Group model building: a decision room approach

Jac A.M. Vennix

Cécile M. Thijssen

Etiënne A.J.A. Rouwette

University of Nijmegen, Dept. of Methodology

PO Box 9104, 6500 HE Nijmegen

The Netherlands

fax: **31-24 3612351

tel: **31-24 3615568

e-mail: J.Vennix@MAW.KUN.NL


Increasingly system dynamicists involve stakeholders in building a model. Group model building has become an accepted way of doing this (Vennix, 1996; Vennix, Richardson and Anderson, 1997). In group model building a group of stakeholders develops the model in one or more structured meetings, with the aid of a facilitator. In this paper a novel approach to group model building is discussed in which a decision room is used. We will describe the implementation of this approach with a group of students, expected results and the research design used for assessing results. The analysis of results is described in a full paper that will be handed out at the conference and be made available through the virtual proceedings.

Group model building in a decision room

The group decision room at Nijmegen University consists of thirteen computers in a network, twelve of which can be used for a maximum of twenty-four participants and one for the chauffeur. On all computers software such as GroupSystems (Ventana, 1994), Vensim and Powersim is installed. The participants sit opposite a video screen on which each computer screen can be projected. There are a number of reasons to explore the possibilities this room offers to support group model building. First of all, it offers the possibility for all participants to give and evaluate data simultaneously (avoiding the waiting time involved in a freely interacting group), to have subgroups working on different tasks and to make plenary presentations of results. This offers a way to expand the number of people involved in a group model building session. Finally, it is often maintainted that new technology puts distributed non-synchronous decision making within reach. As a first step in exploring this claim with regards to system dynamics modeling, we will assess the advances and drawbacks of model building in a decision room in a pilot study.

The participants in this study were thirty graduate students in policy sciences, age 20 to 27, in their 3th to 6th year of study and most of them with no experience in the problem to be modeled. Students' enrollment in this system dynamics course was voluntary. After enrollment it was made clear to them that the grade for the course was dependent on preparation for and participation in all stages of the modeling process.

The course consisted of nine meetings held once every week and some preparation in between, to be done individually or in small groups. In the first meeting an introduction to system dynamics was given, in addition to information about the course and what was expected from participants. The second meeting focused on principles of system dynamics; following an introduction students completed a number of exercises around key concepts. At the end of the second meeting the problem to be modeled was addressed. In order to make the course as realistic as possible, it was decided beforehand to model a problem in a 'real' organization. A regional Dutch hospital proved to be both large enough to allow modeling at an abstract level as well as accessible to both staff and students. The central problem was defined as: how can hospital management keep occupation of hospital beds at the desired level without increasing waiting lists too much? Given the limitations of the network in the decision room, the students worked in two groups of fifteen people from the third session on. In both of these groups model building started by clearly stating the problem to be studied. Participants were then asked to identify concepts in the problem using electronic brainstorming: everybody types concepts individually on their computer which are then immediately added to a plenary list that is visible for the whole group.

In order to make it possible to work simultaneously on different parts of the model, students were asked to categorize variables in a few groups, each to be addressed by a subgroup of three or four participants. The ensuing discussion resulted in four groups of variables: patients; personnel; and two financial categories: costs/benefits and assets/liabilities. In the two following sessions subgroups worked on their model with the aid of three facilitators. They used information about the hospital such as annual accounts. In addition a financial manager of the hospital was available for additional information throughout the course. The sixth meeting consisted of a plenary presentation of structure and dynamics of submodels, after which the rest of the group had the opportunity to criticize and complement the submodel by using devil's advocate. In the rest of session six and in session seven the group connected the submodels two by two in one overall model. In the eight session the overall model was analyzed and policy experiments were conducted. Students wrote a report about the model and, in the final session, presented their conclusions to the hospital management.

Expectations about results

Group model building in a decision room is expected to differ from the traditional approach in a number of ways. In working with subgroups the role of the facilitator(s) is less dominant and more work is done by the members of the subgroup themselves. In subgroups members will interact with only a few others and therefore have more opportunity to make a contribution, so we expect task oriented communication to increase. The smaller group is expected to yield more satisfaction with the process. However, there is a possible downside that more time is spent in smaller groups. Compared to the traditional approach, people might not get as much knowledge of the assumptions of participants outside of their own subgroup. We expect the discussions in the plenary group and the use of devil's advocate to be sufficient to counteract this. The same argument can be made with regard to knowledge about the problem; as students participate only in discussions about one small part of the problem, and learn about the rest only after it has been modeled, wouldn't their learning be limited to part of the problem? Again, we expect plenary discussions about the connection of the submodels and the analysis of the overall model to counteract this.

The expectations about learning as a result of participation in model building can be further specified. Following Vennix (1990), learning is taken to be an increase in the number of concepts, relationships and dynamic characteristics associated with a problem. Apart from the content of what is learned, Verburgh (1994) views an increase in system dynamics-format of knowledge as a potential gain of participation in model building. In addition to these measures of design logic, Richardson et al. (1994) propose a measure of operator logic should be included to determine the amount of learning. Operator knowledge is used in controlling a systems' behavior and consists of knowledge about ends (goals to be reached), means (policy lever and tactics) and heuristics linking means and ends (chunks of strategic insights). Learning in itself is expected to have two additional consequences. First, as a result of activating and structuring a person's knowledge in modeling, Vennix (1990) expects interest in the subject matter to increase. Second, because participants discuss and share their views of the problem and jointly build a system dynamics model, participants' knowledge is expected to grow more aligned; participants are expected to reach more consensus about the problem and commitment to actions to alleviate the problem (although in this study participants are not expected to undertake any action themselves).

Some remarks about research design and data gathering

To explore the consequences of this novel approach to group model building, the classical pretest posttest control group design is best suited (Cook and Campbell, 1979). However, this design was not feasible in this study, because the number of subjects is already quite small and because it was not considered appropriate to have students take two different courses. Instead we chose to employ a one-group pretest posttest design, which also plausibly rules out threats to a valid measurement of the effects of decision room group model building. Results will be compared to the studies of Vennix (1990), Verburgh (1994) and Huz et al. (1996). As much as possible, variables will be operationalised and measured in a way comparable to these studies.

As noted in the introduction, results of the study will be reported in a full paper to be handed out at the conference and made available through the virtual proceedings.


Cook, Th.D. and D.T. Campbell. 1979. Quasi-experimentation: Design and Analysis for Field Settings. Chicago: The University of Chicago Press.

Huz, S., Richardson, G.P, Anderson, D.F. and R. Boothroyd. 1996. Evaluating group model building in mental health and vocational rehabilitation service delivery. In System Dynamics 1996. eds. G.P. Richardson and J.D. Sterman. 233-236.

Richardson, G.P, Anderson, D.F., Maxwell, T.A., and T.R. Stewart. 1994. Foundations of Mental Model Research. In System Dynamics 1994, ed. E.F. Wolstenholme. 181-192. System Dynamics Society, 49 Bedford Rd., Lincoln, MA 01773, U.S.A.

Vennix, J.A.M. 1990. Mental Models and Computer Models. Design and Evaluation Of A Computer-Based Learning Environment For Policy Making. University of Nijmegen.

Vennix, J.A.M. 1996. Group Model-Building: Facilitating Team Learning Using System Dynamics. Chichester: John Wiley & Sons.

Vennix, J.A.M., Richardson, G.P, Anderson, D.F. 1997. Group Model Building. Special issue of System Dynamics Review. Vol. 13 No. 2.

Ventana Corporation. 1994. GroupSystems for Windows reference manual. Tucson, Arizona.

Verburgh, L.D. 1994. Participative Policy Modelling Applied to the Health Care Insurance Industry. University of Nijmegen.

ISDC '97 CD Sponsor