Moran's I & LISA: Detecting Spatial Clusters Interactively

simulator intermediate ~10 min
Loading simulation...
Moran's I = 0.45 — significant clustering (z = 3.2)

With 50 spatial units and autocorrelation strength 0.6, Moran's I = 0.45 with z-score 3.2, indicating highly significant positive spatial autocorrelation. LISA identifies local hot spots (high-high) and cold spots (low-low).

Formula

I = (n/W) Σᵢ Σⱼ wᵢⱼ(xᵢ−x̄)(xⱼ−x̄) / Σ(xᵢ−x̄)²
Iᵢ = (xᵢ−x̄)/s² Σⱼ wᵢⱼ(xⱼ−x̄) (local Moran's)
E[I] = −1/(n−1) under randomization

Is the Pattern Random or Clustered?

When you see a map of disease rates, crime incidents, or mineral grades, the first question is always: is the spatial pattern significant, or could it have occurred by chance? Moran's I statistic, introduced by Patrick Moran in 1950, provides a rigorous answer. It measures the overall degree of spatial autocorrelation — the tendency for nearby locations to have similar (or dissimilar) values. A significantly positive Moran's I confirms what your eye suspects: the pattern is clustered.

From Global to Local

While Moran's I gives a single global summary, Luc Anselin's LISA (1995) reveals where clustering occurs. LISA decomposes the global statistic into local contributions, identifying four distinct spatial regimes. High-high clusters (hot spots) show concentrations of high values — disease outbreak centers, mineral-rich zones, or high-crime neighborhoods. Low-low clusters (cold spots) show concentrations of low values. High-low and low-high spatial outliers mark locations that differ sharply from their surroundings — a wealthy neighborhood in a poor region, or a low-grade drill hole in an otherwise rich ore zone.

Interactive Exploration

In this simulation, each circle represents a spatial unit colored by its value (red=high, blue=low). Lines connect neighbors within the distance threshold. The global Moran's I and z-score update in real time as you adjust the spatial autocorrelation strength. Increasing ρ makes similar values cluster together; decreasing it creates a random salt-and-pepper pattern. LISA significance is shown by circle borders — thick borders indicate locally significant clusters or outliers at the 95% confidence level.

Applications in Science and Policy

Spatial clustering analysis is ubiquitous in modern research and decision-making. Epidemiologists use LISA to identify disease hot spots for targeted intervention. Criminologists detect crime clusters to optimize police patrol allocation. Ecologists find biodiversity hot spots and cold spots. Real estate analysts identify neighborhoods with unusual price patterns. Environmental scientists locate contamination clusters. In each case, the statistical significance test ensures that identified patterns are genuine rather than artifacts of random variation.

FAQ

What is Moran's I?

Moran's I is the most common test for global spatial autocorrelation. It ranges from -1 (perfect dispersion) through 0 (randomness) to +1 (perfect clustering). A significant positive value indicates that similar values tend to be located near each other — spatial clustering exists.

What is LISA?

LISA (Local Indicators of Spatial Association) decomposes global Moran's I into contributions from each location. It identifies four types of local patterns: high-high clusters (hot spots), low-low clusters (cold spots), high-low outliers, and low-high outliers. Each location is tested for local significance.

How is the spatial weight matrix defined?

The spatial weight matrix W defines which locations are neighbors. Common approaches include contiguity (sharing a border), distance threshold (within a specified distance), or k-nearest neighbors. The choice of weight matrix affects both Moran's I and LISA results.

What are hot spots and cold spots?

In LISA analysis, hot spots are locations with high values surrounded by high-value neighbors (high-high). Cold spots have low values surrounded by low-value neighbors (low-low). Spatial outliers are locations that differ markedly from their neighbors — high-low or low-high combinations.

Sources

Embed

<iframe src="https://homo-deus.com/lab/geostatistics/spatial-clustering/embed" width="100%" height="400" frameborder="0"></iframe>
View source on GitHub