Choice of Allocations and Constructs for Attributional or Consequential Life Cycle Assessment and Input‐Output Analysis - 2017 Graedel Prize Blog Post Series

This blog post is part of a series on the 2017 Graedel Prize Winners.
Read the other blog post here: Circular Economy Rebound..

Back to the basics

The question "What are the life-cycle impacts of product x?" is problematic. It does convey the ambition to take into account both direct and indirect environmental impacts "associated" (somehow!) with a product, but it leaves in the dark what this "association" means. This association, of course, may be performed by a life-cycle assessment (LCA), but then our initial research question has become self-referential: by definition, the life-cycle impacts of a product are those that would be calculated by an ideal LCA of this product, and then we need to define ideal LCAs, standard-compliant LCAs, etc. In other words, we cannot afford to confuse research objectives and research tools.

As naive and trivial as this criticism may seem, we are all partly guilty of using such jargon and convoluted definitions to frame our research questions. Yet, we all know that simple formulations of clear research questions are more likely to lead to robust and useful research, as they allow for the identification of the mechanisms that must be captured for a coherent analysis.

In this spirit, my co-authors and I strove to make explicit the distinct research questions that underpin two types of LCAs, so-called attributional and consequential LCAs. We then asked what criteria must be met, and what mechanisms must be captured, to coherently answer each research question, focussing specifically on the modeling of multifunctional processes.

The resulting lists of axiomatic criteria were almost completely different for attributional and consequential analyses. For example, contrary to a rather common perception, maintaining mass balance was not found to be necessary to coherently answer all LCA-based questions: If a study asks what share of the economy's impacts can be attributed to the consumption of a product, the responsibility for these impact can be ascribed in ways that are internally consistent without having anything to do with mass flows.

These lists of criteria then allowed us to clearly identify which coproduction models could appropriately represent different types of multifunctional processes when one strives to answer questions of attribution or of consequences, both through LCA and environmentally extended input-output (EEIO) analysis.

A diversity of models

Why specifically focus on multifunctional processes, and why treat LCA and EEIO together? Both LCA and EEIO analyses need to deal with situations where two products or services are produced together but are not necessarily needed together. For example, grain is required in the value chain of bread, but not straw, although grain and straw are harvested together as coproducts. How to respond to this modeling challenge has proved extremely contentious, notably leading to debates on the use of system expansion, substitution and partition allocations in the LCA community, and the use of industry-, commodity- and byproduct-technology constructs (among others) in the EEIO community. In both communities, desirable properties of these different techniques have been analyzed, often without reference to a specific research question, sometimes leading to competing claims of universal superiority. Our research was driven by the hypothesis that there was no single, universally superior multifunctional model; that the choice of models had to be tweaked depending on the research question and the processes modeled.

There is no one-to-one correspondence between LCA and EEIO coproduction models, and jointly analyzing them has lead to interesting insights. First, the diversity of ways in which LCA partition allocations and substitution models can be applied clearly points to the fact that the dominant EEIO constructs constitute special cases of more flexible models. As national inventories move beyond the exclusive use of a single measurement unit (e.g., dollar) and low-resolution product group classifications, the EEIO community will progressively have access to a much broader range of modeling approaches.

Second, multiple EEIO constructs can be said to "expand the system", demonstrating that LCA's "system expansion" actually regroups under a single name multiple modeling approaches that are not necessarily equivalent. We, therefore, distinguished between different "kinds" of system expansion and evaluated which were most appropriate for what situation.

Key results

We found that multiple models could prove equally compatible with attributional analyses, which points to the potential to further refine the attribution question. Interestingly, while some models that "expand the system" could satisfy all criteria for a coherent attributional analysis in some circumstances, substitution models were outright incompatible. Again, the type of system expansion matters.

For consequential studies, we found that substitution-based modeling was appropriate in cases where coproduction ratios were fully inelastic and coproducts were technologically linked. Conversely, a modeling approach closer to that of EEIO's commodity-technology construct was required for coproducts that display fully elastic coproduction ratios and only a weak technological bond. Critically, though, we found that none of the standard LCA or EEIO coproduction models could satisfactorily represent intermediate situations, where multifunctional processes can partly adapt their coproduction ratios. In other words, we are ill-equipped to assess the environmental consequences of systems where the coproducts are linked technologically but with a certain level of flexibility. Similarly, the standard LCA and EEIO system resolutions cannot appropriately capture the indirect consequences of substitution between non-identical products, which notably hampers the representation of downcycling issues. These two weaknesses both point to the need for a finer parametrization of process models.

A personal note

As a young researcher, the long evolution of this article has been an incredibly positive experience. It was a slow maturing process. Our first attempt at this article really fell short of its own ambitions in terms of scientific rigor and clarity and was soundly rejected from publication at the Journal of Industrial Ecology (JIE) by excellent, dedicated peer reviewers. This rejection was accompanied by highly constructive recommendations, which echoed similar comments by my Ph.D. defense opponents, Professors Helga Weisz, and Yasushi Kondo. Much is owed to these critiques and comments, and to the diversity of perspectives of a broad team of co-authors, in this article's ultimate success and attribution of the Graedel Prize. My last thanks go to the Society for Industrial Ecology, to the JIE team, and to the Graedel Prize evaluation committee for this unique opportunity.