Causal Loop Diagrams: Drawing How Things Connect

So far in this book we have talked a lot about feedback — the idea that the output of a system loops back to become an input, so a system can shape its own future. We have used words like "reinforcing" and "balancing." But words alone get slippery fast. When three or four things all push on each other, a sentence stops being enough to hold the whole picture in your head.

This chapter gives you the single most useful drawing tool in all of systems thinking: the Causal Loop Diagram, or CLD. It is a simple picture — boxes of words joined by labelled arrows — that lets you see the feedback structure of a situation before you act. Once you can draw one, the invisible machinery behind everyday problems (a stress spiral, a growing startup, an arms race) becomes plainly visible.

CLDs grew out of real research. The idea of distinguishing "deviation-amplifying" loops from equilibrium-seeking ones came from Magoroh Maruyama in a 1963 paper. Jay Forrester's MIT work on system dynamics (Industrial Dynamics, 1961) gave them a home discipline. Peter Senge made them famous for managers in The Fifth Discipline (1990), and Donella Meadows made them approachable for everyone in Thinking in Systems (2008).

The three building blocks (that's all there is)

A CLD is made of exactly three kinds of things. There is nothing else to learn about its grammar.

1. Variables

A variable is a quantity that can go up or down over time, written as a noun or noun phrase — Population, Stress Level, Customer Satisfaction. It names what is being measured, never the action of changing it.

2. Causal arrows

A causal arrow is an arrow drawn from a cause to an effect. It means: "if I changed this first variable, and held everything else still, the second variable would change."

3. Polarity signs

Every arrow gets a polarity: a + if the two variables move the same way, or a − if they move in opposite ways.

How to name a variable (the rule beginners always break)

The most common beginner error is writing actions instead of variables. "Hire More Staff," "Reduce Costs," and "Improve Quality" all describe things you do — they are not quantities that drift up and down on their own. You can't ask whether "Hire More Staff" went up or down last month, but you can ask whether Staff Headcount did.

The naming rule, drawn from Daniel H. Kim's 1992 guidelines and writers in The Systems Thinker, gives you two quick tests:

• Does the name fit naturally after "level of…" or "amount of…"? ("level of Customer Satisfaction" works; "level of Improve Quality" does not.)
• Can it go both up and down? A real variable can do both.

Also avoid baking a direction into the name. "Increasing Profits" or "High Staff Turnover" trap you, because then a decrease in "Increasing Profits" is a brain-twister. Strip the direction out: use Profit Level and Staff Turnover Rate. And prefer the positive framing — Job Satisfaction rather than Job Dissatisfaction — so you don't drown in double negatives when reading a loop.

Reading polarity: + and −

A + link (some older texts write it as S, for "same") means: cause goes up → effect goes up; cause goes down → effect goes down. They march together.

A − link (older texts: O, for "opposite") means: cause goes up → effect goes down, and cause goes down → effect goes up. They move against each other.

Crucially, polarity describes the direction of change, not whether the relationship is "good" or "bad." More exercise improving your health is a + link; more exercise reducing your free time is a − link — neither word means "positive" in the moral sense.

Two notations exist (+/− and S/O), but George Richardson's 1986 paper Problems with Causal-Loop Diagrams showed practitioners often muddle S and O, and recommended sticking to +/−. John Sterman uses +/− throughout his authoritative Business Dynamics (2000). We'll use +/− for the rest of this chapter.

The two kinds of loops

The whole point of a CLD is to find closed loops — chains of arrows that eventually circle back to where they started. A change in one variable travels around and comes home, either to amplify itself or to oppose itself. There are exactly two outcomes.

The counting method: trace any closed loop and count the minus signs.

• Even number of minuses (including zero) → Reinforcing loop (label it R). A nudge comes back amplified. This produces exponential growth or exponential decay.
• Odd number of minuses → Balancing loop (label it B). A nudge comes back as a push in the opposite direction. This seeks a goal, plateaus, or oscillates.

Worked example: the Word-of-Mouth Engine (a reinforcing loop)

Let's build one arrow by arrow — the way you always should, like following a conversation: start with one thing, then keep asking "what does that cause?"

1. Start with Number of Users.
2. More users → more Word-of-Mouth Conversations. Arrow with +.
3. More conversations → more Awareness Among Non-Users. Arrow with +.
4. More awareness → more people try the product → back to Number of Users. Arrow with +.
5. Count the minuses: 0. Even → Reinforcing. Call it R1: Word-of-Mouth Engine.

  +---------(R1: Word-of-Mouth Engine)----------+
  |                                             |
  |   +                  +                +     |
  +-> Number of  --> Word-of-Mouth --> Awareness +
     Users          Conversations      (Non-Users)

A tiny user base creates a few conversations, a little awareness, slightly more users — and around it spins, growing exponentially until something limits it.

Worked example: the Sleepless Spiral (two minuses still reinforce)

This one shows why the counting rule is so valuable — it reveals a result your gut might miss.

1. Start with Stress Level.
2. More stress → worse Sleep Quality. They move in opposite directions, so −.
3. Worse sleep → more stress (poor recovery, higher cortisol). Again opposite directions, so −.
4. Count minuses: 2. Even → Reinforcing, a vicious cycle. Call it R1: Sleepless Spiral.

   Stress Level  <-------(−)--------+
        |                          |
       (−)   ||  <- delay          |
        v                          |
   Sleep Quality -----------------+
            (R1: Sleepless Spiral)

Two negatives make a positive — the loop amplifies the original stress. Notice the || marks on one arrow: chronic stress takes days to wreck sleep, so that link carries a delay.

Delays: the mark that predicts oscillation

When an effect lags well behind its cause, draw two short parallel lines (||) across the arrow. Sterman calls delays "critical in creating dynamics." They matter because people act on old information.

When loops compete: Meadows' population model

Real systems usually hold more than one loop, and they fight for control. Meadows' classic example puts Population in the centre with two loops attached:

            +-----(R1: Births)-----+
            |  +              +     |
            v                       |
       Birth Rate            Death Rate
            |                       ^
            +-> Population <---------+
                  ^   |  +
                  +---+
            (B1: Deaths, the loop above carries one − link)

• R1 (Births): Population → Birth Rate (+), Birth Rate → Population (+). Zero minuses → Reinforcing.
• B1 (Deaths): Population → Death Rate (+), Death Rate → Population (−). One minus → Balancing.

When births per person beat deaths per person, R1 wins and population grows exponentially. When deaths win (famine, disease), B1 dominates and population collapses. The identical structure governs business capital: investment feeds a reinforcing stock, depreciation drains it through a balancing one.

This R-plus-B pattern is so common that Senge named it the Limits to Growth archetype. The TQM training example from The Systems Thinker is the same shape: a reinforcing "Capability Building" loop spreads quality activity, while a balancing "Resistance" loop — managers feeling threatened — eventually slows it to a plateau.

Naming every loop

Always give each loop a short, vivid name on top of its R1/B1 label: "Word-of-Mouth Engine," "Burnout," "Haste Makes Waste," "Escalation." Senge's arms-race Escalation archetype is two interlocked reinforcing loops — each nation arming in response to the other — that ratchet upward together. The name turns a tangle of arrows into a story you can talk about, and it points straight at the leverage: if Nation A disarms first, perceived threat falls and the whole loop runs in reverse.

What a CLD can't do

A CLD is qualitative. It shows the direction and shape of relationships — never how fast or how much. Richardson's 1986 paper is blunt: you cannot reliably tell oscillation from growth using a CLD alone, because that depends on which variables are stocks (accumulations) and which are flows (rates). For real numbers you graduate to a stock-and-flow diagram and full simulation — the subject of the next chapter. The CLD's simplicity is both its gift (anyone can draw one in minutes) and its limit (it stays silent on timing).

Key Takeaways

• A CLD has just three parts: variables (nouns that go up and down), causal arrows, and a + or − polarity on each arrow.
• Name variables as neutral nouns ("Profit Level"), never as actions ("Reduce Costs") or directions ("Increasing Profits"). The test: can it follow "level of…" and move both ways?
• Count the minus signs around a closed loop: even (including zero) = Reinforcing (R, amplifies); odd = Balancing (B, opposes). Always double-check by tracing a nudge in your head.
• Two minuses still reinforce — that counter-intuitive result is exactly what the counting rule exists to reveal (the Sleepless Spiral).
• Mark significant time lags with || ; delays cause oscillation because people act on old information, as in Meadows' car-dealer example.
• CLDs show direction and feedback structure, not speed or magnitude — for quantitative dynamics, move on to stock-and-flow diagrams.