Page 218 - Materials Chemistry, Second Edition

216 10. Advancing life cycle sustainability assessment using multiple criteria decision making

some other matters during selection of weighting method such as type of scale use, time
required to collect information and analysis, and DM’s understanding of the domain.
For example, the pair-wise comparison is suitable for a lower number of indicators, but for
a significantly higher number of indicators, the ranking method becomes efficient. Hobbs
(1980) showcased that different weighting methods lead to different results. Therefore, the
decision of an appropriate weighting method is crucial for MADM. Scenario creation in the
decision-making problem helps in formulating case-specific weighting set (Kalbar et al.,
2012), and conducting sensitivity analysis of the weights is also important (Dhiman et al., 2018).

10.5.6 Uncertainty and sensitivity analysis

To validate results from a selected MCDA method, both uncertainty and sensitivity anal-
ysis is essential. Kleijnen (1994) suggests that sensitivity analysis is also known as “what-if”
analysis, which is when the model is subjected to extreme values and limited to a set of
scenarios in which a real system could be analyzed and experimented (Figueira et al.,
2005). For example, what if the queue in a line doubles, or what-if a rule is changed for a
service from first-in-first-out (FIFO) to last-in-first-out (LIFO). Sensitivity analysis answers
two problems. One is to understand criticality of each indicator in an overall change of
results and second is to identify by what extent of alteration could change the overall results
(Kalbar et al., 2012; Triantaphyllou, 2000). Whereas, there will always be inherent uncertainty
involved in a decision problem, as a DM does not know everything with certainty, and
complexity of a decision increases with increased uncertainties, therefore, a DM should
always try to minimize uncertainties associated with different areas of a decision problem
to find out the best solution possible (Nikolaidis et al., 2004).
Nikolaidis et al. (2004) classify uncertainty into two categories: aleatory and epistemic.
Aleatory uncertainty is the uncertainties that are out of the scope of DMs, whereas, epistemic
is fully dependent on the set of choices made by DMs. There are majorly four types of
epistemic uncertainties, which are related to (i) data uncertainty, (ii) weighting uncertainty,
(iii) normalization uncertainty, and (iv) indicator uncertainty (Miller et al., 2017; Beltran et al.,
2016; Clavreul et al., 2013). Data uncertainty involves use of inaccurate data or inputs with
multiple values for analysis, thus resulting in varied models depicting reallife scenarios.
Whereas, there are different approaches to gather weight or assigning weight with varying
levels of stakeholder involvement and analysis (Miller et al., 2017; Zardari et al., 2015).
Data and weighting uncertainties get effected by stochastic, parameter, heterogeneity, and
structural uncertainties, and detail of these uncertainties with examples are provided in
Briggs et al. (2012). A detailed methodology and Monte Carlo simulation is suggested by
Barfod and Salling (2015) to handle data and weighting uncertainty, respectively. Huppes
and van Oers (2011) suggest that weighting is done to compare different types of impacts
in LCSA, however, for comparing different types of impacts, there is a need to convert
different impacts to a same level or unit.
Normalization helps in the conversion of impacts and is a mandatory step in the integra-
tion of MADM with LCSA. However, there are different normalization techniques available
in the literature, mentioned in Section 10.3.2, having different procedures to handle low and
high values (Miller et al., 2017). A good practice to minimize normalization uncertainty is to
use different normalization techniques on the same problem and make a judicious decision on