From the system dynamics of Jay W. Forrester (Industrial Dynamics, 1961), codified by Jorgen Randers (Guidelines for Model Conceptualization, 1980) and John D. Sterman (Business Dynamics, 2000).
Plots how the key variables have actually behaved over time, and how they might behave under different policies, before any model is built.
Name the pattern before building the model: lay each reference shape over the history
Three years of quarterly staff turnover at a chain of veterinary clinics, as a yearly rate (the black line). Slide through the five reference shapes (growth, decline, oscillation, S-shape, and overshoot and collapse) and watch the gap: the shape with the smallest gap names the behavior any explanation has to reproduce. Illustrative.
The reference shapeActual turnover, yearly rate
An example
A chain of veterinary clinics had a staff turnover problem that kept changing its story. When turnover peaked at 18 percent, the explanation was a rival chain opening nearby. When it fell to 10, credit went to a new manager training program. When it climbed back to 18, the blame moved to a pay freeze. Each fix was funded, and none of them held.
Before commissioning another study, the new head of people plotted three years of quarterly turnover on one chart and laid the standard reference shapes over it: growth, decline, oscillation, S-shape, overshoot and collapse. Only one fit. Turnover rose and fell on a cycle of about a year and a half, the signature of a balancing loop with a delay. That pointed to a single cause in place of three: pay was reviewed every eighteen months against a labor market that moved faster, so the clinics caught up just as people left and fell behind again before the next review. The chain moved to a six-month pay review.
Why it works
Teams usually explain a problem with whatever happened just before the latest data point, and each new data point gets a new explanation. Reference mode analysis slows that down. It asks first what the behavior over time looks like, and names it as one of a few patterns that system dynamics links to typical structures. Steady growth points to a reinforcing loop; an S-shape to a reinforcing loop meeting a limit; oscillation to a balancing loop with a delay; overshoot and collapse to a limit that erodes as it is exceeded.
Naming the pattern rules explanations in and out. A one-off cause, such as a rival's opening, cannot produce a repeating cycle. The chosen shape becomes a target any proposed model must reproduce, so a team can test its explanation against the history before it spends money on a fix.
The thinking behind it
Jay W. Forrester Jay W. Forrester, an MIT engineer and management theorist, founded system dynamics, which models organizations as stocks, flows, and feedback loops with delays. Industrial Dynamics (1961) drew on a study of Sprague Electric and gave the first account of what is now called the bullwhip effect, in which delays in a supply chain turn steady demand into swinging orders. Plotting behavior over time and asking what structure produces it begins with his work.
Jørgen Randers Jørgen Randers, a Norwegian futurist and one of the four authors of The Limits to Growth (1972), edited Elements of the System Dynamics Method (1980), a standard collection on the method. His chapter, Guidelines for Model Conceptualization, gives guidance on drawing a problem's reference mode, its behavior over time, before building the model that is meant to explain it.
John D. Sterman John D. Sterman, director of the MIT System Dynamics Group, wrote Business Dynamics: Systems Thinking and Modeling for a Complex World (2000), the standard textbook of the field. It covers stocks, flows, feedback loops, delays, and the typical structures that produce each pattern of behavior. His beer distribution game experiments, reported in Management Science in 1989, showed that managers misread feedback and delays and order in ways that create wild swings, the kind of cycle a reference mode makes visible.
How to run it
- Choose the variables that matter. Pick two or three: the problem itself and what is thought to drive it, such as turnover and pay against the market.
- Plot each one's history. Use as long a stretch as the data allows, at a steady interval; three years of quarters is a minimum for a cycle.
- Name the pattern. Compare the plot with the standard shapes and choose the one that fits best, saying where it does not.
- Write the structure the pattern implies. A cycle needs a balancing loop with a delay; an S-shape needs growth meeting a limit.
- Sketch the future under current policy. Draw what the pattern predicts if nothing changes.
- Sketch the future the team wants. Draw the behavior a new policy should produce, as a second line on the same chart.
- Hold any model to both plots. An explanation that cannot reproduce the history, or reach the target, is not finished.
Where it goes wrong
The common failure is too short a history. A year of data can look like growth or decline when it is the rising half of a cycle. Extend the series back, even with rougher data, before naming the pattern.
A second failure is forcing a fit. Real data are noisy, and two shapes may fit about equally; say so, and keep both explanations alive until more data separate them. The third is stopping at the picture. Naming a cycle is useful only if it leads to the loop and the delay that produce it; ask what is being adjusted, by whom, and how late, and test the answer against the plot.
When to use something else
- Causal Loop Diagram. When the pattern is named and the work is drawing the loops that produce it.
- Stock-and-Flow Models. When the explanation needs numbers: stocks, rates, and delays simulated to reproduce the history.
- Control Chart. When the question is whether the variation is noise or a real change, before any pattern is named.
Where it sits in SOLVE eX
Go to the source
- Business Dynamics: Systems Thinking and Modeling for a Complex World, 2000
- Guidelines for Model Conceptualization, Elements of the System Dynamics Method, 1980
- Industrial Dynamics, 1961