S. (2013). Trends and market forces shaping small community air service in the United States.

WXIX. (2013). Documents reveal cozy relationship between CVG & Delta. Retrieved 02/25/15, from http://tinyurl.com/m8uw6mu.

Zhang, A. (1996). An analysis of fortress hubs in airline networks. Journal of Transport Economics and Policy, 30, 293-307.

Zhang, Y., & Xie, Y. (2005). Small community airport choice behavior analysis: A case study of GTR. Journal of Air Transport Management, 11(6), 442-447.

Zook, M. A., & Brunn, S. D. (2005). Hierarchies, regions and legacies: European cities and global commercial passenger air travel. Journal of Contemporary European Studies, 13(2), 203-220.

Zook, M. A., & Brunn, S. D. (2006). From podes to antipodes: positionalities and global airline geographies. Annals of the Association of American Geographers, 96(3), 471-490.

187

Appendix A Dependent and Independent Variables in LASSO Regression

Model 1 Name Description Destinations Number of direct flight options from CVG to other domestic destinations (by year) Large Hubs (+) Number of large hubs directly connecting to/from CVG Medium Hubs (+) Number of medium hubs directly connecting to/from CVG Small Hubs (+) Number of small hubs directly connecting to/from CVG CVG_FARE (-) Average itinerary fare at CVG CMH_FARE/DAY_FARE/IND_FARE/SDF_FARE/LEX_FARE Average itinerary fare at neighboring airports, respectively (+) CVG_FREQUENCY (+) Market share of frequency counts at CVG CMH_FREQUENCY/DAY_FREQUENCY/IND_FREQUENCY/ Market share of frequency counts at neighboring airports, respectively SDF_FREQUENCY/LEX_FREQUENCY (-) CVG_PASSENGERS (+) Market share of passenger counts at CVG CMH_PASSENGERS/DAY_PASSENGERS/IND_PASSENGER Market share of passenger counts at neighboring airports, respectively S/ SDF_PASSENGERS /LEX_PASSENGERS (-) MILES (-) Average miles flown at CVG Positive influence: (+) Negative influence: (-)

188

Model 2 Name Description Destinations Number of direct flight options from CVG to other domestic destinations (by year) Large Hubs (+) Number of large hubs directly connecting to/from CVG Medium Hubs (+) Number of medium hubs directly connecting to/from CVG Small Hubs (+) Number of small hubs directly connecting to/from CVG CVG_FARE (-) Average itinerary fare at CVG CMH_FARE/DAY_FARE/IND_FARE/SDF_FARE/LEX_FAR Average itinerary fare at neighboring airports, respectively E (+) CVG_FREQUENCY (+) Market share of frequency counts at CVG CMH_FREQUENCY/DAY_FREQUENCY/IND_FREQUENC Market share of frequency counts at neighboring airports, respectively Y/ SDF_FREQUENCY/LEX_FREQUENCY (-) CVG_PASSENGERS (+) Market share of passenger counts at CVG CMH_PASSENGERS/DAY_PASSENGERS/IND_PASSENG Market share of passenger counts at neighboring airports, respectively ERS/ SDF_PASSENGERS /LEX_PASSENGERS (-) MILES (-) Average miles flown at CVG CVG_IT (+) Total number of estimated employee counts in IT sector at CVG CVG_FIRE (+) Total number of estimated employee counts in FIRE sector at CVG CVG_PST (+) Total number of estimated employee counts in PST sector at CVG CVG_TOUR (+) Total number of estimated employee counts in TOUR sector at CVG Positive influence: (+) Negative influence: (-)

189

Model 3 Name Description Destinations Number of direct flight options from CVG to other domestic destinations (by year Large Hubs (+) Number of large hubs directly connecting to/from CVG Medium Hubs (+) Number of medium hubs directly connecting to/from CVG Small Hubs (+) Number of small hubs directly connecting to/from CVG CVG_FARE (-) Average itinerary fare at CVG CMH_FARE/DAY_FARE/IND_FARE/SDF_FARE/LEX_FARE Average itinerary fare at neighboring airports, respectively (+) CVG_FREQUENCY (+) Market share of frequency counts at CVG CMH_FREQUENCY/DAY_FREQUENCY/IND_FREQUENCY/ Market share of frequency counts at neighboring airports, respectively SDF_FREQUENCY/LEX_FREQUENCY (-) CVG_PASSENGERS (+) Market share of passenger counts at CVG CMH_PASSENGERS/DAY_PASSENGERS/IND_PASSENGERS/ Market share of passenger counts at neighboring airports, respectively SDF_PASSENGERS /LEX_PASSENGERS (-) MILES (-) Average miles flown at CVG CVG_L_IT (+) Total number of estimated large establishments in IT sector at CVG CVG_M_IT (+) Total number of estimated medium establishments in IT sector at CVG CVG_S_IT (+) Total number of estimated small establishments in IT sector at CVG CVG_L_FIRE (+) Total number of estimated large establishments in FIRE sector at CVG CVG_M_FIRE (+) Total number of estimated medium establishments in FIRE sector at CVG CVG_S_FIRE (+) Total number of estimated small establishments in FIRE sector at CVG CVG_L_PST (+) Total number of estimated large establishments in PST sector at CVG CVG_M_PST (+) Total number of estimated medium establishments in PST sector at CVG CVG_S_PST (+) Total number of estimated small establishments in PST sector at CVG CVG_L_TOUR (+) Total number of estimated large establishments in TOUR sector at CVG CVG_M_TOUR (+) Total number of estimated medium establishments in TOUR sector at CVG CVG_S_TOUR (+) Total number of estimated small establishments in TOUR sector at CVG Positive influence: (+) Negative influence: (-)

190

Appendix B Fuzzy Clustering Problem (FCP)

( , ) Minimize 𝑛𝑛 𝑛𝑛 𝑟𝑟 𝑟𝑟 (1) ∑𝑖𝑖=1 ∑𝑗𝑗=1 𝑢𝑢𝑖𝑖𝑖𝑖𝑢𝑢𝑗𝑗𝑗𝑗𝑑𝑑 𝑖𝑖 𝑗𝑗 𝑘𝑘 𝑛𝑛 𝑟𝑟 ∑𝑐𝑐=1 2 ∑𝑗𝑗=1 𝑢𝑢𝑗𝑗𝑗𝑗

Subject to

= 1 for all i (2) 𝑘𝑘 𝑐𝑐=1 𝑖𝑖𝑖𝑖 ∑ 𝑢𝑢 0 for all i and k (3)

𝑖𝑖𝑖𝑖 𝑢𝑢 ≥ i, j = index of observations; n = number of observations; k = number of clusters; r = membership exponent d(i,j) = distance between observations i and j;

= membership of observation i in cluster k.

𝑖𝑖𝑖𝑖 𝑢𝑢

Fuzzy clustering aims to minimize the objective function (9). Furthermore, the

objective function is to minimize the distance between observations i and j. The d(i,j) can

be calculated as Euclidean, Manhattan or Squared Euclidean distance. Membership

exponent r ranges from 1 to ∞. When its value is closing to 1, the clustering result can be

converged slowly; when its value is closing to ∞, the clustering result can be completely

fuzziness (Kaufman and Rousseeuw, 1990). r = 2 is usually used for fuzzy clustering

191 problem. Constraint (2) indicates that a constant total membership of each observation is spread out over various clusters and constraint (3) identifies non-negativity of membership

(Grubesic, 2006). Membership coefficients of an observation can be considered as possibilities of belonging to various clusters.

192

Vita

Fangwu Wei Education 2016 Ph.D. in Information Studies. College of Computing & Informatics, Drexel University, Philadelphia, Pennsylvania 2011 Master of Science in Information Systems. College of Information Science and Technology, Drexel University, Philadelphia, Pennsylvania. 2009 Master of Public Administration. Department of Social Sciences, Illinois Institute of Technology, Chicago, Illinois. 2007 Bachelor of Information Management and Information System. School of Management and Economics, Beijing Institute of Technology, Beijing. Honors and Awards 2012 1st Place, Drexel University Research Day in Humanities & Social Sciences Category. 2007 Beijing Institute of Technology Outstanding Graduate Award. Campus-wide award for graduating students who show solid curricular performance and strong personality. Published Journal Articles 2015 Wei, F. and Grubesic, T. A Typology of Rural Airports in the United States: Evaluating Network Accessibility. Wei, F., Grubesic, T.H. and Bishop, B. W. Exploring the GIS Knowledge Domain Using CiteSpace. Wei, F. and Grubesic, T.H. The Dehubbing Cincinnati/Northern Kentucky International Airport (CVG): A Spatiotemporal Panorama. 2014 Grubesic, T., Wei, R., Murray, A. and Wei, F. Essential Air Service in the United States Exploring Strategies to Enhance Spatial and Operational Efficiencies. 2013 Grubesic, T.H. and F. Wei. Essential Air Service: a local, geographic market perspective. 2012 Grubesic, T.H. and F. Wei. Evaluating the Efficiency of the Essential Air Service Program in the United States. Teaching Experience Spring 2016 Instructor. Course: Introduction to Geographic Information Systems (INFO 555, online), College of Computing & Informatics, Drexel University Winter 2016 Instructor. Course: Geographic Information Science (INFO 220), College of Computing & Informatics, Drexel University Journal Reviewer International Regional Science Review (2), Telematics and Informatics (1), Transport Policy (1), Asian Journal of Criminology (1), Journal of Land Use Science (1)