Causality in social life cycle impact assessment (SLCIA)

Abstract

Purpose

The social life cycle impact assessment (SLCIA) incorporates either a type I or type II characterization model. We improved both models by introducing explicit causality by using statistic modeling through development of (1) a quantitative approach to simultaneously identify impact pathways of type II models with multiple impact categories, targeting SLCIA method developers and (2) a new hybrid model to establish causality between inventory indicators and subcategories, targeting social life cycle assessment practitioners.

Methods

Causality establishments for type II impact pathways and the new hybrid model are the core requirements for this study. We used structural equation modeling (SEM) to identify the impact pathways for type II characterization models, therefore resolving the issues of unobservability and unvalidatibility in type II models. Using country-level data from the World Bank, the method was applied to an example impact pathway at macro-scale. We applied Bayesian networks in our hybrid model to address the issues of relevance and representativeness in type I models, assuming the unobservable social performances of an organization are the causes for observable inventory indicators. The method was applied to a hypothetical example for the stakeholder of the worker at company scale. Temporal precedence (i.e., lag effects) was incorporated into both models.

Results and discussion

The results from the confirmatory SEM supported our hypotheses that comprised the impact pathway from economic development to health outcomes, which were fully mediated by health expenditures and health access. A 1-year lag between each impact category resulted in the best model fit. Limitations on the data as well as subjective choice of indicators to represent impact categories are subject to criticism. The results from the hybrid model showed that, depending on the likelihood of the inventory indicators, the posterior probability of subcategories either deviated from their prior probability or behaved similarly. The construction of proper conditional probability tables and the choice of probability distribution for the likelihood are major challenges for the hybrid model.

Conclusions

This study was the first attempt in using statistic causal models to quantitatively identify unobservable impact pathways of the type II model and to develop a hybrid model for SLCIA. A SEM that incorporates temporal precedence enables identification of impact pathways with multiple unobservable impact categories. The hybrid model using Bayesian networks represents the subcategories in posterior probabilities instead of absolute scores, helping companies to better develop instructions for future management practices.