1. Introduction
From a life cycle perspective, it has been acknowledged that household consumption represents over three fifths of the environmental impact of total consumption and contributes the largest proportion of greenhouse gas (GHG) emissions from human activities [
1,
2]. In 2018, 55% of the global population lived in cities, a figure that is expected to increase to 70% by 2050 [
3]. Aside from this, urban metabolism (UM) studies have shown that urban consumption is not only growing in absolute terms, but also per capita [
4,
5,
6,
7,
8,
9]. There is evident need for efforts to reduce the environmental impacts of household consumption, particularly as the growing human population increasingly shifts towards living in urban areas and adopting more consumerist lifestyles. Policies and measures addressing urban household consumption are a potentially effective mitigation strategy.
A number of methods have been used to investigate household consumption at urban scale. Lenzen and Peters used multi-regional input-output (MRIO) based on supply-use regional transaction tables in Sydney and Melbourne [
10]. By supplementing the collated data with spatial detail derived from local land-use maps, pollution inventories and business registers, they were able to produce results comparable to having using life cycle analysis (LCA). In this case, industry and business transaction data from multiple sources was combined in spatial MRIO to characterise factors such as water use and GHG emissions at city level. The method cannot be easily adapted to different geographical regions and would be challenging to implement at district scale due to an inherent lack of industry in residential areas. Only one family archetype was made for each city, giving limited insight into household consumption as differences between households within the same city were not considered.
Miehe and colleagues used MRIO-LCA to study household carbon footprints (HCF) of German households, scaling data provided by individual households through expenditure surveys up to a regional level to give an overall average HCF comparison for different regions [
2]. They found that there were large regional differences in HCF and proposed that considering household consumption at a localised rather than national scale would be more beneficial for developing effective and targeted emission reduction policies. The method is not applicable to different areas due to the development of 578 geographically specific household archetypes that are not valid in other contexts.
Jones and Kammen used IO-LCA combined with household budget survey (HBS) and census data for individual households to find average HCF for each US zip code. Again, differences between households in the same zip code are not accounted for as only an average is found [
11]. Froemelt and colleagues used HBS data and combined it with existing models for building stocks, energy and transport to create a map of average per capita household GHG emissions for all municipalities in Switzerland [
12]. Here, data from individual households was applied to municipal level. As this study offers a comprehensive and bottom-up model based on existing Swiss models and highly specific household archetypes, there is limited geographical applicability which would not be easy to replicate elsewhere.
Conversely, very small-scale studies have demonstrated that tailored solutions can be provided where a lot of detail is available. Greiff et al. studied sixteen households through surveys to calculate material and carbon footprints, Harder et al. illustrated methods for tracking masses of products in individual households consumption and waste generation through entries in an online system or smartphone [
13], and Laakso and Lettenmeier applied MIPS (material input per unit of service) to five Finnish households [
7,
14]. In the study by Laakso and Lettenmeier, participating households positively changed their consumption behaviours as a result of engagement with the researchers and receiving household-specific information. However, potential environmental impact savings from so few households are limited and studies with this level of collaboration are unfeasible at a larger scale. Furthermore, information from small-scale studies cannot be extrapolated to approximate broader consumption patterns. Some authors identify that it is important to provide households with information that can help them make more environmentally responsible choices, but certain groups may need more support or even incentives to adopt more sustainable lifestyles [
15,
16].
With this in mind, identifying consumption patterns for particular household types and providing information or schemes accordingly could be an alternative way to encourage better consumption behaviours. The OECD report “Household Behaviour and the Environment—Reviewing the Evidence” suggests that policies should be targeted to account for differences in household types, but identifies that this may be costly to implement [
15]. This may be more feasible if a low-cost method was available which could help ensure that both the right schemes and information were available to relevant households to assist them in making better decisions.
It is proposed that interventions at city scale could be an effective way to positively influence household consumption in urban areas. However, this may be challenging as cities tend to undergo continuous development, as well as having differences in land-use and high population heterogeneity [
17,
18]. Districts can be seen as urban areas with sufficient homogeneity to allow for straightforward and realistic household consumption estimations, and also large enough populations to support potentially impactful change [
18]. As yet, district scale has rarely been addressed in UM studies and household consumption of goods has not previously been included [
17,
18,
19]. The UM concept describes cities as complex systems that consume resources and produce waste and emissions in order to maintain their functions, much like an organism or ecosystem. Urban Metabolism studies apply material flow analysis and other methods to quantify types and masses of resources consumed, allowing reduction in resource use and consumption-based environmental impacts through monitoring and policy formulation. This approach can also provide decision-makers with quantitative analysis that identifies target product groups or can give indication of which policies might help to reach environmental goals [
20,
21,
22]. To that end, local decision-makers may benefit from quantified insights into household consumption in a particular area that could inform more targeted initiatives.
Existing methods for studying household consumption tend not to be suitable for application at district level. The results are often geographically specific, expressed in monetary units and investigate specific environmental impacts. If the mass of goods can also be derived from monetary data, this would contribute to development of comprehensive district-scale UM accounting. It could also inform development of different schemes for reducing product consumption, in addition to broad impact evaluation. For this reason, a new method is outlined for quantifying household consumption on a mass basis. An example of how the method was applied to districts in the city of Gothenburg, Sweden, is presented to demonstrate that the method is effective at district-level. Application of data outputs from the method are illustrated with two policy-relevant examples: designing sharing economy and to aid in reducing consumption-based environmental impacts including climate change and acidification.
2. Materials and Methods
The developed method is based on the principle that quantifying household consumption on a mass basis offers flexibility for this data to be applied in different ways. Simple archetypes are formed that can be readily adapted to different geographical areas where household expenditure data is available. A diagram of the method developed in this study is presented in
Figure 1. The method can be split into a four-step process, as outlined in the following sections. Socio-economic household archetypes should be set using parameters of interest available in the HBS, such as income and dwelling type. The first step in the method is focused on gathering and collating appropriate datasets. In the second step, the datasets are transformed into mass units to allow consumption per household archetype to be quantified to be defined in the third step. The final step extrapolates the household archetypes into spatial units.
2.1. Step 1—Data Collection
The main data source for this method is the HBS, sometimes called household expenditure survey, which provides expenditure data and associated socio-economic factors and is available in many countries. In this paper, the Swedish HBS from 2012 was used. The Swedish HBS is typically conducted at three year intervals, and Statistics Sweden (SCB) asks survey participants to record their spending on various products and services within a two-week period [
23]. Additional data sources are required as HBS data does not generally include unit prices and masses for individual products. In this study, the consumer price index (CPI), conversion factors from Eurostat (ECF) and supplementary data from retail were used. Socio-economic characteristics (SEC) of households in the study area are needed to enable application of data outputs. In order to harmonise the various datasets used, correspondences between them were defined using standard nomenclature systems.
There are a number of different nomenclature systems used for coding and organising products that are traded, many of which have correspondence tables that are available through sources such as Eurostat RAMON [
24]. In the Swedish HBS, expenditure was reported for products organised by COICOP codes (classification of individual consumption according to purpose) [
25], without indication of the number of units purchased. To connect expenditure with product unit mass, unit prices were extracted from the Swedish CPI (consumer price index) and unit masses from ECF (conversion factors from Eurostat); product data for both is organised by combined nomenclature (CN) codes [
26]. There was no existing equivalence between COICOP and CN, so a new correspondence was developed by linking existing tables with common nomenclature, as shown in
Table 1. See
S1 in the Supplementary Materials for the detailed correspondence.
2.2. Step 2—Data Transformation
The initial data transformation task was to remove services and intangible products from the HBS database. Following this, the HBS was further filtered to exclude products that were not part of the Swedish CPI for the same year as the HBS study. This was because household expenditure on goods was a fixed variable in this study, so it was important to work from a basis of product prices which were reflective of what households participating in the HBS would have paid. As the CPI offers a comprehensive overview of the prices for goods within a specified time frame and geographical region, it was assumed that using average product prices calculated from CPI data would provide consistency in subsequent calculations. The products where expenditure data and average unit price were available in the HBS and CPI databases respectively will be referred to as HBS-CPI products.
To obtain monetary value per unit mass of products, average unit prices and masses needed to be established for each product. For many products this was available in the CPI, however, for products not covered by CPI, price and mass data from other sources had to be combined. Many of the remaining product masses were available via ECF, and all other product masses were found through retail (as shown in
Figure 1). Priority was given to CPI product masses, as these were for exactly the same products that were included in the average product prices. ECF was favoured over retail for the remaining product masses, as the ECF databases already had a large number of product masses collated by CN codes.
S2 in the Supplementary Materials summarises the unit masses for products not commonly sold on a mass basis; S3 outlines the calculated kr/kg (where kr is Swedish crown) for different products and the sources of mass data.
Step 2a—Finding monetary value per unit mass from CPI data
The Swedish CPI product list was filtered to remove products which were not part of the HBS, using the developed correspondence between the COICOP codes used in HBS and CN codes used in CPI. Products where mass was reported alongside price were considered to be the most reliable. In some cases (especially for liquid products such as beverages), the products were instead reported on a volume basis which could then be converted to a mass (with the assumption that 1 litre of product had a mass of 1 kg).
For products where mass data was readily available within CPI, the mean mass per product was calculated, taking into account all products where the CN and COICOP codes matched. When product codes had multiple individual products attributed to them, all relevant product masses were included in the mean for each code. The monetary value per unit mass for each HBS product was then calculated using Equation (1), where standardised mass per unit is the unit of mass (e.g., 1 kg) divided by the mass of one product unit (e.g., the mass in kilograms of a single product).
Step 2b—Finding monetary value per unit mass from ITS data
The ECF database of product masses (available from Eurostat) was used to find mass per unit for products which were not recorded on a basis of physical units in CPI. Within the ECF database, products are described on a net mass basis along with supplementary units and are independent of price. However, the International Trade Statistics (ITS) database did not include masses for all of the products included in the HBS, so it was necessary to supplement the data with product masses from retail.
Step 2c—Finding monetary value per unit mass from retail data
For products that were not available in the CPI or ECF databases, masses were found from websites for retailers that are commonly used in Sweden. For example, homeware and furniture masses were taken from the IKEA website. One product weight was selected for each product listed in the HBS.
An exception to this is for personal vehicles, which can differ in price by several orders of magnitude despite having relatively similar masses. The mass of a car is also generally at least one order of magnitude greater than the mass of other products consumed by households, so it is necessary to adjust for this and prevent reported expenditure on cars introducing skew into the results. When looking at the expenditure data, the cars are noted as either purchased (1) or not purchased (0), using the assumption that households will only buy one car within a year. The proportion of households in each archetype that reported making a purchase is found, and then multiplied by the average mass of a car. This will give an adjusted number of kilograms consumed.
2.3. Step 3—Consumption Patterns
The next stage in the data transformation was to connect expenditure data together with monetary value per unit mass and combine this with socio-economic characteristics (SEC) to form typical consumption archetypes.
Step 3a—Converting HBS data from expenditure to physical units
The calculated monetary value per unit mass for HBS-CPI products were next combined with actual reported expenditure data from Swedish households. A database was created with kr/kg prices for the different products listed alongside the COICOP codes for each product. As the data in the HBS was confidential, this database was imported into the Statistics Sweden’s (SCB) secure system for working with microdata [
27]. The household expenditure microdata and kr/kg pricing were combined using the statistical software R, using the COICOP codes to match the two datasets together. Reported household expenditure in kr was then divided by the kr/kg price for each product in order to give an estimate of the physical amount (in kg) of product that each individual family consumed.
Step 3b—Creating household archetypes
Pre-defined socio-economic archetypes based on parameters collected in the HBS were applied to enable analysis of consumption patterns for different household types. In this study, three simple filters were used to build the archetypes:
Income—households were separated into different income groups based on quartile ranges for income in Sweden. Households in the first quartile were designated as low-income, the second and third quartiles were combined to give the middle-income group, and households in the fourth quartile were classed as high-income.
Children—households were separated into those with children and those without.
Dwelling type—households were separated into those living in multiple-occupancy dwellings (apartments) and those living in single-occupancy dwellings (houses).
Consumption patterns for each archetype were estimated by calculating the mean mass of each product type consumed by all households within the archetype.
2.4. Step 4—Data Outputs
The data outputs from the method are the estimated masses of different products consumed by households within the different archetypes. The archetype consumption patterns derived can be applied to any area with similar socio-economic characteristic data within the geographical bounds of the HBS that the method was used for.
The estimated masses consumed can be linked to environmental impact profiles. The kilogram of products consumed are multiplied by impact per kilogram (e.g., kg CO2-eq), quantified from existing LCA data. The results can then be used to identify primary consumers of high-impact goods, and may even be coupled with geographical districts to find appropriate sharing schemes.
2.5. Assumptions and Limitations
A number of assumptions were needed to derive estimated household archetype consumption patterns using the developed method. One limitation is the accuracy of data obtained through HBS. In Sweden, the HBS is sent out to randomly selected households that need to complete an expenditure diary during a two-week period at a certain point within a year, plus telephone interviews about purchases of durable goods; different two-week periods are allocated to randomly selected households to account for seasonality in purchasing patterns [
23]. The expenditure data quality is therefore reliant on the accuracy of reporting by members of the public.
Assumptions were required when developing nomenclature correspondence. For some products, particularly durable goods, there was no exact correspondence between COICOP and CN product descriptions. However, it was generally possible to cover the COICOP listed in the HBS by attributing multiple CN product codes to a single COICOP code, for example, CN 1412 (milk) was matched with both COICOP 01,141 (milk with fat content >1.5%) and COICOP 01,142 (milk with fat content <1.5%). For these products, it was assumed the average product mass for all relevant CN products could be used in combination with the average CPI price for all relevant CN products to give a representation monetary value per unit mass.
A further limitation was the quality of product mass data. The sequence filtering mass data gave priority to CPI (considered the most accurate due to availability of both product masses and prices), then ECF and lastly retail. Product codes where mass data had to be taken from retail tended to represent a number of different products; although the average mass per unit would not be reflective of any single item, the overall representation should be reasonable, particularly as the expenditure data did not clarify the exact products that had been purchased. It was assumed that the products in the average mass and average price calculations were sufficiently similar to give a fair comparison.
The broad archetypes developed from socio-economic factors with this method are widely applicable. However, using limited socio-economic variables to create the archetypes may have resulted in formation of heterogenous groups, due to potential variations between households within the same archetype. Furthermore, there were inequalities between the number of households in each archetype who participated in the Swedish HBS, meaning that some archetype consumption patterns are based on the mean data from a larger number of households than others. The data for the archetypes with more households in them may be more representative than for the smaller archetypes.
4. Discussion
The method proposed in this paper offers a simple way for local policy-makers to be able to assess the impacts of household consumption within a specific area. Existing methods for evaluating household consumption do not address consumption at district level, even though this resolution has been identified as having potential for positive environmental impact through targeted localised schemes [
2]. Results from the method can be used to inform design of such localised policies with high potential engagement based on the needs of local people, both in terms of reducing the impacts of consumption in wealthier areas and improving access to goods for lower-income residents. This study proposes methodological advances by addressing the issue of combining monetary and physical data, which has been raised as a challenge in other studies [
10]. Additionally, as this method quantifies physical flows of goods consumed by households within a geographical area, it can contribute to accounting district-scale MFA and study of neighbourhood-scale urban metabolism.
By innovatively combining two datasets that are commonly compiled by national statistical offices (HBS and CPI), the method developed in this study derives product cost per kg from unit price and mass. Extraction of data from additional sources was kept to a minimum to ensure that the method would be easily replicable without the need for specialist knowledge, or significant investment of time or money. Furthermore, the chosen socio-economic characteristics are applicable to different geographies and the household archetypes that have been created can help local decision-makers in cities (or even individual districts) to be better informed about household consumption impacts within their sphere of influence.
As the criteria for creating archetypes within this method is very simple, especially compared to other archetype-based studies [
12,
39], similar criteria for identifying patterns of archetype consumption could be applied to HBS data from other countries. Moreover, archetypes can be easily customised to be appropriate to specific areas of interest, within the limits of the socio-economic parameters of HBS. Froemelt and colleagues also used archetypes in their study of Swiss household consumption, but were unsure whether the 578 geographically-specific archetypes they developed would be transferrable to other regions [
12]. The method outlined in this study can be easily applied in different contexts provided that the upper and lower income quartiles are known, so that the household incomes for the archetypes can be categorised as low, middle and high according to local economic conditions. Furthermore, the list of product unit masses compiled in this study would be suitable for combining with HBS and CPI data from other countries to apply this method to other geographies.
Whilst this study is not alone in scaling up consumption from individual households to urban-level [
2,
11,
12], most other studies directly apply scaled data to calculate average HCF or GHG emissions per capita. Our proposed method gives consumption of specific products in physical terms, offering greater flexibility to apply the data in different environmental impact models, as illustrated in this study by estimation of four different impacts using LCA. This also allows evaluation of impacts from products of particular interest, rather than broad average impacts from all consumption as in other studies. Additionally, our method can be used to quantify physical flows of goods consumed by households within a geographical area. These quantified flows are purely reflective of private household consumption, unlike in other methods which tend to use MRIO and therefore derive household consumption from trade and industry flows. Furthermore, because the consumption patterns for different areas generated using our method are based on archetypes, the socio-economic characteristics of consumption reflect each area considered, enabling comparison between districts, including patterns of consumption for individual products in each area. This would be much more difficult to capture with MRIO.
With regard to sharing economy applications, this method offers a straightforward way to evaluate or predict household consumption of products for a specific area. With unique potential for district-scale, this method can be used to identify hotspot districts with high consumption, or districts where consumption of specific products are high. Conversely, districts where residents are resource-poor relative to other areas could be identified. In turn, this can inform the design of localised sharing economy schemes at a district level, for both environmental and social sustainability. This is proposed as an effective approach due to the physical characteristics of particular districts and the tendency for socio-economic similarities between resident of the same district.
This study resulted in the initiation of a project in the city of Gothenburg. It is proposed that Gothenburg could be used as a test-bed for investigating practical applications of this method in more detail, as well as developing it further. Suggestions include quantifying the environmental and economic benefits of sharing economy (for the entire city, districts or of specific initiatives), identifying target districts or household types based on their consumption patterns, or developing customised initiatives for different household types. A case study has been proposed on a new district that currently being planned for development in Gothenburg. The purpose of the case study would be to create a specific plan for partially substituting product demand from households in the new district with shared products, based on the predicted distribution of household types.