Planning, Sustainability and Airport-Led Urban Development

MAA-4-2012-Report excerpts distributed at the meeting

International Planning Studies Vol. 14, No.2, 161-176, May 2009

ROBERT FREESTONE Faculty of the Built Environment, University of New South Wales, Sydney, Australia

ABSTRACT Airports are no longer places where planes just take ofjand land but have evolved into major business enterprises with spatial impacts and functional implications that extend deep into metropolitan areas. They are vital hubs in the global space of flows.. Airport-led urban development, notwithstanding its employment and income generating capabilities and potentials, comes with costs and risks. economic, environmental, and cultural. A host ojplanning issues are raised. Traditional NIMBY reactions against airport expansion are evolving into more fundamental critiques of aviation around issues such as climate change .. Mediating the conflict between the aviation industry's pro-growth stance and more sceptical perspectives is the concept ofsustainable aviation. This may prove an oxymoron but it remains vital to link airport planning to the broader planning ofsustainable communities and regions.

Introduction Without the development ofairports in facilitating the large-scale movement ofpassengers and fr·eight, globalization in its cunent forms would be 'utterly different, possibly non existent' (Uny, 2007). Airports have become vital functional nodes in the world economy and the jousting for regional, national, and international competitiveness., They are growth nodes for local areas and regional economies. They have spawned new urban forms, mostly spontaneously and less rarely deliberative, as their direct and indirect impacts spill over airport boundaries. These trends are impossible to ignore in planning for the future at an international scale, Airport-led urban development, notwithstanding its employment and income generating capabilities and potentials, comes with costs and risks, economic, environmental, and cultural. The aerotropolis model ofdevelopment (Kasarda, 2000, 2001, 2006) while super ficially attractive requires closer interrogation and elaboration as a sustainable urban planning and economic development strategy., These new urban forms, even the most successful ones in connection with planned mega-airports, raise a host of planning issues., There has long been a strong NIMBY reaction against new airports and export expan sion, but this is now morphing into a more fundamental critique of airports and air traffic around the issue of climate change. Airport-led urban development is at the epicentre of

Correspondence Address. Robert Freestone, Planning and Urban Development Program, Faculty of the Built Environment, University of New South Wales, Sydney, NSW 2052, Australia Tel: +61 2 9385 4836; Fax: +61293854507. Email: rfreestone@unsweduau

162 R.. Freestone

this debate. Bringing the sides together requires more consultation, more collaboration, more research, and a far greater awareness at a regional scale fIom all stakeholders that airports are no longer just transport hubs (difficult enough major land uses even in this narrow sense) but underpin diversified mixed used activity centres, zones, and corridors, which need to be comprehended wholly as a new urban form (Guller & Guller, 2002) This paper offers a brief and critical overview of these issues from a broad planning perspective .. It does not attempt a comprehensive survey of the airport-environment interface or offer detailed policy reviews of specific jurisdictions .. Rather, the approach taken is a wide-ranging synoptic one that essentially presents a dialogue between a bullish pro-growth perspective evident in aviation industry and government circles versus a more critical perspective questioning mainstream thinking that seemingly endorses limitless growth potential. The paper has three main parts: a brief survey of the phenomenon and drivers of airport-focused urbanism, a precis of airport planning trends and the aerotropolis concept, and a consideration of economic and environmental airport development critiques with a view to seeking rapprochement in the context of sus tainable development The paper draws flom a broader study examining the urban policy impacts and implications of the 'airport metropolis' in Australia and internationally.l

Airport-Focused Urbanism Transformations in the nature and role ofurban airports are being played out against funda mental changes in forces affecting both cities and aviation. The intersecting trends are towards inexorable growth .. Notwithstanding a scatter of 'shrinking cities', urban populations have been steadily rising and extended metropolitan regions have become a dominant urban form with concomitant demands for transportation infIastructure and connectivity. Similarly, while checked by successive exogenous shocks - terrorism, regional pandemics, surging oil prices, and economic crises - the historical growth in aviation activity has been steadily upward. The nexus between these phenomena rotates around the role of aviation in economic development. There is debate as to whether increased aviation activity is a cause or effect of economic growth, but either way, there is a strong correlation between metropolitan growth and aviation (Brueckner, 2003). In a globalized world, aviation has cemented its place as a dominant transport technology with consequent ramifications for the ordering of urban and regional space economies at different scales .. John Kasarda has memorably captured this emergent reality with his notion of a fourth wave of development in which airports shape business location and urban development in the twenty-first century as much as highways did in the twentieth century, railroads in the nineteenth and seaports in the eighteenth (www.aerotropolis.com). Airports are a remarkable barometer of the historical dynamism of cities through time (Gordon, 2004). They have evolved through a typology of urban and architectural forms in concert with the rising demand for air travel accompanying urban population growth .. The initial grassed aerodromes of the 1920s were formalized as city gateways akin to rail and port terminals by the late 1930s. The development of military-related infIastructure and utilization during the Second World War helped lay a platform for rapid expansion through the democratization of international air travel from the 1950s. The 1960s saw a phase of replication as new airports were developed on the metropolitan fringe to replace or supplement older facilities hemmed in by the spread of urban development The designer airport then took hold as a place-making device with the subsequent struggle MAA-4-2012-Report excerpts distributed at the meeting

Planning, Sustainability and Airport-Led Urban Development 163

for regional and intemational competitive advantage as the era of mass air travel was confirmed. Latterly, the trend has been towards a greater diversity of type and form (mega airports, specialist cargo airports, low cost carrier terminals). Over the same period, airport-related development has also changed in character and increased in scale, Early aviation-linked businesses, often makeshift, opportunistic, and engineering-related were gradually supplemented and supplanted by more extensive commercial development servicing increased passenger and cargo activity such as hotels and height forwarders. In time, where land supply permitted, master planned business parks sprung up because of the proximity advantages and waves of redevelopment began to shape new city and suburban commercial landscapes in line with the rise of the edge city phenomenon (Garreau, 1992). The major trend has been toward 'more upscale, less industrial-based type of development' (Dempsey et at, 1997). The influence of airport access on urban structure has been felt not just in close physical proximity but at a metro politan scale defined by time contours. This broader impact is frequently channelled via development corridors assuming a variety of forms depending on land availability, trans port modalities, and planning strategies, and controls (Schaafsma et at" 2008)., Castells (1996) in his treatise on the rise of informational society introduced the concept of 'spaces of flows', redefining the geography of economic development as less about an amalgam of individual places but more the connections between them, In an increasingly globalized world where flows of goods, people, and services are central, aviation networks help define world connectivity. Increases in routes and traffic have led to the emergence of vital airport hubs and reinforced global city status, Airports are the 'hubs of flows', and have experienced revolutionary change in their operational and strategic environments in the process, Free trade laws and the liberalization of aviation regulations under neo liberal political-economic philosophies have formed a backdrop to these changes. Airports are central to competitive and comparative advantage: 'the airport is perhaps the most important, single piece of infrastructure in the battle between cities and nations for influence in, and the benefits of, growth and development'(O'Connor & Scott, 1992) .. The growth of business travel, discretionary leisure trips, and fleight movements in the world economy has been profound. The expansion of cargo in particular has been driven by development offaster, larger, longer range aircraft, the liberalization ofair height services in many markets, the intemationalization of economic activity, and proliferation of just in-time material management practices (Leinbach & Bowen, 2004). Such trends have reinforced the significance of airports in their regional context and they have now become 'central to the operation and development ofmetropolitan areas' (O'Connor & Scott, 1992). A macro-trend affecting the nature of airports in their urban context has been the pro gressive loosening of their historic ties with the state. Airports have shifted flom being a branch of government to dynamic and commercially-oriented businesses (Doganis, 1992)., Developing this proposition, three key economic trends have transformed the world's air ports over the last quarter century: commercialization, privatization, and globalization (Graham, 2001) .. Airport commercialization means the transformation of a public utility to a commercial enterprise with the adoption ofmore businesslike management practices. Big airports have tumed themselves into shopping malls and progressed into property development. The most recent trend within terminals has been for shops and restaurants to be pushed to the so-called 'airside' zone after passengers have gone through security (so they can relax and consume). This has been complemented by more intensive development on MAA-4-2012-Report excerpts distributed at the meeting

164 R. Freestone vacant airport land not needed for aviation operations and has sometimes extended beyond the actual airport boundary in joint ventures. A recunent phrase heard in aviation-speak captures the guiding philosophy: making the airport 'a destination in its own right'., The new class of Asian airports - such as Singapore Changi developed from 1981 - have all been conceived as 'airport cities' on extensive sites liberated hom the physical constraints faced by older airports in more established built-up areas. Commercialization is also a response to the volatility of the airport business, which is particularly susceptible to downturn from forces beyond the control of the sector itself, Non-aeronautical activities spread the business risk and constitute a source for 'secure sustainable revenue' (Reiss, 2007). In principle, this income stream can be applied to infrastructure investments without steadily raising landing and passenger fees being imposed on the airlines (Doganis, 1992). Non-aeronautical revenues constituted approximately 30% of total airport revenues in 1990. This rose to 46% in 1995, to 51 % in 2000, and to 54% in 2007 (Airport Innovation, 2007)" Non-aviation aspects of airports also tend to be less regulated, offering airport companies greater freedom for entrepreneurial innovation. Airport privatization is the transfer of the management of an airport, and in many cases ownership as well, to the private sector by a variety of methods. These include share flotations, the adoption ofstrategic partnerships, and the introduction ofprivate management contracts. One of the most profound and extensive national privatization exercises has been in Australia where since 1996 a total of 22 airports previously under the control of the Commonwealth Government have been sold on long-term leases to private sector consortia, The policy generated a multi-billion dollar budget windfall for government and has been accompanied by significant investments in infrastructure as desired (and, in some cases, required under the terms of transfer) but has also been marked by aggressive moves into general property development on airport land (Freestone et aZ., 2006), Airport globalization denotes the emergence of global airport companies who operate an increasing number of airports around the world such as BAA, Schiphol, and Macquarie Airports. Some of these global players are traditional airport operators whereas others are new to airport management. Schiphol airport itself has been an industry leader in the possibilities of airport-centric development As early as the 1930s, it was a place that other airports learned flom. Authorities encouraged the public to visit the airport and use it as a 'civic amenity' - 'by visiting the airport citizens could show their support' and become 'more air-minded and, eventually, make the airport more money' (Adey, 2006) Its real influence has been felt since the 1980s with its 24 h a day 'AirportCity' concept based on many different types of commercial and cultural enterprises at the airport, plus various joint building and land development ventures with other corporations and government bodies,. Major airports are assuming a dominant role as transactional spaces in the global economy (Gottdiener, 2001)., They have become key nodes for global production and enterprise systems demanding speed, agility, and accessibility. Airports and their environs offer advantages to business in a globally networked economy where the new credo is what the futurist Alvin Toffler dubbed 'survival of the fastest' (Kasarda, 2006)" They constitute critical gateways and conduits for inter-regional and international travel, trade, and tourism. This means that airport lands and the territory around airports or with good access time-wise have increased significantly in value" However, most econ omic impact studies remain at an aggregate level and despite attempts to measure the more immediate employment impacts of airports (Twomey and Tomkins, 1995; MAA-4-2012-Report excerpts distributed at the meeting

Planning, Sustainability and Airport-Led Urban Development 165

Al Chalbi, 1998), there remain fundamental methodological difficulties in establishing any precise 'causality between the expansion of an airport and wider economic development' (Graham, 2003). The valorization ofairport access appears to be manifested in at least four different ways: through airport-support functions, attraction of time-sensitive activities, concentration of businesses with high-travel demands, and attraction of non-aeronautical development attracted by agglomeration economies. The first ofthese refers to expanded and new activi ties supporting the operations ofthe airport gravitating to the airport environs. These include services directly supporting operations (like flight kitchens and aircraft maintenance), airport-related height services (shipping, height forwarding, customs, foreign trade zones) and services for airline employees and passengers such as hotels, restaurants, and car rental franchises .. Employment mix and growth tends to be proportionate to the scale of airport operations as either an aircraft servicing centre, fr·eight facility, or passenger hub (Weisbrod etat., 1993). Airports and cities compete for major investments like mainten ance operations facilities, a process which can involve commitment ofmillions ofdollars of public funds (Nunn & Schoedel, 1995).. Also drawn to airport areas are time-sensitive goods-processing and distribution func tions such as e-commerce, warehousing, and perishables handling .. Air cargo express has become the preferred mode for shipping high value to weight products, like elec tronics, optics, and pharmaceuticals .. The 'need for speed' means that 'storage has given way to velocity as the emphasis has shifted from static inventory to the movement of goods' (Frej, 2004). As supply chains have become more complex, specialist logistics firms have become key players in inter-regional and international goods assembly and movement. The actual cost of moving materials long distances is a small element in the total cost of shipping door to door (Harris, 1994) .. The real economic savings and efficien cies are to be gained in land-based and inter-modal operations, underlining the importance of airport accessibility. Areas around airports can also be very attractive for development because, if not built out, they are often flat and suitable for big footprint buildings (Blanton, 2004). The rise around transport hubs including airports of new goods assembly and distribution spaces embedded in global supply chains has been described as one of the 'more significant transformations of the built environment over the past decade' (Waldheim & Berger, 2008).. Airport proximity is also a magnet for organizations whose operations require frequent inter-city travel to do face-to-face business. Airports offering fr·equent service to a variety of destinations are thus most highly valued .. 'Airport intensive' economic sectors include insurance, banking and finance, printing and publishing, transport, computers, precision and optical instruments, business services, and R&D (York Aviation, 2004).. Airport corridors have accordingly become attractive for office buildings housing regional corporate headquarters. A familiar global landscape of anonymous towers with familiar corporate logos alongside motorways linking central business districts (CBDs) and air ports has been played out around the world (Hack, 2000). Airports themselves are also places of 'meetingness' (Urry, 2007) offering numerous micro-scale opportunities to help bind together the transactional economy (Knorr-Cetina, 2006).. Finally, the accessibility, visibility, and prestige of an airport address can attract non-airport development, serving other markets through growing economies of scale .. In this and the other ways mentioned, airports clearly have major spatial - and economic - impacts and it is one reason why they are seen as powerful economic MAA-4-2012-Report excerpts distributed at the meeting

166 R. Freestone

development tools. Qualifications can be made but numerous studies report positive growth indicators., One rule of thumb is that a 10% increase in passengers leads approxi mately to a 1% increase in service-related industries (Brueckner, 2003), A study of Schiphol airport estimated that the total multiplier of direct employment is approximately 2: that is, one job on the airport leads to one job in indirect and another in induced employment (Hakfoort et al., 2001). Employment growth within 6 km of an airport can be 2-5 times faster than elsewhere in the suburban ring of the metro area in which it is located (Weisbrod et at" 1993). The market-driven attraction of business activity into airport areas through agglomeration economies raises issues for industrial location policy and the possible need to reserve land for firms requiring air transport access (Warffemius et at" 2009), When Harris and Ullman devised their ideal-typical model of multi-nucleated urban form in 1945, they did not include an airport zone" In an era of DC-3s and flying boats canying 30-40 passengers, the scale of inter-regional and international operations was modest and airports remained a highly specialized and generally fringe land use. Nearly a half century later, Harris (1997) updated this status in a revised 'peripheral model of metropolitan areas' which acknowledged the new 'airport cluster' alongside other new suburban forms such as shopping malls, corporate campuses, and theme parks. Airports can be key development catalysts as cities metamorphose into polycentric urban realms (Hall, 2001), Some theorists have taken this trend to its logical zenith: the aerotropolis.

Airport-Centred Planning The vision of an airport at the city's heart goes back at least to Le Corbusier in the 1920s whose ideal city featured an aerodrome sitting atop a multi-nodal grand central station flanked by vertiginous skyscrapers., Presciently, he conceived the urban form as 'a model city for commerce' and the most literal expression of the philosophy that 'the city which can achieve speed will achieve success' (Le Corbusier, 1987). Other contem porary visions imploded airport structures much less elegantly into the urban fabric (Pearman, 2004)" Latter day versions are characteristically more spatially expansive., Moving beyond the reality of small-scale airport business parks and fly-in residential and leisure communities, Conway (1980) conceived the 'decoplex' as a development ecology complex with jetport alongside industrial sites, offices, hotels, and waste treat ment facilities and then the more heroically scaled 'Jet City' as a 'functionally efficient centre of economic development'., Kasarda's 'aerotropolis' model is the best-known modern model to capture the nexus between planning, sustainability and airport-led urban development. The aerotropolis is a normative urban form 'leveraged by air commerce' (Kasarda, 2001), It is promoted as a means of building competitive advantage into the fabric of cities. The core is an airport city integrating aeronautical and non-aeronautical uses including business offices, hotels, and conference centre. This anchors a more extensive mix of precincts of warehouses, e-fulfilment centres, industrial and office parks, free trade zones, hotel and entertainment districts, all oriented to connecting motorway corridors. Residential districts, the only non-commercial land use, occupy the wedges between the motorways away hum the main flight paths. Three central 'aerotropolis principles' are the desirability of clustered rather than ribbon development, the importance of high-quality design standards, and opportunities for beautification of airport gateways, MAA-4-2012-Report excerpts distributed at the meeting

Planning, Sustainability and Airport-Led Urban Development 167

The aerotropolis model successfully conveys the importance of the airport in urban and suburban business development in a globalized economy, Its appeal from an industry perspective is to encapsulate the economic and community benefits delivered by airports (Siebert, 2008) As a senior executive in Macquarie Airports Management put it: 'a successful airport is clearly a driver of a city's and region's economic performance' (Moore-Wilton, 2007). In these terms the aerotropolis virtually represents a business model. The conceptual weaknesses of the model lie in dependence on a non-renewable resource, over-concentration of critical infrastructure, and a presupposition to the hege mony ofairports in logistics and distribution (Charles etal., 2007)" From a broader planning perspective, the aerotropolis model also tacitly endorses the inevitability if not desirability ofan extensive 'sprawl and scatter' pattern ofsuburban land use development. At worst, the aerotropolis presents an unsustainable urban form, which augments 'already existing externalities of noise, pollution, and increased traffic congestion' (Leinbach, 2004). Most aerotropolis development to date has been 'spontaneous and haphazard' (WWw., aerotropolis.com), mainly because planning and governance structures have been frag mented and have not see the problems - and possibilities - whole" Guidance on better models comes from actual airport master plans, particularly for greenfield projects, which have been conceived as inter-modal and multi-use activity centres, The planning for these projects extends beyond existing airport planning guidelines which relate more to regional scale facilities and a narrower band of environmental impact concerns headed by noise (American Planning Association, 2006) .. On top of solutions to complex engin eering problems, they encompass a range of considerations commensurate with the design of regional communities and town centres, These new airports include Bangkok's Suvarnabhumi, Hong Kong's Chep Lap Kok, and Seoul's Incheon. Master planners of the new Denver International Airport had similarly expansive ideas about integrated land use development (Dempsey et aI., 1997). These mega-projects integrate a mix ofuses in ambi tious schemes, which anticipate and seek to mitigate interface issues with contiguous land uses" Similar ideas will likely involve the many hundreds of new airports or airport expansions planned for India and China over the coming decades. These large integrated projects attempt to internalize spin-off economic activity in a way not systematically captured by airports before the early 1980s (van den Berg et aI., 1996). Moreover, their increasing scale is directly predicated on the presumption that bigger airports ensure greater economic returns and possibilities. Like much property driven development in the DAE, the behemoth Dubai World Central airport takes this model to the extreme (Bagaeen, 2007). However, these sorts of airports are the exception rather than the rule" Most major airports are more complexly integrated into the existing urban environment and must constantly negotiate a set of problematical interactions including noise, air quality, public safety, traffic congestion, and infrastructure provision" Nevertheless, both types of airport still raise questions about the long run sustainability of extensive air travel and the urban forms which come to depend on it This leads to the third issue of concern in this paper,

Airport-Induced Concerns The dominant discourse in the aviation industry is growth and this similarly underpins the aerotropolis concept. Predictions of business travel, tourist flows, and air cargo all point upward in the long term, This in turn translates into the need for more airport MAA-4-2012-Report excerpts distributed at the meeting

168 R. Freestone

capacity and perforce through economic multipliers into continued development in the broader aerotropolis .. However, these aggressive growth scenarios are being contested in many parts of the world on various grounds., The challenges faced by airport-centered development can be discussed under two general sets of concerns: economic and environmentaL Economically, the case for seemingly unrestricted growth of the aviation industry is now questioned on several grounds., Williams (2000) queries much industry-sponsored pro-aviation economic analysis as based on contestable assumptions, uncertain forecasts, aspatial reasoning (hence ignoring redistributional impacts), and inadequate consideration of environmental costs. At the same time, the aviation sector is said to require investment on a massive scale and benefit from heavy subsidies., The implicit assumption in anointing the airport as a regional growth pole is the diffusionist idea of economic growth being imported through airport activity and spreading its material beneficence outward. These ideas are associated with the classic development theory work of Albert Hirschman and Gunnar Myrdal in the 1950s-1960s, yet contemporary regional development theory has clearly moved on (Glasson & Marshall, 2007), In the race to develop airports for regional comparative advantage, the question must be asked as to just how many airport growth poles can be viable? There are certainly examples of major airports without significant area development commensurate to the scale of their aviation activity. For example, contri buting factors in Atlanta's failure to fully tap the development potential of Hartsfield International - the world's busiest airport - include the unavailability of large area development sites, chronic airport noise problems and blight in contiguous commu nities, failure to provide adequate services infrastructure in advance of development, the unattractiveness of the airport area for investors, and breakdowns in regional cooperation and focus (Kramer, 2004)., Other general impediments to the development relate to restrictions on both development of airport land especially for non-aviation purposes and area planning restrictions, such as airports located in green belt zones. Utilizing airports to capture a sizeable slice of the air height market has been seen as a desirable economic development strategy. There are several grounds for this approach, Over the past 30 years, the growth in the value ofcargo shipped by air has significantly out stripped growth of global trade generally and there is further evidence that air cargo traffic recovers at a much quicker rate than passenger flows in an economic downturn (Kasarda & Sullivan, 2005). Much growth has also been fuelled by the rise of integrated operators like DHL, TNT Express, UPS, and FedEx, which are expected to have a global market share of 31 % by 2019 (Gillingwater et ai., 2003). The shining exemplar of height-centered airport development is Memphis, Tennessee, and its rise as a major inland port due to its headquartering ofFedEx (although the immedi ate area around the airport remains blighted and the target ofa new community renaissance movement)., Other places have tried to tap the corporate search for greater and faster supply chain economies as an area development strategy, with development of predomi nantly or exclusively height airports, often converting former military or general aviation facilities. One outcome ofthese developments and the related growth ofthe logistics sector has been the 'global transpark' concept These are more than traditional air cargo facilities, but conceived as concentrations of time-sensitive economic activities based around transport hubs with direct inter-modal loading and unloading capabilities, advanced telecommunication services, and light-handed government taxes, and customs processing., MAA-4-2012-Report excerpts distributed at the meeting

Planning, Sustainability and Airport-Led Urban Development 169

They are seen as economic catalysts 'to render development to a large surrounding hinterland' (Sit, 2004). Although various projects have been proposed in China, Germany, and the USA, none has really taken off, including the Global TransPark at Kinston Airport near Greenville, North Carolina, which remains a conventional and small-scale airport industrial complex, with the novelty of a conference facility. There are general reasons why economic returns from new air freight-led airport devel opments might be overstated. One is that cargo (as opposed to passenger growth) is a poor predictor of employment growth and hence economic development, with warehouse and distribution facilities increasingly automated and offering lower paying jobs (Green, 2007). Second, is an unrealistic faith in a 'build and growth will come' optimism that ignores fundamental parameters such as accessibility and competition from other airports. Third, is a basic business rule ofthe industry that tends to work in favour of existing facili ties: 'airlines go to markets, not airports' (quoted in Erie, 2004) .. Fourth, much height still moves by other modes, including in the holds of passenger aircraft, but in the bigger scheme of things there is still negligible integration of air freight with rail and sea modes (Bowen & Slack, 2007). There is a final problem for some freight airports and that is their environmental impacts. Cargo aircraft are often older planes, less environmen tally efficient and noisy,. Night time operations are important in a 24 h global economy. Problems arise when located close to centres of population and if airport operations are constrained as a result of community concerns with delivery times being lengthened, then just-in-time operations can be threatened (Gillingwater et at, 2003). This links to broader environmental challenges faced by airports and therefore the urban-economic system built around them, namely the aerotropolis. Environmentalist critiques of airports have targeted various operational dimensions such as air pollution, physical, and psychological health, contaminated run-off, and land consumption ('the biggest sprawl of all'). Ayres (2001) argues that 'in its total impact on climate, ecology, and health, today's mega-airport may be one ofthe most ill-conceived forms oflarge-scale infrastructure humankind has ever devised' , as well as one ofthe least accountable.. Historically, environmental protests have been targeted at local concerns about the development of new airports on specific sites or expansion of existing ones. These classic NIMBY campaigns have variously targeted noise, health and amenity; impact on property prices; blight; loss of biodiversity and heritage sites; and accident risk. As a result, many airports have limitations on their activity, e.g., aircraft movement and type limits, flightpath restrictions within noise contours, and night curfews,. Some 66% of European airports have environmental constraints (Hooper, 2007). Beyond oper ational considerations and management efficiencies, airports have an 'environmental capacity' linked to the tolerance of their impacts within the host environment, both human and non-human (Upham et aI.., 2003)., A larger environmental dimension is now evident in the questioning of airport and airport area development. According to Short (2004), 'airports are not just nodes in the global network of flows; they are sites of major environmental impact that highlight the tension between international connectivity and local livability',. Recognition of this nexus between the global and the local is now influencing environmental protests with a shift away from localized campaigns towards broader implications for environmental sustainability and climate change, social justice and economic development. There has emerged 'a more universal struggle aimed at countering airport expansion' (Griggs & Howarth, 2004). In Australia, the issue of privatization lends a special flavour to MAA-4-2012-Report excerpts distributed at the meeting

170 R. Freestone growing public debate on airports with a strongly stated view by some in the community that 'transport is a public responsibility' and that utilization of public land for business enterprises not directly connected to core airport activity reflects only a quest for profits (May & Hill, 2006). More universally, concern at the aviation industry's contribution to global warming has undoubtedly become the key concern. A general consensus appears to be that the aviation industry in terms of global anthropogenic carbon emissions is responsible for 3-5% of the worldwide totaL It is the projections that are most troubling ..

Without ameliorative action, an unconected increase in CO2 aviation emissions by 2050 of up to 300% is now a common prediction (Upham et at, 2003; Macintosh & Downie, 2007). Airports now find themselves at the epicentre of the growing debate about the global environmental consequences of aviation, and hence the target of vigorous criticism and opposition in which local struggles seamlessly mesh with regional, national, and inter national concerns (Griggs & Howarth, 2008). Activist groups like the umbrella organ ization Airport Watch in the UK argue that there are many powerful environmental, equity and economic reasons to oppose the aggressive expansion of aviation .. They bring a more expansive and sophisticated critique compared to isolated past efforts Just as the aviation industry builds coalitions between airlines, airports, and the aero space industry ('freedom to fly'), environmental protest groups are also linking up to form a broader political force which, along with the approaching reality of peak oil, is being heard in government circles .. In the UK, again, the government's independent watchdog, the Sustainable Development Commission, has called for a rethink of aviation policy on environmental and economic grounds (Milmo & Vidal, 2008). An accumulation of concerns lies behind recent opposition to various economic develop ment strategies based on airports, with disquiet over proposals for Hamilton, Ontario a notable example (McGreal, 2008). The opposition that conflunts many airports today inevitably extends into questioning the broader concept of airport-led regional development

Towards Sustainable Airports? At the most extreme in the debate over aviation, diametrically opposed positions have been staked out. The environmentalist position calls for demand management with a tigh tening of government policies and international regulations to constrain demand for air travel. The argument is that aviation is locked too much into a 'predict and provide' mentality.. There is serious discussion about increasing the cost of air travel over and above market forces to dampen demand and offset emissions by adding compulsory environmental charges. The pro-growth position highlights the economic disbenefits of restricting airport capacity. A range of negative impacts are highlighted. One estimate ofthe failure to increase capacity to meet growing demand is a reduction in gross domestic product at a national level by 2.5-3% (York Aviation, 2004). Aviation-linked or orien tated businesses and activities within wider airport catchments must be factored into the discourse of declining patronage, investment, profitability, and job growth .. The concept of sustainable aviation appeals as a middle ground and there are initiatives in this area. For example, Sustainable Aviation in the UK represents a comprehensive industry-led strategy for the long-term sustainability of the UK aviation industry .. Under the group's compact, member airports are committed to a sustainable future through MAA-4-2012-Report excerpts distributed at the meeting

Planning, Sustainability and Airport-Led Urban Development 171 improving local air quality, reducing carbon footprints by implementing energy saving measures, engaging with local communities and other key stakeholders, and sharing environmental management best practices (wwwsustainableaviation.co). The aviation industry is demonstrably seeking to address environmental mitigation through bodies such as Airports Council International (2003) and the International Civil Aviation Organisation (2007). Individual airports have also developed sustainability strategies (e.g. Gatwick Airport Limited, 2000) and environmental management systems (Brisbane Airport Corporation, no date) .. The critical question remains: is there such a thing as sustainable aviation? Environ mental critics distinguish between eco-efficiency and sustainability (Upham et ai., 2003). Eco-efficiency is measured by the environmental impact per unit of business performance (less noise, better fuel economies, less CO2 emissions per passenger kilo metre, etc) but does not imply any constraint on growth in the scale of an activity. From this standpoint, the aim is to mitigate impacts but not at the expense of growth .. However, sustainability, in the fullest sense of the term, connotes precaution and limits to growth .. The paradox for a growth-oriented industry is that savings flom efficiency are likely to be outweighed by the additional resources required to support growth .. The inescapable conclusion in this scenario is that airport growth may be inherently unsustainable. The almost impossible job of trying to reconcile the poles ofenvironment and develop ment has fallen to national aviation strategies. In the UK, a 2003 White Paper set out a future strategic framework for the development of airport capacity (UK Department of Transport, 2003). While taking a 'measured and balanced view', the report was criticized for its expansive predictions for, and accommodation of, air travel demands .. Nevertheless, its starting point was better utilization of existing airport capacity rather than the building of new airports with provision for some additional runways, notably at Stansted and Heathrow in the London region .. The Labour Government's subsequent approval of a third runway for London's Heathrow in January 2009 unleashed a remarkable coalition of opposition including local authorities and communities, environmentalists, the Greater London Council, the Royal Town Planning Institute, the Sustainable Development Commission, and even the Conservative Party.. The White Paper's balancing of the economic and environmental dimensions stressed the need for effective local environ mental controls over matters such as air and noise pollution, loss of landscape and built heritage, biodiversity, water quality, and public transport access while at the same time recognizing the importance of airports in regional economic development There was acknowledgement of wider spatial impacts through support for 'aviation-related business clusters', the potential role of airports in urban regeneration, and proposals to establish Centres of Excellence for aircraft maintenance. More detailed planning was devolved to regional spatial strategies and local development fl·ameworks coordinated with new airport master plans. Development of a national aviation policy has also been in train in Australia through 2008-2009. Coordinated by the Department of Infrastructure, Transport, Regional Development, and Local Government, it has produced consultative issues and green papers en route to releasing a final white paper in the second half of 2009 (Commonwealth ofAustralia, 2008).. The scope ofthis inquiry is wider and also encompasses issues such as security, safety, consumer protection, and workforce training .. Nevertheless, it is being conducted at a time of community concern at the sequestering of local and state MAA-4-2012-Report excerpts distributed at the meeting

172 R.. Freestone

governments from involvement in aggressively commercial airport development since privatization as well as broader concerns about the environmental costs and climate change implications of high carbon transport strategies .. The Green Paper states a basic position that 'The Government will work to ensure that an appropriate balance is main tained between the social, economic, and environmental needs of the community and the development of [airports]'., To date, documentation has eschewed the level of spatial detail in the British policy, instead, pledging a broad commitment to collaborative decision-making and planning, involving better cooperative arrangements with the states and tenitories, more consultative arrangements with communities, and more transparent plan and development assessment. Nevertheless, the Federal Minister will still retain the final decision-making authority for major airport land use planning and development. Aviation is also set to be included in a national emissions trading scheme but specific ideas are also floated including increased operational efficiencies, voluntary offsetting schemes, and better data for comprehensive carbon monitoring. A disappointment of the Green Paper in terms of planning and development is that it still largely treats the airport in a very traditional fashion disconnected from the broader metropolitan and regional fabric. The important role of aviation as a driver of off- but near-airport economic development is muted. Beyond the airport boundary, the Green Paper's interests narrow sharply to noise, building height, and flightpaths., Both national inquiries highlight the policy challenge of truly sustainable aviation. Commercialization, privatization, and globalization trends have exacerbated the dilemma., Airports, like airlines, effectively 'have no rational alternative but to cater to existing markets in ways that generate most profit, while fostering future growth'-a conundrum that compromises 'the entire idea of reconciling aviation growth and sustain ability' (Graham, 2003). The schism so dominates the field that environmental impacts inevitably dominate planning decision making for airport matters, A planning challenge lies ahead in incorporating a vision of planning the airport in its urban context based not just on preventing incompatibility but towards promoting compatibility in the fullest sense of environmentally sustainable development.

Conclusion This is an uncomfortable note on which to end" Urban planning must playa key role in balancing the bullish pro-growth stance evident flom an aviation industry perspective and the more critical perspective questioning current trends flom a sustainability perspective, especially as the scale of airport development impacts way beyond airport boundaries. Kasarda (2000) acknowledges that 'neither the presence of an airport nor planning alone makes for a successful aerotropolis', On economic grounds alone, there are many critical factors, which will impinge on the likely economic success of such a venture., Developing the analysis of Dempsey et al. (1997), these would include the nature and scale of airport activity including its potential to be a hub; the passenger cargo split; support flom air caniers; goodness-of-fit into the overall pattern of metropo litan development; support flom business and political circles; state incentives; an adequate area transportation system and accessibility; room for expansion; the cost of land; control of real estate speculation; and the airport planning and urban planning frameworks. For the aerotropolis concept to even approach its potential as a sustainable development concept, these last two factors must be better synchronized, MAA-4-2012-Report excerpts distributed at the meeting

Planning, Sustainability and Airport-Led Urban Development 173

How is that to be done? Conventional planning approaches seem increasingly problema tical Major planning inquiries on airports rank highly as either disastrous (Hall, 1982) or, at best, protracted and controversial The £80 m inquiry into a fifth terminal at Heathrow (1993-2001) was Britain's longest-ever planning inquiry but even then was just part of the process from design to completion stretching over two decades (UK Competition Commission, 2007). More consultative strategies negotiating 'contracts' between airports and communities have won favour (Graham, 2003), A more opportunistic forum orientated approach has also been advocated (De long, 2008), Likewise, there is no one spatial blueprint to fit all situations" District (Blanton, 2004), smart growth (Lee et at., 2008), and conidor (Schaafsma et at, 2008) strategies have all been advanced" In other instances, the commercial development linked to airports might even be desirably barred from the immediate airport environs for environmental and operational reasons (Government Office for the East Midlands, 2009), Longhurst et at (1996) importantly emphasize the 'triple bottom line' aspects of a sustainable airport that also guides us towards social and economic issues alongside the environmental. Blanton (2004) and Hicks (2007) draw attention to the importance of nestling within broader principles of good urban planning practice" Together with the experiences emerging from actual airport development projects, some fundamental principles towards planning for a sustainable aerotropolis begin to emerge:

• realistic economic forecasts as the basis for development and expansion; • caution about impacts of new development upon the existing environment; • incorporation of aviation into urban and community visions; • a shared sense of responsibility and pUIpose among key stakeholders; • a district wide comprehensive plan that provides for organized land use, environmental protection, and multi-modal mobility; • an economic development and marketing strategy that defines an airport region and provides tools to attract and retain investment; • a governance f1:'amework that facilitates coordination of all relevant public agencies; • an open dialogue and partnership between airport and wider community; and • consistency of relevant plan objectives and territorialities at the airport, local area, region, metropolis, state, and national scales

In short, the stance ofthe UK Town and Country Planning Association (2006) state in their aviation issues paper is compelling if admittedly only flagging the challenges ahead: airport development 'should be fully integrated into the planning and development of sustainable communities'.

Note

1. 'The Airport Metropolis: Managing the interfaces', Australian Research Council Grant LP0775225, Queensland University of Technology with UNSW, Southern Cross University and industry partners, Chief Investigator: Douglas Baker, QUT This paper was OIiginally presented at the 2nd Ajman Urban Planning Conference at the Ajman University of Science and Technology in the United Arab Emirates, March 2008 I am most indebted to Professor Sabah Mushatat of the University of Wolverhampton for his organization ofthat event. I also thank Doug Baker, Mike Jenks, Kym Foster and the referees for their critical commentaries on the written version MAA-4-2012-Report excerpts distributed at the meeting

174 R, Freestone

References

Adey, P (2006) Airports and air-mindedness: Spacing, timing and using the Liverpool Airport, 1929-1939, Social and Cultural Geography, 7, pp, 343-363 Airport Innovation (2007) Ai/port cities & the aerotropolis: New planning models - an interview with Dr.. John D, Kasarda, April, pp, 106-110 Airports Council International (2003) Airports and the environment, ACI Policy Handbook, 4th ed, (Geneva: Airports Council International) Al Chalabi, M, (1998) The impact of a major commercial airport on urban development: A case in Chicago, Ekistics, 65, pp 131-142, American Planning Association (2006) Planning and Design Standards (Chicago: American Planning Association) Ayres, E. (2001) Airports and cities: Can they coexist? World Watch, 14(4), pp 22-33 Bagaeen, S, (2007) Brand Dubai: The instant city; or the instantly recognizable city, International Planning Studies, 12, pp, 173-197 Blanton, W, (2004) On the airfront, Planning, 70(5), pp, 34-35, Bowen, J, T, & Slack, B.. (2007) Shifting modes and spatial flows in North American freight transportation, in T. R Leinbach & C Capineri (Eds) Globalized Freight Transport Intermodality, E-Commerce, Logis tics and Sustainability, pp, 17-53 (Cheltenham: Edward Elgar) Brisbane Ai/port Corporation (no date) The Sustainable Airport (Brisbane: BAC) Available at http://www brisbaneairport,com,au/files/pdfJB_Sustainability_brochure ,pdf (accessed June 2008) Brueckner, J, K (2003) Airline traffic and urban economic development, Urban Studies, 40, pp, 1455-1469 Castells, M, (1996) The Rise ofthe Network Society, L (Oxford: Blackwell) Charles, M, B, Barnes, P, Ryan, N, & Clayton, J (2007) Ai/port futures: Towards a critique of the aerotropolis model, Futures, 39, pp, 1009-1028 Commonwealth of Australia (2008) National Aviation Policy: Green Paper (Canbena: Department ofInfrastruc ture, Transport, Regional Development and Local Government) Conway, H M, (1980) The Airport City, Development Concepts for the 21st Century (Atlanta: Conway Publications Inc) De Jong, B.. (2008) Glocal complexity and airport planning: Towards a more relational planning approach, in D, de Jong, B, Kaashoek & W,-L Zondag (Eds) Blue Skies or Storm Clouds? Essays on Public Policy and Air Transport, pp, 65-71 (Amsterdam: Aerlines Magazine Foundation) Dempsey, P S, Goetz, A, R, & Szyliowivcz, J, S, (1997) Denver International Airport Lessons Learned (New York: McGraw Hill) Doganis, R (1992) The Airport Business (London: Routledge)" Erie, S, P (2004) Globalizing LA. Trade, Infrastructure and Regional Development (Stanford: Stanford University Press), Freestone, R, Williams, P & Bowdon, A (2006) Fly buy cities: Some planning aspects ofairport privatisation in Sydney, Urban Policy and Research, 24, pp, 491-508 Frej, A. B.. (2004) 'Introduction' to Just-in-Time Real Estate, How Trends in Logistics Are Driving Industrial Development, pp 1-2 (Washington DC: Urban Land Institute) Ganeau, J, (1992) Edge City,' Life on the New Frontier (New York: Anchor Books) Gatwick Airport Limited (2000) Gatwick Airport Sustainable Development Strategy (London: BAA Gatwick), Gillingwater, D, Humphreys, L & Watson, R (2003) Air freight and global supply chains: The environmental dimension, in P Upham, J, Maughan, D, Raper & C Thomas (Eds) Towards Sustainable Aviation, pp, 149-161 (London: Earthscan), Glasson, J & Marshall, T. (2007) Regional Planning (London: Routledge), Gordon, A. (2004) Naked Airport. A Cultural History ofthe World's Most Revolutionary Structure (New York: Henry Holt and Co) Gottdiener, M, (2001) Life in the Air Surviving the New Culture ofAir Travel (Lanham, MD: Rowman and Littlefield) Government Office for the East Midlands (2009) East Midlands Regional Plan (London: TSO) Graham, A (2001) Managing Airports, An International Perspective (Oxford: Butterworth Heinemann) Graham, B.. (2003) Air transport policy: Reconciling growth and sustainability?, in I, Docherty & J Shaw (Eds) A New Dealfor Transport? The UK's Struggle with the Sustainable Transport Agenda, pp 198-225 (Oxford: Blackwell) MAA-4-2012-Report excerpts distributed at the meeting

Planning, SU5tainability and Airport-Led Urban Development 175

Green, R K (2007) AirpOlts and economic development, Real Estate Economics, 35, pp. 91-112 Griggs, S. & Howmth, D. (2004) A transfOlmative political campaign? The new rhetmic ofprotest against airpOlt expansion in the UK, Journal ofPolitical Ideologie~, 9, pp. 181-201 Griggs, S. & Howarth, 0. (2008) Populism, localism and environmental politics: The logic and rhetmic of the Stop Stansted Expansion campaign, Planning Theory, 7, pp. 123-144 Guller, M. & Guller, M.. (2002) From Airport to Airport City (Barcelona: Editorial Gustavo Gill) Hack, G (2000) Infiastructure and regional form, in R Simmonds & G. Hack (Eds) Global City Regions Their Emerging Forms, pp. 183-192 (London: Spon Press). HakfoOlt, J., Poot, T & Rietveld, P (2001) The regional economic impact of an airpOlt: The case of Amsterdam Schipol Airport, Regional Studies, 35, pp. 595-604 Hall, P (1982) Great Planning Disa~tefj (Berkeley: University of California Press) Hall, T (2001) Urban Geography, 2nd ed. (London: Routledge). Hanis, N. (1994) The emerging global city: TranspOlt, Citie~, 11, pp .. 332-336 Hanis, C D.. (1997) The nature of cities and urban geography in the last half century, Urban Geography, 18, pp 15-35 Hicks, C (2007) Urban planning and airpOlt issues (2007) Paper delivered to the AirpOlt Cities Conference (Insight Media), Singapore, on disk Hooper, P (2007) Sustainable development: A European perspective on implications fm airports of the future Paper delivered to the Airport Development Conference (Aviation Business), Sydney, on disk International Civil Aviation Organisation (2007) lCAD Environmental Report 2007 (Montreal: ICAO Environ- mental Unit). Kasarda, L (2000) Planning the aerotropolis, Airport World, 5(5), pp. 52-53 Kasarda, L (2001) Logistics and the rise of the aerotropolis, Real Estate Is~ues, 25, pp 43-48 Kasarda, J (2006) The rise of the aerotropolis, The Next American City, 10, pp .. 35-37 Kasarda, J. & Sullivan, D. L (2005) Air cargo, liberalization, and economic development, Annal~ of Air and Space Law, 31, pp. 167-193 Knorr-Cetina, K (2006) Microglobalisation, in L Rossi (Ed .. ) Frontiers of Globaliwtion Research Theoretical and Methodological Approaches, pp 65-92 (New YOlk: Kluwer) Kramer, D. (2004) How airport noise and airpOlt privatisation effect economic development in communities surrounding U.S. airpOlts, Transportation Law Journal, 31, pp 213-247 Le COlbusier (1987) The City of To-morrow and its Planning, first published 1929 (New York: Dover Publications) Lee, R W, Gosling, GD & Irvin, K (2008) Applying Smart Growth Principle~ and Strategies to Resolving Land Use Conflicts around Airports (San Jose: Mineta TranspOltation Institute) Leinbach, T R. (2004) City interactions: The dynamics of passenger and fr·eight flows, in S Hanson & G Giuliono (Eds) The Geography of Urban Transportation, 3rd ed, pp 30-58 (New York: The Guilford Press) Leinbach, T. R& Bowen, J. T, JI (2004) Air cmgo services and the electronics industry in Southeast Asia, Journal of Economic Geography, 4, pp. 299-321 Longhurst, J., Gibbs, D C, Raper, D. W & Conlan, D E. (1996) Towards sustainable airpOlt development, The Environmentalist, 16, pp. 197-202 Macintosh, A & Downie, C (2007) A Flight Risk? Aviation and Climate Change in Au~tralia, Discussion Paper No. 94 (Canberra: The Australia Institute) May, M. & Hill, S. (2006) Questioning airport expansion -A case study of Canberra International AirpOlt, Journal ofTran~port Geography, 14, pp. 437-450. McGreal, R. (2008) Dance, aerotropolis, dance!, Raise The Hammer, June 5. Available at wwwraisetheammer Olg (accessed June 2008) Milmo, D. & Vidal, J (2008) Put UK airpOlt expansion on hold, demands green group, The Guardian, May 21 Moore-Wilton, M.. (2007) Aerotropoli: AirpOlt cities .. Paper to the Capitals Alliance Conference, Canberra, Australia. Nunn, S & Schoedel, C (1995) Cities and airport-based economic development strategies: An analysis ofpublic private deals to provide airline maintenance operating centers in urban airports, Economic Development Quarterly, 9, pp. 159-173 O'Connor, K & Scott, A (1992) Airline services and metropolitan areas in the Asia-Pacific region 1970-1990, Review of Urban and Regional Development Studies, 4, pp. 240-253 Pearman, H. (2004) Airports. A Century of Architecture (London: Laurence King) MAA-4-2012-Report excerpts distributed at the meeting

176 R. Freestone

Reiss, B (2007) Maximising non-aviation revenue for airports: Developing airport cities to optimise real estate and capitalise on land development opportunities, Journal ofAirport Management, 1, pp.. 284-293 Schaafsma, M., Amkreutz, 1. & Gukker, M. (2008) Airport and City. Airport Corridors .. · Drivers of Economic Development (Amsterdam: Schiphol Real Estate). Short, J. R (2004) Global Metropolitan Globalizing Cities in a Capitalist World (London: Routledge) Siebert, L (Ed..) (2008) Airport Cities The Evolution. The Transformation ofAirports into World-Class Airport Cities (London: Insight Media) Sit, V. (2004) Global transpark: New competitiveness for Hong Kong and South China based on Air Logistics, Geografiska Annala, 86, pp.. 145-163 Town and Country Planning Association (2006) Policy Statement. Aviation (London: TCPA) Twomey, J. & Tomkins, L (1995) Development effects at airports: A case study of Manchester Airport, in D.. Bannister (Ed) Transport and Urban Development, pp. 187-211 (London: E & FN Spon) UK Competition Commission (2007) BAA Airports Market Investigation Planning Issues and Competition in the Airports Market (London) UK Department of Transport (2003) The Future ofAir Transport, White Paper (London) Upham, P, Thomas, C, Gillingwater, D. & Raper, 0. (2003) Environmental capacity and airport operations: Cunent issues and future prospects, Journal ofAir Transport Management, 9, pp. 145-151 Urry, J. (2007) Mobilities (Cambridge: Polity). van den Berg, L, Van Klink, H. A. & Pol, P. M. J. (1996) Airports as centres of economic growth, Transport Reviews, 16, pp. 55-65 Waldheim, G. & Berger, A. (2008) Logistics landscape, Landscape Journal, 27, pp .. 219-246 Warffemius, P, van der Hoorn, T. & Klaassen, H. (2009) The dynamic spatial impact of Amsterdam Airport Schiphol, Aerlines, 42, pp. 1-4 Weisbrod, G E., Reed, L S. & Neuwirth, R M (1993) Airport area economic development modeL Paper delivered to the PTRC International Transport Conference, Manchester Williams, N .. (2000) Is aviation and airport development good for the economy? in C. Evans (Ed) The Plane Truth Aviation and the Environment, pp. 23-26 (London: The Ashden Trust). York Aviation (2004) The Social and Economic Impact of Airports in Europe (Geneva: Airports Council International) MAA-4-2012-Report excerpts distributed at the meeting

Journal of AIR TRANSPORT MANAGEMENT PERGAMON Journal 01 Air TianspOlt Management 7 (2001) 159-167 www elsevier.com!locate/jairtraman

The geography of air passenger volume and local employment patterns by US metropolitan core area: 1973-1996 Keith G. Debbage*, Dawn Delk

Department ofGeography, Univer:sity oj North Carolina at Greensboro, Greensboro, NC 27402, USA

Abstract

The pUipose of this paper is to determine if a statistically significant relationship exists between administrative and auxiliary employment levels and air passenger volume for the top 50 urban-airport complexes in the United States from 1973 to 1996 The goal of this paper is a fairly modest one ~ to refine and expand the current literature's focus by conducting a broader investigation of the links that exist between air passenger volume and employment levels within local economies Based on data hom the Federal Aviation Administration (F AA) and the US Census Bureau County Business Patterns, the major findings of this paper were that the correlation between administrative and auxiliary employment and enplaned passenger volume over time are statistically significant at the 1% level. © 2001 Elsevier Science Ltd. All rights reserved

Keywords Air transportation; Economic development; Metropolitan aIeas

1. Introduction just such a sector where the knowledge-economy procliv ities of such workers inevitably trigger disproportionate Since the Airline Deregulation Act of 1978, air trans ly higher propensities to fly (Debbage, 1999; Ivy et aL, portation passenger volume has increased dramatically 1995). The administrative and auxiliary sector primarily in the United States. By 1998, US air passenger volume consists of workers engaged in activities such as manage had risen to a total of just over 600 million passengers ment, research and development, financial services and - an increase of more than 300% on the comparable supporting services such as accounting and data process figures for 1973 (Air Transport Association, 1999).. These ing. In related research, Button and Taylor (2000) cite statistics suggest a radical shift in the absolute and rela consultant reports that suggest that those employed in tive geography of air passenger volume at US airports, 'new economy' activities like information technology, but it is less clear what forces are behind these rapid biotechnology, electronics, and management services will growth patterns 'on the ground' in terms of the corre fly over 1.6 times as much as those in traditional indus sponding shifts in the composition of local and regional tries.. In 1996, the adminstrative and auxiliary sector labor markets. accounted for 1.64 million workers (or 4.2 % of the labor A considerable proportion of airline passengers in the force) in the top 50 metropolitan core economies under United States travel for business purposes suggesting study in this paper.. More importantly, some of the fas that a close relationship exists between business activity test-growing economies in the United States generated on the ground and airline networks in the skies .. Even a disproportionately significant administrative and aux with recent technological innovations that minimize the iliary sector (e.g., in 1996 this included: Atlanta - 8.2% need for direct face-to-face contact, many economic sec of the labor force, Memphis - approximately 9.75%, tors still rely heavily on direct contact with colleagues, and Portland - 8.. 8%). suppliers, customers, and other key employees. Adminis Given this context, the purpose of this paper is to trative and auxiliary employers are a classic example of determine if a statistically significant relationship exists between administrative and auxiliary employment levels * Corresponding author Tel: + 1-336-334-3911; fax: + 1-336-334 and air passenger volume for the top 50 urban-airport 5864 complexes in the United States from 1973 to 1996 .. The E-mail address [email protected] (K G Debbage) goal of this paper is a fairly modest one - to refine and

160 KG Debbage, D Delk / Journal oj Air Transport Management 7 (2001) 159-167 expand the current literature's focus by conducting was occurring in the growth of airline network systems a broader investigation of the links that exist between air over time. By contrast, Smith and Timberlake (1998) passenger volume and employment levels within local focused on the role of air passenger volume and origin economIes. destination links in identifying world cities in the global transportation network, and they argued that cities with major airports play critical roles in serving as "key points 2. Previous studies of exchange in the world economy". In simplistic terms, distance still seems to matter be Transport has long been seen as a strong pOSItIve cause knowledge is more easily exchanged as the level of influence on economic development (Bell and Feitelson, shared experiences increases - a phenomenon that 1991; Button and Lall, 1999; Button et aI, 1999; Button Nooteboom (1999) refers to as 'cultural proximity' .. Fur and Taylor, 2000; Debbage, 1999; Goetz, 1992; Irwin and thermore, cultural proximity can be enhanced by spatial Kasarda, 1991; Ivy et aI, 1995; Van den Berg proximity between firms, suppliers, and customers et aI, 1996). However, the exact role that transportation Nooteboom argues that important keys to knowledge plays in shaping growth and economic development exchange (such as reputation, bonding and trust) are patterns, and how to assess the interactive effects of best achieved when the spatial, cognitive and cultural one on the other, are still the subject of much debate distances are minimized .. Such a phenomenon can be The links that exist between transportation and particularly crucial for administrative and auxiliary economic development can be both direct and indirect employees that are involved in collaborative research For example, efficient transport networks can facilitate and development activities that demand hequent face low shipping costs that can allow wider markets to to-face contacts .. be served and that can also induce economies of scale, Although agglomerative or highly clustered urban scope, and density in an extensive range of activities markets were traditionally intended to minimize Examples of more indirect links include the employment transportation and labor costs, Porter (1998) has argued creation induced when constructing transportation that contemporary metropolitan cluster advantages infrastructure projects and the multiplier effects triggered now "rest on information, transaction costs, comp by the large inputs of raw material and labor needed for lementarities, and incentives as well as 'public' goods that construction .. result hom both public and private investments".. It According to Bell and Feitelson (1991), an efficient is suggested in this paper that airports are part ofPorter's transportation network serves two primary purposes in 'public goods' equation because many airports are any urban hierarchy - it facilitates the movement of operated and managed by quasi-public airport authori goods and services and it allows for the movement of key ties, and they can often exaggerate the competitive ad employees in a timely and reliable manneL These assets vantages oflarge metropolitan markets like New York or can be critical for those elements of the administrative Los Angeles .. and auxiliary sector that require frequent and direct Airports can serve the regional or local agglomeration contact with key personnel in other metropolitan mar in at least three fundamentally different ways - by kets, since an efficient air transportation network can providing access to the air transportation system, by expedite such transactions .. However, the debate about acting as a local employment generator, and by trigger whether transportation guarantees or simply allows for ing or encouraging additional off~site jobs as ancillary the possibility of economic development in general and complementary businesses cluster close to the air - and employment in particular - continues in the port location .. According to Van den Berg et al. (1996), literature. airport regions are becoming attractive locations for Developing a better understanding of how air trans businesses in their own right, making them potential portation networks can shape local employment patterns centers of economic growth with a capacity for is critical because the dominant form of long-distance significant spin-off effects .. passenger transportation in the US is air transportation. Better understanding the role that airports play in any In an analysis of US urban areas hom 1950 to 1986, urban agglomeration is critical because the 'accessibility Goetz (1992) found that prior growth in the population through airports' issue has assumed an elevated role in and employment levels of metropolitan areas partly ex answering the 'how' and 'where' of the geography of plained subsequently higher levels of air passengers per economic activity in the American economy. Irwin and capita .. In attempting to better understand these growth Kasarda (1991) examined the empirical relationships that patterns over time, Irwin and Kasarda (1991) were con existed between airline networks and overall employ cerned with the changing 'centrality' of America's major ment growth rates in 104 US metropolitan areas between airports. They argued that just as industries and people 1950 and 1980. They argued that accessibility levels have were leaving the Northeast/Midwest manufacturing belt changed constantly as new transportation innovations region for the South/West sunbelt regions, a similar shift (e .. g., rail, car, jet engine) have reshaped the competitive MAA-4-2012-Report excerpts distributed at the meeting

KG Debbage, D. Delk / Journal oj Air Transport Management 7 (2001) 159-167 161

advantage of the US spatial economy.. Irwin and urban air transportation network can fundamentally in Kasarda also suggested that in the post-world war 2 era, fluence the locational patterns of this particular sector of air transportation substantially reduced hictional con the economy. straints to long-distance economic interaction to the Debbage (1999) confirmed some of the early research point that new locational advantages were created for conducted by Ivy et at (1995) by analyzing the changing some metropolitan areas, particularly for the manufac administrative and auxiliary employment levels and air turing and producer service sectors of the US economy port passenger volume for the 10 largest airports in the They concluded that "changes in air transportation have US Carolinas. Debbage (1999) concluded that the host altered the competitive advantages of metropolitan counties which "experienced significant gains in air pas areas, and not the reverse" (Irwin and Kasarda, 1991) senger volume and air service connectivity also experi particularly in markets that are centrally located relative enced comparable gains in the employment levels of to existing airline networks .. administrative and auxiliary workers, particularly in the Button and Lall (1999) confirmed that the direction of manufacturing sector". causation was from air service availability to employ However, the research conducted thus far connecting ment growth in an analysis of how US hub airports with administrative and auxiliary employment levels to air international gateways correlate to "new economy" em port-airline operations has had its limitations .. Ivy et al ployment levels in the local economy.. Supporting evid (1995) limited their study to an analysis of airline route ence is provided by Button et al (1999) in a study of connectivity and not airport passenger volume, while hi-tech employment in hub airport markets where the Debbage (1999) limited his analysis to just the US authors found that "hubs create employment rather than Carolinas .. This paper will attempt to expand and update airlines selecting cities as hubs simply because they are this research agenda by studying the 50 largest metro already dynamic". However, Button and Taylor (2000) politan markets in the United States in terms of air temper these findings by suggesting that the benefits of passenger volume from 1973 through 1996 to determine additional international airline connections on "new if changes in air passenger volume correspond to changes economy" employment levels are not infinite. in administrative and auxiliary employment levels over What was less clear in all this research was how these time. By focusing on passenger volume, some insight is changes over time in airline services influenced more provided on the scale of service provided at specific specific sectors ofthe local economy, especially industries airports rather than the variety of destinations served .. highly sensitive to changes in airline connectivity levels Furthermore, passenger volume serves as a reasonable like the administrative and auxiliary sector? Fortunately, proxy for seat capacity with the assumption that large, some research has already been conducted in this area. sophisticated urban agglomerations will offer substantive Ivy et al (1995) argued that changes in air service hub operations that accentuate the significance ofecono connectivity can lead to corresponding changes in ad mies of scale, scope, and density. ministrative and auxiliary employment levels (or what they referred to as "professional employment"). They demonstrated that "significant statistical relationships 3. Definitions and data sources exist between changes in connectivity and professional employment" (Ivy et aI, 1995). They also argued that According to the US Department of Transportation's although large cities are commonly associated with National Transportation Statistics (1996), the top 50 US a number of negative characteristics such as higher land airports accounted for approximately 83% of total pas costs, higher taxes, and increased competition for profes senger enplanements in 1996 .. The data set in this paper, sional labor, they remain attractive both to firms and however, consists of the top 50 airport complexes in the professionals because of the advantages rendered by United States, and not merely the top 50 individual urban agglomerative economies .. Locating in a metro airports, to better reflect the flight and airport choices politan area can give companies "an ample supply of available to an administrative and auxiliary worker in professional workers, a wide variety ofsuppliers, services, any given metropolitan market (e .. g., the New York area and information, not to mention the all-important inh:as includes JFK, La Guardia, and Newark Airports). En tructure .. [including] airports with hequent air service planed passenger volumes were collected from the US to a large variety of destinations" (Ivy et aI, 1995). Ac Federal Aviation Administration (1973, 1983, 1996) cording to Ivy et al (1995), the volume, variety and where an enplaned passenger is defined as any "revenue frequency of air service is important because "access to passenger boarding an aircraft" (FAA, 1996). a large number of destinations facilitates face-to-face Administrative and auxiliary employment data were interaction and helps satisfy corporate travel needs" collected from the US Bureau of the Census (1973, 1983, They suggested that h'equent face-to-face contact can be 1996) County Business Patterns for the host counties that especially important in the administrative and auxiliary make-up each of the 50 largest urban-airport complexes sector to the point that significant restructuring in the under study. If the local built-up area surrounding an MAA-4-2012-Report excerpts distributed at the meeting

162 KG Debbage, D. Delk / Journal oj Air Transport Management 7 (2001) 159-167

urban-airport complex had multiple airports in multiple a disproportionate share of the air passenger market counties then the employment levels were aggregated fOl These included the Chicago area atjust under 16 million the chosen counties (e .. g., the New York area included enplaned passengers (including both O'Hare and Mid JFK and La Guardia airports in Queens County and way) and the New York area with just over 18 million Newark Airport in both Essex and Union Counties, NJ). enplaned passengers (i.e .. , JFK - 7.4 million, La Guardia A second concern arose when the major airports for an - 7J million, and Newark - 3.5 million). Collectively, area were located in a largely peripheral, suburban the 50 urban-airport complexes under study accounted county that was outside the built-up urbanized area forjust under 151 million enplaned passengers or 80% of where the chosen county did not completely capture the all enplanements nationwide.. By contrast, in 1996 the 50 local labor market As a result, additional contiguous largest urban-airport complexes in the United States counties were added to the data set where deemed neces accounted for approximately 488 million passengers or sary to more accurately reflect the employment composi roughly 87% of all US enplanements suggesting that the tion of the local economy (e .. g., the New York area airports under study had elevated their market share of included not just Queens, Essex, and Union counties, the total traffic base (i.e., 1973 - 80%, 1983 - 85%), but also New York County to capture the significant and indicating that an on-going process of spatial number of administrative and auxiliary workers that live concentration was at play. and work in Manhattan even though New York's major By contrast, the market share (%) of the five largest airports are in neighboring Queens and New Jersey). urban-airport complexes dropped slightly hom 1973 to Administrative and auxiliary employment is defined in 1996, indicating that a process of spatial deconcentration the Office of Management and Budget Standard Indus was also underway as the forces of deregulation un trial Classification Manual as any establishment primarily leashed new competitive advantages in locations such as engaged in performing management, supervision, general Atlanta, Dallas, Denver, Houston, Las Vegas, Los administrative functions, and supporting services for Angeles, and Phoenix .. What is less clear is whether or not other establishments of the same company, rather than these profound geographic shifts in air passenger volume for the general public or other business firms. Specific corresponded to equivalent shifts in the administrative examples of auxiliary establishments include central offi and auxiliary sector of the nation's economy.. ces, executive offices, corporate offices, regional offices, Fig. 2 illustrates the geography of administrative and marketing, accounting, public relations, budget, book auxiliary employment for the 50 largest urban-airport keeping, data processing, research and development, test complexes in the United States for 1973 and 1996. ing laboratories, advertising, but also warehousing, and In 1973, the top five urban-airport complexes in terms milk-receiving stations .. In this paper, data were collected of total administrative and auxiliary employment on total aggregate administrative and auxiliary employ included the New York-Newark area with a total of ment levels in each urban-airport complex plus data on 188,000 administrative and auxiliary employees manufacturing-specific administrative and auxiliary em accounting for 63% of total employment in the area.. By ployment levels (which commonly account for one-third contrast, the second-placed Chicago area (which of all administrative and auxiliary employment in most included both Cook and Dupage Counties, and O'Hare metropolitan markets). and Midway airports) generated approximately 176,000 In order to capture the changing relationship between administrative and auxiliary workers or 7..5% of total air passenger volume and administrative and auxiliary employment employment over time, data were collected for 5 years Although one might expect the most heavily trafficked either side of the 1978 Airline Deregulation Act (i.e., 1973 air passenger markets to have the largest total employ and 1983). The most current data available at the time of ment centers, this is not always the case as compared to writing was also included in the data set (i .. e., 1996). By the geography of the administrative and auxiliary hier way of a final caveat, it should be noted that in some archy.. For example, the third and fifth largest adminis cases the published data for administrative and auxiliary trative and auxiliary employment markets in 1973 were employment were reported as a data range for reasons of the Detroit urban area (i.e .. , Wayne County) which gener confidentiality. In those cases, the midpoint of the range ated 84,000 administrative and auxiliary employees and was used to calculate correlation coefficients and mean the Greater Pittsburgh area (i.e .. , Allegheny County) values in an attempt to minimize error and bias .. which generated approximately 57,000 equivalent workers. In both Detroit (9% of total employment) and Pittsburgh (10.4%), the administrative and auxiliary sec 4. Findings tor was proportionally more important to the local econ omy than it was for either New York (6.3%) or Chicago Fig. 1 illustrates the geography of the largest urban (7..5%). Furthermore, a noticeably greater proportion of airport complexes by enplaned passenger volume for the administrative and auxiliary workers in both Detroit 1973 and 1996. In 1973, a select few places captured and Pittsburgh were employed in manufacturing-related MAA-4-2012-Report excerpts distributed at the meeting

KG Debbage, D Delk / Journal ofAir Transport Management 7 (2001) 159-167 163

1973

;.;j». Number of Enplaned Passengers for the 50 Largest U.S. Urban Complexes

Source: FAA Sfafismal Handbook of Avfallon, 1973, 1996 Map prodL.ald byThlmas Triool.

Fig 1 Total enplaned passenger volume for the 50 largest US urban-airport complexes: 1973-1996

activities (i.e., 6..9 and 7.9%, respectively, compared to employment compared to 63% of all employment in just 4% in New York and 2.9% in Chicago) .. However, the New York area .. neither Pittsburgh nor Detroit ranked in the 1973 top 10 Perhaps the most striking finding is the stagnant ad in terms of air passenger volume suggesting that good air ministrative and auxiliary sector in the Atlanta market transport is not always required to attract industry to an (i .. e., Clayton and Fulton County) relative to its third area. Both Pittsburgh and Detroit developed single-sec placed ranking in terms of air passenger volume .. In 1973, tor propulsive industries early on in the 1900s (i.e .. , the Atlanta ranked 13th in administrative and auxiliary steel industry and automobile production, respectively), employment generation with 18,834 workers (or 44% and these industries were spatially fixed, and thus, less of total employment).. Although the "new South" likely to be influenced by changing levels of airline con was to emerge in subsequent years, the Atlanta market nectivity relative to more "footloose" industries. Further did not appear to have a sufficiently skilled labor pool more, both the steel and automobile industry developed to generate a healthy number of administrative and largely before the era where air transportation played auxiliary workers.. Furthermore, much of the traffic such a critical role in shaping the growth of large, metro base in Atlanta was reliant on connecting passengers, politan economies. and thus, local originating traffic was not as significant By contrast, the Los Angeles urban area (described as as in other metropolitan markets. Such findings are a Los Angeles, San Bernardino and Orange Counties in reminder that significant air passenger volume is this study so as to include LAX, Ontario International, not a guarantor of a prosperous regional economy, John Wayne Airport, and Hollywood-Burbank Airport) although things were about to change for Atlanta and seemed to behave in a more conventional fashion in other places. terms of its placement in the urban hierarchy. The In 1973, mean administrative and auxiliary employ Los Angeles area ranked fourth in both enplaned passen ment levels for the 50 urban areas under study was 21,216 ger volume (i.e .. , 9 million) and total number of adminis workers but this had increased in 1983 by more than trative and auxiliary workers (i.e., 72,860 employees in 150% to an average of 33,803 workers by urban area. In 1973).. However, the administrative and auxiliary sector 1983, the major employment centers for administrative played a less significant role in the highly diversified and auxiliary workers remained New York and Chicago Los Angeles economy accounting for only 24% of total which hovered just under the 200,000 mark much like MAA-4-2012-Report excerpts distributed at the meeting

164 KG Debbage, D. Delk / Journal ofAir Transport Management 7 (2001) 159-167

Total Administrative & Auxiliary Employment

0··9,914 • 10,737 ..• 19,502 • 20,658 - 49,776 • 50,602··89,177 e 102,513 "·188,242

Source: County Business PanerflS. 1973. 1996 Map produ:ed byThomaa Tocot

Fig 2 Iotal administrative and auxiliary employment for the 50 largest US urban-airport complexes: 1973 and 1996

in 1973 even though both areas experienced significant Additionally, the first signs of deindustrialization be increases in air passenger volume., Perhaps the most gan to creep into the data set as Pittsburgh experienced interesting departure hom the employment hierarchy a noticeable decline in administrative and auxiliary established in 1973 was the rapid rise of the Los Angeles workers hom 57,000 in 1973 to 44,500 in 1983 (a 22% urban area with a total of approximately 170,000 admin decline). The decrease in administrative and auxiliary istrative and auxiliary workers in 1983 (compared to only workers in Pittsburgh occurred even though air passen 72,860 in 1973), It appeared that large, sophisticated ger volume increased from 3.,6 to 55 million and US urban agglomerations like New York, Chicago, and espe Airways (formerly Allegheny Airlines) began to develop cially Los Angeles tended to attract additional economic a substantial hub operation out of the Pittsburgh activity through a process of circular and cumulative Airport Traditionally, the major propulsive industry in causation whereby economic growth in a region was Pittsburgh has been the steel industry and related manu essentially self-sustaining, Endogenous growth theorists facturing industries" Employers in traditional industries have argued that a significant element of this accelerated like these tend to have a lower propensity to fly relative growth process is infrastructural investment Conse to "new economy" activities like information technology, quently, the proliferation of both established and new electronics and various administrative and auxiliary airport operations in the Los Angeles area (e,g" LAX, functions where a premium is placed on face-to-face Orange County/John Wayne Airport, Ontario Interna contact and collaboration" Some of this rationale may tional, and Hollywood-Burbank Airport) all seemed to partly explain the discrepancy between rising air passen act to exaggerate the competitive advantage of adminis ger volume and declining numbers of administrative and trative and auxiliary establishments based in Los auxiliary workers in the Pittsburgh market As the US Angeles. However, the links between administrative and Airways hub was developed in Pittsburgh, the propor auxiliary employment levels and air passenger volume tion ofconnecting traffic began to rise such that much of are not straightforward, especially given the stagnant the growth in passenger volume had little to do with administrative and auxiliary employment growth rates in events in the local economy, Meanwhile, the manufactur New York and Chicago from 1973 to 1983, even though ing-related administrative and auxiliary sector in air passenger volume increased significantly in both cities Pittsburgh downsized and experienced a period ofsignifi over the same time period, cant job losses, MAA-4-2012-Report excerpts distributed at the meeting

KG Debbage, D Delk / Journal oj Air Transport Management 7 (2001) 159-167 165

Having said this, a significant proportion of adminis states (Atlanta, Dallas, Memphis, and Florida) and the trative and auxiliary workers still tend to be engaged in West Coast (Los Angeles, Portland, Seattle, and San manufacturing-related activities (commonly one-third of Jose).. For example, flom 1983 to 1996, Memphis gained all such workers).. Consequently, as America experienced approximately 27,000 administrative and auxiliary significant manufacturing job losses between 1983 and workers - a 250% increase on 1983 levels - and the 1996, the administrative and auxiliary sector experienced highest percentage growth rate in the study .. It is difficult similar net declines, though to a lesser degree .. By 1996, not to conclude that the establishment of both the FedEx the mean administrative and auxiliary employment levels and Northwest Airlines hub operations in Memphis for the urban areas under study in this paper dropped played some role in triggering this employment growth slightly to 33,324 workers .. The Los Angeles area had Although a cursory examination of Figs. 1 and 2 seem emerged as the leading employment center for adminis to indicate that the geographic changes in air passenger trative and auxiliary workers with nearly 160,000 volume mimic corresponding changes in administrative workers, although the sector still only accounted for and auxiliary employment levels over time, the experien a small proportion of total employment (i.e., 3.2%).. ces in Atlanta, Pittsburgh and other places raise concerns In 1996, three emerging "hot-spots" of administrative about the systematic nature of this relationship. and auxiliary employment were Atlanta (58,001), San To overcome some of these concerns, a Pearson's Jose (49,776), and Seattle (47,778).. Both San Jose and Product Moment Correlation Coefficient was calculated Seattle had only modest airport operations relative to the between the two variables with the assumption that as air other urban areas under study, although American Air passenger volume increases, administrative and auxiliary lines developed a mini-hub operation in San Jose during employment levels will increase in a similar fashion .. the late 1980s. Both Silicon-Chip Valley in San Jose and The correlation coefficient was significant at the 1% Microsoft in Seattle no doubt helped both regions to level for all 3 years under study (i.e .. , 1973 - 0.84, 1983 sustain above average administrative and auxiliary em - 0.83, 1996 - 0.83).. The high, stable and positive ployment levels. In San Jose, 66% of all administrative correlation coefficients suggest that a strong and predict and auxiliary employment were manufacturing-related able linear relationship exists between air passenger - a national anomaly. volume and administrative and auxiliary employment At a national scale, Fig .. 2 reveals that the geographic over time.. A visual inspection of the scatter diagram distribution of administrative and auxiliary employment for 1973 (Fig. 3) illustrates the dominance of New York had spatially de-concentrated away from the traditional and Chicago and the vitality of manufacturing cities northeastern manufacturing belt for places in the sunbelt like Detroit and Pittsburgh. Atlanta stands out as an

200000,..------, NEW YORK C CHICAGO I]

100000 DEIROn I] LA I] PITTSBURGH C I] DAllAS-FT W ORlH I]

I] AIlANIA I]

10000000 20000000

Enplaned Passenger Volume Fig 3 Scatter diagram of air passengers and employment by urban-airpOlt complex, 1973 MAA-4-2012-Report excerpts distributed at the meeting

166 KG Debbage, D. Delk/ Journal ofAir Transport Management 7 (2001) 159-167

180000

160000 1'0 CHICAGO o 140000

120000 DAl.LAS·,FW o NY 100000 o HOUSTONo 80000

60000 ATLANTA SAN JOSE o 0 40000 00 0 PITISBUROH o 00 0 0 o 0... 000 20000 g 0 000 ~ 0 00 o 0 'eP0 0 1ASVEOAS o gpo 0 o 10000000 20000000 30000000 40000000

Enplaned Passenger Volume

Fig 4 Scatter diagram of ail passengers and employment by urban-airport complex, 1996

anomaly because it has been unable to generate adminis opportunities, hence the negligible manufacturing-based trative and auxiliary employment opportunities at the administrative and auxiliary sector in Las Vegas. rate expected for the volume of air passengers generated by Atlanta Hartsfield Airport. The corresponding scatter diagram for 1996 (Fig 4) suggests that the hierarchy of 5. Conclusion administrative and auxiliary employment centers ap peared relatively stable, although the rapid ascendancy of The initial findings in this paper seem to confirm some select 'sunspots' in the Sunbelt and West Coast was of the earlier suppositions put forth by Debbage (1999), apparent (e .. g., Dallas-Fort Worth, Houston, Los Button and Lall (1999), Button et at (1999), Button Angeles, and San Jose). The changing competitive land and Taylor (2000), Goetz (1992), Irwin and Kasarda scape in the airline industry during the post-deregulation (1991), and Ivy et at (1995). Statistically significant era and the evolution of new fortress hub-and-spoke links exist between air transportation and economic systems in places like Dallas (American, Delta, and development, particularly as measured by the ability of Southwest Airlines) and Houston (Continental Airlines) certain metropolitan areas to generate employment may partly account for the significant employment gains opportunities in those sectors of the economy that in these metropolitan areas .. stimulate unusually high propensities to fly due to However, although activity at any airport are closely the crucial importance of face-to-face contact and direct connected to the complex web of urban and regional collaboration .. economic activity SUI rounding the airport region, the As administrative and auxiliary-related jobs and in case of Las Vegas highlights the difficulties encountered dustries shifted away hom the traditional manufacturing when making generalizations on causality. Although the centers of the Northeast and Midwest to the South and Las Vegas airport handled almost 15 million passengers West, the air transportation network appeared to experi in 1996, it generated far fewer administrative and auxili ence a similar geographic shift as it broadened into ary workers than expected .. Some of the explanation may a more deconcentrated air transportation network lie with the substantial tourism companies and hotelj system. The findings in this paper also suggest that while casino complexes that dominate the local economy. a turbulent "job-churn" created a dramatically different Although the tourist economy in Las Vegas generated geography of both employment and air passenger a substantial volume of visitors by air, it failed to spin volume by place, the two variables were closely linked off a significant number of additional employment over time. Air passenger volume behaves much like MAA-4-2012-Report excerpts distributed at the meeting

KG. Debbage, D Delk / Journal oj Air Transport Management 7 (2001) 159-167 167 airline connectIvIty in mimicking the administrative Button, K., Lall, S, Stough, R., Trice, M., 1999 High-technology and auxiliary employment hierarchy of the largest employment and hub airports Journal of Air Transport Manage metropolitan markets of the United States and the ment 5, 53-59 Button, K., Taylor, S, 2000. International air transportation and eco connections between the two variables appear to be nomic development Journal of Air Transport Management 6, remarkably stable over time.. Left unanswered is the 209-222 thorny "chicken or egg" issue - this paper made no Debbage, KG, 1999 Air transportation and urban-economic restruc attempt to unravel the complex casual links that may turing: competitive advantage in the Us. Carolinas. Journal of Air exist between administrative and auxiliary employment Transport Management 5, 211-221 Federal Aviation Administration, 1973 Airport Activity Statistics and air passenger volume, although considerable evid Government Publishing Office, Washington, DC ence exists to suggest that air transportation services can Federal Aviation Administration, 1983 Airport Activity Statistics directly influence employment levels in this sector of the Government Publishing Office, Washington, DC economy. Federal Aviation Administration, 1996. Airport Activity Statistics Government Publishing Office, Washington, DC Goetz, AR, 1992 Air passenger transportation and growth in the US urban system, 1950-1987 Growth and Change 23, 218-242 Acknowledgements Irwin, M n, Kasarda, In., 1991 Air passenger linkages and employ ment growth in U.S metropolitan areas. American Sociological Review 56, 524-537 Professor Debbage acknowledges the generous Ivy, R L., Fik, TJ, Malecki, E 1.,1995. Changes in Air Service Connect support of a University of North Carolina at Greens ivity and Employment Environment and Planning A 27,165-179 boro Faculty Research Assignment which granted Nooteboom, B, 1999. Innovation, learning, and industrial organiza the author research leave at the University of North tion. Cambridge Journal of Economics 23,127-150 Porter, M.E , 1998 Location, clusters, and the 'new' microeconomics of umbria at Newcastle and the University of Surrey competition Business Economics 33, 77-90 in Fall 2000. This publication would not have Smith, D, Timberlake, M., 1998 Cities and the spatial articulation of been possible without the full cooperation of all three the world economy through air travel In: Ciccantell, PS, Bunker, institutions. SO (Eds), Space and Transport in the World-System. Greenwood Press, Westport, CT, pp. 213-240 US Bureau of the Census, 1973 County Business Patterns Government Publishing Office, Washington, DC References US Bureau of the Census, 1983. County Business Patterns Govern ment Publishing Office, Washington, DC Air Transport Association, 1999 Traffic Summary 1960-1998: U.S US Bureau of the Census, 1996 County Business Patterns Scheduled Airlines Available online at http://www.aiItransportmg/ Government Publishing Office, Washington, DC public/industIy/24.. asp US Department of Transportation, 1996 National Transportation Bell, ME, Feitelson, E, 1991 US economic restructuring and demand Statistics Available online at http://wwwbtsgov/btsprod/nts/chpl/ for tIansportation services. Transportation Quarterly 45, 517-538 tbllx13.html Button, K, Lall, S, 1999 The economics of being an airline hub city Van den Berg, L, Van Klink, H A., Pol, PMJ, 1996 Airports as Research in Transport Economics 5, 75-106 centres ofeconomic growth Transport Reviews 16,55-65 MAA-4-2012-Report excerpts distributed at the meeting

The Regional Economic Effects of Airport Infrastructure and Commercial Air Service:

Quasi-Experimental Evaluation of the Economic Effects of Commercial Air Service Near Smaller Airports

Drake E. Warren

Department of Agricultural and Consumer Economics Regional Economics and Public Policy Group University of Illinois at Urbana-Champaign

Abstract

Policymakers, business leaders, media, and the public commonly believe that air ports are the infrastructure critical to connecting a region to the national and inter national economy, especially in knowledge based industries .. They believe that eco nomic development and commercial air service go hand in hand. The perceived eco nomic consequences of poor of air service have led governments to heavily subsidize airports, particularly those small, rural airports that are unlikely to be profitable without government intervention. Billions of dollars are at stake .. This paper builds on previous spatial data analysis to construct and estimate a spatial difference-in differences model of the economic effects of commercial air service near smaller, less populated areas in the post-deregulation era. This model utilizes recent advances in quasi-experimental control group methods to create two control groups to produce more rigorous estimations .. Inference of the average economic effects of commercial air service is conducted using residual block bootstrap and spatial window block bootstrap methods.

There is a general agreement that large commercial airports exert a large economic impact on the surrounding regions due to the connections they fa cilitate with the national and internation economy. Whether this effect exists for smaller airports - those that may have only a few flights each day - is more doubtfuL This paper examines the economic impacts of small airports by con structing quasi-experimental control groups in order to estimate a model of economic impacts .. In order to make this estimation of economic impacts as rigorous as possible, the model utilizes a minimum number of assumptions

* Originally presented at the 54th Annual North American Meetings of the Regional Science Council, Savannah, Georgia, November 8, 2007 Email address.: dewarren©uiuc. edu (Drake K Warren).

26 November 2007 MAA-4-2012-Report excerpts distributed at the meeting

and is estimated with bootstrap methods that can account for the uncertainty caused by violations of traditional assumptions in the error terms. This paper is a condensed version of some of the research in Warren (2008)

The organization of this paper is as follows: Section 1 introduces the research question and reviews some of the existing literature .. Section 2 discusses inno vations in spatial bootstrap methods .. Section 3 outlines the procedures used to create control groups and the results. Section 4 develops an econometric modeL Section 5 presents the results of the estimated modeL Section 6 con cludes

1 Introduction

Policymakers often believe that airline service contributes to the economic health of a region .. This situation is most likely in a large, urban area, especially if that region's airport is an airline hub.. It is least likely in smaller, rural regions. Businesses depend on convenient airline service in order to facilitate face-to-face contact around the world .. This is becoming increasingly important as the information economy becomes more dominate, and therefore airports are increasingly important to a region's economy.. A region without adequate air service cannot sustain modern industries ..

These beliefs appear to be widely held in the political and policy-making world, the media, and with businesses that utilize airports .. Although there is a large quantity of empirical literature about airports and airlines, very little of it tries to assess the economic importance of airports .. The most common type of economic study is an economic impact study, often conducted by a consultant hired by an airport that is considering expansion .. These studies, unfortunately, are incomplete and are not comprehensive .. There are a handful of studies that examine a cross section of metropolitan areas to examine the economic benefits of airports .. Most of these studies ignore smaller airports, which may be the most interesting and enlightening cases due to the substantial variation in service.

The most common form of economic evaluation of airports is economic impact analyses usually prepared for a single airport and a single region that utilize multipliers derived from input-output (10) models .. These analyses often lack objectivity because "the purpose of the study is usually to gain public under standing and support" (Butler and Kiernan, 1992) for airport infrastructure improvements and expansions ..

The FAA has released guidelines on how to perform economic impact analyses using multipliers Regional Input-Output Modeling System (RIMS) II model

2 MAA-4-2012-Report excerpts distributed at the meeting

from the Bureau of Economic Analysis (Butler and Kiernan, 1987, 1992). As mentioned earlier, the FAA does not formally consider macroeconomic bene fits when awarding grants .. The 1992 report explains that these guidelines were "prepared in response to requests from the airport community for FAA guide lines for estimating measures of the importance of individual airports to their surrounding communities," especially for smaller airports that cannot afford a consultant to perform the impact analyses. The real purpose of the guidelines appears to be politicaL The introduction to the 1992 report discusses the im portance of the local political process to airport planning.. Economic impact studies help "the public and their representatives appreciate the economic significance of airports [so they] continue to support them."

Models that predict the economic impact of airports may be useful for predict ing the effects of airport investments .. However, it seems even more important to develop techniques to evaluate the actual, observed economic impacts of air ports rather than making simple predictions .. As the previous section suggests, tenuous theory, researchers' assumptions, and possibly political motivations drives the conclusions of the predicative models .. The validity of these models can be confirmed or rejected by studying regions with airports after airport infrastructure has been constructed and its economic effects felt throughout the regional economy.. Few studies do this

The following papers perform ex post analyses of the economic effects of air ports using cross-sectional econometric estimation. This dissertation will ex tend this line of literature. Green (2002) found 42 economics papers that dealt with airport congestion financing and a few dozen on airport financing, but only one that examined the impact of airports on economic development (Benell and Prentice, 1993) I have also found some more recent papers and papers published outside of economics that are relevant to this research Never theless, there is an obvious dearth of studies about the economic consequences of airports. The studies I found usually fail to examine less populated regions.

A quasi-experimental study of airport investments found that competitor re gions (cities near Indianapolis that have similar characteristics) have higher investments but worse economic outcomes than a control group of exemplar regions (cities further away from Indianapolis that resemble Indianapolis) (Nunn, 2005) Difference in means tests between the two groups found non significant results, but this is likely because he used only nine cities in the study.. Since the control groups were chosen arbitrarily and are small, the pa per is a case study rather than a rigorous econometric evaluation and not useful for a comprehensive evaluation of the economic effects of airports.

Green (2002) develops a simultaneous equation modeL Data comes from the 100 largest airports (by passenger boardings) and the 83 metro areas where those airports are located .. Population growth between 1990 and 2000 and

3 MAA-4-2012-Report excerpts distributed at the meeting

employment growth between 1990 and 1996 are used as economic development dependent variables, and boardings per capita and cargo activity are used as airport activity dependent variables .. Green offers a more careful explanation for his instruments (variables that are on the right hand side of only one equation in the system) and chooses distance to Kansas City and distance to the nearest coastline. Green uses 15 right-hand-side variables .. The relatively small degrees of freedom and the estimation of the simultaneous equation likely lead to noisier estimates .. Nevertheless, he finds strong evidence that enplanements per capita increases population growth and some evidence that it increases employment growth .. Cargo growth, on the other hand, does a poor job at explaining either.. Green justifies truncating his selection of airports to the 100 busiest since small airports "likely have little economic impact" .. This claim needs empirical investigation - continued spending on airports and airline service to less populated regions as well as the results of economic impact studies indicate that the popular consensus is that do have an economic impact.. Furthermore, the addition of more data could strengthen the estimates of the modeL

Cohen and Morrison Paul (2003) evaluates the economic effects of airports in a much different way. They estimate a dynamic generalized Leontief variable cost function model to measure the effects of airport investment (which is represented by "state level air transportation capital outlay" from the Census Bureau) on capital and labor costs for the manufacturing industry of each state.. While this is an elegant economic model derived from the theory of the profit-maximizing firm, the model is not flexible enough to incorporate other important measures of airport quality, travel frequencies, or fares as variables in the modeL The only place where airports enter the model is air port investment, which is imperfect and incomplete because it is simply the sum of construction expenditures at all airports within a state rather than expenditures at individual airports or, better yet, the characteristics of the infrastructure at those airports .. This state-level variable is uninformative to policymakers who need a more comprehensive picture of how individual air ports affect regional economies .. The estimated model is generic enough that the airport investment data can be switched with highway investment data to model the effect of highway infrastructure on manufacturing costs (Cohen and Morrison Paul, 2004). Furthermore, the use of states as the unit of analysis is troubling. This level of aggregation not only reduces value of the results to local and regional policy, but it casts doubt on the validity of their modeL

Finally, Brueckner (2003a) builds upon the research of Green The model ex amines 91 large metro areas and is contemporaneous, i .. e .. the explanatory variables are levels for a single year (1996) rather than the change over sev eral years .. The explanatory variables are various measures of employment and airline traffic, which are both endogenous .. The model includes several instru ments: a dummy variable for whether an airport is a hub, the distance to

4 MAA-4-2012-Report excerpts distributed at the meeting

the population center of gravity in Missouri, a dummy variable for whether a small metro area is within 150 miles of a large airport, a dummy variable for whether the airport is a slot-controlled (capacity constrained) airport, and a dummy for leisure areas .. Two-stage least squares estimations of the model reveal strong evidence that air traffic increases employment in service indus tries, especially when the existence of a hub is used as an instrument instead of the distance to the US. population center. Brueckner uses the results of these regressions to estimate the service employment change that would re sult from the expansion of Chicago's O'Hare airport.. A fifty percent increase in passenger traffic would result in a growth of 185,000 jobs. In a separate paper, Brueckner (2003b) uses the results of the regressions to forecast that the loss of American Airlines' hub will reduce St. Louis' service jobs by 50,000 assuming a 44 percent reduction in traffic. Brueckner's research is clearly a step forward in research on the economic effects of airports. Unfortunately, like previous research, smaller regions have been ignored ..

This paper seeks to econometrically estimate the average economic impact of commercial air service in smaller communities that have been ignored by much of the current research .. Furthermore, this paper strengthens estimates through the use of panel data rather than contemporaneous data used by most of the previous research.

2 Spatial Bootstrap Methods

Bootstrapping (for a review of early development see Efron and Gong (1983)) is a procedure that performs inference on pseudo-datasets created by sampling from observed data .. It is essentially a nonparametric Monte Carlo simulation procedure that uses actual data to dispense with assumptions that may be in doubt or avoid difficult computations. For example, a bootstrap of the mean of a sample would repeatedly calculate the mean of the pseudo-samples to produce an estimate of the distribution of the population mean, rather than relying on the normal distribution and a parametric estimate of the variance of the estimate When used with regressions, bootstrapping can be used to find more accurate distributions of standard errors of model coefficients when parametric assumptions are violated (Freedman and Peters, 1984).

Different bootstrapping methods have been developed for different types of regression .. The block bootstrap (Efron and Tibshirani, 1994) with panel data randomly samples observation units, e.g .. counties, but takes the data from all time periods for each of those sampled units., This ensures that the temporal autocorrelation structure of the data remains in place rather than be broken up by a bootstrap procedure that samples each individual county and year observation separately, The moving block bootstrap (Efron and Tibshirani,

5 MAA-4-2012-Report excerpts distributed at the meeting

1994) is a similar method that is used in time series data with large numbers of time periods., It samples some number of consecutive time periods in order to preserve the autocorrelation structure .. It is considered "moving" because the sampled time periods may overlap, A similar method is the spatial block bootstrap (Hall, 1985), which samples observations that are adjacent spatially.. The trouble with both the moving block bootstrap and the spatial block boot strap is that they do not produce consistent estimates, especially when the data has long range dependence (Garcia-80idan and Hall, 1997) This prob lem is a result of edge effects from sampling only a subset of the data, The data is cutoff at the edges, which makes the pseudo-dataset display substan tially different behavior when dependence occurs over large time periods or distances"

A spatial alternative that avoids these edge effects is residual resampling of a spatial lag model (Anselin, 1990) The spatial structure is preserved in this method by resampling residuals and recalculating the dependent variable by multiplying the estimated spatial multiplier by the perturbed independent variables.

The aim of my estimation is to find the most accurate distribution of model coefficients possible so that policymakers feel safe to make conclusions based on the results" In an effort to allow the data to drive the inference as much as possible and avoid heroic assumptions, I use bootstrap methods to esti mate the distributions of parameters in the modeL Because my model uses spatial panel data, rather than cross sectional data, I must innovate these bootstrap methods by combining both a time block bootstrap and a spatial block bootstrap as outlined in the next subsections.,

2,1 Spatial Window Block Bootstrap Method

The spatial window method requires formulas of the model to be estimated and the fitted values from first stage (IV stage) of the 28L8 procedure for a spatial lag modeL All fitted spatial lags are calculated before the procedure is run .. A spatial structure also needs to be supplied in order to indicate which counties are neighbors of each other. This structure is represented in the form of a spatial neighbors list (see Bivand et aL (2007)),

(1) If necessary, estimate the first stage of 28L8 to get an instrument for the spatial lag, (2) Estimate the model using OL8 and the instrumented spatial lag .. Keep track of the results" (3) Begin a loop of simulations (a) Randomly select a county, giving each county an equal probability of

6 MAA-4-2012-Report excerpts distributed at the meeting

being chosen (b) Using the spatial neighbors information, take all data (ie, every time period) for the selected county and all of its neighbors, (c) Repeat until a number of counties is chosen that is equal to the original data, (4) Estimate the model as before, but use the newly generated pseudo-data Keep track of the results, (5) Repeat 3 and 4 until the desired number of simulations is complete

The result of this procedure will be a list of coefficients for the modeL This list can be used to generate confidence intervals for the parameters and to perform hypothesis testing"

2,,2 Spatial Residual Method

The spatial residual method requires the same inputs as the spatial window method, except it requires a spatial weights matrix rather than a neighbors list"

(1) If necessary, estimate the first stage of 2SLS to get an instrument for the spatial lag (2) Estimate the model using OLS and the instrumented spatial lag, Keep track of the results (3) Save the residuals for sampling, 1 (4) Calculate the spatial multiplier (I - pW)-l using the estimated value of p and the given spatial weights matrix W" (5) Begin a loop of simulations" (a) Create a vector of pseudo-residuals" (i) Randomly select a county, giving each county an equal proba bility of being chosen (ii) Copy the county's residuals, for all years, in order to the pseudo residual vector, (iii) Repeat the random draws, with replacement, until the pseudo residual vector is the same length as the data (b) Calculate a new set of y values, Y; = Yt + (I - pW)-l(e~ - et) V t where Y; is the new vector of pseudo-dependent variables, Yt is the original vector of dependent variables, (I - pW)-l is the previously calculated spatial multiplier, e~ is the new vector of pseudo residuals, and et is the original vector of residuals

1 In the 2SLS procedure, the residuals given by the program will not be the true residuals because the instrumented spatial lag rather than the true spatial lag is used in the regression" These reported residuals are adjusted, similarly to how they are adjusted when calculating standard errors in 2SLS

7 MAA-4-2012-Report excerpts distributed at the meeting

Once again the output will be a matrix of estimated coefficients of the model that can be used to construct confidence intervals and perform hypothesis tests.

2.3 Choice of Spatial Bootstrap Procedure

A series of Monte Carlo experiments was conducted to test the Type I error of the estimates from these bootstrap methods, as well as the traditional block bootstrap, OL8, and 28L8 (Warren, 2008) Despite the theoretical limitations of the spatial window block bootstrap, it was found to be the superior estima tion method when model errors are heteroskedastic or exhibit autocorrelation across time or space .. The spatial residual block bootstrap had similarly strong performance except in the prescence of heteroskedasticity.

When used with control groups, the previous procedures are modified to sample treatment counties and automatically include the matched control county/counties (Warren, 2008).

3 Creation of Control Groups

Quasi-experimental policy evaluation methods seek to find a control group that corresponds to the policy treatment group in order to make inferences about the counterfactual "What would have happened if not for the policy?" There are several methods that attempt to "to make non-random comparison groups truly comparable" despite the inability to perform true experiments (Bartik, 2002) I utilize and extend some of these methods to discover whether the regional airports have had a positive economic impact on the regions where it guarantees air service ..

One of the most basic quasi-experimental methods is constructing a control group and using a difference of means test to test whether the treatment group exhibits a statistically significant difference in some outcome .. Rubin pioneered the creation of control groups (Rubin, 1973). Card and Krueger's pioneering research about the effects of a minimum wage increase used a crude control group by comparing New Jersey fast food restaurants, which experi enced the increase in minimum wage, with neighboring restaurants in Penn-

2 The first stage of 2SLS does not need to be repeated again because the instru mented lag remains the same regardless of the ordering of the residuals ..

8 MAA-4-2012-Report excerpts distributed at the meeting

sylvania, which did not have an increase (Card and Krueger, 1994). Rephann and Isserman (1994) used a more formal control group selection procedure in an infrastructure context with the construction of highways .. Isserman and Rephann (1995) followed this for evaluating the effects of the Appalachian Regional Commission

Another method to make a quasi-experimental control group study more pow erful is to look at multiple control groups for comparison with the treatment group .. Isserman begins on this path by examining the counties in the Lower Mississippi Delta Development Commission (LMDDC) (Isserman, 1997) Mul tiple control groups in observational studies are desirable because they may be used to "reduce the importance of biases or random variation in a single comparison group" (Meyer, 1995). Rosenbaum (1987) gives a theoretical jus tification of why a researcher would find multiple control groups desirable Rosenbaum (1987) shows that the actual treatment effect will fall within a range defined by the two treatment estimates obtained using the two control groups.

Commercial air service at regional airports provides a situation where mul tiple control groups are apparent .. The treatment group are the counties in regions where commercial air service has varied .. These counties have usually lost air service at some point during following deregulation, although other patterns of changing service exist and will be integrated into the final modeL The control groups are those counties that are in regions where commercial air service has always existed since deregulation and counties in regions where commercial air service has never existed since deregulation .. Ideally, procedures to choose control counties would produce control groups with identical distri butions of variables as the treatment counties thereby making the two control groups equivalent.. Unfortunately, reality is not ideal; there are infinite num bers of continuous variables that can be used to construct control groups from a relatively small number of counties .. Therefore, there are likely to be some unknown variables, U, that are not controlled .. The control groups of counties that have always had service and those that have never had service seem to be good candidates for control groups that bracket these unknown variables ..

The control groups used in this paper were developed using a genetic matching algorithm developed by Sekhon (2007) and Diamond and Sekhon (2005) and modified by Warren (2008) to be computationally feasible for large numbers of matching variables. Counties were determined to have air service in a particu lar year if they had at least 500 enplanements. Treatment counties had varied service, while control counties always had service or never had service .. in or der to separate effects from large airports, no county whose centroid is within seventy miles or a large or medium hub in 1979 were included .. Variables used in Isserman and Rephann (1995), for 1979 and 1980, were balanced by the matching procedures .. A second set of control groups also balanced two-digit

9 MAA-4-2012-Report excerpts distributed at the meeting

Table 1 Numbers of counties and numbers of successfully matched between treatment (Tr) and control (Ctr) groups. ------Unmatched Matched No CBP With CBP TT CtT TT CtT TT CtT In County Always 130 256 113 56 79 36 Never 130 1,677 105 83 98 74 40 Miles Always 293 1,171 286 210 275 203 Never 293 599 273 182 274 180 70 Miles Always 141 1,846 117 100 109 92 Never 141 76 121 34 141 45

SIC employment shares from County Business Patterns (CBP) (US. Depart ment of Commerce: Bureau of the Census, 1982). Table 1 shows the number of counties matched when using three distance criteria: airports within a county's borders ("In County"), airports within forty miles of a county's population centroid ("40 Miles"), and airports within seventy miles of a county's popula tion centroid ("70 Miles") Greater details about the matching procedure and the resulting balance between treatment and control counties are in Warren (2008)

Figure 1 shows histograms of all county-years for the combined control and treatment groups for the "in county" case .. Since the "never" control group never has any enplanements, the bottom histogram shows only treated coun ties .. It is apparent that the treated counties have fewer enplanements than the "always" control group because nearly all county-years in the bottom his togram have enplanements below 10,000, while the middle histogram, which includes the control counties, has a long taiL The unbalanced enplanements ex ist because the matching procedure does not match enplanements since it is an interaction with the treatment variable .. Ideally, balancing the other variables would cause enplanements to be balanced between the control and treatment groups. Unfortunately, this is not the case for two reasons. First, different treatments at airports in the treatment and control counties likely leads to different enplanements .. For example, an airport that loses commercial service likely has insufficient subsidies in the preceding years, which may lead to fewer enplanements .. Furthermore, an airport that loses or gains commercial air ser vice in the middle of the year will have a small number of enplanements for that year due to missing service for a number of months .. Second, as mentioned before, there are likely unbalanced variables that factor into the treatment. The possible existence of these variables means that estimations using only the "always" control group should not be trusted alone - estimations using the "never" control group will help bracket the estimates

10 MAA-4-2012-Report excerpts distributed at the meeting

>< [J) co "0..... ;..0 ~ 25'" 0 0 "d G) ~ '"' E S '" 0 0 c

Ok 10k 20k 30k 40k 50k

~ "0..... ;..0 ~ 0 0 ?i c [J) ~ ~ ~ § :;] c

Ok 10k 20k 30k 40k 50k c .~ '" "0..... ;..0 ~ ~ 0 '" 0 c ~..... G) :s

G)> c ~ .~ c

Ok 10k 20k 30k 40k 50k

Fig .. L Frequencies of the projected enplanements for all county-years with commercial air service in treatment and control groups when the "in county" distance criterion is used .. Enplanements over 50,000 are included in last bin ..

4 Model

The development of a model to assess the regional economic effects of airports must start with a choice of a treatment, At first glance, the most obvious choice for treatment is the subsidization of an airport's commercial service through the Essential Air Service (EAS) program, This subsidy is the result of a government policy, and its application to airports is based upon service patterns in the late 1970s right before the transition to deregulation" Unfor tunately, this variable has problems" First, the subsidy is highly endogenous - airports that do well (ie large number of enplanements perhaps caused by a growing economy) will no longer require subsidization to maintain service Second, as outlined previously, the EAS subsidy is just a minor subsidy com pared to those given though the FAA's Airport Improvement Program (AlP) and through the Air Traffic Control (ATC) system. Most small, regional air ports rely on these subsidies in order to maintain commercial air service, even if they do not participate in the EAS program, Third, the quality of data

11 MAA-4-2012-Report excerpts distributed at the meeting

on EAS subsidies is suspect due to a lack of a central database (recall that subsidy definitions were found through tedious research of Congressional Ap propriations reports that were incomplete) Several airports that were part of the EAS program who lost service from an airline (a common occurrence at most EAS airports, especially during times of sharp increases in fuel prices) did not make an effort to renegotiate with the departing airline or solicit new proposals from any other airlines .. Those airports continued to appear on the roll of subsidized airports for several years, even though they did not receive a subsidy.

Rather than subsidization, I consider the presence of scheduled commercial air service to be the indicator of the treatment Most commercial airports operating today were operating before deregulation .. Most small airports have continued to rely upon government subsidies to maintain commercial service. Those airports that have disappeared from the rolls are generally those that decided to close (a decision made by local authorities) or those whose EAS subsidies exceeded the maximum threshold during years when the Department of Transportation decided to reduce the number of subsidized airports .. 3

Data about annual commercial air operations are readily available from the Department of Transportation for the past thirty years .. An airport is consid ered to be treated if it had a minimum of 500 scheduled enplanements in a year. 4

4.. 1 Model Development

My model looks at levels of population, employment, and income just as previ ous regional science models (e .. g. Duffy-Deno and Eberts (1991), Holtz-Eakin and Schwartz (1995), and Boarnet (1998)), but adds a variable for dividends, interest, and rent since it is hypothesized that airports might have a stronger influence on the locations of businesses and people who would receive this income. These variable choices are really just for the sake of convenience (the data are available) and interest. There is a potentially infinite set of variables

3 Losses that occurred as a result of exceeding the maximum allowable subsidy are somewhat worrying because of endogeneity problems, i.e. these high subsidy levels were a result of low enplanement levels that could be caused by or deter mined alongside the regional economy. However, the quasi-experimental matching procedures eliminate most of this worry. 4 A threshold is necessary since many airports have a small number of charter flights each year, but clearly do not have regular, scheduled commercial air service. This threshold was chosen after examining enplanement levels at airports in the EAS program.. Almost all EAS airports exceed this threshold even during years with service operations.

12 MAA-4-2012-Report excerpts distributed at the meeting

that could be used in a simultaneous equations modeL This is demonstrated in the following general equations for the model

F P*i,t - fF(E*i,l ~,t, 1*i,t' D*i,l' \[1*i,t I n i,l ) = I E*t1" fE'L,t (P*t1,,' 1*t,t, D*t'/,,' \[1*2,t nEt)'/"

Itt, = !/t, (P/, t, Ei*t,, D7t,1 \[I7t, I nft)1 D7,t = fi~(Ptt' E~to Ii~t, \[I:,t I nft) where Ptt, Eit, lit, and Dit are equilibrium levels of county population, em ployment, inc~me: and dividends/interest/rent for county i at time t, \[17 tare equilibrium levels of other endogenous variables, and nft, nft, nft, and nPt are county-specific exogenous variables. Any spatial lag~ (or ~ven'time lag~) will be included in the \[I term.

Most regional science models similar to the model above use an adjustment process to account for deviations from the (constant) equilibrium. For ex ample, for population Pi,t = ~,t-l + Ap(Pi - Pi,t-l) when there is a single, time-independent equilibrium. This produces a model with either the change in the endogenous variable on the LHS or a time lag of the endogenous vari able on the RHS. Since these models usually only estimate two periods, models with the differenced variable on the LHS should not suffer from endogeneity problems that arise from the RHS variables being correlated with the error term. However, in a 25-period model like mine, this would be a very likely possibility, which would complicate the modeL Furthermore, the differencing procedure reduces the size of the data making estimates more inefficient.. In addition, differences in differences models (as described in the next section) do not include have differences on the LHS or time lags on the RHS, while spatial methods have been developed most thoroughly for non-dynamic mod els. Therefore, I take a slightly difference approach to developing the model through the use of errors, or deviations from equilibrium

P*t=P'/" it' + EFt1" E~t = Ei,t + Ef,t l 1*t'/" = Ii ' t + E1" t

D*t'/" = Di't + EDtt, making the new structural form of the model

13 MAA-4-2012-Report excerpts distributed at the meeting

~,t = gft(Ei,t, ht, Di,t, Wi,t I Df,t) Ei,t = gft(Pi,t, ht, Di,t, Wi,t I Dtt) Ii,t = gft(pi,t, Ei,t, Di,t, Wi,t I D{,t) Di,t = ge(~,t, Ei,t, Ii,t, Wi,t I Dft) with the observed, non-equilibrium realizations of the endogenous variables. This model does not rely upon an assumed adjustment process or even the existence of equilibrium.

In order to develop a simultaneous model that is estimable and not subject to the difficulties of structural equations models, I solve the above model for the endogenous variables to produce a reduced form model

~,t = hf,t(W~,t, Di,t I Di) (la) Ei,t = htt(W~,t, Di,t I Di) (lb) ht = hft(W~,t, Di,t I Di) (lc) Di,t = hft(W~,t, Di,t I Di) (ld) where W~,t C Wi,t of endogenous variables that remain included in the model (e.g .. if we were interested in the effect of an endogenous variable on the LHS variables), Dit = u(DfnDft,Dft,Dft) are time-varying, county-specific en dogenous vari'ables, and Di '= U(Dr,'Df, D[, Df) where Dfd = Dtj V t, M E {P, E, I, D} are all time-invariant, county-specific exogenous variables .. Since there are few time-varying variables that can realistically be considered ex ogenous, Di,t will be small (eg., the policy variable). 5

4.. 2 Difference-in-Differences Models

Difference in differences models are regression models used to discover the effect of a policy change when there are data for a treatment and control group before and after the treatment is given (Meyer, 1995) A general form for the regression equation estimated in the model is

(2) where dt is a dummy variable that equals one if the observation is post treatment, dj is a dummy variable that equals one if the county is in the treatment group (either before or after treatment), zl,t is a vector of (exoge nous) characteristics of the county that may vary across time periods, and cl,t

5 The assumption of exogeneity of these variables can be relaxed when well constructed control groups are formed (Warren, 2008)

14 MAA-4-2012-Report excerpts distributed at the meeting

is the error term county and time specific error term, The treatment effect is estimated through (3, since it is the coefficient of the interaction between counties that have been treated and in a time that is post-treatment, i,e, dtd j equals one only for counties that have received the treatment,

Meyer (1995) shows that the key identifying assumption for determining an unbiased estimate of (3 is that the error is independent of whether or not the county received the treatment, i,e" E[cl,tldtdj] = 0" This means that the treatment effect is the same regardless of whether a county was treated or not. As I discussed in the previous section, the use of control groups tries to eliminate this endogeneity problem, The additional covariates included in the difference-in-differences model serves to further ensure that the independence of the error term holds, Satisfying this requirement does not ensure trouble free estimation of the treatment effect. The covariates in zl,t6 might be jointly determined with the dependent variable, in which case these endogenous vari ables must be controlled through instrumentation For a thorough discussion of the problems endogeneity plays in estimating treatment effects, see (Besley and Case, 1994),

One criticism leveled against most difference-in-differences research that uses many time periods is that the possibility of serial correlation in the error terms is not acknowledged thus leading to understated standard errors that cannot be trusted (Bertrand et aL, 2004)" Furthermore, in a two-period (be fore and after) difference-in-differences model many researchers use different after dates. These estimates may be similar due to the serial correlation of the explanatory variables and the error term; a string of consecutive years show ing positive effects is not necessarily more convincing than a single year since those estimations are not independent of one another" One of the suggestions of Bertrand et al. (2004) is to use block bootstrapping, which is the method I use to estimate my modeL By keeping all years together for the same observa tion, simulations replicate the serial nature of the data and therefore produce accurate estimations of standard errors,

4 3 Fixed and Random Effects Models

In panel data models with a few time periods and many observations, it is often hypothesized that there is a unique, but constant effect for each observation over all time periods" Furthermore, each time period might have a similar unique effect over all observations" This can be modeled simply as

where Xi,t is a vector of covariates, ai is the unique effect for each observation, at is the unique effect for each time period, and 'f7i,t is the error term, This

15 MAA-4-2012-Report excerpts distributed at the meeting

equation is in fact very similar to equation 2, which suggests that this model may be applicable to difference-in-differences models.

An alternative to the fixed effects model is the random effects modeL In this model, the unique effects are modeled as if they were in the error term For example,

Yi,,, t = Xi t(3 + ti t t ti,t = lYi + IY + TJi,t where those effects and the previous error term are included in a new error term. 8ince the effects are now being modeled in the error term, they could create a situation where the error term is correlated with the explanatory variables Xi,t even if E[TJi,tIXi,t] = o. Therefore, in order to use a random effects model the effects must be uncorrelated with the covariates .. In preparation for the development of my model, I modify this model as

i t IYt ti (3) Y " = Xi t(3 + + ,t ti,t = lYi + TJi,t (4) which is a mixed effects model that reflects a time fixed effect but with a county random effect.. For reasons outlined in Warren (2008), I believe that the county random effect is uncorrelated with the covariates used in the model, but the time fixed effects are correlated .. This is also convenient since my model has 25 time periods, but can have over 500 counties .. To estimate these mixed effects models I integrate the generalized least squares (GL8) estimation proce dure outlined in (Johnston and DiNardo, 1997) with the previously mentioned bootstrap methods.

44 Final Commercial Air Service Models

The final step before estimating the model is transforming the commercial air service model from equations 1a to 1d on page 14 into a mixed effects difference-in-differences modeL The model as it is currently presented could be incredibly complex since it allows for any number of endogenous and exogenous variables in any function form. All of these variables must be included in the proper functional form in order to produce consistent estimates of the modeL Furthermore, all of the endogenous variables need to be instrumented in a procedure like 28L8 in order to produce consistent estimates of the model.

In order to operationalize the model we must ask which variables must be included in the model, which of those variables are endogenous, and what in struments need to be used in a 28L8 modeL If we are honest while answering

16 MAA-4-2012-Report excerpts distributed at the meeting

these questions, estimation of the model might seem impossible .. First, there are many variables that should be included in this modeL There is no way of knowing when we have accounted for all of them. Most importantly, there are many variables that we cannot measure that should likely be included in the model. Luckily, as discussed the previous section, the construction of control groups helps reduce those problems. Second, a good answer to the question of "what is endogenous?" is "what is not endogenous?" An argu ment could be made that just about every variable that changes over time is endogenous.. Instruments need to be found for all of these variables in order to reduce endogeneity bias. As discussed earlier, the construction of control groups mitigates problems caused by the endogeneity of a policy variable .. But it is simply not possible to construct control groups that produce balanced datasets given every possible endogenous variable .. Third, finding sufficient in struments is extremely difficult. Ultimately, we will need to choose variables from the past since those are the only variables that we can safely conclude are exogenous to the modeL Ultimately, with all of these endogenous variables and instruments it becomes much simpler to simply consider a reduced form of this model that includes only a limited number of endogenous variables .. For example, the policy treatment variable can be included since that is the most important variable in this model and its endogeneity has been controlled through construction of control groups ..

The following model modifies equations 1a to 1d by putting the equations into log-linear form, like the difference-in-differences models

where most functions are now time-invariant in order to reduce the dimen sionality of the estimation, \[1~/t are the remaining endogenous covariates in the model, n~,t are the remain'ing time-varying exogenous covariates, n~ are all the instruments included in this reduced form model, and tT,j are the errors. Realizing that there are few exogenous covariates that vary in both time and county, the n~ t can be thought of some economy-wide variables that change over time, but 'are essentially exogenous (especially considering that the coun ties in this model are small in the size of population and economy) .. If we also assume that the function, ~ is the same for all counties and varies only across time, these functions and variables can be represented as a constant effect for each year, i .. e .. at(*) Similarly, the functions of the all of the instruments can be represented as a constant effect for each county, i .. e. a i (*) .. Finally, I leave the policy control as the only remaining endogenous variable and assume that

17 MAA-4-2012-Report excerpts distributed at the meeting

the function is the same over all counties. 6 Making these changes, the model becomes

PiP In [{,t = (3 SER~,t + C

This model closely resembles the difference-in-differences model and the mixed effects model when there are no other covariates .. However, the covariates are not being excluded: they are accounted for the time and the county effects This model becomes less parametric since it does not assume any kind of function form for the functions of these covariates.. Most importantly, any covariate that affects a county's economic outcome can influence the county effects .. Therefore, we do not have to rely on the limited number of instruments that can be observed in order to estimate this modeL This model reduces misspecification errors and should eliminate omitted variables bias ..

I estimate a second set of parametric models that include the spatial lag of the economic variable .. Although the model in equations 5a to 5d should account for the spatial effects via the instruments that are included in the county effect, in many cases it might be beneficial to estimate the effect of space or the effect of some other instrumental variable. This model also includes the covariates that were used in the matching procedure in order to show how these variables can be incorporated into the model and to compare the estimated results. These variables are assumed to having a linear functional form in the following equations:

P P In Pi,t = (3P SER~,t + A In(WPt) + xirS + C

6 An alternative function could be allowed to vary over time since that would remove relatively few degrees of freedom .. However, the estimates of the policy effect could be expected to be much noisier than in the pooled case.

18 MAA-4-2012-Report excerpts distributed at the meeting

enty miles of a given county's population centroid 7 and Xi are all of the variables used to construct control groups (excluding the CBP data due to its large size) plus dummy variables indicating for which control group the county is a member The a's are estimated in the same manner as the previ i ous model, with the exception of a for the residual block bootstrap method. S The spatial lag of the economic variable is estimated with 2SLS using all of the explanatory variables (SERVi,t, Xi, and at but not a i since it is either in the error term or not included in the residual method's estimation) and the natural log of the seventy and 210 mile lags of all four economic variables in 1979 as instruments

5 Results

Both the nonparametric and the parametric models are estimated with the spatial panel bootstrap methods modified to use a mixed effects model with control groups. "OLS" is simply the model with random effects for the counties taking the estimated point estimates and standard errors as true (therefore, it is not true OLS) "2SLS" is the two-stage least squares procedure where the first stage creates instruments for the sum of the economic variables of counties whose population centroids are within seventy miles of the county's population centroid. The second stage is the same as OLS, except using the instrumented spatial lag instead of the true value .. Because 2SLS is used to correct for the endogeneity of the spatial lag it is not used in the nonparametric model. "WinO" is the zero-order spatial window method. It does not draw any neighbors and therefore is equivalent to a nonspatial block bootstrap method .. "Winl" is the first-order spatial window block bootstrap method .. The spatial weights matrix is binary with neighbors that have population centroids within seventy miles of one another Any county in the United States can be drawn as the seed county, but only the counties and the neighbors in the treatment group and its twin(s) will be included in the pseudo-datasets .. "Win22" is the same spatial block bootstrap method except it uses a transposed weights matrix of the 22 nearest neighbors instead of the seventy mile weights matrix, 9 "Res" is the residual block bootstrap .. As mentioned in Warren (2008), it uses

7 Some isolated counties have no counties within seventy miles, In these cases the variable is adjusted so that its natural log is zero in order to avoid an undefined natural log" S Since the residual block bootstrap method can only estimate a i as a fixed effect, a i cannot be included in a model that includes the time-invariant instruments Xi since the combination of the effects and instruments is linearly dependent, 9 See Warren (2008) for an explanation of how the transposed weights matrix is necessary for an unbiased estimate of the average treatment effect when treatment efIects are heterogenous across counties

19 MAA-4-2012-Report excerpts distributed at the meeting

Table 2 Estimates of percentage increase in income due to the presence of commercial air service within the county. Control groups did not use CBP data.. Nonparametric Models Parametric Models Median 95% CI Pseudo Median 95% CI Pseudo Method 50% 2.5% 97.5% p-value 50% 2.5% 97.5% p-value Combined Control Groups FGLS 45% 37% 53% 0000 4.1% 34% 49% 0000 2SLS 44% 3.6% 52% 0.000 WinO 44% 08% 84% 0011 43% 08% 8.1% 0003 Win1 34% 02% 71% 0015 33% 04% 69% 0018 Win22 43% 05% 88% 0.015 43% 0.6% 90% 0.010 Res 36% 0.1% 71% 0018 "Always" Control Group FGLS 5.2% 40% 65% 0000 49% 3.8% 61% 0000 2SLS 52% 40% 65% 0000 WinO 53% 12% 98% 0004 52% 09% 99% 0007 Win1 43% 12% 8.1% 0004 43% 1.0% 81% 0000 Win22 5.1% 1.0% 96% 0006 50% 13% 95% 0005 Res 48% 12% 85% 0002 "Never" Control Group FGLS 36% 25% 47% 0.000 32% 2.3% 4.2% 0.000 2SLS 35% 24% 46% 0000 WinO 36% -03% 74% 0036 32% -0 .. 3% 73% 0037 Win1 25% -07% 66% 0078 23% -11% 64% 0092 Win22 35% -05% 81% 0048 33% -06% 76% 0052 Res 2.3% -1.2% 5.9% 0.103 fixed county effects in the nonparametric model rather than random effects due to the heterogeneity problem of random effects estimation, and it does not use any county effects in the parametric model.

All bootstrap simulations were run 1,000 times for each scenario. Detailed results of the various estimates of the treatment effect from having commer cial air service are included in the upcoming tables .. These include the median estimated value of the treatment effect (which is the point estimate for OL8 and 28L8), the 95 percent confidence interval surrounding the median (de rived from the standard error for OL8 and 28L8), and the pseudo p-value (one-sided), which is the fraction of times the estimated treatment effect was less than zero (derived from the standard error for OL8 and 28L8). In these tables the treatment effect is expressed as a percentage because in this log linear model the coefficient represents the percentage change in the underlying dependent variable due to the presence of commercial air service.

50.1 Results for Commercial Air Service Inside the County

I begin by discussing the results of the model estimation with treatment and control groups that are based on whether a county has commercial air service

20 MAA-4-2012-Report excerpts distributed at the meeting

Table 3 Estimates of percentage increase in income due to the presence of commercial air service within the county. Control groups did use CBP data.. Nonparametric Models Parametric Models Median 95% C.I Pseudo Median 95% CT Pseudo Method 50% 2.5% 97.5% p-value 50% 2.5% 97.5% p-value Combined Control Groups FGLS 49% 41% 5 .. 8% 0.000 47% 39% 56% 0.000 2SLS 49% 41% 58% 0000 WinO 47% 09% 8 .. 9% 0005 47% 1.2% 95% 0003 Win1 45% 14% 87% 0.000 44% 13% 84% 0001 Win22 47% 13% 99% 0006 46% 11% 9.2% 0001 Res 45% 11% 82% 0002 "Always" Control Group FGLS 52% 39% 65% 0000 49% 37% 61% 0.000 2SLS 54% 41% 67% 0.000 WinO 5.2% 02% 107% 0021 53% 06% 107% 0015 Win1 59% 20% 10 .. 7% 0.000 59% 1.9% 103% 0002 Win22 53% 07% 105% 0009 52% 07% 10.2% 0.007 Res 71% 29% 117% 0000 "Never" Control Group FGLS 48% 36% 60% 0.000 43% 33% 54% 0000 2SLS 47% 3.6% 5.9% 0.000 WinO 4.6% 09% 93% 0001 46% 1.0% 93% 0.004 Win1 36% 02% 7.8% 0021 34% 04% 74% 0 .. 012 Win22 48% 09% 94% 0007 45% 06% 91% 0.007 Res 3.4% -0.1% 6.9% 0.028 that is actually within the county boundaries.. If commercial air service does have any sort of effect on its regional economy, we would expect the effect to be largest in the surrounding county since the distance is smallest. This discussion focuses on just the estimates of the treatment effect; estimates of entire regressions and discussion of these results are provided in Warren (2008)

Table 2 begins by showing the results of the model when the natural log of total income is used as the dependent variable and CBP data was not used to construct control groups. The first thing we notice is that the "Always" control group always has higher estimated treatment effects than the "Never" control group .. Regressions on the combined control groups are in between, as we might expect .. This pattern of behavior exists for almost all of the results for all scenarios even though we would expect that the results of any control group to be the same if matching was perfect. Alas, it is impossible for match ing to be perfect .. As mentioned in the control group section, some treatment counties have been thrown out when they are matched poorly. Different coun ties are thrown out when using different control groups .. This means that the composition of the treatment groups is different Since we are estimating the estimated treatment effect on the treated counties we would expect different results due to the difference in composition of the treatment group as long as the distribution of treatment effects is different across different types of coun-

21 MAA-4-2012-Report excerpts distributed at the meeting

Table 4 Estimates of percentage increase in employment due to the presence of commercial air service within the county, Control groups did not use CBP data Nonparametric Models Parametric Models Median 95% CT Pseudo Median 95% CT Pseudo Method 50% 2.5% 97.5% p-value 50% 2.5% 97.5% p-value Combined Control Groups FGL8 57% 49% 64% 0000 5,.2% 45% 5,9% 0,000 28L8 56% 49% 64% 0000 WinO 54% 24% 96% 0000 55% 24% 93% 0,000 Winl 44% L5% 75% 0000 42% L7% 76% 0000 Win22 55% 24% 95% 0000 55% 24% 93% 0000 Res 48% 13% 81% 0004 "Always" Control Group FGL8 63% 52% 74% 0,000 5,9% 48% 69% 0,,000 28L8 6.2% 52% 73% 0,000 WinO 62% 28% 98% 0000 62% 27% 102% 0000 Winl 51% 21% 86% 0001 51% 2,1% 83% 0000 Win22 60% 26% 10.2% 0000 61% 23% 103% 0000 Res 5.2% 20% 88% 0000 "Never" Control Group FGL8 49% 3,9% 59% 0000 44% 34% 53% 0,000 28L8 48% 38% 59% 0,000 WinO 48% L6% 86% 0002 47% 14% 90% 0002 Winl 35% 04% 71% 0015 36% 04% 75% 0015 Win22 46% 16% 90% 0002 47% 14% 85% 0,000 Res 3.9% 0.5% 7.3% 0.013

ties (see Warren (2008)), Another reason for this discrepancy is bracketing, which was discussed in the control group section" Two control groups are often useful if there may be some unobserved set of variables that affects treatment, but cannot be balanced well. It seems likely that in the case of commercial airport service, any unobserved variables are likely to tend in the direction of having service for the "always" control group, while they are likely to be associated with less service in the "never" control group, Therefore, using two control groups that bracket these unobserved variables should help produce a range of possible average values"

Another common behavior in the results is that FGL8 and 28L8 have the smallest confidence intervals (and therefore the lowest p-values) while the first order spatial window block bootstrap method (Win1) and the transposed 22 nearest neighbors (Win22), which tend to have increasingly wider confidence intervals" Confidence intervals for WinO and Win22 are similar for the "in county" case, but in the forty and seventy mile cases the Win22 confidence intervals become much bigger than those of WinO, 10

10 This is because the treatment counties are spread out in the "in county" case, which means that the spatial window block bootstrap usually samples one county at a time, just like ordinary block bootstrap, since treated counties tend not to

22 MAA-4-2012-Report excerpts distributed at the meeting

Table 5 Estimates of percentage increase in employment due to the presence of commercial air service within the county. Control groups did use CBP data. Nonparametric Models Parametric Models Median 95% CI Pseudo Median 95% CI Pseudo Method 50% 2.5% 97.5% p-value 50% 2.5% 97.5% p-value Combined Control Groups FGLS 64% 56% 7.3% 0000 61% 54% 69% 0.000 2SLS 64% 5.6% 7.2% 0.000 WinO 62% 29% 106% 0.000 62% 31% 106% 0000 Winl 54% 27% 9.2% 0000 54% 26% 90% 0.000 Win22 65% 30% 104% 0000 64% 30% 106% 0.000 Res 58% 2.. 3% 92% 0.000 "Always" Control Group FGLS 69% 57% 80% 0000 65% 54% 7.5% 0000 2SLS 7.0% 58% 81% 0000 WinO 68% 30% 112% 0000 69% 31% 112% 0000 Winl 69% 38% 111% 0.000 71% 38% 109% 0000 Win22 66% 27% 113% 0.000 69% 30% 114% 0000 Res 7.6% 35% 122% 0000 "Never" Control Group FGLS 62% 51% 74% 0.000 57% 47% 68% 0000 2SLS 62% 51% 73% 0000 WinO 61% 25% 104% 0.000 59% 25% 104% 0000 Winl 46% 12% 88% 0002 46% 14% 86% 0.000 Win22 61% 27% 107% 0.000 59% 26% 105% 0.000 Res 4.9% 1.9% 8.4% 0.000

Another trend that is observed in Table 2 as well as most other scenarios is that the median value of the WinI estimation is substantially lower than the median value of any other method. As discussed previously, the WinI method may be biased when there are regions with different sized counties and there is spatial heterogeneity. The WinI oversamples counties that are neighbors to many other counties.. If the treatment effect varies over space then the estimates from the WinI method will overweight these counties. The lower median values of WinI suggests that the more densely packed counties that are generally east of the Rockies have a lower average treatment effect than those that are to the west.. Since every county is a neighbor to 22 counties in the Win22 method, this problem of oversampling does not exist and the average treatment effect is unbiased .. Because of this bias, results from the WinI will tend to be ignored in this discussion ..

Another exception to behavior observed in the Monte Carlo experiments in Warren (2008) is the spatial residual block bootstrap method for the para- have treated neighbors .. In the forty and seventy mile distance cases the treated counties are clustered close together .. In these cases Win22 samples many neighbors together and accounts for the uncertainty of the estimates caused by the spatial autocorrelation of the residuals of those neighbors ..

23 MAA-4-2012-Report excerpts distributed at the meeting

Table 6 Estimates of percentage increase in population due to the presence of commercial air service within the county.. Control groups did not use CBP data .. Nonparametric Models Parametric Models Median 95% CJ Pseudo Median 95% CI Pseudo Method 50% 2.5% 97.5% p-value 50% 2.5% 97.5% p-value Combined Control Groups FGL8 38% 31% 4.4% 0.000 33% 27% 39% 0000 28L8 37% 30% 43% 0000 WinO 37% LO% 66% 0001 36% 09% 66% 0004 Winl 31% 09% 59% 0002 30% 09% 53% 0004 Win22 36% 09% 65% 0005 36% 06% 66% 0 .. 011 Res 23% -06% 5.4% 0.062 "Always" Control Group FGL8 4.4% 3.4% 53% 0000 40% 31% 49% 0000 28L8 43% 3.4% 53% 0000 WinO 4.4% 09% 7.7% 0004 42% 07% 79% 0008 Winl 40% 1.4% 68% 0001 40% L3% 69% 0000 Win22 42% 11% 75% 0005 42% 11% 7.8% 0004 Res 35% 03% 6.5% 0018 "Never" Control Group FGL8 30% 22% 39% 0000 2.4% L7% 32% 0.000 28L8 29% 20% 37% 0000 WinO 29% 03% 59% 0.011 28% 02% 59% 0012 Winl 23% -03% 51% 0041 22% -0.4% 49% 0048 Win22 29% 01% 57% 0019 27% 01% 58% 0024 Res 0.2% -2.7% 3.2% 0.452

metric models .. It often has a larger confidence interval and a larger p-value than the other methods .. This behavior seems likely due to two facts .. Most importantly, the estimates of the spatial lag coefficient are nearly zero .. As we saw in the Monte Carlo simulations, when A = 0 and levels (not averages) are used as the dependent variable, the spatial residual block bootstrap does a very poor job of estimating (3, especially with small samples .. 11 Second, "Res" uses parametric regressions that do not have a fixed or random effect, unlike all other methods .. This means that the estimator will tend to be less efficient .. Furthermore, this could create heterogeneity in the residuals, which also causes problems for Res In theory, the spatial residual block bootstrap method will work well in the absence of heteroskedastic errors. In practice (in this application, at least), the residual method is difficult to use .. For these reasons, results for the "Res" method will usually be ignored in the following discussion..

We see strong evidence from Table 2 that, on average, commercial air service

11 Additional trials for some of the scenarios revealed that bootstrap estimates of the residual method produced substantially different results, whereas the estimates from the other methods were stable. This lack of stability echoes the results of the Monte Carlo experiments.

24 MAA-4-2012-Report excerpts distributed at the meeting

Table 7 Estimates of percentage increase in population due to the presence of commercial air service within the county. Control groups did use CBP data Nonparametric Models Parametric Models Median 95% CT Pseudo Median 95% CT Pseudo Method 50% 2.5% 97.5% p-value 50% 2.5% 97.5% p-value Combined Control Groups FGL8 38% 31% 45% 0000 35% 28% 41% 0.000 28L8 37% 31% 44% 0000 WinO 38% 11% 65% 0006 36% 06% 65% 0012 Win1 37% L3% 64% 0000 3.5% L2% 59% 0.000 Win22 38% LO% 6 .. 5% 0003 36% 11% 64% 0003 Res 26% -01% 52% 0028 "Always" Control Group FGL8 37% 27% 47% 0.000 32% 23% 41% 0000 28L8 38% 28% 47% 0000 WinO 38% 00% 71% 0023 36% 00% 70% 0026 Win1 44% L5% 73% 0002 44% L2% 71% 0002 Win22 37% 04% 71% 0012 3.7% 03% 70% 0015 Res 4 .. 1% 06% 78% 0012 "Never" Control Group FGL8 40% 30% 49% 0000 34% 26% 43% 0000 28L8 38% 29% 47% 0000 WinO 38% L2% 72% 0001 37% 11% 67% 0001 Win1 35% 06% 63% 0005 3.2% 06% 59% 0003 Win22 39% L3% 70% 0002 3.7% 11% 65% 0003 Res 2.0% -0.5% 4.8% 0.060 has a positive effect on a county's total income .. In the combined control group regression without CBP data the 95 percent confidence interval ranges from a low of 06 percent for the Win22 method to a high of 90 percent for the Win22 method .. The null hypothesis that the treatment effect is zero can be rejected with strong evidence .. The highest p-value is 0 .. 015 for the Win22 method .. When groups are constructed with CBP data the inference is nearly the same for all estimators.

The least convincing results occur with the control group that never has had commercial air service.. Here the lower end of the Win22 confidence interval dips to -0.6 percent when parametric models used in control group construction and the p-value is 0052, which is still a moderately convincing rejection of the null hypothesis .. The p-values for the nonparametric Win22 method remain below 0.05. When using the control group that always has had service the evidence is overwhelming that commercial air service increases a county's total income. Due to the bracketing of the control groups, it seems a safe bet to make this conclusion Unfortunately, it is difficult to arrive at a sharp estimate of the exact magnitude of this effect, but this problem is likely due in part to the fact that the treatment effect is heterogeneous: different counties are likely to have different treatment effects, so it is difficult to arrive at an average for all treated counties. Nevertheless, it appears that the nonparametric model

25 MAA-4-2012-Report excerpts distributed at the meeting

Table 8 Estimates of percentage increase in dividends, interest, and rent due to the presence of commercial air service within the county. Control groups did not use CBP data. Nonparametric Models Parametric Models Median 95% CI Pseudo Median 95% CI Pseudo Method 50% 2.5% 97.5% p-value 50% 2.5% 97.5% p-value Combined Control Groups FGL8 43% 33% 54% 0.000 3.. 9% 30% 48% 0000 28L8 44% 33% 54% 0.000 WinO 41% -05% 90% 0040 42% -03% 93% 0037 Winl 33% -05% 75% 0046 32% -05% 74% 0046 Win22 43% -08% 96% 0046 43% -02% 94% 0033 Res 51% 11% 93% 0007 "Always" Control Group FGL8 64% 49% 79% 0.000 5.7% 43% 70% 0000 28L8 65% 50% 79% 0000 WinO 63% 11% 115% 0005 64% 13% 121% 0007 Winl 57% L6% 103% 0001 58% 14% 106% 0002 Win22 61% 09% 123% 0012 62% LO% 119% 0007 Res 67% 25% lL2% 0.000 "Never" Control Group FGL8 L8% 04% 33% 0007 23% 11% 34% 0.000 28L8 L9% 05% 33% 0004 WinO L7% -28% 65% 0241 L8% -31% 72% 0220 Winl 07% -32% 54% 0357 09% -35% 59% 0356 Win22 L8% -2.8% 74% 0227 L8% -29% 67% 0230 Res 3.8% -0.2% 7.9% 0.038 arrives at a slightly more compact estimate, likely due to the greater number of degrees of freedom in its estimation .. The next chapter will deal with the heterogeneity issue in depth with an extension of the modeL

Table 3 shows the estimated treatment effect when the CBP data is used. These estimates are generally slightly more dispersed than the estimates with out CBP data.. The estimates are also a little higher.. The highest p-value is now 0021, which occurs with WinO method with the "always" control group. All p-values for Win22 are equal to or less than 0.010 If we believe that the control groups constructed with CBP data have a better balance then these results provide more evidence that commercial air service within a county has a positive effect on total income than the previous set of control groups.

Results in Tables 4 and 5 for the change in employment caused by the pres ence of commercial air service are even more convincing than income.. The null hypothesis of no effect on employment is rejected in the majority of the estimations with a p-value of zero.. The highest p-value (other than the Res and Winl estimates) is 0002 with several of the estimates .. The confidence interval is at is lowest (14 percent) in both the WinO and Win22 estimates with the "never" controls for the groups created with CBP data.. When CBP data was used for matching there was never a simulation where the simulated

26 MAA-4-2012-Report excerpts distributed at the meeting

Table 9 Estimates of percentage increase in dividends, interest, and rent due to the presence of commercial air service within the county. Control groups did use CBP data.. Nonparametric Models Parametric Models Median 95% C.I Pseudo Median 95% CI Pseudo Method 50% 2.5% 91.5% p-value 50% 2.5% 91.5% p-value Combined Control Groups FGLS 53% 41% 65% 0.000 51% 40% 61% 0000 2SLS 54% 42% 6.5% 0.000 WinO 51% 06% 109% 0013 53% 1.3% 101% 0.008 Winl 48% 11% 93% 0005 48% 12% 93% 0006 Win22 53% 1.0% 101% 0009 5.1% 10% 99% 0003 Res 60% 19% 100% 0002 "Always" Control Group FGLS 7..6% 59% 9.2% 0.000 70% 55% 85% 0.000 2SLS 80% 63% 96% 0000 WinO 76% 22% 133% 0003 77% 1.8% 133% 0.003 Winl 87% 43% 138% 0000 90% 46% 140% 0.000 Win22 75% 21% 133% 0001 7.7% 26% 130% 0000 Res 108% 57% 159% 0000 "Never" Control Group FGLS 35% 1.9% 52% 0000 34% 20% 4.8% 0000 2SLS 35% 1.9% 51% 0000 WinO 35% -13% 89% 0.077 35% -11% 87% 0 .. 087 Winl 23% -1.6% 71% 0120 23% -1.7% 68% 0129 Win22 34% -1.2% 9.0% 0.072 32% -11% 85% 0076 Res 3.5% -0.9% 84% 0.051 estimate of the treatment effect was ever less than zero for WinO or Win22. Once again, the CBP control groups tend to produce slightly wider confidence intervals, along with higher estimations of the treatment effect, and the only real difference between the parametric and nonparametric estimates is that the nonparametric estimates tend to have a slightly bigger confidence intervaL

Tables 6 and 7 present the estimates for the treatment effect when county population is the dependent variable .. The results are once again convincing in the rejection of the null hypothesis, although they are not quite as strong as with employment. The confidence intervals for Win22 remain positive. In the combined control group Win22 method with the nonparametric model, the p-value is 0005 when CBP data is not used and 0012 in the same case with CBP data. The size of the confidence intervals once again tends to be smaller without CBP data. The median treatment effect is higher with CBP data with the "never" control and slightly lower with CBP data with the "always" control.

The effect of commercial air service on dividends, interest, and rent may be interesting because it reflects the income of wealthier people who tend to have higher capital derived income. Advocates of airports often claim to be target ing jobs, industries, and corporate headquarters that would attract wealthier

27 MAA-4-2012-Report excerpts distributed at the meeting

people to a region, Dividends, interest, and rent may be a better indication of the spillover effects of commercial air service, Table 8 and 9 show the esti mates of the treatment effect of commercial air service on dividends interest and rent" While the estimates with the "always" control group show that there is a positive treatment effect on dividends, interest, and rent (the highest p value is 0012 for Win22 in the nonparametric regression without CBP data, and the lowest end of the confidence interval is at 1..1 percent for Win22 with out CBP data in the nonparametric regression) the estimates with the "never" control group are inconclusive" The median estimates of the treatment effects are still above zero, The p-values for Win22 with the CBP data still provide mild evidence for rejection of the null hypothesis since they are below 0100 The p-values when CBP data is not used are about three times the size as they are when CBP data is used, so it appears that the choice of covariates is having a large effect on estimation in this case" When the control groups are combined the estimations still tend to reject the null hypothesis, especially when CBP data is used in the matching, The size of the confidence intervals in the cases without CBP data compare less favorably to the cases with the CBP data than they did with previous dependent variables; for example, the size of the Win22 confidence interval remains at 10.2 percent in the "never" nonparametric case with and without CBP data" The overall results, espe cially with the "always" control group, suggest that the effect of commercial air service inside a county is positive on dividends, interest, and rent, This conclusion is somewhat tempered by the "never" results, but still seems to hold"

5,,1 Results for Other Distance Criteria

Results and discussion of model estimates with control groups created using forty and seventy mile distances are in Warren (2008) Results tend to be similar, but smaller, hence less statistically significant., Results for the seventy mile distance are weakest, and in many cases the median treatment effect is negative"

6 Conclusion

This paper documents a methodology for evaluating the economic impacts of commercial air service" Quasi-experimental control groups are created to generate reliable results that do not require treatments to be exogenous, The block bootstrap methods developed here, particularly the spatial window block bootstrap, uses spatial panel data to create estimates of treatment effects, This offers a greater amount of data than the cross-sectional data used in most other

28 MAA-4-2012-Report excerpts distributed at the meeting

research, but also acknowledges the uncertainty caused by heteroskedastic errors, serial autocorrelation in the errors, and spatial error autocorrelation

This paper used these methods to estimate mixed effects models .. One of these models was essentially nonparametric, and therefore required a minimal num ber of assumptions, while the other model incorporated regressors, including spatial lags that were estimated using two-stage least squares .. Results for the two types of models were similar, which leads to the conclusion that the non parametric method is superior because of the reduced number of assumptions .. If the assumption about the endogeneity of possible instruments were relaxed to allow a greater number of instruments (for example, variables flom 1985 were assumed to be exogenous and were used in regressions of 1985) the results of the parametric regressions could differ and be more accurate and precise, provided the endogeneity problem is not too severe ..

The results show that the estimated treatment effects from any type of com mercial air service are almost always positive and statistically significant when the airport is located within a county. The interpretation of this result requires great care to avoid a likely mistaken assumption that the treatment effect is the same in all counties in each year A heterogenous treatment effect could easily exist, as is suggested by the apparent bias of the Winl estimate, which is not robust to heterogeneity. If treatment effects are heterogenous the estimated treatment effect is the average treatment effect over the treated counties .. If this is the case we may conclude that the mere presence of commercial air service has a positive treatment effect on the surrounding county on average, but we cannot say much, if anything, about the effect in individual counties Later research in Warren (2008) modifies the models presented in this paper to allow the treatment effect to vary. For example, treatment effects are found to be stronger as the number of enplanements increases.

Finally, care must be taken to acknowledge identification problems that may exist.. Since the airport infrastructure improvements provided to airports with commercial service may benefit general aviation, the positive treatment effects may be caused by general aviation rather than commercial aviation, as well as the impacts of airport construction and maintenance. Such a result would imply that subsidization of commercial airlines flying to small airports, such as with the Essential Air Service Program, does not cause a positive economic impact; justification of these programs, which often relies upon these assumed economic impacts, may need to be rethought.. Later research in Warren (2008) reveals some evidence that this identification problem is real at airports with low amounts of commercial service ..

29 MAA-4-2012-Report excerpts distributed at the meeting

References

Anselin, L" 1990, Some robust approaches to testing and estimation in spatial econometrics, Regional Science and Urban Economics 20 (2), 141-163 Bartik, T J" 2002, Evaluating the impacts of local economic development poli cies on local economic outcomes: What has been done and what is doable? Staff Working Paper 03-89, W,E. Upjohn Institute for Employment Re search, Benell, D. W, Prentice, B. E., 1993, A regression model for predicting the eco nomic impacts of Canadian airports" Logistics and Transportation Review 29 (2), 139-158, Bertrand, NI, Duflo, E, Mullainathan, S" 2004 How much should we trust differences-in-differences estimates? The Quarterly Journal of Economics 119 (1), 249-275 Besley, T, Case, A" 1994, Unnatural experiments? estimating the incidence of endogenous policies" The Economic Journal 110 (467), 672-694, Bivand, R., Anselin, L" Berke, 0" Bernat, A, Carvalho, NI, Chun, Y, Dor mann, C, Dray, S" Halbersma, R., Lewin-Koh, N" Millo, G, Mueller, W", Ono, H, Peres-Neto, P" Reder, M, Tiefelsdorf, M, , Yu, D, 2007., spdep: Spatial dependence: weighting schemes, statistics and models, R package version 04-2, Boarnet, M, G., 1998" Spillovers and the locational effects of public infrastruc ture Journal of Regional Science 38 (3), 381-400, Brueckner, J K, 2003a, Airline traffic and urban economic development., Ur ban Studies 40 (8), 1455-1469, Brueckner, J. K, 2003b The economic impact of flight cutbacks at the St" Louis airport: A calculation of job losses, Accessed February 14 2006, URL http://www. igpa. uiuc. edu/publications/pdf/stlforum.pdf Butler, S, E", Kiernan, L" J, 1987., Measuring the Regional Economic Signif icance of Airports (Report No, DOT/FAA/PP/87-1) US Dept" of Trans portation, Federal Aviation Administration, Office of Airport Planning and Programming, Washington, DC Butler, S E., Kiernan, L, J, 1992" Estimating the Regional Economic Sig nificance of Airports (Report No" DOT/FAA/PP/92-6), Federal Aviation Administration, Washington, DC Card, D, Krueger, A, B., 1994 Minimum wages and employment: A case study of the fast-food industry in New Jersey and Pennsylvania" The American Economic Review 84 (4), 772-793., Cohen, J P" Morrison Paul, C J", 2003. Airport infrastructure spillovers in a network system, Journal of Urban Economics 54 (3), 459-473, Cohen, J P" Morrison Paul, C J, 2004, Public infrastructure investment, interstate spatial spillovers, and manufacturing costs Review of Economics and Statistics 86 (2), 551-560 Diamond, A, Sekhon, J, S, 2005" Genetic matching for estimating causal effects: A general multivariate matching method for achieving balance in

30 MAA-4-2012-Report excerpts distributed at the meeting

observational studies .. Accessed July 12 2007. URL http://sekhon.berkeley. edu/papers/GenMateh. pdf Duffy-Deno, K T, Eberts, R W, 1991 Public infrastructure and regional economic development: A simultaneous equations approach Journal of Ur ban Economics 30 (3), 329-343 .. Efron, B, Gong, G, 1983. A leisurely look at the bootstrap, the jackknife, and cross-validation .. The American Statistician 37 (1), 36-48. Efron, B., Tibshirani, R J, 1994 An Introduction to the Bootstrap. Chapman & Hall/CRC, New York and London .. Freedman, D. A, Peters, S. C, 1984. Bootstrapping a regression equation: Some empirical results. Journal of the American Statistical Association 79 (385), 97-106 Garcia-Soidan, P H, Hall, P., 1997 On sample reuse methods for spatial data. Biometrics 53 (1), 273-281. Green, R K, 2002 A note on airports and economic development.. Accessed February 14 2006 URL http://www. bus. wise .. edu/realestate/pdf/pdf/ Airport%20and% 20Eeonomie%20Development16.pdf Hall, P., 1985. Resampling a coverage pattern. Stochastic Processes and their Applications 20 (2), 231-246 .. Holtz-Eakin, D., Schwartz, A. E .. , 1995. Spatial productivity spillovers from public infrastructure: Evidence from state highways .. International Tax and Public Finance 2 (3),459-468. Isserman, A. M, April 1997. The national role in rural economic develop ment: Some empirical evidence and policy implications. Benjamin H Hib bard Memorial Lecture Series. Isserman, A. M., Rephann, T, 1995. The economic effects of the Appalachian Regional Commission: An empirical assessment of 26 years of regional de velopment planning. Journal of the American Planning Association 61 (3), 345-364 .. Johnston, J., DiNardo, J, 1997. Econometric Methods: Fourth Edition.. Mc Graw Hill, New York Meyer, B .. D, 1995 .. Natural and quasi-experiments in economics. Journal of Business & Economic Statistics 13 (2), 151-161. Nunn, S., 2005 .. Flight plans for development: Aviation investments and out puts in nine metropolitan regions, 1990 to 2002. Economic Development Quarterly 19 (4), 295-312. Rephann, T., Isserman, A, 1994. New highways as economic-development tools - an evaluation using quasi-experimental matching methods .. Regional Science and Urban Economics 24 (6), 723-751. Rosenbaum, P R, 1987 The role of a second control group in an observational study.. Statistical Science 2 (3), 292-306 Rubin, D.. B, 1973 Matching to remove bias in observational studies. Bio metrics 29 (1), 159-183. Sekhon, J. S, 2007. Alternative balance metrics for bias reduction in matching

31 MAA-4-2012-Report excerpts distributed at the meeting

methods for causal inference Accessed July 12 2007 URL http://sekhon .berkeley. edu/papers/SekhonBalanceMetrics . pdf US Department of Commerce: Bureau of the Census, 1982. County Business Patterns, 1980. U.s.. summary, state, and county data .. ICPSR version Warren, D.. E, 2008 The regional economic effects of airport infrastructure and commercial air service Ph.D.. thesis, University of Illinois at Urbana Champaign.

32 MAA-4-2012-Report excerpts distributed at the meeting

2007 V35 1: pp.. 91-112 REAL ESTATE EeX)N

Airports and Economic Development

",,,'h':'>,'r; K.. ,_,'"c,,,,,y

The popular and local developrnent bOIl)sters huh !Hrnt\rfc as mechanisms for helping metropolitan areas nurnber of of such boosterism below. M~Dl.·t:ov'er, lifp','''\lllrr>c that on the importance inflrasltrm;tUlre in gerlen1ll, airports in particular, to economic developlnent and on the financing of such mlra~;tnJctun::.

so far as I can tell, there have been a limited number that have IO(lkecl at the impact of airports on regional growth (Brueckner 2003 of the many claims that have made {jbom the importance of airports to economic activity, this may seem surprising. Indeed, we may dmw a contrast with the sports stadimn literature, which contains a number of Zimbalist 200 I). That literature, which generally produces evidence that stadiums are not important to eeonornie development, involves a type of infrastructure that is much less costly than airports and that has rlnuch less direct influence on both business consumption than airports. In fact, one things th,lt makes airports inherently is that roughly two million take a cornmerciallflight every day, I so many people can relate to the surrounding

Airports are also controversiaL It seems that, for business leader who wants to see his Ql' her local airport expand, there is a resident in the night path

'Department ofFinanee, School ofBusiness, The u'-".~;'- Washington Unlversltv. Wash· ington, DC 20052 or [email protected] ..

I See the 2006 Bm(~au ofTransportation Statistics at http://www.bts.govJpressreleasesJ 2006Jbts028J)6/htmllbts028J16.html.

2007 American Real Estate and Urban Economics ASSOCiation MAA-4-2012-Report excerpts distributed at the meeting

wants to sec his 01

carl'! Ut MAA-4-2012-Report excerpts distributed at the meeting

had jumped from about 65,000 in 1989 to almost 80,000 in 1996. The study pegged the region's strong hub airport as an important factor in growth. 'Invariably, for nevv cornpanies coming to the area, they cite air service as a critical fflctor,' said Joseph Kramer, vice president ofeconomic developmem for the Greater Cincinnati Chamber of Commerce,

A f'\Llgcles Times story (Herndon I looks enviously at Dallas, when it notes:

DFW, the nation '$ second·busiest airport, has been a catalyst f(J[, some ofthe most impressive employment and population growth in the US since 1970, when construction ofDFW got Airport otllcials here talk about DFWas "engine" of the region's spectacular econOll1k growth. The nunlber ofjobs in Metroplex, as the four count.y region surrounding Ihe airport is known. soared more than I to the lJ Department of ,'ornnlerce. Nationally, employ111ent over the same period

The point H1 with the capacity constraints and Los t'\LI~CIC~ t .. ",u.I»,'c· and the relatively low of at Hopkins Airport many reporters in the popular have a maintained hypothesis matter to economic development, but they have not put the hypothesis 10 a testRemark· ably, as we shall see below, with one no one either.

Literature are four strands ofliteratUlerelevant to the ofairp0l1s and eClcm()mic development: the public finance literature, the economic development literarure, urban literature on theimpacl transportation on economies and the airport literature, r briefly discuss each of in turn.

Puhlic A cornerstone ofthe public finance literature is the identification, provision and evaluation of public goods. Aiqmrts (more specifically runways):! could well qualify as impure public goods because, until they are congested, they are nOll, rival in consumption. As such they are like highways. Therefore, in determining the efficient provision of airports, we must sum individuals' \villingness to pay for such services. In particular, the Samuelson rule tells us that in a world ofN

11 is important to distinguish rtlll\'v'ays from boarding runways

consumers. ic lI1(Jl\lllU;'1IS, U is a ';tanll;.HtI

IS MAA-4-2012-Report excerpts distributed at the meeting

manufacturing),4 places with relatively high concentrations ofcollege graduates perform better than places with relatively low concentrations of college gradu, ates {see Glaeser 1998) and larger places generaiIy perfonn better than smaller places (although there is a at which this advantage becomes exhausted (see Quigley 1(98».

Cienerally, economic performance is measured in terms ofemployment growth, population grmvth and/or income growth, r will on population and em, ployrnellt in this

Dena and ( \99\) rnade a important contribution when they studied the impact of infrastructure on economic developrnenL They found that, over the of time in infrastructure have produced a rate of than private investment capital and that government investment enhances return to private cap.ital. This that the opportunity cost of public the use of taxes that divert resources away private capital) has at times been than the beneHts it has produced. The Deno and Eberts 09(1) result is guite old, 11owever, It IS quite possible that, time since that study 'vvas published, government has developed infrastructure to point its relllrns have fallen to or below'

the levels produced by private \,.,UIJH.'ll.

Agglonleration l:con,'Hllles ,.",•.",,'- works document the relationship between the development of trans portation and the of Douglas North's (1981) book, which discusses the relative importance ofcanals and rails to the ,1,,,,,,,,,l,n1"--. ment the Midwest, and Will iam (199] ) on the development Cll!iCi.::I,f?O are touchstones of this

More recenUy, Gaspar and Glaeser (l and Yandell and Green (200 looked at of the Internet on agglolneration in gcneJraL and they that the Internet could be a complement, as weU as a substitute, for agglomeration. It is certainly that the Internet can replace somemeetings---e-mails and chat rooms enable people to communicate easily without being near one another. On the other hand, because the Internet puts more people in touch with each other, there are more possihle combinations of people who might wish to meet, either for business or social purposes. Consequently, it is possible that the Internet has and \viII continue to generate more air traffic than would exist in its absence. This puts cities that are in a

4 Although an article by Malpezzl, Seah and Shilling (2004) shows that the competi tiveness of individual corporations is even more important than industrial mix .. MAA-4-2012-Report excerpts distributed at the meeting

Do:Sltl.on to '''"i1LU:L lilf demand in a d,-"n,n",,,' POSI!lcm than those are not.

extent I

dala sources for this measure belo\\ MAA-4-2012-Report excerpts distributed at the meeting

J-\ 'I!;r HT<: anel Economic DeVO!'ODrTi8Irll 97

Figure I II Boardings per capita in 1990 and population growth from 1990 to 2000

9000%

80.00%

7000%

6000% .c;; ~ 5000% ~ c o +:: :;'" 0. 30.00% if. 20.00%

1000%,

Boardings per Capita

1990 and growth that occurred But I have not necessarily solved the problem for instance. a Inetl'Opolitan are ..t had compa· that promised to expand once observed adequate air serv Local govelllUneJlll officials would an incentive to hubs and build airport capacity before an expansion was actually Consequently, it would be the ofeconomic developrnent that led to atr and air ~''''··'Jlr'I'' would not the principal cause economic development.

In attempting to deal this possibility, I estimate models using actual mea sures airport service and instrumented measures. One idea for an instrument source comes from Brueckner (1985), \vhich looks at the detenninants of hub airport location. The determinants include such things as population, whether a Metropolitan Statistical (MSA)is a state capital and so forth. One of Brueckner's variables geographic location·~it sirnply the distance an airport is from a 11xed point. This variable has two virtues: it is cleady ex·· ogenous and it presumably has no impact se on economic development As such, the variable allows the impact of airports to be identified because it may included in the structural equation for airport location and excluded from structural equation for economic development.

Another instrument is lagged population growth. Past population grmvth could explain why airport tralfic is heavy in a metropolitan area, but cannot be MAA-4-2012-Report excerpts distributed at the meeting

CCDllonuc (le'velop'nH:::nl measure. x

one In

for me and for ~!",\\f"nrr me how to do MAA-4-2012-Report excerpts distributed at the meeting

The model specification in this article differs from tbe second Brueckner article in two other respects: he assumes the impact of airport traffic on economic development is contemporaneous. while I assume it operates with a lag; he uses one measurement of airport activity. total boardings, while I use boardings per capita. originations per capita. cargo per capita and hub status.

Data and Control Variables

A:1easunng Airport ActiVIty I use measures of airport activity. IlIst ITleasures I use are boardings and originations per capita in each metropolitan area. I begin with data from the Aviation Administration i on number of pass{~n~!er boardings at the I00 airports in the country. I at arbitrary ranking 100 at that point are airports that are small that have little impact and the airports are in \vhere there would alrnost airport

When there is more than one airport in a mctropo.l han area, I combine h("lf(~! lI'''"

MSA are in Table 1, Note that among the leading MSAs are Las Orlando and Atlanta, \vhich are also places that have experienced growth. The quirk is Portland, Maine, a that has had relatively little population growth and yet has an unusually airport for a city ofits BeeauseLas is such an outlier. I run sets that include and exclude it; the impact of excluding Las Vegas is inconsequential.9

The interpretations of hoardings and originations as variables are somewhat different. Hoardings represent total airport activity. but they could in principle

These data come from http://www.,transdata.btsgov. These data ate taken from the 1990 census of population and housing. \) Results excluding Las are available upon request. MAA-4-2012-Report excerpts distributed at the meeting

100

'fable 1 II !)O(Uc!l!H!S pel in 1990 ranked MSi\,

,'\V ' A/ :vISA Portland, ME I\IS:\ MSi\

240

KY IN

()R ,VA ('MSA . North Little AR i'vlSA MSA Ann Flint. Ml eMSA IvlSA Nlc,mes. IA .MSA SprU1,g.fwlcl. 011 MSA MAA-4-2012-Report excerpts distributed at the meeting

LUff""''" ancl Economic Development 101

Table 1 • continued

49 Boston- Worcester - Lawrence, MA - NH- Nfl:: - CT CMSA 1.84 50 Los Angeles- Riverside - Orange County, CA CMSA 1.. 81 51 Colorado Springs, CO MSA lSI 52 Oklahom;l City, OK MSA 1.79 53 Omaha, NE .. IA i\'1SA 1.74 54 Ne," Y<)rk - Northern New Jersey - Long Island, NY NJ· CT - PA CMSA 1.71 55 Jacksonville, FI. MSA 1.70 56 Syracuse, NY MSA 155 57 Huntsville. AI. iVISA 1.52 58 Columbus. OIl MSA 1.43 59 Sacramento Yolo. CA CMSA I 60 Cleveland Akron, OB CMSA 1..38 61 Buffalo - Falls, NY MSA I 62 Charleston North Charleston. SC MSA 1.35 63 WI MSA I 64 Columbia, SC MSA I 65 Birmingham. AL tvlSA 1.3 I Mihvaukce . Racine, WI CMSA I 67 Wichita. MSA I 68 Philadelphia - Wilmington - Allantic PA, NJ . DE --. MD CMSA 1.24 69 .. Fall River Wanvick. RI MA rvlSA 1.21 70 Richmond·· VA MSA 1.15 71 NY MSA I 12 n MSMSA 1.10 Lmilisville. KY - IN MSA I 74 Albany - Schenectady NY I 75 WinstQI1 Salem High Point, MSA IJ) I 76 TN MSA 0.96 77 Virginia Beach- Newpon VA MSA 78 Harrisburg - Lebanon Carlisle, PA MSA 79 Grand Rapids Holland.MItvlSA 0 ..76 80 Bossier LA MSA 0.71 81 Greenville Spartanburg tvISA (Un Fresno. CAMSA 83 Fort Wayne, INMSA 0.58

have little spillover beyond the airp0l1 itself. Imagine an airport that was de signed just to a hub: one without any p

102

measure 10

over Into econonllC 'u'tnllt\,

m;:lrgm,lJ tax nile IS Maryland and Virginia MAA-4-2012-Report excerpts distributed at the meeting

103

Climate \/ariables As a proxy for the mildness ofclimate, I use average heating and cooling degree days for each MSA as reported by the National Oceanic and Atmospheric Administration. Places that are cold have large numbers of heating degree days; places that are warm have large numbers of cooling degree-days., Since World War Ii, there has a major migration to the Sunbelt from the Northeast and Midwest A likely explanation for this is that air conditioning made places such as Houston, Dallas and Phoenix tolerable in the surnmer. Consequently, explaining population growth across American MSAs requires some controls for climate.

HI/man Capital \i(iriables

I'he importance of education to economic develolnnent is well established (Bartik 1999, Glaeser (1998) that cre ate positive externalities for non-college graduates in the labor f()rce Ilh£,.,,,t'C'T'" include as a control variable the of population above the of 25 with high school diplomas and college degrees. 'fhe data come from the 1990 census of population and housing.

Industrial 1.'1'I'IlrI' ,,,.£)

I measure the industrial structure of metropolitan areas data from the County Business Patterns, Specifically, I look at the of employment in manufacturing and the share of employment in the FIRE secwr. I choose man ufacturing employment as 11 variable over the course of the 1990$, manufacturing ernploymenl much melle slowly than overall e.mployment; 1'r)ll1v{'r':I~lv FIRE employment much rnOfe quickly.

"n .'U"~J and Ray (I right-to-work laws can have an impact on employer location c1elClSIOrlS and therefore economic growth. I therefore include a dummy for whether an fv1SA is in a righHo-work state or not.

While large cities reap the benefits of agglomeration economies, after a point they also produce negative externalities. Principal among these is congestion (see, e,g., Tabuchi and Yoshida 20(0). I therefore include average commuting time as measured from the 1990 census as an explanatory variable. MAA-4-2012-Report excerpts distributed at the meeting

104

Results

2. 042:-:

o X06

X3 {} 57 0.12 MAA-4-2012-Report excerpts distributed at the meeting

and Economic Devel,::)O!nO:1t 105

With respect to the boardings, originations and hub first-stage equations, the news is good, With other controls in place. previolls-decade population growth and runway capacity do a good job of eKplaining airport activity On the othCI hand, pIOximity to Kansas City and the nearest coastline do not do a good job of explaining boardings or originations per capita and hub status, Airports 'within 100 miles of a hub activity.. Industrial makeup and the prc:seJ1Ce of a state do not seem to influence passenger activity,

'I'he equation, on the other hand, works only thing that seems to explain cargo trafticwell is the presence of a long runway. But in light that moving become this result should not too surprising.

I now turn to the leatmed re.i~ressjon reSUlts, arc pn;sentc~d in ]hbles Table 3 contains the re.£~re:~sj,.ms use hoardings as an ex, planatory variable. Columns ] and the (oes) and instrumental variables (IV) explaining population growth: columns 3 and 4 oes and reg~re~;Sl(mS explaining are complete sets metropolil.an ar(~as,

\vith population Whether oes or IV are used, it is found that boardings per in 1990 has a substantial impact on population growth from 1990 to 2000 Both OLS and IV regressions have boardings t thai the magnitude if the IV coefficient than the magnitude of the OLS coeft1ciem-.although a Hausman test produces a test statistic on the null of eXJ)g(~nelty of 0,03, suggesting Ihat the OLS coefficient is more eflkient. The coefficient of 0.025 on OLS means that a one-standard-deviation increase in boardings per capita would produce an increase in decadal population growth. In the context of a country that had overall population gHl\vth of 1 over the course of the ]990s, this is a substantial number. Other variables that are statistically significant include the human capital variables, whether an MSA a state capital and heating (waml places grow taster than cold places). This is consistent with past literature (Glaeser and Shapiro 2003).

Now I tum to job growth. Note again that the coefficient on job growth in both the oes regression and the IV regression are significant at the 99CVo level. Once again, the magnitude of the coefficient in the IV regression is higher than it is in the oes regression .. If the OLS regression coefficient is used, it is found MAA-4-2012-Report excerpts distributed at the meeting

106

Table 3 II Hn.'lrrl[llu,,,

POllJulati()Il Growth crnplo'Yl1lcnt Growth 990 2000 19902000 OLS OLS 2SLS -0.157

0028

(L29~

1090

0099

O."XO

-0.846

0..023 0.027

OJ.l0004 I I 0018 0032

OJ1IO

72 0.69

liCliC[()sKecltlcnCll\ concdcd stm1Clard errors In pmclltlJesis.

pnXHJC(:S an 8 MAA-4-2012-Report excerpts distributed at the meeting

107

Table 4 • Originations per capita regressions.

Population (irowth l:mployment Growth 1990·2000 1990··2000 OLS 2SLS OL$ 2SLS

Intercept ·0.157 lOJ HI) Originations per capita 0033 0,028 (0.004) Percent of \\imkforce in 0..228 0269 0.29J lO.1 1 ~O.,706 Percent of 111 -0.415 HRE jobs Stale capital located 0,067 -0067 within CMSA/MSA l) population O.67H 0606 with school (0.1 (0166) and above Percent of POIIlll"lllt>l! -0.307

000002

corp. income tax -0.422 mte - state Icvd personal income tax . (}.270 mte Slate level corp. lax rate HUH :n ,,·OJ)I,:) l 0,0002 0,0003 o 1) Pailurcs ver"us bu\iness 12 starls l) ,·0.00002

IlI.,.!\!\)'I,l) ···0.004 ·0.005 -0.008 ·0.004 ·0004

0:77

similar 11l

Finally, 'nlble 5 contains the that use hub slams as an explanatory variable. MSAs are deemed to have hubs if they were designated hubs by one MAA-4-2012-Report excerpts distributed at the meeting

108

Table 5 II Hub ,c:.;lC"""lD

19902000

0 .. 078 0078

0621

. (IOOOOi .. \10(0»

0,004 {J 02",

0'<)21 fUll6

83 06

or MAA-4-2012-Report excerpts distributed at the meeting

109

Table 6.

Population Growth l:mplo,vmelH Gnn"th 1990~~2000 OLS 2SLS OLS 2SLS lrllen:ept OA53 ·0.390 0..515

per 0024 0.60x 0098 0419 1 I Perc\;~nt of workforce in ~0..065 0037 0.031 0074 manuf~lcturing I) Percent of tn ~ OAn FIRE State capital located 0087 (JOSS 0.090 within CMSA/1vISA Percent of with IUl63 0.857 I 18 school and above Percent of with ·.. 0389 '. 0.15·+ 18) o.oomn 000005

corp. income tax 109 0974 101 rale state lewl personat income tax ·····0.337 ··0498 0.847 nne .. slat..:: level 16) corp income tax 0.010 OJ)13 0.012 late level I -0.015

·0.001 ~ 0.001 rates per $1 ,000 1<'1 t~m·to·W()·1 K Slate fUl36 (U122

Failures versus business 00008 0.005 0.006 0..009 stalls (OJ) 179) (0.018) Hl020) (0.018) 0000002 ··~O.OOOOI .. 0.00000008 ..···0.000004 IfWOO(15) (0.00002000) 10.00001661 ) (0.00002) AVI'rnIW commute time 0,007 0..003 0.001 ... ~ 0,00.01 (0005)

83 83 82 0.61

population growth from 1990 to 2000. Astonishingly, the results imply that hub cities grew between 9 and 16 percentage points faster than nonhub cities.

Job growth in hub cities was similarly impressive. Again, the coefficient on hub status is significant, whether looking at the OLS or the two-stage least squares MAA-4-2012-Report excerpts distributed at the meeting

110 MAA-4-2012-Report excerpts distributed at the meeting

111

case, representatives whose districts include an airport have a strong incenlive to become rnembers of the airport authority. Consequently, decisions abolll air· ports can based on parochial interests, even if the total benefits of the airvort to the economy exceed the cost. Should air tranie be a large determinant of economic success, it is entirely possible that the benefits new or expanded ailvorts exceed costs.

Yet in many places (San Francisco, Boston, have \vorked to inhibit runway expansion, while in other (Chicago), local political squabbles have prevented number of potentially reasonable plans for expanding airport capacity from going forward a timely nmnner All this that airport policy might best be rather than locally. Of course. the pol would to include a scheme for compensating injured But if regional be1ilel/ts airport development are large, the costs fair compensalion should be I() 1111 ,Jl1ce.

Rmmrflf and Kris:en W'llliwlIs research assistance aliI! as lvell as Jan Brueclalcr alld Dm id cmmnentl, All errors are mine. and all OjJliliitH/S are mv 011'/1

References n~,,,,,,... A, 1999 Cincinnati Credit for Fast Cho\vth Plain Dealer Cin·

"\""lU,L October 31 1991 Who !JPllPflf'i Kal:ll11azoo, MI: ,) !J!""" !nSl.ltute

1 the lrnlxlcts 01 Logl.~,IU's and 11IHHr1n"fn't;fl" A Note on the Determinants of Metropolitan Airline Inter IWlSPOr! Economics 12: 175~ 184 I:'CIOmJmIC I;levelcrpn:lcI1L Urban Studies 40: 1455~

Gaspar, J and E., Glaeser, 1998 Information TechlrIollo,12CY and the Puture of Cities. Journal 0/ Economics 43: 136~ 156 Glaeser, E. 1998 Are Journal /',I,on,nn,li' /'j.Jn:nf'f'lll'f'\ 12(2): 139~160 '.''''''''\'', E and J Shapiro" 200.1 Urban Gnl\vth in the I990s: Is City Living Back? Journal ofRegional Science 431: 139-165. Goldberger. A, 1991 A Coune III Econometrics Cambridge. MA: Harvard University Press. MAA-4-2012-Report excerpts distributed at the meeting

112

P and M 1 !:.lrlpllJVrl1eJll {;!llWlh lhe 19XOs:A Shlf!-Sha[<~ Hn;""'" ()~ Herndon, R

and S .. Rosen. 988 l';stmlall:l1g Ve(.~lof~i~,ulof(~gpCSSlmls with 50: 91 118 MIJJ1("j:uaph on Stale and Local Policies Ui'N""""'·f\ihldi:mn Cenlef lor Urban Land Economics

Inrl""ln:\1 Struclure and t:nlp!()Vi!l1eJ11

E'('(1110Jnl( Nonon l),\u>"",l\! and Economic Growth .I0il! nat

Yoshida ';000 :'lep:lI'alln~

t;K: MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth: An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert

CAUSAL RELATIONSHIPS BETWEEN AIRPORT PROVISION, AIR TRAFFIC AND ECONOMIC GROWTH: AN ECONOMETRIC ANALYSIS

Florian Alkoggen, Institute of Transport Economics, University of Muenster/Germany E-mail: [email protected]

Robert Malina, Institute of Transport Economics, University of Muenster/Germany E-mail·[email protected]

ABSTRACT

As globalization progresses, air transport as a means of rapid transportation over long distances, is becoming more important to the development of economies. Thus, the availability of air transportation should exert positive effects on economic growth in the vicinity of an airport. In this paper, we present evidence of such positive economic effects and reveal the causal relationships, by using a production-function approach. The econometric estimation is based on a panel data set of major German airports"

Keywords: airports, economic effects

INTRODUCTION

Airports are the essential link between air and surface transport. In an increasingly globalized world, rapid transportation over long distances is becoming progressively more important to the development of economies. Air transport facilitates fast and intercontinental traffic, thus contributing to economic competitiveness" Consequently, the availability of air transport in the vicinity of airports might exert positive effects on economic output. The impact of airports on productivity and long-run economic growth are, in fact, not their only economic effects. Private individuals benefit from the availability of airports in the vicinity of their homes, through, for example, being able to reach a wide variety of holiday destinations" Of course, these effects of leisure traffic are economic benefits which one has to take into account in order to study the full economic benefits of airports" But as these

1:!h WCTR, July 11-15,2010 - Lisbon, Portugal MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth:' An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert private benefits are not measurable in the output of an airport catchment area, they are not incorporated in the following analysis. Thus, the scope of the study is limited in this respect. Furthermore, the provision of an airport infrastructure and its operations also impact on aggregate demand. These effects are usually identified by employing input-output analysis, based on the national accounts system" Although these effects on aggregate demand cause substantial value-added and employment, they do not include the positive effects which arise due to the macroeconomic input of airport capital stock and increased air traffic services. This analysis focuses on the latter effects. These positive long-run effects are caused mainly by enhanced productivity in the airport region" Due to fast air-traffic connections for business passengers and cargo, as well as reduced travel expenses and connections to distant destinations, enterprises are able to boost their competitiveness through cost reduction and market development. These effects are regional, as they are only generated as far from an airport as their efficient use permits" Macroeconomic regression analysis is often used to research the productivity of public capital, such as airports, streets and railways" Although the capital stock of German airports is not completely public, the production function approach provides a valuable framework for discussing the positive output effects of infrastructure, irrespective of ownership" In this paper, we first consider the research on the productivity effects of public capital and on the growth effects of air traffic" Based on the main issues emerging from this overview, we develop a model for quantifying the assumed positive effects of airports and estimate the model parameters by employing econometric methods. The results of the study are presented and discussed"

PRODUCTIVITY OF PUBLIC CAPITAL AND ECONOMIC EFFECTS OF AIR TRAFFIC

Looking at the literature on the impact of infrastructure and air transport we can distinguish between two main approaches: The first one tries to analyse the effects caused by the provision of infrastructure without explicitly taking into account traffic and traffic patterns on the infrastructure. The second one analyses the impact due to traffic developments at a given level of infrastructure capital. Hence, this analysis does not account for the effects of infrastructure provision"

Growth effects caused by the provision of the infrastructure

Since Aschauer's (1989) work on the productivity of public capital, a broad discussion on econometric methods, measurement of public capital and the economic plausibility of the results has emerged. 1 Aschauer's study is based on a production function approach, linking macroeconomic inputs like labour, private capital and public capital to macroeconomic output. Aschauer found that the output elasticity of core infrastructure such as highways, airports, electrical and gas facilities, water and sewers in the USA is 0.24. 2 This result soon

1 For a comprehensive survey of different approaches see Romp / de Haan (2007) 2 Aschauer(1989), pp,193f

12h WCTR, July 11-15, 2010 - Lisbon, Portugal

2 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth.: An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert became the subject of controversy, as it implies a rate of return on public capital investments higher than 100% in the first year after investment.. This result is considered to be implausibly high. 3 The criticism of Aschauer's results focuses mainly on the methods employed to estimate the output effects of public capital. The alleged problem is that inappropriate methods might lead to upward bias" Parameter endogeneity, non-stationary, but potentially cointegrated time series and measurement errors relating to the public capital stock are the main points of discussion:4 Endogenous regressors in a regression analysis arise from reverse causality. If correlation between public capital and economic output is influenced by causal links in both directions, simple regression analysis (OLS methods) yields biased estimates.5 Because economic growth increases national budgets, the additional funding can be spent on new public investments. Hence, public investment may foster economic growth, and conversely, economic growth may foster public investments" Thus, reverse causality is a relevant source of estimation bias. More recent research uses various different methods to overcome these 6 problems" Some researchers use simultaneous equations , while others include instrument variables in their analysis.? As an alternative means of overcoming the reverse causality issue, some studies use vector autoregressions (VAR) or vector autoregression error correction models (VECM), which do not assume any form of causal relationship between the variables,,8 The growth effects of public capital have been analyzed mainly on the basis of time-series data. If these time series are non-stationary, the estimates might reflect spurious regression" Hence, the estimates could indicate a causal relationship which does not exist. Thus, a test for unit roots in the data is necessary. If there is evidence that the data is non-stationary, the estimation must be conducted in differences or using methods of de-trending,,9 Nevertheless, the non-stationary time series may also be cointegrated. In this case, error correction models can be used in order to develop the model.. In contrast to methods of de-trending or applying an estimation in differences, this approach is superior in identifying long-run relationships between the level of pUblic capital and economic outpUt,lO A third concern about the analysis is the measurement of public capital.. Because public capital is provided by the state, there are no market prices to determine its value. Furthermore, a cost-related valuation of the public capital stock does not provide an unbiased quantification, as the historic costs of producing public infrastructure do not reflect the economic costs, due to inefficiencies in the public sector. 11 Thus, in many investigations, the public capital stock is overvalued. In the context of transportation infrastructure, the loose relationship between the monetary value of public infrastructure and its performance becomes even more problematic, due to different country characteristics. For instance, a certain capital expenditure on highways for a flat region of a country creates a larger highway

3 Gramlich (1994), pp.1185 I 4 Romp / de Haan (2007), pp. 7 I and 12 I. 5 Gramlich (1994), pp.11881 6 See lor instance Esfahani/Ramirez (2003) or Cadot et al. (2006) The latter models the political- economy process in which decisions about public inlrastructure investments are reached 7 Ai / Cassou (1995), Boarnet (1997) or Calderon / Serven (2002) 8 Romp / de Haan (2007), ppA6ff lor a detailed survey 01 these studies 9 Aaron (1990), p..53 10 Munnel/(1992), p .. 193. 11 Sanchez-Robles (1998), p.1 00

12h WCTR, July 11-15, 2010 - Lisbon, Portugal

3 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth:' An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert network than the same capital investment in a mountainous region, because highways need more expensive engineering work for tunnels or bridges in the latter case. 12 In a comprehensive meta-analysis of 76 studies, based on the production function approach, Bam and Lighthart find an output elasticity of 0.086 for public capital, taking into account some of the flaws mentioned above. Moreover, they provide evidence that neglecting cointegration or spurious regressions from non-stationary time series, are an important source of estimation bias,,13 Attempts to quantify the productivity of transportation infrastructure have been conducted particularly with regard to highways" Recent research by Ozbay et ai, quantifies the output elasticity of highway capital at 0,171,14

Growth effects caused by the utilization of the infrastructure

The economic effects of air traffic are mainly discussed with regard to the utilization of airport infrastructure instead of focussing on the mere infrastructure provision" For instance, a study by Brueckner provides evidence that airline traffic positively effects employment in metropolitan areas in the USA. He finds that a 10 percent increase in airline traffic, which is defined as the number of boardings, raises service-related employment by 1,1 percent.. Furthermore, he finds that these effects arise in the service-related sector but not in goods related sectors. Thus, the positive effects of airline services on economic growth are explained with the theory of intercity agglomeration, because high quality airline services facilitate easy face-to-face contact with business partners in other cities" From a methodological point of view, this study is mainly concerned with the problem of endogenous traffic measures, Thus, instrument variable estimation is introduced and tested against ordinary least squares procedures. However, Brueckner does not find evidence that traffic figures are not exogenous. 15 Further research by Green confirms the positive effects of airline traffic on employment growth. Moreover, it provides a more disaggregated view on air traffic. Also accounting for possible reverse causality with the help of two stage least squares estimation procedures, Green reveals that the effects of airline traffic, measured as boardings or originations per capita, on employment is positive. Moreover, he finds that the effects are larger in hub cities and do not exist in case of air cargo,16 In addition to the literature on air traffic, which is explicitly focussed on the utilization of the airport infrastructure, some research on the economic effects of highways accounts for the effects of the utilization of the highways as well. Fernald points out that sectors with high vehicle intensities benefit more from highway investment than sectors with low vehicle intensities. 17 In addition to this result, Boarnet introduces variables of highway congestion into the analysis. He finds that congestion of highways is an important obstacle to economic growth. 18

12 Oi Palma / Mazziotta (2003). p,369 13 Born / Lighthard (2008), p.32 14 Ozbay et aI., (2007), p,327 15 Brueckner (2003) 16 Green (2007) 17 Fernald (1999) 18 Boarnet (1997)

12h WCTR, July 11-15, 2010 - Lisbon, Portugal

4 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth:' An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert The analysis presented in this paper focuses on assessing the positive output effects of airports and air traffic" In addition to the empirical studies mentioned, both effects are included explicitly in the estimation, We use monetary values of airport infrastructure in our analysis instead of physical measures or performance indicators as proposed in the literature" This is due to two reasons: First, contrary to road and rail infrastructure, airports are usually not built at unfavourable and therefore costly locations. Hence, 1 EUR of airport infrastructure can, in principle, represent the same amount of physical infrastructure, Second, all German airports are run as private businesses and some airports are even at least partly privatized. 19 As a consequence, the likelihood of biased capital stocks of German airports is low. As the connectivity of an airport is determined by scheduled traffic and not by individual general aviation flights, we explicitly include air traffic conducted by airlines into our analytical framework, The studies by Brueckner and Green use boardings and originations in order to measure the quality of air traffic services, In our analysis we employ the number of commercial aircraft movements as a measure of quality and quantity of air traffic supply. In contrast to the indicators, which are used by Brueckner and Green, aircraft movements is a supply-side indicator which does not reflect specific demand patterns or the size of aircraft which are used at an airport.

THE MODEL

Macroeconomic production functions model the technical relationship between macroeconomic inputs and macroeconomic output in a given production system., Thus, it is necessary to identify the relevant determinants of macroeconomic output that are correlated to other relevant variables included in our analysis" According to basic macroeconomic theory, we include labor L and capital K., Given that we wish to study the impact of airport capital expenditure on economic output, we also include the capital stock of airports AIRPORTCAP. Furthermore, we need to include a measure of the quantity of air traffic at an airport, so as to avoid omitted variable bias, as shown above. The connectivity of an airport is determined by the number of routes and by the operative frequency of these routes at the airport. Hence, we use the aircraft movements MOVEMENTS in our analysis. Additionally, we include the performance of other surface transportation infrastructure INFRA into our analysis, in order to avoid further omitted variable bias, The basis of the model is a Cobb-Douglas production function. Using the parameters set out above, we formulate the basic model as follows: Y = A· LaKfJ (1a)

¢::::> In(Y) = In(A) + a . In(L) + f3 . In(K)., (1b) According to Romp and de Haan, public capital can be included in the analysis as a macroeconomic input factor or as a determinant of multifactor productivity,,20 As the capital stock K in our data already includes public capital and specifically airport capital, we only

19 Almost 50% of the important German airports of Frankfurt, Dusseldorf and Hamburg are privately owned according to data from the German Airport Association 20 Romp / de Haan (2007), pp, 1Off

12h WCTR, July 11-15, 2010 - Lisbon, Portugal

5 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth.' An Econometric Analysis ALLROGGEN, Florian;' MALINA, Robert model the effect of airport capital and other infrastructure on the multifactor productivity measure. In a general model, this yields:

Y = A (AIRPORTCAP, MOVEMENTS, INFRA) . LaKfJ (2) In accordance with the transformed model, we use the logarithmic multifactor productivity In(A) for further modeling. We employ a second-order trans-log function to specify the multifactor productivity. This methodology is analogous to the modeling of Boarnet, who comparably includes highway capital and congestion measures of highways to quantify the effects of highway investments. 21 Taking into account entity fixed effects and period fixed effects, we can specify the multifactor productivity" Substituting the multifactor productivity into the basic Cobb Douglas production function yields: In(Y) = [Ezln(AIRPORTCAP)Z + E1In(AIRPORTCAP) + (zln(MOVEMENTS)Z + (lln(MOVEMENTS) + lzln(INFRA)Z + L1In(INFRA) + KA,J In(AIRPORTCAP) In(INFRA) +

KA,W In(AIRPORTCAP) In (MOVEMENTS) + KI,W In(INFRA) In(MOVEMENTS) + Lf=l YiDi + Lf=l OtTc] + a In(L) + f3ln(K)" (3)

This modeling includes some universal properties. Thus, our model does not impose restrictive assumptions on the way airports and other transportation infrastructure influence economic output. Exemplarily, the output elasticity of airport capital reveals the possible economic implications which can be considered in our model: ~Y/ Y ~ aln(Y) _ ~AIRPORTCAP/ AIRPORTCAP ~ aIn(AIRPORTCAP)

= 2Ez ln(AIRPORTCAP) + E1 + KA,W In(MOVEMENTS) + KA,I In(INFRA)" (4) The output effects of investments in airports may depend on their capital stock. It is possible that the effects of investments rise (102 > 0) or decline (102 < 0) as the capital stock of an airport increases. The latter implication may be plausible, due to rising opportunity costs of capital. The initial implication may be caused by network effects, which can be generated at a larger airport with a greater capital stock.

A constant level of the output elasticity of the airport capital stock Cl is also considered in the model. The output effects of airport capital can be studied, subject to aircraft movements at the airport and the quality of surface transportation infrastructure in the catchment area of the airport. For instance KA,l > 0 implies that a less competitive surface transportation infrastructure22 might lead to greater positive output effects of airport investments" The reason might be that airports and air traffic constitute a fast link to economic centers, which is even more necessary than within regions with a competitive surface transportation infrastructure. Furthermore, the coefficient KA,W models the interdependencies between

21 Boamet (1997), p.44 22 Because INFRA is a measure of access times in an airport catchment area, a high value of INFRA implies a less competitive infrastructure

1::!h WCTR, July 11-15, 2010 - Lisbon, Portugal

6 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth:' An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert

airport investments and aircraft movements, As an example, KA,W < 0 might reveal that the output effects of airport capital can decline, as air transportation connectivity is enhanced, In short, the model does not contain restrictive economic a priori assumptions on the effects of airports on economic output. Due to the variety of different airports in our sample, which includes first-tier hub airports like Frankfurt as well as third-tier airports like Bremen, a flexible functional form, which can reflect different causal relationships, is necessary for this analysis" Although our economic reasoning confirms the plausibility of the model, we must test the functional specification of the model on the basis of our data set, so as to avoid misspecification.23

ECONOMETRIC ESTIMATION

Definition of "Influenced Areas"

As said before, the positive effects of airports on economic output are generally regarded as being regional.. Hence, they are only generated in areas adjacent to the airport, in which people use the airport.. We call these areas "influenced areas". These areas differ from the catchment area of an airport or its air traffic market because the influenced area only reaches as far as competitive advantage in comparison to the utilization of other airports arises for the passengers at this airport. Although there are further passengers from outside the influenced areas, the use by these people does not cause economic growth as there is no competitive advantage due to the existence of the airport.. Thus, the catchment areas are supposed to be larger than the influenced areas. Influenced areas can be calculated based on different criteria. As an example, they are visualized for Munich Airport in Figure 1:

..- Motorway = Motorway (planned) + Airport -- Border ofa County I City

-- Border of a State

Definition of influenced areas:

-- Iineardistance(50km circle)

access time to the airport

overall connectivity ofthe airport

Figure 1 - Definition of influenced areas: the example of Munich Airport

23 See Estimation Section

12h WCTR, July 11-15, 2010 - Lisbon, Portugal

7 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Aifport Provision, Air Traffic and Economic Growth: An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert 1. Definition based on linear distance As the growth effects of airports and air traffic are believed to be regional, they are generated at locations which are situated "near" the airport Therefore, we define an influenced area based on linear distance to the airport., In Figure 1 "near" is defined as a circle with a 50 km radius around Munich airport.,

2. Definition based on access time to the airport Although it is reasonable to have a basic definition of influenced areas with the help of linear distance to the airport, this definition does not account for access conditions to the airport Thus, the definition of influenced areas should rather be based on access time than linear distance to the airport" In Figure 1 all regions and counties are allocated to the influenced area, if the average access time to the airport does not exceed 45 minutes.24 Not surprisingly, good transport links to the airport, e.g. due to fast motorway connections, lead to an expansion of the influenced area beyond the scope of the definition based on 50 km linear distance.

3. Definition based on the overall connectivity by using the airport Mode and route choice of passengers usually includes total travel time" Therefore, we have to account for total travel time as well when defining the influenced area of an airport. Assuming a threshold of reaching destinations within Europe within one working day,25 we find that the influenced areas of hub airports like Munich are expanded even beyond the scope based on access time. This is due to the fact that direct connections at larger airports provide short travel times compared to smaller airports where most connections require feeder flights to hubs and, thus, have longer overall travel time. However, employing this overall connectivity criterion to smaller airports would, however, lead to implausibly small influenced areas,

Concluding from the evaluation of the criteria we use a combination of the first and the third criterion. It is not necessary to explicitly use criterion two as well, because its indicator (access time) is already part of the third criterion" Thus, we first calculate influenced areas for each German airport based on a maximum linear distance to the airport of 50 km" Because larger airports with high air-traffic connectivity and high intermodal connectivity may affect larger regions, due to their easy access and fast air connections without further transfer times, we expand the airport influenced area of an airport where necessary, by using connectivity measures for the influenced areas. Due to high spatial concentrations of airports in some German regions, the influenced areas of certain airports overlap" As it is not possible to allocate growth effects in regions within the influence area of more than one airport to a specific airport, we cluster airports with overlapping influenced areas" We calculate a common influenced area for the airport cluster and allocate the growth effects to this cluster. Of course, there might be positive or negative spill-over effects to other regions outside the influenced area. However, they cannot be accounted for in our framework. 26

24 See BBSR (201 0) for data source 25 See EGAD (2008) for further elaboration on the enlargement of catchment areas 26 See for instance Ozbay et al. (2007), p .. 325

1:!h WCTR, July 11-15, 2010 - Lisbon, Portugal

8 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth' An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert The Data

The econometric estimation of our model is based on a panel dataset containing 11 airports or airport clusters and a time dimension extending from 1997 to 2006. We choose a rather short time dimension in order to avoid the introduction of structural breaks into our model which might exist due to the reunification in Germany" The influenced areas of the airports are defined as described above. Because the influenced areas of the airports are based on German counties and towns, the relevant data for the estimation of our model can be found in the national accounts. The real gross domestic product in the influenced areas is used as an output measure" The labor input is that of the labor force in the airport region., Given that the labor force is based on the number of individuals not on work input, we standardize the work force in the different counties, using the volume of working hours per work force unit. The macroeconomic capital stock of German counties is not available" The most disaggregated level of the data is available for the German Federal States, divided into six sectors of economic activity" We use the capital intensities for these sectors of the relevant Federal States to approximate the capital stock of the counties in the airport regions,27 Because the defined capital stock includes the entire capital stock of the influenced area, including public capital and the airport capital stock, we quantify the effects of the airport capital stock on output, taking into account the average opportunity costs of capital appropriation in the economy.28 Moreover, the model also includes data about aircraft movements and the airport capital stock, As the macroeconomic capital stock in the German national accounts is defined as the gross capital stock, we use the acquisition and production costs in order to measure the capital stock of an airport. Data for the capital stock of German airports and the number of scheduled flight movements were provided by the German Airports Association, Furthermore, an indicator of the quality of surface transportation infrastructure is included in our model. As explained in the literature overview, performance indicators based on accessibility times for road and rail, rather than monetary valuations of the surface transportation infrastructure are considered. 29 Using weighted averages, the indicators are aggregated into one indicator and to the level of the influenced areas. Given that this indicator is based on accessibility times, a decline in the indicator value reveals a higher infrastructure quality.

Estimation Methods

Based on our panel data set, we perform a two-stage least squares (TSLS) estimation. Thus, we need to identify a valid set of instruments which are strictly exogenous, but relevant. In particular, the aircraft movements at an airport and the quality of public infrastructure are endogenous variables. We use the population in the catchment area of an airport and lagged values of the infrastructure performance indicator as valid instruments" These instruments are considered to be exogenous, as the population in an airport catchment area is not

27 Deitmer (1993), p,,35 28 Canning (1999), 29 We use two indicators for road and two indicators for raiL These indicators are generated by the Federal Office for Building and Regional Planning for all German counties and towns

1:!h WCTR, July 11-15, 2010 - Lisbon, Portugal

9 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth; An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert influenced by variations in economic output3° and lagged values of infrastructure performance cannot be influenced by output variations in a certain period. Furthermore, we add an exogenous measure of airport competitiveness, compared to other German airports.?1 Finally, the airport capital stock is included only as a lagged measure, which eliminates potential endogeneity. The lagged inclusion of the airport capital stock32 is necessary, as the causal effects of airport investments can only arise after the infrastructure becomes operational and when it is actually used by the airlines. Because our definition of capital stock also includes infrastructure under construction, and there is a need for airlines to adapt to new infrastructure, we use lagged values of the airport capital stock in our analysis. Performing panel unit root tests, we find evidence of non-stationary time series in our data. This result is robust with respect to the inclusion of trend variables in the test setup .. Furthermore, the first differences of the time series are stationary.. 33 Hence, these time series are integrated of order one.. As there is no significant evidence of long-run cointegration, we do not construct an error-correction model.. 34 However, this implication may also arise from the short time dimension of our panel.. Consequently, we apply the TSLS method to the data in first differences, which might ignore a positive long-run relationship in the levels of airport capital and economic outpUt. 35 Using the first differences in our model, we add out the entity fixed effects. They are only included implicitly in the first differences of the model. Furthermore, the period fixed effects will be used as period fixed growth effects, rather than period fixed effects, in order to simplify the model. Because we use a panel data set, we have to consider cross-sectional and serial autocorrelation .. Serial autocorrelation is tested with a panel version of the Durbin Watson test statistic.. Cross sectional dependence is researched by means of a cross-sectional dependence test by Pesaran (2004) .. 36 As the tests for autocorrelation require residuals, autocorrelation is considered in the process of model validation. If we find evidence of cross sectional or serial correlation, panel corrected standard errors (PCSE) are used to ensure the econometric validity of the statistical inference in the model.

Estimation

The estimation of the specified model is conducted with the aid of the software package eViews 6.. Table 1 shows the results of the regression analysis .. Model 1 is similar to Aschauer's approach of introducing public capital as a macroeconomic input factor. This model contains neither quadratic summands37 nor cross products. We find that the non-quadratic inputs are insignificant, except for the inputs of labor and capital. In contrast to these results, we show Model 2, which contains all cross products and quadratic terms, but no non-quadratic expressions .. The results become more significant, but, due to some insignificant summands, the model fit is not ameliorated. Model 4 does not contain

30 Nevertheless. output changes are caused by a change in the work force 31 The measure is constructed as the share of the airport's WLU to the total German WLU 32 The lag used is 6 periods 33 Test for panel unit roots are based on ADF Fisher tests for individual unit root processes 34 Cointegration testing is based on Pedroni's residual cointegration test 35 Munnell (1992), p.. 193.. 36 Pesaran (2004) 37 Except for INFRA, which does not yield any plausible results without using its squared form

12h WCTR, July 11-15, 2010 - Lisbon, Portugal

10 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth;' An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert these insignificant regressors, which we treat as irrelevant, so that the model fit rises .. Furthermore, the results of the regression analysis do not change considerably, thus confirming the robustness of the model. Finally, we use the non-quadratic expressions combined with cross products in Model 3, in order to check for further misspecifications of Model 4. As we find a significant decline in the adjusted fit measure R2, we choose Model 4 for our further analysis"

Table 1 - Estimation results Specification ll[ln(BIP)] 1 2 3 4 TSLS, PCSE TSLS,PCSE TSLS,PCSE TSLS,PCSE

Period Fixed X X X X

0697755** 0576832* 0635272** 0649489** ll[ln(labour)] (0,316833) 10.316417) 10.311362) 10.310780) 0.430475** 0,,321206 0475933** 0355760* ll[ln(capital)] 10.196412) 10.228018) 10.216874) 10.181499) 0054637** 0050612** II [In(movements)2] 10.022835) (0.020900) -0021662 -0,,154330 ll[ln(movements)] 10.089194) 10.342204) -0.433538** 0211635 -0411429* 1l[ln(infra)2] 10.216920) 11.485116) (0.229000) 1,260259 ll[ln(infra)] 13.044495) 0018481** 0015469** II [In(airportcap)2] 10.007174) (0.005917) -0006778 -0022011 II [In(airportcap)] 10.021624) 10.167689) -0246746 -0204759 ll[ln(movements) "In (infra)] 10.181654) 10.255780) 0,,164587 0074800 ll[ln(airportcap) ,In (infra)] (0.136280) 10.094829) -0063717* 0,003860 -0,056850** ll[ln(airportcap) In (movements)] (0.024773) 10.016841 ) (0.021825)

F 5135239 4.439324 3,971102 5158973

R2 0440398 0451228 0,443393 0449490

R2 0350862 0336900 0,327434 0352779

Durbin Watson 1954480 1940127 2000412 1959283

CD-Test (p-value\ 0.055819 0.053575 0.053141 0.054571 • Significant at the 10% level .* Significant at the 5% level

1::!h WCTR, July 11-15, 2010 - Lisbon, Portugal

11 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth. An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert A misspecification of the model was analyzed with regard to a lower order of the trans-log specification. Using the RESET method, we also elaborate on other functional forms of the model. The functional form of the analysis cannot be rejected up to a specification of order 5. 38 We thus use Model 4 as the basis for our analysis. We find some important features of the model which highlight its credibility:

1, The model yields almost constant returns to scale, although we did not use this constraint as an explicit assumption. Hence, our results are mainly in line with other studies which use constant returns to scale as an economic assumption39 or which yield comparable results, 40 The constant returns to scale are calculated as the sum of the coefficients of labor and capital, as we use the full capital stock in the analysis.

2" In a Cobb-Douglas production function, the output elasticity of labor input should be roughly the same size as the wage share.41 The adjusted wage share in Germany for 2007 was 64,,6%, which is almost the same size as the coefficient of labor 0,649 in our model.

As shown in Table 1, we also tested for cross-sectional dependence and serial correlation, A value of about two for the Durbin Watson Statistic implies that serial correlation is not considered a relevant problem., The test for cross-sectional dependence reveals econometric evidence of cross-sectional dependence. Therefore, we use panel-corrected standard errors and covariances (PCSE) to estimate the standard errors of the estimators,,42 Because we find comparable results for all models, this method is employed for all presented models, Furthermore, we have already discussed which relevant, but exogenous instruments are included into our analysis. Given that the exogenous instruments are crucial to the validity of the estimation, we apply the overidentifying restrictions test.43 The test yields J~x~ with J = 3.6548, such that the null hypotheses of parameter exogeneity cannot be rejected, even at a level of significance of a = 10%" Thus, we find signs of exogenity of the instruments.

RESULTS

The results of the regression analysis yield insight into the various causal economic relationships that generate the economic effects of airports. On the basis of the estimates, it is possible to distinguish between those capital effects which arise from the provision of airports and the air traffic effects resulting from the connectivity provided by an airport to its influenced area. Furthermore, we can discuss the output effects generated by airport expansion"

38 P-value p = 04447 39 Esfahani/Ramirez (2003), pA47 40 OZbay (2007), p,324, 41 Phelps-Brown (1957) 42 The method is based on Beck/Katz (1995) 43 Hansen (1982)

1Zh WCTR, July 11-15, 2010 - Lisbon, Portugal

12 MAA-4-2012-Report excerpts distributed at the meeting Causal Relationships between Airport Provision, Air Traffic and Economic Growth:' An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert Effects of Airport Provision

The capital effects of airports are caused by the existence of an airport and not by its air traffic" Two lines of economic arguments justify these effects:

1" Entrepreneurs interpret the existence of an airport as a signal of economic strength" The provision of an airport, which is often supported by public agencies, reveals that such agents facilitate economic growth by providing the necessary infrastructure and institutions, Entrepreneurs try to take advantage of these positive conditions in the influenced area. Thus, they locate their business in these regions, causing positive output effects.

2" For the airports which are included into our dataset, we find a strong correlation between the basic "supply" of destinations which are relevant to business travelers and the airport capital stock. Thus, we conclude that a larger airport ensures a higher connectivity for business travelers in the airport region, compared to an airport with a smaller capital stock. Thus, the provision of airports usually provides business travelers with basic air-traffic links. This includes at least hub airports and some direct flights to important commercial centers. Thus, the mere existence of an airport may already generate positive output effects.

The effect of airport provision is analyzed by differentiating the model in terms of logarithms, with respect to the airport capital stock in logarithms" This yields the output elasticity of the provision of airport capital:

al (aln(BTp) ) = 0,03094 ')n(AIRPORTCAPt _ 6 ) - 0,05685 ')n(MOVEMENTS) > ° (Sa) n AIRPORTCAPt _ 6

~ In (MOVEMENTS) < 054421. (5b) In (AIRPORTCAPt _ 6 ) , Interpreting this result, we conclude that the output elasticity of airport capital is only positive for third-tier airports. Thus, only the provision of these airports exerts significant effects on the regional output in their respective influenced area. Furthermore, there is no need for these airports to use the infrastructure to its full capacity. These airports generate their positive output effects by their very existence and through the related supply of basic air transportation services" In addition to this, the leisure traffic at these airports exerts positive economic effects. But as these effects are not part of this study they are not considered in our results. Of course, it is highly implausible that these effects are caused by the mere existence of a huge airport with a low air traffic connectivity" As we have already elaborated above, our data does not provide us with evidence of such a situation. Thus, it is the combination of basic air traffic services and the provision of the airport which creates the positive output effects of airport capital for third-tier airports. In this context, it is important to remember that the production function used for our analysis is substitutional.. Thus, we do not have sunk costs in our model.. Although this assumption is not realistic, it is also possible to evaluate the past capital expenditure" Shifting the focus to first and second-tier airports, we conclude that the influenced areas of these airports do not benefit from the mere provision of these larger airports, although they

1Zh WGTR, July 11-15, 2010 - Lisbon, Portugal

13 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth.:' An Econometric Analysis ALLROGGEN, Florian;' MALINA, Robert benefit from the air traffic which is conducted at these airports 44 Assuming a declining marginal productivity of capital and fixed capital resources, we find that the opportunity costs of providing these airports rises as the airport capital grows, Hence, it can be concluded that the provision of first and second-tier airports becomes too expensive in terms of opportunity cost to justify the capital effects, as shown above,

Effects of Air Traffic

The model provides evidence that the economic significance of airports cannot be justified solely by the provision of an airport. Furthermore, the economic effects of airports are also caused by the air traffic at the airport. These effects are analyzed by using the first derivation of the model in logarithms, with respect to the logarithmic aircraft movements on an airport:

a (aln(Blp) ) = 0,10122 '!n(MOVEMENTS) - 0,05685 '!n(AIRPORTCAPt _ 6) > ° (6a) In MOVEMENTS

¢:::> In (MOVEMENTS) > 0,56162" (6b) In (AIRPORTCAPt _ 6 ) Based on this result, we conclude that first-tier and second-tier airports in our study yield positive output effects of enhanced traffic on the existing infrastructure" Thus, limiting the number of flights by administrative means, on an existing infrastructure at first and second tier airports, comes at an economic cost, as it reduces the connectivity of the influenced area of an airport with regard to air traffic. Furthermore, we reveal that larger airports cause positive output effects through facilitating additional air services. 1,4 ,."',.,..,.,.,"" """"""'" ,.,."",." ..,.",.,.,." ..",..",.." , .

1,3 +""""""'"''''''"''''',''''' ",,,,,,.',,,,,,,,,,,,,,,,,,,"',,,,,,,,,,,,,,,,,,,,,,,--,,,,,,,""

E ilJ a;:' 1,1 o t) -0 ~8 ~!:! 0,9 +" """"""""..",.""",,,,,,,,,,,"',,'""""'"''''''''''''''''"""""''''''''''''''''

0.'" 0,8 +""'''''' '" "" - "" ''',,'''''''''''''''''' '''-- """" '''''''' "'"''

0,7 i"""""""'-"" """""'" """" ""'''''''''' """"""'''''''''''''''' """"""""",", 0,6 "', ''''' "'"" """"""""','",,,,,,,-' ""., , " ,.. 2000 2001 2002 2003 2004 2005 2006 2007

~Business Pax , , , Pax total Figure 2 - Traffic development at German first- and second-tier airports

Traffic figures show that additional traffic at first and second tier airports is mostly traffic which is relevant to business travellers" This conclusion can be drawn from Figure 2 as the development of the sum of business travellers at German first- and second-tier airports nearly matches the development of the sum of the total passenger number at German first and second-tier airports. Thus, additional flights at first- and second-tier airports are usually

44 See next section

12h WCTR, July 11-15, 2010 - Lisbon, Portugal

14 MAA-4-2012-Report excerpts distributed at the meeting Causal Relationships between Airport Provision, Air Traffic and Economic Growth.: An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert direct connections, which are highly attractive to business travellers. These additional flights generate positive effects both on productivity and on economic output. 1 L ..-.-.- ---.------.. - ..-- - - - ..---

12 -..-.--...... -..- - ..-..------.- c. . ~

0,6 .J . 2000 2001 2002 2003 2004 2005 2006 2007

~ Bus ness Pax Pax total

Figure 3 - Traffic development at German third-tier airports

In contrast to this, additional traffic at third-tier airports is mostly leisure traffic" Figure 3 reveals that the development of the sum of business passengers at German airports is lower than the development of the sum of total passengers at German third-tier airports. Hence, additional traffic at third tier airports is generally focused on leisure traffic" As a consequence, it is not surprising that we do not find the positive effects of additional air traffic for third-tier airports. Third-tier airports offer basic air traffic services to hub airports, some direct flights to main European cities and holiday services" Although additional leisure flights at third-tier airports cause positive effects for the inhabitants of the influenced area as they reach their holiday destination at lower time costs, they do not generate economic growth as it is defined in this analysis. Combining the results of capital and air traffic effects, we find that there are airports which do not yield positive effects of airport capital or of air traffic" These are some second-tier airports that yield:

0,54421 < In (MOVEMENTS) < 0,56162. (7) In (AIRPORTCAPt _ 6 ) According to our economic analysis of airport capital and air traffic effects, we can conclude that these airports are too large to generate positive capital effects, due to the high opportunity cost of capital appropriation .. On the other hand, they do not have sufficient traffic to provide a high connectivity for business travellers. Because we have to take into account the high sunk costs of these airports, it is important for them to attract further routes which are attractive to business travellers ..

1Zh WCTR, July 11-15, 2010 - Lisbon, Portugal

15 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Air Traffic and Economic Growth;' An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert Expansion of Airports

Airport expansions often focus on removing infrastructural capacity constraints.. Thus, we have to consider simultaneous variations in aircraft movements and the airport capital stock, in order to evaluate the economic effects of airport expansions" These effects are only relevant to first and second-tier airports, as other airports usually do not have capacity bottlenecks, We use the total differential in order to simulate these effects .. This yields: dey) aIn(Y) . dAIRPORTCAP + aln (Y) • dMovements .. (8) Y aIn(AIRPORTCAP) AIRPORTCAP aIn (MOVEMENTS) Movements . . aIn(HIP) aIn (HIP) As explained above, we find aIn(AIRPORTCAP) < 0 and aIn (MOVEMENTS) > 0 for first and second- tier airports.. Hence, we conclude that positive output effects of airport expansions for first and second-tier airports are only possible, if the positive effects on air traffic movements exceed the negative opportunity costs of providing further airport infrastructure .. Thus, it is possible to calculate a critical "break-even" air traffic development for the airports that enable positive output effects of airport expansions. As we do not verify constant output elasticities in our model, it is not possible to calculate a general relationship .. Simulations show that expansions tailored to fit market needs usually generate positive expansion effects at German first and most second-tier airports. The air traffic growth rates which are necessary to ensure the positive effects of airport expansions increase, as soon as the airport capital grows and there is no commensurate growth in air traffic, Thus, positive output effects diminish for larger airports that are not used sufficiently ..

CONCLUSIONS

Based on a dataset of German airports and their corresponding influenced areas, we find that German airports generate positive effects on economic output For third-tier airports, these positive effects are caused by their very existence, due to signalling of site-specific quality and the provision of basic air services, which mainly includes connections to hub airports and some direct flights to economic centres .. By contrast, the influenced areas of first-tier and second-tier airports benefit from air traffic which facilitates potential for cost reductions and productivity growth in the economy of an influenced area. We also conclude from our analysis that necessary airport expansions tailored to market needs, yield positive output effects.. However, it is necessary to bear in mind the efficient level of capital appropriation for these expansion projects, as high opportunity costs for the airport capital stock may outweigh these positive effects,

12h WCTR, July 11-15, 2010 - Lisbon, Portugal

16 MAA-4-2012-Report excerpts distributed at the meeting Causal Relationships between Airport Provision, Air Traffic and Economic Growth:' An Econometric Analysis ALLROGGEN, Florian; MALINA, Robert REFERENCES

Aaron, H.J. (1990): Why is Infrastructure Important? Discussion. In: Munnell, AH (ed.): Is There a Shortfall in Public Capital Investment? Proceedings of a Conference held at Harwich Port Massachusetts in June 1990, pp"51-62,, Ai, C. I S.P. Cassou (1995): A Normative Analysis of Public Capital. In: Applied Economics, Vol. 27, pp, 1201-1209. Aring (2001): Hamburg Airport - Promoter of Economic Growth and Employment for the Metropolitan Region" In: Pfahler, W. (ed.): Regional Input-Output Analysis" Conceptual Issues, Airport Case Studies and Extensions" HWWA Studies No, 66" Baden-Baden, pp. 157 - 168. Aschauer, D.A. (1989): Is public expenditure productive? In: Journal of Monetary Economics, Vol, 23, No.2, pp.177-200, BBSR (Federal Institute for Research on Building, Urban Affairs and Spatial Development) (2010): Access time to German Airports" Analysis for the Institute of Transport Economics. Beck, N.I Katz (1995): What to do (and not to do) with Time-Series Cross -Section Data, In: The American Political Science Review, Vol. 89, No" 3, pp,,634-647. Boarnet, M.G. (1997): Infrastructure Services and the Productivity of Public Capital: The Case of Streets and Highways" In: National Tax Journal, Vol. 50, No.1, pp"39-57,, Born, P.R.D. I J. Lighthart (2008): How Productive is Public Capital? A Meta-Analysis. CESifo Working Paper No,,22061 CentER Discussion Paper No. 2008-10, Tilburg" Brueckner, J.K. (2003): Airline Traffic and Urban Economic Development In: Urban Studies, Vol. 40, No.8, pp" 1455-1469. Cadot, O. I L.-H. Roller I A. Stephan (2006): Contribution to Productivity or Pork Barrel? The Two Faces of Infrastructure Investment In: Journal of Public Economics, Vol. 90, pp" 1133-1153. Calderon, C.I L. Serven (2002): The Output Cost of Latin America's Infrastructure Gap" Central Bank of Chile Working Papers No. 186, Canning, D. (1999): Infrastructure's Contribution to Aggregate Output World Bank Policy Research Working Paper No. 2246. Deitmer, I. (1993): Effekte der regionalen Strukturpolitik auf Investitionen, Beschaftigung und Wachstum. Munster. Di Palma, M.I C. Mazziotta (2003): Infrastructure, Competitiveness, Growth: the Case of Italy" In: Review of Economic Conditions in Italy, pp"361-389,, ECAD (2008): Katalytische volks- und regionalwirtschaftliche Effekte des Luftverkehrs in Deutschland" Studie im Auftrag der Initiative Luftverkehr fOr Deutschland, der Arbeitsgemeinschaft Deutscher Verkehrsflughafen und des Bundesverbandes der Deutschen Fluggesellschaften" Darmstadt Esfahani, H.S. I M.T. Ramirez (2003): Institutions, Infrastructure and Economic Growth. In: Journal of Development Economics, Vol. 70, No.2, pp" 443-478. Fernald, J.G. (1999): Roads to Prosperity? Assessing the Link Between Public Capital and Productivity" In: American Economic Review, Vol. 89, No" 3, pp, 619-638, Gramlich, E.M. (1994): Infrastructure Investment: A Review Essay. In: Journal of Economic Literature, Vol. 32, pp" 1176-1196,

12'h WCTR, July 11-15, 2010 - Lisbon, Portugal

17 MAA-4-2012-Report excerpts distributed at the meeting

Causal Relationships between Airport Provision, Ai( Traffic and Economic Growth:' An Econometric Analysis ALLROGGEN, Florian, MALINA, Robert Green, R.K. (2007): Airports and Economic Development In: Real Estate Economics, Vol. 35, No, 1, pp, 91-112, Hansen, L.P. (1982): Large Sample Properties of Generalized Method of Moments Estimators. In: Econometrica, Vol. 50, No.4, pp, 1029-1054, Munnell, A.H. (1992): Policy Watch: Infrastructure Investment and Economic Growth" In: The Journal of Economic Perspectives, VoL6, NoA, pp.189-198. Ozbay, K. / D. Ozmen-Ertekin / J. Berechman (2007): Contribution of Transportation Investments to County Output In: Transport Policy, Vol. 14, pp.317-329. Pesaran, M.H. (2004): General Diagnostic Tests for Cross Section Dependence in Panels" CESifo Working Paper No, 1229/ IZA Discussion Paper No, 1240" Phelps-Brown, E.H. (1957): The Meaning of the Fitted Cobb-Douglas Function. In: Quarterly Journal of Economics, Vol. 71, pp" 546-560" Romp, W. / de Haan, J. (2007): Public Capital and Economic Growth: A Critical Survey" In: Perspektiven der Wirtschaftspolitik, Vol. 8 (Special Issue), pp" 6-52. Sanchez-Robles, B. (1998): Infrastructure Investment and Growth: Some Empirical Evidence, In: Contemporary Economic Policy, Vol. 16, pp" 98-108,

12h WCTR, July 11-15, 2010 - Lisbon, Portugal

18 MAA-4-2012-Report excerpts distributed at the meeting

Airport Runway Capacity and Economic Development:

A Dynamic Panel Data Analysis of Metropolitan Statistical Areas*

Derek Tittle **

Patrick McCarthy**

Yuxi Xiao **

Working Paper

Revised August 2010

* The research reported here was performed under contract with the Federal Aviation Administration (FAA), Contract No.,DTFAWA-09-A-80021 ..

**School ofEconomics, Georgia Institute ofTechnology. Ms. Yuxi Xiao is a Graduate Research Assistant in the School ofEconomics .. MAA-4-2012-Report excerpts distributed at the meeting

Abstract

Abstract This study analyzes the economic development impact ofairport capacity in metropolitan areas that have one commercial airport. Based upon panel data for 33 medium and large airports over a 7 year period 2001 - 2007, there is a positive relationship between the number ofrunways and real GMP, all else constant including runway length. A more detailed analysis revealed that an additional runway had differential effects (positive and negative) across the metropolitan areas .. Longer average flight delays were an important determinant ofeconomic development, decreasing gross metropolitan product 2.9% and labor productivity 1.31 %. In a probit analysis ofnew runway additions for the period 1991 - 2007, annual passenger growth rate, freight shipped per runway, and land area ofthe airport increase the likelihood ofa new runway .. This study provides new findings on the effects that airport public capital and, specifically runways, have upon MSA economic development. Yet, more research is required to better understand the linkages between airport capacity and economic development and to expand upon those factors which increase the likelihood that a MSA will add a new runway ..

Keywords airports, airport capacity, economic development, metropolitan statistical area, panel data, public capital, runways

JEL Classification 018, Rll, R4l, R53

2 MAA-4-2012-Report excerpts distributed at the meeting

I Introduction Accounting for economic growth presents a persistent difficulty as many measures exist that reflect growth and economic development. An economic approach to identifying changes in growth and development focuses on available resources and choices, Assuming that individual market participants make decisions in their own best interest, changes in the quantity and productivity of resources at the aggregate level will be consistent with economic development, growth, and economic welfare. In this study, we focus on aggregate measures ofresources, These measures include real wages, employment, and physical capita1. We adopt an aggregate production function to model economic activity within a metropolitan area using traditional factors ofproduction and their contributions in different economic sectors. Within this context, public capital or infrastructure (e.g. highways, water systems) is a factor ofproduction that contributes to economic development and growth. In contrast to most analyses, however, this study focuses upon airport infrastructure, and in particular runways, This study develops and estimates models that analyze the impact ofairport runways on economic development in metropolitan areas, Similar to other aggregate productivity studies that include public capital, runways are an input into a metropolitan area's production function and interact with labor, capital, and other factors to generate metropolitan output. An increasing number ofrunways reduces transactions costs across many margins and facilitates economic development and growth" However, runways are discrete and their use is subject to initial low marginal costs which then begin to rise as congestion sets in and ultimately become infinite as take-offs and landings near the technical limit for safe operations. Once take offs and landings reach this limit, the resource costs ofadditional throughput becomes a choke point whose sustained effects can retard economic growth. For this study, we use real gross metropolitan product, the output ofgoods and services, to measure the level ofeconomic activity in a metropolitan area, Gross output at the metropolitan level is particularly relevant since the objective is to identify the economic impact ofrunways, The strongest effects are likely to occur in the metropolitan area where the airport is located and these effects are expected to diminish with distance from the metropolitan area. 1

1 Similar to gross state and gross domestic product, gross metropolitan product does not include non-market transactions and an area's environmental profile (e,g air pollution, water quality, traffic congestion), Hence, measures ofoutput do not measure economic welfare However, assuming that market participants act in their own selfinterests, these measures

3 MAA-4-2012-Report excerpts distributed at the meeting

II Review ofLiterature Since the 1980s, there has been an increasing number ofpapers analyzing the impact ofpublic capital on economic growth and exploring the extent to which public capital affects total factor productivity and economic growth,2 Researchers use two main approaches to analyze the relationship between the stock or flow ofpublic capital and aggregate or private output: 1) an aggregate production function approach to estimate the impact ofcapital, labor, and public capital on economic growth; and 2) a cost function to estimate the effect ofpublic capital on costs ofprivate production" Costa, Richard, and Martin (1987) estimate the impact ofpublic capital on regional output at the state level using a translog production function" Defining public capital as outlays ofstate and local governments, the study finds that public capital experiences diminishing returns with respect to gross value ofproduction and the results support the inference that labor and public capital are complements. Aschauer (1989) considers the relationship between aggregate productivity and the stock and flow ofgovernment expenditures on public infrastructure over the period from 1949 - 1985. This study estimates a significant private return to public capital where a 1 percent increase in the ratio of public to private capital stocks raises total factor productivity by 0.39 percent.3 Munnell (1990a), building upon Aschauer's findings, explores whether changes in the amount ofpublic capital combine with the growth ofprivate capital and labor to explains the productivity slowdown in the 1970s,4 Assuming that services are proportional to the public sector capital stock and under constant returns to scale, Munnell finds that a 1 percent increase in public capital increases labor productivity by 0.31 to 0.39 percent for total nonmilitary public capital and core infrastructure, respectively5 Munnell (1990b) uses a translog production function approach to estimate the impact of public capital on gross state product (GSP) at the state and regional level. Since no observations on the stock ofprivate or public capital are available on a state-by-state basis, Munnell segregates the

are expected to be positively couelated with economic welfare so that an increase (decrease) in GMP is expected to increase (decrease) economic welfare 2 Aschauer, 1989, p194, 3 Aschauer, 1989, p182 4 Munnell, 1990, p4, 5 Detailed discussion is included in the appendix ofMunnell, AH, Why Productivity Growth Declined? Productivity and Public Investment, New England Economic Review, 1990(Jan./Feb): p 3-22

4 MAA-4-2012-Report excerpts distributed at the meeting

national totals based on input category.6 In this study, Munnell estimates a 0.15 elasticity on public capital, more than halfthe value found in Aschauer (1989) and Munnell (1990a). In a further analysis, Munnell analyzes the impact ofvarious components ofpublic capital on output and finds that the major impact on output from derives from highways (0 .. 06) and water and sewer systems (0.12)7. And in a regional analysis, Munnell reports uniformly positive but varying elasticities ofthe productivity ofpublic capital: 0.07 for the Northeast, 0.12 for North Central states, 0..36 for the South, and 0.08 for the West8 Eisner (1991) utilizes the same dataset as Munnell (1990b) and replicates the calculations using pooled time series, pooled cross section, and first difference regression equations to explore the disparity between the national level and state level results.. Eisner's time series analysis approach does not yield a statistically significant estimate for the elasticity ofpublic capital under the assumption ofconstant retum to scale.. However, his cross section analysis estimates the elasticity of public capital with respect to gross state product at 0.165 9 These results suggest that more public capital generates a larger gross state product. Tatom (1991) presents a theoretical argument critical ofthe existing public capital hypothesis and reviews the claims made by proponents ofthe infrastructure deficit view .. Tatom argues that most ofthe previous literature does not account for nonstationarity in the time series, ignores the trend or broken trend ofproductivity, and overlooks the impact ofchanging energy prices.. Accounting for these will reduce conventional estimates ofelasticity ofprivate capital output to public capital by 30 40 percent to 0 .. 13 percent for a 1 percent change in public capital. Holtz-Eakin (1994) argues that refined empirical methodology reconciles the differences between those who support the hypothesis that public sector capital affects the private sector output and those who do not. Estimates ofproduction function that control for unobserved, state-specific characteristics reveals no role for public capital in affecting private sector productivity. 10 Only estimates ofstate production functions that do not include such controls find substantial productivity Impacts.. II

6 Munnell, 1990b, p12 The Bureau of Labor Statistics publishes the national totals 7 Munnell, 1990b, p17. 8 Eisner, 1991, p47. 9 Eisner, 1991, p48 10 Holtz-Eakin, 1994, p 12. 11 Holtz-Eakin, 1994, p12.

5 MAA-4-2012-Report excerpts distributed at the meeting

Garcia-Mila, McGuire, and Porter (1996) presents the analysis ofthe effect ofpublic capital on gross state product in state-level production function using observations for the 48 contiguous states from 1970-1983. This study tests for random effects, fixed effects, nonstationarity, endogeneity ofthe private inputs, and measurement error, The systematic investigation leads the authors to choose the first difference with fixed state effects as the preferred specification. In the presence ofa statistically significant estimate for private capital and the absence ofa statistically significant estimate for public capital, Garcia-Mila, McGuire, and Porter conclude that only private capital impacts private output within the frame work ofan aggregate production function, This result is consistent with Holtz-Eakin (1994). A number ofrecent studies focus upon explicitly upon airports and economic development. Exploring air passenger travel and urban development, Goetz (1992) finds a positive correlation between increases in per capita passenger flows and past and future urban growth, consistent with the notion that air travel is important for economic development. Hakfoort et aL (2001) and Brueckner (2003) study the impact that airports have upon metropolitan employment. Using an input-output framework to analyze the effects on the Greater Amsterdam region from an expansion at Amsterdam's Schiphol Airport, Hakfoort et aL find a 1-1 relationship, a one job increase at Schiphol producing 1 job from indirect and induced effects" Exploring linkages between employment and air traffic in the Chicago metropolitan area, Brueckner (2003) finds that a 1% increase in passenger enplanements increases employment in service related industries 0" 1%, This has important implications for metropolitan development from airport expansions" Brueckner's results indicate that expanding Chicago O'Hare International Airport would generate 185,000 service related jobs. Green (2007) uses various measures ofairport passenger and cargo activity to analyze the linkage between airports and metropolitan growth" After controlling for various factors, Green finds passenger activity is a strong predictor ofpopulation and employment growth. The current study adds to the literature on the productivity ofpublic capital and, in particular, airports and runway capacity. Further, in focusing upon metropolitan growth and development, this study adds to the developing literature on the role ofairports in metropolitan growth and will have implications for regional, metropolitan, and local policy makers.

6 MAA-4-2012-Report excerpts distributed at the meeting

III Empirical Methodology Consider a MSA aggregate production function with two inputs, Qit = f(Lit, Kit; y), where Qit is

aggregate output, Lit is aggregate labor, and Kit is aggregate private capital for MSA i in year 1. y is the state oftechnology. Assuming standard neoclassical production theory, this framework and the underlying properties ofthe production function are sufficiently general to address a wide variety of questions, depending on one's purpose. For example, including public capital as an explicit input enables one to explore the impact that public capital has on private output (i.e. the productivity of public capital); and in a framework with more than two inputs, one can explore whether pairs of inputs are substitutes or complements in production. Our analysis adopts a commonly employed Cobb-Douglas aggregate production function in order to motivate the empirical modeL A Cobb-Douglas production function is multiplicative in inputs and generates a double log empirical specification,12 Including public capital Rit as a factor of production gives the following Cobb-Douglas specification for metropolitan output:

GMP = A'Lr:2K~3R~4e£it It 1 It It It ' where Ai is a constant (reflects fixed effects), 0,2, 0,3, and 0,4 are parameters to be estimated, and Eit is a stochastic term. Taking the logarithm ofboth sides gives

where o'i = In (AD is a fixed effect for cross section i, Yeart is a trend variable which reflects technological and other unobserved factors that change over time, and Eit is an enor term"

12 The Cobb-Douglas and similar empirical forms have a number ofeconometric problems including endogeneity (GMPit depends on Lib Kilo and Git and each ofthe inputs depends on GMPit as well as the other inputs), multicollinearity among the inputs, and heteroskedasticity (non-constant variance) The source ofsome ofthese problems reflect decisions made at the microeconomic leveL Because plivate or public managers face similar economic environments and resource constraints, they tend to make similar marginal allocations ofproductive inputs" At the individual level, the effects of these decisions are evident as labor and capital decision move together with (private and public) output over time. At the MSA level, gross metropolitan product reflects the cumulative decisions on aggregate labor and capital in the private and public sectors, To the extent possible, this analysis will address these issues

7 MAA-4-2012-Report excerpts distributed at the meeting

IV Data Sources and Descriptive Analysis We develop a panel of35 MSA's with only one commercial airport, identified by the FAA as a medium or large huhI3 For the panel of3.5 MSA's and corresponding airports, Table 1 provides airport and metropolitan descriptive statistics for three groups, the Full Sample, Over Airports (summing across years), and Over Years (summing across airport cross sections)" As seen in Table 1, depending upon the series, the availability ofsome variables ranges from 21 years for some variables to 18 years and 6 years for others.. Entries in the table that list the current year gives information as of 2009 (e.g. airport land area).. For each group, the overall variable means remain the same but the variances differ. Focusing upon airport characteristics for the sample during the period ofanalysis, there is an average of 10.5 thousand domestic annual departures and 4,182 annual international departures.. An annual average of 8.2 million passengers flew on non-stop unlinked segments per airport and airlines carTied, on average, 127 million pounds offreight In 2009, on average, the airports covered over .5,.500 acres on average and there is an average of 3..4 runways per airport 19 airports were large hubs and 16 were medium sized hubs. 14 For the full panel of35 MSAs during the 18-21 year period, the average annual population in the MSAs is just over 2 million persons with an annual average of 1.22 million workers, ofwhich 1.03 million are wage and salary workers.. The average annual real-wage-and-salary disbursement per worker is $19,2.56 and annual average real per capita income is $16,875. The average unemployment rate over all MSAs and observed years is 4.8% and there are just fewer than .51,000 establishments on average per year. Starting in 2001, the Bureau ofEconomic Analysis began reporting gross metropolitan product For the 7-year period from 2001 to 2007, Table 1 reports that annual real GMP averaged $49 .. 0 billion for the full panel. This reflects an average annual per capita GMP of$41,995. On average, for the sample period, annual GMP represents 40% ofgross state product. When disaggregated by type ofactivity, the Finance sector accounts for almost halfofMSA

13 Isolating the effect ofadditional runways on economic development becomes more difficult when a MSA supports multiple airports. In order to avoid this potential problem, this analysis includes only MSAs with a single commercial airport. 14 Large hubs are defined as airports with "I % ofDS Enplanements or more"; medium hubs are defined as "less than 1% but more than 1/4%"

8 MAA-4-2012-Report excerpts distributed at the meeting

Table 1 Panel Data Descriptive Statistics

Standard Deviation Variable Mean Total Across Across Sample Years Airports Airport, Domestic Departures* 105,306 70,301 298,610 113,222 Airport, Freight Shipped, Non-Stop Segments (million pounds)* 126.7 3621 8984 699 .. 0 Airport, International Departures* 4182 5396 22089 8052 Airport, Land Area ofAirport (acres)**** 5589 5804 26977 0 Airport, Large = I, Medium = 2** 1.5 05 23 0 Airport, Number ofDiverted Airport Landings** 171 128 494 232 Airport, Number ofDiverted Airport Take-OfIS** 1261 1531 4902 3962 Airport, Number ofRunways* 3A ]J 5.8 0.9 Airport, Passengers, Non-Stop Unlinked Segments (million)* 82 60 262 81 Metropolitan Area, Employment** 1,220,311 703,863 3,170,836 872,440 Metropolitan Area, Number ofEstablishments** 50,925 31,058 141,243 30,514 Metropolitan Area, Population (persons)** 2,020,215 1,170,432 5,326,622 1,058,136 Metropolitan Area, Real GDP Quantity Index (100=2001)*** 1084 93 237 40.7 Metropolitan Area, Real GMP ($ million)*** 49,011 28,464 119,053 5,999 Metropolitan Area, Real GMP - Education and Health ($ million)*** 6,698 5,598 24,850 4,179 Metropolitan Area, Real GMP - Finance ($ million)*** 20,843 16,210 73,176 7,289 Metropolitan Area, Real GMP - Government ($ million)*** 9,292 4,922 22,804 955 Metropolitan Area, Real GMP - ICT ($ million)*** 5,094 5,924 24,028 4,542 Metropolitan Area, Real GMP - Leisure and Hospitality ($ million)*** 3,729 2,549 11,584 1,244 Metropolitan Area, Real GMP - Private Goods ($ million)*** 15,545 10,334 42,686 5,451 Metropolitan Area, Real GMP - Private Services ($ million)*** 41,757 25,031 102,483 16,201 Metropolitan Area, Real GMP - Profession and Business ($ million)*** 10,917 9,814 40,492 8,230 Metropolitan Area, Real GMP - Transportation and Utilities ($ million)*** 3,645 3,412 13,788 2,617 Metropolitan Area, Real GMP as % ofReal GSP*** 04 04 1.7 02 Metropolitan Area, Real GMP per Capita ($)*** 41,995 6,234 27,866 8,166 Mellopolitan Area, Real Wage and Salary Income per Worker (1982-84 $)** 19,256 2,343 8,454 8,200 Metropolitan Area, Real per Capita Income (1982-84=100, $)** 16,875 2,347 7,720 9,276 Metropolitan Area, Unemployment Rate* 48 ]J 3A 80 Metropolitan Area, Wage and Salary Employment (persons)** 1,026,457 611,375 2,773,262 650,174 State, Higher Education Enrollment** 479,896 427,708 1,961,089 279,628 State, Real GSP (Chained 1982-1984 $million)** 179,023 156,203 700,331 184,590

1990 - 2007,18 years and 35 airports, 630 observations ** 1987 - 2007,21 years and 35 airports, 735 observations *** 2001 - 2006, 6 years and 35 airports, 238 observations **** Data missing for Jacksonville, FL airport

9 MAA-4-2012-Report excerpts distributed at the meeting

GMP, followed by the Profession and Business, Government, and Education and Health sectors. Private goods production is about one-third the contribution ofprivate services to GMP .. Observations by sub-groups display greater heterogeneity between the cross section units than when measuring across time, as is often the case with panel data.. With few exceptions, the standard deviation for all variables is greater, and at times considerably greater, across airports and MSAs than across the years. For example, over the entire sample, non-stop unlinked segments accounted for an annual average of 8..2 million passengers with a standard deviation of 6 million passengers.. When summed over airports so that only the year varies, the standard deviation is 8.1 million passengers, which generally reflects long term passenger trends .. Summing over time to measure the differences between MSA's, however, gives a standard deviation of26..2 million passengers, which reflects the size distribution ofsampled airports .. The two primary exceptions to this variance pattem are real per capita income and the unemployment rate. This deviation from the pattem is to be expected since dividing income by population adjusts for size differences across MSAs and, as a result, real per capita income across MSAs is less heterogeneous than across time, with an observed standard deviation of $9,276 versus $7,720. And, because economic cycles tend to affect all geographic areas to a similar degree, unemployment rates exhibit less heterogeneity across MSAs than across time, with an observed standard deviation of8% over time versus 3.4% over MSAs .. Table 2 identifies the sampled airports in the MSA analysis and the airport's hub status .. There are 19 large hub airports and 16 medium hub airports in the panel data set.

IV. J Empirical Considerations For this analysis, metropolitan area employment is our measure ofaggregate labor.. Because measures of MSA capital are not available, we use the number ofMSA establishments as a proxy for the level of private capital. 15 This analysis includes three variables to capture the effects ofairport public capital. First, the relationship between changes in the number ofrunways and economic development is ofprimary interest. To explore this, we measure the total number ofrunways at an airport. It is

15 In preliminary analyses, we also included population density to capture potential differences in invested capital (as well as other sources ofheterogeneity) across metropolitan areas. This variable added little to the final model's explanatory power ..

10 MAA-4-2012-Report excerpts distributed at the meeting

Table 2 Hub Airports for MSA Analysis

Large Hub Medium Hub

Hartsfield-Jackson Atlanta International, ATL Albuquerque International, ABQ General Edward Lawrence Logan, BaS Austin-Bergstrom International, ADS Baltimore-Washington International, BWI Nashville International, BNA Charlotte/Douglas International, CLT Cleveland-Hopkins International, CLE CincinnatilNorthern Kentucky, CVG Port Columbus International, CMH Denver International, DEN Indianapolis International, IND Detroit Metro Wayne, DTW Jacksonville International, JAX Honolulu International, HNL Kansas City, International, MCI McCanan International, LAS Memphis International, MEM Orlando International, MCO General Mitchell International, MKE Minneapolis-St Paul International, MSP New Orleans International, MSY Philadelphia International, PHL Portland International, PDX Phoenix Sky Harbor International, PHX Raleigh-Durham International, RDU Pittsburgh International, PIT Southwest Florida International, RSW San Diego International, SAN San Antonio International, SAT Seattle-Tacoma International, SEA Sacramento Metro, SMF Salt Lake City International, SLC Lambert-St Louis International, STL Tampa International, TPA

expected that adding an additional runway will increase GMP, all else constant. Second, we include maximum runway length" The longer the runway, the larger the plane a runway can accommodate and this enables the airport to serve more passengers and ship more freight, all else constant. Recognizing that airports can substitute between adding runways to increase the number offlights that can land and lengthening the runways to allow aircraft with higher passenger capacities to land in order to increase total passengers moved, we include a cross product variable between the number of runways and maximum runway length. And third, systemic runway congestion reduces the quality of runways which potentially constrains the extent to which a MSA can sustain economic development. To capture this effect, we include in the model the average flight delay, in minutes. All else constant, an increase in average flight delays is expected to decrease GMPit.

11 MAA-4-2012-Report excerpts distributed at the meeting

In addition to airport public capital, we include two variables as proxies for quantity and productivity ofhighway infrastructure., 16 An increase in the number offreeway and arterial lane miles is expected to reduce market transactions costs and increase metropolitan output. Second, to reflect the quality ofhighway travel, the model includes a road congestion index for each metropolitan area. All else constant, the higher the index the more congested the roads, the more damage on the roads, and the higher the resource costs ofeconomic activity" At the same time, all else constant, congested roads imply a more economically thriving environment which enhances metropolitan output, an effect which is expected to dominate negative effect on output from increased resources devoted to non productive travel during peak periods" Gross metropolitan product measures metropolitan market activity. The Bureau ofLabor Statistics provides data on real gross metropolitan product for the years 2001 - 2007. For this part of the analysis, we omit all observations before 2001 and after 2007. 17 Additionally, because the Bureau ofLabor Statistics does not provide observations for Jacksonville, FL MSA and runway information was missing for Charlotte/Douglas International Airport, we omit Jacksonville (lAX) and Charlotte/Douglas (CLT) International Airports ..

V Metropolitan Statistical Area Estimation Results Vi Gross Metropolitan Product Results Equation (2) identifies the base model for the gross metropolitan product analysis.. Substituting the specific empirical measures for Lit, Kit, and Rit yields the estimating equation:

(2) In (GMPit) = Iai + 0,2 In (GMPi,t-l) + 0,3 In (Employmenti,t_l) +

0,4 In (Establishmentsi,t_l) + 0,5 In (Number ofRunwaysi,t_l) + 0,6 In (Maximum Runway Lengthi,t) + 0,7 In (Number ofRunwaysi,t_1 * Maximum Runway Lengthi,t) + 0,8 (Average Flight DelaYt_l) + 0,9 (Large Hubi,t) + 0,10 In (Lane Milesi,t-d +

all (Road Congestion Indexi,t) + Lj~12 aj Regional Dummy Variablej+

16 We explored alternative measures ofhighway public capital. Lane miles and the congestion index led to better overall fits 17 Gross State Product for a more extended time is available from several sources, However, states possess a large geographic area compared to the MSA which weakens the expected effect, all else constant, ofadditional runway capacity in a given MSA,

12 MAA-4-2012-Report excerpts distributed at the meeting

a 191n (Real Gross Domestic Productt) + Cit (i = 1, ...,33; t = 2002 - 2007)

Equation (2) also includes lagged gross metropolitan product, GMPt-1as an additional explanatory variable to represent a dynamic version ofthe model and to account for serial correlation in the error terms,IS Lagged values for Employment, Establishments, the Number of Runways, Average Flight Delay, and Arterial Streets Daily VMT were used as instmmenta1 variables in order to reduce, ifnot eliminate, concems with endogeneity that often characterize aggregate models. Real GDP serves as a measure ofoverall economic activity and operates to show the impact ofthe period preceding the recession" In addition, standard erTOrs for all parameter estimates are robust to departures from a constant variance assumption. 19 As a proxy for human capital investment, state research and teaching budgets were found to have little explanatory power in exploratory analyses" And extreme collinearity problems precluded a full fixed effects specification" However, by including region variables, it was possible to, at least partially, account for cross section heterogeneity and differences across MSAs from unobserved or omitted variables. In particular, the model included seven FAA region variables: Eastem, Great Lakes, New England, Northwest Mountain, Southem, Southwest, and Westem Pacific" The reference (omitted) region is FAA's Central region.20 Table 4 reports the estimation results and the model fits the data well. The adjusted R2 is .9988, which indicates that the model explains 99.88% ofthe variance in metropolitan GMPit?l As expected, lagged GMPit is a strong determinant ofcurrent GMPit. Increases in lagged employment increase GMPit. All else constant, a 1% increase in lagged employment increases real GMPit in the current period 3.0% or $1.5 billion on average all else constant22 As a proxy for private

18 Exploratory analyses found that serial conelation coefficients ranged from a low oL33 to a high of.99, Theoretically, assuming that gross metropolitan product only partially adjusts to changes in the explanatory variables in the given time period motivates a dynamic version ofequation (1) (Ramanathan, 3rd Edition, 1995) 19 Specifically, the standard eI1'OlS are calculated from a heteroskedastic consistent covariance matrix (Greene, 1997) 20 States included in the regions are: Eastern: DE, MD, NJ, NY, VA, WV; New England: CT, MA, ME, NH, VT; Great Lakes: lN, IL, MI, MN, ND, OH, SD, WI; Southern: AL, FL, GA, KY, MS, NC, SC, TN; Southwest: AR, LA, NM, OK, TX; Northwest Mountain: CO, ID, MT, OR, UT, WA, WY; Western Pacific: AZ, CA, HI, NV; and Central: lA, KS, MO, NE. 21 The reported model provides the best overall fit and the reported estimates were robust to alternative specifications, In other analyses, we substituted two variables, log(current number ofrunways) and a dummy variable for added runway, for log (number ofrunways).I" Also, we estimated various models with up to four years of either the number ofrunways or the dummy variable for the presence ofa new runway, For these alternative specifications, the estimated values of included variables were robust and the lagged values did not show consistent significance and, with GDP in the model, was not significant 22 From Table 1, the sample average real GMP is $49,0 billion, 1A% ofwhich is $695 million,

13 MAA-4-2012-Report excerpts distributed at the meeting

capital, Establishments has the expected positive sign and is statistically significant, with a 1% increase in Establishments raising real GMPit 4 .. 8% or $2,4 billion on average, all else constant23 As expected, the log ofreal GDPit was positive and significant. Associated with a 1 % increase in real GDPit was a 20% increase in real GMPit.24 The variables ofmost interest for this analysis are the Number ofRunways, Maximum Runway Length, the cross product ofthose two variables, and Average Flight Delay. Based upon these results adding a new runway increases real GMP as long as the maximum runway is less than 9958 feet, a result which was significant at the 0 .. 01 level for the number ofrunways and its cross product 25 In addition, and as expected, extending the length ofrunways benefits a metropolitan area's economic development, a result that is statistically significant at the 0.01. While not addressed here, the negative and significant interaction term raises an interesting question on what factors determine whether an airport should increase capacity by extending existing runways (assuming that not all runways have maximum length) or by adding runways. Also consistent with expectations, Average Flight Delay has a negative impact upon economic development All else constant, the results in Table 4 indicate that a 1% increase in average flight delays decreases annual real GMP by 2.9% or $1 ..5 billion on average. Additional results from Table 4 indicate that metropolitan areas with a large hub airport experience, on average, a $934 million (1.9%) benefit per year. As proxies for the quantity and productivity ofmetropolitan highway infrastructure, the quantity oflane-miles and the road congestion index have the expected signs, and are statistically significant at the .03 and .. 01 levels, respectively. The sign and strength ofstatistical significance ofthe road congestion index indicates that the economic benefits flowing from a thriving community more than offsets one ofthe major externalities in metropolitan areas, highway congestion. Also, relative to the FAA Central and all other regions, MSA GMP in the Eastern and New England regions was $790 million and $1,44 billion higher, all else constant And, notwithstanding

23Preliminary analyses found that a dummy variable for the 9-11-2001 terrorist attack had no appreciable effect on GMP

24 In order to explore the potential for reverse causality, we regressed the log ofreal GDPjt against the log ofreal GMPjt The estimated R2 was 09 and, more importantly, the estimated coefficient was not significant at any reasonable level (0.244 p-value). Given these results, reverse causality does not appear to be a significant issue 25 We added a quadratic direct and interaction term for the number ofrunways to explore non-linear effects and whether these would affect the marginal impact ofan additional runway .. These analyses produced insignificant effects for the quadratic terms

14 MAA-4-2012-Report excerpts distributed at the meeting

Table 4 Gross Metropolitan Product Estimation Results 2002-2007

Derendent Variable: In (Gl\APit) Afpox Afpox Variable Estirrate StdElT Pr>ltl

C<::Jrntant -3.. 54254 0.704 <.em1 In (Gl\AP)t-l 0.2 In(Establishrrents)t-l 0.0478 0.022 o.m16 In(Lane Mles)t-l 0.02If7 0.01 o.m25 In (RaidO:xlgestion ~x) 0.0878 0.018 <.em1 In (Nurrber of Runways)t-l 0.9982 0.278 0.s DeClll:l:':stic Prcxfuct) 0.1999 0.059 0.cr09 In (Average Flight Delay)t-l -0.0292 0.008 o.cx:m In (!MDcirrumRunwayLength) 0.2ffi3 0.049 <.em1 In(NuniJer ofRunwaYSit-l * JVbxim.nnRunway Length) -0.1086 o.m 0.

Hub Size (l iflarge hub; 0 otherwise) 0.019 o. 0.0021 EasternRegion 0.016 o. 0.ems Cleat Lakes Region -0.0072 o. 0.2397 NewEngland Region 0.0289 0.01l 0.0088 Northwest M:Juntain Region 0.0038 0.008 0.6347 SouthernRegion 0 .. 0031 0.007 0.6436 Southwest Region -0.0143 0.em 0.1257 Western Pacific Region 0.0188 o.em 0.0404

# observa1:i.orn 198 2 Aqjusted R - 0.9988

Notes 7 years, 35 airports = 245 observations l\1issing data on JAX, CLT=>224observations lose 32observations due to lagging => 198 observations Using current term; for lerrp and lest had little inpact on the results otherthan strongerrejections ofthe null inllDst cases

Authors' calculations The model does not include Jacksonville, FL and Charlotte, NC due to the absence ofdata on some variables. ThiIty-three 2001 observations were not included due to a one-period lag. Standard enors are heteroskedastic consistent covariance matrix (hccm) standard enors ..

15 MAA-4-2012-Report excerpts distributed at the meeting

the relatively short time span for this analysis, GMP increased an average of0.24% or $117 million per year,

V], i A Further Analysis afRunway Effects Table 5 reports estimation results for a model that replaces the Number ofRunways variable with a set ofdummy variables associated with the airports that added an additional runway, This specification enables

Table 5 GMP Estimation Results, 2002 - 2007 Airport Specific Parameter Estimates

Dependent Variable: In (GMPit)

Approx Approx

Variable Estimate Std En PI' > It 1

Atlantat-l -0,,0055 0,,007 0..4155 Bostont-l 0,,0242 0,,006 0.0002 Clevelandt_l -0,,0017 0,,009 0.,8505 Cincinnatk-l -0,,0324 0,,008 <.0001 DenVert_1 -0,,0261 0,,011 0.,0188 Detroitt_l -0,,0268 0,,011 0,,0143 Orlandot-l 0,,0201 0,,014 0,,1450 Minneapolist-l -0,,0001 0,,006 0.,9868 St. Louist-l 0,,0001 0,,006 0,,9846

# observations 198 2 Adjusted R - 0,,9986

Authors' calculations, Except for (Additional RunwaY)t_h the other variables in this model are robust relative to those reported in Table 4, See note below Table 4 for sample information us to determine whether an additional runway reduced GMPit for all MSAs or whether there were differential effects across the MSAs" Because the results for the other variables are qualitatively similar to those reported in Table 4, we present only the airport specific variables in Table 5. 9 airports added runways during the 2001 - 2007 period and the results in Table 5 indicate that the effect ofan additional runway was not uniform across airports. Although the average effect

16 MAA-4-2012-Report excerpts distributed at the meeting

in Table 4 indicated that adding an additional runway decreased GMPit, all else constant, the more detailed results in Table 5 indicate that the effect on GMPit was specific to the airport and varied from a significant negative effect to no effect to a significant positive effect. Orlando and Boston experienced similar positive GMPit effects, amounting to a 2 .. 01 % and 2.42% increase in GMP from an additional runway. The additional runway in Atlanta, Cleveland, Minneapolis, and St. Louis, on the other hand, had neither an appreciable positive nor negative effect on gross metropolitan product In each ofthese cases, we could not rej ect the null hypothesis, at any reasonable level ofsignificance, that the additional runway substantively affected GMP. For Cincinnati, Denver, and Detroit, on the other hand, the effect ofthe additional runway was negative and statistically significant and whose effect ranged between -3 ..2% to -2.6%. Excepting Boston, the absence ofan effect or the weaker positive effect ofan additional runway was not sufficient to offset the estimated negative sign reported in Table 4 .. These results are important for their suggestion that the addition ofa runway per se may have unintended consequences whose net effect may hinder rather than spur economic development. The results presented in Tables 4 and 5 raise interesting questions on what specific factors are most important in determining whether investing in an additional airport will generate net costs or net benefits to the metropolitan area.

VI ii A Probit Analysis ojincreased Capacity In Table 6, we use probit analysis to explore what factors are associated with increased airport landing capacity .. For this analysis, our dependent variable for each airport-year equals 0 ifno additional runways were added in the 1991 - 2007 period and equals 1 ifadditional runways were added. 26 The likelihood ratio statistic strongly rejects the null hypothesis that all estimated coefficients equal 0 and the results are generally consistent with expectations.. There has been an increasing trend toward more runways throughout the sample period. And additional runways are more likely to exist in the Great Lakes, Southern, and Western Pacific Regions relative to other parts ofthe country.

26 Denver, for example, added a runway in 2003 so that the dependent variable for Denver equals 0 for 1991 - 2002, a period ofno new capacity and 1 in 2003 through 2007 when new capacity was available

17 MAA-4-2012-Report excerpts distributed at the meeting

Table 6 Probit Analysis ofRunway Additions 1991 - 2007

Approx Approx Variable Estimate Std Err Pr> ItI

Constant -221.8 31.72 <.,0001 Passengers, Annual Growth (%) 0..4562 0,,704 <.0001 Passengers per Runway -0,.2628 0.,052 <0001 Freight Shipped per Runway 0,0060 0.,002 0,0052 Airport Area (square miles) 0,,0235 0,,007 0,,0006 Great Lakes Region 0,,5086 0 .. 189 0,,0070 Southern Region 0..2358 01703 0 .. 1661 Western Pacific Region 0..3925 0..222 0,,0772 Year 0,.1105 0,,016 <.0001

# observations 561 Log-likelihood at Convergence 229,,79 Log-likelihood at Intercept Only 309,.57 Likelihood Ratio Statistic 159,,55 2 X 05 critical value (9) 16.92

An increase in the rate ofgrowth ofpassenger travel and freight shipped per runway increases the probability ofhaving more runways, all else constant. On the other hand, growth in passengers per runway has a negative and significant effect upon additional runway capacity.. This suggests that, holding constant the growth ofair passenger traffic, increasing passengers per runway more efficiently uses existing capacity and reduces the need for additional runways, These results, however, are exploratory and more research is required to fully understand the relationships that exist between runway capacity and the various runway demands,27 Having the space to grow is also an important determinant ofadditional runway capacity. Airports located on larger parcels ofland are more likely to have additional runways, The metropolitan unemployment rate reflects the economic environment and its positive sign suggests that

27 A number ofpassenger and freight variables (e,g, passengers, freight shipped, passengers per runway, fieight per runway, domestic departures) were included in preliminary estimations and consistently, as reported in Table 6, there appeared trade-offs in that all signs were not uniformly positive, reinforcing the need to better understand the underlying relationships,

18 MAA-4-2012-Report excerpts distributed at the meeting

major expansions at airports, including runways, have job-related economic benefits for the metropolitan area.

V2 Average Product ofLabor Estimates Table 4 reported the results ofa dynamic metropolitan production function that included labor, a proxy for private capital, and various measures ofairport public capitaL In addition to determining the importance ofthese variables to real GMPit, it is also useful to analyze whether the same variables are important determinants ofa metropolitan area's labor productivity, output per labor, and defined here as the (Real GMPit/Employmentit)., Table 7 reports the estimation results for a dynamic metropolitan labor productivity model. Overall, the data fit the model well, explaining 99.7% ofthe variance in the dependent variable and the results are generally consistent with expectations. Past labor productivity is a stronger predictor ofCUITent productivity and, all else constant, increases in employment decrease labor productivity which is consistent with profit maximizing behavior,28 And to the extent that the number ofestablishments is a proxy for private capital, the positive and statistically significant coefficient is consistent with expectations that increases in private capital, all else constant, increases labor productivity. Lane miles and the road congestion index have no impact on labor productivity" Also, during the sample period, labor productivity in these single airport metropolitan areas decreased, on average, ,,35%, Turning to the airport related variables, the results in Table 7 indicate that neither an additional runway, nor runway length, has a direct effect on labor productivity" However, consistent with expectations, Average Flight Delay significantly reduces labor productivity" All else constant, a 1% increase in average delay reduces labor productivity 1.31%. Given an average product oflabor equal to $78, 557, average delays reduce productivity $1,029. Further, labor productivity in metropolitan areas with large hubs, relative to MSAs with medium hub airports, significantly increases (1.12% or $885), suggesting that at least part ofthe increased

28 Profit maximization requires that firms hire labor up to the point where the revenue generated from the marginal product ofthe last laborer hired just equals the resources expended to hire the individuaL When this occurs, employers are effectively on the downward sloping portion oftheir labor demand curves and in this area the marginal product of labor and average product oflabor are falling with increases in employment.

19 MAA-4-2012-Report excerpts distributed at the meeting

Table 7 Metropolitan Average Product ofLabor Estimation Results 2002 - 2007

Dependent Variable: GMPit/Employmentit Approx Approx Variable Estimate Std EIT Pr> ItI

Constant 6..724 L699 0..0001 In (APL)t-l 0 .. 9535 0..021 <0001 In (Employment)t_l -0.. 0320 0.018 0.0780 In (Establishments)t_l 0 .. 0423 0..018 0.0188 In (Lane MileskI -0..0014 0 .. 008 0 .. 8549 In (Road Congestion Index 0..0095 0.016 0 .. 5430 (New RunwaY)t_l -0..0034 0..004 0 .. 3784 In (Average Flight DelaY)t-l -0.. 0131 0 .. 007 0..0738 In (Maximum Runway Length) 0 .. 0150 0 .. 017 0..3842 Hub Size (1 iflarge hub; 0 otherwise) 0..0112 0..005 0.0185 Eastern Region 0..0098 0.005 0 .. 0626 Great Lakes Region 0.0054 0 .. 005 0 .. 3136 New England Region 0.0202 0 .. 009 0.0268 Northwest Mountain Region 0..0010 0.006 0.1015 Southern Region 0.0080 0..005 0.1360 Southwest Region 0.0028 0..009 0.7465 Western Pacific Region 0 .. 0140 0 .. 007 0 .. 0415 Year -0..0035 0.001 <.0001

# observations 198 Adjusted R2 0 .. 9977 Authors' calculations The model does not include Jacksonville, FL and Charlotte, NC due to the absence ofdata on some variables. 2001 data were not included due to a one-period lag.

GMPit associated with a larger scale ofairport activities at large hubs occurs through increased worker productivity. To explore airport specific results, the model in Table 7 was estimated where (New RunwaY)t_

1 was replaced by a set ofdummy variables associate with the specific airport that added a new runway. Similar, although not identical to the results for GMPit, adding a new runway reduced labor

20 MAA-4-2012-Report excerpts distributed at the meeting

productivity 0 .. 94%, 1.8%, and 1A% in Atlanta, Cincinnati, and Denver, respectively. The new runway increased labor productivity in Boston 1.. 6%.

VI Additional Results Based upon a metropolitan production function framework, the primary focus ofprevious sections has been on the effect that the number ofrunways has had upon metropolitan output. There are other measures ofeconomic development and this section summarizes additional estimation results that explore the extent to which airport runway capacity affects alternative measures of development. Specifically, we focus four attributes ofeconomic development: Wage and Salary Compensation, Urban Size, Population Density, and Delay to the Peak Period Traveler.. Panel data for this analysis include the same cross sections as in previous analyses but extends the sample back to 1992. Table 8 presents the two-way fixed effects dynamic model estimation results.. In addition to a lagged dependent variable, the model includes two explanatory variables,

Table 8 Other Economic Development Measures 1992 - 2007

Dependent Variable Wages and Salary Population Delay to Peak Compensation Urban Size Density Period Traveler Estimate p-value Estimate p-value Estimate p-value Estimate p-value Independent Variable Wage and Salary CompensatiOnt-1 Urban Sizet-l Delay to Peak Period Travelen-l Population DensitYt-l RunwaYt-l Unemployment Rate

# observations R2

Authors' calculations .. Two way fixed effects models and all standard enors are heteroskedastic consistent covariance matrix standard enors All variables except RunwaYt_l are in logarithms

21 MAA-4-2012-Report excerpts distributed at the meeting

the number ofrunways in the previous period and the unemployment rate. The model fits the data well with all R2s over 0.97. From these exploratory results, the unemployment rate does not have an impact on the spatial form ofa city .. We cannot reject the null hypothesis that its coefficient in the Urban Size and Population Density equations equals 0 .. However, the unemployment rate does affect the economic character ofa metropolitan area.. The coefficient for unemployment is statistically significant in the Wage and Salary Compensation and Delay to Peak Period Travel equations .. In each ofthese cases, the coefficient for Unemployment is statistically significant and has the expected sign, with increasing unemployment rates decreasing wages and compensation and decreasing travel delays during peak periods .. Turning to the number ofrunways, Table 8 presents mixed results.. Increasing the number of runways has no impact upon labor compensation or upon urban size.. But it does have an effect on population density and traveler delays during the peak period.. All else constant, an additional runway reduces population density 2.5% and peak period delays over 9%.. These results are intriguing and reinforce the notion that runway capacity has a number ofdirect and indirect effects upon metropolitan areas that are not well understood ..

Vl.l Atlanta Hartsfield-Jackson International Airport In comparison with other MSAs with only one airport in this analysis, the scale ofoperations at Atlanta's Hartsfield-Jackson International Airport (ATL) is significantly higher. Because ofthis, in preliminary estimations, we included a number ofAtlanta specific interaction variables to test whether there was a differential airport public capital effect associated with ATL. Consistently, these interaction effects were not determining factors at any reasonable level ofstatistical significance. Also, in Section y'1.i, Table 5, we saw that the new runway at Hartsfield-Jackson had no appreciable impact upon gross metropolitan product Here, we can ask a similar question and explore whether there are differential effects when considering other alternative measures ofeconomic development. The results in Table 8 identified an effect associated with more runways.. Table 9 re-estimates the model in Table 8 but adds a new variable, Atlanta*RunwaYt_l , that interacts Atlanta with (the logarithm of) RunwaYt_l .. In Table 9, the results for RunwaYSt_l are similar to its effect in Table 8, i.e .. no impact upon compensation and urban size but a decreasing effect upon population density and peak period delays. Increasing the number ofrunways at Hartsfield-Jackson has no differential impact upon urban size or peak period traveler delays .. But there is a differential effect on Wage and

22 MAA-4-2012-Report excerpts distributed at the meeting

Salary Compensation and Population Density.. In particular, increasing the number runways at Hartsfield-Jackson increases compensation and the effect is sufficiently large (0.0222) that it which more than offsets the general effect ofrunways (-0.0061), although this latter effect was

Table 9 Other Economic Development Meas ures 1992 - 2007

Dependent Variable Wages and Salary Population Delay to Peak Compemation Urban Size Density Period Traveler Estimate p-value Estimate p-value Estimate p-value Estimate p-value Independent Variable RunwaYt-l -00061 02370 00112 04271 -00265 0.0372 -00955 0.0353 Atlanta*Runwayt-l 00222 01037 -00054 08205 0.1209 <0001 -00165 0.7929

0.9971 0 .. 9991 09967 09752

Authors' calculations. Two way fixed effects models and all standard errors are heteroskedastic consistent covariance matrix standard errors. Other variables in the model are the same as those in Table 8. All variables except RunwaYt_1 are in logarithms not statistically significant. In addition, increasing the number ofrunways increases population density in Atlanta and the effect again more than offsets the general decreasing impact upon population density, 0 .. 1209 versus -0 .. 0265. To the extent that larger airports and more runways reflect and, to a greater or lesser degree generate, more economic activities and metropolitan travel, this result suggests that the net impact may be more urbanization offirms and households ..

VII Concluding Considerations This study explored the economic impact that additional runway capacity has upon a metropolitan growth and economic development. In order to better establish the link between economic development and runway capacity, the sample for this study included MSAs with only one medium or large hub airport, Depending upon the specific analysis the sample period was 2001 2007 or a longer period from 1992 - 2007. Based upon a metropolitan production function framework, a panel data analysis of33 airports over the 7 year period 2001 - 2007 found that that adding a new runway increased annual gross metropolitan product as long as the maximum length ofthe runway present is not longer than

23 MAA-4-2012-Report excerpts distributed at the meeting

9,900 feet and had no effect on labor productivity. Average flight delays were an important determinant ofeconomic development, decreasing gross metropolitan product as well as labor productivity, decreasing GMP by 2.9% ($1.5 billion) and labor productivity by 1,,31 % ($1,029) on average. In addition, increasing maximum runway length increased GMP. A more detailed analysis revealed that an additional runway had differential effects.. In particular, a new runway increased gross metropolitan product in Boston and Orlando; had no appreciable effect in Atlanta, Cleveland, Minneapolis, and St.. Louis; and decreased gross metropolitan product in Cincinnati, Denver, and Detroit. And in an analysis oflabor productivity, adding a new runway significantly increased productivity in Boston but reduced productivity in Atlanta, Cincinnati, and Denver.. From a probit analysis ofnew runway additions for the period 1991 - 2007, annual passenger growth rate, freight shipped per runway, and land area ofthe airport increase the likelihood ofa new runway; and given the passenger growth rate, passengers per runway reduce the likelihood, suggesting more efficient use ofrunway capacity. Additional results found that airports with more runways are associated with MSAs that have lower population densities and lower average delays for the peak period highway traveler. Specific results for the Atlanta-Sandy Springs-Marietta MSA and Atlanta's Hartsfield Jackson International Airport found that, during the period 2001 - 2007, Atlanta's fifth runway had no appreciable impact upon gross metropolitan product but decreased the average product oflabor 0.94%. Over the longer period 1992 - 2007, an increased number ofrunways at Hartsfield-Jackson increased labor compensation as well as population density" As one ofthe busiest airports in the nation during the sample period, Atlanta's Hartsfield Jackson scale ofoperations is unlike that ofother MSA's with one large hub and no medium hubs airports. An area for future research is to determine whether the results obtained in this analysis are robust ifone considers larger MSAs that have multiple airports .. Many ofthe results in this analysis are new and suggestive ofthe effects that public capital, in the form ofairports and, specifically, runways, have upon MSA economic development.. Yet, there needs to be considerably more research in order to better understand the linkages that exist between airport capacity and economic development. The finding that a new runway significantly increases GMP and labor productivity in some areas, significantly decreases these measures in other areas, and

24 MAA-4-2012-Report excerpts distributed at the meeting

has no effect on yet other areas indicates that airports and runway capacity are having diverse effects on metropolitan areas that are not at all well understood .. Last, and related, is a need to improve understanding ofthose factors that increase the likelihood that a MSA will add a new runway, Certainly, increases in the demand for air travel are influential. At the same time, associated with increased capacity are congestion and other effects that may deter economic development. It is important for policy makers to understand the direct and indirect effects ofincreasing airport capacity ifthe nation's systems ofairports are to be engines of economic development. And, as a corollary, an area for future study is to explore the conditions and factors that determine whether new runways or extending existing runways offers the best alternative for increasing capacity"

25 MAA-4-2012-Report excerpts distributed at the meeting

References

L Aschauer, D.. (1989), Is Public Expenditure Productive?, Journal ofMonetary Economics 23, 177-200. 2 .. Brueckner, 1 (2003) Airline Traffic and Urban Economic Development, Urban Studies 40, 1455-69 .. 3.. Costa, Jose da Silva, R. ElIson, R. and R. Martin (1987), Public Capital, Regional Output and Development: Some Empirical Evidence, Journal ofRegional Science 27,419-37. 4.. Greene, W.. (1997). Econometric Analysis Crd Edition).. Prentice Hall: Upper Saddle River, New Jersey .. 5. Eisner, R. (1991), Infrastructure and Regional Economic Performance: Comment, New EnglandEconomic Review, September-October, 47-58 .. 6.. Garcia-Mila, T., T. McGuire, and R. Porter (1996), The Effect ofPublic Capital in State Level Production Functions Reconsidered, Review ofEconomics and Statistics 78, 177-80 .. 7.. Goetz, A. (1992) Air Passenger Transportation and Growth in the U.S .. Urban System, 1950 1987, Growth and Change, Spring, 217-238 .. 8. Green, R. (2007), Airports and Economic Development, Real Estate Economics V35, 91-112. 9.. Hakfoort, 1, 1 Poot, and P. Rietveld (2001), The Regional Economic Impact ofan Airport: The Case ofAmsterdam Schiphol Airport, Regional Studies 35,595-604.. 10. Holtz-Eakin, D. (1994) Public-Sector Capital and the Productivity Puzzle, Review of Economics and Statistics, February 76, 12-21. 11. Munnell, A. (1990a), Is There a Shortfall in Public Capital Investment? An Overview, In Munnell, A. (ed..) Is There a Shortfall in Public Capital Investment? Proceedings ofa conference held at Harwich Port, Massachusetts, June 1990, Conference Series, no.. 34 Boston: Federal Reserve Bank ofBoston, 1-20. 12. Munnell, A. (1990b), Why Has Productivity Growth Declined? Productivity and Public Investment, New England Economic Review, January-February 1990,3-22. 13 .. Ramanathan, R. (1995).. Introductory Econometrics. With Applications .. Fort Worth: Harcourt Brace and Company.. 14. Tatom, 1 (1991), Public Capital and Private Sector Performance, Federal Reserve Bank ofSt Louis Review, May-June 73,3-15 ..

26 MAA-4-2012-Report excerpts distributed at the meeting

Urban Studies, Vol. 40, No.8, 1455-1469, July 2003

Airline Traffic and Urban Economic Development

Jan K. Brueckner

[Paper first received, September 2002, in final form, November 2002J

Summary. This paper provides new evidence on the link between airline traffic and employ ment in US metropolitan areas. The evidence confirms the common view that good airline service is an important factor in urban economic development. Frequent service to a variety of destinations, reflected in a high level of passenger enplanements, facilitates easy face-to-face contact with businesses in other cities, attracting new firms to the metro area and stimulating employment at established enterprises. The empirical results show that a 10 per cent increase in passenger enplanements in a metro area leads approximately to a 1 per cent increase in employment in service-related industries. However, airline traffic has no effect on manufacturing and other goods-related employment, suggesting that air travel is less important for such firms than for service-related businesses. These estimates are generated controlling for reverse causal ity between employment and traffic. The results imply that expansion of Chicago's O'Hare airport would raise service-related employment in the Chicago metro area by 185000 jobs (this impact assumes that expansion raises traffic by 50 per cent). Thus, the expansion of O'Hare airport represents a powerful economic development tool, as argued by its proponents.

1. Introduction Small-town business leaders and government face-to-face contact with business collabora officials sometimes complain that inadequate tors in other cities. This contact, which is airline service is an obstacle to local econ achieved through business travel by the omic development. It is alleged that poor firm's employees, is more costly in terms of service inhibits local employment growth by time and money when airline service is poor. limiting the attractiveness of the city as a While raising the cost of production, these location for new businesses and by reducing higher travel costs may also limit the volume the viability of existing firms. In Cham of face-to-face contacts that the firm under paign-Urbana, Illinois, for example, airline takes. This limitation may in turn impair the service quality has been viewed as a potential viability of the enterprise, especially in high impediment to an attempt by the University tech industries where exchange of infor of Illinois to stimulate high-tech employment mation is critical. through creation of a research park. In effect, by facilitating easy face-to-face The quality of airline service matters to contacts with collaborators in other cities, firms because it affects the cost of achieving good airline service fosters intercity agglom-

Jan K. Brueckner is in the Department of Economics and Institute oj Government and Public Affairs, University of Illinois at Urbana-Champaign, 1206South Sixth Street, Champaign, IL61820, USA. Fax. 2172446678 E-mail' jbrueckn@uiucedu, The author would like to thank Robert Resek and Stuart Rosenthal for helpful discussions and Paul Byrne for research assistance, Any shortcomings in the paper; however; are the author's responsibility

14.56 JAN K. BRUECKNER eration economies. These intercity effects nalize some business services (Pred, 1974, complement the agglomeration economies pp. 205-206). that occur among firms within a given city, whose importance is now firmly established Despite its plausibility and potential import following a rebirth of empirical work on ance, the link between airline service and agglomeration (see, for example, Glaeser et economic development has been the focus of a!., 1992; Rosenthal and Strange, 2001). By only a few research studies. To study this demonstrating the significance of agglomer link, Brueckner (1982) assumed that employ ation effects, this new literature suggests that ment growth in a metropolitan area over a intercity agglomeration economies, which multi-year period depends on the level of are fostered by air travel, may also be im airline traffic in the period's base year. High portant. Because poor airline service limits traffic, indicative of frequent airline service the extent of these economies, it could con to many destinations, was presumed to stitute an impediment to urban economic de stimulate employment growth by attracting velopment. 1 new firms and helping existing firms to pros The geographer Allan Pred commented on per. Unfortunately, the hypothesised relation these issues back in 1977, when air travel ship emerged only weakly in the chosen was arguably less critical for businesses than sample, which consisted of 75 small and it is today. He noted that medium-sized metropolitan areas. The more recent study of Button et a!. (1999) related There are tremendous savings in time, and the level of high-technology employment in hence costs, that accrue from the cluster a sample of over 300 metro areas to a num ing of organizational head offices and an ber of explanatory variables, including a cillary business services in major dummy variable indicating whether the metropolitan areas. The time and cost sav area's airport is one of the nation's 56 ings available in large urban centers are largest. The study confirmed the anticipated compounded by the superior air-transport connection between high-tech employment connections those places possess ... Cen and airport size. Finally, Green (2002) took ters which do not have a wide variety and an approach similar to that of Brueckner great number of daily nonstop flights to (1982), regressing employment as well as the leading metropolitan complexes with a population growth in a metro area on base given system of cities are not particularly year airline traffic. Green's study represents attractive ... because they do not permit an improvement over earlier efforts because nonlocal personal contacts ... to be carried it includes a rich list of additional explana out ... efficiently (Pred, 1977, p. 24). tory variables and it attempts to control for In discussing the success of Boise, Idaho, as the potential endogeneity of airline traffic. a corporate headquarters location, Pred While the study shows the anticipated posi makes the following additional observations: tive connection between growth and base year traffic, the results are marred by the The functioning of major headquarters unexpectedly weak performance of many of office units in Boise has also been made the other explanatory variables. viable by the commercial airline service In attempt to improve further on previous available to that geographically isolated work, the present study offers additional em metropolitan area ... The Boise evidence pirical evidence on the link between airline indicates that lesser metropolitan [areas] traffic and economic development. Following ... possess the potential to house ... div Button et a!. (1999), and in contrast to isional headquarters ... and research and Brueckner (1982) and Green (2002), this link development units ... Good air travel con is assumed to be contemporaneous rather nections would have to be available, and than occurring with a lag. In other words, the the organizations would have to ... inter- level of airline traffic is assumed to affect MAA-4-2012-Report excerpts distributed at the meeting

AIRLINE TRAFFIC AND DEVELOPMENT 1457 metro-area employment in the same year, E = f(T, X; 9) + u (1) rather than boosting employment growth over subsequent years. Part of this contempo where X is a vector of exogenous variables raneous employment effect counts the jobs of that influence employment; 9 is a par'ameter airport workers, which obviously grow in vector; and u is an error term. number as traffic expands, as well as the The relationship in equation (1) can be usual mUltiplier effects of such job growth. viewed as a 'quasi reduced-form' relation But airline traffic's employment effect also ship in that, aside fI'om airline traffic, all has an additional component, which captures potentially endogenous variables that help to the stimulative effect of the intercity agglom determine employment have been eliminated eration economies created by good airline by appropriate substitution fI'Om other struc service. tural equations. For example, although the Given a contemporaneous relationship be wage level might be viewed as a demand tween employment and airline traffic, the side determinant of employment, the wage is endogeneity of traffic becomes a more seri endogenous, being jointly determined along ous issue than in the studies that use a base with E, and it is therefore suppressed in year value to explain subsequent equation (1). Only those variables that can employment growth. In other words, while reasonably be viewed as exogenous determi airline traffic may affect employment, traffic nants of employment appear in X. itself depends partly on the contemporaneous In a separate structural equation determin level of employment in a metro area, which ing airline traffic, T depends on E, on other helps to determine the volume of business endogenous variables such as incomes (and travel. In order to prevent the empirical re hence wages) and on a set of exogenous sults from being contaminated by this poten variables. Through solution of the structural tial reverse causality, satisfactory system, T ultimately depends on a collection instruments, which are variables that affect of exogenous variables that is potentially airline traffic without being strongly corre broader than the set contained in X. In order lated with employment, must be used. While for equation (1) to be identified, allowing the including a number of exogenous variables effect of T on employment to be measured, that help to explain employment, the study this set of exogenous variables must include also makes use of several plausible instru at least one variable (instrument) that does ments. The sample, which pertains to the not already appear as part of X. A challenge year 1996, consists of 91 US metropolitan is to find such instruments, which should be areas covering a wide range of population highly correlated with T but uncorrelated (or sizes. The paper's empirical findings confirm weakly correlated) with the error term u. the hypothesised link between employment Once the instruments are selected, equation and airline traffic. (1) can be estimated by two-stage least The next section of the paper presents the squares (2SLS). empirical framework and discusses the vari To enumerate the actual variables that are ables it contains. Section 3 presents the em used in the empirical model, the discussion pirical results and section 4 uses the results first focuses on those appearing in equation to predict the employment impact of expan (1), with the choice of instruments discussed sion of Chicago's O'Hare airport. Section 5 last. The employment variable E is total non offers conclusions. farm employment in the metropolitan area for 1996, denoted EMP. Other versions of the estimating equation use the disaggregated 2. Empirical Framework and Data employment measures GDSEMP and Letting E denote employment in a metro area SVCEMP, which represent goods-related em and T denote airline traffic, the estimating ployment (manufacturing, construction and equation has the form mining) and service-related employment, re- MAA-4-2012-Report excerpts distributed at the meeting

1458 JAN K. BRUECKNER

spectively. The latter measure includes that a high tax burden deters employment wholesale and retail trade, FIRE (finance, (both from the demand and supply sides), the insurance and real estate), services, govern coefficients of the tax variables should be ment, transport and public utilities employ negative. ment. Airline traffic T is measured by total 1996 The first variable in X is the metro area's passenger enplanements (the number of pas population, denoted POP, Because the con sengers boarding aircraft) in the metro area, temporaneous 1996 population may be denoted TRAFFIC. In metro areas with mul jointly determined along with EMP and tiple airports, such as New York, Los Ange hence endogenous, the metro area's 1990 les and Chicago, TRAFFIC is computed by population is used instead as the POP vari summing enplanements across the relevant able. The predicted sign of this variable's airports. It should be noted that for these coefficient in equation (1) is obviously posi multiple-airport cities and other large urban tive. Since, conditional on POP, the age dis centres, construction of the sample relies on tribution of the metro area's population is the most-encompassing definition of a metro likely to affect employment, the variables area. Thus, when a Consolidated Metropoli YOUNG and OLD appear as part of X, These tan Statistical Area (CMSA) is defined for a variables measure the 1996 shares of the metro area (embracing several smaller units), population with ages 14 years and under and the CMSA is used as the unit of observation. 65 years and over, respectively, and the ef~ For example, in the case of New York, the fects of both variables on EMP are expected unit of observation is the New York to be negative. To measure the possible lure Northern New Jersey-Long Island CMSA of sunbelt locations, the climate variable and the relevant airports are La Guardia, HEATING, which equals average heating de John F. Kennedy, Newark, Islip, White gree days for the metro area over the 1971 Plains and Newburgh (NY). The Appendix 2000 period, appears as part of X. With a provides a list of the 91 metro areas in the high value of HEATING indicating a 'rust sample, indicating the multiple airport cases. belt' location, the traditional view would pre Note that for smaller metro areas, the unit of dict a negative sign for its coefficient. A observation is the MSA.2 metro area with a high level of human capital Turning to the choice of instruments, these is presumably an attractive location for em variables should again be strongly correlated ployers and this level is measured by the with TRAFFIC but unconelated or weakly variable COLGGRAD, which equals the per correlated with the error term in equation (1). centage of the 1990 population over age 25 The model relies on four instruments, which with a college degree. COLGGRAD's are used as a group in the 2SLS estimation. coefficient is likely to be positive. The first instrument, HUB, indicates Following Green (2002), union influence whether the metro area contains a hub air is measured by the dummy variable RTW, port.3 In the case of a single-airport metro which equals one if the metro area is located area, HUB is simply a dummy variable equal in a state with a 'right-to-work' law, which to one if that airport is a hub and zero inhibits the organising efforts of unions. The otherwise. When a metro area has multiple expected sign of RTW's coefficient is posi airports and one is a hub, the HUB variable tive. Finally, again following Green (2002), equals the share of that airport in the metro X includes variables designed to capture the area's total enplanements. In effect, HUB personal and corporate tax burdens in the then equals the zero-one dummy scaled state containing the metro area. PERSTAX down by the hub airport's enplanement equals the maximum marginal rate for the share. In case of New York, for example, state's personal income tax in 1996, while Newark is a hub airport, so that HUB equals CORPTAX equals the maximum marginal Newark's share of the New York metro rate for the state's corporate tax. Assuming area's total enplanements. MAA-4-2012-Report excerpts distributed at the meeting

AIRLINE TRAFFIC AND DEVELOPMENT 1459

To see that HUB has some features of a US hub locations lie in the centre of the suitable instrument, observe that traffic at a country, facilitating both east-west and hub airport consists of local passengers plus north-south travel. Of course, many hub air a typically larger volume of connecting pas ports (Newark, Miami, Washington-Dulles, sengers, whose trip begins and ends else Philadelphia, San Francisco, among others) where. Since both types of passenger constitute exceptions to this central tendency. generate enplanements, the TRAFFIC mea Nevertheless, the effect of centrality is mea sure for a hub will be much larger than for a sured by the distance from the given airport non-hub airport located in a similar-size city, to the US population centre of gravity for the which has little or no connecting traffic, year 1990, which lies in Missouri.4 A de Thus, the HUB variable will be an important crease in this variable, which is denoted determinant of TRAFFIC, satisfying one re CENTRALITY, may raise the likelihood that quirement of a suitable instrument. an airport is a hub, thus increasing TRAFFIC. As for the other requirement, lack of cor However, holding the elements of X fixed, relation with u in equation (1), the suitability CENTRALITY is unlikely to affect EMP, thus of HUB can be questioned. Since the advan being unconelated with u. tage of a large local passenger base means A second instrument attempts to capture that hubs tend to be located in large metro the traffic diversion effect of proximity to a areas, it follows that the raw conelation be large airport. For cities that are relatively tween HUB and EMP is positive. However, close to a metro area containing a large the above correlation requirement says that airport, passengers may drive to it rather than an increase in any of the unobserved determi flying out of their local airport, reducing nants of EMP (which are contained in u) local enplanements. The variable PROXIM should not affect the HUB status of a metro ITY, which captures this traffic diversion ef area's airport(s). Since POP, an observed fect, is set equal to one for smaller metro determinant of EMP, is implicitly held fixed areas (those with enplanements outside the in this thought experiment, the correlation top 26) that are within 150 miles of a metro condition reduces to the following require area containing a large airport (metro areas ment: an increase in an unobserved variable within the top 26).5 A negative correlation that tends to raise a metro area's EMP above between PROXIMITY and TRAFFIC is the typical level for areas of its size will not likely, but the variable is unlikely to be affect the hub status of the airport. If hub correlated with u in equation (1). locations are chosen more on the basis of a Two additional instruments capture the metro area's population than its employment special features of particular airports and level, this requirement may be satisfied. On metro areas. The variable SLOT indicates the other hand, the importance of business whether the metro area has a slot-controlled travel suggests that, among metro area with a airport. Such airports, which operate at ca particular population, those with abnormally pacity and are thus closed to new traffic, high employment levels might be more at should have TRAFFIC levels below those of tractive as hub locations, in which case the otherwise comparable airports that are not conelation requirement would be violated. capacity-constrained. Since each of the slot Even in this instance, however, the cone controlled airports (Chicago-O'Har'e, New lation between HUB and u may not be sub York-La Guar'dia, New York-JFK and stantial. Washington-National) is in a multiple Recognising the potential drawbacks of airport metro area, construction of the slot HUB as an instrument, a variable that may be variable follows that of HUB. The variable is an exogenous determinant of an airport's hub set equal to the metro-ar'ea enplanement status can be substituted in its place. This share of the slot-controlled airport(s) for variable is suggested by the geography of those ar'eas containing such an airport and airline networks, which dictates that the best zero otherwise. SLOT should be negatively MAA-4-2012-Report excerpts distributed at the meeting

1460 JAN K. BRUECKNER

correlated with TRAFFIC but unconelated from zero, indicating that the slot-control with u. status of airports has no effect on TRAFFIC, Finally, two metro areas, Las Vegas and holding its other determinants constant. Orlando, have abnormally high traffic levels Among the X variables, only POP and because of the special leisure attractions that COLGGRAD have significant effects on they offer. To capture the effect of such TRAFFIC. A larger population naturally outliers, the dummy variable LEISURE is set raises enplanements and the coefficient esti equal to one for these two metro areas and mate of 0.979 indicates that the elasticity is zero otherwise. While LEISURE should virtually unitary, with a 1 per cent increase in show a strong positive correlation with POP raising TRAFFIC by 1 per cent. The TRAFFIC, the variable should not exhibit positive coefficient of COLGGRAD shows any particular relationship to employment that a highly educated metro-area generates and hence u (affecting its composition rather more airline traffic than a less-educated area. than its level). This result partly reflects the fact that highly Table I presents definitions and summary educated workers are likely to be employed statistics for all of the variables of the model. in occupations that require business travel. In addition, COLGGRAD's coefficient is likely to capture the effect on leisure travel of an 3. Empirical Findings increase in income, which is suppressed from Table 2 presents the regression results when the empirical framework (see above) but cor HUB is used as an instrument along with related with education. In addition, high edu PROXIMITY, SLOT and LEISURE; and cation may increase the propensity for leisure Table 3 presents the results when CENTRAL air travel, holding income constant. The lack ITY replaces HUB. In the regressions, TRAF of significance of the remaining coefficients FIC, POP and the employment variables shows that the airline traffic does not depend appear in natural log form. In addition, the on the age distribution of a metro area's t-statistics in the regressions are based on population (YOUNG, OLD), on its climate robust standard en'01S (White, 1980) to ac (HEATING) or level of union activity (RTW), count for possible heteroscedasticity in the or on its income-tax burden (PERSTAX, enOl' structure. Once the estimates have been CORPTAX).6 discussed, the results of diagnostic tests for The second column of Table 2 presents the endogeneity of TRAFFIC and the suitability 2SLS estimates of the coefficients of equa of the instruments are presented. tion (1), while the third column presents OLS The first column of Table 2 presents the estimates for comparison. As expected, the first-stage regression of the 2SLS procedure, results show that airline traffic exerts a where TRAFFIC is regressed on the X vari significantly positive effect on total employ ables and the instruments. Focusing first on ment in a metro ar'ea. The point estimate the instruments, the HUB and LEISURE vari shows that the elasticity of this effect is 0.09, ables both have positive coefficients, as pre indicating that a 10 per cent increase in dicted, and the estimates are highly TRAFFIC raises EMP by 0.9 per cent. This significant. Thus, metro areas that contain effect, which shows that a TRAFFIC increase hub airports or are prominent leisure destina translates into higher employment in approx tions have higher TRAFFIC levels. The imately a 10: 1 ratio, is substantial in size. Its PROXIMITY coefficient is negative and magnitude shows that employment gains ex significant, indicating that small and me tend far beyond the airport itself, where dium-sized metro areas close to a large air higher traffic directly generates more jobs. port experience a diversion of traffic, which While these outside employment gains partly lowers local enplanements. While SLOTs reflect the usual multiplier effect of new jobs, coefficient has the predicted negative sign, the gains are also consistent with the exist the coefficient is not significantly different ence of intercity agglomeration economies, MAA-4-2012-Report excerpts distributed at the meeting 0\ ...... j:>...... -J tr1 z ;;::: tr1 tr1 t'"" <: "1:1 0 tl tl 'T1 z ;J> tr1 t'"" 'T1 Z """ ;J> iO ~ n ;J> 1 1 1 1 0.87 edition. 12.0 11.0 author's 34.8 30.9 28.5 721.8 298.3 TRAFFIC: I 1 7976 7595.3 8893.0 SLOT: Administration 1996-97 19480012 34937810 1997-98. 0 0 0 0 0 0 0 704 States, 12.0 15.7 76.6 11.5 77.6 92.2 155 Book, the Atmosphenc PROXIMITY, of 139236 285585 Minimum Maximum and Data Book Area LEISURE, 0.33 0.02 0.02 0.19 6.54 4.73 0.57 12.3 Oceamc 22.0 21.6 162.8 717.7 880.5 3914 Mean 742861 1 5638982 National Governments, Metropolitan CENTRALITY, State airport and of gravity 736.6 statistics HUB, HEATING: large State of of Council 1997. law state state degree centre summary Census, miles the area's and 150 area's of college Carriers, area 1000s) PERSTAX: a 1000s) population metro Air younger older right-to-work metro (in within (in 1000s) metro or or with for US Bureau for has definitions in airport(s) Orlando 14 65 25 (in 1996 areas area the US 1996 rate rate CORPTAX, state to and age age for area over for tax (http://www.nrtw.org/rtws.htm). 1996 Certificated tax of of metro metro OLD: airports Variable of for area in area's 1. metro airport Vegas income at website income smaller Las metro population metro YOUNG, employment Statlstics Table population population largest employment airports for for if for 1990 employment slot-controlled hub 1990 one one POP, one personal corporate area's Foundation days Activity to to to for any any enplanements area's metro-area metro-area of of metro equal equal equal service-related non-farm goods-related degree marginal marginal Airport Defense SVCEMP, 1996 1996 the metro share share population of of of passenger 1996 1996 Legal heating variable variable from variable Statistics, 1996 metro-area metro-area metro-area GDSEMP, Work to Dummy Enplanement Mileage Dummy Average Percentage Maximum Maximum Dummy Enplanement Metro-area Percentage Percentage Total Definition Total Total Total EMP, Right Transportation (http://lwf.noaa.gov/oa/climate/online/ccd/nrmhdd.html). sources; of National PROXIMITY SLOT Data CENTRALITY YOUNG LEISURE HUB COLGGRAD CORPTAX HEATING PERSTAX OLD RTW POP GDSEMP SVCEMP TRAFFIC EMP Variable website Bureau calculations. RTW: MAA-4-2012-Report excerpts distributed at the meeting tr1 ;:r.:l ~ tr1 ttl z ;:r.:l c::: z ?" (') :; 0'1 +>. N I-' 91. = (OLS) 0.991 0.00336 0.00377 0.0000224 0.00583 0.0202 0.0991 0.102 0.0423 0.868 6.552 (0.83) (0.47) (1.60) (3.28) (3.99) (3.07) (4.87) (5.60) SVCEMP - (17.86) - - - observations ) (2SLS) 0.00390 0.00356 0.0000221 0.00549 0.0968 0.0204 0.0429 0.858 0.110 6.509 errors; (0.81 (0.59) (1.54) (3.44) (4.43) (3.16) (4.92) (3.52) SVCEMP - (16.95) (21.56) (35.60) - - - standard robust (OLS) 0.947 0.0192 0.0172 0.0000741 1.023 0.0135 0.164 0.0396 0.0619 0.00768 (1.53) (1.50) (1.21) (4.14) (3.06) (1.96) (2.84) (0.16) (8.56) GDSEMP on ------7.668 based instrument as (2SLS) 0.0174 0.0178 0.0000732 0.0145 0.157 0.0639 0.0168 0.992 (1.66) (1.69) (1.13) (4.25) (3.39) (1.95) (3.07) (0.29) (8.58) GDSEMP - (11.82) (14.54) parenthesis, - -0.0400 - -7.532 HUB in with ) results t-statistics EMP (OLS) 0.992 0.00214 0.0000332 0.00202 0.0233 0.118 0.0460 6.304 0.903 (0.26) (0.62) (0.31 (5.30) (4.28) (4.25) (5.53) (4.34) - (16.79) - - - absolute Regression logs; 2. in EMP (2SLS) 0.00141 0.00135 0.00107 0.0000328 0.00157 0.0235 0.114 0.0886 0.0782 0.0469 0.889 6.246 (0.34) Table (0.48) (0.21) (5.51) (4.82) (4.36) (3.15) (5.72) - (16.60) - - - SVCEMP (OLS) GDSEMP, 1.653 0.872 0.315 0.338 0.922 0.0637 0.00908 0.0000128 0.0540 0.0223 0.0176 0.0437 0.887 0.979 (0.82) (1.96) (3.92) (0.42) (5.76) (3.63) (1.78) (0.32) (0.74) (0.11) (0.86) (0.41) TRAFFIC - - - - (11.63) (24.24) (35.39) - - EMP, POP, variables TRAFFIC, 2 Notes: PROXIMITY R HUB SLOT LEISURE PERSTAX COLGGRAD CORPTAX HEATING YOUNG OLD RTW TRAFFIC INTERCEPT POP Independent MAA-4-2012-Report excerpts distributed at the meeting

AIRLINE TRAFFIC AND DEVELOPMENT 1463

as discussed above. In other words, a high effect on employment in manufacturing and level of TRAFFIC, reflecting frequent airline other goods-related industries. But this service to many destinations, stimulates em finding conforms well to intuition regarding ployment at established firms and attracts the nature of intercity agglomeration econ new employers to the metro area. omies.. Given the routine nature of much The estimates also confirm a number of manufacturing and construction activity, other expectations. Employment is higher in firms in these industries have less need for large metro areas, with the significant POP face-to-face contact with businesses in other elasticity of 0.89 indicating that EMP in cities than do firms engaged in more infor creases slightly less rapidly than population mation-intensive pursuits. As a result, good itself. The significantly negative YOUNG and airline service would constitute much less of OLD coefficients indicate that a metro area an attractive force for goods-related firms has relatively low employment when a larger than for firms in other industries, implying than average share of the population is out the absence of a link between GDSEMP and side the working years. The significantly TRAFFIC. positive coefficient of RTW shows, as ex By contrast, the estimates in column six pected, that metro areas in states with union show that employment in service-related in inhibiting right-to-work laws have higher dustries, SVCEMP, does respond to the level employment than those in states with more of airline traffic in a metro area. The union influence. With RTW being a dummy significant elasiticity estimate equals 0.11, variable, its coefficient shows that a right-to indicating that a 10 per cent increase in work law boosts a metro area's employment TRAFFIC raises SVCEMP by 1.1 per cent. by 11 per cent. Thus, the positive overall impact of TRAF Contrary to expectations, however, PER FIC on EMP arises through the channel of STAX, CORPTAX and COLGGRAD have in service-related employment, a finding that is significant coefficients, indicating that a consistent with the presumed nature of inter metro area's income tax burden and its hu city agglomeration economies .. Note that the man capital level have no effect on employ SVCEMP elasticity is naturally larger than ment. Finally, the significantly positive the overall elasticity of 0.09 from column coefficient of HEATING shows that metro two, a consequence of the fact that TRAFFIC areas in the rustbelt region of the country has a zero impact on the goods have higher employment, other things equal, related component of EMP. than metro areas in the warmer sunbelt. Returning to the other coefficients in Inspection of estimates in the third column column four, POP, YOUNG, OLD, RTW and of Table 2 shows that the OLS results are HEATING have the same qualitative effects very similar to the 2SLS estimates of column on GDSEMP as in the EMP equation. Note, two. While TRAFFIC's coefficient declines however, that a right-to-work law reduces from 0.09 to 0.08, the rest of the significant GDSEMP by a larger 16 per cent. While the coefficients are very close in magnitude to tax and education coefficients are again in the values in column two. This pattern is significant, COLGGRAD's negative impact is discussed further below once the entire set of marginally significant, plausibly indicating results has been considered. that highly educated metro areas are not fa The fourth column of Table 2 presents the voured locations for manufacturing and other 2SLS estimates when EMP is replaced by goods-related firms. By contrast, total goods-related employment, GDSEMP. COLGGRAD's coefficient in the SVCEMP While many of the coefficients mirror the regression in column six is positive and mar EMP results, a striking change is the lack of ginally significant, suggesting that educated significance, as well as the small magnitude, metro areas may attract service-related firms. of the TRAFFIC coefficient. This coefficient The remaining X coefficients in the estimate implies that airline traffic has no SVCEMP equation mirror the qualitative MAA-4-2012-Report excerpts distributed at the meeting

1464 JAN K. BRUECKNER

Table 3. Regression results with CENTRALITY as instrument

TRAFFIC EMP GDSEMP SVCEMP Independent variable (OLS) (2SLS) (2SLS) (2SLS)

INTERCEPT - 4,001 -6338 -8.428 - 6,416 (L66) (14,71) (9.50) (16.11) POP 1..220 0.910 1.197 0,,836 (14.78) (15,,84) (1 Ll9) (15,,64) TRAFFIC 0.0720 - 0,145 0.127 (L56) (L90) (2,94) YOUNG 0,0311 - 0,,0455 -0,0508 -0,0443 (0,50) (5,07) (2,68) (4.98) OLD 0.00755 - 0.0231 - 0.0370 - 0,,0207 (0.21) (4,,72) (3..31) (4.40) RTW 0.280 0.119 0..204 0.0919 (1.24) (4.30) (2.44) (2,,93) HEATING 0.0000681 0,0000335 0,0000794 0,,0000215 (1.23) (5.50) (4,,62) (3,32) COLGGRAD 0.0456 000229 - 0,,00754 0.00476 (250) (0.62) (0.86) (1.28) CORPTAX - 0.0338 - 0.00257 - 0,0287 0,,00507 (0.96) (037) (2,,00) (0,,69) PERSTAX - 0,0401 0,000900 0.0135 - 0.00311 (L64) (0.23) (1.25) (0.72) CENTRALITY 0,000240 (LlO) LEISURE 1.335 (3.32) PROXIMITY - 0.412 (2.49) SLOT - 0.621 (0,80)

R2 0,,840

Notes: TRAFFIC, POP, EMP, GDSEMP, SVCEMP in logs; absolute t-statistics in parenthesis, based on robust standard errors; observations = 91.

results from the other regressions, although contrary to expectations, a central location RTWs smaller coefficient suggests a weaker for a metro area does not boost airline traffic. right-to-work effect for service industries. Recall that such an effect was anticipated Finally, the OLS versions of the GDSEMP because a central location was expected to and SVCEMP regressions, reported in increase the likelihood that an airport serves columns five and seven, are similar to the as a hub. Evidently, the numerous exceptions 2SLS equations. to a hub centrality pattern mask the antici Table 3 presents an equivalent set of re pated effect of this variable on traffic.? gressions where the CENTRALITY variable Although the remaining coefficients in the replaces HUB. Recall that concerns about the regression are similar to those in Table 2, suitability of HUB as an instrument sug a noteworthy change is the rise in gested the use of this alternate variable. The POP's coefficient, which now indicates that first column of Table 3, which contains the TRAFFIC rises faster than population. first-stage regression of TRAFFIC on X and Turning to the 2SLS EMP results in the instruments, shows an insignificant column 2, a notable change is the loss of coefficient for CENTRALITY, indicating that, significance of TRAFFIC's coefficient. The MAA-4-2012-Report excerpts distributed at the meeting

AIRLINE TRAFFIC AND DEVELOPMENT 146.5 magnitude of the coefficient, however, is con'elation between TRAFFIC and the error similar to that in Table 2 and the remaining term u cannot be rejected, implying that the results mirror the earlier ones. In the OLS estimates in Tables 2 and 3 are consist GDSEMP regression in column three, the ent. When this test is canied out, an in TRAFFIC coefficient is again insignificant, significant coefficient does indeed emerge, but its sign is now negative and its t-statistic with the t-statistics well below unity in both is in the marginally significant range. Despite the EMP and SVCEMP regressions, using these divergences from the results of Table 2, either set of instruments. While this con the SVCEMP regression in column four clusion indicates that the OLS estimates in again shows a strongly significant impact of Tables 2 and 3 may be satisfactory, the usual TRAFFIC on employment in service-related danger involved in accepting a null hypoth industries. The estimated elasticity of 0.13 is esis (the unknown probability of doing so close to the value of 0.11 from Table 2, incorI'ectly) suggests that the 2SLS estimates indicating that a 10 per cent increase in are preferable. 8 In any event, the similar TRAFFIC raises service employment by 1.3 magnitudes of the OLS and 2SLS point esti per cent. The remaining results from the mates mean that choice between them is not GDSEMP and SVCEMP regressions are a critical issue. 9 similar to those in Table 2. Use of a different instrument, of course, has no effect on the 4. Applying the Results to Predict the OLS estimates of the three employment Impact of O'Hare's Expansion equations. The magnitudes of TRAFFIC's estimated Many cities view improved airline service as effects on employment are highly robust to a path to economic development. However, changes in the composition of the sample. since the existing level of service is an equi For example, if the two largest metro areas, librium outcome produced by a structural New York and Los Angeles, are dropped equation system that includes equation (1), from the sample, the elasticities of EMP and the desired service increase (and its attendant SVCEMP with respect to TRAFFIC in the employment effect) can only be achieved specification using HUB as an instrument through government intervention that remain at 0.09 and 0.11, respectively. When changes the parameters of the system. For metro areas with 1996 populations below 1 example, one type of intervention consists of million are deleted, reducing the sample to government subsidies for airline service pro 38 observations, the above elasticities are vided to small cities. Such subsidies boost 0.09 and 0.09, respectively. These findings traffic in an exogenous fashion as airlines suggest that TRAFFIC's employment effects reduce small-city fares and increase flights, are stable across a broad range of metro-area and the empirical model predicts a resulting populations. impact on local employment. The similarity of the 2SLS and OLS esti In another intervention, the government mates in Tables 2 and 3 suggests that con can invest resources to increase the size of a cerns about the endogeneity of TRAFFIC capacity-constrained airport. At such an air may be unwarI'anted despite their strong intu port, the desired level of airline operations itive basis. The Hausman-Wu specification exceeds capacity and expansion of the airport test can be used to investigate this issue (see allows extra flights to be accommodated, Davidson and MacKinnon, 1993). To cany leading to an exogenous increase in traffic" out the test, the fitted values of TRAFFIC are Chicago's O'Hare airport is a prominent included as an additional variable (along example of such a capacity-constrained air with the actual TRAFFIC variable) in an port and plans for its expansion have been OLS regression based on equation (l). If the formulated. These plans, which involve a coefficient of the fitted TRAFFIC variable is hard-won agreement between the mayor of insignificant, then the null hypothesis of zero Chicago and the governor of Illinois, call for MAA-4-2012-Report excerpts distributed at the meeting

1466 JAN K. BRUECKNER

an approximate doubling of the airport's predict the resulting level of traffic, a reason flight capacity. Despite the agreement, viru able guess can serve as a basis for estimating lent opposition to the expansion still exists .. the magnitudes of the employment effect of This opposition comes mainly from residents the expansion. living near the airport (and their elected rep To this end, suppose that the new equilib resentatives), who point to the increase in rium level of traffic at O'Hare is 50 per cent noise and other environmental damage that above the current level, a seemingly con will result from a greater volume of flights. servative estimate given the proposed doub Proponents argue that O'Hare's expansion ling of flight capacity. Then, the new will greatly stimulate the economy of the equilibrium level of employment must satisfy Chicago metropolitan area, arguing that the equation (1), with the value of TRAFFIC airport's role as a major economic engine in raised by 50 per cent. Rounding the estimated the region will be strengthened. However, SVCEMP elasticity hom Table 2 down to 0.1 quantitative predictions regarding the poss for simplicity, so that percentage changes are ible magnitude of such an impact are scarce. in a 10: 1 ratio, this 50 per cent traffic increase The current empirical results can be used to is associated with a 5 per cent increase in generate such a prediction, as follows. service-related employment. In 2001, such First, observe that, if a capacity employment equalled 3 698 000 jobs for the constrained airport is still constrained after Chicago CMSA (see Bureau of Labor Statis expansion, then the increase in traffic cone tics, 2002). Five per cent of this value corre sponds exactly to that allowed by the expan sponds to approximately 185 000 jobs. Thus, sion. To understand this point, note that, by if O'Hare expansion raises traffic by 50 per raising metro-area employment, the traffic cent, service-related employment in the increase facilitated by greater capacity will Chicago CMSA is predicted to rise by almost itself produce a further traffic stimulus. How 200 000 jobs. With a zero elasticity for man ever, since the capacity constraint remains ufacturing employment, the results predict no binding, the airport cannot accommodate the effect on the level of such jobs. The service additional desired traffic. As a result, the related job impact is obviously substantial in traffic gain consists only of the exogenous magnitude, testifying to the power of O'Hare initial increase made possible by greater ca expansion as an economic development tool. pacity. Of course, a smaller equilibrium traffic in On the other hand, if the capacity con crease would be associated with a smaller straint ceases to bind after expansion, then employment gain. For example, the service the airport can accommodate a second-round employment gain would be somewhat more traffic gain, which is stimulated by the em that 90 000 jobs if O'Hare's traffic increased ployment increase following the initial gain by only 25 per cent following expansion. in traffic. Progressive rounds of feedback Given that Chicago already has excellent between traffic and employment ultimately airline service, it is natural to ask how addi achieve new equilibrium levels of these vari tional traffic could possibly lead to a further ables. However, predicting these levels re strengthening in intercity agglomeration quires a deeper knowledge of the parameters economies and thus more service-related em of the full equation system governing the ployment. One important avenue for such an variables. effect is through international airline service, Given the magnitude of the proposed ca whose growth is currently restrained by pacity increase at O'Hare, it is doubtful that O'Hare's capacity limit. In addition, one can the new airport will be capacity-constrained, imagine that the already-long list of domestic at least initially. As a result, the relevant destinations served from Chicago would scenario is likely to be the second one above, grow with an expansion of the airport, as where the employment/traffic feedbacks af would the frequency of service to existing fect the outcome. While it is difficult to destinations. MAA-4-2012-Report excerpts distributed at the meeting

AIRLINE TRAFFIC AND DEVELOPMENT 1467

5. Conclusion deleted flom this list, as were metro areas lacking data for any of the variables of the This paper has provided new evidence on the empirical model (usually employment). Fi link between airline traffic and employment nally, some of the listed metro ar'eas were in a metro area. The evidence confirms the consolidated to match the CMSA definitions, yielding a sample size of 91. common view that good airline service is an 3. Hub airports (and the associated airlines) are important factor in urban economic develop as follows: Atlanta (Delta), Charlotte, NC ment. Frequent service to a variety of desti (US Airways), Cleveland (Continental), nations, reflected in a high level of passenger Chicago-O'Hare (United, American), enplanements, facilitates easy face-to-face Cincinnati (Delta), DallaslFort Worth (American), Denver (United), Detroit contact with businesses in other cities, at (Northwest), Houston (Continental), Miami tracting new firms to the metro area and (American), Memphis (Northwest), Min stimulating employment at established enter neapolis/St Paul (Northwest), Newark (Con pnses. tintental), Philadelphia (US Airways), The empirical results show that a 10 per Phoenix (America West), Pittsburgh (US Airways), Salt Lake City (Delta), San Fran cent increase in passenger enplanements in a cisco (United), St Louis (TWA), Washing metro area leads approximately to a 1 per ton-Dulles (United). cent increase in employment in service 4, For multiple-airport cities, the distance from related industries. However, airline traffic the Imgest airport to the centre of gravity is has no effect on manufacturing and other used to compute CENTRALITY 5. These metro areas are designated as 'large goods-related employment, suggesting that hubs' by the Bureau of Transportation Statis air travel is less important for such firms than tics (1997). (In their terminology, 'hub' for service-related businesses. These esti refers to airport size and not to the existence mates are generated controlling for reverse of hub-and-spoke operations.) causality between employment and traffic., 6. Brueckner (1985) estimated a regression like that in column one for a sample of small and The results can be used to predict the medium-sized metro areas,. Among other employment effects of expansion of things, the results show a similar kind of Chicago's O'Hare airport. Assuming that ex proximity effect on traffic, pansion would generate a 50 per cent in 7, In a regression of HUB on POP and CEN crease in traffic, service-related employment TRALITY, the latter variable's coefficient is negative, as expected, but only marginally in the Chicago metro area would grow by significant. 185 000 jobs. Thus, the expansion of O'Hare 8. An intuitive explanation for the evident exo airport represents a powerful economic de geneity of TRAFFIC is not readily apparent velopment tool, as argued by its proponents. On the one hand, TRAFFIC may be deter mined more by population than by employ ment, limiting the extent of simultaneity Notes between TRAFFIC and employment. More over, the substantial variation of TRAFFIC 1, Another related literature studies the effect across hub and non-hub airports generates a of public capital (which includes airport in large, mostly exogenous, variation in this frastructure) on the productivity of private variable that may overwhelm any feedback firms (for the seminal paper, see Aschauer, effects flom employment. 1989, and, for a recent contribution, see 9 Although the outcome of the Hausman-Wu Pereira and Flores de Frutos, 1999; a survey test suggests that the use of instruments in is provided by Munnell, 1992), Since the estimating equation (1) may be unnecessary, studies in this literature are based on aggre it is useful nevertheless to carry out a test for gate capital measures, none shows explicitly overidentifying restrictions, which indicates the effect of investment in airports. whether the chosen instruments are valid. In/ 2. Selection of the sample metro areas was this test (see Davidson and MacKinnon, carried out as follows. The starting-point was 1993), the 2SLS residuals are regressed on a list of the 120 US metro areas with the the X variables and the instruments., Since largest airline traffic volumes, contained in these residuals are analogous to the error Bureau of Transportation Statistics (1997) term U, the explanatory power of this re Metro areas outside the continental US were gression should be low if the instruments are MAA-4-2012-Report excerpts distributed at the meeting

1468 JAN K. BRUECKNER

indeed uncorrelated with u (such conelation BUREAU OF TRANSPORTATION STATISTICS (1997) is absent for the X variables by assumption). Airport Activity Statistics of Certificated Air The test thus relies on the R2 value for this Carriers. Washington, DC: US Department of regression, rejecting the null hypothesis of Transportation. zero conelation if the value is high.. When BUTTON, K, LALL, S., STOUGH, R. and TRICE, R this test is carried out using the instrument (1999) High-technology employment and hub set that includes CENTRALITY, the null hy airports, Journal ofAir Transport Management, pothesis cannot be rejected for both the EMP 5, pp. 53-59 .. and SVCEMP regressions.. However, when COUNCIL OF STATE GOVERNMENTS (1996) Book of HUB replaces CENTRALITY, the null hy the States, 1996-97 edn. Lexington, KY: Coun pothesis is rejected for the EMP regression, cil of State Governments .. while the test-statistic lies just on the margin DAVIDSON, R. and MACKINNON, J. G. (1993) Esti of rejection for the SVCEMP regression. mation and Inference in Econometrics .. New While the overidentification test thus favours York: Oxford University Press. CENTRALITY as an instrument, the simi GLAESER, E, KALLAL, H, SCHEINKMAN, J. and larity of the estimates in Tables 2 and 3 SCHLEIFER, A. (1992) Growth in cities, Journal makes this choice mostly a matter of indif~ of Political Economy, 100, pp .. 1126-1152.. ference GREEN, R K (2002) A note on airports and economic development. Unpublished paper, References University of Wisconsin-Madison. MUNNELL, A. (1992) Infrastructure investment ASCHAUER, D.. (1989) Is public expenditure pro and economic growth, Journal of Economic ductive?, Journal of Monetary Economics, 23, Perspectives, 6, pp. 189-198 pp .. 177-200 PEREIRA, A. M. and FLORES DE FRUTOS, R. (1999) BRUECKNER, J. K (1982) Metropolitan airline Public capital accumulation and private sector traffic' determinants and effects on local em performance, Journal of Urban Economics, 46, ployment growth. Unpublished paper, Univer pp .. 300-322.. sity of Illinois at Urbana-Champaign. PRED, A. (1974) Major Job-providing Organiza BRUECKNER, J. K (1985) A note on the determi tions and Systems of Cities. Washington, DC: nants of metropolitan airline traffic, Inter American Association of Geographers .. national Journal of Transport Economics, 12, PRED, A. (1977) City Systems in Advanced Econ pp. 175-184.. omies .. New York: John Wiley and Sons. BUREAU OF THE CENSUS (1998) State and Metro ROSENTHAL, S. S. and STRANGE, W. C. (2001) The politan Area Data Book, 1997-98. Washington, determinants of agglomeration, Journal of DC: US Department of Commerce. Urban Economics, 50, pp. 191-229. BUREAU OF LABOR STATISTICS (2002) Current Em WHITE, H. (1980) A heteroscedasticity-consistent ployment Statistics. Washington, DC: US De covariance matrix estimator and a direct test for partment of Labor. heteroscedasticity, Econometrica, 53, pp .. 1-16.

Appendix Table At. Sample metro areas

Albany Fort Myers Oklahoma City Albuquerque Grand Rapids Omaha Allentown/BethlehemlEaston Green Bay Orlando Amarillo Greensboro Pensacola Atlanta Greenville, SC Philadelphia Austin Harrisburg Phoenix Baton Rouge Houstone Pittsburgh Birmingham Huntsville Portland Boise Indianapolis Raleigh/Durham Boston Jackson, MS Reno Brownsville Jacksonville Richmond Buffalo Kansas City Rochester, NY Cedar Rapids Knoxville Sacramento Charleston, SC Las Vegas Salt Lake City Charlotte Lexington San Antonio Chicago' Little Rock San Diego MAA-4-2012-Report excerpts distributed at the meeting

AIRLINE TRAFFIC AND DEVELOPMENT 1469

Table Al.-continued

Cincinnati Los Angelesd San Franciscog Cleveland Louisville Sarasota Colorado Springs Lubbock Savannah Columbia, SC Madison Seattle Columbus, OH Melbourne, FL Sioux Falls Corpus Christi Memphis South Bend DallaslFort Worthb Miamie Spokane Dayton Milwaukee St Louis Daytona Beach Minneapolis/St Paul Syracuse Denver Mobile Tampa/St Petersburg Des Moines Nashville Tucson Detroit New Orleans Tulsa El Paso New York! Washington, DCh Eugene Norfolk West Palm Beach Wichita

aAirports are Chicago-O'Hare and Midway. bAirports are DFW and Love Field. cAirports are Houston Intercontinental and Hobby.. dAirports are Los Angeles International, John Wayne, Burbank, Long Beach and Ontario. eAirports are Miami International and Fort Lauderdale. fAirports are La Guardia, John F. Kennedy, Newark, White Plains, Islip and Newburgh. gAirports are San Francisco, Oakland and San Jose .. hAirports are Washington-National, Dulles and Baltimore-Washington.