A red patch on a map is not automatically a hot spot.
It may simply contain a high value. A true statistical hot spot is a location where high values occur together with other high values strongly enough that the local concentration is unlikely under the chosen null model. Low values can form cold spots by the same logic.
Hot spot analysis uses local spatial statistics—commonly including Getis–Ord Gi*—to identify statistically significant clusters of high or low values.
A hot spot is not “the reddest place.” It is a place whose neighbourhood is unusually high under an explicit spatial comparison.
Quick Read: The Hot-Spot Mechanism
VALUES + SPATIAL WEIGHTS → LOCAL NEIGHBOURHOOD SUM → EXPECTED RANDOM PATTERN → Z-SCORE / SIGNIFICANCE → HOT / COLD / NOT SIGNIFICANT
1. High Value Is Not Enough
One extremely high observation surrounded by ordinary values may be a spatial outlier rather than a hot spot. Hot spot statistics examine the local neighbourhood, not only the focal value.
2. Spatial Weights Define the Neighbourhood
Spatial Weights owns the rule defining which places count as neighbours. Change the distance band, contiguity rule or network relationship and the hot-spot result can change.
3. Getis–Ord Gi* Compares Local and Global Structure
Conceptually, Gi* asks whether the values around a location are collectively high or low relative to the study area. Large positive standardised scores indicate high-value clustering; large negative scores indicate low-value clustering.
4. Significance Depends on a Null Model
A p-value or z-score has meaning only under assumptions about how values could have been arranged. Statistical significance is not an intrinsic property painted onto a neighbourhood.
5. Multiple Testing Matters
When hundreds or thousands of locations are tested, some may appear significant by chance. False-discovery-rate or related corrections can help control the number of false positives.
6. Hot Spot Analysis Is Not KDE
KDE smooths point events into an intensity surface. Hot spot analysis tests whether values form statistically unusual local clusters. A KDE peak can exist without a significance test; a hot spot is defined through statistical comparison.
7. Hot Spot Analysis Is Local Autocorrelation’s Close Relative
Spatial Autocorrelation owns the broad question of spatial dependence. Hot spot analysis owns one local inferential task: identifying unusually concentrated high or low neighbourhoods.
8. Counts Need Exposure
High numbers of incidents often occur where more people, vehicles or businesses exist. Raw-count hot spots can therefore be population hot spots in disguise. Rates or exposure-adjusted measures may be more appropriate.
9. Scale Changes the Result
A cluster visible at one kilometre may disappear at five kilometres. Local spatial processes operate at characteristic scales, so neighbourhood size should be tested rather than chosen casually.
10. Primary Geography: High Beside High
Give pupils a grid of numbers. One cell has 10 but all neighbours have 1. Another region contains several 8s and 9s together. Which looks more like a neighbourhood of high values? The distinction between high point and high cluster becomes intuitive.
11. Secondary Geography: Compare Raw and Rate Maps
Students can compare incident counts with incidents per population or traffic volume. A supposed hot spot may weaken once exposure is considered.
12. Advanced Geography: Robustness Across Neighbourhood Definitions
A strong conclusion should not depend entirely on one arbitrary distance. Analysts can test several plausible spatial-weight specifications and report where hot spots persist.
13. Singapore Example: Heat
Neighbouring urban cells with consistently high temperatures may form a statistically significant heat hot spot. The result identifies where high values cluster; urban form, vegetation and ventilation must still explain why.
14. Singapore Example: Traffic Incidents
High crash counts around busy junctions may reflect traffic exposure. Testing rates or modelling expected incidents can distinguish unusually dangerous areas from merely busy ones.
15. Public Health Example
Clusters of high disease rates can guide investigation, but significance does not establish exposure or transmission. Demography, diagnosis and reporting must still be examined.
16. Hostile Test: “It Is Significant, So It Must Be Important”
Statistical significance is not practical importance. A tiny but precisely estimated difference can be significant, while a consequential local pattern may fail significance because data are sparse. Magnitude, uncertainty and decision context all matter.
17. Where Hot-Spot Reasoning Breaks
- Red-equals-hotspot: confusing map colour with statistical clustering.
- High-value collapse: treating one extreme observation as a neighbourhood cluster.
- Default-neighbourhood error: accepting software distance settings without mechanism.
- Multiple-testing blindness: ignoring false positives across many local tests.
- Exposure blindness: mapping counts where denominators vary strongly.
- Significance-cause collapse: treating a significant cluster as an explanation.
18. Ten Questions for Hot Spot Analysis
- What variable is being tested?
- Are values counts, rates or measurements?
- How are neighbours defined?
- What spatial scale is plausible?
- What is the null model?
- Are multiple tests corrected?
- Does exposure vary spatially?
- Do hot spots persist under alternative weights?
- How large is the effect?
- What mechanism could explain the cluster?
19. Where This Fits
Spatial Autocorrelation owns spatial dependence broadly. Spatial Weights owns neighbourhood definition. KDE owns smoothed point intensity. This article owns statistical identification of local high-value and low-value clusters.
The Idea to Keep
A hot spot deserves the name only after the neighbourhood, benchmark and uncertainty have all been made explicit.