Why Cities Are More Compressible Than We Think, and Less Compressible Than We Hope

Apr 28, 2026

11 min read

7 views

91%

The wrong question about cities

What if the real question is not whether cities obey laws, but which parts of a city can be compressed into laws without destroying what makes them work?

That shift matters because urban thinking has long been trapped in a false choice. On one side sits the dream of universal regularity: if we gather enough data, cities will reveal elegant scaling laws, neat predictive equations, and transferable rules. On the other side sits skepticism: every neighborhood is historical, every street corner is contingent, every policy is tangled in institutions, habits, and accidents. The result is usually a stale debate between reductionism and exceptionalism.

A better frame is available. Cities are not fully law-like systems, but neither are they a heap of irreducible anecdotes. They are partially compressible systems. Some features collapse beautifully into general patterns. Others require conditional mechanisms. A final remainder resists compression because it is not noise around the city. It is the city’s historical substrate.

That distinction is not just philosophical. It changes how we interpret scaling laws, land use, mobility, retail access, urban form, and policy failure. It also changes what counts as progress in computational urban science: not total prediction, but increasingly disciplined partitioning of the urban world into what can be generalized, what can be explained conditionally, and what must be treated as inherited structure.

Cities are organized complexity, not simple complexity

Jane Jacobs called cities problems of organized complexity, a phrase that remains one of the best entry points into urban reality. Organized complexity means the city is neither random nor mechanically repetitive. It is full of interacting parts, feedback loops, durable structures, and emergent patterns. Michael Batty’s description of cities as open, bottom-up, far-from-equilibrium systems adds a dynamic vocabulary to the same intuition. Denise Pumain’s evolutionary view pushes further, treating urban systems as historical, path-dependent, open, and unpredictable in full detail, yet still capable of revealing stylized facts.

These are not competing ideas so much as different levels of compression. A city can be studied as a system with recurring regularities, and also as a historically accreted artifact that carries traces of old decisions long after the original conditions have vanished. That is why urban science keeps oscillating between universalism and context. The tension is real because both sides are right.

The deeper insight is that compression is the common thread. When a model captures a recurring pattern across many cities, it is compressing complexity into a smaller description. When a historical account explains why two seemingly similar districts diverged, it is also compressing complexity, but at a different level, by introducing the missing mechanism or path dependence that makes the data intelligible. Even critical approaches that resist totalizing concepts are, in effect, identifying where compression fails and why.

Understanding a city is not the same as flattening it. It is the disciplined search for the shortest description that still respects the structure of the world.

This is why the old distinction between nomothetic and idiographic inquiry matters. Laws and particulars are not mutually exclusive. They are different targets of description, and serious urban research needs both. The best urban theory is not a law machine or a story machine. It is a compression engine with a residue detector.

The grocery store lesson: proximity is not the same as behavior

A useful way to see this is through the puzzle of food deserts and supermarket access. It seems intuitive that if a neighborhood lacks nearby grocery stores, residents will eat worse, and that placing a supermarket there should improve diets. But household travel behavior complicates the story. Many households already travel several miles to shop, and most grocery trips are made by car. In that context, local entry by a new supermarket may change the landscape less than expected.

The point is not that place never matters. It is that distance alone is often a weak proxy for access when mobility systems, vehicle ownership, shopping routines, and price sensitivities already structure behavior. In other words, a neighborhood can look deprived on a map while functioning differently in everyday life. The urban object we think we are measuring is not always the urban object that governs action.

This is a general lesson for urban science: the most visible variable is often not the causal one. A shiny new store, a subway stop, or a new bus route may look decisive in spatial terms but remain behaviorally marginal if the real constraint lies elsewhere, such as reliability, household schedules, trip chaining, regulation, or price gradients. Conversely, a small institutional change can matter enormously because it alters the hidden architecture of choice.

This is where the idea of compressibility becomes powerful. Some urban phenomena are highly compressible into a simple spatial metric, but only if that metric is actually the binding constraint. Other phenomena require a richer model because the urban system is not being shaped by proximity alone, but by an interacting bundle of transportation, household structure, and institutional context.

A neighborhood food story therefore becomes a theory of urban cognition: what appears to be a geographic problem may actually be a coordination problem, a mobility problem, or a governance problem. If we mistake the map for the mechanism, we overestimate what local spatial fixes can achieve.

The urban world splits into three layers of compressibility

A more useful framework is to think of urban reality in three layers.

1. Compressible regularities

These are patterns that recur across many cities with enough consistency to support broad law-like statements. Examples include scaling relations, common density gradients, and certain forms of spatial interaction. Bettencourt’s work and the Santa Fe tradition are important here because they show that some socioeconomic and infrastructural quantities do cluster into recurring universality classes. People, roads, incomes, and built space do not arrange themselves arbitrarily.

Spatial interaction models belong in this layer too. For decades they have reduced movement problems to a parsimonious structure: origin, destination, and separation costs. That is not trivial. It is a genuine success of compression. A large amount of mobility behavior can be described with surprisingly few ingredients.

2. Conditional mechanism classes

The next layer contains patterns that look general only when the relevant conditions are held fixed. Retail vacancies, land use, and transport choice often live here. They are not explained by one causal variable, because they emerge from interactions among regulation, durable buildings, institutional rules, price dynamics, and heterogeneous users.

This is where middle-range theory becomes essential. Robert K. Merton’s idea was not to settle for tiny local observations, nor to leap to universal systems. The goal was to build abstractions that travel, but remain close enough to empirical reality to guide inquiry. In urban science, this means we need theories of specific mechanism classes, such as how zoning shapes commercial vacancy, how transport reliability shapes route choice, or how prior settlement advantages shape current centrality.

3. Historical residue

The final layer is the hardest to compress. It includes durable urban structures, inherited infrastructures, path-dependent land use, and institutionalized governance. Bleakley and Lin’s historical portage sites show the logic clearly: a location can retain importance long after its original advantage disappears because history has solved a coordination problem. Once enough actors expect a place to matter, the expectation itself becomes part of the explanation.

This is the realm where history is not a background condition. It is an active component of the mechanism. Urban property, infrastructure, and governance are repeatedly inscribed into space, and they constrain what later generations can do. In this layer, a city is less like a blank optimization surface and more like a palimpsest, where every new layer is written over old ones that still show through.

The hardest urban facts are often not the most local facts. They are the most durable ones.

This three layer model is useful because it avoids two errors at once. It does not pretend that all urban phenomena are equally law-like. It also does not surrender to the idea that every city is too unique to theorize. Instead, it asks a sharper question: what kind of compressibility does this phenomenon permit?

Why boundary choices matter more than we like to admit

There is another hidden lesson in urban science: sometimes the apparent law depends on how the city is defined in the first place. Population alone is not enough to describe or predict a city’s state, and nonlinear exponents can shift when boundary definitions change. That fact should unsettle anyone who treats a city as if it were a self-evident object.

A city is not merely a place with a population count. It is a set of overlapping definitions, administrative borders, commuting zones, land markets, infrastructure systems, and morphological patterns. Change the boundary, and you change the measurement. Change the measurement, and you change the law you think you discovered.

This is not a weakness of urban science. It is a sign that the city is an object whose structure depends on the scale of observation. A satellite view, a cadastral map, and a commuter network reveal different compressible patterns. The trick is not to find the one true boundary, but to be explicit about what kind of structure each boundary captures.

A good analogy is language. If you compress an essay into an outline, you lose detail but retain argumentative structure. If you compress it further into a headline, you retain only the core claim. Urban models work the same way. A boundary choice is a compression choice. It determines what structure survives and what gets discarded.

This is why the rise of new data fusion, geospatial computation, and classification methods matters. As measurement improves, the frontier of what can be compressed moves outward. Cross-city morphology research using global land-cover data and unsupervised learning can now identify typologies across more than 1,500 cities. That does not mean cities have become less historical. It means we have become better at isolating which historical differences are structural and which are secondary.

Scientific maturity in urban research does not eliminate context. It clarifies the residual.

A better mission for computational urban science

If compression is the organizing principle, then progress in urban science is not about maximizing prediction at any cost. It is about becoming clearer on three things:

What can be generalized across cities and across time.
What can be explained conditionally once the relevant mechanism is specified.
What remains irreducibly historical after every honest attempt at compression.

That third point is crucial. Residuals are often treated as failure. But a residual is not automatically error in the pejorative sense. Sometimes it is the clue that the phenomenon lives in a different causal register. Sometimes it is a sign that the model class is too coarse. Sometimes it is the record of long-accumulated institutional decisions embedded in streets, leases, property rights, or infrastructure.

A mature urban science should therefore publish not only where models fit, but where they do not. It should ask what remains after accounting for scaling, interaction, and mechanism. That remainder is where history becomes visible as structure.

This also resolves a common objection from both sides. The law-seeking camp may worry that historical specificity is endlessly expandable and therefore kills generality. The critical camp may worry that compression is just technocratic reductionism dressed up in formal language. The answer is that compression disciplines explanation instead of flattening it. It forces explicit choices about model class, omitted structure, coding decisions, and residual error.

In practice, that means better comparative data, longer time horizons, more careful boundary definitions, and a willingness to treat inherited infrastructure as a boundary condition rather than a nuisance. It also means accepting that some urban questions will never yield a universal answer, because the point of the question is to explain a historically produced configuration, not to abstract it away.

Key Takeaways

Ask what is compressible before asking what is true. In urban analysis, the first task is often to identify whether a phenomenon is governed by a general law, a conditional mechanism, or historical inheritance.
Do not confuse proximity with causality. A nearby supermarket, transit stop, or new building may matter less than mobility patterns, reliability, regulation, or household routines.
Treat boundaries as theoretical choices, not neutral facts. How a city is defined changes which regularities appear and which disappear.
Use residuals as a research tool. What a model cannot explain is often more informative than what it can.
Think in layers, not absolutes. Cities contain scaling laws, mechanism classes, and durable historical substrates at the same time.

The city is a machine that remembers

The deepest reason urban theory keeps returning to the same debate is that cities are neither pure systems nor pure stories. They are systems that remember. Roads, ports, zoning rules, housing stocks, and commuting habits all preserve the past in present form. At the same time, these inherited structures interact with general patterns of movement, density, and coordination that can be surprisingly regular.

That is why the most useful framework is not “law versus history.” It is law, mechanism, and residue. First compress what can be compressed. Then explain the conditions that make the compression hold. Finally, identify the parts of the city that remain because they were made earlier, and because earlier choices still govern present options.

Seen this way, a city is not an exception to theory. It is a test of whether theory can respect complexity without surrendering to chaos. And if computational urban science succeeds, it will not be because it erased history. It will be because it learned to tell the difference between what history merely decorates and what history actually determines.

That is the real frontier: not a world where cities become fully predictable, but one where we know exactly how far prediction can go, what it depends on, and why the remainder deserves to stay.