Air passenger traffic and local employment: Evidence from Turkey

By providing fast and safe transportation opportunities, air transportation plays an important social and economic role in the development of cities and the nation as a whole. Using Census 2000 data from 31 provinces, this study attempts to figure out the possible effects of air passenger traffic on the local composition of employment in Turkey. Its results suggest that employment in many industries and occupations benefits from the existence of scheduled air passenger traffic. More concretely, we found that a 10% point increase in air passenger traffic per capita would generate 15,013 service-related jobs throughout the 31 provinces analysed.


Introduction
Airports are critical infrastructures that are believed to be essential for economic development because they substantially increase the accessibility of the regions they serve. However, except for several hub airports (e.g., İstanbul Atatürk International Airport and İstanbul Sabiha Gökçen International Airport) and those serving tourist destinations (e.g., Antalya International Airport, Muğla Bodrum-Milas International Airport, and Muğla Dalaman International Airport), airport business is mostly financially unviable in Turkey like the rest of the world. While large airports tend to make profits, smaller and regional airports hardly recover their costs (Doganis, 1992: p.5). The airport industry is characterized by substantial initial infrastructure investment and the operating revenues generally fail to recover annual operating expenses. Therefore, most of the regional and small airports need financial contributions mostly in the form of cross-subsidization from larger profitable airports (Ohta, 1999;Hooper, 2002;Lipovich, 2008;Reinhold et al, 2010).
But airports offer benefits beyond those displayed on simple balance sheets and income statements. Major characteristics of air transportation such as speed, safety, and reliability help sustain critical economic connections. Service industries, where face-to-face interaction is necessary, heavily rely on air transportation to rapidly reach new markets and establish business linkages with customers. In the markets where the life span of the products is too short or the price of the products is relatively Özcan Air passenger traffic and local employment: evidence from Turkey high when compared to unit logistics costs, manufacturing firms need air cargo services to place their products in the shortest and most reliable manner. For Turkey, the role of air transportation is even bigger. It arises as the only viable option, especially for long-distance intercity passenger trips, given the limited scope of high-speed rail infrastructure and the low-quality pavements in the majority of the road network.
For governmental decision makers, to initiate a new green-field, low-traffic small airport project may be challenging, given the poor financial statements of such airports. To make a proper assessment, decision makers must know the economic contribution of airports, the positive economic effects that we cannot identify solely by analysing the financial statements. In other words, operating an airport should be socially profitable meaning that the decision to launch a new airport project that is a candidate to be financially unprofitable or to keep the operations of an already money-losing one should be justified with the economic contributions of the flights.
The goal of this study is to define the effect of air transportation on the local employment in terms of industries and occupations by using econometric models. In Section 2, we will summarize the existing literature on the linkage between air transportation and its effect on economic development and job creation. In Section 3, we will explain our methodology and data. In Section 4, we will discuss our findings; and in section 5, we will present the conclusions.

Literature review
The effect of air transportation on local economies arises in two ways, namely (i) expenditure effects and (ii) transportation effect. Expenditure effects are the benefits that come from wages of employees in the various subsectors of the air transportation industry 2 (for example, air traffic control, airlines, ground handling, maintenance-repair-overhaul, and airport concessions) and expenditures during the construction phase of the airports, such as construction labours' wages and payments to the suppliers of construction materials. However, it is the transportation effect that creates the actual and long-term benefits because the real benefit of transportation investments and services emerge when they are able to reduce the transportation costs, with this reduction in cost originating from the improved accessibility, reliability, and safety and the reduced travel time and emissions (Taylor and Samples, 2002). From the point of view of air transportation, the transportation effect works when the existence of air services reduces the costs of the firms, which eventually increases their competitiveness.
The effect of air traffic on local employment has been attracting considerable academic interest. Hewings et al (1997) calculated that a capacity increase at O'Hare and Midway airports would lead to a total job increase of 522,000 in the Chicago region. Hakfoort et al (2001) estimated Amsterdam Airport Schiphol had a total multiplier of 2, implying that each new airport job would generate an additional job in the form of indirect and induced employment. Using the data from the years 1950 and 1980, Irwin and Kasarda (1991) found that employment growth was stimulated by air passenger traffic in 104 metropolitan areas in the United States (US). Similarly, Goetz (1992) showed that a positive correlation existed between per capita air passenger traffic and both previous and subsequent urban growth in the 50 largest metropolitan areas in US between 1950and 1987. Button and Taylor (2000 suggested that the number of high-technology jobs (including but not limited to Özcan Air passenger traffic and local employment: evidence from Turkey information technology software and services, telecommunication services, and advanced materials) increased as the number of international destinations (European airports in this case) and international passenger traffic increased. Debbage (1999) showed that airports with high traffic growth were more likely to attract parallel growth in administrative and auxiliary jobs in the US Carolinas. Debbage and Delk (2001) tested whether a linkage existed between air passenger traffic and administrative and auxiliary employment and found a high correlation (0.84, 0.83, and 0.83 for 1973, 1983, and 1996, respectively) between them for the 50 largest urban areas in the US. In a similar study on 98 metropolitan areas from the US, Alkaabi and Debbage (2007) documented that a correlation coefficient of 0.90, 0.84, 0.68, and 0.39 existed between air passenger traffic and (i) professional, scientific, and technical establishments; (ii) high-technology establishments; (iii) professional, scientific, and technical employment; and (iv) high-technology employment, respectively. Warren (2007) estimated that counties with commercial air service were more likely to experience higher income, employment, population, dividends, interest, and rent. Rasker et al (2009) found that as counties' distance to a major airport decreased, they tended to have (i) higher per capita income, (ii) more services and professional jobs, (iii) higher mean earning per job, and (iv) lower degree of specialization. The findings of Button et al (2010) suggested that income per capita in the neighbouring area could increase by 0.18-0.4% due to an increase of 10% in air passenger traffic.
One common problem of such comparable studies is that it is difficult to determine the direction of causality between the air traffic and its effect on local employment. On the one hand, the availability of scheduled air service can stimulate local economic growth and employment as it reduces transportation costs because the higher the frequency of flights, the greater the number of destinations served, the greater the transportation benefit. However, increased economic activity, population, and employment, better educational attainment and changes in employment mix can also generate increased air traffic and are able to shape the air traffic concentration as well.
The analysis of Bauer (1987) showed that the greater the population and per capita income, the greater the revenue passenger enplanements would be. Huston and Butler (1991) showed that factors like population, income level, climatic conditions, and being a tourist destination and a business centre were likely to determine whether an airport was a hub or not. In their study to analyse the factors distinguishing between major and minor air traffic markets, Liu et al (2006) found that the population, percentage of workforce in professional, scientific and technical services and management activities, distance to nearest major market, and percentage of workforce in tourism tended to determine whether a metropolitan area was a major air traffic market or not. Fernandes and Pacheco (2010) found that a "unidirectional Granger causal relationship from economic growth to domestic air transport demand" existed in Brazil for the period of 1966-2006. Dobruszkes et al (2011), for example, analysed the factors determining the air traffic volumes of European metropolitan areas. Their analysis revealed that having higher GDP, hosting headquarters of major companies, being a tourism region, and being farther from the nearest main air market tended to increase air traffic of a European metropolitan area.
Such a chicken-and-egg problem makes econometric estimation of this ambiguous relationship between air passenger traffic and local employment challenging. To overcome this problem, scholars apply different techniques. Button et al (1999) analysed the impact of a hub airport on the high-technology employment of a metropolitan statistical area (MSA) that it served, using data across 321 MSAs in the US. Their results showed that a region with a hub airport tended to have an added high-technology Özcan Air passenger traffic and local employment: evidence from Turkey employment of more than 12,000. Button et al (1999) also controlled for the causality using the Granger causality test, and their further analysis confirmed the direction of causality. The empirical study of Brueckner (2003), using data from 91 US metropolitan areas for the year 1996, showed that a 10% increase in air passenger traffic created approximately a 1% increase in employment in service-related industries. Finally, the findings of Green (2007) suggested that both population and employment growth could be predicted using air passenger traffic. Both Brueckner (2003) and Green (2007) used two-stage least-squares estimation to eliminate any causality problem.

Model
The aim of this study is to model the effect of air passenger traffic on local employment. One can assume that the local employment is a function of air traffic, as the existence of air traffic will stimulate local employment through reduced transportation costs. However, it is also expected that the higher employment and the larger the economy of a geographical area (country, state, metropolitan area, province, or city) will lead to higher air traffic figures. So air traffic can also be formulated as a function of local employment. To deal with this problem of causality, this study employed two-stage least-squares (2SLS) estimation, where we used two instrumental variables, joint-use-airport and runway, where joint-use-airport is a dummy variable and equal to 1 for provinces whose airports are joint-use airports (military airports used by civilian commercial flights).
runway is the total length of the runway(s), in natural logs, of the provinces' airport(s). If an airport has more than one runway, runway equals to the sum of their lengths.
Among 31 Turkish provinces analysed, 12 have joint-use airports. These joint-use airports are military airports owned by either the Turkish Air Force or the Turkish Land Forces. General Directorate of State Airports Authority (GDSAA), the state-owned enterprise responsible for the management of the airports and air navigation systems in Turkey, operates civilian facilities such as passenger terminals within these joint-use airports to enable civilian flights. Due to the facts that civilian flights should share the fundamental infrastructures and services (such as runway, aprons, and air traffic control) with the military flights and that the military imposes some restrictions on the civilian flights (in terms of slots, operation schedules, and cockpit crew), air passenger traffic at jointuse airports mostly fails to achieve its full potential; and we can argue that provinces having a jointuse airport tend to have lower air passenger traffic.
On the other hand, runway lengths determine the type of aircrafts able to use the airports. Airports with longer runways can serve wider-body aircrafts and accordingly can handle more air passenger traffic with the same number of flights. While runways lengths were designed based on the thencurrent traffic forecasts and aircraft types at the time of the initial investment, factors such as limited financial resources, natural obstacles, lack of necessary land, and growing environmental concerns tend to prevent or delay the necessary rehabilitation investments (for lengthening the existing runways or building a newer higher-capacity runway) which would enable the shift to wider-body aircrafts and eventually help handle the increased air passenger traffic. As a result, airports lacking the needed runway capacity may fail to achieve its full potential as in the case of joint-use airports. Özcan Air passenger traffic and local employment: evidence from Turkey Based on such characteristics, we believe that both joint-use-airports and runway lengths can be suitable instruments.
As a result, we modelled the first-stage of the estimation as 3 : air passenger traffic = f (joint-use-airport, runway) After running the initial step of the 2SLS estimation, we constructed and ran our model using the following (per capita) specification 4 : local employment = f (air passenger traffic, population, labourforce, GDP) where local employment (per capita) is the ratio of employment figure as of 2000 in each province, in terms of one of the 19 various industry and occupation classifications listed in Table 1

Data
For the air passenger traffic per capita variable, we used the total air passenger traffic at each Turkish province in the year 2000 6 . In the year 2000, among 81 Turkish provinces, 35 had air traffic service at 37 airports. However, we excluded three provinces (Çanakkale, Sinop, and Tokat) each having annual air passenger traffic of fewer than 1,000 passengers per year and one province (Tekirdağ) 3 We have checked for multicollinearity using the variance inflation factor (VIF) scores. For the two independent variables, both the highest and the average VIF score is 1.01. To overcome a possible problem of heteroscedasticity, we used robust standard errors in our regression estimations. We used natural logarithms of runway length to obtain a normal distribution. 4 We have checked for multicollinearity using the variance inflation factor (VIF) scores. For the four independent variables, the highest VIF score is 6.89, and the average VIF score is 4.61. To overcome a possible problem of heteroscedasticity, we used robust standard errors in our regression estimations. We used natural logarithms of population and per capita GDP to obtain a normal distribution. 5 The model assumes that provincial borders limit the airport's catchment areas and no passenger leakage exists between the airports of two neighbour provinces. 6 We preferred to use air passenger demand instead of supply-side figures such as the number of passenger flights or the number of seats supplied since these statistics do not exist in Turkey. One drawback of the air passenger traffic statistics for year 2000 is that they lack connecting air passenger traffic figures. Among 33 airports in 31 provinces in 2000, only two airports, İstanbul Atatürk Airport and Ankara Esenboğa Airport, were hub airports with connecting air passenger traffic. One implication of not being able to adjust the traffic data for connecting passengers is that provinces with these hub airports get higher air passenger traffic per capita values than they should. Özcan Air passenger traffic and local employment: evidence from Turkey having only international charter flights from our data set. As a result, our data set consists of 31 provinces. All of these provinces had scheduled domestic air traffic and 17 of them also experienced international flights. For Muğla and Balıkesir provinces, where each province has two separate operating airports, we simply added the annual total passenger traffic of these two airports to get a single aggregate value.
Turkish censuses classify nine industries and seven occupations among all employment data. In addition to these nine industries and seven occupations, we created three aggregate industry variables as goods-related employment, service-related employment, and nonagricultural employment where  goods-related employment (GOODS) is the total employment in mining and quarrying; the manufacturing industry; construction,  service-related employment (SERVICE) is the total employment in electricity, gas, and water; wholesale and retail trade, restaurants and hotels; transportation, communication, and storage; finance, insurance, real estate, and business services; community, social, and personnel services,  nonagricultural employment (NON-AGR) is the sum of goods-related employment and servicerelated employment. Table 1 describes the 19 employment-related variables, 9 industries, 7 occupations, and 3 aggregate industry variables. Because we use per capita employment as the dependent variable in the second stages of 2SLS estimations, we then divided the employment in each employment classification by the population for each province to get the per capita figures. Table 2 presents the summary statistics of all variables used in our analyses. We would like to thank Dr. Frederic Dobruszkes for his help in preparing the maps. 8 We also calculated the correlation coefficients between the employment in the employment classifications described at Table 1   The major hypothesis of this study is that air passenger traffic can stimulate local employment in most industries and occupations; and accordingly, we suppose air passenger traffic per capita should get a positive coefficient. In regards to population, we think that population of a province can change the composition of the local employment, as some rural and low-population provinces may heavily rely on agricultural activities whereas bigger cities can easily attract bigger businesses and accordingly more white-collar workers. So, the independent variable population should get both positive and negative coefficients, depending on the employment variable used in the regression estimation. As the percentage of the population able to work will directly affect the employment, we anticipate that labourforce should get a positive coefficient. On the other hand, we believe that income levels of the provinces tend to shape their employment mixes and developed provinces are generally more successful in attracting high value-added industries and occupations necessitating qualified human resources; and as a result, we expect that being a high-income province should increase employment in such industries and occupations. The independent variable per capita GDP should therefore get a positive coefficient for such industries and occupations. Özcan Air passenger traffic and local employment: evidence from Turkey   The study might have two possible weaknesses. One weakness may arise due to the fact that the data set is created using the Census 2000, as it is the latest census having the detailed information we needed. It may be relatively old, but we believe that the fundamentals of the relations in the model have remained valid during the years since 2000. Another weakness may stem from the use of relatively fewer observations, as compared with similar studies. Comparable studies focused on US, where (due to its geography, economic activity, and population) one might obtain a considerable number of observations. But we believe that 31 observations from Turkey are adequate to make statistical interpretations.
The most significant advantage of the data set is that it includes so many occupations and industries through which a wide range of analyses and interpretations can be made. Comparable studies focus on specific jobs (like high-technology and managerial jobs) or focus on specific industries (like service industry). In contrast, this paper is able to identify and analyse 19 different groups of occupations and industries, which provides us a large room for conclusions.

Empirical results
The regression results for the first stage of 2SLS estimation (R 2 = 0.443) are as follows: air passenger traffic per capita = -0.790 joint-use-airport + 1.6574 runway -12.623 where the t-statistics (respectively 2.52**;2.37** and 2.31 ***) are based on robust standard errors. The ***, ** and * stand for significance levels at 1%, 5%, and 10%, respectively. The results of the ordinary least-squares (OLS) and 2SLS regressions using joint-use-airport and runway as instrumental variables are shown at Table 3, Table 4, and Table 5. Table 3 presents the results of regressions when local employment per capita in each industry is used as the dependent variable, respectively. In terms of industries, a 1% point increase in air passenger traffic per capita of a province leads to 0.0019% increase in the per capita employment of the construction industry (pcCONS), 0.0098% increase in the per capita employment in wholesale and retail trade; restaurants and hotels (pcWHOLE), 0.0025% increase in the per capita employment in transportation, communication, and storage (pcTRANS), and 0.0026% increase in the per capita employment of finance, insurance, real estate, and business services activities (pcFINAN). Regarding aggregate industries (Table 4), on the other hand, a 1% point increase in the air passenger traffic per capita of a province creates a 0.0250% increase in the per capita employment of aggregate servicerelated industries (pcSERVICE) and a 0.0145% increase in the per capita employment of aggregate nonagricultural industries (pcNON-AGR), which is the sum of the per capita employment of aggregate goods-related industries (pcGOODS) and that of aggregate service-related industries (pcSERVICE).
In terms of occupations (Table 5), a 1% point increase in air passenger traffic per capita of a province leads to a 0.0031% increase in the per capita number of clerical and related workers (pcCLE), a 0.0016% increase in the per capita number of commercial and sales workers (pcCOMME), and a 0.0084 increase in the per capita number of service workers (pcSERV). Özcan Air passenger traffic and local employment: evidence from Turkey  (1) Population figures in natural logs; (2) t-statistics in parenthesis based on robust regressions; (3) ***, **, and * stand for significance levels at 1%, 5%, and 10%, respectively; (4) number of observations = 31. Özcan Air passenger traffic and local employment: evidence from Turkey (2) t-statistics in parenthesis based on robust regressions; (3) ***, **, and * stand for significance levels at 1%, 5%, and 10%, respectively; (4) number of observations = 31. Özcan Air passenger traffic and local employment: evidence from Turkey  (2) t-statistics in parenthesis based on robust regressions; (3) ***, **, and * stand for significance levels at 1%, 5%, and 10%, respectively; (4) number of observations = 31 Özcan Air passenger traffic and local employment: evidence from Turkey These results support the expectation that air traffic plays a positive role in creating new jobs in many industries and occupations. In addition to the air passenger traffic per capita variable, which is the focus of this study, we have also obtained interesting results about our control variables. Regarding population, we found that increasing population tends to decrease local employment per capita in (i) agriculture, hunting, forestry, and fishing activities (pcAGR); (ii) service workers (pcSERV); and (iii) agricultural, animal husbandry, forestry workers, fishers, and hunters (pcAGRICU). Although the negative effect of population on pcAGR and pcAGRICU (as smaller and rural provinces may rely more on agricultural activities and occupations) is quite predictable, our finding that increasing population decreases pcSERVICE is surprising. Though higher population decreases pcAGR, pcSERVICE, and pcAGRICU, it stimulates pcMANUF, pcFINAN, pcGOODS, pcADM, pcCLE, and pcCOMME-implying that bigger and more populous provinces demand more employment at these industries and occupations. As for labourforce, which is a proxy for the percentage of the population able to work, our results reveal that pcCONS, pcGOODS, pcSCI, pcADM, pcCLE, pcCOMME, and pcNON-AGRICU tend to increase with increasing air passenger traffic per capita as we predicted before. Finally, per capita GDP, which controls the regional disparities, had statistically significant and positive coefficients, in parallel with our previous predictions, for pcMANUF, pcWHOLE, pcTRANS, pcGOODS, pcNON-AGR, pcADM, pcCOMME, and pcSERVmeaning that in developed provinces the ratios of total employment in these industries and occupations to total population tend to increase.
So far, we have talked about increases and decreases of employment (and per capita employment) in terms of percentages, which may seem a little abstract. To put our findings in a more concrete way, we also made a basic scenario analysis involving three scenarios. We simply tried to figure out, holding our control variables (population, labourforce, and per capita GDP) constant, how the employment in CONS, WHOLE, TRANS, FINAN, SERVICE, NON-AGR, CLE, COMME, and SERV (the employment classifications affected statistically significantly by air passenger traffic per capita) would change if the air passenger traffic per capita would increase by 3%, 5%, and 10% points, respectively. Table 6, which is divided into four panels, shows the results of our scenario analysis. Panel A represents the base scenario with zero growth in air passenger traffic per capita, while Panel B, Panel C, and Panel D represent 3% point growth, 5% point growth, and 10% point growth scenarios, respectively. In addition to the aggregate statistics of the total of 31 provinces forming our data set, we have also analysed 3 individual provinces (those having the maximum, median, and minimum air passenger traffic per capita) to better understand how individual provinces could be affected by the changes in air passenger traffic per capita.  (GDSAA, 2001). In Panel B, we took the case in which air passenger traffic per capita increased by 3% point, holding the control variables used in our estimations constant. We first revised pcCONS, pcWHOLE, pcTRANS, pcFINAN, pcSERVICE, pcNON-AGR, pcCLE, pcCOMME, and pcSERV based on increased air passenger traffic per capita. We then recalculated the employment figures for CONS, WHOLE, TRANS, FINAN, SERVICE, NON-AGR, CLE, COMME, and SERV based on the changes in pcCONS, pcWHOLE, pcTRANS pcFINAN, pcSERVICE, pcNON-AGR, pcCLE, pcCOMME, and pcSERV, respectively. Last, we calculated the differences between the original CONS, WHOLE, TRANS, FINAN, SERVICE, NON-AGR, CLE, COMME, and SERV listed in Panel A and those revised in Panel B. The numbers in parenthesis in Panel B stand for the generated or lost employment in each of the employment classification due to the increase in air passenger traffic per capita. For example, the number 49 at the last row of the eighth column of Panel B means that the employment in finance, insurance, real Özcan Air passenger traffic and local employment: evidence from Turkey estate, and business services is expected to increase by 49 in the 31 provinces analysed if the air passenger traffic per capita increases by 3% point. We then repeated this methodology for Panel C and Panel D as well.
The changes in various employment classifications are striking. The last row of Panel D shows the employment generated in the 31 provinces when air passenger traffic per capita increases by 10% point. In terms of industries, the employment in the construction industry (CONS), wholesale and retail trade; restaurants and hotels (WHOLE), transportation, communication, and storage (TRANS) and the employment in finance, insurance, real estate, and business services (FINAN) are expected to increase by 143, 1,742, 147, and 164. Although these figures are big, the generated employment for aggregate industries is even bigger. We predicted that growth in employment for SERVICE and NON-AGR, as a result of 10% point increase in air passenger traffic per capita, would be 15,013 and 13,057, respectively.
The last three columns of the last row of PANEL D exhibit the generated employment in three different occupations in the 31 provinces. According to Panel D, a 10% point increase in air passenger traffic per capita, holding the population constant, leads to an increase of 347 in the total number of clerical and related workers (CLE), 183 increase in the total number of commercial and sales workers (COMME), and 1,077 increase in the total number of service workers (SERV).

Conclusion
This paper attempts to find the linkage between air passenger traffic and its possible effects on local employment in Turkey, in terms of both industries and occupations. We used 2SLS estimation to handle the problem of two-way causality between the air passenger traffic and its effect on local employment. The data used in the analyses are from the year 2000, but we think some interpretations can still be derived. 9 The results of this paper support the expectation that air passenger traffic stimulates employment in many industries and occupations. More specifically, the findings here suggest that air passenger traffic per capita raises local employment per capita in industries such as (i) construction industry (pcCONS); (ii) wholesale and retail trade; restaurants and hotels (pcWHOLE); (iii) transportation, communication, and storage (pcTRANS); (iv) finance, insurance, real estate, and business services (pcFINAN); (iii) aggregate service-related industries (pcSERVICE); and (iv) aggregate nonagricultural industries (pcNON-AGR) and occupations such as (i) clerical and related workers (pcCLE); and (ii) commercial and sales workers (pcCOMME); and (iii) service workers (pcSERV). More concretely, a 10% point increase in air passenger traffic per capita, holding population and other variables constant, is likely to generate 15,013 service jobs throughout the 31 provinces analysed.
Whether such an expected job creation justifies the public financing of airports may arise as a policy question. To include the cost of job creation in Turkey within our analysis may help respond to this question. Özcan Air passenger traffic and local employment: evidence from Turkey Note: The numbers in the parentheses are rounded and represent the changes in each employment classification according to base scenario when air passenger traffic per capita increases by 3% point. Özcan Air passenger traffic and local employment: evidence from Turkey Note: The numbers in the parentheses are rounded and represent the changes in each employment classification according to base scenario when air passenger traffic per capita increases by 5% point. Özcan Air passenger traffic and local employment: evidence from Turkey Note: The numbers in the parentheses are rounded and represent the changes in each employment classification according to base scenario when air passenger traffic per capita increases by 10% point.