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

“The availability of Wi-Fi influenced the flight selection process for a majority (66 percent) of those surveyed...and 17 percent have switched from their “preferred" airline to another airline where the odds of a Wi-Fi enabled flight were higher."

Honeywell Aerospace survey (Rabinowitz, 2014)

Abstract

We examine how competition impacts the provision of product quality. Using a unique data set of inflight amenities provided by U.S. airlines, we find that the composition of competition matters. There is significantly higher product quality - Wi-Fi, entertainment, and at-the-seat electrical power outlets - on on more competitive routes (with lower HHI). The presence of Southwest Airlines on the route, however, is shown to reduce product quality offerings. We also find significantly higher product quality on routes with more passengers, tourist destinations, and red-eye flights. We find lower posted base ticket prices on routes with Wi-Fi and entertainment amenities.

Appendix

1.1 Price regression with IV

The price regressions in the main text do not take into account that carriers choose both inflight amenities and ticket prices. There may also be an omitted variables problem. For example, there may be confounding factors beyond our explanatory variables, and these factors may also vary within a carrier or a route and thus are uncontrolled for by the carrier and route fixed effects. As a result, the possibility exists that the amenity variables may be correlated with the error terms, making inflight amenity variables in the price regressions potentially endogenous. We perform Durbin-Wu-Hausman test to check for such a possibility, with the null hypothesis being that the amenity variable is exogenous. Our tests reject that null hypothesis for Seat and Entertainment, but fail to reject the null for Wifi and Power. To deal with the potential endogeneity issue between these two variables, we implement an instrumental variable (IV) estimation method.

Construction of the IVs

Finding a proper external instrument set is often hard. Commonly employed IVs in the IO literature are cost shifters and average characteristics of products in other markets. Unfortunately, these IVs will likely violate the condition that the IV must have an indirect impact on the dependent variable instead of a direct impact. In the absence of external IVs, we construct Lewbel (2012) instruments by exploiting the higher moments of the data.²⁶

The identification strategy of the Lewbel method heavily relies on several assumptions: Suppose that the endogeneity is caused by an omitted variable u so that the error term has the form $e=\alpha \cdot u +v$. The first assumption involves the exogeneity of the independent variables vector X (which does not include the amenity variable): $E(X\cdot e)=0$. This implies that X is uncorrelated with both the omitted variable u and the idiosyncratic error v. Next, let ${\tilde{X}}$ be a sub-vector of X. Assume the ${\tilde{X}}$ is uncorrelated with $u^2$, $v_iv_j$ and $uv_i$, $i\ne j$; and $Cov({\tilde{X}},v^2)\ne 0$. Combined, these assumptions allow the following conditions to be satisfied: (1) $E(X\cdot e)=0$; (2) $cov( {\tilde{X}},e_i\cdot e_j)=0$, $i\ne j$ and (3) $\sigma ^2_i \ne \sigma ^2$: The error terms e are heteroskedastic.²⁷ These conditions allow us to construct the vector of instruments described next.

The Lewbel instruments are constructed as follows: For each inflight amenity variable – Wifi– in the first stage of 2SLS we regress the amenity variable on the exogenous subvector ${\tilde{X}}$.²⁸ Let ${\hat{\varepsilon }}$ denote the residuals. We then use the residuals to interact with the mean-centered exogenous variables – ${\tilde{X}}-E({\tilde{X}})$ – to obtain our instruments Z. Suppose that there are K variables in ${\tilde{X}}$: ${\tilde{X}}=({\tilde{x}}_{1},\cdots ,{\tilde{x}}_{k},\cdots ,{\tilde{x}}_{K})$. For each exogenous variable ${\tilde{x}}_{k}$, we create an instrument $z_{k}$ as follows:

$$\begin{aligned} z_{k}=\left[ {\tilde{x}}_{ik}-E({\tilde{x}}_{k})\right] \cdot {\hat{\varepsilon }} _{i},\quad k=1,\cdots K. \end{aligned}$$

Note that we construct a total of four sets of instruments: one for each of the four amenity variables.²⁹ Although we find that the DWH test suggests that Wifi and Power are less of a concern, nonetheless we use an IV estimation for these two variables as robustness checks. In the second stage, we use $Z=(z_1,\cdots ,z_k,\cdots z_K)$ as instruments to regress the dependent variable (price, or average one-way posted domestic fare) on the exogenous variables X and the amenity variable.

This generated Lewbel instrumental variable technique can be applied to other settings where external instruments are unavailable. This is in the same spirit as Arellano and Bond (1991) which uses lagged endogenous variables as IVs.³⁰

1.2 IV results

The IV results are presented in Table A1. Once again, we find significantly lower baseline posted prices for flights that provide Wi-Fi and Entertainment. The magnitude of these coefficient estimates has increased substantially as we find the presence of Wifi significantly reduces the base posted price of the airline ticket by $\$28$, while Entertain reduces ticket prices by $\$31$. This ticket price reduction clearly exceeds the daily potential revenue from Wifi since carriers charge just $\$16$ for a one-day Wi-Fi access pass. Even if every passenger pays for Wi-Fi, this extra revenue won’t be enough to compensate the loss of base ticket prices.

We believe the following thought experiment may provide some insights on this puzzle: Recall that the U.S. airline industry is replete with price discrimination. Before the airlines began offering Wi-Fi and entertainment inflight amenities, to segment travelers airlines relied on restrictions such as advance purchase, Saturday night stay over, ticket refundability and so on. Whether a passenger pays for inflight amenities provides a much more direct measure to infer consumer type and willingness to pay. So the baseline ticket price is for a more refined group of passengers with the lowest willingness to pay. Since the lowest consumer group is more refined, one would expect that ticket price to drop further. Given that our pricing data selects the lowest posted coach fare, we don’t observe higher quality-offerings such as United’s Economy Plus. Our conjecture is that better refinement of travelers, although reducing the fare for the lowest group, also raises the fare for higher consumer groups: One would expect more expensive upgrades from standard United Economy to, say, Economy Plus. Also, a Honeywell Aerospace survey reveals a growing demand for Wi-Fi by travelers: 17% indicated that they have switched from a “preferred" airline to another carrier that offered greater likelihood of having Wi-Fi on the flight (Rabinowitz, 2014). Accordingly, carriers may be willing to lose a few dollars on Wifi in order to retain (or attract) passengers.

For the other two flight amenity variables, we find that neither Power nor Seat provide any significant explanatory power with respect to the base airline ticket price. This is a change from the OLS estimates as Power was previously found to be associated with slightly higher ($0.81) posted ticket prices.

Table 11 IV estimates of inflight amenities and U.S. domestic airfares