Study Issues

Several concepts must be clarified after reviewing extant literature in the previous chapter. First, the most appropriate observation and measurement of the dependent variable should be discussed. Second, the definitions of airport, route, and network perspectives must be established, and possible determinants need to be selected from the three perspectives as independent variables in the empirical models. Other important factors, apart from the three perspective determinants, should also be determined, discussed, and included as control variables to reduce possible bias of empirical results.

Lastly, the most suitable analysis method for the study variables must be identified.

Issue 1: What is the definition of the dependent variable?

Description of the issue

The literature review indicates that the dependent variables differ among the reviewed studies due to differences in the analyzed observations and measurements (Table 2.6). Such differences result from the subject determinants, data availability, and computational constraints. Thus, the definition of the dependent variable is an important issue.

Research concept

This study aims to explore the determinants of flight delay from airport, route, and network perspectives. Using “route” as the analysis subject is the best option due to the following reasons. First, route determinants can be included. Determinants of origin and destination airports and cities can also be included because a route includes an origin and a destination. Connected airports must also be analyzed. Thus, “route” is the most suitable analysis subject for this study.

This study focuses on determinants from the three perspectives, and the fluctuation of delays throughout a certain period is not the main research interest. Thus, the daily average delay of a route in the second half of 2017 is used as the measurement factor of

42

an observation. In this way, hourly and yearly fluctuations cannot be reflected, but daily, monthly, and seasonal fluctuations can still be observed and analyzed.

In addition, as mentioned in Subsection 1.2.1, arrival delays generally directly affect passengers. They are adopted to measure flight delays in this work.

Therefore, the dependent variable is the daily average arrival delay of a route in the second half of 2017.

Issue 2: What are the three perspective determinants of flight delays in East Asian airports?

Description of the issue

Econometric models are used to explore the possible determinants of flight delays and their effect on flight delays in East Asian airports. Determinants (independent variables) are selected for the subsequent model specification and result interpretation.

Previous studies selected different variables according to their research purposes and study regions. Therefore, selecting variables appropriate for flight delays in East Asian airports is an important issue for this study.

Research concept

This study aims to examine whether the hypotheses in previous studies remain true in East Asian airports, to include characteristics of the air transport system in East Asian airports (obtained from stakeholder interviews), and to investigate the influence of connected airports. Thus, this study focuses on four types of variables, namely, market concentration, hubness, slot control, and air traffic control (ATC) under the three perspectives (airport, route, and network).

1. Airport, route, and network perspectives

The literature review indicates that many and diverse variables affect flight delays.

If variables are not selected systematically, then important variables, such as network variables, may be overlooked. Thus, flight delay determinants are categorized into three

43

Figure 3.2. Concept of airport, route and network perspectives

perspectives, namely, airport, route, and network, following Bendinelli et al. (2016), who categorized determinants of flight delays into two perspectives (airport and route) to obtain a complete and thorough perspective (Figure 3.2).

In this study, “airport determinants” are defined as factors related to the origin or destination airport of a route, which includes characteristics of the origin and destination airport. “Route determinants” are defined as factors regarding the characteristics of the route itself. “Network determinants” are defined as factors related to connected airports and the connecting relationship between connected airports and the origin or destination airport.

“Connected airports” are defined as airports having flights departing from or arriving in the origin or destination airport. To the best of our knowledge, no previous studies on the determinants of flight delays have included network determinants in addition to the delays of connected airports (Hao et al., 2014). Although airports in the entire air traffic network influence a route’s delay, only the immediately connected airports, such as those with non-stop flights connected to the origin or destination airport, are considered in this work to identify the effect of network determinants on flight delays.

2. Market concentration, hubness, slot control, and ATC variables

As shown in Table 2.7 in Subsection 2.2.2, two types of independent variables, namely, competitiveness (concentration) and hubness, were thoroughly discussed in

Connected airports

Connected airports Origin

airport

Destination airport

AIRPORT ROUTE NETWORK

44

Table 3.1. Slot control proportion of three main air transport regions (summer 2017 / winter 2017) (IATA, 2017)

East Asia Europe The Americas^b

Level 2^a 5 / 5 75 / 76 17 / 17

Level 3^a 19 / 19 103 / 75 14 / 14

Total number of slot-controlled airports 24 / 24 178 / 151 31 / 31

a Level 2: schedule facilitated airports; level 3: fully coordinated airports, while others are non-coordinated airports (Santos & Robin, 2010).

b Including North and South America.

previous literature. The exploration of the causal relationships between competitiveness and delay and between hubness and delay aims to examine several hypotheses. In the congestion internalization hypothesis, carriers with high market share have high incentives to internalize delays caused by their flights. In the competition–service quality hypothesis, high competition on routes generates high incentives for carriers to improve their service quality, including on-time performance. In the network benefit hypothesis, hub airports are congested and have many delays because carriers can receive large benefits at such airports in a hub-and-spoke network (Bendinelli et al., 2016; Mayer &

Sinai, 2003). These hypotheses are examined by including market concentration and hubness independent variables in this study.

According to the Worldwide Slot Guidelines of the International Air Transport Association (IATA), airports in East Asia have a low slot control proportion in winter and summer (Table 3.1). Santos and Robin (2010) examined the influence of slot control given that a larger proportion of airports in Europe is slot-controlled compared with the US. They found a positively significant influence of slot control on flight delays. Table 3.1 shows that East Asia has fewer slot-controlled airports than Europe, and this can be viewed as one of the characteristics of the air transport system in East Asia. Thus, this study also includes variables regarding slot control.

Finally, news reports and stakeholder interviews indicate that the high frequency of ATC may be another characteristic of East Asia (Anonymous, 2017; Bergman, 2016).

45

A lot of airports in Taiwan, China, Japan and Vietnam are controlled by military, which have more constraints and thus usually have more delays. … Air traffic control, which usually happens in China and Hong Kong airports, is a highly important determinant of flight delays because under ATC the availability of flight routes sharply reduces and produces more congestion and delays. … High frequency of ATC may be one of the characteristics of Asia Pacific air transport system. (AL001)

Connected airports of Taiwan Taoyuan International Airport such as airports in China and Hong Kong are usually under ATC, which has significant influence on on-time performance of Taiwan Taoyuan International Airport. … The main characteristics of Asia Pacific air transport system compared to Europe and North America are the frequency of ATC, rapid economic growth and lack of abundant facilities due to such rapid growth of air traffic demand.

(AP001)

Thus, this study originally includes the ATC variable to examine its influence on flight delays in East Asia. However, due to data limitation, this variable is subsequently omitted in the regression models (Table 4.1).

Issue 3: What are the control variables of flight delays in East Asian airports?

Description of the issue

Aside from the determinants of the three perspectives, many other factors, such as time and weather, influence flight delays, and the effects of these factors should not be overlooked. Thus, to avoid low fitness or possible biases of the regression models, suitable control variables must be selected.

Research concept

46

According to the literature review and stakeholder interviews, in addition to the determinants mentioned in the second issue, the other possible factors that influence flight delays in this study can be categorized into the following: demand, weather, and time variables. Indicators of demand, such as income per capita, population, and employment rate, were usually included as control variables in the reviewed studies and were found to have significant effects (Greenfield, 2014; Mayer & Sinai, 2003; Prince & Simon, 2009;

Santos & Robin, 2010; Yimga, 2017) possibly because the number of flights increases with increasing demands on air traffic, as explained by one of the interviewees (AL002).

Several of the reviewed studies likewise included different types of weather that could affect delays as control variables because adverse weather conditions, such as thunder and snow, seriously affect air traffic safety (Greenfield, 2014; Mazzeo, 2003; Rupp, 2009);

the interviewees in this study also emphasize the influence of weather (CA001; AP001;

AL001; AL002). In addition, several of the reviewed studies included time variables, such as day of the week, month, season, and year, as control variables or fixed effects to capture time-specific effects, such as special incidents, that may not be reflected by other independent variables (Mayer & Sinai, 2003; Rupp, 2009; Santos & Robin, 2010). The interviewees also accentuate the influence of time, such as Chinese New Year and other holidays (CA001; AP001; AL001). Hence, these influencing factors on air transport are included as control variables.

Issue 4: How can a suitable analysis method for exploring the determinants of flight delays in East Asian airports be selected?

Description of the issue

This study aims to explore the determinants of flight delays, and the original data for study observations include the departure and arrival data of each flight. This study arranges the large amount of data into the daily average delay of a route in 2017.

47

Analyzing these observations through a suitable analysis method is an important issue for this study.

Research concept

As mentioned in Section 1.4, panel data analysis is the most appropriate regression method because this study uses flight data of the same route over a period. Another issue is measuring the delay of each flight because not all flights are delayed; several flights even arrive ahead of schedule (hereafter referred to as early flights). This study uses arrival delay (unit: minutes) as the dependent variable; if early flights are measured as delays with negative values and on-time flights (or flight delays below a certain threshold, such as 15 minutes) are with zero delay, then normal panel data regression can be applied.

However, if only positive values (or flight delays over a certain threshold) are considered, then a panel data truncated model should be applied. Meanwhile, if early and on-time flights (or flight delays below a certain threshold) are coded with zero delays, then the data belong to a censored distribution, and a panel data Tobit model should be applied.

This study chooses the normal panel data model as the preliminary model because it can ideally reflect the original on-time situation of a route.

48