Abstract
Introduction
Most LCA studies are comparative and to an increasing extent the effects uncertainty are included in LCA results. This raises the question how the best option from a set of product alternatives can be selected when the product scores are uncertain. The starting point of this article is a set of Monte Carlo results for a number of alternative products.
Indicators for single product alternatives
First we discuss different ways of expressing results for product alternatives separately. This includes a discussion of centrality (mean, median, geometric mean, etc.) and dispersion (standard deviation, standard error, confidence interval, etc.).
Indicators of difference for two product alternatives
A critical review of approaches to single out the superior option on case of a comparison of two is given. This includes familiar approaches such as $$t$$
t
tests, but also lesser known ones such the Bhattacharyya coefficient and Cohen’s $$d$$
d
. All approaches are defined, discussed, and illustrated with one consistent, downloadable, example.
More than two product alternatives
The findings for two products are generalized for the multi-product situation. In particular, the issue of inflation of type I errors in multiple comparisons is discussed.
Discussion
Two main questions are identified: (1) What is the probability that a randomly selected specimen of product A performs better than a randomly selected specimen of product B? (2) How much will a randomly selected specimen of product A perform better than a randomly selected specimen of product B? These two options can both be relevant, but existing approaches for distinguishing product alternatives address one of these two only, or they even turn out to answer a different, less relevant, question. A proposal for a new indicator that addresses both questions simultaneously is offered and its use is illustrated.