Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

Contents lists available at ScienceDirect

Transportation Research Interdisciplinary Perspectives

journal homepage: [Link]/journal/transportation-
research-interdisciplinary-perspectives

Modeling the online food delivery pricing and waiting time: Evidence from
Davis, Sacramento, and San Francisco
Elham Pourrahmani a, *, Miguel Jaller a, b, Dillon T. Fitch-Polse c
a
Department of Civil and Environmental Engineering, University of California, Davis, CA, USA
b
Sustainable Freight Research Program, Institute of Transportation Studies, University of California, Davis, CA, USA
c
Bicycling Plus Research Collaborative, Institute of Transportation Studies, University of California, Davis, CA, USA

A R T I C L E I N F O A B S T R A C T

Keywords: The proliferation of delivery services has increased researchers’ and practitioners’ interest in understanding their
Online food delivery potential impacts on urban mobility. While there have been studies focusing on the operations and management
Delivery pricing of such services, there is a lack of understanding about the quality of the delivery service and their pricing
Delivery wait time
schemes. In this empirical study, we examine the delivery fee and wait time variations for ordering food using
Mixed-effect ordered-logistic regression
four food delivery applications in the U.S.: DoorDash, Grubhub, Postmates, and Uber Eats. We collected delivery
data featuring a variety of attributes of cost, timing, restaurant type, and location. The analyses cover all pricing
items comprising total cost across the apps and various study areas. Furthermore, we estimated mixed-effect
ordered-logistic regression (ologit) models to identify the variables and assess their association with food de­
livery fees and wait times. Results revealed that delivery fee variation across time of day or day of the week is
negligible; however, delivery distance significantly influences delivery fees, and this effect varies considerably
across apps. The observed effect of requester location reveals potential zone-based delivery pricing schemes by
apps. Delivery wait time is shorter when demand and supply are high. Platforms with more established networks
of users and couriers feature shorter wait times. The results of this study provide insight into user experiences
with food delivery services and the choices available, which can help inform the development of pricing schemes
for food delivery; the development of better optimization and operations management strategies by incorpo­
rating knowledge of key variables.

1. Introduction Nowadays, food delivery digital platforms are increasing in number,

with some platforms delivering almost anything, including food and
Food delivery services are growing in popularity in the US, with two- fresh items. For example, Amazon Flex drivers deliver various Amazon
thirds of Americans now using food delivery applications. According to a products, including Amazon Fresh and Amazon Restaurants (Carter,
2018 survey by Morgan Stanley Research, 65% of respondents had or­ 2019). Postmates delivers a wide range of products, with food deliveries
dered food delivery online in the six prior months. That same survey being the most significant portion of their deliveries. Other services
data also shows that ordering food delivery by phone has decreased, exclusively deliver food from restaurants (e.g., Uber Eats and DoorDash)
indicating that this reduction will continue at a rate of 3% per year or grocery stores (e.g., Rappi, Instacart, and Shipt).
through 2025, while annual online food delivery sales in the U.S. could Some of these delivery platforms are private to one business or
grow 18% by 2025 (Morgan Stanley, 2020; Noah Lichtenstein, 2020). include multiple businesses, and some have direct driver employees or
Food delivery is not a new concept. It dates to 1889, when a pizza rely on driver contractors. Success might depend on the number and
was delivered in Naples, Italy. After that, food delivery evolved for over type of restaurants and establishments a platform represents or, more
a century until 1995, when the first online restaurant delivery service importantly, its logistics performance. Most online food delivery plat­
platform, [Link], was launched in the Bay Area, featuring more forms, such as Uber Eats, Doordash, and Grubhub, match couriers and
than 60 restaurants. The current form of food delivery service began in senders, facilitate navigation and tracking, and set prices (Pourrahmani
2004 when Grubhub was founded in Chicago (Caitlyn Hitt, 2020). & Jaller, 2021). These services provide convenience and access that

* Corresponding author.
E-mail address: epourrahmani@[Link] (E. Pourrahmani).

[Link]
Received 14 January 2023; Received in revised form 25 July 2023; Accepted 25 July 2023
Available online 6 August 2023
2590-1982/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license ([Link]
E. Pourrahmani et al. Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

users are willing to pay for as they weigh the higher cost against the 2. Current state of food delivery pricing
value of their time and effort. However, there is likely a threshold for
how much users are willing to pay for the service and the limited time With the prominent growth in food delivery services in recent years,
they are willing to wait for the food to be delivered to them. Speed of restaurants and food providers have increased the prices of their menu
delivery is a key factor in customer satisfaction (Hirscheberg et al., items offered on delivery apps to respond to the extra costs associated
2016). A survey conducted by U.S. Foods (Lardieri 2019) found that with these third-party platforms. Consequently, users pay more for their
more than 70% of respondents would spend no more than $10 on de­ food when delivered than they would from the original menu (João
livery fees, service fees, and tips. The same survey found that, on Pedreda, 2020). Additionally, there are other factors affecting delivery
average, most people do not want to wait longer than 40 min for their fees. For instance, fees are subject to change due to membership and
food order to be delivered. Accordingly, fees and wait times for food loyalty plans built by platforms to provide more predictable revenue and
delivery play a critical role in requesters’ decisions to order food, and to create an exclusive network of customers. These plans generally
these factors eventually determine the level of service and sustainability include a monthly fee in exchange for reduced service fees and no de­
of such services. On the latter, the shift in demand for deliveries, livery fee for orders valued above a certain minimum.
including food, has caused severe challenges to already congested urban According to Noah Lichtenstein (2020), the variation in pricing
and suburban areas, increasing travel times and exacerbating competi­ schemes causes delivery prices for the same item from the same
tion for curb space parking (Jaller et al., 2021). This has led to further restaurant to vary by 20% or more across different platforms. Their data
problems with safety and air pollution problems that concern both cities showed that consumers pay 17–41% more when using online food de­
and delivery operators (Davani, 2021; Joselow, 2020; Schnieder et al., livery apps than ordering directly from a restaurant. Depending on the
2022). App, consumers might face fees of 12–30% of the total cost of the food.
Moreover, during the COVID-19 pandemic, on-demand delivery Overall, there is much variability in pricing schemes used by different
services became a routine necessity for many households, thus high­ platforms (e.g., DoorDash, Grubhub, Postmates, and Uber Eats). An
lighting the importance of a better understanding of how these services exploratory analysis of various schemes showed that these platforms
operate, not only from the private perspective but also to evaluate the usually include the items below:
effects on home-based accessibility to critical goods and services (Fig­
liozzi and Unnikrishnan, 2021; Silaen et al., 2023). While there have • Menu price: Menu price refers to the price indicated for the food
been studies analyzing the overall impacts of deliveries, customer’s available on the App, excluding any other fees such as taxes, service,
willingness to pay for food delivery services (and for couriers to and delivery fees. In some platforms, this item is subjected to busy/
participate in such services) (Allen et al., 2021; Lewis, 2006; Le and surge pricing and often is higher than the actual menu price that
Ukkusuri, 2019; Li et al., 2020; Pahwa & Jaller, 2022), there is still a could be ordered directly (in-person) at the restaurant. Restaurant
lack of understanding about the pricing schemes implemented by on- owners might increase this price to offset any losses due to the de­
demand delivery services in different areas. The challenge is that there livery, such as commissions they pay to delivery apps.
is little access to primary data resources for research, and delivery • Delivery fee: Delivery fee refers to the price charged for the food
platforms do not publicly provide details about their pricing schemes transported from where it is prepared to the customer. This fee might
and waiting time. In addition to delivery time and distance, food type be subjected to surge (busy time) pricing by some platforms when
and restaurant parameters further influence both delivery cost and wait demand is high.
time. Capturing such influences requires data to have enough variations • Service fee: The service fee is charged by the delivery platform
in those parameters. To overcome these limitations, we collect empirical providing the service. This convenience fee helps the App continue to
data from four popular food delivery apps (DoorDash, Grubhub, Post­ operate and maintain its platform. This fee is a percent of the menu
mates, and Uber Eats), including the five most popular food types, and price or total cost, added to the final bill. There is no clear or
15 restaurants and café located in three Californian cities and analyze consistent percentage rate for service fees across platforms: it varies
the data for delivery fees and platform-estimated wait times. The data from 10 to 18%, with some platforms not charging a service fee at all.
includes delivery pricing and wait times per time of day, day of the • Taxes: Sales tax is applied to orders according to local laws.
week, food type, and delivery distance. The empirical analysis begins Empirical data shows that App’s tax rate fluctuation is about 1.1%.
with characterizing priced items and studying each item in the context • Gratuity or tip: A gratuity is an optional fee paid by the customer
of each delivery app and geographic location. The paper then estimates added to the driver’s income.
mixed effect ordered logistic regression (ologit) models to shed light on • Other fees: Other fees can be charged by restaurants and platforms
factors that might influence food delivery fees (including menu price, to customers. Some of the extra fees included in some (but not all) of
service fees, delivery fees, and other fees), wait times, and their the online food delivery apps are:
magnitude of association. In addition to affecting consumer purchase o Merchant fees: Merchant fees (included in the service fees by
patterns and consumption power, the fees and wait times also determine some delivery platforms) are small fees imposed in addition to the
the remuneration of couriers and the profitability of service platforms by actual food menu price by certain restaurants to help offset the
offsetting costs (Nguyen et al., 2019; Goods et al., 2019). profits lost because of food delivery.
The results of this study contribute to research and practice by i) o Small order fee: A small order fee is an additional fee applied to
providing a better understanding of the pricing schemes implemented by small or low-price orders to make them economically efficient for
popular platforms and how this can affect experiences and choices; ii) delivery. For some platforms, this fee even applies to a member­
informing the development of new and more transparent pricing ship subscription.
schemes to charge customers and compensate couriers maintaining o Delivery minimum: A delivery minimum is a minimum number
service operability and profitability for platforms and food providers; iii) of ordered items or a price minimum set by delivery platforms to
helping improve the design of optimization and operations management be met for the delivery request to be approved.
strategies and models that incorporate knowledge of the key variables o Bag fee: A small fee is sometimes charged per reusable bag
affecting pricing; and iv) supporting with new findings the decision- included with an order. In some delivery platforms, this is included
making on critical topics of on-demand delivery service management in service fees.
in urban areas such as courier wages, right of way and accessibility, o Regulatory response fees: These fees go directly to the app
parking allocation for delivery activities, and home-based accessibility platform to recover revenue loss due to caps on fees the platform
to goods and services for vulnerable populations, especially in dense could charge restaurants per transaction (15% commission cap
urban areas. imposed by at least 68 cities, counties, and states).

2
E. Pourrahmani et al. Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

Table 1
Restaurant information.
Food Restaurants
Type
UberEats DoorDash Postmate Grubhub

Davis Sac SF Davis Sac SF Davis Sac SF Davis Sac SF

Mexican Chipotle Chipotle Chipotle Chipotle Chipotle Chipotle Chipotle Chipotle Chipotle Chipotle Chipotle Chipotle
Indian The Preethi The The Preethi The The Preethi The The Tandoori The
halal Indian halal guys halal Indian halal halal Indian halal Halal grill halal
guys cuisine guys cuisine guys guys cuisine guys Guys guys
American Panera Sourdough Boudin Panera Sourdough Boudin Panera Sourdough Boudin Panera Sourdough Boudin
(Fast bread and Co café and bread and Co café and bread and Co café and bread and Co café and
casual) bakery bakery bakery bakery
American Subway Subway Subway Subway Subway Subway Subway Subway Subway Subway Subway Subway
(Fast
food)
Café Starbucks Starbucks Starbucks Peet’s Peet’s Peet’s Philz’s Jamba juice Philz’s Temple Temple Sextant
drinks coffee coffee coffee coffee coffee coffee coffee coffee
roaster roaster roaster

o Driver benefits: this fee is intended to help cover benefits granted 3.2. Factors associated with delivery food pricing and waiting times
to drivers under Proposition 22, such as healthcare stipend, in­
surance, and guaranteed minimum wages calculated based on The primary purpose of this study is to discern the factors that may
local minimum wages. influence delivery fees and wait times using a mixed-effect ordinal lo­
gistic (ologit) regression model. The variable Delivery Fee is treated as a
The fee structure is unclear to customers, and the factors affecting dependent variable to explore factors associated with the amount the
such fees are even less understood. For example, there is a lack of platforms charge users as delivery fees. The domain of this variable is Dp
transparency on how much of what consumers pay goes to pay the = {0.49, 0.99, 1.49, 1.99, 2.49, 2.99, 3.49, 3.99, 4.49, 4.99, 5.49, 5.99,
couriers, how much is charged by the platform, or even why these vary 6.49, 6.99} categorized into seven classes with $1 intervals (Table 2).
by location (Dablanc et al. 2017; Xu, 2021). Despite increased research Similarly, the variable Delivery Wait Time is treated as a dependent
on operational factors related to crowdshipping, last-mile deliveries, and variable to explore factors associated with user wait times for food de­
pricing work for ride-hailing services, empirical analyses on food de­ livery. This variable falls into four categories: 1 = Low (less than 20
livery pricing are lacking. Tong et al., 2020 and de Oliveira et al., 2022 min); 2 = Medium (20–30 min); 3 = High (30–40 min); and 4 = Very
are among the only studies found trying to provide insights into the High (more than 40 min). See Table 3.
issue, and concentrate on the Chinese and Brazilian markets, respec­ We selected the ologit model to demonstrate the relationship with
tively. This study will provide additional empirical information and potential explanatory variables because delivery fees and wait times
analyses to understand these pricing schemes in general and assess the present an ordinal scale. Because we collected the data from five apps,
impacts of the different factors on such pricing for the study cases in the five food types in three cities, and three request locations in every city,
U.S. the data inside each cluster are correlated, and treating the data as in­
dependent observations causes the standard errors of coefficients to
3. Materials and methods become underestimated, thereby overestimating the statistical signifi­
cance in return. Mixed-effect models recognize these clusters in the data
3.1. Data collection and allow for the existence of residual variance at each cluster level (Wu,
2009).
As mentioned, there is a lack of primary platform data about the
pricing schemes. To overcome this limitation, we complied data for four
popular food delivery apps, Uber Eats, DoorDash, Postmates, and 3.3. Model preliminaries
Grubhub, from mid-July to mid-August 2021 from 15 restaurants and
cafés located in the Northern California regions of Davis (Dav), Sacra­ In the remainder of this section, we describe the background of cu­
mento (Sac), and San Francisco (SF). We collected 1,080 records by mulative logit models for two-level structures where a total of n obser­
submitting delivery requests using a smartphone as a regular user vations (at level 1) is clustered within J groups (at level 2) with nj
without any membership subscription. We randomly chose three loca­ individuals in group j. We denote the ordinal outcome and the explan­
tions for each region to capture the effect of delivery distance and time atory variable by yij and xij, respectively, for observation i in group j. If
on price and wait time. Pricing details and wait time data were collected the ordinal outcome y takes ordinal values from 1 to K, the multilevel
for all pairs of requesters and restaurants at different times of the day: cumulative logit extension is as follows (Equation 1) (Hedeker and
morning (AM), midday (MD), and evening (PM), on both weekends and Gibbons, 1994):
weekdays. To explore the effect of food type, we considered five types of ( ( ))
( ) Pr yij ≤ k ( )
popular meals: Mexican, Indian, American fast casual, American fast Logit(Pr yij ≤ k ) = log ( ) = αk − βxij + uj , k
food, and café drinks (e.g., coffee and tea). While we intended to Pr yij > k
consider the same restaurants for each food type across cities and apps, = 1, ⋯, K − 1 (1)
not all restaurants of interest were available in all cities or across all
apps. For example, there was no Panera Bread in the SF region, and where αk represents the intercepts or cut-off points for the outcome
Starbucks was not operating in the Grubhub App at the time of data probabilities that vary across K categories. This parameter, αk , is inter­
collection. Thus, we considered similar restaurants as replacements for preted as the log-odds that an observation belongs to category k or
these cases. Table 1 summarizes the characteristics of the subjects (food, lower, with x = 0 and u = 0. The parameter β is the effect of one unit
restaurants, Apps, and regions) from which we collected the data. change in explanatory variable × on the expected response variable
level in the ordered log-odds scale. In contrast, the other variables in the
model are held constant after adjusting for cluster effect u. Here, the

3
E. Pourrahmani et al. Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

Table 2
Distribution of delivery fee categories among platforms.
Fee level 1 2 3 4 5 6 7
(<$0.99) [0.99, 1.99) [1.99, 2.99) [2.99, 3.99) [3.99, 4.99) [4.99, 5.99) [5.99, 6.99]

Frequency 71 180 184 331 195 84 35

the model. The predicted cumulative outcome probability for observa­

Table 3
tion i in group j is calculated from the fitted model in Equation 2, where
Distribution of delivery wait time categories among platforms.
sample estimations replace parameters.
Wait time level 1 2 3 4 ( ( ))
Frequency 246 466 304 64 ̂ ij ≤ k) = exp ̂ αk − ̂ βxij + ̂uj
Pr(y ( ( ) ) k = 1, ⋯, K − 1 (2)
1 + exp α ̂k − ̂ βxij + ̂uj

positive values of β imply that one unit increase in x increases the or­
dered log-odds of being in higher categories of outcome y (McCullagh, 3.4. Model estimation
1980). The cluster-level residual uj ~ N(0, σ 2u ) represents the between-
cluster variance due to the unobserved influences on outcome variables, We initiated the model estimation by defining the varying effect
allowing the intercepts and cumulative outcome probabilities to vary structure of the model. Since the requester locations were nested into
among clusters. Note that the relation between the cumulative logit and cities, we considered Requester Location as the varying effect, nested in
explanatory variables is independent of the ordinal cut-off intercepts, City. We also considered Food Type as a varying effect in the models.
which point to the identical slopes or proportional odds-ratio assump­ While food type could have a fixed effect on both delivery fees and wait
tion in ologit model (McCullagh, 1980). times, the food types and restaurants domain in the sample do not fully
To facilitate interpretation, we take the exponentials of the co­ cover variations in the apps, and its inclusion as a fixed effect might lead
efficients, exp(β), and interpret them as odds ratios that compare the to misleading results.
odds of being in a lower outcome category for two observations with x Next, we identified the candidate variables to include in the model
variable one unit apart, holding other explanatory variables constant in considering three factors: 1) study objective, 2) clinical relevance, and
3) collinearity. For collinearity, we conducted a series of statistical tests,

Table 4
Other fees information for all Apps and cities.
Other fees UberEats DoorDash Postmate Grubhub

Davis Sac SF Davis Sac SF Davis Sac SF Davis Sac SF

Service fee1 16% 16% 16% 17%, no 17%, no 15%, no 21%, 21%, 21%, min: 5–23% 2–23% Chipotle
(% of menu (min: (min: $3, (min: more than more than more than min: min: $3 $3 (15%),
cost) $3, max: $5) $3, $15 $15 $15 $3 Boudin
max: max: It can It can It varies It varies (10%),
$5) $5) Chipotle Chipotle Chipotle, increase increase based on based on Halal
(17%), All (17%), All Halal, and up to 40% up to 47% unknown the delivery (7%),
others others Boudin to meet to meet factors. It destination subway
(15%) (15%) (15%), the $3 the $3 generally (10%),
while not while not subway, minimum minimum increases Sextant
exceeding exceeding and café (e.g., (e.g., with coffee
$15 $15 drinks Jamba Philz delivery (21%)
(13%) juice) coffee) distance

Taxes 8% Subway 9% Halal (9%), Peets Chipotle 8% 9% 9% 10–13% 10–13% 10–13%

Grubhub: (12–13%), Peets coffee and Halal
(% of menu Starbucks coffee (0%) and (8%),
cost + (0%), and (0%), and All others Boudin
Small order All others All others (9%) and
fee + (9%) (8%) Subway
Delivery (9%), and
fee) Peets
coffee
All others: (0%)
(% of menu
cost)
Small order 3 (under 3 (under 3 2 2 2 (under
fee ($) $12) $12) (under (under (under $10) $10)
(condition) $12) $10)
Driver 2 2 2 2.5 2.5 2 2.5, 3.53 2.5, 3.5 2.5
benefit2
($)
Regulatory 1 1
response
fee4 ($)
1
This is a convenience fee that helps the App continue to operate and maintain its platform.
2
The is intended to help cover benefits granted to drivers under Proposition 22, such as healthcare stipend, insurance and guaranteed minimum wages calculated
based on local minimum wages.
3
In addition to cover aforementioned benefits, Grubhub drivers receive 30 cents per active mile driven.
4
These fees go directly to the App platform to recover the lost revenue from the 15% commission cap imposed by at least 68 cities, counties, and states.

4
E. Pourrahmani et al. Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

Fig. 1. Share of items in the food delivery pricing scheme for each App and City.

including the Chi-square test, ANOVA test, Kruskal-Wallis test, and restaurant. Other apps automatically choose one location for the
graphical inspection, among pairs of variables. The tests indicate a requesters, depending on the location or working hours of the
correlation between the variables City and Delivery Distance. We did not restaurant.
include highly correlated variables in the same model as fixed effects. • Several restaurants close earlier on some apps while still open on
Further, we created a list of potential interacting variables to examine in others. In Davis, for example, Halal Guys was open on Uber Eats but
the model estimation. We evaluated the extended models against the was closed on DoorDash later in the morning.
base model via the Likelihood Ratio (LR) test. • Several restaurants in SF close operations earlier in the evening
To select the final models, we evaluated more than 20 candidate (about 5:00–6:00 PM) on all apps than they do in Davis and Sacra­
models per subject (delivery fee and delivery wait time), considering the mento. Moreover, some branches of the same restaurants in a city
study objective, relevance of variables, and statistical measures such as have variable working hours: some might start the day earlier while
the Akaike Information Criterion (AIC) and R2. We also compared the others close later in the evening.
models against one another using the LR test wherever applied. We
disregarded singular models1 and those having a large Hessian condition 4.2. Menu cost
number2 from further consideration.
Data analysis on the collected sample from Davis shows that menu
4. Results and analysis pricing is the same across the apps for the same food items. However,
these prices might vary by $1–2 in Sac and SF, depending on the
4.1. Preliminary analysis restaurant branch. For example, a foot-long regular oven-roasted turkey
Subway sandwich ranges in price from $9.50 to $10.39 on the menu on
Before diving into data analysis, a few observations made during Postmates in Sac. In SF, the same sandwich can cost somewhere between
data collection are worth mentioning: $9.99 to $11.49 on DoorDash and Grubhub. While requesters can select
their restaurant branch of choice in Grubhub, the other apps choose it
• In one case, Postmates increased the delivery fee more than five automatically regardless of the menu price. Analyzing the menu price of
times when there were no couriers around to be matched a request food versus the time of day on weekdays or weekends did not indicate a
during an evening weekend in Davis. This observation was removed noticeable relation.
from the dataset for further analysis.
• The studied apps differ in some ways: Grubhub allows requesters to
4.3. Delivery fee
choose from different available locations in a region for the same

Depending on the platform commission rate and settings, the total or

a percentage of the delivery fee is paid to couriers. Delivery fee is in­
1
In singular models, some dimensions of the variance–covariance matrix are dependent of other fees, such as driver benefits or regulatory response
exactly zero. The structure of model is too complex to be explained by the data. fees, which will be discussed later. Delivery fees might vary depending
2
Large Hessian condition number indicates significant numerical instability on delivery distance, time of day, and day of the week (weekends or
in the computation of inverse of Hessian. weekdays).

5
E. Pourrahmani et al. Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

Fig. 2. Food delivery fees as % of food menu price for each App.

Table 5 4.5. Impact of fees on total order cost

Preliminary statistics of food delivery wait time.
Variable Category Min Max Average Standard For the rest of this section, we discuss the average extra fees paid for
(minutes) (minutes) (minutes) Deviation online delivery by users as the share of the total cost according to our
Day Time AM 10 55 28.62 8.43 sample. Depending on the App and the city, consumers might pay
MD 10 52 27.42 8.23 45–50% of the total cost as extra fees (delivery fees, service fees, and
PM 10 61 30.04 9.46 other fees) compared to ordering from a restaurant in-person (see
Day of the Weekend 10 61 29.32 9.06 Fig. 1). The extra fees for small orders of less than $10 in the menu can
Week Weekday 10 55 28.06 8.45
City Davis 12 60 29.59 8.05
be more than two times the food menu price, though this shrinks to
Sac 10 52 28.45 8.45 70–80% for orders with more than $10 in value (Fig. 2). These are
SF 10 61 28.03 9.71 significant increases in access costs for these products, especially when
Food American 10 40 22.89 7.28 delivery becomes the only access channel, as was the case for many
Type Fast Food
during the first few months of the pandemic.
American 15 55 28.18 7.25
Fast Casual
Mexican 14 55 27.99 8.60 4.5. Waiting time
Indian 16 61 35.94 7.41
Café Drink 12 60 28.46 8.06
In addition to pricing information, we collected delivery wait times
for each delivery request. Delivery wait time refers to the expected
The delivery fee for the same item in the same context (app, city, waiting time, announced by the App, to receive the order after con­
requester) varies up to $2.00 in the collected sample. This may be due to firming the request by the app user. Over the entire sample, the collected
restaurant working hours, traffic congestion, changes in demand, and wait times were 10 to 61 min, with an average of 28.70 min and a
supply density. standard deviation of 8.78. It is important to note that these represent
estimated wait times by the platform, and actual waiting times may
4.4. Other fees vary. Table 5 presents preliminary wait time statistics for the data seg­
ments defined by the various variables.
Additional items are commonly categorized as “other fees” in the
receipt. Table 4 lists these other fee items along with details. Note that 4.6. Mixed effect ologit models
not all fees are included for all apps and cities. Regulatory response fees
are only paid by Uber Eats and DoorDash users. Driver benefit fees are All models and analyses were conducted in R statistical software
included in all apps except DoorDash, whose drivers get the full delivery using packages including ordinal, performance, stats, ggplot2, and
fee, tips, and any boosts to their earnings (Helling, 2021). While other sjPlot.
apps are consistent with their service fees, Grubhub differentiates its
pricing based on delivery distance and destination factors. In SF, it uses 4.7. Delivery fee
separate rates for different restaurants. Order tax rates are usually about
8–9% of the menu price. Analysis of our sample showed that the tax rate Table 8 shows the outputs of the best-fitting mixed effect ologit
on Grubhub applies not only to the menu price but also to the delivery model for the delivery fee. The model is compared to the ologit model
fee and any small order fee. without varying effects via the LR test. The result indicates that the
model with varying effects statistically improves prediction (Table 6).

6
E. Pourrahmani et al. Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

Table 6
Likelihood ratio test of delivery fee model.
Model Number of AIC Log Likelihood LR Test Degree of Pr(>Chisq)
Parameters (LL) Freedom

Delivery Distance + App + Delivery Distance:App 13 3148.9 − 1561.4

Delivery Distance + App + Delivery Distance:App + (1|City/ Requester 16 3000.8 − 1484.4 154.04 3 < 2.2e-16
location) + (1|Food Type) ***

Table 7 comparable, these results differ from finding in Brazil that showed
Mixed-effect ologit model of delivery fee with and without interaction effects. varying impacts for the time of day and day of the week for different
Variable Name Log-Odds product categories (de Oliveira et al., 2022), and the lower fees identi­
fied during the weekends in China (Tong et al., 2020).
With Without Interaction
Interaction Fig. 3 illustrates the varying (random) effects via conditional modes
with 95% confidence intervals based on conditional variance. The ef­
Thresholds
1|2 − 5.3 − 4.27
fects of random components are considered small on model parameters.
2|3 − 3.20 − 2.23 The effect of Food Type is significant between some levels. The random
3|4 − 1.75 − 0.87 effect of Indian and Fast Casual food deliveries on delivery fees is
4|5 0.46 1.29 negative compared to the other types (Fig. 3(b)). This can be due to
5|6 2.20 2.99
factors related to restaurant brand and restaurant location in addition to
6|7 3.87 4.57
Coefficients food type. These factors were not fully captured during data collection.
Delivery Distance 0.29 * 0.88 * Fig. 4 shows the curves describing the probability distribution for
App (Reference: DoorDash) each delivery fee level over the delivery distance values. For DoorDash,
Grubhub − 4.86 *** − 3.26 *** the effect of Delivery Distance on probability distribution is imprecise.
Postmate − 0.83 ** − 0.19 **
UberEats − 3.96 *** − 3.30 ***
This contrasts with Grubhub, where distances of 6 miles or more are
App: Delivery Distance more likely to have $6–7 delivery fees or higher, and delivery distances
Grubhub: Delivery Distance − 3.60 *** of 2.5 miles or less are associated with less than $2 in delivery fees.
Postmate: Delivery Distance − 0.09 ** Predicted probability curves for each App are illustrated in Fig. 5,
UberEats: Delivery Distance − 3.26 **
assuming all varying effects are fixed at their mean value. The curves for
Between cluster variance
Nested (Requester Location: City) 0.20 0.17 DoorDash and Postmates are very similar, with delivery fee levels 4 and
City 0.05 0.02 5 being more likely to occur. In contrast, Grubhub and Uber Eats curves
Food Type 0.50 0.42
* p < 0.05 ** p < 0.01 *** p < 0.001
AIC 3000.84 3052.834
Marginal R2 / Conditional R2 0.48/0.58 0.46/0.55 Table 8
LL − 1484.4––1513.4 Mixed-effect ologit model of delivery fee.
LR test: chibar2(3) = 57.993 Prob >= chibar2 = 1.58e-12 ***
Variable Name Parameter

Log-Odds Odds Ratio

Table 7 shows the cross-comparison for the delivery fee model with Thresholds
and without interaction effects. According to the results, the model, 1|2 − 5.273
including the interaction terms, improves (statistically significant) 2|3 − 3.165
model prediction, although the magnitude of the improvement is likely 3|4 − 1.711
4|5 0.500
small. 5|6 2.233
The total exploratory power of the model is 58% (conditional R2), 6|7 3.912
and the part related to the fixed effect alone is 48% (marginal R2).3 The Coefficients
variance partition coefficient (VPC) (Goldstein et al., 2002) indicates Delivery Distance 0.279 * 1.32 *
App (reference: DoorDash)
that 12% (0.5004/(0.2022 + 0.0547 + 0.5004 + 3.294)) and 6%
Grubhub − 4.873 *** 0.01 ***
((0.2022 + 0.05398)/(0.2022 + 0.0547 + 0.5004 + 3.29)) of the total Postmate − 0.842 ** 0.43 **
variations are due to food type and nested (cities and requesters within UberEats − 3.982 *** 0.02 ***
cities) effects, respectively. App: Delivery Distance
The model results (see Table 8) show that the parameters Delivery Grubhub: Delivery Distance − 3.614 *** 0.03 ***
Postmate: Delivery Distance − 0.105 ** 0.90 **
Distance and App, and all the interactions between Delivery Distance and UberEats: Delivery Distance − 3.278 ** 0.04 **
App, are statistically significant, with the magnitude of the effects Period
varying widely. MD 0.031 1.03
Unsurprisingly, the longer delivery distance increases the probability PM 0.176 1.19
Weekend − 0.040 0.96
of paying higher delivery fees. This is expected and consistent with
Between cluster variance
previous findings and the basic pricing structures in ride-hailing that Nested (Requester Location: City) 0.2022
include, at a minimum, time and distance components. However, vari­ City 0.0547
ables for the time of the day (AM, MD, PM) and day of the week Food Type 0.5004
(weekday/weekend) had relatively negligible insignificant coefficients * p < 0.05 ** p < 0.01 *** p < 0.001
Number of observations 1080
in all models explored during the experiments. Although not directly Number of groups Nested (9), Food Type (5)
Number of observations per group 120, 216
AIC 3004.481
3 σ2 3.29
For more information on marginal and conditional R2, it is referred to
Marginal R2 / Conditional R2 0.48 / 0.58
Nakagawa et al., (2013).
4
Residual term in ologit model follows a standard logistic distribution with Due to the inclusion of variables Period and Weekend, the value of parameters slightly
variance π2 /3 = 3.29. differs from the same model with interaction presented in Table 7.

7
E. Pourrahmani et al. Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

Fig. 3. Delivery fee varying effects with 95% confidence intervals a) Requesters nested within the cities, b) Food types, and c) Cities.

Fig. 4. Predicted probability versus delivery distance for each App.

are more similar, with delivery fee levels 2–4 being more likely to occur. The total exploratory power of this model is 44% (conditional R2),
and the part related to the fixed effect alone is 16% (marginal R2). The
VPC indicates that 29% (1.414/4.803) of the total variation is due to
4.8. Delivery wait time
Food Type.
Most of the estimated coefficients are statistically significant in this
Table 10 shows the outcomes of the best-fitting mixed-effect ologit
model. Having Delivery Distance as a varying slope made the fixed effect
model for delivery wait time. The model is compared to the ologit model
of the same variable insignificant but statistically improved the model
without varying effects via the LR test. The result indicates that the
prediction. This is because the travel time of one unit of delivery dis­
model with varying effects statistically improves prediction (see
tance varies in various cases depending on traffic congestion. Hence, the
Table 9).

8
E. Pourrahmani et al. Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

Fig. 5. Delivery pricing level probabilities for the average requester, city, and food type.

influence of delivery distance on delivery wait time varies by requester

location. The predicted probability of waiting longer is about nine times Table 10
higher using Grubhub than DoorDash. This could be because DoorDash Mixed-effect ologit model for delivery wait time.
has a more established fleet of drivers for delivery service (Schroder,
Variable Name Parameter
2022) compared to other apps. The odds of waiting longer for MD de­
liveries are 35% lower than for AM deliveries. This can be attributed to Coefficient Odds Ratio
the higher demand during the midday period compared to earlier in the Thresholds
day, which attracts a higher supply of drivers and reduces wait time in Threshold 1 − 0.665
Threshold 2 2.052
return. Predicted probability curves for each App are illustrated across
Threshold 3 4.995
the wait time levels in Fig. 6, assuming all random effects are fixed at the Coefficients
mean. Unlike DoorDash, the delivery waiting time for Grubhub is more Delivery Distance 0.189 1.21
likely to be high. App
Grubhub 2.281 *** 9.79 ***
Postmate 0.603 *** 1.83 ***
5. Discussion and conclusions
UberEats 0.576 *** 1.78 ***
Period
In this study, we collected fee and platform-estimated wait time data MD − 0.424 ** 0.65 **
from four food delivery apps in three California small to large cities. We PM 0.261 1.30
discussed various fee items that comprise the price paid by users and Between cluster variance
Nested (Requester Location: City)
compared them across apps and cities. We developed mixed-effect ologit Intercept 0.000
models to identify the variables and their magnitude of association with Delivery Distance (Slope) 0.055
food delivery fees and wait times. City
The analyses show the total and itemized fee components variability Intercept 0.038
Delivery Distance (Slope) 0.006
across apps and cities. Some of these fees can be significantly different.
Food Type 1.414
For example, menu prices for food differ across the apps and cities up to * p < 0.05 ** p < 0.01 *** p < 0.001
about $2 (which can represent almost a 25% difference in item price), Number of observations 1080
and not necessarily dependent on the cost of living because other factors Number of groups Nested (9), Food Type (5)
such as competition and demand may be inherently affecting the base Number of observations per group 120, 216
AIC 2162.13
prices of the merchants in the various cities. Other fees include items 3.29
σ2
that may or may not be present in all apps or cities. Service fee also Marginal R2 / Conditional R2 0.158 / 0.441
varies significantly across platforms ranging from 6 to 8% for Grubhub

Table 9
Likelihood ratio test of delivery wait time model.
Model Number of AIC Log Likelihood LR Test Degree of Pr
Parameters (LL) Freedom (>Chisq)

Delivery Distance + App + Period 9 2459.6 − 1220.8

Delivery Distance + App + Period + (1 + Delivery Distance |City/Requester 16 2162.1 − 1065.1 311.52 7 2.2e-16
location) + (1|Food Type) ***

9
E. Pourrahmani et al. Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

Fig. 6. Delivery wait time level probabilities for the average requester.

and between 8% and 18% for the others. The analyses also showed that effect on delivery fees, but more data could help confirm the underlying
there might be tradeoffs between itemized fees in the platform’s fee reasons. It is also important to note that only in one of the apps can users
structure, which affect the total paid price (e.g., a lower service fee but a select the specific store they want their deliveries from, and the analyses
higher item price markup or other fees). This is important considering assumed that the App automatically selects the closest available one.
that fees other than delivery and service can be even more significant The delivery distance was also a significant factor explaining waiting
than those combined, as is the case of Grubhub, which has other fees time variability, though with a lesser effect than determining the de­
between 25 and 33%. livery fee. And this result is expected considering that traffic congestion
In contrast, Postmates has a more evenly split structure. Regardless, and consumer density, which could affect the choice of the courier to
on average, users pay about 50% extra fees for delivery services wait for and consolidate deliveries, vary across cities, amongst other
compared to when they order personally in restaurants or food places factors. Specific to the case studies, the wait time model indicates lower
(though this study did not analyze costs related to time savings, the wait times for DoorDash than other apps, particularly Grubhub, which
utility of dining in, or other factors). Depending on the App they use and could be due to the established network of couriers and requesters
the value of the order, the portion of extra fees can be more than twice registered for DoorDash. Midday deliveries are correlated with lower
the menu price. waiting times compared to earlier times of the day. One potential reason
Although the purpose of this study was not to evaluate the willing­ for this could be the higher demand rate and supply for food deliveries in
ness to pay or the benefits and satisfaction for using or not these delivery the middle of the day in general.
services, the results do show evidence of the added cost of convenience, In summary, the results shed light on the differences in pricing
especially for small orders (e.g., less than $10), for which the user might schemes across apps, highlight the share of the components on the total
pay between 1 and 1.4 times the cost of the desired item in fees alone (i. price paid, and how the total fees can be more than the item itself. We
e., excluding item price). This is because although most of the fee items found delivery distance to be crucial in explaining delivery fees and
are proportional to some order sub-total, some fixed fees are imposed for waiting time, which are critical factors driving the quality of delivery
small orders. It would be interesting to conduct further research to services. The varying effect of cities and request locations on delivery
investigate the effects of membership and other price plans that can help fees in the model indicates the possibility of location-based pricing
offset some of these costs for frequent users. Exploratory analyses on a implemented by delivery service. However, despite the evidence of
subset of platforms showed significant variability in the waived amounts temporal change in delivery demand and traffic, the delivery fees did not
by App and restaurant. show a significant change concerning time.
Additionally, the results showed that despite the discussion that Similar to the case with ride-hail services, these delivery services
some platforms may implement surcharges (e.g., dynamic price multi­ may generate negative externalities in the system, such as traffic
pliers), the overall variation of menu prices and delivery fees across time congestion, reduced safety, and emissions that are proportional to de­
and day was negligible and can be attributable to occasional changes in livery distance mileage and vary by time and location. Similar ap­
restaurant workload, delivery demand, and supply availability. proaches implemented in regulating the shared-mobility industry,
From the various factors considered, delivery distance significantly policies could help internalize such costs (e.g., unified mileage fee).
impacts delivery fee, but the effect varies widely by App. While this Time-varying and location-based pricing policies are effective mecha­
effect is less impactful for DoorDash, the probability distribution of nisms to stabilize the demand and supply and mitigate externalities.
delivery fees positively correlates with delivery distance for Grubhub. Even efforts to foster the use of cleaner vehicles by couriers (e.g., Clean
The significance of the varying effect of requester location in the de­ Miles Standard in California requiring the electrification of ride-hailing
livery fee model indicates the possibility of zone-based pricing imple­ companies starting in 2023) are potential means to mitigate
mentation by some apps. The models also identify a significant food type externalities.

10
E. Pourrahmani et al. Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

The results on wait time also offer additional insights into potential based accessibility for their customers. As the operations behind these
operational consequences. To increase productivity, couriers may try to services generate traffic and require public infrastructure (e.g., roads,
consolidate orders and speed up their travel, thus increasing violations curbside), the new models can help inform planning and policy for
and reducing traffic safety. In places like Brazil, the government has developing strategies to optimally allocate resources and contribute to
enacted Federal Laws prohibiting practices that encourage fast speeds the sustainability of the transportation system.
during deliveries (de Oliveira et al., 2022). In the US, there is a signif­ Finally, although this study provided new empirical data and find­
icant precedent when in 1993, Domino’s Pizza Inc. had to pay more than ings, it has some limitations. Future research could focus on further
$79 million after a jury awarded it to a woman injured when a company analyzing the variation in requester locations and food types; logging
driver crashed into her. As a result, the company stopped promising to real-time delivery travel time between origin and destination at the time
deliver their pizzas within 30 min (or get the product for free). of request; identifying couriers’ spatial density empirically across the
The data collection process and empirical data did not reveal a day; evaluating changes in factors such as menu price, working hours,
transparent allocation of fee revenue for the platform, the courier, the and demand intensity with restaurants in place; enhance the data
restaurant, and others (e.g., tax). While collecting the data, we did not collection with interviews with users and restaurants/food providers;
include an additional amount for tips, and even without this, and it is integrate this type of research with consumer behavior analyses; and
evident that the cost of accessing food through delivery apps is very operationalize the econometric models with other operations manage­
high. ment and optimization tools. Including additional variables in the
As discussed before, during the pandemic, in the absence of dine-in analysis will require larger samples than we obtained in this study.
options, those who could access the apps, which is already an accessi­ Despite the limitations, this study brings us closer to a better under­
bility barrier for some, had to pay extra. This represented an added standing the fees used by on-demand food delivery platforms.
burden to vulnerable populations. As these services become more
ubiquitous in the marketplace, people will be forced to access items
Declaration of Competing Interest
through them, thus affecting limited budgets. Accessibility could also be
hampered by a phenomenon that has grown in different locations,
The authors declare that they have no known competing financial
starting in Europe with fast-delivery service providers and then trans­
interests or personal relationships that could have appeared to influence
lating to the food industry with the use of dark kitchens, cloud kitchens,
the work reported in this paper.
ghost kitchens, or delivery-only restaurants, for which ordering online is
the only option. These locations provide the infrastructure to prepare
Data availability
food for single or multiple restaurants and dispatch orders through the
apps. In 2022, some European cities in France, the Netherlands, and the
Data will be made available on request.
UK, enacted or started to develop policies to manage (e.g., permanently
or temporarily ban new dark stores and kitchens) and regulate these
References
facilities to curb local impacts, improve safety and security, or even to
legally define and classify them (O’brien & Haeck, 2022). There were Allen, J., Piecyk, M., Cherrett, T., Juhari, M.N., McLeod, F., Piotrowska, M., Bates, O.,
discussions in New York City on how to license and regulate these fa­ Bektas, T., Cheliotis, K., Friday, A., Wise, S., 2021. Understanding the transport and
cilities, especially by the Department of Consumer and Worker Protec­ CO2 impacts of on-demand meal deliveries: A London case study. Cities 108,
102973.
tion (Fickenscher & Wayt, 2022). While the industry is still developing, Carter, G. (2019). How To Make Money On The Side With Amazon Flex. Retrieved from
it is essential to identify the potential risks to the various stakeholders, [Link]
including restaurants, couriers, and the final customer. For example, Dablanc, L., Morganti, E., Arvidsson, N., Woxenius, J., Browne, M., & Saidi, N. (2017,
October). The rise of on-demand ‘Instant Deliveries’ in European cities. In Supply
developing strategies and programs that offer worker compensation
Chain Forum: An International Journal (Vol. 18, No. 4, pp. 203-217). Taylor & Francis.
protection and reduce the impact of home-based accessibility to some Davani, A. (2021). To De-Congest Delivery Traffic, Operators and Cities Need to Come
segments of the population may be needed. Similar to other examples in Together. The City Fix.
de Mello, C. A., de Oliveira Carneiro, C. M., Maia, M. L. A., de Oliveira, L. K., & de
the growing shared economy (e.g., ride-hailing and bike-sharing), eq­
Araújo, G. G. F. (2022). Análise da taxa de entrega de aplicativos de entrega
uity-based programs may be necessary to reduce the burden on the most instantânea e a remuneração dos entregadores em cidades brasileiras. Transportes, 30
vulnerable users (Qian & Jaller, 2020; Rodier & Podolsky, 2017). (2), 2641-2641.
Similarly, there is a need for transparency on the actual compensa­ de Oliveira, L.K., de Oliveira, I.K., Nascimento, C.D.O.L., Cordeiro, C.H.D.O.L., Meira, L.
H., Rabay, L., da Silva, F.G.F., 2021. Identification of factors to improve the
tion for the drivers or couriers. Platforms should make it explicit how productivity and working conditions of motorcycle couriers in Belo Horizonte.
much of the paid amount will go toward the driver. Still, it is under­ Brazil. Case studies on transport policy 9 (4), 1737–1745.
standable that based on the ride-hailing experience, platforms might de Oliveira, L.K., Mello, C.A.D., Carneiro, C.M.D.O., Araújo, G.G.F.D., Costa, T.E.R.D.,
Maia, M.L.A., 2022. Identification of Factors that Influence the Delivery Fee Pricing
have additional mechanisms to compensate and incentivize them. As in On-Demand Delivery Service. Front. Future Transp. 39.
discussed elsewhere (de Oliveira et al., 2021; de Mello et al., 2022; Fickenscher L., & T. Wayt (2022). Fast-delivery Apps Face NYC Crackdown over ‘Dark-
Regan & Christie, 2023), compensation for workers might be low stores,’ Worker Safety. [Link]
crackdown-on-fast-delivery-apps-dark-stores-worker-safety/.
through these platforms, which is not consistent with the level of effort, Figliozzi, M., Unnikrishnan, A., 2021. Home-Deliveries before-during COVID-19
costs, or activity risks. Policies may be needed to address these issues. Lockdown: Accessibility, Environmental Justice, Equity, and Policy Implications.
However, it would be essential to consider whether these services are Transp. Res. Part D: Transp. Environ. 93.
Goldstein, H., Browne, W., Rasbash, J., 2002. Partitioning variation in multilevel models.
economically sustainable under the constraints of consumers’ willing­
Understanding Statistics: Statistical Issues in Psychology, Education, and the Social
ness to pay for them if the additional measures further increase the fees. Sciences 1 (4), 223–231.
Whether the restaurants would absorb some of the costs, the platform Goods, C., Veen, A., Barratt, T., 2019. “Is your gig any good?” Analysing job quality in
the Australian platform-based food-delivery sector. J. Ind. Relations 61 (4),
reduces its service fee or combinations of these.
502–527.
Besides these potential regulations, the results of this research will Hedeker, D., Gibbons, R.D., 1994. A random-effects ordinal regression model for
also support other researchers investigating operational aspects of food multilevel analysis. Biometrics. 50, 933–944.
delivery. The models provide empirical evidence of the contributing Helling, B. (2021). Is DoorDash Worth It For Drivers?.
Hirscheberg, C. Rajko, A., Schumacher, T., & Wrulich M. (2016). The changing market
factors to delivery fees and show the relationship with waiting times. for food delivery. mckinsey.
These two variables are critical in the current delivery service pricing Caitlyn Hitt. (2020). The Tasty History of Takeout and Delivery.
models. The findings can enhance the development of new and better Jaller, M., Rodier, C., Zhang, M., Lin, H., & Lewis, K. (2021). Fighting for Curb Space:
Parking, Ride-Hailing, Urban Freight Deliveries, and Other Users. UC Davis: National
static or dynamic pricing schemes for operators looking into maximizing Center for Sustainable Transportation. Retrieved from [Link]
operability and profitability and those interested in improving home- item/3jn371hw.

11
E. Pourrahmani et al. Transportation Research Interdisciplinary Perspectives 21 (2023) 100891

Joselow, M., 2020. Delivery Vehicles Increasingly Choke Cities with Pollution. E&E Qian, X., Jaller, M., 2020. Bikesharing, equity, and disadvantaged communities: A case
news. study in Chicago. Transportation Research Part A: Policy and Practice 140, 354–371.
Le, T.V., Ukkusuri, S.V., 2019. Crowd-shipping services for last mile delivery: Analysis Regan, A., Christie, N., 2023. Workers and the Post-COVID Transportation Gig Economy.
from American survey data. Transp. Res. Interdisciplinary Perspectives 1, 100008. In: Pandemic in the Metropolis. Springer, Cham, pp. 49–59.
Lewis, M., 2006. The effect of shipping fees on customer acquisition, customer retention, Rodier, C., & L. Podolsky (2017). Opportunities for Shared-Use Mobility Services in Rural
and purchase quantities. J. Retailing 82 (1), 13–23. Disadvantaged Communities in California’s San Joaquin Valley: Existing Conditions
Li, C., Mirosa, M., Bremer, P., 2020. Review of online food delivery platforms and their and Conceptual Program Development. UC Davis: National Center for Sustainable
impacts on sustainability. Sustainability 12 (14), 5528. Transportation. Retrieved from [Link]
Noah Lichtenstein. (2020). The hidden cost of food delivery. Schnieder, M., Hinde, C., West, A., 2022. Emission Estimation of On-Demand Meal
McCullagh, P., 1980. Regression models for ordinal data (with discussion). J. R. Stat. Delivery Services Using a Macroscopic Simulation. Int. J. Environ. Res. Public Health
Soc., Series B. 42, 109–142. 2022 (19), 11667.
Nakagawa, S., Schielzeth, H., O’Hara, R.B., 2013. A general and simple method for Schroder, C. (2022). DoorDash’s Strategy Made It the Most Successful Food Delivery
obtaining R2 from generalized linear mixed-effects models. Methods Ecol. Evol. 4 Service in the US.
(2), 133–142. Silaen, M., Crowley, D., Jaller, M., & Circella C. (2023). Exploring the Differences in
Nguyen, D.H., De Leeuw, S., Dullaert, W., Foubert, B.P., 2019. What is the right delivery Type of Commodities and Delivery Methods among Online Shoppers during the
option for you? Consumer preferences for delivery attributes in online retailing. Early Stage of the COVID-19 Pandemic. Transportation Research Board Annual
J. Business Logistics 40 (4), 299–321. Meeting. Washington D.C.
O’brien P., & P. Haeck (2022). ‘Dark Commerce’ Backlash Grows in Cities across Europe. Morgan Stanley. (2020). Can Food Delivery Apps Deliver Profits for Investors?.
[Link] Tong, T., Dai, H., Xiao, Q., Yan, N., 2020. Will dynamic pricing outperform? Theoretical
down/. analysis and empirical evidence from O2O on-demand food service market. Int. J.
Pahwa, A., Jaller, M., 2022. A cost-based comparative analysis of different last-mile Prod. Econ. 219, 375–385.
strategies for e-commerce delivery. Transp. Res. Part E: Log. Transp. Rev. 164, Xu, X. (2021). What are customers commenting on, and how is their satisfaction
102783. affected? Examining online reviews in the on-demand food service context. Decision
João Pedreda. (2020). Restaurant Delivery Pricing Strategies are Evolving to Survive the Support Systems, 142, [Link], Lang. (2009). Mixed Effects Models for Complex
Pressure from Delivery Platforms. Data. Chapman & Hall, Florida, USA.
Pourrahmani, E., Jaller, M., 2021. Crowdshipping in Last Mile Deliveries: Operational
Challenges and Research Opportunities. Socio-Econ. Plann. Sci. 78, 101063.

12