Book Summary: Thinking in Systems by Donella Meadows

This summary of Thinking in Systems: A Primer by Donella Meadows explains how systems thinking can help us better understand and change the systems in our lives.

Buy Thinking in Systems at: Amazon | Kobo (affiliate links)

Key Takeaways from Thinking in Systems

• What is a system?
  • Systems are things where the whole is more than the sum of its parts.
  • There are three essential components: Elements, Interconnections and a Purpose/Function.
  • In the real world, systems don’t have clear boundaries because almost everything is connected. For models to be useful, however, we need to draw boundaries.
• Stocks and flows:
  • Flows are the material or information going in and out of an element.
  • A stock is the history of accumulated flows over time.
• Feedback loops:
  • Balancing feedback loops adjust flows so as to keep stocks within a certain range.
  • Reinforcing feedback loops amplify flows, leading to exponential growth or collapse over time.
  • Resilient systems have multiple feedback loops and some degree of redundancy.
  • Delays in a feedback loop can lead to complexity and oscillations, making it hard to predict a system’s behaviour over time.
  • Complex systems may involve hierarchies of different subsystems, with feedback loops operating a different levels.
• Before intervening in a system, we should try to understand it:
  • Everything we think we know about the world is merely a model of reality.
  • Understanding systems will likely require interdisciplinary thinking.
  • Some common unintuitive features include: non-linearities, causation, limiting factors, and bounded rationality.
  • Common system archetypes include: policy resistance; tragedy of the commons; drift to low performance; rich get richer and others.
• Even when we understand a system, we may not be able to control it or predict its behaviour.
  • There’s a gap between knowing and doing.
  • Systems naturally tend to be resistant to change.
  • We’ll never fully understand systems — but we can redesign and learn from them.

Detailed Summary of Thinking in Systems

What is a system?

A system is a set of things interconnected in such a way that they produce their own pattern of behaviour over time.

Every system has three things:

• Elements
• Interconnections
• Purpose or function

Elements

Elements are the easiest to spot, as many are visible, tangible things. For example, a university system is made up of students, professors, administrators, buildings, books, etc.

Often people blame problems on particular actors or events, even when the root cause is the system’s structure. The elements tend to be the least important part of a system because you can usually change the elements without changing the system’s overall behaviour.

There is an exception to this — sometimes changing an element also changes the interconnections or purpose/goal of the overall system. For example, sometimes the leader at the top of the system may have power to change the system’s overall purpose.

Interconnections

These are the relationships that hold the system together, and they often operate through the flow of information. (See below under “Feedback loops” for more.)

Purpose or function

A system’s purpose or function can be the hardest to spot as it’s not necessarily written down anywhere. The best way to find a system’s purpose is to observe its behaviour over time, rather than its rhetoric or stated goals. (See below under “Events vs behaviours vs structure” for more.)

A system may have a function or purpose even if no individual within the system has that function or purpose. For example, no one deliberately created poverty, drug addiction, or war, and many people want to fix them, but it’s difficult to do so because they are intrinsically systems problems.

A system is more than the sum of its parts

Systems have an integrity or wholeness about them. A set of things that are not interconnected and do not have an overall function is not a system. For example, sand scattered on the side of the road is not a system. You can add or take away sand, but it’s still just sand.

Boundaries

Real-world systems rarely have strict boundaries because everything is connected. As complexity grows, systems will often organise themselves into hierarchies. We often take this self-organisation for granted.

However, some degree of simplification will be inevitable as you try to understand a system — so boundaries are also inevitable. Hierarchies are often “decomposable”, meaning its subsystems can function autonomously when separated from other parts. For example, a doctor can usually treat liver disease by just looking at the liver, not at other organs or your personality. But there are also many exceptions to this — e.g. perhaps your job or lifestyle is causing the harm to the liver.

Drawing the right boundaries can be tricky:

• If boundaries are too narrow, the system will likely surprise you as you’ll fail to account for something outside of your model. The greatest potential for unintended consequences arise at a system’s “boundaries”;
• If boundaries are too broad, your diagram gets too complicated. Systems analysts often fall into this trap, producing diagrams with so much small print it ends up inhibiting understanding rather than aiding it.

Where you draw the boundaries depends on what you want to understand as well as your timeframe (longer timeframes generally require broader boundaries). Ideally, we’d find the right boundary for thinking about each new problem, but we tend to get attached to the boundaries we’re used to.

Stocks and flows

Flows are the material or information going in an out of an element. A stock is the history of accumulated flows that have built up over time. For example, the amount of water in your bathtub is a stock while the water going in and out of it are the flows.

Stocks act as shock absorbers

A system’s “stock” acts as a buffer or shock absorber. For example, even if you turn off the faucet and pull the plug, it takes time for your bathtub to drain. This is one reason why systems are slow to change.

While this slowness can cause problems, it also helps keep the system stable even when flows are unpredictable and there are lags in the system. For example, if you’re a retailer, customer demand may vary and it takes time to receive orders from your supplier. Maintaining a stock (an inventory) allows you to keep selling constantly.

We often focus too much on flows, instead of stocks or a system’s underlying structure

Events are the output of a system

We tend to see the world as a series of events — elections, battles, disasters, economic booms and busts. We ordinarily talk about specific events at specific times and places. News media does this too — they don’t generally help to put news events into the broader context.

Yet events are merely the output of the black box of the system.

Try to understand longer-term behaviour and structure instead

Instead of focusing on events, systems thinkers should first look at a system’s performance over time — its behaviour. Time graphs and data can reveal how elements vary or don’t vary together.

Over the long-term, that behaviour can help us understand the system’s stocks, flows, and feedback loops — its underlying structure. System structure is the source of system behaviour, so is key to understanding why events happen.

The key is to start with behaviour (the facts) and go back and forth between behaviour and structure (the theory). People are often far too confident of theories that are utterly unsupported by any data. We also have a tendency to define a problem not by the system’s behaviour but by leaping to our favourite solution.

Feedback loops

A feedback loop occurs when changes in a stock affects the flows into or out of that same stock. Many systems have feedback loops, but not all do.

Real life systems usually have multiple feedback loops of different strengths interacting in very complex ways. A loop that has a stronger impact on behaviour is said to dominate the other. But dominance can shift over time. Complex behaviour often arises as the relative strengths of feedback loops change.

There are broadly two types of feedback loops: balancing and reinforcing feedback loops.

Balancing feedback loops

Balancing loops adjust inflows or outflows to keep the stock at a certain level or within a certain range. For example, if you see your bank balance is low, you may work extra hours (increasing inflows) or cut your spending (reducing outflows), to keep your bank balance within the desired range.

Balancing loops oppose whatever direction of change is imposed on the system, so are sources of stability in the system.

Example: thermostat — two competing balancing loops

A thermostat regulates your room’s temperature by trying to keep it within a desired range with two competing balancing loops:

• Heating loop. When the temperature falls below the desired range, the thermostat turns on the furnace. When the temperature drops below the desired range, the thermostat turns off the furnace.
• Cooling loop. Because heat leaks, the cooling loop always tries to make the room temperature equal to the temperature outside.

The two loops push in opposite directions. Which one dominates depends on things like the your insulation quality, furnace size, and how cold it is outside.

Once a room reaches the set temperature, the furnace cycles on and off as heat slowly leaks outside. This is a state of dynamic equilibrium, where the system’s stock (the room temperature) does not change even with flows in and out of the system.

Reinforcing feedback loops

Reinforcing loops amplify, self-multiply or snowball, creating cycles that lead to exponential growth or collapse over time. This type of feedback loop generates more input to a stock the more there already is (and less input the less there already is). Examples include the inflation wage-price spiral, population growth and compound interest.

However, any physical, exponentially growing system will eventually run into a counteracting balancing loop. No physical system can grow forever in a finite environment. A popular product will eventually saturate the market; a virus will run out of people to infect.

Growth in a constrained environment is so common that it’s called the “limits-to-growth” archetype. If limits to growth are not self-imposed limits, the system will impose them.

Resilient systems have multiple feedback loops and some redundancy

Resilience is a measure of a system’s ability to persist in the face of change and to repair or restore itself. Our bodies and ecosystems are examples of incredibly resilient systems.

A resilient system will have a structure of feedback loops that can restore it even after a large shock. Systems are more resilient if there are multiple feedback loops operating through different mechanisms and different time scales, with some redundancy. Feedback loops that can learn and evolve more complex restorative structures lead to even more resilience.

A common mistake is to confuse resilience with constancy. Constancy is easy to observe; resilience is not. You might only see if a system is resilient if you push it to its limits or take a whole-system view. But constant systems can be remarkably unresilient, and resilient systems can be very dynamic. Because resilience can be hard to see, people often sacrifice it for more visible measures like stability or efficiency. For example, just-in-time supply chains increase efficiency at the cost of resilience.

Delays in feedback loops

Delays are inevitable

We seem to be constantly surprised by how much time things take. [See also Planning fallacy.] In a system, all stocks are delays. Most flows also have delays (shipping delays, processing delays, etc). A system cannot deliver information or have an impact fast enough to correct the behaviour that drove the current feedback. A flow cannot react instantly to a flow; it can only react to a change in a stock, and only after a slight delay.

A balancing feedback that tries to maintain stock at a particular level must have its goal set to compensate for other inflows or outflows, else it will miss its target. Recall the thermostat example — it takes some time for the furnace to correct for heat loss, so the equilibrium temperature will be slightly lower than the thermostat’s setting. You should therefore set the thermostat slightly higher than your desired temperature.

Which delays you care about will depend on your purpose and timeframe.

Delays can lead to oscillations and complexity

Delays can make a system’s behaviour complex and hard to predict. Many economic models assume that consumption or production can respond instantly to changes. But real economies are extremely complex systems, full of balancing feedback loops with delays. They are inherently oscillatory and don’t behave exactly like the models.

The example below shows how a system can behave counterintuitively when we start trying to change them.

Before intervening in a system, we should try to understand it

Before disturbing or intervening in a system, watch how it behaves and try to “get its beat”. But our understanding will never be perfect. Systems will always surprise us to some extent.

Everything we think we know is a model of reality

Everything we think we know about the world is a model of reality, not reality itself. Our models work in most cases, which is why humans are such a successful species. But they are not perfect. They don’t fully represent the world, which means we will often make mistakes.

There is a lot of value in making your models’ assumptions explicit and consistent and inviting others to challenge them. It’s also a good practice to set it out with diagrams and equations to clarify your thinking. Using words, lists or pictures is fine as well.

Systems require interdisciplinary thinking

Understanding systems will likely cross traditional disciplinary lines. To understand the system, you must learn from various experts — e.g. economists, chemists, psychologists and theologians.

At the same time, you should be careful not to get drawn into the distortions that come from their narrow and incomplete expertise. This can be hard. [In Range, David Epstein similarly talks about how specialists are better at kind problems, while generalists are better at wicked problems.]

Common unintuitive features in systems

Nonlinearities

A linear relationship between two things is represented on a graph by a straight line with constant/propotional effects and returns. Linear thinking comes very naturally to us. But the world is full of non-linear patterns, which confound our expectations.

More importantly, non-linear relationships can change the relative strengths of feedback loops. This can completely flip a system’s behaviour. A dominant reinforcing loop may lead to exponential growth… until the balancing loop suddenly dominates and leads to rapid decline.

Causation

Our minds like to think about single causes producing single effects. At most, we’re comfortable with dealing with a handful of things at a time.

But the real world is much more complex than that. Many causes (or inputs) will often combine to produce many effects (or outputs). [See also Causality is so much harder than we normally think.]

Limiting factors

Every growing thing has multiple limiting factors [aka “bottlenecks”]. We often fail to appreciate the importance of limiting factors. For example, economics evolved at a time when labour and capital were the most common limiting factors to production. Most economic production functions therefore only account for these two factors, and overlook other limiting factors such as clean water/air, dump space or raw materials.

The interaction between growth and its limiting factors is dynamic — i.e. growth itself will change which factor is limiting. When one factor stop being limiting, growth occurs; but growth then changes the relative scarcity of factors until it hits another limiting factor. For example, if a city meets its inhabitants’ needs better than any other city, people will flock there until some limit brings down the city’s ability to satisfy people’s needs.

Bounded rationality

People make reasonable decisions based on the information they have.
However, there are many examples of people acting rationally in their short-term best interests and producing negative results in aggregate. This is because of bounded rationality, due to:

• Imperfect information. We usually have incomplete or delayed information, particularly on more distant parts of the system.
• Satisficing. We tend to satisfice instead of optimise. We don’t see the full range of possibilities, so we stick to choices we can live with instead of seeking a long-term optimum.
• Misperceptions. Even when we have information, we don’t always interpret it correctly. We misperceive risk, discount the future too heavily, and are vulnerable to things like confirmation bias.

Meadows argues that the solution in these cases is rarely to blame the individual or replace them with another person. Instead, we should restructure the system’s information flows, goals, incentives and disincentives so that separate, bounded, rational actions produce outcomes that are in the common interest.

Common system archetypes

Some systems may look very different on the surface but have common feedback loop structures that produce similar dynamic behaviours. These are also known as “archetypes” — common system structures that produce characteristic patterns of behaviours.

I’ve included a very brief summary of the archetypes discussed in the book (Meadows goes into much more detail):

• Policy resistance (fixes that fail). When there are inconsistent subsystem goals, if one subsystem manages to move the system stock in one direction, other subsystems may just increase their efforts to pull it in the other direction.
• Tragedy of the commons. This occurs when there is growth in a commonly shared, erodable environment. The commons structure makes selfish behaviour much more profitable than pro-social behaviour.
• Drift to low performance. In a reinforcing loop, negativity bias causes us to gradually lower our standards, which ends up becoming a self-fulfilling prophecy.
• Escalation. Competing actors trying to get ahead of each other in a reinforcing loop can build exponentially and quickly reach extremes.
• Rich get richer. When the winners of a competition get an advantage that helps them compete even more effectively in the future, it results in a reinforcing loop. If the winners’ gains are at the losers’ expense, the losers are eventually bankrupted or forced out. [Meadows calls this “success to the successful” but that sounds far clunkier imo.]
• Addiction. An actor wants to maintain a stock through a balancing loop and resorts to some intervention (e.g. drugs) to do so. However, the intervention merely fixes the symptoms of the underlying problem. This can cause issues if the intervention undermines the system’s original capability to maintain itself — then more intervention will be needed, leading to more atrophy, etc.
• Rule beating. Rule beating is evasive action that abides by the letter of the law, but not its spirit. It often occurs when the lower levels in a hierarchy don’t understand or agree with rules imposed from above. This becomes a problem if it causes unnatural behaviour that would not make sense in the absence of the rules.
• Seeking the wrong goal. This is just about the limitations of proxy measures and Goodhart’s law. In some ways the opposite of Rule beating.

Even when we understand a system, we may not be able to control it or predict its behaviour

There’s a gap between knowing and doing

New systems thinkers often mistakenly think that understanding a system will allow them to predict and control it. Meadows herself made this mistake when she was a student.

But she quickly discovered that there was a big difference between understanding how to fix a system and actually fixing it. This was as true for their own systems as for broader social systems — they understood the structure of addiction yet couldn’t give up their coffee.

Systems tend to be resistant to change

An important function of almost every system is to ensure its own perpetuation.[I think this is due to selection bias, since systems that are not resistant to change would quickly cease to exist in that form.]

Systems tend to be resistant to change because of:

• Stocks (since, as noted above, stocks act as shock absorbers); and
• Balancing feedback loops.

Meadows suggests a list of 12 possible leverage points, where small changes can cause large shifts in behaviour. Leverage points are often counterintuitive. For example, the Club of Rome (a bunch of business leaders, policymakers, economists and scientists) asked Jay Forrester to develop a complex model to try and understand the nature of the world’s systems and how to solve problems such as poverty, environmental degradation, etc. He did, and came up with a clear leverage point: growth. But decreasing growth (both economic and population growth), not increasing it. World leaders correctly focused on growth as the solution — just in the wrong direction.

In order from lowest leverage to highest, the leverage points are:

12. Numbers and parameters
13. Buffers
14. Stock-and-flow structures
15. Delays
16. Balancing feedback loops
17. Reinforcing feedback loops
18. Information flows
19. Rules
20. Self-organisation
21. Goals
22. Paradigms
23. Transcending paradigms

The book goes through each of these with examples, as Meadows explains why she put it in the position that she did (with the caveat that her list is just a work-in-progress). I won’t attempt to summarise it all here.

My main takeaways were:

• We often focus a lot on “Numbers and parameters”, which tend to be the short-term flows. But since these numbers generally won’t change the system’s underlying structure, they’re last on the list.
• Changing the “Buffers”, “Stock-and-flow structures” and “Delays” have bigger impacts, but they tend to be very hard or costly to change.
• Feedback loops are powerful leverage points. Slowing the growth of a reinforcing loop is usually more effective than strengthening the counteracting balancing loops (of which there could be many), which is why “Reinforcing loops” are higher on the list.
• All the leverage points from “Information flows” up tend to be pretty powerful.
• “Paradigms” and “Transcending paradigms” seem to be about changing your/other people’s mindset or assumptions. (Meadows didn’t clarify, but I assume these only apply to social systems, not biological or physical systems.)

We’ll never fully understand systems — but we can redesign and learn from them

Complex systems are inherently unpredictable. Each new insight raises more questions, many of which people from other disciplines have been asking for years. We can only understand systems in the most general way and control them temporarily, at best.

That said, we can still design and redesign systems. We just have to expect surprises along the way and learn from them.

Faced with dynamic and changing systems, we must be flexible and adapt depending on the state of the system. The best policies should have meta-feedback loops — loops that alter, correct, and expand other feedback loops — so that learning is incorporated into the process. [This reminded me of the construction examples in The Checklist Manifesto and Chris Blattman advocating for taking an incremental approach in Why We Fight].

Other Interesting Points

• Meadows wrote Thinking in Systems in 1993, but the book was not published until 2008, seven years after Meadows unexpectedly died.
• It’s difficult to discuss systems only in words because words follow a linear order. Systems, on the other hand, happen all at once, with connections in multiple directions simultaneously. Meadows uses various diagrams in the book, which I have not reproduced because of copyright.
• Sailboat racing started out being for fun — people just raced whatever boats they already had. The sport became more standardised as people discovered that races are more interesting if competitors are roughly equal in speed and manoeuvrability. People then started designing boats for winning races in particular classes. As races got more serious, rules became stricter, and boats got more and more weird-looking. Today, racing sailboats are nearly unseaworthy — you’d never use them for anything but racing.

My Review of Thinking in Systems

I really enjoyed Thinking in Systems. I’d already thought a lot about many of the ideas in it (bottlenecks, delays, feedback loops, etc) in a loose, scattered way. Meadows’ book helped me put that all together into a more coherent framework. In some ways, it felt like when I started learning basic economics — while some of the ideas may sound kind of “obvious” taken individually (externalities, principal-agent problems, etc), it’s still incredibly useful to have an overarching framework to support your thinking.

Generally, I found Meadows to be a trustworthy writer. She explains at the outset that the book is biased and incomplete, dealing only with the core of systems theory and not the leading edge. Meadows also admits to being regularly surprised by systems herself. And, for the most part, her proposed solutions also contain a good dose of humility.

That said, the book is not perfect. Several of the practical examples (especially the short ones) I thought were a bit more arguable than Meadows made it seem. Some may also accuse Thinking in Systems of being “too left-wing”. This didn’t particularly bother me since Meadows admits at the outset she has biases and I actually agree with her on many issues, though I could see how others with different views may find it annoying. However, I found it interesting that Meadows’ conclusions tend to be more left-wing (e.g. protect the environment, tax the rich), but she reaches them using more conservative lines of reasoning (i.e. don’t disturb systems you don’t understand).

I’ve also posted some detailed thoughts about economic issues and examples raised in Thinking in Systems here.

Have you read Thinking in Systems? Share your thoughts on this summary in the comments below!

Buy Thinking in Systems at: Amazon | Kobo <– These are affiliate links, which means I may earn a small commission if you make a purchase. I’d be grateful if you considered supporting the site in this way! 🙂

If you enjoyed this summary of Thinking in Systems, you may also like:

• Book Summary: Algorithms to Live By by Brian Christian and Tom Griffiths
• Book Summary: Why We Fight by Chris Blattman
• Book Summary: Evicted by Matthew Desmond