Sensitivity to weighting in life cycle impact assessment (LCIA)

Weighting in life cycle assessment (LCA) incorporates stakeholder preferences in the decision-making process of comparative LCAs. Research efforts on this topic are concerned with deriving weights according to different principles, but few studies have evaluated the relationship between normalization and weights and their effect on single scores. We evaluate the sensitivity of aggregation methods to weights in different life cycle impact assessment (LCIA) methods to provide insight on the receptiveness of single score results to value systems. Sensitivity to weights in two LCIA methods is assessed by exploring weight spaces stochastically and evaluating the rank of alternatives via the Rank Acceptability Index (RAI). We assess two aggregation methods: a weighted sum based on externally normalized scores and a method of internal normalization based on outranking across CML-IA and ReCipE midpoint impact assessment. The RAI represents the likelihood that an alternative occupies a certain rank given all possible weight spaces, and it can be used to compare the sensitivity of final ranks to weight values in each aggregation method and LCIA. Evaluation is based on a case study of a comparative LCA of five PV technologies whose inventory is readily available in Ecoinvent. Influence of weights in single scores depend on the scaling/normalization step more than the value of the weight itself. In each LCIA, aggregated results from a weighted sum with external normalization references show a higher weight insensitivity in RAI than outranking-based aggregation because in the former, results are driven by a few dominant impact categories due to the normalization procedure. Differences in sensitivity are caused by the notable variety (up two orders of magnitude) in the scales of normalized values for the weighted sum with external normalization and intrinsic properties of the methods including compensation and a lack of accounting for mutual differences. Contrary to the belief that the choice of weights is decisive in aggregation of LCIA results, in this case study, it is shown that the normalization step has the greatest influence in the results. This point holds for EU and World references in ReCiPe and CML-IA alike. Aggregation consisting of outranking generates rank orderings with a more balanced contribution of impact categories and sensitivity to weights’ values as opposed to weighted sum approaches that rely on external normalization references. Practitioners aiming to include stakeholder values in single scores for LCIA should be aware of how the weights are treated in the aggregation method as to ensure proper representation of values.