Systems Concepts in Action

A Practitioner's Toolkit

Bob Williams, Richard Hummelbrunner

• Publisher: Stanford Business Books
• Year: 2010
• ISBN:
  9780804776554
• Edition: 1
• Topic: Business
• Subtopic: Decision Making
• Index: Business

Systems Concepts in Action: A Practitioner's Toolkit explores the application of systems ideas to investigate, evaluate, and intervene in complex and messy situations. The text serves as a field guide, with each chapter representing a method for describing and analyzing; learning about; or changing and managing a challenge or set of problems.

The book is the first to cover in detail such a wide range of methods from so many different parts of the systems field. The book's Introduction gives an overview of systems thinking, its origins, and its major subfields. In addition, the introductory text to each of the book's three parts provides background information on the selected methods. Systems Concepts in Action may serve as a workbook, offering a selection of tools that readers can use immediately. The approaches presented can also be investigated more profoundly, using the recommended readings provided. While these methods are not intended to serve as "recipes, " they do serve as a menu of options from which to choose. Readers are invited to combine these instruments in a creative manner in order to assemble a mix that is appropriate for their own strategic needs.

PART ONE

DESCRIBING AND ANALYZING SITUATIONS

There is only one thing in life worse than being talked about, and that is not being talked about.

Oscar Wilde

1

CAUSAL LOOP DIAGRAMS

What are the key variables in the situation that interests us?

How do they link to each other?

How do they affect each other? Does each variable have a reinforcing or dampening effect on the variables to which it is linked?

WHAT ARE CAUSAL LOOP DIAGRAMS?

Causal loop diagrams (CLDs) provide a language for articulating our understanding of dynamic, interconnected situations. They can be considered sentences that are constructed by linking together key variables and indicating the causal relationship between them. By connecting several CLDs, a coherent story can be told about a particular situation or issue.

CLDs visualize variables and their relationships over time. They permit us not only to analyze current states and relational patterns but also to make assumptions about the dynamic behavior. They allow us to look beyond individual events and to reach a higher—one might say more systemic—level of understanding, by mapping the structure that is responsible for producing recurring patterns of events over time.

CLDs are based on the concept of “feedback,” which was originally developed in the 1940s as part of the emerging science of cybernetics. Feedback is the transmission or return of information, and a feedback loop is a closed sequence of causes and effects: variable X is affecting Y and Y in turn affecting X. Thus, we cannot study the link between X and Y independently of the link between Y and X. Only by studying the whole feedback loop will we be able to develop a meaningful understanding of the behavior patterns at work in such a system.

Feedback loops are the building blocks of CLDs and appear in two types:

• Reinforcing or positive feedback (see Figure 1.1)) refers to a situation where all the variables respond to each other in the same direction: when A goes up, B goes up as well, which leads to a further increase of A. When A goes down, B goes down as well, which leads to a further decrease of A. Change in one direction is compounded with even more change (sometimes referred to as “cumulative causation”), which can produce both growth and decline. A savings account is an example of positive feedback, as the amount of savings and the interest earned are linked in a reinforcing manner: if savings increase, they will lead to higher interest earnings, which again are added to the savings, and so forth. Contrarily, if savings decrease, the interest earned will also go down, which again affects accumulated savings.

• Negative or balancing feedback (see Figure 1.2)) occurs when at least one variable in the system responds to change in another variable in the opposite direction: when A goes up, B goes down. If the relationship between B and A is positive (i.e., if B goes down, then A goes down too), then change in both variables is attenuated. (See Figure 1.2) in the case application.) An everyday example of negative feedback is a temperature control system. When the temperature in the room rises, the heating is lowered and the temperature begins to fall. As the temperature drops, the heating is increased and the temperature rises again. Provided the limits are close to each other, a steady room temperature is maintained.

Feedback can be used for analyzing the dynamics of a given situation, because it can be associated with specific patterns of behavior:

– Reinforcing feedback leads to exponential growth (or decline) and escalation, whereas balancing feedback results in stabilization.

– Reinforcing feedback by itself is unstable, amplifies small effects, and tends to get out of control. But it is usually counteracted by a balancing feedback of some sort (e.g., spending money, which slows down the growth of a savings account). This might result in s-shaped growth, where initial exponential growth is followed by correctional behavior.

– Delay between variables is the cause of oscillatory behavior, where a variable fluctuates around some level and makes predictions rather difficult. If delays are involved, such oscillations can appear in combination with all the other patterns mentioned above.

The behavior of a system over time is generated through the interaction of these feedback loops and delays. CLDs can also be developed, then, through observing patterns of behavior and identifying the systems structures that are known to cause the pattern. But even if this underlying structure can be mapped quite easily, the resulting behavior is usually far from simple; it is often nonlinear, counterintuitive, and hard to predict. CLDs, then, cannot predict patterns of behavior, nor are they reliable as means of explaining behavioral patterns. To this end, CLDs should be transformed into system dynamic diagrams that allow simulation (see Chapter 2).

DETAILED DESCRIPTION OF THE METHOD AND HOW IT FUNCTIONS

Identifying Relevant Elements

Since CLDs are not an end in themselves but facilitate better understanding, first of all, a situation of interest must be identified, be it an issue, a problem, or an event. This should be as specific as possible because it makes the following choices easier. Next, the time horizon that is considered appropriate should be defined; this should be at least long enough to see the dynamics that are relevant for the situation play out.

Next, the elements that are relevant to the situation of interest (e.g., the factors that explain an event, influence a problem, or are important for achieving a goal) are determined. The elements of CLDs are called variables. It must be possible to state their direction of movement (whether they increase or decrease); that is, they must be able to vary over time. But they do not have to be quantifiable or possess an existing measuring scale.

The choice of variable will depend on the situation and the core relationships to be captured. Most likely this cannot be done in a single attempt and will require an iterative approach that includes several tries, successively eliminating those variables that can be left out without significant effects on the whole. It is also useful at this stage to look at the situation from different perspectives and compare and discuss views on what is considered relevant.

This selection will inevitably require drawing a boundary, in order to stay focused and reduce complexity to a level that can be handled via CLDs. It is important to remember that CLDs are not a holistic undertaking aimed at drawing the whole picture but should only include what is critical for the situation of interest. You are trying to diagram only the problem or issue, not the entire system—a common mistake in working with CLDs. The same holds true for the level of detail to be included, which once more will be guided by the situation of interest and the time horizon.

Determining Relationships

The relations between variables are illustrated by connecting them with arrows, which indicate the direction of influence. A symbol at the end of the arrow indicates the type of causality:

– Positive causal link (marked by a plus sign [+] or “s” to mean “same”) means a positive correlation, where an increase in variable A leads to an increase in variable B, and a decrease in A leads to a decrease in B. It is important to understand that the plus sign at the arrowhead does not necessarily mean that the variables are increasing, only that they are changing in the same direction (i.e., increase or decrease together). (See Figure 1.1))

– Negative causal link (marked by a minus sign [-] or “o” to mean “opposite”) is one of negative or inverse correlation, where an increase in variable A leads to a decrease in variable B, and vice versa. (See Figure 1.2))

Where required, the quality of the relationships can be analyzed still further, for example, in terms of the degree of influence exerted (e.g., strong, medium, weak) or its temporal duration (e.g., short-, medium-, long-term). Other important information is the continuity of an influence over time. Here two factors are of particular importance for understanding dynamic behavior patterns: whether a delay or time lag is to be expected in a link (often denoted by drawing a short line across the causal link) or if effects have a nonlinear sequence (e.g., exponential curves and “threshold” or “tipping-point” effects).

Forming Feedback Loops

The linked variables are further connected with each other to form meaningful and closed loops. This is an important completeness check of the CLD: all the variables should be connected and lead to closed circles of influence. Either the variables without an obvious link to others can be left out of the diagram, or—if they are considered relevant for the situation of interest—the connecting links that are still missing must be investigated before including the variable.

What emerges is a causal network that consists of various interconnected feedback loops. This replaces linear cause-and-effect relations, which are inappropriate for complex and complicated situations because in a connected system one cause can lead to several effects (and vice versa), which in turn become the cause for other effects, and so on. Cause and effect cannot be determined in an unequivocal manner but depend on the respective position: what someone considers a cause, others might consider an effect.

The type of a feedback loop is determined by adding up the negative causal links (-) (see Figure 1.1) and Figure 1.2) in the case application later in this chapter):

– Reinforcing loop (marked with + or “R”) if the sum yields an even number (including zero)

– Balancing loop (marked with - or “B”) if the sum yields an odd number

To determine the type of loop: start with an assumption, for example, “Variable A increases,” and follow the loop around. If at the end of the loop you end up with the same result as the initial assumption, it is a reinforcing loop; if the result contradicts the initial assumption, it is a balancing one.

If the behavior over time for the situation of interest is known (at least qualitatively), you can also proceed the other way around to generate feedback loops: identify whether the existing pattern fits one of the reference behavior patterns (e.g., exponential growth, s-shape, oscillations), identify the corresponding type of feedback, and finally determine the variables and their linkages which make up the loop. But this mainly refers to individual loops. When loops are interacting and contain delays, it is very difficult to match particular patterns of behavior to a CLD.

Analysis of Feedback Loops

This can yield further insights about behavior patterns. Important perspectives are the following:

– Number of feedback loops. A high number suggests that the system is relatively independent and predominantly depends on internal factors, whereas a low number suggests that it is rather a “flow-through system” dominated by external factors.

– Length of the causal links. Lengthy feedback loops involving many elements mean that effects might take place only after a considerable time-span; short feedback loops, on the other hand, could mean fast reactions (unless the links themselves are characterized by considerable delay).

– Connection of the variables. The number of input and output relations indicates the function of the variables (e.g., whether they are more active or more passive); it is also possible to define “start variables” and “end variables,” or the most important interfaces in causal networks.

CASE APPLICATION OF CLD

CLDs were used during the evaluation of a program that provided funding for consulting services to private enterprises. Low use of the fund was evident from data previously collected, and this was identified as the core issue to be explored. To collect explanatory factors for this situation, several group interviews with stakeholders were held (e.g., beneficiary businesses, con sultants providing services, and business institutions acting as delivery points).

Then these factors were arranged and connected to form a CLD in three consecutive steps:

1. Identifying Relevant Elements

First, all of the factors were written on cards and displayed on a pin board. Then two factors were selected as immediate causes for the low use of funds: the number of applications handed in by enterprises and the support delivery by the business institutions involved in the program. The other factors were then grouped around each of these two causal factors, thus resulting in two clusters of factors, each attributed to one “strand”: the beneficiary enterprises or the business support institutions. Factors that could not be directly associated to one of these strands were put aside for the moment.

2. Determining Relationships

Next, the factors were connected according to their respective influence, using the explanations provided during the interviews as background information. The arrows indicate the direction of influence, and + and - are used as symbols to mark whether they are positive or negative causal links. For instance, the ease of the institutions' application procedures and the enterprises' information level on the fund were seen as positively influencing the number of applications. An increase in the information level and easier procedures both meant that more enterprises prepared applications. On the other hand, changes in institutions' personnel had negative effects on the delivery of support. If staff changes became more frequent, service delivery was interrupted, quality management standards were lowered, and subsequently the quality of support declined.

In addition, important time-lags were identified which explained the dynamics of certain relations. For instance, there was a considerable delay between the time when enterprises were informed about the funds and when they actually handed in applications, as this depended on a range of other factors (which were not included in the diagram): their capacity to write applications, the market situation, or the availability of their own financial resources, as the fund only provided cofinancing. These delays were shown in the relationships as a double line across the respective arrow. (SeeFigure ...