VIEW THIS AS

Auto mode follows the Route Engine until you choose a viewpoint.

YOU ARE HERE

ROUTE CHECK

CONNECTED TO

WHAT NEXT

Use the canonical route for this room, or HELP if you are unsure.

How Geography Works | Spatial Association and Causation — Why Two Patterns on the Same Map Do Not Prove One Caused the Other

Map A is dark in the city centre.

Map B is dark in the city centre too.

It is tempting to draw an arrow between them.

Geography asks whether the arrow is real.

Spatial association means two patterns occur together in space. Causation requires an additional claim: changing one process helps produce change in the other through a defensible mechanism.

This is the third pillar beneath How Geography Works | Spatial Thinking. The master owns spatial reasoning broadly. This article owns the inference gate between “the maps overlap” and “we know why.”

Quick Read

Spatial association occurs when values or events show related geographic patterns. But overlapping maps can be produced by a third variable, common population density, shared infrastructure, historical path dependence, measurement bias or boundary choice. Causal reasoning therefore requires temporal order, mechanism, comparison, alternative explanations and tests. Geographic evidence becomes stronger when a proposed mechanism predicts new spatial differences that were not used to invent the explanation.

pattern A + pattern B → spatial association → confounder search → temporal order → mechanism → comparison → prediction → test → bounded causal claim

Overlap Is a Clue, Not a Verdict

Neighbourhoods with more restaurants also have more taxi pickups.

Does restaurant supply cause taxi use?

Possibly partly.

But both may also be responding to:

  • population density;
  • nightlife;
  • tourism;
  • transit hubs;
  • commercial land use.

The overlap is genuine. The first causal story may still be incomplete.

Population Density Is the Classic Spatial Confounder

More people create more opportunities for many things to happen.

Areas with many residents may contain more:

  • crimes;
  • shops;
  • traffic collisions;
  • medical visits;
  • social-media posts;
  • delivery orders.

Two raw-count maps can correlate simply because both measure activity where more people are present.

Rates, exposure measures or matched comparisons may tell a different story.

A Shared Third Process Can Generate Both Patterns

Tree cover is low where surface temperature is high.

That may support a cooling mechanism.

But building density, road surface, shading geometry, traffic and sensor placement may also matter.

The correct question is not whether the first explanation sounds plausible.

It is whether competing explanations can account for the same spatial evidence.

Temporal Order Matters

If A causes B, the relevant change in A must normally precede the change in B.

A cross-sectional map can hide this completely because all values are shown in one time slice.

Historical maps, before-and-after observations or time-stamped spatial data can reveal whether the proposed cause arrived first.

A Mechanism Should Explain Direction and Geography Together

If a factory causes downwind pollution, the mechanism predicts more than simple proximity.

It predicts a directional pattern related to wind.

If a railway station increases nearby retail activity, the mechanism predicts stronger change where access and pedestrian flow improve, not equally in every direction.

A strong mechanism explains why the association has the spatial shape it has.

Mechanisms Generate Counterfactuals

What would we expect if the proposed cause were absent?

What similar place did not receive the intervention?

What happens just beyond the service boundary?

Counterfactual reasoning turns a map from description into testable comparison.

Natural Boundaries Can Create Useful Comparisons

A policy changes on one side of a border.

A transit line opens in one corridor but not a similar nearby corridor.

A flood-control intervention affects one catchment.

These situations can create stronger comparison opportunities when the groups are genuinely comparable and other assumptions are defensible.

But Boundaries Can Also Mislead

Administrative areas may differ in income, history, housing or infrastructure long before a policy is applied.

A visible border is not automatic experimental randomisation.

Scale Can Create or Destroy Association

At national scale, income and urbanisation may move together.

Within one city, the relationship may be weaker, reversed or highly uneven.

Scale owns the deeper problem. Here the causal lesson is that an association is conditional on the level at which it was measured.

The Ecological Fallacy Is a Level Error

Suppose districts with higher average income also have higher average public-transport use.

That does not prove higher-income individuals are the people using public transport more.

Area-level relationships cannot automatically be transferred to individuals inside those areas.

The Opposite Error Also Exists

Individual behaviour does not automatically predict area-level structure.

Thousands of individual decisions can interact through housing markets, networks and policy to create a pattern no single individual intended.

Geography often requires reasoning across levels without collapsing them.

Spatial Autocorrelation Changes Ordinary Statistical Intuition

Nearby observations are often more similar than distant ones because they share environment, infrastructure, history or interaction.

That means observations may not be independent in the way simple statistical models assume.

Formal spatial statistics can model this dependence. The conceptual point for students is simpler:

nearby data points may partly repeat the same underlying spatial information.

Correlation Is Still Useful

“Correlation is not causation” should not be translated into “correlation is useless.”

Association can:

  • find candidate mechanisms;
  • identify anomalies;
  • prioritise fieldwork;
  • generate predictions;
  • show where a theory fails.

How Correlation Works owns association generally. This pillar owns the additional geographic complications created by location, scale and neighbourhood dependence.

Regression Does Not Magically Create Causation Either

Add several control variables.

The estimate may improve.

But omitted variables, measurement error, reverse causation and model misspecification can remain.

How Regression Works owns the general modelling system.

Reverse Causation Can Be Spatial

Do parks attract high-income residents?

Or do high-income communities gain more parks?

Or does planning history create both?

Spatial overlap alone cannot settle direction.

Selection Into Place Matters

People and firms choose locations.

Those choices are rarely random.

A high-performing firm locating beside other high-performing firms may reflect agglomeration benefits, selection of already-successful firms into the cluster, or both.

How the World Works | Sorting explores why similar people, firms and opportunities can cluster together.

Measurement Can Create Association

More sensors are placed near roads.

More events are detected near roads.

Now road proximity appears associated with the phenomenon even if part of the association comes from observation density.

The fourth pillar, Spatial Data, owns this data-generating layer.

A Map Layer Can Be a Proxy

Night-time lights can proxy some economic activity.

Road density can proxy accessibility.

Vegetation indices can proxy aspects of plant condition.

A proxy is useful only when its relationship to the underlying concept is understood and bounded.

Historical Path Dependence Can Mimic Current Cause

Factories cluster near an old railway.

The railway is now minor.

The cluster persists because land use, skills and suppliers accumulated there.

A current map may suggest the present railway causes present industry when the deeper mechanism is inherited history.

Fieldwork Can Test Mechanisms the Map Cannot See

Maps reveal spatial structure.

Fieldwork can inspect:

  • actual pedestrian routes;
  • barriers;
  • microclimate;
  • business relationships;
  • land-use transitions;
  • local explanations.

Spatial thinking becomes stronger when representation returns to the world.

Interventions Are Especially Informative

New station opens.

Road closes.

Flood barrier is built.

Tree cover is added.

Observed change after an intervention—especially relative to credible comparison locations—can provide stronger evidence about mechanism than static overlap alone.

Prediction Is a Hostile Test

If the explanation is correct, where should the next effect appear?

Where should it not appear?

How should it change with distance or direction?

A causal story earns credibility when it survives spatial predictions not used to fit the story initially.

National Geographic’s Spatial-Thinking Standard Supports This Move

National Geographic’s current Geography Standard 1 frames spatial thinking as analysing spatial patterns and organisation using geographic representations and geospatial data. The important word is analysing: seeing overlap is the beginning of reasoning, not the end.

A Better Spatial-Causation Model

observed overlap → denominator check → scale check → data-generation check → temporal order → mechanism → confounder search → comparison → prediction → intervention/world return → bounded causal claim

A 30-Lens Spatial Causation Audit

  1. Pattern A: what is mapped?
  2. Pattern B: what overlaps?
  3. Unit: individual, point, area or surface?
  4. Denominator: are raw counts misleading?
  5. Scale: at what level does association appear?
  6. Boundary: does aggregation create the relationship?
  7. Neighbourhood: how is proximity defined?
  8. Time: which pattern changes first?
  9. Mechanism: what process links them?
  10. Direction: does mechanism predict spatial direction?
  11. Distance: should effect decay with distance?
  12. Barrier: should a physical boundary interrupt it?
  13. Network: should effect follow connections?
  14. Confounder: what third process could create both?
  15. Population: is density driving both?
  16. Selection: do people or firms choose locations?
  17. Reverse cause: could B influence A?
  18. History: is the pattern inherited?
  19. Measurement: where are observations easier to collect?
  20. Proxy: does a layer measure the concept directly?
  21. Ecological level: are area results being applied to individuals?
  22. Dependence: are nearby observations non-independent?
  23. Comparison: what similar area provides a contrast?
  24. Intervention: did something change externally?
  25. Before/after: is temporal response visible?
  26. Prediction: what new pattern should appear?
  27. Falsifier: what result would weaken the theory?
  28. Alternative: which competing mechanism remains?
  29. Uncertainty: how strong is the claim?
  30. World return: did new spatial evidence behave as predicted?

Laboratory 1: Two Dark Maps

Take two fictional choropleth maps that both become darker in dense districts.

List five explanations, including population density, before allowing a causal arrow.

Laboratory 2: Before and After the Station

Map retail activity before and after a new station opens, then compare with a similar corridor without a new station.

What result would support the accessibility mechanism?

Laboratory 3: Area versus Person

Construct an area-level association, then invent individual-level data that contradict it.

Use the exercise to explain ecological fallacy.

For Primary Readers

If umbrellas and puddles appear in the same places, ask what third thing could cause both.

For Secondary Readers

For every spatial association, write one mechanism, one confounder and one observation that could distinguish them.

For Advanced Readers

Model spatial causation as inference under dependence, aggregation and endogenous location choice. Causal identification must survive confounding, reverse causation, spatial autocorrelation and the modifiable geography of measurement units.

Common Misconceptions

  • “If two maps overlap strongly, one caused the other.” Shared causes and population structure can create overlap.
  • “Correlation means nothing.” Association is valuable for generating and testing explanations.
  • “Controlling for several variables proves causation.” Model assumptions and omitted processes still matter.
  • “Area-level patterns tell us about every individual.” This can commit an ecological fallacy.
  • “A current pattern reflects a current cause.” Historical path dependence can preserve old spatial structures.

Research Corridor

Frequently Asked Questions

What is spatial association?

It is a relationship in the geographic arrangement of two or more variables or events—for example, high values occurring in similar places.

Why doesn’t spatial association prove causation?

Because overlapping patterns can arise from shared causes, population density, selection, measurement design, boundaries, history or reverse causation.

What makes a spatial causal claim stronger?

Clear temporal order, a plausible mechanism, credible comparison, attention to confounders and successful predictions about where effects should and should not occur.

Final Thought: A Map Can Draw the Question Before It Draws the Arrow

Two patterns can meet perfectly.

The explanation can still be wrong.

Spatial intelligence is the discipline of enjoying the pattern enough to investigate it—and distrusting it enough to test the mechanism.

SPATIAL THINKING · FOUR PILLAR LEGS

Return to Spatial Thinking, or continue through Spatial Distribution, Mental Maps and Spatial Data. Return to World & Knowledge.

Discover more from eduKate Singapore

Subscribe now to keep reading and get access to the full archive.

Continue reading