Skip to main content

Simulation Modeling to Improve Decision Making in Complex-Dynamic Environments

Final Report Summary - SIMID (Simulation Modeling to Improve Decision Making in Complex-Dynamic Environments)

Summary Description of the Project Objectives

The overall aim of this project is to improve decision making in the presence of dynamic complexity, which is a challenging task. The main reason behind this challenge is the inadequacy of our intuitive skills in coping with complex dynamic decision-making situations. Dynamic complexity often overwhelms human decision makers, leading to poor understanding, which further leads to poor decisions. For the same reasons, effective learning does not typically take place in complex decision-making environments.

We have three sub-objectives:

1. Analyzing the effects of the main elements of dynamic complexity on system performance. (These main elements are the dynamics of stock-flow accumulations, non-linear relations and time delays between variables, and feedback loops).

2. Improving understanding, learning, and decision making of human subjects by training them with controlled manipulations of the elements of dynamic complexity in simulation games.

3. Developing and testing decision-making heuristics (formulations) that will improve management of complex-dynamic systems by simulation modeling.

There are three phases of the project, which correspond to the three objectives listed above.

Phase 1:
Develop simulation models involving the elements of dynamic complexity and analyze the effects of these elements by using simulation experiments. Analytical approaches will also be utilized if necessary.

Phase 2:
Develop simulation games based on the findings of Phase 1. Test the performances of the participants by developing and using simulations of increasing complexity. Simple simulations are expected to help the decision makers to develop an understanding of dynamic complexity, which should in turn improve their performances in more complex simulations.

Phase 3:
Develop models for selected typical dynamic challenges and suggest decision-making heuristics to improve the performances of systems involving those challenges. The improvements in the performances will be demonstrated by simulation experiments. The selected problems will be of generic nature. Therefore, the results will be applicable to many different problems with different contexts, but with similar underlying structure.

Description of the Work Performed since the Beginning of the Project

The broad nature of the project resulted in the diversity of the work completed during the first period (01/12/2010 and 30/11/2012) and the second period (01/12/2012 and 30/11/2014). The work completed during the first period was mainly related to the phases 1 and 3 of the project. In the second period, further studies related to the phases 1 and 3 of the project are carried out. Two studies related to phase 2 are carried out, which are based on the work done in phase 1. Note that, the three phases of this project are closely related to each other and the work carried out in a phase potentially contributes to other phases as well. Phase 2 depends mainly on the results of Phase 1. However, Phase 3 of the project does not follow the first two; it can be carried out simultaneously with the first two phases.
Description of the Main Results Achieved so Far

There are five main achievements:

1. A rule of thumb was suggested for parameter value selection for the well known anchor-and-adjust heuristic used in stock management for a wide range of conditions.

2. A soft landing model was developed and different heuristics were compared in detail.

3. A basic inventory control model and a game based on this model were developed. The design of the experiments is completed and soon the experiments will start.

4. A complex workforce and service backlog management model and a game based on this model were developed. The design of the experiments is completed and the experiments have recently been started.

5. A detailed mathematical model of a famous decision making task known as “The Beer Game” is developed, it is coded in R and simulation experiments are carried out based on this model.

Final Results and their Potential Impact and Use

Improving dynamic decision making is of great importance for very diverse types and levels of managing complex dynamic systems: it is important for individuals trying to control their own weight; for doctors and nurses managing the health of their patients with chronic illnesses; for managers at all levels in both public and private industries; for rectors, deans, and heads of schools managing their universities and schools; for ministers and bureaucrats managing various public institutions in EU, and so forth. In order to improve our ways of dealing with the increasingly complex issues of the modern world, we first need to understand the dynamic complexities of the world we live in. Researchers increasingly argue that systems perspective and understanding of dynamic complexity will play a crucial role in the future of humanity because without such a perspective, the unintended consequences of our decisions about important complex problems will render our efforts ineffective, even harmful. Unless we understand the complex counterintuitive behaviors of dynamic systems that we face in modern world, the ‘solutions’ that we implement today are likely to be our ‘problems’ of tomorrow.

This research contribute to the improvement of dynamic decision making in three different ways:

1. This research serves as a step towards having a better understanding of dynamic complexity.

2. Simulation experiments carried out using artificial decision making agents demonstrate counter-intuitive results. Soon-to-be completed experiments using human decision makers will demonstrate further interesting results as suggested by the pilot studies we have conducted.

3. Some generic dynamic decision making heuristics (i.e. control heuristics) are proposed, which can be used in diverse simulation problems with different contexts but with similar underlying structures.

With additional work, these three classes of research results can all be generalized to different types of real life dynamic decision problems illustrated above.

Project Website Address