Part three systems complexity

 Part three systems complexity

Understanding the behaviour of complex systems

In this part of the book we explore system complexity in greater detail. In Chapter 8 we use the En-ROADS Climate Solutions Simulator as the backbone to introduce systems dynamics as a method for modelling and understanding complex systems. It focuses on highlighting the behavioural dynamics within a system including unintended consequences of decisions and interventions. In the penultimate section, using the knowledge from previous sections, you are guided to use the En-ROADS simulator to understand the climate change problem and develop your own solution. In Chapter 9 we explore the ever-changing nature of complex systems by exploring the levers or leverage points in complex systems that can influence and change the behaviour of the entire system. We introduce different types of levers, specifically structural levers, temporal levers, conceptual levers and boundary levers, and give examples of how they can be used in practice.

This chapter has been developed with help and assistance from Dr. Merve Er, an assistant professor at the Marmara University in Istanbul. In the previous chapters, we have discussed different methods used in modeling soft systems and the complexity associated with them. In particular, we introduced causal loop diagrams in Chapter 6 and causal mapping in Chapter 7 as structured approaches for exploring causalities between different elements of the system interacting with each other. In this chapter, we will take these methods one step further and demonstrate how you can build on qualitative understanding of such systems and construct a quantitative model based on the descriptive qualitative conceptualization of the system, which can then be simulated to explore the system’s behaviour in the future.

The purpose of this chapter is to demonstrate how systems can be modeled and how this modeling to study their future behaviour can help managers learn about the systems as the new systems emerge. It is important to note that when we create these models we make assumptions about how various autonomous parts (organizations/people) would behave in a given circumstance. Over time, as things unfold and we keep on comparing the model to how things are unfolding in reality, we can learn from this experience and refine the model. Therefore, the core value of modeling is not about developing a high-fidelity model. Instead, it is the PROCESS of MODELING that enables the modellers to learn more about the system as it emerges, thus making it more predictable.

The objective of this chapter is to introduce systems dynamics as a method for modeling and understanding complex systems. It will focus on highlighting the behavioral dynamics within a system, including unintended consequences of decisions and interventions. In the penultimate section, using the knowledge from previous sections, the reader will be guided to use the En-ROADS.

simulator, developed by the MIT Sloan School of Management, to understand the climate change problem and an example of a complex and wicked problem and develop their own solution to solving the climate challenge.

LEARNING OUTCOMES

Understand

.the value of modelling future behaviour of systems

purpose of system dynamics

stocks and flows

approaches to and challenges of quantifying relationships in a stock and flow model

Basics of analysing stock and flow model simulation

Learn how to study the system through the process of modelling its future behaviour

Learn about behavioural dynamics of complex systems

8.1 Systems dynamics: an approach to modelling and simulating complex systems

In Chapter 6 we talked about causal loop diagrams as one of the methods used in Soft Systems Methodology. They can be very useful when we try to capture soft aspects of a system. They show us the forces in the system and how they impact various aspects of the system. Although they can give us an idea of how the whole system will behave, they predominantly capture the system’s present behaviour. Although by looking at the interactions between different forces we can deduce what the system’s behaviour may look like in the future, it is much more difficult to predict how the system’s behaviour will change over time as the forces upon the system change.

This is where system dynamics proves useful as the next development in causal loop diagrams. The method was developed by Jay W. Forrester (Forrester, 1958). He proposed using causal loop diagrams as a foundation where you can then apply mathematical/statistical equations to model relationships between different elements of the systems. Then the model can be used to simulate how the system will behave over time in the future. In such simulations, the behaviour that the system exhibits over time is called a dynamic. Systems tend to be dynamic, which means that the behaviour of the system will change over time. If you understand the relationship between different elements in the system, and how they impact each other, you can model the behaviour of the system and then simulate it and see what it will run into in the next 1, 5, 10, 20 years. Such models can also help 'play out' different scenarios, by doing what-if analysis where variables are changed. By changing one or more variables we can further learn how the behaviour of the whole system may change in the future.

The purpose of this chapter is to introduce the foundations of the method and help you understand how to interpret and use system dynamics models. If you would like to learn more about the method and build models of your own, we recommend that you read specialized textbooks dedicated to this method, for instance starting with Donella Meadows, Thinking in Systems (2008).

8.2 Building on soft systems thinking

When trying to build a system dynamics model of a system, the first step is usually conducting a review of what is already known about the system and about the internal and external forces that may impact it. The review will reveal current trends that might lead to changes in the system, variables that might determine these changes, and relationships between variables. With this understanding, you can then begin to represent a system with a causal loop diagram as discussed in Chapter 6. The diagram will give a static representation of the system, in which the future changes cannot be observed, but rather speculated about based on the links between different forces. However, it can help capture all the forces affecting the system.

When a causal loop diagram is built, you will normally need to reduce it to the most significant trends and forces that should be captured and quantified in a system dynamics model. This will require redefining the boundaries of the system as the first step to decide what is indeed important in the simulated system and what should be left outside the boundaries. Although it might be tempting to try and capture all the forces identified, it might lead to an overly complicated model with a lot of noise that might not necessarily be more accurate than a less-complicated version. Quantified relationships are "often at least partly based on assumptions. Each assumption would have errors, and thus incorporating too many assumptions in the model might lead to high accumulated errors in the simulation results, particularly over the longer simulation horizon.

Depending on your area and breadth of expertise, it might be necessary to involve a group of people or external experts in focusing on what are the most essential elements of a causal loop diagram that need to be considered in the simulation model. This might be facilitated with focus groups or using more structured approaches, such as the Delphi method.

Having developed the causal loop diagram and simplified it to focus on the pertinent forces, you can build the simulation model for further analysis of the future of the system in three steps: modelling stocks and flows, quantifying the relationships, and running the simulation and analysing the results. Each of these steps is described below."

Modelling stocks and flows

In Chapter 3 we introduced the concepts of stocks and flows. They are two of the foundation elements of system dynamics. A stock is a variable that is measured at a specific period of time, for example on a specific day. It represents a quantity or a value that exists at that particular point in time. The stock might accumulate or get depleted over time. But at each given period in time, we can measure it. A flow is measured over an interval of time, for example per day, week, or month. The flow changes the amount of stock, i.e., the inflow leads to the accumulation of a stock (is added to the stock), while the outflow leads to its depletion (is subtracted from the stock). The flow rate is related to the flow over time and shows the speed with which the flow flows into the stock or flows out of the stock. The flow that originates from outside the boundaries of the system will originate from a cloud on a system dynamics diagram, and conversely a flow that leaves the system’s boundaries will terminate at a cloud on a systems dynamic diagram. It is assumed that clouds have an infinite capacity that does not constrain the system. And finally, a link is used to indicate dependencies between different variables that determine the rates of the flow. Figure 8.1 illustrates the commonly used symbols and a simple stock and flow diagram for population change.

Let us look at the simple stock and flow diagram in Figure 8.1. In the centre of the diagram you have a population, which is a stock represented as a rectangle. To define the starting point, you need to put an initial value for the population, that is what the population was like at a particular point in time from which the model will start simulating the change in the population. The population is increased by an inflow, which is defined by the birth rate, and reduced by an outflow, which is defined by the death rate. Both are positive values.

The inflow represents a reinforcing loop – the more population is born, the more of the population will give birth in the future, and therefore the more the population will increase in a country. The death rate represents a balancing loop – the more population there is, the more of the population will die, and therefore with this reduction the population will be balanced out. In this simple example we define quite narrow boundaries in the form of clouds, as they do not examine what external factors might define the birth rate and the death rate in this particular system.

In this example, the only two variables that change the behaviour of the system are the birth and death rates, which are assumed as constant. Depending on how they compare to each other, the population will grow, decline, or remain stable. If the birth rate is higher than the death rate, then the population will grow, and if its growth is simulated and plotted, the relationship will be non-linear. If the birth rate is lower than the death rate, the population will decline, and again, the simulation will reveal a non-linear relationship. If the birth and death rates are the same, the population balances out and remains stable.

Using this model as a basis, we can build a more complex system of population change by expanding the boundaries and adding more variables to create a more nuanced representation of relationships. In Figure 8.2, instead of looking at the population as a whole, we can examine stocks of population by different groups. New-born people flow into the stock of young people whose 

volume might reduce as defined by the young mortality rate, or an average rate of how many people will die at a young age. The remaining young will mature to adulthood and inflow into the stock of reproductive adults. Some of them will also get ill or die in accidents at a defined adult death rate. The rest are likely to have children, defined by an adult fertility rate and contributing to the stock of young people. They will also age and move to the stock of non-reproducing adults, which will decrease at a death rate.

If you know these parameters, you can predict more accurately how the population will change over time. For example, if you want to increase the population of a country, but there is a worrying trend observed in this country, you can see where the levers or pressure points discussed in the chapter can be introduced. If you want to further analyse at-risk groups of populations, such as adults that define the reproduction rates in the population, you can further break it down into female and male adults. Such a breakdown might help you further understand the impact of various trends on how the society will reproduce and what the population will be in, let’s say, 20 years’ time. The two examples above show that you can start with a very simple model that describes the behaviour of a system. And then you can add layers of complexity by introducing new variables and expanding the boundaries by considering additional factors that help explain the variables, such as social policies or practices that might influence birth rates. Once you start unravelling all these parameters, the system can become really complex.

Quantifying relationships

Although there are general principles for building system dynamics models (Bala et al., 2017; Forrester, 1994), when it comes to quantifying relationships,

There are no strict rules to follow. Quantification of relationships usually relies on mathematical/statistical equations that define the relationships between different parameters in the model. The equations may be developed using data from a number of sources:

prior studies that have investigated and validated the relationships between these parameters

studies set up specifically to collect data on the relationships

focus groups or Delphi studies with experts in the specific area to help develop assumptions and define the relationships

The first challenge in quantifying the relationships

This presents the first challenge in quantifying the relationships. Complex systems include multiple variables and feedback loops. The number of variables that might need to be considered in a complex system might quickly overwhelm the data available and pose a challenge to choosing from alternative explanations and theories to interpret the behaviour of the system. The assumptions that aim at filling in the gaps in the model and interpretation of its behaviour can vary in degree of accuracy, adding multiple errors in the model. The longer the simulation horizon of the model, the greater the cumulative errors from these assumptions will be. This is particularly true about the assumptions and inferences about the consequences of external events that have not happened before, and thus we do not have reliable evidence of anticipated consequences. Making such assumptions is inevitably distorted with biases due to various cognitive prejudices, such as worldviews, that we all are prone to.

Furthermore, multiple feedback loops might cause many variables to relate with each other, but human ability to unpack these relationships is quite limited, as people are poor judges of correlations due to cognitive biases. Engaging with multiple experts to cross-examine the assumptions will help to reduce the errors in the model.

The second challenge is related to the data and information being fed into the model. For example, if we build a model that simulates the impact of climate change on a system, the data on climate change will come from existing predictions that are publicly available. Quite often the developers of such models have to rely on what is available. However, what is available is not always accurate. Even if data we have is perceived as reliable, we should always be mindful that it is based on delayed measurement, approximations and averages, which in themselves introduce distortions and biases. Other times the data might not be available at all. Similar to quantifying the