Page 353 - Materials Chemistry, Second Edition

5.3 Techniques for Result Analysis 337

‘Monte Carlo’) result in a just as large number of results (mostly per aggregation
as impact indicator result). For a sufficiently high number of simulations (e.g.
1000) a result can be represented as a probability distribution that in turn, in
simple cases, can be characterised by average values and standard deviations. Then
a numerical result like ‘120 kg CO equivalents’ can be provided, for example, as
2
‘120 ± 10 kg CO equivalents’.
2
For reasons of applicability, only inputs as normal distributions (Gauss) with
average value and standard deviation are possible along with even distributions
with highest or lowest value: the triangular distribution and logarithmic normal
distribution for individual parameters have been proposed (Huijbregts, 1998, loc.
cit.). Sometimes discrete single values with defined probabilities are also used as
input. A graphical representation is best suited for an immediate illustration of the
output. Tables of all average values and deviations tend to be confusing. Limitations
to the methodology concern the availability or non-availability of statistical figures
for the input data (which are often mere estimations) and computing times for up
to 10 000 iterations.
Comparative analysis covers the systematic simultaneous listing of product sys-
tems alternatives. Since the most common application of LCA is targeted towards
a comparison of product systems, this method is particularly important. Inventory
and indicator value results often only differ within small margins and one has to
resist the temptation to stress minor advantages of one system in relation to the
other. This would lead to an over-interpretation of results. A comparative analysis
can be made at all levels of results: inventory, characterisation (indicator result),
normalisation and weighting, if required. For this purpose, either absolute values
or presentations on a percentage basis with the highest value corresponding to
100% can be used. By routine, both approaches are already being used in the
impact assessment (see Figure 4.5). In the interpretation phase, a critical anal-
ysis should be made as to whether the first impression complies with the data
analysis.
Discernibility analysis is of special relevance in comparative LCAs as well. Espe-
cially in the case where several product systems need to be compared, a ranking
is targeted. Unfortunately, however, the results are, in the rarest cases, unam-
biguously in favour of one of the analysed systems as proponents of diverse
single-point methods (aggregation of all results into one figure) would like them
to be. A qualitative statement like ‘product A is significantly better concerning the
consumption of fossil resources compared to product B’, statistically characterised
with the (frequently used) significance benchmark of 0.05 should read, ‘With 95%
probability product A in this concern is superior to B’ (Heijungs and Kleijn, 2001,
loc. cit.). Discernibility analysis tends to combine comparative and uncertainty
analyses. The most important tool is therefore the Monte Carlo simulation for as
many results of compared LCAs as possible. This implies the product system to
be ecologically ‘favourable’ with more parameters above a significance threshold
or below the threshold for damage indicators. A quantitative method has been