DEVELOPING MEASURES OF AIRPORT PRODUCTIVITY AND PERFORMANCE: AN APPLICATION OF DATA ENVELOPE ANALYSIS. David Gillen Wilfrid Laurier University

Transcription

1 DEVELOPING MEASURES OF AIRPORT PRODUCTIVITY AND PERFORMANCE: AN APPLICATION OF DATA ENVELOPE ANALYSIS David Gillen Wilfrid Laurier University Ashish Lall Nanyang Technological University Nanyang Business School S3-B1A-04 Singapore

2 Abstract Since deregulation the measurement of productivity performance and profitability of the air transport industry has attracted significant attention. Many studies have been undertaken on the financial and economic productivity of airlines but few have concentrated on the productivity or performance of airports, and how changes in the industry may have affected them. Airports have been quite traditional in their approach to assess their performance. Most measure it in strictly accounting terms by looking at only their total costs and revenue levels and at the resulting surpluses or deficits. Few utilize any type of productivity measures or performance indicators. A broader method of measuring the efficiency and productivity in both financial and physical terms is therefore needed In this paper a new approach to assessing the performance or productivity of airports is developed and estimated. Data Envelope Analysis is used to construct performance indices on the basis of the multiple outputs which airports produce and the multiple inputs which they utilize. In particular we develop productivity measures for terminals and airside operations. The performance measures are then used in a second stage Tobit regression in which environmental, structural and managerial variables are included. The regression results provide a 'net' performance index and also identify which variables the managers have some control over and what the relative importance of each variable is in affecting performance. The data set contains a panel of 21 US airports over a five year period. 1

3 1. INTRODUCTION Over the last two decades a great deal of effort and resources have been expended in developing measures of performance for carriers in the different modes of transportation. This had been stimulated by both deregulation and privatization initiatives. Measures of productivity performance, efficiency and effectiveness are now available for railways, airlines, trucking, and public transit firms. The measures range from relatively simple quantities, such as output per employee, to more sophisticated measures such as TFP (Total Factor Productivity) - a standard which takes account of all inputs in the production process. These measures have been used to assess alternative management actions and strategies in developing, for example, more effective means of satisfying the objectives of the owners or operators. They have also been used to measure technical progress and to rank carriers by their productivity gains. In other cases, measures of cost and service effectiveness have been developed in order to evaluate financing for capital projects and changes in public policy such as deregulation. The motivation for this paper stems from both an evolving trend of 'redefining the way in which government operates' and the growing tendency to shift major capital investments and operation in transportation away from direct government control. This can mean anything from privatizing or commercializing infrastructure to creating incentives for managers so that they pursue particular financial targets and perform in a way which maximizes the objectives of the owners. Worldwide this has taken such forms as airport and roadway privatization as well as commercialization through joint public/private ownership or the contracting out of various services. In many cases these enterprises are break-even or not-for-profit. Under such circumstances standard financial measures of performance such as the rate of return on capital or profits are not meaningful. It is also difficult to define a measure of output or service as well. A major impetus behind the desire to privatize or at least commercialize airports, roadways and ports throughout the world is the lack of investment capital available from governments to meet the needs of these infrastructure to invest in new facilities, terminals and equipment. Furthermore, management is under increasing pressure to wean themselves from government support by becoming more efficient. Airports, in particular, are recognized as mature 'firms' which should be able to stand alone and operate without government support or interference. A further stimulus to improve airport performance comes from the deregulated airline industry. Despite airport charges being only 5-7 percent of total operating costs, airlines operate in highly competitive markets and cannot easily pass rate increases on to customers. 1 They have continued to place pressure on airports to increase their efficiency. A set of performance measures would allow an airport to demonstrate any improvements. 1 These figures are for North America. The costs in Europe and Asia are somewhat higher due to higher fees for navigation and airport rates and charges. 2

4 Airports in Canada and the United States receiving scheduled service from commercial airlines are publicly owned. This ownership and operation takes many forms. Municipal and county ownership is the most common in the US. In Canada the federal government has recently started a program of airport devolution to Local Airport Authorities. These are not private but certainly operate under commercial principles. An airport's objective is not simply to maximize profits but to also provide necessary services while protecting the accessibility of the community. In these cases profit levels do not equate with efficiency. Another form of productivity or performance measurement is needed. However, very little information is currently available about performance measures or indicators now being used by airports. Some airports use measures internally. These indicators (mostly partial factor productivity indices) are used to monitor the airports performance over time. There are almost no inter-airport comparisons. 2 In this research paper we suggest that a method which can be used to assess the performance of the management of transportation infrastructure is Data Envelope Analysis (DEA). We compare this new approach with traditional measures and find that the DEA approach is richer in its ability to handle multiple inputs and outputs but weaker in its menu of statistical tests. We also develop a linkage between the performance measure and management strategies arguing that it is not sufficient to simply describe performance but also to be able to assess it and understand how managers can affect their performance. The focus in this paper is upon air transportation infrastructure but we believe the approach has broad application. 2. AIRPORTS - GENERAL As providers of space and services to facilitate the interchange between air and surface transportation, airports can range from Spartan, with little in the way of comfort or amenities, to plush, with restaurants, hotels, shops, and entertainment. Unlike European airports, where staffing levels can range from very low at broker airports such as Geneva to very high at full service airports, such as Frankfurt and Milan, Canadian and US airports are relatively homogeneous in their provision of physical plant and maintenance only, requiring airlines and concessionaires to provide personnel to staff their own activities. International differences also exist in the design of passenger terminals with a European philosophy of economy of design leading to terminals that are typically small, crowded, and relatively uncomfortable compared to those in North America where passenger comfort and convenience are considered marketing tools (De Neufville, 1976). The competitive environment of the aviation industry in the United States means greater market mobility for carriers and freedom to establish linkages and alliances. Carriers enter and exit markets and change frequency of service and gauge of aircraft. They form partnerships, alliances 2 Doganis and Graham (1987) have pointed out that only in France, the United Kingdom and Germany has some attempt has been made to systematically approach the problem of comparing efficiency among airports. All of these comparisons have been limited to airports within national boundaries. Many airports also seem opposed to any type of inter airport comparisons. They argue that because of the unique characteristics of each airport, there are insurmountable comparability problems which invalidate any measures. This has resulted in a shortage of published data. 3

5 and take equity positions in other national and global carriers. All of these factors have an impact upon airport demands and utilization. With all of the turmoil brought about by consolidation and restructuring of the air carrier industry and the desire to have an efficient aviation system (air carriers and airports) it seems reasonable that the impact of the domestic policy decisions or policies on efficient use of resources should be investigated. 3 Airports are subject to peak demands. To have perfectly satisfied customers (the airlines and their passengers) airports would need to supply sufficient runway and terminal capacity to avoid delays at even the busiest periods, allowing the airlines to maximize fleet utilization and improve load factors by providing service when their customers, the passengers, most desire. Airports, conversely, would like the airlines to spread their flight over the entire day so as to minimize runway and terminal requirements. The advent of hubbing has exacerbated this dichotomy with its concentration of arrivals and departures in narrow time bands. Even at those airports that are not used as hubs by any airline, aircraft movements are not evenly distributed. Among the other factors listed by Ashford (1984) that affect an airport s peaking characteristics are the domestic/international traffic mix as well as the long haul/short haul mix. Doganis (1992) gives an interesting illustration of the pressures being brought to bear on runway capacity. Between 1971 and 1982, a period in which the average annual increase in the number of passengers was 5.2 per cent, all of the growth was accommodated by higher load factors (33%) and increased aircraft size (67%) with the number of departures remaining unchanged. Between 1982 and 1986, however, when the annual average growth rate rose to 8.8%, the increase in demand was met almost entirely (97%) by increasing the number of departures. At most commercial airports in the US, airports and airlines enter into an airport use agreement that spells out the terms and conditions under which the airlines may use the airfield facilities and, in many cases, the terms of the leases for terminal facilities. In addition to financial arrangements, these agreements often include a delineation of the airline s rights regarding approval of airport capital development projects. The many methods of calculating airline rates and charges in use today are essentially variations of two basic methodologies: residual and compensatory. Under the residual approach, airlines agree to pay whatever fees are necessary to cover any deficit remaining after all non-airline sources of revenue (primarily concessions) have been subtracted from the cost of operating the airport. The 3 Even as mere landlords, however, the business of airport planning and management is extremely challenging. As Doganis (1992) points out: Airport authorities must invest substantial capital sums in large and immovable assets that have no alternative use, to satisfy a demand over which they have little control except indirectly. It is the airlines and not the airports who decide where and how the demand for air travel or air freight will be met. Airports merely provide a facility for bring together airlines and their potential customers. Thus, matching the provision of airport capacity with the demand while achieving and maintaining airport profitability and an adequate level of customer satisfaction is a difficult task. It is made particularly difficult because investments to expand airport capacity are lumpy, increasing effective capacity by much more than is needed in the short term, and because they must be planned long in advance. 4

6 charge that is usually used to balance the budget is the landing fee, which is weight-based. 4 In exchange for assuming this financial risk, airlines benefit from lower fees in good economic times and, through majority-in-interest clauses, are able to exercise a significant level of control over capital decisions. Under the compensatory approach, airlines are responsible for covering the costs of the facilities they use, while any shortfall caused by costs associated with concessions, public areas and vacant rentable space must be borne by the airport. The advantage to the airport under this arrangement is that the airport retains control over capital decisions and, during good years, has the potential to make a profit. (Moody s 1991). The key to comparing airports with different financial setups is disaggregation of costs and revenues since even very similar airport in terms of airlines, physical plant, and catchment area, can have substantially different net operating results, with those that use a residual structure typically having much narrower operating margins than those using a compensatory structure. Consequently, historical net revenues may not represent an airport s actual revenue-raising capacity, and a comparison of traditional financial ratios will this not reflect the flexibility or vulnerability of an airports revenue structure (Moody s 1991). 3.MEASURES OF PERFORMANCE: EFFICIENCY AND EFFECTIVENESS Efficiency which is referred to as "productive or "technical" efficiency in the economics/business literature is the relationship between inputs and outputs. Effectiveness, on the other hand, refers to the use of outputs to achieve objectives, or service consumption. There is a need to separate cost efficiency, service effectiveness and cost effectiveness and to develop performance measures. Furthermore, these performance measures must be able to go beyond simply describing efficiency or effectiveness; they must be able to assist planners and policy-makers in developing or adopting strategies which will be more successful in improving performance. While all business enterprises, whether in the public or private sector, need to continuously monitor their performance, it is especially important in the airport industry due to the specific characteristics of airports. Doganis (1992) points out that in a competitive environment, market forces will ensure that optimal performance is equated with profitability. However, the conditions under which airports operate are far from competitive. Regulatory, geographical, economic, social and political constraints all hinder direct competition between airports. At the same time, the extent to which airports can attract other airports traffic with different prices or service levels is also limited. In other words, the demand for airport services is likely to be relatively inelastic. While there is no theoretical reason why this should preclude an airport from being operated and managed in the 4A side note on the negative effect of basing landing fees on aircraft weight is that it exacerbates the capacity problem by encouraging the use of commercial airports by general aviation aircraft which are relatively light and thus pay little or nothing to land. Even at those airports that have a surcharge or minimum for GA aircraft during peak hours (e.g. Boston Logan International - $1.59 per 1,000 pounds or $50 minimum, Denver Stapelton - $1.61 per 1,000 pounds or $16 minimum) the charges are so low as to be insignificant either as a source of revenue or as a deterrent to GA aircraft (Wells 1992). 5

7 same way as a private company, Leibenstein (1978) argued that there is one special consideration as to why we should expect public enterprises to be less efficient in general than private enterprises. The natural selection argument for efficiency in private enterprise is usually not operative in the public sector. To be brief, public enterprises are not usually allowed to die. At the same time that airports should be asked to adhere to private financial standards, they must also be judged in the context of their overall goals which can be diverse, often not clearly articulated, and frequently specified (or influenced) less by professional managers than by public policy and political considerations within various sponsoring governments (U.S. DOT, 1987). Public transportation, for example, must often maintain higher levels of service than would be chosen under private industry standards, often shifting the costs of service from riders to the general taxpayer. In the case of airports, it has been argued, federal support has resulted in facilities that are not so much what is needed as what the government is willing to pay for (Wells, 1992). In their paper Reconciling Diverse Measures of Performance, Bhargava et al. (1993) found very little in the way of goal/objectives as a criterion in much of the work he reviewed, finding instead the assumption that financial measures appropriately capture the objectives of the firm. Given the unique position of airports, profit measures are an inadequate, if not totally misleading means of assessing management performance. Another reason why airports need special consideration is that they differ from most businesses in a very fundamental way. For a manufacturing firm at a constant level of production, a slowdown in sales would be reflected as an increase in inventory and not a decrease in efficiency. If the slowdown were to be anticipated and production reduced, the amount of inputs consumed would likewise be reduced leaving the output/input ratio (i.e., productivity) unchanged (ignoring possible economies of scale). With most airports, however, the factors of production (inputs) usually do not change year to year and there can be no inventory of production. Efficiency, therefore, will suffer anytime there is a slowdown in the economy or by the airlines utilizing the airport, regardless of airport management ability or efforts. A graphic example, though atypical in magnitude, is Anchorage International which has seen its concession revenue shrink from $19.5 million in 1990 to $5.4 million in 1993 due to the cessation of layovers of aircraft flying between the U.S. and Japan. Another example is Dayton International, where passenger traffic has been cut in half between the time of the USAir merger with Piedmont Aviation on August 5, 1989 and the final closure of their Dayton hub in January of Since such exogenous factors do exist, how does one account internally for a change in output? If output is down, does this mean anything under the airports control has become less productive? The answer, obviously, is no. This exogenous slowdown needs to be accounted for in order to provide an accurate measure of managerial performance. In essence we want to determine how much variation in airport performance can be attributed to managerial decision-making and initiatives and what are the important decisions or strategies within that portion of airport performance an airport manager can affect. 6

8 4.0 EFFICIENCY GENERAL Efficiency is a word that means different things to different people. There are various notions of efficiency. Productive efficiency refers to producing a given output by using the minimum possible amounts of inputs, or, producing the maximum possible output using a given amount of inputs. Allocative efficiency on the other hand refers to utilizing resources in their best use. Another notion of efficiency is related to costs of production and in this case cost efficiency may be thought of as producing a given output at the minimum possible cost, given the prices of factors of production. Productive efficiency addresses the question of whether an organization produces its output at a given level of quality at the least cost possible. Deviations from productive efficiency can result from a number of sources. The firm can be on an expansion path that is not least cost; effectively the input mix is inappropriate given the relative prices of inputs in the market and their relative productivity. Second, the firm may be on the least cost expansion path but not at the least cost point given market demand. This would occur, for example, if there were unit cost savings with output expansion arising from more efficient use of capacity. Thirdly, the managers of the firm may simply be wasteful. This later source may be a significant factor in productive [in] efficiency and has been termed x-inefficiency while the former two components have been labeled technical efficiency. X-inefficiency is the notion that people do not work as hard or as effectively as they could. The x- inefficiency concept has been refined and is seen as a part of intra-firm decision-making. While there has been considerable empirical work seeking to quantify x-inefficiency (Frantz, 1988), there is generally a presumption of single output and surplus maximization as the counter-factual basis for comparison. We would like to develop a useful indicator of performance, in particular, to distinguish technological change from efficiency change and to quantify x-inefficiency. Button and Weyman-Jones (1993b) provide an excellent description of the difference between x- inefficiency and technical inefficiency and argue the key basis of x-inefficiency is principal-agent relationships in organizations. Technical inefficiency reflects a choice of production technology or input mix that deviates from the best available. X-inefficiency is simply a failure to minimize costs given the production technology which has been selected, even if it is the non-optimal one; x-inefficiency requires, in essence, a decision by managers to deviate from cost minimization or in other words to simply be wasteful. They further point out that a significant source of x-inefficiency is the lack of external presence. Viewed as a continuous variable one can think of the level of concentration as an indicator of external presence but there may be thresholds or non-linearities which must be accounted for; that is, market structure effects. Regulation and other types of government control or affiliation may allow x- inefficiency to occur. X-inefficiency and rent-seeking behavior may be observationally indistinguishable but if one can measure x-inefficiency it may be possible to develop predictive models of agent behavior such as rent seeking. For example, it would seem important to distinguish the structural and environmental characteristics that may give rise to more or less x-inefficiency and identify those management variables over which the agent has some control. How might efficiency incentives alter this behavior and reduce x-inefficiency? 7

9 Rent seeking behavior may also give rise to x-inefficiency but the question is "is it cause or consequence? Does the opportunity for rent seeking behavior lead to x-inefficiency? The received view is that the lack of constraint leads to x-inefficiency particularly in input use. The sources of x- inefficiency are therefore the market structure, competitiveness, incorporation of or into a public bureaucracy, and the degree to which the firm is subject to regulation or controls. 5.0 DATA ENVELOPMENT ANALYSIS Development of a strategic performance measure requires that the multiple outputs and objectives be accommodated. Furthermore, it must be possible to translate the performance indicator into effective management strategies. Methods of measuring efficiency can be broadly classified into non-parametric and parametric. Non-parametric methods include indexes of partial and total factor productivity, and data envelopment analysis. The latter is essentially a linear programming based method. Parametric methods involve the estimation of neoclassical and stochastic cost and or production functions. There are four widely used methods in measuring efficiency. The data requirements for the various methods differ, as do their ability to inform managerial decisions. The use of partial productivity measures is ubiquitous and though these measures are easy to understand and compute, they can be quite misleading, because they do not reflect differences in factor prices. Consider two countries such as Canada and China. Assume that both countries use labor and capital to produce of shoes. Further assume that labor is relatively cheaper in China. If this information is illustrated using on the usual isoquant diagram, it will show that choice of technique differs across countries, with Canada using a more capital-intensive method of production. Thus, the partial productivity of labor will be higher in Canada while that of capital will be higher in China. Based on these measures, it is difficult to determine which country is more productive or efficient at producing shoes. Using partial productivity measures makes more sense if isoquants are L-shaped because changes in relative factor prices do not affect choice of technique. Partial productivity measures are also unable to handle multiple outputs. If one machine (reflecting common capital costs) produces both black shoes and brown shoes, it is not clear that constructing partial measures of capital productivity makes much sense. One way around this problem is to construct an index of total factor productivity. In addition to data on physical inputs and outputs, this measure also requires information on prices, which is used to aggregate inputs and outputs. While an index of total factor productivity does not suffer from the shortcomings of partial productivity measure, it is not very informative from a managerial viewpoint. Thus the airport manager is unable to determine if they should lay-off some workers or buy some more vehicles in order to be more productive or efficient. Extracting more information from measures of total factor productivity typically requires reliance on estimating parametric neo-classical cost or production functions. These have their own problems. For example, though in theory all flexible functional forms can approximate an unknown production technology, in practice, results may differ quite substantially. Thus choice of functional form becomes an important issue. In addition, flexibility has a price, that is, violation of theoretical consistency requirements for cost minimization. Stochastic or frontier cost and production functions suffer from the same shortcomings though unlike neoclassical functions, they are 8

10 able to distinguish technical progress (movement in the frontier) from technical inefficiency (distance from the frontier). Data Envelopment Analysis (DEA) provides a clear answer to the airport manager s problem. DEA is a [linear] programming based technique and the basic model only requires information on inputs and outputs. Indeed, this is also a major drawback of DEA, as it does not incorporate any information of factor prices or costs of production. DEA can incorporate multiple outputs and inputs; in fact, inputs and outputs can be defined in a very general manner without getting into problems of aggregation. If more of a measure is desirable it can be modeled as output and if less of something is better, it can be interpreted as input. This is an attractive feature as in many service industries such as banking, it is difficult to determine whether something is an input or an output. DEA examines productive efficiency, so it seeks to determine if a given output can be produced by using less inputs, or if given inputs can be used to obtain more output. One advantage of DEA is that it permits the determination of efficiency of firms that consume or produce inputs or outputs, which lack natural prices. This makes it especially suited to measure the productivity of non-profit organizations. DEA cannot be used to analyze or comment on cost efficiency. Firms may be technically efficient but cost inefficient. Furthermore it may be possible that firms ranked technically inefficient by DEA may be able to produce their outputs at a lower cost than those ranked as technically efficient. While it is clearly important for organizations to operate in a productively efficient manner by maximizing outputs from their given inputs, it is also true that most firms (including non-profit organizations) operate with limited, and often severely constrained financial resources. DEA provides a scalar measure of relative efficiency by comparing the efficiency achieved by a decision making unit (DMU) with the efficiency obtained by similar DMUs. The method allows us to obtain a well defined relation between outputs and inputs. In the case of a single output this relation corresponds to a production function in which the output is maximal for the indicated inputs. In the more general case of multiple outputs this relation can be defined as an efficient production possibility surface or frontier. As this production possibility surface or frontier is derived from empirical observations, it measures the relative efficiency of DMUs which can be obtained with the existing technology or management strategy. Technological or managerial change can be evaluated by considering each set of values for different time periods for the same DMU as separate entities (each set of values as a different DMU). If there is a significant change in technology or management strategies this will be reflected in a change in the production possibility surface. Figure 1 below attempts to illustrate the basic notion behind DEA. 5 First consider the five DMUs labeled p1 through p5. All DMUs produce one unit of output using different quantities of the two inputs x1 and x2. Since output is unity, the axes can be interpreted as measuring either quantities of the two factors of production or as input-output coefficients. The efficiency frontier in this case is the locus of the points p5, p1 and p3. This locus can be thought of as a unit isoquant. Since it is piece-wise linear, it 5 Figure 1 is a modified version of Figure I in Bedard (1985). 9

11 implies that the marginal rate of technical substitution is fixed along each of the two linear segments. A point like p4 is inefficient compared to p1, since the former uses one more unit of x2 to produce the same output as the latter. By similar reasoning, p2 is an inefficient point. The base DEA model expresses inefficiency purely in terms of output slacks and excess inputs. The airport manager s question is answered simply as use one less unit of input 1. Thus both DMUs p2 and p4 can become efficient by eliminating excess inputs (as shown by the arrows). However, this is not the only way of getting to the frontier. DMU p4 can reach the frontier by a proportional reduction in the use of both factors of production. The input-oriented DEA model allows the determination of the amount of proportional reduction required. Thus p4 can move a long the ray through the origin to P4 which lies in the cone formed by the two reference points p1 and p5. P4 is a weighted average of p5 and p1 and the weights are positive since it lies in the cone. Orientation in DEA models imposes restrictions on the production technology, as does piece-wise linearity. The usual smooth, twice differentiable, convex to the origin isoquants yield unique and interior solutions. Corners, kinks and holes typically lead to problems. In addition, moving down a ray from the origin implies that choice of technique (or relative input intensity) does not change. In a situation where there are various factors of production, whether a proportional reduction is possible depends upon the production technology. In DEA relative efficiency is quantified using the distance of a particular DMU from the frontier. The measure is a ratio of radial distances thus the efficiency score for p4 would be O p4/o P4. According to this measure, an efficient unit would be accorded a score of unity. Figure 1 The efficient DMUs p3 and p5 form a cone which is also sometimes referred to as the base region. How is a DMU such as p6 that lies outside the base region evaluated? DMU p6 cannot be compared to a weighted average of two efficient DMUs as was the case for p2 and p4. Clearly, compared to p3, p6 is inefficient as it uses an extra unit of input x1. The efficient region is extended in parallel to the x- axis by joining the points p3 and p6, but the two cannot be given an efficiency score of 1. In order to 10

12 get around this problem, the excess input at p6 is accorded a positive weight in the optimization problem. The weight is a small positive number referred to as the non-archimedean epsilon. Thus in base models projected input used equals actual input use less any excess inputs. In the input oriented model there are two sources of inefficiency, one that can be eliminated by a proportional reduction in input use and a residual or excess input. 6 Total inefficiency equals residual inefficiency at the point p6. 7 In view of DMUs such as p6, a DMU is efficient if its efficiency score is unity and there are no output slacks or excess inputs. Despite its many attractive features DEA has many shortcomings, the main ones being that it imposes a-priori restrictions on the production technology, it does not take into account costs of production and prices of factors of production and that it is non-statistical and therefore no hypotheses can be tested DATA AND METHOD The data used in the DEA estimation represents a cross section of airports in the US which have differing ownership, financing and operational characteristics. A brief discussion of some of the variables is useful. Airport financing falls into two categories; residual or compensatory. At a residual financed airport, carriers have the responsibility for meeting any shortfall which occurs after all revenues have been collected from various sources. At a compensatory financed airport, the airport has full responsibility and the carriers simply pay a negotiated fee. The residual airport would be closer to a privatized airport while the compensatory would be akin to a public airport. Airport managers can affect the efficiency of their airport by allowing the runways and facilities to be used in different manners. The access to gates, for example, can affect the number of movements and passengers handled; both measures of output. Noise management, a key issue at major US airports, can be handled in a number of different ways. Each noise management strategy can have a different affect on the number of aircraft using the airport, runways and gates, in a given period of time. A noise budget allows a carrier to emit an aggregate amount of noise. If it uses noisy aircraft it can have fewer flights while use of newer quieter aircraft allows more flights. In the analysis we develop a database which included the top 30 airports in the United States. Our sample uses a subset of 21 of these airports. Information for each airport was assembled for the period We have 114 observations organized in a panel. 9 The first part of our analysis involves deriving a scalar measure of relative efficiency for 114 DMUs. To do this, we define airports as producing two separate classes of services, these being terminal services and movements. Terminal services are modeled as having two outputs number of passengers and pounds of cargo and six inputs 6 Similarly, in the output oriented model the two sources of inefficiency are that which can be eliminated by a proportional augmentation of all outputs and residual inefficiency or output slack. 7 It is also possible to derive an efficiency measure that converts both types of inefficiency into that which can be eliminated by a proportional reduction in all inputs. Using this would imply that p6 move to a point like P6. 8 Other shortcomings reported in the literature include the fact that all DMUs may be ranked as efficient if the number of outputs and inputs approximates the number of DMUs. In addition, proponents of this technique suggest that highly correlated variables must not be included in the optimization, even though this is a non-statistical technique. 9 Data for Dayton Airport are only available for the period

13 number of runways, number of gates, terminal area, number of employees, number of baggage collection belts and number of public parking spots. Movements have two outputs air carrier movements and commuter movements and four inputs airport area, number of runways, runway area and number of employees. 10 Movements are assumed to be produced under constant returns to scale (Gillen, 1994) where as returns to scale are variable in the production of terminal services. 7.0 DEA RESULTS The models used here to obtain a scalar measure of relative efficiency are the output-oriented models. For our purposes, orientation is not of crucial importance and our decision is based on convenience. 11 The output-oriented model fits in more naturally with the second stage of our analysis, which estimates a Tobit model. Movements are modeled under the assumption of constant returns to scale. The model is as follows for a given DMU: 12 min g µ, ν = v X 0 0 s. t. µ Y = 1 µ Y + v X µ 0 v 0 (1) where X 0 is a vector of inputs, Y 0 is a vector of outputs and, µ and ν are weights to be determined by solving the programming problem. Terminal services are modeled using the following variable returns to scale model: 13 min µ, ν, v 0 = v X + ν s. t. µ Y0 = 1 µ Y + v X + v ι 0 f µ 0 v 0 v 0 free (2) 10 We do not have disaggregated data on employees involved in the provision of terminal services and those involved in producing movements. In addition, we do not have data on fuel and other inputs such as materials. 11 Note that if a unit is inefficient according to the input-oriented model, it will also be so according to the outputoriented model. 12 See equation 4 in Charnes, Cooper and Rhodes (1978). 13 See page 34 in Charnes et al., (1994). 12

14 where v 0 is the measure of returns to scale or the constant term in the hyperplane and ι is a unit vector. The measure of efficiency we report in Table 1 represents two types of inefficiency, that which can be removed by a proportional augmentation of all outputs and residual inefficiency. A value of 1 implies that the DMU is efficient and efficiency score increases with increase in relative inefficiency. Table 2 and Table 3 show changes in efficiency between the years 1989 and 93. An examination of Table 1 reveals the following outcomes for the different airports. In the majority of cases the index is falling implying an increase in efficiency (or reduction in inefficiency). This can reflect in part the emergence from the recession in the late 1980's. However there are some interesting differences. Anchorage, for example shows in increase in inefficiency for terminals but a reduction in inefficiency for movements. This reflects the reduction in passengers due to long haul jets being able to bypass this technical stop as well as the growth of hubbing out of the west coast gateways such as San Francisco and Seattle. Table 1 Efficiency Measures from DEA Calculation Airport DMU Terminals Movements Airport DMU Terminals Movements Anchorage ANC Memphis MEM ANC MEM ANC Milwaukee MKE ANC MKE ANC MKE Atlanta ALT MKE ALT MKE ALT Minneapolis - St. MSP Paul ALT MSP ALT MSP Boston BOS MSP BOS MSP BOS Ontario ONT BOS ONT BOS ONT Baltimore / BWI ONT Washington BWI ONT BWI Phoenix PHX BWI PHX BWI PHX Charlotte / Douglas CLT PHX CLT PHX CLT Portland PDX CLT PDX CLT PDX Chicago MDW PDX

15 MDW PDX MDW St. Louis STL MDW STL MDW STL Cincinnati / Northern CVG STL Kentucky CVG STL CVG Salt Lake City SLC CVG SLC CVG SLC Cleveland Int'l CLE SLC CLE SLC CLE San Diego SAN CLE SAN CLE SAN Dayton DAY SAN DAY SAN

16 Table 1 - cont'd DAY San Francisco SFO DAY SFO Fort Lauderdale FLL SFO FLL SFO FLL SFO FLL San Jose SJC FLL SJC Kansas City MCI SJC MCI SJC MCI SJC MCI Seattle - Tacoma SEA MCI SEA Memphis MEM SEA MEM SEA MEM SEA Figure 2 and Figure 3 illustrate the Terminal and Airside efficiency measures contained in Table 1. They provide a better display of both differences between terminal and airside efficiency in an airport but also relative efficiency across airports. The differences are quite striking. Figure Terminal Efficiency Termeff Efficiency Index ANC-89 ALT-90 BOS-91 BWI-92 CLT-93 CVG-89 CLE-90 DAY-91 FLL-93 MKE-89 MSP-90 ONT-91 PHX-92 PDX-93 SAN-89 SFO-90 SJC-91 SEA-92 Airport and Year 15

17 Figure 3 Airside Efficiency Index Airside Efficiency Index ANC-89 ALT-90 BOS-91 BWI-92 CLT-93 CVG-89 CLE-90 DAY-91 FLL-93 MKE-89 MSP-90 ONT-91 PHX-92 PDX-93 SAN-89 SFO-90 SJC-91 SEA-92 Airport and Year Boston's Logan Airport has terminal inefficiency rising and movements inefficiency falling. However, the relative values of efficiency are very different. Terminal efficiency is quite high while airside efficiency was relatively poor but improved markedly over five years. As is true of Salt Lake City but in most of cases the inefficiencies are moving in the same direction. The hub airports exhibit some loose similarities. It appears as hubbing rises the terminal efficiency improves but movements efficiency decreases most likely due to the feed from small commuter aircraft. San Francisco exhibits this result but the level of inefficiency on the movements side is quite startling. San Jose illustrates the impact of American developing their hub there with terminal efficiency rising and movement efficiency gradually falling. Summary tables of the changes in terminal and movements efficiency between were constructed. These results are shown in Table 2 and Table 3. Terminals doing worse are those in which airlines have moved elsewhere or the economy has not picked up. Terminals doing better reflect new and expanding hub operations, and traffic growth from economic growth above the average (e.g. San Diego). Those doing the same are well established hub airports. Table 3 exhibits a clear bifurcation in the results for changes in movement efficiency. Both hub and nonhub airports showed degradation in efficiency. Those airports doing better were those which tended to have more homogenous traffic and fewer wide bodied aircraft. 16

18 Table 2 Terminals -Efficiency in 1993 compared to 1989 Worse Better Same Anchorage Atlanta Chicago Baltimore / Washington Boston Phoenix Dayton Charlotte/Douglas San Francisco Fort Lauderdale Cincinnati / Northern Kentucky Seattle - Tacoma Kansas City Cleveland International Milwaukee Memphis Portland Minneapolis-St. Paul San Jose Ontario St. Louis Salt Lake City San Diego Notes: Bold indicates efficiency of 1 in 1993, bold and italics indicates efficiency of 1 in 1989 and Data for Dayton are for 1989 and Table 3 Movements -Efficiency in 1993 compared to 1989 Worse Better Same Atlanta Anchorage Charlotte/Douglas Baltimore / Washington Boston Chicago Cincinnati / Northern Kentucky Dayton Cleveland International Kansas City Fort Lauderdale Milwaukee Memphis Ontario Minneapolis-St. Paul St. Louis Phoenix Salt Lake City Portland San Francisco San Diego San Jose Seattle-Tacoma Notes: Bold indicates efficiency of 1 in 1993, bold and italics indicates efficiency of 1 in 1989 and Data for Dayton are for 1989 and TOBIT RESULTS At this point we have addressed only the first part of our questions. We have a measure of relative efficiency and we can rank airports by year and to each other by airside as well as terminal efficiency. This is one of the significant benefits of the DEA measure in that it is useful for both inter-temporal as well as cross-sectional comparisons. However, it is also our purpose to explain the variation in performance 17

19 measures both over time and across airports. To do this we develop a set of variables for each airport which we roughly classify as structural (e.g. number of runways, land area and numbers of gates), environmental variables (e.g. annual service volume or ASV) and managerial variables (e.g. use of gates, financing regime, noise strategies, proportion of GA traffic, existence of hubs at airport). We are interested in two issues. What proportion of the variation in efficiency can be explained by the managerial variables and which managerial variables are most important in affecting this proportion of efficiency? To obtain the true or net efficiency index we undertake a second stage of analysis. In this we utilize the efficiency index generated from the DEA methodology and use it as the dependent variable in a regression in order to identify the variables which affect efficiency. The DEA efficiency measure has a lower bound of 1 or, as we have done here, if one take natural logs of the efficiency measure, a lower bound of zero. Thus we use the censored regression or Tobit model. The standard Tobit model can be represented as follows for observation i: y y * i i i i = β x + ε * = 0 if y 0 * * y = y if y > 0 i i i i (3) The estimated coefficients of the Tobit model do not provide the marginal effects. The marginal effect for variable j is derived as follows: 14 E y i x j = β Φ (4) j i In addition, conditional marginal effects are also reported. The conditional marginal effect for variable j is derived as: [ > 0] E y y i x i j β x = βj 1 σ φi φi Φ Φ i i i 2 (5) In the above two equations, φ i and Φ i are the respective probability and cumulative density functions of a standard normal variable evaluated at β x i / σ. Since a unique marginal (and marginal conditional) effect can be calculated for each observation, we report the mean of the marginal. effects. The results of the estimation are reported in Table 4 and 14 See page 963 in Greene (1997) or page 799 in Judge et al. (1988). Note that this is also a good approximation for the marginal effects of dummy variables. 18

20 Table 5, for movements and terminals respectively. Table 4 reports the Tobit regression results for movements (airside) efficiency and Table 5 does the same for Terminal efficiency explanations. The marginal and conditional marginal effects are reported in the last two columns respectively. The two measures are quite similar but the conditional marginal effect is always less. In the movements efficiency regression we are trying to determine which managerial and non-managerial variables have the most impact on the efficiency with which airside can yield aircraft movements. Since the efficiency index has a lower limit of 1, a higher number implies more inefficiency. Once you take natural logs the lower limit is zero. So a positive coefficient implies inefficiency is going up. There are four sets of variables; dummy variables for the time period, dummy variables for hub airports, noise strategy variables and management operational and investment variables. The time dummies are not significant so there is no apparent trend in the data or anomalies in one year or another. The hub dummies show clearly that primary hubs increase the efficient use of the airside. However, there are differences among hubs. At Minneapolis-St. Paul, the hub for Northwest Airlines and Atlanta, the hub for Delta Airlines we see quite similar [positive] affects on efficiency but San Francisco a second hub for United shows a smaller impact. San Francisco is also a gateway and tends to have a greater proportion of wide bodies traffic. The former hubs have a large number of feeder flights hence increasing the level of efficiency. The noise strategy variables are all significant save for limiting operations. Using a preferential flight path or limiting the hours of operation reduces inefficiency while all the remaining noise management strategies tend to increase inefficiency. It appears counterintuitive that limiting the hours an airport is open could reduce inefficiency. However, two explanations are offered. First, if limits on hours and primary hubs are correlated this could explain the result. Airlines will not bring complexes into the airport in the middle of the night to hub since there would be little demand. The second explanation is that given limitations on night use the airport utilizes its' resources very efficiently while those without such constraints do not feel compelled to operate with such efficiency. The management variables contain a broad cross section of strategies and investments. Increasing the number of boarding gates can increase airside efficiency while increasing the number of runways has no impact and increasing the area of airside capacity has the opposite effect, reducing efficiency. Given scarce resources, the airport manager can obtain more cost efficiencies from investing in terminal gates than expanding the size of the airport or increasing the number of runways. Airports that have a number of airlines which hub at the airport have greater cost efficiencies. Some airport managers feel hubs can place an airport in jeopardy but the evidence implies there may be a cost to such a position. The one variable that stands out is the proportion of general aviation traffic at the airport. An increase in this type of traffic has a sizable impact on increasing inefficiency. Such evidence provides support for the introduction of peak pricing for small GA aircraft at major airports. 19