You are on page 1of 44

The impact of online grocery retail sales on food waste and

last-mile CO2 emissions in comparison to traditional


grocery retail

Master thesis in partial fulfillment of the requirements for obtaining an MSc degree in
Transport & Supply Chain Management

VU Amsterdam
School of Business and Economics

Name: Jill van der Tuin


Student number: 2599534
Date: June 30, 2021

First supervisor: dr. A.S. Eruguz


Second supervisor: prof. dr. W.E.H. Dullaert

Word count: 9,995


Abstract
Online grocery retail sales are expected to increase. Especially as a result of the current
coronavirus pandemic, more households are purchasing groceries online. Online grocery shows
the potential to reduce emissions related to food waste and transportation in comparison to
traditional grocery retail, but there is a greater need to study both elements in detail as most
studies focus on a single activity. Therefore, this study aims to investigate the impact of online
grocery retail sales on both food waste and last-mile CO2 emissions in comparison to offline
grocery retail. Food waste and last-mile transportation emissions are estimated using a
comparative simulation study in which an inventory and transportation system are simulated
using data from the literature as well as data collected at Dutch supermarkets. A sensitivity
analysis is performed to test multiple scenarios. The results show that online grocery retail could
reduce emissions from both food waste and last-mile transportation by 32% on average
compared to offline grocery retail. Furthermore, when looking at combined emissions, different
conclusions may be drawn than when looking at a single activity. This study intends to
contribute to the literature on the environmental implications of online grocery retail by
generating a more encompassing view by investigating the impacts of both inventory depletion
and last-mile transportation. For business practitioners, this study provides a model that can be
used to quantify environmental implications for their own cases and it shows the potential of
the concept of inventory pooling to reduce food waste.
Keywords: online grocery retail, food waste, last-mile transportation, simulation

2
1. Introduction
Revenue from online grocery sales in the Netherlands increased by 65% to 2.5 billion euros in
2020, owing largely to the current coronavirus pandemic1. Whereas online groceries had a 0.8%
market share five years ago, last year marked the first time that more than 5% of Dutch
households purchased their groceries online2. According to analysts from the Rabobank3, the
market share of online groceries will continue to grow to 15-20% by 2030. In general, e-grocery
is growing in popularity due to the convenience it provides for people to buy groceries from the
comfort of their own homes (Martín et al., 2019). At the same time, around one-third of all food
produced for human consumption in the world is lost or wasted (FAO, 2013). One of the
Sustainable Development Goals is to reduce global food waste per capita at the retail and
consumer levels by 20304. Therefore, it is worth investigating whether the growing trend of
online grocery retailing can help to achieve this goal by reducing food waste. In addition,
grocery shopping constitutes a significant amount of overall personal car traffic in absolute
terms and is likely to be a dedicated trip (Cairns, 2005). By replacing individual trips to the
grocery store with a more efficient delivery truck route, online grocery shopping has the
potential to eliminate some of that traffic (Belavina, 2016).

There is a large literature on quantifying and comparing general e-commerce to traditional


retailing in terms of environmental impact (Carrillo et al., 2014; Cullinane, 2009). However,
most papers quantify and compare the environmental impact of a single activity, such as last-
mile transportation, warehousing, or packaging (Mangiaracina et al., 2015). Based on this, no
firm conclusions can be drawn about whether e-commerce is more sustainable than traditional
retail. Furthermore, a key distinction between general retail and grocery retail is the perishable
nature of the products, which has environmental implications in terms of food waste. Previous
studies that attempted to quantify the environmental consequences of (online) grocery retailing
also primarily focused on either the carbon footprint of food waste or emissions related to
transportation (Hardi & Wagner, 2019; Scholz et al., 2015). To my knowledge, only the
working paper by Astashkina et al. (2019) has constructed comprehensive research into both
food waste and transport emissions from the advent of online grocery retailing. They propose a
three-echelon perishable inventory model with consumers, retailers and a supplier. Because
their model is so complex, they make numerous simplifications, such as in product lifetime,
lead time, and inventory depletion, which reduces its practical relevance. According to
Cicatiello et al. (2017), the grocery retailer is the point at which many different players in the
food chain collide. Moreover, in influencing both the features of food production and
consumers’ tastes, retailers play a major role. As a result, grocery retailers have a significant
impact on food waste and should be studied further. In terms of transportation, the most notable
difference in environmental impact between online and offline grocery retail can be found in
the last-mile phase (Gee et al., 2020). Therefore, in order to bridge the gap between studies that
focus on either food waste or transportation emissions, the goal of this study is to estimate both
food waste and last-mile CO2 emissions of online versus offline grocery retail where the retailer
is centralized. This leads to the following research question:

“What is the impact of online grocery retail sales on food waste and last-mile CO2 emissions
in comparison to traditional grocery retail?”

1
https://www.businessinsider.nl/online-boodschappenmarkt-ah-jumbo-picnic/
2
https://www.distrifood.nl/food-data/artikel/2020/05/corona-onlineomzet-piekt-op-51-procent-101134282
3
https://www.rabobank.nl/bedrijven/cijfers-en-trends/food/supermarkten/
4
https://sdgs.un.org/goals/goal12

3
The aim of this research is to assess the impact of online grocery retail sales on food waste and
last-mile CO2 emissions in comparison to offline grocery retail. The study's research aim is thus
not to optimize policies to reduce food waste and last-mile CO2 emissions, but rather to explore,
compare, and provide insight into which scenarios online grocery retail is more beneficial or
harmful to the environment in terms of food waste and last-mile CO2 emissions than offline
grocery retail. The method used in order to accomplish this research objective is mathematical
simulation, in which perishable inventory systems and transportation flows of both offline
brick-and-mortar grocery stores and an online pure-play grocery retailer are simulated. This
method is suitable for the research objective as it shows how systems respond when various
types of conditions are considered (Wacker, 1998). These systems are calibrated with data
collected from literature and statistical sources, as well as from first-hand collection at a Dutch
local and an online supermarket to enhance their practical relevance. This distinguishes the
study from other studies on perishable inventory systems because the majority of them consider
hypothesized parameters (Haijema & Minner, 2016; Minner & Transchel, 2010). Moreover,
this study considers the "inventory pooling" effect that online pure players have because
inventory is centralized in one location, as opposed to offline stores where inventory is
decentralized (Chopra & Meindl, 2015). During the literature review, almost no studies that
considered this phenomenon were found, but it was mentioned as a significant difference
between online and offline grocery retail (Hays et al., 2005). The study’s unit of analysis is CO2
emissions from both food waste and last-mile transportation. This way, the combined
environmental elements can be studied. Hence, this can be considered the theoretical
contribution of this study, as the combined effects of these two elements where the retailer is
centralized have not previously been studied.

The remainder of this paper is structured as follows. Section 2 reviews existing literature on
perishable inventory control and grocery retail, including its environmental implications. The
methodology for conducting this research is described in section 3, and the data collection and
simulation model is further discussed in section 4. Section 5 presents an overview of the results
and discusses them, followed by the conclusion of the research, limitations and suggestions for
future research in section 6.

4
2. Literature review
This section presents an overview of existing scientific research regarding perishable inventory
management and the area on which this research is focused: online versus offline grocery retail.
Moreover, the corresponding environmental implications will be discussed, as well as the
methods to quantify these. This overview provides the first step in answering the research
question and contributes to the body of literature that studies the environmental impacts of
online grocery retailing. The research approach for this literature review can be found in
Appendix A.
2.1 Perishable inventory management
In inventory control, most research assumes that products can be kept in stock for an indefinite
period of time. However, for food products such as dairy, meat and vegetables, this is not the
case. These types of products can be characterized as perishable goods: “one that has constant
utility up until an expiration date (which may be known or uncertain), at which point the utility
drops to zero” (Nahmias, 2011, p. viii). Due to the importance and challenges of managing
perishable inventory, the research in this area is quite extensive. The literature on inventory
models for perishable products can be classified based on the different assumptions they make
(see Table 1). There is a distinction between a product's lifetime (random or fixed) and how
demand is modeled (stochastic or deterministic) (Bakker et al., 2012; Goyal & Giri, 2001;
Janssen et al., 2016). Another important classification is the inventory control and
replenishment policy that is in place (Chaudhary et al., 2018; Nahmias, 1982).
Products with a fixed lifetime have a deterministic shelf life, i.e. after x days, the product is
considered to be outdated and has to be thrown away (Goyal & Giri, 2001). Nahmias (1975)
and Fries (1975) were the first ones to introduce periodic inventory models with a fixed lifetime
higher than two periods. They state that a lifetime higher than two periods makes the
computation of optimal policies very complex due to the curse of multidimensionality.
Therefore, simplifications of this assumption are suggested by authors in the field, e.g. to study
a perishable inventory system under a single critical number order policy (Cooper, 2001) or by
using only two periods for the lifetime (Chen et al., 2020). In contrast to the fixed lifetime
assumption, part of the literature on perishable inventory models considers products with a
random lifetime where the exact lifetime is not known while in stock, e.g. a probabilistic
lifetime (Liu, 1990).
According to Chaudhary et al. (2018), research on perishable inventory models is dominated
by papers on deterministic demand. Between 1990 and 2016, 257 papers were published on
deterministic demand, while 161 papers on stochastic demand were published. For example,
Ferguson et al. (2007) consider a variation of the Economic Order Quantity model with a
deterministic constant demand rate. An advantage of assuming deterministic demand is that in
an optimal policy, under fairly general conditions, order quantities will be such that no product
perishes (Nahmias, 1982). However, deterministic demand assumptions are not very realistic
as demand varies and is hard to predict. Therefore, stochastic demand would fit models that
want to mimic the real world better. In these models, demand is regarded as a random variable
that follows a particular type of probability distribution (Chaudhary et al., 2018).

5
The inventory control policy provides an answer to how much and when to order (Axsäter,
2015). The literature on perishable inventory can be classified as stock-level dependent or
stock-age dependent policies. Stock-level dependent policies determine the order quantity
based on the stock on hand without considering the age distribution of the inventory, whereas
a stock-age dependent policy does take this into account (Haijema & Minner, 2016). Several of
these papers consider a policy that continuously reviews inventory and places an order when
the inventory position falls below a certain level (Liu & Shi, 1999; Tekin, 2001). However, it
is not always feasible to continuously monitor the inventory. An alternative is to review the
inventory periodically at given points in time (Axsäter, 2015). Haijema and Minner (2016)
propose such a policy for perishable products where they assume stochastic demand, positive
lead time and lost sales. In their research, they compared different stock-level dependent
policies in terms of average costs. Unlike most authors, they vary the consumer issuing policies,
simulating both First In First Out (FIFO) and Last In First Out (LIFO). A FIFO issuing policy
assumes that the oldest item in stock is sold first, whereas a LIFO issuing policy assumes that
the freshest item is sold first (Minner & Transchel, 2010). Another interesting approach that
has received little attention in the perishable inventory literature is the service-level approach,
in which order quantities are determined to meet service-level constraints. According to Minner
and Transchel (2010), this approach is better suited for perishable inventory control instead of
the cost minimization approach, as optimal replenishment policies are complex and stockout
costs are difficult to implement. Both the studies by Broekmeulen and van Donselaar (2009)
and by Minner and Transchel (2010) developed stock-age dependent policies for perishables
with stochastic demand, positive lead times and lost sales where they take the service-level
approach. Both papers compare LIFO and FIFO issuing policies, but do not combine them. The
results show that the stock-age dependent policies outperform the base-stock and constant-order
policies. It should be noted that age-dependent policies are more computationally complex than
stock-level dependent policies because state-dependent order quantities must be determined in
each period (Minner & Transchel, 2010). Moreover, complete information on the age
distribution is required, which is not always feasible in real-world retail environments
(Broekmeulen & van Donselaar, 2009).

Table 1. Overview papers perishable inventory categorized by assumptions made


Lifetime Demand Replenishment Inventory
control
Deterministic

Continuous

Stock-level
dependent

dependent
Stochastic

Stock-age
Random

Periodic
Fixed

Literature
Nahmias (1975) ● ● ● ●
Fries (1975) ● ● ● ●
Cooper (2001) ● ● ● ●
Chen et al. (2020) ● ● ● ●
Liu (1990) ● ● ● ●
Ferguson et al. (2007) ● ● ● ●
Liu & Shi (1999) ● ● ● ●
Tekin (2001) ● ● ● ● ●
Haijema & Minner (2016) ● ● ● ●
Broekmeulen & van Donselaar (2009) ● ● ● ● ●
Minner & Transchel (2010) ● ● ● ● ●

6
2.2 Grocery retail and environmental implications
The main differences between online and offline grocery retail that have an environmental
effect, as identified in the literature, can be grouped into the following categories: inventory,
consumer behavior, transportation and retail operations (see Table 2).
For most items in grocery retail, the maximum lifetime is known beforehand and many can be
characterized as perishable goods (Janssen et al., 2018). According to van Donselaar et al.
(2006), perishable goods have a shelf life of less than 30 days. This assumption holds true in
both online and offline environments. However, traditional brick-and-mortar stores can sell
products close to or on the day they expire. Moreover, discounts can be applied by the retailer
to give customers an incentive to buy the nearly perished product before it needs to be thrown
out (Filimonau & Gherbin, 2017). This is not possible for the online sale of groceries. To
maintain customer satisfaction, online grocery retailers cannot send products that will expire
the next day and must send products that will be usable for a few more days (Fikar, 2018).
Consequently, online grocery retailers need to throw out products earlier than regular grocery
stores have to, which could lead to higher amounts of food waste. Another difference related to
inventory depletion is how the oldest products in inventory are handled. As in a regular store,
customers can see the expiration date and appearance of products, some customers prefer to
search for the freshest ones while some may choose the product in front (Janssen et al., 2016).
If a store handles active shelf management and applies the FIFO approach, the product in front
is the oldest one (Cicatiello et al., 2020). This indicates a combination of LIFO and FIFO
consumer product withdrawal. In an online setting, a FIFO inventory depletion can be
guaranteed as order pickers select the products instead of customers (Fikar, 2018).
Consequently, traditional grocery retailers might experience higher levels of food waste as older
products would expire sooner (Cicatiello et al., 2020). A last difference to mention concerns
the way the inventory is kept. For online pure players, who have no retail storefronts and just
one or a few warehouses, inventory "pooling" is possible where inventory is centralized and
can be managed more efficiently (Hays et al., 2005). This could result in less spoilage in
comparison to offline grocery retail where inventory is decentralized in both warehouses and
stores (Astashkina et al., 2019).
Besides LIFO and FIFO product withdrawal, another difference in terms of customer behavior
with environmental consequences is basket size. Basket size refers to the number of products a
customer buys in one transaction (Gee et al., 2020). According to Suel and Polak (2017), basket
size tends to be higher for online grocery sales, leading to higher levels of consumer food waste.
However, Belavina et al. (2016) state that the basket size depends on the revenue model that
the online grocery store handles, the per-order or the subscription model. When the per-order
model is handled, consumers pay a delivery fee for each order they place, leading to an increase
in the basket size and a lower order frequency. The subscription model leads to a decrease in
basket size and an increase in order frequency, as customers pay a fixed amount to receive free
delivery of their groceries. In general, higher basket size and a lower frequency of deliveries
lead to an increase in consumer food waste and a decrease in delivery-related emissions
(Belavina et al., 2016).

7
According to Gee et al. (2020), there are not many structural differences in up-stream
transportation between online and offline grocery retail that affect the environment. However,
there are differences to be found in the last-mile phase of grocery retail. In general, e-commerce
reduces environmental impact by replacing consumer trips to the store with consolidated
deliveries to consumers by delivery vans (Edwards et al., 2010; Gee et al., 2020) However,
there are numerous actors to consider, and various retail operations configurations are possible
(Gee et al., 2020). For example, online grocery retailers can operate as online pure players, but
they can also operate from existing brick-and-mortar grocery stores, picking and sending
deliveries from there, resulting in different last-mile transportation emissions
(Shahmohammadi et al., 2020). Furthermore, the mode of transportation used by customers to
get to the grocery store affects the differences in transportation emissions (Siragusa & Tumino,
2021).

Table 2. Overview of literature on differences between online and offline grocery retail and
environmental implications
Differences between online and offline Literature
grocery retail and environmental
implications
Inventory Astashkina et al. (2019), Cicatiello et al. (2020),
Fikar (2018), Filimonau & Gherbin (2017),
Hays et al. (2005), Janssen et al. (2016)
Consumer behavior Belavina et al. (2016), Gee et al. (2020), Suel &
Polak (2017)
Transportation and retail operations Edwards et al. (2010), Gee et al. (2020),
Shahmohammadi et al. (2020), Siragusa &
Tumino (2021)

2.3 Quantification of environmental implications of grocery retail


In the literature, food waste is measured by absolute quantities or by the carbon footprint of the
products (Cicatiello et al., 2017; Scholz et al., 2015). If a product is disposed, all resources and
greenhouse gas (GHG) emissions that flowed from the process of producing, storing and
transporting the product are wasted (Astashkina et al., 2019). This can be seen as the carbon
footprint and is expressed in kgCO2 equivalent (FAO, 2013). According to Scholz et al. (2015),
it is a good measure to quantify the environmental impact of food waste as measuring only the
mass does not provide sufficient information about the environmental consequences. The Life
Cycle Assessment (LCA) method is a popular method for calculating a product's carbon
footprint. For example, Brancoli et al. (2017) conducted an LCA for a Swedish supermarket to
quantify the environmental impacts of supermarket food waste, taking into account all
processes from the cradle to the grave. In contrast, the literature on inventory control policies
for perishables considers food waste, but not the emissions that are related to it (Teller et al.,
2018). It is more concerned with comparing various inventory control policies or model
characteristics, and then comparing the (simulated) quantities of food waste. There is therefore
room for research combining the grocery retailers’ inventory policies while taking into account
the food waste in terms of its carbon footprint.
As previously stated, the biggest difference in transportation between offline and online grocery
retailers is last-mile transportation. In the literature, quantification and comparison of last-mile
transport emissions of grocery delivery and customer pickup has received some attention.
Siikavirta et al. (2002) examined GHG emissions resulting from different home delivery
strategies for e-grocery and compared these findings to GHG emissions that would result from

8
customer pickup by using simulation and data from a grocery shop in Finland. The results show
that it is possible to reduce GHG emissions by 18-87% by substituting shopping trips by car for
grocery delivery, depending on the home delivery strategy used. The study by Hardi and
Wagner (2019) compared grocery delivery and customer pickup by using a modeling approach
based on real-world geodata and Monte Carlo simulations for a district in the city of Munich.
They identified and analyzed break-even points in terms of energy consumption and CO2
emissions. In addition, they considered the mode of transport for customers and the use of
electric vehicles. Their results show that there is potential to reduce energy use and CO2
emissions by substituting part of the individual shopping trips for grocery delivery. The working
paper by Asthaskina et al. (2019) on the environmental impacts of the advent of online grocery
retail adopts the method of Continuous Approximations (CA) for Vehicle Routing Problems
(VRP) formulated by Daganzo (1984) to estimate the average expected distance travelled by a
delivery van to deliver online grocery orders. The aim of CA-based methods is to efficiently
quantify performance metrics with minimal data using simplified analytical forms. For routing
problems, CA could be used to estimate the expected distance of near-optimal delivery routes
by employing geometric probability theory to derive these forms based on the delivery area
(Merchán & Winkenbach, 2019).

9
3. Methodology
This section describes the overall approach used to answer the research question of this study.
First, the model method is discussed. After that, the assumptions and calculations used to
approximate food waste and last-mile transportation emissions are listed.
3.1 Model method
This study used the method of mathematical simulation, which is commonly used in operations
to evaluate inventory and other organizational systems (Meredith et al., 1989). According to
Wacker (1998), the aim of this type of research is to see how models respond when different
types of conditions are considered. Therefore, this method is considered to be a good fit for this
study because the objective is to explore and provide insight into which different scenarios of
online grocery retail are more harmful or beneficial to the environment in terms of food waste
and last-mile CO2 emissions when compared to offline grocery retail. Moreover, simulation
should be used when solving a mathematical model analytically is too complex (Law & Kelton,
2000). Given that the goal of this study is to represent and compare real-world grocery retail
systems with perishable products with a fixed lifetime, deriving an analytical solution will be
complex (Nahmias, 1975), so simulation is a good fit for this study. As a result, this study has
a quantitative research design with an exploratory purpose.
Simulation involves both a conceptual model of what is happening, expressed by analytical
equations, and an element of reality by including realistic parameters (Meredith et al., 1989).
Therefore, this type of research is characterized by a positivistic approach, where the goal is to
explain and predict reality with objective considerations of reality (Mentzer & Kahn, 1995).
The parameters of this study were based on data derived from existing empirical studies and
statistical sources, as well as from first-hand data collection, which enhances the external
validity and includes the element of reality (Meredith et al., 1989).
In particular, the simulation model used in this study is characterized as a stochastic one as it
contains a probabilistic component (Law & Kelton, 2000). The daily demand for perishable
products was modelled as a random input parameter following a theoretical probability
distribution based on historical sales data from a local supermarket. This input parameter was
used to estimate the average food waste and its corresponding carbon footprint in CO2e, as well
as the average last-mile distance traveled and its corresponding CO2 emissions in an offline and
online grocery retail environment. In this way, the two aspects are comparable and joint effects
could be studied. Food waste was estimated by the means of simulating an inventory system,
which is characterized as a dynamic system because inventory levels change over time. The
last-mile distance traveled was estimated by simulating transportation flows, which is classified
as a static system because time does not play a role (Law & Kelton, 2000). The models used
various parameters and equations to differentiate between online and offline grocery retail, as
discussed in the following subsections.

10
3.2 Food waste emissions approximation
3.2.1 Assumptions and notation
For the estimation of food waste and its corresponding emissions, an inventory system with a
stock-level dependent, base-stock policy and periodic review was simulated since it reflects the
rationale behind most order systems used in supermarkets (van Donselaar et al., 2006).
According to Haijema and Minner (2016), costs associated with ordering and storing perishable
goods may be overlooked because the goods are replenished on a regular basis and overstocking
is avoided due to their limited shelf life. Therefore, no such costs were included in the model.
In this study, the service level approach proposed by Minner and Transchel (2010) was
followed, with the goal of determining an order-up-to level that satisfies service level
constraints. The fill rate (β-service level), which is the ratio of satisfied demand from stock,
was used for this purpose (Babiloni et al., 2017).
The inventory of a single perishable product is reviewed daily and each day t, when necessary,
an order qt is placed to raise the inventory position IPt to the order-up-to level S. This order
arrives after a positive deterministic lead time of L days. It is assumed that the supplier has
ample stock and that the products are placed directly on the shelves after arrival. During the
day, the inventory decreases by customer demand Dt. Demand that cannot be met is assumed
to be lost. Here, the first differences between the online and offline environments can be
applied. In this study, it was assumed that an online pure player in grocery retail replaces several
offline brick-and mortar grocery retail stores n with identically distributed demand. Depending
on the degree of demand dependency between offline stores, the online retailer can benefit from
"inventory pooling", which reduces demand uncertainty. If all n stores have demand that is
independent, which means the correlation coefficient between demand of the stores ρ is zero,
the online retailer benefits the most. The aggregated mean is expressed as 𝜇𝑎 = 𝑛𝜇 with an
aggregated standard deviation of 𝜎𝑎𝑖 = √𝑛𝜎. In case the demand is not independent, the
aggregated standard deviation becomes 𝜎𝑎𝑑 = √𝑛𝜎 2 + 2𝜌𝜎 2 𝑛(𝑛−1)
2
(Chopra & Meindl, 2015).
Another difference concerns the customer withdrawal of FIFO and LIFO. In the offline
environment, a fraction of 𝑓 ∈ [0,1] customers withdraw products according to FIFO principles
and a fraction of (1-f) customers withdraw products according to LIFO principles. In the online
environment, FIFO can be assured, so f is equal to one. In both cases, products that will expire
the next day are discarded at the end of the day, which is referred to as food waste FWt. A
representation of the sequence of daily events is visualized in Figure 1.
Because this study only includes products with a printed expiration date, the shelf life in days
m is known and is thus fixed. Here, the third difference between the online and offline
environment can be applied. For the online retailer, products have a guaranteed days of
freshness 𝑔 upon arrival to the customer. Therefore, 𝐼 = 𝑚 − 𝑔 represents the adjusted shelf
life for perishable products. For offline retailers, products can be sold on the last day of their
shelf life and therefore, 𝑔 = 0.

Realize and satisfy Remove any


Receive order placed Place new order qt
Start day demand Dt up to the expired products End day
in prior period qt-L if necessary
extent possible FWt from inventory

Figure 1. Sequence of daily events

11
3.2.2 System description
The state of the system can be defined as “the collection of state variables necessary to describe
the system at a particular time” (Law & Kelton, 2000, p. 3). The state of the system at review
𝑗
point t can be described by the state variable for the inventory level 𝑋𝑡 which consists of batches
of products of 𝑗 ∈ {0, … , 𝐼 − 1} days age and by (𝐿 − 1)+ outstanding orders 𝑞𝑡−1 , … , 𝑞𝑡−𝐿+1 .
Note that when 𝐿 ≤ 1, there are no outstanding orders (Haijema & Minner, 2016). This can be
seen as the inventory position at day t before replenishment and is expressed as:
𝐿−1 𝐼−1
𝑗
𝐼𝑃𝑡 = ∑ 𝑞𝑡−𝑖 + ∑ 𝑋𝑡
𝑖=1 𝑗=0

An order qt is placed when the difference between the order-up-to level S and inventory position
IPt is positive:
𝑞𝑡 = (𝑆 − 𝐼𝑃𝑡 )+
𝑗
The customer withdrawal 𝑊𝑡 of inventory of age j on day t can be derived by the following
recursive equation where the withdrawal is the minimum of the remaining batch size and the
sum of demand that could not be satisfied from older batches in case of f FIFO demand (i.e.,
age j+1,…, I-1) and the demand that could not be satisfied from fresher batches in case of (1-f)
LIFO demand (i.e. age 0,…, j-1) (Broekmeulen & van Donselaar, 2009). To avoid non-integer
customer withdrawals, a random rounding function is applied with an uniformly distributed
random number 𝑟 ∈ [0,1].
𝑗 + 𝑗−1 𝑖 +
𝑗
𝑀𝑖𝑛 {𝑋𝑡 , ((⌊𝑓 ∙ 𝐷𝑡 ⌋ − ∑𝐼−1 𝑖
𝑖=𝑗+1 𝑋𝑡 ) + (⌈(1 − 𝑓) ∙ 𝐷𝑡 ⌉ − ∑𝑖=0 𝑋𝑡 ) )} , 𝑗 = 0, … , 𝐼 − 1 𝑟 < 0.5
𝑊𝑡 = { .
𝑗 𝑖 + 𝑗−1 𝑖 +
𝑀𝑖𝑛 {𝑋𝑡 , ((⌈𝑓 ∙ 𝐷𝑡 ⌉ − ∑𝐼−1
𝑖=𝑗+1 𝑋𝑡 ) + (⌊(1 − 𝑓) ∙ 𝐷𝑡 ⌋ − ∑𝑖=0 𝑋𝑡 ) )} , 𝑗 = 0, … , 𝐼 − 1 𝑟 ≥ 0.5

At the end of the day, the remaining stock with an age of I-1 is removed from the inventory and
discarded:
𝐹𝑊𝑡 = 𝑋𝑡𝐼−1 − 𝑊𝑡𝐼−1
The carbon footprint in CO2e ew is used to estimate the food waste emissions on day t:
𝐸𝐹𝑊𝑡 = 𝐹𝑊𝑡 ∙ 𝑒𝑤
The fill rate FR over a period of T days can be estimated by the sum of all customer withdrawals
of age j divided by the sum of the demand:
𝑗
∑𝑇𝑖=1 ∑𝐼−1
𝑗=0 𝑊𝑖
𝐹𝑅 =
∑𝑇𝑖=1 𝐷𝑖

12
3.3 Last-mile transportation emissions approximation
This section explains how the average last-mile distances driven and corresponding emissions
were estimated for both the offline and online grocery retail environments. Figure 2 visualizes
the differences between online grocery delivery and offline customer pickup.

Figure 2. Last-mile distance grocery retail

Customers in the offline setting travel to the supermarket via their preferred mode of
transportation. In this study, the point of interest is the percentage of people p that use a private
vehicle for their grocery trips as the goal is to estimate the corresponding emissions. Following
Hardi and Wagner (2019), the trips to the supermarket do not need to be optimized as it is not
a circular tour. Therefore, the average total distance driven by cars in kilometers on day t is
expressed as follows:
𝑡
𝑇𝑜𝑓𝑓𝑙𝑖𝑛𝑒 = 2 ∙ 𝑑̅ ∙ 𝑝 ∙ 𝐶𝑡

Where 𝑑̅ is the average one-way distance in kilometers between a customer’s home and the
supermarket. Ct is the total number of customers on day t, which is equal to the daily demand
Dt of all products divided by the basket size B. In this case, the basket size stands for number
of perishable products expected in a basket.
The total emissions emitted by car trips with an emission factor of 𝑒𝑐𝑎𝑟 in kgCO2 per kilometer
on day t is defined as:
𝑡 𝑡
𝐸𝑇𝑜𝑓𝑓𝑙𝑖𝑛𝑒 = 𝑇𝑜𝑓𝑓𝑙𝑖𝑛𝑒 ∙ 𝑒𝑐𝑎𝑟

In the online setting, the pure-play grocery retailer delivers to customers using delivery vans
from a central distribution center. To estimate the distance driven and corresponding emissions,
the method of Daganzo (2005; 1984) was used, which utilizes simple formulas to predict the
distance traveled by fleets of vehicles in physical distribution problems involving a depot and
its area of influence. This method was chosen because it yields near-optimal solutions, requires
minimal data, and has been shown to approximate real-world distances. Moreover, Astashkina
et al. (2019) employ this method and their research objective is similar to that of this study.

13
First, the assumptions underlying this method will be explained, after which the calculations
will be presented. It is assumed that the distribution center is located in the middle of a
rectangular-shaped delivery area R. The online customers 𝐶𝑡 are uniformly distributed
throughout the delivery area and are delivered to by a fleet of vans from the distribution center,
each of which can serve 𝑄̅ customers. These delivery vans must visit all customers while
minimizing the total distance traveled, and they must start and end at the distribution center. To
account for actual road distances rather than straight line distances, the distances are measured
in Manhattan, L1-metric (Eilon et al., 1971). The average total distance of the routes driven by
the delivery vans in kilometers on day t depends on the amount of customers relative to the
capacity of the vehicles and is expressed as:
𝐶𝑡
2𝑙 + 0.73√𝑅 ∙ 𝐶𝑡 𝐶𝑡 > 𝑄̅ 2
𝑡
𝑇𝑜𝑛𝑙𝑖𝑛𝑒 = { 𝑄̅
𝐶𝑡 ≤ 𝑄̅ 2
0.95√𝑅 ∙ 𝐶𝑡
𝐶𝑡
In case 𝐶𝑡 > 𝑄̅ 2, the problem is a capacitated VRP problem, with the first term 2𝑙 𝑄̅
𝐶𝑡
approximating the “line-haul” portion of the route covered by delivery vans. The parameter
𝑄̅
l represents the average distance between the customers and the distribution center which can
(𝑎+𝑏)
be expressed as 𝑙 = where a and b are the sides of the rectangular delivery area R (Eilon
4
𝑏
et al., 1971). The “shape factor” has a significant effect on the distance when it is low, but
𝑎
according to Eilon et al. (1971) becomes relatively unimportant when it is between 0.6 and 1.6,
so this assumption was followed. The second term 0,73√𝑅 ∙ 𝐶𝑡 approximates the local distance
traveled between the customers. In case 𝐶𝑡 ≤ 𝑄̅ 2, the problem reduces to a Travelling Salesman
Problem (TSP) as there are only a few delivery van tours. Daganzo (2005, p. 102) justifies this
by arguing that in this case: “The length of the shortest tour in the area is close to the sum of
the optimal sub-area tours”.
The total emissions emitted by the delivery vans with an emission factor of 𝑒𝑣𝑎𝑛 in kgCO2 per
kilometer on day t is defined as:
𝑡 𝑡
𝐸𝑇𝑜𝑛𝑙𝑖𝑛𝑒 = 𝑇𝑜𝑛𝑙𝑖𝑛𝑒 ∙ 𝑒𝑣𝑎𝑛

14
4. Simulation modelling
This section starts by going over how the data used as input for the simulation model was
gathered. Then, the simulation setup is discussed, followed by the simulation scenarios and the
performance measures used to evaluate the scenarios.
4.1 Data collection
For the estimation of emissions related to food waste of both traditional and online grocery
retail, multiple input data were required. The first step was to gather data on the cradle-to-retail
carbon footprint of perishable retail products through a literature search. Clune et al. (2017)
performed a meta-analysis of LCA studies and provided an overview of the median values of
the carbon footprints of many different perishable retail products. This study was used as a
foundation, and other collected LCA studies or retail food waste studies were compared to those
values to see if they were similar. Based on that, the values shown in Table 3 were selected.
To gather data on the demand and shelf life of perishable products, a local Dutch supermarket
was visited. The goal was to collect data on a sample of perishable products with varying
demand, shelf life and carbon footprint characteristics to represent the perishable food
department in order to estimate food waste emissions. There was no data on actual demand
because this is difficult to track in supermarkets due to lost sales and substitution. According to
the supermarket's management, historical sales data are a reasonable proxy for demand.
Therefore, daily sales data for a selection of perishable products for the first quarter of 2021
was retrieved and used as the demand in this study. This resulted in 75 data points as the store
is closed on Sundays. Given that the supermarket only estimates expected sales based on data
from the previous five weeks for the purpose of reordering, this is a substantial amount of data
and sufficiently representative to use for estimations. Most importantly, the amount of data
should be sufficient to fit theoretical distributions and perform a goodness-of-fit test, such as
the Chi-square test (Law & Kelton, 2000). The procedure and details of this can be found in
Appendix B.
Subsequently, the sample mean and standard deviation were estimated in order to compute the
coefficient of variation (CV), which is useful for comparing variability across datasets (Lee et
al., 2013). When it is greater than one, it indicates that the variability is greater than the mean;
when it is less than one, it indicates that the relative variability is less (Frost, 2020). Therefore,
products with a CV less than one are considered to have a low-variance demand, while those
with a CV greater than one are considered to have a high-variance demand. The products all
have a one-day lead time. Concerning the shelf life, there are no agreements or data for
perishable products when they arrive at the supermarket. Therefore, this data was collected for
the selection of products by going to the supermarket every day for two weeks and keeping
track of the expiration date of the incoming products. From this, an approximation of the shelf
life could be made by taking the average. For online grocery retail, the shelf life is shorter due
to the guarantee of freshness and this data was retrieved from a Dutch online grocery retailer.
Taking into account the above-mentioned goal for the sample, a final selection of six products
was made that acts as input for the simulation models (see Table 3 and Table 4).

15
Table 3. Product selection and data
Product Mean of Standard Coefficient of Shelf Freshness Weight Carbon
daily deviation variation of life guarantee (kg) footprint per
𝝈 kg product
demand of daily daily demand (days) online grocery
(units) μ demand
𝝁
m (days) g (kgCO2e) 𝒆𝒘
(units) σ
Mushrooms 25.88 8.90 0.34 5 3 0.25 0.27a
Chicken slices 8.69 3.04 0.35 22 5 0.2 3.65a
Low-fat yogurt 9.95 4.46 0.45 14 7 1 1.31a
Demi crème fraiche 2.16 2.21 1.02 25 4 0.125 2.7b
Haricot verts 1.08 1.12 1.04 14 3 0.15 0.75a
Salmon fillet 1.55 1.73 1.12 4 2 0.25 3.97c
a
Clune et al. (2017) b Scholz et al. (2015) c Seves et al. (2016)

Table 4. Product selection and characteristics

Shelf life Low-variance High-variance


demand (CV < 1) demand (CV > 1)
Short (1-10 days) Mushrooms Salmon fillet
Medium (11-20 days) Low-fat yogurt Haricot verts
Long (21-30 days) Chicken slices Demi crème fraiche

For some of the input parameters, a baseline and a range needed to be established. A baseline
value in order to represent the “as-is” situation, and a range to test for other possibilities and to
include the baseline value’s uncertainty to evaluate its effects on emissions. The supermarket’s
management believes that roughly half of its customers seek out fresher items, so a baseline for
offline FIFO demand of 0.5 is set with a range adopted from Haijema and Minner (2016) of
[0,1]. According to Broekmeulen and van Donselaar (2009), a fill rate between 90-98% for
perishable products in grocery retail is reasonable, so this range was adopted for the service
level with a baseline of 95%. Because no data for the correlation coefficient between the
demand of the offline stores could be found, independent demand was assumed in the baseline
with a range for sensitivity of [0,1].
For the estimation of emissions related to last-mile transportation, multiple input data were
required. According to an expert opinion from a large Dutch supermarket, the number of offline
stores replaced by online distribution is 20 and this was used as a fixed parameter. As stated in
the literature review, opinions on the customer’s basket size for online grocery compared to
traditional grocery differ. Anesbury et al. (2016) found that on average, 12 products are bought
in a transaction and argued that it can be used for both environments because early adoption of
online grocery is assumed to be modest. However, this number includes non-perishable
products as well, in which food waste does not play a significant role. According to van
Donselaar et al. (2006), approximately 60% of products bought in supermarkets are perishables.
Therefore, 60% of 12 was taken as a basket size, which is approximately 7. This was used as a
fixed parameter because the basket size determines the number of customers in this study, so
the effect of varying the basket size on the emissions cannot be investigated. Data on emission
factors of vehicles was retrieved from the paper of Hardi and Wagner (2019) as their research
was also focused on grocery retail and it was the most recent paper found mentioning numbers
on emission factors. According to an expert from the headquarters of the local Dutch
supermarket, approximately 20 orders can be delivered in a single tour, which was used as a

16
baseline with a range of [5,40] (Astashkina et al., 2019). The Dutch Central Bureau of Statistics
(CBS) was used to obtain data on car use for grocery trips, the average distance from households
to a supermarket, and the size of the delivery area. On average, 47.5% of Dutch citizens use a
car for grocery trips, which was used as a baseline and for sensitivity analysis, a range of [0,1]
was used (CBS, 2020). The one-way distance between households and the nearest supermarket
is approximately 0.9 km. In a city, this distance reduces to 0.5 km and in a rural area it increases
to 1.5 km (CBS, 2021). The size of the delivery area was calculated by averaging the sizes of
Dutch cities classified as "large" by the CBS (2011), which is 60 km2. This size was doubled to
represent a rural area because this leads to a decrease in the density of households due to the
CA-based approximations. The baseline value of the delivery area was set in the middle. All
the parameters mentioned in this subsection can be found in Table 5 in section 4.3.
4.2 Simulation setup
The systems in the simulation can be characterized as steady-state systems as there is no natural
event to specify the end of a run (Law & Kelton, 2000). For steady-state simulation analysis,
Goldsman (2008) recommends using the method of batch means over independent replications
because of the possible initialization bias in each of the replications. Therefore, the batch means
method was adopted in this study. A long simulation run was split into a number of contiguous
batches, each of which contained a number of observations. If the size of each batch is large
enough, we can appeal to the central limit theorem and assume that the batch means are
normally distributed and approximately uncorrelated (Law & Kelton, 2000). To alleviate the
problem of initialization bias and to reach steady-state, the first 100 days were truncated, which
can be seen as the “warm-up” period. The inventory system started empty with zero inventory
and zero demand. A batch consisted of 1000 days and five batches were initialized. As the
service-level approach was followed, for each batch, the fill rate was calculated using equation
(6). After the initialization, a batch was added until the size of the 90% confidence interval for
the mean of the fill rate was smaller than 0.004 (Minner & Transchel, 2010). Then, for all
scenarios and products, the longest simulation run was used for comparison purposes. The
optimal order-up-to level S was determined by starting with a value of zero and increasing it
until the fill rate was reached. Thus, the minimum S to guarantee the achievement of the fill rate
(Babiloni et al., 2017). This procedure was followed in both offline and online environments,
employing the method of common random numbers by using the same seed in both. Otherwise,
the effect of changing one variable could be assigned to differences in random variables (Law
& Kelton, 2000). Because there are no objectives or constraints to fulfill in the transportation
system but the goal is to study the joint environmental effects of both food waste and last-mile
transportation, the transportation system adhered to the seed, batch size and number of batches
of the inventory system simulation. The software used for the simulation was Microsoft Excel.

17
4.3 Simulation scenarios
The goal of this simulation study is to compare the environmental impacts of online grocery
retail sales in terms of food waste and last-mile emissions to those of offline grocery retail. To
evaluate this, the six products from Table 4 and their demand, shelf life and carbon footprint
characteristics were analyzed in both offline and online cases. Baseline values for the “as-is”
situation and ranges were established for the parameters of both the inventory and transportation
systems as discussed in the previous sections (see Table 5). Considering that this study reviews
two systems and aims to study joint effects, the choice was made to perform a sensitivity
analysis including some extensions. In a sensitivity analysis, one parameter is varied while
setting the other parameters to baseline values in order to study the change in the response
variable, emissions in this case (Law & Kelton, 2000). An overview of the scenarios can be
found in Table 6.
Table 5. Parameters simulation

Parameters System Baseline Value Range


Lead time (days) L Inventory 1a Fixed
FIFO withdrawal (offline fraction) f Inventory 0.5 a [0, 0.25, 0.5, 0.75, 1]
Service level (online fraction) β Inventory 0.95 [0.9, 0.95, 0.98]b
Service level (offline fraction) β Inventory 0.95 [0.9, 0.95, 0.98]b
Correlation coefficient offline demand ρ Both 0 [0, 0.5, 1]
Number of offline stores n Both 20c Fixed
Basket size (units) B Transport 7d Fixed
Emission factor of gasoline cars Transport 0.1738e Fixed
(kgCO2/km) ecar
Emission factor of diesel delivery vans Transport 0.196e Fixed
(kgCO2/km) evan
Private car use for grocery trips Transport 0.475f [0, 0.25, 0.475, 0.75, 1]
(fraction) p
Delivery van capacity (deliveries per Transport 20g [5, 10, 20, 40]h
delivery van) 𝑄̅
Average distance from households to the Transport 0.9 [0.5, 0.9, 1.5]i
supermarket (km) 𝑑̅
Delivery area (km2) R Transport 90 [60, 90, 120]j
a
Local Dutch supermarket b Broekmeulen & van Donselaar (2009) c Expert opinion Dutch supermarket d Anesbury et al. (2016) & van Donselaar et al.
(2006) e Hardi & Wagner (2019) f CBS (2020) g Expert opinion headquarters local Dutch supermarket h Astashkina et al. (2019) i CBS (2021) j CBS (2011)

Table 6. Scenarios simulation


Scenario Parameter to be Number of Experiment per product
ranged experiments or aggregated
1. Customer behavior offline stores f (offline) 30 Per product
2. Customer satisfaction β (online and offline) 54 Per product
3. Dependency of demand offline stores ρ 21 Per product and aggregated
4. Private car use for grocery trips p 5 Aggregated
5. Delivery van capacity 𝑄̅ 4 Aggregated
6. City versus rural area 𝑑̅ and R 3 Aggregated
Total number of experiments 117

The first two scenarios concentrate on the input parameters of the inventory system. The first
scenario considered customer behavior in offline grocery retail stores. The parameter f was
ranged to simulate different fractions of offline customers picking products by LIFO or FIFO,
which was compared to the online environment where FIFO can be guaranteed. The second
scenario investigated the level of customer satisfaction. The parameter for the service level β

18
was varied in both the offline and online inventory systems to see the effect of maintaining a
different service level on the amount of emissions. The third scenario considers the effect of
demand dependency between offline stores by varying the correlation coefficient of offline
demand on online grocery emissions. The subsequent scenarios focus on the input parameters
of the transportation system where products are purchased in baskets. For the fourth scenario,
the percentage of people that use cars for grocery trips p was varied and compared to online
grocery delivery. In the fifth scenario, the capacity of the delivery van 𝑄̅ was varied as this
affects transport emissions in the online case, which was then compared to offline consumer
trips. The last scenario considered both city and rural areas by decreasing and increasing the
parameters of the average distance from households to supermarkets 𝑑̅ and the size of the
delivery area R to study the effect of different areas on transport emissions.
4.4 Performance measures
In order to compare the simulated offline and online grocery retail systems, it was necessary to
define performance measures. As the theoretical contribution of this research is to combine the
effects of both food waste and last-mile transport emissions and to compare the offline and
online grocery retail environment, both separate and joint performance measures were
established to adhere to this aim.
For the estimation of the expected emissions of food waste from the inventory system of both
the online and the offline environment, the average over b batches with a size of T days was
used. In the online case, the average daily emissions of food waste in CO2e could be estimated
as follows:
𝑡
∑𝑇𝑡=1 𝐸𝐹𝑊𝑜𝑛𝑙𝑖𝑛𝑒
∑𝑏𝑖=1 ( )
𝑇 𝑖
𝐴𝐸𝐹𝑊𝑜𝑛𝑙𝑖𝑛𝑒 (𝑏) =
𝑏
In case of the offline environment, the n identical stores were considered to estimate the average
daily emissions of food waste in CO2e.
𝑡
∑𝑇 𝐸𝐹𝑊𝑜𝑓𝑓𝑙𝑖𝑛𝑒
∑𝑏𝑖=1 ( 𝑡=1 )
𝑇
𝑖
𝐴𝐸𝐹𝑊𝑜𝑓𝑓𝑙𝑖𝑛𝑒 (𝑏) = 𝑛 ∙
𝑏
For the estimation of the expected emissions from last-mile transportation of both the online
and offline environment, the same procedure was followed as for the food waste. The average
daily emissions of driven distances in CO2 could be estimated as follows:
𝑡
∑𝑇𝑡=1 𝐸𝑇𝑜𝑛𝑙𝑖𝑛𝑒
∑𝑏𝑖=1 ( )
𝑇 𝑖
𝐴𝐸𝑇𝑜𝑛𝑙𝑖𝑛𝑒 (𝑏) =
𝑏
∑𝑇 𝐸𝑇 𝑡
∑𝑏𝑖=1 ( 𝑡=1 𝑜𝑓𝑓𝑙𝑖𝑛𝑒 )
𝑇
𝑖
𝐴𝐸𝑇𝑜𝑓𝑓𝑙𝑖𝑛𝑒 (𝑏) = 𝑛 ∙
𝑏
The joint performance measure for the online as well as for the offline environment is simply
the sum of the average emissions of food waste and the average emissions of distances driven.
Thus, the combined average daily emissions in CO2 could be estimated as follows:
19
𝐴𝐸𝑜𝑓𝑓𝑙𝑖𝑛𝑒 (𝑏) = 𝐴𝐸𝐹𝑊𝑜𝑓𝑓𝑙𝑖𝑛𝑒 (𝑏) + 𝐴𝐸𝑇𝑜𝑓𝑓𝑙𝑖𝑛𝑒 (𝑏)

𝐴𝐸𝑜𝑛𝑙𝑖𝑛𝑒 (𝑏) = 𝐴𝐸𝐹𝑊𝑜𝑛𝑙𝑖𝑛𝑒 (𝑏) + 𝐴𝐸𝑇𝑜𝑛𝑙𝑖𝑛𝑒 (𝑏)


In order to make valid comparisons between the online and offline grocery retail environments,
it is necessary to check if the difference in daily averages is significant. As variances are
considered to be unequal, the Welch t-test was used to accomplish this (Law & Kelton, 2000).
As a significance level, 𝛼 = 0.05 was implemented. The full description of the two-sample t-
test can be found in Appendix C.

20
5. Results and discussion
In this section, the main results of the simulation scenarios are reported in terms of the
performance measures as discussed in section 4.4. Following that, the findings will be
discussed. The results of the t-tests can be found in Appendix D.

5.1 Scenario 1: Customer behavior offline stores

Table 7 shows the average daily food waste emissions when different fractions of FIFO in the
offline environment are considered. When comparing the offline and online environments, it
becomes clear that food waste emissions associated with offline grocery retail are either equal
to, or significantly higher than those associated with online grocery retail for all products in all
FIFO fractions. Even when the fraction of customers withdrawing according to FIFO principles
is 100% and thus equal to the online case, food waste emissions are significantly higher for two
high variable demand products. As the offline FIFO fraction increases, the emissions remain
constant or decrease. The latter is a logical result of more customers purchasing older products,
resulting in less food waste.
Table 7: Average daily food waste emissions in kgCO2e
Fraction of FIFO demand f
0 0.25 0.5 0.75 1 1
Environment Offline Online
Products:
Mushrooms 1.518* 0.210* 0.003* 0.000 0.000 0.000
Low-fat yogurt 1.687* 0.001 0.000 0.000 0.000 0.000
Chicken slices 0.064* 0.000 0.000 0.000 0.000 0.000
Salmon fillet 24.340* 17.544* 16.474* 11.222* 9.500* 0.078
Haricot verts 0.165* 0.011* 0.010* 0.005* 0.001* 0.000
Demi crème fraiche 0.332* 0.005* 0.001* 0.000 0.000 0.000
* indicates a statistically significant difference compared to results online grocery with alpha of 0.05 (see
Appendix D)

Figure 3 shows the combined average daily emissions resulting from both food waste and last-
mile transportation of offline and online grocery retail. In all FIFO fractions, online grocery
becomes more environmentally beneficial than offline grocery.

60
Combined emissions in kgCO2

50

40

30

20

10

0
0 0.25 0.5 0.75 1
Fraction of offline FIFO demand f
Offline Online

Figure 3. Average daily food waste and last-mile transport


emissions in kgCO2

21
5.2 Scenario 2: Customer satisfaction

Table 8 shows the average daily food waste emissions when different service levels are
considered. When evaluating the impact of online grocery in comparison to offline grocery, it
becomes clear that food waste emissions are either similar or less in all service levels. What
stands out is that for five out of six products in the online environment, there are almost no
emissions related to food waste. As the service level increases from 0.9 to 0.98, the emissions
either remain constant or increase, with the latter being a logical result as the order-up-to-level
rises to meet the service level, likely to result in more food waste.
Table 8. Average daily food waste emissions in kgCO2e

Service level β
0.9 0.95 0.98
Environment Offline Online Offline Online Offline Online
Products:
Mushrooms 0.000 0.000 0.003* 0.000 0.029* 0.000
Low-fat yogurt 0.000 0.000 0.000 0.000 0.000 0.000
Chicken slices 0.000 0.000 0.000 0.000 0.000 0.000
Salmon fillet 10.102* 0.015 16.474* 0.078 23.520* 0.295
Haricot verts 0.003* 0.000 0.010* 0.000 0.026* 0.000
Demi crème fraiche 0.000 0.000 0.001* 0.000 0.005* 0.000
* indicates a statistically significant difference compared to the results of online grocery with
alpha of 0.05 (see Appendix D)

When taking into account average daily last-mile transport emissions, online grocery is still
more environmentally beneficial in all cases (see Figure 4). What stands out is that in the case
of offline grocery retail, the percentage increase in combined emissions from a service level of
0.9 to 0.98 is 41.7%, whereas for online grocery this is only 1.3%. This could be partially
explained by the fact that offline food waste emissions accounts for 41.8% of total emissions
on average, whereas online food waste emissions accounts for only 6.3%.

50
Combined emissions in kgCO2

40

30

20

10

0
0.9 0.95 0.98
Service level β
Offline Online

Figure 4. Average daily food waste and last-mile transport


emissions in kgCO2

5.3 Scenario 3: Dependency of demand offline grocery stores

Table 9 shows the average daily food waste emissions when different levels of dependency
between the demand of the offline grocery stores are considered. What stands out is that for
three out of six products, for all correlation coefficient levels, food waste emissions remain zero
for online grocery and are equal or less to the offline case. This means that even though demand

22
uncertainty increases to its extreme, there is still no food waste for online grocery retailers. Only
for the two products that have a short shelf life, there is an increase in emissions when the
correlation coefficient increases from zero to one, resulting in higher emissions for online
grocery compared to offline grocery. An increase in food waste emissions is a logical result of
the order-up-to level needing to increase to cope with the increasing demand uncertainty caused
by a higher dependency between the demand of offline stores in order to meet the service level.
Table 9. Average daily food waste emissions in kgCO2e
Correlation coefficient offline demand ρ
0 0.5 1
Environment Offline Online
Products:
Mushrooms 0.003 0.000* 0.064* 0.750*
Low-fat yogurt 0.000 0.000 0.000 0.000
Chicken slices 0.000 0.000 0.000 0.000
Salmon fillet 16.474 0.078* 15.142* 26.659*
Haricot verts 0.010 0.000* 0.000* 0.002*
Demi crème fraiche 0.001 0.000* 0.000* 0.000*
* indicates a statistically significant difference compared to the results of offline grocery
with alpha of 0.05 (see Appendix D)

Figure 5 shows that when taking into account the combined emissions, online grocery would
have less emissions than offline grocery up until a correlation coefficient of 0.5. In the extreme
case of fully dependent offline demand, online grocery becomes more environmentally harmful.

60
Combined emissions in kgCO2

50

40

30

20

10

0
0 0.5 1
Correlation coefficient offline demand ρ
Offline Online

Figure 5. Average daily food waste and last-mile transport


emissions in kgCO2

5.4 Scenario 4: Private car use for grocery trips

Figure 6a shows the average daily last-mile transport emissions when different fractions of car
use for grocery trips are considered. As could be expected from equation (7) and (8), there is a
linear relationship between the percentage of people using a car and the corresponding
emissions, meaning that if car users double, the emissions double as well. Furthermore, in the
base-scenario where car use is 47.5%, online grocery retail would have 5.7% less emissions
than the offline case. When private car use is 25% or less, online grocery retail becomes more
environmentally harmful in terms of last-mile transport emissions. Taking into account the
emissions related to food waste as well, online grocery has less combined emissions in all cases,
except for when no one uses a car for grocery shopping (see Figure 6b).

23
Last-mile transport emissions in kgCO2 70 70

Combined emissions in kgCO2


60 60

50 50

40 40

30 30

20 20

10 10

0 0
0 0.25 0.475 0.75 1 0 0.25 0.475 0.75 1
Private car use for grocery trips p Private car use for grocery trips p
Offline Online Offline Online

(a) Average daily last-mile transport emissions in kgCO2 (b) Average daily food waste and last-mile transport
emissions in kgCO2
Figure 6. Results scenario 4

5.5 Scenario 5: Delivery van capacity

Figure 7a shows the average daily last-mile transport emissions when different delivery van
capacities are considered. If a delivery van could only serve 10 or less customers, emissions
would rise by 94-215% if online grocery would replace traditional grocery. From a delivery
capacity of 20 onwards, online grocery delivery becomes more environmentally beneficial
compared to traditional grocery. A decrease in transportation emissions when delivery capacity
increases is a logical result because less trucks are needed resulting in less line-haul trips and
possibly more efficient routes. What stands out from the results is that from a delivery capacity
of 20 onwards, increasing delivery van capacity does not result in a reduction in emissions. This
could probably be assigned to the use of CA-based methods for calculating the distance where
the capacitated VRP reduces to a TSP if 𝐶𝑡 ≤ 𝑄̅ 2. When combined emissions from food waste
and last-mile transportation are considered, the results of which grocery retail environment is
more environmentally friendly do not change, but the magnitudes of differences do (see Figure
7b).

80 80
Last-mile transport emissions in kgCO2

70 70
Combined emissions in kgCO2

60 60

50 50

40 40

30 30

20 20

10 10

0 0
5 10 20 40 5 10 20 40
Delivery van capacity Ǭ Delivery van capacity Ǭ
Offline Online Offline Online

(a) Average daily last-mile transport emissions in kgCO2 (b) Average daily food waste and last-mile transport
emissions in kgCO2
Figure 7. Results scenario 5

24
5.6 Scenario 6: City versus rural area

Figure 8a shows the average daily last-mile transport emissions when considering different
areas. When a city and a rural area are compared, the results are diametrically opposed: in a
city, online grocery would increase transport emissions, whereas in a rural area, online grocery
would decrease emissions compared to offline grocery. When average daily food waste
emissions are taken into account, online grocery is more environmentally beneficial in all areas
(see Figure 8b).
60 60
Last-mile transport emissions in kgCO2

Combined emissions in kgCO2


50 50

40 40

30 30

20 20

10 10

0 0
City General Rural City General Rural
Area Area
Offline Online Offline Online

(a) Average daily last-mile transport emissions in kgCO2 (b) Average daily food waste and last-mile transport
emissions in kgCO2
Figure 8. Results scenario 6

5.7 Discussion

This study is aimed at assessing the impact of online grocery retail sales on food waste and last-
mile CO2 emissions in comparison to offline grocery retail. The results show that in the as-is
situation, but also when varying offline FIFO fractions and service levels, emissions related to
food waste decreased or remained the same when online grocery retail is considered. Only when
the correlation coefficient of offline demand exceeds 0.5 are emissions higher for one product
for online grocery compared to offline grocery, and for two products in the extreme case when
the offline demand is perfectly correlated. Despite the fact that online grocery retailers must
deal with a guarantee of freshness, which reduces shelf life, this study's findings indicate that
online grocery has the potential to reduce food waste emissions. This is consistent with the
literature, which states that online grocers can reduce food waste by improving inventory
management through inventory pooling, but also by eliminating customer behavior in the
physical purchasing process (Astashkina et al., 2019; Hays et al., 2005).

In the as-is situation, the results of average daily emissions related to last-mile transportation
differ slightly from what is expected in the literature. When online grocery delivery is
considered, transport emissions are reduced by only 5.7% compared to traditional grocery,
whereas similar studies show a reduction of 18-87% (Cairns, 2005; Siikavirta et al., 2002). This
discrepancy could be assigned to multiple possible reasons. To begin with, the studies of
Siikavirta et al. (2002) and Siragusa and Tumino (2021) assume that everyone uses a car for
grocery trips, which results in higher transportation emissions in the offline case. In real-life, it
is unlikely that everybody uses a car, but the fraction of people could differ per country. In the
United Kingdom, approximately 80% of people use a car for grocery shopping
(Shahmohammadi et al., 2020). This could result in different outcomes when this percentage is
assumed in the as-is situation, as can be seen in Figure 6a. Another possible explanation is

25
related to the capacity of a delivery van. The study of Hardi and Wagner (2019) assumes a
capacity of 200 deliveries, which is high considering grocery orders are large and bulky and,
according to expert views, are expected to be around 20 deliveries per van. This results in fewer
line-haul trips and possibly more efficient routes, potentially resulting in lower emissions for
online grocery delivery. The results of scenario 6, where different areas were considered, are
consistent with previous studies. According to Astashkina et al. (2019) and Siragusa and
Tamino (2021), switching to online grocery delivery can be harmful in a city area where
supermarkets are close to households, whereas it can be beneficial in a rural area where
households are located far from the stores.

The most important contribution of this study was to investigate the combined results of both
food waste and last-mile transport emissions. The results of all scenarios show that online
grocery retail could reduce combined emissions by 32% on average when compared to the
offline case. This is in line with the findings of Astashkina et al. (2019), who also focus on food
waste and transportation emissions and estimate an 8-41% reduction in total emissions if online
grocery retail is considered. Moreover, in some scenarios, the results change in terms of which
retail environment can be considered more environmentally friendly when combined emissions
are taken into account instead of a single activity. In addition, the magnitude of the differences
in emissions between online and offline grocery retail change.

26
6. Conclusion
Due to the upcoming trend of e-grocery and its potential benefits in terms of environmental
impact compared to traditional grocery retail, this study sought to find out the impact of online
grocery retail sales on both food waste and last-mile CO2 emissions in comparison to traditional
grocery retail. By using data collected from empirical studies and statistical sources as well as
from first-hand collection at a local supermarket and an online supermarket, a simulation study
was performed. Inventory systems were simulated in order to estimate daily average emissions
resulting from food waste, and transport flows were simulated in order to estimate daily average
emissions resulting from last-mile transportation for both the offline and online grocery retail
environments. Differences in these environments could be assigned to customer withdrawal,
shelf life, inventory centralization/decentralization and the way in which the products are
obtained. Different scenarios were tested by performing a sensitivity analysis and the results
were compared and tested for significance. Moreover, the combined results of both food waste
and last-mile transportation were evaluated as both were measured in CO2 emissions.

Findings show that online grocery retail has the potential to reduce food waste because food
waste emissions are either similar or less in most scenarios and ranges of parameters considered
compared to traditional grocery retail. An exception to this occurs when the dependency of the
demand of offline stores exceeds 0.5 for products with a short shelf life. The results from the
last-mile transport scenarios show that in the as-is situation, online grocery has a slight decrease
in terms of last-mile transport emissions. Moreover, online grocery retail shows the greatest
potential in rural areas. The results indicate that switching to online grocery retail could reduce
combined emissions from both waste and last-mile transportation by 32% on average.

This study emphasizes the importance of considering more activities when investigating the
environmental implications of a specific retail operation. When only a single activity is
considered, such as transportation or inventory depletion, sometimes different conclusions can
be drawn than when taking both into account. For example, online grocery may be more
environmentally harmful than offline grocery in terms of last-mile transportation emissions in
a city area, but when food waste is taken into account too, it becomes environmentally more
beneficial (see Figure 8). Therefore, this study contributes to the literature investigating the
environmental implications of the relatively new trend of online grocery retailing in a more
encompassing way by taking into account both food waste and last-mile transportation
emissions. This research will offer grocery retail practitioners the opportunity to use the model
presented in this study to investigate their own cases to gain insight into which retail operation
would be more environmentally beneficial, as the results show that different inputs could lead
to different outcomes. Moreover, the concept of inventory pooling investigated in this study
shows the potential to reduce food waste emissions when compared to decentralized inventory
when the dependency of the demand is below 0.5. This information could be useful when
making decisions on how to keep perishable inventory.

There are several limitations that could affect the outcomes of this study. The most significant
one is that, based on the study's findings, no firm conclusions can be drawn about whether
online grocery retail is more or less harmful to the environment in overall terms. This study did
provide a first step towards obtaining a more complete picture by considering both food waste
and last-mile transport emissions, but future research should include other activities where
differences in environmental implications can be identified. For example, in the storage of
perishable products. Traditional grocery retailers need to trade off food preservation and
customer experience by using open refrigerated displays, which use 30% more energy than

27
regular refrigerators, whereas e-grocery warehouses can be designed for optimal storing
purposes (Fricke & Becker, 2010). Moreover, this study assumed complete substitution of the
offline grocery channel for online grocery retail, which might not be realistic. Therefore, future
research should consider different levels of adoption of online grocery retail or even a
combination to investigate the environmental implications. Finally, by performing a sensitivity
analysis, cross-effects could not be studied, nor could it be determined which factor had the
greatest influence on food waste and last-mile transportation emissions. This information may
be useful because some factors, such as the capacity of delivery vans, can be actively managed
in order to possibly reduce environmental consequences. Therefore, a full-factorial design is
recommended to extend the findings of this study. Future research could also look in the effects
of different retail operations, such as an omnichannel grocery retailer, as this could affect the
environmental implications.

This study shows that online grocery retail has the potential to reduce emissions from food
waste and last-mile transportation when compared to traditional grocery retail. As the market
share of online groceries is expected to grow, more emphasis should be placed on research that
investigates the environmental consequences of this trend, taking more than one activity into
account.

28
References
Adan, I., van Eenige, M., & Resing, J. (1995). Fitting discrete distributions on the first two

moments. Probability in the Engineering and Informational Sciences, 9(4), 623–632.

https://doi.org/10.1017/S0269964800004101

Anesbury, Z., Nenycz-Thiel, M., Dawes, J., & Kennedy, R. (2016). How do shoppers behave

online? An observational study of online grocery shopping. Journal of Consumer

Behaviour, 15(3), 261–270. https://doi.org/10.1002/cb.1566

Astashkina, E., Belavina, E., & Marinesi, S. (2019). The environmental impact of the advent

of online grocery retailing (SSRN Scholarly Paper ID 3358664). Social Science

Research Network. https://doi.org/10.2139/ssrn.3358664

Axsäter, S. (2015). Inventory control (Vol. 225). Springer International Publishing.

https://doi.org/10.1007/978-3-319-15729-0

Babiloni, E., Guijarro, E., & Cardós, M. (2017). A reference framework to design inventory

policies using a fill rate criterion in lost sales contexts. In M. Amorim, C. Ferreira, M.

Vieira Junior, & C. Prado (Eds.), Engineering Systems and Networks (pp. 109–116).

Springer International Publishing. https://doi.org/10.1007/978-3-319-45748-2_12

Bakker, M., Riezebos, J., & Teunter, R. H. (2012). Review of inventory systems with

deterioration since 2001. European Journal of Operational Research, 221(2), 275–

284. https://doi.org/10.1016/j.ejor.2012.03.004

Belavina, E., Girotra, K., & Kabra, A. (2016). Online grocery retail: Revenue models and

environmental impact. Management Science, 63(6), 1781–1799.

https://doi.org/10.1287/mnsc.2016.2430

Brancoli, P., Rousta, K., & Bolton, K. (2017). Life cycle assessment of supermarket food

waste. Resources, Conservation and Recycling, 118, 39–46.

https://doi.org/10.1016/j.resconrec.2016.11.024

29
Broekmeulen, R. A. C. M., & van Donselaar, K. H. (2009). A heuristic to manage perishable

inventory with batch ordering, positive lead-times, and time-varying demand.

Computers & Operations Research, 36(11), 3013–3018.

https://doi.org/10.1016/j.cor.2009.01.017

Cairns, S. (2005). Delivering supermarket shopping: More or less traffic? Transport Reviews,

25(1), 51–84. https://doi.org/10.1080/0144164042000218391

Carrillo, J. E., Vakharia, A. J., & Wang, R. (2014). Environmental implications for online

retailing. European Journal of Operational Research, 239(3), 744–755.

https://doi.org/10.1016/j.ejor.2014.05.038

Central Bureau for Statistics (CBS). (2011). Bevolkingskernen 2011; steden en dorpen in

Nederland [Dataset].

https://opendata.cbs.nl/statline/#/CBS/nl/dataset/82271NED/table?dl=55690

Central Bureau for Statistics (CBS). (2020, October 1). Mobiliteit; per persoon,

vervoerwijzen, motieven, regio's [Dataset].

https://opendata.cbs.nl/statline/#/CBS/nl/dataset/84710NED/table?dl=55692

Central Bureau for Statistics (CBS). (2021, February 1). Nabijheid voorzieningen; afstand

locatie, regionale cijfers [Dataset].

https://opendata.cbs.nl/statline/#/CBS/nl/dataset/80305ned/table?dl=556E4

Chaudhary, V., Kulshrestha, R., & Routroy, S. (2018). State-of-the-art literature review on

inventory models for perishable products. Journal of Advances in Management

Research, 15(3), 306–346. https://doi.org/10.1108/JAMR-09-2017-0091

Chen, J., Tian, Z., & Hang, W. (2020). Optimal ordering and pricing policies in managing

perishable products with quality deterioration. International Journal of Production

Research, 0(0), 1–23. https://doi.org/10.1080/00207543.2020.1766715

30
Chopra, S., & Meindl, P. (2015). Supply chain management: Strategy, planning, and

operation (6th ed). Pearson.

Cicatiello, C., Blasi, E., Giordano, C., Martella, A., & Franco, S. (2020). “If only I could

decide”: Opinions of food category managers on in-store food waste. Sustainability,

12(20), 8592. https://doi.org/10.3390/su12208592

Cicatiello, C., Franco, S., Pancino, B., Blasi, E., & Falasconi, L. (2017). The dark side of

retail food waste: Evidences from in-store data. Resources, Conservation and

Recycling, 125, 273–281. https://doi.org/10.1016/j.resconrec.2017.06.010

Clune, S., Crossin, E., & Verghese, K. (2017). Systematic review of greenhouse gas

emissions for different fresh food categories. Journal of Cleaner Production, 140,

766–783. https://doi.org/10.1016/j.jclepro.2016.04.082

Cooper, W. L. (2001). Pathwise properties and performance bounds for a perishable inventory

system. Operations Research, 49(3), 455–466.

https://doi.org/10.1287/opre.49.3.455.11216

Cullinane, S. (2009). From bricks to clicks: The impact of online retailing on transport and the

environment. Transport Reviews, 29(6), 759–776.

https://doi.org/10.1080/01441640902796364

Daganzo, C. (2005). Logistics systems analysis (4th ed). Springer.

Daganzo, C. F. (1984). The distance traveled to visit N points with a maximum of C stops per

vehicle: An analytic model and an application. Transportation Science, 18(4), 331–

350. https://doi.org/10.1287/trsc.18.4.331

Denyer, D., & Tranfield, D. (2009). Producing a systematic review. In The Sage handbook of

organizational research methods (pp. 671–689). Sage Publications Ltd.

Edwards, J. B., McKinnon, A. C., & Cullinane, S. L. (2010). Comparative analysis of the

carbon footprints of conventional and online retailing: A “last mile” perspective.

31
International Journal of Physical Distribution & Logistics Management, 40(1/2),

103–123. https://doi.org/10.1108/09600031011018055

Eilon, S., Watson-Gandy, C. D. T., & Christofides, N. (1971). Distribution management-

mathematical modelling and practical analysis. Griffin.

FAO. (2013). Food wastage footprint: Impacts on natural resources: summary report. FAO.

Ferguson, M., Jayaraman, V., & Souza, G. C. (2007). Note: An application of the EOQ model

with nonlinear holding cost to inventory management of perishables. European

Journal of Operational Research, 180(1), 485–490.

https://doi.org/10.1016/j.ejor.2006.04.031

Fikar, C. (2018). A decision support system to investigate food losses in e-grocery deliveries.

Computers & Industrial Engineering, 117, 282–290.

https://doi.org/10.1016/j.cie.2018.02.014

Filimonau, V., & Gherbin, A. (2017). An exploratory study of food waste management

practices in the UK grocery retail sector. Journal of Cleaner Production, 167, 1184–

1194. https://doi.org/10.1016/j.jclepro.2017.07.229

Fricke, B., & Becker, B. (2010). Energy use of doored and open vertical refrigerated display

cases. International Refrigeration and Air Conditioning Conference, 9.

Fries, B. E. (1975). Optimal ordering policy for a perishable commodity with fixed lifetime.

Operations Research, 23(1), 46–61.

https://doi-org.vu-nl.idm.oclc.org/10.1287/opre.23.1.46

Frost, J. (2020). Coefficient of Variation in Statistics. Statistics by Jim.

https://statisticsbyjim.com/basics/coefficient-variation/

Gee, I. M., Heard, B. R., Webber, M. E., & Miller, S. A. (2020). The future of food:

Environmental lessons from e-commerce. Environmental Science & Technology,

54(23), 14776–14784. https://doi.org/10.1021/acs.est.0c01731

32
Goldsman, D. (2008). Introduction to simulation. 26–37.

https://doi.org/10.1109/WSC.2007.4419585

Goyal, S. K., & Giri, B. C. (2001). Recent trends in modeling of deteriorating inventory.

European Journal of Operational Research, 134(1), 1–16.

https://doi.org/10.1016/S0377-2217(00)00248-4

Haijema, R., & Minner, S. (2016). Stock-level dependent ordering of perishables: A

comparison of hybrid base-stock and constant order policies. International Journal of

Production Economics, 181, 215–225. https://doi.org/10.1016/j.ijpe.2015.10.013

Hardi, L., & Wagner, U. (2019). Grocery delivery or customer pickup—Influences on energy

consumption and CO2 emissions in Munich. Sustainability, 11(3), 641.

https://doi.org/10.3390/su11030641

Hays, T., Keskinocak, P., & de López, V. M. (2005). Strategies and challenges of internet

grocery retailing logistics. In Applications of Supply Chain Management and E-

Commerce Research (pp. 217–252). Springer US.

https://doi.org/10.1007/0-387-23392-X_8

Janssen, L., Claus, T., & Sauer, J. (2016). Literature review of deteriorating inventory models

by key topics from 2012 to 2015. International Journal of Production Economics,

182, 86–112. https://doi.org/10.1016/j.ijpe.2016.08.019

Janssen, L., Sauer, J., Claus, T., & Nehls, U. (2018). Development and simulation analysis of

a new perishable inventory model with a closing days constraint under non-stationary

stochastic demand. Computers & Industrial Engineering, 118, 9–22.

https://doi.org/10.1016/j.cie.2018.02.016

Law, A. M., & Kelton, D. (2000). Simulation modeling and analysis: Vol. Third edition.

McGraw-Hill.

33
Lee, C.-F., Lee, J. C., & Lee, A. C. (2013). Statistics for business and financial economics.

Springer New York. https://doi.org/10.1007/978-1-4614-5897-5

Liu, L. (1990). (S, S) continuous review models for inventory with random lifetimes.

Operations Research Letters, 9(3), 161–167.

https://doi.org/10.1016/0167-6377(90)90014-V

Liu, L., & Shi, D. H. (1999). An (s, S) model for inventory with exponential lifetimes and

renewal demands. Naval Research Logistics (NRL), 46(1), 39–56.

https://doi.org/10.1002/(SICI)1520-6750(199902)46:1<39::AID-NAV3>3.0.CO;2-G

Mangiaracina, R., Marchet, G., Perotti, S., & Tumino, A. (2015). A review of the

environmental implications of B2C e-commerce: A logistics perspective. International

Journal of Physical Distribution & Logistics Management, 45(6), 565–591.

https://doi.org/10.1108/IJPDLM-06-2014-0133

Martín, J. C., Pagliara, F., & Román, C. (2019). The research topics on e-grocery: Trends and

existing gaps. Sustainability, 11(2), 321. https://doi.org/10.3390/su11020321

Mentzer, J. T., & Kahn, K. B. (1995). A framework of logistics research. Journal of Business

Logistics, 16(1), 231–250.

Merchán, D., & Winkenbach, M. (2019). An empirical validation and data-driven extension

of continuum approximation approaches for urban route distances. Networks, 73(4),

418–433. https://doi.org/10.1002/net.21874

Meredith, J. R., Raturi, A., Amoako-Gyampah, K., & Kaplan, B. (1989). Alternative research

paradigms in operations. Journal of Operations Management, 8(4), 297–326.

https://doi.org/10.1016/0272-6963(89)90033-8

Minner, S., & Transchel, S. (2010). Periodic review inventory-control for perishable products

under service-level constraints. OR Spectrum, 32(4), 979–996.

https://doi.org/10.1007/s00291-010-0196-1

34
Nahmias, S. (1975). Optimal ordering policies for perishable inventory-II. Operations

Research, 23(4), 735–749.

Nahmias, S. (1982). Perishable inventory theory: A review. Operations Research, 30(4), 680–

708. https://doi.org/10.1287/opre.30.4.680

Nahmias, S. (2011). Perishable inventory systems (Vol. 160). Springer US.

https://doi.org/10.1007/978-1-4419-7999-5

Saunders, M., Lewis, P., & Thornhill, A. (2015). Research methods for business students

(Seventh edition.). Pearson Education Limited; WorldCat.org.

Scholz, K., Eriksson, M., & Strid, I. (2015). Carbon footprint of supermarket food waste.

Resources, Conservation and Recycling, 94, 56–65.

https://doi.org/10.1016/j.resconrec.2014.11.016

Seves, S. M., Temme, E. H. M., Brosens, M. C. C., Zijp, M. C., Hoekstra, J., & Hollander, A.

(2016). Sustainability aspects and nutritional composition of fish: Evaluation of wild

and cultivated fish species consumed in the Netherlands. Climatic Change, 135(3–4),

597–610. https://doi.org/10.1007/s10584-015-1581-1

Shahmohammadi, S., Steinmann, Z. J. N., Tambjerg, L., van Loon, P., King, J. M. H., &

Huijbregts, M. A. J. (2020). Comparative greenhouse gas footprinting of online versus

traditional shopping for fast-moving consumer goods: A stochastic approach.

Environmental Science & Technology, 54(6), 3499–3509.

https://doi.org/10.1021/acs.est.9b06252

Siikavirta, H., Punakivi, M., Kärkkäinen, M., & Linnanen, L. (2002). Effects of e‐commerce

on greenhouse gas emissions: A case study of grocery home delivery in Finland.

Journal of Industrial Ecology, 6, 83–97.

https://doi.org/10.1162/108819802763471807

35
Siragusa, C., & Tumino, A. (2021). E-grocery: Comparing the environmental impacts of the

online and offline purchasing processes. International Journal of Logistics Research

and Applications, 1–27. https://doi.org/10.1080/13675567.2021.1892041

Suel, E., & Polak, J. W. (2017). Development of joint models for channel, store, and travel

mode choice: Grocery shopping in London. Transportation Research Part A: Policy

and Practice, 99, 147–162. https://doi.org/10.1016/j.tra.2017.03.009

Tekin, E. (2001). Age-based vs. Stock level control policies for a perishable inventory system.

European Journal of Operational Research, 21.

https://doi.org/10.1016/S0377-2217(00)00250-2

Teller, C., Holweg, C., Reiner, G., & Kotzab, H. (2018). Retail store operations and food

waste. Journal of Cleaner Production, 185, 981–997.

https://doi.org/10.1016/j.jclepro.2018.02.280

van Donselaar, K., van Woensel, T., Broekmeulen, R., & Fransoo, J. (2006). Inventory

control of perishables in supermarkets. International Journal of Production

Economics, 104(2), 462–472. https://doi.org/10.1016/j.ijpe.2004.10.019

Wacker, J. G. (1998). A definition of theory: Research guidelines for different theory-building

research methods in operations management. Journal of Operations Management,

16(4), 361–385. https://doi.org/10.1016/S0272-6963(98)00019-9

36
Appendixes
Appendix A: Research approach literature review
This appendix describes in detail the approach used to gather and review the literature included
in this study. In particular, it provides a stepwise overview of all stages of the critical literature
review. According to Saunders et al. (2015), a critical literature review is of high importance as
it provides the context and theoretical framework to further guide your research.
The first phase of the critical literature review was to get familiarized with the main domain
behind the topic: inventory control of perishable goods. This was done by reviewing theses that
were already written on the topic and by collecting and reading five systematic literature
reviews on perishable inventory management that covered the years 1964 until 2016. Moreover,
both the theses and the literature reviews were used to collect papers on perishable inventory
control policies. This method is characterized as backward snowballing and these papers were
selected after looking them up in the online University Library of the VU or Google Scholar
and reading the abstracts to identify their relevance. If a paper seemed relevant, the introduction
and conclusion were read and based on that, the paper was selected or not. The emphasis during
this selection was both on collecting papers on inventory control policies that showed similarity
with real-life implications, but also on collecting different types of inventory control policies
with different assumptions to provide an overview of the literature on perishable inventory
control policies. The quality of the papers was assessed by ensuring that they were peer
reviewed or published in a high-quality journal of which the AI was checked in Eigenfactor.
The selected papers were collected in the reference manager Zotero, fully read and summarized
in an Excel file to provide an overview and a steady information base before the literature
review was written. In total, 11 papers on perishable inventory control policies were selected
for the literature review. Moreover, for some more background information, two books on both
inventory control and perishable inventory systems were included.
The second phase was to collect literature specifically focused on online and offline grocery
retail and its environmental implications. The structured and systematic approach described by
Denyer and Tranfield (2009) was followed to ensure replicability and minimize bias. To locate
the studies, it is necessary to define keywords, formulate a search string and run the search
string into a database. Initial research on the topic of online and offline grocery retail and its
environmental implications was performed in Google Scholar to find relevant keywords and
synonyms used in the literature. Hereafter, the following search string was formulated using
Boolean logic to combine the keywords: TS=("shelf life" OR outdate OR perish* OR "food
waste" OR "food spoilage" OR emission* OR "carbon footprint" OR "environmental impact")
AND TS=("food retail" OR supermarket OR grocery OR "online grocery" OR e-grocery OR
"grocery delivery"). This search string was entered into both the online University Library of
the VU and in the ISI Web of Science database. The initial selection criteria were that the
articles needed to be written in English and published later than 1995 due to the advent of online
grocery retailing5. This resulted in 902 publications that were evaluated based on their titles
and, if they seemed relevant, the abstract was read. As with the previous search, the introduction
and conclusion of papers that still seemed relevant were read and based on that were selected.
The emphasis during this selection was on collecting papers related to the differences between

5
https://www.forbes.com/sites/laurendebter/2020/05/12/webvan-borders-founder-louis-borders-online-grocery-
home-delivery-service/

37
online and offline grocery retail, papers on online grocery retailing and its environmental
implications and papers that combined these. Also, some papers on general e-commerce and its
environmental implications were selected as papers specifically focused on e-grocery were
limited. The selected papers were again checked for their quality, collected in Zotero and were
fully read. For some papers that were of high relevance to the literature review, both forward
and backward snowballing was performed, which resulted in some more relevant papers which
were included. For the analysis and synthesis of the papers, a thematic analysis was used. After
summarizing and analyzing the collected papers, four themes were established into which the
papers could be divided: general online and offline grocery retail, general environmental
implications, food waste and transportation. Again, Excel was used for this and it provided an
overview of the existing literature on online and offline grocery retail and its environmental
implications, as well as a steady information base prior to writing the literature review. In total,
27 papers were selected from the second phase of the literature search.

38
Appendix B: Probability distribution fitting and Chi-square test
To determine the probability distribution for the random demand variables for the six products
(see Table 3), the procedure of Adan et al. (1995), in which discrete distributions are fitted on
the first two moments, was adopted. The method to fit a theoretical distribution to the historical
sales data of the six different products was preferred over using the empirical distribution as the
latter may have certain irregularities and theoretical distributions smooths out the data in case
of smaller datasets. Moreover, it is not possible to generate values outside the range of the
historical data by using the empirical distribution (Law & Kelton, 2000). By using the mean
𝜎2 1
and the variance of the demand of the products, the variable 𝛼 = 𝜇2 − 𝜇 could be calculated of
which the value determines the discrete probability distribution. The range of 𝛼 and
corresponding discrete probability distribution can be found in Table 10. This procedure was
followed for both the offline single store demand based on historical data and the online
aggregated demand.

Table 10. Fitting procedure Adan (1995)


Range for α Discrete probability distribution
-1 < α < 0 Binomial distribution
α=0 Poisson distribution
0<α<1 Negative binomial distribution
α≥1 Geometric distribution

Based on the summary statistics of the products from Table 3 for the offline single store
demand, the alphas were calculated and are listed in Table 11 including the corresponding
probability distribution. Hereafter, a Chi-square goodness-of-fit test was performed in order to
determine if the fitted distribution was indeed a good fit to the data. In other words, to examine
whether the data follows a specific probability distribution based on the frequency distribution
(Lee et al., 2013). The Chi-squared statistic and degrees of freedom were extracted by using the
fitdistrplus package in Rstudio, an open source software for data science, to test the fitted
probability distribution on the historical data. The Chi-squared statistics were compared to the
critical value at a 0.05 significance level. As shown in Table 11, all Chi-squared statistics are
lower than the critical values which indicates that there is no reason to believe that the historical
sales data are not fitted well by the assigned probability distribution by the procedure of Adan
et al. (1995). Moreover, the Chi-square test requires that the expected frequency for each
category is higher or equal to five, which was achieved for all products and it highlights the
sufficiency of the amount of data (Law & Kelton, 2000).

Table 11. Probability distribution fitting and Chi-square values of historical sales data

Product α estimate Probability Chi- Degrees of Critical value at


by Adan distribution squared freedom 0.05 significance
(1995) based on alpha statistic level
Mushrooms 0.0797 Negative binomial 4.024 6 12.592
Chicken slices 0.0076 Negative binomial 1.604 4 9.488
Low-fat yogurt 0.1005 Negative binomial 1.560 4 9.488
Demi crème fraiche 0.5800 Negative binomial 2.835 2 5.991
Haricot verts 0.1576 Negative binomial 2.177 1 3.841
Salmon fillet 0.6000 Negative binomial 1.759 2 5.991

39
For the aggregated mean of demand and corresponding aggregated standard deviation of
demand for the online grocery retailer, the procedure of Adan (1995) was followed as well.
This was done for all levels of the correlation coefficient ρ of the identically distributed demand
of the offline stores. Table 12, Table 13 and Table 14 show the alphas and the corresponding
probability distribution based on Table 10.

Table 12: Aggregated demand online grocery retail and probability distributions (ρ = 0)
Product Aggregated Aggregated α estimate Probability
mean of standard by Adan distribution
daily deviation of daily (1995) based on alpha
demand 𝝁𝒂 demand 𝝈𝒂𝒊
Mushrooms 517.60 39.80 0.0040 Negative binomial
Chicken slices 173.80 13.60 0.0004 Negative binomial
Low-fat yogurt 199.00 19.95 0.0050 Negative binomial
Demi crème fraiche 43.20 9.88 0.0290 Negative binomial
Haricot verts 21.60 5.01 0.0079 Negative binomial
Salmon fillet 31.00 7.74 0.0300 Negative binomial

Table 13. Aggregated demand online grocery retail and probability distributions (ρ = 0.5)
Product Aggregated Aggregated α estimate Probability
mean of standard by Adan distribution
daily deviation of daily (1995) based on alpha
demand 𝝁𝒂 demand 𝝈𝒂𝒅
Mushrooms 517.60 129.02 0.0602 Negative binomial
Chicken slices 173.80 44.12 0.0586 Negative binomial
Low-fat yogurt 199.00 64.63 0.1005 Negative binomial
Demi crème fraiche 43.20 31.97 0.5244 Negative binomial
Haricot verts 21.60 16.29 0.5225 Negative binomial
Salmon fillet 31.00 25.02 0.6219 Negative binomial

Table 14. Aggregated demand online grocery retail and probability distributions (ρ = 1)
Product Aggregated Aggregated α estimate Probability
mean of standard by Adan distribution
daily deviation of daily (1995) based on alpha
demand 𝝁𝒂 demand 𝝈𝒂𝒅
Mushrooms 517.60 178.07 0.1164 Negative binomial
Chicken slices 173.80 60.89 0.1169 Negative binomial
Low-fat yogurt 199.00 89.19 0.1960 Negative binomial
Demi crème fraiche 43.20 44.12 1.0198 Geometric
Haricot verts 21.60 22.48 1.0372 Geometric
Salmon fillet 31.00 34.53 1.2138 Geometric

40
Appendix C: Welch’s t-test
The Welch t-test is characterized as an independent two-sample t-test. In this study, it is used
to check whether the difference in averages between the offline and online environment are
significantly different from one another.
The procedure for Welch’s t-test, as described by Law & Kelton (2000), is followed. Let 𝑋𝑖𝑗 be
the average of the simulation configuration i and batch j and 𝑏𝑖 the number of batches of the
simulation configuration i. The simulation configurations are the online and the offline
environment in this study of which the number of batches is equalized for comparison purposes.
The grand average over all batches 𝑋̅𝑖 (𝑏𝑖 ) is computed as

∑𝑏𝑗=1
𝑖
𝑋𝑖𝑗 (14)
𝑋̅𝑖 (𝑏𝑖 ) =
𝑏𝑖
And the estimator of the variance 𝑆𝑖2 (𝑏𝑖 ) of each of the simulation configuration i as:

∑𝑏𝑗=1
𝑖
[𝑋𝑖𝑗 − 𝑋̅𝑖 (𝑏𝑖 )]2
𝑆𝑖2 (𝑏𝑖 ) = (15)
𝑏𝑖 − 1
The estimated degrees of freedom for i = 1,2 (online, offline) can be computed as:
2
𝑆 2 (𝑏 ) 𝑆 2 (𝑏 )
[ 1 1 + 2 2 ]
𝑏1 𝑏2
𝑓̂ = 2 (𝑏 )/𝑏 2 (16)
(𝑆 ) (𝑆 2 (𝑏 )/𝑏 )2
[ 1 1 1 + 2 2 2 ]
(𝑏1 − 1) (𝑏2 − 1)
To calculate the t-statistic, the following equation is used:
𝑋̅1 − 𝑋̅2
𝑡= (17)
𝑆 𝑆 2 2
√ 1+ 2
𝑏1 𝑏2

When |𝑡| > 𝑡𝑓̂,1−𝛼/2 , the null hypothesis of equal means is rejected. In this case, it can be
concluded that the difference between the averages of the simulation configurations is
statistically significant at a 0.05 significance level.

41
Appendix D: Welch’s t-test results
The results of the Welch’s t-tests for all scenarios of the inventory and the transport systems
can be found in Table 15 and Table 16.

Table 15. T-test results scenarios 1-3 (inventory system)


Grand Grand Variance Variance
mean daily mean daily daily daily
emissions emissions emissions emissions Degrees T-critical
Parameter Parameter offline in online in offline in online in Number of value two-
Scenario Product offline online kgCO2 kgCO2 kgCO2 kgCO2 of batches freedom T-statistic tail Significant?
1-f 0 1 1.5178 5.06E-06 0.0131 2.05E-09 80 79 118.4392 1.9905 Yes
1-f 0.25 1 0.2098 5.06E-06 0.0017 2.05E-09 80 79 45.2536 1.9905 Yes
1-f Mushrooms 0.5 1 0.0029 5.06E-06 1.68E-05 2.05E-09 80 79 6.3225 1.9905 Yes
1-f 0.75 1 3.38E-05 5.06E-06 9.11E-08 2.05E-09 80 83 0.8406 1.9890 No
1-f 1 1 0 5.06E-06 0 2.05E-09 80 79 -1.0000 1.9905 No
1-f 0 1 0.0639 0 0.0017 0 80 79 13.6973 1.9905 Yes
1-f 0.25 1 0 0 0 0 80 - - - -
Chicken 80
1-f slices 0.5 1 0 0 0 0 - - - -
1-f 0.75 1 0 0 0 0 80 - - - -
1-f 1 1 0 0 0 0 80 - - - -
1-f 0 1 1.6869 0 0.1199 0 80 79 43.5718 1.9905 Yes
1-f 0.25 1 0.0007 0 1.69E-05 0 80 79 1.4233 1.9905 No
Low-fat 80
1-f yogurt 0.5 1 0 0 0 0 - - - -
1-f 0.75 1 0 0 0 0 80 - - - -
1-f 1 1 0 0 0 0 80 - - - -
1-f 0 1 0.3324 0 0.0040 0 80 79 47.2460 1.9905 Yes
1-f Demi 0.25 1 0.0051 0 6.63E-05 0 80 79 5.5599 1.9905 Yes
1-f crème 0.5 1 0.0005 0 4.35E-06 0 80 79 2.1699 1.9905 Yes
1-f fraiche 0.75 1 0 0 0 0 80 - - - -
1-f 1 1 0 0 0 0 80 - - - -
1-f 0 1 0.1652 0 0.0003 0 80 79 88.6102 1.9905 Yes
1-f 0.25 1 0.0111 0 3.01E-05 0 80 79 18.0060 1.9905 Yes
Haricot 80
1-f verts 0.5 1 0.0098 0 2.82E-05 0 79 16.5241 1.9905 Yes
1-f 0.75 1 0.0053 0 1.97E-05 0 80 79 10.7803 1.9905 Yes
1-f 1 1 0.0008 0 2.58E-06 0 80 79 4.3885 1.9905 Yes
1-f 0 1 24.3095 0.0785 0.3657 0.0006 80 79 358.0569 1.9905 Yes
1-f 0.25 1 17.5443 0.0785 0.3601 0.0006 80 79 260.0883 1.9905 Yes
Salmon 80
1-f fillet 0.5 1 16.4739 0.0785 0.4024 0.0006 79 230.9898 1.9905 Yes
1-f 0.75 1 11.2217 0.0785 0.4217 0.0006 80 79 153.3679 1.9905 Yes
1-f 1 1 9.4995 0.0785 0.3801 0.0006 80 79 136.5540 1.9905 Yes
2-β 0.9 0.9 0.0002 0 1.05E-06 0 80 79 1.6207 1.9905 No
2-β 0.95 0.95 0.0029 5.06E-06 1.68E-05 2.05E-09 80 79 6.3225 1.9905 Yes
2-β 0.98 0.98 0.0285 7.17E-05 0.0003 1.71E-07 80 79 15.2056 1.9905 Yes
2-β 0.9 0.95 0.0002 5.06E-06 1.05E-06 2.05E-09 80 79 1.5750 1.9905 No
2-β Mushrooms 0.9 0.98 0.0002 7.17E-05 1.05E-06 1.71E-07 80 104 0.9222 1.9830 No
2-β 0.95 0.9 0.0029 0 1.68E-05 0 80 79 6.3339 1.9905 Yes
2-β 0.95 0.98 0.0029 7.17E-05 1.68E-05 1.71E-07 80 81 6.1462 1.9897 Yes
2-β 0.98 0.9 0.0285 0 0.0003 0 80 79 15.2487 1.9905 Yes
2-β 0.98 0.95 0.0285 5.06E-06 0.0003 2.05E-09 80 79 15.2460 1.9905 Yes
2-β 0.9 0.9 0 0 0 0 80 - - - -
2-β 0.95 0.95 0 0 0 0 80 - - - -
2-β 0.98 0.98 0 0 0 0 80 - - - -
2-β 0.9 0.95 0 0 0 0 80 - - - -
Chicken 80
2-β slices 0.9 0.98 0 0 0 0 - - - -
2-β 0.95 0.9 0 0 0 0 80 - - - -
2-β 0.95 0.98 0 0 0 0 80 - - - -
2-β 0.98 0.9 0 0 0 0 80 - - - -
2-β 0.98 0.95 0 0 0 0 80 - - - -
2-β Low-fat 0.9 0.9 0 0 0 0 80 - - - -
2-β yogurt 0.95 0.95 0 0 0 0 80 - - - -

42
2-β 0.98 0.98 0 0 0 0 80 - - - -
2-β 0.9 0.95 0 0 0 0 80 - - - -
2-β 0.9 0.98 0 0 0 0 80 - - - -
2-β 0.95 0.9 0 0 0 0 80 - - - -
2-β 0.95 0.98 0 0 0 0 80 - - - -
2-β 0.98 0.9 0 0 0 0 80 - - - -
2-β 0.98 0.95 0 0 0 0 80 - - - -
2-β 0.9 0.9 0 0 0 0 80 - - - -
2-β 0.95 0.95 0.0005 0 4.35E-06 0 80 79 2.1699 1.9905 Yes
2-β 0.98 0.98 0.0053 0 8.27E-05 0 80 79 5.2282 1.9905 Yes
2-β Demi 0.9 0.95 0 0 0 0 80 - - - -
2-β crème 0.9 0.98 0 0 0 0 80 - - - -
2-β fraiche 0.95 0.9 0.0005 0 4.35E-06 0 80 79 2.1699 1.9905 Yes
2-β 0.95 0.98 0.0005 0 4.35E-06 0 80 79 2.1699 1.9905 Yes
2-β 0.98 0.9 0.0053 0 8.27E-05 0 80 79 5.2282 1.9905 Yes
2-β 0.98 0.95 0.0053 0 8.27E-05 0 80 79 5.2282 1.9905 Yes
2-β 0.9 0.9 0.0030 0 8.86E-06 0 80 79 9.1293 1.9905 Yes
2-β 0.95 0.95 0.0098 0 2.82E-05 0 80 79 16.5241 1.9905 Yes
2-β 0.98 0.98 0.0261 0 8.09E-05 0 80 79 25.9750 1.9905 Yes
2-β 0.9 0.95 0.0030 0 8.86E-06 0 80 79 9.1293 1.9905 Yes
Haricot 80
2-β verts 0.9 0.98 0.0030 0 8.86E-06 0 79 9.1293 1.9905 Yes
2-β 0.95 0.9 0.0098 0 2.82E-05 0 80 79 16.5241 1.9905 Yes
2-β 0.95 0.98 0.0098 0 2.82E-05 0 80 79 16.5241 1.9905 Yes
2-β 0.98 0.9 0.0261 0 8.09E-05 0 80 79 25.9750 1.9905 Yes
2-β 0.98 0.95 0.0261 0 8.09E-05 0 80 79 25.9750 1.9905 Yes
2-β 0.9 0.9 10.1020 0.0145 0.2971 8.76E-05 80 79 165.4940 1.9905 Yes
2-β 0.95 0.95 16.4739 0.0785 0.4024 0.0006 80 79 230.9898 1.9905 Yes
2-β 0.98 0.98 23.5205 0.2945 0.4999 0.0028 80 80 292.9944 1.9901 Yes
2-β 0.9 0.95 10.1020 0.0785 0.2971 0.0006 80 79 164.2927 1.9905 Yes
Salmon 80
2-β fillet 0.9 0.98 10.1020 0.2945 0.2971 0.0028 81 160.1641 1.9897 Yes
2-β 0.95 0.9 16.4739 0.0145 0.4024 8.76E-05 80 79 232.0493 1.9905 Yes
2-β 0.95 0.98 16.4739 0.2945 0.4024 0.0028 80 80 227.3298 1.9901 Yes
2-β 0.98 0.9 23.5205 0.0145 0.4999 8.76E-05 80 79 297.3375 1.9905 Yes
2-β 0.98 0.95 23.5205 0.0785 0.4999 0.0006 80 79 296.3654 1.9905 Yes
3-ρ - 0 0.0029 5.06E-06 1.68E-05 2.05E-09 80 79 6.3225 1.9905 Yes
3-ρ Mushrooms - 0.5 0.0029 0.0635 1.68E-05 0.0005 80 84 -23.0714 1.9886 Yes
3-ρ - 1 0.0029 0.7505 1.68E-05 0.0099 80 79 -67.2152 1.9905 Yes
3-ρ - 0 0 0 0 0 80 - - - -
Chicken 80
3-ρ slices - 0.5 0 0 0 0 - - - -
3-ρ - 1 0 0 0 0 80 - - - -
3-ρ - 0 0 0 0 0 80 - - - -
Low-fat 80
3-ρ yogurt - 0.5 0 0 0 0 - - - -
3-ρ - 1 0 0 0 0 80 - - - -
3-ρ Demi - 0 0.0005 0 4.35E-06 0 80 79 2.1699 1.9905 Yes
3-ρ crème - 0.5 0.0005 0 4.35E-06 0 80 79 2.1699 1.9905 Yes
3-ρ fraiche - 1 0.0005 0 4.35E-06 0 80 79 2.1699 1.9905 Yes
3-ρ - 0 0.0098 0 2.82E-05 0 80 79 16.5241 1.9905 Yes
Haricot 80
3-ρ verts - 0.5 0.0098 4.22E-06 2.82E-05 1.42E-09 79 16.5166 1.9905 Yes
3-ρ - 1 0.0098 0.0022 2.82E-05 4.99E-06 80 106 11.7657 1.9826 Yes
3-ρ - 0 16.4739 0.0785 0.4024 0.0006 80 79 230.9898 1.9905 Yes
Salmon 80
3-ρ fillet - 0.5 16.4739 15.1423 0.4024 0.2594 151 14.6410 1.9758 Yes
3-ρ - 1 16.4739 26.6590 0.4024 0.5021 80 156 -95.7873 1.9753 Yes

43
Table 16. T-test results scenarios 3-6 (transport system)
Grand mean Grand Variance Variance
daily mean daily daily daily
emissions emissions emissions emissions Number Degrees T-critical
Parameter Parameter offline in online in offline in online in of of value two-
Scenario offline online kgCO2 kgCO2 kgCO2 kgCO2 batches freedom T-statistic tail Significant?
3-ρ - 0 22.2469 20.9748 0.0951 0.0012 80 81 36.67 1.9897 Yes
3-ρ - 0.5 22.2469 20.7597 0.0951 0.0118 80 98 40.67 1.9845 Yes
3-ρ - 1 22.2469 20.6119 0.0951 0.0250 80 118 42.20 1.9803 Yes
4-p 0 - 0 20.9748 0 0.0012 80 79 -5525.20 1.9905 Yes
4-p 0.25 - 11.7089 20.9748 0.0263 0.0012 80 86 -499.74 1.9879 Yes
4-p 0.475 - 22.2469 20.9748 0.0951 0.0012 80 81 36.67 1.9897 Yes
4-p 0.75 - 35.1267 20.9748 0.2371 0.0012 80 80 259.30 1.9901 Yes
4-p 1 - 46.8355 20.9748 0.4216 0.0012 80 79 355.75 1.9905 Yes
5-𝑄̅ - 5 22.2469 70.1334 0.0951 0.0403 80 136 -1163.68 1.9776 Yes
5-𝑄̅ - 10 22.2469 43.1187 0.0951 0.0132 80 101 -567.16 1.9837 Yes
5-𝑄̅ - 20 22.2469 20.9748 0.0951 0.0012 80 81 36.67 1.9897 Yes
5-𝑄̅ - 40 22.2469 20.9748 0.0951 0.0012 80 81 36.67 1.9897 Yes
6-𝑑̅ and R 0.5 60 12.3594 17.1258 0.0294 0.0008 80 83 -245.62 1.9890 Yes
6-𝑑̅ and R 0.9 90 22.2469 20.9748 0.0951 0.0012 80 81 36.67 1.9897 Yes
6-𝑑̅ and R 1.5 120 37.0781 24.2196 0.2642 0.0015 80 80 223.09 1.9901 Yes

44

You might also like