[PDF] The Impact of Online Food Delivery Services on - Stanford University PDF Collison.pdf

Finally, conducting a year-over-year analysis shows that the additional sales provided by online food delivery services are decreasing over time This suggests that there is incremental cannibalization of brick-and-mortar restaurant sales as exposure to online food delivery services increases

The main advantage of online food ordering is that it greatly simplifies the ordering process for both the customer and the restaurant, and also simplifies the load at the end of the customer as the whole process of ordering is automated Food distribution can be defined

[PDF] Influence Of Online Food Delivery Apps On The Operations Of The

Items 12 - 24 · These online food delivery services are boosting the option of choosing meals from a wide variety of restaurants with a single tap of our smartphones

[PDF] What factors motivate consumers online food ordering behavior on

understand consumers' motivations for ordering food online, and assist restaurant the effect of review volume and overall ratings to identify factors driving food

[PDF] A Study on Impact of Online Food delivery app on - IJRAR

Offering online ordering lets your guests place an order more conveniently Without feeling pressure to wrap up their order, customers are more inclined to explore

[PDF] Millennial Experience with Online Food Home Delivery: A Lesson

To examine millennial satisfaction towards online food delivery services, in- er, the effect of e-service quality and food quality on customer satisfaction is not

[PDF] effect of public debt on economic growth in lesotho

[PDF] effect of speed on posture

[PDF] effect of water pollution on human life

[PDF] effective academic writing pdf

[PDF] effective address and physical address in 8086

[PDF] effective awk programming 4th edition pdf

[PDF] effective c pdf github

[PDF] effective c++ pdf github

[PDF] effective c++ scott meyers

[PDF] effective c++ third edition pdf

[PDF] effective c: an introduction to professional c programming pdf

[PDF] effective classroom

[PDF] effective conclusion

[PDF] effective decision making pdf

[PDF] effective english writing skills pdf

The Impact of Online Food Delivery Services on Restaurant Sales

Jack Collison

Department of Economics, Stanford University

Advised by Professor Liran Einav

Spring, 2020

Abstract

The rapid growth of online food delivery services has disrupted the traditionally oine restaurant industry. This study presents empirical evidence on the crowding-out eects and market expansions induced by the staggered entry of online food delivery services. Dierence-in-dierences methodology reveals that 30-50 cents of every dollar spent on online food delivery services are incremental, while the rest is drawn away from brick- and-mortar sales. These ndings are statistically signicant at the zip code-level and are heterogeneous across dierent types of consumption, suggesting that convenience and pre-existing spending habits drive the level of substitution. Conducting analyses on a year-by-year basis suggests that there is an increasing level of cannibalization of brick-and-mortar restaurant sales. Back-of-the-envelope calculations show an increase in restaurants' revenues but a decrease in protability. Keywords:E-commerce, market expansion, sales cannibalization E-mail address:jack10@stanford.edu. I am extremely grateful to my advisor, Professor Liran Einav,

for his excellent guidance, input, and advice. I would like to thank Honors program director Marcelo Clerici-

Arias and Sebastian Otero for assisting me in the process of developing a thesis. I would also like to thank

Professor Pete Klenow, Ben Klopack, Toren Fronsdal, and the other faculty and graduate students who have

given me feedback and advice. Finally, I would like to thank Suresh Vaidyanathan, Larry Levin, and all

other members of Visa who have supported me. All errors are my own. 1 \Online delivery is surging, and eating in is the new dining out. Online commerce reduced trac in brick-and-mortar stores, which this year are closing at a record-setting pace... Meal-delivery companies are a symbol of what might be the most powerful force in business today: convenience maximalism."Derek Thompson,The Atlantic(2019)

1 Introduction

The meteoric growth of e-commerce makes it an ever-important area to study. Even in traditional industries, well-established oine rms have adopted new online sales channels that aim to increase their revenue. This has led to the hybridization of strictly brick-and- mortar stores, which now operate both online and oine. In fact, by 2012, more than 80 percent of U.S. retailers sold merchandise through both online and oine channels (Wang, Song, and Yang 2012). One of the most prolic applications of this recent hybridization is in the restaurant industry, with the emergence of online food delivery services. Delivery transactions made up six percent of total US restaurant sales in 2017 and are estimated to reach 40 percent of all restaurant sales by 2020 (Morgan Stanley Research 2017).

1However,

the extent to which these online sales are incremental|causing overall restaurant sales to increase|or, alternatively, drawn away from brick-and-mortar sales, has not been quantied. Online food delivery is a prime example of e-commerce disrupting a traditional market. A ood of new food delivery rms has caused rapid growth in the total number of transactions and revenue for the nascent industry. Although online food delivery services provide extra channels for potential revenue, they also create the risk of cannibalization in which brick- and-mortar sales actually suer because consumers who purchase in-store have transitioned to mostly online purchasing behavior. The purpose of this study is to determine the eects that the entry of these rms|and subsequent hybridization|has had on restaurant sales1

The COVID-19 pandemic may further accelerate this trend because in-person dining has essentially been

shut down. 2 by quantifying the levels of substitution between in-person restaurant sales and online food delivery services. The staggered deployment and expansion of online food delivery services across time and space provides an opportunity to estimate these eects and how they vary across locations and over time. In order to do so, I use the universe of Visa Inc.'s individual-level credit and debit transactional data. Each observation in the main dataset is an individual transaction between a business (the merchant) and a consumer (a cardholder). On the merchant side, the name, zip code, and category of the business are recorded. On the cardholder side, only the anonymized card number is visible. Other variables of interest include the transaction amount and a record of how the purchase was conducted (a physical card swipe, online transaction, phone charge, etc.). The data encapsulates all businesses at which cardholders use their Visa cards. In order to maintain a sample that is useful for the analysis, the data is restricted to purchases that occurred at restaurants in the United States from 2014 through

2017. Cardholder location and family status are imputed based on transactional history.

The analysis explores heterogeneity along these dimensions, as well as other factors of time and space. The primary empirical specication is a standard dierences-in-dierences with a contin- uous treatment variable. I regress total dollars-per-card|brick-and-mortar and online food delivery service sales|spent at restaurants on dollars-per-card spent on online food delivery services with month and zip code xed eects in order to quantify the level cannibalization of restaurant sales. If the two are uncorrelated, this suggests that every dollar spent on online food delivery service is cannibalized; if the two are perfectly correlated, this suggests that online food delivery services are providing purely incremental sales to restaurants. More specically, the coecient on online sales shows the fraction of each dollar spent on online food delivery services that is new; the remainder is drawn away from brick-and-mortar sales. I nd that roughly half of each dollar spent on online food delivery services is new, whereas the remaining half is converted from in-person restaurant sales. This suggests that 3 online food delivery services provide modest incremental sales, but also draw away from traditional sales. Stratifying the regression by the time of transaction shows that while each dollar spent on online food delivery services during lunch, dinner, dierent seasons, and the work week provide some additional sales, online food delivery sales during the weekend are mostly cannibalized from brick-and-mortar sales. The dierences in these estimates suggest that convenience is an important factor in drawing consumers to online food delivery services. Considering time constraints during the week, the ease of online food delivery services may be driving the higher level of substitution between online and oine channels. Separating zip codes into quartiles of average monthly restaurant expenditure shows that the lower quartiles substitute only 10 percent every dollar spent on online food delivery services, whereas the higher quartiles substitute up to half of every dollar spent on online food delivery services. This suggests that online food delivery services are drawing in those who typically do not spend much at restaurants. The distinction of urban and rural regions shows that rural areas provide almost completely new restaurant sales via online food delivery services; the region of the U.S.|northern or southern states|has little eect on the levels of substitution. Finally, conducting a year-over-year analysis shows that the additional sales provided by online food delivery services are decreasing over time. This suggests that there is incremental cannibalization of brick-and-mortar restaurant sales as exposure to online food delivery services increases. Further, back-of-the-envelope calculations show that restaurants' revenues are increasing but their protability is decreasing. There are many descriptive studies that document the growth of online food delivery ser- vices and the characteristics of the customers that use them (Morgan Stanley Research 2017; Technomic Food Trends 2018; Wirth 2018; Zion, Spangler, and Hollmann 2018). However, these studies rely on self-reported qualitative surveys that were sent out to a few thou- sand individuals, which could potentially lead to selection bias, reporting inaccuracies, and attrition rates over time. The analysis presented in this study quanties cannibalization of traditional restaurant sales by online food delivery services using a large, representative 4 network of transactions. This not only contributes to a growing literature on substitution between online and oine sales channels by examining a new and growing industry, but also provides some of the rst empirical evidence on the impact of online food delivery services. Although this study focuses on online food delivery services, the results could potentially be extrapolated to dierent markets. Other traditional markets have many confounding factors such as a less transparent rm-entry pattern, which leads to diculty in quantifying the eects of e-commerce. Restaurants and online food delivery services, however, lend themselves well to studying these eects. While there might be other outside factors in dierent industries that impact the causal eects of e-commerce, the lessons derived here are likely similar elsewhere.

The rest of the paper is organized as follows. In

Section

2 and

Section

3 , I provide background on the evolution of online food delivery services and literature related to online- oine substitution.

Section

4 in troducesthe data, the construction of k eyv ariables,and relevant summary statistics.

Section

5 and

Section

6 describ ethe empirical strategy and results, respectively.

Section

7 concludes br ie y.

2 Background

Online food delivery services have been around for quite some time. Several chain restau- rants created websites to order take-out, but these services were limited to within the chain's own restaurants.

2Individual restaurants followed suit, creating their own websites for de-

livery. Even grocery stores began oering online delivery in the early 21st century (Pozzi

2012; Relihan 2017). However, generalized online food delivery services that oer delivery

from many dierent restaurants have only become popular in the past decade|and they have done so rapidly. By 2018, the online food delivery service industry had an estimated $82 billion in gross2

\PizzaNet," Pizza Hut's original online ordering destination, accepted and delivered the rst online food

delivery in 1994. 5 revenue, and accounted for 6 percent of the restaurant market in 2020 (Frost and Sullivan

2018; Morgan Stanley Research 2020). These rms are backed by revenue growth in excess

of 14 percent over the past four years, and are on track to double their market share by 2025 (Morgan Stanley Research 2020). The rapid expansion of these rms has even in uenced some restaurants to change their entire layouts, and migrate to a \delivery only" model (Bond 2019). It is clear that the restaurant market is evolving. The rst online food delivery rm, Grubhub, was founded in 2004 with the goal of re- placing all paper menus with a single website. Since then, Grubhub has transitioned to connecting delivery drivers from those restaurants in order to deliver to customers. Post- mates, Doordash, and other rms operate slightly dierently from Grubhub. These newer rms|which were founded in 2011 and 2013, respectively|provide menus from restaurants as well as contracting out delivery drivers, much like Uber or Lyft.

3These rms adopted

very similar growth strategies in which they start in select cities and expand to others with their success. 4 Consumers that use online food delivery services also have a few empirically quantied characteristics. Delivery is ordered to the consumer's home 86 percent of the time, and 74 percent of sales occur on weekends (Hirschberger et al. 2017). Further, in 2017, 43 percent of individuals who ordered with online food delivery services say that it replaced an in-person meal at a restaurant. This gure increased from 38 percent just the year before, suggesting that there is incremental cannibalization with the introduction of online channels (Morgan

Stanley 2017).

Online food delivery services often state that they are providing supplementary sales to restaurants. In fact, a survey of several thousand restaurateurs found that oering online delivery has generated additional sales for 60 percent of restaurant operators (Technomic Food Trends 2018). While online food delivery services claim, and actually do, provide incremental sales, the protability of restaurants is declining as online delivery increases3 This background information on specic rms was found on company websites.

4The heterogeneity over time and location is key to the analysis in this study.

6 (Dunn 2018; Thompson 2019). This is mostly due to high fees that online food delivery services charge, not only as service and delivery charges to the consumer, but also to the restaurant. Most online food delivery services charge the restaurant between 20-30 percent of each purchase. Online delivery often represents a large bulk of business for restaurants, so it's not an option to cut online sales channels. In the age of a pandemic, the demand for online food delivery services sales is spiking. In fact, in China, online food delivery service orders surged 20 percent during January alone; rms such as Doordash have even started reducing or eliminating their fees in response to the surge that is beginning in the United States (Keshner 2020). It is expected that consumers will continue to increase their usage of online food delivery services so long as there are stay-at-home orders and sit-down restaurants remain closed, although this likely will not completely replace pre-pandemic restaurant spending. As COVID-19 continues to impact the United States, the demand for non-contact food delivery services will likely follow the example of China and expand greatly. Understanding consumer behavior as it relates to online food delivery services is essential in this rapidly changing environment.

3 Literature Review

Recent studies have described a \retail apocalypse" in which e-commerce has forced brick- and-mortar retail establishments without online channels to shut down across the nation. However, physical stores are not quite nished. The \bricks-and-clicks" hybrid model has become more and more popular|and this trend has not been limited to just retail stores (Hortacsu and Syverson 2015). 5 This study seeks to quantify potential crowding-out eects and market expansions that have occurred due to the entry of online food delivery services and subsequent hybridization of restaurants. \Crowding-out" refers to sales that usually occur in brick-and-mortar stores that are now happening via other channels. Market expansions refer to new sales that5 A thorough review of studies on e-commerce can be found in Lieber and Syverson (2012). 7 are generated by creating an online channel for purchases. Although opening new online channels could potentially increase restaurant revenues and cause overall market expansion, new channels also allow for cannibalization of oine sales, i.e. crowding-out. Firms face a similar trade-o when introducing new products or opening a new store (Shaked and Sutton

1990; Holmes 2011; Mitsukuma 2012). Consumers that would typically purchase meals

in-person are now ordering take-out with online food delivery services. A rich academic literature describes the eects of opening new sales channels, especially relating to e-commerce. There is a particular focus on the investigation of potential mar- ket expansions and substitution eects that could be introduced with online channels in traditional markets. These studies have found signicant substitution eects across dier- ent industries, such as groceries, newspapers, and consumer electronics (Duch-Brown et al.

2017; Wang, Song, and Yang 2013; Pozzi 2013; Gentzkow 2007). The majority of studies

in this literature describe the eects of Internet-based substitutes for traditional goods and services from the consumers' perspective. Electronic goods and computers are found to have relatively sensitive prices between the online and oine purchasing channels (Goolsbee 2001; Prince, 2007). Online presence of advertisements on Craigslist lowered those found in news- papers and even reduced home and rental vacancy rates (Kroft and Pope 2014). However, in the context of restaurants, the impact of introducing online food delivery services is not as well understood. There is limited empirical evidence on the impact of adding an online sales channel to a traditional industry from the rms' perspective. In the newspaper industry, the introduction of online articles caused signicant substitution eects that greatly reduced the readership of print media (Gentzkow 2007). Grocery store sales are only moderately crowded-out with the introduction of an online channel and their overall revenues increase (Pozzi 2012; Relihan

2017). In fact, it is generally found that including an online sales channel provides signicant

increases in sales, inventory, and return on investments, while costs decrease in a sample of more than one hundred publicly traded companies (Xia and Zhang 2010). 8 The literature related specically to online food delivery services is even more limited. These types of rms have been studied only in very narrow contexts. Survey-based descrip- tive statistics show what types of consumers use online food delivery services (Yeo, Goh, and Rezaei 2017 2017). Trac and routing of drivers is studied in order to determine the eects on customer satisfaction (Pigatto et al. 2017). Website quality|estimated by the number of clicks|is quantied, as is the correlation between consumer ratings and brand loyalty (Correa et al. 2019; Ilham 2018). Not only are these studies limited in scope, but they have also been constrained to countries outside of the United States, with the exception of some non-academic survey methodologies. The eects of online food delivery services are not quantied, especially in terms of crowding-out of brick-and-mortar sales. Crowding-out eects, although well understood in some industries, have not been em- pirically studied in the context of restaurants. The case of online food delivery services is especially interesting because a third party oers the delivery service, rather than the indi- vidual restaurant opening its own specialized online channel. Further, the cannibalization of restaurant sales by online food delivery services has recently become a large point of contention.

6This study lls a gap in the literature related to online food delivery services

and their impacts on restaurants, addressing growing concerns in the restaurant industry, especially in light of COVID-19.

4 Data

4.1 Sources

The analysis leverages a proprietary transaction-level dataset provided by Visa Inc., which covers the universe of credit and debit transactions in the Visa network beginning in 2009 and up to and including 2019. Visa Inc. (NYSE: V) is the world's leader in digital payments.6

Elizabeth Dunn describes the case of a restaurant in New York City that is experiencing reduced protability

with increased online food delivery orders due to delivery service fees that are about 20-40 percent of each

order (Dunn 2018). 9 It has a steadily increasing market share of credit card volume, which surpassed 53 percent in 2017, and has maintained a market share of more than 70 percent of debit card volume throughout the sample (Peter 2019).

7The Visa data contains an annual average of 380

million cards, 35.9 billion transactions, and $1.93 trillion in sales. Of these sales, 55 percent were credit transactions and 45 percent were debit transactions. Visa volume has been steadily increasing over time, from approximately 14 percent of consumption in 2009 to almost 22 percent of consumption in 2017 (Dolfen et al. 2019). The unit of observation in the underlying data is a transaction between a cardholder and a merchant. The cardholder is the individual who used their Visa card to purchase a good or service, and the merchant is the business that provided that good or service. On the merchant side, a business name, location, and business category are recorded. The location is recorded as a zip code and the business category is the type of the business (e.g., restaurant, toy store, clothing retailer, etc.) recorded by Visa. On the card side, only a unique card identier is provided. This is the credit or debit card number. The data does not contain information pertaining to the specic goods or services purchased in each transaction, nor does it record the quantity, or price of the items. The sample is completely anonymized, so the name, address, or any other personally identiable information about the cardholder is not observable, other than what can be inferred given the card's transaction history. Cards that are used by the same person or family are not linked to one another. Each transaction has a number of observable characteristics that are pertinent to the analysis, other than the cardholder and merchant identiers. The transaction amount in dollars, date, time, and card type|credit or debit|of transaction are recorded. Further, there are several variables that dene the type of transaction that occurred. Each transaction indicates whether or not it occurred in person. Approximately half of the transactions that did not occur in person are broken down into online, mail order, phone order, and recurring transactions; the remaining transactions only record that the card was not present.7

The remainder of each of these markets is divided between MasterCard, American Express, and Discover.

4.2 Analysis Sample

The analysis sample uses all transactions that occurred at restaurants|characterized by Visa's merchant category variable|between 2014 and 2017 that pass a series of lters. Transactions that are not located in the United States are not included in the sample. Only completed and processed transactions are included. Further, debit-PIN transactions are excluded because of inconsistent routing practices.

8Cardholders that made fewer than ve

purchases at restaurants and spend more than $3,000 dollars per month at restaurants are not included in order to exclude gift cards and corporate cards. Finally, cards that did not transact on online food delivery services are omitted from the sample.

4.3 Variable Construction

The analysis relies on several data constructs that characterize dierent types of trans- actions and cardholders. First, the consumer's preferred shopping location is imputed from their transactional history. Recalling that the ve-digit zip code of the merchant is avail- able for each oine transaction, the modal zip code of oine transactions in which a card transacts at least 20 times is used to dene a cardholder's location. Dolfen et al. (2019) determined the cardholder's preferred shopping location as a latitude-longitude pair given by a transaction-weighted average zip code centroid and found that it is robust to more precise cardholder residential addresses from a large credit rating agency. This lends credibility to the modal zip code estimate, which is a less granular denition of the consumer's location. In order to characterize transactions as restaurant sales and online food delivery sales, the merchant name, card presence, and merchant category variables are used. In the case that the transaction occurred using online food delivery services, the name of the delivery service is recorded as the merchant name. The transactions that do not have a physical card swipe,8 Following the Durbin Amendment in the Dodd-Frank Bill of 2010, payment card networks were no longer

able to restrict how merchants routed PIN-based debit transactions. Therefore, after the bill became law in

2012, massive

uctuations appeared in debit-PIN transactions. In order to maintain a sample with consistent

routing, only credit and debit-signature transactions are included, which make up the vast majority of the

dataset 11 are listed as restaurant purchases by Visa, and have merchant names that map to the top online food delivery services|Doordash, UberEats, Postmates, Caviar, Grubhub, Seamless, and Delivery.com|are classied as online food delivery service sales. Total restaurant sales are all transactions characterized as restaurant purchases by Visa. The cardholder's transactional history is also used to separate dierent types of con- sumers. In particular, individuals that used cards at children's clothing or toy stores|again, characterized by the merchant business category|at some point between 2014-2017 are iden- tied as cardholders that have young children. If the cardholder does not purchase any goods from these types of stores, they are identied as not having children. Roughly one third of unique cards transact at children's clothing or toy stores at some point. The time and date of transactions are also used to characterize dierent types of pur- chases. The date is used to separate weekday and weekend transactions using the day of the week, and the time of the transaction|which is measured down to the second|is used to separate lunch and dinner transactions. Lunch purchases are dened as those that occur prior to 5 pm, and dinner transactions are those that occur after 5 pm. The analysis relies on monthly data aggregated up to the zip code level. The transaction total in dollars for restaurants and online food delivery services are recorded, along with the number of distinct cards. The sample is stratied according to the dierent types of transactions and cardholders delineated above. The main variables of interest for the analysis are measured in dollars-per-card, so the dollar amounts are adjusted accordingly. The sample is augmented by zip code-level characteristics, including voting tendencies dened as the results of the 2016 presidential election, urban-rural characterization based on American Community Survey data, and quartiles of average per-card monthly restaurant expenditure which are generated based on zip code-level average monthly spending at restaurants. 12

4.4 Summary Statistics

Before describing the main analysis of the study, it is useful to consider some initial summary statistics. First, panels of dierent types of cardholders and transactions are broken down in order to describe the number of cards and transactions, measured in millions of cards and swipes, as well as total restaurant sales, and online food delivery sales, measured in millions of dollars. These statistics are presented in

T able

1 Table 1: Panel SummaryPanel Cards Transactions Restaurant Sales OFD Sales

Overall 1.47 60.78 5,584.03 381.39

Lunch 1.46 30.67 3,352.44 195.54

Dinner 1.46 30.11 2,231.59 185.85

Weekend 1.47 24.67 1,527.56 109.39

Weekday 1.47 36.12 4,056.47 272.01

Consumers with Children 0.54 1.68 53.77 4.50

Consumers without Children 1.47 59.10 5,530.26 388.89Note:This table summarizes the number of distinct cards, number of transactions, total restaurant sales,

and total online food delivery service sales in each panel of the data, each of which is measured in millions.

In particular, the data is separated by dierent types of consumers and dierent types of transactions.

Next, zip code-level characteristics are considered. The sample is aggregated at a monthly level by zip code, and contains 478,000 observations. There are roughly 30,000 zip codes observed in the data over the duration of the sample. Within these zip codes, approximately

1.5 million cards are used in approximately 61 million transactions. Descriptive statistics are

presented in

T able

2 , including the monthly expenditure at restaurants and on online food delivery services, as well as monthly per-card spending at restaurants and on online food delivery services. Summary statistics of dierent panels are presented in the Appendix. The Appendix also provides visualizations of the dierences in distributions of total restaurant spending and online food delivery spending between dierent panels of the data. In order to have a valid study design, the entry of online food delivery services must be observable and heterogeneous over time and location. The entry of online food delivery services was validated on a city-by-city basis by examining the growth of the fraction of 13

Table 2: Descriptive Statistics

Mean Median Std. Dev. 10% 25% 75% 90%

Panel A:Zip code-level

Amount ($) 8,716 1,057 37,937 76 253 4,675 16,215

OFD Amount ($) 627 22 4,517 0 0 196 920Panel B:Card-level

Amount ($ per card) 184 163 146 55 106 229 313

OFD Amount ($ per card) 26 22 37 0 0 43 72Note:The descriptive statistics above are generated from cards that transact on online food delivery services

at some point during their lifetimes. Amount refers to the monthly dollar amount spent at restaurants and

OFD amount refers to the monthly dollar amount spent on online food delivery services. The corresponding

per-card entries estimate expenditure on restaurants and online food delivery services on a per-card basis.

restaurant sales conducted on online food delivery services. Various newspaper articles and timelines found on the websites of online food delivery services were used to conrm that the entry is accurate. After conrming that online food delivery services are observable, it is important to understand where they are present and how they are expanding. It is expected that online food delivery services will rst be present in larger metropolitan areas and grow outwards. Dening entry of online food delivery services as three consecutive months of greater than 0.3 percent of total restaurant sales being conducted on online food delivery services, county-level maps are generated for 2014 and 2017 in order to observe the expansion of these rms.

Figure

1 compares the presence of online fo oddeliv eryservices in 2014 and

2017. These maps conrm that online food delivery services entered in counties with large

cities and expanded outwards with their success over time. Trends in online food delivery services as a fraction of total restaurant spending are presented in

Figure

2 . This shows that there is very rapid growth in online food delivery services, even relative to total restaurant expenditure. 14

Figure 1: Trends in Online Food Delivery Sales

Note:These two maps show county-level entry of online food delivery services between 2014 and 2017. The

gold counties are counties that have online food delivery services and the dark blue counties are those that do

not yet have online food delivery services. An active county is dened as a county that has had greater than

0.3 percent of its restaurant sales conducted on online food delivery services in three consecutive months.

It is important to note that the map does not change much with dierent thresholds on the percentage of

restaurant sales conducted on online food delivery services.

Figure 2: Trends in Online Food Delivery SalesNote: This gure shows the trend in the fraction of total restaurant sales spent on online food delivery

services over time from 2014-2017. There is a clear upward trend that shows rapid expansion. 15

5 Empirical Approach

In order to quantify the eects of online food delivery services on restaurant sales, I propose a standard dierence-in-dierences model with a continuous treatment variable that exploits heterogeneity over time and across zip codes:

RestaurantSales

zt=z+OFDSaleszt+t+zt(1) In this equation,RestaurantSalesandOFDSalesare measured in dollars-per-card at the monthly level. Note thatzandtare xed eects of zip codezand montht, which control for seasonality, trends, and variation across zip codes. The interpretation of this regression is relatively straightforward. The key coecient,, reveals the correlation between total restaurant sales and online food delivery service sales. Out of each dollar spent on online delivery services,would be new sales caused by the entry of online food delivery services, whereas 1would become \crowded-out" from brick-and-mortar sales (i.e., 1dollars are displaced from brick-and-mortar sales for each dollar spent on online food delivery services). A threat to the validity of the regression is caused by the non-random expansion of online food delivery services. Since online food delivery services choose to enter zip codes in which the expected demand is higher, results may have a downward bias. However, by sample construction, only cards that transact on online food delivery services are included, so the validity is only threatened by a correlation between demand and the timing of the market expansion. Based on anecdotal evidence of the industry settings and the nature of these rms, this assumption is supported. Further, card-level shocks to demand are not a concern because the data is aggregated at a zip code-level. Examining the regression over dierent cuts of the data yields heterogeneity in the impact of online food delivery services. Time and weekday of transaction are available, so the data is separated into lunch, dinner, weekday, and weekend panels. Dierent seasons| cold and warm months|are also considered in order to determine the potential eect of 16 seasons on online ordering behavior. Further, other card-level spending habits are available to dierentiate consumers. The data is split into consumers that are characterized as those who have young children|determined by their spending at children's clothing and toy stores| and those that do not have children. Then, the impact in conservative states is compared to that in more liberal states, according to the 2016 presidential election results, as well as on a state-by-state basis. Finally, the impact of online food delivery services is quantied year- over-year in order to determine whether dierent levels of exposure to online food delivery services has caused incremental cannibalization of restaurant sales.

6 Results

This section includes the results of overall zip code analysis, as well as an examination of dierent cuts of the data. This includes time of purchase, type of consumer, and zip code-level characteristics with the goal of observing the types of consumers or purchasing behavior that may be driving the cannibalization of traditional restaurant sales. Further, I run the regression on a state-by-state basis and over dierent segments of time in order to determine whether there is incremental cannibalization of brick-and-mortar sales.quotesdbs_dbs14.pdfusesText_20

[PDF] [PDF] The Impact of Online Food Delivery Services on - Stanford University