Large places tend to attract more interaction. Distant places tend to attract less.
That deceptively simple pattern sits underneath a large family of geographic models. Big cities exchange more passengers and goods than tiny settlements. Nearby job centres usually attract more commuters than equally attractive centres much farther away. Large markets pull strongly, but distance adds friction.
A gravity model formalises that intuition by estimating interaction between places from their relative size or attractiveness and the friction of separation between them.
Gravity models turn the geography of attraction and distance into a prediction about flow.
Quick Read: The Gravity Mechanism
ORIGIN MASS × DESTINATION MASS × DISTANCE-FRICTION FUNCTION → EXPECTED INTERACTION
In broad form, interaction between origin i and destination j can be represented as:
Tij = k × f(origin size, destination size, separation)
The exact formula can vary. The geographic job stays the same: estimate how much two places should interact given what each place offers and how hard it is to move between them.
1. Why It Is Called a Gravity Model
The name comes from an analogy with Newtonian gravity. Larger bodies exert stronger attraction; greater separation weakens that attraction. Geography borrows the structure, not the physics. Cities do not pull travellers through a literal gravitational force.
2. The “Mass” of a Place Is Contextual
Population may represent origin size in a migration model. Jobs may represent destination attractiveness in a commuting model. Retail floor space may represent store attraction. GDP may represent economic size in a trade model. There is no universal geographic mass variable.
3. Separation Is More Than Kilometres
Distance can be measured as kilometres, travel time, ticket price, toll, border friction or another impedance. The correct measure depends on what is moving and which barriers matter.
4. Distance Friction Is Usually Nonlinear
Interaction rarely declines one-for-one with distance. A five-minute increase may matter little on a long intercity journey and greatly on a short daily commute. Models therefore use friction functions or exponents that control how rapidly interaction decays with separation.
5. Calibration Turns the Model From Analogy Into Measurement
Parameters are usually estimated from observed flows. Calibration asks what size effects and distance penalties best reproduce real movement in the study system. A commuting model calibrated in one region should not be assumed valid everywhere else.
6. Gravity Models Sit Inside Spatial Interaction
Spatial Interaction owns the broad mechanism of flows between places. Gravity models own one formal modelling family that predicts those flows using origin attributes, destination attributes and separation.
7. Gravity Models Are Not Flow Maps
Flow Mapping owns the visual representation of movement. Gravity models estimate or explain the quantity that may later be drawn as a flow line.
8. Gravity Models Are Not Distance Decay Alone
Distance Decay owns the general weakening of interaction with separation. Gravity models combine that friction with the size or attractiveness of origins and destinations.
9. Complementarity Still Matters
Two large nearby places need a reason to interact. Residential areas supply workers; employment centres supply jobs. Producers supply goods; consumers demand them. Size and distance are powerful predictors only when there is something worth exchanging.
10. Intervening Opportunities Can Break the Simple Prediction
A large attractive destination far away may receive less interaction if a closer alternative provides the same opportunity. Gravity alone can overpredict long-distance flows when intervening choices matter.
11. Transferability Can Break It Too
Even strong attraction is irrelevant if movement is impossible or too costly. Borders, missing infrastructure, congestion, fares and capacity can suppress interaction between otherwise complementary places.
12. Origin-Constrained Models
Sometimes the total number of trips leaving each origin is already known. A constrained gravity model allocates those trips across destinations rather than predicting the origin total from scratch.
13. Destination-Constrained Models
In other applications, destination totals are fixed—for example, known employment capacity or school places. The model distributes origins among those destinations while respecting the destination totals.
14. Doubly Constrained Models
When both origin and destination totals are known, the model estimates the origin–destination matrix that best matches those margins while accounting for distance friction.
15. Gravity Models Can Predict Missing Origin–Destination Flows
Observed origin–destination matrices are expensive to collect and quickly become outdated. Calibrated models can estimate flows where surveys are incomplete or simulate how movement might change after population growth, new infrastructure or altered travel cost.
16. Primary Geography: Bigger and Closer Wins More Often
Give pupils three shops: one small and nearby, one large and farther away, one large and nearby. Ask which might attract more customers and why. They discover that both size and separation matter at the same time.
17. Secondary Geography: Compare Size and Distance
Students can rank likely flows between towns using population and travel distance. Then introduce a new motorway or closer competitor and ask how the predicted pattern changes.
18. Advanced Geography: Friction Functions
Distance deterrence can be represented with power, exponential or other functions. Different functions imply different behavioural sensitivity to separation. The choice should be calibrated against observed interaction rather than selected merely for mathematical convenience.
19. Advanced Geography: Production and Attraction Terms
Modern spatial-interaction models often distinguish the ability of an origin to generate trips from the ability of a destination to attract them. Working-age population may generate commuting; job availability may attract it. The two sides should not be collapsed into one vague measure of “size.”
20. Singapore Example: Commuting
Residential towns generate commuting while employment centres attract it. Travel time, transfer burden and transport quality create separation friction. A gravity-style model can estimate the expected distribution of journeys across origin–destination pairs, then observed flows can reveal where other factors matter.
21. Singapore Example: Retail Attraction
A large retail centre may draw customers from farther away than a small neighbourhood shop. Yet nearer alternatives, transit connectivity and differentiated retail offerings modify the simple size–distance relationship.
22. Migration Example
Large origin and destination populations often generate more migration opportunities, while greater distance suppresses movement. Family networks, immigration rules, income gaps and language can create substantial deviations from a simple gravity prediction.
23. Trade Example
Economic gravity models commonly relate trade between places to economic size and separation. Border institutions, tariffs, shared language, logistics and agreements can alter effective friction beyond physical distance.
24. Airport Example
Large metropolitan regions support more passenger demand, but airline networks do not connect every pair directly. Hub structure, slot availability and route economics mean network topology can matter alongside geographic distance.
25. Hostile Test: “The Biggest Nearby Place Should Receive the Most Flow”
Often—but not always. The model is a baseline. Intervening opportunities, specialised destinations, borders, network structure, capacity and behaviour can overturn the simple ranking.
26. Hostile Test: “The Model Fits, So Distance Causes the Flow”
Fit is not causal proof. Distance may proxy travel time, cost, social separation or infrastructure. Large destinations may correlate with many omitted attractions. The model can predict well while leaving mechanism partly unresolved.
27. Where Gravity-Model Reasoning Breaks
- Physics literalism: treating human mobility as actual gravitational force.
- Mass ambiguity: using population when jobs, capacity or another attraction measure is the real mechanism.
- Euclidean-distance default: ignoring time, price, borders or network impedance.
- Intervening-opportunity blindness: overpredicting distant destinations when closer substitutes exist.
- Calibration transfer: exporting parameters from one region or period without testing them.
- Fit-equals-cause: treating predictive success as causal identification.
- Capacity blindness: allowing a destination to attract more flow than it can actually absorb.
- Static-world assumption: ignoring how new infrastructure or land-use change reshapes the friction surface.
28. Ten Questions for Gravity Models
- What interaction is being modelled?
- What generates flow at the origin?
- What attracts flow at the destination?
- How is separation measured?
- What friction function is used?
- Are origin or destination totals constrained?
- Which intervening opportunities exist?
- What infrastructure or policy changes transferability?
- How was the model calibrated and validated?
- Where do observed flows systematically depart from the prediction?
29. Research Anchor
Transport geography commonly formulates spatial interaction as a function of origin attributes, destination attributes and separation or friction between them. Gravity models are a standard form in which interaction rises with place importance and falls as separation becomes more costly.
30. Where This Fits
Spatial Interaction owns the broad conditions and mechanisms that generate movement. Flow Mapping owns visual representation of observed movement. Distance Decay owns the general weakening of interaction with friction. This article owns the formal size–separation model used to estimate interaction strength between places.
The Idea to Keep
Gravity models are useful because they establish a disciplined baseline: bigger attracts more, friction suppresses more—and the interesting geography often begins where reality departs from that baseline.