*1.1. Cross-Country Di*ff*erences in Household Economies of Scale*

There may be various factors explaining the potential differences in the household economies of scale across EU countries. Some of these are related to the distribution of household size and composition. Adding another member to a household is likely to reduce per capita energy use and carbon footprints at a decreasing rate with rising household size; that is, increasing the household size from one to two members may drastically reduce home energy use and the associated carbon footprint, while a change from three to four members has been shown to produce a smaller effect on average [2]. Furthermore, the household composition—e.g., the age and the gender of the new household member—may also play an important role [3,5].

Several social, political and cultural factors are also likely to influence the effect that an additional household member has on energy and carbon footprints. Widely reported long-term changes include decreasing social trust, concern for others, conformity and religiosity, and increasing individualism, gender egalitarianism, materialistic and extrinsic values [17,18], all of which may have implications for household dynamics and sharing practices. Yet, following the global financial crisis, more recent changes in values towards greater importance of conservation (security, tradition) and concerns for close others (benevolence) have been noted in Europe [17]. As countries with extensive social nets report lower value changes following the financial crisis [17], we discuss social welfare systems as an important country-specific factor that influences the potential for sharing within and between households. Welfare

regimes that promote individual independence, female participation in the labor force, and countries with higher levels of secularization [19] may stand out with lower household sizes, which may also affect the potential for household economies of scale. Differences in consumption patterns across countries, stemming from differences in culture, social norms, geography and climate, infrastructural and institutional contexts, may also explain some of the variation in household economies of scale.

While we cannot test these theories directly in our analysis, they offer potential explanations for the country clustering of household economies of scale in our analysis.

#### *1.2. Interaction between Household Size and Population Density*

Urban areas are associated with high population and employment densities, compact and mixed land uses, and high degrees of connectivity and accessibility [20–22]; as such they have higher potential for collaborative consumption and sharing of resources between households, and more efficient uses of infrastructure [5], the so-called "compact" or "density effect" hypothesis [16]. This is because urban areas with narrower streets and smaller city blocks, compact and connected design, pleasant and safe urban space and mixed land uses generally reduce travel distance and promote active travel (walking and biking) and public transport [5,20,22]. Furthermore, urban dwellings are associated with smaller sizes, a higher proportion of apartments and multi-family houses and the presence of district heating, which are overall less carbon and energy intense per unit of area [3,22]. While there is strong evidence for this density effect on per capita carbon and energy footprints in the European context, this is largely compensated by higher income levels in cities [23]. Urban cores are generally preferred by more affluent and younger adults with greater consumption opportunities and smaller household sizes (and hence higher per capita carbon and energy footprints), while suburban areas benefit from larger household sizes and economies-of-scale effects at the household level [15,16]. This clearly complicates the established view that dense urban environments are more sustainable [5].

Furthermore, household economies of scale are likely to differ between rural and urban areas [15,16]. A recent study from the USA found household economies of scale to be about twice as large in rural compared to urban contexts (up to 8% reduction in per capita carbon emissions when adding an adult in rural contexts compared to 3% reduction in dense urban contexts) [24]. Lower household economies of scale in urban areas have also been found in a European context [25]. An explanation for this trend is that both household and urban economies of scale "are driven by proximity and realized through sharing" [24]. Adding a member in a rural detached house will bring about higher savings through sharing walls, living space and heating and cooling, compared to adding a member in a shared apartment building, where walls are already shared between more households, living space is smaller and common district heating may be present. Urban context is associated with proximity between households and thus higher potential to share resources outside of the household, which may in turn partially offset the household size effect. We explore differences in the household economies of scale between urban and rural context through an interaction term between household size and population density in the model.

#### *1.3. This Study*

In this article, we calculate the total and the average per capita EU carbon and energy footprints for various household sizes. We examine the inter-country differences in household economies of scale across 26 EU countries as a way to uncover sharing opportunities and support reductions in energy use and GHG emissions. This analysis considers differences in effects across consumption domains, as well as between rural and urban areas.

Prior studies generally focus on a single country, while a comparative perspective is lacking. A comparative perspective allows for a more robust discussion of the potential energy and GHG emission cuts that could be achieved through within- and between-household sharing—and may help formulate context-specific initiatives and policies for resource sharing on a regional and country level. Studies usually focus on either carbon or energy. In this study, we examine both in order to enable a wider comparison.

#### **2. Data and Methods**

## *2.1. Databases*

The Household Budget Surveys (HBS), harmonized and disseminated through Eurostat, collect information about household consumption expenditure across EU countries. This study utilized data from 2010, which is the latest available. Price coefficients were used to adjust household expenditure to the reference year of 2010 and EUR/purchasing power standard (PPS) units, thus accounting for price differences across countries and time (for the countries, which collected expenditure in a different year) [26]. A detailed overview of the HBS accuracy (sampling and non-sampling errors), timeliness, comparability and representativeness is provided elsewhere [26]. We transformed household expenditure into per capita expenditure and proceeded with carbon and energy footprint calculations.

We calculated annual carbon and energy footprints on the household level, utilizing the multiregional input-output database EXIOBASE (version 3.7) [27]. We applied the Global Warming Potential (GWP100 [28]) metric to convert various GHGs (carbon dioxide, methane, nitrous oxide and sulphur hexafluoride) to kilograms of CO2-equivalents per year (kgCO2eq). Annual energy use was calculated using the net energy extension measures in terajoules (TJ). There is no double counting with regards to the conversion from primary sources (derived directly from nature, e.g., coal) into secondary sources (coal-generated electricity, for instance) [29]. In this paper, we used the terms "carbon footprints" and "GHG emissions", as well as "energy footprints" and "energy use" interchangeably. We expected that the two environmental indicators would depict similar trends in terms of the effect of household size, as the majority of GHG emissions are related to energy use (e.g., burning of fossil fuels).

The EXIOBASE database covers high sectoral detail (200 products), 49 countries (including all EU countries) and rest-of-the-world regions, and a wide range of environmental and social satellite accounts [27,30]. We matched the HBSs household expenditure in 2010 with the environmental and economic structure in EXIOBASE for the same year. For a detailed overview of the harmonization steps between consumption from HBSs and the environmental intensities from EXIOBASE, see SM1 and elsewhere [4,31].

#### *2.2. The Model*

In order to examine inter-country differences in household size effects, we performed the regression analysis for each EU country *c* separately (see SM4 for a robustness check through a model including all of the countries). We also performed the analysis on EU level. We applied the household weights disseminated by Eurostat. The analysis is conducted on a per capita level for each household *i*, with the following specified model:

$$\begin{aligned} & \quad \ln\{\text{EN}\overline{\text{NF}}\_{\text{cl}}\} \\ &= \beta\_{\text{c0}} + \beta\_{\text{c1}}\{\text{LNINICME}\_{\text{cl}}\} + \beta\_{\text{c2}}\{\text{HISIZE}\_{\text{cl}}\} + \beta\_{\text{c3}}\{\text{DENSE}\_{\text{cl}}\} + \beta\_{\text{c4}}\{\text{INTER}\text{MEDIATE}\_{\text{cl}}\} + \beta\_{\text{c5}}\{\text{HISIZE}\_{\text{cl}}\} \\ &+ \beta\_{\text{c6}}\{\text{ENSE}\_{\text{cl}}\} + \beta\_{\text{c6}}\{\text{REGION}\_{\text{cl}}\} + \epsilon\_{\text{c1}} \end{aligned}$$

*ENVF* stands for the estimated environmental footprint, namely the annual carbon or energy footprint per capita measured in kgCO2eq and TJ, respectively, in logarithmic form. The log-transformation was done to achieve normally distributed regression residuals, which previously had a positively skewed distribution.

*LNINCOME* measures the role of net disposable household income [32] (not equivalized) for the environmental footprint. The income coefficient can be interpreted as income elasticity as both the dependent and independent variables are measured in logarithmic form. As the Italian HBS does not include the income variable used for other countries, we employed the logarithm of total expenditure instead as an independent variable, similar to other studies [14,24].

*HHSIZE* presents the number of household members. The term *household* refers to people with a common use of an address, usually sharing space and practices [9]. In the HBSs, sharing common accommodation and expenses was also central to the household definition.

The dummy variables for population density (*DENSE* and *INTERMEDIATE*) utilize the Eurostat's measure of the degree of urbanization [33], based on Local Administrative Units level 2 (LAU2). LAU are low level administrative divisions below that of a province, region or state [34], where LAU2 is the lowest consisting of municipalities or equivalent units in the 28 EU Member States (formerly NUTS 5 level) [35]. The degree of urbanization defined by Eurostat classifies LAU2 into sparsely, intermediate and densely populated areas, using as a criterion the geographical contiguity in combination with the population density in the different types of areas [33]. A map of the degree of urbanization in 2011 for all of the EU and a detailed explanation of the undertaken steps for the LAU2 classification can be found elsewhere [33]. In this article, variable *DENSE* takes the value of one for households that live in areas with at least 500 inhabitants/km2, and zero otherwise (cities). *INTERMEDIATE* takes the value of one for households that live in areas between 100 and 499 inhabitants/km2, and zero otherwise (towns and suburbs). The base category *SPARSE* is associated with rural or sparsely populated areas with less than 100 inhabitants/km2 according to the HBS classification.

Similar to a prior study [24], we added an interaction term between household size and population density (*HHSIZE*×*DENSE*) in order to explore the potential variability in household economies of scale by urban-rural typology.

We also included spatial controls—a set of regional dummy variables (*REGION*)—aiming to account for regional differences such as technological (e.g., energy efficiency or infrastructure, type of dominant industries) as well as geographical and climatic context [4] (see SM1 for an overview of all regions). The regional distribution is the first-level NUTS of the EU for most countries.

Prior work has discussed the selected variables in the model as key socio-demographic, economic and geographical determinants of environmental footprints [4,5,24]. While additional factors such as dwelling size and type, vehicle ownership, energy sources and prices [3,21] among others are important, the HBSs do not collect such data. We also did not explore the role of household composition, while prior studies found education, gender and age to have small and mixed effects [2–5,36]. For example, females have been found to have lower carbon footprints associated with transport and food, and higher energy use at home [3,36]. Single parent households (mostly headed by women) were found to be more likely to experience fuel and energy poverty [37]. Age has been found to be positively associated with energy use [3,38], although this effect may slow down or even change direction when people reach their later years [2,36]. Education and social status may also redesign preferences towards more or less emission- and energy-intensive consumption [2,4,39].

We estimated the regression model based on household surveys from 25 EU countries (excluding Sweden and the Netherlands due to lack of consumption data and Romania due to lack of population density), with a total sample of 243,911 observations.
