Uncertainties

The results are sometimes criticised by pointing at the uncertainties involved. And indeed uncertainties are large. But before discussing these, one has to distinguish these uncertainties from deviations of current results compared to earlier results as well as from ExternE itself and from other publications. Firstly there has been a substantial methodological development in the last ten years, e.g. from a top-down to a site-dependent bottom-up approach or with regard to the monetary valuation of health effects. So comparisons should include an analysis of whether the chosen methods are appropriate and state of the art and whether the studies are complete. Secondly, new knowledge e.g. about health impacts of course changes the results. For example the emerging knowledge that fine particles can cause chronic diseases resulting in a reduction of life expectancy changed the results considerably. An assessment always reflects current knowledge. That an assessment changes with new knowledge - and also may change due to a change in people's preferences - is natural and not a methodological problem.

Individual sources of uncertainty then have to be identified and quantified. It is appropriate to group them into different categories, even though there may be some overlap:

  1. data uncertainty, e.g. slope of a dose-response function, cost of a day of restricted activity, and deposition velocity of a pollutant;
  2. model uncertainty, e.g. assumptions about causal links between a pollutant and a health impact, assumptions about form of a dose-response function (e.g. with or without threshold), and choice of models for atmospheric dispersion and chemistry;
  3. uncertainty about policy and ethical choices, e.g. discount rate for intergenerational costs, and value of statistical life;
  4. uncertainty about the future e.g. the potential for reducing crop losses by the development of more resistant species;
  5. idiosyncrasies of the analyst e.g. interpretation of ambiguous or incomplete information.

The first two categories (data and model uncertainties) are of a scientific nature and can be analysed by using statistical methods. Results show a geometric standard deviation of ca. 2 to 4, which means that the true value could be 2 to 4 times smaller or larger than the median estimate. The largest uncertainties lie in the exposure-response function for health impacts and the value of a life year lost - current research is directed towards reducing these uncertainties, which reflect our limited knowledge.

Furthermore, certain basic assumptions have to be made e.g. such as the discount rate, the valuation of damage in different parts of the world, the treatment of risks with large impacts or the treatment of gaps in data or scientific knowledge. Here, a sensitivity analysis should be and is carried out demonstrating the impact of different choices on the results. Decisions then would sometimes necessitate a choice of the decision-maker about the assumption to be used for the decision. This would still lead to a decision process that is transparent and, if the same assumptions were used throughout different decisions, these would be consistent with each other. If uncertainties are too large, as currently still is the case for global warming impacts, shadow values could be used as a second best option. Shadow values are inferred from reduction targets or constraints for emissions and estimate the opportunity costs of environmentally harmful activities assuming that a specified reduction target is socially desired.

Despite these uncertainties, the use of the methods described here is seen to be useful, as

  • the knowledge of a possible range of the external costs is obviously a better aid for policy decisions than the alternative - having no quantitative information at all;
  • The relative importance of different impact pathways is identified (e.g. has benzene in street canyons a higher impact on human health as fine particles?);
  • the important parameters or key drivers, that cause high external costs, are identified;
  • the decision making process will become more transparent and comprehensible; a rational discussion of the underlying assumptions and political aims is facilitated;
  • Areas for priority research will be identified.

It is however remarkable that despite these uncertainties certain conclusions or choices are robust, i. e. do not change over the whole range of possible external costs values. Furthermore, it can be shown that the ranking of e.g. electricity production technologies with respect to external costs does not change if assumptions are varied. Thus, the effect of the uncertainty of externalities depends on the application. The key question is: What is the increase in total life cycle cost to society if one makes the wrong choice? A detailed analysis of this question in a specific situation involves the probability distribution of the total social cost for each of the options under consideration, to estimate the expectation value of the social cost or the probability of making the wrong choice.

It should also be noted that gaps can be closed and uncertainties reduced by performing further research (e.g. further contingent valuation studies and epidemiological studies).