Flight cancellations and airline alliances: Empirical evidence from Europe

Transportation Research Part E 116 (2018) 90–101

Contents lists available at ScienceDirect

Transportation Research Part E journal homepage: www.elsevier.com/locate/tre

Flight cancellations and airline alliances: Empirical evidence from Europe Marco Alderighia,b, Alberto A. Gaggeroc, a b c

T

⁎

University of Aosta Valley, Grand Chemin 75/77, Saint Christophe 11020, Italy CERTeT, Bocconi University, Roentgen 1, 20136 Milan, Italy Department of Economics and Management, University of Pavia, Via S. Felice 5, 27100 Pavia, Italy

A R T IC LE I N F O

ABS TRA CT

Keywords: Airline alliance Competition Flight cancellation Oneworld Sky Team Star Alliance

We show that participating in an airline global alliance significantly increases the likelihood of canceling a flight: an airline is more prone to cancel a flight once it can rely on its partners’ network. Alliance membership increases the value of the airline’s own network, enlarges route and hub dominance, and simplifies the re-routing of stranded passengers. Since flight cancellation may be affected by carriers’ behavior, the frequency of cancellations and their implied inconveniences to consumers should be taken into consideration by regulatory authorities.

JEL Classification: C25 D40 L93

1. Introduction When it comes to studying the service quality in the airline industry, most practitioners and researchers base their assessment on the on-time performance of carriers (Mayer and Sinai, 2003; Mazzeo, 2003). It is very common to find in the first pages of many airline annual reports the list of the most punctual airlines, possibly together with an indicator of passenger satisfaction (Foreman and Shea, 1999). Airlines are aware of the importance to their clients of being punctual: a flight delay implies waste of time, loss of opportunities, and disutility to passengers (Sternberg et al., 2016). For these reasons, many airlines emphasize their on-time performance records as a promotional tool to retain clients or attract new passengers (Suzuki, 2000); Ryanair even plays a punctuality jingle inside the aircraft when its airplane lands on time. Flight cancellations cause greater inconvenience to passengers, but have received less attention than flight delays (Sternberg et al., 2017). According to the US Bureau of Transportation Statistics, the major causes of flight cancellations are extreme weather (tornado, blizzard, or hurricane), airline or airport operation problems (mechanical problems, shortage of crew, lost baggage, latearriving aircraft, heavy traffic volume), and security reasons (terminal evacuation due to terrorism threat, re-boarding because of security breach).1 Although the above discussion might lead to the conclusion that disruptive events occur at random, scholars and practitioners in

We are grateful to five anonymous reviewers for very helpful comments which contributed to ameliorate this paper and to the participants of the 2017 AISRe Conference in Cagliari, the 2018 TEM conference in Chambéry. Finally, we thank Branko Bubalo for helping us with the data collection. The usual disclaimer applies. ⁎ Corresponding author. E-mail addresses: [email protected] (M. Alderighi), [email protected] (A.A. Gaggero). 1 Seehttps://www.rita.dot.gov/bts/help/aviation/html/understanding.html#q1. https://doi.org/10.1016/j.tre.2018.05.008 Received 15 October 2017; Received in revised form 21 May 2018; Accepted 22 May 2018 1366-5545/ © 2018 Elsevier Ltd. All rights reserved.

Transportation Research Part E 116 (2018) 90–101

M. Alderighi, A.A. Gaggero

the field agree that airlines act strategically, i.e. they have some leeway in the decision to cancel or not to cancel a flight (Rupp and Holmes, 2006; Seelhorst and Hansen, 2014). The operational research literature provides some guidance to airline dispatchers on how to manage irregular traffic operations (aircraft and crew rescheduling, gate reassignment, and ferry flight allocation). More specifically, flight scheduling research suggests a number of different methods that can minimize the number of cancellations, the overall passenger delay, or the total costs (Xiong and Hansen, 2013; Kohl et al., 2007; Thengvall et al., 2001; Yan and Yang, 1996). Some recent economics literature models disruptive events such as the airline decision to cancel or delay a flight based on the maximization/minimization of a given objective function (Atkinson et al., 2016; Seelhorst and Hansen, 2014). Cancellation decisions are therefore driven by airline preferences, and they can be a way to reduce delays and put an end to irregular operations. In other cases cancellations can be due to low passenger demand or other economic reasons: an airline makes a tradeoff between operating a scheduled flight or canceling it (Rupp and Holmes, 2006; Seelhorst and Hansen, 2014).2 Cancellations and delays may generate additional costs, which may include hotel and accommodation expenses for disrupted passengers and crew; monetary compensations to passengers; and ticket payments to other airlines (EC Regulation 261/2004). In order to better handle irregular operations and reduce these costs, airlines often build some flexibility into their schedules: they allow extra buffers between flights; increase idle capacity; and develop more efficient re-planning methodologies (Atkinson et al., 2016; Kohl et al., 2007; Barnhart et al., 2012). Even though the growing importance of airline alliances is widely recognized (Pels, 2001; Gaggero and Bartolini, 2012), to the best of our knowledge, the empirical link between airline alliances and flight cancellations has not been yet demonstrated. We reckon that there is a combination of different effects, which modify the attitude of airlines toward cancellation when they belong to a global alliance. First, having partner carriers reduces competition and market discipline, thus leading to lower service quality. This result parallels the fact that more concentrated routes tend to experience more delays and cancellations (Rupp and Holmes, 2006). Second, since airlines trade off the marginal costs of disruption against the marginal benefits of hubbing, which in turn are affected by the network size (Mayer and Sinai, 2003), alliance membership amplifies the benefits by enlarging the airline’s number of destinations, and hence makes each member more inclined to sustain greater delay and cancellation costs. Third, participating in an alliance gives the option to rely on the partners’ network to re-route stranded passengers at a lower cost. Indeed, global alliances usually include revenue-sharing agreements with partners (Hu et al., 2013). With this paper we aim to investigate the effect of airline alliances on flight cancellation, using a sample of non-stop flights departing from and landing at the major European airports during the period April 2011 – December 2012. Building on the empirical model illustrated by Rupp and Holmes (2006), we find that belonging to a global alliance increases the odds of flight cancellation. Furthermore, a higher share of partners’ flights on a route is also associated with larger flight cancellations. These results highlight an important regulatory issue, previously ignored by the empirical literature. Since flight cancellation may be affected by carriers’ behavior, the frequency of cancellations and their implied inconvenience to consumers should be taken into consideration by the regulatory authorities when they assess the drawbacks of market concentration. The remainder of the paper is organized as follows. Section 2 provides a review of the literature; the sample and the data collection are presented in Section 3, together with a brief descriptive analysis. The econometric model is described in Section 4, followed by a discussion of the results in Section 5 and the robustness checks in Section 6. Finally, Section 7 summarizes and concludes the paper. 2. Literature review The closest study to ours is the work by Rupp and Holmes (2006), who investigate the determinants of flight cancellation using a probit model on a panel of US domestic flights, observed during the period 1995–2002. They find that flight cancellation is not only determined by a stochastic component (e.g. bad weather), but also by a strategic component (e.g. the airline decision). The main results of the Rupp and Holmes (2006) research are that cancellations are less likely during weekends and on the last flight of the day, since the airline wishes to avoid additional costs, such as those concerning accommodation and meals for the stranded passengers, as well as possible pecuniary compensation. Furthermore, the authors find that other relevant determinants of flight cancellation are: airline profitability on the route; the airport hub status; the route; and airport competition. The fact that flight cancellation can result from the strategic behavior of the airline is also argued by Fukui and Nagata (2014), who study the impact of the introduction of a new rule by the US Department of Transportation (DOT) to reduce the tarmac delay.3 Fukui and Nagata (2014) find that the threat of an investigation by DOT spurs airlines to reduce tarmac delay, but increases gate departure delays and, more importantly, flight cancellations. Cao et al. (2017) apply the fractional response model described by Papke and Wooldridge (2008) to a panel of US flights observed during the period 2005–2012 on a monthly basis. As in Rupp and Holmes (2006),Cao et al. (2017) find that the relationship between 2 The extreme decision by Ryanair to ground thousands of flights between autumn 2017 and spring 2018 provides additional evidence of how airline strategic behavior affects cancellation. The airline’s official explanation for this decision was the poor planning of pilot holidays and the objective to increase punctuality (Economist, 2017a; Economist, 2017b). Other explanations concern the fact low payments and poor working conditions have induced many pilots to leave the budget carrier (Independent, 2017). 3 A tarmac delay is the delay induced by holding an aircraft on the ground before the take-off or after landing with no possibility for the passengers to disembark. The rule introduced by the DOT is called the “Enhancing Airline Passenger Protections”, or “tarmac delay” rule. See also Xiong and Hansen (2013) for further investigation of ground delay.

91

Transportation Research Part E 116 (2018) 90–101

M. Alderighi, A.A. Gaggero

flight cancellation and route concentration is positive; moreover, they find that such a relationship is non-linear, so that the effect of the change in route competition may depend on the initial level of competition.4 A well-documented review of flight-delay prediction models is provided by Sternberg et al. (2017). The authors argue that most of the research concerns the following issues: explaining how delays propagate through the airline or airport network; predicting the route delay; and understanding the process governing flight cancellation. The second stream of research related to our work is on the behavior of airlines in alliance. The literature in this field mainly focuses on some popular antitrust issues, such as how flight frequencies and fares are affected by a merger or by an antitrust immunity decision (Bilotkach and Hueschelrath, 2011; Reitzes and Moss, 2008). By means of a simulation, Brueckner and Whalen (2000) show that members of international alliances charge lower fares by about 25% with respect to the fares charged by non-allied airlines. This result can be described as a positive circular process. Cooperation on lower fares is obtained by the elimination of the double marginalization; lower prices stimulate traffic, which leads to lower marginal costs and, ultimately, to a further reduction of fares. The decrease of fares induced by airline alliances is also documented by Park and Zhang (2000) and Bamberger et al. (2004) for the US market. On the contrary, Brueckner (2001) shows that, on interhub markets, alliances tend to reduce competition and raise fares, although the overall welfare increases due to a better quality of service (higher frequencies). 3. Data The sample used in this analysis combines flight schedule data and meteorological data. The flight schedule data are collected from flightstats.com, which reports daily: the origin; the destination; the airline code; the flight number; the flight status (e.g. “Delayed”, “Canceled”, etc.); and the scheduled and actual departure and arrival times of all the flights departing from and landing at 100 European airports (see Table A.1 in the Appendix). Based on this information, we compute additional variables which measure flight frequency and competition both at the airports and at route level. Our final sample comprises 3,523,470 non-stop flights serving 3,270 European routes operated during the period April 2011–December 2012. Data on the meteorological conditions are obtained from wunderground.com, which publishes the METeorological Air Report (METAR) for each airport or nearby weather stations on a half-hourly basis. The report consists of general weather conditions, temperature, wind direction and speed, humidity, and pressure. The high frequency of METARs during the day makes it possible to link the meteorological situation at the point of origin/destination very close to the time when the airplane is scheduled to take-off/ land. We deem such accurate time correspondence to be very relevant in our analysis because the weather may change so rapidly within the same day that in some hours it may hinder the flight departure or landing, whilst in others it is totally innocuous.5 Table 1 reports the percentage of canceled flights in our sample. We differentiate between flights operated by full-service carriers (FSCs) and low-cost carriers (LCCs).6 We also take in account the membership in three global alliances: Oneworld, Sky Team and Star Alliance. Overall the flights operated by alliance members totals to about 63% of the entire sample. FSCs cancel a larger percentage of flights than LCCs; within the FSC category, alliance members cancel a greater percentage of flights relative to those airlines not participating in any alliance. Among the three alliances, Oneworld and Sky Team have similar rates of cancellation (0.99% and 0.98%, respectively), while the Star Alliance rate is slightly higher (1.24%). Table 2 reports the descriptive statistics of the variables considered in the econometric analysis. A detailed explanation of the variables is provided in the next section. Fig. 1 presents the percentage of flight cancellations during the sample period. Full and dashed lines refer, respectively, to flights operated by airlines belonging or not belonging to a global alliance. The top diagram reports cancellation rates for bad weather conditions at the origin and/or destination airport; the bottom diagram shows these rates for good weather conditions.7 The figure presents two important results. First, the cancellation rate is higher for airlines participating in an alliance, irrespective of the actual weather conditions.8 Second, the path of the two series is similar when the weather is bad, but it differs significantly when the weather is good. Indeed, the path of flight cancellation under good weather conditions is relatively flat for non-alliance carriers, whilst it has a more complex behavior for alliance airlines. This suggests strategic behavior by airlines since the two series in the lower panel of the Figure should have a similar (flat) path in the case of stochastic cancellations. An intuitive representation of the relationship between flight cancellation and airline alliances is provided by Fig. 2, which depicts the fractional polynomial prediction of the percentage of canceled flights over the percentage of flights operated by airlines in an alliance. As the share of flights of alliance airlines rises, the cancellation rate also increases, under both bad and good weather conditions. 4

For a more detailed study on the relationship between competition and airline service quality, see also Greenfield (2014). With the exception of Bubalo and Gaggero (2015), who use half-hourly observations, previous empirical works use daily weather condition (Mazzeo, 2003; Rupp and Holmes, 2006; Forbes and Lederman, 2010). 6 The airlines classified as low-cost are: Aer Lingus, Blue Air, bmibaby, easyJet, Flybe, Germania, Germanwings, Jet2.com, Monarch, Norwegian, Ryanair, Transavia, TUIfly, Vueling, Wizz Air and WOW Air. 7 We classify the weather as “bad” when at the origin and/or at the destination the weather is reported as being foggy, snowy, or thundery, or when at the airport of origin the temperature is below zero degrees Celsius. 8 The spikes of February and September 2012 in both diagrams were mainly due to the occurrence of strikes. More specifically, Frankfurt airport experienced a long strike of ground personnel in the period 16–29 February 2012; France was subject to a day of action on 29 February 2012; on 4 and 7 September Lufthansa cabin crew went on strike, canceling more than a thousand flights (source: Eurocontrol Annual Reports, available athttp://www.eurocontrol.int/publications/coda-delayanalysis-2011-onwards). 5

92

Transportation Research Part E 116 (2018) 90–101

M. Alderighi, A.A. Gaggero

Table 1 Percentage of canceled flights in the sample. Number scheduled flights of

Percentage of canceled flights

Full-service carriers in an alliance Full-service carriers not in an alliance Low-cost carriers

2,203,534 346,331 973,605

1.11 0.87 0.27

Oneworld Sky Team Star Alliance No alliance

531,304 564,872 1,107,358 1,319,936

0.99 0.98 1.24 0.43

Full sample

3,523,470

0.86

Table 2 Descriptive statistics of the regressors (number of observations: 3,523,470). Stochastic (dummy) variables

Mean

Std dev

Strategic variables

Mean

Std dev

Min

Max

Afternoon Late afternoon Evening Sunday Tuesday Wednesday Thursday Friday Saturday Cloudy – destination Foggy – destination Hazy – destination Rainy – destination Snowy – destination Thundery – destination Cloudy – origin Foggy – origin Hazy – origin Rainy – origin Snowy – origin Thundery – origin De-icing operations Strike

0.284 0.265 0.157 0.131 0.146 0.148 0.148 0.147 0.120 0.549 0.017 0.008 0.081 0.018 0.003 0.552 0.021 0.009 0.092 0.008 0.002 0.047 0.003

0.451 0.441 0.364 0.337 0.353 0.355 0.355 0.355 0.324 0.498 0.127 0.090 0.273 0.132 0.051 0.497 0.145 0.094 0.289 0.087 0.049 0.211 0.058

Alliance membership Oneworld Sky Team Star Alliance Oneworld route mkt shr Sky Team route mkt shr Star Alliance route mkt shr Hub – origin Hub – destination Fleet complexity Route average delay Low-cost carrier Nbr of flights Route mkt share Airport volume – orig.

0.625 0.151 0.160 0.314 0.004 0.012 0.045 0.218 0.217 0.495 0.133 0.276 4.339 0.672 2.744

0.484 0.358 0.367 0.464 0.040 0.074 0.146 0.413 0.412 0.303 0.126 0.447 4.827 0.284 1.810

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 1.000 0.016 0.010

1.000 1.000 1.000 1.000 0.917 0.900 0.933 1.000 1.000 0.898 5.608 1.000 69.000 1.000 6.980

4. Econometric analysis Our main estimates are based on the following binary logit model (Greene, 2012, Chapter 17):9

Pr(Cancellation = 1) = F (X ′β ) =

exp(X ′β ) 1 + exp(X ′β )

(1)

where Cancellation is a binary variable, with a value of 1 if the flight is canceled and 0 otherwise; Pr(·) denotes the probability of observing a flight cancellation; F (·) is the cumulative distribution of a logistic random variable, and is given by the right-hand side of the equation; X is the vector of regressors. More specifically, X includes stochastic factors such as weather conditions, strikes and airport attributes; and strategic factors such as flight and airline characteristics. Although both factors have a significant effect on flight cancellation, our attention is mainly devoted to the strategic component and, in particular, to the implications of airline alliances on flight cancellation. We investigate the influence of airline alliances on flight cancellation using various regressors. In some estimates, we include a dummy variable (Alliance membership) equal to 1 if the carrier belongs to one of the aforementioned airline alliances, i.e. Oneworld, Sky Team or Star Alliance. In others, we replace Alliance membership with three alliance specific dummy variables: Oneworld, Sky Team and Star Alliance, with non-allied airlines as the reference group. We also consider partners’ market share on the route for each of the three global alliances: Nbr Oneworld route mkt shr, Nbr Sky Team route mkt shr and Nbr Star Alliance route mkt shr. The Low-cost carrier dummy variable reflects the previous airline type classification (see footNote 6). Route competition variables 9 In Section 6, we also present the estimates from a Probit model (Greene, 2012, Chapter 17), a tobit model (Greene, 2012, Chapter 19), and a fractional logit model (Papke and Wooldridge, 2008).

93

Transportation Research Part E 116 (2018) 90–101

M. Alderighi, A.A. Gaggero

Fig. 1. Flight cancellations during the sample period

Fig. 2. Flight cancellations and airline alliances (route-month percentage).

comprise the number of daily flights operated by the carrier on the route (Nbr of flights) and the market share of the airline on the route (Route mkt share). The airport variables include two dummy variables, Hub – origin and Hub – destination, which, respectively, identify whether the airports of origin or of destination are an airline hub or a main base; and the overall daily number of landing flights at the airport, in units of a hundred (Airport volume – orig.). We also include specific flight characteristics, such as the time and the day of the week when the flight is scheduled to take off. The time of departure is described by four dummy variables: Afternoon (12.00–16.59), Late afternoon (17.00–20.59), and Evening (21.00–23.39), with Morning (0.00–11.59) as the reference group. The day of the week of the flight departure is measured by seven dummy variables, with Monday as the reference category. Strike is a dummy variable equal to 1 if the airports of origin and/or destination are subject to a strike by the airline staff or by the air traffic controllers (see footnote 8). As far as the weather variables are concerned, for both the origin and the destination airports, we introduce six dummy variables

94

Transportation Research Part E 116 (2018) 90–101

M. Alderighi, A.A. Gaggero

Table 3 The determinants of flight cancellation. Model 1

Alliance membership Hub – origin Hub – destination Low-cost carrier Nbr of flights Route mkt share Airport volume – orig. Strike Afternoon Late afternoon Evening Tuesday Wednesday Thursday Friday Saturday Sunday Cloudy – origin Cloudy – destination Foggy – origin Foggy – destination Hazy – origin Hazy – destination Rainy – origin Rainy – destination Snowy – origin Snowy – destination Thundery – origin Thundery – destination De-icing operations Constant Log likelihood Pseudo R2 Observations

Coeff.

Std. error

Odds ratio

0.1608*** −0.2948*** −0.1981*** −1.0856*** 0.0308*** 0.1434*** 0.0254*** 2.7064*** 0.1027*** 0.1853*** 0.0177 −0.0004 −0.1260* −0.1237** 0.0027 −0.7372*** −0.3591*** −0.0180 0.0010 0.7749*** 0.6231*** 0.0820 0.2657*** 0.0698 0.1156*** 0.9094*** 0.4681*** 1.2851*** 0.5412*** 0.4245*** −4.9348*** −161,202 0.071 3,523,470

0.0355 0.0547 0.0234 0.0466 0.0014 0.0377 0.0057 0.1541 0.0231 0.0258 0.0309 0.0600 0.0655 0.0613 0.0651 0.0661 0.0917 0.0392 0.0231 0.0873 0.0445 0.1019 0.0672 0.0474 0.0348 0.1197 0.0507 0.1896 0.1297 0.0653 0.1227

1.1744 0.7447 0.8203 0.3377 1.0313 1.1542 1.0257 14.9751 1.1082 1.2035 1.0179 0.9996 0.8816 0.8836 1.0027 0.4784 0.6983 0.9821 1.0010 2.1703 1.8647 1.0855 1.3043 1.0723 1.1225 2.4827 1.5970 3.6150 1.7181 1.5288 0.0072

(a) Logit estimation. Dependent variable: flight cancellation. Month dummy variables included, but not reported. (b) Standard errors clustered by origin-date; ***, ** and * denote statistical significance at, respectively the 1%, 5% and 10% level.

which capture whether the weather is cloudy, foggy, hazy, rainy, snowy, or thundery at the hour when the flight is scheduled to depart/land. Clear sky is the reference category, which is omitted in the regression. As an additional weather regressor, we include a dummy variable (De-icing operations) equal to 1 if at the hour of departure at the airport of origin the temperature is below zero degrees Celsius, in order to account for possible de-icing operations (Bubalo and Gaggero, 2015). We cluster standard errors by origin-date to allow the residuals of flights leaving from the same airport on the same date to be correlated. Finally, all our regressions include monthly dummy variables to control for possible seasonality effects on flight cancellation.

5. Results Table 3 reports the point estimates, the standard errors and the odds ratios of the baseline model (Model 1), which offers a first explanation of the effect of global alliances on flight cancellation by including, as a main variable of interest, the airline membership dummy.10 Most of the variables listed in Table 2 are added as model’s covariates, and they are discussed below together with the airline membership dummy variable. The first set of variables under consideration are those which can reveal some strategic behavior of the carriers. Model 1 shows that there are significant effects on flight cancellation due to the market dominance on a route: the stronger the presence of an airline on a route, proxied by the number of daily flights and by its market share on the route, the higher the probability of flight cancellation. The positive sign on Nbr of flights suggests that airlines have some leeway in canceling a flight. First, when the number of 10 The odds ratio, i.e. the ratio of the probability of canceling a flight to its complement, provides a measure of the impact of a regressor on the dependent variable. An odds ratio larger (smaller) than 1 indicates that the regressor has a positive (negative) effect on the realization of the event: the farther it is from 1, the stronger the effect.

95

Transportation Research Part E 116 (2018) 90–101

M. Alderighi, A.A. Gaggero

flights is high, the carrier can more easily re-schedule passengers on its own next flight (Pai, 2010; Xiong and Hansen, 2013). Second, airline dominance on a route reduces the quality of the service (Mazzeo, 2003). The negative coefficients on the hub variables confirm what is called ‘hub-effect’: namely, that airlines tend to avoid cancellations of those flights from/to their own hubs. This is because the airline needs to feed its network, which operates via the hub(s). This result (see also Model 7 of Table 5) complements the literature on flight delay, which finds worse on-time performance for flights originating from or directed to the airline’s hub (Mayer and Sinai, 2003; Mazzeo, 2003; Bubalo and Gaggero, 2015). Because hub-andspoke airlines rely heavily on their network to run their business, they strive to operate those flights, which are at most delayed (see Model 7), rather than canceled. This result is related to the literature on airport congestion costs.11 The low-cost carrier status significantly lowers the rate of flight cancellation, in line with the earlier findings of Table 1. This result may stem from one of the main features of the low-cost business model, i.e. the preference for sparse flight frequency on routes in favor of a larger set of destinations and a point-to-point network (Alderighi and Gaggero, 2017). Since the LCC aircraft itinerary usually touches many different airports in sequence, a flight cancellation has a domino effect on the next flights, with little room for recovery.12 The positive and statistical significant sign on the Airport volume – orig. variable indicates that larger airports are more likely to experience high flight-cancellation rates. This finding can be explained by the high levels of congestion observed in larger airports (Mayer and Sinai, 2003). Previous argument on the reluctance of LCCs to cancel flights because of the aircraft itinerary and the greater risk of irregular operations at larger and congested airports provides a clear justification of why LCCs tend to prefer to fly to/ from minor airports. This discussion better clarifies the low cancellation rates of LCCs shown in Table 1. From a joint look at the coefficients on the day of departure dummies, we observe that the rate of flight cancellation is by far the lowest during weekends. This may occur because weekends are characterized by a large proportion of leisure passengers, so canceling a weekend flight would add larger extra costs to the airline in terms of accommodation/meals and compensation for the missed holiday (Rupp and Holmes, 2006; Xiong and Hansen, 2013). Shifting the attention to the scheduled departure time, we find that airlines are more inclined to cancel flights leaving in the middle of the day, while there is no evidence for the lower cancellation of evening flights. This last result contrasts with previous findings in the literature, which explains fewer cancellations for the last flights of day by the fact that the airline wishes to avoid hotel accommodation costs for the stranded passengers and crew (Rupp and Holmes, 2006). As far as the weather variables are concerned, we find a positive coefficient for most of the variables (Clear sky is the reference category), indicating that the probability of a flight being canceled is higher under worse weather conditions. Interestingly, the magnitude and the statistical significance of the coefficients are stronger in those cases where the weather conditions are more likely to cause flight disruption, i.e. foggy, snowy, and thundery conditions.13 Also the De-icing operations variable has the correct positive sign, in line with the aforementioned influence of severe weather on flight cancellation. As expected, strikes have a strong influence on flight cancellations. The main focus of this paper is on the airline participation in a global alliance, which, to the best of our knowledge, has not previously been considered in airline literature. The positive coefficient for Alliance membership indicates that an airline belonging to a global alliance is more prone to cancel a flight, in line with our conjecture. Being a member of a global alliance has three important effects on cancellation. The first effect (dominance) pertains to the fact that airlines can increase their market dominance when forming an alliance, and this has an effect on the quality of the service, i.e. more delays and cancellations. The second effect (network) implies that a global alliance increases the value of each member’s network because it enlarges the number of destinations offered by each member. Thus, an airline can sell its own flights and those of the partners through code-sharing agreements. Since airlines trade off the marginal costs of disruptions against the marginal benefits of hubbing, the increase in the network size due to alliance membership makes each member more inclined to sustain the greater costs associated with delays and cancellations.14 The third effect (re-routing) stems from airline alliances giving their members the option to rely the partners’ network in order to re-route the stranded passengers of the members, which have, therefore, low incentives to contrast delays and cancellations.15

11 The debate on the internalization of airport congestion costs has two main positions. On the one hand, since airlines only partially internalize the congestion costs caused by their flights, airlines with lower airport shares should be charged a higher congestion tax (Brueckner, 2002). On the other hand, if airlines with a lower airport market share have the room to expand their offer at that airport, then dominant airlines have no incentive to reduce congestion, and therefore all airlines should be charged the same congestion tax (Daniel, 1995; Brueckner and Dender, 2008). This latter view is also supported by Mayer and Sinai (2003): because network airlines with a dominant position at the airport tend to organize flights in waves, they prefer to gather their flights in a limited number of time intervals, even if in this way they experience higher delays and produce larger congestion externalities. For this reason, imposing higher congestion charges on airlines with lower airport market shares is unlikely to solve congestion externality problems, which are actually mainly caused by airlines having a dominant position at the airport. Based on this argumentation, our results may provide some support but cannot be conclusive in favor of the second view, since higher delay rates may also emerge as an attempt to internalize congestion costs within a wave system (i.e., if airlines at hub would not internalize congestion costs, the delay could be larger than observed). 12 This feature contrasts with the aircraft allocation strategy adopted by FSCs, which often use the same aircraft back and forth during the same day to link one given spoke to their hub. 13 A similar pattern, but for airline delay, is found by Bubalo and Gaggero (2015). 14 The previous consideration comes from the assumption that airlines have a profit maximizing behavior. The marginal costs of disruption refer to the additional costs sustained by the airlines in terms of meals, accommodation, compensations, and re-protection of the stranded passengers by canceling or delaying a flight, whereas the marginal benefits of hubbing refer to additional benefits that airlines receive from having a better product because of a larger number of frequencies and destinations. 15 Re-routing can be also facilitated, in some cases, by a lower load factor rate on the partners’ flights. Although being in an alliance allows the members to exploit an additional sale channel, which may generate a higher demand, airlines in an alliance tend to increase their frequencies on international routes because of the higher value of the network, and therefore their load factors on these routes tend to be smaller. Empirical evidence shows that the members of an alliance, which are typically FSCs, have lower load factor rates than LCCs (Vidovic et al., 2012).

96

Transportation Research Part E 116 (2018) 90–101

M. Alderighi, A.A. Gaggero

Table 4 The effect of airline alliances on flight cancellation. Model 2 Alliance membership Oneworld Sky Team Star Alliance Oneworld route mkt shr Sky Team route mkt shr Star Alliance route mkt shr Hub – origin Hub – destination Fleet complexity Fleet complexity (Alliance) Route average delay Low-cost carrier Nbr of flights Route mkt share Airport volume – orig. Constant Log likelihood Pseudo R2 Observations

Model 3

Model 4

Model 5

0.1136***

0.1214***

0.0753**

−0.2856*** −0.1843***

1.3926*** 0.4503*** 0.3940*** −0.2906*** −0.1910***

1.3064*** 0.4207*** 0.3885*** −0.3008*** −0.2018*** 0.1586*** −0.1107***

0.9273*** 0.2245* 0.4759*** −0.2839*** −0.2106***

−1.0864*** 0.0318*** 0.1298*** 0.0233*** −4.9176*** −161,179 0.071 3,523,470

−1.0935*** 0.0322*** 0.2612*** 0.0241*** −5.0232*** −161,077 0.072 3,523,470

−1.0635*** 0.0319*** 0.2570*** 0.0247*** −5.0787*** −161,052 0.072 3,523,470

0.0911* 0.1243*** 0.1972***

1.5962*** −1.1267*** 0.0309*** 0.3454*** 0.0220*** −5.2046*** −160,121 0.077 3,523,470

(a) Logit estimation. Dependent variable: flight cancellation. Strike, time departure, day of the week, weather and month dummy variables included, but not reported. (b) Standard errors clustered by origin-date; ***, ** and * denote statistical significance at, respectively the 1%, 5% and 10% level.

In Table 4 we include additional alliance-specific regressors to better describe the impact of alliances on flight cancellation. Owing to lack of space, the table only reports the estimated coefficients of selected covariates. Model 2 replaces Alliance membership with three dummy variables identifying the alliance to which the airline belongs. All the three variables, Oneworld, Sky Team and Star Alliance, are positive and statistically significant, confirming the initial evidence. The magnitude of the effect is higher for Star Alliance and similar for Oneworld and Sky Team, in line with the descriptive statistics of Table 1. Model 3 includes the partners’ route market share for each alliance (i.e. Nbr Oneworld route mkt shr, Nbr Sky Team route mkt shr, and Nbr Star Alliance route mkt shr). The three alliance market share variables are positive and statistically significant, indicating that the rate of cancellation tends to increase when the airline can rely on a steadier flight offer by its partner(s) on the route. The variables referring to the route market share of the airline and the route market share of partners capture the market dominance effect. This effect hinges on the so-called S-curve model: with equal fares, the airline which has higher frequencies receives a more than proportional market share (Douglas and Miller, 1974). Thus, airlines tend to increase their frequencies in order to expand their market shares and/or to charge higher fares (Vaze and Barnhart, 2012; Brueckner and Zhang, 2001; Hansen and Liu, 2015). Since passengers have little choice when an airline has a dominant position (given by its own flights, as well as by those of its allies), we expect that an airline in an alliance to be more inclined to cancel a flight on a dominant route. Indeed, Pai (2010) finds higher rates of cancellation on routes served with high frequency. Moreover, Alderighi and Gaggero (2014) point out that flight frequencies rise as the share of flights offered by airlines belonging to a global alliance increase. These findings also suggest that, in the case of cancellation, the airline can more easily accommodate passengers on other flights using spare capacity (network effect). It is worth noting that, when an airline belongs to a global alliance, it has an additional choice to re-protect its stranded passengers. Due to the large number of connecting passengers at the hub airports (Martin and Voltes-Dorta, 2009; Wei and Hansen, 2006), in the case of flight cancellation, the airline can re-route connecting passengers through a partner’s hub. In this way, each passenger may fly from the same origin to the same destination on an alternative flight connection with minor inconvenience. In Model 4 we extends previous analysis by accounting for the impact of fleet complexity. There are two aspects that point towards higher cancellation rates for airlines that have several aircraft types in their fleet. First, because pilots are trained for specific categories of aircraft models, an airline with a greater degree of fleet complexity has more difficulties in replacing a pilot. Second, when an aircraft has mechanical problems, the likelihood of finding an equivalent aircraft decreases with the complexity of the fleet. To control for fleet complexity, we construct a new regressor as follows. First, we create the “Herfindahl index” (HHI) of fleet composition: on a weekly basis, we calculate the proportion of each aircraft type in the sampled fleet of each airline, i.e. aircraft share, and then we compute the HHI as the sum of the squared aircraft shares. Second, because, as HHI moves towards 1, the extent of fleet complexity diminishes, we set Fleet complexity equal to 1 – HHI. A positive (expected) estimated coefficient for Fleet complexity therefore indicates that the more complex the fleet is, the more likely is flight cancellation. In addition to this, we also includes a dummy variable to identify those alliance carriers with a complex fleet, Fleet complexity (Alliance).16

16

Fleet complexity (Alliance) is set to 1 if HHI is smaller than its median value on the subsample airlines in an alliance.

97

Transportation Research Part E 116 (2018) 90–101

M. Alderighi, A.A. Gaggero

Fig. 3. Odds ratios. Note: Odd ratios of the alliance market share variables (Oneworld route mkt shr, Sky Team route mkt shr and Star Alliance route mkt shr) are computed assuming a 10 per cent variation; Route average delay is on a 10-min scale.

Model 4 shows that Fleet complexity coefficient is positive and significant, confirming our conjecture that airlines with a lower level of fleet heterogeneity tend to cancel a lower number of flights. The coefficient for Fleet complexity (Alliance) is negative. Thus, alliance airlines with a complex fleet have a lower cancellation rate. The explanation for this finding is that global alliances comprise both large and small airlines: the former ones are major airlines (with high fleet complexity), whereas the latter ones are typically regional carriers (with a lower fleet complexity), acting as subsidiaries. Should a flight be canceled, in order to preserve its own reputation, a major airline prefers, if possible, to cut the flight of one of its subsidiaries rather than its own. Although the coefficient for Fleet complexity (Alliance) is negative, the overall effect of flight cancellation of alliance airlines with a complex fleet is positive. In Model 4, the alliance variables maintain the correct sign and level of statistical significance. The conclusions on the other regressors remain unaltered. Model 5 includes the average delay of the airline on a route-month basis. As expected, there is a positive effect of delay on cancellation. More importantly, the alliance variables remain positive and statistically significant. As far as the other regressors are concerned, no notable differences are observed with respect to Table 3: a higher presence of the airline on the route increases the rate of flight cancellation; the hub-effect is confirmed; and LCCs cancel a lower number of flights than FSCs. Although the coefficients are not reported in the table to save space, the weekend-effect is retained; flights scheduled in the middle of the day are subject to more cancellation; and severe weather conditions significantly increase the rate of flight cancellation, with a larger magnitude in those weather situations that can genuinely affect flight cancellation (i.e. fog, storms, and snow). Finally, Fig. 3 is a graphical representation of the odds ratios with their 95% confidence interval for selected covariates obtained from Models 1 to 5. 6. Robustness checks The conclusions of Table 4 are robust to various checks, which are reported in Table 5. Model 6 presents the probit equivalent estimates of Model 3 and confirms the positive effect of airline alliances on flight cancellation. The same qualitative results are also obtained using the probit specifications for the other models presented in this paper (to save space, the estimates are not reported, but are available upon request). 98

Transportation Research Part E 116 (2018) 90–101

M. Alderighi, A.A. Gaggero

Table 5 Robustness Estimator Dependent variable Alliance membership Oneworld route mkt shr Sky Team route mkt shr Star Alliance route mkt shr Hub – origin Hub – destination Departing delay Low-cost carrier Nbr of flights Route mkt share Airport volume – orig. Constant Log likelihood Pseudo R2 Observations

Model 6 Probit Cancellation

Model 7 Tobit Delay

Model 8 Fractional Logit Cancellation

Model 9 Fractional Logit Cancellation

0.0408*** 0.5424*** 0.1948*** 0.1566*** −0.1142*** −0.0703***

5.9087*** 19.6607*** −4.0778*** −3.6541*** 1.0073*** 1.8516*** 0.9675*** −1.2494*** 0.1153*** −4.0006*** 0.0507*** −23.2840*** −7,215,724 0.114 3,493,323

0.3565*** 0.0571 0.7066** 0.6620*** −0.4345*** −0.3435***

0.3164** 0.0962 0.9067*** 0.6073*** −0.3828*** −0.3263***

−0.9492*** 0.0508 0.0498 0.0681** −5.2612*** −3,410 0.071 94,488

−0.9656*** 0.0468 0.0197 0.0688** −5.1465*** −2,746 0.072 75,310

−0.3760*** 0.0134*** 0.0920*** 0.0086*** −2.4896*** −161,033 0.072 3,523,470

(a) Strike, time departure, day of the week, weather and month dummy variables included, but not reported. (b) Standard errors clustered by origindate; ***, ** and * denote statistical significance at, respectively the 1%, 5% and 10% level.

Model 7 shows the tobit estimate of a model in which the dependent variable is the arrival delay of the aircraft (in minutes).17 The effect of airline alliance on flight delay is significantly positive and quantifiable in about 6 minutes, on average. This result points to the same direction of our findings on cancellation: alliance membership reduces the quality of service, whether we measure it as delay or cancellation. In addition to this result, we find an asymmetric effect of the presence of allied partners on the route: for Oneworld members there is a positive effect on delay, whilst for Sky Team and Star Alliance members this effect is negative. The positive sign on the hub variables is in line with the results of other empirical papers on airline delays (Mazzeo, 2003; Mayer and Sinai, 2003); the departing delay, positive and less then 1, confirms the presumption that airlines are able to catch up on some initial delay during their journey (Bubalo and Gaggero, 2015). Models 1–7 have showed that airlines which are in an alliance have higher irregular operation rates (higher cancellations and delays) relative to airlines which do not participate in any alliance. This result, however, cannot be directly employed to provide welfare indications, since we have no information about the utility losses caused by a single delay or cancellation for different airline types. More generally, in evaluating the net utility stemming from a flight operated by different airline categories, in addition to quality of service, we should jointly consider all the aspects concerning flight supply, such as frequencies, fares, ground and onboard services, etc. Model 8 investigates the determinants of flight cancellation using a fractional response model (Papke and Wooldridge, 2008). To be consistent with the previous empirical literature, we average our variables by route-month (Cao et al., 2017). The results are qualitatively similar to those reported in the previous tables and in line with the findings of Cao et al. (2017).18 Before concluding we aim to assess how our estimates fit the data with respect to the previous literature. In Tables 3–5, we report McFadden’s pseudo R-squared for the binary and fractional logit models. The results of this goodness-of-fit test are in line with those found by Rupp and Holmes (2006), and reflect the fact that cancellations are largely unpredictable. In order to test the accuracy of the out-of-sample forecast, we use a hold-out sample approach. Model 9 is estimated using about 80% of the total observations of Model 8 (training sample), excluding the last two months from the analysis (validating sample), as described, for example, by Armstrong (2001). In terms of the mean absolute percentage error (MAPE), we find that the MAPE of the training sample is about 0.672 and that of the validating sample is 0.724. These numbers are in line with the feature that the MAPE of the validating sample must be slightly higher than that of the training sample. Finally, we replicate the whole empirical analysis excluding the LCCs from the sample to be sure that our results are not simply the manifestation of the differences in FSCs vs LCCs business models. We find no qualitative changes on the alliance variables (to save space, the results are not reported and are available from the authors upon request). 7. Discussion and concluding remarks European regulation EC 261/2004 establishes strict rules on compensation to passengers in the event of flight cancellation. If an alternative travel solution is not provided within a short period of time, carriers are obliged to: find a new flight at the earliest point in time; arrange damage compensation including a payment from €250 to €600 depending on the flight distance; and cover the expenses 17 The discrepancy of the observations between Model 6 and Model 7 corresponds to the number of canceled flights, 30,147, for which the delay is obviously not quantifiable. 18 Cao et al. (2017) use the fractional probit estimator, while we use the fractional logit estimator in order to be consistent with the logit approach used throughout the paper. Similar qualitative results are obtained with the fractional probit.

99

Transportation Research Part E 116 (2018) 90–101

M. Alderighi, A.A. Gaggero

sustained by stranded passengers (meal, hotel lodging, etc.). The results found by Rupp and Holmes (2006) for the US market, and confirmed in this paper for the European market, suggest that current regulation on flight cancellation gives carriers some leeway to act strategically. Our analysis has also spotted a new result concerning the strategic practices of flight cancellations: we have found that airlines belonging to global alliances are more likely to cancel a flight. We have also found that these airlines experience higher flight delay. Thus, alliance membership has a negative impact on the quality of service if it comes to schedule adherence. These results do not necessarily imply higher consumer losses for passengers on board alliance airlines, since consumer utility is also affected by how airlines re-protect and compensate their passengers. Nevertheless, our findings open an important regulatory debate regarding airline alliances. Flight cancellations have different implications for passengers and airlines. For the former, cancellation simply translates into a late arrival at their destination, similar to a flight delay. For the latter, cancellation can imply large cost savings, since the airline is spared the operation of one flight, especially when compensation to consumers is limited. The relevance of this issue applies particularly to Europe, which is characterized by a larger number of European carriers within the same alliance (Reitzes and Moss, 2008). Future work could move along three lines. First, it could extend the present research by analyzing in depth regulatory issues concerning airline alliances. Second, it could test whether flight cancellation is also associated with weaker agreements of airline operation such as code-share or other forms of partnership: for instance, it would be interesting to check whether complementary and parallel code-share agreements have a different impact on flight cancellation (Alderighi et al., 2015). Third, it could replicate the present empirical analysis using non-European data to test whether the effect of airline alliances on the quality of service documented in this paper is also observed in other markets. Appendix A

Table A.1 Airports included in the empirical analysis. Aberdeen (ABZ)

Geneva (GVA)

Nice (NCE)

Alghero (AHO) Alicante (ALC) Amsterdam (AMS) Athens (ATH) Barcelona (BCN) Beauvais (BVA) Belfast (BFS) Bergamo (BGY) Berlin-Schoenefeld (SXF) Berlin-Tegel (TXL) Bilbao (BIO) Birmingham (BHX) Bologna (BLQ) Bournemouth (BOH) Bratislava (BTS) Bremen (BRE) Bristol (BRS) Brussels (BRU) Cagliari (CAG) Cardiff (CWL) Catania (CTA) Charleroi (CRL) Cologne (CGN) Copenhagen (CPH) Dortmund (DTM) Dublin (DUB) Durham Tees Valley (MME) Dusseldorf (DUS) East Midlands (EMA) Edinburgh (EDI) Eindhoven (EIN) Florence (FLR) Frankfurt (FRA)

Genoa (GOA) Girona (GRO) Glasgow (GLA) Hahn (HHN) Hamburg (HAM) Hanover (HAJ) Ibiza (IBZ) Istanbul-Ataturk (IST) Istanbul-Gokcen (SAW) Las Palmas (LPA) Leeds (LBA) Leipzig (LEJ) Lille (LIL) Liverpool (LPL) Ljubljana (LJU) London-City (LCY) London-Gatwick (LGW) London–Heathrow (LHR) London-Luton (LTN) London-Stansted (STN) Lubeck (LBC) Lyon (LYS) Madrid (MAD) Malaga (AGP) Malmo (MMX) Malta (MLA) Manchester (MAN) Marseille (MRS) Milan-Linate (LIN) Milan-Malpensa (MXP) Munich (MUC) Naples (NAP) Newcastle (NCL)

Nuremberg (NUE) Olbia (OLB) Oslo Gardermoen (OSL) Oslo-Rygge (RYG) Oslo-Torp(TRF) Palermo (PMO) Palma Mallorca (PMI) Paris-Charles De Gaulle (CDG) Paris-Orly (ORY) Pisa (PSA) Prestwick (PIK) Reus (REU) Riga (RIX) Rome-Ciampino (CIA) Rome-Fiumicino (FCO) Saarbruecken (SCN) Salzburg (SZG) Southampton (SOU) Stockholm-Arlanda (ARN) Stockholm-Bromma (BMA) Stockholm-Skavsta (NYO) Stuttgart (STR) Tenerife-Norte Los Rodeos (TFN) Tenerife-Sur Reina Sofia (TFS) Treviso (TSF) Trieste (TRS) Turin (TRN) Venice (VCE) Vienna (VIE) Weeze (NRN) Zurich (ZRH)

100

Transportation Research Part E 116 (2018) 90–101

M. Alderighi, A.A. Gaggero

Appendix B. Supplementary data Supplementary data associated with this article can be found, in the online version, athttp://dx.doi.org/10.1016/j.tre.2018.05.008.

References Alderighi, M., Gaggero, A.A., 2014. The effects of global alliances on international flight frequencies: some evidence from Italy. J. Air Transp. Manage. 39, 30–33. Alderighi, M., Gaggero, A.A., 2017. Fly and trade: evidence from Italian manufacturing industry. Econ. Transp. 9, 51–60. Alderighi, M., Gaggero, A.A., Piga, C.A., 2015. The effect of code-share agreements on the temporal profile of airline fares. Transp. Res. Part A: Policy Practice 79 (C), 42–54. Armstrong, J.S., 2001. Principles of forecasting: a handbook for researchers and practitioners. Atkinson, S., Ramdas, K., Williams, J., 2016. Robust scheduling practices in the U.S. airline industry: costs, returns, and inefficiencies. Manage. Sci. 62 (11), 3372–3391. Bamberger, G., Carlton, D., Neumann, L., 2004. An empirical investigation of the competitive effects of domestic airline alliances. J. Law Econ. 47 (1), 195–222. Barnhart, C., Fearing, D., Odoni, A., Vaze, V., 2012. Demand and capacity management in air transportation. EURO J. Transp. Logist. 1 (1), 135–155. Bilotkach, V., Hueschelrath, K., 2011. Antitrust immunity for airline alliances. J. Competition Law Econ. 7 (2), 335–380. Brueckner, J.K., 2001. The economics of international codesharing: an analysis of airline alliances. Int. J. Ind. Organ. 19 (10), 1475–1498. Brueckner, J.K., 2002. Airport congestion when carriers have market power. Am. Econ. Rev. 92 (5), 1357–1375. Brueckner, J.K., Dender, K.V., 2008. Atomistic congestion tolls at concentrated airports? Seeking a unified view in the internalization debate. J. Urban Econ. 64 (2), 288–295. Brueckner, J.K., Whalen, W.T., 2000. The price effects of international airline alliances. J. Law Econ. 43 (2), 503–546. Brueckner, J.K., Zhang, Y., 2001. A model of scheduling in airline networks: how a hub-and-spoke system affects flight frequency, fares and welfare. J. Transp. Econ. Policy 35 (2), 195–222. Bubalo, B., Gaggero, A.A., 2015. Low-cost carrier competition and airline service quality in Europe. Transp. Policy 43 (C), 23–31. Cao, K.H., Krier, B., Liu, C.-M., McNamara, B., Sharpe, J., 2017. The nonlinear effects of market structure on service quality: evidence from the U.S. airline industry. Rev. Ind. Organ. 51 (1), 43–73. Daniel, J.I., 1995. Congestion pricing and capacity of large hub airports: a bottleneck model with stochastic queues. Econometrica 63 (2), 327–370. Douglas, G., Miller, J., 1974. Economic Regulation of Domestic Air Transport: Theory and Policy. Brookings Institution, Washigton D.C. Economist, 2017a. Ryanair cancels more than 2000 flights over the next six weeks. Economist, 21st September 2017. Economist, 2017b. Ryanair’s woes continue as it cancels another 18,000 flights. Economist, 28 September 2017. Forbes, S.J., Lederman, M., 2010. Does vertical integration affect firm performance? Evidence from the airline industry. RAND J. Econ. 41 (4), 765–790. Foreman, S.E., Shea, D.G., 1999. Publication of information and market response: the case of airline on-time performance reports. Rev. Ind. Organ. 14 (2), 147–162. Fukui, H., Nagata, K., 2014. Flight cancellation as a reaction to the tarmac delay rule: an unintended consequence of enhanced passenger protection. Econ. Transp. 3 (1), 29–44. Gaggero, A.A., Bartolini, D., 2012. The determinants of airline alliances. J. Transp. Econ. Policy 46 (3), 399–414. Greene, W., 2012. Econometric Analysis, seventh ed. Pearson Education, New York. Greenfield, D., 2014. Competition and service quality: new evidence from the airline industry. Econ. Transp. 3 (1), 80–89. Hansen, M., Liu, Y., 2015. Airline competition and market frequency: a comparison of the S-curve and schedule delay models. Transp. Res. Part B: Methodol. 78, 301–317. Hu, X., Caldentey, R., Vulcano, G., 2013. Revenue sharing in airline alliances. Manage. Sci. 59 (5), 1177–1195. Independent, 2017. Ryanair cancellations: the truth behind why 2,000 flights are due to be scrapped. Independent, 18th September 2017. Kohl, N., Larsen, A., Larsen, J., Ross, A., Tiourine, S., 2007. Airline disruption management perspectives, experiences and outlook. J. Air Transp. Manage. 13 (3), 149–162. Martin, J.C., Voltes-Dorta, A., 2009. A note on how to measure hubbing practices in airline networks. Transp. Res. Part E: Logist. Transp. Rev. 45 (1), 250–254. Mayer, C., Sinai, T., 2003. Network effects, congestion externalities, and air traffic delays: or why not all delays are evil. Am. Econ. Rev. 93 (4), 1194–1215. Mazzeo, M.J., 2003. Competition and service quality in the U.S. airline industry. Rev. Ind. Organ. 22 (4), 275–296. Pai, V., 2010. On the factors that affect airline flight frequency and aircraft size. J. Air Transp. Manage. 16 (4), 169–177. Papke, L.E., Wooldridge, J.M., 2008. Panel data methods for fractional response variables with an application to test pass rates. J. Econometrics 145, 121–133. Park, J.-H., Zhang, A., 2000. An empirical analysis of global airline alliances: cases in North Atlantic markets. Rev. Ind. Organ. 16 (4), 367–384. Pels, E., 2001. A note on airline alliances. J. Air Transp. Manage. 7 (1), 3–7. Reitzes, J., Moss, D., 2008. Airline alliances and systems competition. Houston Law Rev. 45, 294–332. Rupp, N.G., Holmes, G.M., 2006. An investigation into the determinants of flight cancellations. Economica 73 (292), 749–783. Seelhorst, M., Hansen, M., 2014. Flight cancellation behavior and aviation system performance. arXiv preprint arXiv:1703.06118. Sternberg, A., Carvalho, D., Murta, L., Soares, J., Ogasawara, E., 2016. An analysis of Brazilian flight delays based on frequent patterns. Transp. Res. Part E: Logist. Transp. Rev. 95, 282–298. Sternberg, A., de Abreu Soares, J., Carvalho, D., Ogasawara, E.S., 2017. A review on flight delay prediction. CoRR, abs/1703.06118. Suzuki, Y., 2000. The relationship between on-time performance and airline market share: a new approach. Transp. Res. Part E: Logist. Transp. Rev. 36 (2), 139–154. Thengvall, B.G., Yu, G., Bard, J.F., 2001. Multiple fleet aircraft schedule recovery following hub closures. Transp. Res. Part A: Policy Practice 35 (4), 289–308. Vaze, V., Barnhart, C., 2012. Modeling airline frequency competition for airport congestion mitigation. Transp. Sci. 46 (4), 512–535. Vidovic, A., Mihetec, T., Steiner, S., 2012. Low-cost airlines traffic evolution in South-East Europe. J. Soc. Dev. Teaching Bus. Process. New Net Environ. B&H 7 (1), 369–375. Wei, W., Hansen, M., 2006. An aggregate demand model for air passenger traffic in the hub-and-spoke network. Transp. Res. Part A: Policy Practice 40 (10), 841–851. Xiong, J., Hansen, M., 2013. Modelling airline flight cancellation decisions. Transp. Res. Part E: Logist. Transp. Rev. 56, 64–80. Yan, S., Yang, D.-H., 1996. A decision support framework for handling schedule perturbation. Transp. Res. Part B: Methodol. 30 (6), 405–419.

101