Three Perspective Variables

For the three perspective variables, some of the empirical results show that the influencing relationships are not as expected (Table 4.10). Possible explanations are discussed as follows.

(1) Airport variables

For market concentration of airports (HHI_O and HHI_D), different from the congestion internalization hypothesis held by previous research (Bendinelli et al., 2016;

Brueckner, 2002; Mayer & Sinai, 2003; Rupp, 2009; Santos & Robin, 2010), the empirical results of the present study show that either in 2PVMs, which are similar to previous literature, or in 3PVMs, which consider the network variables, routes with origin or destination airports having a high HHI tend to have many delays. Such results are consistent with the findings of Rupp (2009), who concluded that from the passenger’s perspective (i.e., calculating delays as departure or arrival delays, similar to this study), carriers are not internalizing flight delays. This positive relationship between HHI of an airport and delay may due to the poor on-time performance of the major carriers in East Asia, as stated by an interviewee (AP001). It may also be due to monopolized carriers having less incentive to enhance their on-time performance because of the lack of competition.

For the hubness degree of airports (HUB_O and HUB_D), previous research found that airports with high hubness degree have a positive effect on flight delays because hub airports usually have many regular flights and connecting flights (Ater & Orlov, 2015;

Brueckner, 2002; Mayer & Sinai, 2003; Rupp, 2009; Santos & Robin, 2010). However, the empirical results of this study show that in 3PVMs and 2PVMs, a high hubness degree has a negative effect on flight delays. China, Hong Kong, South Korea, and Taiwan are considered developing economies by the United Nations (UN, 2018). As such, their

97

Table 4.9. Number of each slot control level’s airports in the study sample Slot control level Number of

airports

Daily Average Flights

Connected airport’s delays (min) Mean Standard deviation

Level 1 294 39.2528 19.9466 24.2174

Level 2 5 232.4141 17.0845 13.1345

Level 3 19 617.3578 20.6684 17.7035

resources or infrastructures may not be as abundant and complete as those in developed economies. A possible explanation is that in East Asia, international hub airports are given additional resources by governments. These airports also have good managerial efficiency. As a result, they have few delays. They are important airports for a country’s development. The findings of Fan et al. (2014) can support such an explanation; they found that international hub airports in China have better technical efficiency, including good control of flight delays, than regional hub and non-hub airports.

As indicated in Subsections 4.2.2, 4.2.3, and 4.2.4, for the influence of market concentration (HHI_O and HHI_D) and hubness degree (HUB_O and HUB_D), the influencing relationship remains the same. Hence, these determinants and their impacts are the shared determinants of flight delays in East Asia.

For the slot control level of airports (SLOT_O_2, SLOT_O_3, SLOT_D_2, and SLOT_D_3), two previous studies found that airports with a high slot control level usually have many delays (Brueckner, 2002; Santos & Robin, 2010). However, the purpose of applying slot control is to avoid congestion at airports. Santos and Robin (2010) and the interviewees of this study (AL001; AL002) also noted that slot control may improve on-time performance. The empirical results of this study show that in 2PVMs of the full observation models, except for routes with origin airports applying second level slot control, others with airports applying a higher level of slot control have more delays.

However, after adding network variables in 3PVMs, the empirical results show that the influences of slot control on flight delays are insignificant or even negatively significant.

By contrast, in the non-China and international models, several slot control variables

98

remain positively significant. As demonstrated in Table 4.9, the connected airports of level 3 airports have more delays than level 2 airports, and the differences among the levels’ means are significant⁷. This result implies that slot control level applied by airports may be relevant to the delays of their connected airports; that is, if an airport’s connected airports have many delays, then such an airport may choose a higher level of slot control because it expects to have other delays⁸. This mechanism may be the reason for the insignificant slot control variables in 3PVMs after the addition of network variables.

(2) Route variables

Previous literature held the hypothesis that highly competitive routes have minimal delays because carriers try to enhance their service quality, such as on-time performance, to compete with one another (Ater & Orlov, 2015; Bendinelli et al., 2016; Greenfield, 2014; Mazzeo, 2003; Prince & Simon, 2009; Yimga, 2017). However, in the empirical results of the full observation, China, and domestic models, the market concentration of routes (HHI_R) has a negatively significant effect on flight delays in 2PVMs and no significant effect in 3PVMs. This negative effect fits the opinion of an interviewee of this study (AL002) who reasoned that carriers with high market concentration in a route usually have increased coordination capability; thus, few delays should be expected. The insignificant effect also fits the opinion of several interviewees (CA001; AL001) who considered market concentration of routes having no influence on delays. Nevertheless, in the empirical results of the major airport route, non-China, and international models, the effects become positively significant. Such a difference implies that the competition

7 Since through Levene’s test the variances among slot control airport groups are found not equal, this study applied Welch’s and Brown-Forsythe’s tests instead of traditional one-way analysis of variance (ANOVA) test, and found that the difference among the means (or medians in the Brown-Forsythe’s test) of the groups are significant. We also applied post hoc tests using Games-Howell’s test, and found that the differences between every two groups are all significant at least at α = 0.01.

8 Although level 1 airports averagely had more delays then level 3 airports, the amount of flights at level 1 airports was much smaller than at level 2 and level 3 airports, which may not be cost-effective for their airport operators to apply slot control (Table 4.9).

99

hypothesis proposed by previous research may only be applicable for routes between major airports or non-China or international routes in East Asia.

For the number of flights on a route (FLIGHT_R), in the full observation, major airport, China, and domestic and international models, the more flights there are on a route, the less delays the route has, which is contrary to the findings of previous studies (Greenfield, 2014; Mazzeo, 2003). Flights from or to China’s airports account for around 70% of the full observations (Table 4.2). A possible explanation is that China’s air space is scarce⁹ (AL002) (Gu et al., 2018; He, 2013). Carriers give priority to flights on important routes and care less about flights on routes that are not that profitable.

In full observation models, major airport models, and domestic models, routes with long distance (DISTANCE_R) have further delays. Such findings are opposite to the findings of previous studies, which concluded that routes with large distances have few delays because during a long flying journey, flights have a great chance to fly faster to compensate for previous delays (Mazzeo, 2003; Rupp, 2009; Yimga, 2017). However, the interviewees in this study indicated that carriers usually do not choose to accelerate much because they have to consider the cost of fuel (AL001). Table 4.2 in Subsection 4.1.1 shows that the mean of flight distance is around 1,061 km, which is around one-third of the average airport distance of 2,954 km in the US (Paleari et al., 2010), where the above-mentioned studies were conducted. Moreover, the distribution of the study sample is right-skewed (Table 4.2), which implies that numerous observations have comparatively short flying distances. In non-China and international models, DISTANCE_R is insignificant. Thus, previous studies’ findings on the negative relationship between delay and distance may not be applicable in East Asia. An explanation similar to that above may be applied for such results. Carriers care more about

9 According to Gu et al. (2018), in China, airspace available for civil aviation only accounts for 7%, while 93% of it is controlled by military.

100

the on-time performance of routes with higher frequency, which are usually routes with shorter distances. They care less about routes with lower frequency, which are usually routes with longer distances.

(3) Network variables

The significance of the coefficient of network variables shows the importance of considering the effect of connected airports and routes. However, the negative effect of the average delay of connected airports (AVGDELAY_C) in all models and that of the number of flights at connected airports (FLIGHT_C) in the one-way full observation and major airport models are not as expected. A possible explanation is that carriers with routes connected to easily delayed airports may take precautions, such as adjusting their flight schedules or rearranging the deployment of their aircrafts, to reduce possible delays (AL001; AL002; AP001; CA001). Zhang (2016) found that the characteristics of connecting routes have a more direct influence on delays than the characteristics of connected airports themselves. In our empirical results, variables regarding connected routes (AVGDELAY_CR and FLIGHT_CR) have positive effects on flight delays. The coefficients of AVGDELAY_CR are over 1 in the full observation and major airport models, which indicates that a one-minute average delay on connecting routes may propagate to the route they are connected with. It may also cause more than one minute of delay on that route.

101

Table 4.10. Comparison between this study and previous studies on three perspective independent variables’ effects on delays Perspective Independent

102 Perspective Independent

variable

Full observation models

Major airport models

China models Non-China models

Domestic models

International models

Effects found by previous studies

References

3PVM 2PVM 3PVM 2PVM 3PVM 2PVM 3PVM 2PVM 3PVM 2PVM

DISTANCE_R (+) (+) (+) (+) (+) × × (+) (+) × × (−) Mazzeo (2003); Rupp (2009);

Yimga (2017)

Network AVGDELAY_C (−) - (−) (−) - (−) - (−) - (−) - (+) Hao et al. (2014); Yue and

Wei (2014); Zhang (2016)

FLIGHT_C (−)/(+) - (−) (+) - - - (+) - (−)/× - (+) Zhang (2016)

AVGDELAY_CR (+) - (+) (+) - (+) - (+) - (+) - (+) Zhang (2016)

FLIGHT_CR (+) - (+) (+) - (+) - (+) - × - (+) Zhang (2016)

Note: (+) denotes a positive effect; (−) denotes a negative effect; × denotes no effect; - denotes not included in the model.

103