2. Literature Review
2.2 Determinants of Flight Delays
This section reviews studies that focused on exploring the determinants of flight delays and summarizes their contents in Table 2.4. The research reviewed in Section 2.1 may be regarded as being related to this section. However, the research in Section 2.1 only focused on one determinant, which is the delays of other airports. The following studies focused on a certain aspect of possible determinants of flight delays, most of which pertain to the influence of competition and congestion effects either from the airport or route perspective2. Several discussed other determinants, such as Internet or carrier codeshare alliances.
Category 1: Airport perspective – airport congestion internalization
Adopting an airport perspective, the reviewed research mostly focused on examining whether having a higher market share at an airport generates more incentives for a major carrier to internalize its own delays to avoid affecting the on-time performance of other flights of its own, which is called the “congestion internalization hypothesis” (Bendinelli et al., 2016).
Different from road transportation, in which each driver is atomistic, air transportation carriers are non-atomistic. Brueckner (2002) examined congestion theory from road pricing theory, which assumes that each user does not consider the delays he/she imposes on fellow users, to check its applicability in air transportation. Using data from 25 US airports in 1999, research examined the theory through multiple linear regression models. The dependent variable for the regression was the total delays of an airport in 1999, and three measures of airport concentration were used as primary
2 Bendinelli et al. (2016) named determinants regarding an airport’s characteristics as “airport-level”
determinants, and those regarding a route’s characteristics as “route-level” determinants. This study continues using such idea while replacing the term of “level” with “perspective” to prevent the misunderstanding that there is a hierarchical relationship existing between airports and routes.
20
independent variables. He found that in monopoly and oligopoly cases, a carrier internalizes the congestion each flight imposes on other flights it operates, which is different from the assumption of road pricing theory.
However, airports that are dominated by one large carrier in the US still have the largest overall delays. Mayer and Sinai (2003) examined two factors that may induce air traffic congestion and delays, namely, network benefits due to hubbing and congestion externalities. A dataset of flight information on Fridays from January 1988 to November 2000 was used for fixed-effect regression models. Given that carriers can adjust their schedules to compensate for expected delays, the study constructed a measure of delay, that is, actual travel time minus minimum feasible travel time, as the dependent variable.
The Herfindahl–Hirschman index (HHI)3 of origin and destination airports and the hub size of airports and carriers were included as independent variables to examine the two aforementioned factors. The results showed that network benefits due to hubbing make up a more important factor for air traffic congestion than congestion externalities.
Considering that previous literature measured delays from the carriers’ point of view, Rupp (2009) examined previous hypotheses from the perspectives of carriers and passengers by differentiating delay measures as follows: excess travel time as the carriers’
perspective and arrival/departure delay as the passengers’ perspective. Apart from independent variables regarding concentration and hub airports and carriers, Rupp (2009) also considered demand, economic, logistical, and weather variables as control variables.
Through fixed-effect regression models and by using a dataset of randomly selected 1%
flights from January 1995 to December 2004, the study found that carriers are not internalizing the costs of flight delays, especially from passengers’ point of view.
3 Defined as the sum of the squares of the market shares of the firms within the industry, which is a commonly used measure of market concentration.
21
Most previous studies are based on the US, and studies that are suitable for the European setting are limited. Therefore, Santos and Robin (2010) compared the differences between US and European settings and used the variables adopted by Mayer and Sinai (2003) to determine if these variables have the same effects on delays at European airports. The flight data of all domestic and intra-European flights from 2000 to 2004 were used, and the seasonal average delay of a carrier for a route was adopted as the dependent variable. The independent variables included airport concentration, slot control level, and hub size of the airport and carrier. The research results showed that the coefficients of primary independent variables may differ from those in research based on the US possibly due to the local characteristics of Europe.
Category 2: Route perspective – competition and quality
Another category of literature examined the influence of competition on service quality and regarded on-time performance as an indicator of carrier quality. Different from this research category, route perspective research analyzed the route delays of origin–destination pairs. The main hypothesis was that route concentration generates low incentives for dominant carriers to engage in good service quality with respect to on-time performance (Bendinelli et al., 2016).
Mazzeo (2003) conducted the first study that examined the relationship between competition and service quality and considered on-time performance a proxy of service quality. Arrival delay was used as a dependent variable in the base regression. Among the independent variables, competition was measured by monopoly route, market share of the carrier in origin and destination airports, and HHI index; the control variables included weather conditions and flight, airport, and airplane characteristics. Multiple linear regression models and a dataset of 50 major airports in April and July 2000 were used.
22
The empirical results indicated that flights are frequently on-time on routes that are served by few carriers. Carriers schedule long flight times on routes under their monopoly.
While previous studies regarded market structure (i.e., concentration) as exogenously determined with respect to on-time performance, Greenfield (2014) addressed the endogeneity of market structure through two instrument variables, namely, lagged HHI and changes in HHI after the 2008 merger between Delta Air Lines and Northwest Airlines in the US. Greenfield (2014) found the same correlation between concentration and on-time performance by using a panel dataset from 2005:Q4 to 2010:Q3 and a fixed-effect model; the effect of competition on route delays was three times stronger than that in previous studies’ findings.
Bendinelli et al. (2016) examined the hypotheses of previous research from both perspectives and the hypothesis about the impact of low-cost carriers (LCCs) stating that LCC entry affects the airport congestion internalization and on-route service quality improvement of incumbent carriers. A fixed-effect model of flight delays was developed, and a dataset of a panel of 209 routes in Brazil between January 2002 and December 2013 was used for analysis. Combining airport and route variables in the empirical analysis, Bendinelli et al. (2016) discovered that the hypotheses for airport and route variables in previous research, that is, the congestion internalization hypothesis and the competition–
quality hypothesis, are observable in the Brazilian airline market. LCC entry induces internalization from the airport perspective, and they found evidence on incumbents reducing the prevalence of flight delays (but no evidence on the impact on the duration of such delays). However, they did not find sufficient evidence to support an accentuation effect of LCC entry on the competition–quality relationship.
Category 3: Others
23
Previous studies that focused on other determinants of flight delays apart from airport and route factors are placed in the third category.
While previous studies examined the relationship between market structure and service quality, Prince and Simon (2009) were interested in the effect of multimarket contact on the service quality of carriers, with the proposition that firms meeting in multiple markets compete less aggressively to avoid the possible revenge of competitors in other markets. Their hypotheses were that multimarket contact negatively affects service quality, and multimarket contact in competitive markets has minimal effect on service quality. The hypotheses were tested using US domestic flight data on Fridays from January 1995 to August 2001. Different measurements of on-time performance of a carrier on a route in a month were used as dependent variables, and the average multimarket contact of that carrier was employed as independent variables. Applying fixed-effect models, they found strong evidence for the first hypothesis and some support for the second one.
Suspecting that the Internet makes the market highly competitive and motivates firms to reduce prices, Ater and Orlov (2015) examined the effect of the Internet on on-time performance of flights through a fixed-effect regression model and a dataset of one Thursday in each quarter in 1997, 1998, 2000, 2001, 2003, and 2007 of nine carriers.
They found that Internet access leads to long scheduled flight times, and actual flight times and arrival delays increase when many passengers have Internet access.
Yimga (2017) was interested in the influence of code-sharing agreement on on-time performance of carriers by using a fixed-effect regression model and a dataset of the 3rd and 4th quarters of 2002 and 2004 of 19 carriers. The research found that code-sharing agreement improves alliance carriers’ on-time performance, and those that competed before the agreement was built improve more than those who did not.
24
Prince and Simon (2014) investigated how incumbent firms’ service quality changes in response to entry and entry threats. The data they used included the on-time performance of other carriers on routes treaded by Southwest Airline in 1993–2004. They used fixed-effect regression models to address the research question. The dependent variables included different measurements of on-time performance of a carrier on a route in a quarter. The independent variables of interest were dummy variables indicating periods before or after the entry or entry threats by Southwest. The research results showed that incumbent on-time performance worsens in response to entry and entry threats when the entrant is an LCC. This result differs from that in previous research that concluded a positive relationship between service quality and competition (Mazzeo, 2003;
Prince & Simon, 2009; Rupp et al., 2006).
This review reveals that previous studies focused on airport and route determinants, multimarket contact, Internet, competitor’s entry, code-sharing agreement, and entry threats but seemingly overlooked the network determinants on delays.
25 Table 2.4. Literature regarding determinants of flight delays
References Subject; data; study region
Analysis method Dependent variables Independent variables Conclusion
Primary variables Control variables Category 1: Airport perspective – airport congestion internalization Brueckner
Flight share of the airport’s largest carrier
Flight share of the airport’s largest carrier exceeds 65
When an airport is dominated by a monopolist, congestion is fully internalized; when in oligopoly situation, carriers only internalize the congestion they impose on themselves. (6) Difference in excess
travel time between outbound and return flights
(7) Difference in total travel time between themselves) do lead to air traffic delays.
2. However, hubbing effect is the dominant contributor to delays, that delay increases more with the size of a hub.
Rupp (2009) Flight; randomly selected 1% flights
Normalized departure time Flight distance
The use of new delay measures reverses earlier studies’ findings, that carriers do not internalize flight congestion.
26 References Subject; data; study
region
Analysis method Dependent variables Independent variables Conclusion
Primary variables Control variables Carrier
Arrival delay HHI (airport) Slot coordinationa Seasona: Spring & Summer;
Autumn Yeara Carriera
Airport (fixed effect)a
1. The effect of airport hub size does not increase monotonically, which probably because the hub-and-spoke system in Europe is not as extensive as in the US, and that most hub airports in Europe are slot controlled.
2. The result for congestion internalization support previous studies’ hypothesis (Brueckner, 2002; Mayer & Sinai, 2003).
Category 2: Route perspective – competition and quality Mazzeo (2003) Flight; 50 major
airports in January,
Flights delays more on routes that are served by only one carrier and in cases where the carrier’s market share at the airports served are higher. Thus, less competition decreases the service quality (on-time performance) of a route.
27 References Subject; data; study
region
Analysis method Dependent variables Independent variables Conclusion
Primary variables Control variables Greenfield
Total number of flights in airport
Carrier-route (fixed effect)a
The result supports the hypothesis of previous studies, but the effect of competition on carrier’s on-time performance is three times stronger than previous studies suggest.
Num. of flights in congested hours
Num. of flights in uncongested hours HHI (city pair)
LCCb presence (city pair)a HHI (maximum value between endpoint cities) LCC presence (on either of the endpoint cities)a
1. When airport and route
perspective effects are estimated simultaneously, congestion internalization and the
relationship between competition and quality found in previous studies are also observed in the Brazilian airline market.
2. Entry of LCC improves on-time performance. (2) Flights arriving at
least 15 (30) minutes late (%)
(3) Scheduled flight time (4) Travel time
(5) Non-airtime (6) Airtime (7) Departure delay (8) Flights departing at
least 15 (30) minutes late (%)
Average multimarket contact
Average multimarket contact on low, medium and high-concentration routes
28 References Subject; data; study
region
Analysis method Dependent variables Independent variables Conclusion
Primary variables Control variables
airport and arriving in the destination airport of a (3) Scheduled flight time (4) Travel time
Proximity to threat of entry Proximity to entry
Load factor
Carrier’s flights on the route
Num. of flights arriving at destination airport Num. of flights departing from origination airport
Incumbents do not improve on-time performance in response to entry, while this phenomenon can only be observed when the (potential) entrant is a low-cost carrier.
Ater and Orlov
(2) Actual elapsed time (3) Arrival delay
Internet × large × morning
Num. of flights departing from and arriving at the origin and destination airports within the same hour
Average fare
The result shows that increased Internet use led to longer scheduled elapsed flight times and arrival delay, particularly in more performance for the alliance firms, and the effects are larger in markets where the partners competed in prior to the alliance.
29 References Subject; data; study
region
Analysis method Dependent variables Independent variables Conclusion
Primary variables Control variables (4) Departure delay for
more than 15 (30) minutes late (%)
Alliance carrier × post-alliance period × competed prior to the alliance
Carrier (fixed effect) a Time (fixed effect) a
a Dummy variable.
b HHI: Herfindahl-Hirschman index; LCC: low-cost carrier.
30
2.2.2 Comparisons and Analyses (1) Research region
Table 2.4 reveals that most of the study regions are located in the US (Ater & Orlov, 2015; Brueckner, 2002; Mayer & Sinai, 2003; Mazzeo, 2003; Prince & Simon, 2009, 2014; Yimga, 2017). The only two exceptions are Bendinelli et al. (2016), who investigated Brazilian domestic flights, and Santos and Robin (2010), who examined domestic and intra-European flights. Although the effects of determinants on flight delay in each region may differ, as shown by the empirical results of Santos and Robin (2010), a study based on East Asia is necessary because no such research has been conducted to our best knowledge.
(2) Analyzed subject
The analyzed subjects (i.e., observations of each study) differ depending on the perspectives adopted by the researchers. The analyzed subjects in previous research are classified in Table 2.5, which shows that several studies used flights as observations, whereas others used carriers, routes, or airports as analyzed observations and the average value for a certain period as the unit of measurement. The selection of analyzed subjects depends on the perspective adopted by the study and the independent variables used. For example, when a study aims to analyze the difference between carriers, it may use flights or carriers as the analyzed observations; otherwise, it may use routes or airports as its analyzed observations. When a study aims to analyze how flights in the morning or weekdays affect flight delay differently and compare them with non-morning or weekend flights, delay of individual flight may be selected as the definition of measurement. If the study aims to clarify how delay changes among seasons, then it may use seasonal average delay of a carrier as its definition of measurement.
31 Table 2.5. Analyzed subjects of literature in Table 2.4
Analyzed subject Definition of measurement References
Flight Delay of an individual flight Ater and Orlov (2015); Mayer and Sinai (2003); Mazzeo (2003); Rupp (2009) Carrier Monthly average delay of a carrier on a route Prince and Simon (2009);
Yimga (2017) Quarterly (seasonal) average delay of a carrier
on a route
Greenfield (2014); Prince and Simon (2014); Santos and Robin (2010) Route Monthly average delay of a route Bendinelli et al. (2016) Airport Num. of flights delayed at an airport in a year Brueckner (2002)
(3) Dependent variable
Delay can be measured in several ways depending on which perspective is adopted.
Empirical results also differ due to different measurements. For example, from the carriers’ perspective, excess travel time, which is the difference between the actual travel time and minimum travel time of a route, may be an appropriate way to measure delay because carriers establish the schedules themselves. For passengers, the difference between actual departure/arrival time and scheduled departure/arrival time is flight delay.
Several researchers also used scheduled and actual travel times as dependent variables to examine whether a carrier establishes a long scheduled travel time to compensate for possible delays or to investigate which itinerary stage generates numerous delays.
Transformation of the dependent variable was performed in several studies to determine if the empirical results would improve because the relationship between dependent and independent variables may not be linear. The dependent variables shown in Table 2.4 are categorized in Table 2.6.
(4) Analysis method
Table 2.4 shows that most studies used panel data regression models or included fixed-effect parameters to account for cross-sectional heterogeneity (Ater & Orlov, 2015;
Bendinelli et al., 2016; Greenfield, 2014; Mayer & Sinai, 2003; Mazzeo, 2003; Prince &
Simon, 2009, 2014; Rupp, 2009; Santos & Robin, 2010; Yimga, 2017) because most of
32 Table 2.6. Dependent variables of literature in Table 2.4
Dependent variable Variable transformation References
Arrival delay N/A Ater and Orlov (2015); Bendinelli et
al. (2016); Greenfield (2014);
Mazzeo (2003); Prince and Simon (2009, 2014); Rupp (2009); Santos and Robin (2010); Yimga (2017) Log odds Bendinelli et al. (2016); Greenfield
(2014)
Probit Mazzeo (2003)
Proportion Mazzeo (2003); Prince and Simon (2009, 2014); Yimga (2017)
Departure delay N/A Bendinelli et al. (2016); Greenfield
(2014); Mayer and Sinai (2003);
Prince and Simon (2009); Rupp (2009); Yimga (2017)
Log odds Bendinelli et al. (2016); Greenfield (2014)
Total delay N/A Brueckner (2002)
Natural log Brueckner (2002)
Excess travel time N/A Mayer and Sinai (2003); Rupp (2009)
Difference in excess travel time between outbound and return flight
N/A Mayer and Sinai (2003)
Scheduled travel time N/A Ater and Orlov (2015); Mayer and
Sinai (2003); Prince and Simon (2009, 2014)
Shortest scheduled travel time N/A Ater and Orlov (2015)
Actual travel timea N/A Ater and Orlov (2015); Prince and
Simon (2009, 2014); Rupp (2009) Difference in total travel time
between outbound and return flight
N/A Mayer and Sinai (2003)
the studies applied datasets containing information on the same subject (such as a route or an airport) in different times and served by different carriers or at different airports and routes. Only Brueckner (2002) used a linear regression model because he applied total delay in 1999 as the dependent variable.
(5) Independent variables and empirical results
Independent variables include airport and route variables, control variables, and variables regarding specific topics. Only variables that are significant in the empirical results of previous literature are included in this part.
A. Airport and route variables
33
Variables regarding airport and route characteristics in the reviewed studies are shown in Table 2.7 and can be categorized into several types. Each of these variables is explained as follows.
For variables regarding airport characteristics:
1. Market concentration: HHI was the most used indicator of the concentration of
1. Market concentration: HHI was the most used indicator of the concentration of