University of Southern California Dissertations and Theses

Essays on competition and strategy within platform industries

(USC Thesis Other)

Essays on competition and strategy within platform industries

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON COMPETITION AND STRATEGY WITHIN PLATFORM INDUSTRIES
by
Rihyun Park
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)
December 2024
Copyright 2024 Rihyun Park

ii
Acknowledgements
I would like to express I would like to express my deepest appreciation to my advisor,
Professor Ridder Geert, for his unwavering support, guidance, and encouragement throughout my
Ph.D. journey. His insightful feedback and constant motivation have been invaluable in shaping
my research and bringing this dissertation to completion.
I am also grateful to my committee members, Professor Guofu Tan and Professor Davide
Proserpio, for their constructive comments and invaluable advice. Their expertise was truly
instrumental in shaping this work. I extend my thanks to my co-author, Dr. Hae Yeun Park, for the
collaboration that significantly contributed to this research. My heartfelt thanks go to my friends,
whose support and encouragement have been invaluable, especially during the challenging
moments of this journey.
My deepest appreciation goes to my family, whose unconditional love and belief in me
have been my greatest source of strength. Most importantly, I would like to thank my husband, Dr.
Jaehoon Lee. The past five years of our doctorate journey amid challenging times were
unforgettable—full of achievement and support. Without your love, this dissertation would not be
what it is. Thank you for your unwavering belief in me.

iii
Table of Contents
Acknowledgements......................................................................................................................... ii
List of Tables................................................................................................................................... v
List of Figures............................................................................................................................... vii
Abstract......................................................................................................................................... vii
Chapter 1......................................................................................................................................... 1
Commission Structures and Multihoming in Restaurant Platform Selection within
the Food Delivery Industry ............................................................................................................. 1
1.1 Introduction .........................................................................................................................................1
1.2 Related Literature................................................................................................................................4
1.3 Industry Background...........................................................................................................................5
1.3.1 Industry Structure as a Multi-sided Platform .............................................................................................. 6
1.3.2 Commission Dynamics During Covid-19 ................................................................................................... 9
1.4 Data ...................................................................................................................................................15
1.4.1 Listing Data ............................................................................................................................................... 15
1.4.2 SafeGraph Data ......................................................................................................................................... 16
1.5 Model.................................................................................................................................................23
1.5.1 Utility Model............................................................................................................................................. 23
1.6 Estimation..........................................................................................................................................28
1.7 Estimation Results.............................................................................................................................29
1.7.1 Demand Parameters................................................................................................................................... 29
1.7.2 Substitution Patterns.................................................................................................................................. 32
1.8 Counterfactual Analysis ....................................................................................................................35
1.9 Conclusion.........................................................................................................................................37
Chapter 2....................................................................................................................................... 40
Consumer Choice in the Food Delivery Industry ......................................................................... 40
2.1 Introduction .......................................................................................................................................40
2.2 Literature Review..............................................................................................................................41
2.3 Industry Background.........................................................................................................................43
2.3.1 Timing and Structure of the Game ............................................................................................................ 43
2.3.2 Breakdown of Consumer Payments .......................................................................................................... 44
2.3.3 Consumer Subscription Membership ........................................................................................................ 45
2.4 Data ...................................................................................................................................................47
2.5 Model.................................................................................................................................................49
2.6 Estimation..........................................................................................................................................52
2.7 Estimation Results.............................................................................................................................53
2.8 Counterfactual Analysis ....................................................................................................................55

iv
2.9 Linking Restaurant Platform Choices to Consumer Choice Problems .............................................60
2.10 Conclusion.......................................................................................................................................63
Chapter 3....................................................................................................................................... 65
Effect of Code-Share Exit:............................................................................................................ 65
Evidence from Merger of Alaska and Virgin America.................................................................. 65
3.1 Introduction .......................................................................................................................................65
3.2 Literature Review..............................................................................................................................68
3.3 Codeshare Alliance............................................................................................................................71
3.3.1 Definition................................................................................................................................................... 71
3.3.2 Alaska (AS) and American (AA) Codeshare Agreement.......................................................................... 73
3.4 Alaska Airlines’ Acquisition of Virgin America ................................................................................77
3.4.1 Merger Background................................................................................................................................... 77
3.4.2 Remedies................................................................................................................................................... 78
3.5 Data ...................................................................................................................................................81
3.5.1 Data Sources.............................................................................................................................................. 81
3.5.2 Sample Construction ................................................................................................................................. 82
3.6 Descriptive Merger Analysis.............................................................................................................87
3.6.1 Market Level Fare Changes....................................................................................................................... 87
3.6.2 Fare Changes in Overlap and Either Markets........................................................................................... 89
3.7 Codeshare Exit Effects......................................................................................................................91
3.7.1 Empirical Strategy ..................................................................................................................................... 91
3.7.2 Results....................................................................................................................................................... 93
3.7.3 Parallel Trends Assumption....................................................................................................................... 98
3.8 Demand ...........................................................................................................................................100
3.8.1 Model....................................................................................................................................................... 100
3.8.2 Estimation................................................................................................................................................ 103
3.8.3 Results..................................................................................................................................................... 104
3.9 Conclusion.......................................................................................................................................108
References....................................................................................................................................110
Appendices...................................................................................................................................113
Appendix A. Overview of Commission Cap Implementation ..............................................................113
Appendix B. Codeshare Remedy Detail................................................................................................114

v
List of Tables
Table 1. 1 Commission Rate Changes by City and Platform........................................................ 14
Table 1. 2 Summary Statistics of Restaurant, Market, and Platform Characteristics ................... 19
Table 1. 3 Restaurant Platform Choice Model Parameter Estimates............................................ 29
Table 1. 4 Cross-Price Derivatives................................................................................................ 33
Table 1. 5 Cross-Market Share Derivatives.................................................................................. 34
Table 1. 6 Change in Market Shares by Restaurant Choice When Multihoming is removed. ..... 36
Table 2. 1 DoorDash Customer Fees Based on Restaurant Commission Plan and
Customer Subscription Status....................................................................................................... 44
Table 2. 2 Summary Statistics of Order Cost, Restaurant Quality and Platform Characteristic... 48
Table 2. 3 Consumer Restaurant-Platform Choice Model Parameter Estimates .......................... 53
Table 2. 4 Counterfactual Probabilities – by Month and Platform ............................................... 57
Table 2. 5 Original Probabilities – by Month and Platform.......................................................... 57
Table 2. 6 Changes – Probabilities by Month and Platform ......................................................... 57
Table 2. 7 Probabilities of choosing multihoming restaurants given that a specific
platform was chosen ..................................................................................................................... 58
Table 3. 1 Remedy Relevant Markets with Codeshare Products ................................................... 81
Table 3. 2 Regional Airlines affiliated to Major Airlines.............................................................. 84
Table 3. 3 Codeshare Exit Analysis Sample Summary Statistics ................................................. 85
Table 3. 4 Demand Estimation Sample Summary Statistics......................................................... 87
Table 3. 5 Pre & Post Differences: Market Level Fares ............................................................... 88
Table 3. 6 Merger Effects Results.................................................................................................. 90
Table 3. 7 The Codeshare Exit Effects Results (Definition A) ...................................................... 93
Table 3. 8 The Codeshare Exit Effects Results (Definition B)....................................................... 95

vi
Table 3. 9 Parallel Trends Assumption Tests................................................................................. 99
Table 3. 10 Demand Estimation Results.......................................................................................110
Table A. 1 Implementation of Commission Caps by City (Sorted by Start Date) .......................113
Table B. 1 Virgin/American Domestic U.S. Overlap Routes.......................................................115
Table B. 2 Alaska/American Domestic U.S. Overlap Routes.....................................................116

vii
List of Figures
Figure 1. 1 Fee Structure in the Food Delivery Industry ................................................................ 7
Figure 1. 2 Platform Competition and Multihoming in the Food Delivery Industry...................... 9
Figure 1. 3 Commission Rate Evolution Before, During and After COVID-19........................... 14
Figure 1. 4 App Appearance of Restaurants on UberEats and DoorDash .................................... 15
Figure 1. 5 Percentage of Platform Choice ................................................................................... 21
Figure 1. 6 Average Percentage of Restaurant Choices by Period................................................ 22
Figure 1. 7 Average Monthly Restaurant Listings: Original vs Counterfactual Choices.............. 35
Figure 2. 1 Subscriber Behavior across Platforms........................................................................ 47
Figure 2. 2 Changes in Consumer Choices by Delivery App under a Counterfactual
Scenario Removing Multihoming Influence................................................................................. 56
Figure 2. 3 Change in Conditional Probability of Multihoming Restaurants
Given Selected Platforms.............................................................................................................. 59
Figure 2. 4 Comparison of Platform Selection by Restaurants Between Original and
Counterfactual Scenarios .............................................................................................................. 61
Figure 2. 5 Comparison of Consumer Choices over platforms Between Original and
Counterfactual Scenarios ............................................................................................................. 62
Figure 3. 1 US Airlines’ Market Share in 2016............................................................................. 66
Figure 3. 2 Three Types of Airlines Products ............................................................................... 72
Figure 3. 3 AS-AA Codeshare Markets (Traditional)................................................................... 74
Figure 3. 4 AS-AA Codeshare Markets (Virtual).......................................................................... 74
Figure 3. 5 AS-AA Codeshare Total Passengers (Traditional) ..................................................... 76
Figure 3. 6 AS-AA Codeshare Total Passengers (Virtual)............................................................ 76

viii
Figure 3. 7 VX Presence on AS-AA Codeshare Routes ............................................................... 98
Figure 3. 8 The Nesting Structure 1............................................................................................ 101
Figure 3. 9 The Nesting Structure 2............................................................................................ 106

ix
Abstract
This dissertation explores the competitive dynamics and strategic decision-making within
the food delivery sector, emphasizing the multihoming behavior of restaurants and the influence
of commission structures. The first essay develops a discrete choice model to quantify how varying
commission rates, consumer market share, and the number of competitors affect restaurants’
platform choices. Results demonstrate that lower commission rates, a larger consumer base, and
reduced competition among restaurants increase the likelihood of platform adoption, revealing the
hidden competitive benefits of multihoming.
The second essay models consumer behavior within the food delivery market as part of a
sequential game involving platforms, restaurants, and consumers. This analysis considers the
impact of commission plans on consumer and restaurant choices, assessing how platform strategies
affect consumer preferences and market equilibrium. It emphasizes the strategic interdependencies
between stakeholders, enriching the understanding of platform competition.
The third essay investigates the impact of airline mergers on market outcomes, specifically
analyzing the Alaska Airlines-Virgin America merger. Using a Difference-in-Differences approach,
it examines how the termination of a codeshare agreement influenced pricing and demand. Results
show that the cessation led to fare changes and shifts in market competition, offering insights into
the implications of airline alliances and regulatory interventions.
This dissertation mainly contributes to the understanding of competition and strategy in
multi-sided markets, providing valuable implications for platform operators and policymakers. It
employs methodologies such as discrete choice modeling, game theory, and causal inference,
enhancing the analytical framework for examining complex interactions in platform industries.

1
Chapter 1
Commission Structures and Multihoming
in Restaurant Platform Selection within the Food Delivery Industry
1.1 Introduction
Over the past decade, the food delivery industry has undergone significant transformation,
largely driven by the emergence of platform-based business model. The COVID-19 pandemic
further accelerated this shift, as restrictions on in-person dining drove consumers toward online
food delivery services, resulting in the market size of the sector expanding rapidly. For example,
cumulative monthly sales in the United States more than doubled in 2020 compared to 2019.1
Central to this transformation are platforms like UberEats and DoorDash, which operate as
intermediaries connecting restaurants and consumers within a multisided market—a structure
where value is derived from interactions between distinct user groups. (Rochet & Tirole, 2003)
These platforms charge commission fees to restaurants and delivery or service fees to consumers,
creating network externalities where the value for one group increases as the other grows. However,
the prevalent commission structures have placed significant financial strain on restaurants, often
resulting in reduced profit margins or operational challenge.
In response to these challenges, several local governments introduced commission caps in
2020, temporarily limiting the fees platforms could charge restaurants to alleviate financial
1
Janine Perri, "Which company is winning the restaurant food delivery war?" Bloomberg Second Measure, March 14,
2023, https://secondmeasure.com/datapoints/food-delivery-services-grubhub-uber-eats-doordash-postmates.

2
pressures during the pandemic. Following this, platforms like UberEats and DoorDash launched
tiered commission plans in 2021, offering varying rates tied to different service levels. These
changes provide a unique research opportunity to analyze the impact of transparent commission
structures on restaurant platform choices. For this study, I constructed a dataset capturing platformspecific characteristics and restaurant listings across multiple platforms, enabling a detailed
examination of the factors influencing these choices.
This research is particularly timely as platform-based business models proliferate across
industries, with companies adopting diverse pricing plans to attract users. The food delivery
industry offers a distinct lens for this analysis due to the availability of transparent commission
data introduced post-pandemic. Such transparency allows for a nuanced understanding of the
determinants of restaurant platform choices.
Understanding these determinants is essential for both platform providers and
policymakers. Platforms like UberEats and DoorDash can leverage these insights to optimize
commission structures and manage multihoming—the practice of restaurants listing on multiple
platforms— to enhance market share and competitive positioning. Policymakers can use these
findings to design regulatory frameworks that balance platform profitability with restaurant
sustainability, fostering equitable outcomes for all stakeholders.
This dissertation seeks to address several key questions within the food delivery industry,
focusing on the factors that drive restaurants' decisions to list on delivery platforms. It investigates
how market characteristics, platform attributes, and restaurant-specific factors influence the
likelihood of platform adoption, offering insights into restaurants' demand for platform listings.
The study also explores the prevalence of multihoming and examines whether restaurants derive
additional benefits from multihoming compared to single-homing.

3
This research is also evaluating the impact of new commission structures introduced by
platforms like DoorDash and UberEats. Specifically, it analyzes how these tiered commission
plans influence restaurant platform choices, shedding light on the role of pricing strategies in
shaping competitive dynamics. The study also investigates substitution patterns between
DoorDash and UberEats, assessing how changes in one platform's commission rates and consumer
market share affect the likelihood of restaurants choosing the other platform.
Finally, the dissertation includes a counterfactual analysis to examine the market
implications of removing the multihoming option. This analysis explores how restricting
restaurants to a single platform would impact the market shares of UberEats and DoorDash,
providing insights into the potential consequences for competition and platform dominance.
The dissertation is organized as follows: Section 1.2 reviews the theoretical and empirical
literature on platform competition, multihoming, and commission structures. Section 1.3 provides
an overview of the food delivery industry, including the impact of COVID-19 and the introduction
of commission caps and new commission plans. Section 1.4 details the data collection process and
describes the key characteristics of the dataset. Section 1.5 outlines the multinomial logit model
used in the analysis, including the specification of utility functions. Section 1.6 explains the
estimation approach. Section 1.7 presents the results of the model estimation, including demand
parameters and substitution patterns. Section 1.8 explores the counterfactual analysis on the impact
of removing the multihoming option on restaurant platform choices. Finally, Section 1.9
summarizes the key findings, discusses their implications for platform strategy and policy, and
suggests directions for future research.

4
1.2 Related Literature
This project is informed by three relevant streams of literature. First, theoretical analyses
of platform competition are essential for understanding the economics of the platform industry.
Rochet and Tirole (2003) provide a general framework for two-sided markets and explain the
network externalities, how the benefit of one side depends on the size of the other side. Also, when
there are multiple platforms in the industry, competition arises as each side of the platform must
choose which platform to join. Armstrong (2006) models the platform structure with one side
joining a single platform (single homing) while the other joins multiple platforms (multihoming).
This competitive bottleneck model found that platforms have monopoly power on the multihoming
side by providing access to their singlehoming customers, allowing platforms to make a loss on
the singlehoming side and recoup the losses from the multihoming side. Recent studies have started
to explore platform structures where both sides engage in multi-homing (Teh et al., 2023; Bakos
and Halaburda, 2020). Additionally, Tan and Zhou (2021) examine how platform competition and
entry affect prices, profits, and consumer surplus in multi-sided markets, revealing that increased
competition can sometimes lead to higher prices and lower consumer surplus, contrary to
traditional expectations.
Second, many empirical studies have examined platform competition and multihoming
users. The empirical studies in the media industry resemble the food delivery industry, as the main
role of platforms includes listing and marketing. Fan (2013) and Rysman (2004) studied the
newspaper and Yellow Pages industries, respectively, using a Cobb-Douglas form for advertiser
benefits to ease modeling. Song (2021) developed a simultaneous equation model between utilitymaximizing viewers and profit-maximizing advertisers. The main empirical strategies were

5
applied from discrete choice demand models established by Berry (1994) and Berry et al. (1995),
which enabled counterfactual simulations. Gentzkow (2006) extended these models to allow
agents to make multiple choices, an approach applicable to analyzing the multihoming behavior
of individual restaurants in food delivery platforms.
Third, recent research has examined the impact of commission caps on the food delivery
industry using discrete choice models. Sullivan (2024) was the first to assess the impact of
commission caps with such models, finding that while commission caps raise restaurant profits,
they also lead to higher consumer fees and reduced consumer welfare. Commission caps bolster
restaurant profits but at the expense of consumers and platforms. However, the study did not
account for recent developments, such as new commission plans that diversify their delivery
service levels and link to customer subscription models. Li and Wang (2021) show that commission
caps benefit chain restaurants more than independent ones, as platforms respond by promoting
chains and increasing delivery fees. This evolving industry is gaining significant attention from
researchers, reflecting its complex dynamics and the impact of regulatory measures.
In summary, these streams of literature provide a comprehensive framework for
understanding the platform competition, empirical modeling of multihoming behaviors, and the
regulatory impacts on the food delivery industry. This study builds on these foundations to explore
the decision-making processes of restaurants in choosing between major food delivery platforms.
1.3 Industry Background
This chapter will provide an overview of the industry structure, including the fees and
commissions charged by delivery platforms.

6
1.3.1 Industry Structure as a Multi-sided Platform
The food delivery industry is identified as a two-sided or multi-sided platform, as described
by Rochet and Tirole (2003). In contrast to traditional one-sided markets characterized by direct
transactions between buyers and sellers, two-sided markets entail the involvement of multiple user
groups interacting through a platform. Within the food delivery sector, three key participants are
involved: restaurants, customers, and couriers. The value that one group gains from utilizing the
platform is influenced by the number of users present on the opposing side, a phenomenon known
as network externalities. For instance, restaurants are more likely to participate in a food delivery
app as the platform gains popularity among customers. In a parallel manner, an increase in the
number of restaurants on the platform provides customers with a broader array of choices,
enhancing the app’s perceived value.
Food delivery platforms operate as intermediaries, facilitating transactions between
restaurants and consumers. They charge commission fees to restaurants and delivery/service fees
to consumers. This dual-sided fee structure supports the platforms' revenue models while providing
value to both sides of the market. The competition between platforms is driven by their ability to
attract more users on each side, leveraging network externalities to enhance their value
propositions.
Network externalities play a crucial role in the food delivery market. The value of a
platform to one user group (e.g., restaurants) increases with the number of users on the other side
(e.g., consumers). For instance, a larger number of consumers on a platform makes it more
attractive for restaurants to list their services, and vice versa. This interdependency creates a
feedback loop that can amplify the competitive advantages of dominant platforms.

7
To facilitate these transactions, food delivery platforms apply three distinct fees:
commission, service, and delivery fees. Restaurants are charged a commission fee for the services
of providing a payment system, connecting them with drivers, and marketing their business.
Customers are charged a service fee in exchange for the convenience of utilizing the delivery
service. The delivery fee is intended to cover the expenses associated with coordinating the
delivery service and compensating the driver. All pricing is based on a per-order basis. Customers
pay a service fee based on a percentage of the subtotal, as well as a flat delivery fee that varies in
amount. The restaurants are charged a commission rate based on a percentage of the order totals.
(Figure 1. 1)
Figure 1. 1 Fee Structure in the Food Delivery Industry
The diagram illustrates the fee structure in food delivery platforms, highlighting the relationships between restaurants, consumers,
and the platforms. Restaurants pay commission fees to the platforms, which in turn provide marketing, payment processing, and
delivery coordination services. Consumers are charged delivery fees and service fees for the convenience of having food ordered
and delivered to their location. This fee structure supports the revenue model of food delivery platforms, allowing them to facilitate
transactions and grow their user base through network externalities.
For example, when using UberEats to purchase a $30 meal, the service fee, delivery fee,
and tax added during checkout result in an approximate total cost of $40. However, the restaurant
only receives $21 after deducting a 30% commission fee of $9 imposed by the delivery platform.
This structure leaves the restaurant with $21 in revenue, while the platform takes $9 in commission
from the restaurant and $10, excluding tax, from the consumer.

8
The complex structure of multi-sided platforms, along with the associated fees and
commissions levied by food delivery platforms, have significant implications for restaurants,
customers, and delivery drivers.
Restaurants must decide whether to list on DoorDash, UberEats, or both (multihoming).
This decision involves weighing the benefits and costs associated with each platform, including
potential reach, visibility, and operational complexities. Listing on multiple platforms can increase
visibility and demand but also comes with higher costs and operational complexities.
Consumers face decision problems when choosing which food delivery platform to use and
which restaurants to order from, which will be further discussed in the Chapter 2. Consumers
benefit from platforms with a wide selection of restaurants, lower fees, and better service quality.
These decisions involve evaluating different aspects of the platforms and restaurants to make a
choice that best suits their preferences and needs.
The diagram in Figure 1. 2 illustrates the competition of platforms within the food delivery
industry, highlighting the relationships and decision problems faced by restaurants and consumers.

9
Figure 1. 2 Platform Competition and Multihoming in the Food Delivery Industry
The diagram shows the competition of platforms within the food delivery industry with multi-sided participants:
consumers and restaurants. Consumers face decision problems as they choose which platform to use (DoorDash or
UberEats) and which restaurants to order from within those platforms. Similarly, restaurants decide whether to list on
DoorDash, UberEats, or both, a strategy known as multihoming. This competition and decision-making process
highlight the dynamic nature of the food delivery market, where platforms must attract both consumers and restaurants
to succeed. The orange shaded area represents restaurants listed on DoorDash, while the green shaded area represents
those listed on UberEats. Restaurants that are listed on both platforms are depicted in the intersection of the two shaded
areas, indicating their multihoming strategy. Likewise, consumers who order through DoorDash are represented by
the people in the orange shaded area, and vice versa. The paper focuses on the restaurant side's decision-making,
highlighted in bold in the diagram.
1.3.2 Commission Dynamics During Covid-19
This section will examine the evolution of commission dynamics during the COVID-19
pandemic, which provided a unique opportunity to analyze how these rates influence restaurant
platform choices.
1.3.2.1 Covid-19 Lockdown
The COVID-19 pandemic has significantly impacted the food delivery industry, as
restrictions and safety protocols have resulted in a notable increase in the utilization of food

10
delivery services. The market size of the industry saw a substantial growth as more individuals
opted to remain at home and use online food ordering services. This brought significant changes
and challenges for small-scale eateries.
While some restaurants successfully sustained their operations by collaborating with
delivery platforms, others faced difficulties adjusting to the expenses associated with commission
fees, which were individually negotiated and varied widely from 15% to as high as 40%. (Figure
1.3) The imposition of these fees presents a substantial challenge for an industry characterized by
narrow profit margins, especially for smaller local eateries. For instance, as previously discussed
with the example restaurant, if meal preparation costs total approximately $20, the restaurant may
operate with minimal profit margins or even face potential losses.
The COVID-19 outbreak also led to increased overall costs for restaurants, including
higher wages, reduced labor availability, and supply chain disruptions resulting in elevated
ingredient costs. Restaurants attempted to mitigate the influence of commission fees by modifying
their menu prices on the application, but this approach risked decreasing customer demand and
revenue. The combination of commission fees, decreasing customer demand, and rising costs
created challenges for restaurants to operate profitably within this ecosystem.
1.3.2.2 Commission Caps
In response to the adverse financial impact of commission fees imposed by food delivery
platforms, several local governments enacted commission caps to support local restaurants. For
example, San Francisco was the first city to declare a commission cap on food delivery platforms

11
during the COVID-19 pandemic. 2
This initial cap was introduced in April 2020, setting a
maximum commission rate of 15%. San Francisco later made this cap permanent in June 2021.3

New York City also enacted commission caps, with a 20% limit on delivery services and a 5% cap
on marketing fees during the pandemic and made these caps permanent in August 2021. Other
cities such as Los Angeles and Chicago had also considered and implemented commission caps to
support their local restaurant industries.
Within the San Francisco Bay Area, which is the focus of this dissertation, numerous cities
adopted commission caps to support their local eateries. The details about the specific cities and
the start dates of these commission caps are provided in the Appendix A.
A recent academic paper by Sullivan (2024) analyzed the efficacy of commission caps on
food delivery platforms. The study found that commission caps can reduce costs for restaurants,
potentially leading to decreased prices for consumers. However, platforms often impose higher
consumer fees to offset the anticipated decline in revenue caused by these caps, resulting in a
decline in demand and perceived value for restaurants. While commission caps alleviate financial
strain on restaurants, they impose higher costs on consumers.
1.3.2.3 New Structured Commission Plans
In April 2021, DoorDash introduced a new commission structure for restaurant
establishments, representing a substantial change in the food delivery sector. This policy allows
restaurants to choose among three distinct commission rates —15%, 25%, or 30%— based on their
2 Mayor London Breed Announces Delivery Fee Cap to Support San Francisco Restaurants During COVID-19 Pandemic ( 2020,
April 10) Retrieved from “https://sfmayor.org/article/mayor-london-breed-announces-delivery-fee-cap-support-san-franciscorestaurants-during-covid”
3 Tanay Warerkar, “Food App Delivery Commission in S.F. Capped at 15% to Help Restaurants,” San Francisco Chronicle, June
15, 2021, accessed July 9, 2024, https://www.sfchronicle.com/food/restaurants/article/Food-app-delivery-commission-in-S-Fcapped-at-16266468.php.

12
marketing and delivery service needs. This standardized pre-set commission rate policy is the first
of its kind in the food delivery industry.
DoorDash offers three structured commission plans: Basic, Plus, and Premium. The Basic
plan charges a 15% delivery commission per order, providing a limited delivery radius without
marketing services. The Plus plan, with a 25% commission, expands the delivery radius and
includes some app visibility marketing services. The Premium plan, at 30%, offers the largest
delivery radius and extensive marketing services, such as greater app visibility, priority listing,
more promotional tools, and customer incentives, enhancing restaurants' competitiveness.
DoorDash's pre-set pricing options are applied consistently to all restaurants, eliminating
the need for individual commission negotiations as was done previously. (Table 1.1) This
structured pricing approach facilitates the optimization of the pricing process and enhances the
efficiency of businesses in managing their pricing strategies across a wide customer demographic.4
Following DoorDash’s introduction of this new commission structure, UberEats
implemented a comparable commission model in September 2021, allowing restaurants to select
from three commission rates tailored to their specific service needs. Other food delivery platforms
have also adopted similar new commission models as well.
DoorDash's approach aims to provide transparency and flexibility to restaurants. However,
as reported in the Wall Street Journal article "DoorDash Allows Restaurants to Choose
Commissions in Post-Pandemic Future," the implications of this revised commission model for
4 Structured pricing refers to the practice of offering multiple pricing options or packages for customers, based on factors such as
the level of service or features they require. This approach allows businesses to cater to different customer segments without
negotiating individual prices for each transaction. In the case of food delivery companies, structured pricing involves providing
different commission rates based on the level of services that restaurants require. Restaurants can choose from these pre-set
commission rates according to their specific needs and preferences, offering a degree of customization and flexibility.

13
small businesses and customers are still uncertain.5
Reduced commission rates could benefit small
businesses, but they might experience lower visibility on the app, posing challenges in competing
with restaurants that choose higher commission rates and receive preferential treatment.
According to an article in the Wall Street Journal, numerous restaurants feel compelled to
pay higher commission fees on food delivery platforms to increase the volume of their orders.
With the gradual decline of the COVID-19 pandemic and the resumption of in-person dining, there
is a possibility that restaurants will reduce their dependence on food delivery platforms. They may
choose to implement a basic commission structure or decrease their pricing to cultivate a loyal
customer base. Ultimately, the selection of a commission plan will depend on factors such as the
owner's assessment of their cuisine, their customer dining habits, and the effects of COVID-19 on
local enterprises.
The newly introduced commission plans can significantly influence restaurants' decisions
regarding platform adoption. The varying visibility and benefits linked to each commission
structure can directly affect how restaurants choose between different platforms. Therefore, it is
essential to investigate the impact of these new commission structures on restaurants' platform
choices to understand their broader implications on local businesses and customer behavior.
The decision for restaurants to adopt a particular food delivery platform under the new
commission plans is driven by both financial and strategic considerations. While the platforms
allowed restaurants to choose their commission plans, this study assumes that after the introduction
of the new commission plans, all restaurants opted for the plans that were exogenously determined.
5 Rana, Preetika. 2021.“DoorDash Allows Restaurants to Choose Commissions in Post-Pandemic Future”, The Wall Street Journal,
April 27. https://www.wsj.com/articles/doordash-allows-restaurants-to-choose-commissions-in-post-pandemic-future11619517614?mod=Searchresults_pos3&page=1

14
This assumption isolates the effect of the new commission plans on platform adoption.
Understanding these impacts requires a comprehensive analysis of how the new commission
structures affect restaurants' platform preferences, market dynamics, and customer interactions.
The evolution of these commission rates, from pre-pandemic variability to the structured
plans of the post-pandemic era, is summarized in Table 1. 1 and visually depicted in Figure 1.3,
highlighting the key changes that shaped the industry during this period.
Table 1. 1 Commission Rate Changes by City and Platform
Time Period
DoorDash UberEats
Cities Under Cap Cities Without Cap Cities Under Cap Cities Without Cap
2020 Jan – 2020 March Custom Pricing Custom Pricing Custom Pricing Custom Pricing
2020 April – 2021 March Comm. Cap Custom Pricing Comm. Cap Custom Pricing
2021 April – 2021 Aug New Comm. Plan New Comm. Plan Comm. Cap Custom Pricing
2021 Sep – 2021 Dec New Comm. Plan New Comm. Plan New Comm. Plan New Comm. Plan
Note: This table illustrates the commission rate policy for DoorDash and UberEats from January 2020 to December
2021, segmented by cities with and without commission caps. For detailed information about the specific cities and
start dates of these commission caps, refer to the Appendix A. “Comm.” Stands for “Commission”.
• Custom Pricing: Commission individually negotiated; rates not observable.
• Comm. Cap: Commission regulated by local authorities; actual cap rates vary by cities.
• New Comm. Plan: Structured commission rates introduced by platforms; commission rates can be inferred.
Figure 1. 3 Commission Rate Evolution Before, During and After COVID-19
The diagram illustrates the evolution of commission rates before, during, and after the COVID-19 pandemic. Before COVID-19,
commission rates varied widely, reaching up to 40%. During the pandemic, commission caps were introduced, significantly
lowering the rates to as low as 15% to support financially struggling restaurants. With the introduction of new commission plans
by platforms like DoorDash and UberEats, restaurants now have the option to choose from structured rates of 15%, 25%, or 30%,
based on their service needs. This new structure aims to provide more flexibility and transparency for restaurants, although the
implications for small businesses and consumers remain uncertain.

15
1.4 Data
1.4.1 Listing Data
To collect the data required for this research, I gathered all the monthly restaurant listings
from DoorDash and UberEats in the San Francisco Bay Area from January to December in 2021.
I scraped the data twice a month, ensuring a comprehensive and up-to-date dataset. This dataset
contains highly detailed information, including the restaurant's address, name, zip code, cuisine,
rating, latitude, longitude, delivery fee, delivery time, pickup time, service fees, menu prices,
newly added restaurants, memberships (DashPass/Uber One), and any available promotions. I
included restaurants that were listed on at least one platform during each month.
Figure 1. 4 App Appearance of Restaurants on UberEats and DoorDash
The images display the interface of food delivery apps UberEats and DoorDash, showcasing various data points for
each listed restaurant. The information includes restaurant name, address, rating, cuisine category, price indicator,
membership (e.g., DashPass or Uber One), delivery time, delivery fee, pick-up time, and distance. This detailed view
helps in analyzing the factors influencing restaurant platform choices and understanding the user experience on these
platforms.

16
To examine whether each restaurant was listed on both platforms in a given month, the
listings from DoorDash and UberEats were merged. This process generated the multihoming
indicator, which is crucial for analyzing the platform choices of restaurants. To uniquely identify
each restaurant business, a combination of business name, zip code, address, longitude, and
latitude was used. This was particularly challenging due to the different formats used by the two
platforms to present address information. Significant effort was dedicated to standardizing these
formats to ensure consistency across the datasets.
As previously noted, this dataset is unique and detailed, offering tractability of commission
rates for each restaurant. The period covered encompasses both the commission cap and the new
commission plan, facilitating an accurate inference of commission rates. By observing the delivery
fees and platform memberships, it is possible to deduce the commission rate each restaurant is
paying to each platform, as the new commission plan clearly delineates the structure. This level of
detail enables a precise analysis of how changes in commission structures and policies impact
restaurant platform choices.
1.4.2 SafeGraph Data
In addition to the restaurant listings data, I utilized a dataset from SafeGraph, which
provides highly accurate and detailed data about physical locations, known as points of interest
(POI). They collect and curate information about where places are located, their attributes and how
people interact with them. SafeGraph gathers anonymized location data from mobile devices to
track foot traffic and spending patterns at various establishments, ensuring privacy through data
aggregation and anonymization. SafeGraph data covers the same time period as the restaurant
listings, from January to December 2021.

17
The spend data includes aggregated, anonymized credit and debit transaction data at
specific businesses. This dataset provides detailed statistics on transaction numbers and volumes
and includes information on where else consumers spend money.
To find proxies for user market share in this model, I first identified the business with the
highest number of visits in the gas station category within each postal code area. Gas stations were
chosen because they are common and have minimal relation to the food delivery industry, unlike
grocery stores, health and personal care stores, and restaurants, which are partially or wholly
related to the delivery industry (e.g., CVS on DoorDash, 7-Eleven on DoorDash, Sprouts Farmers
Market on UberEats). Therefore, gas stations can serve as a representative indicator of general
consumer spending patterns in an area.
Next, I used the “related_delivery_service_pct” metric, which represents the percentage of
customers who also spent money on specific online delivery services during the month. This metric
allowed me to estimate the market share between UberEats and DoorDash among those who spent
money at gas stations in a given month within the SafeGraph dataset. The
“related_delivery_service_pct” helped estimate the market share for UberEats and DoorDash by
indicating the percentage of gas station customers who also use these food delivery services.
The SafeGraph data also provided monthly foot traffic data, which is the aggregated raw
counts of visits to each business from a panel of mobile devices over a given month. This data
answers questions such as how often people visit, how long they stay, where they come from, and
where else they go. The data is anonymized and aggregated to provide insights into visitor volume
and behavioral patterns.
To quantify economic activity within each market, I focused on aggregating foot traffic
data for essential consumption categories, such as 'General Merchandise Stores' and 'Grocery

18
Stores', since these remained significant even during the COVID-19 lockdown. By summing the
foot traffic data by ZIP code on a monthly basis, I derived a proxy for economic activity levels in
each market. This aggregated foot traffic data serves as a market characteristic for each ZIP code,
providing a robust indicator of local economic activity during the pandemic.
SafeGraph provides additional information on each business, such as brand indicators and
full-service restaurant indicators. The data identifies thousands of distinct brands, including major
global chains and regional brands. With brand information, I could differentiate between chain and
independent local restaurants. Additionally, using NAICS code6
, SafeGraph allows identification
of full-service restaurants, defined as establishments providing table service or dining area meals,
which is also crucial information on restaurant delivery business.
I focused exclusively on the San Francisco Bay Area for this analysis. This area includes
nine counties encompassing multiple cities and numerous zip codes in Northern California. The
San Francisco Bay Area was chosen for this project due to its diverse and dynamic restaurant
industry, high adoption rate of food delivery services by both restaurants and consumers, and its
role as a major urban center with a variety of dining options. Additionally, commission caps have
been implemented at different municipal levels, creating variability in commission rates.
These detailed geographic boundaries ensure a focused analysis on a specific, significant
market. This makes the San Francisco Bay Area an ideal location to study the competitive
dynamics between major delivery platforms like DoorDash and UberEats within a local market.
6 The North American Industry Classification System (NAICS) was developed by the US Census Bureau and consists of numeric
codes up to 6 digits in length. This system is useful for categorizing businesses. The 2017 version of NAICS used in the dataset
includes the code 7225 for restaurants and other eating places, with specific subcategories for full-service restaurants (722511),
limited-service restaurants (722513), Cafeterias, Grill Buffets, and Buffets (722514), snack and nonalcoholic beverage bars
(722515).

19
Instead of expanding the geographic scope to the entire metropolitan area in the US, I chose to
concentrate on a more focused area at the zip code level to better understand the micro-level
platform adoption problem among restaurants. Although DoorDash is dominant in this market,
examining the competitive dynamics remains meaningful to understand the broader implications
of commission changes and market strategies.
The cleaned dataset for this analysis includes balanced data on 17,530 restaurants from the
San Francisco Bay Area, spanning 87 cities and 203 zip codes in California. The time periods
considered are January to March and September to November of 2021, allowing for a comparative
analysis before and after the implementation of new commission plans by major food delivery
platforms. This dataset focuses on restaurants that engaged with at least one of the two major
platforms, DoorDash and UberEats, during the whole periods. The unit of analysis is monthly,
providing a detailed and consistent view of restaurant listings and platform adoption over time.
The cleaned dataset offers a detailed view of restaurant behaviors and choices across different
platforms. Here are some key summary statistics for each variable:
Table 1. 2 Summary Statistics of Restaurant, Market, and Platform Characteristics
Variable Mean Std. Dev. Min Max Obs.
Meal Price ($) 11.869 4.599 5 35 105,180
Rating 4.070 1.218 1 5 105,180
Brand Indicator
(1 = Yes, 0 = No) 0.173 0.378 0 1 105,180
Full-Service Indicator
(1 = Yes, 0 = No) 0.336 0.472 0 1 105,180
Foot Traffic by zip code 5089.899 5173.081 2 32568 105,180
Fee Cap Indicator by City 0.829 0.376 0 1 105,180
New Commission Plan Indicator 0.500 0.500 0 1 105,180
Commission Rate on UberEats (%) 20.955 7.339 15 30 46,898
Commission Rate on DoorDash (%) 22.198 7.494 15 30 91,034
Consumer Market Share on UberEats by zip code 0.115 0.074 0.01 1 46,898
Consumer Market Share on DoorDash by zip code 0.152 0.078 0.01 0.75 91,034
Competitors per Cuisine on UberEats by zip code 8.682 9.190 1 82 46,898
Competitors per Cuisine on DoorDash by zip code 16.265 22.606 1 149 91,034
Premium Plan Share on UberEats by zip code 0.346 0.324 0 1 46,898

20
Premium Plan Share on DoorDash by zip code 0.391 0.400 0 1 91,034
Average Delivery Fee on UberEats by zip code ($) 1.697 0.865 0 4.742 46,898
Average Delivery Fee on DoorDash by zip code ($) 0.212 0.149 0 0.525 91,034
Note: This table provides key summary statistics for various variables related to restaurant, platform, and market characteristics.
The variables include menu prices, ratings, brand indicators, full-service indicators, foot traffic, fee cap indicators, commission
fees on UberEats and DoorDash, and the total number of competitors per cuisine by zipcode.
Restaurant characteristics in the dataset include meal price, rating, brand indicator, and fullservice indicator. Meal price represents the scale of meal prices ($ to $$$$) typically registered by
each restaurant on Google Maps, indicating the cost level of the restaurant, with $ meaning $1 to
$10 per person and $$$$ meaning $30 to $40 per person. Rating refers to the 1 to 5 scale rated by
past consumers on Google Maps, reflecting customer satisfaction. I used meal price and rating
from Google Maps instead of each platform's listings to maintain objectivity as restaurant
characteristics. The brand indicator is a binary variable indicating whether a restaurant is part of
an international or local chain (1) or independent (0), while the full-service indicator is another
binary variable denoting whether a restaurant provides full table service (1) or not (0). Both
indicators are derived from SafeGraph data.
Market characteristics include the foot traffic by zip code and the fee cap indicator by city.
Foot traffic, as mentioned, is sourced from SafeGraph data, providing insights into the number of
visits to each establishment within a specific area. The fee cap indicator by city is a binary variable
that indicates whether a city has implemented a cap on commission rates (1) or not (0), reflecting
local regulatory measures affecting the cost structure for restaurants. The new commission plan
indicator is also included, which is a binary variable showing whether the new commission plan
was in effect (1) or not (0), facilitating the analysis of the impact of these plans on restaurant
behaviors.
Platform characteristics in this model are commission rates, consumer market share, total
number of competitors per cuisine, premium plan share, and average delivery fee. The commission

21
rates reflect the cost of using each platform, ranging from 15% to 30%. Consumer market share is
calculated using lagged data from popular gas stations within each zip code to estimate the local
penetration of UberEats and DoorDash. The total number of competitors per cuisine indicates the
number of competing restaurants with the same cuisine within each zip code, providing insight
into the competitive landscape across different cuisine categories. The premium plan share shows
the fraction of restaurants in the market with a premium plan subscription; higher premium plan
share indicates a larger fraction of restaurants in the market with a premium plan. Finally, average
delivery fees provide insights into the cost structure for customers on each platform. The delivery
fee data utilized represents the delivery fee amount for non-subscribers, as presented on the app.
These variables collectively enable a comprehensive analysis of the impact of commission rates
and other factors on restaurant behaviors and platform dynamics.
Figure 1. 5 Percentage of Platform Choice
The bar graph shows the percentage of restaurants choosing each platform (DoorDash only, UberEats only, or both)
over the entire study period. This visualization provides an overview of the distribution of restaurant listings across
the platforms, indicating a preference for DoorDash and a notable portion of restaurants opting for multihoming.
To illustrate the distribution of restaurant choices across the platforms, two summary plots
were created. Figure 1.5 provides an overview of the distribution of restaurant listings across the
platforms for the entire study period. This bar graph highlights the preference for DoorDash, with
55.4% of restaurants choosing to list exclusively on this platform. In comparison, only 13.5% of

22
restaurants opted to list solely on UberEats. Interestingly, a substantial portion of restaurants,
31.1%, chose to engage in multihoming by listing on both platforms. This data underscores
DoorDash's dominant market presence in the San Francisco Bay Area but also emphasizes the
strategic importance of multihoming for restaurants seeking to maximize their reach and customer
base.
Figure 1. 6 Average Percentage of Restaurant Choices by Period
The bar graph illustrates the average percentage of restaurant choices by platform for two distinct periods: January to
March and September to November. The graph highlights the differences in platform preference before and after the
introduction of new commission plans.
Figure 1. 6 displays the average percentage of restaurants choosing each platform in two
distinct periods: January to March and September to November. This plot reveals significant
changes in platform preferences before and after the introduction of new commission plans.
Notably, DoorDash maintained a dominant position throughout both periods, but there was a
visible shift in restaurant choices. In the first period, 62.0% of restaurants opted for DoorDash,
while this number decreased to 48.5% in the second period. Conversely, the percentage of
restaurants choosing UberEats increased from 12.5% to 14.3%. The multihoming option (choosing
both platforms) also saw an increase from 25.5% to 37.2%. These changes indicate that the new

23
commission plans had a significant impact on restaurant platform choices, encouraging more
restaurants to consider multihoming or switching platforms.
These summary statistics and visualizations provide valuable insights into the restaurant
listing preferences and the competitive landscape of food delivery platforms in the San Francisco
Bay Area. By focusing on a specific, significant market, this analysis captures the nuanced
dynamics of platform adoption and the effects of commission structures on restaurant choices.
1.5 Model
This study investigates the decision-making factors of restaurants in selecting between two
prominent food delivery platforms—UberEats and DoorDash—or opting to list on both. The
analysis employs a multinomial logit framework with individual-specific random effects to model
the utility derived from each choice. This approach accommodates both observable and
unobservable factors influencing restaurant decisions, offering a comprehensive understanding of
their platform selection behavior.
The model is considered static, as the monthly platform choices of restaurants lack
intertemporal dependency. Restaurants are free to opt in or out of the platform each month without
any constraints, which justifies the static nature of the model.
1.5.1 Utility Model
The utility that a restaurant derives from listing on UberEats, DoorDash, or both platforms
is modeled using the multinomial logit framework, incorporating random effects to account for
individual-specific preferences and unobserved heterogeneity. Following the methodologies of

24
Rysman (2004) and Fan (2013), the utility specifications for the three alternatives are defined as
follows:
The utility for restaurant j in market m at time t from listing on UberEats(u) or Doordash(d)
is influenced by factors such as commission rates, consumer market share, platform-specific
attributes, market conditions, and restaurant-specific characteristics. Formally, it is expressed as:
𝑝∙𝑚𝑡
𝑅
represents the commission rate charged by UberEats and Doordash to restaurants at
time t and market m. The commission rate varies by time and market, as discussed in Table 1. 2.
Regulatory interventions led to the imposition of commission rate caps, initially set at 15%.
Following this period, platforms transitioned to charging standardized rates of 15%, 25%, or 30%,
depending on the level of service or contract type. These regulatory caps and subsequent fixed
rates were externally imposed and uniformly applied across various markets, suggesting that they
were not influenced by short-term fluctuations in restaurant demand or other unobserved factors
specific to individual restaurants. The standardized nature of these commission rates, determined
by regulatory requirements and platform-wide pricing strategies rather than market-specific
conditions, supports the argument that these rates can be treated as exogenous in our model.
𝑠∙𝑚𝑡−1
𝐶 denote the market share of UberEats and Doordash among the consumers in market
m at time t-1. Given the multi-sided nature of the platform industry (Song 2021), network
externalities imply that the demand for one group (consumers) affects the demand for the other
(restaurants). To address the potential endogeneity of consumer market share, I used lagged
consumer market share as an explanatory variable. Consumer market share was calculated using

25
lagged data from popular gas stations within each zip code to estimate the local penetration of
UberEats and DoorDash. Although restaurants do not have direct access to current or past market
share information at the time of their decision-making, the lagged consumer market share serves
as a proxy for underlying market trends that indirectly influence restaurant decisions. This
approach helps mitigate simultaneity bias, which occurs when explanatory variables are
simultaneously determined with the dependent variable, leading to biased estimates. By using
lagged market share, I ensure that the variable is exogenous from the perspective of the restaurant's
decision process. Additional instrumental variables or fixed effects could be considered to address
any remaining endogeneity.
𝑥∙𝑚𝑡 are vectors of observable platform-specific characteristics in market m at time t, such
as the number of competitors within the same cuisine in the same platform, premium plan ratio,
and average delivery fee of other restaurants on the same platform. 𝑧𝑚𝑡 represents market
attributes that are constant across different platforms, for example, foot traffic, indicators for
whether the market is under a commission cap or a new commission plan, or both. 𝑤𝑗
includes
restaurant-specific observable characteristics, such as meal price, rating value, chain indicator, and
full-service indicator. 𝛽𝑢 and 𝛽𝑑 are coefficients for restaurant-specific observable characteristics,
capturing the effect of each restaurant’s attributes on the decision to list on UberEats or Doordash.
𝛿𝑢 and 𝛿𝑑 capture platform-specific fixed effects or unobservable attributes of platforms
such as the number of couriers or other unobservable factors. 𝜈𝑗𝑢 and 𝜈𝑗𝑑 represents individualspecific random effects, varying across restaurants but constant over time and specific to each
platform. These effects capture unobserved heterogeneity specific to each restaurant, such as
individual brand loyalty, alignment with platform-specific customer bases, or differing marketing

26
support.7
These random effects are assumed to be normally distributed with mean zero and
variances. This assumption of normality allows for tractable integration over the random effects in
the likelihood estimation. 𝜖𝑗𝑢𝑚𝑡 and 𝜖𝑗𝑑𝑚𝑡 are time-varying shocks for each alternative, assumed
to be i.i.d. type-I extreme value.
The utility for listing on both(b) platforms includes the combined utilities from each
individual platform plus a value-added term:
Where Γ represents the additional utility or disutility derived from listing on both platforms,
reflecting potential synergies or competitive disadvantages.
When businesses opt for multihoming, they may experience several benefits. Listing on an
additional platform can enhance overall demand by attracting more customers and orders,
particularly from those who prefer delivery services. This strategy is especially advantageous for
small businesses, as it allows them to expand their reach beyond their physical capacity.
Additionally, multihoming enhances brand visibility and recognition, contributing to improved
brand awareness. The distinct services offered by each platform can complement each other,
further increasing the overall utility for the restaurant.
7
For example, brand loyalty refers to a restaurant's overall preference for a particular platform based on past experiences or
perceived brand value. Platform-specific customer bases denote alignment with the customer demographics and behavior specific
to each platform. Marketing support reflects the differing levels of marketing and promotional support provided by each platform.
Operational compatibility pertains to the ease of integrating with the platform's operational systems and processes.

27
However, there are significant costs and negative aspects associated with listing on
multiple platforms that may influence some businesses to avoid multihoming. First, some
businesses believe that one dominant platform is sufficient to attract the necessary customer
attention and demand, thus listing on additional platforms may be seen as redundant. Additionally,
the costs associated with commission fees from multiple platforms can be substantial, reducing
overall profitability. There is also a preference for in-person customers, who typically offer higher
profit margins per order compared to online orders.
Furthermore, if the services offered by the platforms are similar, businesses might perceive
no added value from multihoming, seeing it as an unnecessary expense. Existing exclusive
contracts with franchises can restrict the ability to list on multiple platforms, imposing a
contractual cost. Businesses located in popular commercial neighborhoods might already have
maximum customer proximity, making additional platforms less beneficial. Operational concerns
also play a significant role; managing a higher volume of orders can lead to logistical challenges,
potentially harming the business’s reputation if service quality declines. These factors collectively
represent the costs and negative aspects that deter businesses from listing on multiple platforms.
There is no outside option in this model, meaning that restaurants must choose between
UberEats, DoorDash, or both, without the possibility of opting out entirely. This modeling
framework allows for a comprehensive analysis of the factors influencing platform choice
decisions among restaurants and the impact of various market dynamics on these choices.

28
1.6 Estimation
The parameters of the multinomial logit model are estimated using maximum likelihood
estimation (MLE). The likelihood function for restaurant j over all time periods t is given by:
where θ denotes the vector of all model parameters. Here, 𝑦𝑗𝑡 represents the choice outcome for
the restaurant j at time t, indicating the platform alternative it selects. The log-likelihood function
for the sample is:
This formulation integrates over the individual-specific random effects and the
idiosyncratic error terms, assuming a multivariate normal distribution for the random effects and
an i.i.d. type-I extreme value distribution for the error terms. The optimization is performed using
gradient-based methods, ensuring convergence to the maximum likelihood estimates.
The detailed implementation of the model estimation, including the numerical integration
and optimization procedures, is crucial for accurately capturing the restaurant’s platform choice
behavior and understanding the impact of various factors on their decision-making process. To
refine the estimates for the variances, point estimates obtained from the initial optimization are
used as starting guesses. Numerical integration is then conducted to achieve the full set of
parameter estimates.

29
1.7 Estimation Results
1.7.1 Demand Parameters
Table 1. 3 Restaurant Platform Choice Model Parameter Estimates
Variable Estimate Std. Err
Platform-specific Characteristics
Commission Rate (𝛼1
) -0.171 0.011
Lagged Consumer Market Share (𝛼2
) 0.037 0.006
Total Number of Competitors per Cuisine (𝛼3
) -0.367 0.008
Average Premium Plan Ratio (𝛼3
) -3.824 0.022
Average Delivery Fee (𝛼3
) -1.269 0.010
Market-specific Characteristics
Foot Traffic (𝛼4
) 0.124 0.004
New Commission Plan Indicator (𝛼4
) 4.064 0.038
Fee Cap City Indicator (𝛼4
) 0.306 0.026
New Commission Plan Indicator x Fee Cap City Indicator (𝛼4
) -0.236 0.029
Platform-specific Fixed Effects
𝛿𝑢 -7.948 0.045
𝛿𝑑 -2.093 0.029
Γ 3.452 0.024
Restaurant-specific Characteristics
Meal Price (𝛽𝑢) -0.701 0.010
Meal Price (𝛽𝑑) 0.663 0.008
Rating (𝛽𝑢) 1.108 0.022
Rating (𝛽𝑑) 0.821 0.026
Brand indicator (𝛽𝑢) 0.079 0.022
Brand indicator (𝛽𝑑) -0.301 0.019
Full-Service Indicator (𝛽𝑢) -0.032 0.020
Full-Service Indicator (𝛽𝑑) -0.162 0.015
Note: The table reports the estimates and standard errors for the multinomial logit regression.
The table above displays the maximum likelihood estimates of the coefficients on
observable characteristics for the multinomial logit model. Most coefficients in the utility of listing
on platforms are significant, except for the full-service restaurants on UberEats. The results align
closely with expectations, with Γ found to be positive, indicating a synergy effect when restaurants
list on both platforms.

30
Despite the limited variation in commission rates, the coefficient on the commission rate
(𝛼1
) in the regression model was estimated to be negative and statistically significant, which aligns
with theoretical expectations. This result can be justified by the economic intuition that higher
commission rates reduce the net revenue for restaurants, thereby decreasing their likelihood of
choosing a platform with higher commissions. Even with standardized rates, there is still enough
variation to capture the relationship between commission rates and restaurant choices.
The lagged market share coefficient (𝛼2
) is positive and significant, suggesting that
platforms with higher consumer market shares within the zip code are more attractive to restaurants.
This reflects the importance of network externalities in platform selection, where the presence of
more consumers on a platform makes it more appealing.
The negative coefficient on the number of competitors indicates that a higher number of
competitors within the same cuisine on a platform decreases the utility of listing on that platform.
This highlights the competitive pressures within localized markets. The negative coefficient for
the premium plan share suggests that higher premium shares decrease the utility of listing on a
platform. Again, the premium plan share indicates the fraction of restaurants in the market with a
premium plan subscription; a higher premium plan share means a larger fraction of restaurants in
the market have opted for the premium plan. Restaurants fear being overlooked by consumers due
to the presence of numerous highly promoted competing restaurants on delivery apps. This
dynamic can influence restaurant decisions, as a higher concentration of premium plan restaurants
may intensify competition and reduce the perceived benefits of listing on that platform. A higher
average delivery fee within the zip code is associated with lower utility, indicating that higher
delivery fees deter restaurants from listing on that platform.

31
The coefficient on the foot traffic indicator is positive and significant, indicating that higher
foot traffic within a zip code is associated with a higher likelihood of restaurants listing on
platforms. This suggests that a larger potential customer base and higher economic activity
encourages restaurant’s platform participation.
The positive and significant coefficient for the new commission plan indicates that the new
commission plans positively influence platform listing decisions. In cities where the commission
fee was capped by regulators, restaurants were more likely to list on a platform with the
commission cap, fully utilizing the benefits of the commission cap policy.
The negative coefficient on the interaction term indicates that the introduction of new
commission plans in cities with fee caps before, makes restaurants less likely to list on the
platforms compared to other restaurants. This can be interpreted as a difference-in-differences
effect, showing that the combined effect of fee caps and new commission plans reduces platform
participation.
The positive Γ indicates significant value-added effect when restaurants list on both
platforms. This effect means that restaurants derive additional benefits from multihoming, such as
increased visibility, expanded customer base, and higher overall demand, compared to listing on a
single platform.
The coefficients 𝛿𝑢 and 𝛿𝑑 represent the fixed effects for UberEats and DoorDash,
respectively. The negative value of 𝛿𝑢 and 𝛿𝑑 indicates the presence of unobserved factors that
decrease the utility of listing on these platforms. These factors could include platform-specific
characteristics or competitive dynamics not captured by the observable variables.

32
The coefficient on the meal price in UberEats is negative and less than the coefficient for
DoorDash, indicating that restaurants with lower meal prices are inclined to list on UberEats
instead of DoorDash.
The coefficient on ratings in UberEats is positive and significant, indicating that restaurants
with higher ratings are more inclined to list on UberEats. The coefficient for DoorDash is also
positive, but smaller, suggesting a weaker preference for higher-rated restaurants on DoorDash.
The positive coefficient for the chain indicator on UberEats suggests that chain restaurants
are more likely to list on UberEats. The negative coefficient for the full-service indicator indicates
that full-service restaurants are less inclined to list on UberEats platform.
Overall, these findings illustrate the complex interplay of commission rates, market
conditions, and restaurant-specific factors in determining platform participation. The positive Γ
underscores the benefits of multihoming, suggesting that restaurants listed on multiple platforms
derive greater utility compared to single-platform listing. This comprehensive analysis highlights
the nuanced preferences and strategic decisions made by restaurants in choosing between UberEats
and DoorDash, considering both the benefits and costs associated with each platform.
1.7.2 Substitution Patterns
To further explore the estimation results, I compute the substitution patterns by examining
the cross-price and cross-market share derivatives of the listing demand. These derivatives
measure the impact of changes in the commission rate and market share of one platform on the
probability of choosing another platform.
1.7.2.1 Cross-Price Derivatives of Listing Demand
The partial derivatives of the probabilities are calculated as follows:

33
Table 1. 4 Cross-Price Derivatives
Platform Choice Change in Commission Rate Cross-Price Derivative
UberEats only DoorDash -0.00163
DoorDash only UberEats -0.00345
Both DoorDash -0.00117
Both UberEats -0.003
These derivatives show the change in the probability of choosing each platform per 1%
change in the commission rate of listing on the other platform. Given that the platforms are
complements to restaurants, the cross-price derivatives are negative. This indicates that an increase
in the commission rate on one platform decreases the probability of listing on the other platform,
suggesting that restaurants view the platforms as complements. If the cost of listing on DoorDash
increases, they are less likely to choose UberEats as an alternative.
These substitution patterns highlight the complementary nature of UberEats and DoorDash
for restaurants. Changes in the pricing strategy of one platform significantly impact the listing
decisions on the other, emphasizing the interconnected dynamics in the platform market.
Also, the smaller magnitude of the cross-price derivative for UberEats means that the
impact of changes in DoorDash's commission rate on the probability of choosing UberEats is less
than the impact of changes in UberEats' commission rate on the probability of choosing DoorDash.
This suggests that restaurant market share for DoorDash is more sensitive to the change in
UberEats commission rate changes than the UberEats market share to the change of DoorDash
commission rate.
From a platform perspective, when DoorDash changes its commission setting, UberEats
won't be affected significantly. However, DoorDash will be influenced considerably more by
changes in UberEats' price setting. This dynamic indicates that DoorDash's market share is more

34
reactive to competitive pricing strategies implemented by UberEats, highlighting the need for
DoorDash to closely monitor and respond to UberEats' pricing policies to maintain its competitive
position.
1.7.2.2 Cross-Market Share Derivatives of Listing Demand
I also explore the cross-market share derivatives of listing demand, which measure the
impact of changes in the user market share of one platform on the probability that the restaurant
chooses the other platform. Specifically, I examine how a 1% increase in the market share on
DoorDash affects the probability of choosing UberEats and vice versa.
Table 1. 5 Cross-Market Share Derivatives
Platform Choice Change in Consumer Market Share Cross-Market Share Derivative
UberEats DoorDash -0.00311
DoorDash UberEats -0.00659
Both DoorDash -0.00224
Both UberEats -0.00572
These derivatives show the restaurant platform market share change per 1% change in the
consumer market share of the other platform. The cross-market share derivatives are also negative,
indicating that an increase in the consumer market share on one platform decreases the probability
of a restaurant listing on the other platform. As one platform becomes more popular among users
in a market, restaurants in the same market are less likely to choose the other platform as an
alternative.
The larger magnitude of the cross-market share derivative for DoorDash indicates that
DoorDash restaurants are more sensitive to the consumer market share of UberEats compared to
UberEats' market share change in response to the consumer market share of DoorDash. This
suggests that restaurants listing on DoorDash are more likely to react to changes in UberEats'

35
consumer market share, indicating a higher sensitivity and competitive pressure faced by
DoorDash in response to UberEats' market performance.
From a strategic perspective, this implies that DoorDash must be more vigilant and
responsive to shifts in UberEats' consumer base to maintain its restaurant listings and overall
market share.
The substitution patterns highlight the complementary nature of UberEats and DoorDash
for restaurants. Changes in the pricing setting strategy or market share of one platform significantly
impact the listing decisions on the other, emphasizing the competition and interconnected
dynamics in the platform market.
1.8 Counterfactual Analysis
The counterfactual analysis explores the impact of removing the multihoming option for
restaurants, considering the substantial negative utilities imposed by the multihoming alternative.
Figure 1. 7 Average Monthly Restaurant Listings: Original vs Counterfactual Choices
This plot compares the average monthly restaurant listings for DoorDash, UberEats, and both platforms (multihoming)
under original conditions and a counterfactual scenario where multihoming is removed. The left panel shows the
original choices and the right panel displays the counterfactual choices, illustrating a significant shift. This shift
underscores DoorDash's competitive advantage and the impact of removing the multihoming option on market
dynamics. The data highlights the importance of multihoming in maintaining a balanced competitive landscape
between food delivery platforms.

36
The analysis involves setting strong disincentives for choosing both platforms and
examining how restaurant choices would shift if they could only choose either DoorDash or
UberEats. When the multihoming option is removed, the proportion of restaurants choosing both
platforms drops to zero. A substantial increase is observed in the proportion of restaurants choosing
DoorDash, while the proportion choosing UberEats also increases but to a lesser extent.
Table 1. 6 Change in Market Shares by Restaurant Choice When Multihoming is removed.
Month UberEats (%) DoorDash(%) Multihoming (%)
1 7.4 17.6 -25
2 7.1 18.3 -25.5
3 9 17.1 -26.1
9 0 37 -37
10 -0.2 36 -35.8
11 1 36.5 -37.5
Note: This table presents the results of a counterfactual analysis on the restaurant platform choice model, showing the change in
market shares when multihoming is removed or unavailable. The percentages reflect the shifts in market share for UberEats,
DoorDash, and both platforms across various months.
The removal of the multihoming option forces restaurants to choose a single platform. The
data indicates a strong preference for DoorDash when multihoming is not an option. This shift can
be attributed to DoorDash's better-perceived benefits compared to UberEats. The market share for
DoorDash increases significantly (by about 17.1% to 37%), suggesting that many restaurants that
were previously multihoming prefer DoorDash when forced to choose only one platform. This
highlights DoorDash's competitive edge and attractiveness to restaurants.
The market share for UberEats also increases, but to a lesser extent, indicating that while
some restaurants still prefer UberEats, the dominant preference shifts towards DoorDash in the
absence of the multihoming option. This analysis underscores the competitive dynamics between
the two platforms and the significant role that the ability to multihome plays in restaurant platform
choices.

37
The option to use multiple platforms maintains a balanced competitive landscape.
Restricting this option can lead to market dominance by one platform, thereby reducing
competition in the food delivery market. Such an imbalance suggests that interventions restricting
multihoming behavior may result in anti-competitive outcomes, favoring the dominant platform
and limiting consumer choices. In contrast, policies that promote multihoming could foster a more
competitive and equitable marketplace, benefiting both restaurants and consumers by preventing
monopolistic control by any one platform.
1.9 Conclusion
This chapter investigates the decision-making processes of restaurants in the San Francisco
Bay Area when selecting between two major food delivery platforms, UberEats and DoorDash,
employing a multinomial logit model with individual-specific random effects. The analysis focuses
on the impact of commission rates and other factors on restaurant decisions and the hidden benefits
associated with restaurant multihoming.
Key findings reveal that higher commission rates significantly reduce the likelihood of
restaurants listing on a platform, while platforms with higher user market shares attract more
restaurants, illustrating the importance of network externalities. The presence of more competitors
and higher premium plan ratios on a platform diminishes its attractiveness to restaurants. The study
also reveals substantial benefits of multihoming for restaurants, evidenced by the positive Γ in the
model. These benefits include enhanced demand, increased brand visibility, and complementary
services from both platforms.

38
A counterfactual analysis, where multihoming is removed from the choice set, underscores
the competitive edge of DoorDash. Without the ability to list on both platforms, a significant
portion of restaurants would prefer DoorDash, resulting in a substantial increase in its market share.
Allowing restaurants to use multiple platforms promotes a balanced and competitive market,
whereas restricting this option can lead to monopolistic control and reduced efficiency. This
preference highlights the strategic importance of allowing multihoming for restaurants.
These insights can help platforms like UberEats and DoorDash refine their commission
structures to better influence platform competition and user behaviors. Understanding the
synergies associated with multihoming can help platforms design strategies that encourage dual
listings, thereby enhancing their market position. Platforms should also consider differentiation
strategies to make themselves more attractive to both users and sellers, and to enhance the
complementarity of each platform to users. It is essential to promote restaurants more effectively
than multihoming counterparts and retain consumer market share as a key factor. This could
include unique features, better user experiences, or targeted marketing campaigns to highlight the
distinctive benefits of each platform. This strategic focus will help platforms navigate the
competitive landscape more successfully.
Moreover, platforms can estimate the sensitivity of their restaurant market share to changes
in a competitor's commission rate or consumer market share. By understanding these sensitivities,
platforms can anticipate the impact of their competitor's pricing strategies and set their strategies
accordingly to maintain or enhance their market share.
Policymakers can utilize these findings to assess the impact of regulatory interventions,
such as commission caps, on platform competition and restaurant profitability. The evidence
suggests that while commission caps can alleviate financial burdens on restaurants, the

39
introduction of new commission structures needs careful consideration to avoid unintended
negative impacts. Policymakers should also consider whether platforms discourage or prevent
multihoming and if commission plans or other policies harm multihoming and competition.
Future research could expand platform choices to include additional platforms like
Grubhub and Postmates, yielding more comprehensive results. Modeling the choice of service
plans (basic, advanced, premium) selected by restaurants could provide deeper insights into
strategic decisions based on the level of service offered by each platform. Investigating the
implications of the recent removal of commission caps in San Francisco could provide interesting
follow-up opportunities. Observing how these regulatory changes impact the platform adoption
behaviors of restaurants will be beneficial.
The concept of competing platforms acting as complements to seller groups can be
extended to other industries, such as small businesses in retail marketplaces, riders in ride-sharing,
and local guides on online travel agencies. Businesses in these sectors can leverage dual platform
strategies to maximize their reach and profitability.
As the platform industry evolves, it is crucial for platforms to stay agile and responsive to
emerging trends like changing consumer preferences post-pandemic and lifting regulatory
interventions. Adapting commission structures and platform features in response to these trends
will be vital for maintaining competitive advantage and ensuring long-term growth. By applying
these results, stakeholders in the platform industry can better navigate the complexities of platform
competition, optimize their strategic decisions, and enhance the overall ecosystem for both sellers
and consumers.

40
Chapter 2
Consumer Choice in the Food Delivery Industry
2.1 Introduction
The food delivery industry has experienced substantial growth in recent years, with
numerous players vying for market share. Despite this expansion, profitability remains a
significant challenge. The evolving business model is influenced by various factors, including fees,
competition, and consumer preferences.
This study investigates the factors that influence consumers' choices of restaurants and
platforms within the food delivery market, with a particular focus on whether consumers value
multihoming restaurants—where restaurants are available on multiple platforms. By applying a
multinomial logit model based on random utility theory, the research examines how price, quality,
platform characteristics, and the availability of multihoming options shape consumer decisions.
Additionally, a counterfactual analysis is employed to explore hypothetical scenarios, predicting
how changes in these factors might alter consumer behavior.
The dissertation is organized as follows: Section 2.2 reviews the literature on consumer
platform choice, focusing on factors influencing consumer choices, including multihoming and
platform competition. Section 2.3 provides an overview of the food delivery industry's structure.
Section 2.4 describes the dataset, including transaction data and key statistics on consumer

41
behavior. Section 2.5 outlines the multinomial logit model for analyzing consumer choice behavior.
Section 2.6 discusses the estimation methods, focusing on maximum likelihood estimation and
accounting for unobserved heterogeneity. Section 2.7 presents the estimation results, highlighting
key determinants of consumer choices. Section 2.8 conducts a counterfactual analysis to assess the
impact of hypothetical scenarios on consumer choices. Section 2.9 links restaurant platform
choices to consumer behavior, examining how changes in platform selection influence consumer
decisions in the food delivery market through a counterfactual analysis. Section 2.10 concludes
with a summary of key findings, implications of the research.
2.2 Literature Review
The literature review for the consumer platform choice in the US delivery industry
leverages much of the same foundational research used to understand restaurant platform adoption
problem. However, the perspective and application of the findings differ significantly. While
Chapter 1 focused on the decisions made by restaurants to adopt and engage with delivery
platforms, Chapter 2 shifts the focus to consumers' choices among these platforms, a more
prevalent topic in the literature, making it highly relevant and applicable.
In contrast to the previous chapter, where restaurants could choose between two platforms
or opt for both (multihoming), this chapter examines consumer choices where they select only one
alternative from the available options, including the outside option. Furthermore, consumers'
decisions are influenced not only by the platform but also by restaurant-specific characteristics,
adding another layer of complexity to the model.
Rochet and Tirole (2003) provide a foundational understanding of network externalities in
two-sided markets, explaining how the benefit for one side depends on the size of the other side

42
of the platform. This relationship is integral to understanding consumer platform choices as it
underscores the interdependence between platform user bases. The platform structure in the food
delivery industry also differs from traditional models.
Armstrong (2006) and Armstrong and Wright (2007) further elaborate on platform
competition and multihoming. Unlike Armstrong's model, which typically considers one side
joining a single platform while the other joins multiple platforms (creating a competitive
bottleneck), this model allows both restaurants and consumers to multihome. This distinction is
crucial as it negates any monopoly power platforms might have by providing access to both sides,
thereby enabling platforms to charge both consumer fees and commissions to both sides, diverging
from the competitive bottleneck model.
It is worthwhile to revisit the paper by Teh et al. (2023). This paper analyzes platform
competition and equilibrium fees among symmetric two-sided platforms. Similar to our model,
both groups can multihome, and platforms charge per-transaction fees to both sides. In contrast to
our model, the paper assumes that platforms are identical from the sellers' perspective since they
are symmetric and charge the same fees. With a cost for buyers to multihome or switch the platform,
the fraction of multihoming buyers decreases, leading to a decrease in buyer fees but an increase
in seller fees. This insight can also be applied to our model. The logical relationships between fee
structure and group characteristics from this paper are particularly relevant and can provide
valuable context for interpreting our empirical results.
Sullivan (2024) provides a benchmark model for understanding food delivery platform
dynamics. In Sullivan's sequential game model, platforms set commission rates first, followed by
restaurants choosing platforms and setting menu prices, and finally, platforms setting consumer
fees, the total fee that platform charge consumers including service and delivery fee. Consumers

43
then make an order from a pair of a restaurant and a platform. In contrast, my model posits that
platforms set consumer fees and commission schedule first. When restaurants choose platforms,
they choose one of the possible commission rates from the platform’s schedule. Consequently,
restaurants set menu prices, and consumers make their choices. The main difference between the
two models is that in Sullivan's model, platforms set the commission rate, whereas in my model,
restaurants choose the commission rate. This change is reflective of new commission plan launches.
From a data perspective, my dataset includes distinct consumer fees by individual
restaurants, offering a wide variety observed in consumer fees, unlike Sullivan's dataset. This
contrast provides an opportunity to compare the consumer choice problems influenced by the
variability of cost factors.
By integrating these perspectives, this chapter aims to provide a comprehensive
understanding of consumer behavior in the online food delivery market, emphasizing how price
sensitivity, restaurant quality, and platform characteristics drive consumer choices. This nuanced
approach ensures a robust analysis of consumer platform choice dynamics, setting the stage for
exploring various counterfactual scenarios and their implications.
2.3 Industry Background
2.3.1 Timing and Structure of the Game
The structure of the food delivery industry can be understood through a sequential game
framework. Initially, platforms set the commission schedule, determining the commission rates
and consumer fees. Following this, restaurants choose a platform and a specific commission rate,

44
then they set menu prices on each app. Finally, consumers choose a platform and a restaurant based
on various factors, including their subscription status.
As discussed in Chapter 1, delivery platforms implement differentiated commission plans
based on the level of delivery and marketing services provided. A premium plan offers additional
marketing services, priority positioning, dedicated customer service support, and incentives for
subscribers, such as discounted service fees and zero delivery fees. Restaurants can choose from
three types of plans: premium, plus, and basic, each with different commission rates and
corresponding customer fees for subscribers and non-subscribers, as shown in Table 2.1. Higher
commission rates result in lower consumer fees charged by that platform.
Table 2. 1 DoorDash Customer Fees Based on Restaurant Commission Plan and Customer Subscription Status
Restaurant Chooses... Subscriber Non-Subscriber
Delivery Fee Service Fee Delivery Fee Service Fee
Premium (30%) 0 5% 𝑑𝑓𝑎 15%
Plus (25%) 0 5% 𝑑𝑓𝑏 15%
Basic (15%) 𝑑𝑓𝑐 15% 𝑑𝑓𝑐 15%
Note: Subscribers are paying $9.99 per month as a subscription fee and delivery fees for non-subscribers are structured
such that 0 < 𝒅𝒇𝒂 < 𝒅𝒇𝒃 < 𝒅𝒇𝒄< $4.99
This chapter elaborates on the last stage of this game, focusing on the consumer's restaurant
and platform choice problem. It should be noted that the model does not include the customer's
subscription choice problem. I was able to obtain the individual's subscription status from the data;
however, there was not enough comprehensive data on individual characteristics to construct a
subscription choice model.
2.3.2 Breakdown of Consumer Payments
When consumers make a transaction, the total charged amount is broken down into several
components. Beyond the subtotal of the order, users are subject to service fees, delivery fees, and
other charges, which are subsequently conveyed to the delivery platforms. Discounts may apply,

45
benefiting users, while taxes and tips are also included in the final amount. The total charged
amount can be represented as:
Subtotal + Service Fee + Delivery Fee + Other Fee + Taxes – Discount + Tips
Platforms retain the delivery, service, and other fees, while tips are allocated to couriers.
Service Fees Proportional to the subtotal. On DoorDash, service fees are 15% for non-members
and 5% for members. UberEats sets service fees primarily at 15% of the subtotal, with a maximum
of $5.
Delivery Fees Flat, constant amounts, regardless of the order subtotal, and may occasionally be
$0 due to membership benefits.
Other Fees Temporary fees, including small order fees ($3), CA driver benefits ($2), regulatory
compliance fees ($1), and temporary fuel surcharges ($0.45), typically flat rates based on
geolocation and order date.
Taxes Proportional to the subtotal and based on geolocation.
Discounts Amount subtracted from the total charged amount. DoorDash offers $5 off the first
delivery order of $15 or more, and $10 off orders of $20 or more. UberEats provides 5-10% off
orders over $15 for members.
In the empirical model of this chapter, taxes and tips are neglected since they affect every
restaurant the same way in each consumer's decision problem.
2.3.3 Consumer Subscription Membership
Similar to how platforms offer different membership(commission) plans to restaurants,
they also provide subscription plans to consumers, which significantly influence consumer choices

46
and behavior. Indicators for customer subscription status are included in the model to account for
these effects.
DashPass DoorDash’s subscription offers $0 delivery fees and a 5% service fee on orders
over $12 from eligible restaurants.
Uber One UberEats’ subscription provides $0 delivery fees and 5-10% discounts on orders
over $15 from eligible restaurants.
These subscription models aim to enhance customer loyalty and increase order frequency,
providing consistent revenue streams for the platforms while offering savings and convenience to
customers. In Figure 2.1 (a), over 60% of active DoorDash users were subscribed to the DashPass
membership, whereas approximately 40% of UberEats users had Uber One. The loyalty levels
vary significantly between the two memberships; less than 10% of UberEats subscribers renew
their membership beyond the free trial period, while 30% of DoorDash subscribers renew after
free one month and continue their membership for 2 to 5 months. This indicates higher loyalty and
retention rates among DoorDash subscribers compared to UberEats. Analysis of user orders shows
that subscribers tend to place orders more frequently and with larger quantities compared to nonsubscribers.

47
(a) (b)
Figure 2. 1 Subscriber Behavior across Platforms
(a) The trend of subscriber shares among customers by platform. (b) Reverse CDF of Subscriber Subscription Length:
Reverse Cumulative Distribution Function (CDF) of subscriber subscription length, illustrating the duration of user
loyalty

2.4 Data
This study utilizes granular transaction data purchased from Measurable AI, a company
that provides e-receipt data from major food delivery platforms through its own developed
consumer panel. The dataset covers the period from February to March and September to October
of 2021 and includes detailed information such as consumer IDs, subscription status, addresses,
order dates, restaurant information, subtotals, discounts, delivery fees, service fees, other fees, and
total transaction amounts. This comprehensive dataset allows for a thorough examination of the
factors influencing consumer choices and platform dynamics. The data is structured as panel data,
capturing multiple transactions for each consumer over time, though it is not a balanced panel as
some consumers do not have transactions in every observed month.
The sample dataset was constructed by expanding the transaction data to generate all
possible choice sets for each consumer across time and market. The alternative restaurants and
their corresponding characteristics on both platforms were identified using the listing data
presented in Chapter 1. This approach ensures that the analysis captures the full range of options
available to consumers at each point in time, providing a robust foundation for examining their
decision-making processes. Additionally, outside options were also considered in the analysis.
Some key statistics from the dataset will be presented to provide an overview of the
consumer behavior observed. These statistics will help illustrate the diversity and richness of the
data, highlighting trends and patterns relevant to the study's objectives.

48
Table 2. 2 Summary Statistics of Order Cost, Restaurant Quality and Platform Characteristic
Variable Mean Std. Dev. Min Max Observations
Subtotal ($) 42.858 43.845 9.3 120.75 503,392
Delivery Fee ($) 0.895 1.811 0.00 20.00 503,392
Service Fee ($) 5.931 6.346 0.465 18.11 503,392
Other Fee ($) 0.734 0.918 0 3 503,392
Rating 3.900 1.172 0 5 503,392
New Addition Indicator
(1 = New) 0.302 0.459 0 1 503,392
Premium Plan Indicator
(1 = Yes, 0 = No) 0.215 0.411 0 1 503,392
Brand Indicator
(1 = Yes, 0 = No) 0.197 0.397 0 1 503,392
Full Service Indicator
(1 = Yes, 0 = No) 0.314 0.464 0 1 503,392
Multihoming Indicator 0.399 0.490 0 1 503,392
Total Number of Restaurants
on DoorDash by zip code 113.097 113.905 1 496 326,538
Total Number of Restaurants
on UberEats by zip code 63.872 56.879 1 342 176,854
Cuisine Variety
On DoorDash by zip code 23.354 3.163 1 27 326,538
Cuisine Variety
On UberEats by zip code 20.347 4.189 1 29 176,854
Average Delivery Fee
On DoorDash by zip code 1.693 0.830 0.09 4.542 326,538
Average Delivery Fee
On UberEats by zip code 2.084 1.154 0.04 4.651 176,854
Subscription Status
(1 = Yes, 0 = No) 0.673 0.419 0 1 503,392
Regarding the cost of the order, the variables include the subtotal, delivery fee, service fee,
and other fees, all expressed in dollar amounts. The average subtotal is $42.86, the delivery fee is
$0.90, the service fee is $5.93, and other fees average $0.73. Both delivery and service fees are
influenced by the subscription status of the customer.
In terms of restaurant qualities, the consumer ratings on each platform range from 1 to 5. The new
addition indicator shows whether the restaurant was listed on the specific platform for the first
time that month. The premium plan indicator denotes whether a restaurant was on a premium plan
that month on the specific platform. The brand indicator reveals whether the restaurants are part

49
of a chain, while the full-service indicator indicates whether the restaurants offer sit-down dining
service. The multihoming indicator shows whether the restaurants were listed on both platforms
during the month.
Platform characteristics include the total number of restaurants by zip code on each
platform and the variety of cuisines available. DoorDash has an average of 113 restaurants per zip
code, compared to UberEats' 64, indicating a larger restaurant pool on DoorDash. Cuisine variety
is similar between the two platforms, with DoorDash offering an average of 23 different cuisines
per zip code and UberEats offering 20. The cuisine categories on the platforms include options
such as Mexican, Asian, American, Indian, Italian, Japanese, Chinese, Sandwiches, Burgers, Pizza,
Dessert, Coffee, Chicken, Seafood, Healthy, Salads, and more.
Finally, customer characteristics are captured by the subscription status, meaning whether
a customer being subscribed to the specific platform where the restaurant was listed that month.
2.5 Model
This study employs a multinomial logit model to analyze consumer choice behavior in
selecting restaurants through online delivery platforms. The model is grounded in random utility
theory, positing that consumers choose the alternative that provides the highest utility among a set
of available options, including the option to choose nothing.
The empirical model is designed to estimate a customer's restaurant-platform decision
problem by understanding how customers choose between restaurants and platforms based on
various observable factors. Specifically, the decision-making process of customer i at time t in
market m involves selecting a restaurant j and platform f from a set of alternatives, or opting for
an outside option where no restaurant is chosen. This approach rigorously captures the complexity

50
of consumer decision-making in the context of online food delivery services, considering the
interplay of multiple factors influencing their choices.
The choice set for each customer is assumed to be confined to the restaurants available
within their specific postal area, zip code, reflecting the geographical proximity that plays a crucial
role in decision-making. The alternatives that customers face are combinations of restaurants and
platforms, denoted as ( j, f ). If a restaurant j is available on both platforms (e.g., UberEats and
DoorDash), the customer decides which platform to use for their order. If the restaurant is only on
one platform, the choice is straightforward.
The utility that customer i derives from choosing restaurant j on platform f at time t is given
by:
Where 𝑝𝑗𝑓𝑚𝑡 is the meal price at restaurant j on platform f and 𝑐𝑗𝑓𝑚𝑡 is the fee that the customer
pays, excluding tax and tip. Therefore, 𝑝𝑗𝑓𝑚𝑡 + 𝑐𝑗𝑓𝑚𝑡 (subtotal and fee) represents the aggregate
cost of the order, including any delivery fees associated with restaurant.
𝑥𝑗𝑓𝑚𝑡 encompasses quality aspects such as ratings, which reflect consumer group ratings
of the restaurant, indicating quality and consumer satisfaction. An indicator whether the restaurant
is on a premium plan of the platform captures whether the restaurant has joined a membership
program provided by the platforms. The multihoming indicator denotes whether the restaurant is
available on multiple platforms, suggesting its accessibility and reach. A new addition indicator
reveals whether the restaurant was recently added to the platform. The brand indicator identifies if
the restaurant is part of chain, and the full-service indicator shows whether the restaurant provides
full-service dining options or fast food.

51
The term 𝜓𝑖𝑓𝑚𝑡 captures the consumer’s taste for platform f, influenced by platform
characteristics (𝑑𝑓𝑚𝑡) and subscription status (𝑑𝑖𝑓𝑡):
Specifically, 𝑑𝑓𝑚𝑡 includes market-level variables such as the number of restaurants on
platform f in market m at time t, average delivery fee on platform f in market m at time t, and a
measure of cuisine variety on platform f in market m at time t.
The variable 𝑑𝑖𝑓𝑡 captures consumer-specific characteristics, such as whether the
consumer is subscribed to the subscription plan for the platform f that the restaurant is at time t.
Additionally, the random effect 𝜈𝑖𝑗 represents the consumer’s idiosyncratic taste for restaurant j,
capturing correlation over time. 𝜖𝑖𝑗𝑓𝑚𝑡 is an i.i.d. type 1 extreme value error term.
Time and regional variations are accounted for through dummy variables derived from the
transaction month and zip code, allowing for the assessment of temporal trends and geographic
differences in consumer preferences.
Including the outside option in the choice set means that the model also accounts for the
possibility that a customer might choose not to order from any of the available restaurants at a
given time. This enhances the model’s realism by recognizing that customers can decide not to
engage with the platform for various reasons, thereby providing a more comprehensive
understanding of consumer behavior in the online food delivery market.
This empirical model captures the various factors influencing a customer's restaurantplatform choice, considering both observable attributes like prices and quality indicators, and
important platform-specific characteristics. By focusing on these elements, the model reflects a
detailed understanding of the consumer decision-making process in the context of food delivery

52
services, though it abstracts from unobservable market-level and idiosyncratic tastes for
simplification.
2.6 Estimation
The parameters of the model are estimated using maximum likelihood estimation (MLE),
ensuring that the model captures the most probable outcomes given the observed choice data. The
estimation process involves iteratively adjusting the parameters to maximize the likelihood of the
observed choices across the dataset.
To include the outside option, where a customer decides not to order from any of the
available restaurants, an additional row is added for each observation in the dataset. This row is
designed to reflect the nature of the outside option, with specific values set accordingly. Costrelated variables, restaurant quality-related variables, and platform-related variables are all set to
zero, as these attributes do not apply to the outside option. The time and regional dummy variables
remain the same as in the corresponding observation to capture the context in which the decision
is made.
The estimation procedure follows the methods outlined in Chapter 1, adapting them to the
current context. After the initial estimation of the model parameters, numerical integration is
performed to obtain estimates for the idiosyncratic taste parameter. This step refines the model by
accounting for individual-specific unobserved heterogeneity in consumer preferences, thereby
enhancing the accuracy and reliability of the estimated parameters.

53
2.7 Estimation Results
The multinomial logit model estimation results provide insights into the factors influencing
consumer choices among various restaurant-platform combinations. The table below summarizes
the parameter estimates, standard errors for the model. Each coefficient represents the effect of the
corresponding variable on the utility derived from a particular restaurant-platform choice.
Table 2. 3 Consumer Restaurant-Platform Choice Model Parameter Estimates
Variable Coefficient Std Err
Order Cost
Subtotal and fee (𝛼) -0.6237 0.032
Restaurant Quality
Rating (𝛽) 0.3894 0.014
New Add(𝛽) -1.1100 0.060
Premium Plan (𝛽) 0.3459 0.063
Brand (𝛽) 0.1372 0.040
Full-Service (𝛽) -0.1558 0.039
Multihoming (𝛽) 0.4521 0.036
Platform Characteristics
N. Restaurants on DoorDash (𝜆
𝑓
) 0.6476 0.094
N. Restaurants on UberEats (𝜆
𝑓
) 0.2082 0.090
Cuisine Variety on DoorDash (𝜆
𝑓
) 0.0956 0.047
Cuisine Variety on UberEats (𝜆
𝑓
) 0.2321 0.057
Average Delivery Fee on DoorDash (𝜆
𝑓
) -0.1058 0.032
Average Delivery Fee on UberEats (𝜆
𝑓
) -0.1286 0.039
Consumer Characteristics
Consumer Subscription (𝜆
𝑖
) 0.3579 0.043
The coefficient for the order cost, 𝛼, is negative and statistically significant. This negative
coefficient indicates that higher total costs, including meal price and delivery/service fees, reduce
the likelihood of a restaurant being chosen, demonstrating that consumers are sensitive to the
overall cost when make their selection.
Regarding quality-related variables, the coefficient for restaurant rating is positive and
statistically significant. This suggests that higher consumer ratings, indicating better perceived

54
quality and customer satisfaction among past consumers, increase the likelihood of a restaurant
being chosen. For newly added restaurants, the coefficient for the new addition indicator is
negative and statistically significant, implying that newly added restaurants are less likely to be
chosen, possibly due to a lack of established reputation or customer reviews. The coefficient for
premium plan membership is positive and statistically significant, indicating that restaurants on
premium plans, likely offering better services or promotions, are more likely to be chosen by
consumers. The positive coefficient for the brand indicator suggests that chain restaurants or wellknown brands are preferred over independent establishments. The negative coefficient for the fullservice indicator indicates a preference for fast food or quick service restaurants over full-service
dining options. The positive coefficient for multihoming indicates that restaurants available on
multiple platforms are more likely to be chosen, suggesting that greater accessibility increases
consumer preference.
In terms of platform-related variables, the positive coefficients for the total number
of available restaurants on each platform imply that an increase in the number of restaurant listings
on each platform within a zip code increases the likelihood of choosing a restaurant on that
platform, possibly due to larger choice options and a higher market share. The total number of
restaurant listings on each platform within a zip code can serve as a proxy for the restaurant market
share on the platform. This proxy reflects the platform's attractiveness and competitiveness from
the perspective of restaurants in the local market. A higher number of listings suggests that more
restaurants prefer to join the platform, potentially indicating a larger market share and better
platform performance in attracting restaurant partners. The coefficients for cuisine variety on each
platform within a zip code are also positive, indicating that a greater variety of cuisines available
on platforms marginally increases the likelihood of choosing that platform. This suggests that a

55
diverse cuisine offering, as well as a large number of listings on a platform, significantly increases
its attractiveness to consumers. For the average delivery fee within a zip code, the negative
coefficient indicates that higher average delivery fees, which contribute to consumers perceiving
the platform as expensive, reduce the likelihood of choosing that platform.
Finally, the coefficient for the consumer’s subscription indicator is positive,
suggesting that consumers who have a subscription to the platform are more likely to choose it,
highlighting the effectiveness of subscription plans in retaining customer loyalty.
In conclusion, the estimation results reveal that cost, restaurant quality, and
availability on both platforms are significant determinants of consumer choices in online food
delivery platforms. Consumers prefer options with lower order costs, higher ratings, established
brands, and multihoming restaurants. Moreover, the extensive and varied offerings within a
platform positively influence consumer preference, while higher delivery fees deter platform
selection. Subscription plans also play a crucial role in enhancing consumer loyalty. These insights
can guide platform strategies to better align with consumer preferences and improve platform
attractiveness.
2.8 Counterfactual Analysis
This counterfactual analysis assesses how changes in consumer choices might occur under
hypothetical scenarios by altering key model parameters. Specifically, the scenario involves setting
the multihoming coefficient to zero to evaluate the impact on consumer choices, assuming that
consumers become indifferent to whether a restaurant is available on multiple platforms.
To explore this, the model was adjusted to nullify the effect of multihoming, making a
restaurant's presence on multiple platforms irrelevant to consumer choice. This adjustment will

56
help assess the hypothetical situation where multihoming does not influence consumer decisions.
The model was then re-estimated to observe changes in consumer platform preferences and
restaurant selection.
Figure 2. 2 Changes in Consumer Choices by Delivery App under a Counterfactual Scenario Removing
Multihoming Influence
As depicted in the attached graphs, removing the influence of multihoming led to
noticeable shifts in consumer behavior. The total changes in consumer preferences show that
UberEats benefits more from multihoming perceptions, while DoorDash becomes more attractive
based on its intrinsic features, such as restaurant variety or platform-specific characteristics.
Removing the multihoming effect led to a decrease in the likelihood of consumers choosing
both DoorDash and UberEats, with some consumers opting instead for the outside option. If
consumers did not consider whether a restaurant was available on multiple platforms, their choices
would be based solely on other factors such as price, quality, and individual platform
characteristics. This adjustment could result in significant shifts in consumer behavior, with
decisions driven more by the intrinsic attributes of the restaurants and platforms rather than their
availability across multiple services. Consequently, the competitive advantage gained by

57
restaurants through multihoming would be neutralized, potentially leveling the playing field for
singlehoming restaurants.
Table 2. 4 Counterfactual Probabilities – by Month and Platform
Month DoorDash UberEats Outside
2 0.814 0.037 0.149
3 0.812 0.036 0.152
9 0.761 0.091 0.148
10 0.776 0.039 0.185
Table 2. 5 Original Probabilities – by Month and Platform
Month DoorDash UberEats Outside
2 0.801 0.077 0.122
3 0.812 0.061 0.127
9 0.779 0.055 0.166
10 0.800 0.052 0.148
Table 2. 6 Changes – Probabilities by Month and Platform
Month DoorDash UberEats Outside
2 + 0.013 -0.040 + 0.027
3 0.000 -0.025 + 0.025
9 -0.018 +0.036 -0.018
10 -0.024 -0.013 + 0.037
The tables above show the shift in consumer choice probabilities when the influence of
multihoming is removed. The Table 2.4 represents the expected probabilities of choices for
DoorDash, UberEats, and the outside option under the scenario where the multihoming coefficient
is set to zero. The Table 2.5 shows the actual observed probabilities before the adjustment. The
Table 2.6 highlights the differences between the original and counterfactual probabilities.
The results indicate a notable decrease in the probability of consumers choosing UberEats
and a slight decrease in choosing DoorDash when the multihoming effect is removed in Table 2.6.
Conversely, there is a slight increase in the probability of consumers choosing the outside option,
suggesting that some consumers who previously valued multihoming now consolidate their choice
to opt out of using any platform. This shift suggests that UberEats may have been benefiting more

58
from the multihoming effect compared to DoorDash. With the multihoming coefficient set to zero,
DoorDash becomes relatively more attractive to consumers, possibly due to its other restaurant
intrinsic features such as price, quality, or platform characteristics.
To add a valuable dimension to the analysis, I examine the conditional probability of
choosing a multihoming restaurant when a specific platform is selected. This analysis reveals
insights into how the availability of a restaurant on multiple platforms influences consumer choices.
For example, the equation below is the probability of choosing a multihoming restaurant given
that UberEats was chosen. This involves determining how likely it is that a user who chose
UberEats also picks a multihoming restaurant.
where 𝐽𝑚ℎ,𝑈𝐸 is the set of all the restaurants which is multihoming and in the UberEats listings
and 𝑃𝑖𝑗𝑈𝑚𝑡 is the choice probability of choosing a restaurant j such that f = UberEats.
Table 2. 7 Probabilities of choosing multihoming restaurants given that a specific platform was chosen
Original Counterfactual
Month DoorDash UberEats DoorDash UberEats
2 0.483 0.212 0.327 0.400
3 0.460 0.390 0.331 0.375
9 0.462 0.371 0.346 0.224
10 0.432 0.378 0.357 0.321
In the original scenario, the probability of choosing a multihoming restaurant given
DoorDash chosen ranges from approximately 43.2% to 48.3% across the months. On UberEats,
this probability ranges from 21.2% to 39%. These probabilities suggest that multihoming has a

59
notable impact on consumer choices, with consumers showing a preference for multihoming
restaurants when they are aware of their availability on multiple platforms.
In the counterfactual scenario, where the multihoming coefficient is set to zero, the
probabilities of choosing a multihoming restaurant decrease for DoorDash, ranging from 32.7% to
35.7%. For UberEats, the probabilities show varying changes, with an increase in February to
40.0%, but decreases afterwards from 22.4% to 37.5%. This suggests that when consumers do not
consider whether a restaurant is available on multiple platforms, they are less likely to choose
multihoming restaurants on DoorDash and UberEats.
Figure 2. 3 Change in Conditional Probability of Multihoming Restaurants Given Selected Platforms
The conditional probability of consumers choosing a multihoming restaurant when
selecting a specific platform decreased across both DoorDash and UberEats (Figure 2. 3). Without

60
the multihoming influence, consumers are less likely to choose multihoming restaurants, with the
probabilities decreasing more significantly on UberEats than on DoorDash.
The analysis reveals that multihoming has a notable impact on consumer choices. Without
its influence, consumers tend to rely more on intrinsic factors like price, quality, or platformspecific features. The findings suggest that UberEats benefits more from the perception of
multihoming, whereas DoorDash's attractiveness is driven by other factors, making it relatively
more resilient when multihoming is neutralized.
Platforms may need to reconsider their strategies regarding multihoming. Encouraging
exclusive partnerships or enhancing platform-specific features could become more critical if the
influence of multihoming diminishes. Additionally, focusing on improving consumer perception
of individual platform attributes may help attract and retain users. The analysis underscores the
importance of considering underlying qualities such as price, quality, and platform-specific
features when developing strategies to attract and retain both restaurants and users.
2.9 Linking Restaurant Platform Choices to Consumer Choice Problems
As outlined in Section 3, the food delivery industry operates within a complex, multi-stage
framework where decisions by one group of market participants influence the choices and
outcomes of others. This interconnectedness can be effectively analyzed through a sequential game
framework, where platforms, restaurants, and consumers make strategic decisions at different
stages. This section explores how restaurants’ platform selection and menu pricing decisions
directly impact consumer options and behavior.
To examine these dynamics, I conducted a counterfactual analysis in which DoorDash
increases its premium plan commission rate from 30% to 40%, while maintaining a 15% rate for

61
the basic plan. UberEats, in contrast, keeps its commission structure at 15% for the basic plan and
30% for the premium plan. This change is treated as exogenous, without altering other platformspecific characteristics.
Using previously estimated parameters in Chapter 1, I recalculated the utility levels for
each restaurant’s platform choice under the new commission structure. This involved predicting
how restaurants might adjust their strategies, whether by opting for DoorDash, UberEats, or
multihoming.
Figure 2. 4 Comparison of Platform Selection by Restaurants Between Original and Counterfactual Scenarios
The results reveal intriguing patterns in platform selection. Despite the increased premium
commission rate, the proportion of restaurants exclusively listing on DoorDash rose slightly from
55.4% to 57.3%. Additionally, the share of restaurants using both platforms increased from 31.1%
to 32.7%, while the percentage exclusively on UberEats decreased from 13.5% to 10.0%.
This outcome can be attributed to DoorDash's strong market position and network effects,
which appear to outweigh the deterrent effect of higher commissions. Restaurants likely perceive

62
DoorDash as offering superior access to a larger customer base, better delivery coverage, or more
effective platform features compared to UberEats. The positive coefficient for Consumer Market
Share and the smaller fixed effects on DoorDash in Table 1. 3 support this result, indicating that
restaurants are highly responsive to the platform's perceived popularity.
Following these restaurant decisions, I analyzed consumer behavior in response to adjusted
menu prices, assuming that restaurants proportionately increased their prices to reflect the higher
commission rate. The results in Figure 2.5 show that an increase in the number of restaurants on
DoorDash led to a significant rise in consumer preference for the platform. The percentage of
consumers choosing DoorDash rose from 78.3% to 85.6%, while those choosing UberEats dropped
from 6.7% to 5.4%. Additionally, fewer consumers opted for alternatives outside the platforms,
with this figure decreasing from 15.0% to 9.0%.
Figure 2. 5 Comparison of Consumer Choices over platforms Between Original and Counterfactual Scenarios
Interestingly, despite higher menu prices, the expanded restaurant pool on DoorDash
attracted more consumers. The strong coefficient for the Number of Restaurants on DoorDash in
the consumer model (Table 2.3) highlights that consumers are highly responsive to the variety

63
offered. DoorDash’s advantages in restaurant variety and platform features likely drove this shift
in consumer preference, even with higher costs.
The counterfactual analysis in this section indicates that increasing DoorDash’s premium
plan commission rate led to more restaurants choosing the platform and, subsequently, more
consumers favoring it. Specifically, a 3.6 % increase in the DoorDash restaurant pool resulted in
an approximately 9.3% increase in consumer choice for DoorDash. This suggests that platformspecific characteristics, such as market share and restaurant variety, play a crucial role in both
restaurant and consumer decisions, often outweighing the negative effects of higher costs.
These findings demonstrate the two-sidedness of the food delivery market, where strategic
choices made on the restaurant side cascade into consumer decisions, creating a dynamic and
interconnected marketplace. Understanding these links is critical for comprehending the full scope
of platform competition and user behavior in this industry. For platforms, this highlights the
importance of balancing commission rates with value-added services that can enhance restaurant
retention and consumer attraction. It also suggests that DoorDash may have the potential to
increase its market share and margins even with higher commission rates. However, the model's
limitations—such as not differentiating between plan selections and not accounting for profit
margin variations among restaurants—suggest areas for future research to refine the understanding
of these dynamics.
2.10 Conclusion
This study provides valuable insights into the determinants of consumer choices in the
online food delivery market. The analysis reveals that cost, restaurant quality, and multihoming

64
status significantly influence consumer preferences. Higher ratings, established brands, and
premium plans positively impact consumer choices, while higher delivery fees deter platform
selection. The counterfactual analysis, where the multihoming effect is neutralized, indicates
notable shifts in consumer behavior. Without considering multihoming, consumers are less likely
to choose multihoming restaurants on both platforms, suggesting that these restaurants are
appreciated with multihoming and possess favorable characteristics such as better prices and
higher quality.
Combined with Chapter 1, both the consumer choice problem and restaurant platform
adoption problem provide insights into the price-setting strategies employed by platforms. By
comprehensively understanding the impact of commission rates and consumer fees, this study can
be expanded to address the setting of these fees by platforms. Changes in consumer fees will
impact consumer choice, subsequently affecting consumer market share and the probabilities of
restaurant platform adoption. This interconnected analysis reveals how fee adjustments influence
both consumer behavior and restaurant decisions, ultimately guiding platforms in optimizing their
pricing strategies.

65
Chapter 3
Effect of Code-Share Exit:
Evidence from Merger of Alaska and Virgin America8
3.1 Introduction
In 2016, Alaska Airlines (AS) acquired Virgin America (VX). The merger between a fringe
legacy carrier and a low-cost carrier (LCC) resulted in the birth of a strengthened firm capable of
competing with other main airlines in the U.S. (Figure 3.1). The merger between Alaska and Virgin
is unique in that the Department of Justice (DOJ) required the modification of Alaska’s codeshare
alliance with American Airlines (AA). Most remedies in recent airline mergers have been enforced
with the divestiture of gates or slots to the competitors, as structural remedies. However, the
remedies involved in this merger were conduct remedies, or behavioral remedies.9
8 This is joint work, coauthored with Hae Yeun Park.
9
In cases where U.S. federal antitrust enforcement agencies, such as the Antitrust Division of the U.S. Department of Justice (DOJ)
and the Federal Trade Commission (FTC), raise concerns about competitive harm, appropriate remedies may be issued to protect
competition in the market while allowing the merger to proceed. These remedies are typically classified as either structural or
behavioral. Structural remedies involve the divestiture or licensing of the merging firms’ assets, while behavioral remedies impose
restrictions on the merged firm’s conduct (U.S. Department of Justice, Antitrust Division, 2011). Historically, behavioral remedies
have been discouraged due to their implementation challenges, higher monitoring costs, and the risk of circumvention by the
merged firms (Delbaum and Skinner, 2020). For example, in the merger between US Airways and American Airlines, the DOJ
required the merging parties to divest slots and gates at key constrained airports to low-cost carriers, enhancing competition and
benefiting consumers with more choices and competitive airfares (U.S. Department of Justice, 2013).

66
Figure 3. 1 US Airlines’ Market Share in 2016
The government authority was greatly concerned about a potential loss of competition in
markets where Virgin America fiercely competed with American Airlines while Alaska Airlines
was dependent on American Airlines, given their codeshare agreements in many markets. The
modification of codeshare agreement between the merging parties and American was designed to
ensure that Alaska would have additional incentives to vigorously compete with American, just as
Virgin had done before 2016, and would directly impact the codeshare connection in the network
of Alaska and American airlines. Since the required behaviors represented exogenous shocks to
the Alaska-American (AS-AA) codeshare products, this study investigates whether and how
codeshare discontinuation impacted market prices and benefited consumers.
In order to examine the changes associated with the merger and the impact of codeshare
discontinuations, this study begins with a difference-in-differences (DID) fare comparisons
regarding the merger. In this analysis, we compare the market level prices of different types of
markets (types being defined by the presences of the merging airlines on the markets). The results

67
support neither pro-competitive nor anti-competitive effects; the market level fares as well as the
market ticketing level fares exhibit no significant changes following the merger. In the next section,
we use two-way fixed DID regressions to estimate codeshare exit effects. In this analysis, we
compare changes in price of affected markets with those of unaffected markets. The findings show
the codeshare cessation was associated with relative price decreases, especially on the tickets
marketed by the affected airlines (Alaska and American). To examine if these effects are associated
with decreased demand, we estimate the demand of air travel on the relevant markets. Using a set
of relevant dummy variables and their interactions, we conclude that discontinuation of AS-AA
codeshare products did decrease demand for products on the codeshare exit markets, even though
the deceased demand was not specific to AS-AA products.
The remainder of this paper is organized as follows. Section 3.2 presents relevant prior
literature on merger analysis and the competitive effects of airline codeshare alliances. Section 3.3
provides a working definition of codeshare products and briefly discusses AS-AA codeshare
circumstances. Section 3.4 describes the background of the Alaska/Virgin (AS/VX) merger and
the related remedies enforced by the authorities. Section 3.5 describes our data sources and the
construction of our sample. Section 3.6 illustrates how the price changes were compared in terms
of the merger, as well as the results. Section 3.7 describes the empirical methodology for a
codeshare discontinuation analysis and presents the results. Section 3.8 explores the structural
estimation on consumer demand to explain the connection between codeshare and consumer utility.
Finally, the last section of this paper presents a discussion and conclusion.

68
3.2 Literature Review
In this section, we briefly present previous studies focusing on merger effects and
codeshares in airlines industry.
Mergers in the US airlines industry have been extensively studied. There have been
dynamic changes in airlines competition with consecutive huge mergers. This enabled to
investigate the competitive effects of each merger cases and related government enforcement.
Earlier studies generally concerned the anti-competitive effect of the merger. Many argued a loss
of competition, increased fares and harm to customers. (Peters, 2006; Kwoka & Shumilkina, 2010)
Recent studies, however, opened a new possibility with different effects of a merger. Luo
(2014) found that the legacy airlines merger did not significantly change the fares on the affected
routes. He also concluded that the price impact of low-cost carrier competition was larger than that
of legacy carrier competition. Das (2019) studied the merger effect on fares and product quality.
His difference-in-differences analysis showed that the merger between American Airline (AA) and
US Airway (US) has actually decreased the market fare but has not affected on frequency of flights
or the number of seats. Carlton et al. (2019) studied the merger effect on fares and output for three
recent legacy mergers. The paper confirmed all the mergers were pro-competitive, as they found
no adverse effect on nominal fares and a significant increase in passenger traffic.
The codeshare agreements were also widely studied in the airline industry. While there are
many studies on international codeshare, we focus on studies on domestic codeshare agreement in
this section.
Ito and Lee (2005) and Ito and Lee (2007) document definitions of different types of air
travel products in terms of structures of ticketing and operating carriers. After addressing some
stylized facts about codesharing practices in the US airline industry, they show that virtual

69
codesharing tickets are priced lower than those ticketed and operated by a single carrier.10 From
this finding, they conclude that airlines might use a virtual codeshare product as an inferior
substitute to a pure online product for passengers who are more price sensitive.
Gayle (2008) addresses the U.S. Department of Transportation (DOT)’s main concern that
the codeshare alliance among three major airlines, Delta, Continental, and Northwest could
facilitate collusive behaviors of the partners, especially on their overlapping routes. He estimates
actual effects of the codeshare alliance on price and traffic (number of passengers) on the partners’
overlapping routes. He compares the changes in average price or total traffic for city pairs on which
the partners codeshare (alliance city pairs) with those of city pairs on which at least one of the
three provides service but there is no code sharing between any of the three carriers (non-alliance
city pairs). The aggregated analysis shows that, overall, the alliance is associated with price
increases and traffic increases. He also investigates whether different types of codeshares had
different effects. The findings show traditional code sharing is associated with price decreases and
traffic increases, and virtual code sharing is associated with price increases and traffic increases.
From these findings, he concludes the alliance is not associated with collusive behavior in the
partners’ overlapping markets.
Structural models of codesharing between U.S. domestic airlines have also been developed
recently. Gayle (2013) focuses on the traditional codesharing. He takes a vertical relation approach
10 The definitions of a virtual codeshare in Ito and Lee (2005) and Ito and Lee (2007) are slightly different from one in this paper.
In their study, they divide virtual codeshare into two groups: a semi virtual codeshare and a fully virtual codeshare. A semi virtual
codeshare product is an itinerary which is operated by a single carrier, but partly ticketed by another airline. A fully virtual codeshare
product is defined as an itinerary operated by a single carrier and ticketed by another single carrier. A virtual codeshare product in
this study is a fully virtual codeshare in Ito and Lee (2005, 2007), and we exclude a semi virtual codeshare because they involves
multiple ticketing carriers.

70
to test the existence of double marginalization in pricing the codeshare products. Gayle uses a
random coefficients logit demand model, and he allows the supply side of the model for a double
marginalization with two kinds of margins in the prices of codeshare products: an upstream margin
of operating carrier and a downstream margin of marketing carrier. Gayle finds that the upstream
margin persists if the upstream operating carrier simultaneously offers a competing online product
in the same market. It is because the operating carrier optimally chooses not to eliminate its
upstream margin of codeshared products to reduce the intensity of downstream competition for its
own product.
Two other papers adopt a structural model for studying the effect of virtual codesharing.
Gayle and Brown (2014) see whether the virtual codesharing was associated with a collusive
pricing behavior. They study the codeshare contracts among Delta, Continental and Northwest
airlines and finds that the average price increased, and average passengers decreased in their
overlapping market. Demand estimation with a nested logit demand model, however, suggests that
the codeshare alliance has a demand increasing effect, given the alliance creates new opportunities
for passengers to accumulate and redeem the points across airlines. Also, the non-nested statistical
test on price setting behavior assumptions suggests that there was no statistical evidence of
collusive behavior between partners. Shen (2017) studies two codeshare alliances in 2003. He sets
a pricing equation where the ticketing carrier keeps a share of the profits as a commission fee and
the operating carrier acquires the rest of the profits. He finds that codesharing reduces marginal
cost of the airlines, so they can price codeshare products at a lower price. More importantly, he
finds if codesharing creates new products in the market, the demand increases and consumers have
larger surplus.

71
Our analysis differs with those previous literatures in two ways: This paper is the very first
codeshare analysis with the codeshare exit case, while all the others studies codeshare formation.
Also, we could even regard the codeshare exit change as exogenous and its justifiable rationale
behind this assumption is that the decision was formally ordered by the Department of Justice.
3.3 Codeshare Alliance
In this section, we present working definitions for the different types of codeshare products
and describe the specific codeshare agreement between Alaska and American. We then document
the merger and remedy background in the following section.
3.3.1 Definition
A codeshare alliance allows one carrier (“a ticketing/marketing carrier”) to sell and market
the tickets of its partners’ flights using the code of the ticketing/marketing carrier. The partner
carriers, those that actually operate the flight and carry the passengers, are called “operating
carriers”. For example, suppose that American Airlines is selling tickets for a flight, AA999, that
is actually operated by its codeshare alliance partner, Alaska Airlines. In this case, even though a
passenger buys a ticket from American Airlines (AA999), that passenger might actually take a
flight operated by the codeshare partner, Alaska Airlines.
The literature on airline codeshare alliances has defined two types of codeshare itineraries:
(1) traditional codeshare; and (2) virtual codeshare. The different types of products are illustrated
in Figure 3.2. All three tickets represent an example of three types of products marketed by
American (AA) on the market between Los Angeles (LAX) and Boston (BOS). In this figure, the
red airplanes represent flights operated by American, and the blue airplanes represent flight
operated by Alaska (AS).

72
Figure 3. 2 Three Types of Airlines Products
These three tickets represent an example of the three types of products on the market between LAX and BOS. The
first nonstop ticket is marketed and operated by American Airlines, and is an example of a “pure online” product. The
second ticket is operated partially by the marketing carrier (AA) and partially by the carrier (AS) apart from the
marketing carrier (AA) and is defined as a “traditional codeshare” product. The last ticket represents a product wholly
operated by the carrier (AS) apart from the marketing carrier (AA) and is defined as a “virtual codeshare” product.
Traditional codeshare itineraries offer interline operating services from codeshare partners
while the flights are marketed solely by a ticketing carrier who is one of those two distinct
operating carriers.11 This type of product is illustrated as the second ticket in Figure 2.2.
In contrast, virtual codeshare products offer an online operating service for a partner carrier,
meaning that passengers on these itineraries remain on a single operating carrier’s plane(s) for the
entire trip. The ticket for the entire flight is marketed by a ticketing carrier, which is distinct from
the operating carrier in these itineraries. Here, the ticketing carrier is always distinct from the
operating carrier. This type of codeshare itinerary is illustrated as the third ticket in Figure 3.2.
Thus, a key distinction between traditional and virtual codeshares is that the former requires the
passenger to travel on different operating carriers’ planes (interline air travel)—one of them is the
ticketing carrier—on a multi-segment route, while the latter does not involve interline air travel
even when passengers change planes on a multi-segment route.
11 An interline service describes when segments of the itineraries are operated by distinct airlines, requiring passengers to
change airline brands when they change planes at some point in their itinerary. An online service describes when passengers
travel on a single airline brand through the entire trip.

73
The flight itineraries marketed and operated by one airline are known as “pure online”
products. In this case, the marketing carrier and the operating carrier are always identical. The first
ticket in Figure 3.2 illustrates an example of a pure online product.
3.3.2 Alaska (AS) and American (AA) Codeshare Agreement
In 2016, Alaska Airlines maintained codeshare partnerships with two other legacy airlines:
American Airlines (AA) and Delta Airlines (DL). The codeshare agreement between American
and Alaska was implemented in 1999. Since then, the two airlines repeatedly expanded their
arrangement, increasing the number of routes on which each partner could sell on behalf of the
other. As a result of the merger between American (AA) and US Airways (US) in 2013, Alaska
could sell more flights on American’s significantly expanded network in 2015. In April 2016, just
prior to the AS/VX merger, American and Alaska once again expanded their codeshare agreement.
This expansion enabled Alaska to market American flights on over 250 routes and allowed
American to market Alaska flights on approximately 80 routes. This enabled Alaska and American
to sell each other’s flights on certain overlap routes where both partners already offered competing
non- stop services. This situation raised serious concerns regarding potential loss of competition
with the Department of Justice (DOJ). Existing literature has widely discussed how codeshare
partners “may choose to compete less vigorously than they otherwise would” (Ito and Lee, 2007).

74
Figure 3. 3 AS-AA Codeshare Markets (Traditional)
Figure 3. 4 AS-AA Codeshare Markets (Virtual)
Figures 3.3 and 3.4 show in how many routes AS-AA traditional or virtual codeshare
products were marketed, respectively. For these figures, AS/AA indicates codeshare products

75
marketed by Alaska (AS) and operated by American (AA), while AA/AS indicates codeshare
products ticketed by AA and operated by AS. The vertical dashed grey lines indicate 2016Q4, when
the final judgment was proposed, and the solid grey lines indicate 2017Q2, when the final
judgment was revised and finalized. The green, blue, and purple lines are drawn cumulatively, so
the area under the square-connected purple line (AS-AA cs) show the number of markets in which
both AS/AA and AA/AS codeshare products were sold. Similarly, the area between the squareconnected purple line and the triangle-connected blue line (AS/AA cs only) show the number of
markets in which only AS/AA codeshare products were marketed, while the area between the
triangle-connected blue line and the circle-connected green line (AA/AS cs only) show the number
of markets in which only AA/AS codeshare products were marketed. Figures 3.5 and 3.6 show
how many passengers used AS-AA traditional or virtual codeshare products, respectively. The bars
are drawn cumulatively, and the trends are similar to those of the number of markets.12
12 The vertical red lines indicate times when the final judgment was first proposed (dashed) and revised and finalized (solid), similar
to Figure 3.3 and 3.4.

76
Figure 3. 5 AS-AA Codeshare Total Passengers (Traditional)
Figure 3. 6 AS-AA Codeshare Total Passengers (Virtual)

77
3.4 Alaska Airlines’ Acquisition of Virgin America
In this section, we briefly describe the Alaska Airlines’ acquisition of Virgin America and
the remedies associated with the acquisition.13
3.4.1 Merger Background
In April 2016, Alaska Airlines (AS) announced its plans to acquire Virgin America (VX),
and it was approved by the Virgin America’s shareholders in July 2016. While there was a lawsuit
filed against Alaska Airlines by private plaintiffs to block the merger between these two carriers,
the Alaska Air Group was able to settle the lawsuit and the merger was approved by the Department
of Justice (DOJ) in December 2016. About a week after the approval, the merger was officially
closed. Even though there were no asset divestitures required by the authority, the merger was
approved conditional on modifications of the codeshare agreement between Alaska (AS) and
American Airlines (AA), which are described in more detail in the next subsection. The merged
Alaska got a single operating certificate from the Federal Aviation Administration (FAA) in
January 2018. Until the last flight of Virgin America (VX) operated on April 24, 2018, the two
airlines had integrated their various systems gradually.
Even though the merger was approved and consummated eventually, there were concerns
regarding this merger. Many of the concerns were about reductions in competitions. Since the two
airlines’ routes were concentrated on the western U.S., an elimination of one airline could have
reduced competitions in the area significantly. The airlines even shared the same hub airports, Los
Angeles (LAX) and San Francisco (SFO). As perceived from the lawsuit and the DOJ’s remedy,
13 In this study, we do not distinguish an acquisition from a merger because differences between the two are not relevant to the
purpose of this study.

78
the merger was expected to reduce competitions and results in increases in prices in general.
Another concern has to do with processes of integration. Operational inefficiencies are one of the
problems associated with a merger, and the integration between Alaska and Virgin American might
not have been an exception. Despite of these concerns, the two airlines put effort in integrating and
stabilized as a combined airline.
3.4.2 Remedies
In order to resolve the likely competitive harm associated with the merger, US District
court for the District of Columbia constructed the conduct remedies on its approval of the merger
between Alaska and Virgin America. The authorities mainly concerned the loss of competitor
against American Airlines, the world’s largest airlines, as a consequence of this acquisition.
Virgin America had, until the merger, aggressively competed with American (AA) on their
overlap routes in ways that forced American to respond with lower fares and better services. In
fact, markets where Virgin America and American overlapped had expanded significantly in terms
of market total passengers compared to other American and Alaska-American overlap markets.
Moreover, in markets where both American and Virgin operated, they had similar passenger shares,
indicating that Virgin was indeed a strong competitor against American.
Alaska’s codeshare agreement with American, meanwhile, created an incentive for Alaska
to avoid aggressive head-to-head competition in order to preserve its relationship with American.
With constant expansion of the codeshare agreement between the two airlines, they operated codeshare products even on overlap routes where both already offered competing pure online products,
which raised concerns regarding a loss of competition.

79
Taking these issues into account, the DOJ required Alaska to significantly scale back its
code-share agreement with American, identifying four main markets where Alaska and American
should not operate codeshare products.14
First, the DOJ prohibited Alaska and American from
offering their codeshare products on routes where Virgin America and American both offered
competing nonstop services during the pre-merger period.15 Second, the DOJ prohibited Alaska
and American from code sharing on Alaska-American (AS-AA) nonstop overlap routes.16 Third,
the DOJ prevented either Alaska or American from marketing each other’s flights on routes that
included their respective key airports.17 For example, Alaska is prohibited from selling American
code-shared flights on their routes from Seattle, one of Alaska’s key airports. Lastly, the DOJ
prohibited either carrier from codesharing on routes between Los Angeles (LAX) and either
American or Alaska key airports.
14 The final judgement can be found at https://www.justice.gov/atr/casedocument/file/1039436/download. Other related documents can be found at
https://www.justice.gov/opa/pr/ justice-department-requires-alaska-airlines-significantly-scale-back-codeshareagreement and https://www.justice.gov/atr/case/us-v-alaska-air-group-inc-and-virgin-america-inc; U.S.
V. ALASKA AIR GROUP, INC., AND VIRGIN AMERICA INC. (last accessed in March, 2022).
15 The DOJ specified 42 VX-AA U.S. domestic overlap routes as of December 6, 2016 in its final judgement, and the list is in
Appendix Table B.1.
16 The DOJ described 62 AS-AA U.S. domestic overlap routes as of December 6, 2016 in its final judgement for illustrative
purposes, and the list is in Appendix Table B.2. For this definition of overlap routes only, the authority identified the overlap
routes not on the basis of airport, but rather on the basis of city. Thus, airports with the same City Market ID, as identified by
the U.S. Department of Transportation in the Airline Origin and Destination Survey (DB1B), are regarded as serving the same
city, with the following airports excepted: (1) Los Angeles International Airport (LAX), and (2) Norman Y. Mineta San Jose
International Airport (SJC). Thislist of Alaska/American overlap routes may be subject to change as both airlines adjust their
respective schedules.
17
The DOJ defined four Key Alaska Airports as follows: (1) Portland International Airport (PDX); (2) Seattle- Tacoma
International Airport (SEA); (3) San Francisco International Airport (SFO); and (4) Ted Stevens Anchorage International
Airport (ANC). They also identified twelve Key American Airports as follows: (1) Charlotte Douglas International Airport
(CLT); (2) Chicago Midway International Airport (MDW); (3) Chicago O’Hare International Airport (ORD); (4) Dallas/Fort
Worth International Airport(DFW);(5)Dallas Love Field (DAL);(6) FortLauderdale- Hollywood International Airport (FLL);
(7) John F. Kennedy International Airport (JFK); (8) Miami International Airport (MIA); (9) New York LaGuardia Airport
(LGA); (10) Philadelphia International Airport (PHL); (11) Phoenix Sky Harbor International Airport (PHX); and (12)
Washington Reagan National Airport (DCA).

80
In the second quarter of 2016 in the DB1B data, 96 distinct routes can be identified as
affected by these codeshare modification remedies. The four categories of remedy routes
mentioned above were not mutually exclusive. To better understand the affected routes, we
distinguish all the remedy markets into two sets depending on the existence of codeshare products
between Alaska (AS) and American (AA). After exploring the data (Table 3.1), we found a stylized
fact that, in the first category (codeshare regulations on the VX-AA overlap routes), there had been
little appearance of codeshare products between Alaska and American. Inasmuch as the other
remedy markets were directly related to the network of Alaska and American airlines, we found
that the effects of codeshare elimination were relatively stronger in the other three categories of
the remedy markets. More details are provided in Appendix B.

81
Table 3. 1 Remedy Relevant Markets with Codeshare Products
Time
CAT 1 CAT 2 CAT 3 CAT 4 Total
VX/AA Overlap AS/AA Overlap AS KEY AA KEY LAX:KEY Relevant Mkt
ASAA AAAS ASAA AAAS ASAA AAAS ASAA AAAS ASAA AAAS
2015Q1 58 36 1520 5155 30 6615
0 0 5 11 74 26 0 0 74 28
2015Q2 58 38 1593 5270 30 6803
0 0 3 7 71 22 0 1 71 24
2015Q3 58 52 1600 5274 30 6816
0 0 8 11 71 41 0 0 71 44
2015Q4 58 52 1557 5319 30 6816
1 1 7 6 68 35 1 2 68 39
2016Q1 58 48 1511 5207 30 6660
0 0 5 7 49 29 1 1 49 32
2016Q2 58 50 1586 5308 30 6836
1 4 7 18 58 38 1 4 58 45
2016Q3 58 49 1597 5308 30 6845
0 4 4 14 56 46 0 5 56 49
2016Q4 58 64 1560 5299 30 6807
0 4 5 12 59 31 0 4 59 33
2017Q1 58 62 1533 5229 30 6713
0 3 4 12 71 25 0 3 71 27
2017Q2 58 69 1570 5316 30 6840
0 0 6 2 66 30 0 0 68 32
2017Q3 58 67 1584 5279 30 6813
0 0 6 2 65 26 0 0 66 28
2017Q4 58 69 1578 5316 30 6849
0 0 2 0 59 18 0 0 59 18
Each column stands for each category of the codeshare remedies defined on original Final Judgment documented by
the Department of Justice. The first row of each quarter contains the number of remedy markets that document clarified
with definitions and the second row for the number of markets where the codeshare products between two airlines
were observed in data. ASAA stands for the codeshare product marketed by Alaska and operated by American, AAAS
stands for the vice versa.
3.5 Data
3.5.1 Data Sources
The main data used in this paper is the Airline Origin and Destination Survey (DB1B)
dataset from the Bureau of Transportation. This is quarterly 10 percent sample of domestic air
travel tickets reported by reporting carriers. 18
The data contain detailed information on each
18 Usually, a reporting carrier is an airline that operates the first segment of a trip. For example, if a one-stop itinerary is marketed
by American Airline, operated by Alaska Airline, it is likely that a reporting carrier is Alaska Airline.

82
itinerary such as itinerary fares, number of passengers who purchased a ticket at the price,
origin/destination cities, origin/connecting/destination airports. The data also include information
on ticketing (marketing), operating, and reporting carriers.
We supplement the DB1B data with a couple of other data to construct control variables.
Population data are from U.S. Census Bureau.19
Although population might not have changed
dramatically over time, it is included to capture market size as well as potential economies of
traffic density. In order to capture changes in economic conditions, unemployment rate data are
included from U.S. Bureau of Labor Statistics (BLS). One of main cost shifters, jet fuel price data,
is from U.S. Energy Information Administration.
3.5.2 Sample Construction
A basic observation unit—before collapsing—is defined as a roundtrip itinerary on a
market at a particular time (quarter). A market is defined as a directional airport-pair of origin and
destination airports.20 For example, a trip from Los Angeles (LAX) to Dallas (DAL) is treated as
a different market from a trip from DAL to LAX.
Several restrictions have been applied in order to construct the analysis sample. First, for
every analysis in this study, any itinerary that is either bulk fare or suspected to have incredible
fare is not included. Itineraries which include ground transportation or roundtrip itineraries with
19
Annual Estimates of the Residence Population: April 1, 2010 to July 1, 2018: U.S. Census Bureau, Population Division. It
produces estimates of the population for the United States, states, metropolitan and micropolitan statistical areas, counties, cities,
towns.
20 In empirical literature regarding the airline industry, two common ways to define a market are to use routes either between citypairs or airport-pairs.Each definition can be more suitable than the other depending on the specific research questions and
assumptions of any given study. In this paper, we use airport-pairs, because the majority of the remedies introduced by the DOJ (i.e.,
reductions of the codeshare operations between AS and AA in certain markets) were applied at an airport-pair level. One exception
to this is the second category, which specifically targets the city-pair market between AS and AA overlap. This category can be
approximately covered by the airport-pairs.

83
more than two trip breaks are also dropped from the sample. To exclude anomalies, tickets with
too low or too high fares are not included.21 Any tickets with more than two coupons are dropped.
In other words, the tickets included in the sample can have at most one stop per one-way trip.
Itineraries with multiple ticketing carriers or foreign carriers are not included. The sample includes
only pure online products and traditional codeshare products. This decision was made upon
examination of two types of codeshare products between Alaska and American. We confirmed that
the traditional codeshare was the prominent relationship between the two codeshare partners at the
time of the merger and remedy decision. As confirmed from data presented in Figures 3.4 and 3.6,
almost no virtual codeshare products existed between Alaska and American starting from 2015.
Although many of Alaska Airlines (AS)’ routes are from or to airports in the state of Alaska
(AK), airports from outside the contiguous U.S. are not included in this study, mainly because
routes in the contiguous U.S. are likely to have different features than those in non-contiguous
states. Since all airports in the sample are located within the contiguous U.S., any particular
characteristics from non-contiguous states are irrelevant to this study.
Airlines included in this study are mainline carriers such as American Airlines (AA),
Alaska Airlines (AS), and Virgin America (VX).22 Although some other airlines are present on the
relevant markets, most of them experience little routine traffic. In addition, this restriction was
predicted to have an insignificant impact on our results because the included airlines represented
more than 90% of the market shares in most of the sample markets. Moreover, our study recodes
21 Itineraries with fares lower than $50 or greater than $2,000 are dropped from the sample.
22 The included legacy airlines are American (AA), Alaska (AS), Delta (DL), Hawaiian (HA), United (UA), while the included lowcost carriers are JetBlue (B6), Frontier (F9), Allegiant Air (G4), Spirit (NK), Sun Country (SY), Virgin America (VX), and
Southwest (WN).

84
affiliated regional carriers using the codes of the mainline carriers that they serve. In the absence
of such recoding, products with a mainline airline as a ticketing carrier and its affiliated regional
carrier as an operating carrier may be counted as codeshare products, since the operating and
ticketing carriers differ. The main recoding procedure conducted on the sample is clarified in Table
3.2. We found five regional carriers that had been officially affiliated according to the annual report
of the Regional Airline Association (RAA).23
Table 3. 2 Regional Airlines affiliated to Major Airlines
Regional Airlines Major Airlines Note
16, OH AA 16 →US (∼ 2015Q2), OH →AA (2015Q3 ∼ )
b
17, PT AA 17 →US (∼2015Q1), PT →AA (2015Q2 ∼ )
c
MQ AA 1998 ∼
9E DL 2013Q2 ∼
QX AS 1986 ∼
a Source: 2016 Annual Report of Regional Airlines Association, based on 2016 schedules.
b PSA Airlines changed its IATA code from 16 to OH at 20150701.
c Piedmont Airlines changed its IATA code from 17 to PT at 20150401.
3.5.2.1 Codeshare Exit Effects Sample
Following application of the conventional restrictions, we collapse our data into marketticketing carrier-time level observations. For the fare variable, passenger weighted average prices
are calculated. In order to estimate codeshare (CS) discontinuation effects, we focus our analysis
on the relevant AS-AA markets, or markets in which at least one of the AS/AA and AA/AS
traditional codeshare products were operated during the pre-remedy period. For the AS-AA
codeshare relevant markets, two alternative definitions are used. Definition A defines the relevant
markets as routes in which AS-AA traditional codeshare products were operated during all sample
23 The RAA Annual Reports can be found at https://www.raa.org/content-hub/raa-annual-reports (last accessed in March,
2022).

85
pre-remedy periods. Definition B defines the relevant markets as routes in which AS-AA
traditional codeshare products were operated during at least one pre-remedy period. Definition A
is stricter, while Definition B includes more markets. For both definitions, CS Exit markets are
defined as routes for which there were AS-AA codeshare products in the pre-remedy period but no
AS-AA codeshare products in the post-remedy period.
Two quarters from the pre-remedy period and two quarters from the post-remedy period
are included in the sample. Quarters 2 and 3 from 2016 are chosen as the pre-remedy period, and
quarters 3 and 4 from 2017 are included as the post-remedy period. The pre-remedy quarters are
from before the DOJ’s final judgement was released, and the post-remedy quarters are from after
the revised version of the final judgement was released. Summary statistics are presented in Table
3.3.
Table 3. 3 Codeshare Exit Analysis Sample Summary Statistics
Variable Mean Std. Dev. Min Max
Definition A (observations: 1,960)
CS Exit 0.551 0.498 0 1
Post 0.484 0.5 0 1
lnFare 6.23 0.374 4.382 7.489
Fare 540.34 181.70 80 1789
Population: origin (mil.) 3.963 3.047 0.097 19.335
Population: destination (mil.) 3.155 3.669 0.148 19.335
Unemployment rate: origin 4.273 0.644 2.4 7.533
Unemployment rate: destination 4.342 0.807 2 9.167
Definition B (observations: 4,640)
CS Exit 0.657 0.475 0 1
Post 0.49 0.5 0 1
lnFare 6.218 .346 4.043 7.535
Fare 530.48 172.68 57 1872
Population: origin (mil.) 3.717 3.497 0.049 19.335
Population: destination (mil.) 3.315 3.604 0.023 19.335

86
Unemployment rate: origin 4.248 0.708 2 8.267
Unemployment rate: destination 4.31 0.796 2 9.167
Definition A defines codeshare markets as routes where AS-AA codeshare products are operated during the all sample
pre periods and at least one quarter during the post-remedy period. Definition B defines codeshare markets as routes
where AS-AA codeshare products are operated at least one quarter during the pre and post periods, respectively.
3.5.2.2 Demand Estimation Sample
To estimate passengers’ air travel demand, we continued the analysis with product-level
data. A product is defined as a combination of a trip itinerary, a ticketing carrier, and an operating
partner, if the product is a codeshare product. Examples of such products include (1) LAXJFK/AA/AA, (2) LAX-SEA-JFK/AA/AS, and (3) JFK-ATL-LAX/AA/AA. The first product,
LAX-JFK/AA/AA, is a nonstop product from LAX to JFK, ticketed and operated by American
(AA), whereas LAX- SEA-JFK/AA/AS is a onestop codeshare product from LAX to JFK, ticketed
by American, although a part of the trip is operated by Alaska (AS). Even though these two
products are different in number of stops (nonstop vs. onestop) and product types (pure online vs.
traditional codeshare), they are products in the same market, the market of LAX:JFK. The last
example is an onestop pure online product, ticketed and operated by American in the JFK:LAX
market. As stated previously, the LAX:JFK market is different from the market of JFK:LAX.
Therefore, this definition of product determines the product’s observed characteristics, such as the
number of stops, miles flown, direct distance between origin and destination, and product types.
We continue the demand analysis with the stricter given definition of codeshare
discontinuation markets (Definition A in the above subsection). Thus, we have 59 codeshare
discontinuation markets (CS Exit markets) and 49 codeshare continuation markets (CS Keep
markets). We used the same time period to maintain the consistency of the analysis. Summary
statistics are presented in Table 3.4.

87
Table 3. 4 Demand Estimation Sample Summary Statistics
Variable Obs Mean Std. Dev. Min Max
Y (= lnSj −lnS0) 5,143 -13.34 1.45 -16.02 -7.33
Fare (in $100) 5,143 5.5 2.01 0.8 19.98
lnSjg 5,143 -1.7 1.51 -8.43 0
lnSj|hg 5,143 -1.4 1.18 -6.49 0
lnSh|g 5,143 -0.3 1.06 -8.19 0
Nonstop 5,143 0.07 0.26 0 1
Inconv 5,143 1.13 0.15 1 2.21
Tra CS 5,143 0.41 0.49 0 1
CS Exit 5,143 0.57 0.5 0 1
Post 5,143 0.48 0.5 0 1
CS Exit ×Post 5,143 0.27 0.44 0 1
Definition A is used for the sample.
3.6 Descriptive Merger Analysis
Prior to investigating the codeshare exit effects, which is the focus of this study, we
conducted a descriptive analysis on how ticket prices were changed before and after the AlaskaVirgin America (AS-VX) merger. In this section, we conduct market level fare comparison as well
as a comparison of fare changes in the most relevant markets in terms of antitrust issues—routes
the merging parties operated before and after the merger.
3.6.1 Market Level Fare Changes
Similar to the work of Carlton et al. (2019), we focus on market level average prices to
compare changes in different types of markets. To compare market level changes, the markets are
categorized into five exclusive groups: Overlap, Either, Exit, Enter, and Neither markets. Overlap
markets are routes in which both Alaska and Virgin America operated as direct competitors. This
type of market includes most relevant markets in terms of merger effects, since there will be a
direct reduction of airlines in the market, whereas efficiency gain can be realized substantially.

88
Either markets are routes in which only one of the merging airlines operated during the pre- and
post-merger periods. Exit markets are routes in which at least one of the merging airlines operated
prior to the merger but ceased to operate after the merger. Enter markets are routes in which neither
of the merging airlines operated prior to the merger, but in which they began to operate after the
merger. Lastly, Neither markets are routes in which neither of the merging airlines operated before
or after the merger. Because Neither markets are considered to be the least affected by the merger,
subsets of Neither markets are likely to be used as a control group when conducting market level
merger effects analysis. For these different types of markets, market level passenger weighted
average fares are averaged for the pre- and post-merger periods, respectively. These differences
are also compared to the differences of Neither markets as if Neither markets are used as a control
group.
Table 3. 5 Pre & Post Differences: Market Level Fares
Market Def. No. of Markets Pre Post Differences Diff-in-Diff
[(2) - (1)] [(3) - Neither]
(1) (2) (3) (4)
Overlap 69 441.01 435.89 -5.12 13.39
(11.44) (11.35) (16.12) (16.27)
Either 304 401.57 394.49 -7.08 11.43
(4.95) (5.16) (7.15) (7.49)
Exit 15 473.84 476.79 2.95 21.46
(21.53) (28.45 ) (35.68) (35.75)
Enter 40 515.14 479.85 -35.29 -16.78
(13.66) (14.88) (20.2) (20.32)
Neither 2,583 419.68 401.16 -18.51**
(1.53) (1.64) (2.24)
a Standard errors are in parentheses.
b Only markets with at least 900 passengers per quarters and with minimum average number of airlines that is greater
than 2 are included.
c Column (4) difference-in-differences use Neither market as a control group.
d **: 1% significance, *: 5% significance, †: 10% significance.

89
Table 3.5 shows the market level fare comparison results; these results are generally
insignificant. For each of the market types, Column (1) in the table shows the mean of the average
market fares in the pre-merger period; Column (2) shows the mean of the average market fares in
the post-merger period; Column (3) shows the differences between the pre- and post-merger
periods; and Column (4) shows the difference-in-differences that the difference of Neither markets
subtracted from each differences of markets. Even though the average prices dropped in the postmerger period, with the exception of Exit markets, these differences are not significant, and the
differences are not significantly different compared to the difference of Neither markets. The
results suggest that the merger affected markets did not experience higher or lower fare changes
compared to the markets considered the least affected by the merger. More importantly, this merger
appears not to be associated with anti-competitive price effects.
3.6.2 Fare Changes in Overlap and Either Markets
Even though market level merger effects were not detected based on the results, it is
possible that the merging airlines’ prices were affected in specific markets. Since both the Overlap
and Either markets are routes in which the merging airlines presented during the pre- and postmerger periods, we compare pricing changes from the merging airlines to those from the nonmerging airlines on the relevant routes.

90
Table 3. 6 Merger Effects Results
Overlap Markets Either Markets
(1)
OLS
(2)
FE
(3)
OLS
(4)
FE
AS/VX 0.0108 -0.0148
(0.0256) (0.0159)
Post -0.165*** -0.0233*
(0.0245) (0.0128)
AS/VX ×Post -0.00874 -0.0440 -0.00315 0.0155
(0.0403) (0.0337) (0.0224) (0.0137)
No. of Markets 69 304
No. of AS/VX 138 359
No. of the Others 367 1,265
Observations
Adjusted R
2 0.112
1,719
0.019 0.006
5,803
0.004
Market-Carrier FE No Yes No Yes
Time FE No Yes No Yes
Controls Yes Yes Yes Yes
a The dependent variable is the natural log of passenger weighted average ticket fare for the ticketing carrier on the
market.
b All specifications include populations and unemployment rates of the origin and destination cities, but the results are not reported in the table.
c ***: 1% significance, **: 5% significance, *: 10% significance.
The results are reported in Table 3.6. The first two columns of the table show the results of
the Overlap markets, and the last two columns show the results of the Either markets. All
specifications include populations and unemployment rates as controls, although the results for
these variables are not reported in the table. Columns (1) and (3) illustrate the OLS results of
estimating the following equation:
where i is an airline, m is a market, and t is a time. AS/VX is a dummy variable indicating
the merging airlines, Post is a dummy variable indicating the post-merger period, and 𝜀𝑖𝑚𝑡 is an
idiosyncratic shock at market-ticketing carrier-time level. As shown in Table 3.6, no coefficients

91
on AS/VX ×Post are significant. In order words, compared to the fare changes of the non-merging
airlines, the price changes of the merging airlines were neither significantly higher nor significantly
lower on the routes in which the merging airlines competed with each other or in which one of the
merging airlines operated without the other.
Column (2) and (4) represents the results using the market-carrier and time fixed effects
instead of AS/VX and Post dummy variables. The price effects are found to be insignificant in these
specifications as well. Comparing this together with the results from the previous subsection, it is
evident that the merger between Alaska and Virgin America is not associated with significant price
effects.
3.7 Codeshare Exit Effects
3.7.1 Empirical Strategy
In order to estimate the AS-AA codeshare exit effects, we use two-way fixed effects
difference- in-differences regressions. This method has been widely used for examining treatment
effects in various studies, as unobserved time and unit characteristics can be controlled. The main
issue with applying this method is how to define a treatment group and a control group. In our
analysis, the treatment group is defined as markets in which AS-AA ceased their codeshare
products after the remedy, whereas the control group is defined as markets in which AS-AA
continued to operate their codeshare products.
One advantage of using the AS-VX merger to estimate codeshare exit effects is that we can
reasonably treat codeshare exit decisions as exogenous. As stated in Gayle and Brown (2014),
codeshare decisions are usually endogenous, since airlines choose on which routes they offer
codeshare products with which partners. In such cases, endogeneity in codeshare decisions should

92
be addressed. Conversely, AS-AA’s codeshare discontinuation decisions in 2016 were mainly due
to the remedy required by the governmental authorities. As a result, we can treat the codeshare exit
decisions as exogenous, and thus the results are less likely to suffer from selection bias.
A ticket price is specified as follows:
where i is an airline, m is a market (an airport-pair), and t is a time (quarter). The outcome variable
is lnFare, which is the natural log of the passenger weighted average fare of the airline i on the
market m at time t. CS Exit is a dummy variable that takes a value of 1 if the observation is for the
CS Exit markets, as defined in Section 3.5.2. Post is a dummy variable indicating the post-merger
remedy period. Controls are control variables included to capture the potential effects of economic
conditions. We include the population and unemployment rates of the origin and destination cities
as the controls. For the unemployment rates, the rates are used at the level of Metropolitan/
Micropolitan statistical areas (MSA); if the MSA level rates are not available, state level rates are
used instead. 𝛾𝑖𝑚 is market-ticketing carrier fixed effect, and 𝜆𝑡
is time fixed effects. Lastly, 𝜀𝑖𝑚𝑡
is idiosyncratic shock at market-ticketing carrier-time level.
The coefficient of the main interest is the coefficient on CS_Exit×Post, 𝛽1
. 𝛽1
approximates the percentage difference between the changes in fares of observations on markets
in which Alaska and American ceased their codeshare operations, compared to the changes in fare
of observations on the markets in which Alaska or American continued to operate their codeshare
products. This represents the average codeshare exit effect on the airline on the CS Exit markets.

93
3.7.2 Results
Table 3. 7 The Codeshare Exit Effects Results (Definition A)
All Airlines AS - AA
(1)
OLS
(2)
FE
(3)
FE
(4)
OLS
(5)
FE
(6)
FE
Unbalanced Panel
CS Exit -0.182*** -0.159***
(0.0216) (0.0329)
Post 0.0351 0.0635*
(0.0218) (0.0345)
CS Exit ×Post -0.0387 -0.0478** -0.0421* -0.0952* -0.0850** -0.0897**
(0.0322) (0.0200) (0.0238) (0.0490) (0.0361) (0.0434)
No. of Markets 108 108 108 108 108 108
No. of Carriers 537 537 537 189 189 189
Observations 1,960 1,960 1,960 719 719 719
Adjusted R
2 0.071 0.004 0.005 0.091 0.015 0.019
Balanced Panel
CS Exit -0.173*** -0.166***
(0.0220) (0.0328)
Post 0.0316 0.0598*
(0.0212) (0.0340)
CS Exit ×Post -0.0481 -0.0481** -0.0458** -0.0881* -0.0881** -0.0917**
(0.0323) (0.0190) (0.0226) (0.0489) (0.0349) (0.0409)
No. of Markets 107 107 107 104 104 104
No. of Carriers 428 428 428 162 162 162
Observations 1,712 1,712 1,712 648 648 648
Adjusted R
2 0.076 0.006 0.007 0.103 0.016 0.019
Market-Carrier FE No Yes Yes No Yes Yes
Time FE No Yes Yes No Yes Yes
Controls No No Yes No No Yes
a The dependent variable isthe naturallog of passenger weighted average ticketfare ofthe ticketing carrier on the market.
b Definition A defines codeshare markets as routes where AS-AA codeshare products are operated during the
all sample pre periods and at least one quarter during the post-remedy period. Definition B defines codeshare markets
as routes where AS-AA codeshare products are operated at least one quarter during the pre and post periods,
respectively.

94
c Standard error are in parentheses and clustered at the market-airline level.
d *** : 1% significance, ** : 5% significance, * : 10% significance.
The results of this analysis are presented in Tables 3.7 and 3.8. Table 3.7 shows the results
from the Definition A sample. To see whether Alaska and American behaved differently from the
other examined airlines, the last three columns include only Alaska and American observations,
whereas the first three columns include all airlines in the sample. Because airlines enter and exit
markets, and because the DB1B data is a 10% sample, not all airlines are shown in the data for the
entire sample periods. To ensure that the results are robust for airlines present during the entire
sample periods, the lower part of the table shows the results from the balanced sample, which is
balanced in terms of market-carrier level. Since these airlines in the balanced sample were present
in all sample periods, they can be understood as more regular and significant airlines in the relevant
markets.
Columns (1) and (4) in Table 3.7 are the results from the OLS estimation. For these
specifications, dummy variables CS Exit and Post are included, along with their interaction term,
rather than the market-ticketing carrier and time fixed effects for comparison purpose. Columns
(2), (3), (5), and (6) include the unit and time fixed effects, and the controls are included in
Columns (3) and (6). The negative coefficients on CS_Exit in Columns (1) and (4) show that, on
average, ticket prices on the CS Exit markets were lower than the ticket prices on the CS Keep
markets. When the treatment of the Post dummies is controlled, no codeshare exit effects are
detected in the case of all airlines included. When we focus on Alaska and American, however, 10%
significant codeshare exit effects are detected. While the codeshare exit effects are not detected in
Column (1), the fixed effects results show the pro-competitive effects of the codeshare exit. The
results suggest that the price changes on the market in which Alaska and American ceased to

95
operate their codeshare products were lower than the price changes from the airlines on the markets
in which Alaska and American continued to operate their codeshare products. These differences
are approximately 4% when we include all airlines; the differences increase to around 9% when
we focus on Alaska and American observations. In other words, the price changes of AS-AA
products on the CS Exit markets were about 9% lower than the price changes of AS-AA products
on the CS Keep market. From these findings, we can conclude that the AS-AA codeshare
discontinuation is associated with average price decreases among the airlines on the CS Exit
markets.
Table 3. 8 The Codeshare Exit Effects Results (Definition B)
All Airlines AS - AA
(1)
OLS
(2)
FE
(3)
FE
(4)
OLS
(5)
FE
(6)
FE
Unbalanced Panel
CS Exit -0.110*** -0.118***
(0.0141) (0.0216)
Post 0.0241 0.0467*
(0.0164) (0.0255)
CS Exit ×Post -0.0222 -0.0211 -0.0099 -0.0560* -0.0423* -0.0358
(0.0207) (0.0136) (0.0150) (0.0316) (0.0243) (0.0269)
No. of Markets 279 279 279 279 279 279
No. of Carriers 1,303 1,303 1,303 464 464 464
Observations 4,640 4,640 4,640 1,676 1,676 1,676
Adjusted R
2 0.027 0.004 0.007 0.051 0.011 0.012
Balanced Panel
CS Exit -0.101*** -0.117***
(0.0137) (0.0213)
Post 0.0209 0.0482*
(0.0158) (0.0259)

96
CS Exit ×Post -0.0253 -0.0253* -0.0140 -0.0466 -0.0466* -0.0406
(0.0203) (0.0132) (0.0147) (0.0318) (0.0239) (0.0258)
No. of Markets 276 276 276 256 256 256
No. of Carriers 992 992 992 365 365 365
Observations 3,968 3,968 3,968 1,460 1,460 1,460
Adjusted R
2 0.028 0.005 0.008 0.051 0.010 0.010
Market-Carrier FE No Yes Yes No Yes Yes
Time FE No Yes Yes No Yes Yes
Controls No No Yes No No Yes
a The dependent variable is the natural log of passenger weighted average ticket fare of the ticketing carrier on the
market.
b Definition A defines codeshare markets asroutes where AS-AA codeshare products are operated during the all sample
pre periods and at least one quarter during the post-remedy period. Definition B defines codeshare markets as routes
where AS-AA codeshare products are operated at least one quarter during the pre and post periods, respectively.
c Standard error are in parentheses and clustered at the market-airline level.
d *** : 1% significance, ** : 5% significance, * : 10% significance.
Table 3.8 shows the results produced when we include the markets in which AS-AA
codeshare products were included in at least one pre-remedy period on the relevant market, i.e.,
codeshare market Definition B. Similar to the results shown in Table 3.7, the OLS results are
included for comparison, with the first three columns corresponding to the sample in which all
airlines are included, while the last three columns are the results from the sample focused only on
Alaska and American observations. Even though many of the coefficients of interest lost their
significance compared to the results in Table 3.7, the coefficients are all negative, suggesting that
no anti-competitive effects resulted from the codeshare discontinuation. Comparing the results
from the alternative definitions, we can conclude that, at a minimum, the codeshare exit produced
no anti-competitive effects, and is in fact associated with relative price decreases, especially for
the AS and AA products on the CS Exit markets.
While these results support the pro-competitive effects of codeshare discontinuation, a few
critical issues must be addressed. The first such issue is potential measurement error in the
treatment indicator. As stated in Section 3.5.2.1, a misclassified binary regressor is likely to cause

97
estimators to be biased. In the current context, misclassification is mainly due to the fact that the
treatment indicator is classified based on the presence of the relevant products in the data, where
the data are a 10% sample. As a result, some of the markets classified as CS Exit markets might
actually be markets in which AS-AA continued their codeshare products, whereas the real
codeshare exit markets cannot be misclassified as the CS Keep markets. If some of the CS Exit
markets were misclassified, the estimators are likely to suffer from attenuation bias, which means
that the actual effects might have been stronger.
The other potential concern regarding these results is that the codeshare exit effects might
be contaminated by the merger effects. Because the AS-AA codeshare discontinuation was closely
related to the AS-VX merger, the merger effects must be controlled for when examining the
codeshare exit effects. Even though it is difficult to clearly disentangle the two effects, we did
examine whether Virgin (VX) operated in the codeshare relevant markets. If Virgin had offered its
products in these markets, the merger might have had significant effects in the markets. Figure 3.7
shows how many markets were served by Virgin among the AS-AA codeshare relevant markets
from 2014Q1 to 2016Q3. As shown in the figure, the majority of the relevant markets were not
served by Virgin America. Although there might have been some impact from the change of the
entire network due to the merger, the relatively small presence of Virgin America in the relevant
market suggests that the codeshare exit effects are less likely to be sensitive to the merger effects,
if any exist. In fact, the results in Section 3.6 also support the conclusion that the AS-VX merger
generally did not have strong price effects.

98
Figure 3. 7 VX Presence on AS-AA Codeshare Routes
3.7.3 Parallel Trends Assumption
Causal interpretations of two-way fixed effect DID estimates require parallel trends and
constant treatment effects assumptions. One common method for arguing a parallel trends
assumption is to conduct pre-trends test, as in the event study literature (Autor 2003; Sun and
Abraham 2021). Even though the absence of pre-trends does not fully exclude the possibility of
post-trends that may exist without the intervention, we estimate the effects under the event study
approach to determine whether the treatment group and the control group had significantly
different trends prior to the remedy. Specifically, we estimate the following equation:

99
where 𝐷𝑡
is a dummy variable for each time. The last pre-remedy period is excluded as a reference
group. In other words, 𝛽−2
, 𝛽0
, and 𝛽+1
represent relative differences between the airlines on the
CS Exit markets and the airlines on the CS Keep markets, compared to their difference at the last
pre-remedy period.
Table 3. 9 Parallel Trends Assumption Tests
All Airlines AS-AA
(1) (2) (3) (4)
FE Event Study FE Event Study
Definition A
CS Exit ×Post -0.0421* -0.0897**
(0.0238) (0.0434)
Pre -2 -0.0287 -0.0370
(0.0237) (0.0409)
Post +0 -0.0426 -0.0730
(0.0283) (0.0514)
Post +1 -0.0699** -0.147**
(0.0332) (0.0651)
No. of Markets 108 108
No. of Carriers 537 189
Observations 1,960 719
Adjusted R
2 0.005 0.006 0.019 0.021
Definition B
CS Exit ×Post -0.0099 -0.0358
(0.0150) (0.0269)
Pre -2 -0.0166 -0.0190
(0.0156) (0.0261)
Post +0 0.00408 -0.0282
(0.0180) (0.0331)
Post +1 -0.0414** -0.0634
(0.0205) (0.0390)
No. of Markets 279 279

100
No. of Carriers 1,303 464
Observations
Adjusted R
2 0.007
4,640
0.009 0.012
1,676
0.013
a Market-Carrier fixed effects, Time fixed effects, and Controls are included in all specifications.
b Definition A defines codeshare markets as routes where AS-AA codeshare products are operated at least one
quarter during the pre and post periods, re- spectively. Definition B defines codeshare markets as routes where ASAA codeshare products are operated during the all sample periods (two quarters for each pre and post).
c Standard error are in parentheses and clustered at the market-airline level.
d *** : 1% significance, ** : 5% significance, * : 10% significance.
The results of this analysis are reported in Table 3.9. The upper part of the table reports the
results from the Definition A sample, while the lower part of the table reports the results from the
Definition B sample. Columns (1) and (3) are from Columns (3) and (6) of Tables 3.7 and 3.8. The
insignificant coefficients on 𝑃𝑟𝑒−2
show that there were no significant pre-merger trends in the
pre-remedy period.
3.8 Demand
In this section, we present an air travel demand model in order to investigate the effect of
eliminating AS-AA codeshare products. Similar to Verboven (1996), we use a version of the twolevel nested logit model. The model is also similar to the demand model employed in Gayle and
Brown (2014), which examined the demand-increasing effects of virtual codeshare.
3.8.1 Model
In the two-level nested logit model, products are first divided into G + 1 groups, g = 0,
1, ..., G, with each group g further divided into 𝐻𝑔 subgroups, h = 1, ..., 𝐻𝑔. In our model, the first
nest is an identity of the ticketing carrier, while the second nest is whether a product is nonstop or
onestop. The first nest G also include an outside option, g = 0. The outside option includes options
not to travel or to travel using transportation other than airlines. In other words, a passenger first
chooses whether to air travel with a specific airline or not to air travel, then if the passenger decides

101
to travel with the specific airline, the passenger chooses whether to use a direct/nonstop flight or a
connecting/onestop flight. This two-level nested logit model allows preferences to be correlated
among products in a group as well as a subgroup. The nesting structure is illustrated in Figure 3.8.
Figure 3. 8 The Nesting Structure 1
Following Verboven (1996), a passenger i’s utility of choosing a product j on an airportpair m at time t is:
where 𝛿𝑗𝑚𝑡 is a mean-utility that is equal for all passengers and 𝜈𝑖𝑗𝑚𝑡 is an individual specific part.
𝜈𝑖𝑗𝑚𝑡 is specified as 𝜈𝑖𝑗𝑚𝑡 = 𝜀𝑖𝑚𝑡 + 𝜀𝑖𝑔𝑚𝑡 + (1 −𝜎𝑔) 𝜀𝑖ℎ𝑔𝑚𝑡 t + (1 −𝜎ℎ) 𝜀𝑖𝑗𝑚𝑡 , where 𝜎ℎ represents
the correlation of the passenger’s utility across products in the same subgroup, while 𝜎𝑔 captures
the correlation of the utility across products in the same group. Because products in the same
subgroup should be considered closer substitutes compared with ones in other groups, a restriction
of 0 ≤ 𝜎𝑔 ≤ 𝜎ℎ < 1 should hold for this model to be valid. 𝜀𝑖𝑔𝑚𝑡 , 𝜀𝑖ℎ𝑔𝑚𝑡 , and 𝜀𝑖𝑗𝑚𝑡 have unique
Air Travel
Outside
Option AA AS ... Airlinem
nonstop onestop nonstop onestop nonstop onestop

102
distribution, such that 𝜀𝑖𝑔𝑚𝑡 , (1 −𝜎𝑔) 𝜀𝑖ℎ𝑔𝑚𝑡 t + (1 −𝜎ℎ) 𝜀𝑖𝑗𝑚𝑡 and 𝜀𝑖𝑔𝑚𝑡 + (1 −𝜎𝑔) 𝜀𝑖ℎ𝑔𝑚𝑡 t + (1
−𝜎ℎ) 𝜀𝑖𝑗𝑚𝑡 are extreme value random variables.
𝛿𝑗𝑚𝑡 is the mean level of utility of choosing product j on the airport-pair m at time t. It is a
function of the characteristics of product j, such as price, and observed non-price characteristics.
Specifically, 𝛿𝑗𝑚𝑡 is specified as follows:
where 𝑋𝑗𝑚𝑡 is a vector of non-price observed product characteristics. We include dummy variables
for nonstop flights, inconvenience of the product, and an indicator of traditional codeshare as
observed product characteristics. A nonstop dummy takes a value of 1 if the flight is a direct flight.
The coefficient on this variable is expected to be positive because passengers tend to prefer direct
flights over flights with intermediate stops. As suggested in Gayle (2013), onestop flights might
have different effects on utility according to their itinerary distance. In order to account for this
possibility, an inconvenience variable is calculated as the itinerary distance divided by the nonstop
miles of the route. The traditional codeshare dummy takes a value of 1 if the product is a traditional
codeshare of any airline (not just the AA/AS or AS/AA traditional one). 𝑝𝑗𝑚𝑡 is the passenger
weighted average price of the product. CS_Exit takes a value of 1 if the market was served by ASAA traditional codeshare in the pre-merger period but not in the post period. It takes a value of 0
if, in the market, AS-AA codeshare products were operated both in the pre-merger and post-remedy
periods. Post is a dummy variable which takes a value of 1 for the post period and 0 for the pre
period. 𝜑𝑎 , 𝛾𝑚 , and 𝜆𝑡
are ticketing carrier fixed effects, market fixed effects, and time fixed
effects, respectively. Even though the DB1B data contain many detailed information of tickets, it
does not include information about some of ticket-specific restrictions information, which would

103
have affected passengers’ choices. For example, whether the ticket is refundable or not is not
included in the data. In order to account for these kinds of unobserved product quality measures,
𝜉𝑗𝑚𝑡 is included in the specification.
𝛽 is a vector of parameters representing marginal utilities associated with the observed
product characteristics. 𝛼 captures the marginal utility of the price. 𝜃1
represents whether there are
any differences in mean utility between the products in the CS Exit markets and the products in the
markets in which AS-AA codeshare products were operated in both the pre and post periods (the
CS Keep markets). 𝜃2 captures whether there are any changes in mean utility over the pre and post
periods for the products on the CS Keep markets. 𝜃3 captures whether the change in mean utility
over time differs across the CS Exit markets and CS Keep markets; this represents the demand
effects of the AS-AA codeshare discontinuation.
3.8.2 Estimation
The two-level nested logit model explained in the previous subsection can be estimated by
the following equation:
where 𝑆𝑗
is the unconditional probability of product j being selected and 𝑆0
is the probability of an
outside option being chosen. 𝑆𝑗|ℎ𝑔 is the probability of product j being chosen given the selection
of its subgroup. Similarly, 𝑆ℎ|𝑔 is the probability of the subgroup h being selected given the
selection of its group g. These values are used as natural log forms. Each probability is calculated
as the observed market shares. Following Berry and Jia (2010), the market size is calculated as the

104
geometric mean of the origin and destination cities’ populations, and each observed share is
calculated as the total number of passengers divided by the market size.
Since 𝑝𝑗𝑚𝑡 , 𝑙𝑛 𝑆𝑗|ℎ𝑔 , and 𝑙𝑛 𝑆ℎ|𝑔 are likely to be correlated with unobserved product
characteristics 𝜉𝑗𝑚𝑡 , instrumental variables should be used to estimate the parameters. Proper
instrumental variables include cost shifters, which directly affect supply (and thus markup) but not
demand. The characteristics of other products can be also used as instruments. Instrumental
variables should provide variations to identify relative shares of the product type (i.e., nonstop or
onestop) as well as the within group share of each product. The instrumental variables we included
are (1) flown miles, (2) jetfuel price as interacted with the legacy airline dummy,24 (3) the number
of other products offered by the airline on the market, (4) market-airline level average of
inconvenience, and (5) market-airline level average number of stops on the market.25 Using these
instrumental variables, the model is estimated with 2SLS.
3.8.3 Results
The results of this analysis are provided in Columns (1)-(4) of Table 3.10. The first two
columns present the estimation without using the instrumental variables, while Columns (3) and
(4) present the estimation using the instrumental variables. Overall, the results for OLS and 2SLS
are similar even though the exogeneity of the endogenous variables are strongly rejected. For all
specifications, ticketing carrier fixed effects, time fixed effects, and market fixed effects are
included; however, the results for these fixed effects are not reported.
24 Changes in jetfuel price might affect lagacy airlines and low-cost airline differently, because these two types of airlines have
different cost structures.
25 We consider IVs in Gayle (2013), Gayle and Brown (2014), and Shen (2017).

105
The coefficients on the observed product characteristics are as expected. Higher fares are
associated with lower utility. Direct flights are associated with higher utility compared to
connecting flights. Products with higher inconvenience measures are associated with lower utility.
From these results, we can confirm that passengers prefer shorter trips. Traditional codeshare
products are valued less by passengers compared to pure online products. As virtual codeshare
products might be inferior substitutes for pure online products (Ito and Lee, 2007), traditional
codeshare products are likely to be considered as inferior substitutes.
In terms of the nesting parameters, passengers typically recognize products with the same
ticketing carriers as close substitutes with only one nest, as shown in Column (3). The products
with the same number of stops given the ticketing carriers are considered as close substitutes with
two nests, as shown in Column (4). Though the coefficient on 𝑙𝑛 𝑆ℎ|𝑔 is not significant, the
restriction of 0 ≤ 𝜎𝑔 ≤ 𝜎ℎ < 1 is partially satisfied. As the coefficients on 𝑙𝑛 𝑆𝑗|ℎ𝑔 are statistically
significant, we can argue that there are correlations among products in a subgroup. In other words,
passengers perceive products in the same group as closer substitutes compared to ones in other
groups. Given that 𝜎𝑔 is not significant, if we consider it to be 0, the model collapses to a one-level
nested logit with the number of stops (i.e., direct or indirect flight) as nests.
The significantly negative coefficients on CS Exit × Post support the conclusion that the
AS-AA codeshare discontinuation negatively impacted passengers’ utility. In other words, while
the codeshare remedy decreased the average fares on the CS Exit markets, it also might have
decreased utility for those goods. As a result, it is not straightforward to derive welfare impact of
the codeshare cessation.

106
A nested logit model can be sensitive to the nesting structure. Because 𝑙𝑛 𝑆ℎ|𝑔 is
insignificant in Column (4), we conduct a demand estimation with an alternative nesting structure.
In this structure, the first nesting group is composed of whether to take a nonstop flight or an
onestop flight (or choose an outside option), and the subgroup is composed of each ticketing airline.
This nesting structure is illustrated in Figure 3.9. The results are reported in Columns (5)-(8), Table
3.10. The results are generally similar to those of the first nesting structure, with the exception of
the significance of 𝜎𝑔. Note that in this nesting structure, 𝜎𝑔 captures preference correlation among
products with the same number of stops. The results for the main indicator variables are consistent
with the alternative nesting structure, which corroborate the demand-decreasing effects of the
codeshare discontinuation.
Figure 3. 9 The Nesting Structure 2
Air Travel
Outside
Option
Nonstop Onestop
... ...

107

108
3.9 Conclusion
This study examines the effects of Alaska-American (AS-AA) codeshare discontinuation
on prices and demand, as well as the relative price changes associated with the merger between
Alaska (AS) and Virgin America (VX). The results from the market level fare change comparisons
suggest that the AS-VX merger was not associated with any significant impact on prices. This
might be due to their relatively small number of overlapping routes compared to mergers between
larger airlines. It might also be due to the remedies imposed by the governmental authorities to
reduce potential anti-competitive effects from the merger. However, additional research is required
to precisely determine the underlying mechanism.
In terms of the codeshare exit effects, our findings show that the discontinuation of AS-AA
codeshare products is associated with price decreases, since relative price decreases are detected
on the routes for which AS-AA discontinued their codeshare products. While these results may
indicate that the codeshare discontinuation was overall beneficial to consumers in terms of lower
prices, the relative lower prices might be associated with a decrease in demand. In order to address
this issue, we estimate an air travel demand model, concluding that the codeshare discontinuation
did negatively impact consumers’ utility. As a result, welfare implications cannot be directly
derived.
Applying the results of this study to other mergers or industries might not be
straightforward, because each case has important contexts for the purpose of the analysis.
Nevertheless, this study does have important policy implications, as the remedies were determined
to have helped to decrease prices. Given that partnerships between major airlines are crucial

109
strategic choices for the airlines as well as sources of anti-competitive concerns from the DOJ, the
findings of this study can provide useful insight for relevant issues.

110
References
Armstrong, M. (2006). Competition in two‐sided markets. The RAND journal of economics, 37(3),
668-691.
Armstrong, M., & Wright, J. (2007). Two-sided markets, competitive bottlenecks and exclusive
contracts. Economic Theory, 32, 353-380.
Autor, D. H. (2003). Outsourcing at will: The contribution of unjust dismissal doctrine to the
growth of employment outsourcing. Journal of labor economics, 21(1), 1-42.
Bakos, Y., & Halaburda, H. (2020). Platform competition with multihoming on both sides:
Subsidize or not?. Management Science, 66(12), 5599-5607.
Berry, S. T. (1994). Estimating discrete-choice models of product differentiation. The RAND
Journal of Economics, 242-262.
Berry, S., & Jia, P. (2010). Tracing the woes: An empirical analysis of the airline
industry. American Economic Journal: Microeconomics, 2(3), 1-43.
Berry, S., Levinsohn, J., and Pakes, A. (1995). Automobile prices in market equilibrium.
Econometrica, 63(4), 841–890.
Carlton, D., Israel, M., MacSwain, I., & Orlov, E. (2019). Are legacy airline mergers pro-or anticompetitive? Evidence from recent US airline mergers. International Journal of Industrial
Organization, 62, 58-95.
City and County of San Francisco. (2020, April 10). Mayor London Breed announces delivery fee
cap to support San Francisco restaurants during COVID-19 pandemic. San Francisco Mayor's
Office. https://sfmayor.org/article/mayor-london-breed-announces-delivery-fee-cap-support-sanfrancisco-restaurants-during-covid
Das, S. (2019). Effect of merger on market price and product quality: American and US
airways. Review of Industrial Organization, 55(3), 339-374.
Delbaum, J., & Skinner, J. (2020, March 5). Merger remedies in the US: An overview of the leading
cases. e-Competitions US Merger Remedies, Art. N° 922390.
Fan, Y. (2013). Ownership consolidation and product characteristics: A study of the US daily
newspaper market. American Economic Review, 103(5), 1598-1628.
Gayle, P. G. (2008). An empirical analysis of the competitive effects of the
delta/continental/northwest code-share alliance. The Journal of Law and Economics, 51(4), 743-
766.

111
Gayle, P. G. (2013). On the efficiency of codeshare contracts between airlines: is double
marginalization eliminated?. American Economic Journal: Microeconomics, 5(4), 244-273.
Gayle, P. G., & Brown, D. (2014). Airline strategic alliances in overlapping markets: Should
policymakers be concerned?. Economics of Transportation, 3(4), 243-256.
Gentzkow, M. (2007). Valuing new goods in a model with complementarity: Online newspapers.
American Economic Review, 97(3), 713-744.
Ito, H., & Lee, D. (2007). Domestic code sharing, alliances, and airfares in the US airline
industry. The Journal of Law and Economics, 50(2), 355-380.
Ito, H., & Lee, D. (2007). Domestic code sharing, alliances, and airfares in the US airline
industry. The Journal of Law and Economics, 50(2), 355-380.
Kwoka, J., & Shumilkina, E. (2010). The price effect of eliminating potential competition:
Evidence from an airline merger. The journal of industrial economics, 58(4), 767-793.
Li, Z., & Wang, G. (2024). Regulating Powerful Platforms: Evidence from Commission Fee Caps.
Information Systems Research.
Luo, D. (2014). The price effects of the Delta/Northwest airline merger. Review of Industrial
Organization, 44, 27-48.
Perri, J. (2023, March 14). Which company is winning the restaurant food delivery war?
Bloomberg Second Measure. https://secondmeasure.com/datapoints/food-delivery-servicesgrubhub-uber-eats-doordash-postmates
Peters, C. (2006). Evaluating the performance of merger simulation: Evidence from the US airline
industry. The Journal of law and economics, 49(2), 627-649.
Rana, P. (2021, April 27). DoorDash allows restaurants to choose commissions in post-pandemic
future. The Wall Street Journal. https://www.wsj.com/articles/doordash-allows-restaurants-tochoose-commissions-in-post-pandemic-future-11619517614?mod=Searchresults_pos3&page=1
Rochet, J. C., & Tirole, J. (2003). Platform competition in two-sided markets. Journal of the
european economic association, 1(4), 990-1029.
Rysman, M. (2004). Competition between networks: A study of the market for yellow pages. The
Review of Economic Studies, 71(2), 483-512.
Shen, C. (2017). The effects of major US domestic airline code sharing and profit sharing
rule. Journal of Economics & Management Strategy, 26(3), 590-609.
Song, M. (2021). Estimating platform market power in two-sided markets with an application to
magazine advertising. American Economic Journal: Microeconomics, 13(2), 35-67.

112
Sullivan, M. (2024). Price controls in a multi-sided market. Journal of Political Economy.
Manuscript submitted for publication.
Sun, L., & Abraham, S. (2021). Estimating dynamic treatment effects in event studies with
heterogeneous treatment effects. Journal of econometrics, 225(2), 175-199.
Tan, G., & Zhou, J. (2021). The effects of competition and entry in multi-sided markets. The
Review of Economic Studies, 88(2), 1002-1030.
Teh, T. H., Liu, C., Wright, J., & Zhou, J. (2023). Multihoming and oligopolistic platform
competition. American Economic Journal: Microeconomics, 15(4), 68-113.
U.S. Department of Justice, Antitrust Division. (2011). Antitrust division policy guide to merger
remedies. Washington, D.C.: U.S. Department of Justice. https://www.justice.gov/atr/antitrustdivision-policy-guide-merger-remedies-june-2011
U.S. Department of Justice. (2013). Justice Department requires US Airways and American
Airlines to divest facilities at key airports. https://www.justice.gov/opa/pr/justice-departmentrequires-us-airways-and-american-airlines-divest-facilities-seven-key
Verboven, F. L. (1994). International price discrimination in the European car market: An
econometric model of oligopoly behavior with product differentiation.
Warerkar, T. (2021, June 15). Food app delivery commission in S.F. capped at 15% to help
restaurants. San Francisco Chronicle.https://www.sfchronicle.com/food/restaurants/article/Foodapp-delivery-commission-in-S-F-capped-at-16266468.php

113
Appendices
Appendix A. Overview of Commission Cap Implementation
Table A. 1 Implementation of Commission Caps by City (Sorted by Start Date)
County City Start Date County City Start Date
San Francisco San Francisco 2020-04 Contra Costa Danville 2020-12
Alameda Berkeley 2020-07 Santa Clara San Jose 2020-12
Alameda San Leandro 2020-07 Santa Clara Sunnyvale 2020-12
Marin Mill Valley 2020-07 Santa Clara Santa Clara 2020-12
Alameda Fremont 2020-08 Santa Clara Mountain View 2020-12
Alameda Livermore 2020-08 Santa Clara Palo Alto 2020-12
Alameda Oakland 2020-08 Santa Clara Cupertino 2020-12
San Mateo South San Francisco 2020-08 Santa Clara Gilroy 2020-12
Alameda Alameda 2020-09 Santa Clara Milpitas 2020-12
Alameda Hayward 2020-09 Santa Clara Morgan Hill 2020-12
Contra Costa Walnut Creek 2020-10 Santa Clara Campbell 2020-12
San Mateo Millbrae 2020-10 Santa Clara Los Altos 2020-12
Contra Costa Lafayette 2020-11 Santa Clara Monte Sereno 2020-12
San Mateo Atherton 2020-11 Santa Clara Saratoga 2020-12
San Mateo Belmont 2020-11 Contra Costa Antioch 2021-02
San Mateo Brisbane 2020-11 Contra Costa Brentwood 2021-02
San Mateo Burlingame 2020-11 Contra Costa Clayton 2021-02
San Mateo Colma 2020-11 Contra Costa Concord 2021-02
San Mateo Daly City 2020-11 Contra Costa El Cerrito 2021-02
San Mateo East Palo Alto 2020-11 Contra Costa Hercules 2021-02
San Mateo Foster City 2020-11 Contra Costa Martinez 2021-02
San Mateo Half Moon Bay 2020-11 Contra Costa Moraga 2021-02
San Mateo Hillsborough 2020-11 Contra Costa Oakley 2021-02
San Mateo Menlo Park 2020-11 Contra Costa Orinda 2021-02
San Mateo Pacifica 2020-11 Contra Costa Pinole 2021-02
San Mateo Portola Valley 2020-11 Contra Costa Pittsburg 2021-02
San Mateo Redwood City 2020-11 Contra Costa Pleasant Hill 2021-02
San Mateo San Bruno 2020-11 Contra Costa Richmond 2021-02
San Mateo San Carlos 2020-11 Contra Costa San Pablo 2021-02
San Mateo San Mateo 2020-11 Contra Costa San Ramon 2021-02
San Mateo Woodside 2020-11

114
Appendix B. Codeshare Remedy Detail
When the Department of Justice (DOJ) approved the merger between the Alaska Airlines
and Virgin America, the DOJ proposed the modification of AS-AA codeshare conduct through the
final judgment. In the final judgment, the DOJ clarified the market criteria ordering that the
merging parties and American Airlines should not market codeshare products with one other.
The first market criterion was whether Virgin America and American Airlines were competing head-to-head in a market as of December 6, 2016. The second market criterion was whether
Alaska Airlines and American Airlines were competing in the same city market. These two market
criteria prohibited AS-AA codeshare conducts on the Virgin/American overlap route and the
Alaska/American overlap route. In the final judgment, the DOJ defined 31 origin-destination pairs
as “Virgin/American Domestic U.S. Overlap Routes” (Table B.1), and 21 origin-destination pairs
as “Alaska/American Domestic U.S. Overlap Routes” as of December 2016. (Table B.2) The third
market criterion was whether a market originated or terminated at each key airport. The fourth
criterion was indicating the markets between LAX and any key airports.

115
Table B. 1 Virgin/American Domestic U.S. Overlap Routes
Non-Directional Origin - Destination Pairs
Origin Destination
Boston Logan International Airport BOS LAX Los Angeles International Airport
Chicago O’Hare International Airport ORD LAX Los Angeles International Airport
Dallas Love Field Airport DAL LAX Los Angeles International Airport
Dallas/Fort Worth International Airport DFW LAX Los Angeles International Airport
Fort Lauderdale – Hollywood International Airport FLL LAX Los Angeles International Airport
Los Angeles International Airport LAX MIA Miami International Airport
Honolulu International Airport HNL LAX Los Angeles International Airport
McCarran International Airport LAS LAX Los Angeles International Airport
Los Angeles International Airport LAX IAD Washington Dulles International Airport
Los Angeles International Airport LAX DCA Ronald Regan Washington National Airport
Los Angeles International Airport LAX JFK John F. Kennedy International Airport
Los Angeles International Airport LAX EWR Newark Liberty International Airport
Los Angeles International Airport LAX MCO Orlando International Airport
Los Angeles International Airport LAX SEA Seattle – Tacoma International Airport
Dallas Love Field Airport DAL SFO San Francisco International Airport
Dallas/Fort Worth International Airport DFW SFO San Francisco International Airport
Fort Lauderdale – Hollywood International Airport FLL SFO San Francisco International Airport
Miami International Airport MIA SFO San Francisco International Airport
John F. Kennedy International Airport JFK SFO San Francisco International Airport
Los Angeles International Airport LAX SFO San Francisco International Airport
Chicago O’Hare International Airport ORD SFO San Francisco International Airport
Dallas Love Field Airport DAL DCA Ronald Regan Washington National Airport
Dallas/Fort Worth International Airport DFW DCA Ronald Regan Washington National Airport
Dallas Love Field Airport DAL LGA LaGuardia Airport
Dallas/Fort Worth International Airport DFW LGA LaGuardia Airport
Dallas Love Field Airport DAL LAS McCarran International Airport
Dallas/Fort Worth International Airport DFW LAS McCarran International Airport
Fort Lauderdale – Hollywood International Airport FLL JFK John F. Kennedy International Airport
Miami International Airport MIA JFK John F. Kennedy International Airport
Los Angeles International Airport LAX OGG Kahului Airport
McCarran International Airport LAS JFK John F. Kennedy International Airport

116
Table B. 2 Alaska/American Domestic U.S. Overlap Routes
Non-Directional Origin - Destination Pairs
Origin Destination
Ted Stevens Anchorage International Airport ANC LAX Los Angeles International Airport
Ted Stevens Anchorage International Airport ANC PHX Phoenix Sky Harbor International Airport
Chicago O’Hare International Airport ORD PDX Portland International Airport
Chicago O’Hare International Airport ORD SEA Seattle – Tacoma International Airport
Dallas/Fort Worth International Airport DFW PDX Portland International Airport
Dallas/Fort Worth International Airport DFW SEA Seattle – Tacoma International Airport
Los Angeles International Airport LAX PDX Portland International Airport
Los Angeles International Airport LAX SLC Salt Lake City International Airport
Los Angeles International Airport LAX SEA Seattle – Tacoma International Airport
John F. Kennedy International Airport JFK SEA Seattle – Tacoma International Airport
Philadelphia International Airport PHL SEA Seattle – Tacoma International Airport
Phoenix Sky Harbor International Airport PHX SEA Seattle – Tacoma International Airport
Phoenix Sky Harbor International Airport PHX PDX Portland International Airport
Ronald Regan Washington National Airport DCA LAX Los Angeles International Airport
Baltimore – Washington International Airport BWI LAX Los Angeles International Airport
Newark Liberty International Airport EWR SEA Seattle – Tacoma International Airport
John F. Kennedy International Airport JFK SAN San Diego International Airport
Newark Liberty International Airport EWR SAN San Diego International Airport
Miami International Airport MIA SEA Seattle – Tacoma International Airport
Fort Lauderdale–Hollywood International Airport FLL SEA Seattle – Tacoma International Airport
Washington Dulles International Airport IAD LAX Los Angeles International Airport

Abstract (if available)

Abstract During the COVID-19 pandemic, the online food delivery industry experienced significant growth, prompting platforms to introduce new commission plans in response to regulatory changes. This study investigates the platform choice problems faced by restaurants and consumers within this industry. Using a multinomial logit model that accounts for restaurant, platform, and market characteristics, it analyzes how these factors influence both restaurants' and consumers' choices between the two largest platforms. Our findings reveal that restaurants perceive these platforms as complementary, leading to increased multihoming—where restaurants engage with multiple platforms simultaneously. Additionally, consumers demonstrate a preference for restaurants that engage in multihoming. These insights have significant implications for platform competition strategies and future regulatory decisions in the dynamic food delivery industry.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on quality screening in two-sided markets

PDF

Essays on competition and antitrust issues in the airline industry

PDF

Surveying stakeholder views on the current good laboratory practices (GLP) regulations and their proposed revisions

PDF

Essays on competition between multiproduct firms

PDF

Theoretical and data-driven analysis of two-sided markets and strategic behavior in modern economic systems

PDF

Three essays on agent’s strategic behavior on online trading market

PDF

Essays on the unintended consequences of digital platform designs

PDF

Empirical essays on industrial organization

PDF

Essays on revenue management with choice modeling

PDF

Quality investment and advertising: an empirical analysis of the auto industry

PDF

Three essays on industrial organization

PDF

Essays on digital platforms

PDF

Competing across and within platforms: antecedents and consequences of market entries by mobile app developers

PDF

Domestic and international tensions in artificial intelligence policy

PDF

Behavioral approaches to industrial organization

PDF

Essays on development and health economics: social media and education policy

PDF

Reimagining professional learning for early childhood educators of color

PDF

The essays on the optimal information revelation, and multi-stop shopping

PDF

Price competition among firms with a secondary source of revenue

PDF

Competitions in two-sided markets

Asset Metadata

Creator Park, Rihyun (author)

Core Title Essays on competition and strategy within platform industries

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Degree Conferral Date 2024-12

Publication Date 12/04/2024

Defense Date 12/03/2024

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag commission,commission cap,food delivery industry,multi-home,multi-sided platform,OAI-PMH Harvest,platform choice,platform competition,platform selection,tiered commission

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Ridder, Geert (committee chair), Proserpio, Davide (committee member), Tan, Guofu (committee member)

Creator Email rihyunpa@usc.edu,rihyunpark.biz@gmail.com

Unique identifier UC11399E8JB

Identifier etd-ParkRihyun-13670.pdf (filename)

Legacy Identifier etd-ParkRihyun-13670

Document Type Dissertation

Format theses (aat)

Rights Park, Rihyun

Internet Media Type application/pdf

Type texts

Source 20241205-usctheses-batch-1226 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu