Cuzick–Edwards Test¶

A case-control nearest-neighbor significance test for detecting spatial clustering of cases within an already nonuniform background population represented by control locations.

Version: v1 · 2026-09-28 · History
Domain-specific #: 8820
Domain group: Formal Sciences
Origin domain: Experimental Design & Statistics
Subdomains: Spatial Statistics, Spatial Epidemiology → Experimental Design & Statistics

Core Idea¶

The Cuzick–Edwards test asks whether cases cluster spatially beyond the clustering already present in the population from which they arise. It uses control locations to represent that background, then constructs nearest-neighbor statistics such as the number of case neighbors around each case or the number of cases encountered before a specified nearest control. Inference conditions on the observed point pattern and examines whether case labels are more locally aggregated than expected. Inference conditions on the observed point pattern and examines whether case labels are more locally aggregated than expected.

How would you explain it like I'm…

Sick-Dot Neighbor Check

Put red dots on a map for sick people and blue dots for healthy people. People already live bunched up in towns, so bunching by itself tells you nothing. This test asks: is a red dot's closest neighbor red more often than luck would give? If so, the sick people are clumped together, but that still doesn't tell you why.

Nearest-Neighbor Sickness Check

The Cuzick–Edwards test checks whether sick people (cases) are clumped together on a map more than you would expect. Since people already live in crowded towns and empty countryside, it uses healthy people (controls) to show where people normally live. For each case, it looks at that person's nearest neighbors and counts how many are also cases. If cases have case-neighbors much more often than chance would give, the test says the cases are clustered. It doesn't say the illness spreads from person to person or what causes it.

Case–Control Neighbor Clustering Test

The Cuzick–Edwards test asks whether cases of some condition cluster in space beyond the clustering already present in the population they come from. It uses control locations—people without the condition—to represent that background population. For each case, it looks at its k nearest neighbors among all cases and controls and counts how many are cases; summing gives the test statistic. The locations are treated as fixed, and the question is whether the "case" labels are more bunched together than a random labeling would produce. The number of neighbors k sets the spatial scale, and trying several values of k can find more clusters but makes judging significance harder. A significant result shows that case status depends on location, not that the condition is contagious or has a particular environmental cause.

The Cuzick–Edwards test is a nearest-neighbor test for spatial clustering of cases relative to controls. Controls sampled from the same underlying population represent its spatial distribution, so ordinary population clustering is not mistaken for disease clustering. The statistic is built from neighbor relationships—for example, the number of cases among each case's k nearest neighbors, summed over cases, or a variant counting cases met before reaching a specified nearest control. Inference conditions on the observed set of point locations and asks whether the case labels are more locally aggregated than under random allocation of labels to those points. The neighbor order k fixes the spatial scale being probed; testing several k values can increase sensitivity but requires handling multiple comparisons. A significant result is evidence of spatial dependence in case status, not proof of contagion, an environmental cause, or a specific hotspot mechanism.

Scope of Application¶

Use the test with explicit case/control sampling, coordinates, neighbor order, null mechanism, multiple-testing handling, and spatial interpretation. Use the test with explicit case/control sampling, coordinates, neighbor order, null mechanism, multiple-testing handling, and spatial interpretation.

Spatial epidemiology. Tests disease clustering.
Public health. Compares cases with population controls.
Environmental studies. Screens local aggregation.
Demography. Handles nonuniform settlement.
Biostatistics. Develops labeled point-pattern tests.

Clarity¶

Controls are not a nuisance sample; they encode the spatial distribution of the population at risk and therefore determine the null. The closest near miss sets the boundary: A case-only nearest-neighbor test is closest: it can detect point aggregation but cannot adjust in the same way for an underlying population that is itself spatially clustered.

Manages Complexity¶

Selection bias, geocoding error, duplicated households, mobility, scale choice, and multiple k values can affect results. Statistical clustering should be followed by substantive study rather than causal storytelling. The central background adjustment–control quality tradeoff is this: Controls solve population clustering only if they represent the at-risk population. A second scale sensitivity–multiple testing tension matters because Different k values reveal different patterns and inflate exploration.

Abstract Reasoning¶

Use three linked moves: define cases, population at risk, and control sampling; validate spatial coordinates and duplicates; choose neighbor statistic and k values. As a collapse test, the case exits when controls do not represent the background population or the specified case-control neighbor statistic is not used. A fourth check is to generate the conditional null distribution.

Knowledge Transfer¶

Labeled-neighbor comparison transfers to other marked point patterns, but case-control sampling and clustered population risk delimit the test. The nearest stopping boundary is explicit: A case-only nearest-neighbor test is closest: it can detect point aggregation but cannot adjust in the same way for an underlying population that is itself spatially clustered. The inclusion test remains: An analysis is a Cuzick–Edwards test when spatial case labels are compared with control locations through its nearest-neighbor statistic under a conditional case-control null. The structure no longer applies when the case exits when controls do not represent the background population or the specified case-control neighbor statistic is not used. No canonical parent prime is currently asserted; broader structural comparisons remain related-prime analogies until separately adjudicated in the DAG. The alternative is local case aggregation.

Neighborhood in Abstraction Space¶

Cuzick–Edwards Test sits in a moderately populated region (55^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Empirical Measurement & Statistical Inference Methods (50 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08