After lunch, you all decide to sketch out a causal loop diagram of the situation. You’ve heard that as one of the basic tools of systems thinking, CLDs can open up new insights into a problem; you’ve also heard that they work best when you state the variable names in certain ways.

Getting Started

But what makes good variable names, and how do you identify them from a gripe session like the one described above? Here are a few guidelines to get you started (also see “Will the Real Variable Name Please Stand Up,” V9N1):

• Use nouns instead of verbs, action terms, or words suggesting a direction of change.

Example: Number of Products Sold NOT Sell Revenue NOT Increasing Profits By using nouns, you let the CLD arrows show the action.

• Use a neutral or positive term whenever possible.

Example: Morale NOT Bad Feelings Job Satisfaction NOT Job Dissatisfaction Such terms help you avoid confusing double-negatives that can happen when reading a CLD. (For example, “job satisfaction declines” is much easier to grasp conceptually than “job dissatisfaction declines.”)

• Identify hard-to-measure variables, such as “Experience Level” or “Trust,” as well as more concrete variables like “Orders” or “New Hires.” Those intangible variables are often just as important—if not more so—than the quantifiable ones.

Once you’ve properly named the key variables in your problem situation, you can start to identify the causal relationships between them and link them in one or more balancing or reinforcing loops.

So, let’s return to your gripe session with Thomas and Marina. As the three of you prepare to identify key variables from the problem situation, you agree to act as the moderator for this part of the process. Your job is to ask questions that will help Thomas and Marina find potential variable names in your earlier conversation. But don’t stop with just the spoken words—sometimes you can detect perfectly good variable names in unspoken signals like facial expressions or tone of voice.

To begin, let’s work our way through the conversation. All the while, we’ll examine the various comments—and any important nonverbal happenings—for promising variable names. Look again at the first statement that was made: “I don’t get it. Why do people keep leaving?” Your question to Thomas and Marina might be: “What specific, neutral noun phrase could convey the meaning in this statement?” Possible answers could include “Customer Service Turnover” or “Number of Customer Service Resignations.” Congratulations—you’ve nailed down your first variable name!

Now let’s look at the second statement from the conversation, made by Marina: “This is getting bad—I don’t see how the rest of us can deal.” If you were to ask Marina, “What do you mean?” she might respond with, “Well, every time someone in customer service leaves, there are fewer of us left to process the incoming phone orders—our workload just gets worse.” Voilá! You can now formulate another variable name: “Workload.”

TURNOVER IN CUSTOMER SERVICE

As customer service representatives resign, the workload increases for the remaining staff. Morale falls, as the reps scramble to keep up with the incoming calls. Eventually, even more people resign, leading to yet bigger workloads and worse morale.

Let’s next take Thomas’s statement: “It’s worse than that—nobody can ask you folks for anything these days without getting their heads bitten off!” “What’s happening here?” you ask Thomas. He replies, “They’re so irritable!” You nudge him to think of a noun, and he comes up with another variable name: “Morale.” As you reflect on the tension that you sensed in Marina, and on Thomas’s observations about short tempers, these nonverbal clues confirm that morale is indeed a key variable.

From Variables to CLDs

Once you’ve properly named the key variables in your problem situation, you can start to identify the causal relationships between them and link them in one or more balancing or reinforcing loops (see “Causal Loop Construction: The Basics,” V7N3). When you have completed the diagram, walk through the loops and “tell the story” to be sure they accurately capture the behavior being described (see “Turnover in Customer Service” on p. 9).

In real life, identifying variable names involves many more iterations than we’ve shown in this case. However, it’s this very act of refining the names that lets you construct a powerful CLD—one that will yield rich insights into the problem and maybe even point the way to a solution. In this case, Marina might share the loop with others in her department. The group could brainstorm possible interventions, such as accelerating the hiring of replacement workers or temporarily reducing the workload by postponing nonessential tasks. Of course, they will want to map out the potential unintended consequences of each of these proposals as well, to make sure the solution doesn’t make the problem worse!

Lauren Keller Johnson is a freelance writer living in Lincoln, Massachusetts.

The post In Search of the Perfect Causal Loop Variable appeared first on The Systems Thinker.

Perceptions and reality often differ, and it is usually important to capture these differences in your causal diagrams. The true state of affairs can be very different from the perception of that state by the actors in the system.

WHAT IS THE QUALITY OF OUR PRODUCTS?

Distinguishing between actual and perceived conditions adds important information to a model for the management of product quality.

People’s perceptions can be influenced by reporting delays, measurement error, bias, or other distortions, causing people to make different decisions than they would if they had more accurate information.

For example, do the top managers of a firm know the true quality of their products? How do they respond if there is a quality shortfall? In drawing a CLD depicting this situation, separating perceived and actual conditions helps prompt questions such as, How long does it take to measure quality? To change management’s opinion about quality even after data are available? To implement a quality improvement program? To realize results? You might discover that there are significant delays in assessing product quality and in changing management’s opinion about quality (see “What Is the Quality of Our Products?”).

In addition, bias in the reporting system may cause reported quality to differ from quality as experienced by the customer. Customers don’t file warranty claims for all problems or report all defects to their sales representatives. Sales and repair personnel may not report all customer complaints to the home office. Sub ordinates, wishing to avoid delivering bad news, may filter the quality assessment in formation that reaches senior management. The diagram shows these biases, helping to explain how management might come to hold a grossly exaggerated view of product quality. Such a model can serve as the basis for conversation about ways to shorten the delays, overcome the biases, and avoid quality erosion.

Don’t Forget Delays

Delays are critical in creating dynamics. Delays give systems inertia, can cause oscillations, and are often responsible for trade-offs between the short- and long-run effects of our policies. Your CLDs should include delays that are important to the dynamics you are trying to represent or are significant relative to the time horizon relevant to your issue.

For example, when the price of a product rises, supply will tend to increase, but only after significant delays while new capacity is ordered and built and while new businesses enter the market. It’s therefore important to include these delays (by writing the word “delay” or drawing hash marks over the relevant link) because they will affect the system’s behavior over time. Remember that there are both physical, or material, delays, such as the delay between ordering and receiving materials, and information, or perceptual, delays, such as the time required to report sales data, revise forecasts, and make decisions. Both types of delays should be represented in your causal maps.

It’s also useful to remember that delays always involve stock and flow structures. This is why it is doubly important to include delays in your diagrams—they will remind you that there is a stock and flow structure embedded in that particular relationship. You can then explore whether it is useful to make that structure explicit in your diagram. Make stock and flow structures explicit if doing so is important in communicating the basis for the dynamics your map seeks to explain.

Don’t Put All the Loops into One Large Diagram

Our short-term memory can hold from five to nine chunks of information at once. This limits the effective size and complexity of a causal map. Presenting a complex CLD all at once makes it hard to see the loops, understand which are important, or determine how they generate certain behaviors. Resist the temptation to put all the loops you or your client have identified into a single comprehensive diagram.

How then do you communicate the rich feedback structure of a system without oversimplifying? Build up your model in stages, with a series of smaller causal loop diagrams. Each diagram should correspond to one part of the story being told. These diagrams can have enough detail to show how the process actually operates. Then combine simpler versions of the diagrams into a high-level overview to show how they interact with one another. (For more details about ensuring the clarity of your diagrams, see “Tips for CLD Layout.”)

Check to make sure your audience is following the logic of the causal links. If they are not able to follow the thinking without assistance, you may need to include more detail or make some of the intermediate variables more explicit. “Making Links Explicit” shows an example.

MAKING LINKS EXPLICIT

Make intermediate links explicit to clarify a causal relationship.

Once you’ve clarified the logic to the satisfaction of all, you often can “chunk” the more detailed representation into a simpler, more aggregate form. The simpler diagram then serves as shorthand for the richer, underlying causal structure.

Name Your Loops

Whether you use CLDs to elicit clients’ mental models or to communicate the important feedback processes that you believe are responsible for a problem, you will often find yourself trying to keep track of more loops than you can handle. Your diagrams can easily overwhelm the people you are trying to reach. To help your audience navigate the network of loops, give each a number and a name. Numbering the loops R1, R2, B1, B2, and so on helps your audience find each loop as you discuss it. Naming the loops helps your audience understand the function of each loop and provides useful shorthand for discussion.

When working with a client group, it’s often possible to get them to name their own loops. Many times, they will suggest a whimsical phrase or some organization-specific jargon for each loop. For instance, if you have a loop that depicts how working too much overtime might eventually undermine productivity, the group might label it “Burnout.” They might call a loop that shows how schedule pressure can lead to increased errors “Haste Makes Waste.” Loop names make it easy to refer to complex chunks of feedback structure, leading to productive conversations and, ultimately, changes in deeply ingrained behavior.

John D. Sterman is the J. Spencer Standish Professor of Management at the Sloan School of Management of the Massachusetts Institute of Technology and director of MIT’s System Dynamics Group.

This article is part of a 2-part series. Click here to view the first part.

The post Fine-Tuning Your Causal Loop Diagrams—Part II appeared first on The Systems Thinker.

• Quickly capturing your hypotheses about the causes of dynamics;
• Eliciting and capturing the mental models of individuals and teams;
• Communicating the important feedback processes you believe are responsible for a problem.

The conventions for drawing CLDs are simple but should be followed faithfully. Think of CLDs as musical scores: At first, you may find it difficult to construct and interpret these diagrams, but with practice, you will soon be sight-reading. In this article, I present some important guidelines that can help you make sure your CLDs are accurate and effective in capturing and communicating the feedback structure of complex systems.

Avoid Ambiguity in Labeling Causal Links

AMBIGUITY OF LINKS

To be effective, your CLD should not include any ambiguous causal links. Ambiguous polarities usually mean there are multiple causal pathways that you should show separately.

People sometimes argue that a specific link in a CLD can be either positive or negative, depending on other parameters or on where the system is operating. For example, we might draw a diagram that relates a firm’s revenue to the price of its product and then argue that the link between price and company revenue can be either positive or negative, depending on the elasticity of demand (see “Ambiguity of Links”). A higher price means less revenue if a 1 percent increase in price causes demand to fall more than 1 percent. This link would be labeled with a negative sign. But less elastic demand might mean a 1 percent increase in price causes demand to fall less than 1 percent, so revenues would then rise, resulting in a positive link polarity.

When you have trouble assigning a clear and unambiguous sign to a link, it usually means there is more than one causal pathway connecting the two variables. You should make these different pathways explicit in your diagram. The correct diagram for the impact of price on revenue would show that price has at least two effects on revenue: (1) it determines how much revenue is generated per unit sold (a positive link), and (2) it affects the number of units sold (usually a negative link).

Is It Reinforcing or Balancing?

There are two methods for determining whether a loop is reinforcing or balancing: the fast way and the right way. The fast way, which you may have learned when you first started working with CLDs, is to count the number of negative links—represented by “-” or “o”—in the loop (see “‘+’ and ‘-’ Vs. ‘s’ and ‘o’”). If the number is even, the loop is reinforcing; if the number is odd, the loop is balancing. However, this method can sometimes fail, because it is all too easy to mislabel a link’s polarity or miscount the number of negative links.

The right way is to trace the effect of a small change in one of the variables around the loop. Pick any variable in the loop. Now imagine that it has changed (increased or decreased), and trace the effect of this change around the loop. If the change feeds back to reinforce the original change, it is a reinforcing loop. If it opposes the original change, it is a balancing loop. This method works no matter how many variables are in a loop and no matter where you start.

Make the Goals of Balancing Loops Explicit

All balancing loops have goals, which are the system’s desired state. Balancing loops function by comparing the actual state to the goal, then initiating a corrective action in response to the discrepancy between the two. It is often helpful to make the goals of your balancing loops explicit, usually by adding a new variable, such as “desired product quality” (see Desired Product Quality in “Explicit Goals”). The diagram shows a balancing loop that affects the quality of a company’s product: The lower the quality, the more quality improvement programs the company initiates, which, if successful, correct the quality shortfall.

EXPLICIT GOALS

Making goals explicit in balancing loops encourages people to ask questions about how the goals are formed. For example, what drives a company’s desired level of quality?

Making goals explicit encourages people to ask how the goals are formed; for instance, who determines desired product quality and what criteria do they use to make that determination? Hypotheses about the answers to these questions can then be incorporated in the diagram. Goals can vary over time and respond to pressures in the environment, such as customer input or the quality of competing products.

Making the goals of balancing loops explicit is especially important when the loops capture human behavior—showing the goals prompts reflection and conversation about the aspirations and motives of the actors. But often it is important to represent goals explicitly even when the loop doesn’t involve people at all.

Represent Causation Rather Than Correlation

Every link in your diagram must represent what you and your colleagues believe to be causal relationships between the variables. In a causal relationship, one variable has a direct effect on another; for instance, a change in the birth rate alters the total population. You must be careful not to include correlations between variables in your diagrams. Correlations between variables reflect a system’s past behavior, not its underlying structure. If circumstances change, if previously dormant feedback loops become dominant, or if you experiment with new decisions and policies, previously reliable correlations among variables may break down.

ICE-CREAM SALES AND MURDERS

Causal loop diagrams must include only what you believe to be genuine causal relationships, never correlations, no matter how strong.

For example, though sales of ice cream are positively correlated with the murder rate, you may not include a link from ice-cream sales to murder in your CLD. Such a causal link suggests that cutting ice-cream consumption would slash the murder rate and allow society to cut the budget for police and prisons. Obviously, this is not the case: Both ice-cream consumption and violent crime tend to rise in hot weather. But the example illustrates how confusing correlations with causality can lead to terrible misjudgments and policy errors (see “Ice Cream Sales and Murders”).

While few people are likely to attribute murders to the occasional double-dip cone, many correlations are more subtle, and it is often difficult to determine the underlying causal structure. A great deal of scientific research seeks the causal needles in a huge haystack of correlations: Can eating oat bran reduce cholesterol, and if it does, will your risk of a heart attack drop? Does economic growth lead to lower birth rates, or is the lower rate attributable to literacy, education for women, and increasing costs of child-rearing?

Do companies with serious quality improvement programs earn superior returns for stockholders?

Scientists have learned from experience that reliable answers to such questions are hard to come by and require dedication to the scientific method—controlled experiments; randomized, double-blind trials; large samples; long-term followup studies; replication; statistical inference; and so on. In social and human systems, such experiments are difficult, rare, and often impossible. You must take extra care to determine that the relationships in your CLDs are causal, no matter how strong a correlation may be.

John D. Sterman is the J. Spencer Standish Professor of Management at the Sloan School of Management of the Massachusetts Institute of Technology and director of MIT’s System Dynamics Group.

This article is part of a 2-part series. Click here to view the second part.

The post Fine-Tuning Your Causal Loop Diagrams—Part I appeared first on The Systems Thinker.