Empirical Comparison of Input-Output Methods for Life Cycle Assessment


  • Yi Zhang,

  • Erin L. Gibbemeyer,

  • Bhavik R. Bakshi

    Corresponding author
    • Address correspondence to: Prof. Bhavik Bakshi, William G. Lowrie Department of Chemical and Biomolecular Engineering, The Ohio State University, Columbus, OH 43210, USA. Email: bakshi.2@osu.edu

    Search for more papers by this author


This article empirically evaluates the results from input-output (I-O) life cycle assessment (LCA) models of the United States for 2002, based on different sources of data: demand-side survey data; supply-side data with homogeneous prices; and supply-side data with heterogeneous prices. These I-O-based results are also compared with those from process LCA databases. The underlying approach is based on the theoretical insight previously developed. The results show significant differences and offer useful heuristic insight: The survey-based data set seems better for sectors that use a large amount of resources and have large intrasectoral transactions. But, it exhibits many missing values. Homogeneous price-based calculation with supply-side data is easiest because of minimal data needs and simplifying assumption on prices. The result shows relatively complete coverage and obeys mass and energy balances. The heterogeneous price-based data set overcomes the price homogeneity assumption, but the result is sensitive to price accuracy. Overall, it is not possible to declare any currently available data sets to be the best, although survey- and heterogeneous price-based data sets may be improved with more information. This insight is based on comparing direct consumption of fossil resources. Comparison of cumulative consumption reduces the disparities among data sets from error cancellation. Further comparison of process- and I-O-based inventories shows that the former values are often larger than the latter, with the two supply-side data sets being closer to process-based data. These results should be useful for choosing between available data to build the best possible I-O and hybrid LCA models and for further research on enhancing their quality.