Meadows et al., 1972 — the whole world compressed into five stocks and a set of delayed feedbacks
Cited in: Topic 35 Limits to Growth and the S-Curve
01The Question It Poses
Around 1970, almost every argument about resources was a static ledger: divide the reserve of some mineral by this year's consumption and read off "how many years are left". That sum has two large holes. First, consumption itself is growing, so the true answer is much shorter. Second, what drives consumption — population and industry — is in turn affected by resources, pollution and food. The five things bite each other; none of them is exogenous.
Jay Forrester and then the Meadows team asked something other than "when do we run out": if these five things feed back on each other, and every feedback takes decades to complete, what shape does the structure itself produce? World3 is the apparatus for answering that. It was never a prediction machine; it is a shape generator, asking which futures this class of structure can grow, not what happens in which year.
02The Rules
Five stocks (quantities that can only be changed slowly, through flows): population, industrial capital, agricultural capital and arable land, persistent pollution, non-renewable resources.
One set of positive feedbacks: population → labour → industrial output → part reinvested as capital → more output; capital → agricultural inputs → food per capita rises → death rate falls → more population. This loop turns fast and has no ceiling of its own.
One set of negative feedbacks, all with long delays: industrial output → pollution emitted → (decades before it fully bites) → land yield and life expectancy fall; capital → resource extraction → less resource left → more capital needed per unit of resource obtained → less capital available for everything else.
Every stock leaks: capital depreciates, land erodes, pollution degrades, people die.
Twelve scenarios are not twelve models: the structure is identical, only parameters change — double the resource base, add pollution control, double land yield, universalise birth control. The point is how one structure behaves under different assumptions.
The crucial phrase is "long delays" in rule 3. It is not a technical detail, it is the engine of the whole model: the positive loop runs fast, the negative loop arrives slowly, and that speed difference alone is enough to produce overshoot. Nobody has to make a mistake.
The left loop turns in years; the right chain takes decades. That speed difference is itself the source of overshoot — nobody has to make a mistake.
03What You See When It Runs
The most quoted run is the standard run, meaning "no optimistic assumptions in the parameters". What comes out is not one curve but five with staggered peaks: the resource stock slides down throughout; industrial output per capita peaks and turns first; food and services follow; and population turns last, decades later.
The ordering and the lags matter far more than the values. What they say is: by the time you see a problem in population or welfare, the industrial side went over its peak a long time ago. The signals arrive in a queue, not together.
The second observation is sturdier and more often missed. Double the resource base and run it again: the collapse does not disappear — it is postponed by decades, and is then triggered by pollution instead. Relaxing one constraint leaves the shape unchanged and merely swaps which constraint binds. This is the model's most valuable output and its only conclusion that does not depend on specific parameters.
Double the resources and the shape is unchanged, the timing is pushed back, and the binding constraint switches from resources to pollution. This is World3's least parameter-dependent conclusion.
04What It Explains
World3 explains no particular event. What it offers is a catalogue of shapes for an entire class of system: wherever growth has momentum, the constraining feedback is slow, and the thing being consumed recovers slowly, there are only a few possible endings — smooth approach, oscillation around the ceiling, or overshoot followed by settling at a lower level.
It also explains a recurring pattern of policy failure: a single technical fix often just postpones the collapse and changes the cause of death. That is not pessimism about technology; it is a diagnosis of "relaxing exactly one constraint". Unless several loosen together, the system simply jams somewhere else.
As for whether it matches reality: Graham Turner (2008, 2014) and Gaya Herrington (2021) compared observed data against the scenarios and concluded that aggregate variables — population, industrial output, pollution — broadly track the standard run. That conclusion should be read carefully; see the next section.
What It Cannot Explain
It cannot predict timing. The entire planet is aggregated into a single cell: no countries, no geography, no trade, no distribution. The position of any curve along the horizontal axis should not be taken seriously, and "what happens in year X" is exactly what the model refuses to supply.
There are no prices, hence no substitution. This was the core of William Nordhaus's well-known 1973 critique: in reality scarcity raises prices, and higher prices trigger exploration, substitution and efficiency investment. That feedback is strong, and it is essentially absent here. Without it, resource constraints are drawn systematically too hard.
Its parameters cannot be independently calibrated. Many relationships are hand-drawn shape functions (the effect of pollution on lifespan, say) with no independent data source. So "the run looks like reality" is inherently weak evidence — the coarser the aggregation, the more it can fit.
The mineral-depletion table was not a forecast. The famous "years of this mineral remaining" table demonstrated how fast exponential arithmetic eats a reserve. It has been attacked as a forecast for fifty years, which is a critique of something the model never claimed.
Scenarios are not policy, and the model contains no politics. There is no "who decides", no "who bears the cost", no "how agreement is reached". It can tell you what shapes a class of structure produces; it cannot tell you what to do, and it certainly cannot endorse a specific policy.