Before Geography can measure whether neighbours resemble one another, it has to decide who the neighbours are.
Two districts can share a border. Two houses can sit within five hundred metres. Two cities can be connected by a direct flight. Two communities can be separated physically yet linked by a bridge. Each definition creates a different geography of neighbourhood.
Spatial weights are the formal rules geographers use to represent which locations are related and how strongly one location should count when analysing another.
A spatial statistic cannot know what “nearby” means until someone defines nearby.
Quick Read: The Spatial-Weights Mechanism
LOCATIONS → NEIGHBOUR RULE → WEIGHT MATRIX → SPATIAL RELATIONSHIP → STATISTIC / MODEL
1. Contiguity: Sharing a Boundary
Polygon areas can be treated as neighbours when they touch. Rook contiguity requires a shared edge; queen contiguity can also count shared corners. The names come from chess movement and encode different definitions of adjacency.
2. Distance Bands
Locations within a chosen distance can be neighbours. This works well when influence is plausibly local, but the threshold matters. A one-kilometre band and five-kilometre band describe different interaction fields.
3. Inverse Distance
Instead of a hard cutoff, influence can decline with distance. Nearby places receive larger weights; distant places receive smaller ones. This connects naturally to Distance Decay.
4. k-Nearest Neighbours
Each location can be linked to its k closest neighbours. This guarantees every observation has a neighbourhood, useful where point density varies strongly. But in sparse regions the kth neighbour may be very far away.
5. Network Neighbours
Road travel time, rail connection, river flow or airline routes may define relationship better than straight-line distance. Geographic proximity should match the mechanism under study.
6. Direction Can Matter
Upstream can influence downstream more than downstream influences upstream. Downwind exposure is directional. A symmetric weights matrix can therefore be inappropriate for directional processes.
7. Binary and Continuous Weights
A binary matrix says neighbour or not-neighbour. Continuous weights express degrees of relationship. The choice should follow the process rather than convenience.
8. Row Standardisation
Spatial weights are often standardised so each location’s neighbour weights sum to one. This makes the spatial lag interpretable as a weighted neighbourhood average, but standardisation changes the meaning of the raw connections and should be stated.
9. Spatial Weights Sit Under Spatial Autocorrelation
Spatial Autocorrelation asks whether nearby values resemble one another. Spatial weights own the prior modelling decision that defines which observations count as nearby and how much.
10. Different Weight Matrices Can Produce Different Answers
A cluster may look strong under adjacency and weaker under travel-time weights. That is not necessarily a contradiction. The two analyses ask different neighbourhood questions.
11. Islands Need Special Treatment
An area with no shared boundary neighbours can become an isolated row in a contiguity matrix. Analysts may need distance or network rules to represent meaningful connections rather than silently leaving the observation disconnected.
12. Primary Geography: Who Counts as Next Door?
Draw houses on a map and ask children to define neighbours by touching plots, five-minute walking distance or same bus stop. Different rules produce different neighbourhoods from the same houses.
13. Secondary Geography: Compare Adjacency and Travel Time
Students can ask whether two planning areas are neighbours because they share a border or because residents can move easily between them. The exercise shows that geographic relationship is conceptual, not merely geometric.
14. Advanced Geography: Sensitivity to W
Spatial statistics are often written using a weights matrix W. Good analysis tests whether conclusions are robust to several defensible specifications of W rather than selecting one silently because it produces the desired result.
15. Singapore Example: Planning Areas
For local spillovers, shared borders may be useful. For commuting, MRT or bus travel time may be more meaningful. For coastal or atmospheric processes, physical flow may define the relationship differently again.
16. Disease Example
Administrative adjacency may poorly represent infection contact if people travel through transport hubs. Mobility-based weights can sometimes align better with the transmission pathway.
17. River Example
Watersheds create directed upstream–downstream relationships. Euclidean distance alone can connect locations that are close across a ridge but hydrologically unrelated.
18. Hostile Test: “Moran’s I Says It Is Clustered”
Under which spatial weights? A spatial statistic without its neighbourhood definition is incomplete. Change the weights and the answer may change because the geographic hypothesis changed.
19. Where Spatial-Weights Reasoning Breaks
- Default-W fallacy: accepting software defaults without geographic justification.
- Distance worship: using Euclidean distance for a network or flow process.
- Symmetry assumption: ignoring directional influence.
- Threshold arbitrariness: choosing a distance band without sensitivity checks.
- Disconnected-island blindness: leaving observations with no neighbours unintentionally.
- Result shopping: selecting the weight matrix that produces the preferred significance.
20. Ten Questions for Spatial Weights
- What mechanism connects locations?
- Should adjacency or distance define neighbours?
- Would travel time be better?
- Is the relationship directional?
- Should weights be binary or continuous?
- What threshold or k is defensible?
- How are isolated observations handled?
- Are weights standardised?
- Do conclusions survive alternative plausible matrices?
- Does the chosen W match the geographic story?
21. Where This Fits
Spatial Autocorrelation owns whether nearby values resemble one another. Spatial Interaction owns flows between places. This article owns the formal definition of neighbourhood used by spatial statistics and models.
The Idea to Keep
In spatial analysis, “nearby” is not a fact until you explain what kind of nearness the process actually uses.