The Generic Multi-Attribute Analysis (GMAA) System is a Decision Support System (DSS) based on an additive multi-attribute utility model that accounts for incomplete information concerning the inputs. The system is intended to allay many of the operational difficulties involved in the Decision Analysis (DA) cycle, which can be divided into four steps: structuring the problem (which includes specifying objectives, building a value hierarchy and establishing attributes for the lowest-level objectives); identifying the feasible alternatives, their impact and uncertainty (if necessary); quantifying preferences (which includes the assessment of the component attribute utilities as well as the value trade-offs); evaluating strategies and performing Sensitivity Analysis (SA).

The user can interactively create or delete nodes and branches to build or modify the objectives hierarchy. A name, label and description must be entered for each objective, as well as the respective attribute units and ranges for the lowest level objective. The system also accounts for attributes with a subjective scale.

Alternatives and their consequences, in terms of the attributes associated with the lowest-level objectives, can be easily entered by hand or loaded from file. The system admits uncertainty about consequences, which leads to uniformly distributed ranges for each attribute instead of single values. Note that both endpoints being equal would be equivalent to the case under certainty, where the policy effects for an alternative in an attribute are precisely known. Alternatives with missing consequences, that is, alternatives that do not provide values or consequences for some attributes in the hierarchy, can be represented by the respective attribute range. The user can add or delete alternatives and modify their respective consequences.

The next step in DA consists of preference quantification, which involves assessing the DM's component utilities for the attributes and the relative importance of criteria. In both cases, the system admits incomplete information through value intervals as responses to the probability questions the DM is asked, which leads to classes of utility functions and weight intervals, respectively. This is less demanding for a single DM and also makes the system suitable for group decision support, because conflicting individual views or judgments in a group of DMs can easily be captured through imprecise responses.

With respect to component utilities assessment, the system provides three procedures for building an imprecise piecewise linear utility function (providing up to three intermediate points, which can be dragged by the mouse to achieve the right shape), providing imprecise utilities for discrete attribute values, or using a method based on the combination of two slightly modified standard procedures for utility assessment, the Fractile Method and the Extreme Gambles Method.

The system provides two weight elicitation procedures, a direct assignment, which is perhaps more suitable for upper level objectives, which could be more political, and a method based on trade-offs, which is more suitable for the low-level objectives in the hierarchy, because the weight assessment involves a more specific area of knowledge.

Once the DM's preferences have been quantified, the different alternatives under consideration can be evaluated by means of an additive multiattribute utility function. The additive model is used to assess, on the one hand, average overall utilities, on which the ranking of alternatives is based and, on the other, minimum and maximum overall utilities, which give further insight into robustness of such ranking. It is also possible to select another objective to rank by.

The system provides different displays of ranking results: the Stacked Bar Ranking is similar to the alternatives classification but provides more detail of how the alternative's average utilities for the attributes affect the average utility of the Overall Objective, the Measure Utilities for Alternatives displays a bar graph showing performance of a single alternative for the attributes, taking into account average consequences and individual utilities, and where the width of an attribute is proportional to its weight, the Compare Alternatives Graph provides a detailed comparison of the differences between two alternative and the Paired Attributes Correlation display evaluates/compares alternatives component utilities with respect to pairs of selected attributes.

Finally, the SA is performed. The system provides several types of SA, like classical SA, which involves changing the parameters and observing their impact on the ranking of alternatives. Hence, if the DM modifies an average normalized weight, normalized weight interval bound, component utility or alternative consequence, the system takes charge of how these changes are propagated through the objectives hierarchy and automatically recalculates the overall utilities for each alternative and the resulting ranking. Another way of performing SA involves assessing the interval in which the average normalized weight for any objective at any level or branch of the objective hierarchy can vary without affecting the overall ranking of alternatives.

The assessment of non-dominated and potentially optimal alternatives and the application of Monte Carlo simulation techniques take advantage of the useful imprecise information collected during the assignment of the component utilities and weights and the uncertain alternative consequences entered. In some cases, the information obtained from the alternatives evaluation, by means of the additive multiattribute utility model, is not meaningful enough so as to definitively recommend an alternative, i.e., we get very overlapped imprecise overall utilities. In these cases, the above techniques play a very important role. They may provide more meaningful information and an iteration process can be carried out by tightening the respective imprecise alternative consequences, component utilities and weights and reassessing the non-dominated and potentially optimal alternatives or performing the Monte Carlo simulation techniques again, until a dominant strategy is found.