Hot Spot Analysis (Getis-Ord Gi*)
Also known as: Getis-Ord Gi* statistic, spatial hot spot detection, cluster and outlier analysis, HSA
Hot Spot Analysis uses the Getis-Ord Gi* local spatial statistic to identify geographic locations where high or low attribute values cluster together to a degree that is statistically significant. Each feature is evaluated in relation to its neighbours, producing a z-score that flags genuine spatial hot spots and cold spots against a background of random variation.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
+20 more
When to use it
Use Hot Spot Analysis when you have georeferenced data with a continuous or count attribute and you want to pinpoint where statistically significant spatial clusters of high or low values occur — for example, identifying disease burden concentrations, crime clusters, or economic activity hot spots. The method requires at least 30 features for reliable inference and assumes attribute values are not uniformly zero. Avoid it when spatial dependence is irrelevant to the research question, when the spatial support (polygon size) varies wildly across the study area without value standardisation, or when the underlying process is known to be purely random.
Strengths & limitations
- Identifies statistically significant spatial clusters rather than merely mapping high values, reducing false positives from visual inspection.
- Produces an interpretable z-score and confidence-level map that is immediately usable for decision-making and communication.
- Includes the feature itself in the neighbourhood sum (Gi*), making it more sensitive to true hot spots than the earlier Gi variant.
- Flexible neighbourhood definition (distance band, k-NN, contiguity) allows the analyst to match the spatial scale of the hypothesised process.
- Implemented in widely used tools (ArcGIS, GeoDa, R spdep, PySAL), lowering the barrier to application.
- Results are sensitive to the spatial weights specification; different neighbourhood distances can produce materially different hot spot patterns.
- Assumes stationarity of the underlying process across the study area; non-stationarity can produce spurious clusters.
- Multiple simultaneous tests across all features inflate the type I error rate; FDR or Bonferroni corrections are often omitted in practice.
- Does not distinguish between types of spatial patterns (e.g., a cluster of high values vs. a single extreme outlier surrounded by moderate values).
- Requires a reasonably large and spatially distributed sample; sparse or irregular point patterns reduce reliability.
Frequently asked
What is the difference between Hot Spot Analysis and Local Moran's I?
Both are local spatial statistics but they detect different patterns. The Gi* statistic used in Hot Spot Analysis only identifies clusters of high values (hot spots) or low values (cold spots). Local Moran's I additionally identifies spatial outliers — features with high values surrounded by low values, and vice versa — making it more informative for detecting anomalies but harder to communicate simply.
How do I choose the right neighbourhood distance?
A principled approach is to run an Incremental Spatial Autocorrelation analysis: apply Moran's I at increasing distance bands and choose the distance where the global z-score peaks, indicating the spatial scale at which clustering is strongest. An alternative starting point is the mean or median nearest-neighbour distance of the feature set.
Should I correct for multiple comparisons?
Yes. When you test every feature in a large dataset simultaneously at p < 0.05, you expect roughly 5% of features to appear significant by chance. The False Discovery Rate (FDR) correction is widely recommended for Hot Spot Analysis because it balances sensitivity and type I error control better than a strict Bonferroni correction for spatial data.
Can I use Hot Spot Analysis with count data or binary data?
The Gi* statistic can be applied to count attributes (e.g., number of incidents per polygon), but binary (0/1) data rarely produce reliable results because the distribution assumptions are poorly met. For sparse count data, consider aggregating to a coarser spatial unit or using a different approach such as kernel density estimation.
How many features do I need for reliable results?
The general recommendation is a minimum of 30 features, though larger samples (100+) provide more stable z-score distributions. With very few features the normal approximation underlying the z-score is unreliable, and small changes in the data can shift results dramatically.
Sources
- Getis, A., & Ord, J. K. (1992). The analysis of spatial association by use of distance statistics. Geographical Analysis, 24(3), 189-206. DOI: 10.1111/j.1538-4632.1992.tb00261.x ↗
- Ord, J. K., & Getis, A. (1995). Local spatial autocorrelation statistics: Distributional issues and an application. Geographical Analysis, 27(4), 286-306. DOI: 10.1111/j.1538-4632.1995.tb00912.x ↗
How to cite this page
ScholarGate. (2026, June 3). Hot Spot Analysis (Getis-Ord Gi*). ScholarGate. https://scholargate.app/en/spatial-analysis/hot-spot-analysis
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Geographically Weighted RegressionSpatial analysis↔ compare
- Local Indicators of Spatial AssociationSpatial analysis↔ compare
- Local Moran's ISpatial analysis↔ compare
- Moran's ISpatial analysis↔ compare
- Spatial AutocorrelationSpatial analysis↔ compare