Impacts of Behavioral, Organizational, and Spatial Factors on the Carbon Footprint of Traditional Retail and E-commerce in the Paris Region

Abstract

Carbon footprint assessment of retail is necessary to optimize procurement strategies and adopt sustainable shopping habits. However, estimating carbon footprints is a complex task, given the diversity of existing distribution channels. Average values for carbon emissions of “conventional” retail (i.e., purchasing and receiving the product directly at the physical point of sale) found in most studies mask a heterogeneous reality: different retail strategies entail diverse shopping behavior for consumers, as well as varied procurement processes for outlets. In this paper, we propose a methodology to assess greenhouse gas (GHG) impacts of different distribution systems related to the consumption of goods in the Paris Region by coupling traditional transport modeling with a life-cycle assessment (LCA) approach. We model and compare six distribution systems, including five traditional retail formats (hypermarkets, supermarkets, small generalist retail, small food retail, and small nonfood retail) and E-commerce home deliveries. Our model includes warehouse activity, shop and home delivery, shop energy consumption, consumer mobility, and goods packaging. Overall, we conclude that E-commerce emits fewer GHG emissions than retail outlets per kilogram of product purchased. This result is in line with the existing literature on the topic. However, the carbon footprint varies greatly within the case study depending on the characteristics of the logistics procurement processes of outlets, the behavior of shoppers, and spatial characteristics.

Short abstract

In the literature, carbon footprint assessments of traditional retail and E-commerce result in different outcomes depending on assumptions. Our results show the importance of considering behavioral, organizational, and spatial characteristics related to retail.

1. Introduction

There are an increasing number of ways for consumers to purchase goods. The so-called “on-demand economy”¹ has caused E-commerce deliveries to grow rapidly in various forms. This growth has accelerated recently following the COVID pandemic.² However, to date, physical retail remains the most common means of purchasing goods globally.³ Along with the growth of E-commerce, retail formats have evolved, which has led to a diversification in the organization of brick-and-mortar stores.⁴ The increasing complexity of the retail landscape has led scholars studying urban mobility and city logistics to focus on the environmental effects of distribution systems. Hence, the environmental impact of E-commerce deliveries and how it compares to traditional retail has been a topic of discussion in recent studies.⁵⁻⁷

In this paper, we adopt a broad scope of assessment, which is crucial for a comprehensive carbon footprint analysis but can be lacking even in recent papers on the topic. We integrate the life cycle (LC) of distribution, from warehouse to consumer, including transportation, packaging, and buildings. To achieve this, we combine classic transport modeling methods (demand generation, trip distribution, mode choice⁸) with a life-cycle assessment (LCA) approach. Our model, calibrated for the Paris Region, which is our case study, allows us to consider the impacts of diverse distribution systems. Their carbon footprints depend on consumer behavior, procurement organization, and spatial characteristics. This contribution adds new complexity to the debate about the comparison between traditional retail and E-commerce. In Section 2 of this paper, drawing from an exhaustive literature review, we elaborate on our research objectives and hypotheses. We provide an overview of our empirical strategy for combining transport modeling with LCA, and we describe our input data. In Section 3, we outline and discuss our findings.

2. Materials and Methods

2.1. Literature Review

The environmental impact of retail versus E-commerce distribution has been an important topic in the literature since E-commerce began to emerge as a significant distribution channel in the late 1990s.⁹ Based on existing literature reviews in Mangiaracina et al.,⁵ Rai et al.,³ and Siragusa and Tumino,¹⁰ we have compiled and analyzed a corpus of 29 documents addressing the issue. To limit the length of this article, we present the full results of the review in Supporting Information 1 (SI1). The findings of this analysis are synthesized as follows:

• While a majority of studies (19 out of 29) determine that E-commerce has lower GHG impacts than traditional retail, there is no consensus as to the magnitude of impacts in terms of GHG emissions. To understand these inconsistent outcomes, it is essential to evaluate the methodologies of the assessments because they present significant differences in scope and rely on disparate assumptions and entry data.

• Most studies (25 out of 29) consider a single product or product class to assess the impacts of different distribution systems.

• Almost all papers (25 out of 29) consider a comparison of E-commerce vs “traditional” or “conventional” retail, disregarding the heterogeneity of physical retail distribution formats, despite the existence of a body of literature¹¹⁻¹⁴ that shows that different stores rely on different logistics organizations for their restocking, which results in distinct shipment frequencies, sizes, vehicles used, and distances traveled.

• Most studies do not consider the internal spatial heterogeneity of mobility behaviors and logistics organization within their chosen case study. This means that, for example, all modeled consumers are attributed one transport mode split or that stores are attributed one distance traveled for deliveries. In reality, there is always internal heterogeneity in a case study. Failing to account for this heterogeneity may lead to modeling an oversimplified “average” situation, which erases the specificity of different configurations within the case study.

• Papers that measure LC impacts of buildings, packaging, fuel, and vehicles do it in similar ways, relying on generic assumptions (for example, quantity of packaging per quantity of product or per item, or energy consumption per square meter of buildings’ floor space with simple allocation rules defining an average impact per kilogram or per item, etc.), which is in line with standard LCA methodologies applied to complex systems such as urban mobility.

In the following subsection, we explain how to address these gaps.

2.2. Description of the Case Study: Retail in the Paris Region (Ile-de-France)

Using the Paris Region as a case study is interesting in its own right. The Paris Region has 12.3 million inhabitants and 6.3 million jobs over an area of 12,000 km², i.e., an average density of 1016 inhabitants per km² and 507 jobs per km². It accounts for 23% of national employment and 19% of the French population.^15,16 It is one of the most economically dynamic regions in the European Union, generating close to 700 million euros of gross value added (GVA), and it is by far the first European region in terms of GVA generated by wholesale and retail trade activities.¹⁷

In this paper, we estimate the carbon footprint of different distribution systems on the basis of their retail formats. For E-commerce retail, we focus on E-commerce home deliveries. For “traditional” retail, the chosen distribution systems correspond to the way French stakeholders traditionally segment the retail market: hypermarkets (2500–20,000 m², 25,000–40,000 product references), supermarkets (400–2500 m², 3000–5000 product references), and small retail establishments (<400 m²). Small retail outlets can be generalists, meaning that they offer a broad variety of products, or they can focus on more specific categories of products.

E-shopping in France, albeit a dynamic sector, is not quite as popular as it is in other parts of Europe and the world. About 20% of French businesses conduct e-sales, which is just above the European average.²⁰ But the penetration rate of online shopping for physical goods is 76%, which puts France right outside the top 10 in Europe,²¹ although Ile-de-France has the third highest penetration rate of any French region with 80%. Among French cities, the Paris Region has the highest commercial density and is characterized by strong commercial diversity.²² It is well equipped in hypermarkets and supermarkets, as well as in a variety of small generalist stores (grocery stores or épiceries) or specialized stores (food products, clothing stores, bookshops, etc.).

These stores are not distributed equally throughout the Paris Region. This is where spatial heterogeneity plays an important part. The morphology of the Paris Region is typical of Western European cities. It is characterized by a very dense historical urban center (the city of Paris), surrounded by a relatively less densely populated urban ring, itself surrounded by a mix of urban, suburban, and rural areas. For the sake of simplicity, we divide the Paris Region into three distinct concentric zones corresponding to the administrative départements of the Paris region: the city of Paris (département of Paris), the inner suburbs (Hauts-de-Seine, Seine-Saint-Denis, Val-de-Marne), and the outer suburbs (Seine-et-Marne, Yvelines, Essonne, Val-d’Oise), as depicted in Figure 1. Most hypermarkets are located in the outer suburbs, while the majority of small shops are in the city of Paris. More details on the locations of households and stores are given in SI5.

Figure 1.

Population and commercial density in the Paris Region (source: authors, based on French census data¹⁸ and firms’ data¹⁹).

2.3. Scope of Analysis

Having reviewed the literature, we conclude that there are still gaps to be filled in this field of research, as it pertains to behavioral, organizational, and spatial factors and their impacts on the environmental performance of various retail formats. To reflect the heterogeneity of retail establishments, we compare the GHG impacts of six distribution systems: hypermarkets (HMs), supermarkets (SMs), small food retail (SFR), small nonfood retail (SNFR), small generalist retail (SGR), and E-commerce home deliveries (ECHDs).

We consider different types of stores in contrasting spatial contexts within a case study (the Paris Region), relying on modeling outputs that help us consider consumer mobility behavior and specific logistics organizations. We consider indifferently all product groups sold in the considered retail formats and the GHG impacts of the distribution system as a whole (deliveries to stores and consumers, consumer purchase trips, packaging, and buildings). As illustrated in Figure 2, our studied system is composed of warehouses, freight deliveries, shops, consumer mobility, and delivery packaging related to all household consumption goods. It excludes the production of goods, their transportation to local warehouses, their primary packaging, and their use and end-of-life since we consider that these phases are the same regardless of the distribution channel. Given the relatively small impact on GHG emissions, we choose not to consider energy use from web browsing for items, which has been found to represent less than 1% of impacts in the literature.²³ We consider that transport flows start from the final warehouse before delivery to stores for retail activities and from the cross-dock facility before direct delivery to consumers for home deliveries.

Figure 2.

Life-cycle assessment’s study boundaries (source: authors).

In order to compare distribution systems, the chosen functional unit is the mass of the product, expressed in kilograms rather than in product volume or price, which would have led to different results but would have required a large conversion data set. Mass is a relevant functional unit given our modeling approach since the weight of goods will impact vehicles’ load factors and fuel consumption. We chose not to distinguish between types of products in the study, focusing instead on distribution systems within an entire region, with details on store types and locations, delivery vehicles, and routes as a whole. Nonetheless, a detailed description of purchased products in each of the six distribution systems is available in (SI4). Environmental inventories are based on the LCA database EcoInvent 3.7.1 for background processes,²⁴ and GWP characterization factors are based on the ReCiPe v1.13 method.²⁵ The following sections present a synthetic description of our empirical approach.

2.4. Life-Cycle Assessment

This section provides an abbreviated description of the modeling framework and the data that are used for entry parameters. Interested readers can review the Supporting Information for more details on the life-cycle inventories (SI2), modeling framework, and entry data description (SI3).

2.4.1. Freight Delivery Vehicles

Goods are delivered from warehouses to shops and to homes by freight road vehicles, divided into four vehicle sizes based on the gross vehicle weight rating (GVWR): small vans (2.5 t), vans (3.5 t), rigid trucks (19 t), and articulated trucks (32 t). For each size, vehicle production, fuel consumption, exhaust gas emissions, and road use are specified according to vehicle weight and age, as per Euro norms. The French national vehicle fleet description is used to represent the weight and age distributions (details are in SI4). For 2020, as alternatives to diesel vehicles still represent a negligible portion of the French fleet,²⁶ only diesel vehicles are considered.

Impacts of vehicle production and maintenance are a function of the empty weight and the lifespan of each size category. For fuel consumption and exhaust emissions, the emission and consumption model Copert5 is used.²⁷ Its capability to integrate vehicle fleet, traffic, and load characteristics allows us to evaluate each delivery separately, with specific vehicle and traffic conditions (depending on the location of the store or household being serviced) and loads. The impacts of road construction and operation are allocated to freight vehicles based on French national traffic statistics in a similar manner to EcoInvent road allocation.^24,28

2.4.2. Consumer Mobility

In traditional retail formats, consumers travel to stores to purchase goods. Three means of transportation are modeled: active mobility (i.e., walking, biking), public transport, and personal cars. For active mobility, no impacts are estimated. For public transport, we use average impacts per passenger-km on the regional network based on four submodes: bus, tramways, trains, and subways, including manufacturing, fuel supply, and infrastructure.^29,30 For personal cars, we use the car fleet LCA tool, ModEm-ACV, with the regional fleet for the Paris Region.³⁰ This tool is based on the Copert5 emissions and energy consumption model and EcoInvent LCA inventories associated with cars, and it calculates traffic speed LCA impacts per vehicle-km. Car occupancy is used to calculate impacts per passenger-km. Mean traffic speeds vary depending on store location, and we assume 30, 45, and 60 km/h for Paris, inner suburbs, and outer suburbs, respectively (impact factors are in SI2).

2.4.3. Building Impacts

Our study includes warehouses and shops. In both cases, electricity and natural gas consumption are considered. Building construction is also considered and is seldom analyzed in the literature. Three types of warehouses are modeled: regional warehouses, wholesaler warehouses, and home delivery distribution centers, with various sizes, energy consumptions, and yearly goods flows.³¹ Warehouse impacts are allocated per mass of products stored and delivered, expressed in kilograms. Warehouse types are related to the logistics family as defined in SI3 (delivered shop type, vehicle, and logistics operator). The five retail formats are modeled with specific energy consumptions (electricity and natural gas) and sizes. Shop impacts are allocated per commercial surface, expressed in square meters (see SI2).

2.4.4. Packaging Impacts

We model three delivery packings. We exclude the primary packaging that is directly related to the type of product and that varies significantly. Secondary packaging consists of a corrugated board box containing 12.5 kg of products. Tertiary packaging includes wood pallets and a plastic film for a mean load of 500 kg. For home delivery packing, a mix of plastic, paper, and cardboard is used with a packaging rate of 10% in mass.³² LCA models include material production, disposal, and recycling. French end-of-life hypotheses are used to represent sorting rates, disposal facilities, and distances, which are not always considered in the literature. Energy and material recoveries are included through substitution factors related to French or European markets.³³ LC inventories are listed in SI2.

2.5. Modeling Framework for Consumer and Freight Transport

2.5.1. Quantity of Goods Consumed

We start by estimating the quantity of goods consumed, considering the spatial diversity of shopping practices, using the 2017 French Household Budget Consumption Survey (HBCS).³⁴ It allows us to compute the weight of goods purchased by each household per type of store/residential area.

2.5.2. Consumer Trips and Quantity of Goods for Retail Outlets

Then, we estimate the quantity of goods purchased and the number of trips to each commercial outlet from a database of firms in the Paris Region.^19,35 For this, we build origin–destination matrices using the regional household mobility survey (HMS), which describes households’ trips, including shopping trips, according to households’ location, store location, and store types.²⁹ Based on these matrices, we estimate average basket weights in stores and average trip generation depending on the store surface.

2.5.3. Characteristics of Consumer Trips

Using the same database, we calculated distances per transport mode for shopping. Linear regressions are estimated to calculate distances based on store distance to the city center and public transport accessibility according to store type (hypermarket, supermarket, and small shop), transport mode (active mobility, public transport, and personal car), and residential location of consumers.

2.5.4. Freight Generation for Business-to-Business (B2B) Deliveries

The number of deliveries for different retail formats is estimated using linear regression functions based on the Paris Region Urban Goods Movement Survey (UGMS).^36,37 Using the same database, each establishment’s movements is characterized by a “logistic type” depending on the management mode of the delivery, i.e., who is doing the transport (in three classes, e.g., transport operation is performed by a logistics service provider, performed for own-account by the shipment’s sender, or performed for own-account by the shipment’s recipient), the size of the vehicle used in four classes (e.g., small vans, vans, trucks, or articulated trucks), and the type of transport service used for the delivery (full truck-load in direct trips or less than truck-load in freight rounds). The quantity of cargo consumed at each retail outlet is apportioned to each logistics type, depending on the retail format.

2.5.5. Transport Modeling for B2B Deliveries

The distance traveled to perform a given B2B delivery is the sum of two components, the approach distance traveled between the warehouse and the delivery area (divided by the number of deliveries per round) and the average distance between two deliveries within the delivery area. These two components are estimated using multiple linear regression functions estimated using the UGMS and based on calculations by Beziat.^35,38

2.5.6. Parcel Generation for Business-to-Consumer (B2C) Deliveries

B2C deliveries are generated at the level of the household using only publicly available data gathered in Gardrat.³⁹ As specified in Hörl and Puchinger,⁴⁰ parcel demand generation is estimated using marginal data on the number of out-of-household purchases, which depends on the sociodemographic characteristics of the household and its reference person.¹⁸ Then, using data from the same source, we characterize parcel deliveries depending on the nature of the goods. Finally, using the HBCS, we allocate an average weight to each delivery, depending on the nature of the goods being delivered. We consider that all E-commerce-related parcel deliveries are performed by third-party transport service providers. Furthermore, we consider that large parcel deliveries (furniture and household appliances) are delivered by small rigid trucks, while small parcel deliveries (other types of goods) are delivered by light-duty vehicles.

2.5.7. Transport Modeling for B2C Deliveries

Based on experts’ opinions and discussions with stakeholders, we assume that delivery rounds in the city of Paris travel 60 km, 80 km for the inner suburbs, and 120 km for the outer suburbs while delivering an average of 110 parcels per round.

3. Results and Discussion

3.1. Consumer Mobility, Logistics Organizations, and Trip Characteristics

The quantity of goods purchased by households in the Paris Region is estimated at more than 10 million tons per year based on the French HBCS. This represents 843 kg of products per inhabitant. This only counts items that are sold in the types of stores included in our analysis (see SI4 for the list): food products; everyday nonfood items (including hygiene, personal care, and cleaning products); and household products used for daily living, comfort, and convenience (furniture, appliances, electronics, etc.). From other French sources, average food consumption is estimated at 570 kg per person per year,⁴¹ and consumption of everyday products (including food, beverages, and hygiene and cleaning products) is estimated at 700 kg per person per year.⁴²

The analysis of the Paris region mobility survey describes around 230 million trips per week, which represents 19 trips per individual. Of these, 22% are related to a shopping activity, with return trips considered. Table 1 shows the distribution of these trips among store types and locations. We assume that all shopping trips lead to a purchase. Then, combined with the yearly weight of goods purchased in the same region, we estimate average basket weights, with a mean basket weight of 8.4 kg per purchase, with very different results, depending on retail formats. Trip distances and modal shares vary significantly in the studied territory and among different stores. Peripheral stores and hypermarkets generate longer trips with greater use of personal cars. For the whole Paris region, one purchase generates an average trip of 3.6 km, mainly by car (77%). Based on these analyses, linear models are built to estimate the distance per mode for each store based on its distance from the center and on public transport accessibility (see Table 1).

Table 1. Characteristics of Consumer and Freight Trips in the Paris Region^a,b.

			hypermarkets (HM)	supermarkets (SM)	Sm. food ret. (SFR)	Sm. nonfood ret. (SNFR)	Sm. gen. ret. (SGR)	EC home deliveries (ECHD)
tons	total	# k tons	3260	4530	1030	990	540	200
	average	tons/store	14,240	1400	60	30	260
		tons/household	49.8	85.6	13.8	7.1	61.6	0.04
	freight intensity	tons/worker	0.6	0.9	0.2	0.2	0.1
		tons/m2	0.8	1.9	0.5	0.2	1.6
purchase acts	average	# k purchases	205,100	257,800	212,500	445,900	112,600	26,790
		# k purchases/store	895.6	79.7	11.8	13.3	55.4
	Paris	# k purchases	39,400	91,500	66,200	195,500	39,500	7,060
		# k purchases/store	985.0	92.2	12.4	11.9	43.7
	inner suburbs	# k purchases	59,800	73,000	72,500	112,900	35,400	10,710
		# k purchases/store	842.3	65.4	11.1	14.3	59.2
	outer suburbs	# k purchases	105,900	93,300	73,900	137,500	37,700	9,020
		# k purchases/store	897.5	83.0	12.1	15.1	70.9
ave. basket weight	average	kg/purchase	15.9	17.6	4.8	2.2	4.8	7.6
	Paris		10.5	10.7	3.8	1.7	2.9	7.6
	inner suburbs		21.8	23.5	5.6	3	5.5	7.6
	outer suburbs		14.6	19.8	5.1	2.2	6	7.6
consumers traveled distances and modal shares	average	km/purchase	7.6	3.5	2.6	2.6	2.5
		% active/PT/PC	(3/5/92)	(10/11/79)	(14/18/68)	(14/26/60)	(15/19/66)
	Paris	km/purchase	3.4	1.6	1.7	1.9	1.5
		% active/PT/PC	(11/29/60)	(30/40/30)	(25/55/20)	(22/61/17)	(29/56/15)
	inner suburbs	km/purchase	4.7	2.9	2.0	2.1	2.0
		% active/PT/PC	(6/7/87)	(15/12/73)	(19/17/64)	(18/21/61)	(19/19/62)
	outer suburbs	km/purchase	10.8	5.7	4.0	4.0	4.0
		% active/PT/PC	(1/2/97)	(3/2/95)	(7/5/88)	(7/5/88)	(8/4/88)
freight deliveries	total	# k deliveries	1190	1630	5480	94,100	700	26,790
	average	weekly del./store	99.7	9.7	5.9	5.4	6.7
		yearly del./household						5.1
	freight intensity	weekly del./worker	0.4	0.6	1.4	1.3	1. 6
		yearly deliveries/m2	0.3	0.7	2.7	2.3	2.2
	size of shipments	tons/delivery	2.8	2. 8	0.2	0.1	0.8	0.0076
freight transport operator	service provider	% of deliveries	77.7%	43.9%	40.7%	41.4%	43.8%	100.0%
	OA—forwarder		10.4%	38.1%	49.4%	48.8%	48.4%
	OA—recipient		11.8%	18.0%	9.9%	9.8%	7.8%
freight vehicle classes	small vans	% of deliveries	3.9%	5.4%	16.5%	17.2%	15.9%
	vans		10.8%	24.6%	43.9%	41.9%	37.7%	88.0%
	rigid trucks		37.9%	37.9%	37.2%	38.0%	42.2%	12.0%
	articulated trucks		47.3%	32.1%	2.4%	2.9%	4.3%
freight trip modes	freight rounds	% of deliveries	38.5%	52.8%	85.9%	88.0%	81.9%	100.0%
	direct trips		61.5%	47.2%	14.1%	12.0%	18.1%
freight average traveled distance	average	vkm/delivery	25.4	28.5	11.8	10.7	12.7	0.8
	Paris		18.2	22.9	9.3	8.5	10.4	0.6
	inner suburbs		22.0	26.2	11.6	11.1	13.0	0.7
	outer suburbs		28.6	35.1	14.1	14.2	16.8	1.1
	Average	vkm/ton	9.2	10.3	62.8	101.9	116.7	105.6
	Paris		8.0	10.9	59.8	114.5	130.1	71.8
	inner suburbs		6.4	8.6	56.8	74.0	113.4	95.8
	outer suburbs		12.0	11.4	71.4	118.4	132.6	143.6

^aAuthors’ calculations; EGT 2010-STIF-OMNIL-DRIEA;²⁹ ETMV 2011-2012-RIF-DRIEA/DGITM-ADEME.⁴³

^b# k: thousands of; PT: public transport; PC: passenger car; del.: delivery; vkm: vehicle-kilometers; OA: own-account.

Logistics characteristics of the last mile are also sector-specific. Table 1 highlights the important heterogeneity of the retail sector. More specifically, it shows that large retail outlets (supermarkets and hypermarkets) benefit from economies of scale, reflected in several indicators:

• They have a much lower number of deliveries per employee compared to small shops (on average: less than 0.6 weekly deliveries/worker for large retailers vs over 1.3 weekly deliveries/worker for small shops)

• They are delivered more frequently by specialized transport service providers (especially for hypermarkets, with close to 80% of deliveries performed by transport service providers) as opposed to small shops that tend to use suppliers who rely on their own means of transport (close to 60% of deliveries are own-account)

• They are delivered much more by heavy goods vehicles: >70% of deliveries for large outlets vs around 40% for small stores

• A large number of deliveries to hypermarkets (>60% of deliveries) are made through direct trips of full loads, as opposed to small stores, which tend to receive smaller shipments delivered via freight tours (>80% of deliveries).

The total distance traveled per delivery is higher for large outlets because of the number of direct trips. However, the number of vehicle-kilometers traveled to supply 1 ton of cargo is much higher for small stores because the weight of the cargo per delivery is much lower. The same applies to E-commerce home deliveries. The number of kilometers per delivery is very low, but the number of kilometers per ton is significant, equivalent to that of small stores. This is consistent with the characteristics of on-demand logistics. Indeed, parcels are delivered in highly optimized freight rounds to maximize the number of clients within given time constraints, as opposed to the carrying capacity of vehicles.

3.2. Carbon Footprint of Retail and E-commerce in the Paris Region

The chain modeling developed for this paper allows us to compute GHG emissions at a disaggregated level by integrating LCA emissions factors for packaging, warehouses, shops, clients’ trips, and freight vehicles (see SI2 for more information) with specific variables associated with shops and deliveries (e.g., shop types, speed, logistics organization, load factors, etc.). Given the nature of some of our input data, it is possible to estimate an average score at the level of the store (for traditional retail) or the household (for home deliveries) and even for specific products. However, we provide only aggregated results at the level of macroretail sectors and for large zones of the Paris Region, for two reasons. First, this facilitates the interpretation of our results. Second, even though some input data is disaggregated, we use many aggregated and average indicators in the model, so the validity of our results is best ensured at a larger scale. In Figures 3 and 4, the error bar illustrates the variability of results, as it shows the standard deviation from the mean. Depending on the characteristics of the individual commercial outlets, results can vary. It is also possible to disaggregate the results according to the relative contributions of each part of the distribution system: freight trips for deliveries and consumer shopping trips (including vehicle production, fuel production, road use, and exhaust emissions), as well as buildings (stores and warehouses) and packaging. Interested readers can refer to a more comprehensive rendition of Figures 3 and 4 in SI5.

Figure 3.

Average carbon footprints and standard deviation for each retail format (authors’ calculations) (HM = hypermarket, SM = supermarket, SFR = small food retail, SNFR = small nonfood retail, SGR = small generalist retail, RETAIL = all stores, ECHD = E-commerce home delivery) (clients = shopping mobility impacts by consumers, freight = last-mile delivery of goods by transport company, shop = building impacts from retail point of sale, packaging = production of materials for packaging, warehouses = building impacts from goods storage).

Figure 4.

Carbon footprints of different retail formats considering their location (authors’ calculations) (HM = hypermarket, SM = supermarket, SFR = small food retail, SNFR = small nonfood retail, SGR = small generalist retail, RETAIL = all stores, ECHD = E-commerce home delivery) (clients = shopping mobility impacts by consumers, freight = last-mile delivery of goods by transport company, shop = building impacts from retail point of sale, packaging = production of materials for packaging, warehouses = building impacts from goods storage).

Figure 3 illustrates that retail, as a whole, emits slightly more GHG per kilogram of goods than E-commerce (0.22 vs 0.19). However, the distribution of results largely depends on retail formats. In all cases, it should be noted that—as estimated in previous studies on the subject—warehousing impacts contribute very little to GHG emissions, given the massive amounts of cargo that transit through logistics facilities. However, the energy consumption of stores is a major contributor to GHG emissions. For most stores, energy consumption represents about 20–30% of GHG emissions and close to 40% for small food shops, which require a lot of refrigerated storage systems that consume a lot of energy. As we saw in the literature review, some studies focus on the environmental impacts of the last mile, especially on vehicle use. Our GHG results show that this represents a relatively small percentage to emissions in the overall distribution system. Exhaust emissions represent 67% of the total transport emissions (see SI5). The LC of vehicles and fuel and road infrastructure accounts for a third of total transport emissions for freight and consumer trips.

Freight trips account for a small portion of total emissions in the traditional retail sector: between 4% for hypermarkets in the city of Paris, up to 13% for small shops specialized in nonfood products. This considers the fact that hypermarkets and supermarkets enjoy economies of scale due to their size and optimized logistics. The percentage is higher for E-commerce home deliveries (19%) since these deliveries involve no shops and no client trips. Packaging, last, is one of the main contributors of emissions across sectors. Obviously, since E-commerce home delivery involves no client trips or stores, packaging makes up about 80% of total emissions.

Packaging is also one of the main contributors for hypermarkets and supermarkets, with up to 60% of total emissions of supermarkets and more than 40% of emissions of hypermarkets. This is no surprise: packaging is the only impact that is roughly proportional to the tonnage handled by stores. Shopping trip impacts (relative to the quantity of goods) can be reduced depending on the modes used by clients and basket size, freight impacts can be lessened by consolidated and optimized shipments, and building emissions can be mitigated by economies of scale. However, goods have to be packaged, and it is difficult to attenuate these externalities. Since super- and hypermarkets handle a lot of tonnage, packaging is one of the main contributors to their overall impacts. Interestingly, our results also show that the heterogeneity of the results is higher for retail than for E-commerce. The mean shown in the graph masks a diversity of retail situations, particularly evident in small shops (SFR and SNFR). The high dispersion shows that in some contexts, depending on behavioral, organizational, and spatial factors, the GHG emissions of retail can be inferior to that of E-commerce.

Figure 4 illustrates the importance of considering organizational, behavioral, and spatial contrasts to reflect the internal heterogeneity of impacts produced by retail and E-commerce. First, for a given retail format, the carbon footprint is usually higher for a store located in the outer suburbs compared to a store located in the city of Paris. For example, supermarkets in the outer suburbs generate 34% more GHG emissions per kilogram of cargo handled than supermarkets in the city of Paris. Emissions for E-commerce in the outer suburbs are 11% higher than in the city of Paris. This is due to the fact that mobility impacts are less important in denser areas, due to a higher share of active and public trips for shopping, as well as the positive impact of density on the optimization of freight rounds. In the city of Paris, shopping trips account for a very small portion of total impacts (7–17% depending on the retail format) due to the fact that most shoppers use public transit or active travel modes. In the outer suburbs, where most consumers rely on cars to purchase goods, shopping trips can represent up to 50% of total impacts. The inner suburbs characterized by intermediate density, are unsurprisingly in an intermediate situation, with shopping trips accounting for 9–31% of impacts.

Figure 4 also shows that the result of the E-commerce vs retail debate is different depending on the retail format and the location of stores and households. E-commerce deliveries, regardless of location, are more efficient than small retail formats but usually less efficient than large stores such as hypermarkets and supermarkets. Traditional retail produces far greater emissions in less densely populated areas compared to E-commerce (0.26 vs 0.20 in the outer suburbs) due to the weight of car trips for shopping, while retail is more or less equivalent to E-commerce in dense urban areas (0.19 vs 0.18 in Paris and 0.19 vs 0.19 in the inner suburbs). This is despite the fact that E-commerce home deliveries are more efficient in the dense urban center than in the outer suburbs where freight deliveries are optimized because of the high density of consumers. In other words, E-commerce deliveries are more efficient in dense urban areas than in rural areas, but retail is much more inefficient in rural areas than in dense urban areas. Figure 4 highlights that considering the location of stores reduces the intraclass variability, as the standard deviations illustrated by the error bars in the graph are lower than those in Figure 3, in which we only consider retail formats. Another interesting finding is that the dispersion of results is higher in the outer suburbs than in other parts of the Paris Region, which is consistent with the fact that this area is the most spatially diverse (including a mix of low-density urban areas, suburban, and even rural areas).

As mentioned in the literature review (Section 2.1 and SI1), environmental assessments are very sensitive to hypotheses and inputs. Therefore, we opted to perform sensitivity analyses through six scenarios based on reasonable inputs found in recently published studies. The results are listed in Table 2. Two scenarios deal with packaging, two others quantify shop energy consumption, and the last two represent mobility alternatives. We show that overall assessments and conclusions can vary considerably depending on which assumptions are made, even if these assumptions are justified by the authors. One common assumption that many studies make on consumer shopping trips is to consider that all trips are made by cars. In our case, this leads to a significant increase in GHG emissions for retails (+34%). Similarly, scenario 1, which changes the packaging rate, and scenario 3, which uses other shop energy consumption figures, both enlarge the gap between retail and E-commerce systems.

Table 2. Literature-Based Sensitivity Analysis (Authors’ Calculations, More Details Are in the SI).

scenario	literature	description	GHG emissions—kg CO2-eq per kg of product retail vs E-commerce (% difference)	E-commerce has less GHG impacts than retail?
base case			0.22 vs 0.19 (−14%)	yes, less
scenario 1	Kim et al.44	lower packaging rates (3.3% for secondary packaging, 5.3% for home delivery (only cardboard))	0.17 vs 0.10 (−41%)	yes, much less
scenario 2	Tua et al.45	reusable plastic crates for secondary packaging (19 g CO2-eq per kg of product based on 20 rotations) in retail	0.15 vs 0.19 (+27%)	no, much more
scenario 3	Tassou et al.46	electricity consumption for food shops based on commercial surface (+150% compared to the base case)	0.27 vs 0.19 (−42%)	yes, much less
scenario 4	van Ooteghem and Xu47	energy consumption for stores (−50% compared to the base case)	0.20 vs 0.19 (−5%)	yes, but close
scenario 5	many studies (see tables in SI1)	all consumer trips made by cars	0.29 vs 0.19 (−34%)	yes, much less
scenario 6	Sharpe 2019	50% of delivery vehicles and passenger cars are fully electric (see SI2)	0.20 vs 0.18 (−10%)	yes, but close

In scenarios 2, 4, and 6, we simulated mitigation alternatives, with, respectively, reusable plastic crates for secondary packaging, better energy efficiency in stores, and electric vehicles for both consumer cars and delivery vehicles. In our case study, regional average results for traditional retail and E-commerce are either very close or give traditional retail a significantly better environmental performance when employing reusable crates. The electrification of half of the vehicles leads to global decreases of 8 and 6% for retail and E-commerce, respectively. The low sensitivity to a shift to electric motorization reflects the limited contribution of transport (delivery and consumer trips) to total impacts (see Figure 3), as well as the relatively modest gains related to electric vehicles.

3.3. Discussion

Although our methodological approach bridges some gaps identified in the literature, it has its limitations. First, when comparing impacts across retail formats and spatial configurations, we consider only urban freight deliveries and not international or continental transport flows, nor for return logistics. Second, we simplify our modeling of retail strategies. We consider only home deliveries and not out-of-home deliveries. Nor do we consider so-called “phygital” strategies, i.e., physical stores also selling online, or the fact that it is possible to purchase an item at the store and have it home delivered. Finally, we use GHG emissions as a proxy for environmental impacts to simplify our results, even though our LCA model could cover a broader panel of indicators. We do not consider other externalities such as air pollution, noise, or congestion, all of which could change the distribution of impacts between sectors and across spatial configurations.

With that said, we postulate that this study highlights several important results. First, reducing the carbon intensity of freight transportation is important, but transportation is not the first cause of GHG emitted by distribution systems. Consumer mobility plays a larger role, especially in less densely populated areas. Impacts due to packaging and building energy consumption are also very significant. This means that stakeholders aiming to reduce the carbon footprint of distribution systems need to look at impacts along the chain and not just at freight distribution. Retailers need to look at different impacts depending on the scale and nature of their activity: large retailers should concentrate on packaging and client purchasing trips, while small stores should aim to optimize their procurement strategies and energy consumption.

While there is a transfer of negative impacts from retail (buildings and purchasing trips) to home deliveries (packaging and last-mile operations), E-commerce home deliveries emit fewer GHG emissions compared to retail activities overall (0.19 vs 0.22 kg CO₂-eq per kg of product). However, the comparison between E-commerce and retail needs to consider spatial configurations, local behaviors (mobility and purchase practices), and logistics strategies linked to retail formats. When looking at the results in more detail, we show that large retailers usually have a smaller impact compared to E-commerce home deliveries (0.19 for E-commerce vs 0.18 for large retailers), while small stores were found to have a much larger impact (0.19 vs 0.31). Large stores benefit from economies of scale, which manifest in different ways: larger shopping baskets, optimized logistics and shipments, and fewer impacts of buildings relative to the quantity of goods consumed at the store. Furthermore, E-commerce, while having a smaller carbon footprint overall, emits about the same quantity of GHG as traditional retail when you consider the dense urban areas of our case study (0.19 for retail vs 0.18 for E-commerce). In other words, although E-commerce is more efficient in city centers than in peripheral areas, it is more competitive than retail in low-density contexts.

3.4. Research Implications

Our results highlight the fact that the debate over the environmental assessment of E-commerce vs retail needs to be nuanced. The results of our base case illustrate that not all retail formats and not all locations are equal. Future research should consider the internal heterogeneity of behaviors, economic strategies, and spatial configurations in their case study. More retail formats could be taken into consideration, as well, especially given the way the impact of retail systems is distributed. For example, it would be interesting to look in more detail at alternative or emerging retail options, such as open-air markets (less energy consumption compared to stores) or bulk stores (less packaging). The segmentation of retail formats that we chose to analyze in this paper is relevant to the French case. However, similar distinctions could be made for other case studies, which have their own specific retail formats (for example, kiosks located in many Mediterranean cities, bodega stores in New York, street vendors in most cities of the Global South, etc.).

Due to the complexity of models and parameters involved in the analysis, we also have to address uncertainty when comparing distribution formats. First, the heterogeneity of results is due in part to intraclass variability, as illustrated in Figures 3 and 4. Individual outlets have different characteristics that impact their carbon footprint. Second, there are uncertainties linked with the modeling framework and its parameters, which by nature include error terms. Quantifying this uncertainty (using, for example, Monte Carlo methods) will be addressed in future research. Finally, there are local uncertainties in the foreground of the model caused by the chosen parameters. Our sensitivity analysis (Table 2) shows that depending on key assumptions, our results can shift considerably. One problem we find is that most studies, including our own, rely on oversimplified assumptions for calculating different impacts. In this paper, we show that by considering behavioral, organizational, and spatial factors, we can highlight important differences between retail formats. However, in some areas, our hypotheses, while in line with standard research methodologies found in the literature, remain relatively crude. More research is necessary to take into consideration, for example, the heterogeneity of packaging quantities or the energy consumption of buildings across retail formats and spatial configurations. The debate over the environmental assessment of E-commerce vs retail will remain open while these gaps are addressed.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.3c01901.

• Extended literature review of retail versus E-commerce impacts; life-cycle inventories for buildings, packaging, and freight transport; extended transport modeling framework and data description; detailed parameter values used in the transport modeling frameworks; and extended description of results (PDF)

The authors declare no competing financial interest.