Data sources Load models
Impact assessment models
We saw in the previous task that decision making in this planning case is fairly complex, because of the uncertainties involved and the multiple criteria that have to be considered. In order to assist the planners we used the impact model eTransport that gave an approximation of the system’s performances regarding different criteria. In addition we also used another model to capture the preferences of the decision maker regarding the different criteria. This was formally called the preference model and we build it using the multi-attribute utility theory (MAUT).
The way we used in combination these two modelling tools is shown in the figure below. First as we saw in the previous task D, the impact model (in this case eTransport) was run in order to derive innitial cost and emissions information about all alternatives considered. As explained previously, assumptions had to be made for the data used as input into eTransport. Most of the input data are fed into the operations part of the analysis when the eTransport impact model is used to calculate operational attributes (e.g. operational cost, local and global emissions). An algorithm is developed, which does this for all alternatives over all scenarios. The results from the operational analysis are collected in the multi-attribute (MA) achievement matrix (see table obtained in task D) together with attributes which are independent of the operation of the system (e.g. investment cost and visual impact).
In the previous task, it has been discussed that even in the simple example presented here it becomes difficult for the planner to judge the information gathered in the table and provide trade-offs and make risk assessments. Therefore further we have used a formal approach based on decision analysis and MAUT to give the planners the possibility to think more about the planning alternatives and their impacts for the society.
A preference model has been built using MAUT in order to formally incorporate the values and the contribution of the decision makers to the final decision. The model building process consisted of two types of questionnaires. Five persons (planners) involved in the study have been expressing their concerns and expert judgements regarding the planning alternatives analysed, as they would have been the main persons in charge with the decision.
Questionnaire 1:The first type of questions where lottery questions for each of the objectives considered: the decision-maker was asked whether he would prefer an alternative with an uncertain outcome (A) or one with a certain outcome (B). The value of the certain outcome in B was repeatedly modified until the decision-maker became indifferent to these two options - see the figure below.
The outcomes values discussed with the planners were the ones in the table in task D. The purpose here was to extract information about the attitude towards risk of different decision makers. The answers to these questions were collected by the analyst and used to estimate individual utility functions - see the figure below. An exponential form for the single utilities was used.
We observed that all planners involved were risk averse when it comes to investment and operating costs. In contrast, their risk attitude varied more widely for the environmental attributes. For instance, when it comes to NOx-emissions respondents A, B and D were risk averse, E was risk neutral, whereas C was risk prone, as shown in the figure below.
Questionnaire 2:The second type of questions was the trade-off questions. Each planner was first asked which of the criteria analysed was the most important. This criterion was used as reference attribute for the trade-off comparisons. The decision maker was then asked to compare two hypothetical alternatives A and B, measured along the reference attribute and one of the other attributes, as illustrated in the figure below.
The indifference point was found by changing the reference attribute level of alternative B while keeping the level of attribute i at its best (minimum), until the respondent was indifferent between the two alternatives. This was repeated for all criteria except from the reference one.
As a general observation it appeared as if the decision makers tend to be more risk prone about criteria they care less about. In general, we had the impression that decision makers had problems expressing their risk preferences for attributes they were less concerned about.
Deriving results and reccomendations:After derived the decision makers’ preference parameters we calculated the total expected utilities (the theoretical background for these equations can be found here LINKit!) . The results in terms of expected utility and ranking of the four alternatives for the five respondents are shown in the following table:
We can observe that decision makers A, C, D and E ended up with the same ranking of the four alternatives. Alternative 3, which is ranked first for these decision makers, is also the alternative with the least expected cost, as can be seen from the table in task D.
We present in the figures below more details about the preferences of two of the five planners (C and E). The bars represent the total expected utilities for each of the four alternatives analysed. First we can observe that respondent C puts more weight on the local pollution (NOx and heat dump), and therefore ranks alternative 1 first.
We can also see that respondent E is mainly concerned with the cost figures, and do not consider heat dump at all.
The graphs give a good visualisation of how two decision makers in the same position analysing a problem, can have different preferences resulting in different decisions. It might also happen that the resulting ranking of alternatives based on the total expected utilities is the same, even if the respondents’ preferences are different. This is the case for respondents A, B, D, and E in our study.
In this example the preferences of four of the decision makers indicate that a district heating network should be build instead of reinforcing the electricity grid. Separate planning of the electricity and district heating networks could easily result in sub-optimal solutions.
Published August 3, 2012
Contact: Gerd Kjølle, SINTEF Energy Research