When Do Firms Offer Higher Product Quality? Evidence from the Allocation of Inflight Amenities

For example, Mazzeo (2003), Rupp (2009) and Gil and Kim (2021) show that increased competition leads to better on-time performance. On the other hand, Prince and Simon (2015) document worse on-time performance by the network carrier on routes with LCC competition.

See https://​ideaworkscompany​.​com/​airline-ancillary-revenue-begins-recovery-with-a-13-increase-to-65-8-billion-for-2021-press-release/​ accessed on 21 October 2022.

Seat pitch is the measurement of space between one point on an aircraft passenger seat and the same point on the seat in front of it. In coach class, the industry standard is a seat pitch of 31 inches.

The decision of what to include in the base product and what to charge for extra extends beyond the airline industry. For example, hotels need to decide whether to provide Wi-Fi for free or charge for extra [e.g., Lin (2017)]. Rental car companies also charge unlimited miles and drop-off at different locations etc. Also, see Ellison (2005), Santana et al. (2020) and the papers cited there for literature on add-on pricing and drip pricing.

See, for example, Mazzeo (2003), Rupp (2009), Prince and Simon (2015) and Gil and Kim (2021).

The historical on-time performance of the flight is observed but not the realized performance at the time of booking. See Forbes et al. 2019 for a discussion of schedule padding.

See Rupp and Tan (2019), Gayle and Yigma (2018), Bishop et al. (2011), and Mazzeo (2003).

We exclude smaller LCCs since they had no variation in the Wi-Fi amenity as neither Spirit nor Frontier offered Wi-Fi aircraft in our sample, while every Virgin America flight in our sample had Wi-Fi.

The exclusion of Southwest Airline fares from online travel agency data and limiting the sample to direct non-stop flights cause some of the top 500 routes to be dropped from the sample.

During our sample period, the Federal Aviation Administration (FAA) limits scheduled air traffic at four airports– Chicago O’Hara (ORD), Newark (EWR), LaGuardia (LGA), and John F. Kennedy (JFK). The FAA requires each airline to have a departure and arrival slot at these airports.

We define our airline markets as airport-pairs rather than city-pairs since this allows a cleaner comparison across carriers which serve identical airports. In particular, the airport-pair analysis is less of measurement concerns in dealing with the On-Time-Performance variables. For example, the quality of a DFW to DCA (Ronald Reagan Washington National) flight might be perceived differently than a DFW to IAD (Washington Dulles) flight. See Brueckner et al. (2014) for a thorough discussion of the differences between airport-pairs and city-pairs.

As a result each flight appears multiple times, differing by the data collection time.

Both HHI and WN are likely endogenous. In subsequent estimations we use SSR to address endogeneity concerns.

We also conduct a test of the equivalence of the four amenity regressions. The results overwhelmingly reject that the regression coefficients are equivalent across models (1)-(4).

As a robustness check we also considered flight duration (instead of distance) and found very similar results.

While we could have chosen the presence of an airport hub at either endpoint as a measure of competition on the route, we believe that such a measure would not be informative given the overwhelming proportion of network carrier flights are either to/from a hub. In contrast, HHI provides needed variation to identify the competitive environment. Another common competition measure is the number of carriers in a market. We find that HHI and number of carriers are highly correlated.

The geometric means of the metropolitan (MSA) populations of the endpoint cities have been widely used as instruments for HHI since Borenstein and Rose (1994). Larger populations at route endpoints have higher demands and hence strengthen related market competition. Our second instrument is the enplanement ratio, which is the observed carrier’s geometric mean of enplanements at the route endpoints divided by the sum across all carriers of the geometric mean of each carrier’s enplanements at the endpoint airports. The likelihood of entry by a carrier benefits from the synergy of large-scale operations at both airports, which in turn affects the observed carriers’ route shares and HHI. Following Goolsbee and Syverson (2008) or Kim et al. (2021), we employ the following proxy for an airport operating cost by using the mean ticket fares from this airport to all other destinations except the focal destination. Airports with lower operating costs are more likely to attract entry and hence are considered more competitive.

We do not control for aircraft type in the estimations because when travelers search for flights, information about aircraft type is displayed less prominently than amenities such as Wi-Fi. In addition, consumers may not understand the differences among aircraft type or model, while the meaning of a Wi-Fi icon is better understood by most travelers. Also note that when firms make supply-side decisions, they are likely to take into account potential demand side factors. Therefore, our amenity regressions are the results of the interaction of supply-side costs/technologies and firms’ anticipation of consumer demands/preferences.

We are not aware of any quality differences among the free entertainment and paid entertainment offerings by the carriers. All of the major airlines (except JetBlue) charge a fee for Wi-Fi access (as of 27 July 2021).

We also conduct a Likelihood Ratio test to see whether the six carriers have the same estimates, which are rejected by the test results.

Variables such as HHI, Tourist, and Passengers do not vary along the route, and are thus absorbed by route fixed effects. Similarly, bankruptcy status and airline financial variables - as well as the price of \(Wi-Fi\) - are fixed at the airline level and are absorbed by airline fixed effects.

The relationship between lower posted price and days in advance of departure is not monotonic: Posted prices rise rapidly immediately prior to departure. For a thorough look at pricing patterns prior to departure see Escobari et al. (2019).

The preferred specification of the Lewbel generalized IV employs at least one external IV. In our estimation of the Lewbel, however, we are unable to locate an external IV; hence these Lewbel estimations should be considered as an exploratory robustness check.

We use the whole set of exogenous variables: \({\tilde{X}}=X\). We also tried different subsets of the exogenous variables in constructing the generated IVs and ran the same regressions as robustness checks. The results are qualitatively the same.

We perform an under-identification test (KP LM test), an over-identification test (Hansen J test), and weak IV tests (Cragg-Donald Wald F test and Stock-Yogo test) for each of the four sets of instruments. The Lewbel instruments pass most but not all of these tests.

The use of lagged values requires a panel data set. In contrast, the Lewbel et al. (2012) approach works well for cross-sectional data. The tradeoff is that the Lewbel approach requires the heteroscedasticity of the idiosyncratic errors.