Skip to main content

Decision making, systems thinking and decision support

(R.P. Hämäläinen, A. Salo, E. Saarinen, J. Liesiö, P. Mild, J. Mustajoki, A. Punkka, S. Slotte, V. Brummer, M. Lindstedt, M. Marttunen)

The research team is active in a wide forum of decision analysis. Our interests cover the theoretical development of new decision support techniques as well as the practical implementation of decision support tools. Our research topics also include behavioral aspects of the decision making and new techniques for group decision making. For definitions of commonly used words in a decision making context see: A Lexicon of Decision Making.

Preference Programming - Methods for the approximate specification of preferences

(A. Salo, R. P. Hämäläinen, A. Punkka, J. Mustajoki, P. Mild, J. Liesiö)

Different ways of admitting approximate preferences have attracted much attention in the decision analysis literature. Theoretical work has been done to extend methods, such as the analytic hierarchy process (AHP) and value tree analysis, so that the decision maker can express approximate preference statements through interval judgments. The decision maker specifies a range of numbers to indicate the relative importance of two factors at a time. The resulting linear constraints define at each criterion a feasible region which consists of the local priorities that are consistent with the decision maker's statements.

[Figure 2.1.1]
Local weights constrained by interval judgments

Preference programming extends AHP by synthesizing interval preference statements in the hierarchy to derive intervals for the weights of alternatives P[SAL90b, SAL91b, SAL92a, SAL92b, SAL95b], M[SAL03b]. The weight intervals for the alternatives are computed by a series of linear programming problems. PAIRS (Preference Assessment by Imprecise Ratio Statements) is a similar approach for value tree analysis P[SAL92a]. In PAIRS the alternatives' characteristics are modeled through ranges of scores on the lowest twig-level attributes. These methods assume that the ratio statements are consistent with each other and consistency bounds are used to help the decision maker to avoid inconsistencies P[SAL93]. Interval versions of the SMART and SWING procedures are special cases and can be called interval SMART/SWING or simple PAIRS = SPAIRS P[MUS05]. Our WINPRE decision support software runs these methods and can be used free for academic purpose.

In PRIME (Preference Ratios In Multiattribute Evaluation) the decision maker evaluates ratios of value differences P[SAL01b]. The decision maker enters these ratios as intervals of numbers. PRIME is related to the trade-off technique as the weighting procedure explicitly uses the attribute ranges. This normatively attractive feature encourages to develop and operationalize further the PRIME method. PRIME Decisions is the related windows software P[GUS01].

In RICH (Rank Inclusion in Criteria Hierarchies), the decision maker is allowed to specify subsets of attributes which contain the most important attribute or, more generally, to associate a set of rankings with a given set of attributes P[SAL05]. Such preference statements lead to possibly non-convex sets of feasible attribute weights, allowing decision recommendations to be obtained through the computation of dominance relations and decision rules. Support for the use of RICH is provided by a web-based decision support tool, RICH Decisions.

Methods based on interval presentation of approximate preferences lead to a more interactive decision support process than methods which use single number estimates to assess preference statements. The evolution of the priorities can be computed and visualized throughout the process. Our experience suggests that the use of preference intervals can be especially helpful in group decision making P[SAL95a], P[HÄM95c], P[HÄM96a], P[HÄM96b]. The group's preferences can be modeled by taking the union of the members' judgments. The decision makers seek consensus by trying to compromise on their individual judgments and reduce the width of the preference intervals. This process helps to focus the discussion on the key issues and this is expected to increase the efficiency of the negotiation process.

Interval techniques can also be used to carry out global sensitivity analyses. The sensitivity of the model parameters can be studied by extending the point estimates into intervals and examining the consequential variations in the overall values. Intervals can be simultaneously given for all both the preference statements and to the decision outcomes, when the effects of different types of uncertainties can be studied together P[MUS06].

Another application area of interval techniques is to use them to model generally assumed information about the decision makers' preferences. This approach has been applied to support different phases of the even swaps process, which is an elimination process based on value trade-offs. In the approach, the even swaps process is carried out as usual, but in parallel, the evolution of the preferences of the decision maker is modeled with preference programming P[MUS05b]. This model can provide information to the even swaps process to help identify practically dominated alternatives, and to find applicable candidate attributes for the next even swap. The approach does not compromise the original idea of an easy-to-use process. Smart-Swaps web software has been developed to support the approach in practice M[MUS06b].

Dynamic multicriteria choice

(R. P. Hämäläinen)

Multicriteria approaches to dynamic decision making settings have received very little interest in the literature. There are many important practical dynamic problems and the topic needs further research. We testing the use of dynamic goal programming model where both goal points and goal sets can be specified. The definition of goals as sets rather than points is a way to introduce flexibility and robustness into the models. We consider a single family house heating problem (originating from the electricity pricing project described elsewhere in this report) as a test case for evaluating different kinds of multicriteria decision support methodologies. The family faces a dynamic trade-off problem between the hourly comfort (temperature) and heating cost. Different concepts including the interval target setting approach have been implemented. The MOHO (Multicriteria Objective Heating Optimization) software was successfully tested in homes P[HÄM02], P[MÄN97], T[HÄM96]. The heating problem has also been used as an computational tool in examining consumption strategies and tariff coordination in a deregulated electricity market P[HÄM00b], P[PIN00], P[HÄM99b]

[MOHO (Main window)]

Similar dynamic decision making settings are also present in water resources management. In cooperation with Finnish Environment Institute (FEI) we have developed a regulation simulator for Päijänne-Kymijoki lake-river system called ISMO (Interactive analysis of dynamic water regulation Strategies by Multicriteria Optimization) working under the spreadsheet program Excel. The simulator is actively used by FEI to generate regulation strategy alternatives for different stakeholder groups P[HÄM01a], P[HÄM98].

[ISMO (Main window)]

This work is also related to our research on the agent based modelling methods P[HÄM95d], P[HÄM96c] and dynamic games.

Computer support - Decisionarium

See also the project site.

(R. P. Hämäläinen, A. Salo, J. Mustajoki, J. Liesiö)

We are actively developing computer aids for decision making. These software are collected on the Decisionarium Web site P[HÄM03]. In our software development we emphasize the possibilities of the latest developments in information technology, such as multimedia and the Internet.

HIPRE 3+ is known for its user friendliness and it is based on a fully graphic interface. It is the first fully graphical mouse-driven implementation of AHP and value tree analysis. It has many attractive features such as the possibility to customize the scale, i.e. the verbal expressions and numbers used in the ratio comparisons. HIPRE 3+ lets you combine different approaches such as AHP and value functions in one model. It includes different options for the colours of the interface as well as the possibility to use three languages (English, Finnish, Swedish). P[HÄM92c], P[HÄM92d], R[HÄM92].

In response to the development in the Web technology we have produced Web-HIPRE, the first Web-based multiattribute decision analysis tool for value trees and AHP P[MUS00]. It is a Java-applet for multiple criteria decision making based on HIPRE 3+. Being located on the Internet, Web-HIPRE can be accessed from everywhere in the world, which opens up a new dimension in decision support. Web-HIPRE provides a common platform for individual and group decision support. The individual models can be processed at the same or at different times and the results can be easily shared. An essential Web feature is the possibility to define links to other Internet-addresses. These links can refer to graphical or any other kind of information such as sound or video describing the criteria or alternatives. This can improve the quality of decision support. Web-HIPRE supports several weighting methods including AHP, SMART, SWING, SMARTER and value functions. The results are shown by bar graphs and the sensitivity analysis. Web-HIPRE supports also the use of regular HIPRE models. The on-line use of Web-HIPRE can be tested by the sample models.

[Web-HIPRE]
Main Window of Web-HIPRE

We have also developed easy to use software to support the interval techniques. WINPRE is a software for supporting the Preference programming, PAIRS and Interval SMART/SWING, and PRIME Decisions is a software implementation of the PRIME method P[GUS01]. Both these software run under Windows and are avalailable free for academic use. RICH Decisions is a web-based software for supporting the RICH method.

[Winpre]
Main Window of Winpre

[Winpre]
Main Window of Prime Decisions

Smart-Swaps is a web software to support the even swaps process M[MUS06b]. It implements the preference programming approach to provide information for the even swaps process to help identify practically dominated alternatives, and to find applicable candidate attributes for the next even swap P[MUS05b].

The area of Computer Supported Cooperative Working (CSCW) has grown rapidly during the past few years. This area has had its primary focus on computer supported communication whereas the area of Group Decision Support systems (GDSS) studies tools for group decision making. HIPRE 3+ Group Link is a GDSS software which supports on-line group decision making with preference programming P[HÄM92b], P[HÄM94], P[HÄM95a], P[HÄM96b]. With HIPRE 3+ Group Link individual preference models created with HIPRE3+ can be combined into a preference interval model. Through a PC-network the group members can change their own preference models based on the feedback they get from the interval model of the group. This technology enables the use of several different approaches for group decision making (see Figure below). Web-HIPRE provides a possibility to combine individual preferences into a group preferences with an weighted arithmetic mean method. The group model collects the individual preferences directly via the Web allowing the use of a distributed mode. We are also currently exploring the possibilities for distributed multimedia based decision structuring and conferencing.

[HIPRE 3+ Group Link]
Group decision support with HIPRE 3+ Group Link

Decisionarium also provides software for negotiation support.

Behavioral aspects of decision making

(R. P. Hämäläinen, A. Salo, M. Hjelt)

Biases in decision making have been the subject of many recent experiments. Experimental work with decision makers is essential to ensure that decision support techniques can be used reliably. Especially new techniques, such as the use of preference intervals, require careful empirical validation. Our experimental research started from AHP. The supermatrix technique which has been suggested as a remedy for the phenomenon of rank reversals in AHP was shown not to be the final answer R[SAL92b]. In AHP a decision maker can use verbal expressions to facilitate the preference elicitation. Verbal expressions are converted into numbers according to the nine point integer scale which has weaknesses described in P[SAL97]. The use of verbal expressions in decision making and alternative ways of using words in preference elicitation are studied in P[PÖY97].

Our comparative studies include all value tree weighting methods. Previous literature has shown that the commonly used methods yield diverging results. This makes it important to study the origins of these differences. An international experiment in the Internet is run to further study the differences of the weighting methods (AHP, SMART, SWING, Direct weighting and Tradeoff weighting) P[PÖY01a]. The structure of the experiment is shown in the Figure below. The results from the experiment suggest that the differences between the weights derived with different methods are due to response scale effects. All the methods used in value trees have same theoretical foundation and thus we feel that none of the methods can be claimed to be superior.

[The Internet experiment]
The Internet experiment on different weighting procedures.

Our experimental work continues with the biases related to the formulation of a decision problem. There are experiments showing that the structure of a value tree, i.e. the number of the levels in a value tree or the number of attributes, affects the results. Our theoretical considerations show that these problems can have also mathematical origins P[PÖY98]. We run a new experiment studying the decision of attributes in value trees affect the results R[PÖY98a], P[PÖY01b]. New findings are that the decision of attributes either increases or decreases the height and changes the rank of groups of attributes for an individual decision maker. We also suggest that many different attribute weighting biases have the same origins R[PÖY98a], P[PÖY01b]. However, we think that in practice the biases can be avoided or at least their effects can be weakened through interactiveness and teaching R[PÖY98a], P[PÖY00].

Environmental Decision Making

See also the project site.

(R. P. Hämäläinen, M. Marttunen, J. Mustajoki, J. Seppälä)

The decision analysis field has often encountered difficulties in transforming theoretical ideas into practical decision support tools. However, the decision support systems developed in the laboratory have also been widely and succesfully used, for example, in real life environmental problem solving. Water resources management has been one of the main research areas and the laboratory has participated in several research projects on lake regulation. In these projects, we have studied different kind of approaches and tools including decision analysis interviews P[MAR95], M[MAR06] modeling of the lake-river system P[HÄM01c], negotiation P[HÄM99a, HÄM01a], and public participation P[MUS04, MUS06b], M[MUS06].

The other main areas of research projects are energy policy P[HÄM88a, HÄM89b, HÄM90a, HÄM90b, HÄM91a, HÄM92a, HÄM96d], and environmental impact assessment P[HÄM92c, MAR95]. These projects have ranged from a corporate planning systems in artificial intelligence environment P[SAL88a, SAL88b, SAL89, SAL90a, SAL91a, SAL91d] to parliamentary decision making P[HÄM88a, HÄM90a, HÄM90b, HÄM91a, HÄM95c]. Besides the research projects the laboratory offers help for collaborators in many areas. The succesful projects have made us well known as a center of excellence in practical decision support.

Practical problem solving is also supported by experimental studies using the same real problems in a university environment. The first group decision making experiment using preference programming was carried out with Finnish student politicians P[HÄM92b]. They used preference intervals to negotiate on energy policy. The study was continued in a study dealing with a metropolitan traffic planning problem P[HÄM96b]. Here the HIPRE 3+ Group Link environment was used in a computer network. The experiment showed that the preference programming approach can inspire the participants actively communicate about values. The different ways of using preference programming in group decision making is one of the main topics for our future decision support applications.

Life cycle assessment as a decision support tool

Life cycle assessment (LCA) is a method for analysing and assessing the total environmental effects of a product P[LIN95], T[SEP97], P[SEP01], T[SEP03]. The method has both objective and subjective elements. So far, LCA has focused on gathering data on the mass and energy flows of the product over its whole life cycle. However, data alone is insufficient to support decisions without a careful multi-attribute valuation. An explicit decision analytic approach is still missing. In our research we are focusing on the subjective parts of LCA where the largest benefits seem possible. These parts should be affected by the decision maker s preferences and the decision making situation.

We are developing procedures for Goal definition and scoping phase so that it would start from explicit formulation of the related decision problem and identification of the relevant attributes by for example interviewing interest groups P[MAR95]. So far, many LCA researchers have suggested using fixed weights in the product s total environmental impact index. This cannot be recommended, because the attribute ranges and the decision maker's preferences would have no effect to the result. We try to show how to correctly use valuation techniques to help in the data interpretation P[MIE97], P[MIE99].

Decision conferencing, group decision making and decision analysis interviews

( R. P. Hämäläinen, K. Sinkko, J. Mustajoki, M. Lindstedt).

Decision analysis can successfully be used in modern public policy processes. We have developed two new "on the spot" analysis techniques. The decision analysis interview is a way of involving stakeholders and collecting preferential information from citizens groups. Spontaneous decision conferencing P[HÄM95c], P[HÄM96a] is another closely related concept where groups of decision makers spontaneously take a decision analytic approach to their problem solving. We have applied these techniques, for example, in environmental decision making as well as in technology foresight and assessment. We have also tested the applicability of new multicriteria approaches to negotiation settings P[HÄM99a], P[HÄM01a].

One of the application areas is to study how decision support could be used in everyday high level political decision making. We have worked with the parliamentarians P[HÄM88a, HÄM91a, HÄM92a]. For example, two groups of three parliament members used the HIPRE 3+ and ComPAIRS software packages in their work P[HÄM95c], P[HÄM96a]. The aims of the study were to see the benefits of different techniques in everyday decision making and in supporting group decision process. The cases included the decision about Finland's membership in the E.U.

[Join EU?]
An example of the preference profile screen in the problem of whether Finland should join the European Union or not.

Decision analysis interviewing has been applied to environmental planning. Stakeholders and local citizens are interviewed interactively by means of a value tree. The resulting prioritizations are summarized and included in the project material for the final decision making. This approach has proven to be fully operational and it is considered most useful in our studies which dealt with water resource planning and management P[MAR95] .

We have also participated in European joint projects on the testing of decision conference techniques in a major nuclear accident case. In these conferences, we have considered the early phase countermeasures for the public R[FRE96], T[LIN98], R[HÄM98], P[HÄM00a], R[HÄM00], as well as the later phase countermeasures for the milk pathway R[AMM01]. We have also studied different approaches to involve the key players in planning protective actions P[SIN04, MUT06, GELxx] and applied interactive multicriteria group decision support software in these conferences P[MUS01, MUS07].

Systems intelligence, dialogue and philosophy

(E. Saarinen, R.P. Hämäläinen, S. Slotte)

By Systems Intelligence R[BÄC03], R[HÄM04], R[HÄM04b], we refer to intelligent behavior in which one identifies and is able to intelligently encounter structural entities consisting of interaction and feedback. A person who reflects Systems Intelligence takes creatively and appropriately into account his or her environment, himself/herself and the systemic interaction which they create. He or she is able to act in such situations in an intelligent way.

To us, it began to appear that Systems Intelligence is one of the basic elements in human behavioral intelligence. While Systems Thinking observes and models interactions from outside, Systems Intelligence represents active practical thinking in true situations where one is involved. The new elements include the active, personal and existentially relevant contents.

Our project thrives to portray the concept of Systems Intelligence, its different forms of manifestation and its constitutive role in human reality.

By "Dialogue" we refer to a set of techniques designed for the purpose of developing better communication, collaboration and for reaching consensus. We are especially interested in improving and testing various theories and methodologies in differenent situations and for different means (improving communication, conflict management, decision making).

See the Systems Intelligence project.

Technology foresight and assessment

(A. Salo, T. Gustafsson, P. Mild, J.-P. Salmenkaita)

Technology foresight and technology assessment can be seen as interactive processes which seek to produce an enhanced understanding of the potential and likely consequences of deploying new technologies P[SALxxa]. In technology assessment, the emphasis is often on an analysis of the impacts which may be caused by a given technology, including its negative and harmful consequences P[SALxxb]. Technology foresight, on the other hand, can be viewed as an attempt to consider the future for the purpose of identifying research areas, technologies, and joint actions which contribute to the attainment of desired economic and social benefits.

In this area, we have carried out research projects in collaboration with the National Technology Agency (Tekes) and other public organisations in order to support the shaping of innovation policies. Specifically, we have explored the impacts of using group support systems (GSS) in a research and technology development programme on telecommunication research P[SALxxb], as well in the evaluation of a national cluster programme on forestry and forest industry. In these projects, we have promoted participatory approaches where stakeholders are assisted by GSS methodologies in the consideration of past achievement and future opportunities.

More formally, technology assessment and foresight can be assisted by appropriate methodologies, such as scenarios P[BUN93], cross-impact analysis P[SAL95a], and time-series analyses P[BUN96]. However, apart from methodological choices, organisational incentives also need to be considered to ensure that the participants are fully motivated P[SAL01] and the management structures support the validation and dissemination of information P[SALxxa]. The need to bring in elements from technology assessment and foresight to programme evaluation has also been increasingly recognised P[SAMxx].

Risk management is an integral part of the assessment of controversial technologies. If the potential impacts of a new technology are highly uncertain, it may be pertinent to invoke the precautionary principle which - as stated in the 1992 Rio Declaration on Environment and Development - implies that lack of full scientific certainty shall not be used as a reason for postponing cost-effective measures to prevent environmental degradation, if there are threats of serious irreversible damage. The operationalisation of this principle can be supported by multi-criteria tools which capture the uncertainties caused by the lack of scientific evidence or incomplete preference information. These tools help eliminate the limitations of conventional cost-benefit analyses and contribute to the transparency of decision support (see, e.g., P[SAL92a], P[SAL92b], P[SAL95a], P[SAL95b], R[SAL00b]). Multicriteria tools can also be used to modelling uncertainties through fuzzy sets P[SAL96b] or ranges of probability values P[SAL96a].

At the firm level, GSSs hold promise in the context of new product development. They seem particularly helpful during the earliest phases of product development which are focused on the collection and evaluation of customer requirements. Here, GSSs allow distributed organizations to collaborate on the generation, refinement and systematic evaluation of new product ideas, which helps in the shaping of a successful product specification. Our empirical research suggests that groupware is especially useful in large-scale industrial organizations which have intensive communication needs and seek to shorten their product development cycles P[SAL98a], P[SAL99a].

Industrial investment opportunities

(A. Salo, J. Gustafsson, J. Dietrich)

Decision analysis and investment science can be applied to derive estimates for the likely cash flows of industrial investment opportunities and to assess their risks as well. Here, suitable methods include, among others, decision trees, scenario trees, real options analysis, linear optimization models and multicriteria models. These methods can also be employed to estimate the value of new businesses P[GUS01], or to analyse the profitability of major capital investments subject to correlated changes in input prices P[BUN96].

From the viewpoint of investment appraisal, research and development (R&D) projects are particularly challenging, because they involve considerable expenditures and uncertainties; yet, they are central to the development of the firm's future competencies and its competitive position. In this context, a multicriteria models are typically needed, as there is a need to account for several intangible characteristics (e.g., commitment and experience of the project manager) P[RUU89c].

To support the appraisal of R&D projects and other industrial investment opportunities, we have developed a method called Contingent Portfolio Programming (CPP) P[GUSxxa]. In CPP, uncertainties are modelled through a scenario tree while the structure of follow-up and prerequisite projects are captured through a set of linear constraints. Furthermore, CPP employs a mean-risk utility model for portfolio choice under risk. CPP is also computationally appealing, because it leads to linear programming models.