About Shoes, Cucumbers, Beer and Cars: A Bottom-Up Household Consumption LCA-Model

Household consumption is – besides governmental consumption – the main driver of economy. Households are thus ultimately responsible for the environmental impacts which occur over the whole life cycle of products and... [ view full abstract ]

Household consumption is – besides governmental consumption – the main driver of economy. Households are thus ultimately responsible for the environmental impacts which occur over the whole life cycle of products and services they consume. Given that purchase decisions are usually taken on household level and are highly behavior-driven, the derivation of targeted environmental measures requires an understanding of household behavior patterns and the resulting environmental impacts. The goal of this study was thus twofold: First, the Swiss household budget survey (HBS) was analyzed to reveal drivers and causal relationships in the consumption behavior of real households. Second, these insights were then used to develop a spatially resolved bottom-up household model for the whole of Switzerland.

The HBS provides detailed information about approximately 350 different attributes of the nearly 10,000 surveyed households. Among these attributes, there are almost 50 household characteristic features (e.g. age structure, income, size) and about 300 consumption areas, ranging from expenditures for different food and clothing categories (comprising e.g. men’s shoes, cucumbers, beer or similarly specific items) to mobility and dwelling spending. In a first step, the data of the HBS needed to be pre-processed. This also comprised imputing missing information which was conducted by employing different regression techniques such as LASSO-regression, Random-Forest-regression and a K-Nearest-Neighbor approach. In a next step, a Self-Organizing Map (SOM) was applied to the pre-processed dataset in order to recognize different consumption patterns. The SOM pre-conditions the dataset and then supports clustering algorithms to find consumption-based archetypes. Therefore, different clustering approaches were tested on top of the SOM including K-Means, DBSCAN and agglomerative clustering. Finally, consumption-based archetypes were derived based on an agglomerative clustering with Ward-linkage. In a next step, these archetypes were extrapolated to all households of Switzerland based on the national census in order to regionalize the model. However, the national census does not provide all the information which was used to determine the archetypes. Consequently, the assignment of archetypes to households in the census data is fuzzy to a certain degree. Therefore, the probability with which an archetype can be assigned to a particular census household was computed by means of a K-Nearest-Neighbor-classifier. In order to reproduce a realistic variability within a certain region, we sampled for each census household randomly among the archetypes which could be considered for this specific household. In a final step, the simulated expenditures were coupled with detailed environmental background data in order to derive a consumption-based environmental footprint for each Swiss household representing a life cycle perspective.

The resulting bottom-up model quantifies environmental impacts from household consumption with high resolution and reproduces the variability of behavior of the four million individual Swiss households in a realistic manner. At the same time, it also allows for the aggregation of these footprints on any desired regional scale. Therefore, this model constitutes a predestined platform for scenario analysis to support environmental policy makers in their quest for effective measures tailored to the specific problems of a region or municipality.